Merge remote-tracking branch 'origin/master' into soc-2019-npr

1343 files changed, 90367 insertions, 33599 deletions

diff --git a/.clang-format b/.clang-format
index bbbe3f1bac4..b81403c46ce 100644
--- a/.clang-format
+++ b/.clang-format
@@ -224,7 +224,9 @@ ForEachMacros:
   - LISTBASE_CIRCULAR_BACKWARD_BEGIN
   - LISTBASE_CIRCULAR_FORWARD_BEGIN
   - LISTBASE_FOREACH
+  - LISTBASE_FOREACH_BACKWARD
   - LISTBASE_FOREACH_MUTABLE
+  - LISTBASE_FOREACH_BACKWARD_MUTABLE
   - MAN2D_ITER_AXES_BEGIN
   - MAN_ITER_AXES_BEGIN
   - NODE_INSTANCE_HASH_ITER
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 16ac322ebdd..2939d33c4a8 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1430,7 +1430,6 @@ if(CMAKE_COMPILER_IS_GNUCC)
   ADD_CHECK_CXX_COMPILER_FLAG(CXX_WARNINGS CXX_WARN_NO_DIV_BY_ZERO -Wno-div-by-zero)
   ADD_CHECK_CXX_COMPILER_FLAG(CXX_WARNINGS CXX_WARN_TYPE_LIMITS -Wtype-limits)
   ADD_CHECK_CXX_COMPILER_FLAG(CXX_WARNINGS CXX_WARN_ERROR_RETURN_TYPE  -Werror=return-type)
-  ADD_CHECK_CXX_COMPILER_FLAG(CXX_WARNINGS CXX_WARN_ERROR_IMPLICIT_FUNCTION_DECLARATION -Werror=implicit-function-declaration)
   ADD_CHECK_CXX_COMPILER_FLAG(CXX_WARNINGS CXX_WARN_NO_CHAR_SUBSCRIPTS -Wno-char-subscripts)
   ADD_CHECK_CXX_COMPILER_FLAG(CXX_WARNINGS CXX_WARN_NO_UNKNOWN_PRAGMAS -Wno-unknown-pragmas)
   ADD_CHECK_CXX_COMPILER_FLAG(CXX_WARNINGS CXX_WARN_POINTER_ARITH -Wpointer-arith)
diff --git a/GNUmakefile b/GNUmakefile
index 4462a13207e..ef1b6b711d2 100644
--- a/GNUmakefile
+++ b/GNUmakefile
@@ -233,12 +233,19 @@ endif
 ifneq "$(findstring ninja, $(MAKECMDGOALS))" ""
 	BUILD_CMAKE_ARGS:=$(BUILD_CMAKE_ARGS) -G Ninja
 	BUILD_COMMAND:=ninja
+	DEPS_BUILD_COMMAND:=ninja
 else
 	ifneq ("$(wildcard $(BUILD_DIR)/build.ninja)","")
 		BUILD_COMMAND:=ninja
 	else
 		BUILD_COMMAND:=make -s
 	endif
+
+	ifneq ("$(wildcard $(DEPS_BUILD_DIR)/build.ninja)","")
+		DEPS_BUILD_COMMAND:=ninja
+	else
+		DEPS_BUILD_COMMAND:=make -s
+	endif
 endif
 
 # -----------------------------------------------------------------------------
@@ -333,7 +340,7 @@ deps: .FORCE
 
 	@echo
 	@echo Building dependencies ...
-	$(BUILD_COMMAND) -C "$(DEPS_BUILD_DIR)" -j $(NPROCS) $(DEPS_TARGET)
+	$(DEPS_BUILD_COMMAND) -C "$(DEPS_BUILD_DIR)" -j $(NPROCS) $(DEPS_TARGET)
 	@echo
 	@echo Dependencies successfully built and installed to $(DEPS_INSTALL_DIR).
 	@echo
@@ -368,7 +375,7 @@ package_archive: .FORCE
 # Tests
 #
 test: .FORCE
-	cd $(BUILD_DIR) ; ctest . --output-on-failure
+	python3 ./build_files/utils/make_test.py "$(BUILD_DIR)"
 
 # run pep8 check check on scripts we distribute.
 test_pep8: .FORCE
@@ -524,17 +531,7 @@ icons_geom: .FORCE
 	    "$(BLENDER_DIR)/release/datafiles/blender_icons_geom_update.py"
 
 update: .FORCE
-	if [ "$(OS_NCASE)" = "darwin" ] && [ ! -d "../lib/$(OS_NCASE)" ]; then \
-		svn checkout https://svn.blender.org/svnroot/bf-blender/trunk/lib/$(OS_NCASE) ../lib/$(OS_NCASE) ; \
-	fi
-	if [ -d "../lib" ]; then \
-		svn cleanup ../lib/* ; \
-		svn update ../lib/* ; \
-	fi
-	git pull --rebase
-	git submodule update --init --recursive
-	git submodule foreach git checkout master
-	git submodule foreach git pull --rebase origin master
+	python3 ./build_files/utils/make_update.py
 
 format: .FORCE
 	PATH="../lib/${OS_NCASE}/llvm/bin/:$(PATH)" \
diff --git a/build_files/build_environment/CMakeLists.txt b/build_files/build_environment/CMakeLists.txt
index 1b387cb86a2..9756ad28454 100644
--- a/build_files/build_environment/CMakeLists.txt
+++ b/build_files/build_environment/CMakeLists.txt
@@ -43,6 +43,7 @@ project("BlenderDependencies")
 cmake_minimum_required(VERSION 3.5)
 
 include(ExternalProject)
+include(cmake/check_software.cmake)
 include(cmake/options.cmake)
 include(cmake/versions.cmake)
 
@@ -57,7 +58,6 @@ else()
 endif()
 
 include(cmake/zlib.cmake)
-include(cmake/blendthumb.cmake)
 include(cmake/openal.cmake)
 include(cmake/png.cmake)
 include(cmake/jpeg.cmake)
diff --git a/build_files/build_environment/cmake/blendthumb.cmake b/build_files/build_environment/cmake/blendthumb.cmake
deleted file mode 100644
index f4cc6ee9d37..00000000000
--- a/build_files/build_environment/cmake/blendthumb.cmake
+++ /dev/null
@@ -1,67 +0,0 @@
-# ***** BEGIN GPL LICENSE BLOCK *****
-#
-# This program is free software; you can redistribute it and/or
-# modify it under the terms of the GNU General Public License
-# as published by the Free Software Foundation; either version 2
-# of the License, or (at your option) any later version.
-#
-# This program is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program; if not, write to the Free Software Foundation,
-# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
-#
-# ***** END GPL LICENSE BLOCK *****
-
-if(BUILD_MODE STREQUAL Release)
-  if(WIN32)
-    set(THUMB_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../../release/windows/blendthumb)
-
-    ExternalProject_Add(external_zlib_32
-      URL ${ZLIB_URI}
-      CMAKE_GENERATOR ${GENERATOR_32}
-      URL_HASH MD5=${ZLIB_HASH}
-      DOWNLOAD_DIR ${DOWNLOAD_DIR}
-      PREFIX ${BUILD_DIR}/zlib32
-      CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${LIBDIR}/zlib32 ${DEFAULT_CMAKE_FLAGS}
-      INSTALL_DIR ${LIBDIR}/zlib32
-    )
-
-    ExternalProject_Add(external_zlib_64
-      URL ${ZLIB_URI}
-      CMAKE_GENERATOR ${GENERATOR_64}
-      URL_HASH MD5=${ZLIB_HASH}
-      DOWNLOAD_DIR ${DOWNLOAD_DIR}
-      PREFIX ${BUILD_DIR}/zlib64
-      CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${LIBDIR}/zlib64 ${DEFAULT_CMAKE_FLAGS}
-      INSTALL_DIR ${LIBDIR}/zlib64
-    )
-
-    ExternalProject_Add(external_blendthumb_32
-      CMAKE_GENERATOR ${GENERATOR_32}
-      SOURCE_DIR ${THUMB_DIR}
-      PREFIX ${BUILD_DIR}/blendthumb32
-      CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${LIBDIR}/blendThumb32 ${DEFAULT_CMAKE_FLAGS} -DZLIB_INCLUDE=${LIBDIR}/zlib32/include -DZLIB_LIBS=${LIBDIR}/zlib32/lib/zlibstatic.lib
-      INSTALL_DIR ${LIBDIR}/blendthumb32
-    )
-    add_dependencies(
-      external_blendthumb_32
-      external_zlib_32
-    )
-
-    ExternalProject_Add(external_blendthumb_64
-      CMAKE_GENERATOR ${GENERATOR_64}
-      SOURCE_DIR ${THUMB_DIR}
-      PREFIX ${BUILD_DIR}/blendthumb64
-      CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${LIBDIR}/blendThumb64 ${DEFAULT_CMAKE_FLAGS} -DZLIB_INCLUDE=${LIBDIR}/zlib64/include -DZLIB_LIBS=${LIBDIR}/zlib64/lib/zlibstatic.lib
-      INSTALL_DIR ${LIBDIR}/blendthumb64
-    )
-    add_dependencies(
-      external_blendthumb_64
-      external_zlib_64
-    )
-  endif()
-endif()
diff --git a/build_files/build_environment/cmake/check_software.cmake b/build_files/build_environment/cmake/check_software.cmake
new file mode 100644
index 00000000000..30dea754e20
--- /dev/null
+++ b/build_files/build_environment/cmake/check_software.cmake
@@ -0,0 +1,53 @@
+# ***** BEGIN GPL LICENSE BLOCK *****
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License
+# as published by the Free Software Foundation; either version 2
+# of the License, or (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software Foundation,
+# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+#
+# ***** END GPL LICENSE BLOCK *****
+
+if(UNIX)
+  set(_required_software
+    autoconf
+    automake
+    libtoolize
+    nasm
+    yasm
+    tclsh
+  )
+
+  foreach(_software ${_required_software})
+    find_program(_software_find NAMES ${_software})
+    if(NOT _software_find)
+      set(_software_missing "${_software_missing}${_software} ")
+    endif()
+    unset(_software_find CACHE)
+  endforeach()
+
+  if(_software_missing)
+    message(
+      "\n"
+      "Missing software for building Blender dependencies:\n"
+      "  ${_software_missing}\n"
+      "\n"
+      "On Debian and Ubuntu:\n"
+      "  apt install autoconf automake libtool yasm nasm tcl\n"
+      "\n"
+      "On macOS (with homebrew):\n"
+      "  brew install cmake autoconf automake libtool yasm nasm\n"
+      "\n"
+      "Other platforms:\n"
+      "  Install equivalent packages.\n")
+    message(FATAL_ERROR "Install missing software before continuing")
+  endif()
+endif()
diff --git a/build_files/build_environment/cmake/harvest.cmake b/build_files/build_environment/cmake/harvest.cmake
index 97e4a6b69d4..526e72e2e33 100644
--- a/build_files/build_environment/cmake/harvest.cmake
+++ b/build_files/build_environment/cmake/harvest.cmake
@@ -51,9 +51,6 @@ if(BUILD_MODE STREQUAL Release)
         # tiff
         ${CMAKE_COMMAND} -E copy ${LIBDIR}/tiff/lib/tiff.lib ${HARVEST_TARGET}/tiff/lib/libtiff.lib &&
         ${CMAKE_COMMAND} -E copy_directory ${LIBDIR}/tiff/include/ ${HARVEST_TARGET}/tiff/include/ &&
-        # BlendThumb
-        ${CMAKE_COMMAND} -E copy ${LIBDIR}/BlendThumb64/bin/blendthumb.dll ${HARVEST_TARGET}/ThumbHandler/lib/BlendThumb64.dll &&
-        ${CMAKE_COMMAND} -E copy ${LIBDIR}/BlendThumb32/bin/blendthumb.dll ${HARVEST_TARGET}/ThumbHandler/lib/BlendThumb.dll &&
         # hidapi
         ${CMAKE_COMMAND} -E copy_directory ${LIBDIR}/hidapi/ ${HARVEST_TARGET}/hidapi/ &&
         # webp, straight up copy
diff --git a/build_files/build_environment/cmake/osl.cmake b/build_files/build_environment/cmake/osl.cmake
index b8268f90c33..4e0a924530a 100644
--- a/build_files/build_environment/cmake/osl.cmake
+++ b/build_files/build_environment/cmake/osl.cmake
@@ -87,6 +87,7 @@ elseif(APPLE)
   set(OSL_EXTRA_ARGS
     ${OSL_EXTRA_ARGS}
     -DHIDE_SYMBOLS=OFF
+    -DPUGIXML_HOME=${LIBDIR}/pugixml
   )
 endif()
 
diff --git a/build_files/build_environment/install_deps.sh b/build_files/build_environment/install_deps.sh
index 1324616ca35..f594add3a5b 100755
--- a/build_files/build_environment/install_deps.sh
+++ b/build_files/build_environment/install_deps.sh
@@ -26,17 +26,17 @@ ARGS=$( \
 getopt \
 -o s:i:t:h \
 --long source:,install:,tmp:,info:,threads:,help,show-deps,no-sudo,no-build,no-confirm,\
-with-all,with-opencollada,with-jack,with-embree,\
+with-all,with-opencollada,with-jack,with-embree,with-oidn,\
 ver-ocio:,ver-oiio:,ver-llvm:,ver-osl:,ver-osd:,ver-openvdb:,\
 force-all,force-python,force-numpy,force-boost,\
 force-ocio,force-openexr,force-oiio,force-llvm,force-osl,force-osd,force-openvdb,\
-force-ffmpeg,force-opencollada,force-alembic,force-embree,\
+force-ffmpeg,force-opencollada,force-alembic,force-embree,force-oidn,\
 build-all,build-python,build-numpy,build-boost,\
 build-ocio,build-openexr,build-oiio,build-llvm,build-osl,build-osd,build-openvdb,\
-build-ffmpeg,build-opencollada,build-alembic,build-embree,\
+build-ffmpeg,build-opencollada,build-alembic,build-embree,build-oidn,\
 skip-python,skip-numpy,skip-boost,\
 skip-ocio,skip-openexr,skip-oiio,skip-llvm,skip-osl,skip-osd,skip-openvdb,\
-skip-ffmpeg,skip-opencollada,skip-alembic,skip-embree \
+skip-ffmpeg,skip-opencollada,skip-alembic,skip-embree,skip-oidn \
 -- "$@" \
 )
 
@@ -57,6 +57,7 @@ WITH_ALL=false
 # Do not yet enable opencollada or embree, use --with-opencollada/--with-embree (or --with-all) option to try it.
 WITH_OPENCOLLADA=false
 WITH_EMBREE=false
+WITH_OIDN=false
 
 THREADS=$(nproc)
 
@@ -69,6 +70,7 @@ Number of threads for building: \$THREADS (automatically detected, use --threads
 Full install: \$WITH_ALL (use --with-all option to enable it).
 Building OpenCOLLADA: \$WITH_OPENCOLLADA (use --with-opencollada option to enable it).
 Building Embree: \$WITH_EMBREE (use --with-embree option to enable it).
+Building OpenImageDenoise: \$WITH_OIDN (use --with-oidn option to enable it).
 
 Example:
 Full install without OpenCOLLADA: --with-all --skip-opencollada
@@ -118,6 +120,9 @@ ARGUMENTS_INFO="\"COMMAND LINE ARGUMENTS:
     --with-embree
         Build and install the Embree libraries.
 
+    --with-oidn
+        Build and install the OpenImageDenoise libraries.
+
     --with-jack
         Install the jack libraries.
 
@@ -185,6 +190,9 @@ ARGUMENTS_INFO="\"COMMAND LINE ARGUMENTS:
     --build-embree
         Force the build of Embree.
 
+    --build-oidn
+        Force the build of OpenImageDenoise.
+
     --build-ffmpeg
         Force the build of FFMpeg.
 
@@ -240,6 +248,9 @@ ARGUMENTS_INFO="\"COMMAND LINE ARGUMENTS:
     --force-embree
         Force the rebuild of Embree.
 
+    --force-oidn
+        Force the rebuild of OpenImageDenoise.
+
     --force-ffmpeg
         Force the rebuild of FFMpeg.
 
@@ -288,6 +299,9 @@ ARGUMENTS_INFO="\"COMMAND LINE ARGUMENTS:
     --skip-Embree
         Unconditionally skip Embree installation/building.
 
+    --skip-oidn
+        Unconditionally skip OpenImageDenoise installation/building.
+
     --skip-ffmpeg
         Unconditionally skip FFMpeg installation/building.\""
 
@@ -390,6 +404,11 @@ EMBREE_FORCE_BUILD=false
 EMBREE_FORCE_REBUILD=false
 EMBREE_SKIP=false
 
+OIDN_VERSION="1.0.0"
+OIDN_FORCE_BUILD=false
+OIDN_FORCE_REBUILD=false
+OIDN_SKIP=false
+
 FFMPEG_VERSION="4.0.2"
 FFMPEG_VERSION_MIN="2.8.4"
 FFMPEG_FORCE_BUILD=false
@@ -526,6 +545,9 @@ while true; do
     --with-embree)
       WITH_EMBREE=true; shift; continue
     ;;
+    --with-oidn)
+      WITH_OIDN=true; shift; continue
+    ;;
     --with-jack)
       WITH_JACK=true; shift; continue;
     ;;
@@ -572,6 +594,7 @@ while true; do
       OPENVDB_FORCE_BUILD=true
       OPENCOLLADA_FORCE_BUILD=true
       EMBREE_FORCE_BUILD=true
+      OIDN_FORCE_BUILD=true
       FFMPEG_FORCE_BUILD=true
       ALEMBIC_FORCE_BUILD=true
       shift; continue
@@ -616,6 +639,9 @@ while true; do
     --build-embree)
       EMBREE_FORCE_BUILD=true; shift; continue
     ;;
+    --build-oidn)
+      OIDN_FORCE_BUILD=true; shift; continue
+    ;;
     --build-ffmpeg)
       FFMPEG_FORCE_BUILD=true; shift; continue
     ;;
@@ -635,6 +661,7 @@ while true; do
       OPENVDB_FORCE_REBUILD=true
       OPENCOLLADA_FORCE_REBUILD=true
       EMBREE_FORCE_REBUILD=true
+      OIDN_FORCE_REBUILD=true
       FFMPEG_FORCE_REBUILD=true
       ALEMBIC_FORCE_REBUILD=true
       shift; continue
@@ -677,6 +704,9 @@ while true; do
     --force-embree)
       EMBREE_FORCE_REBUILD=true; shift; continue
     ;;
+    --force-oidn)
+      OIDN_FORCE_REBUILD=true; shift; continue
+    ;;
     --force-ffmpeg)
       FFMPEG_FORCE_REBUILD=true; shift; continue
     ;;
@@ -719,6 +749,9 @@ while true; do
     --skip-embree)
       EMBREE_SKIP=true; shift; continue
     ;;
+    --skip-oidn)
+      OIDN_SKIP=true; shift; continue
+    ;;
     --skip-ffmpeg)
       FFMPEG_SKIP=true; shift; continue
     ;;
@@ -746,6 +779,9 @@ fi
 if [ "$WITH_ALL" = true -a "$EMBREE_SKIP" = false ]; then
   WITH_EMBREE=true
 fi
+if [ "$WITH_ALL" = true -a "$OIDN_SKIP" = false ]; then
+  WITH_OIDN=true
+fi
 if [ "$WITH_ALL" = true ]; then
   WITH_JACK=true
 fi
@@ -840,6 +876,11 @@ EMBREE_SOURCE=( "https://github.com/embree/embree/archive/v${EMBREE_VERSION}.tar
 #~ EMBREE_REPO_UID="4a12bfed63c90e85b6eab98b8cdd8dd2a3ba5809"
 #~ EMBREE_REPO_BRANCH="master"
 
+OIDN_USE_REPO=false
+OIDN_SOURCE=( "https://github.com/OpenImageDenoise/oidn/releases/download/v${OIDN_VERSION}/oidn-${OIDN_VERSION}.src.tar.gz" )
+#~ OIDN_SOURCE_REPO=( "https://github.com/OpenImageDenoise/oidn.git" )
+#~ OIDN_REPO_UID="dabfd9c80101edae9d25a710160d12d6d963c591"
+#~ OIDN_REPO_BRANCH="master"
 
 FFMPEG_SOURCE=( "http://ffmpeg.org/releases/ffmpeg-$FFMPEG_VERSION.tar.bz2" )
 
@@ -882,6 +923,7 @@ You may also want to build them yourself (optional ones are [between brackets]):
     * [OpenVDB $OPENVDB_VERSION_MIN] (from $OPENVDB_SOURCE), [Blosc $OPENVDB_BLOSC_VERSION] (from $OPENVDB_BLOSC_SOURCE).
     * [OpenCollada $OPENCOLLADA_VERSION] (from $OPENCOLLADA_SOURCE).
     * [Embree $EMBREE_VERSION] (from $EMBREE_SOURCE).
+    * [OpenImageDenoise $OIDN_VERSION] (from $OIDN_SOURCE).
     * [Alembic $ALEMBIC_VERSION] (from $ALEMBIC_SOURCE).\""
 
 if [ "$DO_SHOW_DEPS" = true ]; then
@@ -1065,13 +1107,14 @@ _create_inst_shortcut() {
 # ldconfig
 run_ldconfig() {
   _lib_path="$INST/$1/lib"
+  _lib64_path="$INST/$1/lib64"
   _ldconf_path="/etc/ld.so.conf.d/$1.conf"
   PRINT ""
   if [ ! $SUDO ]; then
     WARNING "--no-sudo enabled, impossible to run ldconfig for $1, you'll have to do it yourself..."
   else
     INFO "Running ldconfig for $1..."
-    $SUDO sh -c "echo \"$_lib_path\" > $_ldconf_path"
+    $SUDO sh -c "/bin/echo -e \"$_lib_path\n$_lib64_path\" > $_ldconf_path"
     $SUDO /sbin/ldconfig  # XXX OpenSuse does not include sbin in command path with sudo!!!
   fi
   PRINT ""
@@ -1932,7 +1975,8 @@ compile_OSL() {
     cmake_d="$cmake_d -D OSL_BUILD_PLUGINS=OFF"
     cmake_d="$cmake_d -D OSL_BUILD_TESTS=OFF"
     cmake_d="$cmake_d -D USE_SIMD=sse2"
-    cmake_d="$cmake_d -D OSL_BUILD_CPP11=1"
+    cmake_d="$cmake_d -D USE_LLVM_BITCODE=OFF"
+    cmake_d="$cmake_d -D USE_PARTIO=OFF"
 
     #~ cmake_d="$cmake_d -D ILMBASE_VERSION=$ILMBASE_VERSION"
 
@@ -2552,6 +2596,98 @@ compile_Embree() {
   fi
 }
 
+#### Build OpenImageDenoise ####
+_init_oidn() {
+  _src=$SRC/oidn-$OIDN_VERSION
+  _git=true
+  _inst=$INST/oidn-$OIDN_VERSION
+  _inst_shortcut=$INST/oidn
+}
+
+clean_oidn() {
+  _init_oidn
+  _clean
+}
+
+compile_OIDN() {
+  if [ "$NO_BUILD" = true ]; then
+    WARNING "--no-build enabled, OpenImageDenoise will not be compiled!"
+    return
+  fi
+
+  # To be changed each time we make edits that would modify the compiled results!
+  oidn_magic=9
+  _init_oidn
+
+  # Clean install if needed!
+  magic_compile_check oidn-$OIDN_VERSION $oidn_magic
+  if [ $? -eq 1 -o "$OIDN_FORCE_REBUILD" = true ]; then
+    clean_oidn
+  fi
+
+  if [ ! -d $_inst ]; then
+    INFO "Building OpenImageDenoise-$OIDN_VERSION"
+
+    prepare_opt
+
+    if [ ! -d $_src ]; then
+      mkdir -p $SRC
+      if [ "OIDN_USE_REPO" = true ]; then
+        git clone $OIDN_SOURCE_REPO $_src
+      else
+        download OIDN_SOURCE[@] "$_src.tar.gz"
+        INFO "Unpacking OpenImageDenoise-$OIDN_VERSION"
+        tar -C $SRC -xf $_src.tar.gz
+      fi
+    fi
+
+    cd $_src
+
+    if [ "$OIDN_USE_REPO" = true ]; then
+      git pull origin $OIDN_REPO_BRANCH
+
+      # Stick to same rev as windows' libs...
+      git checkout $OIDN_REPO_UID
+      git reset --hard
+    fi
+
+    # Always refresh the whole build!
+    if [ -d build ]; then
+      rm -rf build
+    fi
+    mkdir build
+    cd build
+
+    cmake_d="-D CMAKE_BUILD_TYPE=Release"
+    cmake_d="$cmake_d -D CMAKE_INSTALL_PREFIX=$_inst"
+    cmake_d="$cmake_d -D WITH_EXAMPLE=OFF"
+    cmake_d="$cmake_d -D WITH_TEST=OFF"
+    cmake_d="$cmake_d -D OIDN_STATIC_LIB=ON"
+
+    cmake $cmake_d ../
+
+    make -j$THREADS && make install
+    make clean
+
+    if [ -d $_inst ]; then
+      _create_inst_shortcut
+    else
+      ERROR "OpenImageDenoise-$OIDN_VERSION failed to compile, exiting"
+      exit 1
+    fi
+
+    magic_compile_set oidn-$OIDN_VERSION $oidn_magic
+
+    cd $CWD
+    INFO "Done compiling OpenImageDenoise-$OIDN_VERSION!"
+  else
+    INFO "Own OpenImageDenoise-$OIDN_VERSION is up to date, nothing to do!"
+    INFO "If you want to force rebuild of this lib, use the --force-oidn option."
+  fi
+
+  run_ldconfig "oidn"
+}
+
 #### Build FFMPEG ####
 _init_ffmpeg() {
   _src=$SRC/ffmpeg-$FFMPEG_VERSION
@@ -3148,6 +3284,24 @@ install_DEB() {
     fi
   fi
 
+  if [ "$WITH_OIDN" = true ]; then
+    _do_compile_oidn=false
+    PRINT ""
+    if [ "$OIDN_SKIP" = true ]; then
+      WARNING "Skipping OpenImgeDenoise installation, as requested..."
+    elif [ "$OIDN_FORCE_BUILD" = true ]; then
+      INFO "Forced OpenImageDenoise building, as requested..."
+      _do_compile_oidn=true
+    else
+      # No package currently!
+      _do_compile_oidn=true
+    fi
+
+    if [ "$_do_compile_oidn" = true ]; then
+      compile_OIDN
+    fi
+  fi
+
   PRINT ""
   if [ "$FFMPEG_SKIP" = true ]; then
     WARNING "Skipping FFMpeg installation, as requested..."
@@ -3300,7 +3454,7 @@ install_RPM() {
       $SUDO dnf -y update
 
     elif [ "$RPM" = "RHEL" ]; then
-      if [ "`grep '6\.' /etc/redhat-release`" ]; then
+      if [ "`grep '[^.]6\.' /etc/redhat-release`" ]; then
         ERROR "Building with GCC 4.4 is not supported!"
         exit 1
       else
@@ -3722,6 +3876,24 @@ install_RPM() {
     fi
   fi
 
+  if [ "$WITH_OIDN" = true ]; then
+    _do_compile_oidn=false
+    PRINT ""
+    if [ "$OIDN_SKIP" = true ]; then
+      WARNING "Skipping OpenImgeDenoise installation, as requested..."
+    elif [ "$OIDN_FORCE_BUILD" = true ]; then
+      INFO "Forced OpenImageDenoise building, as requested..."
+      _do_compile_oidn=true
+    else
+      # No package currently!
+      _do_compile_oidn=true
+    fi
+
+    if [ "$_do_compile_oidn" = true ]; then
+      compile_OIDN
+    fi
+  fi
+
   PRINT ""
   if [ "$FFMPEG_SKIP" = true ]; then
     WARNING "Skipping FFMpeg installation, as requested..."
@@ -4186,6 +4358,24 @@ install_ARCH() {
     fi
   fi
 
+  if [ "$WITH_OIDN" = true ]; then
+    _do_compile_oidn=false
+    PRINT ""
+    if [ "$OIDN_SKIP" = true ]; then
+      WARNING "Skipping OpenImgeDenoise installation, as requested..."
+    elif [ "$OIDN_FORCE_BUILD" = true ]; then
+      INFO "Forced OpenImageDenoise building, as requested..."
+      _do_compile_oidn=true
+    else
+      # No package currently!
+      _do_compile_oidn=true
+    fi
+
+    if [ "$_do_compile_oidn" = true ]; then
+      compile_OIDN
+    fi
+  fi
+
   PRINT ""
   if [ "$FFMPEG_SKIP" = true ]; then
     WARNING "Skipping FFMpeg installation, as requested..."
@@ -4372,6 +4562,24 @@ install_OTHER() {
     fi
   fi
 
+  if [ "$WITH_OIDN" = true ]; then
+    _do_compile_oidn=false
+    PRINT ""
+    if [ "$OIDN_SKIP" = true ]; then
+      WARNING "Skipping OpenImgeDenoise installation, as requested..."
+    elif [ "$OIDN_FORCE_BUILD" = true ]; then
+      INFO "Forced OpenImageDenoise building, as requested..."
+      _do_compile_oidn=true
+    else
+      # No package currently!
+      _do_compile_oidn=true
+    fi
+
+    if [ "$_do_compile_oidn" = true ]; then
+      compile_OIDN
+    fi
+  fi
+
   PRINT ""
   if [ "$FFMPEG_SKIP" = true ]; then
     WARNING "Skipping FFMpeg installation, as requested..."
@@ -4587,6 +4795,17 @@ print_info() {
     fi
   fi
 
+  if [ "$WITH_OIDN" = true ]; then
+    _1="-D WITH_OPENIMAGEDENOISE=ON"
+    PRINT "  $_1"
+    _buildargs="$_buildargs $_1"
+    if [ -d $INST/oidn ]; then
+      _1="-D OPENIMAGEDENOISE_ROOT_DIR=$INST/oidn"
+      PRINT "  $_1"
+      _buildargs="$_buildargs $_1"
+    fi
+  fi
+
   if [ "$WITH_JACK" = true ]; then
     _1="-D WITH_JACK=ON"
     _2="-D WITH_JACK_DYNLOAD=ON"
diff --git a/build_files/buildbot/buildbot_utils.py b/build_files/buildbot/buildbot_utils.py
new file mode 100644
index 00000000000..6891b91aa1e
--- /dev/null
+++ b/build_files/buildbot/buildbot_utils.py
@@ -0,0 +1,112 @@
+# ##### BEGIN GPL LICENSE BLOCK #####
+#
+#  This program is free software; you can redistribute it and/or
+#  modify it under the terms of the GNU General Public License
+#  as published by the Free Software Foundation; either version 2
+#  of the License, or (at your option) any later version.
+#
+#  This program is distributed in the hope that it will be useful,
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+#  GNU General Public License for more details.
+#
+#  You should have received a copy of the GNU General Public License
+#  along with this program; if not, write to the Free Software Foundation,
+#  Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+#
+# ##### END GPL LICENSE BLOCK #####
+
+# <pep8 compliant>
+
+import argparse
+import os
+import subprocess
+import sys
+
+class Builder:
+    def __init__(self, name, branch):
+        self.name = name
+        self.branch = branch
+
+        # Buildbot runs from build/ directory
+        self.blender_dir = os.path.abspath(os.path.join('..', 'blender.git'))
+        self.build_dir = os.path.abspath(os.path.join('..', 'build', name))
+        self.install_dir = os.path.abspath(os.path.join('..', 'install', name))
+        self.upload_dir = os.path.abspath(os.path.join('..', 'install'))
+
+        # Detect platform
+        if name.startswith('mac'):
+            self.platform = 'mac'
+            self.command_prefix =  []
+        elif name.startswith('linux'):
+            self.platform = 'linux'
+            self.command_prefix =  ['scl', 'enable', 'devtoolset-6', '--']
+        elif name.startswith('win'):
+            self.platform = 'win'
+            self.command_prefix =  []
+        else:
+            raise ValueError('Unkonw platform for builder ' + self.platform)
+
+        # Always 64 bit now
+        self.bits = 64
+
+def create_builder_from_arguments():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('builder_name')
+    parser.add_argument('branch', default='master', nargs='?')
+    args = parser.parse_args()
+    return Builder(args.builder_name, args.branch)
+
+
+class VersionInfo:
+    def __init__(self, builder):
+        # Get version information
+        buildinfo_h = os.path.join(builder.build_dir, "source", "creator", "buildinfo.h")
+        blender_h = os.path.join(builder.blender_dir, "source", "blender", "blenkernel", "BKE_blender_version.h")
+
+        version_number = int(self._parse_header_file(blender_h, 'BLENDER_VERSION'))
+        self.version = "%d.%d" % (version_number // 100, version_number % 100)
+        self.version_char = self._parse_header_file(blender_h, 'BLENDER_VERSION_CHAR')
+        self.version_cycle = self._parse_header_file(blender_h, 'BLENDER_VERSION_CYCLE')
+        self.version_cycle_number = self._parse_header_file(blender_h, 'BLENDER_VERSION_CYCLE_NUMBER')
+        self.hash = self._parse_header_file(buildinfo_h, 'BUILD_HASH')[1:-1]
+
+        if self.version_cycle == "release":
+            # Final release
+            self.full_version = self.version + self.version_char
+            self.is_development_build = False
+        elif self.version_cycle == "rc":
+            # Release candidate
+            version_cycle = self.version_cycle + self.version_cycle_number
+            if len(self.version_char) == 0:
+                self.full_version = self.version + version_cycle
+            else:
+                self.full_version = self.version + self.version_char + '-' + version_cycle
+            self.is_development_build = False
+        else:
+            # Development build
+            self.full_version = self.version + '-' + self.hash
+            self.is_development_build = True
+
+    def _parse_header_file(self, filename, define):
+        import re
+        regex = re.compile("^#\s*define\s+%s\s+(.*)" % define)
+        with open(filename, "r") as file:
+            for l in file:
+                match = regex.match(l)
+                if match:
+                    return match.group(1)
+        return None
+
+
+def call(cmd, env=None, exit_on_error=True):
+    print(' '.join(cmd))
+
+    # Flush to ensure correct order output on Windows.
+    sys.stdout.flush()
+    sys.stderr.flush()
+
+    retcode = subprocess.call(cmd, env=env)
+    if exit_on_error and retcode != 0:
+        sys.exit(retcode)
+    return retcode
diff --git a/build_files/buildbot/slave_compile.py b/build_files/buildbot/slave_compile.py
index 55543055e5b..84667e663f6 100644
--- a/build_files/buildbot/slave_compile.py
+++ b/build_files/buildbot/slave_compile.py
@@ -18,149 +18,80 @@
 
 # <pep8 compliant>
 
+import buildbot_utils
 import os
-import subprocess
-import sys
 import shutil
 
-# get builder name
-if len(sys.argv) < 2:
-    sys.stderr.write("Not enough arguments, expecting builder name\n")
-    sys.exit(1)
-
-builder = sys.argv[1]
-
-# we run from build/ directory
-blender_dir = os.path.join('..', 'blender.git')
-
-
-def parse_header_file(filename, define):
-    import re
-    regex = re.compile("^#\s*define\s+%s\s+(.*)" % define)
-    with open(filename, "r") as file:
-        for l in file:
-            match = regex.match(l)
-            if match:
-                return match.group(1)
-    return None
-
-if 'cmake' in builder:
-    # cmake
-
-    # Some fine-tuning configuration
-    blender_dir = os.path.abspath(blender_dir)
-    build_dir = os.path.abspath(os.path.join('..', 'build', builder))
-    install_dir = os.path.abspath(os.path.join('..', 'install', builder))
-    targets = ['blender']
-    command_prefix = []
-
-    bits = 64
-
-    # Config file to be used (relative to blender's sources root)
-    cmake_config_file = "build_files/cmake/config/blender_release.cmake"
-
-    # Set build options.
-    cmake_options = []
-    cmake_extra_options = ['-DCMAKE_BUILD_TYPE:STRING=Release',
-                           '-DWITH_GTESTS=ON']
-
-    if builder.startswith('mac'):
-        # Set up OSX architecture
-        if builder.endswith('x86_64_10_9_cmake'):
-            cmake_extra_options.append('-DCMAKE_OSX_ARCHITECTURES:STRING=x86_64')
-        cmake_extra_options.append('-DCMAKE_OSX_DEPLOYMENT_TARGET=10.9')
-
-    elif builder.startswith('win'):
-        if builder.startswith('win64'):
-            cmake_options.extend(['-G', 'Visual Studio 15 2017 Win64'])
-        elif builder.startswith('win32'):
-            bits = 32
-            cmake_options.extend(['-G', 'Visual Studio 15 2017'])
-
-    elif builder.startswith('linux'):
-        cmake_config_file = "build_files/buildbot/config/blender_linux.cmake"
-        tokens = builder.split("_")
-        glibc = tokens[1]
-        if glibc == 'glibc224':
-            deb_name = "stretch"
-            if builder.endswith('x86_64_cmake'):
-                chroot_name = 'buildbot_' + deb_name + '_x86_64'
-            elif builder.endswith('i686_cmake'):
-                bits = 32
-                chroot_name = 'buildbot_' + deb_name + '_i686'
-            command_prefix = ['schroot', '-c', chroot_name, '--']
-        elif glibc == 'glibc217':
-            command_prefix = ['scl', 'enable', 'devtoolset-6', '--']
-
-    cmake_options.append("-C" + os.path.join(blender_dir, cmake_config_file))
-
-    # Prepare CMake options needed to configure cuda binaries compilation, 64bit only.
-    if bits == 64:
-        cmake_options.append("-DWITH_CYCLES_CUDA_BINARIES=ON")
-        cmake_options.append("-DCUDA_64_BIT_DEVICE_CODE=ON")
-    else:
-        cmake_options.append("-DWITH_CYCLES_CUDA_BINARIES=OFF")
+def get_cmake_options(builder):
+    config_file = "build_files/cmake/config/blender_release.cmake"
+    options = ['-DCMAKE_BUILD_TYPE:STRING=Release', '-DWITH_GTESTS=ON']
+
+    if builder.platform == 'mac':
+        options.append('-DCMAKE_OSX_ARCHITECTURES:STRING=x86_64')
+        options.append('-DCMAKE_OSX_DEPLOYMENT_TARGET=10.9')
+    elif builder.platform == 'win':
+        options.extend(['-G', 'Visual Studio 15 2017 Win64'])
+    elif builder.platform == 'linux':
+        config_file = "build_files/buildbot/config/blender_linux.cmake"
+
+    options.append("-C" + os.path.join(builder.blender_dir, config_file))
+    options.append("-DCMAKE_INSTALL_PREFIX=%s" % (builder.install_dir))
+
+    return options
 
-    cmake_options.append("-DCMAKE_INSTALL_PREFIX=%s" % (install_dir))
+def update_git(builder):
+    # Do extra git fetch because not all platform/git/buildbot combinations
+    # update the origin remote, causing buildinfo to detect local changes.
+    os.chdir(builder.blender_dir)
 
-    cmake_options += cmake_extra_options
+    print("Fetching remotes")
+    command = ['git', 'fetch', '--all']
+    buildbot_utils.call(builder.command_prefix + command)
 
+def clean_directories(builder):
     # Make sure no garbage remained from the previous run
-    if os.path.isdir(install_dir):
-        shutil.rmtree(install_dir)
-
-    for target in targets:
-        print("Building target %s" % (target))
-        # Construct build directory name based on the target
-        target_build_dir = build_dir
-        target_command_prefix = command_prefix[:]
-        if target != 'blender':
-            target_build_dir += '_' + target
-        target_name = 'install'
-        # Tweaking CMake options to respect the target
-        target_cmake_options = cmake_options[:]
-        # Do extra git fetch because not all platform/git/buildbot combinations
-        # update the origin remote, causing buildinfo to detect local changes.
-        os.chdir(blender_dir)
-        print("Fetching remotes")
-        command = ['git', 'fetch', '--all']
-        print(command)
-        retcode = subprocess.call(target_command_prefix + command)
-        if retcode != 0:
-            sys.exit(retcode)
-        # Make sure build directory exists and enter it
-        if not os.path.isdir(target_build_dir):
-            os.mkdir(target_build_dir)
-        os.chdir(target_build_dir)
-        # Configure the build
-        print("CMake options:")
-        print(target_cmake_options)
-        if os.path.exists('CMakeCache.txt'):
-            print("Removing CMake cache")
-            os.remove('CMakeCache.txt')
-        # Remove buildinfo files to force buildbot to re-generate them.
-        for buildinfo in ('buildinfo.h', 'buildinfo.h.txt', ):
-            full_path = os.path.join('source', 'creator', buildinfo)
-            if os.path.exists(full_path):
-                print("Removing {}" . format(buildinfo))
-                os.remove(full_path)
-        retcode = subprocess.call(target_command_prefix + ['cmake', blender_dir] + target_cmake_options)
-        if retcode != 0:
-            print('Configuration FAILED!')
-            sys.exit(retcode)
-
-        if 'win32' in builder or 'win64' in builder:
-            command = ['cmake', '--build', '.', '--target', target_name, '--config', 'Release']
-        else:
-            command = ['make', '-s', '-j2', target_name]
-
-        print("Executing command:")
-        print(command)
-        retcode = subprocess.call(target_command_prefix + command)
-
-        if retcode != 0:
-            sys.exit(retcode)
-
-else:
-    print("Unknown building system")
-    sys.exit(1)
+    if os.path.isdir(builder.install_dir):
+        shutil.rmtree(builder.install_dir)
+
+    # Make sure build directory exists and enter it
+    os.makedirs(builder.build_dir, exist_ok=True)
+
+    # Remove buildinfo files to force buildbot to re-generate them.
+    for buildinfo in ('buildinfo.h', 'buildinfo.h.txt', ):
+        full_path = os.path.join(builder.build_dir, 'source', 'creator', buildinfo)
+        if os.path.exists(full_path):
+            print("Removing {}" . format(buildinfo))
+            os.remove(full_path)
+
+def cmake_configure(builder):
+    # CMake configuration
+    os.chdir(builder.build_dir)
+
+    cmake_cache = os.path.join(builder.build_dir, 'CMakeCache.txt')
+    if os.path.exists(cmake_cache):
+        print("Removing CMake cache")
+        os.remove(cmake_cache)
+
+    print("CMake configure:")
+    cmake_options = get_cmake_options(builder)
+    command = ['cmake', builder.blender_dir] + cmake_options
+    buildbot_utils.call(builder.command_prefix + command)
+
+def cmake_build(builder):
+    # CMake build
+    os.chdir(builder.build_dir)
+
+    if builder.platform == 'win':
+        command = ['cmake', '--build', '.', '--target', 'install', '--config', 'Release']
+    else:
+        command = ['make', '-s', '-j2', 'install']
+
+    print("CMake build:")
+    buildbot_utils.call(builder.command_prefix + command)
+
+if __name__ == "__main__":
+    builder = buildbot_utils.create_builder_from_arguments()
+    update_git(builder)
+    clean_directories(builder)
+    cmake_configure(builder)
+    cmake_build(builder)
diff --git a/build_files/buildbot/slave_pack.py b/build_files/buildbot/slave_pack.py
index b29b078ca85..45331cce61f 100644
--- a/build_files/buildbot/slave_pack.py
+++ b/build_files/buildbot/slave_pack.py
@@ -22,47 +22,36 @@
 # system and zipping it into buildbot_upload.zip. This is then uploaded
 # to the master in the next buildbot step.
 
+import buildbot_utils
 import os
-import subprocess
 import sys
-import zipfile
 
-# get builder name
-if len(sys.argv) < 2:
-    sys.stderr.write("Not enough arguments, expecting builder name\n")
-    sys.exit(1)
+def get_package_name(builder, platform=None):
+    info = buildbot_utils.VersionInfo(builder)
 
-builder = sys.argv[1]
-# Never write branch if it is master.
-branch = sys.argv[2] if (len(sys.argv) >= 3 and sys.argv[2] != 'master') else ''
+    package_name = 'blender-' + info.full_version
+    if platform:
+      package_name += '-' + platform
+    if builder.branch != 'master' and info.is_development_build:
+        package_name = builder.branch + "-" + package_name
 
-blender_dir = os.path.join('..', 'blender.git')
-build_dir = os.path.join('..', 'build', builder)
-install_dir = os.path.join('..', 'install', builder)
-buildbot_upload_zip = os.path.abspath(os.path.join(os.path.dirname(install_dir), "buildbot_upload.zip"))
+    return package_name
 
-upload_filename = None  # Name of the archive to be uploaded
-                        # (this is the name of archive which will appear on the
-                        # download page)
-upload_filepath = None  # Filepath to be uploaded to the server
-                        # (this folder will be packed)
+def create_buildbot_upload_zip(builder, package_files):
+    import zipfile
 
+    buildbot_upload_zip = os.path.join(builder.upload_dir, "buildbot_upload.zip")
+    if os.path.exists(buildbot_upload_zip):
+        os.remove(buildbot_upload_zip)
 
-def parse_header_file(filename, define):
-    import re
-    regex = re.compile("^#\s*define\s+%s\s+(.*)" % define)
-    with open(filename, "r") as file:
-        for l in file:
-            match = regex.match(l)
-            if match:
-                return match.group(1)
-    return None
-
-
-# Make sure install directory always exists
-if not os.path.exists(install_dir):
-    os.makedirs(install_dir)
-
+    try:
+        z = zipfile.ZipFile(buildbot_upload_zip, "w", compression=zipfile.ZIP_STORED)
+        for filepath, filename in package_files:
+            z.write(filepath, arcname=filename)
+        z.close()
+    except Exception as ex:
+        sys.stderr.write('Create buildbot_upload.zip failed: ' + str(ex) + '\n')
+        sys.exit(1)
 
 def create_tar_bz2(src, dest, package_name):
     # One extra to remove leading os.sep when cleaning root for package_root
@@ -80,163 +69,120 @@ def create_tar_bz2(src, dest, package_name):
         package.add(entry[0], entry[1], recursive=False)
     package.close()
 
+def cleanup_files(dirpath, extension):
+    for f in os.listdir(dirpath):
+        filepath = os.path.join(dirpath, f)
+        if os.path.isfile(filepath) and f.endswith(extension):
+            os.remove(filepath)
 
-if builder.find('cmake') != -1:
-    # CMake
-    if 'win' in builder or 'mac' in builder:
-        os.chdir(build_dir)
-
-        files = [f for f in os.listdir('.') if os.path.isfile(f) and f.endswith('.zip')]
-        for f in files:
-            os.remove(f)
-        retcode = subprocess.call(['cpack', '-G', 'ZIP'])
-        result_file = [f for f in os.listdir('.') if os.path.isfile(f) and f.endswith('.zip')][0]
-
-        # TODO(sergey): Such magic usually happens in SCon's packaging but we don't have it
-        # in the CMake yet. For until then we do some magic here.
-        tokens = result_file.split('-')
-        blender_version = tokens[1].split('.')
-        blender_full_version = '.'.join(blender_version[0:2])
-        git_hash = tokens[2].split('.')[1]
-        platform = builder.split('_')[0]
-        if platform == 'mac':
-            # Special exception for OSX
-            platform = 'OSX-10.9-'
-            if builder.endswith('x86_64_10_9_cmake'):
-                platform += 'x86_64'
-        if builder.endswith('vc2015'):
-            platform += "-vc14"
-        builderified_name = 'blender-{}-{}-{}'.format(blender_full_version, git_hash, platform)
-        # NOTE: Blender 2.7 is already respected by blender_full_version.
-        if branch != '' and branch != 'blender2.7':
-            builderified_name = branch + "-" + builderified_name
-
-        os.rename(result_file, "{}.zip".format(builderified_name))
-        # create zip file
-        try:
-            if os.path.exists(buildbot_upload_zip):
-                os.remove(buildbot_upload_zip)
-            z = zipfile.ZipFile(buildbot_upload_zip, "w", compression=zipfile.ZIP_STORED)
-            z.write("{}.zip".format(builderified_name))
-            z.close()
-            sys.exit(retcode)
-        except Exception as ex:
-            sys.stderr.write('Create buildbot_upload.zip failed' + str(ex) + '\n')
-            sys.exit(1)
-
-    elif builder.startswith('linux_'):
-        blender = os.path.join(install_dir, 'blender')
-
-        buildinfo_h = os.path.join(build_dir, "source", "creator", "buildinfo.h")
-        blender_h = os.path.join(blender_dir, "source", "blender", "blenkernel", "BKE_blender_version.h")
-
-        # Get version information
-        blender_version = int(parse_header_file(blender_h, 'BLENDER_VERSION'))
-        blender_version = "%d.%d" % (blender_version // 100, blender_version % 100)
-        blender_hash = parse_header_file(buildinfo_h, 'BUILD_HASH')[1:-1]
-        blender_glibc = builder.split('_')[1]
-        command_prefix = []
-        bits = 64
-        blender_arch = 'x86_64'
-
-        if blender_glibc == 'glibc224':
-            if builder.endswith('x86_64_cmake'):
-                chroot_name = 'buildbot_stretch_x86_64'
-            elif builder.endswith('i686_cmake'):
-                chroot_name = 'buildbot_stretch_i686'
-                bits = 32
-                blender_arch = 'i686'
-            command_prefix = ['schroot', '-c', chroot_name, '--']
-        elif blender_glibc == 'glibc217':
-            command_prefix = ['scl', 'enable', 'devtoolset-6', '--']
-
-        # Strip all unused symbols from the binaries
-        print("Stripping binaries...")
-        subprocess.call(command_prefix + ['strip', '--strip-all', blender])
-
-        print("Stripping python...")
-        py_target = os.path.join(install_dir, blender_version)
-        subprocess.call(command_prefix + ['find', py_target, '-iname', '*.so', '-exec', 'strip', '-s', '{}', ';'])
-
-        # Copy all specific files which are too specific to be copied by
-        # the CMake rules themselves
-        print("Copying extra scripts and libs...")
-
-        extra = '/' + os.path.join('home', 'sources', 'release-builder', 'extra')
-        mesalibs = os.path.join(extra, 'mesalibs' + str(bits) + '.tar.bz2')
-        software_gl = os.path.join(blender_dir, 'release', 'bin', 'blender-softwaregl')
-        icons = os.path.join(blender_dir, 'release', 'freedesktop', 'icons')
-
-        os.system('tar -xpf %s -C %s' % (mesalibs, install_dir))
-        os.system('cp %s %s' % (software_gl, install_dir))
-        os.system('cp -r %s %s' % (icons, install_dir))
-        os.system('chmod 755 %s' % (os.path.join(install_dir, 'blender-softwaregl')))
-
-        # Construct archive name
-        package_name = 'blender-%s-%s-linux-%s-%s' % (blender_version,
-                                                      blender_hash,
-                                                      blender_glibc,
-                                                      blender_arch)
-        # NOTE: Blender 2.7 is already respected by blender_full_version.
-        if branch != '' and branch != 'blender2.7':
-            package_name = branch + "-" + package_name
-
-        upload_filename = package_name + ".tar.bz2"
-
-        print("Creating .tar.bz2 archive")
-        upload_filepath = install_dir + '.tar.bz2'
-        create_tar_bz2(install_dir, upload_filepath, package_name)
-else:
-    print("Unknown building system")
-    sys.exit(1)
-
-
-if upload_filepath is None:
-    # clean release directory if it already exists
-    release_dir = 'release'
-
-    if os.path.exists(release_dir):
-        for f in os.listdir(release_dir):
-            if os.path.isfile(os.path.join(release_dir, f)):
-                os.remove(os.path.join(release_dir, f))
-
-    # create release package
-    try:
-        subprocess.call(['make', 'package_archive'])
-    except Exception as ex:
-        sys.stderr.write('Make package release failed' + str(ex) + '\n')
-        sys.exit(1)
 
-    # find release directory, must exist this time
-    if not os.path.exists(release_dir):
-        sys.stderr.write("Failed to find release directory %r.\n" % release_dir)
-        sys.exit(1)
+def pack_mac(builder):
+    info = buildbot_utils.VersionInfo(builder)
 
-    # find release package
-    file = None
-    filepath = None
+    os.chdir(builder.build_dir)
+    cleanup_files(builder.build_dir, '.dmg')
 
-    for f in os.listdir(release_dir):
-        rf = os.path.join(release_dir, f)
-        if os.path.isfile(rf) and f.startswith('blender'):
-            file = f
-            filepath = rf
+    package_name = get_package_name(builder, 'macOS')
+    package_filename = package_name + '.dmg'
+    package_filepath = os.path.join(builder.build_dir, package_filename)
 
-    if not file:
-        sys.stderr.write("Failed to find release package.\n")
-        sys.exit(1)
+    release_dir = os.path.join(builder.blender_dir, 'release', 'darwin')
+    bundle_sh = os.path.join(release_dir, 'bundle.sh')
+    if info.is_development_build:
+        background_image = os.path.join(release_dir, 'buildbot', 'background.tif')
+
+    command = [bundle_sh]
+    command += ['--source', builder.install_dir]
+    command += ['--dmg', package_filepath]
+    command += ['--background-image', background_image]
+    buildbot_utils.call(command)
+
+    create_buildbot_upload_zip(builder, [(package_filepath, package_filename)])
+
+
+def pack_win(builder):
+    info = buildbot_utils.VersionInfo(builder)
+
+    os.chdir(builder.build_dir)
+    cleanup_files(builder.build_dir, '.zip')
+
+    # CPack will add the platform name
+    cpack_name = get_package_name(builder, None)
+    package_name = get_package_name(builder, 'windows' + str(builder.bits))
+
+    command = ['cmake', '-DCPACK_OVERRIDE_PACKAGENAME:STRING=' + cpack_name, '.']
+    buildbot_utils.call(builder.command_prefix + command)
+    command = ['cpack', '-G', 'ZIP']
+    buildbot_utils.call(builder.command_prefix + command)
+
+    package_filename = package_name + '.zip'
+    package_filepath = os.path.join(builder.build_dir, package_filename)
+    package_files = [(package_filepath, package_filename)]
+
+    if info.version_cycle == 'release':
+        # Installer only for final release builds, otherwise will get
+        # 'this product is already installed' messages.
+        command = ['cpack', '-G', 'WIX']
+        buildbot_utils.call(builder.command_prefix + command)
+
+        package_filename = package_name + '.msi'
+        package_filepath = os.path.join(builder.build_dir, package_filename)
+        package_files += [(package_filepath, package_filename)]
+
+    create_buildbot_upload_zip(builder, package_files)
+
+
+def pack_linux(builder):
+    blender_executable = os.path.join(builder.install_dir, 'blender')
+
+    info = buildbot_utils.VersionInfo(builder)
+    blender_glibc = builder.name.split('_')[1]
+    blender_arch = 'x86_64'
+
+    # Strip all unused symbols from the binaries
+    print("Stripping binaries...")
+    buildbot_utils.call(builder.command_prefix + ['strip', '--strip-all', blender_executable])
+
+    print("Stripping python...")
+    py_target = os.path.join(builder.install_dir, info.version)
+    buildbot_utils.call(builder.command_prefix + ['find', py_target, '-iname', '*.so', '-exec', 'strip', '-s', '{}', ';'])
+
+    # Copy all specific files which are too specific to be copied by
+    # the CMake rules themselves
+    print("Copying extra scripts and libs...")
+
+    extra = '/' + os.path.join('home', 'sources', 'release-builder', 'extra')
+    mesalibs = os.path.join(extra, 'mesalibs' + str(builder.bits) + '.tar.bz2')
+    software_gl = os.path.join(builder.blender_dir, 'release', 'bin', 'blender-softwaregl')
+    icons = os.path.join(builder.blender_dir, 'release', 'freedesktop', 'icons')
+
+    os.system('tar -xpf %s -C %s' % (mesalibs, builder.install_dir))
+    os.system('cp %s %s' % (software_gl, builder.install_dir))
+    os.system('cp -r %s %s' % (icons, builder.install_dir))
+    os.system('chmod 755 %s' % (os.path.join(builder.install_dir, 'blender-softwaregl')))
+
+    # Construct package name
+    platform_name = 'linux-' + blender_glibc + '-' + blender_arch
+    package_name = get_package_name(builder, platform_name)
+    package_filename = package_name + ".tar.bz2"
+
+    print("Creating .tar.bz2 archive")
+    package_filepath = builder.install_dir + '.tar.bz2'
+    create_tar_bz2(builder.install_dir, package_filepath, package_name)
+
+    # Create buildbot_upload.zip
+    create_buildbot_upload_zip(builder, [(package_filepath, package_filename)])
+
+
+if __name__ == "__main__":
+    builder = buildbot_utils.create_builder_from_arguments()
+
+    # Make sure install directory always exists
+    os.makedirs(builder.install_dir, exist_ok=True)
 
-    upload_filename = file
-    upload_filepath = filepath
-
-# create zip file
-try:
-    upload_zip = os.path.join(buildbot_upload_zip)
-    if os.path.exists(upload_zip):
-        os.remove(upload_zip)
-    z = zipfile.ZipFile(upload_zip, "w", compression=zipfile.ZIP_STORED)
-    z.write(upload_filepath, arcname=upload_filename)
-    z.close()
-except Exception as ex:
-    sys.stderr.write('Create buildbot_upload.zip failed' + str(ex) + '\n')
-    sys.exit(1)
+    if builder.platform == 'mac':
+        pack_mac(builder)
+    elif builder.platform == 'win':
+        pack_win(builder)
+    elif builder.platform == 'linux':
+        pack_linux(builder)
diff --git a/build_files/buildbot/slave_rsync.py b/build_files/buildbot/slave_rsync.py
index 6936232a495..19f1e67408d 100644
--- a/build_files/buildbot/slave_rsync.py
+++ b/build_files/buildbot/slave_rsync.py
@@ -21,23 +21,17 @@
 # Runs on buildbot slave, rsync zip directly to buildbot server rather
 # than using upload which is much slower
 
+import buildbot_utils
 import os
 import sys
 
-# get builder name
-if len(sys.argv) < 2:
-    sys.stderr.write("Not enough arguments, expecting builder name\n")
-    sys.exit(1)
+if __name__ == "__main__":
+    builder = buildbot_utils.create_builder_from_arguments()
 
-builder = sys.argv[1]
+    # rsync, this assumes ssh keys are setup so no password is needed
+    local_zip = "buildbot_upload.zip"
+    remote_folder = "builder.blender.org:/data/buildbot-master/uploaded/"
+    remote_zip = remote_folder + "buildbot_upload_" + builder.name + ".zip"
 
-# rsync, this assumes ssh keys are setup so no password is needed
-local_zip = "buildbot_upload.zip"
-remote_folder = "builder.blender.org:/data/buildbot-master/uploaded/"
-remote_zip = remote_folder + "buildbot_upload_" + builder + ".zip"
-command = "rsync -avz %s %s" % (local_zip, remote_zip)
-
-print(command)
-
-ret = os.system(command)
-sys.exit(ret)
+    command = ["rsync", "-avz", local_zip, remote_zip]
+    buildbot_utils.call(command)
diff --git a/build_files/buildbot/slave_test.py b/build_files/buildbot/slave_test.py
index ff186d8cd45..1b6b426b621 100644
--- a/build_files/buildbot/slave_test.py
+++ b/build_files/buildbot/slave_test.py
@@ -18,59 +18,39 @@
 
 # <pep8 compliant>
 
-import subprocess
+import buildbot_utils
 import os
 import sys
 
-# get builder name
-if len(sys.argv) < 2:
-    sys.stderr.write("Not enough arguments, expecting builder name\n")
-    sys.exit(1)
+def get_ctest_environment(builder):
+    info = buildbot_utils.VersionInfo(builder)
+    blender_version_dir = os.path.join(builder.install_dir, info.version)
 
-builder = sys.argv[1]
+    env = os.environ.copy()
+    env['BLENDER_SYSTEM_SCRIPTS'] = os.path.join(blender_version_dir, 'scripts')
+    env['BLENDER_SYSTEM_DATAFILES'] = os.path.join(blender_version_dir, 'datafiles')
+    return env
 
-# we run from build/ directory
-blender_dir = '../blender.git'
+def get_ctest_arguments(builder):
+    args = ['--output-on-failure']
+    if builder.platform == 'win':
+        args += ['-C', 'Release']
+    return args
 
-if "cmake" in builder:
-    print("Automated tests are still DISABLED!")
-    sys.exit(0)
-
-    build_dir = os.path.abspath(os.path.join('..', 'build', builder))
-    install_dir = os.path.abspath(os.path.join('..', 'install', builder))
-    # NOTE: For quick test only to see if the approach work.
-    # n the future must be replaced with an actual blender version.
-    blender_version = '2.80'
-    blender_version_dir = os.path.join(install_dir, blender_version)
-    command_prefix = []
-    extra_ctest_args = []
+def test(builder):
+    os.chdir(builder.build_dir)
 
-    if builder.startswith('win'):
-        extra_ctest_args += ['-C', 'Release']
-    elif builder.startswith('linux'):
-        tokens = builder.split("_")
-        glibc = tokens[1]
-        if glibc == 'glibc224':
-            deb_name = "stretch"
-            if builder.endswith('x86_64_cmake'):
-                chroot_name = 'buildbot_' + deb_name + '_x86_64'
-            elif builder.endswith('i686_cmake'):
-                chroot_name = 'buildbot_' + deb_name + '_i686'
-            command_prefix = ['schroot', '--preserve-environment', '-c', chroot_name, '--']
-        elif glibc == 'glibc217':
-            command_prefix = ['scl', 'enable', 'devtoolset-6', '--']
+    command = builder.command_prefix  + ['ctest'] + get_ctest_arguments(builder)
+    ctest_env = get_ctest_environment(builder)
+    buildbot_utils.call(command, env=ctest_env, exit_on_error=False)
 
-    ctest_env = os.environ.copy()
-    ctest_env['BLENDER_SYSTEM_SCRIPTS'] = os.path.join(blender_version_dir, 'scripts')
-    ctest_env['BLENDER_SYSTEM_DATAFILES'] = os.path.join(blender_version_dir, 'datafiles')
+if __name__ == "__main__":
+    print("Automated tests are still DISABLED!")
+    sys.exit(0)
 
-    os.chdir(build_dir)
-    retcode = subprocess.call(command_prefix + ['ctest', '--output-on-failure'] + extra_ctest_args,
-                              env=ctest_env)
+    builder = buildbot_utils.create_builder_from_arguments()
+    test(builder)
 
     # Always exit with a success, for until we know all the tests are passing
     # on all builders.
     sys.exit(0)
-else:
-    print("Unknown building system")
-    sys.exit(1)
diff --git a/build_files/buildbot/slave_update.py b/build_files/buildbot/slave_update.py
new file mode 100644
index 00000000000..42633e1e529
--- /dev/null
+++ b/build_files/buildbot/slave_update.py
@@ -0,0 +1,31 @@
+# ##### BEGIN GPL LICENSE BLOCK #####
+#
+#  This program is free software; you can redistribute it and/or
+#  modify it under the terms of the GNU General Public License
+#  as published by the Free Software Foundation; either version 2
+#  of the License, or (at your option) any later version.
+#
+#  This program is distributed in the hope that it will be useful,
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+#  GNU General Public License for more details.
+#
+#  You should have received a copy of the GNU General Public License
+#  along with this program; if not, write to the Free Software Foundation,
+#  Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+#
+# ##### END GPL LICENSE BLOCK #####
+
+# <pep8 compliant>
+
+import buildbot_utils
+import os
+import sys
+
+if __name__ == "__main__":
+    builder = buildbot_utils.create_builder_from_arguments()
+    os.chdir(builder.blender_dir)
+
+    # Run make update which handles all libraries and submodules.
+    make_update = os.path.join(builder.blender_dir, "build_files", "utils", "make_update.py")
+    buildbot_utils.call([sys.executable, make_update, '--no-blender'])
diff --git a/build_files/cmake/Modules/FindOpenImageDenoise.cmake b/build_files/cmake/Modules/FindOpenImageDenoise.cmake
index 85ba10b14e4..beaf8a0e8f4 100644
--- a/build_files/cmake/Modules/FindOpenImageDenoise.cmake
+++ b/build_files/cmake/Modules/FindOpenImageDenoise.cmake
@@ -46,6 +46,10 @@ FIND_PATH(OPENIMAGEDENOISE_INCLUDE_DIR
 
 SET(_openimagedenoise_FIND_COMPONENTS
   OpenImageDenoise
+)
+
+# These are needed when building statically
+SET(_openimagedenoise_FIND_STATIC_COMPONENTS
   common
   mkldnn
 )
@@ -65,6 +69,23 @@ FOREACH(COMPONENT ${_openimagedenoise_FIND_COMPONENTS})
   LIST(APPEND _openimagedenoise_LIBRARIES "${OPENIMAGEDENOISE_${UPPERCOMPONENT}_LIBRARY}")
 ENDFOREACH()
 
+FOREACH(COMPONENT ${_openimagedenoise_FIND_STATIC_COMPONENTS})
+  STRING(TOUPPER ${COMPONENT} UPPERCOMPONENT)
+
+  FIND_LIBRARY(OPENIMAGEDENOISE_${UPPERCOMPONENT}_LIBRARY
+    NAMES
+      ${COMPONENT}
+    HINTS
+      ${_openimagedenoise_SEARCH_DIRS}
+    PATH_SUFFIXES
+      lib64 lib
+    )
+  MARK_AS_ADVANCED(OPENIMAGEDENOISE_${UPPERCOMPONENT}_LIBRARY)
+  IF(OPENIMAGEDENOISE_${UPPERCOMPONENT}_LIBRARY)
+    LIST(APPEND _openimagedenoise_LIBRARIES "${OPENIMAGEDENOISE_${UPPERCOMPONENT}_LIBRARY}")
+  ENDIF()
+ENDFOREACH()
+
 FIND_LIBRARY(OPENIMAGEDENOISE_LIBRARY
   NAMES
     OpenImageDenoise
diff --git a/build_files/cmake/Modules/GTestTesting.cmake b/build_files/cmake/Modules/GTestTesting.cmake
index 47edbdf37a6..091220bb0e4 100644
--- a/build_files/cmake/Modules/GTestTesting.cmake
+++ b/build_files/cmake/Modules/GTestTesting.cmake
@@ -14,12 +14,15 @@
 
 macro(BLENDER_SRC_GTEST_EX NAME SRC EXTRA_LIBS DO_ADD_TEST)
   if(WITH_GTESTS)
+    set(TARGET_NAME ${NAME}_test)
     get_property(_current_include_directories
                  DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
                  PROPERTY INCLUDE_DIRECTORIES)
     set(TEST_INC
       ${_current_include_directories}
       ${CMAKE_SOURCE_DIR}/tests/gtests
+    )
+    set(TEST_INC_SYS
       ${GLOG_INCLUDE_DIRS}
       ${GFLAGS_INCLUDE_DIRS}
       ${CMAKE_SOURCE_DIR}/extern/gtest/include
@@ -27,8 +30,10 @@ macro(BLENDER_SRC_GTEST_EX NAME SRC EXTRA_LIBS DO_ADD_TEST)
     )
     unset(_current_include_directories)
 
-    add_executable(${NAME}_test ${SRC})
-    target_link_libraries(${NAME}_test
+    add_executable(${TARGET_NAME} ${SRC})
+    target_include_directories(${TARGET_NAME} PUBLIC "${TEST_INC}")
+    target_include_directories(${TARGET_NAME} SYSTEM PUBLIC "${TEST_INC_SYS}")
+    target_link_libraries(${TARGET_NAME}
                           ${EXTRA_LIBS}
                           ${PLATFORM_LINKLIBS}
                           bf_testing_main
@@ -41,20 +46,22 @@ macro(BLENDER_SRC_GTEST_EX NAME SRC EXTRA_LIBS DO_ADD_TEST)
                           ${GLOG_LIBRARIES}
                           ${GFLAGS_LIBRARIES})
     if(WITH_OPENMP_STATIC)
-      target_link_libraries(${NAME}_test ${OpenMP_LIBRARIES})
+      target_link_libraries(${TARGET_NAME} ${OpenMP_LIBRARIES})
     endif()
-    set_target_properties(${NAME}_test PROPERTIES
+    set_target_properties(${TARGET_NAME} PROPERTIES
                           RUNTIME_OUTPUT_DIRECTORY         "${TESTS_OUTPUT_DIR}"
                           RUNTIME_OUTPUT_DIRECTORY_RELEASE "${TESTS_OUTPUT_DIR}"
-                          RUNTIME_OUTPUT_DIRECTORY_DEBUG   "${TESTS_OUTPUT_DIR}"
-                          INCLUDE_DIRECTORIES              "${TEST_INC}")
+                          RUNTIME_OUTPUT_DIRECTORY_DEBUG   "${TESTS_OUTPUT_DIR}")
     if(${DO_ADD_TEST})
-      add_test(NAME ${NAME}_test COMMAND ${TESTS_OUTPUT_DIR}/${NAME}_test WORKING_DIRECTORY $<TARGET_FILE_DIR:blender>)
+      add_test(NAME ${TARGET_NAME} COMMAND ${TESTS_OUTPUT_DIR}/${TARGET_NAME} WORKING_DIRECTORY $<TARGET_FILE_DIR:blender>)
 
       # Don't fail tests on leaks since these often happen in external libraries
       # that we can't fix.
-      set_tests_properties(${NAME}_test PROPERTIES ENVIRONMENT LSAN_OPTIONS=exitcode=0)
+      set_tests_properties(${TARGET_NAME} PROPERTIES ENVIRONMENT LSAN_OPTIONS=exitcode=0)
     endif()
+    unset(TEST_INC)
+    unset(TEST_INC_SYS)
+    unset(TARGET_NAME)
   endif()
 endmacro()
 
diff --git a/build_files/cmake/platform/platform_win32.cmake b/build_files/cmake/platform/platform_win32.cmake
index 42ac285f88d..b2277c440fe 100644
--- a/build_files/cmake/platform/platform_win32.cmake
+++ b/build_files/cmake/platform/platform_win32.cmake
@@ -135,8 +135,8 @@ else()
   set(CMAKE_C_FLAGS     "${CMAKE_C_FLAGS} /nologo /J /Gd /MP /bigobj")
 endif()
 
-set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /MTd")
-set(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} /MTd")
+set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /MTd /ZI")
+set(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} /MTd /ZI")
 set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /MT")
 set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} /MT")
 set(CMAKE_CXX_FLAGS_MINSIZEREL "${CMAKE_CXX_FLAGS_MINSIZEREL} /MT")
diff --git a/build_files/utils/make_test.py b/build_files/utils/make_test.py
new file mode 100755
index 00000000000..0a5e4055a18
--- /dev/null
+++ b/build_files/utils/make_test.py
@@ -0,0 +1,55 @@
+#!/usr/bin/env python3
+#
+# "make test" for all platforms, running automated tests.
+
+import argparse
+import os
+import shutil
+import sys
+
+import make_utils
+from make_utils import call
+
+# Parse arguments
+
+def parse_arguments():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--ctest-command", default="ctest")
+    parser.add_argument("--cmake-command", default="cmake")
+    parser.add_argument("--svn-command", default="svn")
+    parser.add_argument("--git-command", default="git")
+    parser.add_argument("build_directory")
+    return parser.parse_args()
+
+args = parse_arguments()
+git_command = args.git_command
+svn_command = args.svn_command
+ctest_command = args.ctest_command
+cmake_command = args.cmake_command
+build_dir = args.build_directory
+
+if shutil.which(ctest_command) is None:
+    sys.stderr.write("ctest not found, can't run tests\n")
+    sys.exit(1)
+
+# Test if we are building a specific release version.
+release_version = make_utils.git_branch_release_version(git_command)
+lib_tests_dirpath = os.path.join('..', 'lib', "tests")
+
+if not os.path.exists(lib_tests_dirpath):
+    print("Tests files not found, downloading...")
+
+    if shutil.which(svn_command) is None:
+        sys.stderr.write("svn not found, can't checkout test files\n")
+        sys.exit(1)
+
+    svn_url = make_utils.svn_libraries_base_url(release_version) + "/tests"
+    call([svn_command, "checkout", svn_url, lib_tests_dirpath])
+
+    # Run cmake again to detect tests files.
+    os.chdir(build_dir)
+    call([cmake_command, "."])
+
+# Run tests
+os.chdir(build_dir)
+call([ctest_command, ".", "--output-on-failure"])
diff --git a/build_files/utils/make_update.py b/build_files/utils/make_update.py
new file mode 100755
index 00000000000..c089ca96b0c
--- /dev/null
+++ b/build_files/utils/make_update.py
@@ -0,0 +1,117 @@
+#!/usr/bin/env python3
+#
+# "make update" for all platforms, updating svn libraries and tests and Blender
+# git repository and submodules.
+#
+# For release branches, this will check out the appropriate branches of
+# submodules and libraries.
+
+import argparse
+import os
+import shutil
+import sys
+
+import make_utils
+from make_utils import call
+
+# Parse arguments
+
+def parse_arguments():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--no-libraries", action="store_true")
+    parser.add_argument("--no-blender", action="store_true")
+    parser.add_argument("--no-submodules", action="store_true")
+    parser.add_argument("--svn-command", default="svn")
+    parser.add_argument("--git-command", default="git")
+    return parser.parse_args()
+
+args = parse_arguments()
+no_libraries = args.no_libraries
+no_blender = args.no_blender
+no_submodules = args.no_submodules
+git_command = args.git_command
+svn_command = args.svn_command
+svn_non_interactive = [args.svn_command, '--non-interactive']
+
+def print_stage(text):
+    print("")
+    print(text)
+    print("")
+
+# Test if we are building a specific release version.
+release_version = make_utils.git_branch_release_version(git_command)
+
+# Setup for precompiled libraries and tests from svn.
+if not no_libraries:
+    lib_dirpath = os.path.join('..', 'lib')
+    svn_url = make_utils.svn_libraries_base_url(release_version)
+
+    # Checkout precompiled libraries
+    if sys.platform == 'darwin':
+        lib_platform = "darwin"
+    elif sys.platform == 'win32':
+        # Windows checkout is usually handled by bat scripts since python3 to run
+        # this script is bundled as part of the precompiled libraries. However it
+        # is used by the buildbot.
+        lib_platform = "win64_vc14"
+    else:
+        # No precompiled libraries for Linux.
+        lib_platform = None
+
+    if lib_platform:
+        lib_platform_dirpath = os.path.join(lib_dirpath, lib_platform)
+
+        if not os.path.exists(lib_platform_dirpath):
+            print_stage("Checking out Precompiled Libraries")
+
+            if shutil.which(svn_command) is None:
+                sys.stderr.write("svn not found, can't checkout libraries\n")
+                sys.exit(1)
+
+            svn_url_platform = svn_url + lib_platform
+            call(svn_non_interactive + ["checkout", svn_url_platform, lib_platform_dirpath])
+
+    # Update precompiled libraries and tests
+    print_stage("Updating Precompiled Libraries and Tests")
+
+    if os.path.isdir(lib_dirpath):
+      for dirname in os.listdir(lib_dirpath):
+        if dirname == ".svn":
+            continue
+
+        dirpath = os.path.join(lib_dirpath, dirname)
+        svn_dirpath = os.path.join(dirpath, ".svn")
+        svn_root_dirpath = os.path.join(lib_dirpath, ".svn")
+
+        if os.path.isdir(dirpath) and \
+           (os.path.exists(svn_dirpath) or os.path.exists(svn_root_dirpath)):
+            if shutil.which(svn_command) is None:
+                sys.stderr.write("svn not found, can't update libraries\n")
+                sys.exit(1)
+
+            call(svn_non_interactive + ["cleanup", dirpath])
+            call(svn_non_interactive + ["switch", svn_url + dirname, dirpath])
+            call(svn_non_interactive + ["update", dirpath])
+
+# Update blender repository and submodules.
+if not no_blender:
+    print_stage("Updating Blender Git Repository")
+    if shutil.which(git_command) is None:
+        sys.stderr.write("git not found, can't update code\n")
+        sys.exit(1)
+
+    call([git_command, "pull", "--rebase"])
+
+if not no_submodules:
+    print_stage("Updating Submodules")
+    if shutil.which(git_command) is None:
+        sys.stderr.write("git not found, can't update code\n")
+        sys.exit(1)
+
+    call([git_command, "submodule", "update", "--init", "--recursive"])
+    if not release_version:
+        # Update submodules to latest master if not building a specific release.
+        # In that case submodules are set to a specific revision, which is checked
+        # out by running "git submodule update".
+        call([git_command, "submodule", "foreach", "git", "checkout", "master"])
+        call([git_command, "submodule", "foreach", "git", "pull", "--rebase", "origin", "master"])
diff --git a/build_files/utils/make_utils.py b/build_files/utils/make_utils.py
new file mode 100755
index 00000000000..d05c1e6f838
--- /dev/null
+++ b/build_files/utils/make_utils.py
@@ -0,0 +1,39 @@
+#!/usr/bin/env python3
+#
+# Utility functions for make update and make tests.
+
+import re
+import subprocess
+import sys
+
+def call(cmd):
+    print(" ".join(cmd))
+
+    # Flush to ensure correct order output on Windows.
+    sys.stdout.flush()
+    sys.stderr.flush()
+
+    retcode = subprocess.call(cmd)
+    if retcode != 0:
+      sys.exit(retcode)
+
+def git_branch_release_version(git_command):
+    # Test if we are building a specific release version.
+    try:
+        branch = subprocess.check_output([git_command, "rev-parse", "--abbrev-ref", "HEAD"])
+    except subprocess.CalledProcessError as e:
+        sys.stderr.write("Failed to get Blender git branch\n")
+        sys.exit(1)
+
+    branch = branch.strip().decode('utf8')
+    release_version = re.search("^blender-v(.*)-release$", branch)
+    if release_version:
+        release_version = release_version.group(1)
+    return release_version
+
+def svn_libraries_base_url(release_version):
+    if release_version:
+        svn_branch = "tags/blender-" + release_version + "-release"
+    else:
+        svn_branch = "trunk"
+    return "https://svn.blender.org/svnroot/bf-blender/" + svn_branch + "/lib/"
diff --git a/build_files/windows/check_libraries.cmd b/build_files/windows/check_libraries.cmd
index fcae2c90657..b838c7d7d19 100644
--- a/build_files/windows/check_libraries.cmd
+++ b/build_files/windows/check_libraries.cmd
@@ -39,6 +39,15 @@ if NOT EXIST %BUILD_VS_LIBDIR% (
 			)
 		)
 	)
+) else (
+	if NOT EXIST %PYTHON% (
+		if not "%SVN%"=="" (
+			echo.
+			echo Python not found in external libraries, updating to latest version
+			echo.
+			"%SVN%" update %BUILD_VS_LIBDIR%
+		)
+	)
 )
 
 if NOT EXIST %BUILD_VS_LIBDIR% (
diff --git a/build_files/windows/find_dependencies.cmd b/build_files/windows/find_dependencies.cmd
index a861cf1c98b..6a2233ecff7 100644
--- a/build_files/windows/find_dependencies.cmd
+++ b/build_files/windows/find_dependencies.cmd
@@ -2,10 +2,12 @@ REM find all dependencies and set the corresponding environment variables.
 for %%X in (svn.exe) do (set SVN=%%~$PATH:X)
 for %%X in (cmake.exe) do (set CMAKE=%%~$PATH:X)
 for %%X in (git.exe) do (set GIT=%%~$PATH:X)
+set PYTHON=%BLENDER_DIR%\..\lib\win64_vc14\python\37\bin\python.exe
 if NOT "%verbose%" == "" (
-	echo svn   : "%SVN%"
-	echo cmake : "%CMAKE%"
-	echo git   : "%GIT%"
+	echo svn    : "%SVN%"
+	echo cmake  : "%CMAKE%"
+	echo git    : "%GIT%"
+	echo python : "%PYTHON%"
 )
 if "%CMAKE%" == "" (
 	echo Cmake not found in path, required for building, exiting...
diff --git a/build_files/windows/format.cmd b/build_files/windows/format.cmd
index f036257e220..43f8b61727d 100644
--- a/build_files/windows/format.cmd
+++ b/build_files/windows/format.cmd
@@ -9,7 +9,7 @@ exit /b 1
 :detect_done
 echo found clang-format in %CF_PATH%
 
-if EXIST %BLENDER_DIR%\..\lib\win64_vc14\python\37\bin\python.exe (
+if EXIST %PYTHON% (
     set PYTHON=%BLENDER_DIR%\..\lib\win64_vc14\python\37\bin\python.exe
     goto detect_python_done
 )
@@ -25,6 +25,7 @@ set FORMAT_PATHS=%BLENDER_DIR%\source\tools\utils_maintenance\clang_format_paths
 REM The formatting script expects clang-format to be in the current PATH.
 set PATH=%CF_PATH%;%PATH%
 
-%PYTHON% %FORMAT_PATHS% %FORMAT_ARGS%
+REM Use -B to avoid writing __pycache__ in lib directory and causing update conflicts.
+%PYTHON% -B %FORMAT_PATHS% %FORMAT_ARGS%
 
 :EOF
diff --git a/build_files/windows/parse_arguments.cmd b/build_files/windows/parse_arguments.cmd
index 3f40ef1f5ef..acbbc355f57 100644
--- a/build_files/windows/parse_arguments.cmd
+++ b/build_files/windows/parse_arguments.cmd
@@ -82,12 +82,10 @@ if NOT "%1" == "" (
 	REM Non-Build Commands
 	) else if "%1" == "update" (
 		SET BUILD_UPDATE=1
-		set BUILD_UPDATE_SVN=1
-		set BUILD_UPDATE_GIT=1
+		set BUILD_UPDATE_ARGS=
 	) else if "%1" == "code_update" (
 		SET BUILD_UPDATE=1
-		set BUILD_UPDATE_SVN=0
-		set BUILD_UPDATE_GIT=1
+		set BUILD_UPDATE_ARGS="--no-libraries"
 	) else if "%1" == "ninja" (
 		SET BUILD_WITH_NINJA=1
 	) else if "%1" == "clean" (
@@ -108,4 +106,4 @@ if NOT "%1" == "" (
 :EOF
 exit /b 0
 :ERR
-exit /b 1
\ No newline at end of file
+exit /b 1
diff --git a/build_files/windows/update_sources.cmd b/build_files/windows/update_sources.cmd
index 1f571eaf92d..f8fbd383090 100644
--- a/build_files/windows/update_sources.cmd
+++ b/build_files/windows/update_sources.cmd
@@ -1,18 +1,13 @@
-if "%BUILD_UPDATE_SVN%" == "1" (
-	if "%SVN%" == "" (
-		echo svn not found, cannot update libraries
-		goto UPDATE_GIT
-	)
-	"%SVN%" up "%BLENDER_DIR%/../lib/*"
+if EXIST %PYTHON% (
+	goto detect_python_done
 )
-:UPDATE_GIT
 
-if "%BUILD_UPDATE_GIT%" == "1" (
-	if "%GIT%" == "" (
-		echo Git not found, cannot update code
-		goto EOF
-	)
-	"%GIT%" pull --rebase
-	"%GIT%" submodule foreach git pull --rebase origin master
-)
+echo python not found in lib folder
+exit /b 1
+
+:detect_python_done
+
+REM Use -B to avoid writing __pycache__ in lib directory and causing update conflicts.
+%PYTHON% -B %BLENDER_DIR%\build_files\utils\make_update.py --git-command "%GIT%" --svn-command "%SVN%" %BUILD_UPDATE_ARGS%
+
 :EOF
diff --git a/extern/Eigen3/Eigen/Cholesky b/extern/Eigen3/Eigen/Cholesky
index f727f5d89c0..1332b540d83 100644
--- a/extern/Eigen3/Eigen/Cholesky
+++ b/extern/Eigen3/Eigen/Cholesky
@@ -1,7 +1,15 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_CHOLESKY_MODULE_H
 #define EIGEN_CHOLESKY_MODULE_H
 
 #include "Core"
+#include "Jacobi"
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
@@ -10,20 +18,26 @@
   *
   *
   * This module provides two variants of the Cholesky decomposition for selfadjoint (hermitian) matrices.
-  * Those decompositions are accessible via the following MatrixBase methods:
-  *  - MatrixBase::llt(),
+  * Those decompositions are also accessible via the following methods:
+  *  - MatrixBase::llt()
   *  - MatrixBase::ldlt()
+  *  - SelfAdjointView::llt()
+  *  - SelfAdjointView::ldlt()
   *
   * \code
   * #include <Eigen/Cholesky>
   * \endcode
   */
 
-#include "src/misc/Solve.h"
 #include "src/Cholesky/LLT.h"
 #include "src/Cholesky/LDLT.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/Cholesky/LLT_MKL.h"
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/Cholesky/LLT_LAPACKE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/extern/Eigen3/Eigen/CholmodSupport b/extern/Eigen3/Eigen/CholmodSupport
index 745b884e74d..bed8924d31e 100644
--- a/extern/Eigen3/Eigen/CholmodSupport
+++ b/extern/Eigen3/Eigen/CholmodSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_CHOLMODSUPPORT_MODULE_H
 #define EIGEN_CHOLMODSUPPORT_MODULE_H
 
@@ -12,7 +19,7 @@ extern "C" {
 /** \ingroup Support_modules
   * \defgroup CholmodSupport_Module CholmodSupport module
   *
-  * This module provides an interface to the Cholmod library which is part of the <a href="http://www.cise.ufl.edu/research/sparse/SuiteSparse/">suitesparse</a> package.
+  * This module provides an interface to the Cholmod library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
   * It provides the two following main factorization classes:
   * - class CholmodSupernodalLLT: a supernodal LLT Cholesky factorization.
   * - class CholmodDecomposiiton: a general L(D)LT Cholesky factorization with automatic or explicit runtime selection of the underlying factorization method (supernodal or simplicial).
@@ -33,12 +40,8 @@ extern "C" {
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "src/CholmodSupport/CholmodSupport.h"
 
-
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_CHOLMODSUPPORT_MODULE_H
diff --git a/extern/Eigen3/Eigen/Core b/extern/Eigen3/Eigen/Core
index 509c529e13d..b923b8c000c 100644
--- a/extern/Eigen3/Eigen/Core
+++ b/extern/Eigen3/Eigen/Core
@@ -14,6 +14,74 @@
 // first thing Eigen does: stop the compiler from committing suicide
 #include "src/Core/util/DisableStupidWarnings.h"
 
+#if defined(__CUDACC__) && !defined(EIGEN_NO_CUDA)
+  #define EIGEN_CUDACC __CUDACC__
+#endif
+
+#if defined(__CUDA_ARCH__) && !defined(EIGEN_NO_CUDA)
+  #define EIGEN_CUDA_ARCH __CUDA_ARCH__
+#endif
+
+#if defined(__CUDACC_VER_MAJOR__) && (__CUDACC_VER_MAJOR__ >= 9)
+#define EIGEN_CUDACC_VER  ((__CUDACC_VER_MAJOR__ * 10000) + (__CUDACC_VER_MINOR__ * 100))
+#elif defined(__CUDACC_VER__)
+#define EIGEN_CUDACC_VER __CUDACC_VER__
+#else
+#define EIGEN_CUDACC_VER 0
+#endif
+
+// Handle NVCC/CUDA/SYCL
+#if defined(__CUDACC__) || defined(__SYCL_DEVICE_ONLY__)
+  // Do not try asserts on CUDA and SYCL!
+  #ifndef EIGEN_NO_DEBUG
+  #define EIGEN_NO_DEBUG
+  #endif
+
+  #ifdef EIGEN_INTERNAL_DEBUGGING
+  #undef EIGEN_INTERNAL_DEBUGGING
+  #endif
+
+  #ifdef EIGEN_EXCEPTIONS
+  #undef EIGEN_EXCEPTIONS
+  #endif
+
+  // All functions callable from CUDA code must be qualified with __device__
+  #ifdef __CUDACC__
+    // Do not try to vectorize on CUDA and SYCL!
+    #ifndef EIGEN_DONT_VECTORIZE
+    #define EIGEN_DONT_VECTORIZE
+    #endif
+
+    #define EIGEN_DEVICE_FUNC __host__ __device__
+    // We need cuda_runtime.h to ensure that that EIGEN_USING_STD_MATH macro
+    // works properly on the device side
+    #include <cuda_runtime.h>
+  #else
+    #define EIGEN_DEVICE_FUNC
+  #endif
+
+#else
+  #define EIGEN_DEVICE_FUNC
+
+#endif
+
+// When compiling CUDA device code with NVCC, pull in math functions from the
+// global namespace.  In host mode, and when device doee with clang, use the
+// std versions.
+#if defined(__CUDA_ARCH__) && defined(__NVCC__)
+  #define EIGEN_USING_STD_MATH(FUNC) using ::FUNC;
+#else
+  #define EIGEN_USING_STD_MATH(FUNC) using std::FUNC;
+#endif
+
+#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(__CUDA_ARCH__) && !defined(EIGEN_EXCEPTIONS) && !defined(EIGEN_USE_SYCL)
+  #define EIGEN_EXCEPTIONS
+#endif
+
+#ifdef EIGEN_EXCEPTIONS
+  #include <new>
+#endif
+
 // then include this file where all our macros are defined. It's really important to do it first because
 // it's where we do all the alignment settings (platform detection and honoring the user's will if he
 // defined e.g. EIGEN_DONT_ALIGN) so it needs to be done before we do anything with vectorization.
@@ -21,7 +89,7 @@
 
 // Disable the ipa-cp-clone optimization flag with MinGW 6.x or newer (enabled by default with -O3)
 // See http://eigen.tuxfamily.org/bz/show_bug.cgi?id=556 for details.
-#if defined(__MINGW32__) && EIGEN_GNUC_AT_LEAST(4,6)
+#if EIGEN_COMP_MINGW && EIGEN_GNUC_AT_LEAST(4,6)
   #pragma GCC optimize ("-fno-ipa-cp-clone")
 #endif
 
@@ -31,26 +99,26 @@
 // and inclusion of their respective header files
 #include "src/Core/util/MKL_support.h"
 
-// if alignment is disabled, then disable vectorization. Note: EIGEN_ALIGN is the proper check, it takes into
-// account both the user's will (EIGEN_DONT_ALIGN) and our own platform checks
-#if !EIGEN_ALIGN
+// if alignment is disabled, then disable vectorization. Note: EIGEN_MAX_ALIGN_BYTES is the proper check, it takes into
+// account both the user's will (EIGEN_MAX_ALIGN_BYTES,EIGEN_DONT_ALIGN) and our own platform checks
+#if EIGEN_MAX_ALIGN_BYTES==0
   #ifndef EIGEN_DONT_VECTORIZE
     #define EIGEN_DONT_VECTORIZE
   #endif
 #endif
 
-#ifdef _MSC_VER
+#if EIGEN_COMP_MSVC
   #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled
-  #if (_MSC_VER >= 1500) // 2008 or later
+  #if (EIGEN_COMP_MSVC >= 1500) // 2008 or later
     // Remember that usage of defined() in a #define is undefined by the standard.
     // a user reported that in 64-bit mode, MSVC doesn't care to define _M_IX86_FP.
-    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || defined(_M_X64)
+    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || EIGEN_ARCH_x86_64
       #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER
     #endif
   #endif
 #else
   // Remember that usage of defined() in a #define is undefined by the standard
-  #if (defined __SSE2__) && ( (!defined __GNUC__) || (defined __INTEL_COMPILER) || EIGEN_GNUC_AT_LEAST(4,2) )
+  #if (defined __SSE2__) && ( (!EIGEN_COMP_GNUC) || EIGEN_COMP_ICC || EIGEN_GNUC_AT_LEAST(4,2) )
     #define EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC
   #endif
 #endif
@@ -82,6 +150,31 @@
     #ifdef __SSE4_2__
       #define EIGEN_VECTORIZE_SSE4_2
     #endif
+    #ifdef __AVX__
+      #define EIGEN_VECTORIZE_AVX
+      #define EIGEN_VECTORIZE_SSE3
+      #define EIGEN_VECTORIZE_SSSE3
+      #define EIGEN_VECTORIZE_SSE4_1
+      #define EIGEN_VECTORIZE_SSE4_2
+    #endif
+    #ifdef __AVX2__
+      #define EIGEN_VECTORIZE_AVX2
+    #endif
+    #ifdef __FMA__
+      #define EIGEN_VECTORIZE_FMA
+    #endif
+    #if defined(__AVX512F__) && defined(EIGEN_ENABLE_AVX512)
+      #define EIGEN_VECTORIZE_AVX512
+      #define EIGEN_VECTORIZE_AVX2
+      #define EIGEN_VECTORIZE_AVX
+      #define EIGEN_VECTORIZE_FMA
+      #ifdef __AVX512DQ__
+        #define EIGEN_VECTORIZE_AVX512DQ
+      #endif
+      #ifdef __AVX512ER__
+        #define EIGEN_VECTORIZE_AVX512ER
+      #endif
+    #endif
 
     // include files
 
@@ -95,9 +188,10 @@
     extern "C" {
       // In theory we should only include immintrin.h and not the other *mmintrin.h header files directly.
       // Doing so triggers some issues with ICC. However old gcc versions seems to not have this file, thus:
-      #if defined(__INTEL_COMPILER) && __INTEL_COMPILER >= 1110
+      #if EIGEN_COMP_ICC >= 1110
         #include <immintrin.h>
       #else
+        #include <mmintrin.h>
         #include <emmintrin.h>
         #include <xmmintrin.h>
         #ifdef  EIGEN_VECTORIZE_SSE3
@@ -112,8 +206,20 @@
         #ifdef EIGEN_VECTORIZE_SSE4_2
         #include <nmmintrin.h>
         #endif
+        #if defined(EIGEN_VECTORIZE_AVX) || defined(EIGEN_VECTORIZE_AVX512)
+        #include <immintrin.h>
+        #endif
       #endif
     } // end extern "C"
+  #elif defined __VSX__
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_VSX
+    #include <altivec.h>
+    // We need to #undef all these ugly tokens defined in <altivec.h>
+    // => use __vector instead of vector
+    #undef bool
+    #undef vector
+    #undef pixel
   #elif defined __ALTIVEC__
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_ALTIVEC
@@ -123,13 +229,35 @@
     #undef bool
     #undef vector
     #undef pixel
-  #elif defined  __ARM_NEON
+  #elif (defined  __ARM_NEON) || (defined __ARM_NEON__)
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_NEON
     #include <arm_neon.h>
+  #elif (defined __s390x__ && defined __VEC__)
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_ZVECTOR
+    #include <vecintrin.h>
   #endif
 #endif
 
+#if defined(__F16C__) && !defined(EIGEN_COMP_CLANG)
+  // We can use the optimized fp16 to float and float to fp16 conversion routines
+  #define EIGEN_HAS_FP16_C
+#endif
+
+#if defined __CUDACC__
+  #define EIGEN_VECTORIZE_CUDA
+  #include <vector_types.h>
+  #if EIGEN_CUDACC_VER >= 70500
+    #define EIGEN_HAS_CUDA_FP16
+  #endif
+#endif
+
+#if defined EIGEN_HAS_CUDA_FP16
+  #include <host_defines.h>
+  #include <cuda_fp16.h>
+#endif
+
 #if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)
   #define EIGEN_HAS_OPENMP
 #endif
@@ -139,7 +267,7 @@
 #endif
 
 // MSVC for windows mobile does not have the errno.h file
-#if !(defined(_MSC_VER) && defined(_WIN32_WCE)) && !defined(__ARMCC_VERSION)
+#if !(EIGEN_COMP_MSVC && EIGEN_OS_WINCE) && !EIGEN_COMP_ARM
 #define EIGEN_HAS_ERRNO
 #endif
 
@@ -159,29 +287,30 @@
 // for min/max:
 #include <algorithm>
 
+// for std::is_nothrow_move_assignable
+#ifdef EIGEN_INCLUDE_TYPE_TRAITS
+#include <type_traits>
+#endif
+
 // for outputting debug info
 #ifdef EIGEN_DEBUG_ASSIGN
 #include <iostream>
 #endif
 
 // required for __cpuid, needs to be included after cmath
-#if defined(_MSC_VER) && (defined(_M_IX86)||defined(_M_X64)) && (!defined(_WIN32_WCE))
+#if EIGEN_COMP_MSVC && EIGEN_ARCH_i386_OR_x86_64 && !EIGEN_OS_WINCE
   #include <intrin.h>
 #endif
 
-#if defined(_CPPUNWIND) || defined(__EXCEPTIONS)
-  #define EIGEN_EXCEPTIONS
-#endif
-
-#ifdef EIGEN_EXCEPTIONS
-  #include <new>
-#endif
-
 /** \brief Namespace containing all symbols from the %Eigen library. */
 namespace Eigen {
 
 inline static const char *SimdInstructionSetsInUse(void) {
-#if defined(EIGEN_VECTORIZE_SSE4_2)
+#if defined(EIGEN_VECTORIZE_AVX512)
+  return "AVX512, FMA, AVX2, AVX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_AVX)
+  return "AVX SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_SSE4_2)
   return "SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
 #elif defined(EIGEN_VECTORIZE_SSE4_1)
   return "SSE, SSE2, SSE3, SSSE3, SSE4.1";
@@ -193,8 +322,12 @@ inline static const char *SimdInstructionSetsInUse(void) {
   return "SSE, SSE2";
 #elif defined(EIGEN_VECTORIZE_ALTIVEC)
   return "AltiVec";
+#elif defined(EIGEN_VECTORIZE_VSX)
+  return "VSX";
 #elif defined(EIGEN_VECTORIZE_NEON)
   return "ARM NEON";
+#elif defined(EIGEN_VECTORIZE_ZVECTOR)
+  return "S390X ZVECTOR";
 #else
   return "None";
 #endif
@@ -202,42 +335,21 @@ inline static const char *SimdInstructionSetsInUse(void) {
 
 } // end namespace Eigen
 
-#define STAGE10_FULL_EIGEN2_API             10
-#define STAGE20_RESOLVE_API_CONFLICTS       20
-#define STAGE30_FULL_EIGEN3_API             30
-#define STAGE40_FULL_EIGEN3_STRICTNESS      40
-#define STAGE99_NO_EIGEN2_SUPPORT           99
-
-#if   defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS
-  #define EIGEN2_SUPPORT
-  #define EIGEN2_SUPPORT_STAGE STAGE40_FULL_EIGEN3_STRICTNESS
-#elif defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API
-  #define EIGEN2_SUPPORT
-  #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API
-#elif defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS
-  #define EIGEN2_SUPPORT
-  #define EIGEN2_SUPPORT_STAGE STAGE20_RESOLVE_API_CONFLICTS
-#elif defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API
-  #define EIGEN2_SUPPORT
-  #define EIGEN2_SUPPORT_STAGE STAGE10_FULL_EIGEN2_API
-#elif defined EIGEN2_SUPPORT
-  // default to stage 3, that's what it's always meant
-  #define EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API
-  #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API
-#else
-  #define EIGEN2_SUPPORT_STAGE STAGE99_NO_EIGEN2_SUPPORT
+#if defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS || defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API || defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS || defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API || defined EIGEN2_SUPPORT
+// This will generate an error message:
+#error Eigen2-support is only available up to version 3.2. Please go to "http://eigen.tuxfamily.org/index.php?title=Eigen2" for further information
 #endif
 
-#ifdef EIGEN2_SUPPORT
-#undef minor
-#endif
+namespace Eigen {
 
 // we use size_t frequently and we'll never remember to prepend it with std:: everytime just to
 // ensure QNX/QCC support
 using std::size_t;
-// gcc 4.6.0 wants std:: for ptrdiff_t 
+// gcc 4.6.0 wants std:: for ptrdiff_t
 using std::ptrdiff_t;
 
+}
+
 /** \defgroup Core_Module Core module
   * This is the main module of Eigen providing dense matrix and vector support
   * (both fixed and dynamic size) with all the features corresponding to a BLAS library
@@ -249,8 +361,8 @@ using std::ptrdiff_t;
   */
 
 #include "src/Core/util/Constants.h"
-#include "src/Core/util/ForwardDeclarations.h"
 #include "src/Core/util/Meta.h"
+#include "src/Core/util/ForwardDeclarations.h"
 #include "src/Core/util/StaticAssert.h"
 #include "src/Core/util/XprHelper.h"
 #include "src/Core/util/Memory.h"
@@ -258,41 +370,94 @@ using std::ptrdiff_t;
 #include "src/Core/NumTraits.h"
 #include "src/Core/MathFunctions.h"
 #include "src/Core/GenericPacketMath.h"
+#include "src/Core/MathFunctionsImpl.h"
+#include "src/Core/arch/Default/ConjHelper.h"
 
-#if defined EIGEN_VECTORIZE_SSE
+#if defined EIGEN_VECTORIZE_AVX512
+  #include "src/Core/arch/SSE/PacketMath.h"
+  #include "src/Core/arch/AVX/PacketMath.h"
+  #include "src/Core/arch/AVX512/PacketMath.h"
+  #include "src/Core/arch/AVX512/MathFunctions.h"
+#elif defined EIGEN_VECTORIZE_AVX
+  // Use AVX for floats and doubles, SSE for integers
+  #include "src/Core/arch/SSE/PacketMath.h"
+  #include "src/Core/arch/SSE/Complex.h"
+  #include "src/Core/arch/SSE/MathFunctions.h"
+  #include "src/Core/arch/AVX/PacketMath.h"
+  #include "src/Core/arch/AVX/MathFunctions.h"
+  #include "src/Core/arch/AVX/Complex.h"
+  #include "src/Core/arch/AVX/TypeCasting.h"
+  #include "src/Core/arch/SSE/TypeCasting.h"
+#elif defined EIGEN_VECTORIZE_SSE
   #include "src/Core/arch/SSE/PacketMath.h"
   #include "src/Core/arch/SSE/MathFunctions.h"
   #include "src/Core/arch/SSE/Complex.h"
-#elif defined EIGEN_VECTORIZE_ALTIVEC
+  #include "src/Core/arch/SSE/TypeCasting.h"
+#elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX)
   #include "src/Core/arch/AltiVec/PacketMath.h"
+  #include "src/Core/arch/AltiVec/MathFunctions.h"
   #include "src/Core/arch/AltiVec/Complex.h"
 #elif defined EIGEN_VECTORIZE_NEON
   #include "src/Core/arch/NEON/PacketMath.h"
+  #include "src/Core/arch/NEON/MathFunctions.h"
   #include "src/Core/arch/NEON/Complex.h"
+#elif defined EIGEN_VECTORIZE_ZVECTOR
+  #include "src/Core/arch/ZVector/PacketMath.h"
+  #include "src/Core/arch/ZVector/MathFunctions.h"
+  #include "src/Core/arch/ZVector/Complex.h"
+#endif
+
+// Half float support
+#include "src/Core/arch/CUDA/Half.h"
+#include "src/Core/arch/CUDA/PacketMathHalf.h"
+#include "src/Core/arch/CUDA/TypeCasting.h"
+
+#if defined EIGEN_VECTORIZE_CUDA
+  #include "src/Core/arch/CUDA/PacketMath.h"
+  #include "src/Core/arch/CUDA/MathFunctions.h"
 #endif
 
 #include "src/Core/arch/Default/Settings.h"
 
-#include "src/Core/Functors.h"
+#include "src/Core/functors/TernaryFunctors.h"
+#include "src/Core/functors/BinaryFunctors.h"
+#include "src/Core/functors/UnaryFunctors.h"
+#include "src/Core/functors/NullaryFunctors.h"
+#include "src/Core/functors/StlFunctors.h"
+#include "src/Core/functors/AssignmentFunctors.h"
+
+// Specialized functors to enable the processing of complex numbers
+// on CUDA devices
+#include "src/Core/arch/CUDA/Complex.h"
+
+#include "src/Core/IO.h"
 #include "src/Core/DenseCoeffsBase.h"
 #include "src/Core/DenseBase.h"
 #include "src/Core/MatrixBase.h"
 #include "src/Core/EigenBase.h"
 
+#include "src/Core/Product.h"
+#include "src/Core/CoreEvaluators.h"
+#include "src/Core/AssignEvaluator.h"
+
 #ifndef EIGEN_PARSED_BY_DOXYGEN // work around Doxygen bug triggered by Assign.h r814874
                                 // at least confirmed with Doxygen 1.5.5 and 1.5.6
   #include "src/Core/Assign.h"
 #endif
 
+#include "src/Core/ArrayBase.h"
 #include "src/Core/util/BlasUtil.h"
 #include "src/Core/DenseStorage.h"
 #include "src/Core/NestByValue.h"
-#include "src/Core/ForceAlignedAccess.h"
+
+// #include "src/Core/ForceAlignedAccess.h"
+
 #include "src/Core/ReturnByValue.h"
 #include "src/Core/NoAlias.h"
 #include "src/Core/PlainObjectBase.h"
 #include "src/Core/Matrix.h"
 #include "src/Core/Array.h"
+#include "src/Core/CwiseTernaryOp.h"
 #include "src/Core/CwiseBinaryOp.h"
 #include "src/Core/CwiseUnaryOp.h"
 #include "src/Core/CwiseNullaryOp.h"
@@ -300,32 +465,32 @@ using std::ptrdiff_t;
 #include "src/Core/SelfCwiseBinaryOp.h"
 #include "src/Core/Dot.h"
 #include "src/Core/StableNorm.h"
-#include "src/Core/MapBase.h"
 #include "src/Core/Stride.h"
+#include "src/Core/MapBase.h"
 #include "src/Core/Map.h"
+#include "src/Core/Ref.h"
 #include "src/Core/Block.h"
 #include "src/Core/VectorBlock.h"
-#include "src/Core/Ref.h"
 #include "src/Core/Transpose.h"
 #include "src/Core/DiagonalMatrix.h"
 #include "src/Core/Diagonal.h"
 #include "src/Core/DiagonalProduct.h"
-#include "src/Core/PermutationMatrix.h"
-#include "src/Core/Transpositions.h"
 #include "src/Core/Redux.h"
 #include "src/Core/Visitor.h"
 #include "src/Core/Fuzzy.h"
-#include "src/Core/IO.h"
 #include "src/Core/Swap.h"
 #include "src/Core/CommaInitializer.h"
-#include "src/Core/Flagged.h"
-#include "src/Core/ProductBase.h"
 #include "src/Core/GeneralProduct.h"
+#include "src/Core/Solve.h"
+#include "src/Core/Inverse.h"
+#include "src/Core/SolverBase.h"
+#include "src/Core/PermutationMatrix.h"
+#include "src/Core/Transpositions.h"
 #include "src/Core/TriangularMatrix.h"
 #include "src/Core/SelfAdjointView.h"
 #include "src/Core/products/GeneralBlockPanelKernel.h"
 #include "src/Core/products/Parallelizer.h"
-#include "src/Core/products/CoeffBasedProduct.h"
+#include "src/Core/ProductEvaluators.h"
 #include "src/Core/products/GeneralMatrixVector.h"
 #include "src/Core/products/GeneralMatrixMatrix.h"
 #include "src/Core/SolveTriangular.h"
@@ -340,6 +505,7 @@ using std::ptrdiff_t;
 #include "src/Core/products/TriangularSolverVector.h"
 #include "src/Core/BandMatrix.h"
 #include "src/Core/CoreIterators.h"
+#include "src/Core/ConditionEstimator.h"
 
 #include "src/Core/BooleanRedux.h"
 #include "src/Core/Select.h"
@@ -347,18 +513,17 @@ using std::ptrdiff_t;
 #include "src/Core/Random.h"
 #include "src/Core/Replicate.h"
 #include "src/Core/Reverse.h"
-#include "src/Core/ArrayBase.h"
 #include "src/Core/ArrayWrapper.h"
 
 #ifdef EIGEN_USE_BLAS
-#include "src/Core/products/GeneralMatrixMatrix_MKL.h"
-#include "src/Core/products/GeneralMatrixVector_MKL.h"
-#include "src/Core/products/GeneralMatrixMatrixTriangular_MKL.h"
-#include "src/Core/products/SelfadjointMatrixMatrix_MKL.h"
-#include "src/Core/products/SelfadjointMatrixVector_MKL.h"
-#include "src/Core/products/TriangularMatrixMatrix_MKL.h"
-#include "src/Core/products/TriangularMatrixVector_MKL.h"
-#include "src/Core/products/TriangularSolverMatrix_MKL.h"
+#include "src/Core/products/GeneralMatrixMatrix_BLAS.h"
+#include "src/Core/products/GeneralMatrixVector_BLAS.h"
+#include "src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h"
+#include "src/Core/products/SelfadjointMatrixMatrix_BLAS.h"
+#include "src/Core/products/SelfadjointMatrixVector_BLAS.h"
+#include "src/Core/products/TriangularMatrixMatrix_BLAS.h"
+#include "src/Core/products/TriangularMatrixVector_BLAS.h"
+#include "src/Core/products/TriangularSolverMatrix_BLAS.h"
 #endif // EIGEN_USE_BLAS
 
 #ifdef EIGEN_USE_MKL_VML
@@ -369,8 +534,4 @@ using std::ptrdiff_t;
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
-#ifdef EIGEN2_SUPPORT
-#include "Eigen2Support"
-#endif
-
 #endif // EIGEN_CORE_H
diff --git a/extern/Eigen3/Eigen/Eigen b/extern/Eigen3/Eigen/Eigen
index 19b40ea4e7e..654c8dc6380 100644
--- a/extern/Eigen3/Eigen/Eigen
+++ b/extern/Eigen3/Eigen/Eigen
@@ -1,2 +1,2 @@
 #include "Dense"
-//#include "Sparse"
+#include "Sparse"
diff --git a/extern/Eigen3/Eigen/Eigenvalues b/extern/Eigen3/Eigen/Eigenvalues
index 53c5a73a278..f3f661b074b 100644
--- a/extern/Eigen3/Eigen/Eigenvalues
+++ b/extern/Eigen3/Eigen/Eigenvalues
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_EIGENVALUES_MODULE_H
 #define EIGEN_EIGENVALUES_MODULE_H
 
@@ -25,6 +32,7 @@
   * \endcode
   */
 
+#include "src/misc/RealSvd2x2.h"
 #include "src/Eigenvalues/Tridiagonalization.h"
 #include "src/Eigenvalues/RealSchur.h"
 #include "src/Eigenvalues/EigenSolver.h"
@@ -37,9 +45,14 @@
 #include "src/Eigenvalues/GeneralizedEigenSolver.h"
 #include "src/Eigenvalues/MatrixBaseEigenvalues.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/Eigenvalues/RealSchur_MKL.h"
-#include "src/Eigenvalues/ComplexSchur_MKL.h"
-#include "src/Eigenvalues/SelfAdjointEigenSolver_MKL.h"
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/Eigenvalues/RealSchur_LAPACKE.h"
+#include "src/Eigenvalues/ComplexSchur_LAPACKE.h"
+#include "src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/extern/Eigen3/Eigen/Geometry b/extern/Eigen3/Eigen/Geometry
index efd9d4504cb..716d529529a 100644
--- a/extern/Eigen3/Eigen/Geometry
+++ b/extern/Eigen3/Eigen/Geometry
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_GEOMETRY_MODULE_H
 #define EIGEN_GEOMETRY_MODULE_H
 
@@ -9,21 +16,17 @@
 #include "LU"
 #include <limits>
 
-#ifndef M_PI
-#define M_PI 3.14159265358979323846
-#endif
-
 /** \defgroup Geometry_Module Geometry module
   *
-  *
-  *
   * This module provides support for:
   *  - fixed-size homogeneous transformations
   *  - translation, scaling, 2D and 3D rotations
-  *  - quaternions
-  *  - \ref MatrixBase::cross() "cross product"
-  *  - \ref MatrixBase::unitOrthogonal() "orthognal vector generation"
-  *  - some linear components: parametrized-lines and hyperplanes
+  *  - \link Quaternion quaternions \endlink
+  *  - cross products (\ref MatrixBase::cross, \ref MatrixBase::cross3)
+  *  - orthognal vector generation (\ref MatrixBase::unitOrthogonal)
+  *  - some linear components: \link ParametrizedLine parametrized-lines \endlink and \link Hyperplane hyperplanes \endlink
+  *  - \link AlignedBox axis aligned bounding boxes \endlink
+  *  - \link umeyama least-square transformation fitting \endlink
   *
   * \code
   * #include <Eigen/Geometry>
@@ -33,27 +36,23 @@
 #include "src/Geometry/OrthoMethods.h"
 #include "src/Geometry/EulerAngles.h"
 
-#if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
-  #include "src/Geometry/Homogeneous.h"
-  #include "src/Geometry/RotationBase.h"
-  #include "src/Geometry/Rotation2D.h"
-  #include "src/Geometry/Quaternion.h"
-  #include "src/Geometry/AngleAxis.h"
-  #include "src/Geometry/Transform.h"
-  #include "src/Geometry/Translation.h"
-  #include "src/Geometry/Scaling.h"
-  #include "src/Geometry/Hyperplane.h"
-  #include "src/Geometry/ParametrizedLine.h"
-  #include "src/Geometry/AlignedBox.h"
-  #include "src/Geometry/Umeyama.h"
-
-  #if defined EIGEN_VECTORIZE_SSE
-    #include "src/Geometry/arch/Geometry_SSE.h"
-  #endif
-#endif
-
-#ifdef EIGEN2_SUPPORT
-#include "src/Eigen2Support/Geometry/All.h"
+#include "src/Geometry/Homogeneous.h"
+#include "src/Geometry/RotationBase.h"
+#include "src/Geometry/Rotation2D.h"
+#include "src/Geometry/Quaternion.h"
+#include "src/Geometry/AngleAxis.h"
+#include "src/Geometry/Transform.h"
+#include "src/Geometry/Translation.h"
+#include "src/Geometry/Scaling.h"
+#include "src/Geometry/Hyperplane.h"
+#include "src/Geometry/ParametrizedLine.h"
+#include "src/Geometry/AlignedBox.h"
+#include "src/Geometry/Umeyama.h"
+
+// Use the SSE optimized version whenever possible. At the moment the
+// SSE version doesn't compile when AVX is enabled
+#if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX
+#include "src/Geometry/arch/Geometry_SSE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/extern/Eigen3/Eigen/Householder b/extern/Eigen3/Eigen/Householder
index 6e348db5c43..89cd81b1afb 100644
--- a/extern/Eigen3/Eigen/Householder
+++ b/extern/Eigen3/Eigen/Householder
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_HOUSEHOLDER_MODULE_H
 #define EIGEN_HOUSEHOLDER_MODULE_H
 
diff --git a/extern/Eigen3/Eigen/IterativeLinearSolvers b/extern/Eigen3/Eigen/IterativeLinearSolvers
index 0f4159dc19f..957d5750b2c 100644
--- a/extern/Eigen3/Eigen/IterativeLinearSolvers
+++ b/extern/Eigen3/Eigen/IterativeLinearSolvers
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_ITERATIVELINEARSOLVERS_MODULE_H
 #define EIGEN_ITERATIVELINEARSOLVERS_MODULE_H
 
@@ -12,28 +19,29 @@
   * This module currently provides iterative methods to solve problems of the form \c A \c x = \c b, where \c A is a squared matrix, usually very large and sparse.
   * Those solvers are accessible via the following classes:
   *  - ConjugateGradient for selfadjoint (hermitian) matrices,
+  *  - LeastSquaresConjugateGradient for rectangular least-square problems,
   *  - BiCGSTAB for general square matrices.
   *
   * These iterative solvers are associated with some preconditioners:
   *  - IdentityPreconditioner - not really useful
-  *  - DiagonalPreconditioner - also called JAcobi preconditioner, work very well on diagonal dominant matrices.
-  *  - IncompleteILUT - incomplete LU factorization with dual thresholding
+  *  - DiagonalPreconditioner - also called Jacobi preconditioner, work very well on diagonal dominant matrices.
+  *  - IncompleteLUT - incomplete LU factorization with dual thresholding
   *
   * Such problems can also be solved using the direct sparse decomposition modules: SparseCholesky, CholmodSupport, UmfPackSupport, SuperLUSupport.
   *
-  * \code
-  * #include <Eigen/IterativeLinearSolvers>
-  * \endcode
+    \code
+    #include <Eigen/IterativeLinearSolvers>
+    \endcode
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
+#include "src/IterativeLinearSolvers/SolveWithGuess.h"
 #include "src/IterativeLinearSolvers/IterativeSolverBase.h"
 #include "src/IterativeLinearSolvers/BasicPreconditioners.h"
 #include "src/IterativeLinearSolvers/ConjugateGradient.h"
+#include "src/IterativeLinearSolvers/LeastSquareConjugateGradient.h"
 #include "src/IterativeLinearSolvers/BiCGSTAB.h"
 #include "src/IterativeLinearSolvers/IncompleteLUT.h"
+#include "src/IterativeLinearSolvers/IncompleteCholesky.h"
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
diff --git a/extern/Eigen3/Eigen/Jacobi b/extern/Eigen3/Eigen/Jacobi
index ba8a4dc36a5..17c1d785a16 100644
--- a/extern/Eigen3/Eigen/Jacobi
+++ b/extern/Eigen3/Eigen/Jacobi
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_JACOBI_MODULE_H
 #define EIGEN_JACOBI_MODULE_H
 
diff --git a/extern/Eigen3/Eigen/LU b/extern/Eigen3/Eigen/LU
index db579550448..6418a86e192 100644
--- a/extern/Eigen3/Eigen/LU
+++ b/extern/Eigen3/Eigen/LU
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_LU_MODULE_H
 #define EIGEN_LU_MODULE_H
 
@@ -16,25 +23,27 @@
   * \endcode
   */
 
-#include "src/misc/Solve.h"
 #include "src/misc/Kernel.h"
 #include "src/misc/Image.h"
 #include "src/LU/FullPivLU.h"
 #include "src/LU/PartialPivLU.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/LU/PartialPivLU_MKL.h"
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/LU/PartialPivLU_LAPACKE.h"
 #endif
 #include "src/LU/Determinant.h"
-#include "src/LU/Inverse.h"
+#include "src/LU/InverseImpl.h"
 
-#if defined EIGEN_VECTORIZE_SSE
+// Use the SSE optimized version whenever possible. At the moment the
+// SSE version doesn't compile when AVX is enabled
+#if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX
   #include "src/LU/arch/Inverse_SSE.h"
 #endif
 
-#ifdef EIGEN2_SUPPORT
-  #include "src/Eigen2Support/LU.h"
-#endif
-
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_LU_MODULE_H
diff --git a/extern/Eigen3/Eigen/MetisSupport b/extern/Eigen3/Eigen/MetisSupport
index 6a113f7a878..85c41bf3400 100644
--- a/extern/Eigen3/Eigen/MetisSupport
+++ b/extern/Eigen3/Eigen/MetisSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_METISSUPPORT_MODULE_H
 #define EIGEN_METISSUPPORT_MODULE_H
 
diff --git a/extern/Eigen3/Eigen/OrderingMethods b/extern/Eigen3/Eigen/OrderingMethods
index 7c0f1fffff6..d8ea3619366 100644
--- a/extern/Eigen3/Eigen/OrderingMethods
+++ b/extern/Eigen3/Eigen/OrderingMethods
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_ORDERINGMETHODS_MODULE_H
 #define EIGEN_ORDERINGMETHODS_MODULE_H
 
diff --git a/extern/Eigen3/Eigen/PaStiXSupport b/extern/Eigen3/Eigen/PaStiXSupport
index 7c616ee5eac..de3a63b4d12 100644
--- a/extern/Eigen3/Eigen/PaStiXSupport
+++ b/extern/Eigen3/Eigen/PaStiXSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_PASTIXSUPPORT_MODULE_H
 #define EIGEN_PASTIXSUPPORT_MODULE_H
 
@@ -5,7 +12,6 @@
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
-#include <complex.h>
 extern "C" {
 #include <pastix_nompi.h>
 #include <pastix.h>
@@ -35,12 +41,8 @@ extern "C" {
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "src/PaStiXSupport/PaStiXSupport.h"
 
-
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_PASTIXSUPPORT_MODULE_H
diff --git a/extern/Eigen3/Eigen/PardisoSupport b/extern/Eigen3/Eigen/PardisoSupport
index 99330ce7a7d..340edf51fe2 100644
--- a/extern/Eigen3/Eigen/PardisoSupport
+++ b/extern/Eigen3/Eigen/PardisoSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_PARDISOSUPPORT_MODULE_H
 #define EIGEN_PARDISOSUPPORT_MODULE_H
 
@@ -7,8 +14,6 @@
 
 #include <mkl_pardiso.h>
 
-#include <unsupported/Eigen/SparseExtra>
-
 /** \ingroup Support_modules
   * \defgroup PardisoSupport_Module PardisoSupport module
   *
diff --git a/extern/Eigen3/Eigen/QR b/extern/Eigen3/Eigen/QR
index ac5b0269354..c7e91446994 100644
--- a/extern/Eigen3/Eigen/QR
+++ b/extern/Eigen3/Eigen/QR
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_QR_MODULE_H
 #define EIGEN_QR_MODULE_H
 
@@ -15,31 +22,30 @@
   *
   * This module provides various QR decompositions
   * This module also provides some MatrixBase methods, including:
-  *  - MatrixBase::qr(),
+  *  - MatrixBase::householderQr()
+  *  - MatrixBase::colPivHouseholderQr()
+  *  - MatrixBase::fullPivHouseholderQr()
   *
   * \code
   * #include <Eigen/QR>
   * \endcode
   */
 
-#include "src/misc/Solve.h"
 #include "src/QR/HouseholderQR.h"
 #include "src/QR/FullPivHouseholderQR.h"
 #include "src/QR/ColPivHouseholderQR.h"
+#include "src/QR/CompleteOrthogonalDecomposition.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/QR/HouseholderQR_MKL.h"
-#include "src/QR/ColPivHouseholderQR_MKL.h"
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
 #endif
-
-#ifdef EIGEN2_SUPPORT
-#include "src/Eigen2Support/QR.h"
+#include "src/QR/HouseholderQR_LAPACKE.h"
+#include "src/QR/ColPivHouseholderQR_LAPACKE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
-#ifdef EIGEN2_SUPPORT
-#include "Eigenvalues"
-#endif
-
 #endif // EIGEN_QR_MODULE_H
 /* vim: set filetype=cpp et sw=2 ts=2 ai: */
diff --git a/extern/Eigen3/Eigen/QtAlignedMalloc b/extern/Eigen3/Eigen/QtAlignedMalloc
index 46f7d83b70f..4f07df02ae9 100644
--- a/extern/Eigen3/Eigen/QtAlignedMalloc
+++ b/extern/Eigen3/Eigen/QtAlignedMalloc
@@ -1,3 +1,9 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_QTMALLOC_MODULE_H
 #define EIGEN_QTMALLOC_MODULE_H
@@ -8,7 +14,7 @@
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
-void *qMalloc(size_t size)
+void *qMalloc(std::size_t size)
 {
   return Eigen::internal::aligned_malloc(size);
 }
@@ -18,10 +24,10 @@ void qFree(void *ptr)
   Eigen::internal::aligned_free(ptr);
 }
 
-void *qRealloc(void *ptr, size_t size)
+void *qRealloc(void *ptr, std::size_t size)
 {
   void* newPtr = Eigen::internal::aligned_malloc(size);
-  memcpy(newPtr, ptr, size);
+  std::memcpy(newPtr, ptr, size);
   Eigen::internal::aligned_free(ptr);
   return newPtr;
 }
diff --git a/extern/Eigen3/Eigen/SPQRSupport b/extern/Eigen3/Eigen/SPQRSupport
index 77016442ee7..f70390c1766 100644
--- a/extern/Eigen3/Eigen/SPQRSupport
+++ b/extern/Eigen3/Eigen/SPQRSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SPQRSUPPORT_MODULE_H
 #define EIGEN_SPQRSUPPORT_MODULE_H
 
@@ -10,7 +17,7 @@
 /** \ingroup Support_modules
   * \defgroup SPQRSupport_Module SuiteSparseQR module
   * 
-  * This module provides an interface to the SPQR library, which is part of the <a href="http://www.cise.ufl.edu/research/sparse/SuiteSparse/">suitesparse</a> package.
+  * This module provides an interface to the SPQR library, which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
   *
   * \code
   * #include <Eigen/SPQRSupport>
@@ -21,8 +28,6 @@
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
 #include "src/CholmodSupport/CholmodSupport.h"
 #include "src/SPQRSupport/SuiteSparseQRSupport.h"
 
diff --git a/extern/Eigen3/Eigen/SVD b/extern/Eigen3/Eigen/SVD
index fd310017ad1..5d0e75f7f75 100644
--- a/extern/Eigen3/Eigen/SVD
+++ b/extern/Eigen3/Eigen/SVD
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SVD_MODULE_H
 #define EIGEN_SVD_MODULE_H
 
@@ -12,23 +19,30 @@
   *
   *
   * This module provides SVD decomposition for matrices (both real and complex).
-  * This decomposition is accessible via the following MatrixBase method:
+  * Two decomposition algorithms are provided:
+  *  - JacobiSVD implementing two-sided Jacobi iterations is numerically very accurate, fast for small matrices, but very slow for larger ones.
+  *  - BDCSVD implementing a recursive divide & conquer strategy on top of an upper-bidiagonalization which remains fast for large problems.
+  * These decompositions are accessible via the respective classes and following MatrixBase methods:
   *  - MatrixBase::jacobiSvd()
+  *  - MatrixBase::bdcSvd()
   *
   * \code
   * #include <Eigen/SVD>
   * \endcode
   */
 
-#include "src/misc/Solve.h"
+#include "src/misc/RealSvd2x2.h"
+#include "src/SVD/UpperBidiagonalization.h"
+#include "src/SVD/SVDBase.h"
 #include "src/SVD/JacobiSVD.h"
+#include "src/SVD/BDCSVD.h"
 #if defined(EIGEN_USE_LAPACKE) && !defined(EIGEN_USE_LAPACKE_STRICT)
-#include "src/SVD/JacobiSVD_MKL.h"
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
 #endif
-#include "src/SVD/UpperBidiagonalization.h"
-
-#ifdef EIGEN2_SUPPORT
-#include "src/Eigen2Support/SVD.h"
+#include "src/SVD/JacobiSVD_LAPACKE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/extern/Eigen3/Eigen/Sparse b/extern/Eigen3/Eigen/Sparse
index 7cc9c09133a..136e681a1f6 100644
--- a/extern/Eigen3/Eigen/Sparse
+++ b/extern/Eigen3/Eigen/Sparse
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SPARSE_MODULE_H
 #define EIGEN_SPARSE_MODULE_H
 
@@ -11,14 +18,16 @@
   * - \ref SparseQR_Module
   * - \ref IterativeLinearSolvers_Module
   *
-  * \code
-  * #include <Eigen/Sparse>
-  * \endcode
+    \code
+    #include <Eigen/Sparse>
+    \endcode
   */
 
 #include "SparseCore"
 #include "OrderingMethods"
+#ifndef EIGEN_MPL2_ONLY
 #include "SparseCholesky"
+#endif
 #include "SparseLU"
 #include "SparseQR"
 #include "IterativeLinearSolvers"
diff --git a/extern/Eigen3/Eigen/SparseCholesky b/extern/Eigen3/Eigen/SparseCholesky
index 9f5056aa1a1..b6a320c4027 100644
--- a/extern/Eigen3/Eigen/SparseCholesky
+++ b/extern/Eigen3/Eigen/SparseCholesky
@@ -34,8 +34,6 @@
 #error The SparseCholesky module has nothing to offer in MPL2 only mode
 #endif
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
 #include "src/SparseCholesky/SimplicialCholesky.h"
 
 #ifndef EIGEN_MPL2_ONLY
diff --git a/extern/Eigen3/Eigen/SparseCore b/extern/Eigen3/Eigen/SparseCore
index 24bcf0156b3..76966c4c4cb 100644
--- a/extern/Eigen3/Eigen/SparseCore
+++ b/extern/Eigen3/Eigen/SparseCore
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SPARSECORE_MODULE_H
 #define EIGEN_SPARSECORE_MODULE_H
 
@@ -26,37 +33,35 @@
   * This module depends on: Core.
   */
 
-namespace Eigen {
-
-/** The type used to identify a general sparse storage. */
-struct Sparse {};
-
-}
-
 #include "src/SparseCore/SparseUtil.h"
 #include "src/SparseCore/SparseMatrixBase.h"
+#include "src/SparseCore/SparseAssign.h"
 #include "src/SparseCore/CompressedStorage.h"
 #include "src/SparseCore/AmbiVector.h"
+#include "src/SparseCore/SparseCompressedBase.h"
 #include "src/SparseCore/SparseMatrix.h"
+#include "src/SparseCore/SparseMap.h"
 #include "src/SparseCore/MappedSparseMatrix.h"
 #include "src/SparseCore/SparseVector.h"
-#include "src/SparseCore/SparseBlock.h"
-#include "src/SparseCore/SparseTranspose.h"
+#include "src/SparseCore/SparseRef.h"
 #include "src/SparseCore/SparseCwiseUnaryOp.h"
 #include "src/SparseCore/SparseCwiseBinaryOp.h"
+#include "src/SparseCore/SparseTranspose.h"
+#include "src/SparseCore/SparseBlock.h"
 #include "src/SparseCore/SparseDot.h"
-#include "src/SparseCore/SparsePermutation.h"
 #include "src/SparseCore/SparseRedux.h"
-#include "src/SparseCore/SparseFuzzy.h"
+#include "src/SparseCore/SparseView.h"
+#include "src/SparseCore/SparseDiagonalProduct.h"
 #include "src/SparseCore/ConservativeSparseSparseProduct.h"
 #include "src/SparseCore/SparseSparseProductWithPruning.h"
 #include "src/SparseCore/SparseProduct.h"
 #include "src/SparseCore/SparseDenseProduct.h"
-#include "src/SparseCore/SparseDiagonalProduct.h"
-#include "src/SparseCore/SparseTriangularView.h"
 #include "src/SparseCore/SparseSelfAdjointView.h"
+#include "src/SparseCore/SparseTriangularView.h"
 #include "src/SparseCore/TriangularSolver.h"
-#include "src/SparseCore/SparseView.h"
+#include "src/SparseCore/SparsePermutation.h"
+#include "src/SparseCore/SparseFuzzy.h"
+#include "src/SparseCore/SparseSolverBase.h"
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
diff --git a/extern/Eigen3/Eigen/SparseLU b/extern/Eigen3/Eigen/SparseLU
index 8527a49bd86..38b38b531d1 100644
--- a/extern/Eigen3/Eigen/SparseLU
+++ b/extern/Eigen3/Eigen/SparseLU
@@ -20,9 +20,6 @@
   * Please, see the documentation of the SparseLU class for more details.
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 // Ordering interface
 #include "OrderingMethods"
 
diff --git a/extern/Eigen3/Eigen/SparseQR b/extern/Eigen3/Eigen/SparseQR
index 4ee42065eed..a6f3b7f7d75 100644
--- a/extern/Eigen3/Eigen/SparseQR
+++ b/extern/Eigen3/Eigen/SparseQR
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SPARSEQR_MODULE_H
 #define EIGEN_SPARSEQR_MODULE_H
 
@@ -21,9 +28,6 @@
   * 
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "OrderingMethods"
 #include "src/SparseCore/SparseColEtree.h"
 #include "src/SparseQR/SparseQR.h"
diff --git a/extern/Eigen3/Eigen/StdDeque b/extern/Eigen3/Eigen/StdDeque
index f27234778f4..bc68397be25 100644
--- a/extern/Eigen3/Eigen/StdDeque
+++ b/extern/Eigen3/Eigen/StdDeque
@@ -14,7 +14,7 @@
 #include "Core"
 #include <deque>
 
-#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...)
 
diff --git a/extern/Eigen3/Eigen/StdList b/extern/Eigen3/Eigen/StdList
index 225c1e18f8e..4c6262c08cc 100644
--- a/extern/Eigen3/Eigen/StdList
+++ b/extern/Eigen3/Eigen/StdList
@@ -13,7 +13,7 @@
 #include "Core"
 #include <list>
 
-#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */    
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...)
 
diff --git a/extern/Eigen3/Eigen/StdVector b/extern/Eigen3/Eigen/StdVector
index 6b22627f6f6..0c4697ad5be 100644
--- a/extern/Eigen3/Eigen/StdVector
+++ b/extern/Eigen3/Eigen/StdVector
@@ -14,7 +14,7 @@
 #include "Core"
 #include <vector>
 
-#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_VECTOR_SPECIALIZATION(...)
 
diff --git a/extern/Eigen3/Eigen/SuperLUSupport b/extern/Eigen3/Eigen/SuperLUSupport
index 575e14fbc29..59312a82db0 100644
--- a/extern/Eigen3/Eigen/SuperLUSupport
+++ b/extern/Eigen3/Eigen/SuperLUSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SUPERLUSUPPORT_MODULE_H
 #define EIGEN_SUPERLUSUPPORT_MODULE_H
 
@@ -36,6 +43,8 @@ namespace Eigen { struct SluMatrix; }
   * - class SuperLU: a supernodal sequential LU factorization.
   * - class SuperILU: a supernodal sequential incomplete LU factorization (to be used as a preconditioner for iterative methods).
   *
+  * \warning This wrapper requires at least versions 4.0 of SuperLU. The 3.x versions are not supported.
+  *
   * \warning When including this module, you have to use SUPERLU_EMPTY instead of EMPTY which is no longer defined because it is too polluting.
   *
   * \code
@@ -48,12 +57,8 @@ namespace Eigen { struct SluMatrix; }
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "src/SuperLUSupport/SuperLUSupport.h"
 
-
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_SUPERLUSUPPORT_MODULE_H
diff --git a/extern/Eigen3/Eigen/UmfPackSupport b/extern/Eigen3/Eigen/UmfPackSupport
index 984f64a8419..00eec80875f 100644
--- a/extern/Eigen3/Eigen/UmfPackSupport
+++ b/extern/Eigen3/Eigen/UmfPackSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_UMFPACKSUPPORT_MODULE_H
 #define EIGEN_UMFPACKSUPPORT_MODULE_H
 
@@ -12,7 +19,7 @@ extern "C" {
 /** \ingroup Support_modules
   * \defgroup UmfPackSupport_Module UmfPackSupport module
   *
-  * This module provides an interface to the UmfPack library which is part of the <a href="http://www.cise.ufl.edu/research/sparse/SuiteSparse/">suitesparse</a> package.
+  * This module provides an interface to the UmfPack library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
   * It provides the following factorization class:
   * - class UmfPackLU: a multifrontal sequential LU factorization.
   *
@@ -26,9 +33,6 @@ extern "C" {
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "src/UmfPackSupport/UmfPackSupport.h"
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/extern/Eigen3/Eigen/src/Cholesky/LDLT.h b/extern/Eigen3/Eigen/src/Cholesky/LDLT.h
index abd30bd916d..15ccf24f144 100644
--- a/extern/Eigen3/Eigen/src/Cholesky/LDLT.h
+++ b/extern/Eigen3/Eigen/src/Cholesky/LDLT.h
@@ -13,7 +13,7 @@
 #ifndef EIGEN_LDLT_H
 #define EIGEN_LDLT_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
   template<typename MatrixType, int UpLo> struct LDLT_Traits;
@@ -28,8 +28,8 @@ namespace internal {
   *
   * \brief Robust Cholesky decomposition of a matrix with pivoting
   *
-  * \param MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition
-  * \param UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
+  * \tparam _MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition
+  * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
   *             The other triangular part won't be read.
   *
   * Perform a robust Cholesky decomposition of a positive semidefinite or negative semidefinite
@@ -43,7 +43,9 @@ namespace internal {
   * Remember that Cholesky decompositions are not rank-revealing. Also, do not use a Cholesky
   * decomposition to determine whether a system of equations has a solution.
   *
-  * \sa MatrixBase::ldlt(), class LLT
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
+  * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt(), class LLT
   */
 template<typename _MatrixType, int _UpLo> class LDLT
 {
@@ -52,15 +54,15 @@ template<typename _MatrixType, int _UpLo> class LDLT
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options & ~RowMajorBit, // these are the options for the TmpMatrixType, we need a ColMajor matrix here!
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
       UpLo = _UpLo
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef Matrix<Scalar, RowsAtCompileTime, 1, Options, MaxRowsAtCompileTime, 1> TmpMatrixType;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef Matrix<Scalar, RowsAtCompileTime, 1, 0, MaxRowsAtCompileTime, 1> TmpMatrixType;
 
     typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
     typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationType;
@@ -72,11 +74,11 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * The default constructor is useful in cases in which the user intends to
       * perform decompositions via LDLT::compute(const MatrixType&).
       */
-    LDLT() 
-      : m_matrix(), 
-        m_transpositions(), 
+    LDLT()
+      : m_matrix(),
+        m_transpositions(),
         m_sign(internal::ZeroSign),
-        m_isInitialized(false) 
+        m_isInitialized(false)
     {}
 
     /** \brief Default Constructor with memory preallocation
@@ -85,7 +87,7 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * according to the specified problem \a size.
       * \sa LDLT()
       */
-    LDLT(Index size)
+    explicit LDLT(Index size)
       : m_matrix(size, size),
         m_transpositions(size),
         m_temporary(size),
@@ -96,16 +98,35 @@ template<typename _MatrixType, int _UpLo> class LDLT
     /** \brief Constructor with decomposition
       *
       * This calculates the decomposition for the input \a matrix.
+      *
       * \sa LDLT(Index size)
       */
-    LDLT(const MatrixType& matrix)
+    template<typename InputType>
+    explicit LDLT(const EigenBase<InputType>& matrix)
       : m_matrix(matrix.rows(), matrix.cols()),
         m_transpositions(matrix.rows()),
         m_temporary(matrix.rows()),
         m_sign(internal::ZeroSign),
         m_isInitialized(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+    /** \brief Constructs a LDLT factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+      *
+      * \sa LDLT(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit LDLT(EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
+        m_transpositions(matrix.rows()),
+        m_temporary(matrix.rows()),
+        m_sign(internal::ZeroSign),
+        m_isInitialized(false)
+    {
+      compute(matrix.derived());
     }
 
     /** Clear any existing decomposition
@@ -151,13 +172,6 @@ template<typename _MatrixType, int _UpLo> class LDLT
       eigen_assert(m_isInitialized && "LDLT is not initialized.");
       return m_sign == internal::PositiveSemiDef || m_sign == internal::ZeroSign;
     }
-    
-    #ifdef EIGEN2_SUPPORT
-    inline bool isPositiveDefinite() const
-    {
-      return isPositive();
-    }
-    #endif
 
     /** \returns true if the matrix is negative (semidefinite) */
     inline bool isNegative(void) const
@@ -173,37 +187,38 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * \note_about_checking_solutions
       *
       * More precisely, this method solves \f$ A x = b \f$ using the decomposition \f$ A = P^T L D L^* P \f$
-      * by solving the systems \f$ P^T y_1 = b \f$, \f$ L y_2 = y_1 \f$, \f$ D y_3 = y_2 \f$, 
+      * by solving the systems \f$ P^T y_1 = b \f$, \f$ L y_2 = y_1 \f$, \f$ D y_3 = y_2 \f$,
       * \f$ L^* y_4 = y_3 \f$ and \f$ P x = y_4 \f$ in succession. If the matrix \f$ A \f$ is singular, then
       * \f$ D \f$ will also be singular (all the other matrices are invertible). In that case, the
       * least-square solution of \f$ D y_3 = y_2 \f$ is computed. This does not mean that this function
       * computes the least-square solution of \f$ A x = b \f$ is \f$ A \f$ is singular.
       *
-      * \sa MatrixBase::ldlt()
+      * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt()
       */
     template<typename Rhs>
-    inline const internal::solve_retval<LDLT, Rhs>
+    inline const Solve<LDLT, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "LDLT is not initialized.");
       eigen_assert(m_matrix.rows()==b.rows()
                 && "LDLT::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<LDLT, Rhs>(*this, b.derived());
+      return Solve<LDLT, Rhs>(*this, b.derived());
     }
 
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived, typename ResultType>
-    bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const
-    {
-      *result = this->solve(b);
-      return true;
-    }
-    #endif
-
     template<typename Derived>
     bool solveInPlace(MatrixBase<Derived> &bAndX) const;
 
-    LDLT& compute(const MatrixType& matrix);
+    template<typename InputType>
+    LDLT& compute(const EigenBase<InputType>& matrix);
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of
+     *  which \c *this is the LDLT decomposition.
+     */
+    RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
+    }
 
     template <typename Derived>
     LDLT& rankUpdate(const MatrixBase<Derived>& w, const RealScalar& alpha=1);
@@ -220,22 +235,35 @@ template<typename _MatrixType, int _UpLo> class LDLT
 
     MatrixType reconstructedMatrix() const;
 
+    /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
+      *
+      * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
+      * \code x = decomposition.adjoint().solve(b) \endcode
+      */
+    const LDLT& adjoint() const { return *this; };
+
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
 
     /** \brief Reports whether previous computation was successful.
       *
       * \returns \c Success if computation was succesful,
-      *          \c NumericalIssue if the matrix.appears to be negative.
+      *          \c NumericalIssue if the factorization failed because of a zero pivot.
       */
     ComputationInfo info() const
     {
       eigen_assert(m_isInitialized && "LDLT is not initialized.");
-      return Success;
+      return m_info;
     }
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
+
   protected:
-    
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
@@ -248,10 +276,12 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * is not stored), and the diagonal entries correspond to D.
       */
     MatrixType m_matrix;
+    RealScalar m_l1_norm;
     TranspositionType m_transpositions;
     TmpMatrixType m_temporary;
     internal::SignMatrix m_sign;
     bool m_isInitialized;
+    ComputationInfo m_info;
 };
 
 namespace internal {
@@ -266,15 +296,18 @@ template<> struct ldlt_inplace<Lower>
     using std::abs;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename TranspositionType::StorageIndex IndexType;
     eigen_assert(mat.rows()==mat.cols());
     const Index size = mat.rows();
+    bool found_zero_pivot = false;
+    bool ret = true;
 
     if (size <= 1)
     {
       transpositions.setIdentity();
-      if (numext::real(mat.coeff(0,0)) > 0) sign = PositiveSemiDef;
-      else if (numext::real(mat.coeff(0,0)) < 0) sign = NegativeSemiDef;
+      if(size==0) sign = ZeroSign;
+      else if (numext::real(mat.coeff(0,0)) > static_cast<RealScalar>(0) ) sign = PositiveSemiDef;
+      else if (numext::real(mat.coeff(0,0)) < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
       else sign = ZeroSign;
       return true;
     }
@@ -286,7 +319,7 @@ template<> struct ldlt_inplace<Lower>
       mat.diagonal().tail(size-k).cwiseAbs().maxCoeff(&index_of_biggest_in_corner);
       index_of_biggest_in_corner += k;
 
-      transpositions.coeffRef(k) = index_of_biggest_in_corner;
+      transpositions.coeffRef(k) = IndexType(index_of_biggest_in_corner);
       if(k != index_of_biggest_in_corner)
       {
         // apply the transposition while taking care to consider only
@@ -295,7 +328,7 @@ template<> struct ldlt_inplace<Lower>
         mat.row(k).head(k).swap(mat.row(index_of_biggest_in_corner).head(k));
         mat.col(k).tail(s).swap(mat.col(index_of_biggest_in_corner).tail(s));
         std::swap(mat.coeffRef(k,k),mat.coeffRef(index_of_biggest_in_corner,index_of_biggest_in_corner));
-        for(int i=k+1;i<index_of_biggest_in_corner;++i)
+        for(Index i=k+1;i<index_of_biggest_in_corner;++i)
         {
           Scalar tmp = mat.coeffRef(i,k);
           mat.coeffRef(i,k) = numext::conj(mat.coeffRef(index_of_biggest_in_corner,i));
@@ -321,26 +354,46 @@ template<> struct ldlt_inplace<Lower>
         if(rs>0)
           A21.noalias() -= A20 * temp.head(k);
       }
-      
+
       // In some previous versions of Eigen (e.g., 3.2.1), the scaling was omitted if the pivot
-      // was smaller than the cutoff value. However, soince LDLT is not rank-revealing
-      // we should only make sure we do not introduce INF or NaN values.
-      // LAPACK also uses 0 as the cutoff value.
+      // was smaller than the cutoff value. However, since LDLT is not rank-revealing
+      // we should only make sure that we do not introduce INF or NaN values.
+      // Remark that LAPACK also uses 0 as the cutoff value.
       RealScalar realAkk = numext::real(mat.coeffRef(k,k));
-      if((rs>0) && (abs(realAkk) > RealScalar(0)))
+      bool pivot_is_valid = (abs(realAkk) > RealScalar(0));
+
+      if(k==0 && !pivot_is_valid)
+      {
+        // The entire diagonal is zero, there is nothing more to do
+        // except filling the transpositions, and checking whether the matrix is zero.
+        sign = ZeroSign;
+        for(Index j = 0; j<size; ++j)
+        {
+          transpositions.coeffRef(j) = IndexType(j);
+          ret = ret && (mat.col(j).tail(size-j-1).array()==Scalar(0)).all();
+        }
+        return ret;
+      }
+
+      if((rs>0) && pivot_is_valid)
         A21 /= realAkk;
+      else if(rs>0)
+        ret = ret && (A21.array()==Scalar(0)).all();
+
+      if(found_zero_pivot && pivot_is_valid) ret = false; // factorization failed
+      else if(!pivot_is_valid) found_zero_pivot = true;
 
       if (sign == PositiveSemiDef) {
-        if (realAkk < 0) sign = Indefinite;
+        if (realAkk < static_cast<RealScalar>(0)) sign = Indefinite;
       } else if (sign == NegativeSemiDef) {
-        if (realAkk > 0) sign = Indefinite;
+        if (realAkk > static_cast<RealScalar>(0)) sign = Indefinite;
       } else if (sign == ZeroSign) {
-        if (realAkk > 0) sign = PositiveSemiDef;
-        else if (realAkk < 0) sign = NegativeSemiDef;
+        if (realAkk > static_cast<RealScalar>(0)) sign = PositiveSemiDef;
+        else if (realAkk < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
       }
     }
 
-    return true;
+    return ret;
   }
 
   // Reference for the algorithm: Davis and Hager, "Multiple Rank
@@ -356,7 +409,6 @@ template<> struct ldlt_inplace<Lower>
     using numext::isfinite;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
 
     const Index size = mat.rows();
     eigen_assert(mat.cols() == size && w.size()==size);
@@ -420,16 +472,16 @@ template<typename MatrixType> struct LDLT_Traits<MatrixType,Lower>
 {
   typedef const TriangularView<const MatrixType, UnitLower> MatrixL;
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitUpper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m; }
-  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
 };
 
 template<typename MatrixType> struct LDLT_Traits<MatrixType,Upper>
 {
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitLower> MatrixL;
   typedef const TriangularView<const MatrixType, UnitUpper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m.adjoint(); }
-  static inline MatrixU getU(const MatrixType& m) { return m; }
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
 };
 
 } // end namespace internal
@@ -437,21 +489,35 @@ template<typename MatrixType> struct LDLT_Traits<MatrixType,Upper>
 /** Compute / recompute the LDLT decomposition A = L D L^* = U^* D U of \a matrix
   */
 template<typename MatrixType, int _UpLo>
-LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const MatrixType& a)
+template<typename InputType>
+LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
 {
   check_template_parameters();
-  
+
   eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
 
-  m_matrix = a;
+  m_matrix = a.derived();
+
+  // Compute matrix L1 norm = max abs column sum.
+  m_l1_norm = RealScalar(0);
+  // TODO move this code to SelfAdjointView
+  for (Index col = 0; col < size; ++col) {
+    RealScalar abs_col_sum;
+    if (_UpLo == Lower)
+      abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
+    else
+      abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
+    if (abs_col_sum > m_l1_norm)
+      m_l1_norm = abs_col_sum;
+  }
 
   m_transpositions.resize(size);
   m_isInitialized = false;
   m_temporary.resize(size);
   m_sign = internal::ZeroSign;
 
-  internal::ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, m_sign);
+  m_info = internal::ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, m_sign) ? Success : NumericalIssue;
 
   m_isInitialized = true;
   return *this;
@@ -466,18 +532,19 @@ template<typename MatrixType, int _UpLo>
 template<typename Derived>
 LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Derived>& w, const typename LDLT<MatrixType,_UpLo>::RealScalar& sigma)
 {
+  typedef typename TranspositionType::StorageIndex IndexType;
   const Index size = w.rows();
   if (m_isInitialized)
   {
     eigen_assert(m_matrix.rows()==size);
   }
   else
-  {    
+  {
     m_matrix.resize(size,size);
     m_matrix.setZero();
     m_transpositions.resize(size);
     for (Index i = 0; i < size; i++)
-      m_transpositions.coeffRef(i) = i;
+      m_transpositions.coeffRef(i) = IndexType(i);
     m_temporary.resize(size);
     m_sign = sigma>=0 ? internal::PositiveSemiDef : internal::NegativeSemiDef;
     m_isInitialized = true;
@@ -488,53 +555,46 @@ LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Deri
   return *this;
 }
 
-namespace internal {
-template<typename _MatrixType, int _UpLo, typename Rhs>
-struct solve_retval<LDLT<_MatrixType,_UpLo>, Rhs>
-  : solve_retval_base<LDLT<_MatrixType,_UpLo>, Rhs>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType, int _UpLo>
+template<typename RhsType, typename DstType>
+void LDLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
 {
-  typedef LDLT<_MatrixType,_UpLo> LDLTType;
-  EIGEN_MAKE_SOLVE_HELPERS(LDLTType,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
+  eigen_assert(rhs.rows() == rows());
+  // dst = P b
+  dst = m_transpositions * rhs;
+
+  // dst = L^-1 (P b)
+  matrixL().solveInPlace(dst);
+
+  // dst = D^-1 (L^-1 P b)
+  // more precisely, use pseudo-inverse of D (see bug 241)
+  using std::abs;
+  const typename Diagonal<const MatrixType>::RealReturnType vecD(vectorD());
+  // In some previous versions, tolerance was set to the max of 1/highest (or rather numeric_limits::min())
+  // and the maximal diagonal entry * epsilon as motivated by LAPACK's xGELSS:
+  // RealScalar tolerance = numext::maxi(vecD.array().abs().maxCoeff() * NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
+  // However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
+  // diagonal element is not well justified and leads to numerical issues in some cases.
+  // Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
+  // Using numeric_limits::min() gives us more robustness to denormals.
+  RealScalar tolerance = (std::numeric_limits<RealScalar>::min)();
+
+  for (Index i = 0; i < vecD.size(); ++i)
   {
-    eigen_assert(rhs().rows() == dec().matrixLDLT().rows());
-    // dst = P b
-    dst = dec().transpositionsP() * rhs();
-
-    // dst = L^-1 (P b)
-    dec().matrixL().solveInPlace(dst);
-
-    // dst = D^-1 (L^-1 P b)
-    // more precisely, use pseudo-inverse of D (see bug 241)
-    using std::abs;
-    using std::max;
-    typedef typename LDLTType::MatrixType MatrixType;
-    typedef typename LDLTType::RealScalar RealScalar;
-    const typename Diagonal<const MatrixType>::RealReturnType vectorD(dec().vectorD());
-    // In some previous versions, tolerance was set to the max of 1/highest and the maximal diagonal entry * epsilon
-    // as motivated by LAPACK's xGELSS:
-    // RealScalar tolerance = (max)(vectorD.array().abs().maxCoeff() *NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
-    // However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
-    // diagonal element is not well justified and to numerical issues in some cases.
-    // Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
-    RealScalar tolerance = RealScalar(1) / NumTraits<RealScalar>::highest();
-    
-    for (Index i = 0; i < vectorD.size(); ++i) {
-      if(abs(vectorD(i)) > tolerance)
-        dst.row(i) /= vectorD(i);
-      else
-        dst.row(i).setZero();
-    }
+    if(abs(vecD(i)) > tolerance)
+      dst.row(i) /= vecD(i);
+    else
+      dst.row(i).setZero();
+  }
 
-    // dst = L^-T (D^-1 L^-1 P b)
-    dec().matrixU().solveInPlace(dst);
+  // dst = L^-T (D^-1 L^-1 P b)
+  matrixU().solveInPlace(dst);
 
-    // dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
-    dst = dec().transpositionsP().transpose() * dst;
-  }
-};
+  // dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
+  dst = m_transpositions.transpose() * dst;
 }
+#endif
 
 /** \internal use x = ldlt_object.solve(x);
   *
@@ -588,6 +648,7 @@ MatrixType LDLT<MatrixType,_UpLo>::reconstructedMatrix() const
 
 /** \cholesky_module
   * \returns the Cholesky decomposition with full pivoting without square root of \c *this
+  * \sa MatrixBase::ldlt()
   */
 template<typename MatrixType, unsigned int UpLo>
 inline const LDLT<typename SelfAdjointView<MatrixType, UpLo>::PlainObject, UpLo>
@@ -598,6 +659,7 @@ SelfAdjointView<MatrixType, UpLo>::ldlt() const
 
 /** \cholesky_module
   * \returns the Cholesky decomposition with full pivoting without square root of \c *this
+  * \sa SelfAdjointView::ldlt()
   */
 template<typename Derived>
 inline const LDLT<typename MatrixBase<Derived>::PlainObject>
diff --git a/extern/Eigen3/Eigen/src/Cholesky/LLT.h b/extern/Eigen3/Eigen/src/Cholesky/LLT.h
index 7c11a2dc29a..e1624d21b69 100644
--- a/extern/Eigen3/Eigen/src/Cholesky/LLT.h
+++ b/extern/Eigen3/Eigen/src/Cholesky/LLT.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_LLT_H
 #define EIGEN_LLT_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal{
 template<typename MatrixType, int UpLo> struct LLT_Traits;
@@ -22,9 +22,9 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
   *
   * \brief Standard Cholesky decomposition (LL^T) of a matrix and associated features
   *
-  * \param MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
-  * \param UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
-  *             The other triangular part won't be read.
+  * \tparam _MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
+  * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
+  *               The other triangular part won't be read.
   *
   * This class performs a LL^T Cholesky decomposition of a symmetric, positive definite
   * matrix A such that A = LL^* = U^*U, where L is lower triangular.
@@ -40,12 +40,18 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
   *
   * Example: \include LLT_example.cpp
   * Output: \verbinclude LLT_example.out
-  *    
-  * \sa MatrixBase::llt(), class LDLT
-  */
- /* HEY THIS DOX IS DISABLED BECAUSE THERE's A BUG EITHER HERE OR IN LDLT ABOUT THAT (OR BOTH)
-  * Note that during the decomposition, only the upper triangular part of A is considered. Therefore,
-  * the strict lower part does not have to store correct values.
+  *
+  * \b Performance: for best performance, it is recommended to use a column-major storage format
+  * with the Lower triangular part (the default), or, equivalently, a row-major storage format
+  * with the Upper triangular part. Otherwise, you might get a 20% slowdown for the full factorization
+  * step, and rank-updates can be up to 3 times slower.
+  *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  *
+  * Note that during the decomposition, only the lower (or upper, as defined by _UpLo) triangular part of A is considered.
+  * Therefore, the strict lower part does not have to store correct values.
+  *
+  * \sa MatrixBase::llt(), SelfAdjointView::llt(), class LDLT
   */
 template<typename _MatrixType, int _UpLo> class LLT
 {
@@ -54,12 +60,12 @@ template<typename _MatrixType, int _UpLo> class LLT
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
 
     enum {
       PacketSize = internal::packet_traits<Scalar>::size,
@@ -83,14 +89,30 @@ template<typename _MatrixType, int _UpLo> class LLT
       * according to the specified problem \a size.
       * \sa LLT()
       */
-    LLT(Index size) : m_matrix(size, size),
+    explicit LLT(Index size) : m_matrix(size, size),
                     m_isInitialized(false) {}
 
-    LLT(const MatrixType& matrix)
+    template<typename InputType>
+    explicit LLT(const EigenBase<InputType>& matrix)
       : m_matrix(matrix.rows(), matrix.cols()),
         m_isInitialized(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+    /** \brief Constructs a LDLT factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when
+      * \c MatrixType is a Eigen::Ref.
+      *
+      * \sa LLT(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit LLT(EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
+        m_isInitialized(false)
+    {
+      compute(matrix.derived());
     }
 
     /** \returns a view of the upper triangular matrix U */
@@ -115,33 +137,33 @@ template<typename _MatrixType, int _UpLo> class LLT
       * Example: \include LLT_solve.cpp
       * Output: \verbinclude LLT_solve.out
       *
-      * \sa solveInPlace(), MatrixBase::llt()
+      * \sa solveInPlace(), MatrixBase::llt(), SelfAdjointView::llt()
       */
     template<typename Rhs>
-    inline const internal::solve_retval<LLT, Rhs>
+    inline const Solve<LLT, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "LLT is not initialized.");
       eigen_assert(m_matrix.rows()==b.rows()
                 && "LLT::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<LLT, Rhs>(*this, b.derived());
-    }
-
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived, typename ResultType>
-    bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const
-    {
-      *result = this->solve(b);
-      return true;
+      return Solve<LLT, Rhs>(*this, b.derived());
     }
-    
-    bool isPositiveDefinite() const { return true; }
-    #endif
 
     template<typename Derived>
-    void solveInPlace(MatrixBase<Derived> &bAndX) const;
+    void solveInPlace(const MatrixBase<Derived> &bAndX) const;
 
-    LLT& compute(const MatrixType& matrix);
+    template<typename InputType>
+    LLT& compute(const EigenBase<InputType>& matrix);
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of
+      *  which \c *this is the Cholesky decomposition.
+      */
+    RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(m_info == Success && "LLT failed because matrix appears to be negative");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
+    }
 
     /** \returns the LLT decomposition matrix
       *
@@ -159,7 +181,7 @@ template<typename _MatrixType, int _UpLo> class LLT
     /** \brief Reports whether previous computation was successful.
       *
       * \returns \c Success if computation was succesful,
-      *          \c NumericalIssue if the matrix.appears to be negative.
+      *          \c NumericalIssue if the matrix.appears not to be positive definite.
       */
     ComputationInfo info() const
     {
@@ -167,24 +189,38 @@ template<typename _MatrixType, int _UpLo> class LLT
       return m_info;
     }
 
+    /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
+      *
+      * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
+      * \code x = decomposition.adjoint().solve(b) \endcode
+      */
+    const LLT& adjoint() const { return *this; };
+
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
 
     template<typename VectorType>
     LLT rankUpdate(const VectorType& vec, const RealScalar& sigma = 1);
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
+
   protected:
-    
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
-    
+
     /** \internal
       * Used to compute and store L
       * The strict upper part is not used and even not initialized.
       */
     MatrixType m_matrix;
+    RealScalar m_l1_norm;
     bool m_isInitialized;
     ComputationInfo m_info;
 };
@@ -194,12 +230,11 @@ namespace internal {
 template<typename Scalar, int UpLo> struct llt_inplace;
 
 template<typename MatrixType, typename VectorType>
-static typename MatrixType::Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma)
+static Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma)
 {
   using std::sqrt;
   typedef typename MatrixType::Scalar Scalar;
   typedef typename MatrixType::RealScalar RealScalar;
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::ColXpr ColXpr;
   typedef typename internal::remove_all<ColXpr>::type ColXprCleaned;
   typedef typename ColXprCleaned::SegmentReturnType ColXprSegment;
@@ -268,11 +303,10 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
   template<typename MatrixType>
-  static typename MatrixType::Index unblocked(MatrixType& mat)
+  static Index unblocked(MatrixType& mat)
   {
     using std::sqrt;
-    typedef typename MatrixType::Index Index;
-    
+
     eigen_assert(mat.rows()==mat.cols());
     const Index size = mat.rows();
     for(Index k = 0; k < size; ++k)
@@ -295,9 +329,8 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
   }
 
   template<typename MatrixType>
-  static typename MatrixType::Index blocked(MatrixType& m)
+  static Index blocked(MatrixType& m)
   {
-    typedef typename MatrixType::Index Index;
     eigen_assert(m.rows()==m.cols());
     Index size = m.rows();
     if(size<32)
@@ -322,36 +355,36 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
       Index ret;
       if((ret=unblocked(A11))>=0) return k+ret;
       if(rs>0) A11.adjoint().template triangularView<Upper>().template solveInPlace<OnTheRight>(A21);
-      if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,-1); // bottleneck
+      if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,typename NumTraits<RealScalar>::Literal(-1)); // bottleneck
     }
     return -1;
   }
 
   template<typename MatrixType, typename VectorType>
-  static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
   {
     return Eigen::internal::llt_rank_update_lower(mat, vec, sigma);
   }
 };
-  
+
 template<typename Scalar> struct llt_inplace<Scalar, Upper>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
 
   template<typename MatrixType>
-  static EIGEN_STRONG_INLINE typename MatrixType::Index unblocked(MatrixType& mat)
+  static EIGEN_STRONG_INLINE Index unblocked(MatrixType& mat)
   {
     Transpose<MatrixType> matt(mat);
     return llt_inplace<Scalar, Lower>::unblocked(matt);
   }
   template<typename MatrixType>
-  static EIGEN_STRONG_INLINE typename MatrixType::Index blocked(MatrixType& mat)
+  static EIGEN_STRONG_INLINE Index blocked(MatrixType& mat)
   {
     Transpose<MatrixType> matt(mat);
     return llt_inplace<Scalar, Lower>::blocked(matt);
   }
   template<typename MatrixType, typename VectorType>
-  static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
   {
     Transpose<MatrixType> matt(mat);
     return llt_inplace<Scalar, Lower>::rankUpdate(matt, vec.conjugate(), sigma);
@@ -362,8 +395,8 @@ template<typename MatrixType> struct LLT_Traits<MatrixType,Lower>
 {
   typedef const TriangularView<const MatrixType, Lower> MatrixL;
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, Upper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m; }
-  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
   static bool inplace_decomposition(MatrixType& m)
   { return llt_inplace<typename MatrixType::Scalar, Lower>::blocked(m)==-1; }
 };
@@ -372,8 +405,8 @@ template<typename MatrixType> struct LLT_Traits<MatrixType,Upper>
 {
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, Lower> MatrixL;
   typedef const TriangularView<const MatrixType, Upper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m.adjoint(); }
-  static inline MatrixU getU(const MatrixType& m) { return m; }
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
   static bool inplace_decomposition(MatrixType& m)
   { return llt_inplace<typename MatrixType::Scalar, Upper>::blocked(m)==-1; }
 };
@@ -388,14 +421,29 @@ template<typename MatrixType> struct LLT_Traits<MatrixType,Upper>
   * Output: \verbinclude TutorialLinAlgComputeTwice.out
   */
 template<typename MatrixType, int _UpLo>
-LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const MatrixType& a)
+template<typename InputType>
+LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
 {
   check_template_parameters();
-  
+
   eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
   m_matrix.resize(size, size);
-  m_matrix = a;
+  if (!internal::is_same_dense(m_matrix, a.derived()))
+    m_matrix = a.derived();
+
+  // Compute matrix L1 norm = max abs column sum.
+  m_l1_norm = RealScalar(0);
+  // TODO move this code to SelfAdjointView
+  for (Index col = 0; col < size; ++col) {
+    RealScalar abs_col_sum;
+    if (_UpLo == Lower)
+      abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
+    else
+      abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
+    if (abs_col_sum > m_l1_norm)
+      m_l1_norm = abs_col_sum;
+  }
 
   m_isInitialized = true;
   bool ok = Traits::inplace_decomposition(m_matrix);
@@ -423,39 +471,33 @@ LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, c
 
   return *this;
 }
-    
-namespace internal {
-template<typename _MatrixType, int UpLo, typename Rhs>
-struct solve_retval<LLT<_MatrixType, UpLo>, Rhs>
-  : solve_retval_base<LLT<_MatrixType, UpLo>, Rhs>
-{
-  typedef LLT<_MatrixType,UpLo> LLTType;
-  EIGEN_MAKE_SOLVE_HELPERS(LLTType,Rhs)
 
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dst = rhs();
-    dec().solveInPlace(dst);
-  }
-};
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType,int _UpLo>
+template<typename RhsType, typename DstType>
+void LLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
+{
+  dst = rhs;
+  solveInPlace(dst);
 }
+#endif
 
 /** \internal use x = llt_object.solve(x);
-  * 
+  *
   * This is the \em in-place version of solve().
   *
   * \param bAndX represents both the right-hand side matrix b and result x.
   *
-  * \returns true always! If you need to check for existence of solutions, use another decomposition like LU, QR, or SVD.
+  * This version avoids a copy when the right hand side matrix b is not needed anymore.
   *
-  * This version avoids a copy when the right hand side matrix b is not
-  * needed anymore.
+  * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
+  * This function will const_cast it, so constness isn't honored here.
   *
   * \sa LLT::solve(), MatrixBase::llt()
   */
 template<typename MatrixType, int _UpLo>
 template<typename Derived>
-void LLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
+void LLT<MatrixType,_UpLo>::solveInPlace(const MatrixBase<Derived> &bAndX) const
 {
   eigen_assert(m_isInitialized && "LLT is not initialized.");
   eigen_assert(m_matrix.rows()==bAndX.rows());
@@ -475,6 +517,7 @@ MatrixType LLT<MatrixType,_UpLo>::reconstructedMatrix() const
 
 /** \cholesky_module
   * \returns the LLT decomposition of \c *this
+  * \sa SelfAdjointView::llt()
   */
 template<typename Derived>
 inline const LLT<typename MatrixBase<Derived>::PlainObject>
@@ -485,6 +528,7 @@ MatrixBase<Derived>::llt() const
 
 /** \cholesky_module
   * \returns the LLT decomposition of \c *this
+  * \sa SelfAdjointView::llt()
   */
 template<typename MatrixType, unsigned int UpLo>
 inline const LLT<typename SelfAdjointView<MatrixType, UpLo>::PlainObject, UpLo>
diff --git a/extern/Eigen3/Eigen/src/Cholesky/LLT_LAPACKE.h b/extern/Eigen3/Eigen/src/Cholesky/LLT_LAPACKE.h
new file mode 100644
index 00000000000..bc6489e69a9
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Cholesky/LLT_LAPACKE.h
@@ -0,0 +1,99 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *     LLt decomposition based on LAPACKE_?potrf function.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_LLT_LAPACKE_H
+#define EIGEN_LLT_LAPACKE_H
+
+namespace Eigen { 
+
+namespace internal {
+
+template<typename Scalar> struct lapacke_llt;
+
+#define EIGEN_LAPACKE_LLT(EIGTYPE, BLASTYPE, LAPACKE_PREFIX) \
+template<> struct lapacke_llt<EIGTYPE> \
+{ \
+  template<typename MatrixType> \
+  static inline Index potrf(MatrixType& m, char uplo) \
+  { \
+    lapack_int matrix_order; \
+    lapack_int size, lda, info, StorageOrder; \
+    EIGTYPE* a; \
+    eigen_assert(m.rows()==m.cols()); \
+    /* Set up parameters for ?potrf */ \
+    size = convert_index<lapack_int>(m.rows()); \
+    StorageOrder = MatrixType::Flags&RowMajorBit?RowMajor:ColMajor; \
+    matrix_order = StorageOrder==RowMajor ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \
+    a = &(m.coeffRef(0,0)); \
+    lda = convert_index<lapack_int>(m.outerStride()); \
+\
+    info = LAPACKE_##LAPACKE_PREFIX##potrf( matrix_order, uplo, size, (BLASTYPE*)a, lda ); \
+    info = (info==0) ? -1 : info>0 ? info-1 : size; \
+    return info; \
+  } \
+}; \
+template<> struct llt_inplace<EIGTYPE, Lower> \
+{ \
+  template<typename MatrixType> \
+  static Index blocked(MatrixType& m) \
+  { \
+    return lapacke_llt<EIGTYPE>::potrf(m, 'L'); \
+  } \
+  template<typename MatrixType, typename VectorType> \
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
+  { return Eigen::internal::llt_rank_update_lower(mat, vec, sigma); } \
+}; \
+template<> struct llt_inplace<EIGTYPE, Upper> \
+{ \
+  template<typename MatrixType> \
+  static Index blocked(MatrixType& m) \
+  { \
+    return lapacke_llt<EIGTYPE>::potrf(m, 'U'); \
+  } \
+  template<typename MatrixType, typename VectorType> \
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
+  { \
+    Transpose<MatrixType> matt(mat); \
+    return llt_inplace<EIGTYPE, Lower>::rankUpdate(matt, vec.conjugate(), sigma); \
+  } \
+};
+
+EIGEN_LAPACKE_LLT(double, double, d)
+EIGEN_LAPACKE_LLT(float, float, s)
+EIGEN_LAPACKE_LLT(dcomplex, lapack_complex_double, z)
+EIGEN_LAPACKE_LLT(scomplex, lapack_complex_float, c)
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_LLT_LAPACKE_H
diff --git a/extern/Eigen3/Eigen/src/CholmodSupport/CholmodSupport.h b/extern/Eigen3/Eigen/src/CholmodSupport/CholmodSupport.h
index 99dbe171c36..57197202383 100644
--- a/extern/Eigen3/Eigen/src/CholmodSupport/CholmodSupport.h
+++ b/extern/Eigen3/Eigen/src/CholmodSupport/CholmodSupport.h
@@ -14,46 +14,52 @@ namespace Eigen {
 
 namespace internal {
 
-template<typename Scalar, typename CholmodType>
-void cholmod_configure_matrix(CholmodType& mat)
-{
-  if (internal::is_same<Scalar,float>::value)
-  {
-    mat.xtype = CHOLMOD_REAL;
-    mat.dtype = CHOLMOD_SINGLE;
-  }
-  else if (internal::is_same<Scalar,double>::value)
-  {
+template<typename Scalar> struct cholmod_configure_matrix;
+
+template<> struct cholmod_configure_matrix<double> {
+  template<typename CholmodType>
+  static void run(CholmodType& mat) {
     mat.xtype = CHOLMOD_REAL;
     mat.dtype = CHOLMOD_DOUBLE;
   }
-  else if (internal::is_same<Scalar,std::complex<float> >::value)
-  {
-    mat.xtype = CHOLMOD_COMPLEX;
-    mat.dtype = CHOLMOD_SINGLE;
-  }
-  else if (internal::is_same<Scalar,std::complex<double> >::value)
-  {
+};
+
+template<> struct cholmod_configure_matrix<std::complex<double> > {
+  template<typename CholmodType>
+  static void run(CholmodType& mat) {
     mat.xtype = CHOLMOD_COMPLEX;
     mat.dtype = CHOLMOD_DOUBLE;
   }
-  else
-  {
-    eigen_assert(false && "Scalar type not supported by CHOLMOD");
-  }
-}
+};
+
+// Other scalar types are not yet suppotred by Cholmod
+// template<> struct cholmod_configure_matrix<float> {
+//   template<typename CholmodType>
+//   static void run(CholmodType& mat) {
+//     mat.xtype = CHOLMOD_REAL;
+//     mat.dtype = CHOLMOD_SINGLE;
+//   }
+// };
+//
+// template<> struct cholmod_configure_matrix<std::complex<float> > {
+//   template<typename CholmodType>
+//   static void run(CholmodType& mat) {
+//     mat.xtype = CHOLMOD_COMPLEX;
+//     mat.dtype = CHOLMOD_SINGLE;
+//   }
+// };
 
 } // namespace internal
 
 /** Wraps the Eigen sparse matrix \a mat into a Cholmod sparse matrix object.
   * Note that the data are shared.
   */
-template<typename _Scalar, int _Options, typename _Index>
-cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
+template<typename _Scalar, int _Options, typename _StorageIndex>
+cholmod_sparse viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_StorageIndex> > mat)
 {
   cholmod_sparse res;
   res.nzmax   = mat.nonZeros();
-  res.nrow    = mat.rows();;
+  res.nrow    = mat.rows();
   res.ncol    = mat.cols();
   res.p       = mat.outerIndexPtr();
   res.i       = mat.innerIndexPtr();
@@ -74,11 +80,11 @@ cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
   res.dtype   = 0;
   res.stype   = -1;
   
-  if (internal::is_same<_Index,int>::value)
+  if (internal::is_same<_StorageIndex,int>::value)
   {
     res.itype = CHOLMOD_INT;
   }
-  else if (internal::is_same<_Index,SuiteSparse_long>::value)
+  else if (internal::is_same<_StorageIndex,long>::value)
   {
     res.itype = CHOLMOD_LONG;
   }
@@ -88,7 +94,7 @@ cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
   }
 
   // setup res.xtype
-  internal::cholmod_configure_matrix<_Scalar>(res);
+  internal::cholmod_configure_matrix<_Scalar>::run(res);
   
   res.stype = 0;
   
@@ -98,16 +104,23 @@ cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
 template<typename _Scalar, int _Options, typename _Index>
 const cholmod_sparse viewAsCholmod(const SparseMatrix<_Scalar,_Options,_Index>& mat)
 {
-  cholmod_sparse res = viewAsCholmod(mat.const_cast_derived());
+  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
+  return res;
+}
+
+template<typename _Scalar, int _Options, typename _Index>
+const cholmod_sparse viewAsCholmod(const SparseVector<_Scalar,_Options,_Index>& mat)
+{
+  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
   return res;
 }
 
 /** Returns a view of the Eigen sparse matrix \a mat as Cholmod sparse matrix.
   * The data are not copied but shared. */
 template<typename _Scalar, int _Options, typename _Index, unsigned int UpLo>
-cholmod_sparse viewAsCholmod(const SparseSelfAdjointView<SparseMatrix<_Scalar,_Options,_Index>, UpLo>& mat)
+cholmod_sparse viewAsCholmod(const SparseSelfAdjointView<const SparseMatrix<_Scalar,_Options,_Index>, UpLo>& mat)
 {
-  cholmod_sparse res = viewAsCholmod(mat.matrix().const_cast_derived());
+  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.matrix().const_cast_derived()));
   
   if(UpLo==Upper) res.stype =  1;
   if(UpLo==Lower) res.stype = -1;
@@ -131,19 +144,19 @@ cholmod_dense viewAsCholmod(MatrixBase<Derived>& mat)
   res.x      = (void*)(mat.derived().data());
   res.z      = 0;
 
-  internal::cholmod_configure_matrix<Scalar>(res);
+  internal::cholmod_configure_matrix<Scalar>::run(res);
 
   return res;
 }
 
 /** Returns a view of the Cholmod sparse matrix \a cm as an Eigen sparse matrix.
   * The data are not copied but shared. */
-template<typename Scalar, int Flags, typename Index>
-MappedSparseMatrix<Scalar,Flags,Index> viewAsEigen(cholmod_sparse& cm)
+template<typename Scalar, int Flags, typename StorageIndex>
+MappedSparseMatrix<Scalar,Flags,StorageIndex> viewAsEigen(cholmod_sparse& cm)
 {
-  return MappedSparseMatrix<Scalar,Flags,Index>
-         (cm.nrow, cm.ncol, static_cast<Index*>(cm.p)[cm.ncol],
-          static_cast<Index*>(cm.p), static_cast<Index*>(cm.i),static_cast<Scalar*>(cm.x) );
+  return MappedSparseMatrix<Scalar,Flags,StorageIndex>
+         (cm.nrow, cm.ncol, static_cast<StorageIndex*>(cm.p)[cm.ncol],
+          static_cast<StorageIndex*>(cm.p), static_cast<StorageIndex*>(cm.i),static_cast<Scalar*>(cm.x) );
 }
 
 enum CholmodMode {
@@ -157,29 +170,39 @@ enum CholmodMode {
   * \sa class CholmodSupernodalLLT, class CholmodSimplicialLDLT, class CholmodSimplicialLLT
   */
 template<typename _MatrixType, int _UpLo, typename Derived>
-class CholmodBase : internal::noncopyable
+class CholmodBase : public SparseSolverBase<Derived>
 {
+  protected:
+    typedef SparseSolverBase<Derived> Base;
+    using Base::derived;
+    using Base::m_isInitialized;
   public:
     typedef _MatrixType MatrixType;
     enum { UpLo = _UpLo };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
     typedef MatrixType CholMatrixType;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
 
   public:
 
     CholmodBase()
-      : m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
+      : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
     {
-      m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
+      EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
+      m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
       cholmod_start(&m_cholmod);
     }
 
-    CholmodBase(const MatrixType& matrix)
-      : m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
+    explicit CholmodBase(const MatrixType& matrix)
+      : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
     {
-      m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
+      EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
+      m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
       cholmod_start(&m_cholmod);
       compute(matrix);
     }
@@ -191,11 +214,8 @@ class CholmodBase : internal::noncopyable
       cholmod_finish(&m_cholmod);
     }
     
-    inline Index cols() const { return m_cholmodFactor->n; }
-    inline Index rows() const { return m_cholmodFactor->n; }
-    
-    Derived& derived() { return *static_cast<Derived*>(this); }
-    const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    inline StorageIndex cols() const { return internal::convert_index<StorageIndex, Index>(m_cholmodFactor->n); }
+    inline StorageIndex rows() const { return internal::convert_index<StorageIndex, Index>(m_cholmodFactor->n); }
     
     /** \brief Reports whether previous computation was successful.
       *
@@ -216,34 +236,6 @@ class CholmodBase : internal::noncopyable
       return derived();
     }
     
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<CholmodBase, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "LLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<CholmodBase, Rhs>(*this, b.derived());
-    }
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<CholmodBase, Rhs>
-    solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "LLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<CholmodBase, Rhs>(*this, b.derived());
-    }
-    
     /** Performs a symbolic decomposition on the sparsity pattern of \a matrix.
       *
       * This function is particularly useful when solving for several problems having the same structure.
@@ -277,7 +269,7 @@ class CholmodBase : internal::noncopyable
       eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
       cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
       cholmod_factorize_p(&A, m_shiftOffset, 0, 0, m_cholmodFactor, &m_cholmod);
-      
+
       // If the factorization failed, minor is the column at which it did. On success minor == n.
       this->m_info = (m_cholmodFactor->minor == m_cholmodFactor->n ? Success : NumericalIssue);
       m_factorizationIsOk = true;
@@ -290,20 +282,22 @@ class CholmodBase : internal::noncopyable
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
     {
       eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
       const Index size = m_cholmodFactor->n;
       EIGEN_UNUSED_VARIABLE(size);
       eigen_assert(size==b.rows());
+      
+      // Cholmod needs column-major stoarge without inner-stride, which corresponds to the default behavior of Ref.
+      Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b.derived());
 
-      // note: cd stands for Cholmod Dense
-      Rhs& b_ref(b.const_cast_derived());
       cholmod_dense b_cd = viewAsCholmod(b_ref);
       cholmod_dense* x_cd = cholmod_solve(CHOLMOD_A, m_cholmodFactor, &b_cd, &m_cholmod);
       if(!x_cd)
       {
         this->m_info = NumericalIssue;
+        return;
       }
       // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
       dest = Matrix<Scalar,Dest::RowsAtCompileTime,Dest::ColsAtCompileTime>::Map(reinterpret_cast<Scalar*>(x_cd->x),b.rows(),b.cols());
@@ -311,8 +305,8 @@ class CholmodBase : internal::noncopyable
     }
     
     /** \internal */
-    template<typename RhsScalar, int RhsOptions, typename RhsIndex, typename DestScalar, int DestOptions, typename DestIndex>
-    void _solve(const SparseMatrix<RhsScalar,RhsOptions,RhsIndex> &b, SparseMatrix<DestScalar,DestOptions,DestIndex> &dest) const
+    template<typename RhsDerived, typename DestDerived>
+    void _solve_impl(const SparseMatrixBase<RhsDerived> &b, SparseMatrixBase<DestDerived> &dest) const
     {
       eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
       const Index size = m_cholmodFactor->n;
@@ -320,14 +314,16 @@ class CholmodBase : internal::noncopyable
       eigen_assert(size==b.rows());
 
       // note: cs stands for Cholmod Sparse
-      cholmod_sparse b_cs = viewAsCholmod(b);
+      Ref<SparseMatrix<typename RhsDerived::Scalar,ColMajor,typename RhsDerived::StorageIndex> > b_ref(b.const_cast_derived());
+      cholmod_sparse b_cs = viewAsCholmod(b_ref);
       cholmod_sparse* x_cs = cholmod_spsolve(CHOLMOD_A, m_cholmodFactor, &b_cs, &m_cholmod);
       if(!x_cs)
       {
         this->m_info = NumericalIssue;
+        return;
       }
       // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
-      dest = viewAsEigen<DestScalar,DestOptions,DestIndex>(*x_cs);
+      dest.derived() = viewAsEigen<typename DestDerived::Scalar,ColMajor,typename DestDerived::StorageIndex>(*x_cs);
       cholmod_free_sparse(&x_cs, &m_cholmod);
     }
     #endif // EIGEN_PARSED_BY_DOXYGEN
@@ -344,10 +340,61 @@ class CholmodBase : internal::noncopyable
       */
     Derived& setShift(const RealScalar& offset)
     {
-      m_shiftOffset[0] = offset;
+      m_shiftOffset[0] = double(offset);
       return derived();
     }
     
+    /** \returns the determinant of the underlying matrix from the current factorization */
+    Scalar determinant() const
+    {
+      using std::exp;
+      return exp(logDeterminant());
+    }
+
+    /** \returns the log determinant of the underlying matrix from the current factorization */
+    Scalar logDeterminant() const
+    {
+      using std::log;
+      using numext::real;
+      eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
+
+      RealScalar logDet = 0;
+      Scalar *x = static_cast<Scalar*>(m_cholmodFactor->x);
+      if (m_cholmodFactor->is_super)
+      {
+        // Supernodal factorization stored as a packed list of dense column-major blocs,
+        // as described by the following structure:
+
+        // super[k] == index of the first column of the j-th super node
+        StorageIndex *super = static_cast<StorageIndex*>(m_cholmodFactor->super);
+        // pi[k] == offset to the description of row indices
+        StorageIndex *pi = static_cast<StorageIndex*>(m_cholmodFactor->pi);
+        // px[k] == offset to the respective dense block
+        StorageIndex *px = static_cast<StorageIndex*>(m_cholmodFactor->px);
+
+        Index nb_super_nodes = m_cholmodFactor->nsuper;
+        for (Index k=0; k < nb_super_nodes; ++k)
+        {
+          StorageIndex ncols = super[k + 1] - super[k];
+          StorageIndex nrows = pi[k + 1] - pi[k];
+
+          Map<const Array<Scalar,1,Dynamic>, 0, InnerStride<> > sk(x + px[k], ncols, InnerStride<>(nrows+1));
+          logDet += sk.real().log().sum();
+        }
+      }
+      else
+      {
+        // Simplicial factorization stored as standard CSC matrix.
+        StorageIndex *p = static_cast<StorageIndex*>(m_cholmodFactor->p);
+        Index size = m_cholmodFactor->n;
+        for (Index k=0; k<size; ++k)
+          logDet += log(real( x[p[k]] ));
+      }
+      if (m_cholmodFactor->is_ll)
+        logDet *= 2.0;
+      return logDet;
+    };
+
     template<typename Stream>
     void dumpMemory(Stream& /*s*/)
     {}
@@ -355,9 +402,8 @@ class CholmodBase : internal::noncopyable
   protected:
     mutable cholmod_common m_cholmod;
     cholmod_factor* m_cholmodFactor;
-    RealScalar m_shiftOffset[2];
+    double m_shiftOffset[2];
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     int m_factorizationIsOk;
     int m_analysisIsOk;
 };
@@ -376,9 +422,13 @@ class CholmodBase : internal::noncopyable
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
+  * \implsparsesolverconcept
+  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \sa \ref TutorialSparseDirectSolvers, class CholmodSupernodalLLT, class SimplicialLLT
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLLT
   */
 template<typename _MatrixType, int _UpLo = Lower>
 class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLLT<_MatrixType, _UpLo> >
@@ -395,7 +445,7 @@ class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimpl
     CholmodSimplicialLLT(const MatrixType& matrix) : Base()
     {
       init();
-      Base::compute(matrix);
+      this->compute(matrix);
     }
 
     ~CholmodSimplicialLLT() {}
@@ -423,9 +473,13 @@ class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimpl
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
+  * \implsparsesolverconcept
+  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \sa \ref TutorialSparseDirectSolvers, class CholmodSupernodalLLT, class SimplicialLDLT
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLDLT
   */
 template<typename _MatrixType, int _UpLo = Lower>
 class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLDLT<_MatrixType, _UpLo> >
@@ -442,7 +496,7 @@ class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimp
     CholmodSimplicialLDLT(const MatrixType& matrix) : Base()
     {
       init();
-      Base::compute(matrix);
+      this->compute(matrix);
     }
 
     ~CholmodSimplicialLDLT() {}
@@ -468,9 +522,13 @@ class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimp
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
+  * \implsparsesolverconcept
+  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept
   */
 template<typename _MatrixType, int _UpLo = Lower>
 class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSupernodalLLT<_MatrixType, _UpLo> >
@@ -487,7 +545,7 @@ class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSuper
     CholmodSupernodalLLT(const MatrixType& matrix) : Base()
     {
       init();
-      Base::compute(matrix);
+      this->compute(matrix);
     }
 
     ~CholmodSupernodalLLT() {}
@@ -515,9 +573,13 @@ class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSuper
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
+  * \implsparsesolverconcept
+  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept
   */
 template<typename _MatrixType, int _UpLo = Lower>
 class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecomposition<_MatrixType, _UpLo> >
@@ -534,7 +596,7 @@ class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecom
     CholmodDecomposition(const MatrixType& matrix) : Base()
     {
       init();
-      Base::compute(matrix);
+      this->compute(matrix);
     }
 
     ~CholmodDecomposition() {}
@@ -572,36 +634,6 @@ class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecom
     }
 };
 
-namespace internal {
-  
-template<typename _MatrixType, int _UpLo, typename Derived, typename Rhs>
-struct solve_retval<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
-  : solve_retval_base<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
-{
-  typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, int _UpLo, typename Derived, typename Rhs>
-struct sparse_solve_retval<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
-  : sparse_solve_retval_base<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
-{
-  typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_CHOLMODSUPPORT_H
diff --git a/extern/Eigen3/Eigen/src/Core/Array.h b/extern/Eigen3/Eigen/src/Core/Array.h
index 0b9c38c8219..16770fc7b3a 100644
--- a/extern/Eigen3/Eigen/src/Core/Array.h
+++ b/extern/Eigen3/Eigen/src/Core/Array.h
@@ -12,7 +12,16 @@
 
 namespace Eigen {
 
-/** \class Array 
+namespace internal {
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+{
+  typedef ArrayXpr XprKind;
+  typedef ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > XprBase;
+};
+}
+
+/** \class Array
   * \ingroup Core_Module
   *
   * \brief General-purpose arrays with easy API for coefficient-wise operations
@@ -24,20 +33,14 @@ namespace Eigen {
   * API for the %Matrix class provides easy access to linear-algebra
   * operations.
   *
+  * See documentation of class Matrix for detailed information on the template parameters
+  * storage layout.
+  *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
   *
-  * \sa \ref TutorialArrayClass, \ref TopicClassHierarchy
+  * \sa \blank \ref TutorialArrayClass, \ref TopicClassHierarchy
   */
-namespace internal {
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
-{
-  typedef ArrayXpr XprKind;
-  typedef ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > XprBase;
-};
-}
-
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 class Array
   : public PlainObjectBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
@@ -69,11 +72,27 @@ class Array
       * the usage of 'using'. This should be done only for operator=.
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array& operator=(const EigenBase<OtherDerived> &other)
     {
       return Base::operator=(other);
     }
 
+    /** Set all the entries to \a value.
+      * \sa DenseBase::setConstant(), DenseBase::fill()
+      */
+    /* This overload is needed because the usage of
+      *   using Base::operator=;
+      * fails on MSVC. Since the code below is working with GCC and MSVC, we skipped
+      * the usage of 'using'. This should be done only for operator=.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array& operator=(const Scalar &value)
+    {
+      Base::setConstant(value);
+      return *this;
+    }
+
     /** Copies the value of the expression \a other into \c *this with automatic resizing.
       *
       * *this might be resized to match the dimensions of \a other. If *this was a null matrix (not already initialized),
@@ -84,7 +103,8 @@ class Array
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array& operator=(const ArrayBase<OtherDerived>& other)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array& operator=(const DenseBase<OtherDerived>& other)
     {
       return Base::_set(other);
     }
@@ -92,11 +112,12 @@ class Array
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array& operator=(const Array& other)
     {
       return Base::_set(other);
     }
-
+    
     /** Default constructor.
       *
       * For fixed-size matrices, does nothing.
@@ -107,6 +128,7 @@ class Array
       *
       * \sa resize(Index,Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array() : Base()
     {
       Base::_check_template_params();
@@ -116,6 +138,7 @@ class Array
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     // FIXME is it still needed ??
     /** \internal */
+    EIGEN_DEVICE_FUNC
     Array(internal::constructor_without_unaligned_array_assert)
       : Base(internal::constructor_without_unaligned_array_assert())
     {
@@ -124,56 +147,62 @@ class Array
     }
 #endif
 
-#ifdef EIGEN_HAVE_RVALUE_REFERENCES
-    Array(Array&& other)
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    Array(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
       : Base(std::move(other))
     {
       Base::_check_template_params();
-      if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic)
-        Base::_set_noalias(other);
     }
-    Array& operator=(Array&& other)
+    EIGEN_DEVICE_FUNC
+    Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
     {
       other.swap(*this);
       return *this;
     }
 #endif
 
-    /** Constructs a vector or row-vector with given dimension. \only_for_vectors
-      *
-      * Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
-      * it is redundant to pass the dimension here, so it makes more sense to use the default
-      * constructor Matrix() instead.
-      */
-    EIGEN_STRONG_INLINE explicit Array(Index dim)
-      : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE explicit Array(const T& x)
     {
       Base::_check_template_params();
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Array)
-      eigen_assert(dim >= 0);
-      eigen_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim);
-      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+      Base::template _init1<T>(x);
     }
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const T0& val0, const T1& val1)
     {
       Base::_check_template_params();
       this->template _init2<T0,T1>(val0, val1);
     }
     #else
-    /** constructs an uninitialized matrix with \a rows rows and \a cols columns.
+    /** \brief Constructs a fixed-sized array initialized with coefficients starting at \a data */
+    EIGEN_DEVICE_FUNC explicit Array(const Scalar *data);
+    /** Constructs a vector or row-vector with given dimension. \only_for_vectors
+      *
+      * Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
+      * it is redundant to pass the dimension here, so it makes more sense to use the default
+      * constructor Array() instead.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE explicit Array(Index dim);
+    /** constructs an initialized 1x1 Array with the given coefficient */
+    Array(const Scalar& value);
+    /** constructs an uninitialized array with \a rows rows and \a cols columns.
       *
-      * This is useful for dynamic-size matrices. For fixed-size matrices,
+      * This is useful for dynamic-size arrays. For fixed-size arrays,
       * it is redundant to pass these parameters, so one should use the default constructor
-      * Matrix() instead. */
+      * Array() instead. */
     Array(Index rows, Index cols);
     /** constructs an initialized 2D vector with given coefficients */
     Array(const Scalar& val0, const Scalar& val1);
     #endif
 
     /** constructs an initialized 3D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2)
     {
       Base::_check_template_params();
@@ -183,6 +212,7 @@ class Array
       m_storage.data()[2] = val2;
     }
     /** constructs an initialized 4D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2, const Scalar& val3)
     {
       Base::_check_template_params();
@@ -193,51 +223,27 @@ class Array
       m_storage.data()[3] = val3;
     }
 
-    explicit Array(const Scalar *data);
-
-    /** Constructor copying the value of the expression \a other */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array(const ArrayBase<OtherDerived>& other)
-             : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
     /** Copy constructor */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Array& other)
-            : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
-    /** Copy constructor with in-place evaluation */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array(const ReturnByValue<OtherDerived>& other)
-    {
-      Base::_check_template_params();
-      Base::resize(other.rows(), other.cols());
-      other.evalTo(*this);
-    }
+            : Base(other)
+    { }
 
-    /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other)
-      : Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
-    {
-      Base::_check_template_params();
-      Base::_resize_to_match(other);
-      *this = other;
-    }
+  private:
+    struct PrivateType {};
+  public:
 
-    /** Override MatrixBase::swap() since for dynamic-sized matrices of same type it is enough to swap the
-      * data pointers.
-      */
+    /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
     template<typename OtherDerived>
-    void swap(ArrayBase<OtherDerived> const & other)
-    { this->_swap(other.derived()); }
-
-    inline Index innerStride() const { return 1; }
-    inline Index outerStride() const { return this->innerSize(); }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other,
+                              typename internal::enable_if<internal::is_convertible<typename OtherDerived::Scalar,Scalar>::value,
+                                                           PrivateType>::type = PrivateType())
+      : Base(other.derived())
+    { }
+
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return this->innerSize(); }
 
     #ifdef EIGEN_ARRAY_PLUGIN
     #include EIGEN_ARRAY_PLUGIN
diff --git a/extern/Eigen3/Eigen/src/Core/ArrayBase.h b/extern/Eigen3/Eigen/src/Core/ArrayBase.h
index 33ff553712e..3dbc7084cda 100644
--- a/extern/Eigen3/Eigen/src/Core/ArrayBase.h
+++ b/extern/Eigen3/Eigen/src/Core/ArrayBase.h
@@ -32,7 +32,7 @@ template<typename ExpressionType> class MatrixWrapper;
   * \tparam Derived is the derived type, e.g., an array or an expression type.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
   *
   * \sa class MatrixBase, \ref TopicClassHierarchy
   */
@@ -47,13 +47,11 @@ template<typename Derived> class ArrayBase
     typedef ArrayBase Eigen_BaseClassForSpecializationOfGlobalMathFuncImpl;
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
     typedef DenseBase<Derived> Base;
-    using Base::operator*;
     using Base::RowsAtCompileTime;
     using Base::ColsAtCompileTime;
     using Base::SizeAtCompileTime;
@@ -62,8 +60,7 @@ template<typename Derived> class ArrayBase
     using Base::MaxSizeAtCompileTime;
     using Base::IsVectorAtCompileTime;
     using Base::Flags;
-    using Base::CoeffReadCost;
-
+    
     using Base::derived;
     using Base::const_cast_derived;
     using Base::rows;
@@ -83,25 +80,14 @@ template<typename Derived> class ArrayBase
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-    /** \internal the plain matrix type corresponding to this expression. Note that is not necessarily
-      * exactly the return type of eval(): in the case of plain matrices, the return type of eval() is a const
-      * reference to a matrix, not a matrix! It is however guaranteed that the return type of eval() is either
-      * PlainObject or const PlainObject&.
-      */
-    typedef Array<typename internal::traits<Derived>::Scalar,
-                internal::traits<Derived>::RowsAtCompileTime,
-                internal::traits<Derived>::ColsAtCompileTime,
-                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
-                internal::traits<Derived>::MaxRowsAtCompileTime,
-                internal::traits<Derived>::MaxColsAtCompileTime
-          > PlainObject;
-
+    typedef typename Base::PlainObject PlainObject;
 
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::ArrayBase
+#define EIGEN_DOC_UNARY_ADDONS(X,Y)
 #   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
 #   include "../plugins/ArrayCwiseUnaryOps.h"
@@ -112,44 +98,62 @@ template<typename Derived> class ArrayBase
 #     include EIGEN_ARRAYBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
+#undef EIGEN_DOC_UNARY_ADDONS
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const ArrayBase& other)
     {
-      return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
+      internal::call_assignment(derived(), other.derived());
+      return derived();
     }
-
-    Derived& operator+=(const Scalar& scalar)
-    { return *this = derived() + scalar; }
-    Derived& operator-=(const Scalar& scalar)
-    { return *this = derived() - scalar; }
+    
+    /** Set all the entries to \a value.
+      * \sa DenseBase::setConstant(), DenseBase::fill() */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator=(const Scalar &value)
+    { Base::setConstant(value); return derived(); }
+
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator+=(const Scalar& scalar);
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator-=(const Scalar& scalar);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator+=(const ArrayBase<OtherDerived>& other);
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator-=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator*=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator/=(const ArrayBase<OtherDerived>& other);
 
   public:
+    EIGEN_DEVICE_FUNC
     ArrayBase<Derived>& array() { return *this; }
+    EIGEN_DEVICE_FUNC
     const ArrayBase<Derived>& array() const { return *this; }
 
     /** \returns an \link Eigen::MatrixBase Matrix \endlink expression of this array
       * \sa MatrixBase::array() */
-    MatrixWrapper<Derived> matrix() { return derived(); }
-    const MatrixWrapper<const Derived> matrix() const { return derived(); }
+    EIGEN_DEVICE_FUNC
+    MatrixWrapper<Derived> matrix() { return MatrixWrapper<Derived>(derived()); }
+    EIGEN_DEVICE_FUNC
+    const MatrixWrapper<const Derived> matrix() const { return MatrixWrapper<const Derived>(derived()); }
 
 //     template<typename Dest>
 //     inline void evalTo(Dest& dst) const { dst = matrix(); }
 
   protected:
+    EIGEN_DEVICE_FUNC
     ArrayBase() : Base() {}
 
   private:
@@ -171,11 +175,10 @@ template<typename Derived> class ArrayBase
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
 {
-  SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
@@ -185,11 +188,10 @@ ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
@@ -199,11 +201,10 @@ ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<internal::scalar_product_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::mul_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
@@ -213,11 +214,10 @@ ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator/=(const ArrayBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<internal::scalar_quotient_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::div_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
diff --git a/extern/Eigen3/Eigen/src/Core/ArrayWrapper.h b/extern/Eigen3/Eigen/src/Core/ArrayWrapper.h
index b4641e2a01f..688aadd6260 100644
--- a/extern/Eigen3/Eigen/src/Core/ArrayWrapper.h
+++ b/extern/Eigen3/Eigen/src/Core/ArrayWrapper.h
@@ -32,7 +32,8 @@ struct traits<ArrayWrapper<ExpressionType> >
   // Let's remove NestByRefBit
   enum {
     Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
-    Flags = Flags0 & ~NestByRefBit
+    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
+    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
   };
 };
 }
@@ -44,6 +45,7 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
     typedef ArrayBase<ArrayWrapper> Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(ArrayWrapper)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ArrayWrapper)
+    typedef typename internal::remove_all<ExpressionType>::type NestedExpression;
 
     typedef typename internal::conditional<
                        internal::is_lvalue<ExpressionType>::value,
@@ -51,76 +53,45 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
+    typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
 
-    inline ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
+    using Base::coeffRef;
 
+    EIGEN_DEVICE_FUNC
+    explicit EIGEN_STRONG_INLINE ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
+
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return m_expression.innerStride(); }
 
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
+    EIGEN_DEVICE_FUNC
     inline const Scalar* data() const { return m_expression.data(); }
 
-    inline CoeffReturnType coeff(Index rowId, Index colId) const
-    {
-      return m_expression.coeff(rowId, colId);
-    }
-
-    inline Scalar& coeffRef(Index rowId, Index colId)
-    {
-      return m_expression.const_cast_derived().coeffRef(rowId, colId);
-    }
-
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
-      return m_expression.const_cast_derived().coeffRef(rowId, colId);
-    }
-
-    inline CoeffReturnType coeff(Index index) const
-    {
-      return m_expression.coeff(index);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_expression.const_cast_derived().coeffRef(index);
+      return m_expression.coeffRef(rowId, colId);
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
-      return m_expression.const_cast_derived().coeffRef(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index rowId, Index colId) const
-    {
-      return m_expression.template packet<LoadMode>(rowId, colId);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
-    {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(rowId, colId, val);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return m_expression.template packet<LoadMode>(index);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& val)
-    {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(index, val);
+      return m_expression.coeffRef(index);
     }
 
     template<typename Dest>
+    EIGEN_DEVICE_FUNC
     inline void evalTo(Dest& dst) const { dst = m_expression; }
 
     const typename internal::remove_all<NestedExpressionType>::type& 
+    EIGEN_DEVICE_FUNC
     nestedExpression() const 
     {
       return m_expression;
@@ -128,10 +99,12 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
 
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index)  */
-    void resize(Index newSize) { m_expression.const_cast_derived().resize(newSize); }
+    EIGEN_DEVICE_FUNC
+    void resize(Index newSize) { m_expression.resize(newSize); }
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index,Index)*/
-    void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); }
+    EIGEN_DEVICE_FUNC
+    void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
 
   protected:
     NestedExpressionType m_expression;
@@ -157,7 +130,8 @@ struct traits<MatrixWrapper<ExpressionType> >
   // Let's remove NestByRefBit
   enum {
     Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
-    Flags = Flags0 & ~NestByRefBit
+    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
+    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
   };
 };
 }
@@ -169,6 +143,7 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
     typedef MatrixBase<MatrixWrapper<ExpressionType> > Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(MatrixWrapper)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(MatrixWrapper)
+    typedef typename internal::remove_all<ExpressionType>::type NestedExpression;
 
     typedef typename internal::conditional<
                        internal::is_lvalue<ExpressionType>::value,
@@ -176,72 +151,40 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
+    typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
 
-    inline MatrixWrapper(ExpressionType& a_matrix) : m_expression(a_matrix) {}
+    using Base::coeffRef;
 
+    EIGEN_DEVICE_FUNC
+    explicit inline MatrixWrapper(ExpressionType& matrix) : m_expression(matrix) {}
+
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return m_expression.innerStride(); }
 
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
+    EIGEN_DEVICE_FUNC
     inline const Scalar* data() const { return m_expression.data(); }
 
-    inline CoeffReturnType coeff(Index rowId, Index colId) const
-    {
-      return m_expression.coeff(rowId, colId);
-    }
-
-    inline Scalar& coeffRef(Index rowId, Index colId)
-    {
-      return m_expression.const_cast_derived().coeffRef(rowId, colId);
-    }
-
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
       return m_expression.derived().coeffRef(rowId, colId);
     }
 
-    inline CoeffReturnType coeff(Index index) const
-    {
-      return m_expression.coeff(index);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_expression.const_cast_derived().coeffRef(index);
-    }
-
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
-      return m_expression.const_cast_derived().coeffRef(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index rowId, Index colId) const
-    {
-      return m_expression.template packet<LoadMode>(rowId, colId);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
-    {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(rowId, colId, val);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return m_expression.template packet<LoadMode>(index);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& val)
-    {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(index, val);
+      return m_expression.coeffRef(index);
     }
 
+    EIGEN_DEVICE_FUNC
     const typename internal::remove_all<NestedExpressionType>::type& 
     nestedExpression() const 
     {
@@ -250,10 +193,12 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
 
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index)  */
-    void resize(Index newSize) { m_expression.const_cast_derived().resize(newSize); }
+    EIGEN_DEVICE_FUNC
+    void resize(Index newSize) { m_expression.resize(newSize); }
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index,Index)*/
-    void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); }
+    EIGEN_DEVICE_FUNC
+    void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
 
   protected:
     NestedExpressionType m_expression;
diff --git a/extern/Eigen3/Eigen/src/Core/Assign.h b/extern/Eigen3/Eigen/src/Core/Assign.h
index f4817317279..53806ba33c4 100644
--- a/extern/Eigen3/Eigen/src/Core/Assign.h
+++ b/extern/Eigen3/Eigen/src/Core/Assign.h
@@ -14,478 +14,6 @@
 
 namespace Eigen {
 
-namespace internal {
-
-/***************************************************************************
-* Part 1 : the logic deciding a strategy for traversal and unrolling       *
-***************************************************************************/
-
-template <typename Derived, typename OtherDerived>
-struct assign_traits
-{
-public:
-  enum {
-    DstIsAligned = Derived::Flags & AlignedBit,
-    DstHasDirectAccess = Derived::Flags & DirectAccessBit,
-    SrcIsAligned = OtherDerived::Flags & AlignedBit,
-    JointAlignment = bool(DstIsAligned) && bool(SrcIsAligned) ? Aligned : Unaligned
-  };
-
-private:
-  enum {
-    InnerSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::SizeAtCompileTime)
-              : int(Derived::Flags)&RowMajorBit ? int(Derived::ColsAtCompileTime)
-              : int(Derived::RowsAtCompileTime),
-    InnerMaxSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::MaxSizeAtCompileTime)
-              : int(Derived::Flags)&RowMajorBit ? int(Derived::MaxColsAtCompileTime)
-              : int(Derived::MaxRowsAtCompileTime),
-    MaxSizeAtCompileTime = Derived::SizeAtCompileTime,
-    PacketSize = packet_traits<typename Derived::Scalar>::size
-  };
-
-  enum {
-    StorageOrdersAgree = (int(Derived::IsRowMajor) == int(OtherDerived::IsRowMajor)),
-    MightVectorize = StorageOrdersAgree
-                  && (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit),
-    MayInnerVectorize  = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0
-                       && int(DstIsAligned) && int(SrcIsAligned),
-    MayLinearize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit),
-    MayLinearVectorize = MightVectorize && MayLinearize && DstHasDirectAccess
-                       && (DstIsAligned || MaxSizeAtCompileTime == Dynamic),
-      /* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
-         so it's only good for large enough sizes. */
-    MaySliceVectorize  = MightVectorize && DstHasDirectAccess
-                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=3*PacketSize)
-      /* slice vectorization can be slow, so we only want it if the slices are big, which is
-         indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
-         in a fixed-size matrix */
-  };
-
-public:
-  enum {
-    Traversal = int(MayInnerVectorize)  ? int(InnerVectorizedTraversal)
-              : int(MayLinearVectorize) ? int(LinearVectorizedTraversal)
-              : int(MaySliceVectorize)  ? int(SliceVectorizedTraversal)
-              : int(MayLinearize)       ? int(LinearTraversal)
-                                        : int(DefaultTraversal),
-    Vectorized = int(Traversal) == InnerVectorizedTraversal
-              || int(Traversal) == LinearVectorizedTraversal
-              || int(Traversal) == SliceVectorizedTraversal
-  };
-
-private:
-  enum {
-    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * (Vectorized ? int(PacketSize) : 1),
-    MayUnrollCompletely = int(Derived::SizeAtCompileTime) != Dynamic
-                       && int(OtherDerived::CoeffReadCost) != Dynamic
-                       && int(Derived::SizeAtCompileTime) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit),
-    MayUnrollInner      = int(InnerSize) != Dynamic
-                       && int(OtherDerived::CoeffReadCost) != Dynamic
-                       && int(InnerSize) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit)
-  };
-
-public:
-  enum {
-    Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
-                ? (
-                    int(MayUnrollCompletely) ? int(CompleteUnrolling)
-                  : int(MayUnrollInner)      ? int(InnerUnrolling)
-                                             : int(NoUnrolling)
-                  )
-              : int(Traversal) == int(LinearVectorizedTraversal)
-                ? ( bool(MayUnrollCompletely) && bool(DstIsAligned) ? int(CompleteUnrolling) : int(NoUnrolling) )
-              : int(Traversal) == int(LinearTraversal)
-                ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) : int(NoUnrolling) )
-              : int(NoUnrolling)
-  };
-
-#ifdef EIGEN_DEBUG_ASSIGN
-  static void debug()
-  {
-    EIGEN_DEBUG_VAR(DstIsAligned)
-    EIGEN_DEBUG_VAR(SrcIsAligned)
-    EIGEN_DEBUG_VAR(JointAlignment)
-    EIGEN_DEBUG_VAR(InnerSize)
-    EIGEN_DEBUG_VAR(InnerMaxSize)
-    EIGEN_DEBUG_VAR(PacketSize)
-    EIGEN_DEBUG_VAR(StorageOrdersAgree)
-    EIGEN_DEBUG_VAR(MightVectorize)
-    EIGEN_DEBUG_VAR(MayLinearize)
-    EIGEN_DEBUG_VAR(MayInnerVectorize)
-    EIGEN_DEBUG_VAR(MayLinearVectorize)
-    EIGEN_DEBUG_VAR(MaySliceVectorize)
-    EIGEN_DEBUG_VAR(Traversal)
-    EIGEN_DEBUG_VAR(UnrollingLimit)
-    EIGEN_DEBUG_VAR(MayUnrollCompletely)
-    EIGEN_DEBUG_VAR(MayUnrollInner)
-    EIGEN_DEBUG_VAR(Unrolling)
-  }
-#endif
-};
-
-/***************************************************************************
-* Part 2 : meta-unrollers
-***************************************************************************/
-
-/************************
-*** Default traversal ***
-************************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_DefaultTraversal_CompleteUnrolling
-{
-  enum {
-    outer = Index / Derived1::InnerSizeAtCompileTime,
-    inner = Index % Derived1::InnerSizeAtCompileTime
-  };
-
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.copyCoeffByOuterInner(outer, inner, src);
-    assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_DefaultTraversal_InnerUnrolling
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer)
-  {
-    dst.copyCoeffByOuterInner(outer, Index, src);
-    assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src, outer);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {}
-};
-
-/***********************
-*** Linear traversal ***
-***********************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_LinearTraversal_CompleteUnrolling
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.copyCoeff(Index, src);
-    assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_innervec_CompleteUnrolling
-{
-  enum {
-    outer = Index / Derived1::InnerSizeAtCompileTime,
-    inner = Index % Derived1::InnerSizeAtCompileTime,
-    JointAlignment = assign_traits<Derived1,Derived2>::JointAlignment
-  };
-
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.template copyPacketByOuterInner<Derived2, Aligned, JointAlignment>(outer, inner, src);
-    assign_innervec_CompleteUnrolling<Derived1, Derived2,
-      Index+packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_innervec_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_innervec_InnerUnrolling
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer)
-  {
-    dst.template copyPacketByOuterInner<Derived2, Aligned, Aligned>(outer, Index, src);
-    assign_innervec_InnerUnrolling<Derived1, Derived2,
-      Index+packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src, outer);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_innervec_InnerUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {}
-};
-
-/***************************************************************************
-* Part 3 : implementation of all cases
-***************************************************************************/
-
-template<typename Derived1, typename Derived2,
-         int Traversal = assign_traits<Derived1, Derived2>::Traversal,
-         int Unrolling = assign_traits<Derived1, Derived2>::Unrolling,
-         int Version = Specialized>
-struct assign_impl;
-
-/************************
-*** Default traversal ***
-************************/
-
-template<typename Derived1, typename Derived2, int Unrolling, int Version>
-struct assign_impl<Derived1, Derived2, InvalidTraversal, Unrolling, Version>
-{
-  static inline void run(Derived1 &, const Derived2 &) { }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, DefaultTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      for(Index inner = 0; inner < innerSize; ++inner)
-        dst.copyCoeffByOuterInner(outer, inner, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, DefaultTraversal, CompleteUnrolling, Version>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, DefaultTraversal, InnerUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
-        ::run(dst, src, outer);
-  }
-};
-
-/***********************
-*** Linear traversal ***
-***********************/
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, LinearTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index size = dst.size();
-    for(Index i = 0; i < size; ++i)
-      dst.copyCoeff(i, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, LinearTraversal, CompleteUnrolling, Version>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    const Index packetSize = packet_traits<typename Derived1::Scalar>::size;
-    for(Index outer = 0; outer < outerSize; ++outer)
-      for(Index inner = 0; inner < innerSize; inner+=packetSize)
-        dst.template copyPacketByOuterInner<Derived2, Aligned, Aligned>(outer, inner, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, CompleteUnrolling, Version>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, InnerUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      assign_innervec_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
-        ::run(dst, src, outer);
-  }
-};
-
-/***************************
-*** Linear vectorization ***
-***************************/
-
-template <bool IsAligned = false>
-struct unaligned_assign_impl
-{
-  template <typename Derived, typename OtherDerived>
-  static EIGEN_STRONG_INLINE void run(const Derived&, OtherDerived&, typename Derived::Index, typename Derived::Index) {}
-};
-
-template <>
-struct unaligned_assign_impl<false>
-{
-  // MSVC must not inline this functions. If it does, it fails to optimize the
-  // packet access path.
-#ifdef _MSC_VER
-  template <typename Derived, typename OtherDerived>
-  static EIGEN_DONT_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end)
-#else
-  template <typename Derived, typename OtherDerived>
-  static EIGEN_STRONG_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end)
-#endif
-  {
-    for (typename Derived::Index index = start; index < end; ++index)
-      dst.copyCoeff(index, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index size = dst.size();
-    typedef packet_traits<typename Derived1::Scalar> PacketTraits;
-    enum {
-      packetSize = PacketTraits::size,
-      dstAlignment = PacketTraits::AlignedOnScalar ? Aligned : int(assign_traits<Derived1,Derived2>::DstIsAligned) ,
-      srcAlignment = assign_traits<Derived1,Derived2>::JointAlignment
-    };
-    const Index alignedStart = assign_traits<Derived1,Derived2>::DstIsAligned ? 0
-                             : internal::first_aligned(&dst.coeffRef(0), size);
-    const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
-
-    unaligned_assign_impl<assign_traits<Derived1,Derived2>::DstIsAligned!=0>::run(src,dst,0,alignedStart);
-
-    for(Index index = alignedStart; index < alignedEnd; index += packetSize)
-    {
-      dst.template copyPacket<Derived2, dstAlignment, srcAlignment>(index, src);
-    }
-
-    unaligned_assign_impl<>::run(src,dst,alignedEnd,size);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, CompleteUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    enum { size = Derived1::SizeAtCompileTime,
-           packetSize = packet_traits<typename Derived1::Scalar>::size,
-           alignedSize = (size/packetSize)*packetSize };
-
-    assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, alignedSize>::run(dst, src);
-    assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, alignedSize, size>::run(dst, src);
-  }
-};
-
-/**************************
-*** Slice vectorization ***
-***************************/
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, SliceVectorizedTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    typedef typename Derived1::Scalar Scalar;
-    typedef packet_traits<Scalar> PacketTraits;
-    enum {
-      packetSize = PacketTraits::size,
-      alignable = PacketTraits::AlignedOnScalar,
-      dstIsAligned = assign_traits<Derived1,Derived2>::DstIsAligned,
-      dstAlignment = alignable ? Aligned : int(dstIsAligned),
-      srcAlignment = assign_traits<Derived1,Derived2>::JointAlignment
-    };
-    const Scalar *dst_ptr = &dst.coeffRef(0,0);
-    if((!bool(dstIsAligned)) && (size_t(dst_ptr) % sizeof(Scalar))>0)
-    {
-      // the pointer is not aligend-on scalar, so alignment is not possible
-      return assign_impl<Derived1,Derived2,DefaultTraversal,NoUnrolling>::run(dst, src);
-    }
-    const Index packetAlignedMask = packetSize - 1;
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    const Index alignedStep = alignable ? (packetSize - dst.outerStride() % packetSize) & packetAlignedMask : 0;
-    Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned(dst_ptr, innerSize);
-
-    for(Index outer = 0; outer < outerSize; ++outer)
-    {
-      const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
-      // do the non-vectorizable part of the assignment
-      for(Index inner = 0; inner<alignedStart ; ++inner)
-        dst.copyCoeffByOuterInner(outer, inner, src);
-
-      // do the vectorizable part of the assignment
-      for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
-        dst.template copyPacketByOuterInner<Derived2, dstAlignment, Unaligned>(outer, inner, src);
-
-      // do the non-vectorizable part of the assignment
-      for(Index inner = alignedEnd; inner<innerSize ; ++inner)
-        dst.copyCoeffByOuterInner(outer, inner, src);
-
-      alignedStart = std::min<Index>((alignedStart+alignedStep)%packetSize, innerSize);
-    }
-  }
-};
-
-} // end namespace internal
-
-/***************************************************************************
-* Part 4 : implementation of DenseBase methods
-***************************************************************************/
-
 template<typename Derived>
 template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived& DenseBase<Derived>
@@ -499,90 +27,62 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>
   EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived)
   EIGEN_STATIC_ASSERT(SameType,YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
 
-#ifdef EIGEN_DEBUG_ASSIGN
-  internal::assign_traits<Derived, OtherDerived>::debug();
-#endif
   eigen_assert(rows() == other.rows() && cols() == other.cols());
-  internal::assign_impl<Derived, OtherDerived, int(SameType) ? int(internal::assign_traits<Derived, OtherDerived>::Traversal)
-                                                       : int(InvalidTraversal)>::run(derived(),other.derived());
-#ifndef EIGEN_NO_DEBUG
-  checkTransposeAliasing(other.derived());
-#endif
+  internal::call_assignment_no_alias(derived(),other.derived());
+  
   return derived();
 }
 
-namespace internal {
-
-template<typename Derived, typename OtherDerived,
-         bool EvalBeforeAssigning = (int(internal::traits<OtherDerived>::Flags) & EvalBeforeAssigningBit) != 0,
-         bool NeedToTranspose = ((int(Derived::RowsAtCompileTime) == 1 && int(OtherDerived::ColsAtCompileTime) == 1)
-                              |   // FIXME | instead of || to please GCC 4.4.0 stupid warning "suggest parentheses around &&".
-                                  // revert to || as soon as not needed anymore.
-                                  (int(Derived::ColsAtCompileTime) == 1 && int(OtherDerived::RowsAtCompileTime) == 1))
-                              && int(Derived::SizeAtCompileTime) != 1>
-struct assign_selector;
-
-template<typename Derived, typename OtherDerived>
-struct assign_selector<Derived,OtherDerived,false,false> {
-  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); }
-  template<typename ActualDerived, typename ActualOtherDerived>
-  static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { other.evalTo(dst); return dst; }
-};
-template<typename Derived, typename OtherDerived>
-struct assign_selector<Derived,OtherDerived,true,false> {
-  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.eval()); }
-};
-template<typename Derived, typename OtherDerived>
-struct assign_selector<Derived,OtherDerived,false,true> {
-  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose()); }
-  template<typename ActualDerived, typename ActualOtherDerived>
-  static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { Transpose<ActualDerived> dstTrans(dst); other.evalTo(dstTrans); return dst; }
-};
-template<typename Derived, typename OtherDerived>
-struct assign_selector<Derived,OtherDerived,true,true> {
-  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose().eval()); }
-};
-
-} // end namespace internal
-
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
 {
-  return internal::assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase& other)
 {
-  return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const MatrixBase& other)
 {
-  return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
 template <typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
 {
-  return internal::assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
 template <typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const EigenBase<OtherDerived>& other)
 {
-  return internal::assign_selector<Derived,OtherDerived,false>::evalTo(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
 {
-  return internal::assign_selector<Derived,OtherDerived,false>::evalTo(derived(), other.derived());
+  other.derived().evalTo(derived());
+  return derived();
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/AssignEvaluator.h b/extern/Eigen3/Eigen/src/Core/AssignEvaluator.h
new file mode 100644
index 00000000000..dbe435d86b3
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/AssignEvaluator.h
@@ -0,0 +1,935 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ASSIGN_EVALUATOR_H
+#define EIGEN_ASSIGN_EVALUATOR_H
+
+namespace Eigen {
+
+// This implementation is based on Assign.h
+
+namespace internal {
+  
+/***************************************************************************
+* Part 1 : the logic deciding a strategy for traversal and unrolling       *
+***************************************************************************/
+
+// copy_using_evaluator_traits is based on assign_traits
+
+template <typename DstEvaluator, typename SrcEvaluator, typename AssignFunc>
+struct copy_using_evaluator_traits
+{
+  typedef typename DstEvaluator::XprType Dst;
+  typedef typename Dst::Scalar DstScalar;
+  
+  enum {
+    DstFlags = DstEvaluator::Flags,
+    SrcFlags = SrcEvaluator::Flags
+  };
+  
+public:
+  enum {
+    DstAlignment = DstEvaluator::Alignment,
+    SrcAlignment = SrcEvaluator::Alignment,
+    DstHasDirectAccess = (DstFlags & DirectAccessBit) == DirectAccessBit,
+    JointAlignment = EIGEN_PLAIN_ENUM_MIN(DstAlignment,SrcAlignment)
+  };
+
+private:
+  enum {
+    InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
+              : int(DstFlags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
+              : int(Dst::RowsAtCompileTime),
+    InnerMaxSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime)
+              : int(DstFlags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime)
+              : int(Dst::MaxRowsAtCompileTime),
+    OuterStride = int(outer_stride_at_compile_time<Dst>::ret),
+    MaxSizeAtCompileTime = Dst::SizeAtCompileTime
+  };
+
+  // TODO distinguish between linear traversal and inner-traversals
+  typedef typename find_best_packet<DstScalar,Dst::SizeAtCompileTime>::type LinearPacketType;
+  typedef typename find_best_packet<DstScalar,InnerSize>::type InnerPacketType;
+
+  enum {
+    LinearPacketSize = unpacket_traits<LinearPacketType>::size,
+    InnerPacketSize = unpacket_traits<InnerPacketType>::size
+  };
+
+public:
+  enum {
+    LinearRequiredAlignment = unpacket_traits<LinearPacketType>::alignment,
+    InnerRequiredAlignment = unpacket_traits<InnerPacketType>::alignment
+  };
+
+private:
+  enum {
+    DstIsRowMajor = DstFlags&RowMajorBit,
+    SrcIsRowMajor = SrcFlags&RowMajorBit,
+    StorageOrdersAgree = (int(DstIsRowMajor) == int(SrcIsRowMajor)),
+    MightVectorize = bool(StorageOrdersAgree)
+                  && (int(DstFlags) & int(SrcFlags) & ActualPacketAccessBit)
+                  && bool(functor_traits<AssignFunc>::PacketAccess),
+    MayInnerVectorize  = MightVectorize
+                       && int(InnerSize)!=Dynamic && int(InnerSize)%int(InnerPacketSize)==0
+                       && int(OuterStride)!=Dynamic && int(OuterStride)%int(InnerPacketSize)==0
+                       && (EIGEN_UNALIGNED_VECTORIZE  || int(JointAlignment)>=int(InnerRequiredAlignment)),
+    MayLinearize = bool(StorageOrdersAgree) && (int(DstFlags) & int(SrcFlags) & LinearAccessBit),
+    MayLinearVectorize = bool(MightVectorize) && bool(MayLinearize) && bool(DstHasDirectAccess)
+                       && (EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)) || MaxSizeAtCompileTime == Dynamic),
+      /* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
+         so it's only good for large enough sizes. */
+    MaySliceVectorize  = bool(MightVectorize) && bool(DstHasDirectAccess)
+                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=(EIGEN_UNALIGNED_VECTORIZE?InnerPacketSize:(3*InnerPacketSize)))
+      /* slice vectorization can be slow, so we only want it if the slices are big, which is
+         indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
+         in a fixed-size matrix
+         However, with EIGEN_UNALIGNED_VECTORIZE and unrolling, slice vectorization is still worth it */
+  };
+
+public:
+  enum {
+    Traversal = int(MayLinearVectorize) && (LinearPacketSize>InnerPacketSize) ? int(LinearVectorizedTraversal)
+              : int(MayInnerVectorize)   ? int(InnerVectorizedTraversal)
+              : int(MayLinearVectorize)  ? int(LinearVectorizedTraversal)
+              : int(MaySliceVectorize)   ? int(SliceVectorizedTraversal)
+              : int(MayLinearize)        ? int(LinearTraversal)
+                                         : int(DefaultTraversal),
+    Vectorized = int(Traversal) == InnerVectorizedTraversal
+              || int(Traversal) == LinearVectorizedTraversal
+              || int(Traversal) == SliceVectorizedTraversal
+  };
+
+  typedef typename conditional<int(Traversal)==LinearVectorizedTraversal, LinearPacketType, InnerPacketType>::type PacketType;
+
+private:
+  enum {
+    ActualPacketSize    = int(Traversal)==LinearVectorizedTraversal ? LinearPacketSize
+                        : Vectorized ? InnerPacketSize
+                        : 1,
+    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * ActualPacketSize,
+    MayUnrollCompletely = int(Dst::SizeAtCompileTime) != Dynamic
+                       && int(Dst::SizeAtCompileTime) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit),
+    MayUnrollInner      = int(InnerSize) != Dynamic
+                       && int(InnerSize) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit)
+  };
+
+public:
+  enum {
+    Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
+                ? (
+                    int(MayUnrollCompletely) ? int(CompleteUnrolling)
+                  : int(MayUnrollInner)      ? int(InnerUnrolling)
+                                             : int(NoUnrolling)
+                  )
+              : int(Traversal) == int(LinearVectorizedTraversal)
+                ? ( bool(MayUnrollCompletely) && ( EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)))
+                          ? int(CompleteUnrolling)
+                          : int(NoUnrolling) )
+              : int(Traversal) == int(LinearTraversal)
+                ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) 
+                                              : int(NoUnrolling) )
+#if EIGEN_UNALIGNED_VECTORIZE
+              : int(Traversal) == int(SliceVectorizedTraversal)
+                ? ( bool(MayUnrollInner) ? int(InnerUnrolling)
+                                         : int(NoUnrolling) )
+#endif
+              : int(NoUnrolling)
+  };
+
+#ifdef EIGEN_DEBUG_ASSIGN
+  static void debug()
+  {
+    std::cerr << "DstXpr: " << typeid(typename DstEvaluator::XprType).name() << std::endl;
+    std::cerr << "SrcXpr: " << typeid(typename SrcEvaluator::XprType).name() << std::endl;
+    std::cerr.setf(std::ios::hex, std::ios::basefield);
+    std::cerr << "DstFlags" << " = " << DstFlags << " (" << demangle_flags(DstFlags) << " )" << std::endl;
+    std::cerr << "SrcFlags" << " = " << SrcFlags << " (" << demangle_flags(SrcFlags) << " )" << std::endl;
+    std::cerr.unsetf(std::ios::hex);
+    EIGEN_DEBUG_VAR(DstAlignment)
+    EIGEN_DEBUG_VAR(SrcAlignment)
+    EIGEN_DEBUG_VAR(LinearRequiredAlignment)
+    EIGEN_DEBUG_VAR(InnerRequiredAlignment)
+    EIGEN_DEBUG_VAR(JointAlignment)
+    EIGEN_DEBUG_VAR(InnerSize)
+    EIGEN_DEBUG_VAR(InnerMaxSize)
+    EIGEN_DEBUG_VAR(LinearPacketSize)
+    EIGEN_DEBUG_VAR(InnerPacketSize)
+    EIGEN_DEBUG_VAR(ActualPacketSize)
+    EIGEN_DEBUG_VAR(StorageOrdersAgree)
+    EIGEN_DEBUG_VAR(MightVectorize)
+    EIGEN_DEBUG_VAR(MayLinearize)
+    EIGEN_DEBUG_VAR(MayInnerVectorize)
+    EIGEN_DEBUG_VAR(MayLinearVectorize)
+    EIGEN_DEBUG_VAR(MaySliceVectorize)
+    std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
+    EIGEN_DEBUG_VAR(SrcEvaluator::CoeffReadCost)
+    EIGEN_DEBUG_VAR(UnrollingLimit)
+    EIGEN_DEBUG_VAR(MayUnrollCompletely)
+    EIGEN_DEBUG_VAR(MayUnrollInner)
+    std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl;
+    std::cerr << std::endl;
+  }
+#endif
+};
+
+/***************************************************************************
+* Part 2 : meta-unrollers
+***************************************************************************/
+
+/************************
+*** Default traversal ***
+************************/
+
+template<typename Kernel, int Index, int Stop>
+struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling
+{
+  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
+  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
+  typedef typename DstEvaluatorType::XprType DstXprType;
+  
+  enum {
+    outer = Index / DstXprType::InnerSizeAtCompileTime,
+    inner = Index % DstXprType::InnerSizeAtCompileTime
+  };
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    kernel.assignCoeffByOuterInner(outer, inner);
+    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
+};
+
+template<typename Kernel, int Index_, int Stop>
+struct copy_using_evaluator_DefaultTraversal_InnerUnrolling
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
+  {
+    kernel.assignCoeffByOuterInner(outer, Index_);
+    copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Index_+1, Stop>::run(kernel, outer);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index) { }
+};
+
+/***********************
+*** Linear traversal ***
+***********************/
+
+template<typename Kernel, int Index, int Stop>
+struct copy_using_evaluator_LinearTraversal_CompleteUnrolling
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel& kernel)
+  {
+    kernel.assignCoeff(Index);
+    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
+};
+
+/**************************
+*** Inner vectorization ***
+**************************/
+
+template<typename Kernel, int Index, int Stop>
+struct copy_using_evaluator_innervec_CompleteUnrolling
+{
+  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
+  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
+  typedef typename DstEvaluatorType::XprType DstXprType;
+  typedef typename Kernel::PacketType PacketType;
+  
+  enum {
+    outer = Index / DstXprType::InnerSizeAtCompileTime,
+    inner = Index % DstXprType::InnerSizeAtCompileTime,
+    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
+    DstAlignment = Kernel::AssignmentTraits::DstAlignment
+  };
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
+    enum { NextIndex = Index + unpacket_traits<PacketType>::size };
+    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, NextIndex, Stop>::run(kernel);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_innervec_CompleteUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
+};
+
+template<typename Kernel, int Index_, int Stop, int SrcAlignment, int DstAlignment>
+struct copy_using_evaluator_innervec_InnerUnrolling
+{
+  typedef typename Kernel::PacketType PacketType;
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
+  {
+    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, Index_);
+    enum { NextIndex = Index_ + unpacket_traits<PacketType>::size };
+    copy_using_evaluator_innervec_InnerUnrolling<Kernel, NextIndex, Stop, SrcAlignment, DstAlignment>::run(kernel, outer);
+  }
+};
+
+template<typename Kernel, int Stop, int SrcAlignment, int DstAlignment>
+struct copy_using_evaluator_innervec_InnerUnrolling<Kernel, Stop, Stop, SrcAlignment, DstAlignment>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &, Index) { }
+};
+
+/***************************************************************************
+* Part 3 : implementation of all cases
+***************************************************************************/
+
+// dense_assignment_loop is based on assign_impl
+
+template<typename Kernel,
+         int Traversal = Kernel::AssignmentTraits::Traversal,
+         int Unrolling = Kernel::AssignmentTraits::Unrolling>
+struct dense_assignment_loop;
+
+/************************
+*** Default traversal ***
+************************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, DefaultTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel &kernel)
+  {
+    for(Index outer = 0; outer < kernel.outerSize(); ++outer) {
+      for(Index inner = 0; inner < kernel.innerSize(); ++inner) {
+        kernel.assignCoeffByOuterInner(outer, inner);
+      }
+    }
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, DefaultTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, DefaultTraversal, InnerUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+
+    const Index outerSize = kernel.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer)
+      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime>::run(kernel, outer);
+  }
+};
+
+/***************************
+*** Linear vectorization ***
+***************************/
+
+
+// The goal of unaligned_dense_assignment_loop is simply to factorize the handling
+// of the non vectorizable beginning and ending parts
+
+template <bool IsAligned = false>
+struct unaligned_dense_assignment_loop
+{
+  // if IsAligned = true, then do nothing
+  template <typename Kernel>
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index, Index) {}
+};
+
+template <>
+struct unaligned_dense_assignment_loop<false>
+{
+  // MSVC must not inline this functions. If it does, it fails to optimize the
+  // packet access path.
+  // FIXME check which version exhibits this issue
+#if EIGEN_COMP_MSVC
+  template <typename Kernel>
+  static EIGEN_DONT_INLINE void run(Kernel &kernel,
+                                    Index start,
+                                    Index end)
+#else
+  template <typename Kernel>
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel,
+                                      Index start,
+                                      Index end)
+#endif
+  {
+    for (Index index = start; index < end; ++index)
+      kernel.assignCoeff(index);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    const Index size = kernel.size();
+    typedef typename Kernel::Scalar Scalar;
+    typedef typename Kernel::PacketType PacketType;
+    enum {
+      requestedAlignment = Kernel::AssignmentTraits::LinearRequiredAlignment,
+      packetSize = unpacket_traits<PacketType>::size,
+      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
+      dstAlignment = packet_traits<Scalar>::AlignedOnScalar ? int(requestedAlignment)
+                                                            : int(Kernel::AssignmentTraits::DstAlignment),
+      srcAlignment = Kernel::AssignmentTraits::JointAlignment
+    };
+    const Index alignedStart = dstIsAligned ? 0 : internal::first_aligned<requestedAlignment>(kernel.dstDataPtr(), size);
+    const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
+
+    unaligned_dense_assignment_loop<dstIsAligned!=0>::run(kernel, 0, alignedStart);
+
+    for(Index index = alignedStart; index < alignedEnd; index += packetSize)
+      kernel.template assignPacket<dstAlignment, srcAlignment, PacketType>(index);
+
+    unaligned_dense_assignment_loop<>::run(kernel, alignedEnd, size);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    typedef typename Kernel::PacketType PacketType;
+    
+    enum { size = DstXprType::SizeAtCompileTime,
+           packetSize =unpacket_traits<PacketType>::size,
+           alignedSize = (size/packetSize)*packetSize };
+
+    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, alignedSize>::run(kernel);
+    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, alignedSize, size>::run(kernel);
+  }
+};
+
+/**************************
+*** Inner vectorization ***
+**************************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, NoUnrolling>
+{
+  typedef typename Kernel::PacketType PacketType;
+  enum {
+    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
+    DstAlignment = Kernel::AssignmentTraits::DstAlignment
+  };
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    const Index innerSize = kernel.innerSize();
+    const Index outerSize = kernel.outerSize();
+    const Index packetSize = unpacket_traits<PacketType>::size;
+    for(Index outer = 0; outer < outerSize; ++outer)
+      for(Index inner = 0; inner < innerSize; inner+=packetSize)
+        kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, InnerUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    typedef typename Kernel::AssignmentTraits Traits;
+    const Index outerSize = kernel.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer)
+      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime,
+                                                   Traits::SrcAlignment, Traits::DstAlignment>::run(kernel, outer);
+  }
+};
+
+/***********************
+*** Linear traversal ***
+***********************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    const Index size = kernel.size();
+    for(Index i = 0; i < size; ++i)
+      kernel.assignCoeff(i);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
+  }
+};
+
+/**************************
+*** Slice vectorization ***
+***************************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::Scalar Scalar;
+    typedef typename Kernel::PacketType PacketType;
+    enum {
+      packetSize = unpacket_traits<PacketType>::size,
+      requestedAlignment = int(Kernel::AssignmentTraits::InnerRequiredAlignment),
+      alignable = packet_traits<Scalar>::AlignedOnScalar || int(Kernel::AssignmentTraits::DstAlignment)>=sizeof(Scalar),
+      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
+      dstAlignment = alignable ? int(requestedAlignment)
+                               : int(Kernel::AssignmentTraits::DstAlignment)
+    };
+    const Scalar *dst_ptr = kernel.dstDataPtr();
+    if((!bool(dstIsAligned)) && (UIntPtr(dst_ptr) % sizeof(Scalar))>0)
+    {
+      // the pointer is not aligend-on scalar, so alignment is not possible
+      return dense_assignment_loop<Kernel,DefaultTraversal,NoUnrolling>::run(kernel);
+    }
+    const Index packetAlignedMask = packetSize - 1;
+    const Index innerSize = kernel.innerSize();
+    const Index outerSize = kernel.outerSize();
+    const Index alignedStep = alignable ? (packetSize - kernel.outerStride() % packetSize) & packetAlignedMask : 0;
+    Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned<requestedAlignment>(dst_ptr, innerSize);
+
+    for(Index outer = 0; outer < outerSize; ++outer)
+    {
+      const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
+      // do the non-vectorizable part of the assignment
+      for(Index inner = 0; inner<alignedStart ; ++inner)
+        kernel.assignCoeffByOuterInner(outer, inner);
+
+      // do the vectorizable part of the assignment
+      for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
+        kernel.template assignPacketByOuterInner<dstAlignment, Unaligned, PacketType>(outer, inner);
+
+      // do the non-vectorizable part of the assignment
+      for(Index inner = alignedEnd; inner<innerSize ; ++inner)
+        kernel.assignCoeffByOuterInner(outer, inner);
+
+      alignedStart = numext::mini((alignedStart+alignedStep)%packetSize, innerSize);
+    }
+  }
+};
+
+#if EIGEN_UNALIGNED_VECTORIZE
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, InnerUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    typedef typename Kernel::PacketType PacketType;
+
+    enum { size = DstXprType::InnerSizeAtCompileTime,
+           packetSize =unpacket_traits<PacketType>::size,
+           vectorizableSize = (size/packetSize)*packetSize };
+
+    for(Index outer = 0; outer < kernel.outerSize(); ++outer)
+    {
+      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, vectorizableSize, 0, 0>::run(kernel, outer);
+      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, vectorizableSize, size>::run(kernel, outer);
+    }
+  }
+};
+#endif
+
+
+/***************************************************************************
+* Part 4 : Generic dense assignment kernel
+***************************************************************************/
+
+// This class generalize the assignment of a coefficient (or packet) from one dense evaluator
+// to another dense writable evaluator.
+// It is parametrized by the two evaluators, and the actual assignment functor.
+// This abstraction level permits to keep the evaluation loops as simple and as generic as possible.
+// One can customize the assignment using this generic dense_assignment_kernel with different
+// functors, or by completely overloading it, by-passing a functor.
+template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version = Specialized>
+class generic_dense_assignment_kernel
+{
+protected:
+  typedef typename DstEvaluatorTypeT::XprType DstXprType;
+  typedef typename SrcEvaluatorTypeT::XprType SrcXprType;
+public:
+  
+  typedef DstEvaluatorTypeT DstEvaluatorType;
+  typedef SrcEvaluatorTypeT SrcEvaluatorType;
+  typedef typename DstEvaluatorType::Scalar Scalar;
+  typedef copy_using_evaluator_traits<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor> AssignmentTraits;
+  typedef typename AssignmentTraits::PacketType PacketType;
+  
+  
+  EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
+    : m_dst(dst), m_src(src), m_functor(func), m_dstExpr(dstExpr)
+  {
+    #ifdef EIGEN_DEBUG_ASSIGN
+    AssignmentTraits::debug();
+    #endif
+  }
+  
+  EIGEN_DEVICE_FUNC Index size() const        { return m_dstExpr.size(); }
+  EIGEN_DEVICE_FUNC Index innerSize() const   { return m_dstExpr.innerSize(); }
+  EIGEN_DEVICE_FUNC Index outerSize() const   { return m_dstExpr.outerSize(); }
+  EIGEN_DEVICE_FUNC Index rows() const        { return m_dstExpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const        { return m_dstExpr.cols(); }
+  EIGEN_DEVICE_FUNC Index outerStride() const { return m_dstExpr.outerStride(); }
+  
+  EIGEN_DEVICE_FUNC DstEvaluatorType& dstEvaluator() { return m_dst; }
+  EIGEN_DEVICE_FUNC const SrcEvaluatorType& srcEvaluator() const { return m_src; }
+  
+  /// Assign src(row,col) to dst(row,col) through the assignment functor.
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index row, Index col)
+  {
+    m_functor.assignCoeff(m_dst.coeffRef(row,col), m_src.coeff(row,col));
+  }
+  
+  /// \sa assignCoeff(Index,Index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index index)
+  {
+    m_functor.assignCoeff(m_dst.coeffRef(index), m_src.coeff(index));
+  }
+  
+  /// \sa assignCoeff(Index,Index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeffByOuterInner(Index outer, Index inner)
+  {
+    Index row = rowIndexByOuterInner(outer, inner); 
+    Index col = colIndexByOuterInner(outer, inner); 
+    assignCoeff(row, col);
+  }
+  
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index row, Index col)
+  {
+    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(row,col), m_src.template packet<LoadMode,PacketType>(row,col));
+  }
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index index)
+  {
+    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(index), m_src.template packet<LoadMode,PacketType>(index));
+  }
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacketByOuterInner(Index outer, Index inner)
+  {
+    Index row = rowIndexByOuterInner(outer, inner); 
+    Index col = colIndexByOuterInner(outer, inner);
+    assignPacket<StoreMode,LoadMode,PacketType>(row, col);
+  }
+  
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner)
+  {
+    typedef typename DstEvaluatorType::ExpressionTraits Traits;
+    return int(Traits::RowsAtCompileTime) == 1 ? 0
+      : int(Traits::ColsAtCompileTime) == 1 ? inner
+      : int(DstEvaluatorType::Flags)&RowMajorBit ? outer
+      : inner;
+  }
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner)
+  {
+    typedef typename DstEvaluatorType::ExpressionTraits Traits;
+    return int(Traits::ColsAtCompileTime) == 1 ? 0
+      : int(Traits::RowsAtCompileTime) == 1 ? inner
+      : int(DstEvaluatorType::Flags)&RowMajorBit ? inner
+      : outer;
+  }
+
+  EIGEN_DEVICE_FUNC const Scalar* dstDataPtr() const
+  {
+    return m_dstExpr.data();
+  }
+  
+protected:
+  DstEvaluatorType& m_dst;
+  const SrcEvaluatorType& m_src;
+  const Functor &m_functor;
+  // TODO find a way to avoid the needs of the original expression
+  DstXprType& m_dstExpr;
+};
+
+/***************************************************************************
+* Part 5 : Entry point for dense rectangular assignment
+***************************************************************************/
+
+template<typename DstXprType,typename SrcXprType, typename Functor>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const Functor &/*func*/)
+{
+  EIGEN_ONLY_USED_FOR_DEBUG(dst);
+  EIGEN_ONLY_USED_FOR_DEBUG(src);
+  eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+}
+
+template<typename DstXprType,typename SrcXprType, typename T1, typename T2>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const internal::assign_op<T1,T2> &/*func*/)
+{
+  Index dstRows = src.rows();
+  Index dstCols = src.cols();
+  if(((dst.rows()!=dstRows) || (dst.cols()!=dstCols)))
+    dst.resize(dstRows, dstCols);
+  eigen_assert(dst.rows() == dstRows && dst.cols() == dstCols);
+}
+
+template<typename DstXprType, typename SrcXprType, typename Functor>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func)
+{
+  typedef evaluator<DstXprType> DstEvaluatorType;
+  typedef evaluator<SrcXprType> SrcEvaluatorType;
+
+  SrcEvaluatorType srcEvaluator(src);
+
+  // NOTE To properly handle A = (A*A.transpose())/s with A rectangular,
+  // we need to resize the destination after the source evaluator has been created.
+  resize_if_allowed(dst, src, func);
+
+  DstEvaluatorType dstEvaluator(dst);
+    
+  typedef generic_dense_assignment_kernel<DstEvaluatorType,SrcEvaluatorType,Functor> Kernel;
+  Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());
+
+  dense_assignment_loop<Kernel>::run(kernel);
+}
+
+template<typename DstXprType, typename SrcXprType>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src)
+{
+  call_dense_assignment_loop(dst, src, internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>());
+}
+
+/***************************************************************************
+* Part 6 : Generic assignment
+***************************************************************************/
+
+// Based on the respective shapes of the destination and source,
+// the class AssignmentKind determine the kind of assignment mechanism.
+// AssignmentKind must define a Kind typedef.
+template<typename DstShape, typename SrcShape> struct AssignmentKind;
+
+// Assignement kind defined in this file:
+struct Dense2Dense {};
+struct EigenBase2EigenBase {};
+
+template<typename,typename> struct AssignmentKind { typedef EigenBase2EigenBase Kind; };
+template<> struct AssignmentKind<DenseShape,DenseShape> { typedef Dense2Dense Kind; };
+    
+// This is the main assignment class
+template< typename DstXprType, typename SrcXprType, typename Functor,
+          typename Kind = typename AssignmentKind< typename evaluator_traits<DstXprType>::Shape , typename evaluator_traits<SrcXprType>::Shape >::Kind,
+          typename EnableIf = void>
+struct Assignment;
+
+
+// The only purpose of this call_assignment() function is to deal with noalias() / "assume-aliasing" and automatic transposition.
+// Indeed, I (Gael) think that this concept of "assume-aliasing" was a mistake, and it makes thing quite complicated.
+// So this intermediate function removes everything related to "assume-aliasing" such that Assignment
+// does not has to bother about these annoying details.
+
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src)
+{
+  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(const Dst& dst, const Src& src)
+{
+  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+                     
+// Deal with "assume-aliasing"
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if< evaluator_assume_aliasing<Src>::value, void*>::type = 0)
+{
+  typename plain_matrix_type<Src>::type tmp(src);
+  call_assignment_no_alias(dst, tmp, func);
+}
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<!evaluator_assume_aliasing<Src>::value, void*>::type = 0)
+{
+  call_assignment_no_alias(dst, src, func);
+}
+
+// by-pass "assume-aliasing"
+// When there is no aliasing, we require that 'dst' has been properly resized
+template<typename Dst, template <typename> class StorageBase, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(NoAlias<Dst,StorageBase>& dst, const Src& src, const Func& func)
+{
+  call_assignment_no_alias(dst.expression(), src, func);
+}
+
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
+{
+  enum {
+    NeedToTranspose = (    (int(Dst::RowsAtCompileTime) == 1 && int(Src::ColsAtCompileTime) == 1)
+                        || (int(Dst::ColsAtCompileTime) == 1 && int(Src::RowsAtCompileTime) == 1)
+                      ) && int(Dst::SizeAtCompileTime) != 1
+  };
+
+  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst>::type ActualDstTypeCleaned;
+  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst&>::type ActualDstType;
+  ActualDstType actualDst(dst);
+  
+  // TODO check whether this is the right place to perform these checks:
+  EIGEN_STATIC_ASSERT_LVALUE(Dst)
+  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(ActualDstTypeCleaned,Src)
+  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename ActualDstTypeCleaned::Scalar,typename Src::Scalar);
+  
+  Assignment<ActualDstTypeCleaned,Src,Func>::run(actualDst, src, func);
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias(Dst& dst, const Src& src)
+{
+  call_assignment_no_alias(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func)
+{
+  // TODO check whether this is the right place to perform these checks:
+  EIGEN_STATIC_ASSERT_LVALUE(Dst)
+  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Dst,Src)
+  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar);
+
+  Assignment<Dst,Src,Func>::run(dst, src, func);
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src)
+{
+  call_assignment_no_alias_no_transpose(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+
+// forward declaration
+template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, const Src &src);
+
+// Generic Dense to Dense assignment
+// Note that the last template argument "Weak" is needed to make it possible to perform
+// both partial specialization+SFINAE without ambiguous specialization
+template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
+struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Weak>
+{
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+#ifndef EIGEN_NO_DEBUG
+    internal::check_for_aliasing(dst, src);
+#endif
+    
+    call_dense_assignment_loop(dst, src, func);
+  }
+};
+
+// Generic assignment through evalTo.
+// TODO: not sure we have to keep that one, but it helps porting current code to new evaluator mechanism.
+// Note that the last template argument "Weak" is needed to make it possible to perform
+// both partial specialization+SFINAE without ambiguous specialization
+template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
+struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Weak>
+{
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.evalTo(dst);
+  }
+
+  // NOTE The following two functions are templated to avoid their instanciation if not needed
+  //      This is needed because some expressions supports evalTo only and/or have 'void' as scalar type.
+  template<typename SrcScalarType>
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.addTo(dst);
+  }
+
+  template<typename SrcScalarType>
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.subTo(dst);
+  }
+};
+
+} // namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_ASSIGN_EVALUATOR_H
diff --git a/extern/Eigen3/Eigen/src/Core/Assign_MKL.h b/extern/Eigen3/Eigen/src/Core/Assign_MKL.h
index 7772951b915..6866095bf8a 100644
--- a/extern/Eigen3/Eigen/src/Core/Assign_MKL.h
+++ b/extern/Eigen3/Eigen/src/Core/Assign_MKL.h
@@ -1,6 +1,7 @@
 /*
  Copyright (c) 2011, Intel Corporation. All rights reserved.
-
+ Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+ 
  Redistribution and use in source and binary forms, with or without modification,
  are permitted provided that the following conditions are met:
 
@@ -37,17 +38,13 @@ namespace Eigen {
 
 namespace internal {
 
-template<typename Op> struct vml_call
-{ enum { IsSupported = 0 }; };
-
-template<typename Dst, typename Src, typename UnaryOp>
+template<typename Dst, typename Src>
 class vml_assign_traits
 {
   private:
     enum {
       DstHasDirectAccess = Dst::Flags & DirectAccessBit,
       SrcHasDirectAccess = Src::Flags & DirectAccessBit,
-
       StorageOrdersAgree = (int(Dst::IsRowMajor) == int(Src::IsRowMajor)),
       InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
                 : int(Dst::Flags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
@@ -57,165 +54,122 @@ class vml_assign_traits
                     : int(Dst::MaxRowsAtCompileTime),
       MaxSizeAtCompileTime = Dst::SizeAtCompileTime,
 
-      MightEnableVml =  vml_call<UnaryOp>::IsSupported && StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess
-                     && Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1,
+      MightEnableVml = StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess && Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1,
       MightLinearize = MightEnableVml && (int(Dst::Flags) & int(Src::Flags) & LinearAccessBit),
       VmlSize = MightLinearize ? MaxSizeAtCompileTime : InnerMaxSize,
-      LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD,
-      MayEnableVml = MightEnableVml && LargeEnough,
-      MayLinearize = MayEnableVml && MightLinearize
+      LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD
     };
   public:
     enum {
-      Traversal = MayLinearize ? LinearVectorizedTraversal
-                : MayEnableVml ? InnerVectorizedTraversal
-                : DefaultTraversal
+      EnableVml = MightEnableVml && LargeEnough,
+      Traversal = MightLinearize ? LinearTraversal : DefaultTraversal
     };
 };
 
-template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling,
-         int VmlTraversal = vml_assign_traits<Derived1, Derived2, UnaryOp>::Traversal >
-struct vml_assign_impl
-  : assign_impl<Derived1, Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>
-{
-};
-
-template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling>
-struct vml_assign_impl<Derived1, Derived2, UnaryOp, Traversal, Unrolling, InnerVectorizedTraversal>
-{
-  typedef typename Derived1::Scalar Scalar;
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1& dst, const CwiseUnaryOp<UnaryOp, Derived2>& src)
-  {
-    // in case we want to (or have to) skip VML at runtime we can call:
-    // assign_impl<Derived1,Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>::run(dst,src);
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer) {
-      const Scalar *src_ptr = src.IsRowMajor ?  &(src.nestedExpression().coeffRef(outer,0)) :
-                                                &(src.nestedExpression().coeffRef(0, outer));
-      Scalar *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));
-      vml_call<UnaryOp>::run(src.functor(), innerSize, src_ptr, dst_ptr );
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling>
-struct vml_assign_impl<Derived1, Derived2, UnaryOp, Traversal, Unrolling, LinearVectorizedTraversal>
-{
-  static inline void run(Derived1& dst, const CwiseUnaryOp<UnaryOp, Derived2>& src)
-  {
-    // in case we want to (or have to) skip VML at runtime we can call:
-    // assign_impl<Derived1,Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>::run(dst,src);
-    vml_call<UnaryOp>::run(src.functor(), dst.size(), src.nestedExpression().data(), dst.data() );
-  }
-};
-
-// Macroses
-
-#define EIGEN_MKL_VML_SPECIALIZE_ASSIGN(TRAVERSAL,UNROLLING) \
-  template<typename Derived1, typename Derived2, typename UnaryOp> \
-  struct assign_impl<Derived1, Eigen::CwiseUnaryOp<UnaryOp, Derived2>, TRAVERSAL, UNROLLING, Specialized>  {  \
-    static inline void run(Derived1 &dst, const Eigen::CwiseUnaryOp<UnaryOp, Derived2> &src) { \
-      vml_assign_impl<Derived1,Derived2,UnaryOp,TRAVERSAL,UNROLLING>::run(dst, src); \
-    } \
-  };
-
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,NoUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,CompleteUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,InnerUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,NoUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,CompleteUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,NoUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,CompleteUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,InnerUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,CompleteUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,NoUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(SliceVectorizedTraversal,NoUnrolling)
-
-
+#define EIGEN_PP_EXPAND(ARG) ARG
 #if !defined (EIGEN_FAST_MATH) || (EIGEN_FAST_MATH != 1)
-#define  EIGEN_MKL_VML_MODE VML_HA
+#define EIGEN_VMLMODE_EXPAND_LA , VML_HA
 #else
-#define  EIGEN_MKL_VML_MODE VML_LA
+#define EIGEN_VMLMODE_EXPAND_LA , VML_LA
 #endif
 
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE)     \
-  template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > {               \
-    enum { IsSupported = 1 };                                                    \
-    static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& /*func*/,        \
-                            int size, const EIGENTYPE* src, EIGENTYPE* dst) {    \
-      VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst);                           \
-    }                                                                            \
+#define EIGEN_VMLMODE_EXPAND__ 
+
+#define EIGEN_VMLMODE_PREFIX_LA vm
+#define EIGEN_VMLMODE_PREFIX__  v
+#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_,VMLMODE)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
+  template< typename DstXprType, typename SrcXprNested>                                                                         \
+  struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE,EIGENTYPE>,   \
+                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {              \
+    typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType;                                            \
+    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) {                       \
+      resize_if_allowed(dst, src, func);                                                                                        \
+      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                       \
+      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) {                                              \
+        VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(),                                                        \
+              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                           \
+      } else {                                                                                                                  \
+        const Index outerSize = dst.outerSize();                                                                                \
+        for(Index outer = 0; outer < outerSize; ++outer) {                                                                      \
+          const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.nestedExpression().coeffRef(outer,0)) :                             \
+                                                      &(src.nestedExpression().coeffRef(0, outer));                             \
+          EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                           \
+          VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr,                                                                      \
+                (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                             \
+        }                                                                                                                       \
+      }                                                                                                                         \
+    }                                                                                                                           \
+  };                                                                                                                            \
+
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                         \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),s##VMLOP), float, float, VMLMODE)           \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),d##VMLOP), double, double, VMLMODE)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)                                                         \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),c##VMLOP), scomplex, MKL_Complex8, VMLMODE) \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),z##VMLOP), dcomplex, MKL_Complex16, VMLMODE)
+  
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP, VMLMODE)                                                              \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                               \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)
+
+  
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sin,   Sin,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(asin,  Asin,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sinh,  Sinh,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cos,   Cos,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(acos,  Acos,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cosh,  Cosh,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tan,   Tan,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(atan,  Atan,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tanh,  Tanh,  LA)
+// EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs,   Abs,    _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(exp,   Exp,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log,   Ln,    LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log10, Log10, LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sqrt,  Sqrt,  _)
+
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr,   _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(arg, Arg,      _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(round, Round,  _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(floor, Floor,  _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil,  Ceil,   _)
+
+#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
+  template< typename DstXprType, typename SrcXprNested, typename Plain>                                                       \
+  struct Assignment<DstXprType, CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                       \
+                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> >, assign_op<EIGENTYPE,EIGENTYPE>,    \
+                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {            \
+    typedef CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                                           \
+                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> > SrcXprType;                         \
+    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) {                     \
+      resize_if_allowed(dst, src, func);                                                                                      \
+      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                     \
+      VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.rhs().functor().m_other);                                       \
+      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal)                                              \
+      {                                                                                                                       \
+        VMLOP( dst.size(), (const VMLTYPE*)src.lhs().data(), exponent,                                                        \
+              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                         \
+      } else {                                                                                                                \
+        const Index outerSize = dst.outerSize();                                                                              \
+        for(Index outer = 0; outer < outerSize; ++outer) {                                                                    \
+          const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.lhs().coeffRef(outer,0)) :                                        \
+                                                      &(src.lhs().coeffRef(0, outer));                                        \
+          EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                         \
+          VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, exponent,                                                          \
+                 (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                          \
+        }                                                                                                                     \
+      }                                                                                                                       \
+    }                                                                                                                         \
   };
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE)  \
-  template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > {               \
-    enum { IsSupported = 1 };                                                    \
-    static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& /*func*/,        \
-                            int size, const EIGENTYPE* src, EIGENTYPE* dst) {    \
-      MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE;                                    \
-      VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst, vmlMode);                  \
-    }                                                                            \
-  };
-
-#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE)       \
-  template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > {               \
-    enum { IsSupported = 1 };                                                    \
-    static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& func,        \
-                          int size, const EIGENTYPE* src, EIGENTYPE* dst) {      \
-      EIGENTYPE exponent = func.m_exponent;                                      \
-      MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE;                                    \
-      VMLOP(&size, (const VMLTYPE*)src, (const VMLTYPE*)&exponent,               \
-                        (VMLTYPE*)dst, &vmlMode);                                \
-    }                                                                            \
-  };
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP)                   \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vs##VMLOP, float, float)             \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vd##VMLOP, double, double)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP)                \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vc##VMLOP, scomplex, MKL_Complex8)   \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vz##VMLOP, dcomplex, MKL_Complex16)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP)                        \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP)                         \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP)
-
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP)                \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vms##VMLOP, float, float)         \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmd##VMLOP, double, double)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP)             \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmc##VMLOP, scomplex, MKL_Complex8)  \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmz##VMLOP, dcomplex, MKL_Complex16)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(EIGENOP, VMLOP)                     \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP)                      \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP)
-
-
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sin,  Sin)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(asin, Asin)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(cos,  Cos)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(acos, Acos)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(tan,  Tan)
-//EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs,  Abs)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(exp,  Exp)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(log,  Ln)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sqrt, Sqrt)
-
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr)
-
-// The vm*powx functions are not avaibale in the windows version of MKL.
-#ifndef _WIN32
-EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmspowx_, float, float)
-EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdpowx_, double, double)
-EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcpowx_, scomplex, MKL_Complex8)
-EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzpowx_, dcomplex, MKL_Complex16)
-#endif
+  
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmsPowx, float,    float,         LA)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdPowx, double,   double,        LA)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcPowx, scomplex, MKL_Complex8,  LA)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzPowx, dcomplex, MKL_Complex16, LA)
 
 } // end namespace internal
 
diff --git a/extern/Eigen3/Eigen/src/Core/BandMatrix.h b/extern/Eigen3/Eigen/src/Core/BandMatrix.h
index ffd7fe8b301..4978c914057 100644
--- a/extern/Eigen3/Eigen/src/Core/BandMatrix.h
+++ b/extern/Eigen3/Eigen/src/Core/BandMatrix.h
@@ -32,7 +32,7 @@ class BandMatrixBase : public EigenBase<Derived>
     };
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> DenseMatrixType;
-    typedef typename DenseMatrixType::Index Index;
+    typedef typename DenseMatrixType::StorageIndex StorageIndex;
     typedef typename internal::traits<Derived>::CoefficientsType CoefficientsType;
     typedef EigenBase<Derived> Base;
 
@@ -161,15 +161,15 @@ class BandMatrixBase : public EigenBase<Derived>
   *
   * \brief Represents a rectangular matrix with a banded storage
   *
-  * \param _Scalar Numeric type, i.e. float, double, int
-  * \param Rows Number of rows, or \b Dynamic
-  * \param Cols Number of columns, or \b Dynamic
-  * \param Supers Number of super diagonal
-  * \param Subs Number of sub diagonal
-  * \param _Options A combination of either \b #RowMajor or \b #ColMajor, and of \b #SelfAdjoint
-  *                 The former controls \ref TopicStorageOrders "storage order", and defaults to
-  *                 column-major. The latter controls whether the matrix represents a selfadjoint 
-  *                 matrix in which case either Supers of Subs have to be null.
+  * \tparam _Scalar Numeric type, i.e. float, double, int
+  * \tparam _Rows Number of rows, or \b Dynamic
+  * \tparam _Cols Number of columns, or \b Dynamic
+  * \tparam _Supers Number of super diagonal
+  * \tparam _Subs Number of sub diagonal
+  * \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of \b #SelfAdjoint
+  *                  The former controls \ref TopicStorageOrders "storage order", and defaults to
+  *                  column-major. The latter controls whether the matrix represents a selfadjoint
+  *                  matrix in which case either Supers of Subs have to be null.
   *
   * \sa class TridiagonalMatrix
   */
@@ -179,7 +179,7 @@ struct traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
 {
   typedef _Scalar Scalar;
   typedef Dense StorageKind;
-  typedef DenseIndex Index;
+  typedef Eigen::Index StorageIndex;
   enum {
     CoeffReadCost = NumTraits<Scalar>::ReadCost,
     RowsAtCompileTime = _Rows,
@@ -201,10 +201,10 @@ class BandMatrix : public BandMatrixBase<BandMatrix<_Scalar,Rows,Cols,Supers,Sub
   public:
 
     typedef typename internal::traits<BandMatrix>::Scalar Scalar;
-    typedef typename internal::traits<BandMatrix>::Index Index;
+    typedef typename internal::traits<BandMatrix>::StorageIndex StorageIndex;
     typedef typename internal::traits<BandMatrix>::CoefficientsType CoefficientsType;
 
-    inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs)
+    explicit inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs)
       : m_coeffs(1+supers+subs,cols),
         m_rows(rows), m_supers(supers), m_subs(subs)
     {
@@ -241,7 +241,7 @@ struct traits<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Opt
 {
   typedef typename _CoefficientsType::Scalar Scalar;
   typedef typename _CoefficientsType::StorageKind StorageKind;
-  typedef typename _CoefficientsType::Index Index;
+  typedef typename _CoefficientsType::StorageIndex StorageIndex;
   enum {
     CoeffReadCost = internal::traits<_CoefficientsType>::CoeffReadCost,
     RowsAtCompileTime = _Rows,
@@ -264,9 +264,9 @@ class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsT
 
     typedef typename internal::traits<BandMatrixWrapper>::Scalar Scalar;
     typedef typename internal::traits<BandMatrixWrapper>::CoefficientsType CoefficientsType;
-    typedef typename internal::traits<BandMatrixWrapper>::Index Index;
+    typedef typename internal::traits<BandMatrixWrapper>::StorageIndex StorageIndex;
 
-    inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs)
+    explicit inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs)
       : m_coeffs(coeffs),
         m_rows(rows), m_supers(supers), m_subs(subs)
     {
@@ -302,9 +302,9 @@ class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsT
   *
   * \brief Represents a tridiagonal matrix with a compact banded storage
   *
-  * \param _Scalar Numeric type, i.e. float, double, int
-  * \param Size Number of rows and cols, or \b Dynamic
-  * \param _Options Can be 0 or \b SelfAdjoint
+  * \tparam Scalar Numeric type, i.e. float, double, int
+  * \tparam Size Number of rows and cols, or \b Dynamic
+  * \tparam Options Can be 0 or \b SelfAdjoint
   *
   * \sa class BandMatrix
   */
@@ -312,9 +312,9 @@ template<typename Scalar, int Size, int Options>
 class TridiagonalMatrix : public BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor>
 {
     typedef BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor> Base;
-    typedef typename Base::Index Index;
+    typedef typename Base::StorageIndex StorageIndex;
   public:
-    TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {}
+    explicit TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {}
 
     inline typename Base::template DiagonalIntReturnType<1>::Type super()
     { return Base::template diagonal<1>(); }
@@ -327,6 +327,25 @@ class TridiagonalMatrix : public BandMatrix<Scalar,Size,Size,Options&SelfAdjoint
   protected:
 };
 
+
+struct BandShape {};
+
+template<typename _Scalar, int _Rows, int _Cols, int _Supers, int _Subs, int _Options>
+struct evaluator_traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
+  : public evaluator_traits_base<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
+{
+  typedef BandShape Shape;
+};
+
+template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
+struct evaluator_traits<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
+  : public evaluator_traits_base<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
+{
+  typedef BandShape Shape;
+};
+
+template<> struct AssignmentKind<DenseShape,BandShape> { typedef EigenBase2EigenBase Kind; };
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/Block.h b/extern/Eigen3/Eigen/src/Core/Block.h
index 82789444327..11de45c2ecb 100644
--- a/extern/Eigen3/Eigen/src/Core/Block.h
+++ b/extern/Eigen3/Eigen/src/Core/Block.h
@@ -13,38 +13,6 @@
 
 namespace Eigen { 
 
-/** \class Block
-  * \ingroup Core_Module
-  *
-  * \brief Expression of a fixed-size or dynamic-size block
-  *
-  * \param XprType the type of the expression in which we are taking a block
-  * \param BlockRows the number of rows of the block we are taking at compile time (optional)
-  * \param BlockCols the number of columns of the block we are taking at compile time (optional)
-  *
-  * This class represents an expression of either a fixed-size or dynamic-size block. It is the return
-  * type of DenseBase::block(Index,Index,Index,Index) and DenseBase::block<int,int>(Index,Index) and
-  * most of the time this is the only way it is used.
-  *
-  * However, if you want to directly maniputate block expressions,
-  * for instance if you want to write a function returning such an expression, you
-  * will need to use this class.
-  *
-  * Here is an example illustrating the dynamic case:
-  * \include class_Block.cpp
-  * Output: \verbinclude class_Block.out
-  *
-  * \note Even though this expression has dynamic size, in the case where \a XprType
-  * has fixed size, this expression inherits a fixed maximal size which means that evaluating
-  * it does not cause a dynamic memory allocation.
-  *
-  * Here is an example illustrating the fixed-size case:
-  * \include class_FixedBlock.cpp
-  * Output: \verbinclude class_FixedBlock.out
-  *
-  * \sa DenseBase::block(Index,Index,Index,Index), DenseBase::block(Index,Index), class VectorBlock
-  */
-
 namespace internal {
 template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
 struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprType>
@@ -52,7 +20,7 @@ struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprTyp
   typedef typename traits<XprType>::Scalar Scalar;
   typedef typename traits<XprType>::StorageKind StorageKind;
   typedef typename traits<XprType>::XprKind XprKind;
-  typedef typename nested<XprType>::type XprTypeNested;
+  typedef typename ref_selector<XprType>::type XprTypeNested;
   typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
   enum{
     MatrixRows = traits<XprType>::RowsAtCompileTime,
@@ -65,10 +33,10 @@ struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprTyp
     MaxColsAtCompileTime = BlockCols==0 ? 0
                          : ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime)
                          : int(traits<XprType>::MaxColsAtCompileTime),
+
     XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0,
-    IsDense = is_same<StorageKind,Dense>::value,
-    IsRowMajor = (IsDense&&MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
-               : (IsDense&&MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
+    IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
+               : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
                : XprTypeIsRowMajor,
     HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor),
     InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
@@ -78,18 +46,16 @@ struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprTyp
     OuterStrideAtCompileTime = HasSameStorageOrderAsXprType
                              ? int(outer_stride_at_compile_time<XprType>::ret)
                              : int(inner_stride_at_compile_time<XprType>::ret),
-    MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % packet_traits<Scalar>::size) == 0)
-                       && (InnerStrideAtCompileTime == 1)
-                        ? PacketAccessBit : 0,
-    MaskAlignedBit = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % 16) == 0)) ? AlignedBit : 0,
-    FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (traits<XprType>::Flags&LinearAccessBit))) ? LinearAccessBit : 0,
+
+    // FIXME, this traits is rather specialized for dense object and it needs to be cleaned further
     FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
     FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
-    Flags0 = traits<XprType>::Flags & ( (HereditaryBits & ~RowMajorBit) |
-                                        DirectAccessBit |
-                                        MaskPacketAccessBit |
-                                        MaskAlignedBit),
-    Flags = Flags0 | FlagsLinearAccessBit | FlagsLvalueBit | FlagsRowMajorBit
+    Flags = (traits<XprType>::Flags & (DirectAccessBit | (InnerPanel?CompressedAccessBit:0))) | FlagsLvalueBit | FlagsRowMajorBit,
+    // FIXME DirectAccessBit should not be handled by expressions
+    // 
+    // Alignment is needed by MapBase's assertions
+    // We can sefely set it to false here. Internal alignment errors will be detected by an eigen_internal_assert in the respective evaluator
+    Alignment = 0
   };
 };
 
@@ -100,6 +66,40 @@ template<typename XprType, int BlockRows=Dynamic, int BlockCols=Dynamic, bool In
 
 template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, typename StorageKind> class BlockImpl;
 
+/** \class Block
+  * \ingroup Core_Module
+  *
+  * \brief Expression of a fixed-size or dynamic-size block
+  *
+  * \tparam XprType the type of the expression in which we are taking a block
+  * \tparam BlockRows the number of rows of the block we are taking at compile time (optional)
+  * \tparam BlockCols the number of columns of the block we are taking at compile time (optional)
+  * \tparam InnerPanel is true, if the block maps to a set of rows of a row major matrix or
+  *         to set of columns of a column major matrix (optional). The parameter allows to determine
+  *         at compile time whether aligned access is possible on the block expression.
+  *
+  * This class represents an expression of either a fixed-size or dynamic-size block. It is the return
+  * type of DenseBase::block(Index,Index,Index,Index) and DenseBase::block<int,int>(Index,Index) and
+  * most of the time this is the only way it is used.
+  *
+  * However, if you want to directly maniputate block expressions,
+  * for instance if you want to write a function returning such an expression, you
+  * will need to use this class.
+  *
+  * Here is an example illustrating the dynamic case:
+  * \include class_Block.cpp
+  * Output: \verbinclude class_Block.out
+  *
+  * \note Even though this expression has dynamic size, in the case where \a XprType
+  * has fixed size, this expression inherits a fixed maximal size which means that evaluating
+  * it does not cause a dynamic memory allocation.
+  *
+  * Here is an example illustrating the fixed-size case:
+  * \include class_FixedBlock.cpp
+  * Output: \verbinclude class_FixedBlock.out
+  *
+  * \sa DenseBase::block(Index,Index,Index,Index), DenseBase::block(Index,Index), class VectorBlock
+  */
 template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class Block
   : public BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, typename internal::traits<XprType>::StorageKind>
 {
@@ -109,9 +109,12 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
     typedef Impl Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(Block)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Block)
+    
+    typedef typename internal::remove_all<XprType>::type NestedExpression;
   
     /** Column or Row constructor
       */
+    EIGEN_DEVICE_FUNC
     inline Block(XprType& xpr, Index i) : Impl(xpr,i)
     {
       eigen_assert( (i>=0) && (
@@ -121,25 +124,27 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
 
     /** Fixed-size constructor
       */
-    inline Block(XprType& xpr, Index a_startRow, Index a_startCol)
-      : Impl(xpr, a_startRow, a_startCol)
+    EIGEN_DEVICE_FUNC
+    inline Block(XprType& xpr, Index startRow, Index startCol)
+      : Impl(xpr, startRow, startCol)
     {
       EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE)
-      eigen_assert(a_startRow >= 0 && BlockRows >= 1 && a_startRow + BlockRows <= xpr.rows()
-             && a_startCol >= 0 && BlockCols >= 1 && a_startCol + BlockCols <= xpr.cols());
+      eigen_assert(startRow >= 0 && BlockRows >= 0 && startRow + BlockRows <= xpr.rows()
+             && startCol >= 0 && BlockCols >= 0 && startCol + BlockCols <= xpr.cols());
     }
 
     /** Dynamic-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline Block(XprType& xpr,
-          Index a_startRow, Index a_startCol,
+          Index startRow, Index startCol,
           Index blockRows, Index blockCols)
-      : Impl(xpr, a_startRow, a_startCol, blockRows, blockCols)
+      : Impl(xpr, startRow, startCol, blockRows, blockCols)
     {
       eigen_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==blockRows)
           && (ColsAtCompileTime==Dynamic || ColsAtCompileTime==blockCols));
-      eigen_assert(a_startRow >= 0 && blockRows >= 0 && a_startRow  <= xpr.rows() - blockRows
-          && a_startCol >= 0 && blockCols >= 0 && a_startCol <= xpr.cols() - blockCols);
+      eigen_assert(startRow >= 0 && blockRows >= 0 && startRow  <= xpr.rows() - blockRows
+          && startCol >= 0 && blockCols >= 0 && startCol <= xpr.cols() - blockCols);
     }
 };
          
@@ -150,14 +155,15 @@ class BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, Dense>
   : public internal::BlockImpl_dense<XprType, BlockRows, BlockCols, InnerPanel>
 {
     typedef internal::BlockImpl_dense<XprType, BlockRows, BlockCols, InnerPanel> Impl;
-    typedef typename XprType::Index Index;
+    typedef typename XprType::StorageIndex StorageIndex;
   public:
     typedef Impl Base;
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
-    inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
-    inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol) : Impl(xpr, a_startRow, a_startCol) {}
-    inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol, Index blockRows, Index blockCols)
-      : Impl(xpr, a_startRow, a_startCol, blockRows, blockCols) {}
+    EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
+    EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {}
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+      : Impl(xpr, startRow, startCol, blockRows, blockCols) {}
 };
 
 namespace internal {
@@ -167,16 +173,18 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
   : public internal::dense_xpr_base<Block<XprType, BlockRows, BlockCols, InnerPanel> >::type
 {
     typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
+    typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
   public:
 
     typedef typename internal::dense_xpr_base<BlockType>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(BlockType)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl_dense)
 
-    class InnerIterator;
+    // class InnerIterator; // FIXME apparently never used
 
     /** Column or Row constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, Index i)
       : m_xpr(xpr),
         // It is a row if and only if BlockRows==1 and BlockCols==XprType::ColsAtCompileTime,
@@ -191,75 +199,76 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
 
     /** Fixed-size constructor
       */
-    inline BlockImpl_dense(XprType& xpr, Index a_startRow, Index a_startCol)
-      : m_xpr(xpr), m_startRow(a_startRow), m_startCol(a_startCol),
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
+      : m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
                     m_blockRows(BlockRows), m_blockCols(BlockCols)
     {}
 
     /** Dynamic-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr,
-          Index a_startRow, Index a_startCol,
+          Index startRow, Index startCol,
           Index blockRows, Index blockCols)
-      : m_xpr(xpr), m_startRow(a_startRow), m_startCol(a_startCol),
+      : m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
                     m_blockRows(blockRows), m_blockCols(blockCols)
     {}
 
-    inline Index rows() const { return m_blockRows.value(); }
-    inline Index cols() const { return m_blockCols.value(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_blockRows.value(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_blockCols.value(); }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index rowId, Index colId)
     {
       EIGEN_STATIC_ASSERT_LVALUE(XprType)
-      return m_xpr.const_cast_derived()
-               .coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
+      return m_xpr.coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
-      return m_xpr.derived()
-               .coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
+      return m_xpr.derived().coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index rowId, Index colId) const
     {
       return m_xpr.coeff(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index index)
     {
       EIGEN_STATIC_ASSERT_LVALUE(XprType)
-      return m_xpr.const_cast_derived()
-             .coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                       m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_xpr.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                            m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
-      return m_xpr.const_cast_derived()
-             .coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                       m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_xpr.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                            m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
+    EIGEN_DEVICE_FUNC
     inline const CoeffReturnType coeff(Index index) const
     {
-      return m_xpr
-             .coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                    m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_xpr.coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                         m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
     template<int LoadMode>
     inline PacketScalar packet(Index rowId, Index colId) const
     {
-      return m_xpr.template packet<Unaligned>
-              (rowId + m_startRow.value(), colId + m_startCol.value());
+      return m_xpr.template packet<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
     template<int LoadMode>
     inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
     {
-      m_xpr.const_cast_derived().template writePacket<Unaligned>
-              (rowId + m_startRow.value(), colId + m_startCol.value(), val);
+      m_xpr.template writePacket<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value(), val);
     }
 
     template<int LoadMode>
@@ -273,40 +282,46 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
     template<int LoadMode>
     inline void writePacket(Index index, const PacketScalar& val)
     {
-      m_xpr.const_cast_derived().template writePacket<Unaligned>
+      m_xpr.template writePacket<Unaligned>
          (m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
           m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0), val);
     }
 
     #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \sa MapBase::data() */
-    inline const Scalar* data() const;
-    inline Index innerStride() const;
-    inline Index outerStride() const;
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const;
+    EIGEN_DEVICE_FUNC inline Index innerStride() const;
+    EIGEN_DEVICE_FUNC inline Index outerStride() const;
     #endif
 
-    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const 
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
     { 
       return m_xpr; 
     }
+
+    EIGEN_DEVICE_FUNC
+    XprType& nestedExpression() { return m_xpr; }
       
-    Index startRow() const 
+    EIGEN_DEVICE_FUNC
+    StorageIndex startRow() const
     { 
       return m_startRow.value(); 
     }
       
-    Index startCol() const 
+    EIGEN_DEVICE_FUNC
+    StorageIndex startCol() const
     { 
       return m_startCol.value(); 
     }
 
   protected:
 
-    const typename XprType::Nested m_xpr;
-    const internal::variable_if_dynamic<Index, XprType::RowsAtCompileTime == 1 ? 0 : Dynamic> m_startRow;
-    const internal::variable_if_dynamic<Index, XprType::ColsAtCompileTime == 1 ? 0 : Dynamic> m_startCol;
-    const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_blockRows;
-    const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_blockCols;
+    XprTypeNested m_xpr;
+    const internal::variable_if_dynamic<StorageIndex, (XprType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
+    const internal::variable_if_dynamic<StorageIndex, (XprType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
+    const internal::variable_if_dynamic<StorageIndex, RowsAtCompileTime> m_blockRows;
+    const internal::variable_if_dynamic<StorageIndex, ColsAtCompileTime> m_blockCols;
 };
 
 /** \internal Internal implementation of dense Blocks in the direct access case.*/
@@ -315,6 +330,10 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
   : public MapBase<Block<XprType, BlockRows, BlockCols, InnerPanel> >
 {
     typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
+    typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
+    enum {
+      XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0
+    };
   public:
 
     typedef MapBase<BlockType> Base;
@@ -323,42 +342,52 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     /** Column or Row constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, Index i)
-      : Base(internal::const_cast_ptr(&xpr.coeffRef(
-              (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0,
-              (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0)),
+      : Base(xpr.data() + i * (    ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && (!XprTypeIsRowMajor)) 
+                                || ((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && ( XprTypeIsRowMajor)) ? xpr.innerStride() : xpr.outerStride()),
              BlockRows==1 ? 1 : xpr.rows(),
              BlockCols==1 ? 1 : xpr.cols()),
-        m_xpr(xpr)
+        m_xpr(xpr),
+        m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0),
+        m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0)
     {
       init();
     }
 
     /** Fixed-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
-      : Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol))), m_xpr(xpr)
+      : Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol)),
+        m_xpr(xpr), m_startRow(startRow), m_startCol(startCol)
     {
       init();
     }
 
     /** Dynamic-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr,
           Index startRow, Index startCol,
           Index blockRows, Index blockCols)
-      : Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol)), blockRows, blockCols),
-        m_xpr(xpr)
+      : Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol), blockRows, blockCols),
+        m_xpr(xpr), m_startRow(startRow), m_startCol(startCol)
     {
       init();
     }
 
-    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const 
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
     { 
       return m_xpr; 
     }
+
+    EIGEN_DEVICE_FUNC
+    XprType& nestedExpression() { return m_xpr; }
       
     /** \sa MapBase::innerStride() */
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return internal::traits<BlockType>::HasSameStorageOrderAsXprType
@@ -367,11 +396,24 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
     }
 
     /** \sa MapBase::outerStride() */
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return m_outerStride;
     }
 
+    EIGEN_DEVICE_FUNC
+    StorageIndex startRow() const
+    {
+      return m_startRow.value();
+    }
+
+    EIGEN_DEVICE_FUNC
+    StorageIndex startCol() const
+    {
+      return m_startCol.value();
+    }
+
   #ifndef __SUNPRO_CC
   // FIXME sunstudio is not friendly with the above friend...
   // META-FIXME there is no 'friend' keyword around here. Is this obsolete?
@@ -380,6 +422,7 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal used by allowAligned() */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
       : Base(data, blockRows, blockCols), m_xpr(xpr)
     {
@@ -388,6 +431,7 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
     #endif
 
   protected:
+    EIGEN_DEVICE_FUNC
     void init()
     {
       m_outerStride = internal::traits<BlockType>::HasSameStorageOrderAsXprType
@@ -395,7 +439,9 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
                     : m_xpr.innerStride();
     }
 
-    typename XprType::Nested m_xpr;
+    XprTypeNested m_xpr;
+    const internal::variable_if_dynamic<StorageIndex, (XprType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
+    const internal::variable_if_dynamic<StorageIndex, (XprType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
     Index m_outerStride;
 };
 
diff --git a/extern/Eigen3/Eigen/src/Core/BooleanRedux.h b/extern/Eigen3/Eigen/src/Core/BooleanRedux.h
index be9f48a8c71..8409d8749ad 100644
--- a/extern/Eigen3/Eigen/src/Core/BooleanRedux.h
+++ b/extern/Eigen3/Eigen/src/Core/BooleanRedux.h
@@ -17,9 +17,10 @@ namespace internal {
 template<typename Derived, int UnrollCount>
 struct all_unroller
 {
+  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived::RowsAtCompileTime
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
   };
 
   static inline bool run(const Derived &mat)
@@ -43,11 +44,12 @@ struct all_unroller<Derived, Dynamic>
 template<typename Derived, int UnrollCount>
 struct any_unroller
 {
+  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived::RowsAtCompileTime
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
   };
-
+  
   static inline bool run(const Derived &mat)
   {
     return any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
@@ -78,19 +80,19 @@ struct any_unroller<Derived, Dynamic>
 template<typename Derived>
 inline bool DenseBase<Derived>::all() const
 {
+  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && CoeffReadCost != Dynamic
-          && NumTraits<Scalar>::AddCost != Dynamic
-          && SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
+  Evaluator evaluator(derived());
   if(unroll)
-    return internal::all_unroller<Derived, unroll ? int(SizeAtCompileTime) : Dynamic>::run(derived());
+    return internal::all_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
   else
   {
     for(Index j = 0; j < cols(); ++j)
       for(Index i = 0; i < rows(); ++i)
-        if (!coeff(i, j)) return false;
+        if (!evaluator.coeff(i, j)) return false;
     return true;
   }
 }
@@ -102,19 +104,19 @@ inline bool DenseBase<Derived>::all() const
 template<typename Derived>
 inline bool DenseBase<Derived>::any() const
 {
+  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && CoeffReadCost != Dynamic
-          && NumTraits<Scalar>::AddCost != Dynamic
-          && SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
+  Evaluator evaluator(derived());
   if(unroll)
-    return internal::any_unroller<Derived, unroll ? int(SizeAtCompileTime) : Dynamic>::run(derived());
+    return internal::any_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
   else
   {
     for(Index j = 0; j < cols(); ++j)
       for(Index i = 0; i < rows(); ++i)
-        if (coeff(i, j)) return true;
+        if (evaluator.coeff(i, j)) return true;
     return false;
   }
 }
@@ -124,7 +126,7 @@ inline bool DenseBase<Derived>::any() const
   * \sa all(), any()
   */
 template<typename Derived>
-inline typename DenseBase<Derived>::Index DenseBase<Derived>::count() const
+inline Eigen::Index DenseBase<Derived>::count() const
 {
   return derived().template cast<bool>().template cast<Index>().sum();
 }
@@ -136,7 +138,11 @@ inline typename DenseBase<Derived>::Index DenseBase<Derived>::count() const
 template<typename Derived>
 inline bool DenseBase<Derived>::hasNaN() const
 {
+#if EIGEN_COMP_MSVC || (defined __FAST_MATH__)
+  return derived().array().isNaN().any();
+#else
   return !((derived().array()==derived().array()).all());
+#endif
 }
 
 /** \returns true if \c *this contains only finite numbers, i.e., no NaN and no +/-INF values.
@@ -146,7 +152,11 @@ inline bool DenseBase<Derived>::hasNaN() const
 template<typename Derived>
 inline bool DenseBase<Derived>::allFinite() const
 {
+#if EIGEN_COMP_MSVC || (defined __FAST_MATH__)
+  return derived().array().isFinite().all();
+#else
   return !((derived()-derived()).hasNaN());
+#endif
 }
     
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/CommaInitializer.h b/extern/Eigen3/Eigen/src/Core/CommaInitializer.h
index a036d8c3bc9..d218e98143f 100644
--- a/extern/Eigen3/Eigen/src/Core/CommaInitializer.h
+++ b/extern/Eigen3/Eigen/src/Core/CommaInitializer.h
@@ -22,14 +22,14 @@ namespace Eigen {
   * the return type of MatrixBase::operator<<, and most of the time this is the only
   * way it is used.
   *
-  * \sa \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
+  * \sa \blank \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
   */
 template<typename XprType>
 struct CommaInitializer
 {
   typedef typename XprType::Scalar Scalar;
-  typedef typename XprType::Index Index;
 
+  EIGEN_DEVICE_FUNC
   inline CommaInitializer(XprType& xpr, const Scalar& s)
     : m_xpr(xpr), m_row(0), m_col(1), m_currentBlockRows(1)
   {
@@ -37,6 +37,7 @@ struct CommaInitializer
   }
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   inline CommaInitializer(XprType& xpr, const DenseBase<OtherDerived>& other)
     : m_xpr(xpr), m_row(0), m_col(other.cols()), m_currentBlockRows(other.rows())
   {
@@ -46,6 +47,7 @@ struct CommaInitializer
   /* Copy/Move constructor which transfers ownership. This is crucial in 
    * absence of return value optimization to avoid assertions during destruction. */
   // FIXME in C++11 mode this could be replaced by a proper RValue constructor
+  EIGEN_DEVICE_FUNC
   inline CommaInitializer(const CommaInitializer& o)
   : m_xpr(o.m_xpr), m_row(o.m_row), m_col(o.m_col), m_currentBlockRows(o.m_currentBlockRows) {
     // Mark original object as finished. In absence of R-value references we need to const_cast:
@@ -55,6 +57,7 @@ struct CommaInitializer
   }
 
   /* inserts a scalar value in the target matrix */
+  EIGEN_DEVICE_FUNC
   CommaInitializer& operator,(const Scalar& s)
   {
     if (m_col==m_xpr.cols())
@@ -74,11 +77,10 @@ struct CommaInitializer
 
   /* inserts a matrix expression in the target matrix */
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   CommaInitializer& operator,(const DenseBase<OtherDerived>& other)
   {
-    if(other.cols()==0 || other.rows()==0)
-      return *this;
-    if (m_col==m_xpr.cols())
+    if (m_col==m_xpr.cols() && (other.cols()!=0 || other.rows()!=m_currentBlockRows))
     {
       m_row+=m_currentBlockRows;
       m_col = 0;
@@ -86,24 +88,22 @@ struct CommaInitializer
       eigen_assert(m_row+m_currentBlockRows<=m_xpr.rows()
         && "Too many rows passed to comma initializer (operator<<)");
     }
-    eigen_assert(m_col<m_xpr.cols()
+    eigen_assert((m_col + other.cols() <= m_xpr.cols())
       && "Too many coefficients passed to comma initializer (operator<<)");
     eigen_assert(m_currentBlockRows==other.rows());
-    if (OtherDerived::SizeAtCompileTime != Dynamic)
-      m_xpr.template block<OtherDerived::RowsAtCompileTime != Dynamic ? OtherDerived::RowsAtCompileTime : 1,
-                              OtherDerived::ColsAtCompileTime != Dynamic ? OtherDerived::ColsAtCompileTime : 1>
-                    (m_row, m_col) = other;
-    else
-      m_xpr.block(m_row, m_col, other.rows(), other.cols()) = other;
+    m_xpr.template block<OtherDerived::RowsAtCompileTime, OtherDerived::ColsAtCompileTime>
+                    (m_row, m_col, other.rows(), other.cols()) = other;
     m_col += other.cols();
     return *this;
   }
 
+  EIGEN_DEVICE_FUNC
   inline ~CommaInitializer()
+#if defined VERIFY_RAISES_ASSERT && (!defined EIGEN_NO_ASSERTION_CHECKING) && defined EIGEN_EXCEPTIONS
+  EIGEN_EXCEPTION_SPEC(Eigen::eigen_assert_exception)
+#endif
   {
-    eigen_assert((m_row+m_currentBlockRows) == m_xpr.rows()
-         && m_col == m_xpr.cols()
-         && "Too few coefficients passed to comma initializer (operator<<)");
+      finished();
   }
 
   /** \returns the built matrix once all its coefficients have been set.
@@ -113,9 +113,15 @@ struct CommaInitializer
     * quaternion.fromRotationMatrix((Matrix3f() << axis0, axis1, axis2).finished());
     * \endcode
     */
-  inline XprType& finished() { return m_xpr; }
+  EIGEN_DEVICE_FUNC
+  inline XprType& finished() {
+      eigen_assert(((m_row+m_currentBlockRows) == m_xpr.rows() || m_xpr.cols() == 0)
+           && m_col == m_xpr.cols()
+           && "Too few coefficients passed to comma initializer (operator<<)");
+      return m_xpr;
+  }
 
-  XprType& m_xpr;   // target expression
+  XprType& m_xpr;           // target expression
   Index m_row;              // current row id
   Index m_col;              // current col id
   Index m_currentBlockRows; // current block height
diff --git a/extern/Eigen3/Eigen/src/Core/ConditionEstimator.h b/extern/Eigen3/Eigen/src/Core/ConditionEstimator.h
new file mode 100644
index 00000000000..51a2e5f1b6f
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/ConditionEstimator.h
@@ -0,0 +1,175 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Rasmus Munk Larsen (rmlarsen@google.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_CONDITIONESTIMATOR_H
+#define EIGEN_CONDITIONESTIMATOR_H
+
+namespace Eigen {
+
+namespace internal {
+
+template <typename Vector, typename RealVector, bool IsComplex>
+struct rcond_compute_sign {
+  static inline Vector run(const Vector& v) {
+    const RealVector v_abs = v.cwiseAbs();
+    return (v_abs.array() == static_cast<typename Vector::RealScalar>(0))
+            .select(Vector::Ones(v.size()), v.cwiseQuotient(v_abs));
+  }
+};
+
+// Partial specialization to avoid elementwise division for real vectors.
+template <typename Vector>
+struct rcond_compute_sign<Vector, Vector, false> {
+  static inline Vector run(const Vector& v) {
+    return (v.array() < static_cast<typename Vector::RealScalar>(0))
+           .select(-Vector::Ones(v.size()), Vector::Ones(v.size()));
+  }
+};
+
+/**
+  * \returns an estimate of ||inv(matrix)||_1 given a decomposition of
+  * \a matrix that implements .solve() and .adjoint().solve() methods.
+  *
+  * This function implements Algorithms 4.1 and 5.1 from
+  *   http://www.maths.manchester.ac.uk/~higham/narep/narep135.pdf
+  * which also forms the basis for the condition number estimators in
+  * LAPACK. Since at most 10 calls to the solve method of dec are
+  * performed, the total cost is O(dims^2), as opposed to O(dims^3)
+  * needed to compute the inverse matrix explicitly.
+  *
+  * The most common usage is in estimating the condition number
+  * ||matrix||_1 * ||inv(matrix)||_1. The first term ||matrix||_1 can be
+  * computed directly in O(n^2) operations.
+  *
+  * Supports the following decompositions: FullPivLU, PartialPivLU, LDLT, and
+  * LLT.
+  *
+  * \sa FullPivLU, PartialPivLU, LDLT, LLT.
+  */
+template <typename Decomposition>
+typename Decomposition::RealScalar rcond_invmatrix_L1_norm_estimate(const Decomposition& dec)
+{
+  typedef typename Decomposition::MatrixType MatrixType;
+  typedef typename Decomposition::Scalar Scalar;
+  typedef typename Decomposition::RealScalar RealScalar;
+  typedef typename internal::plain_col_type<MatrixType>::type Vector;
+  typedef typename internal::plain_col_type<MatrixType, RealScalar>::type RealVector;
+  const bool is_complex = (NumTraits<Scalar>::IsComplex != 0);
+
+  eigen_assert(dec.rows() == dec.cols());
+  const Index n = dec.rows();
+  if (n == 0)
+    return 0;
+
+  // Disable Index to float conversion warning
+#ifdef __INTEL_COMPILER
+  #pragma warning push
+  #pragma warning ( disable : 2259 )
+#endif
+  Vector v = dec.solve(Vector::Ones(n) / Scalar(n));
+#ifdef __INTEL_COMPILER
+  #pragma warning pop
+#endif
+
+  // lower_bound is a lower bound on
+  //   ||inv(matrix)||_1  = sup_v ||inv(matrix) v||_1 / ||v||_1
+  // and is the objective maximized by the ("super-") gradient ascent
+  // algorithm below.
+  RealScalar lower_bound = v.template lpNorm<1>();
+  if (n == 1)
+    return lower_bound;
+
+  // Gradient ascent algorithm follows: We know that the optimum is achieved at
+  // one of the simplices v = e_i, so in each iteration we follow a
+  // super-gradient to move towards the optimal one.
+  RealScalar old_lower_bound = lower_bound;
+  Vector sign_vector(n);
+  Vector old_sign_vector;
+  Index v_max_abs_index = -1;
+  Index old_v_max_abs_index = v_max_abs_index;
+  for (int k = 0; k < 4; ++k)
+  {
+    sign_vector = internal::rcond_compute_sign<Vector, RealVector, is_complex>::run(v);
+    if (k > 0 && !is_complex && sign_vector == old_sign_vector) {
+      // Break if the solution stagnated.
+      break;
+    }
+    // v_max_abs_index = argmax |real( inv(matrix)^T * sign_vector )|
+    v = dec.adjoint().solve(sign_vector);
+    v.real().cwiseAbs().maxCoeff(&v_max_abs_index);
+    if (v_max_abs_index == old_v_max_abs_index) {
+      // Break if the solution stagnated.
+      break;
+    }
+    // Move to the new simplex e_j, where j = v_max_abs_index.
+    v = dec.solve(Vector::Unit(n, v_max_abs_index));  // v = inv(matrix) * e_j.
+    lower_bound = v.template lpNorm<1>();
+    if (lower_bound <= old_lower_bound) {
+      // Break if the gradient step did not increase the lower_bound.
+      break;
+    }
+    if (!is_complex) {
+      old_sign_vector = sign_vector;
+    }
+    old_v_max_abs_index = v_max_abs_index;
+    old_lower_bound = lower_bound;
+  }
+  // The following calculates an independent estimate of ||matrix||_1 by
+  // multiplying matrix by a vector with entries of slowly increasing
+  // magnitude and alternating sign:
+  //   v_i = (-1)^{i} (1 + (i / (dim-1))), i = 0,...,dim-1.
+  // This improvement to Hager's algorithm above is due to Higham. It was
+  // added to make the algorithm more robust in certain corner cases where
+  // large elements in the matrix might otherwise escape detection due to
+  // exact cancellation (especially when op and op_adjoint correspond to a
+  // sequence of backsubstitutions and permutations), which could cause
+  // Hager's algorithm to vastly underestimate ||matrix||_1.
+  Scalar alternating_sign(RealScalar(1));
+  for (Index i = 0; i < n; ++i) {
+    // The static_cast is needed when Scalar is a complex and RealScalar implements expression templates
+    v[i] = alternating_sign * static_cast<RealScalar>(RealScalar(1) + (RealScalar(i) / (RealScalar(n - 1))));
+    alternating_sign = -alternating_sign;
+  }
+  v = dec.solve(v);
+  const RealScalar alternate_lower_bound = (2 * v.template lpNorm<1>()) / (3 * RealScalar(n));
+  return numext::maxi(lower_bound, alternate_lower_bound);
+}
+
+/** \brief Reciprocal condition number estimator.
+  *
+  * Computing a decomposition of a dense matrix takes O(n^3) operations, while
+  * this method estimates the condition number quickly and reliably in O(n^2)
+  * operations.
+  *
+  * \returns an estimate of the reciprocal condition number
+  * (1 / (||matrix||_1 * ||inv(matrix)||_1)) of matrix, given ||matrix||_1 and
+  * its decomposition. Supports the following decompositions: FullPivLU,
+  * PartialPivLU, LDLT, and LLT.
+  *
+  * \sa FullPivLU, PartialPivLU, LDLT, LLT.
+  */
+template <typename Decomposition>
+typename Decomposition::RealScalar
+rcond_estimate_helper(typename Decomposition::RealScalar matrix_norm, const Decomposition& dec)
+{
+  typedef typename Decomposition::RealScalar RealScalar;
+  eigen_assert(dec.rows() == dec.cols());
+  if (dec.rows() == 0)              return NumTraits<RealScalar>::infinity();
+  if (matrix_norm == RealScalar(0)) return RealScalar(0);
+  if (dec.rows() == 1)              return RealScalar(1);
+  const RealScalar inverse_matrix_norm = rcond_invmatrix_L1_norm_estimate(dec);
+  return (inverse_matrix_norm == RealScalar(0) ? RealScalar(0)
+                                               : (RealScalar(1) / inverse_matrix_norm) / matrix_norm);
+}
+
+}  // namespace internal
+
+}  // namespace Eigen
+
+#endif
diff --git a/extern/Eigen3/Eigen/src/Core/CoreEvaluators.h b/extern/Eigen3/Eigen/src/Core/CoreEvaluators.h
new file mode 100644
index 00000000000..910889efa70
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/CoreEvaluators.h
@@ -0,0 +1,1688 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+
+#ifndef EIGEN_COREEVALUATORS_H
+#define EIGEN_COREEVALUATORS_H
+
+namespace Eigen {
+  
+namespace internal {
+
+// This class returns the evaluator kind from the expression storage kind.
+// Default assumes index based accessors
+template<typename StorageKind>
+struct storage_kind_to_evaluator_kind {
+  typedef IndexBased Kind;
+};
+
+// This class returns the evaluator shape from the expression storage kind.
+// It can be Dense, Sparse, Triangular, Diagonal, SelfAdjoint, Band, etc.
+template<typename StorageKind> struct storage_kind_to_shape;
+
+template<> struct storage_kind_to_shape<Dense>                  { typedef DenseShape Shape;           };
+template<> struct storage_kind_to_shape<SolverStorage>          { typedef SolverShape Shape;           };
+template<> struct storage_kind_to_shape<PermutationStorage>     { typedef PermutationShape Shape;     };
+template<> struct storage_kind_to_shape<TranspositionsStorage>  { typedef TranspositionsShape Shape;  };
+
+// Evaluators have to be specialized with respect to various criteria such as:
+//  - storage/structure/shape
+//  - scalar type
+//  - etc.
+// Therefore, we need specialization of evaluator providing additional template arguments for each kind of evaluators.
+// We currently distinguish the following kind of evaluators:
+// - unary_evaluator    for expressions taking only one arguments (CwiseUnaryOp, CwiseUnaryView, Transpose, MatrixWrapper, ArrayWrapper, Reverse, Replicate)
+// - binary_evaluator   for expression taking two arguments (CwiseBinaryOp)
+// - ternary_evaluator   for expression taking three arguments (CwiseTernaryOp)
+// - product_evaluator  for linear algebra products (Product); special case of binary_evaluator because it requires additional tags for dispatching.
+// - mapbase_evaluator  for Map, Block, Ref
+// - block_evaluator    for Block (special dispatching to a mapbase_evaluator or unary_evaluator)
+
+template< typename T,
+          typename Arg1Kind   = typename evaluator_traits<typename T::Arg1>::Kind,
+          typename Arg2Kind   = typename evaluator_traits<typename T::Arg2>::Kind,
+          typename Arg3Kind   = typename evaluator_traits<typename T::Arg3>::Kind,
+          typename Arg1Scalar = typename traits<typename T::Arg1>::Scalar,
+          typename Arg2Scalar = typename traits<typename T::Arg2>::Scalar,
+          typename Arg3Scalar = typename traits<typename T::Arg3>::Scalar> struct ternary_evaluator;
+
+template< typename T,
+          typename LhsKind   = typename evaluator_traits<typename T::Lhs>::Kind,
+          typename RhsKind   = typename evaluator_traits<typename T::Rhs>::Kind,
+          typename LhsScalar = typename traits<typename T::Lhs>::Scalar,
+          typename RhsScalar = typename traits<typename T::Rhs>::Scalar> struct binary_evaluator;
+
+template< typename T,
+          typename Kind   = typename evaluator_traits<typename T::NestedExpression>::Kind,
+          typename Scalar = typename T::Scalar> struct unary_evaluator;
+          
+// evaluator_traits<T> contains traits for evaluator<T> 
+
+template<typename T>
+struct evaluator_traits_base
+{
+  // by default, get evaluator kind and shape from storage
+  typedef typename storage_kind_to_evaluator_kind<typename traits<T>::StorageKind>::Kind Kind;
+  typedef typename storage_kind_to_shape<typename traits<T>::StorageKind>::Shape Shape;
+};
+
+// Default evaluator traits
+template<typename T>
+struct evaluator_traits : public evaluator_traits_base<T>
+{
+};
+
+template<typename T, typename Shape = typename evaluator_traits<T>::Shape >
+struct evaluator_assume_aliasing {
+  static const bool value = false;
+};
+
+// By default, we assume a unary expression:
+template<typename T>
+struct evaluator : public unary_evaluator<T>
+{
+  typedef unary_evaluator<T> Base;
+  EIGEN_DEVICE_FUNC explicit evaluator(const T& xpr) : Base(xpr) {}
+};
+
+
+// TODO: Think about const-correctness
+template<typename T>
+struct evaluator<const T>
+  : evaluator<T>
+{
+  EIGEN_DEVICE_FUNC
+  explicit evaluator(const T& xpr) : evaluator<T>(xpr) {}
+};
+
+// ---------- base class for all evaluators ----------
+
+template<typename ExpressionType>
+struct evaluator_base : public noncopyable
+{
+  // TODO that's not very nice to have to propagate all these traits. They are currently only needed to handle outer,inner indices.
+  typedef traits<ExpressionType> ExpressionTraits;
+  
+  enum {
+    Alignment = 0
+  };
+};
+
+// -------------------- Matrix and Array --------------------
+//
+// evaluator<PlainObjectBase> is a common base class for the
+// Matrix and Array evaluators.
+// Here we directly specialize evaluator. This is not really a unary expression, and it is, by definition, dense,
+// so no need for more sophisticated dispatching.
+
+template<typename Derived>
+struct evaluator<PlainObjectBase<Derived> >
+  : evaluator_base<Derived>
+{
+  typedef PlainObjectBase<Derived> PlainObjectType;
+  typedef typename PlainObjectType::Scalar Scalar;
+  typedef typename PlainObjectType::CoeffReturnType CoeffReturnType;
+
+  enum {
+    IsRowMajor = PlainObjectType::IsRowMajor,
+    IsVectorAtCompileTime = PlainObjectType::IsVectorAtCompileTime,
+    RowsAtCompileTime = PlainObjectType::RowsAtCompileTime,
+    ColsAtCompileTime = PlainObjectType::ColsAtCompileTime,
+    
+    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Flags = traits<Derived>::EvaluatorFlags,
+    Alignment = traits<Derived>::Alignment
+  };
+  
+  EIGEN_DEVICE_FUNC evaluator()
+    : m_data(0),
+      m_outerStride(IsVectorAtCompileTime  ? 0 
+                                           : int(IsRowMajor) ? ColsAtCompileTime 
+                                           : RowsAtCompileTime)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const PlainObjectType& m)
+    : m_data(m.data()), m_outerStride(IsVectorAtCompileTime ? 0 : m.outerStride()) 
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    if (IsRowMajor)
+      return m_data[row * m_outerStride.value() + col];
+    else
+      return m_data[row + col * m_outerStride.value()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_data[index];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    if (IsRowMajor)
+      return const_cast<Scalar*>(m_data)[row * m_outerStride.value() + col];
+    else
+      return const_cast<Scalar*>(m_data)[row + col * m_outerStride.value()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return const_cast<Scalar*>(m_data)[index];
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    if (IsRowMajor)
+      return ploadt<PacketType, LoadMode>(m_data + row * m_outerStride.value() + col);
+    else
+      return ploadt<PacketType, LoadMode>(m_data + row + col * m_outerStride.value());
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return ploadt<PacketType, LoadMode>(m_data + index);
+  }
+
+  template<int StoreMode,typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    if (IsRowMajor)
+      return pstoret<Scalar, PacketType, StoreMode>
+	            (const_cast<Scalar*>(m_data) + row * m_outerStride.value() + col, x);
+    else
+      return pstoret<Scalar, PacketType, StoreMode>
+                    (const_cast<Scalar*>(m_data) + row + col * m_outerStride.value(), x);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    return pstoret<Scalar, PacketType, StoreMode>(const_cast<Scalar*>(m_data) + index, x);
+  }
+
+protected:
+  const Scalar *m_data;
+
+  // We do not need to know the outer stride for vectors
+  variable_if_dynamic<Index, IsVectorAtCompileTime  ? 0 
+                                                    : int(IsRowMajor) ? ColsAtCompileTime 
+                                                    : RowsAtCompileTime> m_outerStride;
+};
+
+template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
+struct evaluator<Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
+  : evaluator<PlainObjectBase<Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> > >
+{
+  typedef Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> XprType;
+  
+  EIGEN_DEVICE_FUNC evaluator() {}
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& m)
+    : evaluator<PlainObjectBase<XprType> >(m) 
+  { }
+};
+
+template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
+struct evaluator<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
+  : evaluator<PlainObjectBase<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> > >
+{
+  typedef Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> XprType;
+
+  EIGEN_DEVICE_FUNC evaluator() {}
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& m)
+    : evaluator<PlainObjectBase<XprType> >(m) 
+  { }
+};
+
+// -------------------- Transpose --------------------
+
+template<typename ArgType>
+struct unary_evaluator<Transpose<ArgType>, IndexBased>
+  : evaluator_base<Transpose<ArgType> >
+{
+  typedef Transpose<ArgType> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,    
+    Flags = evaluator<ArgType>::Flags ^ RowMajorBit,
+    Alignment = evaluator<ArgType>::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& t) : m_argImpl(t.nestedExpression()) {}
+
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_argImpl.coeff(col, row);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_argImpl.coeff(index);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_argImpl.coeffRef(col, row);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  typename XprType::Scalar& coeffRef(Index index)
+  {
+    return m_argImpl.coeffRef(index);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_argImpl.template packet<LoadMode,PacketType>(col, row);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_argImpl.template packet<LoadMode,PacketType>(index);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    m_argImpl.template writePacket<StoreMode,PacketType>(col, row, x);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    m_argImpl.template writePacket<StoreMode,PacketType>(index, x);
+  }
+
+protected:
+  evaluator<ArgType> m_argImpl;
+};
+
+// -------------------- CwiseNullaryOp --------------------
+// Like Matrix and Array, this is not really a unary expression, so we directly specialize evaluator.
+// Likewise, there is not need to more sophisticated dispatching here.
+
+template<typename Scalar,typename NullaryOp,
+         bool has_nullary = has_nullary_operator<NullaryOp>::value,
+         bool has_unary   = has_unary_operator<NullaryOp>::value,
+         bool has_binary  = has_binary_operator<NullaryOp>::value>
+struct nullary_wrapper
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const { return op(i,j); }
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const { return op(i); }
+
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const { return op.template packetOp<T>(i,j); }
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const { return op.template packetOp<T>(i); }
+};
+
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,true,false,false>
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType=0, IndexType=0) const { return op(); }
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType=0, IndexType=0) const { return op.template packetOp<T>(); }
+};
+
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,false,false,true>
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j=0) const { return op(i,j); }
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j=0) const { return op.template packetOp<T>(i,j); }
+};
+
+// We need the following specialization for vector-only functors assigned to a runtime vector,
+// for instance, using linspace and assigning a RowVectorXd to a MatrixXd or even a row of a MatrixXd.
+// In this case, i==0 and j is used for the actual iteration.
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,false,true,false>
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const {
+    eigen_assert(i==0 || j==0);
+    return op(i+j);
+  }
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const {
+    eigen_assert(i==0 || j==0);
+    return op.template packetOp<T>(i+j);
+  }
+
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const { return op(i); }
+  template <typename T, typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const { return op.template packetOp<T>(i); }
+};
+
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,false,false,false> {};
+
+#if 0 && EIGEN_COMP_MSVC>0
+// Disable this ugly workaround. This is now handled in traits<Ref>::match,
+// but this piece of code might still become handly if some other weird compilation
+// erros pop up again.
+
+// MSVC exhibits a weird compilation error when
+// compiling:
+//    Eigen::MatrixXf A = MatrixXf::Random(3,3);
+//    Ref<const MatrixXf> R = 2.f*A;
+// and that has_*ary_operator<scalar_constant_op<float>> have not been instantiated yet.
+// The "problem" is that evaluator<2.f*A> is instantiated by traits<Ref>::match<2.f*A>
+// and at that time has_*ary_operator<T> returns true regardless of T.
+// Then nullary_wrapper is badly instantiated as nullary_wrapper<.,.,true,true,true>.
+// The trick is thus to defer the proper instantiation of nullary_wrapper when coeff(),
+// and packet() are really instantiated as implemented below:
+
+// This is a simple wrapper around Index to enforce the re-instantiation of
+// has_*ary_operator when needed.
+template<typename T> struct nullary_wrapper_workaround_msvc {
+  nullary_wrapper_workaround_msvc(const T&);
+  operator T()const;
+};
+
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,true,true,true>
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const {
+    return nullary_wrapper<Scalar,NullaryOp,
+    has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().operator()(op,i,j);
+  }
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const {
+    return nullary_wrapper<Scalar,NullaryOp,
+    has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().operator()(op,i);
+  }
+
+  template <typename T, typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const {
+    return nullary_wrapper<Scalar,NullaryOp,
+    has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().template packetOp<T>(op,i,j);
+  }
+  template <typename T, typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const {
+    return nullary_wrapper<Scalar,NullaryOp,
+    has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().template packetOp<T>(op,i);
+  }
+};
+#endif // MSVC workaround
+
+template<typename NullaryOp, typename PlainObjectType>
+struct evaluator<CwiseNullaryOp<NullaryOp,PlainObjectType> >
+  : evaluator_base<CwiseNullaryOp<NullaryOp,PlainObjectType> >
+{
+  typedef CwiseNullaryOp<NullaryOp,PlainObjectType> XprType;
+  typedef typename internal::remove_all<PlainObjectType>::type PlainObjectTypeCleaned;
+  
+  enum {
+    CoeffReadCost = internal::functor_traits<NullaryOp>::Cost,
+    
+    Flags = (evaluator<PlainObjectTypeCleaned>::Flags
+          &  (  HereditaryBits
+              | (functor_has_linear_access<NullaryOp>::ret  ? LinearAccessBit : 0)
+              | (functor_traits<NullaryOp>::PacketAccess    ? PacketAccessBit : 0)))
+          | (functor_traits<NullaryOp>::IsRepeatable ? 0 : EvalBeforeNestingBit),
+    Alignment = AlignedMax
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& n)
+    : m_functor(n.functor()), m_wrapper()
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(IndexType row, IndexType col) const
+  {
+    return m_wrapper(m_functor, row, col);
+  }
+
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(IndexType index) const
+  {
+    return m_wrapper(m_functor,index);
+  }
+
+  template<int LoadMode, typename PacketType, typename IndexType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(IndexType row, IndexType col) const
+  {
+    return m_wrapper.template packetOp<PacketType>(m_functor, row, col);
+  }
+
+  template<int LoadMode, typename PacketType, typename IndexType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(IndexType index) const
+  {
+    return m_wrapper.template packetOp<PacketType>(m_functor, index);
+  }
+
+protected:
+  const NullaryOp m_functor;
+  const internal::nullary_wrapper<CoeffReturnType,NullaryOp> m_wrapper;
+};
+
+// -------------------- CwiseUnaryOp --------------------
+
+template<typename UnaryOp, typename ArgType>
+struct unary_evaluator<CwiseUnaryOp<UnaryOp, ArgType>, IndexBased >
+  : evaluator_base<CwiseUnaryOp<UnaryOp, ArgType> >
+{
+  typedef CwiseUnaryOp<UnaryOp, ArgType> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
+    
+    Flags = evaluator<ArgType>::Flags
+          & (HereditaryBits | LinearAccessBit | (functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0)),
+    Alignment = evaluator<ArgType>::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  explicit unary_evaluator(const XprType& op)
+    : m_functor(op.functor()), 
+      m_argImpl(op.nestedExpression()) 
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_functor(m_argImpl.coeff(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_functor(m_argImpl.coeff(index));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(row, col));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(index));
+  }
+
+protected:
+  const UnaryOp m_functor;
+  evaluator<ArgType> m_argImpl;
+};
+
+// -------------------- CwiseTernaryOp --------------------
+
+// this is a ternary expression
+template<typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
+struct evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >
+  : public ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >
+{
+  typedef CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> XprType;
+  typedef ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> > Base;
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+template<typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
+struct ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3>, IndexBased, IndexBased>
+  : evaluator_base<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >
+{
+  typedef CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<Arg1>::CoeffReadCost + evaluator<Arg2>::CoeffReadCost + evaluator<Arg3>::CoeffReadCost + functor_traits<TernaryOp>::Cost,
+    
+    Arg1Flags = evaluator<Arg1>::Flags,
+    Arg2Flags = evaluator<Arg2>::Flags,
+    Arg3Flags = evaluator<Arg3>::Flags,
+    SameType = is_same<typename Arg1::Scalar,typename Arg2::Scalar>::value && is_same<typename Arg1::Scalar,typename Arg3::Scalar>::value,
+    StorageOrdersAgree = (int(Arg1Flags)&RowMajorBit)==(int(Arg2Flags)&RowMajorBit) && (int(Arg1Flags)&RowMajorBit)==(int(Arg3Flags)&RowMajorBit),
+    Flags0 = (int(Arg1Flags) | int(Arg2Flags) | int(Arg3Flags)) & (
+        HereditaryBits
+        | (int(Arg1Flags) & int(Arg2Flags) & int(Arg3Flags) &
+           ( (StorageOrdersAgree ? LinearAccessBit : 0)
+           | (functor_traits<TernaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
+           )
+        )
+     ),
+    Flags = (Flags0 & ~RowMajorBit) | (Arg1Flags & RowMajorBit),
+    Alignment = EIGEN_PLAIN_ENUM_MIN(
+        EIGEN_PLAIN_ENUM_MIN(evaluator<Arg1>::Alignment, evaluator<Arg2>::Alignment),
+        evaluator<Arg3>::Alignment)
+  };
+
+  EIGEN_DEVICE_FUNC explicit ternary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_arg1Impl(xpr.arg1()), 
+      m_arg2Impl(xpr.arg2()), 
+      m_arg3Impl(xpr.arg3())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<TernaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_functor(m_arg1Impl.coeff(row, col), m_arg2Impl.coeff(row, col), m_arg3Impl.coeff(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_functor(m_arg1Impl.coeff(index), m_arg2Impl.coeff(index), m_arg3Impl.coeff(index));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_functor.packetOp(m_arg1Impl.template packet<LoadMode,PacketType>(row, col),
+                              m_arg2Impl.template packet<LoadMode,PacketType>(row, col),
+                              m_arg3Impl.template packet<LoadMode,PacketType>(row, col));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_functor.packetOp(m_arg1Impl.template packet<LoadMode,PacketType>(index),
+                              m_arg2Impl.template packet<LoadMode,PacketType>(index),
+                              m_arg3Impl.template packet<LoadMode,PacketType>(index));
+  }
+
+protected:
+  const TernaryOp m_functor;
+  evaluator<Arg1> m_arg1Impl;
+  evaluator<Arg2> m_arg2Impl;
+  evaluator<Arg3> m_arg3Impl;
+};
+
+// -------------------- CwiseBinaryOp --------------------
+
+// this is a binary expression
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+  : public binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> > Base;
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IndexBased>
+  : evaluator_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    
+    LhsFlags = evaluator<Lhs>::Flags,
+    RhsFlags = evaluator<Rhs>::Flags,
+    SameType = is_same<typename Lhs::Scalar,typename Rhs::Scalar>::value,
+    StorageOrdersAgree = (int(LhsFlags)&RowMajorBit)==(int(RhsFlags)&RowMajorBit),
+    Flags0 = (int(LhsFlags) | int(RhsFlags)) & (
+        HereditaryBits
+      | (int(LhsFlags) & int(RhsFlags) &
+           ( (StorageOrdersAgree ? LinearAccessBit : 0)
+           | (functor_traits<BinaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
+           )
+        )
+     ),
+    Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit),
+    Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<Lhs>::Alignment,evaluator<Rhs>::Alignment)
+  };
+
+  EIGEN_DEVICE_FUNC explicit binary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_functor(m_lhsImpl.coeff(row, col), m_rhsImpl.coeff(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_functor(m_lhsImpl.coeff(index), m_rhsImpl.coeff(index));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(row, col),
+                              m_rhsImpl.template packet<LoadMode,PacketType>(row, col));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(index),
+                              m_rhsImpl.template packet<LoadMode,PacketType>(index));
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<Lhs> m_lhsImpl;
+  evaluator<Rhs> m_rhsImpl;
+};
+
+// -------------------- CwiseUnaryView --------------------
+
+template<typename UnaryOp, typename ArgType>
+struct unary_evaluator<CwiseUnaryView<UnaryOp, ArgType>, IndexBased>
+  : evaluator_base<CwiseUnaryView<UnaryOp, ArgType> >
+{
+  typedef CwiseUnaryView<UnaryOp, ArgType> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
+    
+    Flags = (evaluator<ArgType>::Flags & (HereditaryBits | LinearAccessBit | DirectAccessBit)),
+    
+    Alignment = 0 // FIXME it is not very clear why alignment is necessarily lost...
+  };
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& op)
+    : m_unaryOp(op.functor()), 
+      m_argImpl(op.nestedExpression()) 
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_unaryOp(m_argImpl.coeff(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_unaryOp(m_argImpl.coeff(index));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_unaryOp(m_argImpl.coeffRef(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_unaryOp(m_argImpl.coeffRef(index));
+  }
+
+protected:
+  const UnaryOp m_unaryOp;
+  evaluator<ArgType> m_argImpl;
+};
+
+// -------------------- Map --------------------
+
+// FIXME perhaps the PlainObjectType could be provided by Derived::PlainObject ?
+// but that might complicate template specialization
+template<typename Derived, typename PlainObjectType>
+struct mapbase_evaluator;
+
+template<typename Derived, typename PlainObjectType>
+struct mapbase_evaluator : evaluator_base<Derived>
+{
+  typedef Derived  XprType;
+  typedef typename XprType::PointerType PointerType;
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  
+  enum {
+    IsRowMajor = XprType::RowsAtCompileTime,
+    ColsAtCompileTime = XprType::ColsAtCompileTime,
+    CoeffReadCost = NumTraits<Scalar>::ReadCost
+  };
+
+  EIGEN_DEVICE_FUNC explicit mapbase_evaluator(const XprType& map)
+    : m_data(const_cast<PointerType>(map.data())),
+      m_innerStride(map.innerStride()),
+      m_outerStride(map.outerStride())
+  {
+    EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(evaluator<Derived>::Flags&PacketAccessBit, internal::inner_stride_at_compile_time<Derived>::ret==1),
+                        PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_data[col * colStride() + row * rowStride()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_data[index * m_innerStride.value()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_data[col * colStride() + row * rowStride()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_data[index * m_innerStride.value()];
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    PointerType ptr = m_data + row * rowStride() + col * colStride();
+    return internal::ploadt<PacketType, LoadMode>(ptr);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return internal::ploadt<PacketType, LoadMode>(m_data + index * m_innerStride.value());
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    PointerType ptr = m_data + row * rowStride() + col * colStride();
+    return internal::pstoret<Scalar, PacketType, StoreMode>(ptr, x);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    internal::pstoret<Scalar, PacketType, StoreMode>(m_data + index * m_innerStride.value(), x);
+  }
+protected:
+  EIGEN_DEVICE_FUNC
+  inline Index rowStride() const { return XprType::IsRowMajor ? m_outerStride.value() : m_innerStride.value(); }
+  EIGEN_DEVICE_FUNC
+  inline Index colStride() const { return XprType::IsRowMajor ? m_innerStride.value() : m_outerStride.value(); }
+
+  PointerType m_data;
+  const internal::variable_if_dynamic<Index, XprType::InnerStrideAtCompileTime> m_innerStride;
+  const internal::variable_if_dynamic<Index, XprType::OuterStrideAtCompileTime> m_outerStride;
+};
+
+template<typename PlainObjectType, int MapOptions, typename StrideType> 
+struct evaluator<Map<PlainObjectType, MapOptions, StrideType> >
+  : public mapbase_evaluator<Map<PlainObjectType, MapOptions, StrideType>, PlainObjectType>
+{
+  typedef Map<PlainObjectType, MapOptions, StrideType> XprType;
+  typedef typename XprType::Scalar Scalar;
+  // TODO: should check for smaller packet types once we can handle multi-sized packet types
+  typedef typename packet_traits<Scalar>::type PacketScalar;
+  
+  enum {
+    InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
+                             ? int(PlainObjectType::InnerStrideAtCompileTime)
+                             : int(StrideType::InnerStrideAtCompileTime),
+    OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
+                             ? int(PlainObjectType::OuterStrideAtCompileTime)
+                             : int(StrideType::OuterStrideAtCompileTime),
+    HasNoInnerStride = InnerStrideAtCompileTime == 1,
+    HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0,
+    HasNoStride = HasNoInnerStride && HasNoOuterStride,
+    IsDynamicSize = PlainObjectType::SizeAtCompileTime==Dynamic,
+    
+    PacketAccessMask = bool(HasNoInnerStride) ? ~int(0) : ~int(PacketAccessBit),
+    LinearAccessMask = bool(HasNoStride) || bool(PlainObjectType::IsVectorAtCompileTime) ? ~int(0) : ~int(LinearAccessBit),
+    Flags = int( evaluator<PlainObjectType>::Flags) & (LinearAccessMask&PacketAccessMask),
+    
+    Alignment = int(MapOptions)&int(AlignedMask)
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& map)
+    : mapbase_evaluator<XprType, PlainObjectType>(map) 
+  { }
+};
+
+// -------------------- Ref --------------------
+
+template<typename PlainObjectType, int RefOptions, typename StrideType> 
+struct evaluator<Ref<PlainObjectType, RefOptions, StrideType> >
+  : public mapbase_evaluator<Ref<PlainObjectType, RefOptions, StrideType>, PlainObjectType>
+{
+  typedef Ref<PlainObjectType, RefOptions, StrideType> XprType;
+  
+  enum {
+    Flags = evaluator<Map<PlainObjectType, RefOptions, StrideType> >::Flags,
+    Alignment = evaluator<Map<PlainObjectType, RefOptions, StrideType> >::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& ref)
+    : mapbase_evaluator<XprType, PlainObjectType>(ref) 
+  { }
+};
+
+// -------------------- Block --------------------
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel,
+         bool HasDirectAccess = internal::has_direct_access<ArgType>::ret> struct block_evaluator;
+         
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel> 
+struct evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
+  : block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel>
+{
+  typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
+  typedef typename XprType::Scalar Scalar;
+  // TODO: should check for smaller packet types once we can handle multi-sized packet types
+  typedef typename packet_traits<Scalar>::type PacketScalar;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    
+    RowsAtCompileTime = traits<XprType>::RowsAtCompileTime,
+    ColsAtCompileTime = traits<XprType>::ColsAtCompileTime,
+    MaxRowsAtCompileTime = traits<XprType>::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = traits<XprType>::MaxColsAtCompileTime,
+    
+    ArgTypeIsRowMajor = (int(evaluator<ArgType>::Flags)&RowMajorBit) != 0,
+    IsRowMajor = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? 1
+               : (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0
+               : ArgTypeIsRowMajor,
+    HasSameStorageOrderAsArgType = (IsRowMajor == ArgTypeIsRowMajor),
+    InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
+    InnerStrideAtCompileTime = HasSameStorageOrderAsArgType
+                             ? int(inner_stride_at_compile_time<ArgType>::ret)
+                             : int(outer_stride_at_compile_time<ArgType>::ret),
+    OuterStrideAtCompileTime = HasSameStorageOrderAsArgType
+                             ? int(outer_stride_at_compile_time<ArgType>::ret)
+                             : int(inner_stride_at_compile_time<ArgType>::ret),
+    MaskPacketAccessBit = (InnerStrideAtCompileTime == 1 || HasSameStorageOrderAsArgType) ? PacketAccessBit : 0,
+    
+    FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (evaluator<ArgType>::Flags&LinearAccessBit))) ? LinearAccessBit : 0,    
+    FlagsRowMajorBit = XprType::Flags&RowMajorBit,
+    Flags0 = evaluator<ArgType>::Flags & ( (HereditaryBits & ~RowMajorBit) |
+                                           DirectAccessBit |
+                                           MaskPacketAccessBit),
+    Flags = Flags0 | FlagsLinearAccessBit | FlagsRowMajorBit,
+    
+    PacketAlignment = unpacket_traits<PacketScalar>::alignment,
+    Alignment0 = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic)
+                             && (OuterStrideAtCompileTime!=0)
+                             && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % int(PacketAlignment)) == 0)) ? int(PacketAlignment) : 0,
+    Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ArgType>::Alignment, Alignment0)
+  };
+  typedef block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel> block_evaluator_type;
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& block) : block_evaluator_type(block)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+};
+
+// no direct-access => dispatch to a unary evaluator
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+struct block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel, /*HasDirectAccess*/ false>
+  : unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
+{
+  typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
+
+  EIGEN_DEVICE_FUNC explicit block_evaluator(const XprType& block)
+    : unary_evaluator<XprType>(block) 
+  {}
+};
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBased>
+  : evaluator_base<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
+{
+  typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& block)
+    : m_argImpl(block.nestedExpression()), 
+      m_startRow(block.startRow()), 
+      m_startCol(block.startCol()),
+      m_linear_offset(InnerPanel?(XprType::IsRowMajor ? block.startRow()*block.cols() : block.startCol()*block.rows()):0)
+  { }
+ 
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  enum {
+    RowsAtCompileTime = XprType::RowsAtCompileTime,
+    ForwardLinearAccess = InnerPanel && bool(evaluator<ArgType>::Flags&LinearAccessBit)
+  };
+ 
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  { 
+    return m_argImpl.coeff(m_startRow.value() + row, m_startCol.value() + col); 
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  { 
+    if (ForwardLinearAccess)
+      return m_argImpl.coeff(m_linear_offset.value() + index); 
+    else
+      return coeff(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  { 
+    return m_argImpl.coeffRef(m_startRow.value() + row, m_startCol.value() + col); 
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  { 
+    if (ForwardLinearAccess)
+      return m_argImpl.coeffRef(m_linear_offset.value() + index); 
+    else
+      return coeffRef(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
+  }
+ 
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const 
+  { 
+    return m_argImpl.template packet<LoadMode,PacketType>(m_startRow.value() + row, m_startCol.value() + col); 
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const 
+  { 
+    if (ForwardLinearAccess)
+      return m_argImpl.template packet<LoadMode,PacketType>(m_linear_offset.value() + index);
+    else
+      return packet<LoadMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
+                                         RowsAtCompileTime == 1 ? index : 0);
+  }
+  
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x) 
+  {
+    return m_argImpl.template writePacket<StoreMode,PacketType>(m_startRow.value() + row, m_startCol.value() + col, x); 
+  }
+  
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x) 
+  {
+    if (ForwardLinearAccess)
+      return m_argImpl.template writePacket<StoreMode,PacketType>(m_linear_offset.value() + index, x);
+    else
+      return writePacket<StoreMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
+                                              RowsAtCompileTime == 1 ? index : 0,
+                                              x);
+  }
+ 
+protected:
+  evaluator<ArgType> m_argImpl;
+  const variable_if_dynamic<Index, (ArgType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
+  const variable_if_dynamic<Index, (ArgType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
+  const variable_if_dynamic<Index, InnerPanel ? Dynamic : 0> m_linear_offset;
+};
+
+// TODO: This evaluator does not actually use the child evaluator; 
+// all action is via the data() as returned by the Block expression.
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel> 
+struct block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel, /* HasDirectAccess */ true>
+  : mapbase_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>,
+                      typename Block<ArgType, BlockRows, BlockCols, InnerPanel>::PlainObject>
+{
+  typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
+  typedef typename XprType::Scalar Scalar;
+
+  EIGEN_DEVICE_FUNC explicit block_evaluator(const XprType& block)
+    : mapbase_evaluator<XprType, typename XprType::PlainObject>(block) 
+  {
+    // TODO: for the 3.3 release, this should be turned to an internal assertion, but let's keep it as is for the beta lifetime
+    eigen_assert(((internal::UIntPtr(block.data()) % EIGEN_PLAIN_ENUM_MAX(1,evaluator<XprType>::Alignment)) == 0) && "data is not aligned");
+  }
+};
+
+
+// -------------------- Select --------------------
+// NOTE shall we introduce a ternary_evaluator?
+
+// TODO enable vectorization for Select
+template<typename ConditionMatrixType, typename ThenMatrixType, typename ElseMatrixType>
+struct evaluator<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
+  : evaluator_base<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
+{
+  typedef Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> XprType;
+  enum {
+    CoeffReadCost = evaluator<ConditionMatrixType>::CoeffReadCost
+                  + EIGEN_PLAIN_ENUM_MAX(evaluator<ThenMatrixType>::CoeffReadCost,
+                                         evaluator<ElseMatrixType>::CoeffReadCost),
+
+    Flags = (unsigned int)evaluator<ThenMatrixType>::Flags & evaluator<ElseMatrixType>::Flags & HereditaryBits,
+    
+    Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ThenMatrixType>::Alignment, evaluator<ElseMatrixType>::Alignment)
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& select)
+    : m_conditionImpl(select.conditionMatrix()),
+      m_thenImpl(select.thenMatrix()),
+      m_elseImpl(select.elseMatrix())
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+ 
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    if (m_conditionImpl.coeff(row, col))
+      return m_thenImpl.coeff(row, col);
+    else
+      return m_elseImpl.coeff(row, col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    if (m_conditionImpl.coeff(index))
+      return m_thenImpl.coeff(index);
+    else
+      return m_elseImpl.coeff(index);
+  }
+ 
+protected:
+  evaluator<ConditionMatrixType> m_conditionImpl;
+  evaluator<ThenMatrixType> m_thenImpl;
+  evaluator<ElseMatrixType> m_elseImpl;
+};
+
+
+// -------------------- Replicate --------------------
+
+template<typename ArgType, int RowFactor, int ColFactor> 
+struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
+  : evaluator_base<Replicate<ArgType, RowFactor, ColFactor> >
+{
+  typedef Replicate<ArgType, RowFactor, ColFactor> XprType;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  enum {
+    Factor = (RowFactor==Dynamic || ColFactor==Dynamic) ? Dynamic : RowFactor*ColFactor
+  };
+  typedef typename internal::nested_eval<ArgType,Factor>::type ArgTypeNested;
+  typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgTypeNestedCleaned>::CoeffReadCost,
+    LinearAccessMask = XprType::IsVectorAtCompileTime ? LinearAccessBit : 0,
+    Flags = (evaluator<ArgTypeNestedCleaned>::Flags & (HereditaryBits|LinearAccessMask) & ~RowMajorBit) | (traits<XprType>::Flags & RowMajorBit),
+    
+    Alignment = evaluator<ArgTypeNestedCleaned>::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& replicate)
+    : m_arg(replicate.nestedExpression()),
+      m_argImpl(m_arg),
+      m_rows(replicate.nestedExpression().rows()),
+      m_cols(replicate.nestedExpression().cols())
+  {}
+ 
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    // try to avoid using modulo; this is a pure optimization strategy
+    const Index actual_row = internal::traits<XprType>::RowsAtCompileTime==1 ? 0
+                           : RowFactor==1 ? row
+                           : row % m_rows.value();
+    const Index actual_col = internal::traits<XprType>::ColsAtCompileTime==1 ? 0
+                           : ColFactor==1 ? col
+                           : col % m_cols.value();
+    
+    return m_argImpl.coeff(actual_row, actual_col);
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    // try to avoid using modulo; this is a pure optimization strategy
+    const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
+                                  ? (ColFactor==1 ?  index : index%m_cols.value())
+                                  : (RowFactor==1 ?  index : index%m_rows.value());
+    
+    return m_argImpl.coeff(actual_index);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    const Index actual_row = internal::traits<XprType>::RowsAtCompileTime==1 ? 0
+                           : RowFactor==1 ? row
+                           : row % m_rows.value();
+    const Index actual_col = internal::traits<XprType>::ColsAtCompileTime==1 ? 0
+                           : ColFactor==1 ? col
+                           : col % m_cols.value();
+
+    return m_argImpl.template packet<LoadMode,PacketType>(actual_row, actual_col);
+  }
+  
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
+                                  ? (ColFactor==1 ?  index : index%m_cols.value())
+                                  : (RowFactor==1 ?  index : index%m_rows.value());
+
+    return m_argImpl.template packet<LoadMode,PacketType>(actual_index);
+  }
+ 
+protected:
+  const ArgTypeNested m_arg;
+  evaluator<ArgTypeNestedCleaned> m_argImpl;
+  const variable_if_dynamic<Index, ArgType::RowsAtCompileTime> m_rows;
+  const variable_if_dynamic<Index, ArgType::ColsAtCompileTime> m_cols;
+};
+
+
+// -------------------- PartialReduxExpr --------------------
+
+template< typename ArgType, typename MemberOp, int Direction>
+struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
+  : evaluator_base<PartialReduxExpr<ArgType, MemberOp, Direction> >
+{
+  typedef PartialReduxExpr<ArgType, MemberOp, Direction> XprType;
+  typedef typename internal::nested_eval<ArgType,1>::type ArgTypeNested;
+  typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
+  typedef typename ArgType::Scalar InputScalar;
+  typedef typename XprType::Scalar Scalar;
+  enum {
+    TraversalSize = Direction==int(Vertical) ? int(ArgType::RowsAtCompileTime) :  int(ArgType::ColsAtCompileTime)
+  };
+  typedef typename MemberOp::template Cost<InputScalar,int(TraversalSize)> CostOpType;
+  enum {
+    CoeffReadCost = TraversalSize==Dynamic ? HugeCost
+                  : TraversalSize * evaluator<ArgType>::CoeffReadCost + int(CostOpType::value),
+    
+    Flags = (traits<XprType>::Flags&RowMajorBit) | (evaluator<ArgType>::Flags&(HereditaryBits&(~RowMajorBit))) | LinearAccessBit,
+    
+    Alignment = 0 // FIXME this will need to be improved once PartialReduxExpr is vectorized
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType xpr)
+    : m_arg(xpr.nestedExpression()), m_functor(xpr.functor())
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(TraversalSize==Dynamic ? HugeCost : int(CostOpType::value));
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar coeff(Index i, Index j) const
+  {
+    if (Direction==Vertical)
+      return m_functor(m_arg.col(j));
+    else
+      return m_functor(m_arg.row(i));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar coeff(Index index) const
+  {
+    if (Direction==Vertical)
+      return m_functor(m_arg.col(index));
+    else
+      return m_functor(m_arg.row(index));
+  }
+
+protected:
+  typename internal::add_const_on_value_type<ArgTypeNested>::type m_arg;
+  const MemberOp m_functor;
+};
+
+
+// -------------------- MatrixWrapper and ArrayWrapper --------------------
+//
+// evaluator_wrapper_base<T> is a common base class for the
+// MatrixWrapper and ArrayWrapper evaluators.
+
+template<typename XprType>
+struct evaluator_wrapper_base
+  : evaluator_base<XprType>
+{
+  typedef typename remove_all<typename XprType::NestedExpressionType>::type ArgType;
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    Flags = evaluator<ArgType>::Flags,
+    Alignment = evaluator<ArgType>::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator_wrapper_base(const ArgType& arg) : m_argImpl(arg) {}
+
+  typedef typename ArgType::Scalar Scalar;
+  typedef typename ArgType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_argImpl.coeff(row, col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_argImpl.coeff(index);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_argImpl.coeffRef(row, col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_argImpl.coeffRef(index);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_argImpl.template packet<LoadMode,PacketType>(row, col);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_argImpl.template packet<LoadMode,PacketType>(index);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    m_argImpl.template writePacket<StoreMode>(row, col, x);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    m_argImpl.template writePacket<StoreMode>(index, x);
+  }
+
+protected:
+  evaluator<ArgType> m_argImpl;
+};
+
+template<typename TArgType>
+struct unary_evaluator<MatrixWrapper<TArgType> >
+  : evaluator_wrapper_base<MatrixWrapper<TArgType> >
+{
+  typedef MatrixWrapper<TArgType> XprType;
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& wrapper)
+    : evaluator_wrapper_base<MatrixWrapper<TArgType> >(wrapper.nestedExpression())
+  { }
+};
+
+template<typename TArgType>
+struct unary_evaluator<ArrayWrapper<TArgType> >
+  : evaluator_wrapper_base<ArrayWrapper<TArgType> >
+{
+  typedef ArrayWrapper<TArgType> XprType;
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& wrapper)
+    : evaluator_wrapper_base<ArrayWrapper<TArgType> >(wrapper.nestedExpression())
+  { }
+};
+
+
+// -------------------- Reverse --------------------
+
+// defined in Reverse.h:
+template<typename PacketType, bool ReversePacket> struct reverse_packet_cond;
+
+template<typename ArgType, int Direction>
+struct unary_evaluator<Reverse<ArgType, Direction> >
+  : evaluator_base<Reverse<ArgType, Direction> >
+{
+  typedef Reverse<ArgType, Direction> XprType;
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  enum {
+    IsRowMajor = XprType::IsRowMajor,
+    IsColMajor = !IsRowMajor,
+    ReverseRow = (Direction == Vertical)   || (Direction == BothDirections),
+    ReverseCol = (Direction == Horizontal) || (Direction == BothDirections),
+    ReversePacket = (Direction == BothDirections)
+                    || ((Direction == Vertical)   && IsColMajor)
+                    || ((Direction == Horizontal) && IsRowMajor),
+                    
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    
+    // let's enable LinearAccess only with vectorization because of the product overhead
+    // FIXME enable DirectAccess with negative strides?
+    Flags0 = evaluator<ArgType>::Flags,
+    LinearAccess = ( (Direction==BothDirections) && (int(Flags0)&PacketAccessBit) )
+                  || ((ReverseRow && XprType::ColsAtCompileTime==1) || (ReverseCol && XprType::RowsAtCompileTime==1))
+                 ? LinearAccessBit : 0,
+
+    Flags = int(Flags0) & (HereditaryBits | PacketAccessBit | LinearAccess),
+    
+    Alignment = 0 // FIXME in some rare cases, Alignment could be preserved, like a Vector4f.
+  };
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& reverse)
+    : m_argImpl(reverse.nestedExpression()),
+      m_rows(ReverseRow ? reverse.nestedExpression().rows() : 1),
+      m_cols(ReverseCol ? reverse.nestedExpression().cols() : 1)
+  { }
+ 
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_argImpl.coeff(ReverseRow ? m_rows.value() - row - 1 : row,
+                           ReverseCol ? m_cols.value() - col - 1 : col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_argImpl.coeff(m_rows.value() * m_cols.value() - index - 1);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_argImpl.coeffRef(ReverseRow ? m_rows.value() - row - 1 : row,
+                              ReverseCol ? m_cols.value() - col - 1 : col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_argImpl.coeffRef(m_rows.value() * m_cols.value() - index - 1);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    enum {
+      PacketSize = unpacket_traits<PacketType>::size,
+      OffsetRow  = ReverseRow && IsColMajor ? PacketSize : 1,
+      OffsetCol  = ReverseCol && IsRowMajor ? PacketSize : 1
+    };
+    typedef internal::reverse_packet_cond<PacketType,ReversePacket> reverse_packet;
+    return reverse_packet::run(m_argImpl.template packet<LoadMode,PacketType>(
+                                  ReverseRow ? m_rows.value() - row - OffsetRow : row,
+                                  ReverseCol ? m_cols.value() - col - OffsetCol : col));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    enum { PacketSize = unpacket_traits<PacketType>::size };
+    return preverse(m_argImpl.template packet<LoadMode,PacketType>(m_rows.value() * m_cols.value() - index - PacketSize));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    // FIXME we could factorize some code with packet(i,j)
+    enum {
+      PacketSize = unpacket_traits<PacketType>::size,
+      OffsetRow  = ReverseRow && IsColMajor ? PacketSize : 1,
+      OffsetCol  = ReverseCol && IsRowMajor ? PacketSize : 1
+    };
+    typedef internal::reverse_packet_cond<PacketType,ReversePacket> reverse_packet;
+    m_argImpl.template writePacket<LoadMode>(
+                                  ReverseRow ? m_rows.value() - row - OffsetRow : row,
+                                  ReverseCol ? m_cols.value() - col - OffsetCol : col,
+                                  reverse_packet::run(x));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    enum { PacketSize = unpacket_traits<PacketType>::size };
+    m_argImpl.template writePacket<LoadMode>
+      (m_rows.value() * m_cols.value() - index - PacketSize, preverse(x));
+  }
+ 
+protected:
+  evaluator<ArgType> m_argImpl;
+
+  // If we do not reverse rows, then we do not need to know the number of rows; same for columns
+  // Nonetheless, in this case it is important to set to 1 such that the coeff(index) method works fine for vectors.
+  const variable_if_dynamic<Index, ReverseRow ? ArgType::RowsAtCompileTime : 1> m_rows;
+  const variable_if_dynamic<Index, ReverseCol ? ArgType::ColsAtCompileTime : 1> m_cols;
+};
+
+
+// -------------------- Diagonal --------------------
+
+template<typename ArgType, int DiagIndex>
+struct evaluator<Diagonal<ArgType, DiagIndex> >
+  : evaluator_base<Diagonal<ArgType, DiagIndex> >
+{
+  typedef Diagonal<ArgType, DiagIndex> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    
+    Flags = (unsigned int)(evaluator<ArgType>::Flags & (HereditaryBits | DirectAccessBit) & ~RowMajorBit) | LinearAccessBit,
+    
+    Alignment = 0
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& diagonal)
+    : m_argImpl(diagonal.nestedExpression()),
+      m_index(diagonal.index())
+  { }
+ 
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index) const
+  {
+    return m_argImpl.coeff(row + rowOffset(), row + colOffset());
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_argImpl.coeff(index + rowOffset(), index + colOffset());
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index)
+  {
+    return m_argImpl.coeffRef(row + rowOffset(), row + colOffset());
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_argImpl.coeffRef(index + rowOffset(), index + colOffset());
+  }
+
+protected:
+  evaluator<ArgType> m_argImpl;
+  const internal::variable_if_dynamicindex<Index, XprType::DiagIndex> m_index;
+
+private:
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value() > 0 ? 0 : -m_index.value(); }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value() > 0 ? m_index.value() : 0; }
+};
+
+
+//----------------------------------------------------------------------
+// deprecated code
+//----------------------------------------------------------------------
+
+// -------------------- EvalToTemp --------------------
+
+// expression class for evaluating nested expression to a temporary
+
+template<typename ArgType> class EvalToTemp;
+
+template<typename ArgType>
+struct traits<EvalToTemp<ArgType> >
+  : public traits<ArgType>
+{ };
+
+template<typename ArgType>
+class EvalToTemp
+  : public dense_xpr_base<EvalToTemp<ArgType> >::type
+{
+ public:
+ 
+  typedef typename dense_xpr_base<EvalToTemp>::type Base;
+  EIGEN_GENERIC_PUBLIC_INTERFACE(EvalToTemp)
+ 
+  explicit EvalToTemp(const ArgType& arg)
+    : m_arg(arg)
+  { }
+ 
+  const ArgType& arg() const
+  {
+    return m_arg;
+  }
+
+  Index rows() const 
+  {
+    return m_arg.rows();
+  }
+
+  Index cols() const 
+  {
+    return m_arg.cols();
+  }
+
+ private:
+  const ArgType& m_arg;
+};
+ 
+template<typename ArgType>
+struct evaluator<EvalToTemp<ArgType> >
+  : public evaluator<typename ArgType::PlainObject>
+{
+  typedef EvalToTemp<ArgType>                   XprType;
+  typedef typename ArgType::PlainObject         PlainObject;
+  typedef evaluator<PlainObject> Base;
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr)
+    : m_result(xpr.arg())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+  }
+
+  // This constructor is used when nesting an EvalTo evaluator in another evaluator
+  EIGEN_DEVICE_FUNC evaluator(const ArgType& arg)
+    : m_result(arg)
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+  }
+
+protected:
+  PlainObject m_result;
+};
+
+} // namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_COREEVALUATORS_H
diff --git a/extern/Eigen3/Eigen/src/Core/CoreIterators.h b/extern/Eigen3/Eigen/src/Core/CoreIterators.h
index 6da4683d2c2..4eb42b93af1 100644
--- a/extern/Eigen3/Eigen/src/Core/CoreIterators.h
+++ b/extern/Eigen3/Eigen/src/Core/CoreIterators.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -15,47 +15,113 @@ namespace Eigen {
 /* This file contains the respective InnerIterator definition of the expressions defined in Eigen/Core
  */
 
-/** \ingroup SparseCore_Module
-  * \class InnerIterator
-  * \brief An InnerIterator allows to loop over the element of a sparse (or dense) matrix or expression
-  *
-  * todo
+namespace internal {
+
+template<typename XprType, typename EvaluatorKind>
+class inner_iterator_selector;
+
+}
+
+/** \class InnerIterator
+  * \brief An InnerIterator allows to loop over the element of any matrix expression.
+  * 
+  * \warning To be used with care because an evaluator is constructed every time an InnerIterator iterator is constructed.
+  * 
+  * TODO: add a usage example
   */
+template<typename XprType>
+class InnerIterator
+{
+protected:
+  typedef internal::inner_iterator_selector<XprType, typename internal::evaluator_traits<XprType>::Kind> IteratorType;
+  typedef internal::evaluator<XprType> EvaluatorType;
+  typedef typename internal::traits<XprType>::Scalar Scalar;
+public:
+  /** Construct an iterator over the \a outerId -th row or column of \a xpr */
+  InnerIterator(const XprType &xpr, const Index &outerId)
+    : m_eval(xpr), m_iter(m_eval, outerId, xpr.innerSize())
+  {}
+  
+  /// \returns the value of the current coefficient.
+  EIGEN_STRONG_INLINE Scalar value() const          { return m_iter.value(); }
+  /** Increment the iterator \c *this to the next non-zero coefficient.
+    * Explicit zeros are not skipped over. To skip explicit zeros, see class SparseView
+    */
+  EIGEN_STRONG_INLINE InnerIterator& operator++()   { m_iter.operator++(); return *this; }
+  /// \returns the column or row index of the current coefficient.
+  EIGEN_STRONG_INLINE Index index() const           { return m_iter.index(); }
+  /// \returns the row index of the current coefficient.
+  EIGEN_STRONG_INLINE Index row() const             { return m_iter.row(); }
+  /// \returns the column index of the current coefficient.
+  EIGEN_STRONG_INLINE Index col() const             { return m_iter.col(); }
+  /// \returns \c true if the iterator \c *this still references a valid coefficient.
+  EIGEN_STRONG_INLINE operator bool() const         { return m_iter; }
+  
+protected:
+  EvaluatorType m_eval;
+  IteratorType m_iter;
+private:
+  // If you get here, then you're not using the right InnerIterator type, e.g.:
+  //   SparseMatrix<double,RowMajor> A;
+  //   SparseMatrix<double>::InnerIterator it(A,0);
+  template<typename T> InnerIterator(const EigenBase<T>&,Index outer);
+};
+
+namespace internal {
 
-// generic version for dense matrix and expressions
-template<typename Derived> class DenseBase<Derived>::InnerIterator
+// Generic inner iterator implementation for dense objects
+template<typename XprType>
+class inner_iterator_selector<XprType, IndexBased>
 {
-  protected:
-    typedef typename Derived::Scalar Scalar;
-    typedef typename Derived::Index Index;
-
-    enum { IsRowMajor = (Derived::Flags&RowMajorBit)==RowMajorBit };
-  public:
-    EIGEN_STRONG_INLINE InnerIterator(const Derived& expr, Index outer)
-      : m_expression(expr), m_inner(0), m_outer(outer), m_end(expr.innerSize())
-    {}
-
-    EIGEN_STRONG_INLINE Scalar value() const
-    {
-      return (IsRowMajor) ? m_expression.coeff(m_outer, m_inner)
-                          : m_expression.coeff(m_inner, m_outer);
-    }
-
-    EIGEN_STRONG_INLINE InnerIterator& operator++() { m_inner++; return *this; }
-
-    EIGEN_STRONG_INLINE Index index() const { return m_inner; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; }
-
-  protected:
-    const Derived& m_expression;
-    Index m_inner;
-    const Index m_outer;
-    const Index m_end;
+protected:
+  typedef evaluator<XprType> EvaluatorType;
+  typedef typename traits<XprType>::Scalar Scalar;
+  enum { IsRowMajor = (XprType::Flags&RowMajorBit)==RowMajorBit };
+  
+public:
+  EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &innerSize)
+    : m_eval(eval), m_inner(0), m_outer(outerId), m_end(innerSize)
+  {}
+
+  EIGEN_STRONG_INLINE Scalar value() const
+  {
+    return (IsRowMajor) ? m_eval.coeff(m_outer, m_inner)
+                        : m_eval.coeff(m_inner, m_outer);
+  }
+
+  EIGEN_STRONG_INLINE inner_iterator_selector& operator++() { m_inner++; return *this; }
+
+  EIGEN_STRONG_INLINE Index index() const { return m_inner; }
+  inline Index row() const { return IsRowMajor ? m_outer : index(); }
+  inline Index col() const { return IsRowMajor ? index() : m_outer; }
+
+  EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; }
+
+protected:
+  const EvaluatorType& m_eval;
+  Index m_inner;
+  const Index m_outer;
+  const Index m_end;
 };
 
+// For iterator-based evaluator, inner-iterator is already implemented as
+// evaluator<>::InnerIterator
+template<typename XprType>
+class inner_iterator_selector<XprType, IteratorBased>
+ : public evaluator<XprType>::InnerIterator
+{
+protected:
+  typedef typename evaluator<XprType>::InnerIterator Base;
+  typedef evaluator<XprType> EvaluatorType;
+  
+public:
+  EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &/*innerSize*/)
+    : Base(eval, outerId)
+  {}  
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_COREITERATORS_H
diff --git a/extern/Eigen3/Eigen/src/Core/CwiseBinaryOp.h b/extern/Eigen3/Eigen/src/Core/CwiseBinaryOp.h
index 519a866e605..a36765e396b 100644
--- a/extern/Eigen3/Eigen/src/Core/CwiseBinaryOp.h
+++ b/extern/Eigen3/Eigen/src/Core/CwiseBinaryOp.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -13,26 +13,6 @@
 
 namespace Eigen {
 
-/** \class CwiseBinaryOp
-  * \ingroup Core_Module
-  *
-  * \brief Generic expression where a coefficient-wise binary operator is applied to two expressions
-  *
-  * \param BinaryOp template functor implementing the operator
-  * \param Lhs the type of the left-hand side
-  * \param Rhs the type of the right-hand side
-  *
-  * This class represents an expression  where a coefficient-wise binary operator is applied to two expressions.
-  * It is the return type of binary operators, by which we mean only those binary operators where
-  * both the left-hand side and the right-hand side are Eigen expressions.
-  * For example, the return type of matrix1+matrix2 is a CwiseBinaryOp.
-  *
-  * Most of the time, this is the only way that it is used, so you typically don't have to name
-  * CwiseBinaryOp types explicitly.
-  *
-  * \sa MatrixBase::binaryExpr(const MatrixBase<OtherDerived> &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp
-  */
-
 namespace internal {
 template<typename BinaryOp, typename Lhs, typename Rhs>
 struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
@@ -52,77 +32,75 @@ struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
   // we still want to handle the case when the result type is different.
   typedef typename result_of<
                      BinaryOp(
-                       typename Lhs::Scalar,
-                       typename Rhs::Scalar
+                       const typename Lhs::Scalar&,
+                       const typename Rhs::Scalar&
                      )
                    >::type Scalar;
-  typedef typename promote_storage_type<typename traits<Lhs>::StorageKind,
-                                           typename traits<Rhs>::StorageKind>::ret StorageKind;
-  typedef typename promote_index_type<typename traits<Lhs>::Index,
-                                         typename traits<Rhs>::Index>::type Index;
+  typedef typename cwise_promote_storage_type<typename traits<Lhs>::StorageKind,
+                                              typename traits<Rhs>::StorageKind,
+                                              BinaryOp>::ret StorageKind;
+  typedef typename promote_index_type<typename traits<Lhs>::StorageIndex,
+                                      typename traits<Rhs>::StorageIndex>::type StorageIndex;
   typedef typename Lhs::Nested LhsNested;
   typedef typename Rhs::Nested RhsNested;
   typedef typename remove_reference<LhsNested>::type _LhsNested;
   typedef typename remove_reference<RhsNested>::type _RhsNested;
   enum {
-    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
-    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
-    LhsFlags = _LhsNested::Flags,
-    RhsFlags = _RhsNested::Flags,
-    SameType = is_same<typename _LhsNested::Scalar,typename _RhsNested::Scalar>::value,
-    StorageOrdersAgree = (int(Lhs::Flags)&RowMajorBit)==(int(Rhs::Flags)&RowMajorBit),
-    Flags0 = (int(LhsFlags) | int(RhsFlags)) & (
-        HereditaryBits
-      | (int(LhsFlags) & int(RhsFlags) &
-           ( AlignedBit
-           | (StorageOrdersAgree ? LinearAccessBit : 0)
-           | (functor_traits<BinaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
-           )
-        )
-     ),
-    Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit),
-    Cost0 = EIGEN_ADD_COST(LhsCoeffReadCost,RhsCoeffReadCost),
-    CoeffReadCost = EIGEN_ADD_COST(Cost0,functor_traits<BinaryOp>::Cost)
+    Flags = cwise_promote_storage_order<typename traits<Lhs>::StorageKind,typename traits<Rhs>::StorageKind,_LhsNested::Flags & RowMajorBit,_RhsNested::Flags & RowMajorBit>::value
   };
 };
 } // end namespace internal
 
-// we require Lhs and Rhs to have the same scalar type. Currently there is no example of a binary functor
-// that would take two operands of different types. If there were such an example, then this check should be
-// moved to the BinaryOp functors, on a per-case basis. This would however require a change in the BinaryOp functors, as
-// currently they take only one typename Scalar template parameter.
-// It is tempting to always allow mixing different types but remember that this is often impossible in the vectorized paths.
-// So allowing mixing different types gives very unexpected errors when enabling vectorization, when the user tries to
-// add together a float matrix and a double matrix.
-#define EIGEN_CHECK_BINARY_COMPATIBILIY(BINOP,LHS,RHS) \
-  EIGEN_STATIC_ASSERT((internal::functor_is_product_like<BINOP>::ret \
-                        ? int(internal::scalar_product_traits<LHS, RHS>::Defined) \
-                        : int(internal::is_same<LHS, RHS>::value)), \
-    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
 template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
 class CwiseBinaryOpImpl;
 
-template<typename BinaryOp, typename Lhs, typename Rhs>
-class CwiseBinaryOp : internal::no_assignment_operator,
+/** \class CwiseBinaryOp
+  * \ingroup Core_Module
+  *
+  * \brief Generic expression where a coefficient-wise binary operator is applied to two expressions
+  *
+  * \tparam BinaryOp template functor implementing the operator
+  * \tparam LhsType the type of the left-hand side
+  * \tparam RhsType the type of the right-hand side
+  *
+  * This class represents an expression  where a coefficient-wise binary operator is applied to two expressions.
+  * It is the return type of binary operators, by which we mean only those binary operators where
+  * both the left-hand side and the right-hand side are Eigen expressions.
+  * For example, the return type of matrix1+matrix2 is a CwiseBinaryOp.
+  *
+  * Most of the time, this is the only way that it is used, so you typically don't have to name
+  * CwiseBinaryOp types explicitly.
+  *
+  * \sa MatrixBase::binaryExpr(const MatrixBase<OtherDerived> &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp
+  */
+template<typename BinaryOp, typename LhsType, typename RhsType>
+class CwiseBinaryOp : 
   public CwiseBinaryOpImpl<
-          BinaryOp, Lhs, Rhs,
-          typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
-                                           typename internal::traits<Rhs>::StorageKind>::ret>
+          BinaryOp, LhsType, RhsType,
+          typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
+                                                        typename internal::traits<RhsType>::StorageKind,
+                                                        BinaryOp>::ret>,
+  internal::no_assignment_operator
 {
   public:
+    
+    typedef typename internal::remove_all<BinaryOp>::type Functor;
+    typedef typename internal::remove_all<LhsType>::type Lhs;
+    typedef typename internal::remove_all<RhsType>::type Rhs;
 
     typedef typename CwiseBinaryOpImpl<
-        BinaryOp, Lhs, Rhs,
-        typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
-                                         typename internal::traits<Rhs>::StorageKind>::ret>::Base Base;
+        BinaryOp, LhsType, RhsType,
+        typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
+                                                      typename internal::traits<Rhs>::StorageKind,
+                                                      BinaryOp>::ret>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseBinaryOp)
 
-    typedef typename internal::nested<Lhs>::type LhsNested;
-    typedef typename internal::nested<Rhs>::type RhsNested;
+    typedef typename internal::ref_selector<LhsType>::type LhsNested;
+    typedef typename internal::ref_selector<RhsType>::type RhsNested;
     typedef typename internal::remove_reference<LhsNested>::type _LhsNested;
     typedef typename internal::remove_reference<RhsNested>::type _RhsNested;
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp())
       : m_lhs(aLhs), m_rhs(aRhs), m_functor(func)
     {
@@ -132,6 +110,7 @@ class CwiseBinaryOp : internal::no_assignment_operator,
       eigen_assert(aLhs.rows() == aRhs.rows() && aLhs.cols() == aRhs.cols());
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rows() const {
       // return the fixed size type if available to enable compile time optimizations
       if (internal::traits<typename internal::remove_all<LhsNested>::type>::RowsAtCompileTime==Dynamic)
@@ -139,6 +118,7 @@ class CwiseBinaryOp : internal::no_assignment_operator,
       else
         return m_lhs.rows();
     }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index cols() const {
       // return the fixed size type if available to enable compile time optimizations
       if (internal::traits<typename internal::remove_all<LhsNested>::type>::ColsAtCompileTime==Dynamic)
@@ -148,10 +128,13 @@ class CwiseBinaryOp : internal::no_assignment_operator,
     }
 
     /** \returns the left hand side nested expression */
+    EIGEN_DEVICE_FUNC
     const _LhsNested& lhs() const { return m_lhs; }
     /** \returns the right hand side nested expression */
+    EIGEN_DEVICE_FUNC
     const _RhsNested& rhs() const { return m_rhs; }
     /** \returns the functor representing the binary operation */
+    EIGEN_DEVICE_FUNC
     const BinaryOp& functor() const { return m_functor; }
 
   protected:
@@ -160,41 +143,13 @@ class CwiseBinaryOp : internal::no_assignment_operator,
     const BinaryOp m_functor;
 };
 
-template<typename BinaryOp, typename Lhs, typename Rhs>
-class CwiseBinaryOpImpl<BinaryOp, Lhs, Rhs, Dense>
-  : public internal::dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
+// Generic API dispatcher
+template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
+class CwiseBinaryOpImpl
+  : public internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
 {
-    typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> Derived;
-  public:
-
-    typedef typename internal::dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE( Derived )
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
-    {
-      return derived().functor()(derived().lhs().coeff(rowId, colId),
-                                 derived().rhs().coeff(rowId, colId));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
-    {
-      return derived().functor().packetOp(derived().lhs().template packet<LoadMode>(rowId, colId),
-                                          derived().rhs().template packet<LoadMode>(rowId, colId));
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
-    {
-      return derived().functor()(derived().lhs().coeff(index),
-                                 derived().rhs().coeff(index));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
-    {
-      return derived().functor().packetOp(derived().lhs().template packet<LoadMode>(index),
-                                          derived().rhs().template packet<LoadMode>(index));
-    }
+public:
+  typedef typename internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
 };
 
 /** replaces \c *this by \c *this - \a other.
@@ -206,8 +161,7 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator-=(const MatrixBase<OtherDerived> &other)
 {
-  SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
@@ -220,11 +174,11 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator+=(const MatrixBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
 } // end namespace Eigen
 
 #endif // EIGEN_CWISE_BINARY_OP_H
+
diff --git a/extern/Eigen3/Eigen/src/Core/CwiseNullaryOp.h b/extern/Eigen3/Eigen/src/Core/CwiseNullaryOp.h
index a93bab2d0f9..ddd607e3832 100644
--- a/extern/Eigen3/Eigen/src/Core/CwiseNullaryOp.h
+++ b/extern/Eigen3/Eigen/src/Core/CwiseNullaryOp.h
@@ -12,13 +12,24 @@
 
 namespace Eigen {
 
+namespace internal {
+template<typename NullaryOp, typename PlainObjectType>
+struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
+{
+  enum {
+    Flags = traits<PlainObjectType>::Flags & RowMajorBit
+  };
+};
+
+} // namespace internal
+
 /** \class CwiseNullaryOp
   * \ingroup Core_Module
   *
   * \brief Generic expression of a matrix where all coefficients are defined by a functor
   *
-  * \param NullaryOp template functor implementing the operator
-  * \param PlainObjectType the underlying plain matrix/array type
+  * \tparam NullaryOp template functor implementing the operator
+  * \tparam PlainObjectType the underlying plain matrix/array type
   *
   * This class represents an expression of a generic nullary operator.
   * It is the return type of the Ones(), Zero(), Constant(), Identity() and Random() methods,
@@ -27,68 +38,49 @@ namespace Eigen {
   * However, if you want to write a function returning such an expression, you
   * will need to use this class.
   *
-  * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr()
+  * The functor NullaryOp must expose one of the following method:
+    <table class="manual">
+    <tr            ><td>\c operator()() </td><td>if the procedural generation does not depend on the coefficient entries (e.g., random numbers)</td></tr>
+    <tr class="alt"><td>\c operator()(Index i)</td><td>if the procedural generation makes sense for vectors only and that it depends on the coefficient index \c i (e.g., linspace) </td></tr>
+    <tr            ><td>\c operator()(Index i,Index j)</td><td>if the procedural generation depends on the matrix coordinates \c i, \c j (e.g., to generate a checkerboard with 0 and 1)</td></tr>
+    </table>
+  * It is also possible to expose the last two operators if the generation makes sense for matrices but can be optimized for vectors.
+  *
+  * See DenseBase::NullaryExpr(Index,const CustomNullaryOp&) for an example binding
+  * C++11 random number generators.
+  *
+  * A nullary expression can also be used to implement custom sophisticated matrix manipulations
+  * that cannot be covered by the existing set of natively supported matrix manipulations.
+  * See this \ref TopicCustomizing_NullaryExpr "page" for some examples and additional explanations
+  * on the behavior of CwiseNullaryOp.
+  *
+  * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr
   */
-
-namespace internal {
 template<typename NullaryOp, typename PlainObjectType>
-struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
-{
-  enum {
-    Flags = (traits<PlainObjectType>::Flags
-      & (  HereditaryBits
-         | (functor_has_linear_access<NullaryOp>::ret ? LinearAccessBit : 0)
-         | (functor_traits<NullaryOp>::PacketAccess ? PacketAccessBit : 0)))
-      | (functor_traits<NullaryOp>::IsRepeatable ? 0 : EvalBeforeNestingBit),
-    CoeffReadCost = functor_traits<NullaryOp>::Cost
-  };
-};
-}
-
-template<typename NullaryOp, typename PlainObjectType>
-class CwiseNullaryOp : internal::no_assignment_operator,
-  public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type
+class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type, internal::no_assignment_operator
 {
   public:
 
     typedef typename internal::dense_xpr_base<CwiseNullaryOp>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(CwiseNullaryOp)
 
-    CwiseNullaryOp(Index nbRows, Index nbCols, const NullaryOp& func = NullaryOp())
-      : m_rows(nbRows), m_cols(nbCols), m_functor(func)
+    EIGEN_DEVICE_FUNC
+    CwiseNullaryOp(Index rows, Index cols, const NullaryOp& func = NullaryOp())
+      : m_rows(rows), m_cols(cols), m_functor(func)
     {
-      eigen_assert(nbRows >= 0
-            && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == nbRows)
-            &&  nbCols >= 0
-            && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == nbCols));
+      eigen_assert(rows >= 0
+            && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
+            &&  cols >= 0
+            && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rows() const { return m_rows.value(); }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index cols() const { return m_cols.value(); }
 
-    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
-    {
-      return m_functor(rowId, colId);
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
-    {
-      return m_functor.packetOp(rowId, colId);
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
-    {
-      return m_functor(index);
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
-    {
-      return m_functor.packetOp(index);
-    }
-
     /** \returns the functor representing the nullary operation */
+    EIGEN_DEVICE_FUNC
     const NullaryOp& functor() const { return m_functor; }
 
   protected:
@@ -113,10 +105,10 @@ class CwiseNullaryOp : internal::no_assignment_operator,
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
 DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func)
 {
-  return CwiseNullaryOp<CustomNullaryOp, Derived>(rows, cols, func);
+  return CwiseNullaryOp<CustomNullaryOp, PlainObject>(rows, cols, func);
 }
 
 /** \returns an expression of a matrix defined by a custom functor \a func
@@ -132,16 +124,19 @@ DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& f
   *
   * The template parameter \a CustomNullaryOp is the type of the functor.
   *
+  * Here is an example with C++11 random generators: \include random_cpp11.cpp
+  * Output: \verbinclude random_cpp11.out
+  * 
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
+EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
 DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, Derived>(1, size, func);
-  else return CwiseNullaryOp<CustomNullaryOp, Derived>(size, 1, func);
+  if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, PlainObject>(1, size, func);
+  else return CwiseNullaryOp<CustomNullaryOp, PlainObject>(size, 1, func);
 }
 
 /** \returns an expression of a matrix defined by a custom functor \a func
@@ -155,19 +150,19 @@ DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
 DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
 {
-  return CwiseNullaryOp<CustomNullaryOp, Derived>(RowsAtCompileTime, ColsAtCompileTime, func);
+  return CwiseNullaryOp<CustomNullaryOp, PlainObject>(RowsAtCompileTime, ColsAtCompileTime, func);
 }
 
 /** \returns an expression of a constant matrix of value \a value
   *
-  * The parameters \a nbRows and \a nbCols are the number of rows and of columns of
+  * The parameters \a rows and \a cols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this DenseBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
-  * it is redundant to pass \a nbRows and \a nbCols as arguments, so Zero() should be used
+  * it is redundant to pass \a rows and \a cols as arguments, so Zero() should be used
   * instead.
   *
   * The template parameter \a CustomNullaryOp is the type of the functor.
@@ -176,9 +171,9 @@ DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Constant(Index nbRows, Index nbCols, const Scalar& value)
+DenseBase<Derived>::Constant(Index rows, Index cols, const Scalar& value)
 {
-  return DenseBase<Derived>::NullaryExpr(nbRows, nbCols, internal::scalar_constant_op<Scalar>(value));
+  return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_constant_op<Scalar>(value));
 }
 
 /** \returns an expression of a constant matrix of value \a value
@@ -197,7 +192,7 @@ DenseBase<Derived>::Constant(Index nbRows, Index nbCols, const Scalar& value)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(Index size, const Scalar& value)
 {
   return DenseBase<Derived>::NullaryExpr(size, internal::scalar_constant_op<Scalar>(value));
@@ -213,53 +208,40 @@ DenseBase<Derived>::Constant(Index size, const Scalar& value)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(const Scalar& value)
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
   return DenseBase<Derived>::NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_constant_op<Scalar>(value));
 }
 
-/**
-  * \brief Sets a linearly space vector.
-  *
-  * The function generates 'size' equally spaced values in the closed interval [low,high].
-  * This particular version of LinSpaced() uses sequential access, i.e. vector access is
-  * assumed to be a(0), a(1), ..., a(size). This assumption allows for better vectorization
-  * and yields faster code than the random access version.
-  *
-  * When size is set to 1, a vector of length 1 containing 'high' is returned.
-  *
-  * \only_for_vectors
-  *
-  * Example: \include DenseBase_LinSpaced_seq.cpp
-  * Output: \verbinclude DenseBase_LinSpaced_seq.out
+/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(Index,const Scalar&,const Scalar&)
   *
-  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Index,Scalar,Scalar), CwiseNullaryOp
+  * \sa LinSpaced(Index,Scalar,Scalar), setLinSpaced(Index,const Scalar&,const Scalar&)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::SequentialLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,false>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
 }
 
-/**
-  * \copydoc DenseBase::LinSpaced(Sequential_t, Index, const Scalar&, const Scalar&)
-  * Special version for fixed size types which does not require the size parameter.
+/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(const Scalar&,const Scalar&)
+  *
+  * \sa LinSpaced(Scalar,Scalar)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::SequentialLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,false>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
 }
 
 /**
-  * \brief Sets a linearly space vector.
+  * \brief Sets a linearly spaced vector.
   *
   * The function generates 'size' equally spaced values in the closed interval [low,high].
   * When size is set to 1, a vector of length 1 containing 'high' is returned.
@@ -269,14 +251,24 @@ DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& hig
   * Example: \include DenseBase_LinSpaced.cpp
   * Output: \verbinclude DenseBase_LinSpaced.out
   *
-  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Sequential_t,Index,const Scalar&,const Scalar&,Index), CwiseNullaryOp
+  * For integer scalar types, an even spacing is possible if and only if the length of the range,
+  * i.e., \c high-low is a scalar multiple of \c size-1, or if \c size is a scalar multiple of the
+  * number of values \c high-low+1 (meaning each value can be repeated the same number of time).
+  * If one of these two considions is not satisfied, then \c high is lowered to the largest value
+  * satisfying one of this constraint.
+  * Here are some examples:
+  *
+  * Example: \include DenseBase_LinSpacedInt.cpp
+  * Output: \verbinclude DenseBase_LinSpacedInt.out
+  *
+  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,true>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
 }
 
 /**
@@ -284,22 +276,23 @@ DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
   * Special version for fixed size types which does not require the size parameter.
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,true>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
 }
 
 /** \returns true if all coefficients in this matrix are approximately equal to \a val, to within precision \a prec */
 template<typename Derived>
-bool DenseBase<Derived>::isApproxToConstant
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isApproxToConstant
 (const Scalar& val, const RealScalar& prec) const
 {
+  typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index j = 0; j < cols(); ++j)
     for(Index i = 0; i < rows(); ++i)
-      if(!internal::isApprox(this->coeff(i, j), val, prec))
+      if(!internal::isApprox(self.coeff(i, j), val, prec))
         return false;
   return true;
 }
@@ -308,7 +301,7 @@ bool DenseBase<Derived>::isApproxToConstant
   *
   * \returns true if all coefficients in this matrix are approximately equal to \a value, to within precision \a prec */
 template<typename Derived>
-bool DenseBase<Derived>::isConstant
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isConstant
 (const Scalar& val, const RealScalar& prec) const
 {
   return isApproxToConstant(val, prec);
@@ -319,22 +312,22 @@ bool DenseBase<Derived>::isConstant
   * \sa setConstant(), Constant(), class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
 {
   setConstant(val);
 }
 
-/** Sets all coefficients in this expression to \a value.
+/** Sets all coefficients in this expression to value \a val.
   *
   * \sa fill(), setConstant(Index,const Scalar&), setConstant(Index,Index,const Scalar&), setZero(), setOnes(), Constant(), class CwiseNullaryOp, setZero(), setOnes()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
 {
   return derived() = Constant(rows(), cols(), val);
 }
 
-/** Resizes to the given \a size, and sets all coefficients in this expression to the given \a value.
+/** Resizes to the given \a size, and sets all coefficients in this expression to the given value \a val.
   *
   * \only_for_vectors
   *
@@ -344,17 +337,17 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
   * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
 {
   resize(size);
   return setConstant(val);
 }
 
-/** Resizes to the given size, and sets all coefficients in this expression to the given \a value.
+/** Resizes to the given size, and sets all coefficients in this expression to the given value \a val.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   * \param val the value to which all coefficients are set
   *
   * Example: \include Matrix_setConstant_int_int.cpp
@@ -363,15 +356,15 @@ PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
   * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setConstant(Index nbRows, Index nbCols, const Scalar& val)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
 {
-  resize(nbRows, nbCols);
+  resize(rows, cols);
   return setConstant(val);
 }
 
 /**
-  * \brief Sets a linearly space vector.
+  * \brief Sets a linearly spaced vector.
   *
   * The function generates 'size' equally spaced values in the closed interval [low,high].
   * When size is set to 1, a vector of length 1 containing 'high' is returned.
@@ -381,27 +374,33 @@ PlainObjectBase<Derived>::setConstant(Index nbRows, Index nbCols, const Scalar&
   * Example: \include DenseBase_setLinSpaced.cpp
   * Output: \verbinclude DenseBase_setLinSpaced.out
   *
-  * \sa CwiseNullaryOp
+  * For integer scalar types, do not miss the explanations on the definition
+  * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
+  *
+  * \sa LinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,false>(low,high,newSize));
+  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,PacketScalar>(low,high,newSize));
 }
 
 /**
-  * \brief Sets a linearly space vector.
+  * \brief Sets a linearly spaced vector.
   *
-  * The function fill *this with equally spaced values in the closed interval [low,high].
+  * The function fills \c *this with equally spaced values in the closed interval [low,high].
   * When size is set to 1, a vector of length 1 containing 'high' is returned.
   *
   * \only_for_vectors
   *
-  * \sa setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
+  * For integer scalar types, do not miss the explanations on the definition
+  * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
+  *
+  * \sa LinSpaced(Index,const Scalar&,const Scalar&), setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return setLinSpaced(size(), low, high);
@@ -424,10 +423,10 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low,
   * \sa Zero(), Zero(Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Zero(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Zero(Index rows, Index cols)
 {
-  return Constant(nbRows, nbCols, Scalar(0));
+  return Constant(rows, cols, Scalar(0));
 }
 
 /** \returns an expression of a zero vector.
@@ -447,7 +446,7 @@ DenseBase<Derived>::Zero(Index nbRows, Index nbCols)
   * \sa Zero(), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Zero(Index size)
 {
   return Constant(size, Scalar(0));
@@ -464,7 +463,7 @@ DenseBase<Derived>::Zero(Index size)
   * \sa Zero(Index), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Zero()
 {
   return Constant(Scalar(0));
@@ -479,11 +478,12 @@ DenseBase<Derived>::Zero()
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-bool DenseBase<Derived>::isZero(const RealScalar& prec) const
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isZero(const RealScalar& prec) const
 {
+  typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index j = 0; j < cols(); ++j)
     for(Index i = 0; i < rows(); ++i)
-      if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast<Scalar>(1), prec))
+      if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<Scalar>(1), prec))
         return false;
   return true;
 }
@@ -496,7 +496,7 @@ bool DenseBase<Derived>::isZero(const RealScalar& prec) const
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
 {
   return setConstant(Scalar(0));
 }
@@ -511,7 +511,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
   * \sa DenseBase::setZero(), setZero(Index,Index), class CwiseNullaryOp, DenseBase::Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setZero(Index newSize)
 {
   resize(newSize);
@@ -520,8 +520,8 @@ PlainObjectBase<Derived>::setZero(Index newSize)
 
 /** Resizes to the given size, and sets all coefficients in this expression to zero.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   *
   * Example: \include Matrix_setZero_int_int.cpp
   * Output: \verbinclude Matrix_setZero_int_int.out
@@ -529,10 +529,10 @@ PlainObjectBase<Derived>::setZero(Index newSize)
   * \sa DenseBase::setZero(), setZero(Index), class CwiseNullaryOp, DenseBase::Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setZero(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setZero(Index rows, Index cols)
 {
-  resize(nbRows, nbCols);
+  resize(rows, cols);
   return setConstant(Scalar(0));
 }
 
@@ -540,7 +540,7 @@ PlainObjectBase<Derived>::setZero(Index nbRows, Index nbCols)
 
 /** \returns an expression of a matrix where all coefficients equal one.
   *
-  * The parameters \a nbRows and \a nbCols are the number of rows and of columns of
+  * The parameters \a rows and \a cols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this MatrixBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
@@ -553,10 +553,10 @@ PlainObjectBase<Derived>::setZero(Index nbRows, Index nbCols)
   * \sa Ones(), Ones(Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Ones(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Ones(Index rows, Index cols)
 {
-  return Constant(nbRows, nbCols, Scalar(1));
+  return Constant(rows, cols, Scalar(1));
 }
 
 /** \returns an expression of a vector where all coefficients equal one.
@@ -576,7 +576,7 @@ DenseBase<Derived>::Ones(Index nbRows, Index nbCols)
   * \sa Ones(), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Ones(Index newSize)
 {
   return Constant(newSize, Scalar(1));
@@ -593,7 +593,7 @@ DenseBase<Derived>::Ones(Index newSize)
   * \sa Ones(Index), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Ones()
 {
   return Constant(Scalar(1));
@@ -608,7 +608,7 @@ DenseBase<Derived>::Ones()
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-bool DenseBase<Derived>::isOnes
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isOnes
 (const RealScalar& prec) const
 {
   return isApproxToConstant(Scalar(1), prec);
@@ -622,7 +622,7 @@ bool DenseBase<Derived>::isOnes
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
 {
   return setConstant(Scalar(1));
 }
@@ -637,7 +637,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
   * \sa MatrixBase::setOnes(), setOnes(Index,Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setOnes(Index newSize)
 {
   resize(newSize);
@@ -646,8 +646,8 @@ PlainObjectBase<Derived>::setOnes(Index newSize)
 
 /** Resizes to the given size, and sets all coefficients in this expression to one.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   *
   * Example: \include Matrix_setOnes_int_int.cpp
   * Output: \verbinclude Matrix_setOnes_int_int.out
@@ -655,10 +655,10 @@ PlainObjectBase<Derived>::setOnes(Index newSize)
   * \sa MatrixBase::setOnes(), setOnes(Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setOnes(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
 {
-  resize(nbRows, nbCols);
+  resize(rows, cols);
   return setConstant(Scalar(1));
 }
 
@@ -666,7 +666,7 @@ PlainObjectBase<Derived>::setOnes(Index nbRows, Index nbCols)
 
 /** \returns an expression of the identity matrix (not necessarily square).
   *
-  * The parameters \a nbRows and \a nbCols are the number of rows and of columns of
+  * The parameters \a rows and \a cols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this MatrixBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
@@ -679,10 +679,10 @@ PlainObjectBase<Derived>::setOnes(Index nbRows, Index nbCols)
   * \sa Identity(), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
-MatrixBase<Derived>::Identity(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+MatrixBase<Derived>::Identity(Index rows, Index cols)
 {
-  return DenseBase<Derived>::NullaryExpr(nbRows, nbCols, internal::scalar_identity_op<Scalar>());
+  return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_identity_op<Scalar>());
 }
 
 /** \returns an expression of the identity matrix (not necessarily square).
@@ -696,7 +696,7 @@ MatrixBase<Derived>::Identity(Index nbRows, Index nbCols)
   * \sa Identity(Index,Index), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
 MatrixBase<Derived>::Identity()
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
@@ -716,18 +716,19 @@ template<typename Derived>
 bool MatrixBase<Derived>::isIdentity
 (const RealScalar& prec) const
 {
+  typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index j = 0; j < cols(); ++j)
   {
     for(Index i = 0; i < rows(); ++i)
     {
       if(i == j)
       {
-        if(!internal::isApprox(this->coeff(i, j), static_cast<Scalar>(1), prec))
+        if(!internal::isApprox(self.coeff(i, j), static_cast<Scalar>(1), prec))
           return false;
       }
       else
       {
-        if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast<RealScalar>(1), prec))
+        if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<RealScalar>(1), prec))
           return false;
       }
     }
@@ -740,6 +741,7 @@ namespace internal {
 template<typename Derived, bool Big = (Derived::SizeAtCompileTime>=16)>
 struct setIdentity_impl
 {
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Derived& run(Derived& m)
   {
     return m = Derived::Identity(m.rows(), m.cols());
@@ -749,11 +751,11 @@ struct setIdentity_impl
 template<typename Derived>
 struct setIdentity_impl<Derived, true>
 {
-  typedef typename Derived::Index Index;
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Derived& run(Derived& m)
   {
     m.setZero();
-    const Index size = (std::min)(m.rows(), m.cols());
+    const Index size = numext::mini(m.rows(), m.cols());
     for(Index i = 0; i < size; ++i) m.coeffRef(i,i) = typename Derived::Scalar(1);
     return m;
   }
@@ -769,15 +771,15 @@ struct setIdentity_impl<Derived, true>
   * \sa class CwiseNullaryOp, Identity(), Identity(Index,Index), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
 {
   return internal::setIdentity_impl<Derived>::run(derived());
 }
 
 /** \brief Resizes to the given size, and writes the identity expression (not necessarily square) into *this.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   *
   * Example: \include Matrix_setIdentity_int_int.cpp
   * Output: \verbinclude Matrix_setIdentity_int_int.out
@@ -785,9 +787,9 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
   * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Identity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index rows, Index cols)
 {
-  derived().resize(nbRows, nbCols);
+  derived().resize(rows, cols);
   return setIdentity();
 }
 
@@ -798,7 +800,7 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index nbRows, Inde
   * \sa MatrixBase::Unit(Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, Index i)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, Index i)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return BasisReturnType(SquareMatrixType::Identity(newSize,newSize), i);
@@ -813,7 +815,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return BasisReturnType(SquareMatrixType::Identity(),i);
@@ -826,7 +828,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
 { return Derived::Unit(0); }
 
 /** \returns an expression of the Y axis unit vector (0,1{,0}^*)
@@ -836,7 +838,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
 { return Derived::Unit(1); }
 
 /** \returns an expression of the Z axis unit vector (0,0,1{,0}^*)
@@ -846,7 +848,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
 { return Derived::Unit(2); }
 
 /** \returns an expression of the W axis unit vector (0,0,0,1)
@@ -856,7 +858,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
 { return Derived::Unit(3); }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/CwiseTernaryOp.h b/extern/Eigen3/Eigen/src/Core/CwiseTernaryOp.h
new file mode 100644
index 00000000000..9f3576fecef
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/CwiseTernaryOp.h
@@ -0,0 +1,197 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2016 Eugene Brevdo <ebrevdo@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_CWISE_TERNARY_OP_H
+#define EIGEN_CWISE_TERNARY_OP_H
+
+namespace Eigen {
+
+namespace internal {
+template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
+struct traits<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> > {
+  // we must not inherit from traits<Arg1> since it has
+  // the potential to cause problems with MSVC
+  typedef typename remove_all<Arg1>::type Ancestor;
+  typedef typename traits<Ancestor>::XprKind XprKind;
+  enum {
+    RowsAtCompileTime = traits<Ancestor>::RowsAtCompileTime,
+    ColsAtCompileTime = traits<Ancestor>::ColsAtCompileTime,
+    MaxRowsAtCompileTime = traits<Ancestor>::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = traits<Ancestor>::MaxColsAtCompileTime
+  };
+
+  // even though we require Arg1, Arg2, and Arg3 to have the same scalar type
+  // (see CwiseTernaryOp constructor),
+  // we still want to handle the case when the result type is different.
+  typedef typename result_of<TernaryOp(
+      const typename Arg1::Scalar&, const typename Arg2::Scalar&,
+      const typename Arg3::Scalar&)>::type Scalar;
+
+  typedef typename internal::traits<Arg1>::StorageKind StorageKind;
+  typedef typename internal::traits<Arg1>::StorageIndex StorageIndex;
+
+  typedef typename Arg1::Nested Arg1Nested;
+  typedef typename Arg2::Nested Arg2Nested;
+  typedef typename Arg3::Nested Arg3Nested;
+  typedef typename remove_reference<Arg1Nested>::type _Arg1Nested;
+  typedef typename remove_reference<Arg2Nested>::type _Arg2Nested;
+  typedef typename remove_reference<Arg3Nested>::type _Arg3Nested;
+  enum { Flags = _Arg1Nested::Flags & RowMajorBit };
+};
+}  // end namespace internal
+
+template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3,
+          typename StorageKind>
+class CwiseTernaryOpImpl;
+
+/** \class CwiseTernaryOp
+  * \ingroup Core_Module
+  *
+  * \brief Generic expression where a coefficient-wise ternary operator is
+ * applied to two expressions
+  *
+  * \tparam TernaryOp template functor implementing the operator
+  * \tparam Arg1Type the type of the first argument
+  * \tparam Arg2Type the type of the second argument
+  * \tparam Arg3Type the type of the third argument
+  *
+  * This class represents an expression where a coefficient-wise ternary
+ * operator is applied to three expressions.
+  * It is the return type of ternary operators, by which we mean only those
+ * ternary operators where
+  * all three arguments are Eigen expressions.
+  * For example, the return type of betainc(matrix1, matrix2, matrix3) is a
+ * CwiseTernaryOp.
+  *
+  * Most of the time, this is the only way that it is used, so you typically
+ * don't have to name
+  * CwiseTernaryOp types explicitly.
+  *
+  * \sa MatrixBase::ternaryExpr(const MatrixBase<Argument2> &, const
+ * MatrixBase<Argument3> &, const CustomTernaryOp &) const, class CwiseBinaryOp,
+ * class CwiseUnaryOp, class CwiseNullaryOp
+  */
+template <typename TernaryOp, typename Arg1Type, typename Arg2Type,
+          typename Arg3Type>
+class CwiseTernaryOp : public CwiseTernaryOpImpl<
+                           TernaryOp, Arg1Type, Arg2Type, Arg3Type,
+                           typename internal::traits<Arg1Type>::StorageKind>,
+                       internal::no_assignment_operator
+{
+ public:
+  typedef typename internal::remove_all<Arg1Type>::type Arg1;
+  typedef typename internal::remove_all<Arg2Type>::type Arg2;
+  typedef typename internal::remove_all<Arg3Type>::type Arg3;
+
+  typedef typename CwiseTernaryOpImpl<
+      TernaryOp, Arg1Type, Arg2Type, Arg3Type,
+      typename internal::traits<Arg1Type>::StorageKind>::Base Base;
+  EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseTernaryOp)
+
+  typedef typename internal::ref_selector<Arg1Type>::type Arg1Nested;
+  typedef typename internal::ref_selector<Arg2Type>::type Arg2Nested;
+  typedef typename internal::ref_selector<Arg3Type>::type Arg3Nested;
+  typedef typename internal::remove_reference<Arg1Nested>::type _Arg1Nested;
+  typedef typename internal::remove_reference<Arg2Nested>::type _Arg2Nested;
+  typedef typename internal::remove_reference<Arg3Nested>::type _Arg3Nested;
+
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE CwiseTernaryOp(const Arg1& a1, const Arg2& a2,
+                                     const Arg3& a3,
+                                     const TernaryOp& func = TernaryOp())
+      : m_arg1(a1), m_arg2(a2), m_arg3(a3), m_functor(func) {
+    // require the sizes to match
+    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg2)
+    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg3)
+
+    // The index types should match
+    EIGEN_STATIC_ASSERT((internal::is_same<
+                         typename internal::traits<Arg1Type>::StorageKind,
+                         typename internal::traits<Arg2Type>::StorageKind>::value),
+                        STORAGE_KIND_MUST_MATCH)
+    EIGEN_STATIC_ASSERT((internal::is_same<
+                         typename internal::traits<Arg1Type>::StorageKind,
+                         typename internal::traits<Arg3Type>::StorageKind>::value),
+                        STORAGE_KIND_MUST_MATCH)
+
+    eigen_assert(a1.rows() == a2.rows() && a1.cols() == a2.cols() &&
+                 a1.rows() == a3.rows() && a1.cols() == a3.cols());
+  }
+
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE Index rows() const {
+    // return the fixed size type if available to enable compile time
+    // optimizations
+    if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                RowsAtCompileTime == Dynamic &&
+        internal::traits<typename internal::remove_all<Arg2Nested>::type>::
+                RowsAtCompileTime == Dynamic)
+      return m_arg3.rows();
+    else if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                     RowsAtCompileTime == Dynamic &&
+             internal::traits<typename internal::remove_all<Arg3Nested>::type>::
+                     RowsAtCompileTime == Dynamic)
+      return m_arg2.rows();
+    else
+      return m_arg1.rows();
+  }
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE Index cols() const {
+    // return the fixed size type if available to enable compile time
+    // optimizations
+    if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                ColsAtCompileTime == Dynamic &&
+        internal::traits<typename internal::remove_all<Arg2Nested>::type>::
+                ColsAtCompileTime == Dynamic)
+      return m_arg3.cols();
+    else if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                     ColsAtCompileTime == Dynamic &&
+             internal::traits<typename internal::remove_all<Arg3Nested>::type>::
+                     ColsAtCompileTime == Dynamic)
+      return m_arg2.cols();
+    else
+      return m_arg1.cols();
+  }
+
+  /** \returns the first argument nested expression */
+  EIGEN_DEVICE_FUNC
+  const _Arg1Nested& arg1() const { return m_arg1; }
+  /** \returns the first argument nested expression */
+  EIGEN_DEVICE_FUNC
+  const _Arg2Nested& arg2() const { return m_arg2; }
+  /** \returns the third argument nested expression */
+  EIGEN_DEVICE_FUNC
+  const _Arg3Nested& arg3() const { return m_arg3; }
+  /** \returns the functor representing the ternary operation */
+  EIGEN_DEVICE_FUNC
+  const TernaryOp& functor() const { return m_functor; }
+
+ protected:
+  Arg1Nested m_arg1;
+  Arg2Nested m_arg2;
+  Arg3Nested m_arg3;
+  const TernaryOp m_functor;
+};
+
+// Generic API dispatcher
+template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3,
+          typename StorageKind>
+class CwiseTernaryOpImpl
+    : public internal::generic_xpr_base<
+          CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >::type {
+ public:
+  typedef typename internal::generic_xpr_base<
+      CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >::type Base;
+};
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_CWISE_TERNARY_OP_H
diff --git a/extern/Eigen3/Eigen/src/Core/CwiseUnaryOp.h b/extern/Eigen3/Eigen/src/Core/CwiseUnaryOp.h
index f7ee60e9879..1d2dd19f2b1 100644
--- a/extern/Eigen3/Eigen/src/Core/CwiseUnaryOp.h
+++ b/extern/Eigen3/Eigen/src/Core/CwiseUnaryOp.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -13,41 +13,18 @@
 
 namespace Eigen { 
 
-/** \class CwiseUnaryOp
-  * \ingroup Core_Module
-  *
-  * \brief Generic expression where a coefficient-wise unary operator is applied to an expression
-  *
-  * \param UnaryOp template functor implementing the operator
-  * \param XprType the type of the expression to which we are applying the unary operator
-  *
-  * This class represents an expression where a unary operator is applied to an expression.
-  * It is the return type of all operations taking exactly 1 input expression, regardless of the
-  * presence of other inputs such as scalars. For example, the operator* in the expression 3*matrix
-  * is considered unary, because only the right-hand side is an expression, and its
-  * return type is a specialization of CwiseUnaryOp.
-  *
-  * Most of the time, this is the only way that it is used, so you typically don't have to name
-  * CwiseUnaryOp types explicitly.
-  *
-  * \sa MatrixBase::unaryExpr(const CustomUnaryOp &) const, class CwiseBinaryOp, class CwiseNullaryOp
-  */
-
 namespace internal {
 template<typename UnaryOp, typename XprType>
 struct traits<CwiseUnaryOp<UnaryOp, XprType> >
  : traits<XprType>
 {
   typedef typename result_of<
-                     UnaryOp(typename XprType::Scalar)
+                     UnaryOp(const typename XprType::Scalar&)
                    >::type Scalar;
   typedef typename XprType::Nested XprTypeNested;
   typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
   enum {
-    Flags = _XprTypeNested::Flags & (
-      HereditaryBits | LinearAccessBit | AlignedBit
-      | (functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0)),
-    CoeffReadCost = EIGEN_ADD_COST(_XprTypeNested::CoeffReadCost, functor_traits<UnaryOp>::Cost)
+    Flags = _XprTypeNested::Flags & RowMajorBit 
   };
 };
 }
@@ -55,70 +32,70 @@ struct traits<CwiseUnaryOp<UnaryOp, XprType> >
 template<typename UnaryOp, typename XprType, typename StorageKind>
 class CwiseUnaryOpImpl;
 
+/** \class CwiseUnaryOp
+  * \ingroup Core_Module
+  *
+  * \brief Generic expression where a coefficient-wise unary operator is applied to an expression
+  *
+  * \tparam UnaryOp template functor implementing the operator
+  * \tparam XprType the type of the expression to which we are applying the unary operator
+  *
+  * This class represents an expression where a unary operator is applied to an expression.
+  * It is the return type of all operations taking exactly 1 input expression, regardless of the
+  * presence of other inputs such as scalars. For example, the operator* in the expression 3*matrix
+  * is considered unary, because only the right-hand side is an expression, and its
+  * return type is a specialization of CwiseUnaryOp.
+  *
+  * Most of the time, this is the only way that it is used, so you typically don't have to name
+  * CwiseUnaryOp types explicitly.
+  *
+  * \sa MatrixBase::unaryExpr(const CustomUnaryOp &) const, class CwiseBinaryOp, class CwiseNullaryOp
+  */
 template<typename UnaryOp, typename XprType>
-class CwiseUnaryOp : internal::no_assignment_operator,
-  public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal::traits<XprType>::StorageKind>
+class CwiseUnaryOp : public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal::traits<XprType>::StorageKind>, internal::no_assignment_operator
 {
   public:
 
     typedef typename CwiseUnaryOpImpl<UnaryOp, XprType,typename internal::traits<XprType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryOp)
+    typedef typename internal::ref_selector<XprType>::type XprTypeNested;
+    typedef typename internal::remove_all<XprType>::type NestedExpression;
 
-    inline CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    explicit CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
       : m_xpr(xpr), m_functor(func) {}
 
-    EIGEN_STRONG_INLINE Index rows() const { return m_xpr.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_xpr.cols(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Index rows() const { return m_xpr.rows(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Index cols() const { return m_xpr.cols(); }
 
     /** \returns the functor representing the unary operation */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     const UnaryOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    const typename internal::remove_all<typename XprType::Nested>::type&
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    const typename internal::remove_all<XprTypeNested>::type&
     nestedExpression() const { return m_xpr; }
 
     /** \returns the nested expression */
-    typename internal::remove_all<typename XprType::Nested>::type&
-    nestedExpression() { return m_xpr.const_cast_derived(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    typename internal::remove_all<XprTypeNested>::type&
+    nestedExpression() { return m_xpr; }
 
   protected:
-    typename XprType::Nested m_xpr;
+    XprTypeNested m_xpr;
     const UnaryOp m_functor;
 };
 
-// This is the generic implementation for dense storage.
-// It can be used for any expression types implementing the dense concept.
-template<typename UnaryOp, typename XprType>
-class CwiseUnaryOpImpl<UnaryOp,XprType,Dense>
-  : public internal::dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
+// Generic API dispatcher
+template<typename UnaryOp, typename XprType, typename StorageKind>
+class CwiseUnaryOpImpl
+  : public internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
 {
-  public:
-
-    typedef CwiseUnaryOp<UnaryOp, XprType> Derived;
-    typedef typename internal::dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(rowId, colId));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
-    {
-      return derived().functor().packetOp(derived().nestedExpression().template packet<LoadMode>(rowId, colId));
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(index));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
-    {
-      return derived().functor().packetOp(derived().nestedExpression().template packet<LoadMode>(index));
-    }
+public:
+  typedef typename internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
 };
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/CwiseUnaryView.h b/extern/Eigen3/Eigen/src/Core/CwiseUnaryView.h
index b2638d3265b..27103305629 100644
--- a/extern/Eigen3/Eigen/src/Core/CwiseUnaryView.h
+++ b/extern/Eigen3/Eigen/src/Core/CwiseUnaryView.h
@@ -12,33 +12,19 @@
 
 namespace Eigen {
 
-/** \class CwiseUnaryView
-  * \ingroup Core_Module
-  *
-  * \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector
-  *
-  * \param ViewOp template functor implementing the view
-  * \param MatrixType the type of the matrix we are applying the unary operator
-  *
-  * This class represents a lvalue expression of a generic unary view operator of a matrix or a vector.
-  * It is the return type of real() and imag(), and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp
-  */
-
 namespace internal {
 template<typename ViewOp, typename MatrixType>
 struct traits<CwiseUnaryView<ViewOp, MatrixType> >
  : traits<MatrixType>
 {
   typedef typename result_of<
-                     ViewOp(typename traits<MatrixType>::Scalar)
+                     ViewOp(const typename traits<MatrixType>::Scalar&)
                    >::type Scalar;
   typedef typename MatrixType::Nested MatrixTypeNested;
   typedef typename remove_all<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
-    Flags = (traits<_MatrixTypeNested>::Flags & (HereditaryBits | LvalueBit | LinearAccessBit | DirectAccessBit)),
-    CoeffReadCost = traits<_MatrixTypeNested>::CoeffReadCost + functor_traits<ViewOp>::Cost,
+    FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
+    Flags = traits<_MatrixTypeNested>::Flags & (RowMajorBit | FlagsLvalueBit | DirectAccessBit), // FIXME DirectAccessBit should not be handled by expressions
     MatrixTypeInnerStride =  inner_stride_at_compile_time<MatrixType>::ret,
     // need to cast the sizeof's from size_t to int explicitly, otherwise:
     // "error: no integral type can represent all of the enumerator values
@@ -55,6 +41,19 @@ struct traits<CwiseUnaryView<ViewOp, MatrixType> >
 template<typename ViewOp, typename MatrixType, typename StorageKind>
 class CwiseUnaryViewImpl;
 
+/** \class CwiseUnaryView
+  * \ingroup Core_Module
+  *
+  * \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector
+  *
+  * \tparam ViewOp template functor implementing the view
+  * \tparam MatrixType the type of the matrix we are applying the unary operator
+  *
+  * This class represents a lvalue expression of a generic unary view operator of a matrix or a vector.
+  * It is the return type of real() and imag(), and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp
+  */
 template<typename ViewOp, typename MatrixType>
 class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename internal::traits<MatrixType>::StorageKind>
 {
@@ -62,8 +61,10 @@ class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename in
 
     typedef typename CwiseUnaryViewImpl<ViewOp, MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryView)
+    typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
-    inline CwiseUnaryView(const MatrixType& mat, const ViewOp& func = ViewOp())
+    explicit inline CwiseUnaryView(MatrixType& mat, const ViewOp& func = ViewOp())
       : m_matrix(mat), m_functor(func) {}
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryView)
@@ -75,19 +76,27 @@ class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename in
     const ViewOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    const typename internal::remove_all<typename MatrixType::Nested>::type&
+    const typename internal::remove_all<MatrixTypeNested>::type&
     nestedExpression() const { return m_matrix; }
 
     /** \returns the nested expression */
-    typename internal::remove_all<typename MatrixType::Nested>::type&
+    typename internal::remove_reference<MatrixTypeNested>::type&
     nestedExpression() { return m_matrix.const_cast_derived(); }
 
   protected:
-    // FIXME changed from MatrixType::Nested because of a weird compilation error with sun CC
-    typename internal::nested<MatrixType>::type m_matrix;
+    MatrixTypeNested m_matrix;
     ViewOp m_functor;
 };
 
+// Generic API dispatcher
+template<typename ViewOp, typename XprType, typename StorageKind>
+class CwiseUnaryViewImpl
+  : public internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type
+{
+public:
+  typedef typename internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type Base;
+};
+
 template<typename ViewOp, typename MatrixType>
 class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
   : public internal::dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type
@@ -100,38 +109,18 @@ class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
     EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryViewImpl)
     
-    inline Scalar* data() { return &coeffRef(0); }
-    inline const Scalar* data() const { return &coeff(0); }
+    EIGEN_DEVICE_FUNC inline Scalar* data() { return &(this->coeffRef(0)); }
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return &(this->coeff(0)); }
 
-    inline Index innerStride() const
+    EIGEN_DEVICE_FUNC inline Index innerStride() const
     {
       return derived().nestedExpression().innerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
 
-    inline Index outerStride() const
+    EIGEN_DEVICE_FUNC inline Index outerStride() const
     {
       return derived().nestedExpression().outerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
-
-    EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(row, col));
-    }
-
-    EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(index));
-    }
-
-    EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
-    {
-      return derived().functor()(const_cast_derived().nestedExpression().coeffRef(row, col));
-    }
-
-    EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
-    {
-      return derived().functor()(const_cast_derived().nestedExpression().coeffRef(index));
-    }
 };
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/DenseBase.h b/extern/Eigen3/Eigen/src/Core/DenseBase.h
index 4b371b075b8..90066ae73fd 100644
--- a/extern/Eigen3/Eigen/src/Core/DenseBase.h
+++ b/extern/Eigen3/Eigen/src/Core/DenseBase.h
@@ -34,37 +34,45 @@ static inline void check_DenseIndex_is_signed() {
   * \tparam Derived is the derived type, e.g., a matrix type or an expression.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN.
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived> class DenseBase
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-  : public internal::special_scalar_op_base<Derived, typename internal::traits<Derived>::Scalar,
-                                            typename NumTraits<typename internal::traits<Derived>::Scalar>::Real,
-                                            DenseCoeffsBase<Derived> >
-#else
   : public DenseCoeffsBase<Derived>
+#else
+  : public DenseCoeffsBase<Derived,DirectWriteAccessors>
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 {
   public:
 
-    class InnerIterator;
+    /** Inner iterator type to iterate over the coefficients of a row or column.
+      * \sa class InnerIterator
+      */
+    typedef Eigen::InnerIterator<Derived> InnerIterator;
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
 
-    /** \brief The type of indices 
-      * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
-      * \sa \ref TopicPreprocessorDirectives.
-      */
-    typedef typename internal::traits<Derived>::Index Index; 
+    /**
+      * \brief The type used to store indices
+      * \details This typedef is relevant for types that store multiple indices such as
+      *          PermutationMatrix or Transpositions, otherwise it defaults to Eigen::Index
+      * \sa \blank \ref TopicPreprocessorDirectives, Eigen::Index, SparseMatrixBase.
+     */
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
 
+    /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc. */
     typedef typename internal::traits<Derived>::Scalar Scalar;
-    typedef typename internal::packet_traits<Scalar>::type PacketScalar;
+    
+    /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc.
+      *
+      * It is an alias for the Scalar type */
+    typedef Scalar value_type;
+    
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef internal::special_scalar_op_base<Derived,Scalar,RealScalar, DenseCoeffsBase<Derived> > Base;
+    typedef DenseCoeffsBase<Derived> Base;
 
-    using Base::operator*;
     using Base::derived;
     using Base::const_cast_derived;
     using Base::rows;
@@ -74,16 +82,6 @@ template<typename Derived> class DenseBase
     using Base::colIndexByOuterInner;
     using Base::coeff;
     using Base::coeffByOuterInner;
-    using Base::packet;
-    using Base::packetByOuterInner;
-    using Base::writePacket;
-    using Base::writePacketByOuterInner;
-    using Base::coeffRef;
-    using Base::coeffRefByOuterInner;
-    using Base::copyCoeff;
-    using Base::copyCoeffByOuterInner;
-    using Base::copyPacket;
-    using Base::copyPacketByOuterInner;
     using Base::operator();
     using Base::operator[];
     using Base::x;
@@ -169,19 +167,46 @@ template<typename Derived> class DenseBase
       InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime)
                              : int(IsRowMajor) ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
 
-      CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
-        /**< This is a rough measure of how expensive it is to read one coefficient from
-          * this expression.
-          */
-
       InnerStrideAtCompileTime = internal::inner_stride_at_compile_time<Derived>::ret,
       OuterStrideAtCompileTime = internal::outer_stride_at_compile_time<Derived>::ret
     };
+    
+    typedef typename internal::find_best_packet<Scalar,SizeAtCompileTime>::type PacketScalar;
 
-    enum { ThisConstantIsPrivateInPlainObjectBase };
+    enum { IsPlainObjectBase = 0 };
+    
+    /** The plain matrix type corresponding to this expression.
+      * \sa PlainObject */
+    typedef Matrix<typename internal::traits<Derived>::Scalar,
+                internal::traits<Derived>::RowsAtCompileTime,
+                internal::traits<Derived>::ColsAtCompileTime,
+                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                internal::traits<Derived>::MaxRowsAtCompileTime,
+                internal::traits<Derived>::MaxColsAtCompileTime
+          > PlainMatrix;
+    
+    /** The plain array type corresponding to this expression.
+      * \sa PlainObject */
+    typedef Array<typename internal::traits<Derived>::Scalar,
+                internal::traits<Derived>::RowsAtCompileTime,
+                internal::traits<Derived>::ColsAtCompileTime,
+                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                internal::traits<Derived>::MaxRowsAtCompileTime,
+                internal::traits<Derived>::MaxColsAtCompileTime
+          > PlainArray;
+
+    /** \brief The plain matrix or array type corresponding to this expression.
+      *
+      * This is not necessarily exactly the return type of eval(). In the case of plain matrices,
+      * the return type of eval() is a const reference to a matrix, not a matrix! It is however guaranteed
+      * that the return type of eval() is either PlainObject or const PlainObject&.
+      */
+    typedef typename internal::conditional<internal::is_same<typename internal::traits<Derived>::XprKind,MatrixXpr >::value,
+                                 PlainMatrix, PlainArray>::type PlainObject;
 
     /** \returns the number of nonzero coefficients which is in practice the number
       * of stored coefficients. */
+    EIGEN_DEVICE_FUNC
     inline Index nonZeros() const { return size(); }
 
     /** \returns the outer size.
@@ -189,6 +214,7 @@ template<typename Derived> class DenseBase
       * \note For a vector, this returns just 1. For a matrix (non-vector), this is the major dimension
       * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of columns for a
       * column-major matrix, and the number of rows for a row-major matrix. */
+    EIGEN_DEVICE_FUNC
     Index outerSize() const
     {
       return IsVectorAtCompileTime ? 1
@@ -200,6 +226,7 @@ template<typename Derived> class DenseBase
       * \note For a vector, this is just the size. For a matrix (non-vector), this is the minor dimension
       * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of rows for a 
       * column-major matrix, and the number of columns for a row-major matrix. */
+    EIGEN_DEVICE_FUNC
     Index innerSize() const
     {
       return IsVectorAtCompileTime ? this->size()
@@ -210,6 +237,7 @@ template<typename Derived> class DenseBase
       * Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does
       * nothing else.
       */
+    EIGEN_DEVICE_FUNC
     void resize(Index newSize)
     {
       EIGEN_ONLY_USED_FOR_DEBUG(newSize);
@@ -220,22 +248,22 @@ template<typename Derived> class DenseBase
       * Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does
       * nothing else.
       */
-    void resize(Index nbRows, Index nbCols)
+    EIGEN_DEVICE_FUNC
+    void resize(Index rows, Index cols)
     {
-      EIGEN_ONLY_USED_FOR_DEBUG(nbRows);
-      EIGEN_ONLY_USED_FOR_DEBUG(nbCols);
-      eigen_assert(nbRows == this->rows() && nbCols == this->cols()
+      EIGEN_ONLY_USED_FOR_DEBUG(rows);
+      EIGEN_ONLY_USED_FOR_DEBUG(cols);
+      eigen_assert(rows == this->rows() && cols == this->cols()
                 && "DenseBase::resize() does not actually allow to resize.");
     }
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
-    /** \internal Represents a vector with linearly spaced coefficients that allows sequential access only. */
-    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,false>,Derived> SequentialLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
+    /** \internal \deprecated Represents a vector with linearly spaced coefficients that allows sequential access only. */
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar>,PlainObject> SequentialLinSpacedReturnType;
     /** \internal Represents a vector with linearly spaced coefficients that allows random access. */
-    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,true>,Derived> RandomAccessLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar>,PlainObject> RandomAccessLinSpacedReturnType;
     /** \internal the return type of MatrixBase::eigenvalues() */
     typedef Matrix<typename NumTraits<typename internal::traits<Derived>::Scalar>::Real, internal::traits<Derived>::ColsAtCompileTime, 1> EigenvaluesReturnType;
 
@@ -243,120 +271,133 @@ template<typename Derived> class DenseBase
 
     /** Copies \a other into *this. \returns a reference to *this. */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase<OtherDerived>& other);
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const EigenBase<OtherDerived> &other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator+=(const EigenBase<OtherDerived> &other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator-=(const EigenBase<OtherDerived> &other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const ReturnByValue<OtherDerived>& func);
 
-    /** \internal Copies \a other into *this without evaluating other. \returns a reference to *this. */
+    /** \internal
+      * Copies \a other into *this without evaluating other. \returns a reference to *this.
+      * \deprecated */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& lazyAssign(const DenseBase<OtherDerived>& other);
 
-    /** \internal Evaluates \a other into *this. \returns a reference to *this. */
-    template<typename OtherDerived>
-    Derived& lazyAssign(const ReturnByValue<OtherDerived>& other);
-
+    EIGEN_DEVICE_FUNC
     CommaInitializer<Derived> operator<< (const Scalar& s);
 
+    /** \deprecated it now returns \c *this */
     template<unsigned int Added,unsigned int Removed>
-    const Flagged<Derived, Added, Removed> flagged() const;
+    EIGEN_DEPRECATED
+    const Derived& flagged() const
+    { return derived(); }
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     CommaInitializer<Derived> operator<< (const DenseBase<OtherDerived>& other);
 
-    Eigen::Transpose<Derived> transpose();
-	typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
+    typedef Transpose<Derived> TransposeReturnType;
+    EIGEN_DEVICE_FUNC
+    TransposeReturnType transpose();
+    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
+    EIGEN_DEVICE_FUNC
     ConstTransposeReturnType transpose() const;
+    EIGEN_DEVICE_FUNC
     void transposeInPlace();
-#ifndef EIGEN_NO_DEBUG
-  protected:
-    template<typename OtherDerived>
-    void checkTransposeAliasing(const OtherDerived& other) const;
-  public:
-#endif
-
 
-    static const ConstantReturnType
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(Index rows, Index cols, const Scalar& value);
-    static const ConstantReturnType
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(Index size, const Scalar& value);
-    static const ConstantReturnType
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(const Scalar& value);
 
-    static const SequentialLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high);
-    static const RandomAccessLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
     LinSpaced(Index size, const Scalar& low, const Scalar& high);
-    static const SequentialLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
-    static const RandomAccessLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
     LinSpaced(const Scalar& low, const Scalar& high);
 
-    template<typename CustomNullaryOp>
-    static const CwiseNullaryOp<CustomNullaryOp, Derived>
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
     NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func);
-    template<typename CustomNullaryOp>
-    static const CwiseNullaryOp<CustomNullaryOp, Derived>
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
     NullaryExpr(Index size, const CustomNullaryOp& func);
-    template<typename CustomNullaryOp>
-    static const CwiseNullaryOp<CustomNullaryOp, Derived>
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
     NullaryExpr(const CustomNullaryOp& func);
 
-    static const ConstantReturnType Zero(Index rows, Index cols);
-    static const ConstantReturnType Zero(Index size);
-    static const ConstantReturnType Zero();
-    static const ConstantReturnType Ones(Index rows, Index cols);
-    static const ConstantReturnType Ones(Index size);
-    static const ConstantReturnType Ones();
-
-    void fill(const Scalar& value);
-    Derived& setConstant(const Scalar& value);
-    Derived& setLinSpaced(Index size, const Scalar& low, const Scalar& high);
-    Derived& setLinSpaced(const Scalar& low, const Scalar& high);
-    Derived& setZero();
-    Derived& setOnes();
-    Derived& setRandom();
-
-    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index size);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero();
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index size);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones();
+
+    EIGEN_DEVICE_FUNC void fill(const Scalar& value);
+    EIGEN_DEVICE_FUNC Derived& setConstant(const Scalar& value);
+    EIGEN_DEVICE_FUNC Derived& setLinSpaced(Index size, const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC Derived& setLinSpaced(const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC Derived& setZero();
+    EIGEN_DEVICE_FUNC Derived& setOnes();
+    EIGEN_DEVICE_FUNC Derived& setRandom();
+
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC
     bool isApprox(const DenseBase<OtherDerived>& other,
                   const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC 
     bool isMuchSmallerThan(const RealScalar& other,
                            const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    template<typename OtherDerived>
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC
     bool isMuchSmallerThan(const DenseBase<OtherDerived>& other,
                            const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
 
-    bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isZero(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isOnes(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isZero(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isOnes(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
     
     inline bool hasNaN() const;
     inline bool allFinite() const;
 
-    inline Derived& operator*=(const Scalar& other);
-    inline Derived& operator/=(const Scalar& other);
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator*=(const Scalar& other);
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator/=(const Scalar& other);
 
     typedef typename internal::add_const_on_value_type<typename internal::eval<Derived>::type>::type EvalReturnType;
     /** \returns the matrix or vector obtained by evaluating this expression.
       *
       * Notice that in the case of a plain matrix or vector (not an expression) this function just returns
       * a const reference, in order to avoid a useless copy.
+      * 
+      * \warning Be carefull with eval() and the auto C++ keyword, as detailed in this \link TopicPitfalls_auto_keyword page \endlink.
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE EvalReturnType eval() const
     {
       // Even though MSVC does not honor strong inlining when the return type
@@ -364,61 +405,78 @@ template<typename Derived> class DenseBase
       // size types on MSVC.
       return typename internal::eval<Derived>::type(derived());
     }
-
+    
     /** swaps *this with the expression \a other.
       *
       */
     template<typename OtherDerived>
-    void swap(const DenseBase<OtherDerived>& other,
-              int = OtherDerived::ThisConstantIsPrivateInPlainObjectBase)
+    EIGEN_DEVICE_FUNC
+    void swap(const DenseBase<OtherDerived>& other)
     {
-      SwapWrapper<Derived>(derived()).lazyAssign(other.derived());
+      EIGEN_STATIC_ASSERT(!OtherDerived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      eigen_assert(rows()==other.rows() && cols()==other.cols());
+      call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
     }
 
     /** swaps *this with the matrix or array \a other.
       *
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     void swap(PlainObjectBase<OtherDerived>& other)
     {
-      SwapWrapper<Derived>(derived()).lazyAssign(other.derived());
+      eigen_assert(rows()==other.rows() && cols()==other.cols());
+      call_assignment(derived(), other.derived(), internal::swap_assign_op<Scalar>());
     }
 
+    EIGEN_DEVICE_FUNC inline const NestByValue<Derived> nestByValue() const;
+    EIGEN_DEVICE_FUNC inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
+    EIGEN_DEVICE_FUNC inline ForceAlignedAccess<Derived> forceAlignedAccess();
+    template<bool Enable> EIGEN_DEVICE_FUNC
+    inline const typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf() const;
+    template<bool Enable> EIGEN_DEVICE_FUNC
+    inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
 
-    inline const NestByValue<Derived> nestByValue() const;
-    inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
-    inline ForceAlignedAccess<Derived> forceAlignedAccess();
-    template<bool Enable> inline const typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf() const;
-    template<bool Enable> inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
-
-    Scalar sum() const;
-    Scalar mean() const;
-    Scalar trace() const;
+    EIGEN_DEVICE_FUNC Scalar sum() const;
+    EIGEN_DEVICE_FUNC Scalar mean() const;
+    EIGEN_DEVICE_FUNC Scalar trace() const;
 
-    Scalar prod() const;
+    EIGEN_DEVICE_FUNC Scalar prod() const;
 
-    typename internal::traits<Derived>::Scalar minCoeff() const;
-    typename internal::traits<Derived>::Scalar maxCoeff() const;
+    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar minCoeff() const;
+    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar maxCoeff() const;
 
-    template<typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const;
-    template<typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const;
-    template<typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const;
-    template<typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const;
 
     template<typename BinaryOp>
-    typename internal::result_of<BinaryOp(typename internal::traits<Derived>::Scalar)>::type
-    redux(const BinaryOp& func) const;
+    EIGEN_DEVICE_FUNC
+    Scalar redux(const BinaryOp& func) const;
 
     template<typename Visitor>
+    EIGEN_DEVICE_FUNC
     void visit(Visitor& func) const;
 
-    inline const WithFormat<Derived> format(const IOFormat& fmt) const;
+    /** \returns a WithFormat proxy object allowing to print a matrix the with given
+      * format \a fmt.
+      *
+      * See class IOFormat for some examples.
+      *
+      * \sa class IOFormat, class WithFormat
+      */
+    inline const WithFormat<Derived> format(const IOFormat& fmt) const
+    {
+      return WithFormat<Derived>(derived(), fmt);
+    }
 
     /** \returns the unique coefficient of a 1x1 expression */
+    EIGEN_DEVICE_FUNC
     CoeffReturnType value() const
     {
       EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
@@ -426,23 +484,44 @@ template<typename Derived> class DenseBase
       return derived().coeff(0,0);
     }
 
-    bool all(void) const;
-    bool any(void) const;
-    Index count() const;
+    EIGEN_DEVICE_FUNC bool all() const;
+    EIGEN_DEVICE_FUNC bool any() const;
+    EIGEN_DEVICE_FUNC Index count() const;
 
     typedef VectorwiseOp<Derived, Horizontal> RowwiseReturnType;
     typedef const VectorwiseOp<const Derived, Horizontal> ConstRowwiseReturnType;
     typedef VectorwiseOp<Derived, Vertical> ColwiseReturnType;
     typedef const VectorwiseOp<const Derived, Vertical> ConstColwiseReturnType;
 
-    ConstRowwiseReturnType rowwise() const;
-    RowwiseReturnType rowwise();
-    ConstColwiseReturnType colwise() const;
-    ColwiseReturnType colwise();
+    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
+    *
+    * Example: \include MatrixBase_rowwise.cpp
+    * Output: \verbinclude MatrixBase_rowwise.out
+    *
+    * \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
+    */
+    //Code moved here due to a CUDA compiler bug
+    EIGEN_DEVICE_FUNC inline ConstRowwiseReturnType rowwise() const {
+      return ConstRowwiseReturnType(derived());
+    }
+    EIGEN_DEVICE_FUNC RowwiseReturnType rowwise();
+
+    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
+    *
+    * Example: \include MatrixBase_colwise.cpp
+    * Output: \verbinclude MatrixBase_colwise.out
+    *
+    * \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
+    */
+    EIGEN_DEVICE_FUNC inline ConstColwiseReturnType colwise() const {
+      return ConstColwiseReturnType(derived());
+    }
+    EIGEN_DEVICE_FUNC ColwiseReturnType colwise();
 
-    static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random(Index rows, Index cols);
-    static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random(Index size);
-    static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random();
+    typedef CwiseNullaryOp<internal::scalar_random_op<Scalar>,PlainObject> RandomReturnType;
+    static const RandomReturnType Random(Index rows, Index cols);
+    static const RandomReturnType Random(Index size);
+    static const RandomReturnType Random();
 
     template<typename ThenDerived,typename ElseDerived>
     const Select<Derived,ThenDerived,ElseDerived>
@@ -460,45 +539,56 @@ template<typename Derived> class DenseBase
     template<int p> RealScalar lpNorm() const;
 
     template<int RowFactor, int ColFactor>
-    inline const Replicate<Derived,RowFactor,ColFactor> replicate() const;
-    
-    typedef Replicate<Derived,Dynamic,Dynamic> ReplicateReturnType;
-    inline const ReplicateReturnType replicate(Index rowFacor,Index colFactor) const;
+    EIGEN_DEVICE_FUNC
+    const Replicate<Derived,RowFactor,ColFactor> replicate() const;
+    /**
+    * \return an expression of the replication of \c *this
+    *
+    * Example: \include MatrixBase_replicate_int_int.cpp
+    * Output: \verbinclude MatrixBase_replicate_int_int.out
+    *
+    * \sa VectorwiseOp::replicate(), DenseBase::replicate<int,int>(), class Replicate
+    */
+    //Code moved here due to a CUDA compiler bug
+    EIGEN_DEVICE_FUNC
+    const Replicate<Derived, Dynamic, Dynamic> replicate(Index rowFactor, Index colFactor) const
+    {
+      return Replicate<Derived, Dynamic, Dynamic>(derived(), rowFactor, colFactor);
+    }
 
     typedef Reverse<Derived, BothDirections> ReverseReturnType;
     typedef const Reverse<const Derived, BothDirections> ConstReverseReturnType;
-    ReverseReturnType reverse();
-    ConstReverseReturnType reverse() const;
-    void reverseInPlace();
+    EIGEN_DEVICE_FUNC ReverseReturnType reverse();
+    /** This is the const version of reverse(). */
+    //Code moved here due to a CUDA compiler bug
+    EIGEN_DEVICE_FUNC ConstReverseReturnType reverse() const
+    {
+      return ConstReverseReturnType(derived());
+    }
+    EIGEN_DEVICE_FUNC void reverseInPlace();
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::DenseBase
+#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND)
 #   include "../plugins/BlockMethods.h"
 #   ifdef EIGEN_DENSEBASE_PLUGIN
 #     include EIGEN_DENSEBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
-
-#ifdef EIGEN2_SUPPORT
-
-    Block<Derived> corner(CornerType type, Index cRows, Index cCols);
-    const Block<Derived> corner(CornerType type, Index cRows, Index cCols) const;
-    template<int CRows, int CCols>
-    Block<Derived, CRows, CCols> corner(CornerType type);
-    template<int CRows, int CCols>
-    const Block<Derived, CRows, CCols> corner(CornerType type) const;
-
-#endif // EIGEN2_SUPPORT
-
+#undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF
 
     // disable the use of evalTo for dense objects with a nice compilation error
-    template<typename Dest> inline void evalTo(Dest& ) const
+    template<typename Dest>
+    EIGEN_DEVICE_FUNC
+    inline void evalTo(Dest& ) const
     {
       EIGEN_STATIC_ASSERT((internal::is_same<Dest,void>::value),THE_EVAL_EVALTO_FUNCTION_SHOULD_NEVER_BE_CALLED_FOR_DENSE_OBJECTS);
     }
 
   protected:
     /** Default constructor. Do nothing. */
-    DenseBase()
+    EIGEN_DEVICE_FUNC DenseBase()
     {
       /* Just checks for self-consistency of the flags.
        * Only do it when debugging Eigen, as this borders on paranoiac and could slow compilation down
@@ -511,9 +601,9 @@ template<typename Derived> class DenseBase
     }
 
   private:
-    explicit DenseBase(int);
-    DenseBase(int,int);
-    template<typename OtherDerived> explicit DenseBase(const DenseBase<OtherDerived>&);
+    EIGEN_DEVICE_FUNC explicit DenseBase(int);
+    EIGEN_DEVICE_FUNC DenseBase(int,int);
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit DenseBase(const DenseBase<OtherDerived>&);
 };
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/DenseCoeffsBase.h b/extern/Eigen3/Eigen/src/Core/DenseCoeffsBase.h
index 3c890f21590..c4af48ab699 100644
--- a/extern/Eigen3/Eigen/src/Core/DenseCoeffsBase.h
+++ b/extern/Eigen3/Eigen/src/Core/DenseCoeffsBase.h
@@ -35,7 +35,6 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
 {
   public:
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
 
@@ -61,6 +60,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
     using Base::size;
     using Base::derived;
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner) const
     {
       return int(Derived::RowsAtCompileTime) == 1 ? 0
@@ -69,6 +69,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
           : inner;
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner) const
     {
       return int(Derived::ColsAtCompileTime) == 1 ? 0
@@ -91,13 +92,15 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       *
       * \sa operator()(Index,Index) const, coeffRef(Index,Index), coeff(Index) const
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
     {
       eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      return derived().coeff(row, col);
+                         && col >= 0 && col < cols());
+      return internal::evaluator<Derived>(derived()).coeff(row,col);
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
     {
       return coeff(rowIndexByOuterInner(outer, inner),
@@ -108,11 +111,12 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       *
       * \sa operator()(Index,Index), operator[](Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType operator()(Index row, Index col) const
     {
       eigen_assert(row >= 0 && row < rows()
           && col >= 0 && col < cols());
-      return derived().coeff(row, col);
+      return coeff(row, col);
     }
 
     /** Short version: don't use this function, use
@@ -130,11 +134,14 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       * \sa operator[](Index) const, coeffRef(Index), coeff(Index,Index) const
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     coeff(Index index) const
     {
+      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
+                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
       eigen_internal_assert(index >= 0 && index < size());
-      return derived().coeff(index);
+      return internal::evaluator<Derived>(derived()).coeff(index);
     }
 
 
@@ -146,15 +153,14 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       * z() const, w() const
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     operator[](Index index) const
     {
-      #ifndef EIGEN2_SUPPORT
       EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
                           THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
-      #endif
       eigen_assert(index >= 0 && index < size());
-      return derived().coeff(index);
+      return coeff(index);
     }
 
     /** \returns the coefficient at given index.
@@ -167,32 +173,49 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       * z() const, w() const
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     operator()(Index index) const
     {
       eigen_assert(index >= 0 && index < size());
-      return derived().coeff(index);
+      return coeff(index);
     }
 
     /** equivalent to operator[](0).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     x() const { return (*this)[0]; }
 
     /** equivalent to operator[](1).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
-    y() const { return (*this)[1]; }
+    y() const
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
+      return (*this)[1];
+    }
 
     /** equivalent to operator[](2).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
-    z() const { return (*this)[2]; }
+    z() const
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
+      return (*this)[2];
+    }
 
     /** equivalent to operator[](3).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
-    w() const { return (*this)[3]; }
+    w() const
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
+      return (*this)[3];
+    }
 
     /** \internal
       * \returns the packet of coefficients starting at the given row and column. It is your responsibility
@@ -207,9 +230,9 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
     template<int LoadMode>
     EIGEN_STRONG_INLINE PacketReturnType packet(Index row, Index col) const
     {
-      eigen_internal_assert(row >= 0 && row < rows()
-                      && col >= 0 && col < cols());
-      return derived().template packet<LoadMode>(row,col);
+      typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
+      eigen_internal_assert(row >= 0 && row < rows() && col >= 0 && col < cols());
+      return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(row,col);
     }
 
 
@@ -234,8 +257,11 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
     template<int LoadMode>
     EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
     {
+      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
+                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
+      typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
       eigen_internal_assert(index >= 0 && index < size());
-      return derived().template packet<LoadMode>(index);
+      return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(index);
     }
 
   protected:
@@ -278,7 +304,6 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
     typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -311,13 +336,15 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       *
       * \sa operator()(Index,Index), coeff(Index, Index) const, coeffRef(Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
     {
       eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      return derived().coeffRef(row, col);
+                         && col >= 0 && col < cols());
+      return internal::evaluator<Derived>(derived()).coeffRef(row,col);
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     coeffRefByOuterInner(Index outer, Index inner)
     {
@@ -330,12 +357,13 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       * \sa operator[](Index)
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     operator()(Index row, Index col)
     {
       eigen_assert(row >= 0 && row < rows()
           && col >= 0 && col < cols());
-      return derived().coeffRef(row, col);
+      return coeffRef(row, col);
     }
 
 
@@ -354,11 +382,14 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       * \sa operator[](Index), coeff(Index) const, coeffRef(Index,Index)
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     coeffRef(Index index)
     {
+      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
+                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
       eigen_internal_assert(index >= 0 && index < size());
-      return derived().coeffRef(index);
+      return internal::evaluator<Derived>(derived()).coeffRef(index);
     }
 
     /** \returns a reference to the coefficient at given index.
@@ -368,15 +399,14 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       * \sa operator[](Index) const, operator()(Index,Index), x(), y(), z(), w()
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     operator[](Index index)
     {
-      #ifndef EIGEN2_SUPPORT
       EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
                           THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
-      #endif
       eigen_assert(index >= 0 && index < size());
-      return derived().coeffRef(index);
+      return coeffRef(index);
     }
 
     /** \returns a reference to the coefficient at given index.
@@ -388,167 +418,49 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       * \sa operator[](Index) const, operator()(Index,Index), x(), y(), z(), w()
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     operator()(Index index)
     {
       eigen_assert(index >= 0 && index < size());
-      return derived().coeffRef(index);
+      return coeffRef(index);
     }
 
     /** equivalent to operator[](0).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     x() { return (*this)[0]; }
 
     /** equivalent to operator[](1).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
-    y() { return (*this)[1]; }
-
-    /** equivalent to operator[](2).  */
-
-    EIGEN_STRONG_INLINE Scalar&
-    z() { return (*this)[2]; }
-
-    /** equivalent to operator[](3).  */
-
-    EIGEN_STRONG_INLINE Scalar&
-    w() { return (*this)[3]; }
-
-    /** \internal
-      * Stores the given packet of coefficients, at the given row and column of this expression. It is your responsibility
-      * to ensure that a packet really starts there. This method is only available on expressions having the
-      * PacketAccessBit.
-      *
-      * The \a LoadMode parameter may have the value \a #Aligned or \a #Unaligned. Its effect is to select
-      * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
-      * starting at an address which is a multiple of the packet size.
-      */
-
-    template<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacket
-    (Index row, Index col, const typename internal::packet_traits<Scalar>::type& val)
+    y()
     {
-      eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      derived().template writePacket<StoreMode>(row,col,val);
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
+      return (*this)[1];
     }
 
+    /** equivalent to operator[](2).  */
 
-    /** \internal */
-    template<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacketByOuterInner
-    (Index outer, Index inner, const typename internal::packet_traits<Scalar>::type& val)
-    {
-      writePacket<StoreMode>(rowIndexByOuterInner(outer, inner),
-                            colIndexByOuterInner(outer, inner),
-                            val);
-    }
-
-    /** \internal
-      * Stores the given packet of coefficients, at the given index in this expression. It is your responsibility
-      * to ensure that a packet really starts there. This method is only available on expressions having the
-      * PacketAccessBit and the LinearAccessBit.
-      *
-      * The \a LoadMode parameter may have the value \a Aligned or \a Unaligned. Its effect is to select
-      * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
-      * starting at an address which is a multiple of the packet size.
-      */
-    template<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacket
-    (Index index, const typename internal::packet_traits<Scalar>::type& val)
-    {
-      eigen_internal_assert(index >= 0 && index < size());
-      derived().template writePacket<StoreMode>(index,val);
-    }
-
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-
-    /** \internal Copies the coefficient at position (row,col) of other into *this.
-      *
-      * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-      * with usual assignments.
-      *
-      * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-      */
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void copyCoeff(Index row, Index col, const DenseBase<OtherDerived>& other)
-    {
-      eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      derived().coeffRef(row, col) = other.derived().coeff(row, col);
-    }
-
-    /** \internal Copies the coefficient at the given index of other into *this.
-      *
-      * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-      * with usual assignments.
-      *
-      * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-      */
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
-    {
-      eigen_internal_assert(index >= 0 && index < size());
-      derived().coeffRef(index) = other.derived().coeff(index);
-    }
-
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void copyCoeffByOuterInner(Index outer, Index inner, const DenseBase<OtherDerived>& other)
-    {
-      const Index row = rowIndexByOuterInner(outer,inner);
-      const Index col = colIndexByOuterInner(outer,inner);
-      // derived() is important here: copyCoeff() may be reimplemented in Derived!
-      derived().copyCoeff(row, col, other);
-    }
-
-    /** \internal Copies the packet at position (row,col) of other into *this.
-      *
-      * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-      * with usual assignments.
-      *
-      * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-      */
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    EIGEN_STRONG_INLINE void copyPacket(Index row, Index col, const DenseBase<OtherDerived>& other)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    z()
     {
-      eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      derived().template writePacket<StoreMode>(row, col,
-        other.derived().template packet<LoadMode>(row, col));
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
+      return (*this)[2];
     }
 
-    /** \internal Copies the packet at the given index of other into *this.
-      *
-      * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-      * with usual assignments.
-      *
-      * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-      */
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    EIGEN_STRONG_INLINE void copyPacket(Index index, const DenseBase<OtherDerived>& other)
-    {
-      eigen_internal_assert(index >= 0 && index < size());
-      derived().template writePacket<StoreMode>(index,
-        other.derived().template packet<LoadMode>(index));
-    }
+    /** equivalent to operator[](3).  */
 
-    /** \internal */
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    EIGEN_STRONG_INLINE void copyPacketByOuterInner(Index outer, Index inner, const DenseBase<OtherDerived>& other)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    w()
     {
-      const Index row = rowIndexByOuterInner(outer,inner);
-      const Index col = colIndexByOuterInner(outer,inner);
-      // derived() is important here: copyCoeff() may be reimplemented in Derived!
-      derived().template copyPacket< OtherDerived, StoreMode, LoadMode>(row, col, other);
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
+      return (*this)[3];
     }
-#endif
-
 };
 
 /** \brief Base class providing direct read-only coefficient access to matrices and arrays.
@@ -560,7 +472,7 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
   * inherits DenseCoeffsBase<Derived, ReadOnlyAccessors> which defines functions to access entries read-only using
   * \c operator() .
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived>
 class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived, ReadOnlyAccessors>
@@ -568,7 +480,6 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
   public:
 
     typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
@@ -581,6 +492,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa outerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return derived().innerStride();
@@ -591,6 +503,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return derived().outerStride();
@@ -606,6 +519,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
@@ -615,6 +529,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
@@ -630,7 +545,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
   * inherits DenseCoeffsBase<Derived, WriteAccessors> which defines functions to access entries read/write using
   * \c operator().
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived>
 class DenseCoeffsBase<Derived, DirectWriteAccessors>
@@ -639,7 +554,6 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
   public:
 
     typedef DenseCoeffsBase<Derived, WriteAccessors> Base;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
@@ -652,6 +566,7 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa outerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return derived().innerStride();
@@ -662,6 +577,7 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return derived().outerStride();
@@ -677,6 +593,7 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), outerStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
@@ -686,6 +603,7 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
@@ -694,33 +612,42 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
 
 namespace internal {
 
-template<typename Derived, bool JustReturnZero>
+template<int Alignment, typename Derived, bool JustReturnZero>
 struct first_aligned_impl
 {
-  static inline typename Derived::Index run(const Derived&)
+  static inline Index run(const Derived&)
   { return 0; }
 };
 
-template<typename Derived>
-struct first_aligned_impl<Derived, false>
+template<int Alignment, typename Derived>
+struct first_aligned_impl<Alignment, Derived, false>
 {
-  static inline typename Derived::Index run(const Derived& m)
+  static inline Index run(const Derived& m)
   {
-    return internal::first_aligned(&m.const_cast_derived().coeffRef(0,0), m.size());
+    return internal::first_aligned<Alignment>(m.data(), m.size());
   }
 };
 
-/** \internal \returns the index of the first element of the array that is well aligned for vectorization.
+/** \internal \returns the index of the first element of the array stored by \a m that is properly aligned with respect to \a Alignment for vectorization.
+  *
+  * \tparam Alignment requested alignment in Bytes.
   *
   * There is also the variant first_aligned(const Scalar*, Integer) defined in Memory.h. See it for more
   * documentation.
   */
+template<int Alignment, typename Derived>
+static inline Index first_aligned(const DenseBase<Derived>& m)
+{
+  enum { ReturnZero = (int(evaluator<Derived>::Alignment) >= Alignment) || !(Derived::Flags & DirectAccessBit) };
+  return first_aligned_impl<Alignment, Derived, ReturnZero>::run(m.derived());
+}
+
 template<typename Derived>
-static inline typename Derived::Index first_aligned(const Derived& m)
+static inline Index first_default_aligned(const DenseBase<Derived>& m)
 {
-  return first_aligned_impl
-          <Derived, (Derived::Flags & AlignedBit) || !(Derived::Flags & DirectAccessBit)>
-          ::run(m);
+  typedef typename Derived::Scalar Scalar;
+  typedef typename packet_traits<Scalar>::type DefaultPacketType;
+  return internal::first_aligned<int(unpacket_traits<DefaultPacketType>::alignment),Derived>(m);
 }
 
 template<typename Derived, bool HasDirectAccess = has_direct_access<Derived>::ret>
diff --git a/extern/Eigen3/Eigen/src/Core/DenseStorage.h b/extern/Eigen3/Eigen/src/Core/DenseStorage.h
index 568493cbae0..7958feeb9c0 100644
--- a/extern/Eigen3/Eigen/src/Core/DenseStorage.h
+++ b/extern/Eigen3/Eigen/src/Core/DenseStorage.h
@@ -3,7 +3,7 @@
 //
 // Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2010 Hauke Heibel <hauke.heibel@gmail.com>
+// Copyright (C) 2010-2013 Hauke Heibel <hauke.heibel@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -13,9 +13,9 @@
 #define EIGEN_MATRIXSTORAGE_H
 
 #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN EIGEN_DENSE_STORAGE_CTOR_PLUGIN;
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X) X; EIGEN_DENSE_STORAGE_CTOR_PLUGIN;
 #else
-  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X)
 #endif
 
 namespace Eigen {
@@ -24,7 +24,9 @@ namespace internal {
 
 struct constructor_without_unaligned_array_assert {};
 
-template<typename T, int Size> void check_static_allocation_size()
+template<typename T, int Size>
+EIGEN_DEVICE_FUNC
+void check_static_allocation_size()
 {
   // if EIGEN_STACK_ALLOCATION_LIMIT is defined to 0, then no limit
   #if EIGEN_STACK_ALLOCATION_LIMIT
@@ -38,18 +40,19 @@ template<typename T, int Size> void check_static_allocation_size()
   */
 template <typename T, int Size, int MatrixOrArrayOptions,
           int Alignment = (MatrixOrArrayOptions&DontAlign) ? 0
-                        : (((Size*sizeof(T))%16)==0) ? 16
-                        : 0 >
+                        : compute_default_alignment<T,Size>::value >
 struct plain_array
 {
   T array[Size];
 
-  plain_array() 
+  EIGEN_DEVICE_FUNC
+  plain_array()
   { 
     check_static_allocation_size<T,Size>();
   }
 
-  plain_array(constructor_without_unaligned_array_assert) 
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert)
   { 
     check_static_allocation_size<T,Size>();
   }
@@ -64,29 +67,88 @@ struct plain_array
   template<typename PtrType>
   EIGEN_ALWAYS_INLINE PtrType eigen_unaligned_array_assert_workaround_gcc47(PtrType array) { return array; }
   #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
-    eigen_assert((reinterpret_cast<size_t>(eigen_unaligned_array_assert_workaround_gcc47(array)) & sizemask) == 0 \
+    eigen_assert((internal::UIntPtr(eigen_unaligned_array_assert_workaround_gcc47(array)) & (sizemask)) == 0 \
               && "this assertion is explained here: " \
               "http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \
               " **** READ THIS WEB PAGE !!! ****");
 #else
   #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
-    eigen_assert((reinterpret_cast<size_t>(array) & sizemask) == 0 \
+    eigen_assert((internal::UIntPtr(array) & (sizemask)) == 0 \
               && "this assertion is explained here: " \
               "http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \
               " **** READ THIS WEB PAGE !!! ****");
 #endif
 
 template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 8>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(8) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  {
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(7);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int Size, int MatrixOrArrayOptions>
 struct plain_array<T, Size, MatrixOrArrayOptions, 16>
 {
-  EIGEN_USER_ALIGN16 T array[Size];
+  EIGEN_ALIGN_TO_BOUNDARY(16) T array[Size];
 
+  EIGEN_DEVICE_FUNC
   plain_array() 
   { 
-    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(0xf);
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(15);
     check_static_allocation_size<T,Size>();
   }
 
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 32>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(32) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  {
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(31);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 64>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(64) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  { 
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(63);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
   plain_array(constructor_without_unaligned_array_assert) 
   { 
     check_static_allocation_size<T,Size>();
@@ -96,9 +158,9 @@ struct plain_array<T, Size, MatrixOrArrayOptions, 16>
 template <typename T, int MatrixOrArrayOptions, int Alignment>
 struct plain_array<T, 0, MatrixOrArrayOptions, Alignment>
 {
-  EIGEN_USER_ALIGN16 T array[1];
-  plain_array() {}
-  plain_array(constructor_without_unaligned_array_assert) {}
+  T array[1];
+  EIGEN_DEVICE_FUNC plain_array() {}
+  EIGEN_DEVICE_FUNC plain_array(constructor_without_unaligned_array_assert) {}
 };
 
 } // end namespace internal
@@ -122,41 +184,54 @@ template<typename T, int Size, int _Rows, int _Cols, int _Options> class DenseSt
 {
     internal::plain_array<T,Size,_Options> m_data;
   public:
-    DenseStorage() {}
-    DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
+    }
+    EIGEN_DEVICE_FUNC
+    explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()) {}
-    DenseStorage(const DenseStorage& other) : m_data(other.m_data) {}
+    EIGEN_DEVICE_FUNC 
+    DenseStorage(const DenseStorage& other) : m_data(other.m_data) {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
+    }
+    EIGEN_DEVICE_FUNC 
     DenseStorage& operator=(const DenseStorage& other)
-    {
+    { 
       if (this != &other) m_data = other.m_data;
-      return *this;
+      return *this; 
     }
-    DenseStorage(DenseIndex,DenseIndex,DenseIndex) {}
-    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); }
-    static DenseIndex rows(void) {return _Rows;}
-    static DenseIndex cols(void) {return _Cols;}
-    void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {}
-    void resize(DenseIndex,DenseIndex,DenseIndex) {}
-    const T *data() const { return m_data.array; }
-    T *data() { return m_data.array; }
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows==_Rows && cols==_Cols);
+      EIGEN_UNUSED_VARIABLE(size);
+      EIGEN_UNUSED_VARIABLE(rows);
+      EIGEN_UNUSED_VARIABLE(cols);
+    }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); }
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
 };
 
 // null matrix
 template<typename T, int _Rows, int _Cols, int _Options> class DenseStorage<T, 0, _Rows, _Cols, _Options>
 {
   public:
-    DenseStorage() {}
-    DenseStorage(internal::constructor_without_unaligned_array_assert) {}
-    DenseStorage(const DenseStorage&) {}
-    DenseStorage& operator=(const DenseStorage&) { return *this; }
-    DenseStorage(DenseIndex,DenseIndex,DenseIndex) {}
-    void swap(DenseStorage& ) {}
-    static DenseIndex rows(void) {return _Rows;}
-    static DenseIndex cols(void) {return _Cols;}
-    void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {}
-    void resize(DenseIndex,DenseIndex,DenseIndex) {}
-    const T *data() const { return 0; }
-    T *data() { return 0; }
+    EIGEN_DEVICE_FUNC DenseStorage() {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) {}
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage&) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) { return *this; }
+    EIGEN_DEVICE_FUNC DenseStorage(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& ) {}
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC const T *data() const { return 0; }
+    EIGEN_DEVICE_FUNC T *data() { return 0; }
 };
 
 // more specializations for null matrices; these are necessary to resolve ambiguities
@@ -173,74 +248,74 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, 0, Dynamic,
 template<typename T, int Size, int _Options> class DenseStorage<T, Size, Dynamic, Dynamic, _Options>
 {
     internal::plain_array<T,Size,_Options> m_data;
-    DenseIndex m_rows;
-    DenseIndex m_cols;
+    Index m_rows;
+    Index m_cols;
   public:
-    DenseStorage() : m_rows(0), m_cols(0) {}
-    DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0), m_cols(0) {}
-    DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows), m_cols(other.m_cols) {}
-    DenseStorage& operator=(const DenseStorage& other)
-    {
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows), m_cols(other.m_cols) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) 
+    { 
       if (this != &other)
       {
         m_data = other.m_data;
         m_rows = other.m_rows;
         m_cols = other.m_cols;
       }
-      return *this;
+      return *this; 
     }
-    DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) : m_rows(nbRows), m_cols(nbCols) {}
-    void swap(DenseStorage& other)
+    EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index cols) : m_rows(rows), m_cols(cols) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
     { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-    DenseIndex rows() const {return m_rows;}
-    DenseIndex cols() const {return m_cols;}
-    void conservativeResize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; }
-    void resize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; }
-    const T *data() const { return m_data.array; }
-    T *data() { return m_data.array; }
+    EIGEN_DEVICE_FUNC Index rows() const {return m_rows;}
+    EIGEN_DEVICE_FUNC Index cols() const {return m_cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
+    EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
 };
 
 // dynamic-size matrix with fixed-size storage and fixed width
 template<typename T, int Size, int _Cols, int _Options> class DenseStorage<T, Size, Dynamic, _Cols, _Options>
 {
     internal::plain_array<T,Size,_Options> m_data;
-    DenseIndex m_rows;
+    Index m_rows;
   public:
-    DenseStorage() : m_rows(0) {}
-    DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0) {}
-    DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows) {}
-    DenseStorage& operator=(const DenseStorage& other)
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) 
     {
       if (this != &other)
       {
         m_data = other.m_data;
         m_rows = other.m_rows;
       }
-      return *this;
+      return *this; 
     }
-    DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex) : m_rows(nbRows) {}
-    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-    DenseIndex rows(void) const {return m_rows;}
-    DenseIndex cols(void) const {return _Cols;}
-    void conservativeResize(DenseIndex, DenseIndex nbRows, DenseIndex) { m_rows = nbRows; }
-    void resize(DenseIndex, DenseIndex nbRows, DenseIndex) { m_rows = nbRows; }
-    const T *data() const { return m_data.array; }
-    T *data() { return m_data.array; }
+    EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index) : m_rows(rows) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return _Cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index) { m_rows = rows; }
+    EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index) { m_rows = rows; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
 };
 
 // dynamic-size matrix with fixed-size storage and fixed height
 template<typename T, int Size, int _Rows, int _Options> class DenseStorage<T, Size, _Rows, Dynamic, _Options>
 {
     internal::plain_array<T,Size,_Options> m_data;
-    DenseIndex m_cols;
+    Index m_cols;
   public:
-    DenseStorage() : m_cols(0) {}
-    DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_cols(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()), m_cols(0) {}
-    DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_cols(other.m_cols) {}
-    DenseStorage& operator=(const DenseStorage& other)
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_cols(other.m_cols) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
     {
       if (this != &other)
       {
@@ -249,38 +324,62 @@ template<typename T, int Size, int _Rows, int _Options> class DenseStorage<T, Si
       }
       return *this;
     }
-    DenseStorage(DenseIndex, DenseIndex, DenseIndex nbCols) : m_cols(nbCols) {}
-    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-    DenseIndex rows(void) const {return _Rows;}
-    DenseIndex cols(void) const {return m_cols;}
-    void conservativeResize(DenseIndex, DenseIndex, DenseIndex nbCols) { m_cols = nbCols; }
-    void resize(DenseIndex, DenseIndex, DenseIndex nbCols) { m_cols = nbCols; }
-    const T *data() const { return m_data.array; }
-    T *data() { return m_data.array; }
+    EIGEN_DEVICE_FUNC DenseStorage(Index, Index, Index cols) : m_cols(cols) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return _Rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
+    void conservativeResize(Index, Index, Index cols) { m_cols = cols; }
+    void resize(Index, Index, Index cols) { m_cols = cols; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
 };
 
 // purely dynamic matrix.
 template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynamic, _Options>
 {
     T *m_data;
-    DenseIndex m_rows;
-    DenseIndex m_cols;
+    Index m_rows;
+    Index m_cols;
   public:
-    DenseStorage() : m_data(0), m_rows(0), m_cols(0) {}
-    DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
        : m_data(0), m_rows(0), m_cols(0) {}
-    DenseStorage(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols)
-      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(nbRows), m_cols(nbCols)
-    { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN }
-#ifdef EIGEN_HAVE_RVALUE_REFERENCES
-    DenseStorage(DenseStorage&& other)
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows), m_cols(cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows>=0 && cols >=0);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*other.m_cols))
+      , m_rows(other.m_rows)
+      , m_cols(other.m_cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*m_cols)
+      internal::smart_copy(other.m_data, other.m_data+other.m_rows*other.m_cols, m_data);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        DenseStorage tmp(other);
+        this->swap(tmp);
+      }
+      return *this;
+    }
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
       : m_data(std::move(other.m_data))
       , m_rows(std::move(other.m_rows))
       , m_cols(std::move(other.m_cols))
     {
       other.m_data = nullptr;
+      other.m_rows = 0;
+      other.m_cols = 0;
     }
-    DenseStorage& operator=(DenseStorage&& other)
+    EIGEN_DEVICE_FUNC
+    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
     {
       using std::swap;
       swap(m_data, other.m_data);
@@ -289,18 +388,18 @@ template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynam
       return *this;
     }
 #endif
-    ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
-    void swap(DenseStorage& other)
+    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
     { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-    DenseIndex rows(void) const {return m_rows;}
-    DenseIndex cols(void) const {return m_cols;}
-    void conservativeResize(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols)
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
+    void conservativeResize(Index size, Index rows, Index cols)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*m_cols);
-      m_rows = nbRows;
-      m_cols = nbCols;
+      m_rows = rows;
+      m_cols = cols;
     }
-    void resize(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols)
+    EIGEN_DEVICE_FUNC void resize(Index size, Index rows, Index cols)
     {
       if(size != m_rows*m_cols)
       {
@@ -309,36 +408,56 @@ template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynam
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       }
-      m_rows = nbRows;
-      m_cols = nbCols;
+      m_rows = rows;
+      m_cols = cols;
     }
-    const T *data() const { return m_data; }
-    T *data() { return m_data; }
-  private:
-    DenseStorage(const DenseStorage&);
-    DenseStorage& operator=(const DenseStorage&);
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data; }
 };
 
 // matrix with dynamic width and fixed height (so that matrix has dynamic size).
 template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Rows, Dynamic, _Options>
 {
     T *m_data;
-    DenseIndex m_cols;
+    Index m_cols;
   public:
-    DenseStorage() : m_data(0), m_cols(0) {}
-    DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
-    DenseStorage(DenseIndex size, DenseIndex, DenseIndex nbCols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_cols(nbCols)
-    { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN }
-#ifdef EIGEN_HAVE_RVALUE_REFERENCES
-    DenseStorage(DenseStorage&& other)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_cols(0) {}
+    explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_cols(cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows==_Rows && cols >=0);
+      EIGEN_UNUSED_VARIABLE(rows);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(_Rows*other.m_cols))
+      , m_cols(other.m_cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_cols*_Rows)
+      internal::smart_copy(other.m_data, other.m_data+_Rows*m_cols, m_data);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        DenseStorage tmp(other);
+        this->swap(tmp);
+      }
+      return *this;
+    }    
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
       : m_data(std::move(other.m_data))
       , m_cols(std::move(other.m_cols))
     {
       other.m_data = nullptr;
+      other.m_cols = 0;
     }
-    DenseStorage& operator=(DenseStorage&& other)
+    EIGEN_DEVICE_FUNC
+    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
     {
       using std::swap;
       swap(m_data, other.m_data);
@@ -346,16 +465,16 @@ template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Ro
       return *this;
     }
 #endif
-    ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
-    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-    static DenseIndex rows(void) {return _Rows;}
-    DenseIndex cols(void) const {return m_cols;}
-    void conservativeResize(DenseIndex size, DenseIndex, DenseIndex nbCols)
+    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index size, Index, Index cols)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, _Rows*m_cols);
-      m_cols = nbCols;
+      m_cols = cols;
     }
-    EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex, DenseIndex nbCols)
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index, Index cols)
     {
       if(size != _Rows*m_cols)
       {
@@ -364,35 +483,55 @@ template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Ro
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       }
-      m_cols = nbCols;
+      m_cols = cols;
     }
-    const T *data() const { return m_data; }
-    T *data() { return m_data; }
-  private:
-    DenseStorage(const DenseStorage&);
-    DenseStorage& operator=(const DenseStorage&);
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data; }
 };
 
 // matrix with dynamic height and fixed width (so that matrix has dynamic size).
 template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dynamic, _Cols, _Options>
 {
     T *m_data;
-    DenseIndex m_rows;
+    Index m_rows;
   public:
-    DenseStorage() : m_data(0), m_rows(0) {}
-    DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
-    DenseStorage(DenseIndex size, DenseIndex nbRows, DenseIndex) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(nbRows)
-    { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN }
-#ifdef EIGEN_HAVE_RVALUE_REFERENCES
-    DenseStorage(DenseStorage&& other)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0) {}
+    explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows>=0 && cols == _Cols);
+      EIGEN_UNUSED_VARIABLE(cols);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*_Cols))
+      , m_rows(other.m_rows)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*_Cols)
+      internal::smart_copy(other.m_data, other.m_data+other.m_rows*_Cols, m_data);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        DenseStorage tmp(other);
+        this->swap(tmp);
+      }
+      return *this;
+    }    
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
       : m_data(std::move(other.m_data))
       , m_rows(std::move(other.m_rows))
     {
       other.m_data = nullptr;
+      other.m_rows = 0;
     }
-    DenseStorage& operator=(DenseStorage&& other)
+    EIGEN_DEVICE_FUNC
+    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
     {
       using std::swap;
       swap(m_data, other.m_data);
@@ -400,16 +539,16 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
       return *this;
     }
 #endif
-    ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
-    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-    DenseIndex rows(void) const {return m_rows;}
-    static DenseIndex cols(void) {return _Cols;}
-    void conservativeResize(DenseIndex size, DenseIndex nbRows, DenseIndex)
+    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
+    void conservativeResize(Index size, Index rows, Index)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*_Cols);
-      m_rows = nbRows;
+      m_rows = rows;
     }
-    EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex nbRows, DenseIndex)
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index rows, Index)
     {
       if(size != m_rows*_Cols)
       {
@@ -418,15 +557,12 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       }
-      m_rows = nbRows;
+      m_rows = rows;
     }
-    const T *data() const { return m_data; }
-    T *data() { return m_data; }
-  private:
-    DenseStorage(const DenseStorage&);
-    DenseStorage& operator=(const DenseStorage&);
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data; }
 };
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/Diagonal.h b/extern/Eigen3/Eigen/src/Core/Diagonal.h
index 68cf6d4b044..afcaf357567 100644
--- a/extern/Eigen3/Eigen/src/Core/Diagonal.h
+++ b/extern/Eigen3/Eigen/src/Core/Diagonal.h
@@ -21,7 +21,7 @@ namespace Eigen {
   * \param MatrixType the type of the object in which we are taking a sub/main/super diagonal
   * \param DiagIndex the index of the sub/super diagonal. The default is 0 and it means the main diagonal.
   *              A positive value means a superdiagonal, a negative value means a subdiagonal.
-  *              You can also use Dynamic so the index can be set at runtime.
+  *              You can also use DynamicIndex so the index can be set at runtime.
   *
   * The matrix is not required to be square.
   *
@@ -37,7 +37,7 @@ template<typename MatrixType, int DiagIndex>
 struct traits<Diagonal<MatrixType,DiagIndex> >
  : traits<MatrixType>
 {
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   typedef typename MatrixType::StorageKind StorageKind;
   enum {
@@ -52,8 +52,7 @@ struct traits<Diagonal<MatrixType,DiagIndex> >
                                                  MatrixType::MaxColsAtCompileTime - EIGEN_PLAIN_ENUM_MAX( DiagIndex, 0))),
     MaxColsAtCompileTime = 1,
     MaskLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
-    Flags = (unsigned int)_MatrixTypeNested::Flags & (HereditaryBits | LinearAccessBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit,
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost,
+    Flags = (unsigned int)_MatrixTypeNested::Flags & (RowMajorBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit, // FIXME DirectAccessBit should not be handled by expressions
     MatrixTypeOuterStride = outer_stride_at_compile_time<MatrixType>::ret,
     InnerStrideAtCompileTime = MatrixTypeOuterStride == Dynamic ? Dynamic : MatrixTypeOuterStride+1,
     OuterStrideAtCompileTime = 0
@@ -70,20 +69,31 @@ template<typename MatrixType, int _DiagIndex> class Diagonal
     typedef typename internal::dense_xpr_base<Diagonal>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Diagonal)
 
-    inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index)
+    {
+      eigen_assert( a_index <= m_matrix.cols() && -a_index <= m_matrix.rows() );
+    }
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Diagonal)
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const
-    { return m_index.value()<0 ? (std::min<Index>)(m_matrix.cols(),m_matrix.rows()+m_index.value()) : (std::min<Index>)(m_matrix.rows(),m_matrix.cols()-m_index.value()); }
+    {
+      return m_index.value()<0 ? numext::mini<Index>(m_matrix.cols(),m_matrix.rows()+m_index.value())
+                               : numext::mini<Index>(m_matrix.rows(),m_matrix.cols()-m_index.value());
+    }
 
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return 1; }
 
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return m_matrix.outerStride() + 1;
     }
 
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return 0;
@@ -95,62 +105,75 @@ template<typename MatrixType, int _DiagIndex> class Diagonal
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.const_cast_derived().coeffRef(rowOffset(), colOffset())); }
-    inline const Scalar* data() const { return &(m_matrix.const_cast_derived().coeffRef(rowOffset(), colOffset())); }
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
+    EIGEN_DEVICE_FUNC
+    inline const Scalar* data() const { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index row, Index)
     {
       EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return m_matrix.const_cast_derived().coeffRef(row+rowOffset(), row+colOffset());
+      return m_matrix.coeffRef(row+rowOffset(), row+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index row, Index) const
     {
-      return m_matrix.const_cast_derived().coeffRef(row+rowOffset(), row+colOffset());
+      return m_matrix.coeffRef(row+rowOffset(), row+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline CoeffReturnType coeff(Index row, Index) const
     {
       return m_matrix.coeff(row+rowOffset(), row+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index idx)
     {
       EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return m_matrix.const_cast_derived().coeffRef(idx+rowOffset(), idx+colOffset());
+      return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index idx) const
     {
-      return m_matrix.const_cast_derived().coeffRef(idx+rowOffset(), idx+colOffset());
+      return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline CoeffReturnType coeff(Index idx) const
     {
       return m_matrix.coeff(idx+rowOffset(), idx+colOffset());
     }
 
-    const typename internal::remove_all<typename MatrixType::Nested>::type& 
+    EIGEN_DEVICE_FUNC
+    inline const typename internal::remove_all<typename MatrixType::Nested>::type& 
     nestedExpression() const 
     {
       return m_matrix;
     }
 
-    int index() const
+    EIGEN_DEVICE_FUNC
+    inline Index index() const
     {
       return m_index.value();
     }
 
   protected:
-    typename MatrixType::Nested m_matrix;
+    typename internal::ref_selector<MatrixType>::non_const_type m_matrix;
     const internal::variable_if_dynamicindex<Index, DiagIndex> m_index;
 
   private:
     // some compilers may fail to optimize std::max etc in case of compile-time constants...
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index absDiagIndex() const { return m_index.value()>0 ? m_index.value() : -m_index.value(); }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value()>0 ? 0 : -m_index.value(); }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value()>0 ? m_index.value() : 0; }
-    // triger a compile time error is someone try to call packet
+    // trigger a compile-time error if someone try to call packet
     template<int LoadMode> typename MatrixType::PacketReturnType packet(Index) const;
     template<int LoadMode> typename MatrixType::PacketReturnType packet(Index,Index) const;
 };
@@ -167,7 +190,7 @@ template<typename Derived>
 inline typename MatrixBase<Derived>::DiagonalReturnType
 MatrixBase<Derived>::diagonal()
 {
-  return derived();
+  return DiagonalReturnType(derived());
 }
 
 /** This is the const version of diagonal(). */
@@ -216,20 +239,20 @@ MatrixBase<Derived>::diagonal(Index index) const
   *
   * \sa MatrixBase::diagonal(), class Diagonal */
 template<typename Derived>
-template<int Index>
-inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index>::Type
+template<int Index_>
+inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index_>::Type
 MatrixBase<Derived>::diagonal()
 {
-  return derived();
+  return typename DiagonalIndexReturnType<Index_>::Type(derived());
 }
 
 /** This is the const version of diagonal<int>(). */
 template<typename Derived>
-template<int Index>
-inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index>::Type
+template<int Index_>
+inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index_>::Type
 MatrixBase<Derived>::diagonal() const
 {
-  return derived();
+  return typename ConstDiagonalIndexReturnType<Index_>::Type(derived());
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/DiagonalMatrix.h b/extern/Eigen3/Eigen/src/Core/DiagonalMatrix.h
index e6c220f4195..ecfdce8efa1 100644
--- a/extern/Eigen3/Eigen/src/Core/DiagonalMatrix.h
+++ b/extern/Eigen3/Eigen/src/Core/DiagonalMatrix.h
@@ -22,7 +22,7 @@ class DiagonalBase : public EigenBase<Derived>
     typedef typename DiagonalVectorType::Scalar Scalar;
     typedef typename DiagonalVectorType::RealScalar RealScalar;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
 
     enum {
       RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
@@ -30,79 +30,61 @@ class DiagonalBase : public EigenBase<Derived>
       MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
       MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
       IsVectorAtCompileTime = 0,
-      Flags = 0
+      Flags = NoPreferredStorageOrderBit
     };
 
     typedef Matrix<Scalar, RowsAtCompileTime, ColsAtCompileTime, 0, MaxRowsAtCompileTime, MaxColsAtCompileTime> DenseMatrixType;
     typedef DenseMatrixType DenseType;
     typedef DiagonalMatrix<Scalar,DiagonalVectorType::SizeAtCompileTime,DiagonalVectorType::MaxSizeAtCompileTime> PlainObject;
 
+    EIGEN_DEVICE_FUNC
     inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    EIGEN_DEVICE_FUNC
     inline Derived& derived() { return *static_cast<Derived*>(this); }
 
+    EIGEN_DEVICE_FUNC
     DenseMatrixType toDenseMatrix() const { return derived(); }
-    template<typename DenseDerived>
-    void evalTo(MatrixBase<DenseDerived> &other) const;
-    template<typename DenseDerived>
-    void addTo(MatrixBase<DenseDerived> &other) const
-    { other.diagonal() += diagonal(); }
-    template<typename DenseDerived>
-    void subTo(MatrixBase<DenseDerived> &other) const
-    { other.diagonal() -= diagonal(); }
-
+    
+    EIGEN_DEVICE_FUNC
     inline const DiagonalVectorType& diagonal() const { return derived().diagonal(); }
+    EIGEN_DEVICE_FUNC
     inline DiagonalVectorType& diagonal() { return derived().diagonal(); }
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return diagonal().size(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return diagonal().size(); }
 
-    /** \returns the diagonal matrix product of \c *this by the matrix \a matrix.
-      */
     template<typename MatrixDerived>
-    const DiagonalProduct<MatrixDerived, Derived, OnTheLeft>
+    EIGEN_DEVICE_FUNC
+    const Product<Derived,MatrixDerived,LazyProduct>
     operator*(const MatrixBase<MatrixDerived> &matrix) const
     {
-      return DiagonalProduct<MatrixDerived, Derived, OnTheLeft>(matrix.derived(), derived());
+      return Product<Derived, MatrixDerived, LazyProduct>(derived(),matrix.derived());
     }
 
-    inline const DiagonalWrapper<const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const DiagonalVectorType> >
+    typedef DiagonalWrapper<const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const DiagonalVectorType> > InverseReturnType;
+    EIGEN_DEVICE_FUNC
+    inline const InverseReturnType
     inverse() const
     {
-      return diagonal().cwiseInverse();
+      return InverseReturnType(diagonal().cwiseInverse());
     }
     
-    inline const DiagonalWrapper<const CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const DiagonalVectorType> >
+    EIGEN_DEVICE_FUNC
+    inline const DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >
     operator*(const Scalar& scalar) const
     {
-      return diagonal() * scalar;
+      return DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >(diagonal() * scalar);
     }
-    friend inline const DiagonalWrapper<const CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const DiagonalVectorType> >
+    EIGEN_DEVICE_FUNC
+    friend inline const DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >
     operator*(const Scalar& scalar, const DiagonalBase& other)
     {
-      return other.diagonal() * scalar;
-    }
-    
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived>
-    bool isApprox(const DiagonalBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
-    {
-      return diagonal().isApprox(other.diagonal(), precision);
+      return DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >(scalar * other.diagonal());
     }
-    template<typename OtherDerived>
-    bool isApprox(const MatrixBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
-    {
-      return toDenseMatrix().isApprox(other, precision);
-    }
-    #endif
 };
 
-template<typename Derived>
-template<typename DenseDerived>
-void DiagonalBase<Derived>::evalTo(MatrixBase<DenseDerived> &other) const
-{
-  other.setZero();
-  other.diagonal() = diagonal();
-}
 #endif
 
 /** \class DiagonalMatrix
@@ -124,10 +106,9 @@ struct traits<DiagonalMatrix<_Scalar,SizeAtCompileTime,MaxSizeAtCompileTime> >
  : traits<Matrix<_Scalar,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
 {
   typedef Matrix<_Scalar,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1> DiagonalVectorType;
-  typedef Dense StorageKind;
-  typedef DenseIndex Index;
+  typedef DiagonalShape StorageKind;
   enum {
-    Flags = LvalueBit
+    Flags = LvalueBit | NoPreferredStorageOrderBit
   };
 };
 }
@@ -141,7 +122,7 @@ class DiagonalMatrix
     typedef const DiagonalMatrix& Nested;
     typedef _Scalar Scalar;
     typedef typename internal::traits<DiagonalMatrix>::StorageKind StorageKind;
-    typedef typename internal::traits<DiagonalMatrix>::Index Index;
+    typedef typename internal::traits<DiagonalMatrix>::StorageIndex StorageIndex;
     #endif
 
   protected:
@@ -151,24 +132,31 @@ class DiagonalMatrix
   public:
 
     /** const version of diagonal(). */
+    EIGEN_DEVICE_FUNC
     inline const DiagonalVectorType& diagonal() const { return m_diagonal; }
     /** \returns a reference to the stored vector of diagonal coefficients. */
+    EIGEN_DEVICE_FUNC
     inline DiagonalVectorType& diagonal() { return m_diagonal; }
 
     /** Default constructor without initialization */
+    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix() {}
 
     /** Constructs a diagonal matrix with given dimension  */
-    inline DiagonalMatrix(Index dim) : m_diagonal(dim) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline DiagonalMatrix(Index dim) : m_diagonal(dim) {}
 
     /** 2D constructor. */
+    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix(const Scalar& x, const Scalar& y) : m_diagonal(x,y) {}
 
     /** 3D constructor. */
+    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix(const Scalar& x, const Scalar& y, const Scalar& z) : m_diagonal(x,y,z) {}
 
     /** Copy constructor. */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix(const DiagonalBase<OtherDerived>& other) : m_diagonal(other.diagonal()) {}
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
@@ -178,11 +166,13 @@ class DiagonalMatrix
 
     /** generic constructor from expression of the diagonal coefficients */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     explicit inline DiagonalMatrix(const MatrixBase<OtherDerived>& other) : m_diagonal(other)
     {}
 
     /** Copy operator. */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     DiagonalMatrix& operator=(const DiagonalBase<OtherDerived>& other)
     {
       m_diagonal = other.diagonal();
@@ -193,6 +183,7 @@ class DiagonalMatrix
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
       */
+    EIGEN_DEVICE_FUNC
     DiagonalMatrix& operator=(const DiagonalMatrix& other)
     {
       m_diagonal = other.diagonal();
@@ -201,14 +192,19 @@ class DiagonalMatrix
     #endif
 
     /** Resizes to given size. */
+    EIGEN_DEVICE_FUNC
     inline void resize(Index size) { m_diagonal.resize(size); }
     /** Sets all coefficients to zero. */
+    EIGEN_DEVICE_FUNC
     inline void setZero() { m_diagonal.setZero(); }
     /** Resizes and sets all coefficients to zero. */
+    EIGEN_DEVICE_FUNC
     inline void setZero(Index size) { m_diagonal.setZero(size); }
     /** Sets this matrix to be the identity matrix of the current size. */
+    EIGEN_DEVICE_FUNC
     inline void setIdentity() { m_diagonal.setOnes(); }
     /** Sets this matrix to be the identity matrix of the given size. */
+    EIGEN_DEVICE_FUNC
     inline void setIdentity(Index size) { m_diagonal.setOnes(size); }
 };
 
@@ -232,14 +228,15 @@ struct traits<DiagonalWrapper<_DiagonalVectorType> >
 {
   typedef _DiagonalVectorType DiagonalVectorType;
   typedef typename DiagonalVectorType::Scalar Scalar;
-  typedef typename DiagonalVectorType::Index Index;
-  typedef typename DiagonalVectorType::StorageKind StorageKind;
+  typedef typename DiagonalVectorType::StorageIndex StorageIndex;
+  typedef DiagonalShape StorageKind;
+  typedef typename traits<DiagonalVectorType>::XprKind XprKind;
   enum {
     RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
     ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    MaxRowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    MaxColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    Flags =  traits<DiagonalVectorType>::Flags & LvalueBit
+    MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+    MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+    Flags =  (traits<DiagonalVectorType>::Flags & LvalueBit) | NoPreferredStorageOrderBit
   };
 };
 }
@@ -255,9 +252,11 @@ class DiagonalWrapper
     #endif
 
     /** Constructor from expression of diagonal coefficients to wrap. */
-    inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {}
 
     /** \returns a const reference to the wrapped expression of diagonal coefficients. */
+    EIGEN_DEVICE_FUNC
     const DiagonalVectorType& diagonal() const { return m_diagonal; }
 
   protected:
@@ -277,7 +276,7 @@ template<typename Derived>
 inline const DiagonalWrapper<const Derived>
 MatrixBase<Derived>::asDiagonal() const
 {
-  return derived();
+  return DiagonalWrapper<const Derived>(derived());
 }
 
 /** \returns true if *this is approximately equal to a diagonal matrix,
@@ -291,12 +290,11 @@ MatrixBase<Derived>::asDiagonal() const
 template<typename Derived>
 bool MatrixBase<Derived>::isDiagonal(const RealScalar& prec) const
 {
-  using std::abs;
   if(cols() != rows()) return false;
   RealScalar maxAbsOnDiagonal = static_cast<RealScalar>(-1);
   for(Index j = 0; j < cols(); ++j)
   {
-    RealScalar absOnDiagonal = abs(coeff(j,j));
+    RealScalar absOnDiagonal = numext::abs(coeff(j,j));
     if(absOnDiagonal > maxAbsOnDiagonal) maxAbsOnDiagonal = absOnDiagonal;
   }
   for(Index j = 0; j < cols(); ++j)
@@ -308,6 +306,38 @@ bool MatrixBase<Derived>::isDiagonal(const RealScalar& prec) const
   return true;
 }
 
+namespace internal {
+
+template<> struct storage_kind_to_shape<DiagonalShape> { typedef DiagonalShape Shape; };
+
+struct Diagonal2Dense {};
+
+template<> struct AssignmentKind<DenseShape,DiagonalShape> { typedef Diagonal2Dense Kind; };
+
+// Diagonal matrix to Dense assignment
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Dense>
+{
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    
+    dst.setZero();
+    dst.diagonal() = src.diagonal();
+  }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() += src.diagonal(); }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() -= src.diagonal(); }
+};
+
+} // namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_DIAGONALMATRIX_H
diff --git a/extern/Eigen3/Eigen/src/Core/DiagonalProduct.h b/extern/Eigen3/Eigen/src/Core/DiagonalProduct.h
index cc6b536e199..d372b938f65 100644
--- a/extern/Eigen3/Eigen/src/Core/DiagonalProduct.h
+++ b/extern/Eigen3/Eigen/src/Core/DiagonalProduct.h
@@ -13,117 +13,14 @@
 
 namespace Eigen { 
 
-namespace internal {
-template<typename MatrixType, typename DiagonalType, int ProductOrder>
-struct traits<DiagonalProduct<MatrixType, DiagonalType, ProductOrder> >
- : traits<MatrixType>
-{
-  typedef typename scalar_product_traits<typename MatrixType::Scalar, typename DiagonalType::Scalar>::ReturnType Scalar;
-  enum {
-    RowsAtCompileTime = MatrixType::RowsAtCompileTime,
-    ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-
-    _StorageOrder = MatrixType::Flags & RowMajorBit ? RowMajor : ColMajor,
-    _ScalarAccessOnDiag =  !((int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheLeft)
-                          ||(int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheRight)),
-    _SameTypes = is_same<typename MatrixType::Scalar, typename DiagonalType::Scalar>::value,
-    // FIXME currently we need same types, but in the future the next rule should be the one
-    //_Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && ((!_PacketOnDiag) || (_SameTypes && bool(int(DiagonalType::DiagonalVectorType::Flags)&PacketAccessBit))),
-    _Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && _SameTypes && (_ScalarAccessOnDiag || (bool(int(DiagonalType::DiagonalVectorType::Flags)&PacketAccessBit))),
-    _LinearAccessMask = (RowsAtCompileTime==1 || ColsAtCompileTime==1) ? LinearAccessBit : 0,
-
-    Flags = ((HereditaryBits|_LinearAccessMask|AlignedBit) & (unsigned int)(MatrixType::Flags)) | (_Vectorizable ? PacketAccessBit : 0),//(int(MatrixType::Flags)&int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit),
-    Cost0 = EIGEN_ADD_COST(NumTraits<Scalar>::MulCost, MatrixType::CoeffReadCost),
-    CoeffReadCost = EIGEN_ADD_COST(Cost0,DiagonalType::DiagonalVectorType::CoeffReadCost)
-  };
-};
-}
-
-template<typename MatrixType, typename DiagonalType, int ProductOrder>
-class DiagonalProduct : internal::no_assignment_operator,
-                        public MatrixBase<DiagonalProduct<MatrixType, DiagonalType, ProductOrder> >
-{
-  public:
-
-    typedef MatrixBase<DiagonalProduct> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(DiagonalProduct)
-
-    inline DiagonalProduct(const MatrixType& matrix, const DiagonalType& diagonal)
-      : m_matrix(matrix), m_diagonal(diagonal)
-    {
-      eigen_assert(diagonal.diagonal().size() == (ProductOrder == OnTheLeft ? matrix.rows() : matrix.cols()));
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return m_matrix.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_matrix.cols(); }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
-    {
-      return m_diagonal.diagonal().coeff(ProductOrder == OnTheLeft ? row : col) * m_matrix.coeff(row, col);
-    }
-    
-    EIGEN_STRONG_INLINE const Scalar coeff(Index idx) const
-    {
-      enum {
-        StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor
-      };
-      return coeff(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index row, Index col) const
-    {
-      enum {
-        StorageOrder = Flags & RowMajorBit ? RowMajor : ColMajor
-      };
-      const Index indexInDiagonalVector = ProductOrder == OnTheLeft ? row : col;
-      return packet_impl<LoadMode>(row,col,indexInDiagonalVector,typename internal::conditional<
-        ((int(StorageOrder) == RowMajor && int(ProductOrder) == OnTheLeft)
-       ||(int(StorageOrder) == ColMajor && int(ProductOrder) == OnTheRight)), internal::true_type, internal::false_type>::type());
-    }
-    
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index idx) const
-    {
-      enum {
-        StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor
-      };
-      return packet<LoadMode>(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
-    }
-
-  protected:
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::true_type) const
-    {
-      return internal::pmul(m_matrix.template packet<LoadMode>(row, col),
-                     internal::pset1<PacketScalar>(m_diagonal.diagonal().coeff(id)));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::false_type) const
-    {
-      enum {
-        InnerSize = (MatrixType::Flags & RowMajorBit) ? MatrixType::ColsAtCompileTime : MatrixType::RowsAtCompileTime,
-        DiagonalVectorPacketLoadMode = (LoadMode == Aligned && (((InnerSize%16) == 0) || (int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit)==AlignedBit) ? Aligned : Unaligned)
-      };
-      return internal::pmul(m_matrix.template packet<LoadMode>(row, col),
-                     m_diagonal.diagonal().template packet<DiagonalVectorPacketLoadMode>(id));
-    }
-
-    typename MatrixType::Nested m_matrix;
-    typename DiagonalType::Nested m_diagonal;
-};
-
 /** \returns the diagonal matrix product of \c *this by the diagonal matrix \a diagonal.
   */
 template<typename Derived>
 template<typename DiagonalDerived>
-inline const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
+inline const Product<Derived, DiagonalDerived, LazyProduct>
 MatrixBase<Derived>::operator*(const DiagonalBase<DiagonalDerived> &a_diagonal) const
 {
-  return DiagonalProduct<Derived, DiagonalDerived, OnTheRight>(derived(), a_diagonal.derived());
+  return Product<Derived, DiagonalDerived, LazyProduct>(derived(),a_diagonal.derived());
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/Dot.h b/extern/Eigen3/Eigen/src/Core/Dot.h
index 9d7651f1fd4..1fe7a84a48d 100644
--- a/extern/Eigen3/Eigen/src/Core/Dot.h
+++ b/extern/Eigen3/Eigen/src/Core/Dot.h
@@ -28,26 +28,33 @@ template<typename T, typename U,
 >
 struct dot_nocheck
 {
-  typedef typename scalar_product_traits<typename traits<T>::Scalar,typename traits<U>::Scalar>::ReturnType ResScalar;
-  static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
+  typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
+  typedef typename conj_prod::result_type ResScalar;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE
+  static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    return a.template binaryExpr<scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> >(b).sum();
+    return a.template binaryExpr<conj_prod>(b).sum();
   }
 };
 
 template<typename T, typename U>
 struct dot_nocheck<T, U, true>
 {
-  typedef typename scalar_product_traits<typename traits<T>::Scalar,typename traits<U>::Scalar>::ReturnType ResScalar;
-  static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
+  typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
+  typedef typename conj_prod::result_type ResScalar;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE
+  static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    return a.transpose().template binaryExpr<scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> >(b).sum();
+    return a.transpose().template binaryExpr<conj_prod>(b).sum();
   }
 };
 
 } // end namespace internal
 
-/** \returns the dot product of *this with other.
+/** \fn MatrixBase::dot
+  * \returns the dot product of *this with other.
   *
   * \only_for_vectors
   *
@@ -59,55 +66,31 @@ struct dot_nocheck<T, U, true>
   */
 template<typename Derived>
 template<typename OtherDerived>
-typename internal::scalar_product_traits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE
+typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
 MatrixBase<Derived>::dot(const MatrixBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
   EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived)
+#if !(defined(EIGEN_NO_STATIC_ASSERT) && defined(EIGEN_NO_DEBUG))
   typedef internal::scalar_conj_product_op<Scalar,typename OtherDerived::Scalar> func;
   EIGEN_CHECK_BINARY_COMPATIBILIY(func,Scalar,typename OtherDerived::Scalar);
-
+#endif
+  
   eigen_assert(size() == other.size());
 
   return internal::dot_nocheck<Derived,OtherDerived>::run(*this, other);
 }
 
-#ifdef EIGEN2_SUPPORT
-/** \returns the dot product of *this with other, with the Eigen2 convention that the dot product is linear in the first variable
-  * (conjugating the second variable). Of course this only makes a difference in the complex case.
-  *
-  * This method is only available in EIGEN2_SUPPORT mode.
-  *
-  * \only_for_vectors
-  *
-  * \sa dot()
-  */
-template<typename Derived>
-template<typename OtherDerived>
-typename internal::traits<Derived>::Scalar
-MatrixBase<Derived>::eigen2_dot(const MatrixBase<OtherDerived>& other) const
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
-  EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived)
-  EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
-    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
-  eigen_assert(size() == other.size());
-
-  return internal::dot_nocheck<OtherDerived,Derived>::run(other,*this);
-}
-#endif
-
-
 //---------- implementation of L2 norm and related functions ----------
 
 /** \returns, for vectors, the squared \em l2 norm of \c *this, and for matrices the Frobenius norm.
   * In both cases, it consists in the sum of the square of all the matrix entries.
   * For vectors, this is also equals to the dot product of \c *this with itself.
   *
-  * \sa dot(), norm()
+  * \sa dot(), norm(), lpNorm()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::squaredNorm() const
@@ -119,41 +102,98 @@ EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scala
   * In both cases, it consists in the square root of the sum of the square of all the matrix entries.
   * For vectors, this is also equals to the square root of the dot product of \c *this with itself.
   *
-  * \sa dot(), squaredNorm()
+  * \sa lpNorm(), dot(), squaredNorm()
   */
 template<typename Derived>
-inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
+EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
 {
-  using std::sqrt;
-  return sqrt(squaredNorm());
+  return numext::sqrt(squaredNorm());
 }
 
-/** \returns an expression of the quotient of *this by its own norm.
+/** \returns an expression of the quotient of \c *this by its own norm.
+  *
+  * \warning If the input vector is too small (i.e., this->norm()==0),
+  *          then this function returns a copy of the input.
   *
   * \only_for_vectors
   *
   * \sa norm(), normalize()
   */
 template<typename Derived>
-inline const typename MatrixBase<Derived>::PlainObject
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
 MatrixBase<Derived>::normalized() const
 {
-  typedef typename internal::nested<Derived>::type Nested;
-  typedef typename internal::remove_reference<Nested>::type _Nested;
+  typedef typename internal::nested_eval<Derived,2>::type _Nested;
   _Nested n(derived());
-  return n / n.norm();
+  RealScalar z = n.squaredNorm();
+  // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
+  if(z>RealScalar(0))
+    return n / numext::sqrt(z);
+  else
+    return n;
 }
 
 /** Normalizes the vector, i.e. divides it by its own norm.
   *
   * \only_for_vectors
   *
+  * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
+  *
   * \sa norm(), normalized()
   */
 template<typename Derived>
-inline void MatrixBase<Derived>::normalize()
+EIGEN_STRONG_INLINE void MatrixBase<Derived>::normalize()
 {
-  *this /= norm();
+  RealScalar z = squaredNorm();
+  // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
+  if(z>RealScalar(0))
+    derived() /= numext::sqrt(z);
+}
+
+/** \returns an expression of the quotient of \c *this by its own norm while avoiding underflow and overflow.
+  *
+  * \only_for_vectors
+  *
+  * This method is analogue to the normalized() method, but it reduces the risk of
+  * underflow and overflow when computing the norm.
+  *
+  * \warning If the input vector is too small (i.e., this->norm()==0),
+  *          then this function returns a copy of the input.
+  *
+  * \sa stableNorm(), stableNormalize(), normalized()
+  */
+template<typename Derived>
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
+MatrixBase<Derived>::stableNormalized() const
+{
+  typedef typename internal::nested_eval<Derived,3>::type _Nested;
+  _Nested n(derived());
+  RealScalar w = n.cwiseAbs().maxCoeff();
+  RealScalar z = (n/w).squaredNorm();
+  if(z>RealScalar(0))
+    return n / (numext::sqrt(z)*w);
+  else
+    return n;
+}
+
+/** Normalizes the vector while avoid underflow and overflow
+  *
+  * \only_for_vectors
+  *
+  * This method is analogue to the normalize() method, but it reduces the risk of
+  * underflow and overflow when computing the norm.
+  *
+  * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
+  *
+  * \sa stableNorm(), stableNormalized(), normalize()
+  */
+template<typename Derived>
+EIGEN_STRONG_INLINE void MatrixBase<Derived>::stableNormalize()
+{
+  RealScalar w = cwiseAbs().maxCoeff();
+  RealScalar z = (derived()/w).squaredNorm();
+  if(z>RealScalar(0))
+    derived() /= numext::sqrt(z)*w;
 }
 
 //---------- implementation of other norms ----------
@@ -164,9 +204,10 @@ template<typename Derived, int p>
 struct lpNorm_selector
 {
   typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const MatrixBase<Derived>& m)
   {
-    using std::pow;
+    EIGEN_USING_STD_MATH(pow)
     return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p);
   }
 };
@@ -174,6 +215,7 @@ struct lpNorm_selector
 template<typename Derived>
 struct lpNorm_selector<Derived, 1>
 {
+  EIGEN_DEVICE_FUNC
   static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
   {
     return m.cwiseAbs().sum();
@@ -183,6 +225,7 @@ struct lpNorm_selector<Derived, 1>
 template<typename Derived>
 struct lpNorm_selector<Derived, 2>
 {
+  EIGEN_DEVICE_FUNC
   static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
   {
     return m.norm();
@@ -192,23 +235,35 @@ struct lpNorm_selector<Derived, 2>
 template<typename Derived>
 struct lpNorm_selector<Derived, Infinity>
 {
-  static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
+  typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const MatrixBase<Derived>& m)
   {
+    if(Derived::SizeAtCompileTime==0 || (Derived::SizeAtCompileTime==Dynamic && m.size()==0))
+      return RealScalar(0);
     return m.cwiseAbs().maxCoeff();
   }
 };
 
 } // end namespace internal
 
-/** \returns the \f$ \ell^p \f$ norm of *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values
-  *          of the coefficients of *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$
-  *          norm, that is the maximum of the absolute values of the coefficients of *this.
+/** \returns the \b coefficient-wise \f$ \ell^p \f$ norm of \c *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values
+  *          of the coefficients of \c *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$
+  *          norm, that is the maximum of the absolute values of the coefficients of \c *this.
+  *
+  * In all cases, if \c *this is empty, then the value 0 is returned.
+  *
+  * \note For matrices, this function does not compute the <a href="https://en.wikipedia.org/wiki/Operator_norm">operator-norm</a>. That is, if \c *this is a matrix, then its coefficients are interpreted as a 1D vector. Nonetheless, you can easily compute the 1-norm and \f$\infty\f$-norm matrix operator norms using \link TutorialReductionsVisitorsBroadcastingReductionsNorm partial reductions \endlink.
   *
   * \sa norm()
   */
 template<typename Derived>
 template<int p>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
+#else
+MatrixBase<Derived>::RealScalar
+#endif
 MatrixBase<Derived>::lpNorm() const
 {
   return internal::lpNorm_selector<Derived, p>::run(*this);
@@ -227,8 +282,8 @@ template<typename OtherDerived>
 bool MatrixBase<Derived>::isOrthogonal
 (const MatrixBase<OtherDerived>& other, const RealScalar& prec) const
 {
-  typename internal::nested<Derived,2>::type nested(derived());
-  typename internal::nested<OtherDerived,2>::type otherNested(other.derived());
+  typename internal::nested_eval<Derived,2>::type nested(derived());
+  typename internal::nested_eval<OtherDerived,2>::type otherNested(other.derived());
   return numext::abs2(nested.dot(otherNested)) <= prec * prec * nested.squaredNorm() * otherNested.squaredNorm();
 }
 
@@ -246,13 +301,13 @@ bool MatrixBase<Derived>::isOrthogonal
 template<typename Derived>
 bool MatrixBase<Derived>::isUnitary(const RealScalar& prec) const
 {
-  typename Derived::Nested nested(derived());
+  typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index i = 0; i < cols(); ++i)
   {
-    if(!internal::isApprox(nested.col(i).squaredNorm(), static_cast<RealScalar>(1), prec))
+    if(!internal::isApprox(self.col(i).squaredNorm(), static_cast<RealScalar>(1), prec))
       return false;
     for(Index j = 0; j < i; ++j)
-      if(!internal::isMuchSmallerThan(nested.col(i).dot(nested.col(j)), static_cast<Scalar>(1), prec))
+      if(!internal::isMuchSmallerThan(self.col(i).dot(self.col(j)), static_cast<Scalar>(1), prec))
         return false;
   }
   return true;
diff --git a/extern/Eigen3/Eigen/src/Core/EigenBase.h b/extern/Eigen3/Eigen/src/Core/EigenBase.h
index fadb45852f6..b195506a91e 100644
--- a/extern/Eigen3/Eigen/src/Core/EigenBase.h
+++ b/extern/Eigen3/Eigen/src/Core/EigenBase.h
@@ -13,7 +13,10 @@
 
 namespace Eigen {
 
-/** Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T).
+/** \class EigenBase
+  * \ingroup Core_Module
+  * 
+  * Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T).
   *
   * In other words, an EigenBase object is an object that can be copied into a MatrixBase.
   *
@@ -21,39 +24,57 @@ namespace Eigen {
   *
   * Notice that this class is trivial, it is only used to disambiguate overloaded functions.
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived> struct EigenBase
 {
 //   typedef typename internal::plain_matrix_type<Derived>::type PlainObject;
-
+  
+  /** \brief The interface type of indices
+    * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
+    * \deprecated Since Eigen 3.3, its usage is deprecated. Use Eigen::Index instead.
+    * \sa StorageIndex, \ref TopicPreprocessorDirectives.
+    */
+  typedef Eigen::Index Index;
+
+  // FIXME is it needed?
   typedef typename internal::traits<Derived>::StorageKind StorageKind;
-  typedef typename internal::traits<Derived>::Index Index;
 
   /** \returns a reference to the derived object */
+  EIGEN_DEVICE_FUNC
   Derived& derived() { return *static_cast<Derived*>(this); }
   /** \returns a const reference to the derived object */
+  EIGEN_DEVICE_FUNC
   const Derived& derived() const { return *static_cast<const Derived*>(this); }
 
+  EIGEN_DEVICE_FUNC
   inline Derived& const_cast_derived() const
   { return *static_cast<Derived*>(const_cast<EigenBase*>(this)); }
+  EIGEN_DEVICE_FUNC
   inline const Derived& const_derived() const
   { return *static_cast<const Derived*>(this); }
 
   /** \returns the number of rows. \sa cols(), RowsAtCompileTime */
+  EIGEN_DEVICE_FUNC
   inline Index rows() const { return derived().rows(); }
   /** \returns the number of columns. \sa rows(), ColsAtCompileTime*/
+  EIGEN_DEVICE_FUNC
   inline Index cols() const { return derived().cols(); }
   /** \returns the number of coefficients, which is rows()*cols().
     * \sa rows(), cols(), SizeAtCompileTime. */
+  EIGEN_DEVICE_FUNC
   inline Index size() const { return rows() * cols(); }
 
   /** \internal Don't use it, but do the equivalent: \code dst = *this; \endcode */
-  template<typename Dest> inline void evalTo(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC
+  inline void evalTo(Dest& dst) const
   { derived().evalTo(dst); }
 
   /** \internal Don't use it, but do the equivalent: \code dst += *this; \endcode */
-  template<typename Dest> inline void addTo(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC
+  inline void addTo(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -63,7 +84,9 @@ template<typename Derived> struct EigenBase
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst -= *this; \endcode */
-  template<typename Dest> inline void subTo(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC
+  inline void subTo(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -73,7 +96,8 @@ template<typename Derived> struct EigenBase
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheRight(*this); \endcode */
-  template<typename Dest> inline void applyThisOnTheRight(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC inline void applyThisOnTheRight(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -81,7 +105,8 @@ template<typename Derived> struct EigenBase
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheLeft(*this); \endcode */
-  template<typename Dest> inline void applyThisOnTheLeft(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC inline void applyThisOnTheLeft(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -104,25 +129,28 @@ template<typename Derived> struct EigenBase
   */
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
 {
-  other.derived().evalTo(derived());
+  call_assignment(derived(), other.derived());
   return derived();
 }
 
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
 {
-  other.derived().addTo(derived());
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
 {
-  other.derived().subTo(derived());
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
diff --git a/extern/Eigen3/Eigen/src/Core/ForceAlignedAccess.h b/extern/Eigen3/Eigen/src/Core/ForceAlignedAccess.h
index 807c7a29346..7b08b45e67c 100644
--- a/extern/Eigen3/Eigen/src/Core/ForceAlignedAccess.h
+++ b/extern/Eigen3/Eigen/src/Core/ForceAlignedAccess.h
@@ -39,29 +39,29 @@ template<typename ExpressionType> class ForceAlignedAccess
     typedef typename internal::dense_xpr_base<ForceAlignedAccess>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(ForceAlignedAccess)
 
-    inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
+    EIGEN_DEVICE_FUNC explicit inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    inline Index rows() const { return m_expression.rows(); }
-    inline Index cols() const { return m_expression.cols(); }
-    inline Index outerStride() const { return m_expression.outerStride(); }
-    inline Index innerStride() const { return m_expression.innerStride(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); }
 
-    inline const CoeffReturnType coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
     {
       return m_expression.coeff(row, col);
     }
 
-    inline Scalar& coeffRef(Index row, Index col)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
     {
       return m_expression.const_cast_derived().coeffRef(row, col);
     }
 
-    inline const CoeffReturnType coeff(Index index) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
     {
       return m_expression.coeff(index);
     }
 
-    inline Scalar& coeffRef(Index index)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
     {
       return m_expression.const_cast_derived().coeffRef(index);
     }
@@ -90,7 +90,7 @@ template<typename ExpressionType> class ForceAlignedAccess
       m_expression.const_cast_derived().template writePacket<Aligned>(index, x);
     }
 
-    operator const ExpressionType&() const { return m_expression; }
+    EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
 
   protected:
     const ExpressionType& m_expression;
@@ -127,7 +127,7 @@ template<bool Enable>
 inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type
 MatrixBase<Derived>::forceAlignedAccessIf() const
 {
-  return derived();
+  return derived();  // FIXME This should not work but apparently is never used
 }
 
 /** \returns an expression of *this with forced aligned access if \a Enable is true.
@@ -138,7 +138,7 @@ template<bool Enable>
 inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type
 MatrixBase<Derived>::forceAlignedAccessIf()
 {
-  return derived();
+  return derived();  // FIXME This should not work but apparently is never used
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/Fuzzy.h b/extern/Eigen3/Eigen/src/Core/Fuzzy.h
index fe63bd2984d..3e403a09d92 100644
--- a/extern/Eigen3/Eigen/src/Core/Fuzzy.h
+++ b/extern/Eigen3/Eigen/src/Core/Fuzzy.h
@@ -19,18 +19,19 @@ namespace internal
 template<typename Derived, typename OtherDerived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
 struct isApprox_selector
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec)
   {
-    using std::min;
-    typename internal::nested<Derived,2>::type nested(x);
-    typename internal::nested<OtherDerived,2>::type otherNested(y);
-    return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * (min)(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().sum());
+    typename internal::nested_eval<Derived,2>::type nested(x);
+    typename internal::nested_eval<OtherDerived,2>::type otherNested(y);
+    return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * numext::mini(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().sum());
   }
 };
 
 template<typename Derived, typename OtherDerived>
 struct isApprox_selector<Derived, OtherDerived, true>
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar&)
   {
     return x.matrix() == y.matrix();
@@ -40,6 +41,7 @@ struct isApprox_selector<Derived, OtherDerived, true>
 template<typename Derived, typename OtherDerived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
 struct isMuchSmallerThan_object_selector
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec)
   {
     return x.cwiseAbs2().sum() <= numext::abs2(prec) * y.cwiseAbs2().sum();
@@ -49,6 +51,7 @@ struct isMuchSmallerThan_object_selector
 template<typename Derived, typename OtherDerived>
 struct isMuchSmallerThan_object_selector<Derived, OtherDerived, true>
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const OtherDerived&, const typename Derived::RealScalar&)
   {
     return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix();
@@ -58,6 +61,7 @@ struct isMuchSmallerThan_object_selector<Derived, OtherDerived, true>
 template<typename Derived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
 struct isMuchSmallerThan_scalar_selector
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const typename Derived::RealScalar& y, const typename Derived::RealScalar& prec)
   {
     return x.cwiseAbs2().sum() <= numext::abs2(prec * y);
@@ -67,6 +71,7 @@ struct isMuchSmallerThan_scalar_selector
 template<typename Derived>
 struct isMuchSmallerThan_scalar_selector<Derived, true>
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const typename Derived::RealScalar&, const typename Derived::RealScalar&)
   {
     return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix();
diff --git a/extern/Eigen3/Eigen/src/Core/GeneralProduct.h b/extern/Eigen3/Eigen/src/Core/GeneralProduct.h
index 0eae529909f..6f0cc80e940 100644
--- a/extern/Eigen3/Eigen/src/Core/GeneralProduct.h
+++ b/extern/Eigen3/Eigen/src/Core/GeneralProduct.h
@@ -11,29 +11,7 @@
 #ifndef EIGEN_GENERAL_PRODUCT_H
 #define EIGEN_GENERAL_PRODUCT_H
 
-namespace Eigen { 
-
-/** \class GeneralProduct
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the product of two general matrices or vectors
-  *
-  * \param LhsNested the type used to store the left-hand side
-  * \param RhsNested the type used to store the right-hand side
-  * \param ProductMode the type of the product
-  *
-  * This class represents an expression of the product of two general matrices.
-  * We call a general matrix, a dense matrix with full storage. For instance,
-  * This excludes triangular, selfadjoint, and sparse matrices.
-  * It is the return type of the operator* between general matrices. Its template
-  * arguments are determined automatically by ProductReturnType. Therefore,
-  * GeneralProduct should never be used direclty. To determine the result type of a
-  * function which involves a matrix product, use ProductReturnType::Type.
-  *
-  * \sa ProductReturnType, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
-  */
-template<typename Lhs, typename Rhs, int ProductType = internal::product_type<Lhs,Rhs>::value>
-class GeneralProduct;
+namespace Eigen {
 
 enum {
   Large = 2,
@@ -46,11 +24,17 @@ template<int Rows, int Cols, int Depth> struct product_type_selector;
 
 template<int Size, int MaxSize> struct product_size_category
 {
-  enum { is_large = MaxSize == Dynamic ||
-                    Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD,
-         value = is_large  ? Large
-               : Size == 1 ? 1
-                           : Small
+  enum {
+    #ifndef EIGEN_CUDA_ARCH
+    is_large = MaxSize == Dynamic ||
+               Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
+               (Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
+    #else
+    is_large = 0,
+    #endif
+    value = is_large  ? Large
+          : Size == 1 ? 1
+                      : Small
   };
 };
 
@@ -59,15 +43,14 @@ template<typename Lhs, typename Rhs> struct product_type
   typedef typename remove_all<Lhs>::type _Lhs;
   typedef typename remove_all<Rhs>::type _Rhs;
   enum {
-    MaxRows  = _Lhs::MaxRowsAtCompileTime,
-    Rows  = _Lhs::RowsAtCompileTime,
-    MaxCols  = _Rhs::MaxColsAtCompileTime,
-    Cols  = _Rhs::ColsAtCompileTime,
-    MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::MaxColsAtCompileTime,
-                                           _Rhs::MaxRowsAtCompileTime),
-    Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::ColsAtCompileTime,
-                                        _Rhs::RowsAtCompileTime),
-    LargeThreshold = EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+    MaxRows = traits<_Lhs>::MaxRowsAtCompileTime,
+    Rows    = traits<_Lhs>::RowsAtCompileTime,
+    MaxCols = traits<_Rhs>::MaxColsAtCompileTime,
+    Cols    = traits<_Rhs>::ColsAtCompileTime,
+    MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::MaxColsAtCompileTime,
+                                           traits<_Rhs>::MaxRowsAtCompileTime),
+    Depth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::ColsAtCompileTime,
+                                        traits<_Rhs>::RowsAtCompileTime)
   };
 
   // the splitting into different lines of code here, introducing the _select enums and the typedef below,
@@ -82,7 +65,8 @@ private:
 
 public:
   enum {
-    value = selector::ret
+    value = selector::ret,
+    ret = selector::ret
   };
 #ifdef EIGEN_DEBUG_PRODUCT
   static void debug()
@@ -98,12 +82,13 @@ public:
 #endif
 };
 
-
 /* The following allows to select the kind of product at compile time
  * based on the three dimensions of the product.
  * This is a compile time mapping from {1,Small,Large}^3 -> {product types} */
 // FIXME I'm not sure the current mapping is the ideal one.
 template<int M, int N>  struct product_type_selector<M,N,1>              { enum { ret = OuterProduct }; };
+template<int M>         struct product_type_selector<M, 1, 1>            { enum { ret = LazyCoeffBasedProductMode }; };
+template<int N>         struct product_type_selector<1, N, 1>            { enum { ret = LazyCoeffBasedProductMode }; };
 template<int Depth>     struct product_type_selector<1,    1,    Depth>  { enum { ret = InnerProduct }; };
 template<>              struct product_type_selector<1,    1,    1>      { enum { ret = InnerProduct }; };
 template<>              struct product_type_selector<Small,1,    Small>  { enum { ret = CoeffBasedProductMode }; };
@@ -122,60 +107,12 @@ template<>              struct product_type_selector<Small,Small,Large>  { enum
 template<>              struct product_type_selector<Large,Small,Large>  { enum { ret = GemmProduct }; };
 template<>              struct product_type_selector<Small,Large,Large>  { enum { ret = GemmProduct }; };
 template<>              struct product_type_selector<Large,Large,Large>  { enum { ret = GemmProduct }; };
-template<>              struct product_type_selector<Large,Small,Small>  { enum { ret = GemmProduct }; };
-template<>              struct product_type_selector<Small,Large,Small>  { enum { ret = GemmProduct }; };
+template<>              struct product_type_selector<Large,Small,Small>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Small,Large,Small>  { enum { ret = CoeffBasedProductMode }; };
 template<>              struct product_type_selector<Large,Large,Small>  { enum { ret = GemmProduct }; };
 
 } // end namespace internal
 
-/** \class ProductReturnType
-  * \ingroup Core_Module
-  *
-  * \brief Helper class to get the correct and optimized returned type of operator*
-  *
-  * \param Lhs the type of the left-hand side
-  * \param Rhs the type of the right-hand side
-  * \param ProductMode the type of the product (determined automatically by internal::product_mode)
-  *
-  * This class defines the typename Type representing the optimized product expression
-  * between two matrix expressions. In practice, using ProductReturnType<Lhs,Rhs>::Type
-  * is the recommended way to define the result type of a function returning an expression
-  * which involve a matrix product. The class Product should never be
-  * used directly.
-  *
-  * \sa class Product, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
-  */
-template<typename Lhs, typename Rhs, int ProductType>
-struct ProductReturnType
-{
-  // TODO use the nested type to reduce instanciations ????
-//   typedef typename internal::nested<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
-//   typedef typename internal::nested<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
-
-  typedef GeneralProduct<Lhs/*Nested*/, Rhs/*Nested*/, ProductType> Type;
-};
-
-template<typename Lhs, typename Rhs>
-struct ProductReturnType<Lhs,Rhs,CoeffBasedProductMode>
-{
-  typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
-  typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
-  typedef CoeffBasedProduct<LhsNested, RhsNested, EvalBeforeAssigningBit | EvalBeforeNestingBit> Type;
-};
-
-template<typename Lhs, typename Rhs>
-struct ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
-{
-  typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
-  typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
-  typedef CoeffBasedProduct<LhsNested, RhsNested, NestByRefBit> Type;
-};
-
-// this is a workaround for sun CC
-template<typename Lhs, typename Rhs>
-struct LazyProductReturnType : public ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
-{};
-
 /***********************************************************************
 *  Implementation of Inner Vector Vector Product
 ***********************************************************************/
@@ -187,119 +124,10 @@ struct LazyProductReturnType : public ProductReturnType<Lhs,Rhs,LazyCoeffBasedPr
 // product ends up to a row-vector times col-vector product... To tackle this use
 // case, we could have a specialization for Block<MatrixType,1,1> with: operator=(Scalar x);
 
-namespace internal {
-
-template<typename Lhs, typename Rhs>
-struct traits<GeneralProduct<Lhs,Rhs,InnerProduct> >
- : traits<Matrix<typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> >
-{};
-
-}
-
-template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, InnerProduct>
-  : internal::no_assignment_operator,
-    public Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1>
-{
-    typedef Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> Base;
-  public:
-    GeneralProduct(const Lhs& lhs, const Rhs& rhs)
-    {
-      EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
-      Base::coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
-    }
-
-    /** Convertion to scalar */
-    operator const typename Base::Scalar() const {
-      return Base::coeff(0,0);
-    }
-};
-
 /***********************************************************************
 *  Implementation of Outer Vector Vector Product
 ***********************************************************************/
 
-namespace internal {
-
-// Column major
-template<typename ProductType, typename Dest, typename Func>
-EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const false_type&)
-{
-  typedef typename Dest::Index Index;
-  // FIXME make sure lhs is sequentially stored
-  // FIXME not very good if rhs is real and lhs complex while alpha is real too
-  const Index cols = dest.cols();
-  for (Index j=0; j<cols; ++j)
-    func(dest.col(j), prod.rhs().coeff(0,j) * prod.lhs());
-}
-
-// Row major
-template<typename ProductType, typename Dest, typename Func>
-EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const true_type&) {
-  typedef typename Dest::Index Index;
-  // FIXME make sure rhs is sequentially stored
-  // FIXME not very good if lhs is real and rhs complex while alpha is real too
-  const Index rows = dest.rows();
-  for (Index i=0; i<rows; ++i)
-    func(dest.row(i), prod.lhs().coeff(i,0) * prod.rhs());
-}
-
-template<typename Lhs, typename Rhs>
-struct traits<GeneralProduct<Lhs,Rhs,OuterProduct> >
- : traits<ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs> >
-{};
-
-}
-
-template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, OuterProduct>
-  : public ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs>
-{
-    template<typename T> struct is_row_major : internal::conditional<(int(T::Flags)&RowMajorBit), internal::true_type, internal::false_type>::type {};
-    
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
-
-    GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {
-      EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-    }
-    
-    struct set  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived()  = src; } };
-    struct add  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } };
-    struct sub  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } };
-    struct adds {
-      Scalar m_scale;
-      adds(const Scalar& s) : m_scale(s) {}
-      template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const {
-        dst.const_cast_derived() += m_scale * src;
-      }
-    };
-    
-    template<typename Dest>
-    inline void evalTo(Dest& dest) const {
-      internal::outer_product_selector_run(*this, dest, set(), is_row_major<Dest>());
-    }
-    
-    template<typename Dest>
-    inline void addTo(Dest& dest) const {
-      internal::outer_product_selector_run(*this, dest, add(), is_row_major<Dest>());
-    }
-
-    template<typename Dest>
-    inline void subTo(Dest& dest) const {
-      internal::outer_product_selector_run(*this, dest, sub(), is_row_major<Dest>());
-    }
-
-    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
-    {
-      internal::outer_product_selector_run(*this, dest, adds(alpha), is_row_major<Dest>());
-    }
-};
-
 /***********************************************************************
 *  Implementation of General Matrix Vector Product
 ***********************************************************************/
@@ -313,60 +141,13 @@ class GeneralProduct<Lhs, Rhs, OuterProduct>
  */
 namespace internal {
 
-template<typename Lhs, typename Rhs>
-struct traits<GeneralProduct<Lhs,Rhs,GemvProduct> >
- : traits<ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs> >
-{};
-
 template<int Side, int StorageOrder, bool BlasCompatible>
-struct gemv_selector;
+struct gemv_dense_selector;
 
 } // end namespace internal
 
-template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, GemvProduct>
-  : public ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs>
-{
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
-
-    typedef typename Lhs::Scalar LhsScalar;
-    typedef typename Rhs::Scalar RhsScalar;
-
-    GeneralProduct(const Lhs& a_lhs, const Rhs& a_rhs) : Base(a_lhs,a_rhs)
-    {
-//       EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::Scalar, typename Rhs::Scalar>::value),
-//         YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-    }
-
-    enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
-    typedef typename internal::conditional<int(Side)==OnTheRight,_LhsNested,_RhsNested>::type MatrixType;
-
-    template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
-    {
-      eigen_assert(m_lhs.rows() == dst.rows() && m_rhs.cols() == dst.cols());
-      internal::gemv_selector<Side,(int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
-                       bool(internal::blas_traits<MatrixType>::HasUsableDirectAccess)>::run(*this, dst, alpha);
-    }
-};
-
 namespace internal {
 
-// The vector is on the left => transposition
-template<int StorageOrder, bool BlasCompatible>
-struct gemv_selector<OnTheLeft,StorageOrder,BlasCompatible>
-{
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
-  {
-    Transpose<Dest> destT(dest);
-    enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
-    gemv_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
-      ::run(GeneralProduct<Transpose<const typename ProductType::_RhsNested>,Transpose<const typename ProductType::_LhsNested>, GemvProduct>
-        (prod.rhs().transpose(), prod.lhs().transpose()), destT, alpha);
-  }
-};
-
 template<typename Scalar,int Size,int MaxSize,bool Cond> struct gemv_static_vector_if;
 
 template<typename Scalar,int Size,int MaxSize>
@@ -384,126 +165,161 @@ struct gemv_static_vector_if<Scalar,Size,Dynamic,true>
 template<typename Scalar,int Size,int MaxSize>
 struct gemv_static_vector_if<Scalar,Size,MaxSize,true>
 {
-  #if EIGEN_ALIGN_STATICALLY
-  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0> m_data;
-  EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; }
-  #else
-  // Some architectures cannot align on the stack,
-  // => let's manually enforce alignment by allocating more data and return the address of the first aligned element.
   enum {
     ForceAlignment  = internal::packet_traits<Scalar>::Vectorizable,
     PacketSize      = internal::packet_traits<Scalar>::size
   };
-  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?PacketSize:0),0> m_data;
+  #if EIGEN_MAX_STATIC_ALIGN_BYTES!=0
+  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0,EIGEN_PLAIN_ENUM_MIN(AlignedMax,PacketSize)> m_data;
+  EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; }
+  #else
+  // Some architectures cannot align on the stack,
+  // => let's manually enforce alignment by allocating more data and return the address of the first aligned element.
+  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?EIGEN_MAX_ALIGN_BYTES:0),0> m_data;
   EIGEN_STRONG_INLINE Scalar* data() {
     return ForceAlignment
-            ? reinterpret_cast<Scalar*>((reinterpret_cast<size_t>(m_data.array) & ~(size_t(15))) + 16)
+            ? reinterpret_cast<Scalar*>((internal::UIntPtr(m_data.array) & ~(std::size_t(EIGEN_MAX_ALIGN_BYTES-1))) + EIGEN_MAX_ALIGN_BYTES)
             : m_data.array;
   }
   #endif
 };
 
-template<> struct gemv_selector<OnTheRight,ColMajor,true>
+// The vector is on the left => transposition
+template<int StorageOrder, bool BlasCompatible>
+struct gemv_dense_selector<OnTheLeft,StorageOrder,BlasCompatible>
+{
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
+  {
+    Transpose<Dest> destT(dest);
+    enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
+    gemv_dense_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
+      ::run(rhs.transpose(), lhs.transpose(), destT, alpha);
+  }
+};
+
+template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
 {
-  template<typename ProductType, typename Dest>
-  static inline void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static inline void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::LhsScalar   LhsScalar;
-    typedef typename ProductType::RhsScalar   RhsScalar;
-    typedef typename ProductType::Scalar      ResScalar;
-    typedef typename ProductType::RealScalar  RealScalar;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
-    typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
-
-    ActualLhsType actualLhs = LhsBlasTraits::extract(prod.lhs());
-    ActualRhsType actualRhs = RhsBlasTraits::extract(prod.rhs());
-
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
+    typedef typename Lhs::Scalar   LhsScalar;
+    typedef typename Rhs::Scalar   RhsScalar;
+    typedef typename Dest::Scalar  ResScalar;
+    typedef typename Dest::RealScalar  RealScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+  
+    typedef Map<Matrix<ResScalar,Dynamic,1>, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits<ResScalar>::size)> MappedDest;
+
+    ActualLhsType actualLhs = LhsBlasTraits::extract(lhs);
+    ActualRhsType actualRhs = RhsBlasTraits::extract(rhs);
+
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
+                                  * RhsBlasTraits::extractScalarFactor(rhs);
+
+    // make sure Dest is a compile-time vector type (bug 1166)
+    typedef typename conditional<Dest::IsVectorAtCompileTime, Dest, typename Dest::ColXpr>::type ActualDest;
 
     enum {
       // FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
       // on, the other hand it is good for the cache to pack the vector anyways...
-      EvalToDestAtCompileTime = Dest::InnerStrideAtCompileTime==1,
+      EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1),
       ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
-      MightCannotUseDest = (Dest::InnerStrideAtCompileTime!=1) || ComplexByReal
+      MightCannotUseDest = (!EvalToDestAtCompileTime) || ComplexByReal
     };
 
-    gemv_static_vector_if<ResScalar,Dest::SizeAtCompileTime,Dest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
-
-    bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
-    bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
-    
+    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
     RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
 
-    ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
-                                                  evalToDest ? dest.data() : static_dest.data());
-    
-    if(!evalToDest)
+    if(!MightCannotUseDest)
     {
-      #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = dest.size();
-      EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      #endif
-      if(!alphaIsCompatible)
+      // shortcut if we are sure to be able to use dest directly,
+      // this ease the compiler to generate cleaner and more optimzized code for most common cases
+      general_matrix_vector_product
+          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          dest.data(), 1,
+          compatibleAlpha);
+    }
+    else
+    {
+      gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
+
+      const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
+      const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
+
+      ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
+                                                    evalToDest ? dest.data() : static_dest.data());
+
+      if(!evalToDest)
       {
-        MappedDest(actualDestPtr, dest.size()).setZero();
-        compatibleAlpha = RhsScalar(1);
+        #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+        Index size = dest.size();
+        EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+        #endif
+        if(!alphaIsCompatible)
+        {
+          MappedDest(actualDestPtr, dest.size()).setZero();
+          compatibleAlpha = RhsScalar(1);
+        }
+        else
+          MappedDest(actualDestPtr, dest.size()) = dest;
       }
-      else
-        MappedDest(actualDestPtr, dest.size()) = dest;
-    }
 
-    general_matrix_vector_product
-      <Index,LhsScalar,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
-        actualLhs.rows(), actualLhs.cols(),
-        actualLhs.data(), actualLhs.outerStride(),
-        actualRhs.data(), actualRhs.innerStride(),
-        actualDestPtr, 1,
-        compatibleAlpha);
+      general_matrix_vector_product
+          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          actualDestPtr, 1,
+          compatibleAlpha);
 
-    if (!evalToDest)
-    {
-      if(!alphaIsCompatible)
-        dest += actualAlpha * MappedDest(actualDestPtr, dest.size());
-      else
-        dest = MappedDest(actualDestPtr, dest.size());
+      if (!evalToDest)
+      {
+        if(!alphaIsCompatible)
+          dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
+        else
+          dest = MappedDest(actualDestPtr, dest.size());
+      }
     }
   }
 };
 
-template<> struct gemv_selector<OnTheRight,RowMajor,true>
+template<> struct gemv_dense_selector<OnTheRight,RowMajor,true>
 {
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef typename ProductType::LhsScalar LhsScalar;
-    typedef typename ProductType::RhsScalar RhsScalar;
-    typedef typename ProductType::Scalar    ResScalar;
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::_ActualRhsType _ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
-
-    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
-    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
-
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
+    typedef typename Lhs::Scalar   LhsScalar;
+    typedef typename Rhs::Scalar   RhsScalar;
+    typedef typename Dest::Scalar  ResScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
+
+    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
+    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
+
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
+                                  * RhsBlasTraits::extractScalarFactor(rhs);
 
     enum {
       // FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
       // on, the other hand it is good for the cache to pack the vector anyways...
-      DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1
+      DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1
     };
 
-    gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
+    gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
 
     ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
         DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
@@ -511,45 +327,48 @@ template<> struct gemv_selector<OnTheRight,RowMajor,true>
     if(!DirectlyUseRhs)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = actualRhs.size();
+      Index size = actualRhs.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
-      Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
+      Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
     }
 
+    typedef const_blas_data_mapper<LhsScalar,Index,RowMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,ColMajor> RhsMapper;
     general_matrix_vector_product
-      <Index,LhsScalar,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
+        <Index,LhsScalar,LhsMapper,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
         actualLhs.rows(), actualLhs.cols(),
-        actualLhs.data(), actualLhs.outerStride(),
-        actualRhsPtr, 1,
-        dest.data(), dest.innerStride(),
+        LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+        RhsMapper(actualRhsPtr, 1),
+        dest.data(), dest.col(0).innerStride(), //NOTE  if dest is not a vector at compile-time, then dest.innerStride() might be wrong. (bug 1166)
         actualAlpha);
   }
 };
 
-template<> struct gemv_selector<OnTheRight,ColMajor,false>
+template<> struct gemv_dense_selector<OnTheRight,ColMajor,false>
 {
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef typename Dest::Index Index;
-    // TODO makes sure dest is sequentially stored in memory, otherwise use a temp
-    const Index size = prod.rhs().rows();
+    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
+    // TODO if rhs is large enough it might be beneficial to make sure that dest is sequentially stored in memory, otherwise use a temp
+    typename nested_eval<Rhs,1>::type actual_rhs(rhs);
+    const Index size = rhs.rows();
     for(Index k=0; k<size; ++k)
-      dest += (alpha*prod.rhs().coeff(k)) * prod.lhs().col(k);
+      dest += (alpha*actual_rhs.coeff(k)) * lhs.col(k);
   }
 };
 
-template<> struct gemv_selector<OnTheRight,RowMajor,false>
+template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
 {
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef typename Dest::Index Index;
-    // TODO makes sure rhs is sequentially stored in memory, otherwise use a temp
-    const Index rows = prod.rows();
+    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
+    typename nested_eval<Rhs,Lhs::RowsAtCompileTime>::type actual_rhs(rhs);
+    const Index rows = dest.rows();
     for(Index i=0; i<rows; ++i)
-      dest.coeffRef(i) += alpha * (prod.lhs().row(i).cwiseProduct(prod.rhs().transpose())).sum();
+      dest.coeffRef(i) += alpha * (lhs.row(i).cwiseProduct(actual_rhs.transpose())).sum();
   }
 };
 
@@ -567,7 +386,7 @@ template<> struct gemv_selector<OnTheRight,RowMajor,false>
   */
 template<typename Derived>
 template<typename OtherDerived>
-inline const typename ProductReturnType<Derived, OtherDerived>::Type
+inline const Product<Derived, OtherDerived>
 MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
 {
   // A note regarding the function declaration: In MSVC, this function will sometimes
@@ -592,7 +411,8 @@ MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
 #ifdef EIGEN_DEBUG_PRODUCT
   internal::product_type<Derived,OtherDerived>::debug();
 #endif
-  return typename ProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
+
+  return Product<Derived, OtherDerived>(derived(), other.derived());
 }
 
 /** \returns an expression of the matrix product of \c *this and \a other without implicit evaluation.
@@ -608,7 +428,7 @@ MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
   */
 template<typename Derived>
 template<typename OtherDerived>
-const typename LazyProductReturnType<Derived,OtherDerived>::Type
+const Product<Derived,OtherDerived,LazyProduct>
 MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
 {
   enum {
@@ -627,7 +447,7 @@ MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
     INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
   EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
 
-  return typename LazyProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
+  return Product<Derived,OtherDerived,LazyProduct>(derived(), other.derived());
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/GenericPacketMath.h b/extern/Eigen3/Eigen/src/Core/GenericPacketMath.h
index 5f783ebeee8..029f8ac36fb 100644
--- a/extern/Eigen3/Eigen/src/Core/GenericPacketMath.h
+++ b/extern/Eigen3/Eigen/src/Core/GenericPacketMath.h
@@ -42,21 +42,28 @@ namespace internal {
 struct default_packet_traits
 {
   enum {
+    HasHalfPacket = 0,
+
     HasAdd    = 1,
     HasSub    = 1,
     HasMul    = 1,
     HasNegate = 1,
     HasAbs    = 1,
+    HasArg    = 0,
     HasAbs2   = 1,
     HasMin    = 1,
     HasMax    = 1,
     HasConj   = 1,
     HasSetLinear = 1,
+    HasBlend  = 0,
 
     HasDiv    = 0,
     HasSqrt   = 0,
+    HasRsqrt  = 0,
     HasExp    = 0,
     HasLog    = 0,
+    HasLog1p  = 0,
+    HasLog10  = 0,
     HasPow    = 0,
 
     HasSin    = 0,
@@ -64,17 +71,37 @@ struct default_packet_traits
     HasTan    = 0,
     HasASin   = 0,
     HasACos   = 0,
-    HasATan   = 0
+    HasATan   = 0,
+    HasSinh   = 0,
+    HasCosh   = 0,
+    HasTanh   = 0,
+    HasLGamma = 0,
+    HasDiGamma = 0,
+    HasZeta = 0,
+    HasPolygamma = 0,
+    HasErf = 0,
+    HasErfc = 0,
+    HasIGamma = 0,
+    HasIGammac = 0,
+    HasBetaInc = 0,
+
+    HasRound  = 0,
+    HasFloor  = 0,
+    HasCeil   = 0,
+
+    HasSign   = 0
   };
 };
 
 template<typename T> struct packet_traits : default_packet_traits
 {
   typedef T type;
+  typedef T half;
   enum {
     Vectorizable = 0,
     size = 1,
-    AlignedOnScalar = 0
+    AlignedOnScalar = 0,
+    HasHalfPacket = 0
   };
   enum {
     HasAdd    = 0,
@@ -90,135 +117,239 @@ template<typename T> struct packet_traits : default_packet_traits
   };
 };
 
+template<typename T> struct packet_traits<const T> : packet_traits<T> { };
+
+template <typename Src, typename Tgt> struct type_casting_traits {
+  enum {
+    VectorizedCast = 0,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+
+/** \internal \returns static_cast<TgtType>(a) (coeff-wise) */
+template <typename SrcPacket, typename TgtPacket>
+EIGEN_DEVICE_FUNC inline TgtPacket
+pcast(const SrcPacket& a) {
+  return static_cast<TgtPacket>(a);
+}
+template <typename SrcPacket, typename TgtPacket>
+EIGEN_DEVICE_FUNC inline TgtPacket
+pcast(const SrcPacket& a, const SrcPacket& /*b*/) {
+  return static_cast<TgtPacket>(a);
+}
+
+template <typename SrcPacket, typename TgtPacket>
+EIGEN_DEVICE_FUNC inline TgtPacket
+pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/) {
+  return static_cast<TgtPacket>(a);
+}
+
 /** \internal \returns a + b (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 padd(const Packet& a,
         const Packet& b) { return a+b; }
 
 /** \internal \returns a - b (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 psub(const Packet& a,
         const Packet& b) { return a-b; }
 
 /** \internal \returns -a (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pnegate(const Packet& a) { return -a; }
 
 /** \internal \returns conj(a) (coeff-wise) */
-template<typename Packet> inline Packet
+
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pconj(const Packet& a) { return numext::conj(a); }
 
 /** \internal \returns a * b (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pmul(const Packet& a,
         const Packet& b) { return a*b; }
 
 /** \internal \returns a / b (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pdiv(const Packet& a,
         const Packet& b) { return a/b; }
 
 /** \internal \returns the min of \a a and \a b  (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pmin(const Packet& a,
-        const Packet& b) { using std::min; return (min)(a, b); }
+        const Packet& b) { return numext::mini(a, b); }
 
 /** \internal \returns the max of \a a and \a b  (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pmax(const Packet& a,
-        const Packet& b) { using std::max; return (max)(a, b); }
+        const Packet& b) { return numext::maxi(a, b); }
 
 /** \internal \returns the absolute value of \a a */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pabs(const Packet& a) { using std::abs; return abs(a); }
 
+/** \internal \returns the phase angle of \a a */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+parg(const Packet& a) { using numext::arg; return arg(a); }
+
 /** \internal \returns the bitwise and of \a a and \a b */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pand(const Packet& a, const Packet& b) { return a & b; }
 
 /** \internal \returns the bitwise or of \a a and \a b */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 por(const Packet& a, const Packet& b) { return a | b; }
 
 /** \internal \returns the bitwise xor of \a a and \a b */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pxor(const Packet& a, const Packet& b) { return a ^ b; }
 
 /** \internal \returns the bitwise andnot of \a a and \a b */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pandnot(const Packet& a, const Packet& b) { return a & (!b); }
 
 /** \internal \returns a packet version of \a *from, from must be 16 bytes aligned */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pload(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
 /** \internal \returns a packet version of \a *from, (un-aligned load) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 ploadu(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
+/** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
+
+/** \internal \returns a packet with constant coefficients \a a[0], e.g.: (a[0],a[0],a[0],a[0]) */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pload1(const typename unpacket_traits<Packet>::type  *a) { return pset1<Packet>(*a); }
+
 /** \internal \returns a packet with elements of \a *from duplicated.
-  * For instance, for a packet of 8 elements, 4 scalar will be read from \a *from and
-  * duplicated to form: {from[0],from[0],from[1],from[1],,from[2],from[2],,from[3],from[3]}
+  * For instance, for a packet of 8 elements, 4 scalars will be read from \a *from and
+  * duplicated to form: {from[0],from[0],from[1],from[1],from[2],from[2],from[3],from[3]}
   * Currently, this function is only used for scalar * complex products.
- */
-template<typename Packet> inline Packet
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
 ploaddup(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
-/** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */
-template<typename Packet> inline Packet
-pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
+/** \internal \returns a packet with elements of \a *from quadrupled.
+  * For instance, for a packet of 8 elements, 2 scalars will be read from \a *from and
+  * replicated to form: {from[0],from[0],from[0],from[0],from[1],from[1],from[1],from[1]}
+  * Currently, this function is only used in matrix products.
+  * For packet-size smaller or equal to 4, this function is equivalent to pload1 
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+ploadquad(const typename unpacket_traits<Packet>::type* from)
+{ return pload1<Packet>(from); }
+
+/** \internal equivalent to
+  * \code
+  * a0 = pload1(a+0);
+  * a1 = pload1(a+1);
+  * a2 = pload1(a+2);
+  * a3 = pload1(a+3);
+  * \endcode
+  * \sa pset1, pload1, ploaddup, pbroadcast2
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC
+inline void pbroadcast4(const typename unpacket_traits<Packet>::type *a,
+                        Packet& a0, Packet& a1, Packet& a2, Packet& a3)
+{
+  a0 = pload1<Packet>(a+0);
+  a1 = pload1<Packet>(a+1);
+  a2 = pload1<Packet>(a+2);
+  a3 = pload1<Packet>(a+3);
+}
+
+/** \internal equivalent to
+  * \code
+  * a0 = pload1(a+0);
+  * a1 = pload1(a+1);
+  * \endcode
+  * \sa pset1, pload1, ploaddup, pbroadcast4
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC
+inline void pbroadcast2(const typename unpacket_traits<Packet>::type *a,
+                        Packet& a0, Packet& a1)
+{
+  a0 = pload1<Packet>(a+0);
+  a1 = pload1<Packet>(a+1);
+}
 
 /** \internal \brief Returns a packet with coefficients (a,a+1,...,a+packet_size-1). */
-template<typename Scalar> inline typename packet_traits<Scalar>::type
-plset(const Scalar& a) { return a; }
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
+plset(const typename unpacket_traits<Packet>::type& a) { return a; }
 
 /** \internal copy the packet \a from to \a *to, \a to must be 16 bytes aligned */
-template<typename Scalar, typename Packet> inline void pstore(Scalar* to, const Packet& from)
+template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstore(Scalar* to, const Packet& from)
 { (*to) = from; }
 
 /** \internal copy the packet \a from to \a *to, (un-aligned store) */
-template<typename Scalar, typename Packet> inline void pstoreu(Scalar* to, const Packet& from)
-{ (*to) = from; }
+template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu(Scalar* to, const Packet& from)
+{  (*to) = from; }
+
+ template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline Packet pgather(const Scalar* from, Index /*stride*/)
+ { return ploadu<Packet>(from); }
+
+ template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pscatter(Scalar* to, const Packet& from, Index /*stride*/)
+ { pstore(to, from); }
 
 /** \internal tries to do cache prefetching of \a addr */
-template<typename Scalar> inline void prefetch(const Scalar* addr)
+template<typename Scalar> EIGEN_DEVICE_FUNC inline void prefetch(const Scalar* addr)
 {
-#if !defined(_MSC_VER)
-__builtin_prefetch(addr);
+#ifdef __CUDA_ARCH__
+#if defined(__LP64__)
+  // 64-bit pointer operand constraint for inlined asm
+  asm(" prefetch.L1 [ %1 ];" : "=l"(addr) : "l"(addr));
+#else
+  // 32-bit pointer operand constraint for inlined asm
+  asm(" prefetch.L1 [ %1 ];" : "=r"(addr) : "r"(addr));
+#endif
+#elif (!EIGEN_COMP_MSVC) && (EIGEN_COMP_GNUC || EIGEN_COMP_CLANG || EIGEN_COMP_ICC)
+  __builtin_prefetch(addr);
 #endif
 }
 
 /** \internal \returns the first element of a packet */
-template<typename Packet> inline typename unpacket_traits<Packet>::type pfirst(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type pfirst(const Packet& a)
 { return a; }
 
 /** \internal \returns a packet where the element i contains the sum of the packet of \a vec[i] */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 preduxp(const Packet* vecs) { return vecs[0]; }
 
 /** \internal \returns the sum of the elements of \a a*/
-template<typename Packet> inline typename unpacket_traits<Packet>::type predux(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux(const Packet& a)
+{ return a; }
+
+/** \internal \returns the sum of the elements of \a a by block of 4 elements.
+  * For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7}
+  * For packet-size smaller or equal to 4, this boils down to a noop.
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC inline
+typename conditional<(unpacket_traits<Packet>::size%8)==0,typename unpacket_traits<Packet>::half,Packet>::type
+predux_downto4(const Packet& a)
 { return a; }
 
 /** \internal \returns the product of the elements of \a a*/
-template<typename Packet> inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a)
 { return a; }
 
 /** \internal \returns the min of the elements of \a a*/
-template<typename Packet> inline typename unpacket_traits<Packet>::type predux_min(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(const Packet& a)
 { return a; }
 
 /** \internal \returns the max of the elements of \a a*/
-template<typename Packet> inline typename unpacket_traits<Packet>::type predux_max(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(const Packet& a)
 { return a; }
 
 /** \internal \returns the reversed elements of \a a*/
-template<typename Packet> inline Packet preverse(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet preverse(const Packet& a)
 { return a; }
 
-
 /** \internal \returns \a a with real and imaginary part flipped (for complex type only) */
-template<typename Packet> inline Packet pcplxflip(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet pcplxflip(const Packet& a)
 {
   // FIXME: uncomment the following in case we drop the internal imag and real functions.
 //   using std::imag;
@@ -250,6 +381,22 @@ Packet pasin(const Packet& a) { using std::asin; return asin(a); }
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet pacos(const Packet& a) { using std::acos; return acos(a); }
 
+/** \internal \returns the arc tangent of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet patan(const Packet& a) { using std::atan; return atan(a); }
+
+/** \internal \returns the hyperbolic sine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet psinh(const Packet& a) { using std::sinh; return sinh(a); }
+
+/** \internal \returns the hyperbolic cosine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pcosh(const Packet& a) { using std::cosh; return cosh(a); }
+
+/** \internal \returns the hyperbolic tan of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet ptanh(const Packet& a) { using std::tanh; return tanh(a); }
+
 /** \internal \returns the exp of \a a (coeff-wise) */
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet pexp(const Packet& a) { using std::exp; return exp(a); }
@@ -258,10 +405,36 @@ Packet pexp(const Packet& a) { using std::exp; return exp(a); }
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet plog(const Packet& a) { using std::log; return log(a); }
 
+/** \internal \returns the log1p of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet plog1p(const Packet& a) { return numext::log1p(a); }
+
+/** \internal \returns the log10 of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet plog10(const Packet& a) { using std::log10; return log10(a); }
+
 /** \internal \returns the square-root of \a a (coeff-wise) */
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet psqrt(const Packet& a) { using std::sqrt; return sqrt(a); }
 
+/** \internal \returns the reciprocal square-root of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet prsqrt(const Packet& a) {
+  return pdiv(pset1<Packet>(1), psqrt(a));
+}
+
+/** \internal \returns the rounded value of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pround(const Packet& a) { using numext::round; return round(a); }
+
+/** \internal \returns the floor of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pfloor(const Packet& a) { using numext::floor; return floor(a); }
+
+/** \internal \returns the ceil of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pceil(const Packet& a) { using numext::ceil; return ceil(a); }
+
 /***************************************************************************
 * The following functions might not have to be overwritten for vectorized types
 ***************************************************************************/
@@ -275,34 +448,45 @@ inline void pstore1(typename unpacket_traits<Packet>::type* to, const typename u
 }
 
 /** \internal \returns a * b + c (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pmadd(const Packet&  a,
          const Packet&  b,
          const Packet&  c)
 { return padd(pmul(a, b),c); }
 
 /** \internal \returns a packet version of \a *from.
-  * If LoadMode equals #Aligned, \a from must be 16 bytes aligned */
-template<typename Packet, int LoadMode>
-inline Packet ploadt(const typename unpacket_traits<Packet>::type* from)
+  * The pointer \a from must be aligned on a \a Alignment bytes boundary. */
+template<typename Packet, int Alignment>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt(const typename unpacket_traits<Packet>::type* from)
 {
-  if(LoadMode == Aligned)
+  if(Alignment >= unpacket_traits<Packet>::alignment)
     return pload<Packet>(from);
   else
     return ploadu<Packet>(from);
 }
 
 /** \internal copy the packet \a from to \a *to.
-  * If StoreMode equals #Aligned, \a to must be 16 bytes aligned */
-template<typename Scalar, typename Packet, int LoadMode>
-inline void pstoret(Scalar* to, const Packet& from)
+  * The pointer \a from must be aligned on a \a Alignment bytes boundary. */
+template<typename Scalar, typename Packet, int Alignment>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void pstoret(Scalar* to, const Packet& from)
 {
-  if(LoadMode == Aligned)
+  if(Alignment >= unpacket_traits<Packet>::alignment)
     pstore(to, from);
   else
     pstoreu(to, from);
 }
 
+/** \internal \returns a packet version of \a *from.
+  * Unlike ploadt, ploadt_ro takes advantage of the read-only memory path on the
+  * hardware if available to speedup the loading of data that won't be modified
+  * by the current computation.
+  */
+template<typename Packet, int LoadMode>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt_ro(const typename unpacket_traits<Packet>::type* from)
+{
+  return ploadt<Packet, LoadMode>(from);
+}
+
 /** \internal default implementation of palign() allowing partial specialization */
 template<int Offset,typename PacketType>
 struct palign_impl
@@ -336,15 +520,74 @@ inline void palign(PacketType& first, const PacketType& second)
 * Fast complex products (GCC generates a function call which is very slow)
 ***************************************************************************/
 
+// Eigen+CUDA does not support complexes.
+#ifndef __CUDACC__
+
 template<> inline std::complex<float> pmul(const std::complex<float>& a, const std::complex<float>& b)
 { return std::complex<float>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
 
 template<> inline std::complex<double> pmul(const std::complex<double>& a, const std::complex<double>& b)
 { return std::complex<double>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
 
+#endif
+
+
+/***************************************************************************
+ * PacketBlock, that is a collection of N packets where the number of words
+ * in the packet is a multiple of N.
+***************************************************************************/
+template <typename Packet,int N=unpacket_traits<Packet>::size> struct PacketBlock {
+  Packet packet[N];
+};
+
+template<typename Packet> EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet,1>& /*kernel*/) {
+  // Nothing to do in the scalar case, i.e. a 1x1 matrix.
+}
+
+/***************************************************************************
+ * Selector, i.e. vector of N boolean values used to select (i.e. blend)
+ * words from 2 packets.
+***************************************************************************/
+template <size_t N> struct Selector {
+  bool select[N];
+};
+
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pblend(const Selector<unpacket_traits<Packet>::size>& ifPacket, const Packet& thenPacket, const Packet& elsePacket) {
+  return ifPacket.select[0] ? thenPacket : elsePacket;
+}
+
+/** \internal \returns \a a with the first coefficient replaced by the scalar b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pinsertfirst(const Packet& a, typename unpacket_traits<Packet>::type b)
+{
+  // Default implementation based on pblend.
+  // It must be specialized for higher performance.
+  Selector<unpacket_traits<Packet>::size> mask;
+  mask.select[0] = true;
+  // This for loop should be optimized away by the compiler.
+  for(Index i=1; i<unpacket_traits<Packet>::size; ++i)
+    mask.select[i] = false;
+  return pblend(mask, pset1<Packet>(b), a);
+}
+
+/** \internal \returns \a a with the last coefficient replaced by the scalar b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pinsertlast(const Packet& a, typename unpacket_traits<Packet>::type b)
+{
+  // Default implementation based on pblend.
+  // It must be specialized for higher performance.
+  Selector<unpacket_traits<Packet>::size> mask;
+  // This for loop should be optimized away by the compiler.
+  for(Index i=0; i<unpacket_traits<Packet>::size-1; ++i)
+    mask.select[i] = false;
+  mask.select[unpacket_traits<Packet>::size-1] = true;
+  return pblend(mask, pset1<Packet>(b), a);
+}
+
 } // end namespace internal
 
 } // end namespace Eigen
 
 #endif // EIGEN_GENERIC_PACKET_MATH_H
-
diff --git a/extern/Eigen3/Eigen/src/Core/GlobalFunctions.h b/extern/Eigen3/Eigen/src/Core/GlobalFunctions.h
index 2acf9772334..769dc255c20 100644
--- a/extern/Eigen3/Eigen/src/Core/GlobalFunctions.h
+++ b/extern/Eigen3/Eigen/src/Core/GlobalFunctions.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2010-2012 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2010-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -11,13 +11,30 @@
 #ifndef EIGEN_GLOBAL_FUNCTIONS_H
 #define EIGEN_GLOBAL_FUNCTIONS_H
 
-#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR) \
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+
+#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
+  /** \returns an expression of the coefficient-wise DOC_OP of \a x
+
+    DOC_DETAILS
+
+    \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_##NAME">Math functions</a>, class CwiseUnaryOp
+    */ \
+  template<typename Derived> \
+  inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
+  NAME(const Eigen::ArrayBase<Derived>& x);
+
+#else
+
+#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
   template<typename Derived> \
   inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
-  NAME(const Eigen::ArrayBase<Derived>& x) { \
-    return x.derived(); \
+  (NAME)(const Eigen::ArrayBase<Derived>& x) { \
+    return Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived>(x.derived()); \
   }
 
+#endif // EIGEN_PARSED_BY_DOXYGEN
+
 #define EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(NAME,FUNCTOR) \
   \
   template<typename Derived> \
@@ -30,55 +47,133 @@
   { \
     static inline typename NAME##_retval<ArrayBase<Derived> >::type run(const Eigen::ArrayBase<Derived>& x) \
     { \
-      return x.derived(); \
+      return typename NAME##_retval<ArrayBase<Derived> >::type(x.derived()); \
     } \
   };
 
-
 namespace Eigen
 {
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op,real part,\sa ArrayBase::real)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op,imaginary part,\sa ArrayBase::imag)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op,complex conjugate,\sa ArrayBase::conjugate)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(inverse,scalar_inverse_op,inverse,\sa ArrayBase::inverse)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op,sine,\sa ArrayBase::sin)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op,cosine,\sa ArrayBase::cos)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op,tangent,\sa ArrayBase::tan)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atan,scalar_atan_op,arc-tangent,\sa ArrayBase::atan)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op,arc-sine,\sa ArrayBase::asin)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op,arc-consine,\sa ArrayBase::acos)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sinh,scalar_sinh_op,hyperbolic sine,\sa ArrayBase::sinh)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cosh,scalar_cosh_op,hyperbolic cosine,\sa ArrayBase::cosh)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tanh,scalar_tanh_op,hyperbolic tangent,\sa ArrayBase::tanh)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(lgamma,scalar_lgamma_op,natural logarithm of the gamma function,\sa ArrayBase::lgamma)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(digamma,scalar_digamma_op,derivative of lgamma,\sa ArrayBase::digamma)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erf,scalar_erf_op,error function,\sa ArrayBase::erf)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erfc,scalar_erfc_op,complement error function,\sa ArrayBase::erfc)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op,exponential,\sa ArrayBase::exp)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op,natural logarithm,\sa Eigen::log10 DOXCOMMA ArrayBase::log)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log1p,scalar_log1p_op,natural logarithm of 1 plus the value,\sa ArrayBase::log1p)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op,base 10 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op,absolute value,\sa ArrayBase::abs DOXCOMMA MatrixBase::cwiseAbs)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs2,scalar_abs2_op,squared absolute value,\sa ArrayBase::abs2 DOXCOMMA MatrixBase::cwiseAbs2)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op,complex argument,\sa ArrayBase::arg)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op,square root,\sa ArrayBase::sqrt DOXCOMMA MatrixBase::cwiseSqrt)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rsqrt,scalar_rsqrt_op,reciprocal square root,\sa ArrayBase::rsqrt)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(square,scalar_square_op,square (power 2),\sa Eigen::abs2 DOXCOMMA Eigen::pow DOXCOMMA ArrayBase::square)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cube,scalar_cube_op,cube (power 3),\sa Eigen::pow DOXCOMMA ArrayBase::cube)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(round,scalar_round_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(floor,scalar_floor_op,nearest integer not greater than the giben value,\sa Eigen::ceil DOXCOMMA ArrayBase::floor)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ceil,scalar_ceil_op,nearest integer not less than the giben value,\sa Eigen::floor DOXCOMMA ArrayBase::ceil)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isnan,scalar_isnan_op,not-a-number test,\sa Eigen::isinf DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isnan)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isinf,scalar_isinf_op,infinite value test,\sa Eigen::isnan DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isinf)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isfinite,scalar_isfinite_op,finite value test,\sa Eigen::isinf DOXCOMMA Eigen::isnan DOXCOMMA ArrayBase::isfinite)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sign,scalar_sign_op,sign (or 0),\sa ArrayBase::sign)
   
+  /** \returns an expression of the coefficient-wise power of \a x to the given constant \a exponent.
+    *
+    * \tparam ScalarExponent is the scalar type of \a exponent. It must be compatible with the scalar type of the given expression (\c Derived::Scalar).
+    *
+    * \sa ArrayBase::pow()
+    *
+    * \relates ArrayBase
+    */
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+  template<typename Derived,typename ScalarExponent>
+  inline const CwiseBinaryOp<internal::scalar_pow_op<Derived::Scalar,ScalarExponent>,Derived,Constant<ScalarExponent> >
+  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent);
+#else
+  template<typename Derived,typename ScalarExponent>
+  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,ScalarExponent>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent),
+          const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,ScalarExponent,pow) >::type
+  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent) {
+    return x.derived().pow(exponent);
+  }
+
   template<typename Derived>
-  inline const Eigen::CwiseUnaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar>, const Derived>
+  inline const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename Derived::Scalar,pow)
   pow(const Eigen::ArrayBase<Derived>& x, const typename Derived::Scalar& exponent) {
     return x.derived().pow(exponent);
   }
+#endif
 
-  template<typename Derived>
-  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_binary_pow_op<typename Derived::Scalar, typename Derived::Scalar>, const Derived, const Derived>
-  pow(const Eigen::ArrayBase<Derived>& x, const Eigen::ArrayBase<Derived>& exponents) 
+  /** \returns an expression of the coefficient-wise power of \a x to the given array of \a exponents.
+    *
+    * This function computes the coefficient-wise power.
+    *
+    * Example: \include Cwise_array_power_array.cpp
+    * Output: \verbinclude Cwise_array_power_array.out
+    * 
+    * \sa ArrayBase::pow()
+    *
+    * \relates ArrayBase
+    */
+  template<typename Derived,typename ExponentDerived>
+  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>
+  pow(const Eigen::ArrayBase<Derived>& x, const Eigen::ArrayBase<ExponentDerived>& exponents) 
   {
-    return Eigen::CwiseBinaryOp<Eigen::internal::scalar_binary_pow_op<typename Derived::Scalar, typename Derived::Scalar>, const Derived, const Derived>(
+    return Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>(
       x.derived(),
       exponents.derived()
     );
   }
   
-  /**
-  * \brief Component-wise division of a scalar by array elements.
-  **/
-  template <typename Derived>
-  inline const Eigen::CwiseUnaryOp<Eigen::internal::scalar_inverse_mult_op<typename Derived::Scalar>, const Derived>
-    operator/(const typename Derived::Scalar& s, const Eigen::ArrayBase<Derived>& a)
+  /** \returns an expression of the coefficient-wise power of the scalar \a x to the given array of \a exponents.
+    *
+    * This function computes the coefficient-wise power between a scalar and an array of exponents.
+    *
+    * \tparam Scalar is the scalar type of \a x. It must be compatible with the scalar type of the given array expression (\c Derived::Scalar).
+    *
+    * Example: \include Cwise_scalar_power_array.cpp
+    * Output: \verbinclude Cwise_scalar_power_array.out
+    * 
+    * \sa ArrayBase::pow()
+    *
+    * \relates ArrayBase
+    */
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+  template<typename Scalar,typename Derived>
+  inline const CwiseBinaryOp<internal::scalar_pow_op<Scalar,Derived::Scalar>,Constant<Scalar>,Derived>
+  pow(const Scalar& x,const Eigen::ArrayBase<Derived>& x);
+#else
+  template<typename Scalar, typename Derived>
+  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,Scalar>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar),
+          const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow) >::type
+  pow(const Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
   {
-    return Eigen::CwiseUnaryOp<Eigen::internal::scalar_inverse_mult_op<typename Derived::Scalar>, const Derived>(
-      a.derived(),
-      Eigen::internal::scalar_inverse_mult_op<typename Derived::Scalar>(s)  
-    );
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow)(
+            typename internal::plain_constant_type<Derived,Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
   }
 
+  template<typename Derived>
+  inline const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)
+  pow(const typename Derived::Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
+  {
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)(
+      typename internal::plain_constant_type<Derived,typename Derived::Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
+  }
+#endif
+
+
   namespace internal
   {
     EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(real,scalar_real_op)
diff --git a/extern/Eigen3/Eigen/src/Core/IO.h b/extern/Eigen3/Eigen/src/Core/IO.h
index 8d4bc59e9dd..da7fd6cce2e 100644
--- a/extern/Eigen3/Eigen/src/Core/IO.h
+++ b/extern/Eigen3/Eigen/src/Core/IO.h
@@ -49,7 +49,7 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
   */
 struct IOFormat
 {
-  /** Default contructor, see class IOFormat for the meaning of the parameters */
+  /** Default constructor, see class IOFormat for the meaning of the parameters */
   IOFormat(int _precision = StreamPrecision, int _flags = 0,
     const std::string& _coeffSeparator = " ",
     const std::string& _rowSeparator = "\n", const std::string& _rowPrefix="", const std::string& _rowSuffix="",
@@ -57,6 +57,10 @@ struct IOFormat
   : matPrefix(_matPrefix), matSuffix(_matSuffix), rowPrefix(_rowPrefix), rowSuffix(_rowSuffix), rowSeparator(_rowSeparator),
     rowSpacer(""), coeffSeparator(_coeffSeparator), precision(_precision), flags(_flags)
   {
+    // TODO check if rowPrefix, rowSuffix or rowSeparator contains a newline
+    // don't add rowSpacer if columns are not to be aligned
+    if((flags & DontAlignCols))
+      return;
     int i = int(matSuffix.length())-1;
     while (i>=0 && matSuffix[i]!='\n')
     {
@@ -76,7 +80,7 @@ struct IOFormat
   *
   * \brief Pseudo expression providing matrix output with given format
   *
-  * \param ExpressionType the type of the object on which IO stream operations are performed
+  * \tparam ExpressionType the type of the object on which IO stream operations are performed
   *
   * This class represents an expression with stream operators controlled by a given IOFormat.
   * It is the return type of DenseBase::format()
@@ -101,52 +105,24 @@ class WithFormat
     }
 
   protected:
-    const typename ExpressionType::Nested m_matrix;
+    typename ExpressionType::Nested m_matrix;
     IOFormat m_format;
 };
 
-/** \returns a WithFormat proxy object allowing to print a matrix the with given
-  * format \a fmt.
-  *
-  * See class IOFormat for some examples.
-  *
-  * \sa class IOFormat, class WithFormat
-  */
-template<typename Derived>
-inline const WithFormat<Derived>
-DenseBase<Derived>::format(const IOFormat& fmt) const
-{
-  return WithFormat<Derived>(derived(), fmt);
-}
-
 namespace internal {
 
-template<typename Scalar, bool IsInteger>
-struct significant_decimals_default_impl
-{
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  static inline int run()
-  {
-    using std::ceil;
-    using std::log;
-    return cast<RealScalar,int>(ceil(-log(NumTraits<RealScalar>::epsilon())/log(RealScalar(10))));
-  }
-};
-
+// NOTE: This helper is kept for backward compatibility with previous code specializing
+//       this internal::significant_decimals_impl structure. In the future we should directly
+//       call digits10() which has been introduced in July 2016 in 3.3.
 template<typename Scalar>
-struct significant_decimals_default_impl<Scalar, true>
+struct significant_decimals_impl
 {
   static inline int run()
   {
-    return 0;
+    return NumTraits<Scalar>::digits10();
   }
 };
 
-template<typename Scalar>
-struct significant_decimals_impl
-  : significant_decimals_default_impl<Scalar, NumTraits<Scalar>::IsInteger>
-{};
-
 /** \internal
   * print the matrix \a _m to the output stream \a s using the output format \a fmt */
 template<typename Derived>
@@ -160,7 +136,6 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
   
   typename Derived::Nested m = _m;
   typedef typename Derived::Scalar Scalar;
-  typedef typename Derived::Index Index;
 
   Index width = 0;
 
diff --git a/extern/Eigen3/Eigen/src/Core/Inverse.h b/extern/Eigen3/Eigen/src/Core/Inverse.h
new file mode 100644
index 00000000000..b76f0439d80
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/Inverse.h
@@ -0,0 +1,118 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_INVERSE_H
+#define EIGEN_INVERSE_H
+
+namespace Eigen { 
+
+template<typename XprType,typename StorageKind> class InverseImpl;
+
+namespace internal {
+
+template<typename XprType>
+struct traits<Inverse<XprType> >
+  : traits<typename XprType::PlainObject>
+{
+  typedef typename XprType::PlainObject PlainObject;
+  typedef traits<PlainObject> BaseTraits;
+  enum {
+    Flags = BaseTraits::Flags & RowMajorBit
+  };
+};
+
+} // end namespace internal
+
+/** \class Inverse
+  *
+  * \brief Expression of the inverse of another expression
+  *
+  * \tparam XprType the type of the expression we are taking the inverse
+  *
+  * This class represents an abstract expression of A.inverse()
+  * and most of the time this is the only way it is used.
+  *
+  */
+template<typename XprType>
+class Inverse : public InverseImpl<XprType,typename internal::traits<XprType>::StorageKind>
+{
+public:
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename XprType::PlainObject                       PlainObject;
+  typedef typename XprType::Scalar                            Scalar;
+  typedef typename internal::ref_selector<XprType>::type      XprTypeNested;
+  typedef typename internal::remove_all<XprTypeNested>::type  XprTypeNestedCleaned;
+  typedef typename internal::ref_selector<Inverse>::type Nested;
+  typedef typename internal::remove_all<XprType>::type NestedExpression;
+  
+  explicit EIGEN_DEVICE_FUNC Inverse(const XprType &xpr)
+    : m_xpr(xpr)
+  {}
+
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
+
+  EIGEN_DEVICE_FUNC const XprTypeNestedCleaned& nestedExpression() const { return m_xpr; }
+
+protected:
+  XprTypeNested m_xpr;
+};
+
+// Generic API dispatcher
+template<typename XprType, typename StorageKind>
+class InverseImpl
+  : public internal::generic_xpr_base<Inverse<XprType> >::type
+{
+public:
+  typedef typename internal::generic_xpr_base<Inverse<XprType> >::type Base;
+  typedef typename XprType::Scalar Scalar;
+private:
+
+  Scalar coeff(Index row, Index col) const;
+  Scalar coeff(Index i) const;
+};
+
+namespace internal {
+
+/** \internal
+  * \brief Default evaluator for Inverse expression.
+  * 
+  * This default evaluator for Inverse expression simply evaluate the inverse into a temporary
+  * by a call to internal::call_assignment_no_alias.
+  * Therefore, inverse implementers only have to specialize Assignment<Dst,Inverse<...>, ...> for
+  * there own nested expression.
+  *
+  * \sa class Inverse
+  */
+template<typename ArgType>
+struct unary_evaluator<Inverse<ArgType> >
+  : public evaluator<typename Inverse<ArgType>::PlainObject>
+{
+  typedef Inverse<ArgType> InverseType;
+  typedef typename InverseType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+  
+  enum { Flags = Base::Flags | EvalBeforeNestingBit };
+
+  unary_evaluator(const InverseType& inv_xpr)
+    : m_result(inv_xpr.rows(), inv_xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    internal::call_assignment_no_alias(m_result, inv_xpr);
+  }
+  
+protected:
+  PlainObject m_result;
+};
+  
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_INVERSE_H
diff --git a/extern/Eigen3/Eigen/src/Core/Map.h b/extern/Eigen3/Eigen/src/Core/Map.h
index f804c89d63e..548bf9a2d55 100644
--- a/extern/Eigen3/Eigen/src/Core/Map.h
+++ b/extern/Eigen3/Eigen/src/Core/Map.h
@@ -13,13 +13,41 @@
 
 namespace Eigen { 
 
+namespace internal {
+template<typename PlainObjectType, int MapOptions, typename StrideType>
+struct traits<Map<PlainObjectType, MapOptions, StrideType> >
+  : public traits<PlainObjectType>
+{
+  typedef traits<PlainObjectType> TraitsBase;
+  enum {
+    PlainObjectTypeInnerSize = ((traits<PlainObjectType>::Flags&RowMajorBit)==RowMajorBit)
+                             ? PlainObjectType::ColsAtCompileTime
+                             : PlainObjectType::RowsAtCompileTime,
+
+    InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
+                             ? int(PlainObjectType::InnerStrideAtCompileTime)
+                             : int(StrideType::InnerStrideAtCompileTime),
+    OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
+                             ? (InnerStrideAtCompileTime==Dynamic || PlainObjectTypeInnerSize==Dynamic
+                                ? Dynamic
+                                : int(InnerStrideAtCompileTime) * int(PlainObjectTypeInnerSize))
+                             : int(StrideType::OuterStrideAtCompileTime),
+    Alignment = int(MapOptions)&int(AlignedMask),
+    Flags0 = TraitsBase::Flags & (~NestByRefBit),
+    Flags = is_lvalue<PlainObjectType>::value ? int(Flags0) : (int(Flags0) & ~LvalueBit)
+  };
+private:
+  enum { Options }; // Expressions don't have Options
+};
+}
+
 /** \class Map
   * \ingroup Core_Module
   *
   * \brief A matrix or vector expression mapping an existing array of data.
   *
   * \tparam PlainObjectType the equivalent matrix type of the mapped data
-  * \tparam MapOptions specifies whether the pointer is \c #Aligned, or \c #Unaligned.
+  * \tparam MapOptions specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
   *                The default is \c #Unaligned.
   * \tparam StrideType optionally specifies strides. By default, Map assumes the memory layout
   *                   of an ordinary, contiguous array. This can be overridden by specifying strides.
@@ -63,44 +91,6 @@ namespace Eigen {
   *
   * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
   */
-
-namespace internal {
-template<typename PlainObjectType, int MapOptions, typename StrideType>
-struct traits<Map<PlainObjectType, MapOptions, StrideType> >
-  : public traits<PlainObjectType>
-{
-  typedef traits<PlainObjectType> TraitsBase;
-  typedef typename PlainObjectType::Index Index;
-  typedef typename PlainObjectType::Scalar Scalar;
-  enum {
-    InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
-                             ? int(PlainObjectType::InnerStrideAtCompileTime)
-                             : int(StrideType::InnerStrideAtCompileTime),
-    OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
-                             ? int(PlainObjectType::OuterStrideAtCompileTime)
-                             : int(StrideType::OuterStrideAtCompileTime),
-    HasNoInnerStride = InnerStrideAtCompileTime == 1,
-    HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0,
-    HasNoStride = HasNoInnerStride && HasNoOuterStride,
-    IsAligned = bool(EIGEN_ALIGN) && ((int(MapOptions)&Aligned)==Aligned),
-    IsDynamicSize = PlainObjectType::SizeAtCompileTime==Dynamic,
-    KeepsPacketAccess = bool(HasNoInnerStride)
-                        && ( bool(IsDynamicSize)
-                           || HasNoOuterStride
-                           || ( OuterStrideAtCompileTime!=Dynamic
-                           && ((static_cast<int>(sizeof(Scalar))*OuterStrideAtCompileTime)%16)==0 ) ),
-    Flags0 = TraitsBase::Flags & (~NestByRefBit),
-    Flags1 = IsAligned ? (int(Flags0) | AlignedBit) : (int(Flags0) & ~AlignedBit),
-    Flags2 = (bool(HasNoStride) || bool(PlainObjectType::IsVectorAtCompileTime))
-           ? int(Flags1) : int(Flags1 & ~LinearAccessBit),
-    Flags3 = is_lvalue<PlainObjectType>::value ? int(Flags2) : (int(Flags2) & ~LvalueBit),
-    Flags = KeepsPacketAccess ? int(Flags3) : (int(Flags3) & ~PacketAccessBit)
-  };
-private:
-  enum { Options }; // Expressions don't have Options
-};
-}
-
 template<typename PlainObjectType, int MapOptions, typename StrideType> class Map
   : public MapBase<Map<PlainObjectType, MapOptions, StrideType> >
 {
@@ -110,34 +100,34 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     EIGEN_DENSE_PUBLIC_INTERFACE(Map)
 
     typedef typename Base::PointerType PointerType;
-#if EIGEN2_SUPPORT_STAGE <= STAGE30_FULL_EIGEN3_API
-    typedef const Scalar* PointerArgType;
-    inline PointerType cast_to_pointer_type(PointerArgType ptr) { return const_cast<PointerType>(ptr); }
-#else
     typedef PointerType PointerArgType;
+    EIGEN_DEVICE_FUNC
     inline PointerType cast_to_pointer_type(PointerArgType ptr) { return ptr; }
-#endif
 
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
     }
 
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
-      return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
-           : IsVectorAtCompileTime ? this->size()
-           : int(Flags)&RowMajorBit ? this->cols()
-           : this->rows();
+      return int(StrideType::OuterStrideAtCompileTime) != 0 ? m_stride.outer()
+           : int(internal::traits<Map>::OuterStrideAtCompileTime) != Dynamic ? Index(internal::traits<Map>::OuterStrideAtCompileTime)
+           : IsVectorAtCompileTime ? (this->size() * innerStride())
+           : (int(Flags)&RowMajorBit) ? (this->cols() * innerStride())
+           : (this->rows() * innerStride());
     }
 
     /** Constructor in the fixed-size case.
       *
       * \param dataPtr pointer to the array to map
-      * \param a_stride optional Stride object, passing the strides.
+      * \param stride optional Stride object, passing the strides.
       */
-    inline Map(PointerArgType dataPtr, const StrideType& a_stride = StrideType())
-      : Base(cast_to_pointer_type(dataPtr)), m_stride(a_stride)
+    EIGEN_DEVICE_FUNC
+    explicit inline Map(PointerArgType dataPtr, const StrideType& stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr)), m_stride(stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
@@ -145,11 +135,12 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     /** Constructor in the dynamic-size vector case.
       *
       * \param dataPtr pointer to the array to map
-      * \param a_size the size of the vector expression
-      * \param a_stride optional Stride object, passing the strides.
+      * \param size the size of the vector expression
+      * \param stride optional Stride object, passing the strides.
       */
-    inline Map(PointerArgType dataPtr, Index a_size, const StrideType& a_stride = StrideType())
-      : Base(cast_to_pointer_type(dataPtr), a_size), m_stride(a_stride)
+    EIGEN_DEVICE_FUNC
+    inline Map(PointerArgType dataPtr, Index size, const StrideType& stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr), size), m_stride(stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
@@ -157,12 +148,13 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     /** Constructor in the dynamic-size matrix case.
       *
       * \param dataPtr pointer to the array to map
-      * \param nbRows the number of rows of the matrix expression
-      * \param nbCols the number of columns of the matrix expression
-      * \param a_stride optional Stride object, passing the strides.
+      * \param rows the number of rows of the matrix expression
+      * \param cols the number of columns of the matrix expression
+      * \param stride optional Stride object, passing the strides.
       */
-    inline Map(PointerArgType dataPtr, Index nbRows, Index nbCols, const StrideType& a_stride = StrideType())
-      : Base(cast_to_pointer_type(dataPtr), nbRows, nbCols), m_stride(a_stride)
+    EIGEN_DEVICE_FUNC
+    inline Map(PointerArgType dataPtr, Index rows, Index cols, const StrideType& stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr), rows, cols), m_stride(stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
@@ -173,19 +165,6 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     StrideType m_stride;
 };
 
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-inline Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>
-  ::Array(const Scalar *data)
-{
-  this->_set_noalias(Eigen::Map<const Array>(data));
-}
-
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-inline Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>
-  ::Matrix(const Scalar *data)
-{
-  this->_set_noalias(Eigen::Map<const Matrix>(data));
-}
 
 } // end namespace Eigen
 
diff --git a/extern/Eigen3/Eigen/src/Core/MapBase.h b/extern/Eigen3/Eigen/src/Core/MapBase.h
index a9828f7f4b2..668922ffccb 100644
--- a/extern/Eigen3/Eigen/src/Core/MapBase.h
+++ b/extern/Eigen3/Eigen/src/Core/MapBase.h
@@ -12,15 +12,25 @@
 #define EIGEN_MAPBASE_H
 
 #define EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) \
-      EIGEN_STATIC_ASSERT((int(internal::traits<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
+      EIGEN_STATIC_ASSERT((int(internal::evaluator<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
                           YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT)
 
 namespace Eigen { 
 
-/** \class MapBase
-  * \ingroup Core_Module
+/** \ingroup Core_Module
   *
-  * \brief Base class for Map and Block expression with direct access
+  * \brief Base class for dense Map and Block expression with direct access
+  *
+  * This base class provides the const low-level accessors (e.g. coeff, coeffRef) of dense
+  * Map and Block objects with direct access.
+  * Typical users do not have to directly deal with this class.
+  *
+  * This class can be extended by through the macro plugin \c EIGEN_MAPBASE_PLUGIN.
+  * See \link TopicCustomizing_Plugins customizing Eigen \endlink for details.
+  *
+  * The \c Derived class has to provide the following two methods describing the memory layout:
+  *  \code Index innerStride() const; \endcode
+  *  \code Index outerStride() const; \endcode
   *
   * \sa class Map, class Block
   */
@@ -33,11 +43,11 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
     enum {
       RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
       ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
+      InnerStrideAtCompileTime = internal::traits<Derived>::InnerStrideAtCompileTime,
       SizeAtCompileTime = Base::SizeAtCompileTime
     };
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -76,8 +86,10 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
 
     typedef typename Base::CoeffReturnType CoeffReturnType;
 
-    inline Index rows() const { return m_rows.value(); }
-    inline Index cols() const { return m_cols.value(); }
+    /** \copydoc DenseBase::rows() */
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_rows.value(); }
+    /** \copydoc DenseBase::cols() */
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_cols.value(); }
 
     /** Returns a pointer to the first coefficient of the matrix or vector.
       *
@@ -85,30 +97,39 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
       *
       * \sa innerStride(), outerStride()
       */
-    inline const Scalar* data() const { return m_data; }
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return m_data; }
 
+    /** \copydoc PlainObjectBase::coeff(Index,Index) const */
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeff(Index rowId, Index colId) const
     {
       return m_data[colId * colStride() + rowId * rowStride()];
     }
 
+    /** \copydoc PlainObjectBase::coeff(Index) const */
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeff(Index index) const
     {
       EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
       return m_data[index * innerStride()];
     }
 
+    /** \copydoc PlainObjectBase::coeffRef(Index,Index) const */
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
       return this->m_data[colId * colStride() + rowId * rowStride()];
     }
 
+    /** \copydoc PlainObjectBase::coeffRef(Index) const */
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
       EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
       return this->m_data[index * innerStride()];
     }
 
+    /** \internal */
     template<int LoadMode>
     inline PacketScalar packet(Index rowId, Index colId) const
     {
@@ -116,6 +137,7 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
                (m_data + (colId * colStride() + rowId * rowStride()));
     }
 
+    /** \internal */
     template<int LoadMode>
     inline PacketScalar packet(Index index) const
     {
@@ -123,12 +145,16 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
       return internal::ploadt<PacketScalar, LoadMode>(m_data + index * innerStride());
     }
 
+    /** \internal Constructor for fixed size matrices or vectors */
+    EIGEN_DEVICE_FUNC
     explicit inline MapBase(PointerType dataPtr) : m_data(dataPtr), m_rows(RowsAtCompileTime), m_cols(ColsAtCompileTime)
     {
       EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-      checkSanity();
+      checkSanity<Derived>();
     }
 
+    /** \internal Constructor for dynamically sized vectors */
+    EIGEN_DEVICE_FUNC
     inline MapBase(PointerType dataPtr, Index vecSize)
             : m_data(dataPtr),
               m_rows(RowsAtCompileTime == Dynamic ? vecSize : Index(RowsAtCompileTime)),
@@ -137,34 +163,59 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
       eigen_assert(vecSize >= 0);
       eigen_assert(dataPtr == 0 || SizeAtCompileTime == Dynamic || SizeAtCompileTime == vecSize);
-      checkSanity();
+      checkSanity<Derived>();
     }
 
-    inline MapBase(PointerType dataPtr, Index nbRows, Index nbCols)
-            : m_data(dataPtr), m_rows(nbRows), m_cols(nbCols)
+    /** \internal Constructor for dynamically sized matrices */
+    EIGEN_DEVICE_FUNC
+    inline MapBase(PointerType dataPtr, Index rows, Index cols)
+            : m_data(dataPtr), m_rows(rows), m_cols(cols)
     {
       eigen_assert( (dataPtr == 0)
-              || (   nbRows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == nbRows)
-                  && nbCols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == nbCols)));
-      checkSanity();
+              || (   rows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
+                  && cols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols)));
+      checkSanity<Derived>();
     }
 
+    #ifdef EIGEN_MAPBASE_PLUGIN
+    #include EIGEN_MAPBASE_PLUGIN
+    #endif
+
   protected:
 
-    void checkSanity() const
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    void checkSanity(typename internal::enable_if<(internal::traits<T>::Alignment>0),void*>::type = 0) const
     {
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(internal::traits<Derived>::Flags&PacketAccessBit,
-                                        internal::inner_stride_at_compile_time<Derived>::ret==1),
-                          PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
-      eigen_assert(EIGEN_IMPLIES(internal::traits<Derived>::Flags&AlignedBit, (size_t(m_data) % 16) == 0)
-                   && "input pointer is not aligned on a 16 byte boundary");
+#if EIGEN_MAX_ALIGN_BYTES>0
+      // innerStride() is not set yet when this function is called, so we optimistically assume the lowest plausible value:
+      const Index minInnerStride = InnerStrideAtCompileTime == Dynamic ? 1 : Index(InnerStrideAtCompileTime);
+      EIGEN_ONLY_USED_FOR_DEBUG(minInnerStride);
+      eigen_assert((   ((internal::UIntPtr(m_data) % internal::traits<Derived>::Alignment) == 0)
+                    || (cols() * rows() * minInnerStride * sizeof(Scalar)) < internal::traits<Derived>::Alignment ) && "data is not aligned");
+#endif
     }
 
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    void checkSanity(typename internal::enable_if<internal::traits<T>::Alignment==0,void*>::type = 0) const
+    {}
+
     PointerType m_data;
     const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_rows;
     const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_cols;
 };
 
+/** \ingroup Core_Module
+  *
+  * \brief Base class for non-const dense Map and Block expression with direct access
+  *
+  * This base class provides the non-const low-level accessors (e.g. coeff and coeffRef) of
+  * dense Map and Block objects with direct access.
+  * It inherits MapBase<Derived, ReadOnlyAccessors> which defines the const variant for reading specific entries.
+  *
+  * \sa class Map, class Block
+  */
 template<typename Derived> class MapBase<Derived, WriteAccessors>
   : public MapBase<Derived, ReadOnlyAccessors>
 {
@@ -175,7 +226,7 @@ template<typename Derived> class MapBase<Derived, WriteAccessors>
 
     typedef typename Base::Scalar Scalar;
     typedef typename Base::PacketScalar PacketScalar;
-    typedef typename Base::Index Index;
+    typedef typename Base::StorageIndex StorageIndex;
     typedef typename Base::PointerType PointerType;
 
     using Base::derived;
@@ -196,14 +247,18 @@ template<typename Derived> class MapBase<Derived, WriteAccessors>
                     const Scalar
                   >::type ScalarWithConstIfNotLvalue;
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar* data() const { return this->m_data; }
+    EIGEN_DEVICE_FUNC
     inline ScalarWithConstIfNotLvalue* data() { return this->m_data; } // no const-cast here so non-const-correct code will give a compile error
 
+    EIGEN_DEVICE_FUNC
     inline ScalarWithConstIfNotLvalue& coeffRef(Index row, Index col)
     {
       return this->m_data[col * colStride() + row * rowStride()];
     }
 
+    EIGEN_DEVICE_FUNC
     inline ScalarWithConstIfNotLvalue& coeffRef(Index index)
     {
       EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
@@ -225,10 +280,11 @@ template<typename Derived> class MapBase<Derived, WriteAccessors>
                 (this->m_data + index * innerStride(), val);
     }
 
-    explicit inline MapBase(PointerType dataPtr) : Base(dataPtr) {}
-    inline MapBase(PointerType dataPtr, Index vecSize) : Base(dataPtr, vecSize) {}
-    inline MapBase(PointerType dataPtr, Index nbRows, Index nbCols) : Base(dataPtr, nbRows, nbCols) {}
+    EIGEN_DEVICE_FUNC explicit inline MapBase(PointerType dataPtr) : Base(dataPtr) {}
+    EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index vecSize) : Base(dataPtr, vecSize) {}
+    EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index rows, Index cols) : Base(dataPtr, rows, cols) {}
 
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const MapBase& other)
     {
       ReadOnlyMapBase::Base::operator=(other);
diff --git a/extern/Eigen3/Eigen/src/Core/MathFunctions.h b/extern/Eigen3/Eigen/src/Core/MathFunctions.h
index adf2f9c511b..b249ce0c8b0 100644
--- a/extern/Eigen3/Eigen/src/Core/MathFunctions.h
+++ b/extern/Eigen3/Eigen/src/Core/MathFunctions.h
@@ -10,11 +10,25 @@
 #ifndef EIGEN_MATHFUNCTIONS_H
 #define EIGEN_MATHFUNCTIONS_H
 
+// source: http://www.geom.uiuc.edu/~huberty/math5337/groupe/digits.html
+// TODO this should better be moved to NumTraits
+#define EIGEN_PI 3.141592653589793238462643383279502884197169399375105820974944592307816406L
+
+
 namespace Eigen {
 
+// On WINCE, std::abs is defined for int only, so let's defined our own overloads:
+// This issue has been confirmed with MSVC 2008 only, but the issue might exist for more recent versions too.
+#if EIGEN_OS_WINCE && EIGEN_COMP_MSVC && EIGEN_COMP_MSVC<=1500
+long        abs(long        x) { return (labs(x));  }
+double      abs(double      x) { return (fabs(x));  }
+float       abs(float       x) { return (fabsf(x)); }
+long double abs(long double x) { return (fabsl(x)); }
+#endif
+
 namespace internal {
 
-/** \internal \struct global_math_functions_filtering_base
+/** \internal \class global_math_functions_filtering_base
   *
   * What it does:
   * Defines a typedef 'type' as follows:
@@ -62,6 +76,7 @@ template<typename Scalar, bool IsComplex = NumTraits<Scalar>::IsComplex>
 struct real_default_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
     return x;
@@ -72,6 +87,7 @@ template<typename Scalar>
 struct real_default_impl<Scalar,true>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
     using std::real;
@@ -81,13 +97,25 @@ struct real_default_impl<Scalar,true>
 
 template<typename Scalar> struct real_impl : real_default_impl<Scalar> {};
 
+#ifdef __CUDA_ARCH__
+template<typename T>
+struct real_impl<std::complex<T> >
+{
+  typedef T RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline T run(const std::complex<T>& x)
+  {
+    return x.real();
+  }
+};
+#endif
+
 template<typename Scalar>
 struct real_retval
 {
   typedef typename NumTraits<Scalar>::Real type;
 };
 
-
 /****************************************************************************
 * Implementation of imag                                                 *
 ****************************************************************************/
@@ -96,6 +124,7 @@ template<typename Scalar, bool IsComplex = NumTraits<Scalar>::IsComplex>
 struct imag_default_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar&)
   {
     return RealScalar(0);
@@ -106,6 +135,7 @@ template<typename Scalar>
 struct imag_default_impl<Scalar,true>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
     using std::imag;
@@ -115,6 +145,19 @@ struct imag_default_impl<Scalar,true>
 
 template<typename Scalar> struct imag_impl : imag_default_impl<Scalar> {};
 
+#ifdef __CUDA_ARCH__
+template<typename T>
+struct imag_impl<std::complex<T> >
+{
+  typedef T RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline T run(const std::complex<T>& x)
+  {
+    return x.imag();
+  }
+};
+#endif
+
 template<typename Scalar>
 struct imag_retval
 {
@@ -129,10 +172,12 @@ template<typename Scalar>
 struct real_ref_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar& run(Scalar& x)
   {
     return reinterpret_cast<RealScalar*>(&x)[0];
   }
+  EIGEN_DEVICE_FUNC
   static inline const RealScalar& run(const Scalar& x)
   {
     return reinterpret_cast<const RealScalar*>(&x)[0];
@@ -153,10 +198,12 @@ template<typename Scalar, bool IsComplex>
 struct imag_ref_default_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar& run(Scalar& x)
   {
     return reinterpret_cast<RealScalar*>(&x)[1];
   }
+  EIGEN_DEVICE_FUNC
   static inline const RealScalar& run(const Scalar& x)
   {
     return reinterpret_cast<RealScalar*>(&x)[1];
@@ -166,10 +213,12 @@ struct imag_ref_default_impl
 template<typename Scalar>
 struct imag_ref_default_impl<Scalar, false>
 {
+  EIGEN_DEVICE_FUNC
   static inline Scalar run(Scalar&)
   {
     return Scalar(0);
   }
+  EIGEN_DEVICE_FUNC
   static inline const Scalar run(const Scalar&)
   {
     return Scalar(0);
@@ -192,6 +241,7 @@ struct imag_ref_retval
 template<typename Scalar, bool IsComplex = NumTraits<Scalar>::IsComplex>
 struct conj_impl
 {
+  EIGEN_DEVICE_FUNC
   static inline Scalar run(const Scalar& x)
   {
     return x;
@@ -201,6 +251,7 @@ struct conj_impl
 template<typename Scalar>
 struct conj_impl<Scalar,true>
 {
+  EIGEN_DEVICE_FUNC
   static inline Scalar run(const Scalar& x)
   {
     using std::conj;
@@ -218,26 +269,40 @@ struct conj_retval
 * Implementation of abs2                                                 *
 ****************************************************************************/
 
-template<typename Scalar>
-struct abs2_impl
+template<typename Scalar,bool IsComplex>
+struct abs2_impl_default
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
     return x*x;
   }
 };
 
-template<typename RealScalar>
-struct abs2_impl<std::complex<RealScalar> >
+template<typename Scalar>
+struct abs2_impl_default<Scalar, true> // IsComplex
 {
-  static inline RealScalar run(const std::complex<RealScalar>& x)
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const Scalar& x)
   {
     return real(x)*real(x) + imag(x)*imag(x);
   }
 };
 
 template<typename Scalar>
+struct abs2_impl
+{
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const Scalar& x)
+  {
+    return abs2_impl_default<Scalar,NumTraits<Scalar>::IsComplex>::run(x);
+  }
+};
+
+template<typename Scalar>
 struct abs2_retval
 {
   typedef typename NumTraits<Scalar>::Real type;
@@ -251,9 +316,10 @@ template<typename Scalar, bool IsComplex>
 struct norm1_default_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
-    using std::abs;
+    EIGEN_USING_STD_MATH(abs);
     return abs(real(x)) + abs(imag(x));
   }
 };
@@ -261,9 +327,10 @@ struct norm1_default_impl
 template<typename Scalar>
 struct norm1_default_impl<Scalar, false>
 {
+  EIGEN_DEVICE_FUNC
   static inline Scalar run(const Scalar& x)
   {
-    using std::abs;
+    EIGEN_USING_STD_MATH(abs);
     return abs(x);
   }
 };
@@ -281,25 +348,7 @@ struct norm1_retval
 * Implementation of hypot                                                *
 ****************************************************************************/
 
-template<typename Scalar>
-struct hypot_impl
-{
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  static inline RealScalar run(const Scalar& x, const Scalar& y)
-  {
-    using std::max;
-    using std::min;
-    using std::abs;
-    using std::sqrt;
-    RealScalar _x = abs(x);
-    RealScalar _y = abs(y);
-    RealScalar p = (max)(_x, _y);
-    if(p==RealScalar(0)) return RealScalar(0);
-    RealScalar q = (min)(_x, _y);
-    RealScalar qp = q/p;
-    return p * sqrt(RealScalar(1) + qp*qp);
-  }
-};
+template<typename Scalar> struct hypot_impl;
 
 template<typename Scalar>
 struct hypot_retval
@@ -314,6 +363,7 @@ struct hypot_retval
 template<typename OldType, typename NewType>
 struct cast_impl
 {
+  EIGEN_DEVICE_FUNC
   static inline NewType run(const OldType& x)
   {
     return static_cast<NewType>(x);
@@ -323,48 +373,124 @@ struct cast_impl
 // here, for once, we're plainly returning NewType: we don't want cast to do weird things.
 
 template<typename OldType, typename NewType>
+EIGEN_DEVICE_FUNC
 inline NewType cast(const OldType& x)
 {
   return cast_impl<OldType, NewType>::run(x);
 }
 
 /****************************************************************************
-* Implementation of atanh2                                                *
+* Implementation of round                                                   *
 ****************************************************************************/
 
-template<typename Scalar, bool IsInteger>
-struct atanh2_default_impl
-{
-  typedef Scalar retval;
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  static inline Scalar run(const Scalar& x, const Scalar& y)
+#if EIGEN_HAS_CXX11_MATH
+  template<typename Scalar>
+  struct round_impl {
+    static inline Scalar run(const Scalar& x)
+    {
+      EIGEN_STATIC_ASSERT((!NumTraits<Scalar>::IsComplex), NUMERIC_TYPE_MUST_BE_REAL)
+      using std::round;
+      return round(x);
+    }
+  };
+#else
+  template<typename Scalar>
+  struct round_impl
   {
-    using std::abs;
-    using std::log;
-    using std::sqrt;
-    Scalar z = x / y;
-    if (y == Scalar(0) || abs(z) > sqrt(NumTraits<RealScalar>::epsilon()))
-      return RealScalar(0.5) * log((y + x) / (y - x));
-    else
-      return z + z*z*z / RealScalar(3);
-  }
+    static inline Scalar run(const Scalar& x)
+    {
+      EIGEN_STATIC_ASSERT((!NumTraits<Scalar>::IsComplex), NUMERIC_TYPE_MUST_BE_REAL)
+      EIGEN_USING_STD_MATH(floor);
+      EIGEN_USING_STD_MATH(ceil);
+      return (x > Scalar(0)) ? floor(x + Scalar(0.5)) : ceil(x - Scalar(0.5));
+    }
+  };
+#endif
+
+template<typename Scalar>
+struct round_retval
+{
+  typedef Scalar type;
 };
 
+/****************************************************************************
+* Implementation of arg                                                     *
+****************************************************************************/
+
+#if EIGEN_HAS_CXX11_MATH
+  template<typename Scalar>
+  struct arg_impl {
+    static inline Scalar run(const Scalar& x)
+    {
+      EIGEN_USING_STD_MATH(arg);
+      return arg(x);
+    }
+  };
+#else
+  template<typename Scalar, bool IsComplex = NumTraits<Scalar>::IsComplex>
+  struct arg_default_impl
+  {
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    EIGEN_DEVICE_FUNC
+    static inline RealScalar run(const Scalar& x)
+    {
+      return (x < Scalar(0)) ? Scalar(EIGEN_PI) : Scalar(0); }
+  };
+
+  template<typename Scalar>
+  struct arg_default_impl<Scalar,true>
+  {
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    EIGEN_DEVICE_FUNC
+    static inline RealScalar run(const Scalar& x)
+    {
+      EIGEN_USING_STD_MATH(arg);
+      return arg(x);
+    }
+  };
+
+  template<typename Scalar> struct arg_impl : arg_default_impl<Scalar> {};
+#endif
+
 template<typename Scalar>
-struct atanh2_default_impl<Scalar, true>
+struct arg_retval
 {
-  static inline Scalar run(const Scalar&, const Scalar&)
+  typedef typename NumTraits<Scalar>::Real type;
+};
+
+/****************************************************************************
+* Implementation of log1p                                                   *
+****************************************************************************/
+
+namespace std_fallback {
+  // fallback log1p implementation in case there is no log1p(Scalar) function in namespace of Scalar,
+  // or that there is no suitable std::log1p function available
+  template<typename Scalar>
+  EIGEN_DEVICE_FUNC inline Scalar log1p(const Scalar& x) {
+    EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar)
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    EIGEN_USING_STD_MATH(log);
+    Scalar x1p = RealScalar(1) + x;
+    return numext::equal_strict(x1p, Scalar(1)) ? x : x * ( log(x1p) / (x1p - RealScalar(1)) );
+  }
+}
+
+template<typename Scalar>
+struct log1p_impl {
+  static inline Scalar run(const Scalar& x)
   {
     EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar)
-    return Scalar(0);
+    #if EIGEN_HAS_CXX11_MATH
+    using std::log1p;
+    #endif
+    using std_fallback::log1p;
+    return log1p(x);
   }
 };
 
-template<typename Scalar>
-struct atanh2_impl : atanh2_default_impl<Scalar, NumTraits<Scalar>::IsInteger> {};
 
 template<typename Scalar>
-struct atanh2_retval
+struct log1p_retval
 {
   typedef Scalar type;
 };
@@ -373,24 +499,26 @@ struct atanh2_retval
 * Implementation of pow                                                  *
 ****************************************************************************/
 
-template<typename Scalar, bool IsInteger>
-struct pow_default_impl
+template<typename ScalarX,typename ScalarY, bool IsInteger = NumTraits<ScalarX>::IsInteger&&NumTraits<ScalarY>::IsInteger>
+struct pow_impl
 {
-  typedef Scalar retval;
-  static inline Scalar run(const Scalar& x, const Scalar& y)
+  //typedef Scalar retval;
+  typedef typename ScalarBinaryOpTraits<ScalarX,ScalarY,internal::scalar_pow_op<ScalarX,ScalarY> >::ReturnType result_type;
+  static EIGEN_DEVICE_FUNC inline result_type run(const ScalarX& x, const ScalarY& y)
   {
-    using std::pow;
+    EIGEN_USING_STD_MATH(pow);
     return pow(x, y);
   }
 };
 
-template<typename Scalar>
-struct pow_default_impl<Scalar, true>
+template<typename ScalarX,typename ScalarY>
+struct pow_impl<ScalarX,ScalarY, true>
 {
-  static inline Scalar run(Scalar x, Scalar y)
+  typedef ScalarX result_type;
+  static EIGEN_DEVICE_FUNC inline ScalarX run(ScalarX x, ScalarY y)
   {
-    Scalar res(1);
-    eigen_assert(!NumTraits<Scalar>::IsSigned || y >= 0);
+    ScalarX res(1);
+    eigen_assert(!NumTraits<ScalarY>::IsSigned || y >= 0);
     if(y & 1) res *= x;
     y >>= 1;
     while(y)
@@ -403,15 +531,6 @@ struct pow_default_impl<Scalar, true>
   }
 };
 
-template<typename Scalar>
-struct pow_impl : pow_default_impl<Scalar, NumTraits<Scalar>::IsInteger> {};
-
-template<typename Scalar>
-struct pow_retval
-{
-  typedef Scalar type;
-};
-
 /****************************************************************************
 * Implementation of random                                               *
 ****************************************************************************/
@@ -447,48 +566,48 @@ struct random_default_impl<Scalar, false, false>
 };
 
 enum {
-  floor_log2_terminate,
-  floor_log2_move_up,
-  floor_log2_move_down,
-  floor_log2_bogus
+  meta_floor_log2_terminate,
+  meta_floor_log2_move_up,
+  meta_floor_log2_move_down,
+  meta_floor_log2_bogus
 };
 
-template<unsigned int n, int lower, int upper> struct floor_log2_selector
+template<unsigned int n, int lower, int upper> struct meta_floor_log2_selector
 {
   enum { middle = (lower + upper) / 2,
-         value = (upper <= lower + 1) ? int(floor_log2_terminate)
-               : (n < (1 << middle)) ? int(floor_log2_move_down)
-               : (n==0) ? int(floor_log2_bogus)
-               : int(floor_log2_move_up)
+         value = (upper <= lower + 1) ? int(meta_floor_log2_terminate)
+               : (n < (1 << middle)) ? int(meta_floor_log2_move_down)
+               : (n==0) ? int(meta_floor_log2_bogus)
+               : int(meta_floor_log2_move_up)
   };
 };
 
 template<unsigned int n,
          int lower = 0,
          int upper = sizeof(unsigned int) * CHAR_BIT - 1,
-         int selector = floor_log2_selector<n, lower, upper>::value>
-struct floor_log2 {};
+         int selector = meta_floor_log2_selector<n, lower, upper>::value>
+struct meta_floor_log2 {};
 
 template<unsigned int n, int lower, int upper>
-struct floor_log2<n, lower, upper, floor_log2_move_down>
+struct meta_floor_log2<n, lower, upper, meta_floor_log2_move_down>
 {
-  enum { value = floor_log2<n, lower, floor_log2_selector<n, lower, upper>::middle>::value };
+  enum { value = meta_floor_log2<n, lower, meta_floor_log2_selector<n, lower, upper>::middle>::value };
 };
 
 template<unsigned int n, int lower, int upper>
-struct floor_log2<n, lower, upper, floor_log2_move_up>
+struct meta_floor_log2<n, lower, upper, meta_floor_log2_move_up>
 {
-  enum { value = floor_log2<n, floor_log2_selector<n, lower, upper>::middle, upper>::value };
+  enum { value = meta_floor_log2<n, meta_floor_log2_selector<n, lower, upper>::middle, upper>::value };
 };
 
 template<unsigned int n, int lower, int upper>
-struct floor_log2<n, lower, upper, floor_log2_terminate>
+struct meta_floor_log2<n, lower, upper, meta_floor_log2_terminate>
 {
   enum { value = (n >= ((unsigned int)(1) << (lower+1))) ? lower+1 : lower };
 };
 
 template<unsigned int n, int lower, int upper>
-struct floor_log2<n, lower, upper, floor_log2_bogus>
+struct meta_floor_log2<n, lower, upper, meta_floor_log2_bogus>
 {
   // no value, error at compile time
 };
@@ -496,11 +615,31 @@ struct floor_log2<n, lower, upper, floor_log2_bogus>
 template<typename Scalar>
 struct random_default_impl<Scalar, false, true>
 {
-  typedef typename NumTraits<Scalar>::NonInteger NonInteger;
-
   static inline Scalar run(const Scalar& x, const Scalar& y)
   {
-    return x + Scalar((NonInteger(y)-x+1) * std::rand() / (RAND_MAX + NonInteger(1)));
+    if (y <= x)
+      return x;
+    // ScalarU is the unsigned counterpart of Scalar, possibly Scalar itself.
+    typedef typename make_unsigned<Scalar>::type ScalarU;
+    // ScalarX is the widest of ScalarU and unsigned int.
+    // We'll deal only with ScalarX and unsigned int below thus avoiding signed
+    // types and arithmetic and signed overflows (which are undefined behavior).
+    typedef typename conditional<(ScalarU(-1) > unsigned(-1)), ScalarU, unsigned>::type ScalarX;
+    // The following difference doesn't overflow, provided our integer types are two's
+    // complement and have the same number of padding bits in signed and unsigned variants.
+    // This is the case in most modern implementations of C++.
+    ScalarX range = ScalarX(y) - ScalarX(x);
+    ScalarX offset = 0;
+    ScalarX divisor = 1;
+    ScalarX multiplier = 1;
+    const unsigned rand_max = RAND_MAX;
+    if (range <= rand_max) divisor = (rand_max + 1) / (range + 1);
+    else                   multiplier = 1 + range / (rand_max + 1);
+    // Rejection sampling.
+    do {
+      offset = (unsigned(std::rand()) * multiplier) / divisor;
+    } while (offset > range);
+    return Scalar(ScalarX(x) + offset);
   }
 
   static inline Scalar run()
@@ -508,7 +647,7 @@ struct random_default_impl<Scalar, false, true>
 #ifdef EIGEN_MAKING_DOCS
     return run(Scalar(NumTraits<Scalar>::IsSigned ? -10 : 0), Scalar(10));
 #else
-    enum { rand_bits = floor_log2<(unsigned int)(RAND_MAX)+1>::value,
+    enum { rand_bits = meta_floor_log2<(unsigned int)(RAND_MAX)+1>::value,
            scalar_bits = sizeof(Scalar) * CHAR_BIT,
            shift = EIGEN_PLAIN_ENUM_MAX(0, int(rand_bits) - int(scalar_bits)),
            offset = NumTraits<Scalar>::IsSigned ? (1 << (EIGEN_PLAIN_ENUM_MIN(rand_bits,scalar_bits)-1)) : 0
@@ -545,97 +684,602 @@ inline EIGEN_MATHFUNC_RETVAL(random, Scalar) random()
   return EIGEN_MATHFUNC_IMPL(random, Scalar)::run();
 }
 
+// Implementatin of is* functions
+
+// std::is* do not work with fast-math and gcc, std::is* are available on MSVC 2013 and newer, as well as in clang.
+#if (EIGEN_HAS_CXX11_MATH && !(EIGEN_COMP_GNUC_STRICT && __FINITE_MATH_ONLY__)) || (EIGEN_COMP_MSVC>=1800) || (EIGEN_COMP_CLANG)
+#define EIGEN_USE_STD_FPCLASSIFY 1
+#else
+#define EIGEN_USE_STD_FPCLASSIFY 0
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<internal::is_integral<T>::value,bool>::type
+isnan_impl(const T&) { return false; }
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<internal::is_integral<T>::value,bool>::type
+isinf_impl(const T&) { return false; }
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<internal::is_integral<T>::value,bool>::type
+isfinite_impl(const T&) { return true; }
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<(!internal::is_integral<T>::value)&&(!NumTraits<T>::IsComplex),bool>::type
+isfinite_impl(const T& x)
+{
+  #ifdef __CUDA_ARCH__
+    return (::isfinite)(x);
+  #elif EIGEN_USE_STD_FPCLASSIFY
+    using std::isfinite;
+    return isfinite EIGEN_NOT_A_MACRO (x);
+  #else
+    return x<=NumTraits<T>::highest() && x>=NumTraits<T>::lowest();
+  #endif
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<(!internal::is_integral<T>::value)&&(!NumTraits<T>::IsComplex),bool>::type
+isinf_impl(const T& x)
+{
+  #ifdef __CUDA_ARCH__
+    return (::isinf)(x);
+  #elif EIGEN_USE_STD_FPCLASSIFY
+    using std::isinf;
+    return isinf EIGEN_NOT_A_MACRO (x);
+  #else
+    return x>NumTraits<T>::highest() || x<NumTraits<T>::lowest();
+  #endif
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<(!internal::is_integral<T>::value)&&(!NumTraits<T>::IsComplex),bool>::type
+isnan_impl(const T& x)
+{
+  #ifdef __CUDA_ARCH__
+    return (::isnan)(x);
+  #elif EIGEN_USE_STD_FPCLASSIFY
+    using std::isnan;
+    return isnan EIGEN_NOT_A_MACRO (x);
+  #else
+    return x != x;
+  #endif
+}
+
+#if (!EIGEN_USE_STD_FPCLASSIFY)
+
+#if EIGEN_COMP_MSVC
+
+template<typename T> EIGEN_DEVICE_FUNC bool isinf_msvc_helper(T x)
+{
+  return _fpclass(x)==_FPCLASS_NINF || _fpclass(x)==_FPCLASS_PINF;
+}
+
+//MSVC defines a _isnan builtin function, but for double only
+EIGEN_DEVICE_FUNC inline bool isnan_impl(const long double& x) { return _isnan(x)!=0; }
+EIGEN_DEVICE_FUNC inline bool isnan_impl(const double& x)      { return _isnan(x)!=0; }
+EIGEN_DEVICE_FUNC inline bool isnan_impl(const float& x)       { return _isnan(x)!=0; }
+
+EIGEN_DEVICE_FUNC inline bool isinf_impl(const long double& x) { return isinf_msvc_helper(x); }
+EIGEN_DEVICE_FUNC inline bool isinf_impl(const double& x)      { return isinf_msvc_helper(x); }
+EIGEN_DEVICE_FUNC inline bool isinf_impl(const float& x)       { return isinf_msvc_helper(x); }
+
+#elif (defined __FINITE_MATH_ONLY__ && __FINITE_MATH_ONLY__ && EIGEN_COMP_GNUC)
+
+#if EIGEN_GNUC_AT_LEAST(5,0)
+  #define EIGEN_TMP_NOOPT_ATTRIB EIGEN_DEVICE_FUNC inline __attribute__((optimize("no-finite-math-only")))
+#else
+  // NOTE the inline qualifier and noinline attribute are both needed: the former is to avoid linking issue (duplicate symbol),
+  //      while the second prevent too aggressive optimizations in fast-math mode:
+  #define EIGEN_TMP_NOOPT_ATTRIB EIGEN_DEVICE_FUNC inline __attribute__((noinline,optimize("no-finite-math-only")))
+#endif
+
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isnan_impl(const long double& x) { return __builtin_isnan(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isnan_impl(const double& x)      { return __builtin_isnan(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isnan_impl(const float& x)       { return __builtin_isnan(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const double& x)      { return __builtin_isinf(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const float& x)       { return __builtin_isinf(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const long double& x) { return __builtin_isinf(x); }
+
+#undef EIGEN_TMP_NOOPT_ATTRIB
+
+#endif
+
+#endif
+
+// The following overload are defined at the end of this file
+template<typename T> EIGEN_DEVICE_FUNC bool isfinite_impl(const std::complex<T>& x);
+template<typename T> EIGEN_DEVICE_FUNC bool isnan_impl(const std::complex<T>& x);
+template<typename T> EIGEN_DEVICE_FUNC bool isinf_impl(const std::complex<T>& x);
+
+template<typename T> T generic_fast_tanh_float(const T& a_x);
+
 } // end namespace internal
 
 /****************************************************************************
-* Generic math function                                                    *
+* Generic math functions                                                    *
 ****************************************************************************/
 
 namespace numext {
 
+#ifndef __CUDA_ARCH__
+template<typename T>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE T mini(const T& x, const T& y)
+{
+  EIGEN_USING_STD_MATH(min);
+  return min EIGEN_NOT_A_MACRO (x,y);
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE T maxi(const T& x, const T& y)
+{
+  EIGEN_USING_STD_MATH(max);
+  return max EIGEN_NOT_A_MACRO (x,y);
+}
+#else
+template<typename T>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE T mini(const T& x, const T& y)
+{
+  return y < x ? y : x;
+}
+template<>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE float mini(const float& x, const float& y)
+{
+  return fminf(x, y);
+}
+template<typename T>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE T maxi(const T& x, const T& y)
+{
+  return x < y ? y : x;
+}
+template<>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE float maxi(const float& x, const float& y)
+{
+  return fmaxf(x, y);
+}
+#endif
+
+
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(real, Scalar) real(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(real, Scalar)::run(x);
-}  
+}
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline typename internal::add_const_on_value_type< EIGEN_MATHFUNC_RETVAL(real_ref, Scalar) >::type real_ref(const Scalar& x)
 {
   return internal::real_ref_impl<Scalar>::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(real_ref, Scalar) real_ref(Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(real_ref, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(imag, Scalar) imag(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(imag, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
+inline EIGEN_MATHFUNC_RETVAL(arg, Scalar) arg(const Scalar& x)
+{
+  return EIGEN_MATHFUNC_IMPL(arg, Scalar)::run(x);
+}
+
+template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline typename internal::add_const_on_value_type< EIGEN_MATHFUNC_RETVAL(imag_ref, Scalar) >::type imag_ref(const Scalar& x)
 {
   return internal::imag_ref_impl<Scalar>::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(imag_ref, Scalar) imag_ref(Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(imag_ref, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(conj, Scalar) conj(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(conj, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(abs2, Scalar) abs2(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(abs2, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(norm1, Scalar) norm1(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(norm1, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(hypot, Scalar) hypot(const Scalar& x, const Scalar& y)
 {
   return EIGEN_MATHFUNC_IMPL(hypot, Scalar)::run(x, y);
 }
 
 template<typename Scalar>
-inline EIGEN_MATHFUNC_RETVAL(atanh2, Scalar) atanh2(const Scalar& x, const Scalar& y)
+EIGEN_DEVICE_FUNC
+inline EIGEN_MATHFUNC_RETVAL(log1p, Scalar) log1p(const Scalar& x)
 {
-  return EIGEN_MATHFUNC_IMPL(atanh2, Scalar)::run(x, y);
+  return EIGEN_MATHFUNC_IMPL(log1p, Scalar)::run(x);
 }
 
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float log1p(const float &x) { return ::log1pf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double log1p(const double &x) { return ::log1p(x); }
+#endif
+
+template<typename ScalarX,typename ScalarY>
+EIGEN_DEVICE_FUNC
+inline typename internal::pow_impl<ScalarX,ScalarY>::result_type pow(const ScalarX& x, const ScalarY& y)
+{
+  return internal::pow_impl<ScalarX,ScalarY>::run(x, y);
+}
+
+template<typename T> EIGEN_DEVICE_FUNC bool (isnan)   (const T &x) { return internal::isnan_impl(x); }
+template<typename T> EIGEN_DEVICE_FUNC bool (isinf)   (const T &x) { return internal::isinf_impl(x); }
+template<typename T> EIGEN_DEVICE_FUNC bool (isfinite)(const T &x) { return internal::isfinite_impl(x); }
+
 template<typename Scalar>
-inline EIGEN_MATHFUNC_RETVAL(pow, Scalar) pow(const Scalar& x, const Scalar& y)
+EIGEN_DEVICE_FUNC
+inline EIGEN_MATHFUNC_RETVAL(round, Scalar) round(const Scalar& x)
+{
+  return EIGEN_MATHFUNC_IMPL(round, Scalar)::run(x);
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+T (floor)(const T& x)
+{
+  EIGEN_USING_STD_MATH(floor);
+  return floor(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float floor(const float &x) { return ::floorf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double floor(const double &x) { return ::floor(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+T (ceil)(const T& x)
+{
+  EIGEN_USING_STD_MATH(ceil);
+  return ceil(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float ceil(const float &x) { return ::ceilf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double ceil(const double &x) { return ::ceil(x); }
+#endif
+
+
+/** Log base 2 for 32 bits positive integers.
+  * Conveniently returns 0 for x==0. */
+inline int log2(int x)
 {
-  return EIGEN_MATHFUNC_IMPL(pow, Scalar)::run(x, y);
+  eigen_assert(x>=0);
+  unsigned int v(x);
+  static const int table[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+  v |= v >> 1;
+  v |= v >> 2;
+  v |= v >> 4;
+  v |= v >> 8;
+  v |= v >> 16;
+  return table[(v * 0x07C4ACDDU) >> 27];
 }
 
-// std::isfinite is non standard, so let's define our own version,
-// even though it is not very efficient.
-template<typename T> bool (isfinite)(const T& x)
+/** \returns the square root of \a x.
+  *
+  * It is essentially equivalent to
+  * \code using std::sqrt; return sqrt(x); \endcode
+  * but slightly faster for float/double and some compilers (e.g., gcc), thanks to
+  * specializations when SSE is enabled.
+  *
+  * It's usage is justified in performance critical functions, like norm/normalize.
+  */
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T sqrt(const T &x)
 {
-  return x<NumTraits<T>::highest() && x>NumTraits<T>::lowest();
+  EIGEN_USING_STD_MATH(sqrt);
+  return sqrt(x);
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T log(const T &x) {
+  EIGEN_USING_STD_MATH(log);
+  return log(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float log(const float &x) { return ::logf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double log(const double &x) { return ::log(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+typename internal::enable_if<NumTraits<T>::IsSigned || NumTraits<T>::IsComplex,typename NumTraits<T>::Real>::type
+abs(const T &x) {
+  EIGEN_USING_STD_MATH(abs);
+  return abs(x);
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+typename internal::enable_if<!(NumTraits<T>::IsSigned || NumTraits<T>::IsComplex),typename NumTraits<T>::Real>::type
+abs(const T &x) {
+  return x;
+}
+
+#if defined(__SYCL_DEVICE_ONLY__)
+EIGEN_ALWAYS_INLINE float   abs(float x) { return cl::sycl::fabs(x); }
+EIGEN_ALWAYS_INLINE double  abs(double x) { return cl::sycl::fabs(x); }
+#endif // defined(__SYCL_DEVICE_ONLY__)
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float abs(const float &x) { return ::fabsf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double abs(const double &x) { return ::fabs(x); }
+
+template <> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float abs(const std::complex<float>& x) {
+  return ::hypotf(x.real(), x.imag());
+}
+
+template <> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double abs(const std::complex<double>& x) {
+  return ::hypot(x.real(), x.imag());
+}
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T exp(const T &x) {
+  EIGEN_USING_STD_MATH(exp);
+  return exp(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float exp(const float &x) { return ::expf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double exp(const double &x) { return ::exp(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T cos(const T &x) {
+  EIGEN_USING_STD_MATH(cos);
+  return cos(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float cos(const float &x) { return ::cosf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double cos(const double &x) { return ::cos(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T sin(const T &x) {
+  EIGEN_USING_STD_MATH(sin);
+  return sin(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float sin(const float &x) { return ::sinf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double sin(const double &x) { return ::sin(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T tan(const T &x) {
+  EIGEN_USING_STD_MATH(tan);
+  return tan(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float tan(const float &x) { return ::tanf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double tan(const double &x) { return ::tan(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T acos(const T &x) {
+  EIGEN_USING_STD_MATH(acos);
+  return acos(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float acos(const float &x) { return ::acosf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double acos(const double &x) { return ::acos(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T asin(const T &x) {
+  EIGEN_USING_STD_MATH(asin);
+  return asin(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float asin(const float &x) { return ::asinf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double asin(const double &x) { return ::asin(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T atan(const T &x) {
+  EIGEN_USING_STD_MATH(atan);
+  return atan(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float atan(const float &x) { return ::atanf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double atan(const double &x) { return ::atan(x); }
+#endif
+
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T cosh(const T &x) {
+  EIGEN_USING_STD_MATH(cosh);
+  return cosh(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float cosh(const float &x) { return ::coshf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double cosh(const double &x) { return ::cosh(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T sinh(const T &x) {
+  EIGEN_USING_STD_MATH(sinh);
+  return sinh(x);
 }
 
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float sinh(const float &x) { return ::sinhf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double sinh(const double &x) { return ::sinh(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T tanh(const T &x) {
+  EIGEN_USING_STD_MATH(tanh);
+  return tanh(x);
+}
+
+#if (!defined(__CUDACC__)) && EIGEN_FAST_MATH
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float tanh(float x) { return internal::generic_fast_tanh_float(x); }
+#endif
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float tanh(const float &x) { return ::tanhf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double tanh(const double &x) { return ::tanh(x); }
+#endif
+
+template <typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T fmod(const T& a, const T& b) {
+  EIGEN_USING_STD_MATH(fmod);
+  return fmod(a, b);
+}
+
+#ifdef __CUDACC__
+template <>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float fmod(const float& a, const float& b) {
+  return ::fmodf(a, b);
+}
+
+template <>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double fmod(const double& a, const double& b) {
+  return ::fmod(a, b);
+}
+#endif
+
 } // end namespace numext
 
 namespace internal {
 
+template<typename T>
+EIGEN_DEVICE_FUNC bool isfinite_impl(const std::complex<T>& x)
+{
+  return (numext::isfinite)(numext::real(x)) && (numext::isfinite)(numext::imag(x));
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC bool isnan_impl(const std::complex<T>& x)
+{
+  return (numext::isnan)(numext::real(x)) || (numext::isnan)(numext::imag(x));
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC bool isinf_impl(const std::complex<T>& x)
+{
+  return ((numext::isinf)(numext::real(x)) || (numext::isinf)(numext::imag(x))) && (!(numext::isnan)(x));
+}
+
 /****************************************************************************
 * Implementation of fuzzy comparisons                                       *
 ****************************************************************************/
@@ -649,18 +1293,17 @@ template<typename Scalar>
 struct scalar_fuzzy_default_impl<Scalar, false, false>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  template<typename OtherScalar>
+  template<typename OtherScalar> EIGEN_DEVICE_FUNC
   static inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y, const RealScalar& prec)
   {
-    using std::abs;
-    return abs(x) <= abs(y) * prec;
+    return numext::abs(x) <= numext::abs(y) * prec;
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar& prec)
   {
-    using std::min;
-    using std::abs;
-    return abs(x - y) <= (min)(abs(x), abs(y)) * prec;
+    return numext::abs(x - y) <= numext::mini(numext::abs(x), numext::abs(y)) * prec;
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y, const RealScalar& prec)
   {
     return x <= y || isApprox(x, y, prec);
@@ -671,15 +1314,17 @@ template<typename Scalar>
 struct scalar_fuzzy_default_impl<Scalar, false, true>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  template<typename OtherScalar>
+  template<typename OtherScalar> EIGEN_DEVICE_FUNC
   static inline bool isMuchSmallerThan(const Scalar& x, const Scalar&, const RealScalar&)
   {
     return x == Scalar(0);
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar&)
   {
     return x == y;
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y, const RealScalar&)
   {
     return x <= y;
@@ -690,38 +1335,38 @@ template<typename Scalar>
 struct scalar_fuzzy_default_impl<Scalar, true, false>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  template<typename OtherScalar>
+  template<typename OtherScalar> EIGEN_DEVICE_FUNC
   static inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y, const RealScalar& prec)
   {
     return numext::abs2(x) <= numext::abs2(y) * prec * prec;
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar& prec)
   {
-    using std::min;
-    return numext::abs2(x - y) <= (min)(numext::abs2(x), numext::abs2(y)) * prec * prec;
+    return numext::abs2(x - y) <= numext::mini(numext::abs2(x), numext::abs2(y)) * prec * prec;
   }
 };
 
 template<typename Scalar>
 struct scalar_fuzzy_impl : scalar_fuzzy_default_impl<Scalar, NumTraits<Scalar>::IsComplex, NumTraits<Scalar>::IsInteger> {};
 
-template<typename Scalar, typename OtherScalar>
+template<typename Scalar, typename OtherScalar> EIGEN_DEVICE_FUNC
 inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y,
-                                   typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision())
+                              const typename NumTraits<Scalar>::Real &precision = NumTraits<Scalar>::dummy_precision())
 {
   return scalar_fuzzy_impl<Scalar>::template isMuchSmallerThan<OtherScalar>(x, y, precision);
 }
 
-template<typename Scalar>
+template<typename Scalar> EIGEN_DEVICE_FUNC
 inline bool isApprox(const Scalar& x, const Scalar& y,
-                          typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision())
+                     const typename NumTraits<Scalar>::Real &precision = NumTraits<Scalar>::dummy_precision())
 {
   return scalar_fuzzy_impl<Scalar>::isApprox(x, y, precision);
 }
 
-template<typename Scalar>
+template<typename Scalar> EIGEN_DEVICE_FUNC
 inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y,
-                                    typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision())
+                               const typename NumTraits<Scalar>::Real &precision = NumTraits<Scalar>::dummy_precision())
 {
   return scalar_fuzzy_impl<Scalar>::isApproxOrLessThan(x, y, precision);
 }
@@ -742,17 +1387,19 @@ template<> struct scalar_fuzzy_impl<bool>
 {
   typedef bool RealScalar;
   
-  template<typename OtherScalar>
+  template<typename OtherScalar> EIGEN_DEVICE_FUNC
   static inline bool isMuchSmallerThan(const bool& x, const bool&, const bool&)
   {
     return !x;
   }
   
+  EIGEN_DEVICE_FUNC
   static inline bool isApprox(bool x, bool y, bool)
   {
     return x == y;
   }
 
+  EIGEN_DEVICE_FUNC
   static inline bool isApproxOrLessThan(const bool& x, const bool& y, const bool&)
   {
     return (!x) || y;
diff --git a/extern/Eigen3/Eigen/src/Core/MathFunctionsImpl.h b/extern/Eigen3/Eigen/src/Core/MathFunctionsImpl.h
new file mode 100644
index 00000000000..9c1ceb0eb0f
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/MathFunctionsImpl.h
@@ -0,0 +1,101 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Pedro Gonnet (pedro.gonnet@gmail.com)
+// Copyright (C) 2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_MATHFUNCTIONSIMPL_H
+#define EIGEN_MATHFUNCTIONSIMPL_H
+
+namespace Eigen {
+
+namespace internal {
+
+/** \internal \returns the hyperbolic tan of \a a (coeff-wise)
+    Doesn't do anything fancy, just a 13/6-degree rational interpolant which
+    is accurate up to a couple of ulp in the range [-9, 9], outside of which
+    the tanh(x) = +/-1.
+
+    This implementation works on both scalars and packets.
+*/
+template<typename T>
+T generic_fast_tanh_float(const T& a_x)
+{
+  // Clamp the inputs to the range [-9, 9] since anything outside
+  // this range is +/-1.0f in single-precision.
+  const T plus_9 = pset1<T>(9.f);
+  const T minus_9 = pset1<T>(-9.f);
+  // NOTE GCC prior to 6.3 might improperly optimize this max/min
+  //      step such that if a_x is nan, x will be either 9 or -9,
+  //      and tanh will return 1 or -1 instead of nan.
+  //      This is supposed to be fixed in gcc6.3,
+  //      see: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=72867
+  const T x = pmax(minus_9,pmin(plus_9,a_x));
+  // The monomial coefficients of the numerator polynomial (odd).
+  const T alpha_1 = pset1<T>(4.89352455891786e-03f);
+  const T alpha_3 = pset1<T>(6.37261928875436e-04f);
+  const T alpha_5 = pset1<T>(1.48572235717979e-05f);
+  const T alpha_7 = pset1<T>(5.12229709037114e-08f);
+  const T alpha_9 = pset1<T>(-8.60467152213735e-11f);
+  const T alpha_11 = pset1<T>(2.00018790482477e-13f);
+  const T alpha_13 = pset1<T>(-2.76076847742355e-16f);
+
+  // The monomial coefficients of the denominator polynomial (even).
+  const T beta_0 = pset1<T>(4.89352518554385e-03f);
+  const T beta_2 = pset1<T>(2.26843463243900e-03f);
+  const T beta_4 = pset1<T>(1.18534705686654e-04f);
+  const T beta_6 = pset1<T>(1.19825839466702e-06f);
+
+  // Since the polynomials are odd/even, we need x^2.
+  const T x2 = pmul(x, x);
+
+  // Evaluate the numerator polynomial p.
+  T p = pmadd(x2, alpha_13, alpha_11);
+  p = pmadd(x2, p, alpha_9);
+  p = pmadd(x2, p, alpha_7);
+  p = pmadd(x2, p, alpha_5);
+  p = pmadd(x2, p, alpha_3);
+  p = pmadd(x2, p, alpha_1);
+  p = pmul(x, p);
+
+  // Evaluate the denominator polynomial p.
+  T q = pmadd(x2, beta_6, beta_4);
+  q = pmadd(x2, q, beta_2);
+  q = pmadd(x2, q, beta_0);
+
+  // Divide the numerator by the denominator.
+  return pdiv(p, q);
+}
+
+template<typename RealScalar>
+EIGEN_STRONG_INLINE
+RealScalar positive_real_hypot(const RealScalar& x, const RealScalar& y)
+{
+  EIGEN_USING_STD_MATH(sqrt);
+  RealScalar p, qp;
+  p = numext::maxi(x,y);
+  if(p==RealScalar(0)) return RealScalar(0);
+  qp = numext::mini(y,x) / p;    
+  return p * sqrt(RealScalar(1) + qp*qp);
+}
+
+template<typename Scalar>
+struct hypot_impl
+{
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+  static inline RealScalar run(const Scalar& x, const Scalar& y)
+  {
+    EIGEN_USING_STD_MATH(abs);
+    return positive_real_hypot<RealScalar>(abs(x), abs(y));
+  }
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_MATHFUNCTIONSIMPL_H
diff --git a/extern/Eigen3/Eigen/src/Core/Matrix.h b/extern/Eigen3/Eigen/src/Core/Matrix.h
index 02be142d8cc..7f4a7af93c6 100644
--- a/extern/Eigen3/Eigen/src/Core/Matrix.h
+++ b/extern/Eigen3/Eigen/src/Core/Matrix.h
@@ -13,6 +13,45 @@
 
 namespace Eigen {
 
+namespace internal {
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+{
+private:
+  enum { size = internal::size_at_compile_time<_Rows,_Cols>::ret };
+  typedef typename find_best_packet<_Scalar,size>::type PacketScalar;
+  enum {
+      row_major_bit = _Options&RowMajor ? RowMajorBit : 0,
+      is_dynamic_size_storage = _MaxRows==Dynamic || _MaxCols==Dynamic,
+      max_size = is_dynamic_size_storage ? Dynamic : _MaxRows*_MaxCols,
+      default_alignment = compute_default_alignment<_Scalar,max_size>::value,
+      actual_alignment = ((_Options&DontAlign)==0) ? default_alignment : 0,
+      required_alignment = unpacket_traits<PacketScalar>::alignment,
+      packet_access_bit = (packet_traits<_Scalar>::Vectorizable && (EIGEN_UNALIGNED_VECTORIZE || (actual_alignment>=required_alignment))) ? PacketAccessBit : 0
+    };
+    
+public:
+  typedef _Scalar Scalar;
+  typedef Dense StorageKind;
+  typedef Eigen::Index StorageIndex;
+  typedef MatrixXpr XprKind;
+  enum {
+    RowsAtCompileTime = _Rows,
+    ColsAtCompileTime = _Cols,
+    MaxRowsAtCompileTime = _MaxRows,
+    MaxColsAtCompileTime = _MaxCols,
+    Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
+    Options = _Options,
+    InnerStrideAtCompileTime = 1,
+    OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime,
+    
+    // FIXME, the following flag in only used to define NeedsToAlign in PlainObjectBase
+    EvaluatorFlags = LinearAccessBit | DirectAccessBit | packet_access_bit | row_major_bit,
+    Alignment = actual_alignment
+  };
+};
+}
+
 /** \class Matrix
   * \ingroup Core_Module
   *
@@ -24,13 +63,13 @@ namespace Eigen {
   * The %Matrix class encompasses \em both fixed-size and dynamic-size objects (\ref fixedsize "note").
   *
   * The first three template parameters are required:
-  * \tparam _Scalar \anchor matrix_tparam_scalar Numeric type, e.g. float, double, int or std::complex<float>.
-  *                 User defined sclar types are supported as well (see \ref user_defined_scalars "here").
+  * \tparam _Scalar Numeric type, e.g. float, double, int or std::complex<float>.
+  *                 User defined scalar types are supported as well (see \ref user_defined_scalars "here").
   * \tparam _Rows Number of rows, or \b Dynamic
   * \tparam _Cols Number of columns, or \b Dynamic
   *
   * The remaining template parameters are optional -- in most cases you don't have to worry about them.
-  * \tparam _Options \anchor matrix_tparam_options A combination of either \b #RowMajor or \b #ColMajor, and of either
+  * \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of either
   *                 \b #AutoAlign or \b #DontAlign.
   *                 The former controls \ref TopicStorageOrders "storage order", and defaults to column-major. The latter controls alignment, which is required
   *                 for vectorization. It defaults to aligning matrices except for fixed sizes that aren't a multiple of the packet size.
@@ -67,7 +106,7 @@ namespace Eigen {
   * \endcode
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN.
   *
   * <i><b>Some notes:</b></i>
   *
@@ -97,32 +136,44 @@ namespace Eigen {
   * are the dimensions of the original matrix, while _Rows and _Cols are Dynamic.</dd>
   * </dl>
   *
-  * \see MatrixBase for the majority of the API methods for matrices, \ref TopicClassHierarchy, 
-  * \ref TopicStorageOrders 
+  * <i><b>ABI and storage layout</b></i>
+  *
+  * The table below summarizes the ABI of some possible Matrix instances which is fixed thorough the lifetime of Eigen 3.
+  * <table  class="manual">
+  * <tr><th>Matrix type</th><th>Equivalent C structure</th></tr>
+  * <tr><td>\code Matrix<T,Dynamic,Dynamic> \endcode</td><td>\code
+  * struct {
+  *   T *data;                  // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0
+  *   Eigen::Index rows, cols;
+  *  };
+  * \endcode</td></tr>
+  * <tr class="alt"><td>\code
+  * Matrix<T,Dynamic,1>
+  * Matrix<T,1,Dynamic> \endcode</td><td>\code
+  * struct {
+  *   T *data;                  // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0
+  *   Eigen::Index size;
+  *  };
+  * \endcode</td></tr>
+  * <tr><td>\code Matrix<T,Rows,Cols> \endcode</td><td>\code
+  * struct {
+  *   T data[Rows*Cols];        // with (size_t(data)%A(Rows*Cols*sizeof(T)))==0
+  *  };
+  * \endcode</td></tr>
+  * <tr class="alt"><td>\code Matrix<T,Dynamic,Dynamic,0,MaxRows,MaxCols> \endcode</td><td>\code
+  * struct {
+  *   T data[MaxRows*MaxCols];  // with (size_t(data)%A(MaxRows*MaxCols*sizeof(T)))==0
+  *   Eigen::Index rows, cols;
+  *  };
+  * \endcode</td></tr>
+  * </table>
+  * Note that in this table Rows, Cols, MaxRows and MaxCols are all positive integers. A(S) is defined to the largest possible power-of-two
+  * smaller to EIGEN_MAX_STATIC_ALIGN_BYTES.
+  *
+  * \see MatrixBase for the majority of the API methods for matrices, \ref TopicClassHierarchy,
+  * \ref TopicStorageOrders
   */
 
-namespace internal {
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
-{
-  typedef _Scalar Scalar;
-  typedef Dense StorageKind;
-  typedef DenseIndex Index;
-  typedef MatrixXpr XprKind;
-  enum {
-    RowsAtCompileTime = _Rows,
-    ColsAtCompileTime = _Cols,
-    MaxRowsAtCompileTime = _MaxRows,
-    MaxColsAtCompileTime = _MaxCols,
-    Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
-    CoeffReadCost = NumTraits<Scalar>::ReadCost,
-    Options = _Options,
-    InnerStrideAtCompileTime = 1,
-    OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime
-  };
-};
-}
-
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 class Matrix
   : public PlainObjectBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
@@ -151,6 +202,7 @@ class Matrix
       *
       * \callgraph
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const Matrix& other)
     {
       return Base::_set(other);
@@ -167,7 +219,8 @@ class Matrix
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Matrix& operator=(const MatrixBase<OtherDerived>& other)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Matrix& operator=(const DenseBase<OtherDerived>& other)
     {
       return Base::_set(other);
     }
@@ -179,12 +232,14 @@ class Matrix
       * \copydetails DenseBase::operator=(const EigenBase<OtherDerived> &other)
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const EigenBase<OtherDerived> &other)
     {
       return Base::operator=(other);
     }
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const ReturnByValue<OtherDerived>& func)
     {
       return Base::operator=(func);
@@ -200,6 +255,7 @@ class Matrix
       *
       * \sa resize(Index,Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix() : Base()
     {
       Base::_check_template_params();
@@ -207,60 +263,85 @@ class Matrix
     }
 
     // FIXME is it still needed
-    Matrix(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC
+    explicit Matrix(internal::constructor_without_unaligned_array_assert)
       : Base(internal::constructor_without_unaligned_array_assert())
     { Base::_check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED }
 
-#ifdef EIGEN_HAVE_RVALUE_REFERENCES
-    Matrix(Matrix&& other)
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    Matrix(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
       : Base(std::move(other))
     {
       Base::_check_template_params();
-      if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic)
-        Base::_set_noalias(other);
     }
-    Matrix& operator=(Matrix&& other)
+    EIGEN_DEVICE_FUNC
+    Matrix& operator=(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
     {
       other.swap(*this);
       return *this;
     }
 #endif
 
-    /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors
-      *
-      * Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
-      * it is redundant to pass the dimension here, so it makes more sense to use the default
-      * constructor Matrix() instead.
-      */
-    EIGEN_STRONG_INLINE explicit Matrix(Index dim)
-      : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+
+    // This constructor is for both 1x1 matrices and dynamic vectors
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE explicit Matrix(const T& x)
     {
       Base::_check_template_params();
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Matrix)
-      eigen_assert(dim >= 0);
-      eigen_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim);
-      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+      Base::template _init1<T>(x);
     }
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const T0& x, const T1& y)
     {
       Base::_check_template_params();
       Base::template _init2<T0,T1>(x, y);
     }
     #else
+    /** \brief Constructs a fixed-sized matrix initialized with coefficients starting at \a data */
+    EIGEN_DEVICE_FUNC
+    explicit Matrix(const Scalar *data);
+
+    /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors
+      *
+      * This is useful for dynamic-size vectors. For fixed-size vectors,
+      * it is redundant to pass these parameters, so one should use the default constructor
+      * Matrix() instead.
+      * 
+      * \warning This constructor is disabled for fixed-size \c 1x1 matrices. For instance,
+      * calling Matrix<double,1,1>(1) will call the initialization constructor: Matrix(const Scalar&).
+      * For fixed-size \c 1x1 matrices it is therefore recommended to use the default
+      * constructor Matrix() instead, especially when using one of the non standard
+      * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
+      */
+    EIGEN_STRONG_INLINE explicit Matrix(Index dim);
+    /** \brief Constructs an initialized 1x1 matrix with the given coefficient */
+    Matrix(const Scalar& x);
     /** \brief Constructs an uninitialized matrix with \a rows rows and \a cols columns.
       *
       * This is useful for dynamic-size matrices. For fixed-size matrices,
       * it is redundant to pass these parameters, so one should use the default constructor
-      * Matrix() instead. */
+      * Matrix() instead.
+      * 
+      * \warning This constructor is disabled for fixed-size \c 1x2 and \c 2x1 vectors. For instance,
+      * calling Matrix2f(2,1) will call the initialization constructor: Matrix(const Scalar& x, const Scalar& y).
+      * For fixed-size \c 1x2 or \c 2x1 vectors it is therefore recommended to use the default
+      * constructor Matrix() instead, especially when using one of the non standard
+      * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
+      */
+    EIGEN_DEVICE_FUNC
     Matrix(Index rows, Index cols);
+    
     /** \brief Constructs an initialized 2D vector with given coefficients */
     Matrix(const Scalar& x, const Scalar& y);
     #endif
 
     /** \brief Constructs an initialized 3D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z)
     {
       Base::_check_template_params();
@@ -270,6 +351,7 @@ class Matrix
       m_storage.data()[2] = z;
     }
     /** \brief Constructs an initialized 4D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z, const Scalar& w)
     {
       Base::_check_template_params();
@@ -280,76 +362,33 @@ class Matrix
       m_storage.data()[3] = w;
     }
 
-    explicit Matrix(const Scalar *data);
 
-    /** \brief Constructor copying the value of the expression \a other */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Matrix(const MatrixBase<OtherDerived>& other)
-             : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      // This test resides here, to bring the error messages closer to the user. Normally, these checks
-      // are performed deeply within the library, thus causing long and scary error traces.
-      EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
     /** \brief Copy constructor */
-    EIGEN_STRONG_INLINE Matrix(const Matrix& other)
-            : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
-    /** \brief Copy constructor with in-place evaluation */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Matrix(const ReturnByValue<OtherDerived>& other)
-    {
-      Base::_check_template_params();
-      Base::resize(other.rows(), other.cols());
-      other.evalTo(*this);
-    }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Matrix(const Matrix& other) : Base(other)
+    { }
 
     /** \brief Copy constructor for generic expressions.
       * \sa MatrixBase::operator=(const EigenBase<OtherDerived>&)
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const EigenBase<OtherDerived> &other)
-      : Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
-    {
-      Base::_check_template_params();
-      Base::_resize_to_match(other);
-      // FIXME/CHECK: isn't *this = other.derived() more efficient. it allows to
-      //              go for pure _set() implementations, right?
-      *this = other;
-    }
-
-    /** \internal
-      * \brief Override MatrixBase::swap() since for dynamic-sized matrices
-      * of same type it is enough to swap the data pointers.
-      */
-    template<typename OtherDerived>
-    void swap(MatrixBase<OtherDerived> const & other)
-    { this->_swap(other.derived()); }
+      : Base(other.derived())
+    { }
 
-    inline Index innerStride() const { return 1; }
-    inline Index outerStride() const { return this->innerSize(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return this->innerSize(); }
 
     /////////// Geometry module ///////////
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     explicit Matrix(const RotationBase<OtherDerived,ColsAtCompileTime>& r);
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Matrix& operator=(const RotationBase<OtherDerived,ColsAtCompileTime>& r);
 
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived>
-    explicit Matrix(const eigen2_RotationBase<OtherDerived,ColsAtCompileTime>& r);
-    template<typename OtherDerived>
-    Matrix& operator=(const eigen2_RotationBase<OtherDerived,ColsAtCompileTime>& r);
-    #endif
-
     // allow to extend Matrix outside Eigen
     #ifdef EIGEN_MATRIX_PLUGIN
     #include EIGEN_MATRIX_PLUGIN
diff --git a/extern/Eigen3/Eigen/src/Core/MatrixBase.h b/extern/Eigen3/Eigen/src/Core/MatrixBase.h
index e83ef4dc056..e6c35907c39 100644
--- a/extern/Eigen3/Eigen/src/Core/MatrixBase.h
+++ b/extern/Eigen3/Eigen/src/Core/MatrixBase.h
@@ -41,9 +41,9 @@ namespace Eigen {
   * \endcode
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN.
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived> class MatrixBase
   : public DenseBase<Derived>
@@ -52,7 +52,7 @@ template<typename Derived> class MatrixBase
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef MatrixBase StorageBaseType;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -66,7 +66,6 @@ template<typename Derived> class MatrixBase
     using Base::MaxSizeAtCompileTime;
     using Base::IsVectorAtCompileTime;
     using Base::Flags;
-    using Base::CoeffReadCost;
 
     using Base::derived;
     using Base::const_cast_derived;
@@ -98,25 +97,14 @@ template<typename Derived> class MatrixBase
 
     /** \returns the size of the main diagonal, which is min(rows(),cols()).
       * \sa rows(), cols(), SizeAtCompileTime. */
-    inline Index diagonalSize() const { return (std::min)(rows(),cols()); }
+    EIGEN_DEVICE_FUNC
+    inline Index diagonalSize() const { return (numext::mini)(rows(),cols()); }
 
-    /** \brief The plain matrix type corresponding to this expression.
-      *
-      * This is not necessarily exactly the return type of eval(). In the case of plain matrices,
-      * the return type of eval() is a const reference to a matrix, not a matrix! It is however guaranteed
-      * that the return type of eval() is either PlainObject or const PlainObject&.
-      */
-    typedef Matrix<typename internal::traits<Derived>::Scalar,
-                internal::traits<Derived>::RowsAtCompileTime,
-                internal::traits<Derived>::ColsAtCompileTime,
-                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
-                internal::traits<Derived>::MaxRowsAtCompileTime,
-                internal::traits<Derived>::MaxColsAtCompileTime
-          > PlainObject;
+    typedef typename Base::PlainObject PlainObject;
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
     /** \internal the return type of MatrixBase::adjoint() */
     typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
                         CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, ConstTransposeReturnType>,
@@ -125,7 +113,7 @@ template<typename Derived> class MatrixBase
     /** \internal Return type of eigenvalues() */
     typedef Matrix<std::complex<RealScalar>, internal::traits<Derived>::ColsAtCompileTime, 1, ColMajor> EigenvaluesReturnType;
     /** \internal the return type of identity */
-    typedef CwiseNullaryOp<internal::scalar_identity_op<Scalar>,Derived> IdentityReturnType;
+    typedef CwiseNullaryOp<internal::scalar_identity_op<Scalar>,PlainObject> IdentityReturnType;
     /** \internal the return type of unit vectors */
     typedef Block<const CwiseNullaryOp<internal::scalar_identity_op<Scalar>, SquareMatrixType>,
                   internal::traits<Derived>::RowsAtCompileTime,
@@ -133,6 +121,7 @@ template<typename Derived> class MatrixBase
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::MatrixBase
+#define EIGEN_DOC_UNARY_ADDONS(X,Y)
 #   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/CommonCwiseBinaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
@@ -141,41 +130,44 @@ template<typename Derived> class MatrixBase
 #     include EIGEN_MATRIXBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
+#undef EIGEN_DOC_UNARY_ADDONS
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const MatrixBase& other);
 
     // We cannot inherit here via Base::operator= since it is causing
     // trouble with MSVC.
 
     template <typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase<OtherDerived>& other);
 
     template <typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const EigenBase<OtherDerived>& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const ReturnByValue<OtherDerived>& other);
 
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    Derived& lazyAssign(const ProductBase<ProductDerived, Lhs,Rhs>& other);
-
-    template<typename MatrixPower, typename Lhs, typename Rhs>
-    Derived& lazyAssign(const MatrixPowerProduct<MatrixPower, Lhs,Rhs>& other);
-
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator+=(const MatrixBase<OtherDerived>& other);
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator-=(const MatrixBase<OtherDerived>& other);
 
     template<typename OtherDerived>
-    const typename ProductReturnType<Derived,OtherDerived>::Type
+    EIGEN_DEVICE_FUNC
+    const Product<Derived,OtherDerived>
     operator*(const MatrixBase<OtherDerived> &other) const;
 
     template<typename OtherDerived>
-    const typename LazyProductReturnType<Derived,OtherDerived>::Type
+    EIGEN_DEVICE_FUNC
+    const Product<Derived,OtherDerived,LazyProduct>
     lazyProduct(const MatrixBase<OtherDerived> &other) const;
 
     template<typename OtherDerived>
@@ -188,84 +180,93 @@ template<typename Derived> class MatrixBase
     void applyOnTheRight(const EigenBase<OtherDerived>& other);
 
     template<typename DiagonalDerived>
-    const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
+    EIGEN_DEVICE_FUNC
+    const Product<Derived, DiagonalDerived, LazyProduct>
     operator*(const DiagonalBase<DiagonalDerived> &diagonal) const;
 
     template<typename OtherDerived>
-    typename internal::scalar_product_traits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
+    EIGEN_DEVICE_FUNC
+    typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
     dot(const MatrixBase<OtherDerived>& other) const;
 
-    #ifdef EIGEN2_SUPPORT
-      template<typename OtherDerived>
-      Scalar eigen2_dot(const MatrixBase<OtherDerived>& other) const;
-    #endif
-
-    RealScalar squaredNorm() const;
-    RealScalar norm() const;
+    EIGEN_DEVICE_FUNC RealScalar squaredNorm() const;
+    EIGEN_DEVICE_FUNC RealScalar norm() const;
     RealScalar stableNorm() const;
     RealScalar blueNorm() const;
     RealScalar hypotNorm() const;
-    const PlainObject normalized() const;
-    void normalize();
+    EIGEN_DEVICE_FUNC const PlainObject normalized() const;
+    EIGEN_DEVICE_FUNC const PlainObject stableNormalized() const;
+    EIGEN_DEVICE_FUNC void normalize();
+    EIGEN_DEVICE_FUNC void stableNormalize();
 
-    const AdjointReturnType adjoint() const;
-    void adjointInPlace();
+    EIGEN_DEVICE_FUNC const AdjointReturnType adjoint() const;
+    EIGEN_DEVICE_FUNC void adjointInPlace();
 
     typedef Diagonal<Derived> DiagonalReturnType;
+    EIGEN_DEVICE_FUNC
     DiagonalReturnType diagonal();
+
     typedef typename internal::add_const<Diagonal<const Derived> >::type ConstDiagonalReturnType;
+    EIGEN_DEVICE_FUNC
     ConstDiagonalReturnType diagonal() const;
 
     template<int Index> struct DiagonalIndexReturnType { typedef Diagonal<Derived,Index> Type; };
     template<int Index> struct ConstDiagonalIndexReturnType { typedef const Diagonal<const Derived,Index> Type; };
 
-    template<int Index> typename DiagonalIndexReturnType<Index>::Type diagonal();
-    template<int Index> typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
-    
+    template<int Index>
+    EIGEN_DEVICE_FUNC
+    typename DiagonalIndexReturnType<Index>::Type diagonal();
+
+    template<int Index>
+    EIGEN_DEVICE_FUNC
+    typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
+
     typedef Diagonal<Derived,DynamicIndex> DiagonalDynamicIndexReturnType;
     typedef typename internal::add_const<Diagonal<const Derived,DynamicIndex> >::type ConstDiagonalDynamicIndexReturnType;
 
+    EIGEN_DEVICE_FUNC
     DiagonalDynamicIndexReturnType diagonal(Index index);
+    EIGEN_DEVICE_FUNC
     ConstDiagonalDynamicIndexReturnType diagonal(Index index) const;
 
-    #ifdef EIGEN2_SUPPORT
-    template<unsigned int Mode> typename internal::eigen2_part_return_type<Derived, Mode>::type part();
-    template<unsigned int Mode> const typename internal::eigen2_part_return_type<Derived, Mode>::type part() const;
-    
-    // huuuge hack. make Eigen2's matrix.part<Diagonal>() work in eigen3. Problem: Diagonal is now a class template instead
-    // of an integer constant. Solution: overload the part() method template wrt template parameters list.
-    template<template<typename T, int N> class U>
-    const DiagonalWrapper<ConstDiagonalReturnType> part() const
-    { return diagonal().asDiagonal(); }
-    #endif // EIGEN2_SUPPORT
-
     template<unsigned int Mode> struct TriangularViewReturnType { typedef TriangularView<Derived, Mode> Type; };
     template<unsigned int Mode> struct ConstTriangularViewReturnType { typedef const TriangularView<const Derived, Mode> Type; };
 
-    template<unsigned int Mode> typename TriangularViewReturnType<Mode>::Type triangularView();
-    template<unsigned int Mode> typename ConstTriangularViewReturnType<Mode>::Type triangularView() const;
+    template<unsigned int Mode>
+    EIGEN_DEVICE_FUNC
+    typename TriangularViewReturnType<Mode>::Type triangularView();
+    template<unsigned int Mode>
+    EIGEN_DEVICE_FUNC
+    typename ConstTriangularViewReturnType<Mode>::Type triangularView() const;
 
     template<unsigned int UpLo> struct SelfAdjointViewReturnType { typedef SelfAdjointView<Derived, UpLo> Type; };
     template<unsigned int UpLo> struct ConstSelfAdjointViewReturnType { typedef const SelfAdjointView<const Derived, UpLo> Type; };
 
-    template<unsigned int UpLo> typename SelfAdjointViewReturnType<UpLo>::Type selfadjointView();
-    template<unsigned int UpLo> typename ConstSelfAdjointViewReturnType<UpLo>::Type selfadjointView() const;
+    template<unsigned int UpLo>
+    EIGEN_DEVICE_FUNC
+    typename SelfAdjointViewReturnType<UpLo>::Type selfadjointView();
+    template<unsigned int UpLo>
+    EIGEN_DEVICE_FUNC
+    typename ConstSelfAdjointViewReturnType<UpLo>::Type selfadjointView() const;
 
     const SparseView<Derived> sparseView(const Scalar& m_reference = Scalar(0),
                                          const typename NumTraits<Scalar>::Real& m_epsilon = NumTraits<Scalar>::dummy_precision()) const;
-    static const IdentityReturnType Identity();
-    static const IdentityReturnType Identity(Index rows, Index cols);
-    static const BasisReturnType Unit(Index size, Index i);
-    static const BasisReturnType Unit(Index i);
-    static const BasisReturnType UnitX();
-    static const BasisReturnType UnitY();
-    static const BasisReturnType UnitZ();
-    static const BasisReturnType UnitW();
-
+    EIGEN_DEVICE_FUNC static const IdentityReturnType Identity();
+    EIGEN_DEVICE_FUNC static const IdentityReturnType Identity(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index size, Index i);
+    EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index i);
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitX();
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitY();
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitZ();
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitW();
+
+    EIGEN_DEVICE_FUNC
     const DiagonalWrapper<const Derived> asDiagonal() const;
     const PermutationWrapper<const Derived> asPermutation() const;
 
+    EIGEN_DEVICE_FUNC
     Derived& setIdentity();
+    EIGEN_DEVICE_FUNC
     Derived& setIdentity(Index rows, Index cols);
 
     bool isIdentity(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
@@ -284,7 +285,7 @@ template<typename Derived> class MatrixBase
       *          fuzzy comparison such as isApprox()
       * \sa isApprox(), operator!= */
     template<typename OtherDerived>
-    inline bool operator==(const MatrixBase<OtherDerived>& other) const
+    EIGEN_DEVICE_FUNC inline bool operator==(const MatrixBase<OtherDerived>& other) const
     { return cwiseEqual(other).all(); }
 
     /** \returns true if at least one pair of coefficients of \c *this and \a other are not exactly equal to each other.
@@ -292,64 +293,50 @@ template<typename Derived> class MatrixBase
       *          fuzzy comparison such as isApprox()
       * \sa isApprox(), operator== */
     template<typename OtherDerived>
-    inline bool operator!=(const MatrixBase<OtherDerived>& other) const
+    EIGEN_DEVICE_FUNC inline bool operator!=(const MatrixBase<OtherDerived>& other) const
     { return cwiseNotEqual(other).any(); }
 
     NoAlias<Derived,Eigen::MatrixBase > noalias();
 
-    inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
-    inline ForceAlignedAccess<Derived> forceAlignedAccess();
-    template<bool Enable> inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type forceAlignedAccessIf() const;
-    template<bool Enable> inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
-
-    Scalar trace() const;
+    // TODO forceAlignedAccess is temporarily disabled
+    // Need to find a nicer workaround.
+    inline const Derived& forceAlignedAccess() const { return derived(); }
+    inline Derived& forceAlignedAccess() { return derived(); }
+    template<bool Enable> inline const Derived& forceAlignedAccessIf() const { return derived(); }
+    template<bool Enable> inline Derived& forceAlignedAccessIf() { return derived(); }
 
-/////////// Array module ///////////
+    EIGEN_DEVICE_FUNC Scalar trace() const;
 
-    template<int p> RealScalar lpNorm() const;
+    template<int p> EIGEN_DEVICE_FUNC RealScalar lpNorm() const;
 
-    MatrixBase<Derived>& matrix() { return *this; }
-    const MatrixBase<Derived>& matrix() const { return *this; }
+    EIGEN_DEVICE_FUNC MatrixBase<Derived>& matrix() { return *this; }
+    EIGEN_DEVICE_FUNC const MatrixBase<Derived>& matrix() const { return *this; }
 
     /** \returns an \link Eigen::ArrayBase Array \endlink expression of this matrix
       * \sa ArrayBase::matrix() */
-    ArrayWrapper<Derived> array() { return derived(); }
-    const ArrayWrapper<const Derived> array() const { return derived(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ArrayWrapper<Derived> array() { return ArrayWrapper<Derived>(derived()); }
+    /** \returns a const \link Eigen::ArrayBase Array \endlink expression of this matrix
+      * \sa ArrayBase::matrix() */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const ArrayWrapper<const Derived> array() const { return ArrayWrapper<const Derived>(derived()); }
 
 /////////// LU module ///////////
 
-    const FullPivLU<PlainObject> fullPivLu() const;
-    const PartialPivLU<PlainObject> partialPivLu() const;
-
-    #if EIGEN2_SUPPORT_STAGE < STAGE20_RESOLVE_API_CONFLICTS
-    const LU<PlainObject> lu() const;
-    #endif
+    inline const FullPivLU<PlainObject> fullPivLu() const;
+    inline const PartialPivLU<PlainObject> partialPivLu() const;
 
-    #ifdef EIGEN2_SUPPORT
-    const LU<PlainObject> eigen2_lu() const;
-    #endif
+    inline const PartialPivLU<PlainObject> lu() const;
 
-    #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
-    const PartialPivLU<PlainObject> lu() const;
-    #endif
-    
-    #ifdef EIGEN2_SUPPORT
-    template<typename ResultType>
-    void computeInverse(MatrixBase<ResultType> *result) const {
-      *result = this->inverse();
-    }
-    #endif
+    inline const Inverse<Derived> inverse() const;
 
-    const internal::inverse_impl<Derived> inverse() const;
     template<typename ResultType>
-    void computeInverseAndDetWithCheck(
+    inline void computeInverseAndDetWithCheck(
       ResultType& inverse,
       typename ResultType::Scalar& determinant,
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
     ) const;
     template<typename ResultType>
-    void computeInverseWithCheck(
+    inline void computeInverseWithCheck(
       ResultType& inverse,
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
@@ -358,65 +345,70 @@ template<typename Derived> class MatrixBase
 
 /////////// Cholesky module ///////////
 
-    const LLT<PlainObject>  llt() const;
-    const LDLT<PlainObject> ldlt() const;
+    inline const LLT<PlainObject>  llt() const;
+    inline const LDLT<PlainObject> ldlt() const;
 
 /////////// QR module ///////////
 
-    const HouseholderQR<PlainObject> householderQr() const;
-    const ColPivHouseholderQR<PlainObject> colPivHouseholderQr() const;
-    const FullPivHouseholderQR<PlainObject> fullPivHouseholderQr() const;
-    
-    #ifdef EIGEN2_SUPPORT
-    const QR<PlainObject> qr() const;
-    #endif
+    inline const HouseholderQR<PlainObject> householderQr() const;
+    inline const ColPivHouseholderQR<PlainObject> colPivHouseholderQr() const;
+    inline const FullPivHouseholderQR<PlainObject> fullPivHouseholderQr() const;
+    inline const CompleteOrthogonalDecomposition<PlainObject> completeOrthogonalDecomposition() const;
 
-    EigenvaluesReturnType eigenvalues() const;
-    RealScalar operatorNorm() const;
+/////////// Eigenvalues module ///////////
 
-/////////// SVD module ///////////
+    inline EigenvaluesReturnType eigenvalues() const;
+    inline RealScalar operatorNorm() const;
 
-    JacobiSVD<PlainObject> jacobiSvd(unsigned int computationOptions = 0) const;
+/////////// SVD module ///////////
 
-    #ifdef EIGEN2_SUPPORT
-    SVD<PlainObject> svd() const;
-    #endif
+    inline JacobiSVD<PlainObject> jacobiSvd(unsigned int computationOptions = 0) const;
+    inline BDCSVD<PlainObject>    bdcSvd(unsigned int computationOptions = 0) const;
 
 /////////// Geometry module ///////////
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /// \internal helper struct to form the return type of the cross product
     template<typename OtherDerived> struct cross_product_return_type {
-      typedef typename internal::scalar_product_traits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType Scalar;
+      typedef typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType Scalar;
       typedef Matrix<Scalar,MatrixBase::RowsAtCompileTime,MatrixBase::ColsAtCompileTime> type;
     };
     #endif // EIGEN_PARSED_BY_DOXYGEN
     template<typename OtherDerived>
-    typename cross_product_return_type<OtherDerived>::type
+    EIGEN_DEVICE_FUNC
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    inline typename cross_product_return_type<OtherDerived>::type
+#else
+    inline PlainObject
+#endif
     cross(const MatrixBase<OtherDerived>& other) const;
+
     template<typename OtherDerived>
-    PlainObject cross3(const MatrixBase<OtherDerived>& other) const;
-    PlainObject unitOrthogonal(void) const;
-    Matrix<Scalar,3,1> eulerAngles(Index a0, Index a1, Index a2) const;
-    
-    #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
-    ScalarMultipleReturnType operator*(const UniformScaling<Scalar>& s) const;
+    EIGEN_DEVICE_FUNC
+    inline PlainObject cross3(const MatrixBase<OtherDerived>& other) const;
+
+    EIGEN_DEVICE_FUNC
+    inline PlainObject unitOrthogonal(void) const;
+
+    EIGEN_DEVICE_FUNC
+    inline Matrix<Scalar,3,1> eulerAngles(Index a0, Index a1, Index a2) const;
+
     // put this as separate enum value to work around possible GCC 4.3 bug (?)
-    enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1?Vertical:Horizontal };
+    enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1&&RowsAtCompileTime==1 ? ((internal::traits<Derived>::Flags&RowMajorBit)==RowMajorBit ? Horizontal : Vertical)
+                                          : ColsAtCompileTime==1 ? Vertical : Horizontal };
     typedef Homogeneous<Derived, HomogeneousReturnTypeDirection> HomogeneousReturnType;
-    HomogeneousReturnType homogeneous() const;
-    #endif
-    
+    EIGEN_DEVICE_FUNC
+    inline HomogeneousReturnType homogeneous() const;
+
     enum {
       SizeMinusOne = SizeAtCompileTime==Dynamic ? Dynamic : SizeAtCompileTime-1
     };
     typedef Block<const Derived,
                   internal::traits<Derived>::ColsAtCompileTime==1 ? SizeMinusOne : 1,
                   internal::traits<Derived>::ColsAtCompileTime==1 ? 1 : SizeMinusOne> ConstStartMinusOne;
-    typedef CwiseUnaryOp<internal::scalar_quotient1_op<typename internal::traits<Derived>::Scalar>,
-                const ConstStartMinusOne > HNormalizedReturnType;
-
-    const HNormalizedReturnType hnormalized() const;
+    typedef EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(ConstStartMinusOne,Scalar,quotient) HNormalizedReturnType;
+    EIGEN_DEVICE_FUNC
+    inline const HNormalizedReturnType hnormalized() const;
 
 ////////// Householder module ///////////
 
@@ -452,58 +444,32 @@ template<typename Derived> class MatrixBase
 ///////// MatrixFunctions module /////////
 
     typedef typename internal::stem_function<Scalar>::type StemFunction;
-    const MatrixExponentialReturnValue<Derived> exp() const;
+#define EIGEN_MATRIX_FUNCTION(ReturnType, Name, Description) \
+    /** \returns an expression of the matrix Description of \c *this. \brief This function requires the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>. To compute the coefficient-wise Description use ArrayBase::##Name . */ \
+    const ReturnType<Derived> Name() const;
+#define EIGEN_MATRIX_FUNCTION_1(ReturnType, Name, Description, Argument) \
+    /** \returns an expression of the matrix Description of \c *this. \brief This function requires the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>. To compute the coefficient-wise Description use ArrayBase::##Name . */ \
+    const ReturnType<Derived> Name(Argument) const;
+
+    EIGEN_MATRIX_FUNCTION(MatrixExponentialReturnValue, exp, exponential)
+    /** \brief Helper function for the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>.*/
     const MatrixFunctionReturnValue<Derived> matrixFunction(StemFunction f) const;
-    const MatrixFunctionReturnValue<Derived> cosh() const;
-    const MatrixFunctionReturnValue<Derived> sinh() const;
-    const MatrixFunctionReturnValue<Derived> cos() const;
-    const MatrixFunctionReturnValue<Derived> sin() const;
-    const MatrixSquareRootReturnValue<Derived> sqrt() const;
-    const MatrixLogarithmReturnValue<Derived> log() const;
-    const MatrixPowerReturnValue<Derived> pow(const RealScalar& p) const;
-
-#ifdef EIGEN2_SUPPORT
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    Derived& operator+=(const Flagged<ProductBase<ProductDerived, Lhs,Rhs>, 0,
-                                      EvalBeforeAssigningBit>& other);
-
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    Derived& operator-=(const Flagged<ProductBase<ProductDerived, Lhs,Rhs>, 0,
-                                      EvalBeforeAssigningBit>& other);
-
-    /** \deprecated because .lazy() is deprecated
-      * Overloaded for cache friendly product evaluation */
-    template<typename OtherDerived>
-    Derived& lazyAssign(const Flagged<OtherDerived, 0, EvalBeforeAssigningBit>& other)
-    { return lazyAssign(other._expression()); }
-
-    template<unsigned int Added>
-    const Flagged<Derived, Added, 0> marked() const;
-    const Flagged<Derived, 0, EvalBeforeAssigningBit> lazy() const;
-
-    inline const Cwise<Derived> cwise() const;
-    inline Cwise<Derived> cwise();
-
-    VectorBlock<Derived> start(Index size);
-    const VectorBlock<const Derived> start(Index size) const;
-    VectorBlock<Derived> end(Index size);
-    const VectorBlock<const Derived> end(Index size) const;
-    template<int Size> VectorBlock<Derived,Size> start();
-    template<int Size> const VectorBlock<const Derived,Size> start() const;
-    template<int Size> VectorBlock<Derived,Size> end();
-    template<int Size> const VectorBlock<const Derived,Size> end() const;
-
-    Minor<Derived> minor(Index row, Index col);
-    const Minor<Derived> minor(Index row, Index col) const;
-#endif
+    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cosh, hyperbolic cosine)
+    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sinh, hyperbolic sine)
+    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cos, cosine)
+    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sin, sine)
+    EIGEN_MATRIX_FUNCTION(MatrixSquareRootReturnValue, sqrt, square root)
+    EIGEN_MATRIX_FUNCTION(MatrixLogarithmReturnValue, log, logarithm)
+    EIGEN_MATRIX_FUNCTION_1(MatrixPowerReturnValue,        pow, power to \c p, const RealScalar& p)
+    EIGEN_MATRIX_FUNCTION_1(MatrixComplexPowerReturnValue, pow, power to \c p, const std::complex<RealScalar>& p)
 
   protected:
-    MatrixBase() : Base() {}
+    EIGEN_DEVICE_FUNC MatrixBase() : Base() {}
 
   private:
-    explicit MatrixBase(int);
-    MatrixBase(int,int);
-    template<typename OtherDerived> explicit MatrixBase(const MatrixBase<OtherDerived>&);
+    EIGEN_DEVICE_FUNC explicit MatrixBase(int);
+    EIGEN_DEVICE_FUNC MatrixBase(int,int);
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit MatrixBase(const MatrixBase<OtherDerived>&);
   protected:
     // mixing arrays and matrices is not legal
     template<typename OtherDerived> Derived& operator+=(const ArrayBase<OtherDerived>& )
diff --git a/extern/Eigen3/Eigen/src/Core/NestByValue.h b/extern/Eigen3/Eigen/src/Core/NestByValue.h
index a893b1761b5..13adf070e89 100644
--- a/extern/Eigen3/Eigen/src/Core/NestByValue.h
+++ b/extern/Eigen3/Eigen/src/Core/NestByValue.h
@@ -13,25 +13,24 @@
 
 namespace Eigen {
 
+namespace internal {
+template<typename ExpressionType>
+struct traits<NestByValue<ExpressionType> > : public traits<ExpressionType>
+{};
+}
+
 /** \class NestByValue
   * \ingroup Core_Module
   *
   * \brief Expression which must be nested by value
   *
-  * \param ExpressionType the type of the object of which we are requiring nesting-by-value
+  * \tparam ExpressionType the type of the object of which we are requiring nesting-by-value
   *
   * This class is the return type of MatrixBase::nestByValue()
   * and most of the time this is the only way it is used.
   *
   * \sa MatrixBase::nestByValue()
   */
-
-namespace internal {
-template<typename ExpressionType>
-struct traits<NestByValue<ExpressionType> > : public traits<ExpressionType>
-{};
-}
-
 template<typename ExpressionType> class NestByValue
   : public internal::dense_xpr_base< NestByValue<ExpressionType> >::type
 {
@@ -40,29 +39,29 @@ template<typename ExpressionType> class NestByValue
     typedef typename internal::dense_xpr_base<NestByValue>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(NestByValue)
 
-    inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
+    EIGEN_DEVICE_FUNC explicit inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    inline Index rows() const { return m_expression.rows(); }
-    inline Index cols() const { return m_expression.cols(); }
-    inline Index outerStride() const { return m_expression.outerStride(); }
-    inline Index innerStride() const { return m_expression.innerStride(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); }
 
-    inline const CoeffReturnType coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
     {
       return m_expression.coeff(row, col);
     }
 
-    inline Scalar& coeffRef(Index row, Index col)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
     {
       return m_expression.const_cast_derived().coeffRef(row, col);
     }
 
-    inline const CoeffReturnType coeff(Index index) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
     {
       return m_expression.coeff(index);
     }
 
-    inline Scalar& coeffRef(Index index)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
     {
       return m_expression.const_cast_derived().coeffRef(index);
     }
@@ -91,7 +90,7 @@ template<typename ExpressionType> class NestByValue
       m_expression.const_cast_derived().template writePacket<LoadMode>(index, x);
     }
 
-    operator const ExpressionType&() const { return m_expression; }
+    EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
 
   protected:
     const ExpressionType m_expression;
diff --git a/extern/Eigen3/Eigen/src/Core/NoAlias.h b/extern/Eigen3/Eigen/src/Core/NoAlias.h
index 768bfb18ca4..33908010b4f 100644
--- a/extern/Eigen3/Eigen/src/Core/NoAlias.h
+++ b/extern/Eigen3/Eigen/src/Core/NoAlias.h
@@ -17,7 +17,7 @@ namespace Eigen {
   *
   * \brief Pseudo expression providing an operator = assuming no aliasing
   *
-  * \param ExpressionType the type of the object on which to do the lazy assignment
+  * \tparam ExpressionType the type of the object on which to do the lazy assignment
   *
   * This class represents an expression with special assignment operators
   * assuming no aliasing between the target expression and the source expression.
@@ -30,62 +30,36 @@ namespace Eigen {
 template<typename ExpressionType, template <typename> class StorageBase>
 class NoAlias
 {
-    typedef typename ExpressionType::Scalar Scalar;
   public:
-    NoAlias(ExpressionType& expression) : m_expression(expression) {}
-
-    /** Behaves like MatrixBase::lazyAssign(other)
-      * \sa MatrixBase::lazyAssign() */
+    typedef typename ExpressionType::Scalar Scalar;
+    
+    explicit NoAlias(ExpressionType& expression) : m_expression(expression) {}
+    
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE ExpressionType& operator=(const StorageBase<OtherDerived>& other)
-    { return internal::assign_selector<ExpressionType,OtherDerived,false>::run(m_expression,other.derived()); }
-
-    /** \sa MatrixBase::operator+= */
+    {
+      call_assignment_no_alias(m_expression, other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+      return m_expression;
+    }
+    
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE ExpressionType& operator+=(const StorageBase<OtherDerived>& other)
     {
-      typedef SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
-      SelfAdder tmp(m_expression);
-      typedef typename internal::nested<OtherDerived>::type OtherDerivedNested;
-      typedef typename internal::remove_all<OtherDerivedNested>::type _OtherDerivedNested;
-      internal::assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
+      call_assignment_no_alias(m_expression, other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
       return m_expression;
     }
-
-    /** \sa MatrixBase::operator-= */
+    
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase<OtherDerived>& other)
     {
-      typedef SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
-      SelfAdder tmp(m_expression);
-      typedef typename internal::nested<OtherDerived>::type OtherDerivedNested;
-      typedef typename internal::remove_all<OtherDerivedNested>::type _OtherDerivedNested;
-      internal::assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
+      call_assignment_no_alias(m_expression, other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
       return m_expression;
     }
 
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE ExpressionType& operator+=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    { other.derived().addTo(m_expression); return m_expression; }
-
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE ExpressionType& operator-=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    { other.derived().subTo(m_expression); return m_expression; }
-
-    template<typename Lhs, typename Rhs, int NestingFlags>
-    EIGEN_STRONG_INLINE ExpressionType& operator+=(const CoeffBasedProduct<Lhs,Rhs,NestingFlags>& other)
-    { return m_expression.derived() += CoeffBasedProduct<Lhs,Rhs,NestByRefBit>(other.lhs(), other.rhs()); }
-
-    template<typename Lhs, typename Rhs, int NestingFlags>
-    EIGEN_STRONG_INLINE ExpressionType& operator-=(const CoeffBasedProduct<Lhs,Rhs,NestingFlags>& other)
-    { return m_expression.derived() -= CoeffBasedProduct<Lhs,Rhs,NestByRefBit>(other.lhs(), other.rhs()); }
-    
-    template<typename OtherDerived>
-    ExpressionType& operator=(const ReturnByValue<OtherDerived>& func)
-    { return m_expression = func; }
-#endif
-
+    EIGEN_DEVICE_FUNC
     ExpressionType& expression() const
     {
       return m_expression;
@@ -126,7 +100,7 @@ class NoAlias
 template<typename Derived>
 NoAlias<Derived,MatrixBase> MatrixBase<Derived>::noalias()
 {
-  return derived();
+  return NoAlias<Derived, Eigen::MatrixBase >(derived());
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/NumTraits.h b/extern/Eigen3/Eigen/src/Core/NumTraits.h
index bac9e50b857..daf48987898 100644
--- a/extern/Eigen3/Eigen/src/Core/NumTraits.h
+++ b/extern/Eigen3/Eigen/src/Core/NumTraits.h
@@ -12,24 +12,57 @@
 
 namespace Eigen {
 
+namespace internal {
+
+// default implementation of digits10(), based on numeric_limits if specialized,
+// 0 for integer types, and log10(epsilon()) otherwise.
+template< typename T,
+          bool use_numeric_limits = std::numeric_limits<T>::is_specialized,
+          bool is_integer = NumTraits<T>::IsInteger>
+struct default_digits10_impl
+{
+  static int run() { return std::numeric_limits<T>::digits10; }
+};
+
+template<typename T>
+struct default_digits10_impl<T,false,false> // Floating point
+{
+  static int run() {
+    using std::log10;
+    using std::ceil;
+    typedef typename NumTraits<T>::Real Real;
+    return int(ceil(-log10(NumTraits<Real>::epsilon())));
+  }
+};
+
+template<typename T>
+struct default_digits10_impl<T,false,true> // Integer
+{
+  static int run() { return 0; }
+};
+
+} // end namespace internal
+
 /** \class NumTraits
   * \ingroup Core_Module
   *
   * \brief Holds information about the various numeric (i.e. scalar) types allowed by Eigen.
   *
-  * \param T the numeric type at hand
+  * \tparam T the numeric type at hand
   *
   * This class stores enums, typedefs and static methods giving information about a numeric type.
   *
   * The provided data consists of:
-  * \li A typedef \a Real, giving the "real part" type of \a T. If \a T is already real,
-  *     then \a Real is just a typedef to \a T. If \a T is \c std::complex<U> then \a Real
+  * \li A typedef \c Real, giving the "real part" type of \a T. If \a T is already real,
+  *     then \c Real is just a typedef to \a T. If \a T is \c std::complex<U> then \c Real
   *     is a typedef to \a U.
-  * \li A typedef \a NonInteger, giving the type that should be used for operations producing non-integral values,
+  * \li A typedef \c NonInteger, giving the type that should be used for operations producing non-integral values,
   *     such as quotients, square roots, etc. If \a T is a floating-point type, then this typedef just gives
   *     \a T again. Note however that many Eigen functions such as internal::sqrt simply refuse to
   *     take integers. Outside of a few cases, Eigen doesn't do automatic type promotion. Thus, this typedef is
   *     only intended as a helper for code that needs to explicitly promote types.
+  * \li A typedef \c Literal giving the type to use for numeric literals such as "2" or "0.5". For instance, for \c std::complex<U>, Literal is defined as \c U.
+  *     Of course, this type must be fully compatible with \a T. In doubt, just use \a T here.
   * \li A typedef \a Nested giving the type to use to nest a value inside of the expression tree. If you don't know what
   *     this means, just use \a T here.
   * \li An enum value \a IsComplex. It is equal to 1 if \a T is a \c std::complex
@@ -42,10 +75,14 @@ namespace Eigen {
   * \li An enum value \a IsSigned. It is equal to \c 1 if \a T is a signed type and to 0 if \a T is unsigned.
   * \li An enum value \a RequireInitialization. It is equal to \c 1 if the constructor of the numeric type \a T must
   *     be called, and to 0 if it is safe not to call it. Default is 0 if \a T is an arithmetic type, and 1 otherwise.
-  * \li An epsilon() function which, unlike std::numeric_limits::epsilon(), returns a \a Real instead of a \a T.
+  * \li An epsilon() function which, unlike <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/epsilon">std::numeric_limits::epsilon()</a>,
+  *     it returns a \a Real instead of a \a T.
   * \li A dummy_precision() function returning a weak epsilon value. It is mainly used as a default
   *     value by the fuzzy comparison operators.
   * \li highest() and lowest() functions returning the highest and lowest possible values respectively.
+  * \li digits10() function returning the number of decimal digits that can be represented without change. This is
+  *     the analogue of <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/digits10">std::numeric_limits<T>::digits10</a>
+  *     which is used as the default implementation if specialized.
   */
 
 template<typename T> struct GenericNumTraits
@@ -67,22 +104,47 @@ template<typename T> struct GenericNumTraits
                      T
                    >::type NonInteger;
   typedef T Nested;
+  typedef T Literal;
+
+  EIGEN_DEVICE_FUNC
+  static inline Real epsilon()
+  {
+    return numext::numeric_limits<T>::epsilon();
+  }
+
+  EIGEN_DEVICE_FUNC
+  static inline int digits10()
+  {
+    return internal::default_digits10_impl<T>::run();
+  }
 
-  static inline Real epsilon() { return std::numeric_limits<T>::epsilon(); }
+  EIGEN_DEVICE_FUNC
   static inline Real dummy_precision()
   {
     // make sure to override this for floating-point types
     return Real(0);
   }
-  static inline T highest() { return (std::numeric_limits<T>::max)(); }
-  static inline T lowest()  { return IsInteger ? (std::numeric_limits<T>::min)() : (-(std::numeric_limits<T>::max)()); }
-  
-#ifdef EIGEN2_SUPPORT
-  enum {
-    HasFloatingPoint = !IsInteger
-  };
-  typedef NonInteger FloatingPoint;
-#endif
+
+
+  EIGEN_DEVICE_FUNC
+  static inline T highest() {
+    return (numext::numeric_limits<T>::max)();
+  }
+
+  EIGEN_DEVICE_FUNC
+  static inline T lowest()  {
+    return IsInteger ? (numext::numeric_limits<T>::min)() : (-(numext::numeric_limits<T>::max)());
+  }
+
+  EIGEN_DEVICE_FUNC
+  static inline T infinity() {
+    return numext::numeric_limits<T>::infinity();
+  }
+
+  EIGEN_DEVICE_FUNC
+  static inline T quiet_NaN() {
+    return numext::numeric_limits<T>::quiet_NaN();
+  }
 };
 
 template<typename T> struct NumTraits : GenericNumTraits<T>
@@ -91,11 +153,13 @@ template<typename T> struct NumTraits : GenericNumTraits<T>
 template<> struct NumTraits<float>
   : GenericNumTraits<float>
 {
+  EIGEN_DEVICE_FUNC
   static inline float dummy_precision() { return 1e-5f; }
 };
 
 template<> struct NumTraits<double> : GenericNumTraits<double>
 {
+  EIGEN_DEVICE_FUNC
   static inline double dummy_precision() { return 1e-12; }
 };
 
@@ -109,6 +173,7 @@ template<typename _Real> struct NumTraits<std::complex<_Real> >
   : GenericNumTraits<std::complex<_Real> >
 {
   typedef _Real Real;
+  typedef typename NumTraits<_Real>::Literal Literal;
   enum {
     IsComplex = 1,
     RequireInitialization = NumTraits<_Real>::RequireInitialization,
@@ -117,8 +182,12 @@ template<typename _Real> struct NumTraits<std::complex<_Real> >
     MulCost = 4 * NumTraits<Real>::MulCost + 2 * NumTraits<Real>::AddCost
   };
 
+  EIGEN_DEVICE_FUNC
   static inline Real epsilon() { return NumTraits<Real>::epsilon(); }
+  EIGEN_DEVICE_FUNC
   static inline Real dummy_precision() { return NumTraits<Real>::dummy_precision(); }
+  EIGEN_DEVICE_FUNC
+  static inline int digits10() { return NumTraits<Real>::digits10(); }
 };
 
 template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
@@ -130,21 +199,50 @@ struct NumTraits<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
   typedef typename NumTraits<Scalar>::NonInteger NonIntegerScalar;
   typedef Array<NonIntegerScalar, Rows, Cols, Options, MaxRows, MaxCols> NonInteger;
   typedef ArrayType & Nested;
-  
+  typedef typename NumTraits<Scalar>::Literal Literal;
+
   enum {
     IsComplex = NumTraits<Scalar>::IsComplex,
     IsInteger = NumTraits<Scalar>::IsInteger,
     IsSigned  = NumTraits<Scalar>::IsSigned,
     RequireInitialization = 1,
-    ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::ReadCost,
-    AddCost  = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::AddCost,
-    MulCost  = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::MulCost
+    ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::ReadCost,
+    AddCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::AddCost,
+    MulCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::MulCost
   };
-  
+
+  EIGEN_DEVICE_FUNC
   static inline RealScalar epsilon() { return NumTraits<RealScalar>::epsilon(); }
+  EIGEN_DEVICE_FUNC
   static inline RealScalar dummy_precision() { return NumTraits<RealScalar>::dummy_precision(); }
+
+  static inline int digits10() { return NumTraits<Scalar>::digits10(); }
+};
+
+template<> struct NumTraits<std::string>
+  : GenericNumTraits<std::string>
+{
+  enum {
+    RequireInitialization = 1,
+    ReadCost = HugeCost,
+    AddCost  = HugeCost,
+    MulCost  = HugeCost
+  };
+
+  static inline int digits10() { return 0; }
+
+private:
+  static inline std::string epsilon();
+  static inline std::string dummy_precision();
+  static inline std::string lowest();
+  static inline std::string highest();
+  static inline std::string infinity();
+  static inline std::string quiet_NaN();
 };
 
+// Empty specialization for void to allow template specialization based on NumTraits<T>::Real with T==void and SFINAE.
+template<> struct NumTraits<void> {};
+
 } // end namespace Eigen
 
 #endif // EIGEN_NUMTRAITS_H
diff --git a/extern/Eigen3/Eigen/src/Core/PermutationMatrix.h b/extern/Eigen3/Eigen/src/Core/PermutationMatrix.h
index 85ffae2653b..b1fb455b98c 100644
--- a/extern/Eigen3/Eigen/src/Core/PermutationMatrix.h
+++ b/extern/Eigen3/Eigen/src/Core/PermutationMatrix.h
@@ -2,7 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2009-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -13,14 +13,18 @@
 
 namespace Eigen { 
 
-template<int RowCol,typename IndicesType,typename MatrixType, typename StorageKind> class PermutedImpl;
+namespace internal {
+
+enum PermPermProduct_t {PermPermProduct};
+
+} // end namespace internal
 
 /** \class PermutationBase
   * \ingroup Core_Module
   *
   * \brief Base class for permutations
   *
-  * \param Derived the derived class
+  * \tparam Derived the derived class
   *
   * This class is the base class for all expressions representing a permutation matrix,
   * internally stored as a vector of integers.
@@ -38,17 +42,6 @@ template<int RowCol,typename IndicesType,typename MatrixType, typename StorageKi
   *
   * \sa class PermutationMatrix, class PermutationWrapper
   */
-
-namespace internal {
-
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed=false>
-struct permut_matrix_product_retval;
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed=false>
-struct permut_sparsematrix_product_retval;
-enum PermPermProduct_t {PermPermProduct};
-
-} // end namespace internal
-
 template<typename Derived>
 class PermutationBase : public EigenBase<Derived>
 {
@@ -60,19 +53,20 @@ class PermutationBase : public EigenBase<Derived>
     typedef typename Traits::IndicesType IndicesType;
     enum {
       Flags = Traits::Flags,
-      CoeffReadCost = Traits::CoeffReadCost,
       RowsAtCompileTime = Traits::RowsAtCompileTime,
       ColsAtCompileTime = Traits::ColsAtCompileTime,
       MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = Traits::MaxColsAtCompileTime
     };
-    typedef typename Traits::Scalar Scalar;
-    typedef typename Traits::Index Index;
-    typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime,0,MaxRowsAtCompileTime,MaxColsAtCompileTime>
+    typedef typename Traits::StorageIndex StorageIndex;
+    typedef Matrix<StorageIndex,RowsAtCompileTime,ColsAtCompileTime,0,MaxRowsAtCompileTime,MaxColsAtCompileTime>
             DenseMatrixType;
-    typedef PermutationMatrix<IndicesType::SizeAtCompileTime,IndicesType::MaxSizeAtCompileTime,Index>
+    typedef PermutationMatrix<IndicesType::SizeAtCompileTime,IndicesType::MaxSizeAtCompileTime,StorageIndex>
             PlainPermutationType;
+    typedef PlainPermutationType PlainObject;
     using Base::derived;
+    typedef Inverse<Derived> InverseReturnType;
+    typedef void Scalar;
     #endif
 
     /** Copies the other permutation into *this */
@@ -118,7 +112,7 @@ class PermutationBase : public EigenBase<Derived>
     void evalTo(MatrixBase<DenseDerived>& other) const
     {
       other.setZero();
-      for (int i=0; i<rows();++i)
+      for (Index i=0; i<rows(); ++i)
         other.coeffRef(indices().coeff(i),i) = typename DenseDerived::Scalar(1);
     }
     #endif
@@ -147,7 +141,8 @@ class PermutationBase : public EigenBase<Derived>
     /** Sets *this to be the identity permutation matrix */
     void setIdentity()
     {
-      for(Index i = 0; i < size(); ++i)
+      StorageIndex n = StorageIndex(size());
+      for(StorageIndex i = 0; i < n; ++i)
         indices().coeffRef(i) = i;
     }
 
@@ -163,18 +158,18 @@ class PermutationBase : public EigenBase<Derived>
       *
       * \returns a reference to *this.
       *
-      * \warning This is much slower than applyTranspositionOnTheRight(int,int):
+      * \warning This is much slower than applyTranspositionOnTheRight(Index,Index):
       * this has linear complexity and requires a lot of branching.
       *
-      * \sa applyTranspositionOnTheRight(int,int)
+      * \sa applyTranspositionOnTheRight(Index,Index)
       */
     Derived& applyTranspositionOnTheLeft(Index i, Index j)
     {
       eigen_assert(i>=0 && j>=0 && i<size() && j<size());
       for(Index k = 0; k < size(); ++k)
       {
-        if(indices().coeff(k) == i) indices().coeffRef(k) = j;
-        else if(indices().coeff(k) == j) indices().coeffRef(k) = i;
+        if(indices().coeff(k) == i) indices().coeffRef(k) = StorageIndex(j);
+        else if(indices().coeff(k) == j) indices().coeffRef(k) = StorageIndex(i);
       }
       return derived();
     }
@@ -185,7 +180,7 @@ class PermutationBase : public EigenBase<Derived>
       *
       * This is a fast operation, it only consists in swapping two indices.
       *
-      * \sa applyTranspositionOnTheLeft(int,int)
+      * \sa applyTranspositionOnTheLeft(Index,Index)
       */
     Derived& applyTranspositionOnTheRight(Index i, Index j)
     {
@@ -196,16 +191,16 @@ class PermutationBase : public EigenBase<Derived>
 
     /** \returns the inverse permutation matrix.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
-    inline Transpose<PermutationBase> inverse() const
-    { return derived(); }
+    inline InverseReturnType inverse() const
+    { return InverseReturnType(derived()); }
     /** \returns the tranpose permutation matrix.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
-    inline Transpose<PermutationBase> transpose() const
-    { return derived(); }
+    inline InverseReturnType transpose() const
+    { return InverseReturnType(derived()); }
 
     /**** multiplication helpers to hopefully get RVO ****/
 
@@ -215,13 +210,13 @@ class PermutationBase : public EigenBase<Derived>
     template<typename OtherDerived>
     void assignTranspose(const PermutationBase<OtherDerived>& other)
     {
-      for (int i=0; i<rows();++i) indices().coeffRef(other.indices().coeff(i)) = i;
+      for (Index i=0; i<rows();++i) indices().coeffRef(other.indices().coeff(i)) = i;
     }
     template<typename Lhs,typename Rhs>
     void assignProduct(const Lhs& lhs, const Rhs& rhs)
     {
       eigen_assert(lhs.cols() == rhs.rows());
-      for (int i=0; i<rows();++i) indices().coeffRef(i) = lhs.indices().coeff(rhs.indices().coeff(i));
+      for (Index i=0; i<rows();++i) indices().coeffRef(i) = lhs.indices().coeff(rhs.indices().coeff(i));
     }
 #endif
 
@@ -229,7 +224,7 @@ class PermutationBase : public EigenBase<Derived>
 
     /** \returns the product permutation matrix.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
     template<typename Other>
     inline PlainPermutationType operator*(const PermutationBase<Other>& other) const
@@ -237,18 +232,18 @@ class PermutationBase : public EigenBase<Derived>
 
     /** \returns the product of a permutation with another inverse permutation.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
     template<typename Other>
-    inline PlainPermutationType operator*(const Transpose<PermutationBase<Other> >& other) const
+    inline PlainPermutationType operator*(const InverseImpl<Other,PermutationStorage>& other) const
     { return PlainPermutationType(internal::PermPermProduct, *this, other.eval()); }
 
     /** \returns the product of an inverse permutation with another permutation.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
     template<typename Other> friend
-    inline PlainPermutationType operator*(const Transpose<PermutationBase<Other> >& other, const PermutationBase& perm)
+    inline PlainPermutationType operator*(const InverseImpl<Other, PermutationStorage>& other, const PermutationBase& perm)
     { return PlainPermutationType(internal::PermPermProduct, other.eval(), perm); }
     
     /** \returns the determinant of the permutation matrix, which is either 1 or -1 depending on the parity of the permutation.
@@ -284,39 +279,43 @@ class PermutationBase : public EigenBase<Derived>
 
 };
 
+namespace internal {
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+struct traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
+ : traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
+{
+  typedef PermutationStorage StorageKind;
+  typedef Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
+  typedef _StorageIndex StorageIndex;
+  typedef void Scalar;
+};
+}
+
 /** \class PermutationMatrix
   * \ingroup Core_Module
   *
   * \brief Permutation matrix
   *
-  * \param SizeAtCompileTime the number of rows/cols, or Dynamic
-  * \param MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
-  * \param IndexType the interger type of the indices
+  * \tparam SizeAtCompileTime the number of rows/cols, or Dynamic
+  * \tparam MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
+  * \tparam _StorageIndex the integer type of the indices
   *
   * This class represents a permutation matrix, internally stored as a vector of integers.
   *
   * \sa class PermutationBase, class PermutationWrapper, class DiagonalMatrix
   */
-
-namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
-struct traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType> >
- : traits<Matrix<IndexType,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
-{
-  typedef IndexType Index;
-  typedef Matrix<IndexType, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
-};
-}
-
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
-class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
 {
     typedef PermutationBase<PermutationMatrix> Base;
     typedef internal::traits<PermutationMatrix> Traits;
   public:
 
+    typedef const PermutationMatrix& Nested;
+
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef typename Traits::IndicesType IndicesType;
+    typedef typename Traits::StorageIndex StorageIndex;
     #endif
 
     inline PermutationMatrix()
@@ -324,8 +323,10 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
 
     /** Constructs an uninitialized permutation matrix of given size.
       */
-    inline PermutationMatrix(int size) : m_indices(size)
-    {}
+    explicit inline PermutationMatrix(Index size) : m_indices(size)
+    {
+      eigen_internal_assert(size <= NumTraits<StorageIndex>::highest());
+    }
 
     /** Copy constructor. */
     template<typename OtherDerived>
@@ -346,7 +347,7 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
       * array's size.
       */
     template<typename Other>
-    explicit inline PermutationMatrix(const MatrixBase<Other>& a_indices) : m_indices(a_indices)
+    explicit inline PermutationMatrix(const MatrixBase<Other>& indices) : m_indices(indices)
     {}
 
     /** Convert the Transpositions \a tr to a permutation matrix */
@@ -393,10 +394,13 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename Other>
-    PermutationMatrix(const Transpose<PermutationBase<Other> >& other)
-      : m_indices(other.nestedPermutation().size())
+    PermutationMatrix(const InverseImpl<Other,PermutationStorage>& other)
+      : m_indices(other.derived().nestedExpression().size())
     {
-      for (int i=0; i<m_indices.size();++i) m_indices.coeffRef(other.nestedPermutation().indices().coeff(i)) = i;
+      eigen_internal_assert(m_indices.size() <= NumTraits<StorageIndex>::highest());
+      StorageIndex end = StorageIndex(m_indices.size());
+      for (StorageIndex i=0; i<end;++i)
+        m_indices.coeffRef(other.derived().nestedExpression().indices().coeff(i)) = i;
     }
     template<typename Lhs,typename Rhs>
     PermutationMatrix(internal::PermPermProduct_t, const Lhs& lhs, const Rhs& rhs)
@@ -413,18 +417,20 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
 
 
 namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int _PacketAccess>
-struct traits<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,_PacketAccess> >
- : traits<Matrix<IndexType,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
+struct traits<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
+ : traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
 {
-  typedef IndexType Index;
-  typedef Map<const Matrix<IndexType, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1>, _PacketAccess> IndicesType;
+  typedef PermutationStorage StorageKind;
+  typedef Map<const Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1>, _PacketAccess> IndicesType;
+  typedef _StorageIndex StorageIndex;
+  typedef void Scalar;
 };
 }
 
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int _PacketAccess>
-class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,_PacketAccess>
-  : public PermutationBase<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,_PacketAccess> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
+class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess>
+  : public PermutationBase<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
 {
     typedef PermutationBase<Map> Base;
     typedef internal::traits<Map> Traits;
@@ -432,14 +438,14 @@ class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
     #endif
 
-    inline Map(const Index* indicesPtr)
+    inline Map(const StorageIndex* indicesPtr)
       : m_indices(indicesPtr)
     {}
 
-    inline Map(const Index* indicesPtr, Index size)
+    inline Map(const StorageIndex* indicesPtr, Index size)
       : m_indices(indicesPtr,size)
     {}
 
@@ -474,40 +480,36 @@ class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,
     IndicesType m_indices;
 };
 
-/** \class PermutationWrapper
-  * \ingroup Core_Module
-  *
-  * \brief Class to view a vector of integers as a permutation matrix
-  *
-  * \param _IndicesType the type of the vector of integer (can be any compatible expression)
-  *
-  * This class allows to view any vector expression of integers as a permutation matrix.
-  *
-  * \sa class PermutationBase, class PermutationMatrix
-  */
-
-struct PermutationStorage {};
-
 template<typename _IndicesType> class TranspositionsWrapper;
 namespace internal {
 template<typename _IndicesType>
 struct traits<PermutationWrapper<_IndicesType> >
 {
   typedef PermutationStorage StorageKind;
-  typedef typename _IndicesType::Scalar Scalar;
-  typedef typename _IndicesType::Scalar Index;
+  typedef void Scalar;
+  typedef typename _IndicesType::Scalar StorageIndex;
   typedef _IndicesType IndicesType;
   enum {
     RowsAtCompileTime = _IndicesType::SizeAtCompileTime,
     ColsAtCompileTime = _IndicesType::SizeAtCompileTime,
-    MaxRowsAtCompileTime = IndicesType::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = IndicesType::MaxColsAtCompileTime,
-    Flags = 0,
-    CoeffReadCost = _IndicesType::CoeffReadCost
+    MaxRowsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
+    MaxColsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
+    Flags = 0
   };
 };
 }
 
+/** \class PermutationWrapper
+  * \ingroup Core_Module
+  *
+  * \brief Class to view a vector of integers as a permutation matrix
+  *
+  * \tparam _IndicesType the type of the vector of integer (can be any compatible expression)
+  *
+  * This class allows to view any vector expression of integers as a permutation matrix.
+  *
+  * \sa class PermutationBase, class PermutationMatrix
+  */
 template<typename _IndicesType>
 class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesType> >
 {
@@ -519,8 +521,8 @@ class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesTyp
     typedef typename Traits::IndicesType IndicesType;
     #endif
 
-    inline PermutationWrapper(const IndicesType& a_indices)
-      : m_indices(a_indices)
+    inline PermutationWrapper(const IndicesType& indices)
+      : m_indices(indices)
     {}
 
     /** const version of indices(). */
@@ -532,182 +534,86 @@ class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesTyp
     typename IndicesType::Nested m_indices;
 };
 
+
 /** \returns the matrix with the permutation applied to the columns.
   */
-template<typename Derived, typename PermutationDerived>
-inline const internal::permut_matrix_product_retval<PermutationDerived, Derived, OnTheRight>
-operator*(const MatrixBase<Derived>& matrix,
-          const PermutationBase<PermutationDerived> &permutation)
+template<typename MatrixDerived, typename PermutationDerived>
+EIGEN_DEVICE_FUNC
+const Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
+operator*(const MatrixBase<MatrixDerived> &matrix,
+          const PermutationBase<PermutationDerived>& permutation)
 {
-  return internal::permut_matrix_product_retval
-           <PermutationDerived, Derived, OnTheRight>
-           (permutation.derived(), matrix.derived());
+  return Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
+            (matrix.derived(), permutation.derived());
 }
 
 /** \returns the matrix with the permutation applied to the rows.
   */
-template<typename Derived, typename PermutationDerived>
-inline const internal::permut_matrix_product_retval
-               <PermutationDerived, Derived, OnTheLeft>
+template<typename PermutationDerived, typename MatrixDerived>
+EIGEN_DEVICE_FUNC
+const Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
 operator*(const PermutationBase<PermutationDerived> &permutation,
-          const MatrixBase<Derived>& matrix)
+          const MatrixBase<MatrixDerived>& matrix)
 {
-  return internal::permut_matrix_product_retval
-           <PermutationDerived, Derived, OnTheLeft>
-           (permutation.derived(), matrix.derived());
+  return Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
+            (permutation.derived(), matrix.derived());
 }
 
-namespace internal {
 
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct traits<permut_matrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
+template<typename PermutationType>
+class InverseImpl<PermutationType, PermutationStorage>
+  : public EigenBase<Inverse<PermutationType> >
 {
-  typedef typename MatrixType::PlainObject ReturnType;
-};
-
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct permut_matrix_product_retval
- : public ReturnByValue<permut_matrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
-{
-    typedef typename remove_all<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
-    typedef typename MatrixType::Index Index;
-
-    permut_matrix_product_retval(const PermutationType& perm, const MatrixType& matrix)
-      : m_permutation(perm), m_matrix(matrix)
-    {}
-
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
-
-    template<typename Dest> inline void evalTo(Dest& dst) const
-    {
-      const Index n = Side==OnTheLeft ? rows() : cols();
-      // FIXME we need an is_same for expression that is not sensitive to constness. For instance
-      // is_same_xpr<Block<const Matrix>, Block<Matrix> >::value should be true.
-      if(    is_same<MatrixTypeNestedCleaned,Dest>::value
-          && blas_traits<MatrixTypeNestedCleaned>::HasUsableDirectAccess
-          && blas_traits<Dest>::HasUsableDirectAccess
-          && extract_data(dst) == extract_data(m_matrix))
-      {
-        // apply the permutation inplace
-        Matrix<bool,PermutationType::RowsAtCompileTime,1,0,PermutationType::MaxRowsAtCompileTime> mask(m_permutation.size());
-        mask.fill(false);
-        Index r = 0;
-        while(r < m_permutation.size())
-        {
-          // search for the next seed
-          while(r<m_permutation.size() && mask[r]) r++;
-          if(r>=m_permutation.size())
-            break;
-          // we got one, let's follow it until we are back to the seed
-          Index k0 = r++;
-          Index kPrev = k0;
-          mask.coeffRef(k0) = true;
-          for(Index k=m_permutation.indices().coeff(k0); k!=k0; k=m_permutation.indices().coeff(k))
-          {
-                  Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>(dst, k)
-            .swap(Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
-                       (dst,((Side==OnTheLeft) ^ Transposed) ? k0 : kPrev));
-
-            mask.coeffRef(k) = true;
-            kPrev = k;
-          }
-        }
-      }
-      else
-      {
-        for(int i = 0; i < n; ++i)
-        {
-          Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
-               (dst, ((Side==OnTheLeft) ^ Transposed) ? m_permutation.indices().coeff(i) : i)
-
-          =
-
-          Block<const MatrixTypeNestedCleaned,Side==OnTheLeft ? 1 : MatrixType::RowsAtCompileTime,Side==OnTheRight ? 1 : MatrixType::ColsAtCompileTime>
-               (m_matrix, ((Side==OnTheRight) ^ Transposed) ? m_permutation.indices().coeff(i) : i);
-        }
-      }
-    }
-
-  protected:
-    const PermutationType& m_permutation;
-    typename MatrixType::Nested m_matrix;
-};
-
-/* Template partial specialization for transposed/inverse permutations */
-
-template<typename Derived>
-struct traits<Transpose<PermutationBase<Derived> > >
- : traits<Derived>
-{};
-
-} // end namespace internal
-
-template<typename Derived>
-class Transpose<PermutationBase<Derived> >
-  : public EigenBase<Transpose<PermutationBase<Derived> > >
-{
-    typedef Derived PermutationType;
-    typedef typename PermutationType::IndicesType IndicesType;
     typedef typename PermutationType::PlainPermutationType PlainPermutationType;
+    typedef internal::traits<PermutationType> PermTraits;
+  protected:
+    InverseImpl() {}
   public:
+    typedef Inverse<PermutationType> InverseType;
+    using EigenBase<Inverse<PermutationType> >::derived;
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
-    typedef internal::traits<PermutationType> Traits;
-    typedef typename Derived::DenseMatrixType DenseMatrixType;
+    typedef typename PermutationType::DenseMatrixType DenseMatrixType;
     enum {
-      Flags = Traits::Flags,
-      CoeffReadCost = Traits::CoeffReadCost,
-      RowsAtCompileTime = Traits::RowsAtCompileTime,
-      ColsAtCompileTime = Traits::ColsAtCompileTime,
-      MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
-      MaxColsAtCompileTime = Traits::MaxColsAtCompileTime
+      RowsAtCompileTime = PermTraits::RowsAtCompileTime,
+      ColsAtCompileTime = PermTraits::ColsAtCompileTime,
+      MaxRowsAtCompileTime = PermTraits::MaxRowsAtCompileTime,
+      MaxColsAtCompileTime = PermTraits::MaxColsAtCompileTime
     };
-    typedef typename Traits::Scalar Scalar;
     #endif
 
-    Transpose(const PermutationType& p) : m_permutation(p) {}
-
-    inline int rows() const { return m_permutation.rows(); }
-    inline int cols() const { return m_permutation.cols(); }
-
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename DenseDerived>
     void evalTo(MatrixBase<DenseDerived>& other) const
     {
       other.setZero();
-      for (int i=0; i<rows();++i)
-        other.coeffRef(i, m_permutation.indices().coeff(i)) = typename DenseDerived::Scalar(1);
+      for (Index i=0; i<derived().rows();++i)
+        other.coeffRef(i, derived().nestedExpression().indices().coeff(i)) = typename DenseDerived::Scalar(1);
     }
     #endif
 
     /** \return the equivalent permutation matrix */
-    PlainPermutationType eval() const { return *this; }
+    PlainPermutationType eval() const { return derived(); }
 
-    DenseMatrixType toDenseMatrix() const { return *this; }
+    DenseMatrixType toDenseMatrix() const { return derived(); }
 
     /** \returns the matrix with the inverse permutation applied to the columns.
       */
     template<typename OtherDerived> friend
-    inline const internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheRight, true>
-    operator*(const MatrixBase<OtherDerived>& matrix, const Transpose& trPerm)
+    const Product<OtherDerived, InverseType, AliasFreeProduct>
+    operator*(const MatrixBase<OtherDerived>& matrix, const InverseType& trPerm)
     {
-      return internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheRight, true>(trPerm.m_permutation, matrix.derived());
+      return Product<OtherDerived, InverseType, AliasFreeProduct>(matrix.derived(), trPerm.derived());
     }
 
     /** \returns the matrix with the inverse permutation applied to the rows.
       */
     template<typename OtherDerived>
-    inline const internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheLeft, true>
+    const Product<InverseType, OtherDerived, AliasFreeProduct>
     operator*(const MatrixBase<OtherDerived>& matrix) const
     {
-      return internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheLeft, true>(m_permutation, matrix.derived());
+      return Product<InverseType, OtherDerived, AliasFreeProduct>(derived(), matrix.derived());
     }
-
-    const PermutationType& nestedPermutation() const { return m_permutation; }
-
-  protected:
-    const PermutationType& m_permutation;
 };
 
 template<typename Derived>
@@ -716,6 +622,12 @@ const PermutationWrapper<const Derived> MatrixBase<Derived>::asPermutation() con
   return derived();
 }
 
+namespace internal {
+
+template<> struct AssignmentKind<DenseShape,PermutationShape> { typedef EigenBase2EigenBase Kind; };
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_PERMUTATIONMATRIX_H
diff --git a/extern/Eigen3/Eigen/src/Core/PlainObjectBase.h b/extern/Eigen3/Eigen/src/Core/PlainObjectBase.h
index a4e4af4a7b2..1dc7e223af4 100644
--- a/extern/Eigen3/Eigen/src/Core/PlainObjectBase.h
+++ b/extern/Eigen3/Eigen/src/Core/PlainObjectBase.h
@@ -28,6 +28,7 @@ namespace internal {
 
 template<int MaxSizeAtCompileTime> struct check_rows_cols_for_overflow {
   template<typename Index>
+  EIGEN_DEVICE_FUNC
   static EIGEN_ALWAYS_INLINE void run(Index, Index)
   {
   }
@@ -35,11 +36,12 @@ template<int MaxSizeAtCompileTime> struct check_rows_cols_for_overflow {
 
 template<> struct check_rows_cols_for_overflow<Dynamic> {
   template<typename Index>
+  EIGEN_DEVICE_FUNC
   static EIGEN_ALWAYS_INLINE void run(Index rows, Index cols)
   {
     // http://hg.mozilla.org/mozilla-central/file/6c8a909977d3/xpcom/ds/CheckedInt.h#l242
     // we assume Index is signed
-    Index max_index = (size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed
+    Index max_index = (std::size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed
     bool error = (rows == 0 || cols == 0) ? false
                : (rows > max_index / cols);
     if (error)
@@ -56,33 +58,41 @@ template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers> struct m
 
 } // end namespace internal
 
-/** \class PlainObjectBase
-  * \brief %Dense storage base class for matrices and arrays.
-  *
-  * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_PLAINOBJECTBASE_PLUGIN.
-  *
-  * \sa \ref TopicClassHierarchy
-  */
 #ifdef EIGEN_PARSED_BY_DOXYGEN
-namespace internal {
+namespace doxygen {
 
-// this is a warkaround to doxygen not being able to understand the inheritence logic
+// This is a workaround to doxygen not being able to understand the inheritance logic
 // when it is hidden by the dense_xpr_base helper struct.
-template<typename Derived> struct dense_xpr_base_dispatcher_for_doxygen;// : public MatrixBase<Derived> {};
+// Moreover, doxygen fails to include members that are not documented in the declaration body of
+// MatrixBase if we inherits MatrixBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >,
+// this is why we simply inherits MatrixBase, though this does not make sense.
+
+/** This class is just a workaround for Doxygen and it does not not actually exist. */
+template<typename Derived> struct dense_xpr_base_dispatcher;
 /** This class is just a workaround for Doxygen and it does not not actually exist. */
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct dense_xpr_base_dispatcher_for_doxygen<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
-    : public MatrixBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > {};
+struct dense_xpr_base_dispatcher<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+    : public MatrixBase {};
 /** This class is just a workaround for Doxygen and it does not not actually exist. */
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct dense_xpr_base_dispatcher_for_doxygen<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
-    : public ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > {};
+struct dense_xpr_base_dispatcher<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+    : public ArrayBase {};
 
-} // namespace internal
+} // namespace doxygen
 
+/** \class PlainObjectBase
+  * \ingroup Core_Module
+  * \brief %Dense storage base class for matrices and arrays.
+  *
+  * This class can be extended with the help of the plugin mechanism described on the page
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_PLAINOBJECTBASE_PLUGIN.
+  *
+  * \tparam Derived is the derived type, e.g., a Matrix or Array
+  *
+  * \sa \ref TopicClassHierarchy
+  */
 template<typename Derived>
-class PlainObjectBase : public internal::dense_xpr_base_dispatcher_for_doxygen<Derived>
+class PlainObjectBase : public doxygen::dense_xpr_base_dispatcher<Derived>
 #else
 template<typename Derived>
 class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
@@ -93,8 +103,8 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     typedef typename internal::dense_xpr_base<Derived>::type Base;
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
+    
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
     typedef Derived DenseType;
@@ -113,28 +123,40 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     typedef Eigen::Map<Derived, Unaligned>  MapType;
     friend  class Eigen::Map<const Derived, Unaligned>;
     typedef const Eigen::Map<const Derived, Unaligned> ConstMapType;
-    friend  class Eigen::Map<Derived, Aligned>;
-    typedef Eigen::Map<Derived, Aligned> AlignedMapType;
-    friend  class Eigen::Map<const Derived, Aligned>;
-    typedef const Eigen::Map<const Derived, Aligned> ConstAlignedMapType;
+#if EIGEN_MAX_ALIGN_BYTES>0
+    // for EIGEN_MAX_ALIGN_BYTES==0, AlignedMax==Unaligned, and many compilers generate warnings for friend-ing a class twice.
+    friend  class Eigen::Map<Derived, AlignedMax>;
+    friend  class Eigen::Map<const Derived, AlignedMax>;
+#endif
+    typedef Eigen::Map<Derived, AlignedMax> AlignedMapType;
+    typedef const Eigen::Map<const Derived, AlignedMax> ConstAlignedMapType;
     template<typename StrideType> struct StridedMapType { typedef Eigen::Map<Derived, Unaligned, StrideType> type; };
     template<typename StrideType> struct StridedConstMapType { typedef Eigen::Map<const Derived, Unaligned, StrideType> type; };
-    template<typename StrideType> struct StridedAlignedMapType { typedef Eigen::Map<Derived, Aligned, StrideType> type; };
-    template<typename StrideType> struct StridedConstAlignedMapType { typedef Eigen::Map<const Derived, Aligned, StrideType> type; };
+    template<typename StrideType> struct StridedAlignedMapType { typedef Eigen::Map<Derived, AlignedMax, StrideType> type; };
+    template<typename StrideType> struct StridedConstAlignedMapType { typedef Eigen::Map<const Derived, AlignedMax, StrideType> type; };
 
   protected:
     DenseStorage<Scalar, Base::MaxSizeAtCompileTime, Base::RowsAtCompileTime, Base::ColsAtCompileTime, Options> m_storage;
 
   public:
-    enum { NeedsToAlign = SizeAtCompileTime != Dynamic && (internal::traits<Derived>::Flags & AlignedBit) != 0 };
+    enum { NeedsToAlign = (SizeAtCompileTime != Dynamic) && (internal::traits<Derived>::Alignment>0) };
     EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
 
+    EIGEN_DEVICE_FUNC
     Base& base() { return *static_cast<Base*>(this); }
+    EIGEN_DEVICE_FUNC
     const Base& base() const { return *static_cast<const Base*>(this); }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rows() const { return m_storage.rows(); }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index cols() const { return m_storage.cols(); }
 
+    /** This is an overloaded version of DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index,Index) const
+      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
+      *
+      * See DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index) const for details. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeff(Index rowId, Index colId) const
     {
       if(Flags & RowMajorBit)
@@ -143,11 +165,21 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
         return m_storage.data()[rowId + colId * m_storage.rows()];
     }
 
+    /** This is an overloaded version of DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index) const
+      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
+      *
+      * See DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index) const for details. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeff(Index index) const
     {
       return m_storage.data()[index];
     }
 
+    /** This is an overloaded version of DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index,Index) const
+      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
+      *
+      * See DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index,Index) const for details. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar& coeffRef(Index rowId, Index colId)
     {
       if(Flags & RowMajorBit)
@@ -156,11 +188,19 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
         return m_storage.data()[rowId + colId * m_storage.rows()];
     }
 
+    /** This is an overloaded version of DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index) const
+      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
+      *
+      * See DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index) const for details. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
     {
       return m_storage.data()[index];
     }
 
+    /** This is the const version of coeffRef(Index,Index) which is thus synonym of coeff(Index,Index).
+      * It is provided for convenience. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeffRef(Index rowId, Index colId) const
     {
       if(Flags & RowMajorBit)
@@ -169,6 +209,9 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
         return m_storage.data()[rowId + colId * m_storage.rows()];
     }
 
+    /** This is the const version of coeffRef(Index) which is thus synonym of coeff(Index).
+      * It is provided for convenience. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeffRef(Index index) const
     {
       return m_storage.data()[index];
@@ -209,11 +252,11 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     }
 
     /** \returns a const pointer to the data array of this matrix */
-    EIGEN_STRONG_INLINE const Scalar *data() const
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar *data() const
     { return m_storage.data(); }
 
     /** \returns a pointer to the data array of this matrix */
-    EIGEN_STRONG_INLINE Scalar *data()
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar *data()
     { return m_storage.data(); }
 
     /** Resizes \c *this to a \a rows x \a cols matrix.
@@ -232,22 +275,22 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \sa resize(Index) for vectors, resize(NoChange_t, Index), resize(Index, NoChange_t)
       */
-    EIGEN_STRONG_INLINE void resize(Index nbRows, Index nbCols)
-    {
-      eigen_assert(   EIGEN_IMPLIES(RowsAtCompileTime!=Dynamic,nbRows==RowsAtCompileTime)
-                   && EIGEN_IMPLIES(ColsAtCompileTime!=Dynamic,nbCols==ColsAtCompileTime)
-                   && EIGEN_IMPLIES(RowsAtCompileTime==Dynamic && MaxRowsAtCompileTime!=Dynamic,nbRows<=MaxRowsAtCompileTime)
-                   && EIGEN_IMPLIES(ColsAtCompileTime==Dynamic && MaxColsAtCompileTime!=Dynamic,nbCols<=MaxColsAtCompileTime)
-                   && nbRows>=0 && nbCols>=0 && "Invalid sizes when resizing a matrix or array.");
-      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(nbRows, nbCols);
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void resize(Index rows, Index cols)
+    {
+      eigen_assert(   EIGEN_IMPLIES(RowsAtCompileTime!=Dynamic,rows==RowsAtCompileTime)
+                   && EIGEN_IMPLIES(ColsAtCompileTime!=Dynamic,cols==ColsAtCompileTime)
+                   && EIGEN_IMPLIES(RowsAtCompileTime==Dynamic && MaxRowsAtCompileTime!=Dynamic,rows<=MaxRowsAtCompileTime)
+                   && EIGEN_IMPLIES(ColsAtCompileTime==Dynamic && MaxColsAtCompileTime!=Dynamic,cols<=MaxColsAtCompileTime)
+                   && rows>=0 && cols>=0 && "Invalid sizes when resizing a matrix or array.");
+      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(rows, cols);
       #ifdef EIGEN_INITIALIZE_COEFFS
-        Index size = nbRows*nbCols;
+        Index size = rows*cols;
         bool size_changed = size != this->size();
-        m_storage.resize(size, nbRows, nbCols);
+        m_storage.resize(size, rows, cols);
         if(size_changed) EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
       #else
-        internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(nbRows, nbCols);
-        m_storage.resize(nbRows*nbCols, nbRows, nbCols);
+        m_storage.resize(rows*cols, rows, cols);
       #endif
     }
 
@@ -262,6 +305,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \sa resize(Index,Index), resize(NoChange_t, Index), resize(Index, NoChange_t)
       */
+    EIGEN_DEVICE_FUNC
     inline void resize(Index size)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(PlainObjectBase)
@@ -286,9 +330,10 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \sa resize(Index,Index)
       */
-    inline void resize(NoChange_t, Index nbCols)
+    EIGEN_DEVICE_FUNC
+    inline void resize(NoChange_t, Index cols)
     {
-      resize(rows(), nbCols);
+      resize(rows(), cols);
     }
 
     /** Resizes the matrix, changing only the number of rows. For the parameter of type NoChange_t, just pass the special value \c NoChange
@@ -299,9 +344,10 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \sa resize(Index,Index)
       */
-    inline void resize(Index nbRows, NoChange_t)
+    EIGEN_DEVICE_FUNC
+    inline void resize(Index rows, NoChange_t)
     {
-      resize(nbRows, cols());
+      resize(rows, cols());
     }
 
     /** Resizes \c *this to have the same dimensions as \a other.
@@ -312,6 +358,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE void resizeLike(const EigenBase<OtherDerived>& _other)
     {
       const OtherDerived& other = _other.derived();
@@ -339,9 +386,10 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * Matrices are resized relative to the top-left element. In case values need to be 
       * appended to the matrix they will be uninitialized.
       */
-    EIGEN_STRONG_INLINE void conservativeResize(Index nbRows, Index nbCols)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(Index rows, Index cols)
     {
-      internal::conservative_resize_like_impl<Derived>::run(*this, nbRows, nbCols);
+      internal::conservative_resize_like_impl<Derived>::run(*this, rows, cols);
     }
 
     /** Resizes the matrix to \a rows x \a cols while leaving old values untouched.
@@ -351,10 +399,11 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * In case the matrix is growing, new rows will be uninitialized.
       */
-    EIGEN_STRONG_INLINE void conservativeResize(Index nbRows, NoChange_t)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(Index rows, NoChange_t)
     {
       // Note: see the comment in conservativeResize(Index,Index)
-      conservativeResize(nbRows, cols());
+      conservativeResize(rows, cols());
     }
 
     /** Resizes the matrix to \a rows x \a cols while leaving old values untouched.
@@ -364,10 +413,11 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * In case the matrix is growing, new columns will be uninitialized.
       */
-    EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index nbCols)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index cols)
     {
       // Note: see the comment in conservativeResize(Index,Index)
-      conservativeResize(rows(), nbCols);
+      conservativeResize(rows(), cols);
     }
 
     /** Resizes the vector to \a size while retaining old values.
@@ -378,6 +428,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * When values are appended, they will be uninitialized.
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void conservativeResize(Index size)
     {
       internal::conservative_resize_like_impl<Derived>::run(*this, size);
@@ -393,6 +444,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * appended to the matrix they will copied from \c other.
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void conservativeResizeLike(const DenseBase<OtherDerived>& other)
     {
       internal::conservative_resize_like_impl<Derived,OtherDerived>::run(*this, other);
@@ -401,6 +453,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Derived& operator=(const PlainObjectBase& other)
     {
       return _set(other);
@@ -408,6 +461,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
 
     /** \sa MatrixBase::lazyAssign() */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Derived& lazyAssign(const DenseBase<OtherDerived>& other)
     {
       _resize_to_match(other);
@@ -415,12 +469,18 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     }
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Derived& operator=(const ReturnByValue<OtherDerived>& func)
     {
       resize(func.rows(), func.cols());
       return Base::operator=(func);
     }
 
+    // Prevent user from trying to instantiate PlainObjectBase objects
+    // by making all its constructor protected. See bug 1074.
+  protected:
+
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE PlainObjectBase() : m_storage()
     {
 //       _check_template_params();
@@ -430,20 +490,23 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     // FIXME is it still needed ?
     /** \internal */
-    PlainObjectBase(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC
+    explicit PlainObjectBase(internal::constructor_without_unaligned_array_assert)
       : m_storage(internal::constructor_without_unaligned_array_assert())
     {
 //       _check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 #endif
 
-#ifdef EIGEN_HAVE_RVALUE_REFERENCES
-    PlainObjectBase(PlainObjectBase&& other)
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    PlainObjectBase(PlainObjectBase&& other) EIGEN_NOEXCEPT
       : m_storage( std::move(other.m_storage) )
     {
     }
 
-    PlainObjectBase& operator=(PlainObjectBase&& other)
+    EIGEN_DEVICE_FUNC
+    PlainObjectBase& operator=(PlainObjectBase&& other) EIGEN_NOEXCEPT
     {
       using std::swap;
       swap(m_storage, other.m_storage);
@@ -452,31 +515,56 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
 #endif
 
     /** Copy constructor */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE PlainObjectBase(const PlainObjectBase& other)
-      : m_storage()
+      : Base(), m_storage(other.m_storage) { }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(Index size, Index rows, Index cols)
+      : m_storage(size, rows, cols)
     {
-      _check_template_params();
-      lazyAssign(other);
+//       _check_template_params();
+//       EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 
+    /** \sa PlainObjectBase::operator=(const EigenBase<OtherDerived>&) */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE PlainObjectBase(const DenseBase<OtherDerived> &other)
       : m_storage()
     {
       _check_template_params();
-      lazyAssign(other);
+      resizeLike(other);
+      _set_noalias(other);
     }
 
-    EIGEN_STRONG_INLINE PlainObjectBase(Index a_size, Index nbRows, Index nbCols)
-      : m_storage(a_size, nbRows, nbCols)
+    /** \sa PlainObjectBase::operator=(const EigenBase<OtherDerived>&) */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const EigenBase<OtherDerived> &other)
+      : m_storage()
     {
-//       _check_template_params();
-//       EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+      _check_template_params();
+      resizeLike(other);
+      *this = other.derived();
+    }
+    /** \brief Copy constructor with in-place evaluation */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const ReturnByValue<OtherDerived>& other)
+    {
+      _check_template_params();
+      // FIXME this does not automatically transpose vectors if necessary
+      resize(other.rows(), other.cols());
+      other.evalTo(this->derived());
     }
 
-    /** \copydoc MatrixBase::operator=(const EigenBase<OtherDerived>&)
+  public:
+
+    /** \brief Copies the generic expression \a other into *this.
+      * \copydetails DenseBase::operator=(const EigenBase<OtherDerived> &other)
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE Derived& operator=(const EigenBase<OtherDerived> &other)
     {
       _resize_to_match(other);
@@ -484,21 +572,15 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       return this->derived();
     }
 
-    /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE PlainObjectBase(const EigenBase<OtherDerived> &other)
-      : m_storage(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
-    {
-      _check_template_params();
-      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(other.derived().rows(), other.derived().cols());
-      Base::operator=(other.derived());
-    }
-
     /** \name Map
       * These are convenience functions returning Map objects. The Map() static functions return unaligned Map objects,
       * while the AlignedMap() functions return aligned Map objects and thus should be called only with 16-byte-aligned
       * \a data pointers.
       *
+      * Here is an example using strides:
+      * \include Matrix_Map_stride.cpp
+      * Output: \verbinclude Matrix_Map_stride.out
+      *
       * \see class Map
       */
     //@{
@@ -568,16 +650,16 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     //@}
 
     using Base::setConstant;
-    Derived& setConstant(Index size, const Scalar& value);
-    Derived& setConstant(Index rows, Index cols, const Scalar& value);
+    EIGEN_DEVICE_FUNC Derived& setConstant(Index size, const Scalar& val);
+    EIGEN_DEVICE_FUNC Derived& setConstant(Index rows, Index cols, const Scalar& val);
 
     using Base::setZero;
-    Derived& setZero(Index size);
-    Derived& setZero(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC Derived& setZero(Index size);
+    EIGEN_DEVICE_FUNC Derived& setZero(Index rows, Index cols);
 
     using Base::setOnes;
-    Derived& setOnes(Index size);
-    Derived& setOnes(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC Derived& setOnes(Index size);
+    EIGEN_DEVICE_FUNC Derived& setOnes(Index rows, Index cols);
 
     using Base::setRandom;
     Derived& setRandom(Index size);
@@ -596,6 +678,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE void _resize_to_match(const EigenBase<OtherDerived>& other)
     {
       #ifdef EIGEN_NO_AUTOMATIC_RESIZING
@@ -603,8 +686,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
                  : (rows() == other.rows() && cols() == other.cols())))
         && "Size mismatch. Automatic resizing is disabled because EIGEN_NO_AUTOMATIC_RESIZING is defined");
       EIGEN_ONLY_USED_FOR_DEBUG(other);
-      if(this->size()==0)
-        resizeLike(other);
       #else
       resizeLike(other);
       #endif
@@ -624,25 +705,23 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \internal
       */
+    // aliasing is dealt once in internall::call_assignment
+    // so at this stage we have to assume aliasing... and resising has to be done later.
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE Derived& _set(const DenseBase<OtherDerived>& other)
     {
-      _set_selector(other.derived(), typename internal::conditional<static_cast<bool>(int(OtherDerived::Flags) & EvalBeforeAssigningBit), internal::true_type, internal::false_type>::type());
+      internal::call_assignment(this->derived(), other.derived());
       return this->derived();
     }
 
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::true_type&) { _set_noalias(other.eval()); }
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::false_type&) { _set_noalias(other); }
-
     /** \internal Like _set() but additionally makes the assumption that no aliasing effect can happen (which
       * is the case when creating a new matrix) so one can enforce lazy evaluation.
       *
       * \sa operator=(const MatrixBase<OtherDerived>&), _set()
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE Derived& _set_noalias(const DenseBase<OtherDerived>& other)
     {
       // I don't think we need this resize call since the lazyAssign will anyways resize
@@ -650,40 +729,175 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       //_resize_to_match(other);
       // the 'false' below means to enforce lazy evaluation. We don't use lazyAssign() because
       // it wouldn't allow to copy a row-vector into a column-vector.
-      return internal::assign_selector<Derived,OtherDerived,false>::run(this->derived(), other.derived());
+      internal::call_assignment_no_alias(this->derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+      return this->derived();
     }
 
     template<typename T0, typename T1>
-    EIGEN_STRONG_INLINE void _init2(Index nbRows, Index nbCols, typename internal::enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init2(Index rows, Index cols, typename internal::enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
     {
       EIGEN_STATIC_ASSERT(bool(NumTraits<T0>::IsInteger) &&
                           bool(NumTraits<T1>::IsInteger),
                           FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
-      resize(nbRows,nbCols);
+      resize(rows,cols);
+    }
+    
+    template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC 
+    EIGEN_STRONG_INLINE void _init2(const T0& val0, const T1& val1, typename internal::enable_if<Base::SizeAtCompileTime==2,T0>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2)
+      m_storage.data()[0] = Scalar(val0);
+      m_storage.data()[1] = Scalar(val1);
     }
+    
     template<typename T0, typename T1>
-    EIGEN_STRONG_INLINE void _init2(const Scalar& val0, const Scalar& val1, typename internal::enable_if<Base::SizeAtCompileTime==2,T0>::type* = 0)
+    EIGEN_DEVICE_FUNC 
+    EIGEN_STRONG_INLINE void _init2(const Index& val0, const Index& val1,
+                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
+                                                                  && (internal::is_same<T0,Index>::value)
+                                                                  && (internal::is_same<T1,Index>::value)
+                                                                  && Base::SizeAtCompileTime==2,T1>::type* = 0)
     {
       EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2)
+      m_storage.data()[0] = Scalar(val0);
+      m_storage.data()[1] = Scalar(val1);
+    }
+
+    // The argument is convertible to the Index type and we either have a non 1x1 Matrix, or a dynamic-sized Array,
+    // then the argument is meant to be the size of the object.
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(Index size, typename internal::enable_if<    (Base::SizeAtCompileTime!=1 || !internal::is_convertible<T, Scalar>::value)
+                                                                              && ((!internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value || Base::SizeAtCompileTime==Dynamic)),T>::type* = 0)
+    {
+      // NOTE MSVC 2008 complains if we directly put bool(NumTraits<T>::IsInteger) as the EIGEN_STATIC_ASSERT argument.
+      const bool is_integer = NumTraits<T>::IsInteger;
+      EIGEN_UNUSED_VARIABLE(is_integer);
+      EIGEN_STATIC_ASSERT(is_integer,
+                          FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
+      resize(size);
+    }
+    
+    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type can be implicitely converted)
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Scalar& val0, typename internal::enable_if<Base::SizeAtCompileTime==1 && internal::is_convertible<T, Scalar>::value,T>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1)
       m_storage.data()[0] = val0;
-      m_storage.data()[1] = val1;
+    }
+    
+    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type match the index type)
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Index& val0,
+                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
+                                                                  && (internal::is_same<Index,T>::value)
+                                                                  && Base::SizeAtCompileTime==1
+                                                                  && internal::is_convertible<T, Scalar>::value,T*>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1)
+      m_storage.data()[0] = Scalar(val0);
     }
 
+    // Initialize a fixed size matrix from a pointer to raw data
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Scalar* data){
+      this->_set_noalias(ConstMapType(data));
+    }
+
+    // Initialize an arbitrary matrix from a dense expression
+    template<typename T, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const DenseBase<OtherDerived>& other){
+      this->_set_noalias(other);
+    }
+
+    // Initialize an arbitrary matrix from an object convertible to the Derived type.
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Derived& other){
+      this->_set_noalias(other);
+    }
+
+    // Initialize an arbitrary matrix from a generic Eigen expression
+    template<typename T, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const EigenBase<OtherDerived>& other){
+      this->derived() = other;
+    }
+
+    template<typename T, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const ReturnByValue<OtherDerived>& other)
+    {
+      resize(other.rows(), other.cols());
+      other.evalTo(this->derived());
+    }
+
+    template<typename T, typename OtherDerived, int ColsAtCompileTime>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
+    {
+      this->derived() = r;
+    }
+    
+    // For fixed-size Array<Scalar,...>
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Scalar& val0,
+                                    typename internal::enable_if<    Base::SizeAtCompileTime!=Dynamic
+                                                                  && Base::SizeAtCompileTime!=1
+                                                                  && internal::is_convertible<T, Scalar>::value
+                                                                  && internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T>::type* = 0)
+    {
+      Base::setConstant(val0);
+    }
+    
+    // For fixed-size Array<Index,...>
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Index& val0,
+                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
+                                                                  && (internal::is_same<Index,T>::value)
+                                                                  && Base::SizeAtCompileTime!=Dynamic
+                                                                  && Base::SizeAtCompileTime!=1
+                                                                  && internal::is_convertible<T, Scalar>::value
+                                                                  && internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T*>::type* = 0)
+    {
+      Base::setConstant(val0);
+    }
+    
     template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
     friend struct internal::matrix_swap_impl;
 
-    /** \internal generic implementation of swap for dense storage since for dynamic-sized matrices of same type it is enough to swap the
-      * data pointers.
+  public:
+    
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** \internal
+      * \brief Override DenseBase::swap() since for dynamic-sized matrices
+      * of same type it is enough to swap the data pointers.
       */
     template<typename OtherDerived>
-    void _swap(DenseBase<OtherDerived> const & other)
+    EIGEN_DEVICE_FUNC
+    void swap(DenseBase<OtherDerived> & other)
     {
       enum { SwapPointers = internal::is_same<Derived, OtherDerived>::value && Base::SizeAtCompileTime==Dynamic };
-      internal::matrix_swap_impl<Derived, OtherDerived, bool(SwapPointers)>::run(this->derived(), other.const_cast_derived());
+      internal::matrix_swap_impl<Derived, OtherDerived, bool(SwapPointers)>::run(this->derived(), other.derived());
     }
-
-  public:
-#ifndef EIGEN_PARSED_BY_DOXYGEN
+    
+    /** \internal
+      * \brief const version forwarded to DenseBase::swap
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void swap(DenseBase<OtherDerived> const & other)
+    { Base::swap(other.derived()); }
+    
+    EIGEN_DEVICE_FUNC 
     static EIGEN_STRONG_INLINE void _check_template_params()
     {
       EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, (Options&RowMajor)==RowMajor)
@@ -697,10 +911,9 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
                         && (Options & (DontAlign|RowMajor)) == Options),
         INVALID_MATRIX_TEMPLATE_PARAMETERS)
     }
-#endif
 
-private:
-    enum { ThisConstantIsPrivateInPlainObjectBase };
+    enum { IsPlainObjectBase = 1 };
+#endif
 };
 
 namespace internal {
@@ -708,7 +921,6 @@ namespace internal {
 template <typename Derived, typename OtherDerived, bool IsVector>
 struct conservative_resize_like_impl
 {
-  typedef typename Derived::Index Index;
   static void run(DenseBase<Derived>& _this, Index rows, Index cols)
   {
     if (_this.rows() == rows && _this.cols() == cols) return;
@@ -724,8 +936,8 @@ struct conservative_resize_like_impl
     {
       // The storage order does not allow us to use reallocation.
       typename Derived::PlainObject tmp(rows,cols);
-      const Index common_rows = (std::min)(rows, _this.rows());
-      const Index common_cols = (std::min)(cols, _this.cols());
+      const Index common_rows = numext::mini(rows, _this.rows());
+      const Index common_cols = numext::mini(cols, _this.cols());
       tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
       _this.derived().swap(tmp);
     }
@@ -758,8 +970,8 @@ struct conservative_resize_like_impl
     {
       // The storage order does not allow us to use reallocation.
       typename Derived::PlainObject tmp(other);
-      const Index common_rows = (std::min)(tmp.rows(), _this.rows());
-      const Index common_cols = (std::min)(tmp.cols(), _this.cols());
+      const Index common_rows = numext::mini(tmp.rows(), _this.rows());
+      const Index common_cols = numext::mini(tmp.cols(), _this.cols());
       tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
       _this.derived().swap(tmp);
     }
@@ -774,7 +986,6 @@ struct conservative_resize_like_impl<Derived,OtherDerived,true>
 {
   using conservative_resize_like_impl<Derived,OtherDerived,false>::run;
   
-  typedef typename Derived::Index Index;
   static void run(DenseBase<Derived>& _this, Index size)
   {
     const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : size;
@@ -800,6 +1011,7 @@ struct conservative_resize_like_impl<Derived,OtherDerived,true>
 template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
 struct matrix_swap_impl
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(MatrixTypeA& a, MatrixTypeB& b)
   {
     a.base().swap(b);
@@ -809,6 +1021,7 @@ struct matrix_swap_impl
 template<typename MatrixTypeA, typename MatrixTypeB>
 struct matrix_swap_impl<MatrixTypeA, MatrixTypeB, true>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(MatrixTypeA& a, MatrixTypeB& b)
   {
     static_cast<typename MatrixTypeA::Base&>(a).m_storage.swap(static_cast<typename MatrixTypeB::Base&>(b).m_storage);
diff --git a/extern/Eigen3/Eigen/src/Core/Product.h b/extern/Eigen3/Eigen/src/Core/Product.h
index 30aa8943b4c..676c4802779 100644
--- a/extern/Eigen3/Eigen/src/Core/Product.h
+++ b/extern/Eigen3/Eigen/src/Core/Product.h
@@ -3,96 +3,184 @@
 //
 // Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// This Source Code Form is subject to the terms of the Mozilla Public
-// License, v. 2.0. If a copy of the MPL was not distributed with this
-// file, You can obtain one at http://mozilla.org/MPL/2.0/.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_PRODUCT_H
 #define EIGEN_PRODUCT_H
 
-template<typename Lhs, typename Rhs> class Product;
-template<typename Lhs, typename Rhs, typename StorageKind> class ProductImpl;
+namespace Eigen {
 
-/** \class Product
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the product of two arbitrary matrices or vectors
-  *
-  * \param Lhs the type of the left-hand side expression
-  * \param Rhs the type of the right-hand side expression
-  *
-  * This class represents an expression of the product of two arbitrary matrices.
-  *
-  */
+template<typename Lhs, typename Rhs, int Option, typename StorageKind> class ProductImpl;
 
 namespace internal {
-template<typename Lhs, typename Rhs>
-struct traits<Product<Lhs, Rhs> >
+
+template<typename Lhs, typename Rhs, int Option>
+struct traits<Product<Lhs, Rhs, Option> >
 {
-  typedef MatrixXpr XprKind;
   typedef typename remove_all<Lhs>::type LhsCleaned;
   typedef typename remove_all<Rhs>::type RhsCleaned;
-  typedef typename scalar_product_traits<typename traits<LhsCleaned>::Scalar, typename traits<RhsCleaned>::Scalar>::ReturnType Scalar;
-  typedef typename promote_storage_type<typename traits<LhsCleaned>::StorageKind,
-                                        typename traits<RhsCleaned>::StorageKind>::ret StorageKind;
-  typedef typename promote_index_type<typename traits<LhsCleaned>::Index,
-                                      typename traits<RhsCleaned>::Index>::type Index;
+  typedef traits<LhsCleaned> LhsTraits;
+  typedef traits<RhsCleaned> RhsTraits;
+  
+  typedef MatrixXpr XprKind;
+  
+  typedef typename ScalarBinaryOpTraits<typename traits<LhsCleaned>::Scalar, typename traits<RhsCleaned>::Scalar>::ReturnType Scalar;
+  typedef typename product_promote_storage_type<typename LhsTraits::StorageKind,
+                                                typename RhsTraits::StorageKind,
+                                                internal::product_type<Lhs,Rhs>::ret>::ret StorageKind;
+  typedef typename promote_index_type<typename LhsTraits::StorageIndex,
+                                      typename RhsTraits::StorageIndex>::type StorageIndex;
+  
   enum {
-    RowsAtCompileTime = LhsCleaned::RowsAtCompileTime,
-    ColsAtCompileTime = RhsCleaned::ColsAtCompileTime,
-    MaxRowsAtCompileTime = LhsCleaned::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = RhsCleaned::MaxColsAtCompileTime,
-    Flags = (MaxRowsAtCompileTime==1 ? RowMajorBit : 0), // TODO should be no storage order
-    CoeffReadCost = 0 // TODO CoeffReadCost should not be part of the expression traits
+    RowsAtCompileTime    = LhsTraits::RowsAtCompileTime,
+    ColsAtCompileTime    = RhsTraits::ColsAtCompileTime,
+    MaxRowsAtCompileTime = LhsTraits::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = RhsTraits::MaxColsAtCompileTime,
+    
+    // FIXME: only needed by GeneralMatrixMatrixTriangular
+    InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsTraits::ColsAtCompileTime, RhsTraits::RowsAtCompileTime),
+    
+    // The storage order is somewhat arbitrary here. The correct one will be determined through the evaluator.
+    Flags = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? RowMajorBit
+          : (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0
+          : (   ((LhsTraits::Flags&NoPreferredStorageOrderBit) && (RhsTraits::Flags&RowMajorBit))
+             || ((RhsTraits::Flags&NoPreferredStorageOrderBit) && (LhsTraits::Flags&RowMajorBit)) ) ? RowMajorBit
+          : NoPreferredStorageOrderBit
   };
 };
-} // end namespace internal
 
+} // end namespace internal
 
-template<typename Lhs, typename Rhs>
-class Product : public ProductImpl<Lhs,Rhs,typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
-                                                                            typename internal::traits<Rhs>::StorageKind>::ret>
+/** \class Product
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the product of two arbitrary matrices or vectors
+  *
+  * \tparam _Lhs the type of the left-hand side expression
+  * \tparam _Rhs the type of the right-hand side expression
+  *
+  * This class represents an expression of the product of two arbitrary matrices.
+  *
+  * The other template parameters are:
+  * \tparam Option     can be DefaultProduct, AliasFreeProduct, or LazyProduct
+  *
+  */
+template<typename _Lhs, typename _Rhs, int Option>
+class Product : public ProductImpl<_Lhs,_Rhs,Option,
+                                   typename internal::product_promote_storage_type<typename internal::traits<_Lhs>::StorageKind,
+                                                                                   typename internal::traits<_Rhs>::StorageKind,
+                                                                                   internal::product_type<_Lhs,_Rhs>::ret>::ret>
 {
   public:
     
+    typedef _Lhs Lhs;
+    typedef _Rhs Rhs;
+    
     typedef typename ProductImpl<
-        Lhs, Rhs,
-        typename internal::promote_storage_type<typename Lhs::StorageKind,
-                                                typename Rhs::StorageKind>::ret>::Base Base;
+        Lhs, Rhs, Option,
+        typename internal::product_promote_storage_type<typename internal::traits<Lhs>::StorageKind,
+                                                        typename internal::traits<Rhs>::StorageKind,
+                                                        internal::product_type<Lhs,Rhs>::ret>::ret>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
 
-    typedef typename Lhs::Nested LhsNested;
-    typedef typename Rhs::Nested RhsNested;
+    typedef typename internal::ref_selector<Lhs>::type LhsNested;
+    typedef typename internal::ref_selector<Rhs>::type RhsNested;
     typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
     typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
 
-    Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs)
+    EIGEN_DEVICE_FUNC Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs)
     {
       eigen_assert(lhs.cols() == rhs.rows()
         && "invalid matrix product"
         && "if you wanted a coeff-wise or a dot product use the respective explicit functions");
     }
 
-    inline Index rows() const { return m_lhs.rows(); }
-    inline Index cols() const { return m_rhs.cols(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
 
-    const LhsNestedCleaned& lhs() const { return m_lhs; }
-    const RhsNestedCleaned& rhs() const { return m_rhs; }
+    EIGEN_DEVICE_FUNC const LhsNestedCleaned& lhs() const { return m_lhs; }
+    EIGEN_DEVICE_FUNC const RhsNestedCleaned& rhs() const { return m_rhs; }
 
   protected:
 
-    const LhsNested m_lhs;
-    const RhsNested m_rhs;
+    LhsNested m_lhs;
+    RhsNested m_rhs;
+};
+
+namespace internal {
+  
+template<typename Lhs, typename Rhs, int Option, int ProductTag = internal::product_type<Lhs,Rhs>::ret>
+class dense_product_base
+ : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
+{};
+
+/** Convertion to scalar for inner-products */
+template<typename Lhs, typename Rhs, int Option>
+class dense_product_base<Lhs, Rhs, Option, InnerProduct>
+ : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
+{
+  typedef Product<Lhs,Rhs,Option> ProductXpr;
+  typedef typename internal::dense_xpr_base<ProductXpr>::type Base;
+public:
+  using Base::derived;
+  typedef typename Base::Scalar Scalar;
+  
+  EIGEN_STRONG_INLINE operator const Scalar() const
+  {
+    return internal::evaluator<ProductXpr>(derived()).coeff(0,0);
+  }
 };
 
-template<typename Lhs, typename Rhs>
-class ProductImpl<Lhs,Rhs,Dense> : public internal::dense_xpr_base<Product<Lhs,Rhs> >::type
+} // namespace internal
+
+// Generic API dispatcher
+template<typename Lhs, typename Rhs, int Option, typename StorageKind>
+class ProductImpl : public internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type
 {
-    typedef Product<Lhs, Rhs> Derived;
   public:
+    typedef typename internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type Base;
+};
 
-    typedef typename internal::dense_xpr_base<Product<Lhs, Rhs> >::type Base;
+template<typename Lhs, typename Rhs, int Option>
+class ProductImpl<Lhs,Rhs,Option,Dense>
+  : public internal::dense_product_base<Lhs,Rhs,Option>
+{
+    typedef Product<Lhs, Rhs, Option> Derived;
+    
+  public:
+    
+    typedef typename internal::dense_product_base<Lhs, Rhs, Option> Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
+  protected:
+    enum {
+      IsOneByOne = (RowsAtCompileTime == 1 || RowsAtCompileTime == Dynamic) && 
+                   (ColsAtCompileTime == 1 || ColsAtCompileTime == Dynamic),
+      EnableCoeff = IsOneByOne || Option==LazyProduct
+    };
+    
+  public:
+  
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index row, Index col) const
+    {
+      EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
+      eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
+      
+      return internal::evaluator<Derived>(derived()).coeff(row,col);
+    }
+
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index i) const
+    {
+      EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
+      eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
+      
+      return internal::evaluator<Derived>(derived()).coeff(i);
+    }
+    
+  
 };
 
+} // end namespace Eigen
+
 #endif // EIGEN_PRODUCT_H
diff --git a/extern/Eigen3/Eigen/src/Core/ProductEvaluators.h b/extern/Eigen3/Eigen/src/Core/ProductEvaluators.h
new file mode 100644
index 00000000000..9b99bd76962
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/ProductEvaluators.h
@@ -0,0 +1,1112 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011 Jitse Niesen <jitse@maths.leeds.ac.uk>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+
+#ifndef EIGEN_PRODUCTEVALUATORS_H
+#define EIGEN_PRODUCTEVALUATORS_H
+
+namespace Eigen {
+  
+namespace internal {
+
+/** \internal
+  * Evaluator of a product expression.
+  * Since products require special treatments to handle all possible cases,
+  * we simply deffer the evaluation logic to a product_evaluator class
+  * which offers more partial specialization possibilities.
+  * 
+  * \sa class product_evaluator
+  */
+template<typename Lhs, typename Rhs, int Options>
+struct evaluator<Product<Lhs, Rhs, Options> > 
+ : public product_evaluator<Product<Lhs, Rhs, Options> >
+{
+  typedef Product<Lhs, Rhs, Options> XprType;
+  typedef product_evaluator<XprType> Base;
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr) : Base(xpr) {}
+};
+ 
+// Catch "scalar * ( A * B )" and transform it to "(A*scalar) * B"
+// TODO we should apply that rule only if that's really helpful
+template<typename Lhs, typename Rhs, typename Scalar1, typename Scalar2, typename Plain1>
+struct evaluator_assume_aliasing<CwiseBinaryOp<internal::scalar_product_op<Scalar1,Scalar2>,
+                                               const CwiseNullaryOp<internal::scalar_constant_op<Scalar1>, Plain1>,
+                                               const Product<Lhs, Rhs, DefaultProduct> > >
+{
+  static const bool value = true;
+};
+template<typename Lhs, typename Rhs, typename Scalar1, typename Scalar2, typename Plain1>
+struct evaluator<CwiseBinaryOp<internal::scalar_product_op<Scalar1,Scalar2>,
+                               const CwiseNullaryOp<internal::scalar_constant_op<Scalar1>, Plain1>,
+                               const Product<Lhs, Rhs, DefaultProduct> > >
+ : public evaluator<Product<EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar1,Lhs,product), Rhs, DefaultProduct> >
+{
+  typedef CwiseBinaryOp<internal::scalar_product_op<Scalar1,Scalar2>,
+                               const CwiseNullaryOp<internal::scalar_constant_op<Scalar1>, Plain1>,
+                               const Product<Lhs, Rhs, DefaultProduct> > XprType;
+  typedef evaluator<Product<EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar1,Lhs,product), Rhs, DefaultProduct> > Base;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr)
+    : Base(xpr.lhs().functor().m_other * xpr.rhs().lhs() * xpr.rhs().rhs())
+  {}
+};
+
+
+template<typename Lhs, typename Rhs, int DiagIndex>
+struct evaluator<Diagonal<const Product<Lhs, Rhs, DefaultProduct>, DiagIndex> > 
+ : public evaluator<Diagonal<const Product<Lhs, Rhs, LazyProduct>, DiagIndex> >
+{
+  typedef Diagonal<const Product<Lhs, Rhs, DefaultProduct>, DiagIndex> XprType;
+  typedef evaluator<Diagonal<const Product<Lhs, Rhs, LazyProduct>, DiagIndex> > Base;
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr)
+    : Base(Diagonal<const Product<Lhs, Rhs, LazyProduct>, DiagIndex>(
+        Product<Lhs, Rhs, LazyProduct>(xpr.nestedExpression().lhs(), xpr.nestedExpression().rhs()),
+        xpr.index() ))
+  {}
+};
+
+
+// Helper class to perform a matrix product with the destination at hand.
+// Depending on the sizes of the factors, there are different evaluation strategies
+// as controlled by internal::product_type.
+template< typename Lhs, typename Rhs,
+          typename LhsShape = typename evaluator_traits<Lhs>::Shape,
+          typename RhsShape = typename evaluator_traits<Rhs>::Shape,
+          int ProductType = internal::product_type<Lhs,Rhs>::value>
+struct generic_product_impl;
+
+template<typename Lhs, typename Rhs>
+struct evaluator_assume_aliasing<Product<Lhs, Rhs, DefaultProduct> > {
+  static const bool value = true;
+};
+
+// This is the default evaluator implementation for products:
+// It creates a temporary and call generic_product_impl
+template<typename Lhs, typename Rhs, int Options, int ProductTag, typename LhsShape, typename RhsShape>
+struct product_evaluator<Product<Lhs, Rhs, Options>, ProductTag, LhsShape, RhsShape>
+  : public evaluator<typename Product<Lhs, Rhs, Options>::PlainObject>
+{
+  typedef Product<Lhs, Rhs, Options> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+  enum {
+    Flags = Base::Flags | EvalBeforeNestingBit
+  };
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  explicit product_evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    
+// FIXME shall we handle nested_eval here?,
+// if so, then we must take care at removing the call to nested_eval in the specializations (e.g., in permutation_matrix_product, transposition_matrix_product, etc.)
+//     typedef typename internal::nested_eval<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
+//     typedef typename internal::nested_eval<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
+//     typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
+//     typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
+//     
+//     const LhsNested lhs(xpr.lhs());
+//     const RhsNested rhs(xpr.rhs());
+//   
+//     generic_product_impl<LhsNestedCleaned, RhsNestedCleaned>::evalTo(m_result, lhs, rhs);
+
+    generic_product_impl<Lhs, Rhs, LhsShape, RhsShape, ProductTag>::evalTo(m_result, xpr.lhs(), xpr.rhs());
+  }
+  
+protected:  
+  PlainObject m_result;
+};
+
+// The following three shortcuts are enabled only if the scalar types match excatly.
+// TODO: we could enable them for different scalar types when the product is not vectorized.
+
+// Dense = Product
+template< typename DstXprType, typename Lhs, typename Rhs, int Options, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::assign_op<Scalar,Scalar>, Dense2Dense,
+  typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type>
+{
+  typedef Product<Lhs,Rhs,Options> SrcXprType;
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    // FIXME shall we handle nested_eval here?
+    generic_product_impl<Lhs, Rhs>::evalTo(dst, src.lhs(), src.rhs());
+  }
+};
+
+// Dense += Product
+template< typename DstXprType, typename Lhs, typename Rhs, int Options, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::add_assign_op<Scalar,Scalar>, Dense2Dense,
+  typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type>
+{
+  typedef Product<Lhs,Rhs,Options> SrcXprType;
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<Scalar,Scalar> &)
+  {
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    // FIXME shall we handle nested_eval here?
+    generic_product_impl<Lhs, Rhs>::addTo(dst, src.lhs(), src.rhs());
+  }
+};
+
+// Dense -= Product
+template< typename DstXprType, typename Lhs, typename Rhs, int Options, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::sub_assign_op<Scalar,Scalar>, Dense2Dense,
+  typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type>
+{
+  typedef Product<Lhs,Rhs,Options> SrcXprType;
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<Scalar,Scalar> &)
+  {
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    // FIXME shall we handle nested_eval here?
+    generic_product_impl<Lhs, Rhs>::subTo(dst, src.lhs(), src.rhs());
+  }
+};
+
+
+// Dense ?= scalar * Product
+// TODO we should apply that rule if that's really helpful
+// for instance, this is not good for inner products
+template< typename DstXprType, typename Lhs, typename Rhs, typename AssignFunc, typename Scalar, typename ScalarBis, typename Plain>
+struct Assignment<DstXprType, CwiseBinaryOp<internal::scalar_product_op<ScalarBis,Scalar>, const CwiseNullaryOp<internal::scalar_constant_op<ScalarBis>,Plain>,
+                                           const Product<Lhs,Rhs,DefaultProduct> >, AssignFunc, Dense2Dense>
+{
+  typedef CwiseBinaryOp<internal::scalar_product_op<ScalarBis,Scalar>,
+                        const CwiseNullaryOp<internal::scalar_constant_op<ScalarBis>,Plain>,
+                        const Product<Lhs,Rhs,DefaultProduct> > SrcXprType;
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const AssignFunc& func)
+  {
+    call_assignment_no_alias(dst, (src.lhs().functor().m_other * src.rhs().lhs())*src.rhs().rhs(), func);
+  }
+};
+
+//----------------------------------------
+// Catch "Dense ?= xpr + Product<>" expression to save one temporary
+// FIXME we could probably enable these rules for any product, i.e., not only Dense and DefaultProduct
+
+template<typename OtherXpr, typename Lhs, typename Rhs>
+struct evaluator_assume_aliasing<CwiseBinaryOp<internal::scalar_sum_op<typename OtherXpr::Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, const OtherXpr,
+                                               const Product<Lhs,Rhs,DefaultProduct> >, DenseShape > {
+  static const bool value = true;
+};
+
+template<typename OtherXpr, typename Lhs, typename Rhs>
+struct evaluator_assume_aliasing<CwiseBinaryOp<internal::scalar_difference_op<typename OtherXpr::Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, const OtherXpr,
+                                               const Product<Lhs,Rhs,DefaultProduct> >, DenseShape > {
+  static const bool value = true;
+};
+
+template<typename DstXprType, typename OtherXpr, typename ProductType, typename Func1, typename Func2>
+struct assignment_from_xpr_op_product
+{
+  template<typename SrcXprType, typename InitialFunc>
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const InitialFunc& /*func*/)
+  {
+    call_assignment_no_alias(dst, src.lhs(), Func1());
+    call_assignment_no_alias(dst, src.rhs(), Func2());
+  }
+};
+
+#define EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(ASSIGN_OP,BINOP,ASSIGN_OP2) \
+  template< typename DstXprType, typename OtherXpr, typename Lhs, typename Rhs, typename DstScalar, typename SrcScalar, typename OtherScalar,typename ProdScalar> \
+  struct Assignment<DstXprType, CwiseBinaryOp<internal::BINOP<OtherScalar,ProdScalar>, const OtherXpr, \
+                                            const Product<Lhs,Rhs,DefaultProduct> >, internal::ASSIGN_OP<DstScalar,SrcScalar>, Dense2Dense> \
+    : assignment_from_xpr_op_product<DstXprType, OtherXpr, Product<Lhs,Rhs,DefaultProduct>, internal::ASSIGN_OP<DstScalar,OtherScalar>, internal::ASSIGN_OP2<DstScalar,ProdScalar> > \
+  {}
+
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(assign_op,    scalar_sum_op,add_assign_op);
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(add_assign_op,scalar_sum_op,add_assign_op);
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(sub_assign_op,scalar_sum_op,sub_assign_op);
+
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(assign_op,    scalar_difference_op,sub_assign_op);
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(add_assign_op,scalar_difference_op,sub_assign_op);
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(sub_assign_op,scalar_difference_op,add_assign_op);
+
+//----------------------------------------
+
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,InnerProduct>
+{
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    dst.coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    dst.coeffRef(0,0) += (lhs.transpose().cwiseProduct(rhs)).sum();
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  { dst.coeffRef(0,0) -= (lhs.transpose().cwiseProduct(rhs)).sum(); }
+};
+
+
+/***********************************************************************
+*  Implementation of outer dense * dense vector product
+***********************************************************************/
+
+// Column major result
+template<typename Dst, typename Lhs, typename Rhs, typename Func>
+void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const false_type&)
+{
+  evaluator<Rhs> rhsEval(rhs);
+  typename nested_eval<Lhs,Rhs::SizeAtCompileTime>::type actual_lhs(lhs);
+  // FIXME if cols is large enough, then it might be useful to make sure that lhs is sequentially stored
+  // FIXME not very good if rhs is real and lhs complex while alpha is real too
+  const Index cols = dst.cols();
+  for (Index j=0; j<cols; ++j)
+    func(dst.col(j), rhsEval.coeff(Index(0),j) * actual_lhs);
+}
+
+// Row major result
+template<typename Dst, typename Lhs, typename Rhs, typename Func>
+void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const true_type&)
+{
+  evaluator<Lhs> lhsEval(lhs);
+  typename nested_eval<Rhs,Lhs::SizeAtCompileTime>::type actual_rhs(rhs);
+  // FIXME if rows is large enough, then it might be useful to make sure that rhs is sequentially stored
+  // FIXME not very good if lhs is real and rhs complex while alpha is real too
+  const Index rows = dst.rows();
+  for (Index i=0; i<rows; ++i)
+    func(dst.row(i), lhsEval.coeff(i,Index(0)) * actual_rhs);
+}
+
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,OuterProduct>
+{
+  template<typename T> struct is_row_major : internal::conditional<(int(T::Flags)&RowMajorBit), internal::true_type, internal::false_type>::type {};
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  // TODO it would be nice to be able to exploit our *_assign_op functors for that purpose
+  struct set  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived()  = src; } };
+  struct add  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } };
+  struct sub  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } };
+  struct adds {
+    Scalar m_scale;
+    explicit adds(const Scalar& s) : m_scale(s) {}
+    template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const {
+      dst.const_cast_derived() += m_scale * src;
+    }
+  };
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    internal::outer_product_selector_run(dst, lhs, rhs, set(), is_row_major<Dst>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    internal::outer_product_selector_run(dst, lhs, rhs, add(), is_row_major<Dst>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    internal::outer_product_selector_run(dst, lhs, rhs, sub(), is_row_major<Dst>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    internal::outer_product_selector_run(dst, lhs, rhs, adds(alpha), is_row_major<Dst>());
+  }
+  
+};
+
+
+// This base class provides default implementations for evalTo, addTo, subTo, in terms of scaleAndAddTo
+template<typename Lhs, typename Rhs, typename Derived>
+struct generic_product_impl_base
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  { dst.setZero(); scaleAndAddTo(dst, lhs, rhs, Scalar(1)); }
+
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  { scaleAndAddTo(dst,lhs, rhs, Scalar(1)); }
+
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  { scaleAndAddTo(dst, lhs, rhs, Scalar(-1)); }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  { Derived::scaleAndAddTo(dst,lhs,rhs,alpha); }
+
+};
+
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemvProduct>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemvProduct> >
+{
+  typedef typename nested_eval<Lhs,1>::type LhsNested;
+  typedef typename nested_eval<Rhs,1>::type RhsNested;
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
+  typedef typename internal::remove_all<typename internal::conditional<int(Side)==OnTheRight,LhsNested,RhsNested>::type>::type MatrixType;
+
+  template<typename Dest>
+  static EIGEN_STRONG_INLINE void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    LhsNested actual_lhs(lhs);
+    RhsNested actual_rhs(rhs);
+    internal::gemv_dense_selector<Side,
+                            (int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
+                            bool(internal::blas_traits<MatrixType>::HasUsableDirectAccess)
+                           >::run(actual_lhs, actual_rhs, dst, alpha);
+  }
+};
+
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode> 
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    // Same as: dst.noalias() = lhs.lazyProduct(rhs);
+    // but easier on the compiler side
+    call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::assign_op<typename Dst::Scalar,Scalar>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    // dst.noalias() += lhs.lazyProduct(rhs);
+    call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::add_assign_op<typename Dst::Scalar,Scalar>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    // dst.noalias() -= lhs.lazyProduct(rhs);
+    call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::sub_assign_op<typename Dst::Scalar,Scalar>());
+  }
+  
+//   template<typename Dst>
+//   static inline void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+//   { dst.noalias() += alpha * lhs.lazyProduct(rhs); }
+};
+
+// This specialization enforces the use of a coefficient-based evaluation strategy
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,LazyCoeffBasedProductMode>
+  : generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode> {};
+
+// Case 2: Evaluate coeff by coeff
+//
+// This is mostly taken from CoeffBasedProduct.h
+// The main difference is that we add an extra argument to the etor_product_*_impl::run() function
+// for the inner dimension of the product, because evaluator object do not know their size.
+
+template<int Traversal, int UnrollingIndex, typename Lhs, typename Rhs, typename RetScalar>
+struct etor_product_coeff_impl;
+
+template<int StorageOrder, int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl;
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape, DenseShape>
+    : evaluator_base<Product<Lhs, Rhs, LazyProduct> >
+{
+  typedef Product<Lhs, Rhs, LazyProduct> XprType;
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  explicit product_evaluator(const XprType& xpr)
+    : m_lhs(xpr.lhs()),
+      m_rhs(xpr.rhs()),
+      m_lhsImpl(m_lhs),     // FIXME the creation of the evaluator objects should result in a no-op, but check that!
+      m_rhsImpl(m_rhs),     //       Moreover, they are only useful for the packet path, so we could completely disable them when not needed,
+                            //       or perhaps declare them on the fly on the packet method... We have experiment to check what's best.
+      m_innerDim(xpr.lhs().cols())
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits<Scalar>::MulCost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits<Scalar>::AddCost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+#if 0
+    std::cerr << "LhsOuterStrideBytes=  " << LhsOuterStrideBytes << "\n";
+    std::cerr << "RhsOuterStrideBytes=  " << RhsOuterStrideBytes << "\n";
+    std::cerr << "LhsAlignment=         " << LhsAlignment << "\n";
+    std::cerr << "RhsAlignment=         " << RhsAlignment << "\n";
+    std::cerr << "CanVectorizeLhs=      " << CanVectorizeLhs << "\n";
+    std::cerr << "CanVectorizeRhs=      " << CanVectorizeRhs << "\n";
+    std::cerr << "CanVectorizeInner=    " << CanVectorizeInner << "\n";
+    std::cerr << "EvalToRowMajor=       " << EvalToRowMajor << "\n";
+    std::cerr << "Alignment=            " << Alignment << "\n";
+    std::cerr << "Flags=                " << Flags << "\n";
+#endif
+  }
+
+  // Everything below here is taken from CoeffBasedProduct.h
+
+  typedef typename internal::nested_eval<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
+  typedef typename internal::nested_eval<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
+  
+  typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
+  typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
+
+  typedef evaluator<LhsNestedCleaned> LhsEtorType;
+  typedef evaluator<RhsNestedCleaned> RhsEtorType;
+
+  enum {
+    RowsAtCompileTime = LhsNestedCleaned::RowsAtCompileTime,
+    ColsAtCompileTime = RhsNestedCleaned::ColsAtCompileTime,
+    InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsNestedCleaned::ColsAtCompileTime, RhsNestedCleaned::RowsAtCompileTime),
+    MaxRowsAtCompileTime = LhsNestedCleaned::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = RhsNestedCleaned::MaxColsAtCompileTime
+  };
+
+  typedef typename find_best_packet<Scalar,RowsAtCompileTime>::type LhsVecPacketType;
+  typedef typename find_best_packet<Scalar,ColsAtCompileTime>::type RhsVecPacketType;
+
+  enum {
+      
+    LhsCoeffReadCost = LhsEtorType::CoeffReadCost,
+    RhsCoeffReadCost = RhsEtorType::CoeffReadCost,
+    CoeffReadCost = InnerSize==0 ? NumTraits<Scalar>::ReadCost
+                  : InnerSize == Dynamic ? HugeCost
+                  : InnerSize * (NumTraits<Scalar>::MulCost + LhsCoeffReadCost + RhsCoeffReadCost)
+                    + (InnerSize - 1) * NumTraits<Scalar>::AddCost,
+
+    Unroll = CoeffReadCost <= EIGEN_UNROLLING_LIMIT,
+    
+    LhsFlags = LhsEtorType::Flags,
+    RhsFlags = RhsEtorType::Flags,
+    
+    LhsRowMajor = LhsFlags & RowMajorBit,
+    RhsRowMajor = RhsFlags & RowMajorBit,
+
+    LhsVecPacketSize = unpacket_traits<LhsVecPacketType>::size,
+    RhsVecPacketSize = unpacket_traits<RhsVecPacketType>::size,
+
+    // Here, we don't care about alignment larger than the usable packet size.
+    LhsAlignment = EIGEN_PLAIN_ENUM_MIN(LhsEtorType::Alignment,LhsVecPacketSize*int(sizeof(typename LhsNestedCleaned::Scalar))),
+    RhsAlignment = EIGEN_PLAIN_ENUM_MIN(RhsEtorType::Alignment,RhsVecPacketSize*int(sizeof(typename RhsNestedCleaned::Scalar))),
+      
+    SameType = is_same<typename LhsNestedCleaned::Scalar,typename RhsNestedCleaned::Scalar>::value,
+
+    CanVectorizeRhs = bool(RhsRowMajor) && (RhsFlags & PacketAccessBit) && (ColsAtCompileTime!=1),
+    CanVectorizeLhs = (!LhsRowMajor) && (LhsFlags & PacketAccessBit) && (RowsAtCompileTime!=1),
+
+    EvalToRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
+                    : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
+                    : (bool(RhsRowMajor) && !CanVectorizeLhs),
+
+    Flags = ((unsigned int)(LhsFlags | RhsFlags) & HereditaryBits & ~RowMajorBit)
+          | (EvalToRowMajor ? RowMajorBit : 0)
+          // TODO enable vectorization for mixed types
+          | (SameType && (CanVectorizeLhs || CanVectorizeRhs) ? PacketAccessBit : 0)
+          | (XprType::IsVectorAtCompileTime ? LinearAccessBit : 0),
+          
+    LhsOuterStrideBytes = int(LhsNestedCleaned::OuterStrideAtCompileTime) * int(sizeof(typename LhsNestedCleaned::Scalar)),
+    RhsOuterStrideBytes = int(RhsNestedCleaned::OuterStrideAtCompileTime) * int(sizeof(typename RhsNestedCleaned::Scalar)),
+
+    Alignment = bool(CanVectorizeLhs) ? (LhsOuterStrideBytes<=0 || (int(LhsOuterStrideBytes) % EIGEN_PLAIN_ENUM_MAX(1,LhsAlignment))!=0 ? 0 : LhsAlignment)
+              : bool(CanVectorizeRhs) ? (RhsOuterStrideBytes<=0 || (int(RhsOuterStrideBytes) % EIGEN_PLAIN_ENUM_MAX(1,RhsAlignment))!=0 ? 0 : RhsAlignment)
+              : 0,
+
+    /* CanVectorizeInner deserves special explanation. It does not affect the product flags. It is not used outside
+     * of Product. If the Product itself is not a packet-access expression, there is still a chance that the inner
+     * loop of the product might be vectorized. This is the meaning of CanVectorizeInner. Since it doesn't affect
+     * the Flags, it is safe to make this value depend on ActualPacketAccessBit, that doesn't affect the ABI.
+     */
+    CanVectorizeInner =    SameType
+                        && LhsRowMajor
+                        && (!RhsRowMajor)
+                        && (LhsFlags & RhsFlags & ActualPacketAccessBit)
+                        && (InnerSize % packet_traits<Scalar>::size == 0)
+  };
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index row, Index col) const
+  {
+    return (m_lhs.row(row).transpose().cwiseProduct( m_rhs.col(col) )).sum();
+  }
+
+  /* Allow index-based non-packet access. It is impossible though to allow index-based packed access,
+   * which is why we don't set the LinearAccessBit.
+   * TODO: this seems possible when the result is a vector
+   */
+  EIGEN_DEVICE_FUNC const CoeffReturnType coeff(Index index) const
+  {
+    const Index row = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? 0 : index;
+    const Index col = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? index : 0;
+    return (m_lhs.row(row).transpose().cwiseProduct( m_rhs.col(col) )).sum();
+  }
+
+  template<int LoadMode, typename PacketType>
+  const PacketType packet(Index row, Index col) const
+  {
+    PacketType res;
+    typedef etor_product_packet_impl<bool(int(Flags)&RowMajorBit) ? RowMajor : ColMajor,
+                                     Unroll ? int(InnerSize) : Dynamic,
+                                     LhsEtorType, RhsEtorType, PacketType, LoadMode> PacketImpl;
+    PacketImpl::run(row, col, m_lhsImpl, m_rhsImpl, m_innerDim, res);
+    return res;
+  }
+
+  template<int LoadMode, typename PacketType>
+  const PacketType packet(Index index) const
+  {
+    const Index row = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? 0 : index;
+    const Index col = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? index : 0;
+    return packet<LoadMode,PacketType>(row,col);
+  }
+
+protected:
+  typename internal::add_const_on_value_type<LhsNested>::type m_lhs;
+  typename internal::add_const_on_value_type<RhsNested>::type m_rhs;
+  
+  LhsEtorType m_lhsImpl;
+  RhsEtorType m_rhsImpl;
+
+  // TODO: Get rid of m_innerDim if known at compile time
+  Index m_innerDim;
+};
+
+template<typename Lhs, typename Rhs>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, LazyCoeffBasedProductMode, DenseShape, DenseShape>
+  : product_evaluator<Product<Lhs, Rhs, LazyProduct>, CoeffBasedProductMode, DenseShape, DenseShape>
+{
+  typedef Product<Lhs, Rhs, DefaultProduct> XprType;
+  typedef Product<Lhs, Rhs, LazyProduct> BaseProduct;
+  typedef product_evaluator<BaseProduct, CoeffBasedProductMode, DenseShape, DenseShape> Base;
+  enum {
+    Flags = Base::Flags | EvalBeforeNestingBit
+  };
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(BaseProduct(xpr.lhs(),xpr.rhs()))
+  {}
+};
+
+/****************************************
+*** Coeff based product, Packet path  ***
+****************************************/
+
+template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<RowMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res)
+  {
+    etor_product_packet_impl<RowMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, innerDim, res);
+    res =  pmadd(pset1<Packet>(lhs.coeff(row, Index(UnrollingIndex-1))), rhs.template packet<LoadMode,Packet>(Index(UnrollingIndex-1), col), res);
+  }
+};
+
+template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<ColMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res)
+  {
+    etor_product_packet_impl<ColMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, innerDim, res);
+    res =  pmadd(lhs.template packet<LoadMode,Packet>(row, Index(UnrollingIndex-1)), pset1<Packet>(rhs.coeff(Index(UnrollingIndex-1), col)), res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<RowMajor, 1, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res)
+  {
+    res = pmul(pset1<Packet>(lhs.coeff(row, Index(0))),rhs.template packet<LoadMode,Packet>(Index(0), col));
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<ColMajor, 1, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res)
+  {
+    res = pmul(lhs.template packet<LoadMode,Packet>(row, Index(0)), pset1<Packet>(rhs.coeff(Index(0), col)));
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<RowMajor, 0, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res)
+  {
+    res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<ColMajor, 0, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res)
+  {
+    res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<RowMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res)
+  {
+    res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
+    for(Index i = 0; i < innerDim; ++i)
+      res =  pmadd(pset1<Packet>(lhs.coeff(row, i)), rhs.template packet<LoadMode,Packet>(i, col), res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<ColMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res)
+  {
+    res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
+    for(Index i = 0; i < innerDim; ++i)
+      res =  pmadd(lhs.template packet<LoadMode,Packet>(row, i), pset1<Packet>(rhs.coeff(i, col)), res);
+  }
+};
+
+
+/***************************************************************************
+* Triangular products
+***************************************************************************/
+template<int Mode, bool LhsIsTriangular,
+         typename Lhs, bool LhsIsVector,
+         typename Rhs, bool RhsIsVector>
+struct triangular_product_impl;
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct generic_product_impl<Lhs,Rhs,TriangularShape,DenseShape,ProductTag>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,TriangularShape,DenseShape,ProductTag> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    triangular_product_impl<Lhs::Mode,true,typename Lhs::MatrixType,false,Rhs, Rhs::ColsAtCompileTime==1>
+        ::run(dst, lhs.nestedExpression(), rhs, alpha);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct generic_product_impl<Lhs,Rhs,DenseShape,TriangularShape,ProductTag>
+: generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,TriangularShape,ProductTag> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    triangular_product_impl<Rhs::Mode,false,Lhs,Lhs::RowsAtCompileTime==1, typename Rhs::MatrixType, false>::run(dst, lhs, rhs.nestedExpression(), alpha);
+  }
+};
+
+
+/***************************************************************************
+* SelfAdjoint products
+***************************************************************************/
+template <typename Lhs, int LhsMode, bool LhsIsVector,
+          typename Rhs, int RhsMode, bool RhsIsVector>
+struct selfadjoint_product_impl;
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct generic_product_impl<Lhs,Rhs,SelfAdjointShape,DenseShape,ProductTag>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,SelfAdjointShape,DenseShape,ProductTag> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    selfadjoint_product_impl<typename Lhs::MatrixType,Lhs::Mode,false,Rhs,0,Rhs::IsVectorAtCompileTime>::run(dst, lhs.nestedExpression(), rhs, alpha);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct generic_product_impl<Lhs,Rhs,DenseShape,SelfAdjointShape,ProductTag>
+: generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,SelfAdjointShape,ProductTag> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    selfadjoint_product_impl<Lhs,0,Lhs::IsVectorAtCompileTime,typename Rhs::MatrixType,Rhs::Mode,false>::run(dst, lhs, rhs.nestedExpression(), alpha);
+  }
+};
+
+
+/***************************************************************************
+* Diagonal products
+***************************************************************************/
+  
+template<typename MatrixType, typename DiagonalType, typename Derived, int ProductOrder>
+struct diagonal_product_evaluator_base
+  : evaluator_base<Derived>
+{
+   typedef typename ScalarBinaryOpTraits<typename MatrixType::Scalar, typename DiagonalType::Scalar>::ReturnType Scalar;
+public:
+  enum {
+    CoeffReadCost = NumTraits<Scalar>::MulCost + evaluator<MatrixType>::CoeffReadCost + evaluator<DiagonalType>::CoeffReadCost,
+    
+    MatrixFlags = evaluator<MatrixType>::Flags,
+    DiagFlags = evaluator<DiagonalType>::Flags,
+    _StorageOrder = MatrixFlags & RowMajorBit ? RowMajor : ColMajor,
+    _ScalarAccessOnDiag =  !((int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheLeft)
+                           ||(int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheRight)),
+    _SameTypes = is_same<typename MatrixType::Scalar, typename DiagonalType::Scalar>::value,
+    // FIXME currently we need same types, but in the future the next rule should be the one
+    //_Vectorizable = bool(int(MatrixFlags)&PacketAccessBit) && ((!_PacketOnDiag) || (_SameTypes && bool(int(DiagFlags)&PacketAccessBit))),
+    _Vectorizable = bool(int(MatrixFlags)&PacketAccessBit) && _SameTypes && (_ScalarAccessOnDiag || (bool(int(DiagFlags)&PacketAccessBit))),
+    _LinearAccessMask = (MatrixType::RowsAtCompileTime==1 || MatrixType::ColsAtCompileTime==1) ? LinearAccessBit : 0,
+    Flags = ((HereditaryBits|_LinearAccessMask) & (unsigned int)(MatrixFlags)) | (_Vectorizable ? PacketAccessBit : 0),
+    Alignment = evaluator<MatrixType>::Alignment,
+
+    AsScalarProduct =     (DiagonalType::SizeAtCompileTime==1)
+                      ||  (DiagonalType::SizeAtCompileTime==Dynamic && MatrixType::RowsAtCompileTime==1 && ProductOrder==OnTheLeft)
+                      ||  (DiagonalType::SizeAtCompileTime==Dynamic && MatrixType::ColsAtCompileTime==1 && ProductOrder==OnTheRight)
+  };
+  
+  diagonal_product_evaluator_base(const MatrixType &mat, const DiagonalType &diag)
+    : m_diagImpl(diag), m_matImpl(mat)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits<Scalar>::MulCost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index idx) const
+  {
+    if(AsScalarProduct)
+      return m_diagImpl.coeff(0) * m_matImpl.coeff(idx);
+    else
+      return m_diagImpl.coeff(idx) * m_matImpl.coeff(idx);
+  }
+  
+protected:
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet_impl(Index row, Index col, Index id, internal::true_type) const
+  {
+    return internal::pmul(m_matImpl.template packet<LoadMode,PacketType>(row, col),
+                          internal::pset1<PacketType>(m_diagImpl.coeff(id)));
+  }
+  
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet_impl(Index row, Index col, Index id, internal::false_type) const
+  {
+    enum {
+      InnerSize = (MatrixType::Flags & RowMajorBit) ? MatrixType::ColsAtCompileTime : MatrixType::RowsAtCompileTime,
+      DiagonalPacketLoadMode = EIGEN_PLAIN_ENUM_MIN(LoadMode,((InnerSize%16) == 0) ? int(Aligned16) : int(evaluator<DiagonalType>::Alignment)) // FIXME hardcoded 16!!
+    };
+    return internal::pmul(m_matImpl.template packet<LoadMode,PacketType>(row, col),
+                          m_diagImpl.template packet<DiagonalPacketLoadMode,PacketType>(id));
+  }
+  
+  evaluator<DiagonalType> m_diagImpl;
+  evaluator<MatrixType>   m_matImpl;
+};
+
+// diagonal * dense
+template<typename Lhs, typename Rhs, int ProductKind, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, ProductKind>, ProductTag, DiagonalShape, DenseShape>
+  : diagonal_product_evaluator_base<Rhs, typename Lhs::DiagonalVectorType, Product<Lhs, Rhs, LazyProduct>, OnTheLeft>
+{
+  typedef diagonal_product_evaluator_base<Rhs, typename Lhs::DiagonalVectorType, Product<Lhs, Rhs, LazyProduct>, OnTheLeft> Base;
+  using Base::m_diagImpl;
+  using Base::m_matImpl;
+  using Base::coeff;
+  typedef typename Base::Scalar Scalar;
+  
+  typedef Product<Lhs, Rhs, ProductKind> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  
+  enum {
+    StorageOrder = int(Rhs::Flags) & RowMajorBit ? RowMajor : ColMajor
+  };
+
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(xpr.rhs(), xpr.lhs().diagonal())
+  {
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
+  {
+    return m_diagImpl.coeff(row) * m_matImpl.coeff(row, col);
+  }
+  
+#ifndef __CUDACC__
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet(Index row, Index col) const
+  {
+    // FIXME: NVCC used to complain about the template keyword, but we have to check whether this is still the case.
+    // See also similar calls below.
+    return this->template packet_impl<LoadMode,PacketType>(row,col, row,
+                                 typename internal::conditional<int(StorageOrder)==RowMajor, internal::true_type, internal::false_type>::type());
+  }
+  
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet(Index idx) const
+  {
+    return packet<LoadMode,PacketType>(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
+  }
+#endif
+};
+
+// dense * diagonal
+template<typename Lhs, typename Rhs, int ProductKind, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, ProductKind>, ProductTag, DenseShape, DiagonalShape>
+  : diagonal_product_evaluator_base<Lhs, typename Rhs::DiagonalVectorType, Product<Lhs, Rhs, LazyProduct>, OnTheRight>
+{
+  typedef diagonal_product_evaluator_base<Lhs, typename Rhs::DiagonalVectorType, Product<Lhs, Rhs, LazyProduct>, OnTheRight> Base;
+  using Base::m_diagImpl;
+  using Base::m_matImpl;
+  using Base::coeff;
+  typedef typename Base::Scalar Scalar;
+  
+  typedef Product<Lhs, Rhs, ProductKind> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  
+  enum { StorageOrder = int(Lhs::Flags) & RowMajorBit ? RowMajor : ColMajor };
+
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(xpr.lhs(), xpr.rhs().diagonal())
+  {
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
+  {
+    return m_matImpl.coeff(row, col) * m_diagImpl.coeff(col);
+  }
+  
+#ifndef __CUDACC__
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet(Index row, Index col) const
+  {
+    return this->template packet_impl<LoadMode,PacketType>(row,col, col,
+                                 typename internal::conditional<int(StorageOrder)==ColMajor, internal::true_type, internal::false_type>::type());
+  }
+  
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet(Index idx) const
+  {
+    return packet<LoadMode,PacketType>(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
+  }
+#endif
+};
+
+/***************************************************************************
+* Products with permutation matrices
+***************************************************************************/
+
+/** \internal
+  * \class permutation_matrix_product
+  * Internal helper class implementing the product between a permutation matrix and a matrix.
+  * This class is specialized for DenseShape below and for SparseShape in SparseCore/SparsePermutation.h
+  */
+template<typename ExpressionType, int Side, bool Transposed, typename ExpressionShape>
+struct permutation_matrix_product;
+
+template<typename ExpressionType, int Side, bool Transposed>
+struct permutation_matrix_product<ExpressionType, Side, Transposed, DenseShape>
+{
+    typedef typename nested_eval<ExpressionType, 1>::type MatrixType;
+    typedef typename remove_all<MatrixType>::type MatrixTypeCleaned;
+
+    template<typename Dest, typename PermutationType>
+    static inline void run(Dest& dst, const PermutationType& perm, const ExpressionType& xpr)
+    {
+      MatrixType mat(xpr);
+      const Index n = Side==OnTheLeft ? mat.rows() : mat.cols();
+      // FIXME we need an is_same for expression that is not sensitive to constness. For instance
+      // is_same_xpr<Block<const Matrix>, Block<Matrix> >::value should be true.
+      //if(is_same<MatrixTypeCleaned,Dest>::value && extract_data(dst) == extract_data(mat))
+      if(is_same_dense(dst, mat))
+      {
+        // apply the permutation inplace
+        Matrix<bool,PermutationType::RowsAtCompileTime,1,0,PermutationType::MaxRowsAtCompileTime> mask(perm.size());
+        mask.fill(false);
+        Index r = 0;
+        while(r < perm.size())
+        {
+          // search for the next seed
+          while(r<perm.size() && mask[r]) r++;
+          if(r>=perm.size())
+            break;
+          // we got one, let's follow it until we are back to the seed
+          Index k0 = r++;
+          Index kPrev = k0;
+          mask.coeffRef(k0) = true;
+          for(Index k=perm.indices().coeff(k0); k!=k0; k=perm.indices().coeff(k))
+          {
+                  Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>(dst, k)
+            .swap(Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
+                       (dst,((Side==OnTheLeft) ^ Transposed) ? k0 : kPrev));
+
+            mask.coeffRef(k) = true;
+            kPrev = k;
+          }
+        }
+      }
+      else
+      {
+        for(Index i = 0; i < n; ++i)
+        {
+          Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
+               (dst, ((Side==OnTheLeft) ^ Transposed) ? perm.indices().coeff(i) : i)
+
+          =
+
+          Block<const MatrixTypeCleaned,Side==OnTheLeft ? 1 : MatrixTypeCleaned::RowsAtCompileTime,Side==OnTheRight ? 1 : MatrixTypeCleaned::ColsAtCompileTime>
+               (mat, ((Side==OnTheRight) ^ Transposed) ? perm.indices().coeff(i) : i);
+        }
+      }
+    }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Rhs, PermutationShape, MatrixShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    permutation_matrix_product<Rhs, OnTheLeft, false, MatrixShape>::run(dst, lhs, rhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Rhs, MatrixShape, PermutationShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    permutation_matrix_product<Lhs, OnTheRight, false, MatrixShape>::run(dst, rhs, lhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Inverse<Lhs>, Rhs, PermutationShape, MatrixShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Inverse<Lhs>& lhs, const Rhs& rhs)
+  {
+    permutation_matrix_product<Rhs, OnTheLeft, true, MatrixShape>::run(dst, lhs.nestedExpression(), rhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Inverse<Rhs>, MatrixShape, PermutationShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Inverse<Rhs>& rhs)
+  {
+    permutation_matrix_product<Lhs, OnTheRight, true, MatrixShape>::run(dst, rhs.nestedExpression(), lhs);
+  }
+};
+
+
+/***************************************************************************
+* Products with transpositions matrices
+***************************************************************************/
+
+// FIXME could we unify Transpositions and Permutation into a single "shape"??
+
+/** \internal
+  * \class transposition_matrix_product
+  * Internal helper class implementing the product between a permutation matrix and a matrix.
+  */
+template<typename ExpressionType, int Side, bool Transposed, typename ExpressionShape>
+struct transposition_matrix_product
+{
+  typedef typename nested_eval<ExpressionType, 1>::type MatrixType;
+  typedef typename remove_all<MatrixType>::type MatrixTypeCleaned;
+  
+  template<typename Dest, typename TranspositionType>
+  static inline void run(Dest& dst, const TranspositionType& tr, const ExpressionType& xpr)
+  {
+    MatrixType mat(xpr);
+    typedef typename TranspositionType::StorageIndex StorageIndex;
+    const Index size = tr.size();
+    StorageIndex j = 0;
+
+    if(!is_same_dense(dst,mat))
+      dst = mat;
+
+    for(Index k=(Transposed?size-1:0) ; Transposed?k>=0:k<size ; Transposed?--k:++k)
+      if(Index(j=tr.coeff(k))!=k)
+      {
+        if(Side==OnTheLeft)        dst.row(k).swap(dst.row(j));
+        else if(Side==OnTheRight)  dst.col(k).swap(dst.col(j));
+      }
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Rhs, TranspositionsShape, MatrixShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    transposition_matrix_product<Rhs, OnTheLeft, false, MatrixShape>::run(dst, lhs, rhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Rhs, MatrixShape, TranspositionsShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    transposition_matrix_product<Lhs, OnTheRight, false, MatrixShape>::run(dst, rhs, lhs);
+  }
+};
+
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Transpose<Lhs>, Rhs, TranspositionsShape, MatrixShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Transpose<Lhs>& lhs, const Rhs& rhs)
+  {
+    transposition_matrix_product<Rhs, OnTheLeft, true, MatrixShape>::run(dst, lhs.nestedExpression(), rhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Transpose<Rhs>, MatrixShape, TranspositionsShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Transpose<Rhs>& rhs)
+  {
+    transposition_matrix_product<Lhs, OnTheRight, true, MatrixShape>::run(dst, rhs.nestedExpression(), lhs);
+  }
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PRODUCT_EVALUATORS_H
diff --git a/extern/Eigen3/Eigen/src/Core/Random.h b/extern/Eigen3/Eigen/src/Core/Random.h
index 480fea408d0..6faf789c761 100644
--- a/extern/Eigen3/Eigen/src/Core/Random.h
+++ b/extern/Eigen3/Eigen/src/Core/Random.h
@@ -16,8 +16,7 @@ namespace internal {
 
 template<typename Scalar> struct scalar_random_op {
   EIGEN_EMPTY_STRUCT_CTOR(scalar_random_op)
-  template<typename Index>
-  inline const Scalar operator() (Index, Index = 0) const { return random<Scalar>(); }
+  inline const Scalar operator() () const { return random<Scalar>(); }
 };
 
 template<typename Scalar>
@@ -28,12 +27,18 @@ struct functor_traits<scalar_random_op<Scalar> >
 
 /** \returns a random matrix expression
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
   * The parameters \a rows and \a cols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this MatrixBase type.
   *
+  * \not_reentrant
+  * 
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
   * it is redundant to pass \a rows and \a cols as arguments, so Random() should be used
   * instead.
+  * 
   *
   * Example: \include MatrixBase_random_int_int.cpp
   * Output: \verbinclude MatrixBase_random_int_int.out
@@ -41,11 +46,13 @@ struct functor_traits<scalar_random_op<Scalar> >
   * This expression has the "evaluate before nesting" flag so that it will be evaluated into
   * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
+  * 
+  * See DenseBase::NullaryExpr(Index, const CustomNullaryOp&) for an example using C++11 random generators.
   *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index), MatrixBase::Random()
+  * \sa DenseBase::setRandom(), DenseBase::Random(Index), DenseBase::Random()
   */
 template<typename Derived>
-inline const CwiseNullaryOp<internal::scalar_random_op<typename internal::traits<Derived>::Scalar>, Derived>
+inline const typename DenseBase<Derived>::RandomReturnType
 DenseBase<Derived>::Random(Index rows, Index cols)
 {
   return NullaryExpr(rows, cols, internal::scalar_random_op<Scalar>());
@@ -53,10 +60,14 @@ DenseBase<Derived>::Random(Index rows, Index cols)
 
 /** \returns a random vector expression
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  *
   * The parameter \a size is the size of the returned vector.
   * Must be compatible with this MatrixBase type.
   *
   * \only_for_vectors
+  * \not_reentrant
   *
   * This variant is meant to be used for dynamic-size vector types. For fixed-size types,
   * it is redundant to pass \a size as argument, so Random() should be used
@@ -69,10 +80,10 @@ DenseBase<Derived>::Random(Index rows, Index cols)
   * a temporary vector whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
   *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random()
+  * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random()
   */
 template<typename Derived>
-inline const CwiseNullaryOp<internal::scalar_random_op<typename internal::traits<Derived>::Scalar>, Derived>
+inline const typename DenseBase<Derived>::RandomReturnType
 DenseBase<Derived>::Random(Index size)
 {
   return NullaryExpr(size, internal::scalar_random_op<Scalar>());
@@ -80,6 +91,9 @@ DenseBase<Derived>::Random(Index size)
 
 /** \returns a fixed-size random matrix or vector expression
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
   * This variant is only for fixed-size MatrixBase types. For dynamic-size types, you
   * need to use the variants taking size arguments.
   *
@@ -89,11 +103,13 @@ DenseBase<Derived>::Random(Index size)
   * This expression has the "evaluate before nesting" flag so that it will be evaluated into
   * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
+  * 
+  * \not_reentrant
   *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random(Index)
+  * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random(Index)
   */
 template<typename Derived>
-inline const CwiseNullaryOp<internal::scalar_random_op<typename internal::traits<Derived>::Scalar>, Derived>
+inline const typename DenseBase<Derived>::RandomReturnType
 DenseBase<Derived>::Random()
 {
   return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_random_op<Scalar>());
@@ -101,6 +117,11 @@ DenseBase<Derived>::Random()
 
 /** Sets all coefficients in this expression to random values.
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
+  * \not_reentrant
+  * 
   * Example: \include MatrixBase_setRandom.cpp
   * Output: \verbinclude MatrixBase_setRandom.out
   *
@@ -114,12 +135,16 @@ inline Derived& DenseBase<Derived>::setRandom()
 
 /** Resizes to the given \a newSize, and sets all coefficients in this expression to random values.
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
   * \only_for_vectors
+  * \not_reentrant
   *
   * Example: \include Matrix_setRandom_int.cpp
   * Output: \verbinclude Matrix_setRandom_int.out
   *
-  * \sa MatrixBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, MatrixBase::Random()
+  * \sa DenseBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, DenseBase::Random()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
@@ -131,19 +156,24 @@ PlainObjectBase<Derived>::setRandom(Index newSize)
 
 /** Resizes to the given size, and sets all coefficients in this expression to random values.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  *
+  * \not_reentrant
+  * 
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   *
   * Example: \include Matrix_setRandom_int_int.cpp
   * Output: \verbinclude Matrix_setRandom_int_int.out
   *
-  * \sa MatrixBase::setRandom(), setRandom(Index), class CwiseNullaryOp, MatrixBase::Random()
+  * \sa DenseBase::setRandom(), setRandom(Index), class CwiseNullaryOp, DenseBase::Random()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setRandom(Index nbRows, Index nbCols)
+PlainObjectBase<Derived>::setRandom(Index rows, Index cols)
 {
-  resize(nbRows, nbCols);
+  resize(rows, cols);
   return setRandom();
 }
 
diff --git a/extern/Eigen3/Eigen/src/Core/Redux.h b/extern/Eigen3/Eigen/src/Core/Redux.h
index 9b8662a6f9a..760e9f86154 100644
--- a/extern/Eigen3/Eigen/src/Core/Redux.h
+++ b/extern/Eigen3/Eigen/src/Core/Redux.h
@@ -27,8 +27,9 @@ template<typename Func, typename Derived>
 struct redux_traits
 {
 public:
+    typedef typename find_best_packet<typename Derived::Scalar,Derived::SizeAtCompileTime>::type PacketType;
   enum {
-    PacketSize = packet_traits<typename Derived::Scalar>::size,
+    PacketSize = unpacket_traits<PacketType>::size,
     InnerMaxSize = int(Derived::IsRowMajor)
                  ? Derived::MaxColsAtCompileTime
                  : Derived::MaxRowsAtCompileTime
@@ -37,8 +38,8 @@ public:
   enum {
     MightVectorize = (int(Derived::Flags)&ActualPacketAccessBit)
                   && (functor_traits<Func>::PacketAccess),
-    MayLinearVectorize = MightVectorize && (int(Derived::Flags)&LinearAccessBit),
-    MaySliceVectorize  = MightVectorize && int(InnerMaxSize)>=3*PacketSize
+    MayLinearVectorize = bool(MightVectorize) && (int(Derived::Flags)&LinearAccessBit),
+    MaySliceVectorize  = bool(MightVectorize) && int(InnerMaxSize)>=3*PacketSize
   };
 
 public:
@@ -50,21 +51,34 @@ public:
 
 public:
   enum {
-    Cost = (  Derived::SizeAtCompileTime == Dynamic
-           || Derived::CoeffReadCost == Dynamic
-           || (Derived::SizeAtCompileTime!=1 && functor_traits<Func>::Cost == Dynamic)
-           ) ? Dynamic
-           : Derived::SizeAtCompileTime * Derived::CoeffReadCost
-               + (Derived::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
+    Cost = Derived::SizeAtCompileTime == Dynamic ? HugeCost
+         : Derived::SizeAtCompileTime * Derived::CoeffReadCost + (Derived::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
     UnrollingLimit = EIGEN_UNROLLING_LIMIT * (int(Traversal) == int(DefaultTraversal) ? 1 : int(PacketSize))
   };
 
 public:
   enum {
-    Unrolling = Cost != Dynamic && Cost <= UnrollingLimit
-              ? CompleteUnrolling
-              : NoUnrolling
+    Unrolling = Cost <= UnrollingLimit ? CompleteUnrolling : NoUnrolling
   };
+  
+#ifdef EIGEN_DEBUG_ASSIGN
+  static void debug()
+  {
+    std::cerr << "Xpr: " << typeid(typename Derived::XprType).name() << std::endl;
+    std::cerr.setf(std::ios::hex, std::ios::basefield);
+    EIGEN_DEBUG_VAR(Derived::Flags)
+    std::cerr.unsetf(std::ios::hex);
+    EIGEN_DEBUG_VAR(InnerMaxSize)
+    EIGEN_DEBUG_VAR(PacketSize)
+    EIGEN_DEBUG_VAR(MightVectorize)
+    EIGEN_DEBUG_VAR(MayLinearVectorize)
+    EIGEN_DEBUG_VAR(MaySliceVectorize)
+    EIGEN_DEBUG_VAR(Traversal)
+    EIGEN_DEBUG_VAR(UnrollingLimit)
+    EIGEN_DEBUG_VAR(Unrolling)
+    std::cerr << std::endl;
+  }
+#endif
 };
 
 /***************************************************************************
@@ -82,6 +96,7 @@ struct redux_novec_unroller
 
   typedef typename Derived::Scalar Scalar;
 
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
     return func(redux_novec_unroller<Func, Derived, Start, HalfLength>::run(mat,func),
@@ -99,6 +114,7 @@ struct redux_novec_unroller<Func, Derived, Start, 1>
 
   typedef typename Derived::Scalar Scalar;
 
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func&)
   {
     return mat.coeffByOuterInner(outer, inner);
@@ -112,6 +128,7 @@ template<typename Func, typename Derived, int Start>
 struct redux_novec_unroller<Func, Derived, Start, 0>
 {
   typedef typename Derived::Scalar Scalar;
+  EIGEN_DEVICE_FUNC 
   static EIGEN_STRONG_INLINE Scalar run(const Derived&, const Func&) { return Scalar(); }
 };
 
@@ -121,12 +138,12 @@ template<typename Func, typename Derived, int Start, int Length>
 struct redux_vec_unroller
 {
   enum {
-    PacketSize = packet_traits<typename Derived::Scalar>::size,
+    PacketSize = redux_traits<Func, Derived>::PacketSize,
     HalfLength = Length/2
   };
 
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
 
   static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func& func)
   {
@@ -140,18 +157,18 @@ template<typename Func, typename Derived, int Start>
 struct redux_vec_unroller<Func, Derived, Start, 1>
 {
   enum {
-    index = Start * packet_traits<typename Derived::Scalar>::size,
+    index = Start * redux_traits<Func, Derived>::PacketSize,
     outer = index / int(Derived::InnerSizeAtCompileTime),
     inner = index % int(Derived::InnerSizeAtCompileTime),
-    alignment = (Derived::Flags & AlignedBit) ? Aligned : Unaligned
+    alignment = Derived::Alignment
   };
 
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
 
   static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func&)
   {
-    return mat.template packetByOuterInner<alignment>(outer, inner);
+    return mat.template packetByOuterInner<alignment,PacketScalar>(outer, inner);
   }
 };
 
@@ -169,8 +186,8 @@ template<typename Func, typename Derived>
 struct redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename Derived::Index Index;
-  static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func)
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
     eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
     Scalar res;
@@ -193,19 +210,19 @@ template<typename Func, typename Derived>
 struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
-  typedef typename Derived::Index Index;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
 
-  static Scalar run(const Derived& mat, const Func& func)
+  static Scalar run(const Derived &mat, const Func& func)
   {
     const Index size = mat.size();
-    eigen_assert(size && "you are using an empty matrix");
-    const Index packetSize = packet_traits<Scalar>::size;
-    const Index alignedStart = internal::first_aligned(mat);
+    
+    const Index packetSize = redux_traits<Func, Derived>::PacketSize;
+    const int packetAlignment = unpacket_traits<PacketScalar>::alignment;
     enum {
-      alignment = bool(Derived::Flags & DirectAccessBit) || bool(Derived::Flags & AlignedBit)
-                ? Aligned : Unaligned
+      alignment0 = (bool(Derived::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) ? int(packetAlignment) : int(Unaligned),
+      alignment = EIGEN_PLAIN_ENUM_MAX(alignment0, Derived::Alignment)
     };
+    const Index alignedStart = internal::first_default_aligned(mat.nestedExpression());
     const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize);
     const Index alignedSize = ((size-alignedStart)/(packetSize))*(packetSize);
     const Index alignedEnd2 = alignedStart + alignedSize2;
@@ -213,19 +230,19 @@ struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
     Scalar res;
     if(alignedSize)
     {
-      PacketScalar packet_res0 = mat.template packet<alignment>(alignedStart);
+      PacketScalar packet_res0 = mat.template packet<alignment,PacketScalar>(alignedStart);
       if(alignedSize>packetSize) // we have at least two packets to partly unroll the loop
       {
-        PacketScalar packet_res1 = mat.template packet<alignment>(alignedStart+packetSize);
+        PacketScalar packet_res1 = mat.template packet<alignment,PacketScalar>(alignedStart+packetSize);
         for(Index index = alignedStart + 2*packetSize; index < alignedEnd2; index += 2*packetSize)
         {
-          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment>(index));
-          packet_res1 = func.packetOp(packet_res1, mat.template packet<alignment>(index+packetSize));
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(index));
+          packet_res1 = func.packetOp(packet_res1, mat.template packet<alignment,PacketScalar>(index+packetSize));
         }
 
         packet_res0 = func.packetOp(packet_res0,packet_res1);
         if(alignedEnd>alignedEnd2)
-          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment>(alignedEnd2));
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(alignedEnd2));
       }
       res = func.predux(packet_res0);
 
@@ -252,25 +269,24 @@ template<typename Func, typename Derived, int Unrolling>
 struct redux_impl<Func, Derived, SliceVectorizedTraversal, Unrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
-  typedef typename Derived::Index Index;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketType;
 
-  static Scalar run(const Derived& mat, const Func& func)
+  EIGEN_DEVICE_FUNC static Scalar run(const Derived &mat, const Func& func)
   {
     eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
     const Index innerSize = mat.innerSize();
     const Index outerSize = mat.outerSize();
     enum {
-      packetSize = packet_traits<Scalar>::size
+      packetSize = redux_traits<Func, Derived>::PacketSize
     };
     const Index packetedInnerSize = ((innerSize)/packetSize)*packetSize;
     Scalar res;
     if(packetedInnerSize)
     {
-      PacketScalar packet_res = mat.template packet<Unaligned>(0,0);
+      PacketType packet_res = mat.template packet<Unaligned,PacketType>(0,0);
       for(Index j=0; j<outerSize; ++j)
         for(Index i=(j==0?packetSize:0); i<packetedInnerSize; i+=Index(packetSize))
-          packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned>(j,i));
+          packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned,PacketType>(j,i));
 
       res = func.predux(packet_res);
       for(Index j=0; j<outerSize; ++j)
@@ -291,22 +307,90 @@ template<typename Func, typename Derived>
 struct redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
+
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
   enum {
-    PacketSize = packet_traits<Scalar>::size,
+    PacketSize = redux_traits<Func, Derived>::PacketSize,
     Size = Derived::SizeAtCompileTime,
     VectorizedSize = (Size / PacketSize) * PacketSize
   };
-  static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func)
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
     eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
-    Scalar res = func.predux(redux_vec_unroller<Func, Derived, 0, Size / PacketSize>::run(mat,func));
-    if (VectorizedSize != Size)
-      res = func(res,redux_novec_unroller<Func, Derived, VectorizedSize, Size-VectorizedSize>::run(mat,func));
-    return res;
+    if (VectorizedSize > 0) {
+      Scalar res = func.predux(redux_vec_unroller<Func, Derived, 0, Size / PacketSize>::run(mat,func));
+      if (VectorizedSize != Size)
+        res = func(res,redux_novec_unroller<Func, Derived, VectorizedSize, Size-VectorizedSize>::run(mat,func));
+      return res;
+    }
+    else {
+      return redux_novec_unroller<Func, Derived, 0, Size>::run(mat,func);
+    }
   }
 };
 
+// evaluator adaptor
+template<typename _XprType>
+class redux_evaluator
+{
+public:
+  typedef _XprType XprType;
+  EIGEN_DEVICE_FUNC explicit redux_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
+  
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  typedef typename XprType::PacketScalar PacketScalar;
+  typedef typename XprType::PacketReturnType PacketReturnType;
+  
+  enum {
+    MaxRowsAtCompileTime = XprType::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = XprType::MaxColsAtCompileTime,
+    // TODO we should not remove DirectAccessBit and rather find an elegant way to query the alignment offset at runtime from the evaluator
+    Flags = evaluator<XprType>::Flags & ~DirectAccessBit,
+    IsRowMajor = XprType::IsRowMajor,
+    SizeAtCompileTime = XprType::SizeAtCompileTime,
+    InnerSizeAtCompileTime = XprType::InnerSizeAtCompileTime,
+    CoeffReadCost = evaluator<XprType>::CoeffReadCost,
+    Alignment = evaluator<XprType>::Alignment
+  };
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
+  EIGEN_DEVICE_FUNC Index size() const { return m_xpr.size(); }
+  EIGEN_DEVICE_FUNC Index innerSize() const { return m_xpr.innerSize(); }
+  EIGEN_DEVICE_FUNC Index outerSize() const { return m_xpr.outerSize(); }
+
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeff(Index row, Index col) const
+  { return m_evaluator.coeff(row, col); }
+
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeff(Index index) const
+  { return m_evaluator.coeff(index); }
+
+  template<int LoadMode, typename PacketType>
+  PacketType packet(Index row, Index col) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(row, col); }
+
+  template<int LoadMode, typename PacketType>
+  PacketType packet(Index index) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(index); }
+  
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
+  { return m_evaluator.coeff(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
+  
+  template<int LoadMode, typename PacketType>
+  PacketType packetByOuterInner(Index outer, Index inner) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
+  
+  const XprType & nestedExpression() const { return m_xpr; }
+  
+protected:
+  internal::evaluator<XprType> m_evaluator;
+  const XprType &m_xpr;
+};
+
 } // end namespace internal
 
 /***************************************************************************
@@ -317,18 +401,21 @@ struct redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
 /** \returns the result of a full redux operation on the whole matrix or vector using \a func
   *
   * The template parameter \a BinaryOp is the type of the functor \a func which must be
-  * an associative operator. Both current STL and TR1 functor styles are handled.
+  * an associative operator. Both current C++98 and C++11 functor styles are handled.
   *
   * \sa DenseBase::sum(), DenseBase::minCoeff(), DenseBase::maxCoeff(), MatrixBase::colwise(), MatrixBase::rowwise()
   */
 template<typename Derived>
 template<typename Func>
-EIGEN_STRONG_INLINE typename internal::result_of<Func(typename internal::traits<Derived>::Scalar)>::type
+EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::redux(const Func& func) const
 {
-  typedef typename internal::remove_all<typename Derived::Nested>::type ThisNested;
-  return internal::redux_impl<Func, ThisNested>
-            ::run(derived(), func);
+  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
+
+  typedef typename internal::redux_evaluator<Derived> ThisEvaluator;
+  ThisEvaluator thisEval(derived());
+  
+  return internal::redux_impl<Func, ThisEvaluator>::run(thisEval, func);
 }
 
 /** \returns the minimum of all coefficients of \c *this.
@@ -338,7 +425,7 @@ template<typename Derived>
 EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff() const
 {
-  return this->redux(Eigen::internal::scalar_min_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_min_op<Scalar,Scalar>());
 }
 
 /** \returns the maximum of all coefficients of \c *this.
@@ -348,10 +435,12 @@ template<typename Derived>
 EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff() const
 {
-  return this->redux(Eigen::internal::scalar_max_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_max_op<Scalar,Scalar>());
 }
 
-/** \returns the sum of all coefficients of *this
+/** \returns the sum of all coefficients of \c *this
+  *
+  * If \c *this is empty, then the value 0 is returned.
   *
   * \sa trace(), prod(), mean()
   */
@@ -361,7 +450,7 @@ DenseBase<Derived>::sum() const
 {
   if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
     return Scalar(0);
-  return this->redux(Eigen::internal::scalar_sum_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_sum_op<Scalar,Scalar>());
 }
 
 /** \returns the mean of all coefficients of *this
@@ -372,7 +461,14 @@ template<typename Derived>
 EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::mean() const
 {
-  return Scalar(this->redux(Eigen::internal::scalar_sum_op<Scalar>())) / Scalar(this->size());
+#ifdef __INTEL_COMPILER
+  #pragma warning push
+  #pragma warning ( disable : 2259 )
+#endif
+  return Scalar(derived().redux(Eigen::internal::scalar_sum_op<Scalar,Scalar>())) / Scalar(this->size());
+#ifdef __INTEL_COMPILER
+  #pragma warning pop
+#endif
 }
 
 /** \returns the product of all coefficients of *this
@@ -388,7 +484,7 @@ DenseBase<Derived>::prod() const
 {
   if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
     return Scalar(1);
-  return this->redux(Eigen::internal::scalar_product_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_product_op<Scalar>());
 }
 
 /** \returns the trace of \c *this, i.e. the sum of the coefficients on the main diagonal.
diff --git a/extern/Eigen3/Eigen/src/Core/Ref.h b/extern/Eigen3/Eigen/src/Core/Ref.h
index 7a3becaf882..9c6e3c5d9b0 100644
--- a/extern/Eigen3/Eigen/src/Core/Ref.h
+++ b/extern/Eigen3/Eigen/src/Core/Ref.h
@@ -12,79 +12,6 @@
 
 namespace Eigen { 
 
-template<typename Derived> class RefBase;
-template<typename PlainObjectType, int Options = 0,
-         typename StrideType = typename internal::conditional<PlainObjectType::IsVectorAtCompileTime,InnerStride<1>,OuterStride<> >::type > class Ref;
-
-/** \class Ref
-  * \ingroup Core_Module
-  *
-  * \brief A matrix or vector expression mapping an existing expressions
-  *
-  * \tparam PlainObjectType the equivalent matrix type of the mapped data
-  * \tparam Options specifies whether the pointer is \c #Aligned, or \c #Unaligned.
-  *                The default is \c #Unaligned.
-  * \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1),
-  *                   but accept a variable outer stride (leading dimension).
-  *                   This can be overridden by specifying strides.
-  *                   The type passed here must be a specialization of the Stride template, see examples below.
-  *
-  * This class permits to write non template functions taking Eigen's object as parameters while limiting the number of copies.
-  * A Ref<> object can represent either a const expression or a l-value:
-  * \code
-  * // in-out argument:
-  * void foo1(Ref<VectorXf> x);
-  *
-  * // read-only const argument:
-  * void foo2(const Ref<const VectorXf>& x);
-  * \endcode
-  *
-  * In the in-out case, the input argument must satisfies the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered.
-  * By default, a Ref<VectorXf> can reference any dense vector expression of float having a contiguous memory layout.
-  * Likewise, a Ref<MatrixXf> can reference any column major dense matrix expression of float whose column's elements are contiguously stored with
-  * the possibility to have a constant space inbetween each column, i.e.: the inner stride mmust be equal to 1, but the outer-stride (or leading dimension),
-  * can be greater than the number of rows.
-  *
-  * In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function.
-  * Here are some examples:
-  * \code
-  * MatrixXf A;
-  * VectorXf a;
-  * foo1(a.head());             // OK
-  * foo1(A.col());              // OK
-  * foo1(A.row());              // compilation error because here innerstride!=1
-  * foo2(A.row());              // The row is copied into a contiguous temporary
-  * foo2(2*a);                  // The expression is evaluated into a temporary
-  * foo2(A.col().segment(2,4)); // No temporary
-  * \endcode
-  *
-  * The range of inputs that can be referenced without temporary can be enlarged using the last two template parameter.
-  * Here is an example accepting an innerstride!=1:
-  * \code
-  * // in-out argument:
-  * void foo3(Ref<VectorXf,0,InnerStride<> > x);
-  * foo3(A.row());              // OK
-  * \endcode
-  * The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involved more
-  * expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overloads internally calling a
-  * template function, e.g.:
-  * \code
-  * // in the .h:
-  * void foo(const Ref<MatrixXf>& A);
-  * void foo(const Ref<MatrixXf,0,Stride<> >& A);
-  *
-  * // in the .cpp:
-  * template<typename TypeOfA> void foo_impl(const TypeOfA& A) {
-  *     ... // crazy code goes here
-  * }
-  * void foo(const Ref<MatrixXf>& A) { foo_impl(A); }
-  * void foo(const Ref<MatrixXf,0,Stride<> >& A) { foo_impl(A); }
-  * \endcode
-  *
-  *
-  * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
-  */
-
 namespace internal {
 
 template<typename _PlainObjectType, int _Options, typename _StrideType>
@@ -95,7 +22,8 @@ struct traits<Ref<_PlainObjectType, _Options, _StrideType> >
   typedef _StrideType StrideType;
   enum {
     Options = _Options,
-    Flags = traits<Map<_PlainObjectType, _Options, _StrideType> >::Flags | NestByRefBit
+    Flags = traits<Map<_PlainObjectType, _Options, _StrideType> >::Flags | NestByRefBit,
+    Alignment = traits<Map<_PlainObjectType, _Options, _StrideType> >::Alignment
   };
 
   template<typename Derived> struct match {
@@ -107,7 +35,13 @@ struct traits<Ref<_PlainObjectType, _Options, _StrideType> >
                       || (int(StrideType::InnerStrideAtCompileTime)==0 && int(Derived::InnerStrideAtCompileTime)==1),
       OuterStrideMatch = Derived::IsVectorAtCompileTime
                       || int(StrideType::OuterStrideAtCompileTime)==int(Dynamic) || int(StrideType::OuterStrideAtCompileTime)==int(Derived::OuterStrideAtCompileTime),
-      AlignmentMatch = (_Options!=Aligned) || ((PlainObjectType::Flags&AlignedBit)==0) || ((traits<Derived>::Flags&AlignedBit)==AlignedBit),
+      // NOTE, this indirection of evaluator<Derived>::Alignment is needed
+      // to workaround a very strange bug in MSVC related to the instantiation
+      // of has_*ary_operator in evaluator<CwiseNullaryOp>.
+      // This line is surprisingly very sensitive. For instance, simply adding parenthesis
+      // as "DerivedAlignment = (int(evaluator<Derived>::Alignment))," will make MSVC fail...
+      DerivedAlignment = int(evaluator<Derived>::Alignment),
+      AlignmentMatch = (int(traits<PlainObjectType>::Alignment)==int(Unaligned)) || (DerivedAlignment >= int(Alignment)), // FIXME the first condition is not very clear, it should be replaced by the required alignment
       ScalarTypeMatch = internal::is_same<typename PlainObjectType::Scalar, typename Derived::Scalar>::value,
       MatchAtCompileTime = HasDirectAccess && StorageOrderMatch && InnerStrideMatch && OuterStrideMatch && AlignmentMatch && ScalarTypeMatch
     };
@@ -132,12 +66,12 @@ public:
   typedef MapBase<Derived> Base;
   EIGEN_DENSE_PUBLIC_INTERFACE(RefBase)
 
-  inline Index innerStride() const
+  EIGEN_DEVICE_FUNC inline Index innerStride() const
   {
     return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
   }
 
-  inline Index outerStride() const
+  EIGEN_DEVICE_FUNC inline Index outerStride() const
   {
     return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
          : IsVectorAtCompileTime ? this->size()
@@ -145,7 +79,7 @@ public:
          : this->rows();
   }
 
-  RefBase()
+  EIGEN_DEVICE_FUNC RefBase()
     : Base(0,RowsAtCompileTime==Dynamic?0:RowsAtCompileTime,ColsAtCompileTime==Dynamic?0:ColsAtCompileTime),
       // Stride<> does not allow default ctor for Dynamic strides, so let' initialize it with dummy values:
       m_stride(StrideType::OuterStrideAtCompileTime==Dynamic?0:StrideType::OuterStrideAtCompileTime,
@@ -159,8 +93,10 @@ protected:
   typedef Stride<StrideType::OuterStrideAtCompileTime,StrideType::InnerStrideAtCompileTime> StrideBase;
 
   template<typename Expression>
-  void construct(Expression& expr)
+  EIGEN_DEVICE_FUNC void construct(Expression& expr)
   {
+    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(PlainObjectType,Expression);
+
     if(PlainObjectType::RowsAtCompileTime==1)
     {
       eigen_assert(expr.rows()==1 || expr.cols()==1);
@@ -184,15 +120,83 @@ protected:
   StrideBase m_stride;
 };
 
-
+/** \class Ref
+  * \ingroup Core_Module
+  *
+  * \brief A matrix or vector expression mapping an existing expression
+  *
+  * \tparam PlainObjectType the equivalent matrix type of the mapped data
+  * \tparam Options specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
+  *                 The default is \c #Unaligned.
+  * \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1),
+  *                   but accepts a variable outer stride (leading dimension).
+  *                   This can be overridden by specifying strides.
+  *                   The type passed here must be a specialization of the Stride template, see examples below.
+  *
+  * This class provides a way to write non-template functions taking Eigen objects as parameters while limiting the number of copies.
+  * A Ref<> object can represent either a const expression or a l-value:
+  * \code
+  * // in-out argument:
+  * void foo1(Ref<VectorXf> x);
+  *
+  * // read-only const argument:
+  * void foo2(const Ref<const VectorXf>& x);
+  * \endcode
+  *
+  * In the in-out case, the input argument must satisfy the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered.
+  * By default, a Ref<VectorXf> can reference any dense vector expression of float having a contiguous memory layout.
+  * Likewise, a Ref<MatrixXf> can reference any column-major dense matrix expression of float whose column's elements are contiguously stored with
+  * the possibility to have a constant space in-between each column, i.e. the inner stride must be equal to 1, but the outer stride (or leading dimension)
+  * can be greater than the number of rows.
+  *
+  * In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function.
+  * Here are some examples:
+  * \code
+  * MatrixXf A;
+  * VectorXf a;
+  * foo1(a.head());             // OK
+  * foo1(A.col());              // OK
+  * foo1(A.row());              // Compilation error because here innerstride!=1
+  * foo2(A.row());              // Compilation error because A.row() is a 1xN object while foo2 is expecting a Nx1 object
+  * foo2(A.row().transpose());  // The row is copied into a contiguous temporary
+  * foo2(2*a);                  // The expression is evaluated into a temporary
+  * foo2(A.col().segment(2,4)); // No temporary
+  * \endcode
+  *
+  * The range of inputs that can be referenced without temporary can be enlarged using the last two template parameters.
+  * Here is an example accepting an innerstride!=1:
+  * \code
+  * // in-out argument:
+  * void foo3(Ref<VectorXf,0,InnerStride<> > x);
+  * foo3(A.row());              // OK
+  * \endcode
+  * The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involve more
+  * expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overload internally calling a
+  * template function, e.g.:
+  * \code
+  * // in the .h:
+  * void foo(const Ref<MatrixXf>& A);
+  * void foo(const Ref<MatrixXf,0,Stride<> >& A);
+  *
+  * // in the .cpp:
+  * template<typename TypeOfA> void foo_impl(const TypeOfA& A) {
+  *     ... // crazy code goes here
+  * }
+  * void foo(const Ref<MatrixXf>& A) { foo_impl(A); }
+  * void foo(const Ref<MatrixXf,0,Stride<> >& A) { foo_impl(A); }
+  * \endcode
+  *
+  *
+  * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
+  */
 template<typename PlainObjectType, int Options, typename StrideType> class Ref
   : public RefBase<Ref<PlainObjectType, Options, StrideType> >
 {
   private:
     typedef internal::traits<Ref> Traits;
     template<typename Derived>
-    inline Ref(const PlainObjectBase<Derived>& expr,
-               typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0);
+    EIGEN_DEVICE_FUNC inline Ref(const PlainObjectBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0);
   public:
 
     typedef RefBase<Ref> Base;
@@ -201,23 +205,24 @@ template<typename PlainObjectType, int Options, typename StrideType> class Ref
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename Derived>
-    inline Ref(PlainObjectBase<Derived>& expr,
-               typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
+    EIGEN_DEVICE_FUNC inline Ref(PlainObjectBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
     {
-      EIGEN_STATIC_ASSERT(static_cast<bool>(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
       Base::construct(expr.derived());
     }
     template<typename Derived>
-    inline Ref(const DenseBase<Derived>& expr,
-               typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
+    EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
     #else
+    /** Implicit constructor from any dense expression */
     template<typename Derived>
     inline Ref(DenseBase<Derived>& expr)
     #endif
     {
-      EIGEN_STATIC_ASSERT(static_cast<bool>(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
-      EIGEN_STATIC_ASSERT(static_cast<bool>(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
-      enum { THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY = Derived::ThisConstantIsPrivateInPlainObjectBase};
+      EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      EIGEN_STATIC_ASSERT(!Derived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
       Base::construct(expr.const_cast_derived());
     }
 
@@ -236,36 +241,36 @@ template<typename TPlainObjectType, int Options, typename StrideType> class Ref<
     EIGEN_DENSE_PUBLIC_INTERFACE(Ref)
 
     template<typename Derived>
-    inline Ref(const DenseBase<Derived>& expr,
-               typename internal::enable_if<bool(Traits::template match<Derived>::ScalarTypeMatch),Derived>::type* = 0)
+    EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::ScalarTypeMatch),Derived>::type* = 0)
     {
 //      std::cout << match_helper<Derived>::HasDirectAccess << "," << match_helper<Derived>::OuterStrideMatch << "," << match_helper<Derived>::InnerStrideMatch << "\n";
 //      std::cout << int(StrideType::OuterStrideAtCompileTime) << " - " << int(Derived::OuterStrideAtCompileTime) << "\n";
 //      std::cout << int(StrideType::InnerStrideAtCompileTime) << " - " << int(Derived::InnerStrideAtCompileTime) << "\n";
       construct(expr.derived(), typename Traits::template match<Derived>::type());
     }
-    
-    inline Ref(const Ref& other) : Base(other) {
+
+    EIGEN_DEVICE_FUNC inline Ref(const Ref& other) : Base(other) {
       // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
     }
 
     template<typename OtherRef>
-    inline Ref(const RefBase<OtherRef>& other) {
+    EIGEN_DEVICE_FUNC inline Ref(const RefBase<OtherRef>& other) {
       construct(other.derived(), typename Traits::template match<OtherRef>::type());
     }
 
   protected:
 
     template<typename Expression>
-    void construct(const Expression& expr,internal::true_type)
+    EIGEN_DEVICE_FUNC void construct(const Expression& expr,internal::true_type)
     {
       Base::construct(expr);
     }
 
     template<typename Expression>
-    void construct(const Expression& expr, internal::false_type)
+    EIGEN_DEVICE_FUNC void construct(const Expression& expr, internal::false_type)
     {
-      m_object.lazyAssign(expr);
+      internal::call_assignment_no_alias(m_object,expr,internal::assign_op<Scalar,Scalar>());
       Base::construct(m_object);
     }
 
diff --git a/extern/Eigen3/Eigen/src/Core/Replicate.h b/extern/Eigen3/Eigen/src/Core/Replicate.h
index ac4537c1422..9960ef884ef 100644
--- a/extern/Eigen3/Eigen/src/Core/Replicate.h
+++ b/extern/Eigen3/Eigen/src/Core/Replicate.h
@@ -12,21 +12,6 @@
 
 namespace Eigen { 
 
-/**
-  * \class Replicate
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the multiple replication of a matrix or vector
-  *
-  * \param MatrixType the type of the object we are replicating
-  *
-  * This class represents an expression of the multiple replication of a matrix or vector.
-  * It is the return type of DenseBase::replicate() and most of the time
-  * this is the only way it is used.
-  *
-  * \sa DenseBase::replicate()
-  */
-
 namespace internal {
 template<typename MatrixType,int RowFactor,int ColFactor>
 struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
@@ -35,10 +20,7 @@ struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
   typedef typename MatrixType::Scalar Scalar;
   typedef typename traits<MatrixType>::StorageKind StorageKind;
   typedef typename traits<MatrixType>::XprKind XprKind;
-  enum {
-    Factor = (RowFactor==Dynamic || ColFactor==Dynamic) ? Dynamic : RowFactor*ColFactor
-  };
-  typedef typename nested<MatrixType,Factor>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsAtCompileTime = RowFactor==Dynamic || int(MatrixType::RowsAtCompileTime)==Dynamic
@@ -53,12 +35,29 @@ struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
     IsRowMajor = MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1 ? 1
                : MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1 ? 0
                : (MatrixType::Flags & RowMajorBit) ? 1 : 0,
-    Flags = (_MatrixTypeNested::Flags & HereditaryBits & ~RowMajorBit) | (IsRowMajor ? RowMajorBit : 0),
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost
+    
+    // FIXME enable DirectAccess with negative strides?
+    Flags = IsRowMajor ? RowMajorBit : 0
   };
 };
 }
 
+/**
+  * \class Replicate
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the multiple replication of a matrix or vector
+  *
+  * \tparam MatrixType the type of the object we are replicating
+  * \tparam RowFactor number of repetitions at compile time along the vertical direction, can be Dynamic.
+  * \tparam ColFactor number of repetitions at compile time along the horizontal direction, can be Dynamic.
+  *
+  * This class represents an expression of the multiple replication of a matrix or vector.
+  * It is the return type of DenseBase::replicate() and most of the time
+  * this is the only way it is used.
+  *
+  * \sa DenseBase::replicate()
+  */
 template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
   : public internal::dense_xpr_base< Replicate<MatrixType,RowFactor,ColFactor> >::type
 {
@@ -68,10 +67,12 @@ template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
 
     typedef typename internal::dense_xpr_base<Replicate>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Replicate)
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
     template<typename OriginalMatrixType>
-    inline explicit Replicate(const OriginalMatrixType& a_matrix)
-      : m_matrix(a_matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor)
+    EIGEN_DEVICE_FUNC
+    inline explicit Replicate(const OriginalMatrixType& matrix)
+      : m_matrix(matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor)
     {
       EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
                           THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
@@ -79,41 +80,20 @@ template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
     }
 
     template<typename OriginalMatrixType>
-    inline Replicate(const OriginalMatrixType& a_matrix, Index rowFactor, Index colFactor)
-      : m_matrix(a_matrix), m_rowFactor(rowFactor), m_colFactor(colFactor)
+    EIGEN_DEVICE_FUNC
+    inline Replicate(const OriginalMatrixType& matrix, Index rowFactor, Index colFactor)
+      : m_matrix(matrix), m_rowFactor(rowFactor), m_colFactor(colFactor)
     {
       EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
                           THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
     }
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_matrix.rows() * m_rowFactor.value(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_matrix.cols() * m_colFactor.value(); }
 
-    inline Scalar coeff(Index rowId, Index colId) const
-    {
-      // try to avoid using modulo; this is a pure optimization strategy
-      const Index actual_row  = internal::traits<MatrixType>::RowsAtCompileTime==1 ? 0
-                            : RowFactor==1 ? rowId
-                            : rowId%m_matrix.rows();
-      const Index actual_col  = internal::traits<MatrixType>::ColsAtCompileTime==1 ? 0
-                            : ColFactor==1 ? colId
-                            : colId%m_matrix.cols();
-
-      return m_matrix.coeff(actual_row, actual_col);
-    }
-    template<int LoadMode>
-    inline PacketScalar packet(Index rowId, Index colId) const
-    {
-      const Index actual_row  = internal::traits<MatrixType>::RowsAtCompileTime==1 ? 0
-                            : RowFactor==1 ? rowId
-                            : rowId%m_matrix.rows();
-      const Index actual_col  = internal::traits<MatrixType>::ColsAtCompileTime==1 ? 0
-                            : ColFactor==1 ? colId
-                            : colId%m_matrix.cols();
-
-      return m_matrix.template packet<LoadMode>(actual_row, actual_col);
-    }
-
+    EIGEN_DEVICE_FUNC
     const _MatrixTypeNested& nestedExpression() const
     { 
       return m_matrix; 
@@ -142,21 +122,6 @@ DenseBase<Derived>::replicate() const
 }
 
 /**
-  * \return an expression of the replication of \c *this
-  *
-  * Example: \include MatrixBase_replicate_int_int.cpp
-  * Output: \verbinclude MatrixBase_replicate_int_int.out
-  *
-  * \sa VectorwiseOp::replicate(), DenseBase::replicate<int,int>(), class Replicate
-  */
-template<typename Derived>
-const typename DenseBase<Derived>::ReplicateReturnType
-DenseBase<Derived>::replicate(Index rowFactor,Index colFactor) const
-{
-  return Replicate<Derived,Dynamic,Dynamic>(derived(),rowFactor,colFactor);
-}
-
-/**
   * \return an expression of the replication of each column (or row) of \c *this
   *
   * Example: \include DirectionWise_replicate_int.cpp
diff --git a/extern/Eigen3/Eigen/src/Core/ReturnByValue.h b/extern/Eigen3/Eigen/src/Core/ReturnByValue.h
index f635598dccf..c44b7673bb3 100644
--- a/extern/Eigen3/Eigen/src/Core/ReturnByValue.h
+++ b/extern/Eigen3/Eigen/src/Core/ReturnByValue.h
@@ -13,11 +13,6 @@
 
 namespace Eigen {
 
-/** \class ReturnByValue
-  * \ingroup Core_Module
-  *
-  */
-
 namespace internal {
 
 template<typename Derived>
@@ -38,17 +33,22 @@ struct traits<ReturnByValue<Derived> >
  * So internal::nested always gives the plain return matrix type.
  *
  * FIXME: I don't understand why we need this specialization: isn't this taken care of by the EvalBeforeNestingBit ??
+ * Answer: EvalBeforeNestingBit should be deprecated since we have the evaluators
  */
 template<typename Derived,int n,typename PlainObject>
-struct nested<ReturnByValue<Derived>, n, PlainObject>
+struct nested_eval<ReturnByValue<Derived>, n, PlainObject>
 {
   typedef typename traits<Derived>::ReturnType type;
 };
 
 } // end namespace internal
 
+/** \class ReturnByValue
+  * \ingroup Core_Module
+  *
+  */
 template<typename Derived> class ReturnByValue
-  : internal::no_assignment_operator, public internal::dense_xpr_base< ReturnByValue<Derived> >::type
+  : public internal::dense_xpr_base< ReturnByValue<Derived> >::type, internal::no_assignment_operator
 {
   public:
     typedef typename internal::traits<Derived>::ReturnType ReturnType;
@@ -57,10 +57,11 @@ template<typename Derived> class ReturnByValue
     EIGEN_DENSE_PUBLIC_INTERFACE(ReturnByValue)
 
     template<typename Dest>
+    EIGEN_DEVICE_FUNC
     inline void evalTo(Dest& dst) const
     { static_cast<const Derived*>(this)->evalTo(dst); }
-    inline Index rows() const { return static_cast<const Derived*>(this)->rows(); }
-    inline Index cols() const { return static_cast<const Derived*>(this)->cols(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return static_cast<const Derived*>(this)->rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return static_cast<const Derived*>(this)->cols(); }
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 #define Unusable YOU_ARE_TRYING_TO_ACCESS_A_SINGLE_COEFFICIENT_IN_A_SPECIAL_EXPRESSION_WHERE_THAT_IS_NOT_ALLOWED_BECAUSE_THAT_WOULD_BE_INEFFICIENT
@@ -72,8 +73,7 @@ template<typename Derived> class ReturnByValue
     const Unusable& coeff(Index,Index) const { return *reinterpret_cast<const Unusable*>(this); }
     Unusable& coeffRef(Index) { return *reinterpret_cast<Unusable*>(this); }
     Unusable& coeffRef(Index,Index) { return *reinterpret_cast<Unusable*>(this); }
-    template<int LoadMode>  Unusable& packet(Index) const;
-    template<int LoadMode>  Unusable& packet(Index, Index) const;
+#undef Unusable
 #endif
 };
 
@@ -85,14 +85,32 @@ Derived& DenseBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
   return derived();
 }
 
+namespace internal {
+
+// Expression is evaluated in a temporary; default implementation of Assignment is bypassed so that
+// when a ReturnByValue expression is assigned, the evaluator is not constructed.
+// TODO: Finalize port to new regime; ReturnByValue should not exist in the expression world
+  
 template<typename Derived>
-template<typename OtherDerived>
-Derived& DenseBase<Derived>::lazyAssign(const ReturnByValue<OtherDerived>& other)
+struct evaluator<ReturnByValue<Derived> >
+  : public evaluator<typename internal::traits<Derived>::ReturnType>
 {
-  other.evalTo(derived());
-  return derived();
-}
+  typedef ReturnByValue<Derived> XprType;
+  typedef typename internal::traits<Derived>::ReturnType PlainObject;
+  typedef evaluator<PlainObject> Base;
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    xpr.evalTo(m_result);
+  }
+
+protected:
+  PlainObject m_result;
+};
 
+} // end namespace internal
 
 } // end namespace Eigen
 
diff --git a/extern/Eigen3/Eigen/src/Core/Reverse.h b/extern/Eigen3/Eigen/src/Core/Reverse.h
index e30ae3d281b..0640cda2a15 100644
--- a/extern/Eigen3/Eigen/src/Core/Reverse.h
+++ b/extern/Eigen3/Eigen/src/Core/Reverse.h
@@ -14,20 +14,6 @@
 
 namespace Eigen { 
 
-/** \class Reverse
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the reverse of a vector or matrix
-  *
-  * \param MatrixType the type of the object of which we are taking the reverse
-  *
-  * This class represents an expression of the reverse of a vector.
-  * It is the return type of MatrixBase::reverse() and VectorwiseOp::reverse()
-  * and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::reverse(), VectorwiseOp::reverse()
-  */
-
 namespace internal {
 
 template<typename MatrixType, int Direction>
@@ -37,36 +23,43 @@ struct traits<Reverse<MatrixType, Direction> >
   typedef typename MatrixType::Scalar Scalar;
   typedef typename traits<MatrixType>::StorageKind StorageKind;
   typedef typename traits<MatrixType>::XprKind XprKind;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsAtCompileTime = MatrixType::RowsAtCompileTime,
     ColsAtCompileTime = MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-
-    // let's enable LinearAccess only with vectorization because of the product overhead
-    LinearAccess = ( (Direction==BothDirections) && (int(_MatrixTypeNested::Flags)&PacketAccessBit) )
-                 ? LinearAccessBit : 0,
-
-    Flags = int(_MatrixTypeNested::Flags) & (HereditaryBits | LvalueBit | PacketAccessBit | LinearAccess),
-
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost
+    Flags = _MatrixTypeNested::Flags & (RowMajorBit | LvalueBit)
   };
 };
 
-template<typename PacketScalar, bool ReversePacket> struct reverse_packet_cond
+template<typename PacketType, bool ReversePacket> struct reverse_packet_cond
 {
-  static inline PacketScalar run(const PacketScalar& x) { return preverse(x); }
+  static inline PacketType run(const PacketType& x) { return preverse(x); }
 };
 
-template<typename PacketScalar> struct reverse_packet_cond<PacketScalar,false>
+template<typename PacketType> struct reverse_packet_cond<PacketType,false>
 {
-  static inline PacketScalar run(const PacketScalar& x) { return x; }
+  static inline PacketType run(const PacketType& x) { return x; }
 };
 
 } // end namespace internal 
 
+/** \class Reverse
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the reverse of a vector or matrix
+  *
+  * \tparam MatrixType the type of the object of which we are taking the reverse
+  * \tparam Direction defines the direction of the reverse operation, can be Vertical, Horizontal, or BothDirections
+  *
+  * This class represents an expression of the reverse of a vector.
+  * It is the return type of MatrixBase::reverse() and VectorwiseOp::reverse()
+  * and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::reverse(), VectorwiseOp::reverse()
+  */
 template<typename MatrixType, int Direction> class Reverse
   : public internal::dense_xpr_base< Reverse<MatrixType, Direction> >::type
 {
@@ -74,12 +67,9 @@ template<typename MatrixType, int Direction> class Reverse
 
     typedef typename internal::dense_xpr_base<Reverse>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Reverse)
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
     using Base::IsRowMajor;
 
-    // next line is necessary because otherwise const version of operator()
-    // is hidden by non-const version defined in this file
-    using Base::operator(); 
-
   protected:
     enum {
       PacketSize = internal::packet_traits<Scalar>::size,
@@ -95,82 +85,19 @@ template<typename MatrixType, int Direction> class Reverse
     typedef internal::reverse_packet_cond<PacketScalar,ReversePacket> reverse_packet;
   public:
 
-    inline Reverse(const MatrixType& matrix) : m_matrix(matrix) { }
+    EIGEN_DEVICE_FUNC explicit inline Reverse(const MatrixType& matrix) : m_matrix(matrix) { }
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Reverse)
 
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_matrix.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_matrix.cols(); }
 
-    inline Index innerStride() const
+    EIGEN_DEVICE_FUNC inline Index innerStride() const
     {
       return -m_matrix.innerStride();
     }
 
-    inline Scalar& operator()(Index row, Index col)
-    {
-      eigen_assert(row >= 0 && row < rows() && col >= 0 && col < cols());
-      return coeffRef(row, col);
-    }
-
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      return m_matrix.const_cast_derived().coeffRef(ReverseRow ? m_matrix.rows() - row - 1 : row,
-                                                    ReverseCol ? m_matrix.cols() - col - 1 : col);
-    }
-
-    inline CoeffReturnType coeff(Index row, Index col) const
-    {
-      return m_matrix.coeff(ReverseRow ? m_matrix.rows() - row - 1 : row,
-                            ReverseCol ? m_matrix.cols() - col - 1 : col);
-    }
-
-    inline CoeffReturnType coeff(Index index) const
-    {
-      return m_matrix.coeff(m_matrix.size() - index - 1);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_matrix.const_cast_derived().coeffRef(m_matrix.size() - index - 1);
-    }
-
-    inline Scalar& operator()(Index index)
-    {
-      eigen_assert(index >= 0 && index < m_matrix.size());
-      return coeffRef(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index row, Index col) const
-    {
-      return reverse_packet::run(m_matrix.template packet<LoadMode>(
-                                    ReverseRow ? m_matrix.rows() - row - OffsetRow : row,
-                                    ReverseCol ? m_matrix.cols() - col - OffsetCol : col));
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index row, Index col, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(
-                                      ReverseRow ? m_matrix.rows() - row - OffsetRow : row,
-                                      ReverseCol ? m_matrix.cols() - col - OffsetCol : col,
-                                      reverse_packet::run(x));
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return internal::preverse(m_matrix.template packet<LoadMode>( m_matrix.size() - index - PacketSize ));
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(m_matrix.size() - index - PacketSize, internal::preverse(x));
-    }
-
-    const typename internal::remove_all<typename MatrixType::Nested>::type& 
+    EIGEN_DEVICE_FUNC const typename internal::remove_all<typename MatrixType::Nested>::type&
     nestedExpression() const 
     {
       return m_matrix;
@@ -190,33 +117,93 @@ template<typename Derived>
 inline typename DenseBase<Derived>::ReverseReturnType
 DenseBase<Derived>::reverse()
 {
-  return derived();
+  return ReverseReturnType(derived());
 }
 
-/** This is the const version of reverse(). */
-template<typename Derived>
-inline const typename DenseBase<Derived>::ConstReverseReturnType
-DenseBase<Derived>::reverse() const
-{
-  return derived();
-}
+
+//reverse const overload moved DenseBase.h due to a CUDA compiler bug
 
 /** This is the "in place" version of reverse: it reverses \c *this.
   *
   * In most cases it is probably better to simply use the reversed expression
   * of a matrix. However, when reversing the matrix data itself is really needed,
   * then this "in-place" version is probably the right choice because it provides
-  * the following additional features:
+  * the following additional benefits:
   *  - less error prone: doing the same operation with .reverse() requires special care:
   *    \code m = m.reverse().eval(); \endcode
-  *  - this API allows to avoid creating a temporary (the current implementation creates a temporary, but that could be avoided using swap)
+  *  - this API enables reverse operations without the need for a temporary
   *  - it allows future optimizations (cache friendliness, etc.)
   *
-  * \sa reverse() */
+  * \sa VectorwiseOp::reverseInPlace(), reverse() */
 template<typename Derived>
 inline void DenseBase<Derived>::reverseInPlace()
 {
-  derived() = derived().reverse().eval();
+  if(cols()>rows())
+  {
+    Index half = cols()/2;
+    leftCols(half).swap(rightCols(half).reverse());
+    if((cols()%2)==1)
+    {
+      Index half2 = rows()/2;
+      col(half).head(half2).swap(col(half).tail(half2).reverse());
+    }
+  }
+  else
+  {
+    Index half = rows()/2;
+    topRows(half).swap(bottomRows(half).reverse());
+    if((rows()%2)==1)
+    {
+      Index half2 = cols()/2;
+      row(half).head(half2).swap(row(half).tail(half2).reverse());
+    }
+  }
+}
+
+namespace internal {
+  
+template<int Direction>
+struct vectorwise_reverse_inplace_impl;
+
+template<>
+struct vectorwise_reverse_inplace_impl<Vertical>
+{
+  template<typename ExpressionType>
+  static void run(ExpressionType &xpr)
+  {
+    Index half = xpr.rows()/2;
+    xpr.topRows(half).swap(xpr.bottomRows(half).colwise().reverse());
+  }
+};
+
+template<>
+struct vectorwise_reverse_inplace_impl<Horizontal>
+{
+  template<typename ExpressionType>
+  static void run(ExpressionType &xpr)
+  {
+    Index half = xpr.cols()/2;
+    xpr.leftCols(half).swap(xpr.rightCols(half).rowwise().reverse());
+  }
+};
+
+} // end namespace internal
+
+/** This is the "in place" version of VectorwiseOp::reverse: it reverses each column or row of \c *this.
+  *
+  * In most cases it is probably better to simply use the reversed expression
+  * of a matrix. However, when reversing the matrix data itself is really needed,
+  * then this "in-place" version is probably the right choice because it provides
+  * the following additional benefits:
+  *  - less error prone: doing the same operation with .reverse() requires special care:
+  *    \code m = m.reverse().eval(); \endcode
+  *  - this API enables reverse operations without the need for a temporary
+  *
+  * \sa DenseBase::reverseInPlace(), reverse() */
+template<typename ExpressionType, int Direction>
+void VectorwiseOp<ExpressionType,Direction>::reverseInPlace()
+{
+  internal::vectorwise_reverse_inplace_impl<Direction>::run(_expression().const_cast_derived());
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/Select.h b/extern/Eigen3/Eigen/src/Core/Select.h
index 87993bbb553..79eec1b5b0e 100644
--- a/extern/Eigen3/Eigen/src/Core/Select.h
+++ b/extern/Eigen3/Eigen/src/Core/Select.h
@@ -43,23 +43,21 @@ struct traits<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
     ColsAtCompileTime = ConditionMatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = ConditionMatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = ConditionMatrixType::MaxColsAtCompileTime,
-    Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & HereditaryBits,
-    CoeffReadCost = traits<typename remove_all<ConditionMatrixNested>::type>::CoeffReadCost
-                  + EIGEN_SIZE_MAX(traits<typename remove_all<ThenMatrixNested>::type>::CoeffReadCost,
-                                   traits<typename remove_all<ElseMatrixNested>::type>::CoeffReadCost)
+    Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & RowMajorBit
   };
 };
 }
 
 template<typename ConditionMatrixType, typename ThenMatrixType, typename ElseMatrixType>
-class Select : internal::no_assignment_operator,
-  public internal::dense_xpr_base< Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >::type
+class Select : public internal::dense_xpr_base< Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >::type,
+               internal::no_assignment_operator
 {
   public:
 
     typedef typename internal::dense_xpr_base<Select>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Select)
 
+    inline EIGEN_DEVICE_FUNC
     Select(const ConditionMatrixType& a_conditionMatrix,
            const ThenMatrixType& a_thenMatrix,
            const ElseMatrixType& a_elseMatrix)
@@ -69,9 +67,10 @@ class Select : internal::no_assignment_operator,
       eigen_assert(m_condition.cols() == m_then.cols() && m_condition.cols() == m_else.cols());
     }
 
-    Index rows() const { return m_condition.rows(); }
-    Index cols() const { return m_condition.cols(); }
+    inline EIGEN_DEVICE_FUNC Index rows() const { return m_condition.rows(); }
+    inline EIGEN_DEVICE_FUNC Index cols() const { return m_condition.cols(); }
 
+    inline EIGEN_DEVICE_FUNC
     const Scalar coeff(Index i, Index j) const
     {
       if (m_condition.coeff(i,j))
@@ -80,6 +79,7 @@ class Select : internal::no_assignment_operator,
         return m_else.coeff(i,j);
     }
 
+    inline EIGEN_DEVICE_FUNC
     const Scalar coeff(Index i) const
     {
       if (m_condition.coeff(i))
@@ -88,17 +88,17 @@ class Select : internal::no_assignment_operator,
         return m_else.coeff(i);
     }
 
-    const ConditionMatrixType& conditionMatrix() const
+    inline EIGEN_DEVICE_FUNC const ConditionMatrixType& conditionMatrix() const
     {
       return m_condition;
     }
 
-    const ThenMatrixType& thenMatrix() const
+    inline EIGEN_DEVICE_FUNC const ThenMatrixType& thenMatrix() const
     {
       return m_then;
     }
 
-    const ElseMatrixType& elseMatrix() const
+    inline EIGEN_DEVICE_FUNC const ElseMatrixType& elseMatrix() const
     {
       return m_else;
     }
diff --git a/extern/Eigen3/Eigen/src/Core/SelfAdjointView.h b/extern/Eigen3/Eigen/src/Core/SelfAdjointView.h
index 6fa7cd15e6f..b2e51f37ac4 100644
--- a/extern/Eigen3/Eigen/src/Core/SelfAdjointView.h
+++ b/extern/Eigen3/Eigen/src/Core/SelfAdjointView.h
@@ -32,54 +32,62 @@ namespace internal {
 template<typename MatrixType, unsigned int UpLo>
 struct traits<SelfAdjointView<MatrixType, UpLo> > : traits<MatrixType>
 {
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::non_const_type MatrixTypeNested;
   typedef typename remove_all<MatrixTypeNested>::type MatrixTypeNestedCleaned;
   typedef MatrixType ExpressionType;
-  typedef typename MatrixType::PlainObject DenseMatrixType;
+  typedef typename MatrixType::PlainObject FullMatrixType;
   enum {
     Mode = UpLo | SelfAdjoint,
-    Flags =  MatrixTypeNestedCleaned::Flags & (HereditaryBits)
-           & (~(PacketAccessBit | DirectAccessBit | LinearAccessBit)), // FIXME these flags should be preserved
-    CoeffReadCost = MatrixTypeNestedCleaned::CoeffReadCost
+    FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
+    Flags =  MatrixTypeNestedCleaned::Flags & (HereditaryBits|FlagsLvalueBit)
+           & (~(PacketAccessBit | DirectAccessBit | LinearAccessBit)) // FIXME these flags should be preserved
   };
 };
 }
 
-template <typename Lhs, int LhsMode, bool LhsIsVector,
-          typename Rhs, int RhsMode, bool RhsIsVector>
-struct SelfadjointProductMatrix;
 
-// FIXME could also be called SelfAdjointWrapper to be consistent with DiagonalWrapper ??
-template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
-  : public TriangularBase<SelfAdjointView<MatrixType, UpLo> >
+template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
+  : public TriangularBase<SelfAdjointView<_MatrixType, UpLo> >
 {
   public:
 
+    typedef _MatrixType MatrixType;
     typedef TriangularBase<SelfAdjointView> Base;
     typedef typename internal::traits<SelfAdjointView>::MatrixTypeNested MatrixTypeNested;
     typedef typename internal::traits<SelfAdjointView>::MatrixTypeNestedCleaned MatrixTypeNestedCleaned;
+    typedef MatrixTypeNestedCleaned NestedExpression;
 
     /** \brief The type of coefficients in this matrix */
     typedef typename internal::traits<SelfAdjointView>::Scalar Scalar; 
-
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type MatrixConjugateReturnType;
 
     enum {
-      Mode = internal::traits<SelfAdjointView>::Mode
+      Mode = internal::traits<SelfAdjointView>::Mode,
+      Flags = internal::traits<SelfAdjointView>::Flags,
+      TransposeMode = ((Mode & Upper) ? Lower : 0) | ((Mode & Lower) ? Upper : 0)
     };
     typedef typename MatrixType::PlainObject PlainObject;
 
-    inline SelfAdjointView(MatrixType& matrix) : m_matrix(matrix)
-    {}
+    EIGEN_DEVICE_FUNC
+    explicit inline SelfAdjointView(MatrixType& matrix) : m_matrix(matrix)
+    {
+      EIGEN_STATIC_ASSERT(UpLo==Lower || UpLo==Upper,SELFADJOINTVIEW_ACCEPTS_UPPER_AND_LOWER_MODE_ONLY);
+    }
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_matrix.rows(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_matrix.cols(); }
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return m_matrix.outerStride(); }
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return m_matrix.innerStride(); }
 
     /** \sa MatrixBase::coeff()
       * \warning the coordinates must fit into the referenced triangular part
       */
+    EIGEN_DEVICE_FUNC
     inline Scalar coeff(Index row, Index col) const
     {
       Base::check_coordinates_internal(row, col);
@@ -89,36 +97,46 @@ template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
     /** \sa MatrixBase::coeffRef()
       * \warning the coordinates must fit into the referenced triangular part
       */
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index row, Index col)
     {
+      EIGEN_STATIC_ASSERT_LVALUE(SelfAdjointView);
       Base::check_coordinates_internal(row, col);
-      return m_matrix.const_cast_derived().coeffRef(row, col);
+      return m_matrix.coeffRef(row, col);
     }
 
     /** \internal */
+    EIGEN_DEVICE_FUNC
     const MatrixTypeNestedCleaned& _expression() const { return m_matrix; }
 
+    EIGEN_DEVICE_FUNC
     const MatrixTypeNestedCleaned& nestedExpression() const { return m_matrix; }
-    MatrixTypeNestedCleaned& nestedExpression() { return *const_cast<MatrixTypeNestedCleaned*>(&m_matrix); }
+    EIGEN_DEVICE_FUNC
+    MatrixTypeNestedCleaned& nestedExpression() { return m_matrix; }
 
-    /** Efficient self-adjoint matrix times vector/matrix product */
+    /** Efficient triangular matrix times vector/matrix product */
     template<typename OtherDerived>
-    SelfadjointProductMatrix<MatrixType,Mode,false,OtherDerived,0,OtherDerived::IsVectorAtCompileTime>
+    EIGEN_DEVICE_FUNC
+    const Product<SelfAdjointView,OtherDerived>
     operator*(const MatrixBase<OtherDerived>& rhs) const
     {
-      return SelfadjointProductMatrix
-              <MatrixType,Mode,false,OtherDerived,0,OtherDerived::IsVectorAtCompileTime>
-              (m_matrix, rhs.derived());
+      return Product<SelfAdjointView,OtherDerived>(*this, rhs.derived());
     }
 
-    /** Efficient vector/matrix times self-adjoint matrix product */
+    /** Efficient vector/matrix times triangular matrix product */
     template<typename OtherDerived> friend
-    SelfadjointProductMatrix<OtherDerived,0,OtherDerived::IsVectorAtCompileTime,MatrixType,Mode,false>
+    EIGEN_DEVICE_FUNC
+    const Product<OtherDerived,SelfAdjointView>
     operator*(const MatrixBase<OtherDerived>& lhs, const SelfAdjointView& rhs)
     {
-      return SelfadjointProductMatrix
-              <OtherDerived,0,OtherDerived::IsVectorAtCompileTime,MatrixType,Mode,false>
-              (lhs.derived(),rhs.m_matrix);
+      return Product<OtherDerived,SelfAdjointView>(lhs.derived(),rhs);
+    }
+    
+    friend EIGEN_DEVICE_FUNC
+    const SelfAdjointView<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,MatrixType,product),UpLo>
+    operator*(const Scalar& s, const SelfAdjointView& mat)
+    {
+      return (s*mat.nestedExpression()).template selfadjointView<UpLo>();
     }
 
     /** Perform a symmetric rank 2 update of the selfadjoint matrix \c *this:
@@ -132,6 +150,7 @@ template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
       * \sa rankUpdate(const MatrixBase<DerivedU>&, Scalar)
       */
     template<typename DerivedU, typename DerivedV>
+    EIGEN_DEVICE_FUNC
     SelfAdjointView& rankUpdate(const MatrixBase<DerivedU>& u, const MatrixBase<DerivedV>& v, const Scalar& alpha = Scalar(1));
 
     /** Perform a symmetric rank K update of the selfadjoint matrix \c *this:
@@ -145,8 +164,74 @@ template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
       * \sa rankUpdate(const MatrixBase<DerivedU>&, const MatrixBase<DerivedV>&, Scalar)
       */
     template<typename DerivedU>
+    EIGEN_DEVICE_FUNC
     SelfAdjointView& rankUpdate(const MatrixBase<DerivedU>& u, const Scalar& alpha = Scalar(1));
 
+    /** \returns an expression of a triangular view extracted from the current selfadjoint view of a given triangular part
+      *
+      * The parameter \a TriMode can have the following values: \c #Upper, \c #StrictlyUpper, \c #UnitUpper,
+      * \c #Lower, \c #StrictlyLower, \c #UnitLower.
+      *
+      * If \c TriMode references the same triangular part than \c *this, then this method simply return a \c TriangularView of the nested expression,
+      * otherwise, the nested expression is first transposed, thus returning a \c TriangularView<Transpose<MatrixType>> object.
+      *
+      * \sa MatrixBase::triangularView(), class TriangularView
+      */
+    template<unsigned int TriMode>
+    EIGEN_DEVICE_FUNC
+    typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)),
+                                   TriangularView<MatrixType,TriMode>,
+                                   TriangularView<typename MatrixType::AdjointReturnType,TriMode> >::type
+    triangularView() const
+    {
+      typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)), MatrixType&, typename MatrixType::ConstTransposeReturnType>::type tmp1(m_matrix);
+      typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)), MatrixType&, typename MatrixType::AdjointReturnType>::type tmp2(tmp1);
+      return typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)),
+                                   TriangularView<MatrixType,TriMode>,
+                                   TriangularView<typename MatrixType::AdjointReturnType,TriMode> >::type(tmp2);
+    }
+
+    typedef SelfAdjointView<const MatrixConjugateReturnType,UpLo> ConjugateReturnType;
+    /** \sa MatrixBase::conjugate() const */
+    EIGEN_DEVICE_FUNC
+    inline const ConjugateReturnType conjugate() const
+    { return ConjugateReturnType(m_matrix.conjugate()); }
+
+    typedef SelfAdjointView<const typename MatrixType::AdjointReturnType,TransposeMode> AdjointReturnType;
+    /** \sa MatrixBase::adjoint() const */
+    EIGEN_DEVICE_FUNC
+    inline const AdjointReturnType adjoint() const
+    { return AdjointReturnType(m_matrix.adjoint()); }
+
+    typedef SelfAdjointView<typename MatrixType::TransposeReturnType,TransposeMode> TransposeReturnType;
+     /** \sa MatrixBase::transpose() */
+    EIGEN_DEVICE_FUNC
+    inline TransposeReturnType transpose()
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
+      typename MatrixType::TransposeReturnType tmp(m_matrix);
+      return TransposeReturnType(tmp);
+    }
+
+    typedef SelfAdjointView<const typename MatrixType::ConstTransposeReturnType,TransposeMode> ConstTransposeReturnType;
+    /** \sa MatrixBase::transpose() const */
+    EIGEN_DEVICE_FUNC
+    inline const ConstTransposeReturnType transpose() const
+    {
+      return ConstTransposeReturnType(m_matrix.transpose());
+    }
+
+    /** \returns a const expression of the main diagonal of the matrix \c *this
+      *
+      * This method simply returns the diagonal of the nested expression, thus by-passing the SelfAdjointView decorator.
+      *
+      * \sa MatrixBase::diagonal(), class Diagonal */
+    EIGEN_DEVICE_FUNC
+    typename MatrixType::ConstDiagonalReturnType diagonal() const
+    {
+      return typename MatrixType::ConstDiagonalReturnType(m_matrix);
+    }
+
 /////////// Cholesky module ///////////
 
     const LLT<PlainObject, UpLo> llt() const;
@@ -159,31 +244,10 @@ template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
     /** Return type of eigenvalues() */
     typedef Matrix<RealScalar, internal::traits<MatrixType>::ColsAtCompileTime, 1> EigenvaluesReturnType;
 
+    EIGEN_DEVICE_FUNC
     EigenvaluesReturnType eigenvalues() const;
+    EIGEN_DEVICE_FUNC
     RealScalar operatorNorm() const;
-    
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived>
-    SelfAdjointView& operator=(const MatrixBase<OtherDerived>& other)
-    {
-      enum {
-        OtherPart = UpLo == Upper ? StrictlyLower : StrictlyUpper
-      };
-      m_matrix.const_cast_derived().template triangularView<UpLo>() = other;
-      m_matrix.const_cast_derived().template triangularView<OtherPart>() = other.adjoint();
-      return *this;
-    }
-    template<typename OtherMatrixType, unsigned int OtherMode>
-    SelfAdjointView& operator=(const TriangularView<OtherMatrixType, OtherMode>& other)
-    {
-      enum {
-        OtherPart = UpLo == Upper ? StrictlyLower : StrictlyUpper
-      };
-      m_matrix.const_cast_derived().template triangularView<UpLo>() = other.toDenseMatrix();
-      m_matrix.const_cast_derived().template triangularView<OtherPart>() = other.toDenseMatrix().adjoint();
-      return *this;
-    }
-    #endif
 
   protected:
     MatrixTypeNested m_matrix;
@@ -201,90 +265,54 @@ template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
 
 namespace internal {
 
-template<typename Derived1, typename Derived2, int UnrollCount, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Upper), UnrollCount, ClearOpposite>
-{
-  enum {
-    col = (UnrollCount-1) / Derived1::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived1::RowsAtCompileTime
-  };
-
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Upper), UnrollCount-1, ClearOpposite>::run(dst, src);
-
-    if(row == col)
-      dst.coeffRef(row, col) = numext::real(src.coeff(row, col));
-    else if(row < col)
-      dst.coeffRef(col, row) = numext::conj(dst.coeffRef(row, col) = src.coeff(row, col));
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Upper, 0, ClearOpposite>
+// TODO currently a selfadjoint expression has the form SelfAdjointView<.,.>
+//      in the future selfadjoint-ness should be defined by the expression traits
+//      such that Transpose<SelfAdjointView<.,.> > is valid. (currently TriangularBase::transpose() is overloaded to make it work)
+template<typename MatrixType, unsigned int Mode>
+struct evaluator_traits<SelfAdjointView<MatrixType,Mode> >
 {
-  static inline void run(Derived1 &, const Derived2 &) {}
+  typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
+  typedef SelfAdjointShape Shape;
 };
 
-template<typename Derived1, typename Derived2, int UnrollCount, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Lower), UnrollCount, ClearOpposite>
+template<int UpLo, int SetOpposite, typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version>
+class triangular_dense_assignment_kernel<UpLo,SelfAdjoint,SetOpposite,DstEvaluatorTypeT,SrcEvaluatorTypeT,Functor,Version>
+  : public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version>
 {
-  enum {
-    col = (UnrollCount-1) / Derived1::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived1::RowsAtCompileTime
-  };
-
-  static inline void run(Derived1 &dst, const Derived2 &src)
+protected:
+  typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version> Base;
+  typedef typename Base::DstXprType DstXprType;
+  typedef typename Base::SrcXprType SrcXprType;
+  using Base::m_dst;
+  using Base::m_src;
+  using Base::m_functor;
+public:
+  
+  typedef typename Base::DstEvaluatorType DstEvaluatorType;
+  typedef typename Base::SrcEvaluatorType SrcEvaluatorType;
+  typedef typename Base::Scalar Scalar;
+  typedef typename Base::AssignmentTraits AssignmentTraits;
+  
+  
+  EIGEN_DEVICE_FUNC triangular_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
+    : Base(dst, src, func, dstExpr)
+  {}
+  
+  EIGEN_DEVICE_FUNC void assignCoeff(Index row, Index col)
   {
-    triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Lower), UnrollCount-1, ClearOpposite>::run(dst, src);
-
-    if(row == col)
-      dst.coeffRef(row, col) = numext::real(src.coeff(row, col));
-    else if(row > col)
-      dst.coeffRef(col, row) = numext::conj(dst.coeffRef(row, col) = src.coeff(row, col));
+    eigen_internal_assert(row!=col);
+    Scalar tmp = m_src.coeff(row,col);
+    m_functor.assignCoeff(m_dst.coeffRef(row,col), tmp);
+    m_functor.assignCoeff(m_dst.coeffRef(col,row), numext::conj(tmp));
   }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Lower, 0, ClearOpposite>
-{
-  static inline void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Upper, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
+  
+  EIGEN_DEVICE_FUNC void assignDiagonalCoeff(Index id)
   {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      for(Index i = 0; i < j; ++i)
-      {
-        dst.copyCoeff(i, j, src);
-        dst.coeffRef(j,i) = numext::conj(dst.coeff(i,j));
-      }
-      dst.copyCoeff(j, j, src);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Lower, Dynamic, ClearOpposite>
-{
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-  typedef typename Derived1::Index Index;
-    for(Index i = 0; i < dst.rows(); ++i)
-    {
-      for(Index j = 0; j < i; ++j)
-      {
-        dst.copyCoeff(i, j, src);
-        dst.coeffRef(j,i) = numext::conj(dst.coeff(i,j));
-      }
-      dst.copyCoeff(i, i, src);
-    }
+    Base::assignCoeff(id,id);
   }
+  
+  EIGEN_DEVICE_FUNC void assignOppositeCoeff(Index, Index)
+  { eigen_internal_assert(false && "should never be called"); }
 };
 
 } // end namespace internal
@@ -293,20 +321,30 @@ struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Lower, Dyn
 * Implementation of MatrixBase methods
 ***************************************************************************/
 
+/** This is the const version of MatrixBase::selfadjointView() */
 template<typename Derived>
 template<unsigned int UpLo>
 typename MatrixBase<Derived>::template ConstSelfAdjointViewReturnType<UpLo>::Type
 MatrixBase<Derived>::selfadjointView() const
 {
-  return derived();
+  return typename ConstSelfAdjointViewReturnType<UpLo>::Type(derived());
 }
 
+/** \returns an expression of a symmetric/self-adjoint view extracted from the upper or lower triangular part of the current matrix
+  *
+  * The parameter \a UpLo can be either \c #Upper or \c #Lower
+  *
+  * Example: \include MatrixBase_selfadjointView.cpp
+  * Output: \verbinclude MatrixBase_selfadjointView.out
+  *
+  * \sa class SelfAdjointView
+  */
 template<typename Derived>
 template<unsigned int UpLo>
 typename MatrixBase<Derived>::template SelfAdjointViewReturnType<UpLo>::Type
 MatrixBase<Derived>::selfadjointView()
 {
-  return derived();
+  return typename SelfAdjointViewReturnType<UpLo>::Type(derived());
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/SelfCwiseBinaryOp.h b/extern/Eigen3/Eigen/src/Core/SelfCwiseBinaryOp.h
index 0956475af51..7c89c2e23c6 100644
--- a/extern/Eigen3/Eigen/src/Core/SelfCwiseBinaryOp.h
+++ b/extern/Eigen3/Eigen/src/Core/SelfCwiseBinaryOp.h
@@ -12,177 +12,33 @@
 
 namespace Eigen { 
 
-/** \class SelfCwiseBinaryOp
-  * \ingroup Core_Module
-  *
-  * \internal
-  *
-  * \brief Internal helper class for optimizing operators like +=, -=
-  *
-  * This is a pseudo expression class re-implementing the copyCoeff/copyPacket
-  * method to directly performs a +=/-= operations in an optimal way. In particular,
-  * this allows to make sure that the input/output data are loaded only once using
-  * aligned packet loads.
-  *
-  * \sa class SwapWrapper for a similar trick.
-  */
+// TODO generalize the scalar type of 'other'
 
-namespace internal {
-template<typename BinaryOp, typename Lhs, typename Rhs>
-struct traits<SelfCwiseBinaryOp<BinaryOp,Lhs,Rhs> >
-  : traits<CwiseBinaryOp<BinaryOp,Lhs,Rhs> >
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator*=(const Scalar& other)
 {
-  enum {
-    // Note that it is still a good idea to preserve the DirectAccessBit
-    // so that assign can correctly align the data.
-    Flags = traits<CwiseBinaryOp<BinaryOp,Lhs,Rhs> >::Flags | (Lhs::Flags&DirectAccessBit) | (Lhs::Flags&LvalueBit),
-    OuterStrideAtCompileTime = Lhs::OuterStrideAtCompileTime,
-    InnerStrideAtCompileTime = Lhs::InnerStrideAtCompileTime
-  };
-};
+  internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::mul_assign_op<Scalar,Scalar>());
+  return derived();
 }
 
-template<typename BinaryOp, typename Lhs, typename Rhs> class SelfCwiseBinaryOp
-  : public internal::dense_xpr_base< SelfCwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator+=(const Scalar& other)
 {
-  public:
-
-    typedef typename internal::dense_xpr_base<SelfCwiseBinaryOp>::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(SelfCwiseBinaryOp)
-
-    typedef typename internal::packet_traits<Scalar>::type Packet;
-
-    inline SelfCwiseBinaryOp(Lhs& xpr, const BinaryOp& func = BinaryOp()) : m_matrix(xpr), m_functor(func) {}
-
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
-    inline Index outerStride() const { return m_matrix.outerStride(); }
-    inline Index innerStride() const { return m_matrix.innerStride(); }
-    inline const Scalar* data() const { return m_matrix.data(); }
-
-    // note that this function is needed by assign to correctly align loads/stores
-    // TODO make Assign use .data()
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      EIGEN_STATIC_ASSERT_LVALUE(Lhs)
-      return m_matrix.const_cast_derived().coeffRef(row, col);
-    }
-    inline const Scalar& coeffRef(Index row, Index col) const
-    {
-      return m_matrix.coeffRef(row, col);
-    }
-
-    // note that this function is needed by assign to correctly align loads/stores
-    // TODO make Assign use .data()
-    inline Scalar& coeffRef(Index index)
-    {
-      EIGEN_STATIC_ASSERT_LVALUE(Lhs)
-      return m_matrix.const_cast_derived().coeffRef(index);
-    }
-    inline const Scalar& coeffRef(Index index) const
-    {
-      return m_matrix.const_cast_derived().coeffRef(index);
-    }
-
-    template<typename OtherDerived>
-    void copyCoeff(Index row, Index col, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(row >= 0 && row < rows()
-                         && col >= 0 && col < cols());
-      Scalar& tmp = m_matrix.coeffRef(row,col);
-      tmp = m_functor(tmp, _other.coeff(row,col));
-    }
-
-    template<typename OtherDerived>
-    void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(index >= 0 && index < m_matrix.size());
-      Scalar& tmp = m_matrix.coeffRef(index);
-      tmp = m_functor(tmp, _other.coeff(index));
-    }
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    void copyPacket(Index row, Index col, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      m_matrix.template writePacket<StoreMode>(row, col,
-        m_functor.packetOp(m_matrix.template packet<StoreMode>(row, col),_other.template packet<LoadMode>(row, col)) );
-    }
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    void copyPacket(Index index, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(index >= 0 && index < m_matrix.size());
-      m_matrix.template writePacket<StoreMode>(index,
-        m_functor.packetOp(m_matrix.template packet<StoreMode>(index),_other.template packet<LoadMode>(index)) );
-    }
-
-    // reimplement lazyAssign to handle complex *= real
-    // see CwiseBinaryOp ctor for details
-    template<typename RhsDerived>
-    EIGEN_STRONG_INLINE SelfCwiseBinaryOp& lazyAssign(const DenseBase<RhsDerived>& rhs)
-    {
-      EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Lhs,RhsDerived)
-      EIGEN_CHECK_BINARY_COMPATIBILIY(BinaryOp,typename Lhs::Scalar,typename RhsDerived::Scalar);
-      
-    #ifdef EIGEN_DEBUG_ASSIGN
-      internal::assign_traits<SelfCwiseBinaryOp, RhsDerived>::debug();
-    #endif
-      eigen_assert(rows() == rhs.rows() && cols() == rhs.cols());
-      internal::assign_impl<SelfCwiseBinaryOp, RhsDerived>::run(*this,rhs.derived());
-    #ifndef EIGEN_NO_DEBUG
-      this->checkTransposeAliasing(rhs.derived());
-    #endif
-      return *this;
-    }
-    
-    // overloaded to honor evaluation of special matrices
-    // maybe another solution would be to not use SelfCwiseBinaryOp
-    // at first...
-    SelfCwiseBinaryOp& operator=(const Rhs& _rhs)
-    {
-      typename internal::nested<Rhs>::type rhs(_rhs);
-      return Base::operator=(rhs);
-    }
-
-    Lhs& expression() const 
-    { 
-      return m_matrix;
-    }
-
-    const BinaryOp& functor() const 
-    { 
-      return m_functor;
-    }
-
-  protected:
-    Lhs& m_matrix;
-    const BinaryOp& m_functor;
-
-  private:
-    SelfCwiseBinaryOp& operator=(const SelfCwiseBinaryOp&);
-};
+  internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::add_assign_op<Scalar,Scalar>());
+  return derived();
+}
 
 template<typename Derived>
-inline Derived& DenseBase<Derived>::operator*=(const Scalar& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator-=(const Scalar& other)
 {
-  typedef typename Derived::PlainObject PlainObject;
-  SelfCwiseBinaryOp<internal::scalar_product_op<Scalar>, Derived, typename PlainObject::ConstantReturnType> tmp(derived());
-  tmp = PlainObject::Constant(rows(),cols(),other);
+  internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::sub_assign_op<Scalar,Scalar>());
   return derived();
 }
 
 template<typename Derived>
-inline Derived& DenseBase<Derived>::operator/=(const Scalar& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator/=(const Scalar& other)
 {
-  typedef typename Derived::PlainObject PlainObject;
-  SelfCwiseBinaryOp<internal::scalar_quotient_op<Scalar>, Derived, typename PlainObject::ConstantReturnType> tmp(derived());
-  tmp = PlainObject::Constant(rows(),cols(), other);
+  internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::div_assign_op<Scalar,Scalar>());
   return derived();
 }
 
diff --git a/extern/Eigen3/Eigen/src/Core/Solve.h b/extern/Eigen3/Eigen/src/Core/Solve.h
new file mode 100644
index 00000000000..a8daea51135
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/Solve.h
@@ -0,0 +1,188 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SOLVE_H
+#define EIGEN_SOLVE_H
+
+namespace Eigen {
+
+template<typename Decomposition, typename RhsType, typename StorageKind> class SolveImpl;
+  
+/** \class Solve
+  * \ingroup Core_Module
+  *
+  * \brief Pseudo expression representing a solving operation
+  *
+  * \tparam Decomposition the type of the matrix or decomposion object
+  * \tparam Rhstype the type of the right-hand side
+  *
+  * This class represents an expression of A.solve(B)
+  * and most of the time this is the only way it is used.
+  *
+  */
+namespace internal {
+
+// this solve_traits class permits to determine the evaluation type with respect to storage kind (Dense vs Sparse)
+template<typename Decomposition, typename RhsType,typename StorageKind> struct solve_traits;
+
+template<typename Decomposition, typename RhsType>
+struct solve_traits<Decomposition,RhsType,Dense>
+{
+  typedef typename make_proper_matrix_type<typename RhsType::Scalar,
+                 Decomposition::ColsAtCompileTime,
+                 RhsType::ColsAtCompileTime,
+                 RhsType::PlainObject::Options,
+                 Decomposition::MaxColsAtCompileTime,
+                 RhsType::MaxColsAtCompileTime>::type PlainObject;
+};
+
+template<typename Decomposition, typename RhsType>
+struct traits<Solve<Decomposition, RhsType> >
+  : traits<typename solve_traits<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>::PlainObject>
+{
+  typedef typename solve_traits<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>::PlainObject PlainObject;
+  typedef typename promote_index_type<typename Decomposition::StorageIndex, typename RhsType::StorageIndex>::type StorageIndex;
+  typedef traits<PlainObject> BaseTraits;
+  enum {
+    Flags = BaseTraits::Flags & RowMajorBit,
+    CoeffReadCost = HugeCost
+  };
+};
+
+}
+
+
+template<typename Decomposition, typename RhsType>
+class Solve : public SolveImpl<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>
+{
+public:
+  typedef typename internal::traits<Solve>::PlainObject PlainObject;
+  typedef typename internal::traits<Solve>::StorageIndex StorageIndex;
+  
+  Solve(const Decomposition &dec, const RhsType &rhs)
+    : m_dec(dec), m_rhs(rhs)
+  {}
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return m_dec.cols(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_rhs.cols(); }
+
+  EIGEN_DEVICE_FUNC const Decomposition& dec() const { return m_dec; }
+  EIGEN_DEVICE_FUNC const RhsType&       rhs() const { return m_rhs; }
+
+protected:
+  const Decomposition &m_dec;
+  const RhsType       &m_rhs;
+};
+
+
+// Specialization of the Solve expression for dense results
+template<typename Decomposition, typename RhsType>
+class SolveImpl<Decomposition,RhsType,Dense>
+  : public MatrixBase<Solve<Decomposition,RhsType> >
+{
+  typedef Solve<Decomposition,RhsType> Derived;
+  
+public:
+  
+  typedef MatrixBase<Solve<Decomposition,RhsType> > Base;
+  EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
+
+private:
+  
+  Scalar coeff(Index row, Index col) const;
+  Scalar coeff(Index i) const;
+};
+
+// Generic API dispatcher
+template<typename Decomposition, typename RhsType, typename StorageKind>
+class SolveImpl : public internal::generic_xpr_base<Solve<Decomposition,RhsType>, MatrixXpr, StorageKind>::type
+{
+  public:
+    typedef typename internal::generic_xpr_base<Solve<Decomposition,RhsType>, MatrixXpr, StorageKind>::type Base;
+};
+
+namespace internal {
+
+// Evaluator of Solve -> eval into a temporary
+template<typename Decomposition, typename RhsType>
+struct evaluator<Solve<Decomposition,RhsType> >
+  : public evaluator<typename Solve<Decomposition,RhsType>::PlainObject>
+{
+  typedef Solve<Decomposition,RhsType> SolveType;
+  typedef typename SolveType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  enum { Flags = Base::Flags | EvalBeforeNestingBit };
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const SolveType& solve)
+    : m_result(solve.rows(), solve.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    solve.dec()._solve_impl(solve.rhs(), m_result);
+  }
+  
+protected:  
+  PlainObject m_result;
+};
+
+// Specialization for "dst = dec.solve(rhs)"
+// NOTE we need to specialize it for Dense2Dense to avoid ambiguous specialization error and a Sparse2Sparse specialization must exist somewhere
+template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
+struct Assignment<DstXprType, Solve<DecType,RhsType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
+{
+  typedef Solve<DecType,RhsType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    src.dec()._solve_impl(src.rhs(), dst);
+  }
+};
+
+// Specialization for "dst = dec.transpose().solve(rhs)"
+template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
+struct Assignment<DstXprType, Solve<Transpose<const DecType>,RhsType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
+{
+  typedef Solve<Transpose<const DecType>,RhsType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    src.dec().nestedExpression().template _solve_impl_transposed<false>(src.rhs(), dst);
+  }
+};
+
+// Specialization for "dst = dec.adjoint().solve(rhs)"
+template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
+struct Assignment<DstXprType, Solve<CwiseUnaryOp<internal::scalar_conjugate_op<typename DecType::Scalar>, const Transpose<const DecType> >,RhsType>,
+                  internal::assign_op<Scalar,Scalar>, Dense2Dense>
+{
+  typedef Solve<CwiseUnaryOp<internal::scalar_conjugate_op<typename DecType::Scalar>, const Transpose<const DecType> >,RhsType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    
+    src.dec().nestedExpression().nestedExpression().template _solve_impl_transposed<true>(src.rhs(), dst);
+  }
+};
+
+} // end namepsace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SOLVE_H
diff --git a/extern/Eigen3/Eigen/src/Core/SolveTriangular.h b/extern/Eigen3/Eigen/src/Core/SolveTriangular.h
index ef17f288e29..4652e2e19ff 100644
--- a/extern/Eigen3/Eigen/src/Core/SolveTriangular.h
+++ b/extern/Eigen3/Eigen/src/Core/SolveTriangular.h
@@ -68,7 +68,7 @@ struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,1>
     if(!useRhsDirectly)
       MappedRhs(actualRhs,rhs.size()) = rhs;
 
-    triangular_solve_vector<LhsScalar, RhsScalar, typename Lhs::Index, Side, Mode, LhsProductTraits::NeedToConjugate,
+    triangular_solve_vector<LhsScalar, RhsScalar, Index, Side, Mode, LhsProductTraits::NeedToConjugate,
                             (int(Lhs::Flags) & RowMajorBit) ? RowMajor : ColMajor>
       ::run(actualLhs.cols(), actualLhs.data(), actualLhs.outerStride(), actualRhs);
 
@@ -82,7 +82,6 @@ template<typename Lhs, typename Rhs, int Side, int Mode>
 struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,Dynamic>
 {
   typedef typename Rhs::Scalar Scalar;
-  typedef typename Rhs::Index Index;
   typedef blas_traits<Lhs> LhsProductTraits;
   typedef typename LhsProductTraits::DirectLinearAccessType ActualLhsType;
 
@@ -96,7 +95,7 @@ struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,Dynamic>
     typedef internal::gemm_blocking_space<(Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor,Scalar,Scalar,
               Rhs::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime, Lhs::MaxRowsAtCompileTime,4> BlockingType;
 
-    BlockingType blocking(rhs.rows(), rhs.cols(), size);
+    BlockingType blocking(rhs.rows(), rhs.cols(), size, 1, false);
 
     triangular_solve_matrix<Scalar,Index,Side,Mode,LhsProductTraits::NeedToConjugate,(int(Lhs::Flags) & RowMajorBit) ? RowMajor : ColMajor,
                                (Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor>
@@ -108,32 +107,32 @@ struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,Dynamic>
 * meta-unrolling implementation
 ***************************************************************************/
 
-template<typename Lhs, typename Rhs, int Mode, int Index, int Size,
-         bool Stop = Index==Size>
+template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size,
+         bool Stop = LoopIndex==Size>
 struct triangular_solver_unroller;
 
-template<typename Lhs, typename Rhs, int Mode, int Index, int Size>
-struct triangular_solver_unroller<Lhs,Rhs,Mode,Index,Size,false> {
+template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size>
+struct triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex,Size,false> {
   enum {
     IsLower = ((Mode&Lower)==Lower),
-    I = IsLower ? Index : Size - Index - 1,
-    S = IsLower ? 0     : I+1
+    DiagIndex  = IsLower ? LoopIndex : Size - LoopIndex - 1,
+    StartIndex = IsLower ? 0         : DiagIndex+1
   };
   static void run(const Lhs& lhs, Rhs& rhs)
   {
-    if (Index>0)
-      rhs.coeffRef(I) -= lhs.row(I).template segment<Index>(S).transpose()
-                         .cwiseProduct(rhs.template segment<Index>(S)).sum();
+    if (LoopIndex>0)
+      rhs.coeffRef(DiagIndex) -= lhs.row(DiagIndex).template segment<LoopIndex>(StartIndex).transpose()
+                                .cwiseProduct(rhs.template segment<LoopIndex>(StartIndex)).sum();
 
     if(!(Mode & UnitDiag))
-      rhs.coeffRef(I) /= lhs.coeff(I,I);
+      rhs.coeffRef(DiagIndex) /= lhs.coeff(DiagIndex,DiagIndex);
 
-    triangular_solver_unroller<Lhs,Rhs,Mode,Index+1,Size>::run(lhs,rhs);
+    triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex+1,Size>::run(lhs,rhs);
   }
 };
 
-template<typename Lhs, typename Rhs, int Mode, int Index, int Size>
-struct triangular_solver_unroller<Lhs,Rhs,Mode,Index,Size,true> {
+template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size>
+struct triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex,Size,true> {
   static void run(const Lhs&, Rhs&) {}
 };
 
@@ -162,61 +161,38 @@ struct triangular_solver_selector<Lhs,Rhs,OnTheRight,Mode,CompleteUnrolling,1> {
 * TriangularView methods
 ***************************************************************************/
 
-/** "in-place" version of TriangularView::solve() where the result is written in \a other
-  *
-  * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
-  * This function will const_cast it, so constness isn't honored here.
-  *
-  * See TriangularView:solve() for the details.
-  */
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 template<typename MatrixType, unsigned int Mode>
 template<int Side, typename OtherDerived>
-void TriangularView<MatrixType,Mode>::solveInPlace(const MatrixBase<OtherDerived>& _other) const
+void TriangularViewImpl<MatrixType,Mode,Dense>::solveInPlace(const MatrixBase<OtherDerived>& _other) const
 {
   OtherDerived& other = _other.const_cast_derived();
-  eigen_assert( cols() == rows() && ((Side==OnTheLeft && cols() == other.rows()) || (Side==OnTheRight && cols() == other.cols())) );
+  eigen_assert( derived().cols() == derived().rows() && ((Side==OnTheLeft && derived().cols() == other.rows()) || (Side==OnTheRight && derived().cols() == other.cols())) );
   eigen_assert((!(Mode & ZeroDiag)) && bool(Mode & (Upper|Lower)));
+  // If solving for a 0x0 matrix, nothing to do, simply return.
+  if (derived().cols() == 0)
+    return;
 
-  enum { copy = internal::traits<OtherDerived>::Flags & RowMajorBit  && OtherDerived::IsVectorAtCompileTime };
+  enum { copy = (internal::traits<OtherDerived>::Flags & RowMajorBit)  && OtherDerived::IsVectorAtCompileTime && OtherDerived::SizeAtCompileTime!=1};
   typedef typename internal::conditional<copy,
     typename internal::plain_matrix_type_column_major<OtherDerived>::type, OtherDerived&>::type OtherCopy;
   OtherCopy otherCopy(other);
 
   internal::triangular_solver_selector<MatrixType, typename internal::remove_reference<OtherCopy>::type,
-    Side, Mode>::run(nestedExpression(), otherCopy);
+    Side, Mode>::run(derived().nestedExpression(), otherCopy);
 
   if (copy)
     other = otherCopy;
 }
 
-/** \returns the product of the inverse of \c *this with \a other, \a *this being triangular.
-  *
-  * This function computes the inverse-matrix matrix product inverse(\c *this) * \a other if
-  * \a Side==OnTheLeft (the default), or the right-inverse-multiply  \a other * inverse(\c *this) if
-  * \a Side==OnTheRight.
-  *
-  * The matrix \c *this must be triangular and invertible (i.e., all the coefficients of the
-  * diagonal must be non zero). It works as a forward (resp. backward) substitution if \c *this
-  * is an upper (resp. lower) triangular matrix.
-  *
-  * Example: \include MatrixBase_marked.cpp
-  * Output: \verbinclude MatrixBase_marked.out
-  *
-  * This function returns an expression of the inverse-multiply and can works in-place if it is assigned
-  * to the same matrix or vector \a other.
-  *
-  * For users coming from BLAS, this function (and more specifically solveInPlace()) offer
-  * all the operations supported by the \c *TRSV and \c *TRSM BLAS routines.
-  *
-  * \sa TriangularView::solveInPlace()
-  */
 template<typename Derived, unsigned int Mode>
 template<int Side, typename Other>
 const internal::triangular_solve_retval<Side,TriangularView<Derived,Mode>,Other>
-TriangularView<Derived,Mode>::solve(const MatrixBase<Other>& other) const
+TriangularViewImpl<Derived,Mode,Dense>::solve(const MatrixBase<Other>& other) const
 {
-  return internal::triangular_solve_retval<Side,TriangularView,Other>(*this, other.derived());
+  return internal::triangular_solve_retval<Side,TriangularViewType,Other>(derived(), other.derived());
 }
+#endif
 
 namespace internal {
 
@@ -232,7 +208,6 @@ template<int Side, typename TriangularType, typename Rhs> struct triangular_solv
 {
   typedef typename remove_all<typename Rhs::Nested>::type RhsNestedCleaned;
   typedef ReturnByValue<triangular_solve_retval> Base;
-  typedef typename Base::Index Index;
 
   triangular_solve_retval(const TriangularType& tri, const Rhs& rhs)
     : m_triangularMatrix(tri), m_rhs(rhs)
@@ -243,7 +218,7 @@ template<int Side, typename TriangularType, typename Rhs> struct triangular_solv
 
   template<typename Dest> inline void evalTo(Dest& dst) const
   {
-    if(!(is_same<RhsNestedCleaned,Dest>::value && extract_data(dst) == extract_data(m_rhs)))
+    if(!is_same_dense(dst,m_rhs))
       dst = m_rhs;
     m_triangularMatrix.template solveInPlace<Side>(dst);
   }
diff --git a/extern/Eigen3/Eigen/src/Core/SolverBase.h b/extern/Eigen3/Eigen/src/Core/SolverBase.h
new file mode 100644
index 00000000000..8a4adc22973
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/SolverBase.h
@@ -0,0 +1,130 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SOLVERBASE_H
+#define EIGEN_SOLVERBASE_H
+
+namespace Eigen {
+
+namespace internal {
+
+
+
+} // end namespace internal
+
+/** \class SolverBase
+  * \brief A base class for matrix decomposition and solvers
+  *
+  * \tparam Derived the actual type of the decomposition/solver.
+  *
+  * Any matrix decomposition inheriting this base class provide the following API:
+  *
+  * \code
+  * MatrixType A, b, x;
+  * DecompositionType dec(A);
+  * x = dec.solve(b);             // solve A   * x = b
+  * x = dec.transpose().solve(b); // solve A^T * x = b
+  * x = dec.adjoint().solve(b);   // solve A'  * x = b
+  * \endcode
+  *
+  * \warning Currently, any other usage of transpose() and adjoint() are not supported and will produce compilation errors.
+  *
+  * \sa class PartialPivLU, class FullPivLU
+  */
+template<typename Derived>
+class SolverBase : public EigenBase<Derived>
+{
+  public:
+
+    typedef EigenBase<Derived> Base;
+    typedef typename internal::traits<Derived>::Scalar Scalar;
+    typedef Scalar CoeffReturnType;
+
+    enum {
+      RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
+      ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
+      SizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::RowsAtCompileTime,
+                                                          internal::traits<Derived>::ColsAtCompileTime>::ret),
+      MaxRowsAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime,
+      MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime,
+      MaxSizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::MaxRowsAtCompileTime,
+                                                             internal::traits<Derived>::MaxColsAtCompileTime>::ret),
+      IsVectorAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime == 1
+                           || internal::traits<Derived>::MaxColsAtCompileTime == 1
+    };
+
+    /** Default constructor */
+    SolverBase()
+    {}
+
+    ~SolverBase()
+    {}
+
+    using Base::derived;
+
+    /** \returns an expression of the solution x of \f$ A x = b \f$ using the current decomposition of A.
+      */
+    template<typename Rhs>
+    inline const Solve<Derived, Rhs>
+    solve(const MatrixBase<Rhs>& b) const
+    {
+      eigen_assert(derived().rows()==b.rows() && "solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<Derived, Rhs>(derived(), b.derived());
+    }
+
+    /** \internal the return type of transpose() */
+    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
+    /** \returns an expression of the transposed of the factored matrix.
+      *
+      * A typical usage is to solve for the transposed problem A^T x = b:
+      * \code x = dec.transpose().solve(b); \endcode
+      *
+      * \sa adjoint(), solve()
+      */
+    inline ConstTransposeReturnType transpose() const
+    {
+      return ConstTransposeReturnType(derived());
+    }
+
+    /** \internal the return type of adjoint() */
+    typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
+                        CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, ConstTransposeReturnType>,
+                        ConstTransposeReturnType
+                     >::type AdjointReturnType;
+    /** \returns an expression of the adjoint of the factored matrix
+      *
+      * A typical usage is to solve for the adjoint problem A' x = b:
+      * \code x = dec.adjoint().solve(b); \endcode
+      *
+      * For real scalar types, this function is equivalent to transpose().
+      *
+      * \sa transpose(), solve()
+      */
+    inline AdjointReturnType adjoint() const
+    {
+      return AdjointReturnType(derived().transpose());
+    }
+
+  protected:
+};
+
+namespace internal {
+
+template<typename Derived>
+struct generic_xpr_base<Derived, MatrixXpr, SolverStorage>
+{
+  typedef SolverBase<Derived> type;
+
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SOLVERBASE_H
diff --git a/extern/Eigen3/Eigen/src/Core/StableNorm.h b/extern/Eigen3/Eigen/src/Core/StableNorm.h
index 389d9427539..88c8d989024 100644
--- a/extern/Eigen3/Eigen/src/Core/StableNorm.h
+++ b/extern/Eigen3/Eigen/src/Core/StableNorm.h
@@ -17,10 +17,9 @@ namespace internal {
 template<typename ExpressionType, typename Scalar>
 inline void stable_norm_kernel(const ExpressionType& bl, Scalar& ssq, Scalar& scale, Scalar& invScale)
 {
-  using std::max;
   Scalar maxCoeff = bl.cwiseAbs().maxCoeff();
   
-  if (maxCoeff>scale)
+  if(maxCoeff>scale)
   {
     ssq = ssq * numext::abs2(scale/maxCoeff);
     Scalar tmp = Scalar(1)/maxCoeff;
@@ -29,12 +28,21 @@ inline void stable_norm_kernel(const ExpressionType& bl, Scalar& ssq, Scalar& sc
       invScale = NumTraits<Scalar>::highest();
       scale = Scalar(1)/invScale;
     }
+    else if(maxCoeff>NumTraits<Scalar>::highest()) // we got a INF
+    {
+      invScale = Scalar(1);
+      scale = maxCoeff;
+    }
     else
     {
       scale = maxCoeff;
       invScale = tmp;
     }
   }
+  else if(maxCoeff!=maxCoeff) // we got a NaN
+  {
+    scale = maxCoeff;
+  }
   
   // TODO if the maxCoeff is much much smaller than the current scale,
   // then we can neglect this sub vector
@@ -47,15 +55,12 @@ inline typename NumTraits<typename traits<Derived>::Scalar>::Real
 blueNorm_impl(const EigenBase<Derived>& _vec)
 {
   typedef typename Derived::RealScalar RealScalar;  
-  typedef typename Derived::Index Index;
   using std::pow;
-  using std::min;
-  using std::max;
   using std::sqrt;
   using std::abs;
   const Derived& vec(_vec.derived());
   static bool initialized = false;
-  static RealScalar b1, b2, s1m, s2m, overfl, rbig, relerr;
+  static RealScalar b1, b2, s1m, s2m, rbig, relerr;
   if(!initialized)
   {
     int ibeta, it, iemin, iemax, iexp;
@@ -84,7 +89,6 @@ blueNorm_impl(const EigenBase<Derived>& _vec)
     iexp  = - ((iemax+it)/2);
     s2m   = RealScalar(pow(RealScalar(ibeta),RealScalar(iexp)));    // scaling factor for upper range
 
-    overfl  = rbig*s2m;                                             // overflow boundary for abig
     eps     = RealScalar(pow(double(ibeta), 1-it));
     relerr  = sqrt(eps);                                            // tolerance for neglecting asml
     initialized = true;
@@ -101,13 +105,13 @@ blueNorm_impl(const EigenBase<Derived>& _vec)
     else if(ax < b1) asml += numext::abs2(ax*s1m);
     else             amed += numext::abs2(ax);
   }
+  if(amed!=amed)
+    return amed;  // we got a NaN
   if(abig > RealScalar(0))
   {
     abig = sqrt(abig);
-    if(abig > overfl)
-    {
-      return rbig;
-    }
+    if(abig > rbig) // overflow, or *this contains INF values
+      return abig;  // return INF
     if(amed > RealScalar(0))
     {
       abig = abig/s2m;
@@ -128,8 +132,8 @@ blueNorm_impl(const EigenBase<Derived>& _vec)
   }
   else
     return sqrt(amed);
-  asml = (min)(abig, amed);
-  abig = (max)(abig, amed);
+  asml = numext::mini(abig, amed);
+  abig = numext::maxi(abig, amed);
   if(asml <= abig*relerr)
     return abig;
   else
@@ -152,21 +156,35 @@ template<typename Derived>
 inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
 MatrixBase<Derived>::stableNorm() const
 {
-  using std::min;
   using std::sqrt;
+  using std::abs;
   const Index blockSize = 4096;
   RealScalar scale(0);
   RealScalar invScale(1);
   RealScalar ssq(0); // sum of square
+  
+  typedef typename internal::nested_eval<Derived,2>::type DerivedCopy;
+  typedef typename internal::remove_all<DerivedCopy>::type DerivedCopyClean;
+  const DerivedCopy copy(derived());
+  
   enum {
-    Alignment = (int(Flags)&DirectAccessBit) || (int(Flags)&AlignedBit) ? 1 : 0
+    CanAlign = (   (int(DerivedCopyClean::Flags)&DirectAccessBit)
+                || (int(internal::evaluator<DerivedCopyClean>::Alignment)>0) // FIXME Alignment)>0 might not be enough
+               ) && (blockSize*sizeof(Scalar)*2<EIGEN_STACK_ALLOCATION_LIMIT)
+                 && (EIGEN_MAX_STATIC_ALIGN_BYTES>0) // if we cannot allocate on the stack, then let's not bother about this optimization
   };
+  typedef typename internal::conditional<CanAlign, Ref<const Matrix<Scalar,Dynamic,1,0,blockSize,1>, internal::evaluator<DerivedCopyClean>::Alignment>,
+                                                   typename DerivedCopyClean::ConstSegmentReturnType>::type SegmentWrapper;
   Index n = size();
-  Index bi = internal::first_aligned(derived());
+  
+  if(n==1)
+    return abs(this->coeff(0));
+  
+  Index bi = internal::first_default_aligned(copy);
   if (bi>0)
-    internal::stable_norm_kernel(this->head(bi), ssq, scale, invScale);
+    internal::stable_norm_kernel(copy.head(bi), ssq, scale, invScale);
   for (; bi<n; bi+=blockSize)
-    internal::stable_norm_kernel(this->segment(bi,(min)(blockSize, n - bi)).template forceAlignedAccessIf<Alignment>(), ssq, scale, invScale);
+    internal::stable_norm_kernel(SegmentWrapper(copy.segment(bi,numext::mini(blockSize, n - bi))), ssq, scale, invScale);
   return scale * sqrt(ssq);
 }
 
diff --git a/extern/Eigen3/Eigen/src/Core/Stride.h b/extern/Eigen3/Eigen/src/Core/Stride.h
index 1e3f5fe9fff..513742f34b9 100644
--- a/extern/Eigen3/Eigen/src/Core/Stride.h
+++ b/extern/Eigen3/Eigen/src/Core/Stride.h
@@ -31,8 +31,8 @@ namespace Eigen {
   * arguments to the constructor.
   *
   * Indeed, this class takes two template parameters:
-  *  \param _OuterStrideAtCompileTime the outer stride, or Dynamic if you want to specify it at runtime.
-  *  \param _InnerStrideAtCompileTime the inner stride, or Dynamic if you want to specify it at runtime.
+  *  \tparam _OuterStrideAtCompileTime the outer stride, or Dynamic if you want to specify it at runtime.
+  *  \tparam _InnerStrideAtCompileTime the inner stride, or Dynamic if you want to specify it at runtime.
   *
   * Here is an example:
   * \include Map_general_stride.cpp
@@ -44,13 +44,14 @@ template<int _OuterStrideAtCompileTime, int _InnerStrideAtCompileTime>
 class Stride
 {
   public:
-    typedef DenseIndex Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
     enum {
       InnerStrideAtCompileTime = _InnerStrideAtCompileTime,
       OuterStrideAtCompileTime = _OuterStrideAtCompileTime
     };
 
     /** Default constructor, for use when strides are fixed at compile time */
+    EIGEN_DEVICE_FUNC
     Stride()
       : m_outer(OuterStrideAtCompileTime), m_inner(InnerStrideAtCompileTime)
     {
@@ -58,6 +59,7 @@ class Stride
     }
 
     /** Constructor allowing to pass the strides at runtime */
+    EIGEN_DEVICE_FUNC
     Stride(Index outerStride, Index innerStride)
       : m_outer(outerStride), m_inner(innerStride)
     {
@@ -65,13 +67,16 @@ class Stride
     }
 
     /** Copy constructor */
+    EIGEN_DEVICE_FUNC
     Stride(const Stride& other)
       : m_outer(other.outer()), m_inner(other.inner())
     {}
 
     /** \returns the outer stride */
+    EIGEN_DEVICE_FUNC
     inline Index outer() const { return m_outer.value(); }
     /** \returns the inner stride */
+    EIGEN_DEVICE_FUNC
     inline Index inner() const { return m_inner.value(); }
 
   protected:
@@ -81,26 +86,24 @@ class Stride
 
 /** \brief Convenience specialization of Stride to specify only an inner stride
   * See class Map for some examples */
-template<int Value = Dynamic>
+template<int Value>
 class InnerStride : public Stride<0, Value>
 {
     typedef Stride<0, Value> Base;
   public:
-    typedef DenseIndex Index;
-    InnerStride() : Base() {}
-    InnerStride(Index v) : Base(0, v) {}
+    EIGEN_DEVICE_FUNC InnerStride() : Base() {}
+    EIGEN_DEVICE_FUNC InnerStride(Index v) : Base(0, v) {} // FIXME making this explicit could break valid code
 };
 
 /** \brief Convenience specialization of Stride to specify only an outer stride
   * See class Map for some examples */
-template<int Value = Dynamic>
+template<int Value>
 class OuterStride : public Stride<Value, 0>
 {
     typedef Stride<Value, 0> Base;
   public:
-    typedef DenseIndex Index;
-    OuterStride() : Base() {}
-    OuterStride(Index v) : Base(v,0) {}
+    EIGEN_DEVICE_FUNC OuterStride() : Base() {}
+    EIGEN_DEVICE_FUNC OuterStride(Index v) : Base(v,0) {} // FIXME making this explicit could break valid code
 };
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/Swap.h b/extern/Eigen3/Eigen/src/Core/Swap.h
index bf58bd5997d..d702009185e 100644
--- a/extern/Eigen3/Eigen/src/Core/Swap.h
+++ b/extern/Eigen3/Eigen/src/Core/Swap.h
@@ -12,115 +12,56 @@
 
 namespace Eigen { 
 
-/** \class SwapWrapper
-  * \ingroup Core_Module
-  *
-  * \internal
-  *
-  * \brief Internal helper class for swapping two expressions
-  */
 namespace internal {
-template<typename ExpressionType>
-struct traits<SwapWrapper<ExpressionType> > : traits<ExpressionType> {};
-}
 
-template<typename ExpressionType> class SwapWrapper
-  : public internal::dense_xpr_base<SwapWrapper<ExpressionType> >::type
+// Overload default assignPacket behavior for swapping them
+template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT>
+class generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, Specialized>
+ : public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, BuiltIn>
 {
-  public:
-
-    typedef typename internal::dense_xpr_base<SwapWrapper>::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(SwapWrapper)
-    typedef typename internal::packet_traits<Scalar>::type Packet;
-
-    inline SwapWrapper(ExpressionType& xpr) : m_expression(xpr) {}
-
-    inline Index rows() const { return m_expression.rows(); }
-    inline Index cols() const { return m_expression.cols(); }
-    inline Index outerStride() const { return m_expression.outerStride(); }
-    inline Index innerStride() const { return m_expression.innerStride(); }
-    
-    typedef typename internal::conditional<
-                       internal::is_lvalue<ExpressionType>::value,
-                       Scalar,
-                       const Scalar
-                     >::type ScalarWithConstIfNotLvalue;
-                     
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
-    inline const Scalar* data() const { return m_expression.data(); }
-
-    inline Scalar& coeffRef(Index rowId, Index colId)
-    {
-      return m_expression.const_cast_derived().coeffRef(rowId, colId);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_expression.const_cast_derived().coeffRef(index);
-    }
-
-    inline Scalar& coeffRef(Index rowId, Index colId) const
-    {
-      return m_expression.coeffRef(rowId, colId);
-    }
-
-    inline Scalar& coeffRef(Index index) const
-    {
-      return m_expression.coeffRef(index);
-    }
-
-    template<typename OtherDerived>
-    void copyCoeff(Index rowId, Index colId, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(rowId >= 0 && rowId < rows()
-                         && colId >= 0 && colId < cols());
-      Scalar tmp = m_expression.coeff(rowId, colId);
-      m_expression.coeffRef(rowId, colId) = _other.coeff(rowId, colId);
-      _other.coeffRef(rowId, colId) = tmp;
-    }
-
-    template<typename OtherDerived>
-    void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(index >= 0 && index < m_expression.size());
-      Scalar tmp = m_expression.coeff(index);
-      m_expression.coeffRef(index) = _other.coeff(index);
-      _other.coeffRef(index) = tmp;
-    }
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    void copyPacket(Index rowId, Index colId, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(rowId >= 0 && rowId < rows()
-                        && colId >= 0 && colId < cols());
-      Packet tmp = m_expression.template packet<StoreMode>(rowId, colId);
-      m_expression.template writePacket<StoreMode>(rowId, colId,
-        _other.template packet<LoadMode>(rowId, colId)
-      );
-      _other.template writePacket<LoadMode>(rowId, colId, tmp);
-    }
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    void copyPacket(Index index, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(index >= 0 && index < m_expression.size());
-      Packet tmp = m_expression.template packet<StoreMode>(index);
-      m_expression.template writePacket<StoreMode>(index,
-        _other.template packet<LoadMode>(index)
-      );
-      _other.template writePacket<LoadMode>(index, tmp);
-    }
-
-    ExpressionType& expression() const { return m_expression; }
-
-  protected:
-    ExpressionType& m_expression;
+protected:
+  typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, BuiltIn> Base;
+  using Base::m_dst;
+  using Base::m_src;
+  using Base::m_functor;
+  
+public:
+  typedef typename Base::Scalar Scalar;
+  typedef typename Base::DstXprType DstXprType;
+  typedef swap_assign_op<Scalar> Functor;
+  
+  EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr)
+    : Base(dst, src, func, dstExpr)
+  {}
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  void assignPacket(Index row, Index col)
+  {
+    PacketType tmp = m_src.template packet<LoadMode,PacketType>(row,col);
+    const_cast<SrcEvaluatorTypeT&>(m_src).template writePacket<LoadMode>(row,col, m_dst.template packet<StoreMode,PacketType>(row,col));
+    m_dst.template writePacket<StoreMode>(row,col,tmp);
+  }
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  void assignPacket(Index index)
+  {
+    PacketType tmp = m_src.template packet<LoadMode,PacketType>(index);
+    const_cast<SrcEvaluatorTypeT&>(m_src).template writePacket<LoadMode>(index, m_dst.template packet<StoreMode,PacketType>(index));
+    m_dst.template writePacket<StoreMode>(index,tmp);
+  }
+  
+  // TODO find a simple way not to have to copy/paste this function from generic_dense_assignment_kernel, by simple I mean no CRTP (Gael)
+  template<int StoreMode, int LoadMode, typename PacketType>
+  void assignPacketByOuterInner(Index outer, Index inner)
+  {
+    Index row = Base::rowIndexByOuterInner(outer, inner); 
+    Index col = Base::colIndexByOuterInner(outer, inner);
+    assignPacket<StoreMode,LoadMode,PacketType>(row, col);
+  }
 };
 
+} // namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SWAP_H
diff --git a/extern/Eigen3/Eigen/src/Core/Transpose.h b/extern/Eigen3/Eigen/src/Core/Transpose.h
index 22096ea2fde..79b767bccae 100644
--- a/extern/Eigen3/Eigen/src/Core/Transpose.h
+++ b/extern/Eigen3/Eigen/src/Core/Transpose.h
@@ -2,7 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -13,39 +13,21 @@
 
 namespace Eigen { 
 
-/** \class Transpose
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the transpose of a matrix
-  *
-  * \param MatrixType the type of the object of which we are taking the transpose
-  *
-  * This class represents an expression of the transpose of a matrix.
-  * It is the return type of MatrixBase::transpose() and MatrixBase::adjoint()
-  * and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::transpose(), MatrixBase::adjoint()
-  */
-
 namespace internal {
 template<typename MatrixType>
-struct traits<Transpose<MatrixType> > : traits<MatrixType>
+struct traits<Transpose<MatrixType> > : public traits<MatrixType>
 {
-  typedef typename MatrixType::Scalar Scalar;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type MatrixTypeNestedPlain;
-  typedef typename traits<MatrixType>::StorageKind StorageKind;
-  typedef typename traits<MatrixType>::XprKind XprKind;
   enum {
     RowsAtCompileTime = MatrixType::ColsAtCompileTime,
     ColsAtCompileTime = MatrixType::RowsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxColsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
-    Flags0 = MatrixTypeNestedPlain::Flags & ~(LvalueBit | NestByRefBit),
+    Flags0 = traits<MatrixTypeNestedPlain>::Flags & ~(LvalueBit | NestByRefBit),
     Flags1 = Flags0 | FlagsLvalueBit,
     Flags = Flags1 ^ RowMajorBit,
-    CoeffReadCost = MatrixTypeNestedPlain::CoeffReadCost,
     InnerStrideAtCompileTime = inner_stride_at_compile_time<MatrixType>::ret,
     OuterStrideAtCompileTime = outer_stride_at_compile_time<MatrixType>::ret
   };
@@ -54,31 +36,55 @@ struct traits<Transpose<MatrixType> > : traits<MatrixType>
 
 template<typename MatrixType, typename StorageKind> class TransposeImpl;
 
+/** \class Transpose
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the transpose of a matrix
+  *
+  * \tparam MatrixType the type of the object of which we are taking the transpose
+  *
+  * This class represents an expression of the transpose of a matrix.
+  * It is the return type of MatrixBase::transpose() and MatrixBase::adjoint()
+  * and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::transpose(), MatrixBase::adjoint()
+  */
 template<typename MatrixType> class Transpose
   : public TransposeImpl<MatrixType,typename internal::traits<MatrixType>::StorageKind>
 {
   public:
 
+    typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
+
     typedef typename TransposeImpl<MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(Transpose)
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
-    inline Transpose(MatrixType& a_matrix) : m_matrix(a_matrix) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline Transpose(MatrixType& matrix) : m_matrix(matrix) {}
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Transpose)
 
-    inline Index rows() const { return m_matrix.cols(); }
-    inline Index cols() const { return m_matrix.rows(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_matrix.cols(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_matrix.rows(); }
 
     /** \returns the nested expression */
-    const typename internal::remove_all<typename MatrixType::Nested>::type&
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<MatrixTypeNested>::type&
     nestedExpression() const { return m_matrix; }
 
     /** \returns the nested expression */
-    typename internal::remove_all<typename MatrixType::Nested>::type&
-    nestedExpression() { return m_matrix.const_cast_derived(); }
+    EIGEN_DEVICE_FUNC
+    typename internal::remove_reference<MatrixTypeNested>::type&
+    nestedExpression() { return m_matrix; }
+
+    /** \internal */
+    void resize(Index nrows, Index ncols) {
+      m_matrix.resize(ncols,nrows);
+    }
 
   protected:
-    typename MatrixType::Nested m_matrix;
+    typename internal::ref_selector<MatrixType>::non_const_type m_matrix;
 };
 
 namespace internal {
@@ -97,17 +103,27 @@ struct TransposeImpl_base<MatrixType, false>
 
 } // end namespace internal
 
+// Generic API dispatcher
+template<typename XprType, typename StorageKind>
+class TransposeImpl
+  : public internal::generic_xpr_base<Transpose<XprType> >::type
+{
+public:
+  typedef typename internal::generic_xpr_base<Transpose<XprType> >::type Base;
+};
+
 template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
   : public internal::TransposeImpl_base<MatrixType>::type
 {
   public:
 
     typedef typename internal::TransposeImpl_base<MatrixType>::type Base;
+    using Base::coeffRef;
     EIGEN_DENSE_PUBLIC_INTERFACE(Transpose<MatrixType>)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(TransposeImpl)
 
-    inline Index innerStride() const { return derived().nestedExpression().innerStride(); }
-    inline Index outerStride() const { return derived().nestedExpression().outerStride(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return derived().nestedExpression().innerStride(); }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return derived().nestedExpression().outerStride(); }
 
     typedef typename internal::conditional<
                        internal::is_lvalue<MatrixType>::value,
@@ -115,64 +131,21 @@ template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    inline ScalarWithConstIfNotLvalue* data() { return derived().nestedExpression().data(); }
-    inline const Scalar* data() const { return derived().nestedExpression().data(); }
-
-    inline ScalarWithConstIfNotLvalue& coeffRef(Index rowId, Index colId)
-    {
-      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return derived().nestedExpression().const_cast_derived().coeffRef(colId, rowId);
-    }
-
-    inline ScalarWithConstIfNotLvalue& coeffRef(Index index)
-    {
-      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return derived().nestedExpression().const_cast_derived().coeffRef(index);
-    }
+    EIGEN_DEVICE_FUNC inline ScalarWithConstIfNotLvalue* data() { return derived().nestedExpression().data(); }
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return derived().nestedExpression().data(); }
 
+    // FIXME: shall we keep the const version of coeffRef?
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
       return derived().nestedExpression().coeffRef(colId, rowId);
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
       return derived().nestedExpression().coeffRef(index);
     }
-
-    inline CoeffReturnType coeff(Index rowId, Index colId) const
-    {
-      return derived().nestedExpression().coeff(colId, rowId);
-    }
-
-    inline CoeffReturnType coeff(Index index) const
-    {
-      return derived().nestedExpression().coeff(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index rowId, Index colId) const
-    {
-      return derived().nestedExpression().template packet<LoadMode>(colId, rowId);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index rowId, Index colId, const PacketScalar& x)
-    {
-      derived().nestedExpression().const_cast_derived().template writePacket<LoadMode>(colId, rowId, x);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return derived().nestedExpression().template packet<LoadMode>(index);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& x)
-    {
-      derived().nestedExpression().const_cast_derived().template writePacket<LoadMode>(index, x);
-    }
 };
 
 /** \returns an expression of the transpose of *this.
@@ -198,7 +171,7 @@ template<typename Derived>
 inline Transpose<Derived>
 DenseBase<Derived>::transpose()
 {
-  return derived();
+  return TransposeReturnType(derived());
 }
 
 /** This is the const version of transpose().
@@ -236,8 +209,7 @@ template<typename Derived>
 inline const typename MatrixBase<Derived>::AdjointReturnType
 MatrixBase<Derived>::adjoint() const
 {
-  return this->transpose(); // in the complex case, the .conjugate() is be implicit here
-                            // due to implicit conversion to return type
+  return AdjointReturnType(this->transpose());
 }
 
 /***************************************************************************
@@ -247,18 +219,38 @@ MatrixBase<Derived>::adjoint() const
 namespace internal {
 
 template<typename MatrixType,
-  bool IsSquare = (MatrixType::RowsAtCompileTime == MatrixType::ColsAtCompileTime) && MatrixType::RowsAtCompileTime!=Dynamic>
+  bool IsSquare = (MatrixType::RowsAtCompileTime == MatrixType::ColsAtCompileTime) && MatrixType::RowsAtCompileTime!=Dynamic,
+  bool MatchPacketSize =
+        (int(MatrixType::RowsAtCompileTime) == int(internal::packet_traits<typename MatrixType::Scalar>::size))
+    &&  (internal::evaluator<MatrixType>::Flags&PacketAccessBit) >
 struct inplace_transpose_selector;
 
 template<typename MatrixType>
-struct inplace_transpose_selector<MatrixType,true> { // square matrix
+struct inplace_transpose_selector<MatrixType,true,false> { // square matrix
   static void run(MatrixType& m) {
     m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose());
   }
 };
 
+// TODO: vectorized path is currently limited to LargestPacketSize x LargestPacketSize cases only.
 template<typename MatrixType>
-struct inplace_transpose_selector<MatrixType,false> { // non square matrix
+struct inplace_transpose_selector<MatrixType,true,true> { // PacketSize x PacketSize
+  static void run(MatrixType& m) {
+    typedef typename MatrixType::Scalar Scalar;
+    typedef typename internal::packet_traits<typename MatrixType::Scalar>::type Packet;
+    const Index PacketSize = internal::packet_traits<Scalar>::size;
+    const Index Alignment = internal::evaluator<MatrixType>::Alignment;
+    PacketBlock<Packet> A;
+    for (Index i=0; i<PacketSize; ++i)
+      A.packet[i] = m.template packetByOuterInner<Alignment>(i,0);
+    internal::ptranspose(A);
+    for (Index i=0; i<PacketSize; ++i)
+      m.template writePacket<Alignment>(m.rowIndexByOuterInner(i,0), m.colIndexByOuterInner(i,0), A.packet[i]);
+  }
+};
+
+template<typename MatrixType,bool MatchPacketSize>
+struct inplace_transpose_selector<MatrixType,false,MatchPacketSize> { // non square matrix
   static void run(MatrixType& m) {
     if (m.rows()==m.cols())
       m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose());
@@ -331,14 +323,6 @@ inline void MatrixBase<Derived>::adjointInPlace()
 
 namespace internal {
 
-template<typename BinOp,typename NestedXpr,typename Rhs>
-struct blas_traits<SelfCwiseBinaryOp<BinOp,NestedXpr,Rhs> >
- : blas_traits<NestedXpr>
-{
-  typedef SelfCwiseBinaryOp<BinOp,NestedXpr,Rhs> XprType;
-  static inline const XprType extract(const XprType& x) { return x; }
-};
-
 template<bool DestIsTransposed, typename OtherDerived>
 struct check_transpose_aliasing_compile_time_selector
 {
@@ -404,15 +388,15 @@ struct checkTransposeAliasing_impl<Derived, OtherDerived, false>
     }
 };
 
-} // end namespace internal
-
-template<typename Derived>
-template<typename OtherDerived>
-void DenseBase<Derived>::checkTransposeAliasing(const OtherDerived& other) const
+template<typename Dst, typename Src>
+void check_for_aliasing(const Dst &dst, const Src &src)
 {
-    internal::checkTransposeAliasing_impl<Derived, OtherDerived>::run(derived(), other);
+  internal::checkTransposeAliasing_impl<Dst, Src>::run(dst, src);
 }
-#endif
+
+} // end namespace internal
+
+#endif // EIGEN_NO_DEBUG
 
 } // end namespace Eigen
 
diff --git a/extern/Eigen3/Eigen/src/Core/Transpositions.h b/extern/Eigen3/Eigen/src/Core/Transpositions.h
index e4ba0756fa6..86da5af5936 100644
--- a/extern/Eigen3/Eigen/src/Core/Transpositions.h
+++ b/extern/Eigen3/Eigen/src/Core/Transpositions.h
@@ -12,39 +12,6 @@
 
 namespace Eigen { 
 
-/** \class Transpositions
-  * \ingroup Core_Module
-  *
-  * \brief Represents a sequence of transpositions (row/column interchange)
-  *
-  * \param SizeAtCompileTime the number of transpositions, or Dynamic
-  * \param MaxSizeAtCompileTime the maximum number of transpositions, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
-  *
-  * This class represents a permutation transformation as a sequence of \em n transpositions
-  * \f$[T_{n-1} \ldots T_{i} \ldots T_{0}]\f$. It is internally stored as a vector of integers \c indices.
-  * Each transposition \f$ T_{i} \f$ applied on the left of a matrix (\f$ T_{i} M\f$) interchanges
-  * the rows \c i and \c indices[i] of the matrix \c M.
-  * A transposition applied on the right (e.g., \f$ M T_{i}\f$) yields a column interchange.
-  *
-  * Compared to the class PermutationMatrix, such a sequence of transpositions is what is
-  * computed during a decomposition with pivoting, and it is faster when applying the permutation in-place.
-  * 
-  * To apply a sequence of transpositions to a matrix, simply use the operator * as in the following example:
-  * \code
-  * Transpositions tr;
-  * MatrixXf mat;
-  * mat = tr * mat;
-  * \endcode
-  * In this example, we detect that the matrix appears on both side, and so the transpositions
-  * are applied in-place without any temporary or extra copy.
-  *
-  * \sa class PermutationMatrix
-  */
-
-namespace internal {
-template<typename TranspositionType, typename MatrixType, int Side, bool Transposed=false> struct transposition_matrix_product_retval;
-}
-
 template<typename Derived>
 class TranspositionsBase
 {
@@ -53,7 +20,8 @@ class TranspositionsBase
   public:
 
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     Derived& derived() { return *static_cast<Derived*>(this); }
     const Derived& derived() const { return *static_cast<const Derived*>(this); }
@@ -65,7 +33,7 @@ class TranspositionsBase
       indices() = other.indices();
       return derived();
     }
-
+    
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
@@ -78,20 +46,24 @@ class TranspositionsBase
     #endif
 
     /** \returns the number of transpositions */
-    inline Index size() const { return indices().size(); }
+    Index size() const { return indices().size(); }
+    /** \returns the number of rows of the equivalent permutation matrix */
+    Index rows() const { return indices().size(); }
+    /** \returns the number of columns of the equivalent permutation matrix */
+    Index cols() const { return indices().size(); }
 
     /** Direct access to the underlying index vector */
-    inline const Index& coeff(Index i) const { return indices().coeff(i); }
+    inline const StorageIndex& coeff(Index i) const { return indices().coeff(i); }
     /** Direct access to the underlying index vector */
-    inline Index& coeffRef(Index i) { return indices().coeffRef(i); }
+    inline StorageIndex& coeffRef(Index i) { return indices().coeffRef(i); }
     /** Direct access to the underlying index vector */
-    inline const Index& operator()(Index i) const { return indices()(i); }
+    inline const StorageIndex& operator()(Index i) const { return indices()(i); }
     /** Direct access to the underlying index vector */
-    inline Index& operator()(Index i) { return indices()(i); }
+    inline StorageIndex& operator()(Index i) { return indices()(i); }
     /** Direct access to the underlying index vector */
-    inline const Index& operator[](Index i) const { return indices()(i); }
+    inline const StorageIndex& operator[](Index i) const { return indices()(i); }
     /** Direct access to the underlying index vector */
-    inline Index& operator[](Index i) { return indices()(i); }
+    inline StorageIndex& operator[](Index i) { return indices()(i); }
 
     /** const version of indices(). */
     const IndicesType& indices() const { return derived().indices(); }
@@ -99,7 +71,7 @@ class TranspositionsBase
     IndicesType& indices() { return derived().indices(); }
 
     /** Resizes to given size. */
-    inline void resize(int newSize)
+    inline void resize(Index newSize)
     {
       indices().resize(newSize);
     }
@@ -107,7 +79,7 @@ class TranspositionsBase
     /** Sets \c *this to represents an identity transformation */
     void setIdentity()
     {
-      for(int i = 0; i < indices().size(); ++i)
+      for(StorageIndex i = 0; i < indices().size(); ++i)
         coeffRef(i) = i;
     }
 
@@ -144,23 +116,53 @@ class TranspositionsBase
 };
 
 namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
-struct traits<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+struct traits<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
+ : traits<PermutationMatrix<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
 {
-  typedef IndexType Index;
-  typedef Matrix<Index, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
+  typedef Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
+  typedef TranspositionsStorage StorageKind;
 };
 }
 
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
-class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType> >
+/** \class Transpositions
+  * \ingroup Core_Module
+  *
+  * \brief Represents a sequence of transpositions (row/column interchange)
+  *
+  * \tparam SizeAtCompileTime the number of transpositions, or Dynamic
+  * \tparam MaxSizeAtCompileTime the maximum number of transpositions, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
+  *
+  * This class represents a permutation transformation as a sequence of \em n transpositions
+  * \f$[T_{n-1} \ldots T_{i} \ldots T_{0}]\f$. It is internally stored as a vector of integers \c indices.
+  * Each transposition \f$ T_{i} \f$ applied on the left of a matrix (\f$ T_{i} M\f$) interchanges
+  * the rows \c i and \c indices[i] of the matrix \c M.
+  * A transposition applied on the right (e.g., \f$ M T_{i}\f$) yields a column interchange.
+  *
+  * Compared to the class PermutationMatrix, such a sequence of transpositions is what is
+  * computed during a decomposition with pivoting, and it is faster when applying the permutation in-place.
+  *
+  * To apply a sequence of transpositions to a matrix, simply use the operator * as in the following example:
+  * \code
+  * Transpositions tr;
+  * MatrixXf mat;
+  * mat = tr * mat;
+  * \endcode
+  * In this example, we detect that the matrix appears on both side, and so the transpositions
+  * are applied in-place without any temporary or extra copy.
+  *
+  * \sa class PermutationMatrix
+  */
+
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
 {
     typedef internal::traits<Transpositions> Traits;
   public:
 
     typedef TranspositionsBase<Transpositions> Base;
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
 
     inline Transpositions() {}
 
@@ -177,7 +179,7 @@ class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTim
 
     /** Generic constructor from expression of the transposition indices. */
     template<typename Other>
-    explicit inline Transpositions(const MatrixBase<Other>& a_indices) : m_indices(a_indices)
+    explicit inline Transpositions(const MatrixBase<Other>& indices) : m_indices(indices)
     {}
 
     /** Copies the \a other transpositions into \c *this */
@@ -215,30 +217,32 @@ class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTim
 
 
 namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int _PacketAccess>
-struct traits<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType>,_PacketAccess> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
+struct traits<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,_PacketAccess> >
+ : traits<PermutationMatrix<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
 {
-  typedef IndexType Index;
-  typedef Map<const Matrix<Index,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1>, _PacketAccess> IndicesType;
+  typedef Map<const Matrix<_StorageIndex,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1>, _PacketAccess> IndicesType;
+  typedef _StorageIndex StorageIndex;
+  typedef TranspositionsStorage StorageKind;
 };
 }
 
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int PacketAccess>
-class Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType>,PacketAccess>
- : public TranspositionsBase<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType>,PacketAccess> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int PacketAccess>
+class Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,PacketAccess>
+ : public TranspositionsBase<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,PacketAccess> >
 {
     typedef internal::traits<Map> Traits;
   public:
 
     typedef TranspositionsBase<Map> Base;
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
 
-    inline Map(const Index* indicesPtr)
+    explicit inline Map(const StorageIndex* indicesPtr)
       : m_indices(indicesPtr)
     {}
 
-    inline Map(const Index* indicesPtr, Index size)
+    inline Map(const StorageIndex* indicesPtr, Index size)
       : m_indices(indicesPtr,size)
     {}
 
@@ -274,9 +278,9 @@ class Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType>,Packe
 namespace internal {
 template<typename _IndicesType>
 struct traits<TranspositionsWrapper<_IndicesType> >
+ : traits<PermutationWrapper<_IndicesType> >
 {
-  typedef typename _IndicesType::Scalar Index;
-  typedef _IndicesType IndicesType;
+  typedef TranspositionsStorage StorageKind;
 };
 }
 
@@ -289,10 +293,10 @@ class TranspositionsWrapper
 
     typedef TranspositionsBase<TranspositionsWrapper> Base;
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
 
-    inline TranspositionsWrapper(IndicesType& a_indices)
-      : m_indices(a_indices)
+    explicit inline TranspositionsWrapper(IndicesType& indices)
+      : m_indices(indices)
     {}
 
     /** Copies the \a other transpositions into \c *this */
@@ -321,83 +325,46 @@ class TranspositionsWrapper
 
   protected:
 
-    const typename IndicesType::Nested m_indices;
+    typename IndicesType::Nested m_indices;
 };
 
+
+
 /** \returns the \a matrix with the \a transpositions applied to the columns.
   */
-template<typename Derived, typename TranspositionsDerived>
-inline const internal::transposition_matrix_product_retval<TranspositionsDerived, Derived, OnTheRight>
-operator*(const MatrixBase<Derived>& matrix,
-          const TranspositionsBase<TranspositionsDerived> &transpositions)
+template<typename MatrixDerived, typename TranspositionsDerived>
+EIGEN_DEVICE_FUNC
+const Product<MatrixDerived, TranspositionsDerived, AliasFreeProduct>
+operator*(const MatrixBase<MatrixDerived> &matrix,
+          const TranspositionsBase<TranspositionsDerived>& transpositions)
 {
-  return internal::transposition_matrix_product_retval
-           <TranspositionsDerived, Derived, OnTheRight>
-           (transpositions.derived(), matrix.derived());
+  return Product<MatrixDerived, TranspositionsDerived, AliasFreeProduct>
+            (matrix.derived(), transpositions.derived());
 }
 
 /** \returns the \a matrix with the \a transpositions applied to the rows.
   */
-template<typename Derived, typename TranspositionDerived>
-inline const internal::transposition_matrix_product_retval
-               <TranspositionDerived, Derived, OnTheLeft>
-operator*(const TranspositionsBase<TranspositionDerived> &transpositions,
-          const MatrixBase<Derived>& matrix)
+template<typename TranspositionsDerived, typename MatrixDerived>
+EIGEN_DEVICE_FUNC
+const Product<TranspositionsDerived, MatrixDerived, AliasFreeProduct>
+operator*(const TranspositionsBase<TranspositionsDerived> &transpositions,
+          const MatrixBase<MatrixDerived>& matrix)
 {
-  return internal::transposition_matrix_product_retval
-           <TranspositionDerived, Derived, OnTheLeft>
-           (transpositions.derived(), matrix.derived());
+  return Product<TranspositionsDerived, MatrixDerived, AliasFreeProduct>
+            (transpositions.derived(), matrix.derived());
 }
 
-namespace internal {
-
-template<typename TranspositionType, typename MatrixType, int Side, bool Transposed>
-struct traits<transposition_matrix_product_retval<TranspositionType, MatrixType, Side, Transposed> >
-{
-  typedef typename MatrixType::PlainObject ReturnType;
-};
-
-template<typename TranspositionType, typename MatrixType, int Side, bool Transposed>
-struct transposition_matrix_product_retval
- : public ReturnByValue<transposition_matrix_product_retval<TranspositionType, MatrixType, Side, Transposed> >
-{
-    typedef typename remove_all<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
-    typedef typename TranspositionType::Index Index;
-
-    transposition_matrix_product_retval(const TranspositionType& tr, const MatrixType& matrix)
-      : m_transpositions(tr), m_matrix(matrix)
-    {}
+// Template partial specialization for transposed/inverse transpositions
 
-    inline int rows() const { return m_matrix.rows(); }
-    inline int cols() const { return m_matrix.cols(); }
-
-    template<typename Dest> inline void evalTo(Dest& dst) const
-    {
-      const int size = m_transpositions.size();
-      Index j = 0;
-
-      if(!(is_same<MatrixTypeNestedCleaned,Dest>::value && extract_data(dst) == extract_data(m_matrix)))
-        dst = m_matrix;
-
-      for(int k=(Transposed?size-1:0) ; Transposed?k>=0:k<size ; Transposed?--k:++k)
-        if((j=m_transpositions.coeff(k))!=k)
-        {
-          if(Side==OnTheLeft)
-            dst.row(k).swap(dst.row(j));
-          else if(Side==OnTheRight)
-            dst.col(k).swap(dst.col(j));
-        }
-    }
+namespace internal {
 
-  protected:
-    const TranspositionType& m_transpositions;
-    typename MatrixType::Nested m_matrix;
-};
+template<typename Derived>
+struct traits<Transpose<TranspositionsBase<Derived> > >
+ : traits<Derived>
+{};
 
 } // end namespace internal
 
-/* Template partial specialization for transposed/inverse transpositions */
-
 template<typename TranspositionsDerived>
 class Transpose<TranspositionsBase<TranspositionsDerived> >
 {
@@ -405,27 +372,31 @@ class Transpose<TranspositionsBase<TranspositionsDerived> >
     typedef typename TranspositionType::IndicesType IndicesType;
   public:
 
-    Transpose(const TranspositionType& t) : m_transpositions(t) {}
+    explicit Transpose(const TranspositionType& t) : m_transpositions(t) {}
 
-    inline int size() const { return m_transpositions.size(); }
+    Index size() const { return m_transpositions.size(); }
+    Index rows() const { return m_transpositions.size(); }
+    Index cols() const { return m_transpositions.size(); }
 
     /** \returns the \a matrix with the inverse transpositions applied to the columns.
       */
-    template<typename Derived> friend
-    inline const internal::transposition_matrix_product_retval<TranspositionType, Derived, OnTheRight, true>
-    operator*(const MatrixBase<Derived>& matrix, const Transpose& trt)
+    template<typename OtherDerived> friend
+    const Product<OtherDerived, Transpose, AliasFreeProduct>
+    operator*(const MatrixBase<OtherDerived>& matrix, const Transpose& trt)
     {
-      return internal::transposition_matrix_product_retval<TranspositionType, Derived, OnTheRight, true>(trt.m_transpositions, matrix.derived());
+      return Product<OtherDerived, Transpose, AliasFreeProduct>(matrix.derived(), trt);
     }
 
     /** \returns the \a matrix with the inverse transpositions applied to the rows.
       */
-    template<typename Derived>
-    inline const internal::transposition_matrix_product_retval<TranspositionType, Derived, OnTheLeft, true>
-    operator*(const MatrixBase<Derived>& matrix) const
+    template<typename OtherDerived>
+    const Product<Transpose, OtherDerived, AliasFreeProduct>
+    operator*(const MatrixBase<OtherDerived>& matrix) const
     {
-      return internal::transposition_matrix_product_retval<TranspositionType, Derived, OnTheLeft, true>(m_transpositions, matrix.derived());
+      return Product<Transpose, OtherDerived, AliasFreeProduct>(*this, matrix.derived());
     }
+    
+    const TranspositionType& nestedExpression() const { return m_transpositions; }
 
   protected:
     const TranspositionType& m_transpositions;
diff --git a/extern/Eigen3/Eigen/src/Core/TriangularMatrix.h b/extern/Eigen3/Eigen/src/Core/TriangularMatrix.h
index 4d65392c685..667ef09dc1a 100644
--- a/extern/Eigen3/Eigen/src/Core/TriangularMatrix.h
+++ b/extern/Eigen3/Eigen/src/Core/TriangularMatrix.h
@@ -19,9 +19,7 @@ template<int Side, typename TriangularType, typename Rhs> struct triangular_solv
   
 }
 
-/** \internal
-  *
-  * \class TriangularBase
+/** \class TriangularBase
   * \ingroup Core_Module
   *
   * \brief Base class for triangular part in a matrix
@@ -32,41 +30,69 @@ template<typename Derived> class TriangularBase : public EigenBase<Derived>
 
     enum {
       Mode = internal::traits<Derived>::Mode,
-      CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
       RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
       ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
       MaxRowsAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime,
-      MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime
+      MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime,
+      
+      SizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::RowsAtCompileTime,
+                                                   internal::traits<Derived>::ColsAtCompileTime>::ret),
+      /**< This is equal to the number of coefficients, i.e. the number of
+          * rows times the number of columns, or to \a Dynamic if this is not
+          * known at compile-time. \sa RowsAtCompileTime, ColsAtCompileTime */
+      
+      MaxSizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::MaxRowsAtCompileTime,
+                                                   internal::traits<Derived>::MaxColsAtCompileTime>::ret)
+        
     };
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
-    typedef typename internal::traits<Derived>::DenseMatrixType DenseMatrixType;
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+    typedef typename internal::traits<Derived>::FullMatrixType DenseMatrixType;
     typedef DenseMatrixType DenseType;
+    typedef Derived const& Nested;
 
+    EIGEN_DEVICE_FUNC
     inline TriangularBase() { eigen_assert(!((Mode&UnitDiag) && (Mode&ZeroDiag))); }
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return derived().rows(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return derived().cols(); }
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return derived().outerStride(); }
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return derived().innerStride(); }
+    
+    // dummy resize function
+    void resize(Index rows, Index cols)
+    {
+      EIGEN_UNUSED_VARIABLE(rows);
+      EIGEN_UNUSED_VARIABLE(cols);
+      eigen_assert(rows==this->rows() && cols==this->cols());
+    }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar coeff(Index row, Index col) const  { return derived().coeff(row,col); }
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index row, Index col) { return derived().coeffRef(row,col); }
 
     /** \see MatrixBase::copyCoeff(row,col)
       */
     template<typename Other>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void copyCoeff(Index row, Index col, Other& other)
     {
       derived().coeffRef(row, col) = other.coeff(row, col);
     }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar operator()(Index row, Index col) const
     {
       check_coordinates(row, col);
       return coeff(row,col);
     }
+    EIGEN_DEVICE_FUNC
     inline Scalar& operator()(Index row, Index col)
     {
       check_coordinates(row, col);
@@ -74,15 +100,20 @@ template<typename Derived> class TriangularBase : public EigenBase<Derived>
     }
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
+    EIGEN_DEVICE_FUNC
     inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    EIGEN_DEVICE_FUNC
     inline Derived& derived() { return *static_cast<Derived*>(this); }
     #endif // not EIGEN_PARSED_BY_DOXYGEN
 
     template<typename DenseDerived>
+    EIGEN_DEVICE_FUNC
     void evalTo(MatrixBase<DenseDerived> &other) const;
     template<typename DenseDerived>
+    EIGEN_DEVICE_FUNC
     void evalToLazy(MatrixBase<DenseDerived> &other) const;
 
+    EIGEN_DEVICE_FUNC
     DenseMatrixType toDenseMatrix() const
     {
       DenseMatrixType res(rows(), cols());
@@ -119,17 +150,17 @@ template<typename Derived> class TriangularBase : public EigenBase<Derived>
 /** \class TriangularView
   * \ingroup Core_Module
   *
-  * \brief Base class for triangular part in a matrix
+  * \brief Expression of a triangular part in a matrix
   *
   * \param MatrixType the type of the object in which we are taking the triangular part
   * \param Mode the kind of triangular matrix expression to construct. Can be #Upper,
   *             #Lower, #UnitUpper, #UnitLower, #StrictlyUpper, or #StrictlyLower.
   *             This is in fact a bit field; it must have either #Upper or #Lower, 
-  *             and additionnaly it may have #UnitDiag or #ZeroDiag or neither.
+  *             and additionally it may have #UnitDiag or #ZeroDiag or neither.
   *
   * This class represents a triangular part of a matrix, not necessarily square. Strictly speaking, for rectangular
   * matrices one should speak of "trapezoid" parts. This class is the return type
-  * of MatrixBase::triangularView() and most of the time this is the only way it is used.
+  * of MatrixBase::triangularView() and SparseMatrixBase::triangularView(), and most of the time this is the only way it is used.
   *
   * \sa MatrixBase::triangularView()
   */
@@ -137,499 +168,405 @@ namespace internal {
 template<typename MatrixType, unsigned int _Mode>
 struct traits<TriangularView<MatrixType, _Mode> > : traits<MatrixType>
 {
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::non_const_type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type MatrixTypeNestedNonRef;
   typedef typename remove_all<MatrixTypeNested>::type MatrixTypeNestedCleaned;
+  typedef typename MatrixType::PlainObject FullMatrixType;
   typedef MatrixType ExpressionType;
-  typedef typename MatrixType::PlainObject DenseMatrixType;
   enum {
     Mode = _Mode,
-    Flags = (MatrixTypeNestedCleaned::Flags & (HereditaryBits) & (~(PacketAccessBit | DirectAccessBit | LinearAccessBit))) | Mode,
-    CoeffReadCost = MatrixTypeNestedCleaned::CoeffReadCost
+    FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
+    Flags = (MatrixTypeNestedCleaned::Flags & (HereditaryBits | FlagsLvalueBit) & (~(PacketAccessBit | DirectAccessBit | LinearAccessBit)))
   };
 };
 }
 
-template<int Mode, bool LhsIsTriangular,
-         typename Lhs, bool LhsIsVector,
-         typename Rhs, bool RhsIsVector>
-struct TriangularProduct;
+template<typename _MatrixType, unsigned int _Mode, typename StorageKind> class TriangularViewImpl;
 
 template<typename _MatrixType, unsigned int _Mode> class TriangularView
-  : public TriangularBase<TriangularView<_MatrixType, _Mode> >
+  : public TriangularViewImpl<_MatrixType, _Mode, typename internal::traits<_MatrixType>::StorageKind >
 {
   public:
 
-    typedef TriangularBase<TriangularView> Base;
+    typedef TriangularViewImpl<_MatrixType, _Mode, typename internal::traits<_MatrixType>::StorageKind > Base;
     typedef typename internal::traits<TriangularView>::Scalar Scalar;
-
     typedef _MatrixType MatrixType;
-    typedef typename internal::traits<TriangularView>::DenseMatrixType DenseMatrixType;
-    typedef DenseMatrixType PlainObject;
 
   protected:
     typedef typename internal::traits<TriangularView>::MatrixTypeNested MatrixTypeNested;
     typedef typename internal::traits<TriangularView>::MatrixTypeNestedNonRef MatrixTypeNestedNonRef;
-    typedef typename internal::traits<TriangularView>::MatrixTypeNestedCleaned MatrixTypeNestedCleaned;
 
     typedef typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type MatrixConjugateReturnType;
     
   public:
-    using Base::evalToLazy;
-  
 
     typedef typename internal::traits<TriangularView>::StorageKind StorageKind;
-    typedef typename internal::traits<TriangularView>::Index Index;
+    typedef typename internal::traits<TriangularView>::MatrixTypeNestedCleaned NestedExpression;
 
     enum {
       Mode = _Mode,
+      Flags = internal::traits<TriangularView>::Flags,
       TransposeMode = (Mode & Upper ? Lower : 0)
                     | (Mode & Lower ? Upper : 0)
                     | (Mode & (UnitDiag))
-                    | (Mode & (ZeroDiag))
+                    | (Mode & (ZeroDiag)),
+      IsVectorAtCompileTime = false
     };
 
-    inline TriangularView(const MatrixType& matrix) : m_matrix(matrix)
+    EIGEN_DEVICE_FUNC
+    explicit inline TriangularView(MatrixType& matrix) : m_matrix(matrix)
     {}
+    
+    using Base::operator=;
+    TriangularView& operator=(const TriangularView &other)
+    { return Base::operator=(other); }
 
+    /** \copydoc EigenBase::rows() */
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_matrix.rows(); }
+    /** \copydoc EigenBase::cols() */
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_matrix.cols(); }
-    inline Index outerStride() const { return m_matrix.outerStride(); }
-    inline Index innerStride() const { return m_matrix.innerStride(); }
+
+    /** \returns a const reference to the nested expression */
+    EIGEN_DEVICE_FUNC
+    const NestedExpression& nestedExpression() const { return m_matrix; }
+
+    /** \returns a reference to the nested expression */
+    EIGEN_DEVICE_FUNC
+    NestedExpression& nestedExpression() { return m_matrix; }
+    
+    typedef TriangularView<const MatrixConjugateReturnType,Mode> ConjugateReturnType;
+    /** \sa MatrixBase::conjugate() const */
+    EIGEN_DEVICE_FUNC
+    inline const ConjugateReturnType conjugate() const
+    { return ConjugateReturnType(m_matrix.conjugate()); }
+
+    typedef TriangularView<const typename MatrixType::AdjointReturnType,TransposeMode> AdjointReturnType;
+    /** \sa MatrixBase::adjoint() const */
+    EIGEN_DEVICE_FUNC
+    inline const AdjointReturnType adjoint() const
+    { return AdjointReturnType(m_matrix.adjoint()); }
+
+    typedef TriangularView<typename MatrixType::TransposeReturnType,TransposeMode> TransposeReturnType;
+     /** \sa MatrixBase::transpose() */
+    EIGEN_DEVICE_FUNC
+    inline TransposeReturnType transpose()
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
+      typename MatrixType::TransposeReturnType tmp(m_matrix);
+      return TransposeReturnType(tmp);
+    }
+    
+    typedef TriangularView<const typename MatrixType::ConstTransposeReturnType,TransposeMode> ConstTransposeReturnType;
+    /** \sa MatrixBase::transpose() const */
+    EIGEN_DEVICE_FUNC
+    inline const ConstTransposeReturnType transpose() const
+    {
+      return ConstTransposeReturnType(m_matrix.transpose());
+    }
+
+    template<typename Other>
+    EIGEN_DEVICE_FUNC
+    inline const Solve<TriangularView, Other> 
+    solve(const MatrixBase<Other>& other) const
+    { return Solve<TriangularView, Other>(*this, other.derived()); }
+    
+  // workaround MSVC ICE
+  #if EIGEN_COMP_MSVC
+    template<int Side, typename Other>
+    EIGEN_DEVICE_FUNC
+    inline const internal::triangular_solve_retval<Side,TriangularView, Other>
+    solve(const MatrixBase<Other>& other) const
+    { return Base::template solve<Side>(other); }
+  #else
+    using Base::solve;
+  #endif
+
+    /** \returns a selfadjoint view of the referenced triangular part which must be either \c #Upper or \c #Lower.
+      *
+      * This is a shortcut for \code this->nestedExpression().selfadjointView<(*this)::Mode>() \endcode
+      * \sa MatrixBase::selfadjointView() */
+    EIGEN_DEVICE_FUNC
+    SelfAdjointView<MatrixTypeNestedNonRef,Mode> selfadjointView()
+    {
+      EIGEN_STATIC_ASSERT((Mode&(UnitDiag|ZeroDiag))==0,PROGRAMMING_ERROR);
+      return SelfAdjointView<MatrixTypeNestedNonRef,Mode>(m_matrix);
+    }
+
+    /** This is the const version of selfadjointView() */
+    EIGEN_DEVICE_FUNC
+    const SelfAdjointView<MatrixTypeNestedNonRef,Mode> selfadjointView() const
+    {
+      EIGEN_STATIC_ASSERT((Mode&(UnitDiag|ZeroDiag))==0,PROGRAMMING_ERROR);
+      return SelfAdjointView<MatrixTypeNestedNonRef,Mode>(m_matrix);
+    }
+
+
+    /** \returns the determinant of the triangular matrix
+      * \sa MatrixBase::determinant() */
+    EIGEN_DEVICE_FUNC
+    Scalar determinant() const
+    {
+      if (Mode & UnitDiag)
+        return 1;
+      else if (Mode & ZeroDiag)
+        return 0;
+      else
+        return m_matrix.diagonal().prod();
+    }
+      
+  protected:
+
+    MatrixTypeNested m_matrix;
+};
+
+/** \ingroup Core_Module
+  *
+  * \brief Base class for a triangular part in a \b dense matrix
+  *
+  * This class is an abstract base class of class TriangularView, and objects of type TriangularViewImpl cannot be instantiated.
+  * It extends class TriangularView with additional methods which available for dense expressions only.
+  *
+  * \sa class TriangularView, MatrixBase::triangularView()
+  */
+template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_MatrixType,_Mode,Dense>
+  : public TriangularBase<TriangularView<_MatrixType, _Mode> >
+{
+  public:
+
+    typedef TriangularView<_MatrixType, _Mode> TriangularViewType;
+    typedef TriangularBase<TriangularViewType> Base;
+    typedef typename internal::traits<TriangularViewType>::Scalar Scalar;
+
+    typedef _MatrixType MatrixType;
+    typedef typename MatrixType::PlainObject DenseMatrixType;
+    typedef DenseMatrixType PlainObject;
+
+  public:
+    using Base::evalToLazy;
+    using Base::derived;
+
+    typedef typename internal::traits<TriangularViewType>::StorageKind StorageKind;
+
+    enum {
+      Mode = _Mode,
+      Flags = internal::traits<TriangularViewType>::Flags
+    };
+
+    /** \returns the outer-stride of the underlying dense matrix
+      * \sa DenseCoeffsBase::outerStride() */
+    EIGEN_DEVICE_FUNC
+    inline Index outerStride() const { return derived().nestedExpression().outerStride(); }
+    /** \returns the inner-stride of the underlying dense matrix
+      * \sa DenseCoeffsBase::innerStride() */
+    EIGEN_DEVICE_FUNC
+    inline Index innerStride() const { return derived().nestedExpression().innerStride(); }
 
     /** \sa MatrixBase::operator+=() */
-    template<typename Other> TriangularView&  operator+=(const DenseBase<Other>& other) { return *this = m_matrix + other.derived(); }
+    template<typename Other>
+    EIGEN_DEVICE_FUNC
+    TriangularViewType&  operator+=(const DenseBase<Other>& other) {
+      internal::call_assignment_no_alias(derived(), other.derived(), internal::add_assign_op<Scalar,typename Other::Scalar>());
+      return derived();
+    }
     /** \sa MatrixBase::operator-=() */
-    template<typename Other> TriangularView&  operator-=(const DenseBase<Other>& other) { return *this = m_matrix - other.derived(); }
+    template<typename Other>
+    EIGEN_DEVICE_FUNC
+    TriangularViewType&  operator-=(const DenseBase<Other>& other) {
+      internal::call_assignment_no_alias(derived(), other.derived(), internal::sub_assign_op<Scalar,typename Other::Scalar>());
+      return derived();
+    }
+    
     /** \sa MatrixBase::operator*=() */
-    TriangularView&  operator*=(const typename internal::traits<MatrixType>::Scalar& other) { return *this = m_matrix * other; }
-    /** \sa MatrixBase::operator/=() */
-    TriangularView&  operator/=(const typename internal::traits<MatrixType>::Scalar& other) { return *this = m_matrix / other; }
+    EIGEN_DEVICE_FUNC
+    TriangularViewType&  operator*=(const typename internal::traits<MatrixType>::Scalar& other) { return *this = derived().nestedExpression() * other; }
+    /** \sa DenseBase::operator/=() */
+    EIGEN_DEVICE_FUNC
+    TriangularViewType&  operator/=(const typename internal::traits<MatrixType>::Scalar& other) { return *this = derived().nestedExpression() / other; }
 
     /** \sa MatrixBase::fill() */
+    EIGEN_DEVICE_FUNC
     void fill(const Scalar& value) { setConstant(value); }
     /** \sa MatrixBase::setConstant() */
-    TriangularView& setConstant(const Scalar& value)
-    { return *this = MatrixType::Constant(rows(), cols(), value); }
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& setConstant(const Scalar& value)
+    { return *this = MatrixType::Constant(derived().rows(), derived().cols(), value); }
     /** \sa MatrixBase::setZero() */
-    TriangularView& setZero() { return setConstant(Scalar(0)); }
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& setZero() { return setConstant(Scalar(0)); }
     /** \sa MatrixBase::setOnes() */
-    TriangularView& setOnes() { return setConstant(Scalar(1)); }
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& setOnes() { return setConstant(Scalar(1)); }
 
     /** \sa MatrixBase::coeff()
       * \warning the coordinates must fit into the referenced triangular part
       */
+    EIGEN_DEVICE_FUNC
     inline Scalar coeff(Index row, Index col) const
     {
       Base::check_coordinates_internal(row, col);
-      return m_matrix.coeff(row, col);
+      return derived().nestedExpression().coeff(row, col);
     }
 
     /** \sa MatrixBase::coeffRef()
       * \warning the coordinates must fit into the referenced triangular part
       */
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index row, Index col)
     {
+      EIGEN_STATIC_ASSERT_LVALUE(TriangularViewType);
       Base::check_coordinates_internal(row, col);
-      return m_matrix.const_cast_derived().coeffRef(row, col);
+      return derived().nestedExpression().coeffRef(row, col);
     }
 
-    const MatrixTypeNestedCleaned& nestedExpression() const { return m_matrix; }
-    MatrixTypeNestedCleaned& nestedExpression() { return *const_cast<MatrixTypeNestedCleaned*>(&m_matrix); }
-
     /** Assigns a triangular matrix to a triangular part of a dense matrix */
     template<typename OtherDerived>
-    TriangularView& operator=(const TriangularBase<OtherDerived>& other);
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& operator=(const TriangularBase<OtherDerived>& other);
 
+    /** Shortcut for\code *this = other.other.triangularView<(*this)::Mode>() \endcode */
     template<typename OtherDerived>
-    TriangularView& operator=(const MatrixBase<OtherDerived>& other);
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& operator=(const MatrixBase<OtherDerived>& other);
 
-    TriangularView& operator=(const TriangularView& other)
-    { return *this = other.nestedExpression(); }
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& operator=(const TriangularViewImpl& other)
+    { return *this = other.derived().nestedExpression(); }
 
+    /** \deprecated */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     void lazyAssign(const TriangularBase<OtherDerived>& other);
 
+    /** \deprecated */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     void lazyAssign(const MatrixBase<OtherDerived>& other);
-
-    /** \sa MatrixBase::conjugate() */
-    inline TriangularView<MatrixConjugateReturnType,Mode> conjugate()
-    { return m_matrix.conjugate(); }
-    /** \sa MatrixBase::conjugate() const */
-    inline const TriangularView<MatrixConjugateReturnType,Mode> conjugate() const
-    { return m_matrix.conjugate(); }
-
-    /** \sa MatrixBase::adjoint() const */
-    inline const TriangularView<const typename MatrixType::AdjointReturnType,TransposeMode> adjoint() const
-    { return m_matrix.adjoint(); }
-
-    /** \sa MatrixBase::transpose() */
-    inline TriangularView<Transpose<MatrixType>,TransposeMode> transpose()
-    {
-      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return m_matrix.const_cast_derived().transpose();
-    }
-    /** \sa MatrixBase::transpose() const */
-    inline const TriangularView<Transpose<MatrixType>,TransposeMode> transpose() const
-    {
-      return m_matrix.transpose();
-    }
+#endif
 
     /** Efficient triangular matrix times vector/matrix product */
     template<typename OtherDerived>
-    TriangularProduct<Mode, true, MatrixType, false, OtherDerived, OtherDerived::ColsAtCompileTime==1>
+    EIGEN_DEVICE_FUNC
+    const Product<TriangularViewType,OtherDerived>
     operator*(const MatrixBase<OtherDerived>& rhs) const
     {
-      return TriangularProduct
-              <Mode, true, MatrixType, false, OtherDerived, OtherDerived::ColsAtCompileTime==1>
-              (m_matrix, rhs.derived());
+      return Product<TriangularViewType,OtherDerived>(derived(), rhs.derived());
     }
 
     /** Efficient vector/matrix times triangular matrix product */
     template<typename OtherDerived> friend
-    TriangularProduct<Mode, false, OtherDerived, OtherDerived::RowsAtCompileTime==1, MatrixType, false>
-    operator*(const MatrixBase<OtherDerived>& lhs, const TriangularView& rhs)
-    {
-      return TriangularProduct
-              <Mode, false, OtherDerived, OtherDerived::RowsAtCompileTime==1, MatrixType, false>
-              (lhs.derived(),rhs.m_matrix);
-    }
-
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived>
-    struct eigen2_product_return_type
-    {
-      typedef typename TriangularView<MatrixType,Mode>::DenseMatrixType DenseMatrixType;
-      typedef typename OtherDerived::PlainObject::DenseType OtherPlainObject;
-      typedef typename ProductReturnType<DenseMatrixType, OtherPlainObject>::Type ProdRetType;
-      typedef typename ProdRetType::PlainObject type;
-    };
-    template<typename OtherDerived>
-    const typename eigen2_product_return_type<OtherDerived>::type
-    operator*(const EigenBase<OtherDerived>& rhs) const
-    {
-      typename OtherDerived::PlainObject::DenseType rhsPlainObject;
-      rhs.evalTo(rhsPlainObject);
-      return this->toDenseMatrix() * rhsPlainObject;
-    }
-    template<typename OtherMatrixType>
-    bool isApprox(const TriangularView<OtherMatrixType, Mode>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
-    {
-      return this->toDenseMatrix().isApprox(other.toDenseMatrix(), precision);
-    }
-    template<typename OtherDerived>
-    bool isApprox(const MatrixBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
-    {
-      return this->toDenseMatrix().isApprox(other, precision);
-    }
-    #endif // EIGEN2_SUPPORT
-
+    EIGEN_DEVICE_FUNC
+    const Product<OtherDerived,TriangularViewType>
+    operator*(const MatrixBase<OtherDerived>& lhs, const TriangularViewImpl& rhs)
+    {
+      return Product<OtherDerived,TriangularViewType>(lhs.derived(),rhs.derived());
+    }
+
+    /** \returns the product of the inverse of \c *this with \a other, \a *this being triangular.
+      *
+      * This function computes the inverse-matrix matrix product inverse(\c *this) * \a other if
+      * \a Side==OnTheLeft (the default), or the right-inverse-multiply  \a other * inverse(\c *this) if
+      * \a Side==OnTheRight.
+      *
+      * Note that the template parameter \c Side can be ommitted, in which case \c Side==OnTheLeft
+      *
+      * The matrix \c *this must be triangular and invertible (i.e., all the coefficients of the
+      * diagonal must be non zero). It works as a forward (resp. backward) substitution if \c *this
+      * is an upper (resp. lower) triangular matrix.
+      *
+      * Example: \include Triangular_solve.cpp
+      * Output: \verbinclude Triangular_solve.out
+      *
+      * This function returns an expression of the inverse-multiply and can works in-place if it is assigned
+      * to the same matrix or vector \a other.
+      *
+      * For users coming from BLAS, this function (and more specifically solveInPlace()) offer
+      * all the operations supported by the \c *TRSV and \c *TRSM BLAS routines.
+      *
+      * \sa TriangularView::solveInPlace()
+      */
     template<int Side, typename Other>
-    inline const internal::triangular_solve_retval<Side,TriangularView, Other>
+    EIGEN_DEVICE_FUNC
+    inline const internal::triangular_solve_retval<Side,TriangularViewType, Other>
     solve(const MatrixBase<Other>& other) const;
 
+    /** "in-place" version of TriangularView::solve() where the result is written in \a other
+      *
+      * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
+      * This function will const_cast it, so constness isn't honored here.
+      *
+      * Note that the template parameter \c Side can be ommitted, in which case \c Side==OnTheLeft
+      *
+      * See TriangularView:solve() for the details.
+      */
     template<int Side, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     void solveInPlace(const MatrixBase<OtherDerived>& other) const;
 
-    template<typename Other>
-    inline const internal::triangular_solve_retval<OnTheLeft,TriangularView, Other> 
-    solve(const MatrixBase<Other>& other) const
-    { return solve<OnTheLeft>(other); }
-
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     void solveInPlace(const MatrixBase<OtherDerived>& other) const
     { return solveInPlace<OnTheLeft>(other); }
 
-    const SelfAdjointView<MatrixTypeNestedNonRef,Mode> selfadjointView() const
-    {
-      EIGEN_STATIC_ASSERT((Mode&UnitDiag)==0,PROGRAMMING_ERROR);
-      return SelfAdjointView<MatrixTypeNestedNonRef,Mode>(m_matrix);
-    }
-    SelfAdjointView<MatrixTypeNestedNonRef,Mode> selfadjointView()
-    {
-      EIGEN_STATIC_ASSERT((Mode&UnitDiag)==0,PROGRAMMING_ERROR);
-      return SelfAdjointView<MatrixTypeNestedNonRef,Mode>(m_matrix);
-    }
-
+    /** Swaps the coefficients of the common triangular parts of two matrices */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+    void swap(TriangularBase<OtherDerived> &other)
+#else
     void swap(TriangularBase<OtherDerived> const & other)
+#endif
     {
-      TriangularView<SwapWrapper<MatrixType>,Mode>(const_cast<MatrixType&>(m_matrix)).lazyAssign(other.derived());
+      EIGEN_STATIC_ASSERT_LVALUE(OtherDerived);
+      call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
     }
 
+    /** \deprecated
+      * Shortcut for \code (*this).swap(other.triangularView<(*this)::Mode>()) \endcode */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     void swap(MatrixBase<OtherDerived> const & other)
     {
-      SwapWrapper<MatrixType> swaper(const_cast<MatrixType&>(m_matrix));
-      TriangularView<SwapWrapper<MatrixType>,Mode>(swaper).lazyAssign(other.derived());
+      EIGEN_STATIC_ASSERT_LVALUE(OtherDerived);
+      call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
     }
 
-    Scalar determinant() const
-    {
-      if (Mode & UnitDiag)
-        return 1;
-      else if (Mode & ZeroDiag)
-        return 0;
-      else
-        return m_matrix.diagonal().prod();
-    }
-    
-    // TODO simplify the following:
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE TriangularView& operator=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    {
-      setZero();
-      return assignProduct(other.derived(),1);
-    }
-    
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE TriangularView& operator+=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    {
-      return assignProduct(other.derived(),1);
-    }
-    
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE TriangularView& operator-=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    {
-      return assignProduct(other.derived(),-1);
-    }
-    
-    
-    template<typename ProductDerived>
-    EIGEN_STRONG_INLINE TriangularView& operator=(const ScaledProduct<ProductDerived>& other)
-    {
-      setZero();
-      return assignProduct(other.derived(),other.alpha());
-    }
-    
-    template<typename ProductDerived>
-    EIGEN_STRONG_INLINE TriangularView& operator+=(const ScaledProduct<ProductDerived>& other)
-    {
-      return assignProduct(other.derived(),other.alpha());
-    }
-    
-    template<typename ProductDerived>
-    EIGEN_STRONG_INLINE TriangularView& operator-=(const ScaledProduct<ProductDerived>& other)
-    {
-      return assignProduct(other.derived(),-other.alpha());
-    }
-    
-  protected:
-    
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE TriangularView& assignProduct(const ProductBase<ProductDerived, Lhs,Rhs>& prod, const Scalar& alpha);
-    
-    template<int Mode, bool LhsIsTriangular,
-         typename Lhs, bool LhsIsVector,
-         typename Rhs, bool RhsIsVector>
-    EIGEN_STRONG_INLINE TriangularView& assignProduct(const TriangularProduct<Mode, LhsIsTriangular, Lhs, LhsIsVector, Rhs, RhsIsVector>& prod, const Scalar& alpha)
-    {
-      lazyAssign(alpha*prod.eval());
-      return *this;
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _solve_impl(const RhsType &rhs, DstType &dst) const {
+      if(!internal::is_same_dense(dst,rhs))
+        dst = rhs;
+      this->solveInPlace(dst);
     }
 
-    MatrixTypeNested m_matrix;
+    template<typename ProductType>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE TriangularViewType& _assignProduct(const ProductType& prod, const Scalar& alpha, bool beta);
 };
 
 /***************************************************************************
 * Implementation of triangular evaluation/assignment
 ***************************************************************************/
 
-namespace internal {
-
-template<typename Derived1, typename Derived2, unsigned int Mode, int UnrollCount, bool ClearOpposite>
-struct triangular_assignment_selector
-{
-  enum {
-    col = (UnrollCount-1) / Derived1::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived1::RowsAtCompileTime
-  };
-  
-  typedef typename Derived1::Scalar Scalar;
-
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    triangular_assignment_selector<Derived1, Derived2, Mode, UnrollCount-1, ClearOpposite>::run(dst, src);
-
-    eigen_assert( Mode == Upper || Mode == Lower
-            || Mode == StrictlyUpper || Mode == StrictlyLower
-            || Mode == UnitUpper || Mode == UnitLower);
-    if((Mode == Upper && row <= col)
-    || (Mode == Lower && row >= col)
-    || (Mode == StrictlyUpper && row < col)
-    || (Mode == StrictlyLower && row > col)
-    || (Mode == UnitUpper && row < col)
-    || (Mode == UnitLower && row > col))
-      dst.copyCoeff(row, col, src);
-    else if(ClearOpposite)
-    {
-      if (Mode&UnitDiag && row==col)
-        dst.coeffRef(row, col) = Scalar(1);
-      else
-        dst.coeffRef(row, col) = Scalar(0);
-    }
-  }
-};
-
-// prevent buggy user code from causing an infinite recursion
-template<typename Derived1, typename Derived2, unsigned int Mode, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, Mode, 0, ClearOpposite>
-{
-  static inline void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, Upper, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  typedef typename Derived1::Scalar Scalar;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      Index maxi = (std::min)(j, dst.rows()-1);
-      for(Index i = 0; i <= maxi; ++i)
-        dst.copyCoeff(i, j, src);
-      if (ClearOpposite)
-        for(Index i = maxi+1; i < dst.rows(); ++i)
-          dst.coeffRef(i, j) = Scalar(0);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, Lower, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      for(Index i = j; i < dst.rows(); ++i)
-        dst.copyCoeff(i, j, src);
-      Index maxi = (std::min)(j, dst.rows());
-      if (ClearOpposite)
-        for(Index i = 0; i < maxi; ++i)
-          dst.coeffRef(i, j) = static_cast<typename Derived1::Scalar>(0);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, StrictlyUpper, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  typedef typename Derived1::Scalar Scalar;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      Index maxi = (std::min)(j, dst.rows());
-      for(Index i = 0; i < maxi; ++i)
-        dst.copyCoeff(i, j, src);
-      if (ClearOpposite)
-        for(Index i = maxi; i < dst.rows(); ++i)
-          dst.coeffRef(i, j) = Scalar(0);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, StrictlyLower, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      for(Index i = j+1; i < dst.rows(); ++i)
-        dst.copyCoeff(i, j, src);
-      Index maxi = (std::min)(j, dst.rows()-1);
-      if (ClearOpposite)
-        for(Index i = 0; i <= maxi; ++i)
-          dst.coeffRef(i, j) = static_cast<typename Derived1::Scalar>(0);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, UnitUpper, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      Index maxi = (std::min)(j, dst.rows());
-      for(Index i = 0; i < maxi; ++i)
-        dst.copyCoeff(i, j, src);
-      if (ClearOpposite)
-      {
-        for(Index i = maxi+1; i < dst.rows(); ++i)
-          dst.coeffRef(i, j) = 0;
-      }
-    }
-    dst.diagonal().setOnes();
-  }
-};
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, UnitLower, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      Index maxi = (std::min)(j, dst.rows());
-      for(Index i = maxi+1; i < dst.rows(); ++i)
-        dst.copyCoeff(i, j, src);
-      if (ClearOpposite)
-      {
-        for(Index i = 0; i < maxi; ++i)
-          dst.coeffRef(i, j) = 0;
-      }
-    }
-    dst.diagonal().setOnes();
-  }
-};
-
-} // end namespace internal
-
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 // FIXME should we keep that possibility
 template<typename MatrixType, unsigned int Mode>
 template<typename OtherDerived>
 inline TriangularView<MatrixType, Mode>&
-TriangularView<MatrixType, Mode>::operator=(const MatrixBase<OtherDerived>& other)
+TriangularViewImpl<MatrixType, Mode, Dense>::operator=(const MatrixBase<OtherDerived>& other)
 {
-  if(OtherDerived::Flags & EvalBeforeAssigningBit)
-  {
-    typename internal::plain_matrix_type<OtherDerived>::type other_evaluated(other.rows(), other.cols());
-    other_evaluated.template triangularView<Mode>().lazyAssign(other.derived());
-    lazyAssign(other_evaluated);
-  }
-  else
-    lazyAssign(other.derived());
-  return *this;
+  internal::call_assignment_no_alias(derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
 }
 
 // FIXME should we keep that possibility
 template<typename MatrixType, unsigned int Mode>
 template<typename OtherDerived>
-void TriangularView<MatrixType, Mode>::lazyAssign(const MatrixBase<OtherDerived>& other)
+void TriangularViewImpl<MatrixType, Mode, Dense>::lazyAssign(const MatrixBase<OtherDerived>& other)
 {
-  enum {
-    unroll = MatrixType::SizeAtCompileTime != Dynamic
-          && internal::traits<OtherDerived>::CoeffReadCost != Dynamic
-          && MatrixType::SizeAtCompileTime*internal::traits<OtherDerived>::CoeffReadCost/2 <= EIGEN_UNROLLING_LIMIT
-  };
-  eigen_assert(m_matrix.rows() == other.rows() && m_matrix.cols() == other.cols());
-
-  internal::triangular_assignment_selector
-    <MatrixType, OtherDerived, int(Mode),
-    unroll ? int(MatrixType::SizeAtCompileTime) : Dynamic,
-    false // do not change the opposite triangular part
-    >::run(m_matrix.const_cast_derived(), other.derived());
+  internal::call_assignment_no_alias(derived(), other.template triangularView<Mode>());
 }
 
 
@@ -637,38 +574,21 @@ void TriangularView<MatrixType, Mode>::lazyAssign(const MatrixBase<OtherDerived>
 template<typename MatrixType, unsigned int Mode>
 template<typename OtherDerived>
 inline TriangularView<MatrixType, Mode>&
-TriangularView<MatrixType, Mode>::operator=(const TriangularBase<OtherDerived>& other)
+TriangularViewImpl<MatrixType, Mode, Dense>::operator=(const TriangularBase<OtherDerived>& other)
 {
   eigen_assert(Mode == int(OtherDerived::Mode));
-  if(internal::traits<OtherDerived>::Flags & EvalBeforeAssigningBit)
-  {
-    typename OtherDerived::DenseMatrixType other_evaluated(other.rows(), other.cols());
-    other_evaluated.template triangularView<Mode>().lazyAssign(other.derived().nestedExpression());
-    lazyAssign(other_evaluated);
-  }
-  else
-    lazyAssign(other.derived().nestedExpression());
-  return *this;
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename MatrixType, unsigned int Mode>
 template<typename OtherDerived>
-void TriangularView<MatrixType, Mode>::lazyAssign(const TriangularBase<OtherDerived>& other)
+void TriangularViewImpl<MatrixType, Mode, Dense>::lazyAssign(const TriangularBase<OtherDerived>& other)
 {
-  enum {
-    unroll = MatrixType::SizeAtCompileTime != Dynamic
-                   && internal::traits<OtherDerived>::CoeffReadCost != Dynamic
-                   && MatrixType::SizeAtCompileTime * internal::traits<OtherDerived>::CoeffReadCost / 2
-                        <= EIGEN_UNROLLING_LIMIT
-  };
-  eigen_assert(m_matrix.rows() == other.rows() && m_matrix.cols() == other.cols());
-
-  internal::triangular_assignment_selector
-    <MatrixType, OtherDerived, int(Mode),
-    unroll ? int(MatrixType::SizeAtCompileTime) : Dynamic,
-    false // preserve the opposite triangular part
-    >::run(m_matrix.const_cast_derived(), other.derived().nestedExpression());
+  eigen_assert(Mode == int(OtherDerived::Mode));
+  internal::call_assignment_no_alias(derived(), other.derived());
 }
+#endif
 
 /***************************************************************************
 * Implementation of TriangularBase methods
@@ -680,35 +600,7 @@ template<typename Derived>
 template<typename DenseDerived>
 void TriangularBase<Derived>::evalTo(MatrixBase<DenseDerived> &other) const
 {
-  if(internal::traits<Derived>::Flags & EvalBeforeAssigningBit)
-  {
-    typename internal::plain_matrix_type<Derived>::type other_evaluated(rows(), cols());
-    evalToLazy(other_evaluated);
-    other.derived().swap(other_evaluated);
-  }
-  else
-    evalToLazy(other.derived());
-}
-
-/** Assigns a triangular or selfadjoint matrix to a dense matrix.
-  * If the matrix is triangular, the opposite part is set to zero. */
-template<typename Derived>
-template<typename DenseDerived>
-void TriangularBase<Derived>::evalToLazy(MatrixBase<DenseDerived> &other) const
-{
-  enum {
-    unroll = DenseDerived::SizeAtCompileTime != Dynamic
-                   && internal::traits<Derived>::CoeffReadCost != Dynamic
-                   && DenseDerived::SizeAtCompileTime * internal::traits<Derived>::CoeffReadCost / 2
-                        <= EIGEN_UNROLLING_LIMIT
-  };
-  other.derived().resize(this->rows(), this->cols());
-
-  internal::triangular_assignment_selector
-    <DenseDerived, typename internal::traits<Derived>::MatrixTypeNestedCleaned, Derived::Mode,
-    unroll ? int(DenseDerived::SizeAtCompileTime) : Dynamic,
-    true // clear the opposite triangular part
-    >::run(other.derived(), derived().nestedExpression());
+  evalToLazy(other.derived());
 }
 
 /***************************************************************************
@@ -719,49 +611,14 @@ void TriangularBase<Derived>::evalToLazy(MatrixBase<DenseDerived> &other) const
 * Implementation of MatrixBase methods
 ***************************************************************************/
 
-#ifdef EIGEN2_SUPPORT
-
-// implementation of part<>(), including the SelfAdjoint case.
-
-namespace internal {
-template<typename MatrixType, unsigned int Mode>
-struct eigen2_part_return_type
-{
-  typedef TriangularView<MatrixType, Mode> type;
-};
-
-template<typename MatrixType>
-struct eigen2_part_return_type<MatrixType, SelfAdjoint>
-{
-  typedef SelfAdjointView<MatrixType, Upper> type;
-};
-}
-
-/** \deprecated use MatrixBase::triangularView() */
-template<typename Derived>
-template<unsigned int Mode>
-const typename internal::eigen2_part_return_type<Derived, Mode>::type MatrixBase<Derived>::part() const
-{
-  return derived();
-}
-
-/** \deprecated use MatrixBase::triangularView() */
-template<typename Derived>
-template<unsigned int Mode>
-typename internal::eigen2_part_return_type<Derived, Mode>::type MatrixBase<Derived>::part()
-{
-  return derived();
-}
-#endif
-
 /**
   * \returns an expression of a triangular view extracted from the current matrix
   *
   * The parameter \a Mode can have the following values: \c #Upper, \c #StrictlyUpper, \c #UnitUpper,
   * \c #Lower, \c #StrictlyLower, \c #UnitLower.
   *
-  * Example: \include MatrixBase_extract.cpp
-  * Output: \verbinclude MatrixBase_extract.out
+  * Example: \include MatrixBase_triangularView.cpp
+  * Output: \verbinclude MatrixBase_triangularView.out
   *
   * \sa class TriangularView
   */
@@ -770,7 +627,7 @@ template<unsigned int Mode>
 typename MatrixBase<Derived>::template TriangularViewReturnType<Mode>::Type
 MatrixBase<Derived>::triangularView()
 {
-  return derived();
+  return typename TriangularViewReturnType<Mode>::Type(derived());
 }
 
 /** This is the const version of MatrixBase::triangularView() */
@@ -779,7 +636,7 @@ template<unsigned int Mode>
 typename MatrixBase<Derived>::template ConstTriangularViewReturnType<Mode>::Type
 MatrixBase<Derived>::triangularView() const
 {
-  return derived();
+  return typename ConstTriangularViewReturnType<Mode>::Type(derived());
 }
 
 /** \returns true if *this is approximately equal to an upper triangular matrix,
@@ -790,21 +647,20 @@ MatrixBase<Derived>::triangularView() const
 template<typename Derived>
 bool MatrixBase<Derived>::isUpperTriangular(const RealScalar& prec) const
 {
-  using std::abs;
   RealScalar maxAbsOnUpperPart = static_cast<RealScalar>(-1);
   for(Index j = 0; j < cols(); ++j)
   {
-    Index maxi = (std::min)(j, rows()-1);
+    Index maxi = numext::mini(j, rows()-1);
     for(Index i = 0; i <= maxi; ++i)
     {
-      RealScalar absValue = abs(coeff(i,j));
+      RealScalar absValue = numext::abs(coeff(i,j));
       if(absValue > maxAbsOnUpperPart) maxAbsOnUpperPart = absValue;
     }
   }
   RealScalar threshold = maxAbsOnUpperPart * prec;
   for(Index j = 0; j < cols(); ++j)
     for(Index i = j+1; i < rows(); ++i)
-      if(abs(coeff(i, j)) > threshold) return false;
+      if(numext::abs(coeff(i, j)) > threshold) return false;
   return true;
 }
 
@@ -816,24 +672,312 @@ bool MatrixBase<Derived>::isUpperTriangular(const RealScalar& prec) const
 template<typename Derived>
 bool MatrixBase<Derived>::isLowerTriangular(const RealScalar& prec) const
 {
-  using std::abs;
   RealScalar maxAbsOnLowerPart = static_cast<RealScalar>(-1);
   for(Index j = 0; j < cols(); ++j)
     for(Index i = j; i < rows(); ++i)
     {
-      RealScalar absValue = abs(coeff(i,j));
+      RealScalar absValue = numext::abs(coeff(i,j));
       if(absValue > maxAbsOnLowerPart) maxAbsOnLowerPart = absValue;
     }
   RealScalar threshold = maxAbsOnLowerPart * prec;
   for(Index j = 1; j < cols(); ++j)
   {
-    Index maxi = (std::min)(j, rows()-1);
+    Index maxi = numext::mini(j, rows()-1);
     for(Index i = 0; i < maxi; ++i)
-      if(abs(coeff(i, j)) > threshold) return false;
+      if(numext::abs(coeff(i, j)) > threshold) return false;
   }
   return true;
 }
 
+
+/***************************************************************************
+****************************************************************************
+* Evaluators and Assignment of triangular expressions
+***************************************************************************
+***************************************************************************/
+
+namespace internal {
+
+  
+// TODO currently a triangular expression has the form TriangularView<.,.>
+//      in the future triangular-ness should be defined by the expression traits
+//      such that Transpose<TriangularView<.,.> > is valid. (currently TriangularBase::transpose() is overloaded to make it work)
+template<typename MatrixType, unsigned int Mode>
+struct evaluator_traits<TriangularView<MatrixType,Mode> >
+{
+  typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
+  typedef typename glue_shapes<typename evaluator_traits<MatrixType>::Shape, TriangularShape>::type Shape;
+};
+
+template<typename MatrixType, unsigned int Mode>
+struct unary_evaluator<TriangularView<MatrixType,Mode>, IndexBased>
+ : evaluator<typename internal::remove_all<MatrixType>::type>
+{
+  typedef TriangularView<MatrixType,Mode> XprType;
+  typedef evaluator<typename internal::remove_all<MatrixType>::type> Base;
+  unary_evaluator(const XprType &xpr) : Base(xpr.nestedExpression()) {}
+};
+
+// Additional assignment kinds:
+struct Triangular2Triangular    {};
+struct Triangular2Dense         {};
+struct Dense2Triangular         {};
+
+
+template<typename Kernel, unsigned int Mode, int UnrollCount, bool ClearOpposite> struct triangular_assignment_loop;
+
+ 
+/** \internal Specialization of the dense assignment kernel for triangular matrices.
+  * The main difference is that the triangular, diagonal, and opposite parts are processed through three different functions.
+  * \tparam UpLo must be either Lower or Upper
+  * \tparam Mode must be either 0, UnitDiag, ZeroDiag, or SelfAdjoint
+  */
+template<int UpLo, int Mode, int SetOpposite, typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version = Specialized>
+class triangular_dense_assignment_kernel : public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version>
+{
+protected:
+  typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version> Base;
+  typedef typename Base::DstXprType DstXprType;
+  typedef typename Base::SrcXprType SrcXprType;
+  using Base::m_dst;
+  using Base::m_src;
+  using Base::m_functor;
+public:
+  
+  typedef typename Base::DstEvaluatorType DstEvaluatorType;
+  typedef typename Base::SrcEvaluatorType SrcEvaluatorType;
+  typedef typename Base::Scalar Scalar;
+  typedef typename Base::AssignmentTraits AssignmentTraits;
+  
+  
+  EIGEN_DEVICE_FUNC triangular_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
+    : Base(dst, src, func, dstExpr)
+  {}
+  
+#ifdef EIGEN_INTERNAL_DEBUGGING
+  EIGEN_DEVICE_FUNC void assignCoeff(Index row, Index col)
+  {
+    eigen_internal_assert(row!=col);
+    Base::assignCoeff(row,col);
+  }
+#else
+  using Base::assignCoeff;
+#endif
+  
+  EIGEN_DEVICE_FUNC void assignDiagonalCoeff(Index id)
+  {
+         if(Mode==UnitDiag && SetOpposite) m_functor.assignCoeff(m_dst.coeffRef(id,id), Scalar(1));
+    else if(Mode==ZeroDiag && SetOpposite) m_functor.assignCoeff(m_dst.coeffRef(id,id), Scalar(0));
+    else if(Mode==0)                       Base::assignCoeff(id,id);
+  }
+  
+  EIGEN_DEVICE_FUNC void assignOppositeCoeff(Index row, Index col)
+  { 
+    eigen_internal_assert(row!=col);
+    if(SetOpposite)
+      m_functor.assignCoeff(m_dst.coeffRef(row,col), Scalar(0));
+  }
+};
+
+template<int Mode, bool SetOpposite, typename DstXprType, typename SrcXprType, typename Functor>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_triangular_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func)
+{
+  typedef evaluator<DstXprType> DstEvaluatorType;
+  typedef evaluator<SrcXprType> SrcEvaluatorType;
+
+  SrcEvaluatorType srcEvaluator(src);
+
+  Index dstRows = src.rows();
+  Index dstCols = src.cols();
+  if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+    dst.resize(dstRows, dstCols);
+  DstEvaluatorType dstEvaluator(dst);
+    
+  typedef triangular_dense_assignment_kernel< Mode&(Lower|Upper),Mode&(UnitDiag|ZeroDiag|SelfAdjoint),SetOpposite,
+                                              DstEvaluatorType,SrcEvaluatorType,Functor> Kernel;
+  Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());
+  
+  enum {
+      unroll = DstXprType::SizeAtCompileTime != Dynamic
+            && SrcEvaluatorType::CoeffReadCost < HugeCost
+            && DstXprType::SizeAtCompileTime * (DstEvaluatorType::CoeffReadCost+SrcEvaluatorType::CoeffReadCost) / 2 <= EIGEN_UNROLLING_LIMIT
+    };
+  
+  triangular_assignment_loop<Kernel, Mode, unroll ? int(DstXprType::SizeAtCompileTime) : Dynamic, SetOpposite>::run(kernel);
+}
+
+template<int Mode, bool SetOpposite, typename DstXprType, typename SrcXprType>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_triangular_assignment_loop(DstXprType& dst, const SrcXprType& src)
+{
+  call_triangular_assignment_loop<Mode,SetOpposite>(dst, src, internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>());
+}
+
+template<> struct AssignmentKind<TriangularShape,TriangularShape> { typedef Triangular2Triangular Kind; };
+template<> struct AssignmentKind<DenseShape,TriangularShape>      { typedef Triangular2Dense      Kind; };
+template<> struct AssignmentKind<TriangularShape,DenseShape>      { typedef Dense2Triangular      Kind; };
+
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Triangular2Triangular>
+{
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+    eigen_assert(int(DstXprType::Mode) == int(SrcXprType::Mode));
+    
+    call_triangular_assignment_loop<DstXprType::Mode, false>(dst, src, func);  
+  }
+};
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Triangular2Dense>
+{
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+    call_triangular_assignment_loop<SrcXprType::Mode, (SrcXprType::Mode&SelfAdjoint)==0>(dst, src, func);  
+  }
+};
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Dense2Triangular>
+{
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+    call_triangular_assignment_loop<DstXprType::Mode, false>(dst, src, func);  
+  }
+};
+
+
+template<typename Kernel, unsigned int Mode, int UnrollCount, bool SetOpposite>
+struct triangular_assignment_loop
+{
+  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
+  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
+  typedef typename DstEvaluatorType::XprType DstXprType;
+  
+  enum {
+    col = (UnrollCount-1) / DstXprType::RowsAtCompileTime,
+    row = (UnrollCount-1) % DstXprType::RowsAtCompileTime
+  };
+  
+  typedef typename Kernel::Scalar Scalar;
+
+  EIGEN_DEVICE_FUNC
+  static inline void run(Kernel &kernel)
+  {
+    triangular_assignment_loop<Kernel, Mode, UnrollCount-1, SetOpposite>::run(kernel);
+    
+    if(row==col)
+      kernel.assignDiagonalCoeff(row);
+    else if( ((Mode&Lower) && row>col) || ((Mode&Upper) && row<col) )
+      kernel.assignCoeff(row,col);
+    else if(SetOpposite)
+      kernel.assignOppositeCoeff(row,col);
+  }
+};
+
+// prevent buggy user code from causing an infinite recursion
+template<typename Kernel, unsigned int Mode, bool SetOpposite>
+struct triangular_assignment_loop<Kernel, Mode, 0, SetOpposite>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(Kernel &) {}
+};
+
+
+
+// TODO: experiment with a recursive assignment procedure splitting the current
+//       triangular part into one rectangular and two triangular parts.
+
+
+template<typename Kernel, unsigned int Mode, bool SetOpposite>
+struct triangular_assignment_loop<Kernel, Mode, Dynamic, SetOpposite>
+{
+  typedef typename Kernel::Scalar Scalar;
+  EIGEN_DEVICE_FUNC
+  static inline void run(Kernel &kernel)
+  {
+    for(Index j = 0; j < kernel.cols(); ++j)
+    {
+      Index maxi = numext::mini(j, kernel.rows());
+      Index i = 0;
+      if (((Mode&Lower) && SetOpposite) || (Mode&Upper))
+      {
+        for(; i < maxi; ++i)
+          if(Mode&Upper) kernel.assignCoeff(i, j);
+          else           kernel.assignOppositeCoeff(i, j);
+      }
+      else
+        i = maxi;
+      
+      if(i<kernel.rows()) // then i==j
+        kernel.assignDiagonalCoeff(i++);
+      
+      if (((Mode&Upper) && SetOpposite) || (Mode&Lower))
+      {
+        for(; i < kernel.rows(); ++i)
+          if(Mode&Lower) kernel.assignCoeff(i, j);
+          else           kernel.assignOppositeCoeff(i, j);
+      }
+    }
+  }
+};
+
+} // end namespace internal
+
+/** Assigns a triangular or selfadjoint matrix to a dense matrix.
+  * If the matrix is triangular, the opposite part is set to zero. */
+template<typename Derived>
+template<typename DenseDerived>
+void TriangularBase<Derived>::evalToLazy(MatrixBase<DenseDerived> &other) const
+{
+  other.derived().resize(this->rows(), this->cols());
+  internal::call_triangular_assignment_loop<Derived::Mode,(Derived::Mode&SelfAdjoint)==0 /* SetOpposite */>(other.derived(), derived().nestedExpression());
+}
+
+namespace internal {
+  
+// Triangular = Product
+template< typename DstXprType, typename Lhs, typename Rhs, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::assign_op<Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, Dense2Triangular>
+{
+  typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename SrcXprType::Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    dst._assignProduct(src, 1, 0);
+  }
+};
+
+// Triangular += Product
+template< typename DstXprType, typename Lhs, typename Rhs, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::add_assign_op<Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, Dense2Triangular>
+{
+  typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<Scalar,typename SrcXprType::Scalar> &)
+  {
+    dst._assignProduct(src, 1, 1);
+  }
+};
+
+// Triangular -= Product
+template< typename DstXprType, typename Lhs, typename Rhs, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::sub_assign_op<Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, Dense2Triangular>
+{
+  typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<Scalar,typename SrcXprType::Scalar> &)
+  {
+    dst._assignProduct(src, -1, 1);
+  }
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_TRIANGULARMATRIX_H
diff --git a/extern/Eigen3/Eigen/src/Core/VectorBlock.h b/extern/Eigen3/Eigen/src/Core/VectorBlock.h
index 1a7330f3cfc..d72fbf7e99d 100644
--- a/extern/Eigen3/Eigen/src/Core/VectorBlock.h
+++ b/extern/Eigen3/Eigen/src/Core/VectorBlock.h
@@ -13,13 +13,23 @@
 
 namespace Eigen { 
 
+namespace internal {
+template<typename VectorType, int Size>
+struct traits<VectorBlock<VectorType, Size> >
+  : public traits<Block<VectorType,
+                     traits<VectorType>::Flags & RowMajorBit ? 1 : Size,
+                     traits<VectorType>::Flags & RowMajorBit ? Size : 1> >
+{
+};
+}
+
 /** \class VectorBlock
   * \ingroup Core_Module
   *
   * \brief Expression of a fixed-size or dynamic-size sub-vector
   *
-  * \param VectorType the type of the object in which we are taking a sub-vector
-  * \param Size size of the sub-vector we are taking at compile time (optional)
+  * \tparam VectorType the type of the object in which we are taking a sub-vector
+  * \tparam Size size of the sub-vector we are taking at compile time (optional)
   *
   * This class represents an expression of either a fixed-size or dynamic-size sub-vector.
   * It is the return type of DenseBase::segment(Index,Index) and DenseBase::segment<int>(Index) and
@@ -43,17 +53,6 @@ namespace Eigen {
   *
   * \sa class Block, DenseBase::segment(Index,Index,Index,Index), DenseBase::segment(Index,Index)
   */
-
-namespace internal {
-template<typename VectorType, int Size>
-struct traits<VectorBlock<VectorType, Size> >
-  : public traits<Block<VectorType,
-                     traits<VectorType>::Flags & RowMajorBit ? 1 : Size,
-                     traits<VectorType>::Flags & RowMajorBit ? Size : 1> >
-{
-};
-}
-
 template<typename VectorType, int Size> class VectorBlock
   : public Block<VectorType,
                      internal::traits<VectorType>::Flags & RowMajorBit ? 1 : Size,
@@ -72,6 +71,7 @@ template<typename VectorType, int Size> class VectorBlock
 
     /** Dynamic-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline VectorBlock(VectorType& vector, Index start, Index size)
       : Base(vector,
              IsColVector ? start : 0, IsColVector ? 0 : start,
@@ -82,6 +82,7 @@ template<typename VectorType, int Size> class VectorBlock
 
     /** Fixed-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline VectorBlock(VectorType& vector, Index start)
       : Base(vector, IsColVector ? start : 0, IsColVector ? 0 : start)
     {
diff --git a/extern/Eigen3/Eigen/src/Core/VectorwiseOp.h b/extern/Eigen3/Eigen/src/Core/VectorwiseOp.h
index d5ab036647e..4fe267e9f11 100644
--- a/extern/Eigen3/Eigen/src/Core/VectorwiseOp.h
+++ b/extern/Eigen3/Eigen/src/Core/VectorwiseOp.h
@@ -11,7 +11,7 @@
 #ifndef EIGEN_PARTIAL_REDUX_H
 #define EIGEN_PARTIAL_REDUX_H
 
-namespace Eigen { 
+namespace Eigen {
 
 /** \class PartialReduxExpr
   * \ingroup Core_Module
@@ -41,64 +41,43 @@ struct traits<PartialReduxExpr<MatrixType, MemberOp, Direction> >
   typedef typename traits<MatrixType>::StorageKind StorageKind;
   typedef typename traits<MatrixType>::XprKind XprKind;
   typedef typename MatrixType::Scalar InputScalar;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
-  typedef typename remove_all<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsAtCompileTime = Direction==Vertical   ? 1 : MatrixType::RowsAtCompileTime,
     ColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = Direction==Vertical   ? 1 : MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::MaxColsAtCompileTime,
-    Flags0 = (unsigned int)_MatrixTypeNested::Flags & HereditaryBits,
-    Flags = (Flags0 & ~RowMajorBit) | (RowsAtCompileTime == 1 ? RowMajorBit : 0),
+    Flags = RowsAtCompileTime == 1 ? RowMajorBit : 0,
     TraversalSize = Direction==Vertical ? MatrixType::RowsAtCompileTime :  MatrixType::ColsAtCompileTime
   };
-  #if EIGEN_GNUC_AT_LEAST(3,4)
-  typedef typename MemberOp::template Cost<InputScalar,int(TraversalSize)> CostOpType;
-  #else
-  typedef typename MemberOp::template Cost<InputScalar,TraversalSize> CostOpType;
-  #endif
-  enum {
-    CoeffReadCost = TraversalSize==Dynamic ? Dynamic
-                  : TraversalSize * traits<_MatrixTypeNested>::CoeffReadCost + int(CostOpType::value)
-  };
 };
 }
 
 template< typename MatrixType, typename MemberOp, int Direction>
-class PartialReduxExpr : internal::no_assignment_operator,
-  public internal::dense_xpr_base< PartialReduxExpr<MatrixType, MemberOp, Direction> >::type
+class PartialReduxExpr : public internal::dense_xpr_base< PartialReduxExpr<MatrixType, MemberOp, Direction> >::type,
+                         internal::no_assignment_operator
 {
   public:
 
     typedef typename internal::dense_xpr_base<PartialReduxExpr>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(PartialReduxExpr)
-    typedef typename internal::traits<PartialReduxExpr>::MatrixTypeNested MatrixTypeNested;
-    typedef typename internal::traits<PartialReduxExpr>::_MatrixTypeNested _MatrixTypeNested;
 
-    PartialReduxExpr(const MatrixType& mat, const MemberOp& func = MemberOp())
+    EIGEN_DEVICE_FUNC
+    explicit PartialReduxExpr(const MatrixType& mat, const MemberOp& func = MemberOp())
       : m_matrix(mat), m_functor(func) {}
 
+    EIGEN_DEVICE_FUNC
     Index rows() const { return (Direction==Vertical   ? 1 : m_matrix.rows()); }
+    EIGEN_DEVICE_FUNC
     Index cols() const { return (Direction==Horizontal ? 1 : m_matrix.cols()); }
 
-    EIGEN_STRONG_INLINE const Scalar coeff(Index i, Index j) const
-    {
-      if (Direction==Vertical)
-        return m_functor(m_matrix.col(j));
-      else
-        return m_functor(m_matrix.row(i));
-    }
+    EIGEN_DEVICE_FUNC
+    typename MatrixType::Nested nestedExpression() const { return m_matrix; }
 
-    const Scalar coeff(Index index) const
-    {
-      if (Direction==Vertical)
-        return m_functor(m_matrix.col(index));
-      else
-        return m_functor(m_matrix.row(index));
-    }
+    EIGEN_DEVICE_FUNC
+    const MemberOp& functor() const { return m_functor; }
 
   protected:
-    MatrixTypeNested m_matrix;
+    typename MatrixType::Nested m_matrix;
     const MemberOp m_functor;
 };
 
@@ -110,7 +89,8 @@ class PartialReduxExpr : internal::no_assignment_operator,
     template<typename Scalar, int Size> struct Cost                     \
     { enum { value = COST }; };                                         \
     template<typename XprType>                                          \
-    EIGEN_STRONG_INLINE ResultType operator()(const XprType& mat) const \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE                               \
+    ResultType operator()(const XprType& mat) const                     \
     { return mat.MEMBER(); } \
   }
 
@@ -130,17 +110,27 @@ EIGEN_MEMBER_FUNCTOR(any, (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(count, (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(prod, (Size-1)*NumTraits<Scalar>::MulCost);
 
+template <int p, typename ResultType>
+struct member_lpnorm {
+  typedef ResultType result_type;
+  template<typename Scalar, int Size> struct Cost
+  { enum { value = (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost }; };
+  EIGEN_DEVICE_FUNC member_lpnorm() {}
+  template<typename XprType>
+  EIGEN_DEVICE_FUNC inline ResultType operator()(const XprType& mat) const
+  { return mat.template lpNorm<p>(); }
+};
 
 template <typename BinaryOp, typename Scalar>
 struct member_redux {
   typedef typename result_of<
-                     BinaryOp(Scalar)
+                     BinaryOp(const Scalar&,const Scalar&)
                    >::type  result_type;
   template<typename _Scalar, int Size> struct Cost
   { enum { value = (Size-1) * functor_traits<BinaryOp>::Cost }; };
-  member_redux(const BinaryOp func) : m_functor(func) {}
+  EIGEN_DEVICE_FUNC explicit member_redux(const BinaryOp func) : m_functor(func) {}
   template<typename Derived>
-  inline result_type operator()(const DenseBase<Derived>& mat) const
+  EIGEN_DEVICE_FUNC inline result_type operator()(const DenseBase<Derived>& mat) const
   { return mat.redux(m_functor); }
   const BinaryOp m_functor;
 };
@@ -151,8 +141,8 @@ struct member_redux {
   *
   * \brief Pseudo expression providing partial reduction operations
   *
-  * \param ExpressionType the type of the object on which to do partial reductions
-  * \param Direction indicates the direction of the redux (#Vertical or #Horizontal)
+  * \tparam ExpressionType the type of the object on which to do partial reductions
+  * \tparam Direction indicates the direction of the redux (#Vertical or #Horizontal)
   *
   * This class represents a pseudo expression with partial reduction features.
   * It is the return type of DenseBase::colwise() and DenseBase::rowwise()
@@ -169,16 +159,15 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
 
     typedef typename ExpressionType::Scalar Scalar;
     typedef typename ExpressionType::RealScalar RealScalar;
-    typedef typename ExpressionType::Index Index;
-    typedef typename internal::conditional<internal::must_nest_by_value<ExpressionType>::ret,
-        ExpressionType, ExpressionType&>::type ExpressionTypeNested;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename internal::ref_selector<ExpressionType>::non_const_type ExpressionTypeNested;
     typedef typename internal::remove_all<ExpressionTypeNested>::type ExpressionTypeNestedCleaned;
 
     template<template<typename _Scalar> class Functor,
-                      typename Scalar=typename internal::traits<ExpressionType>::Scalar> struct ReturnType
+                      typename Scalar_=Scalar> struct ReturnType
     {
       typedef PartialReduxExpr<ExpressionType,
-                               Functor<Scalar>,
+                               Functor<Scalar_>,
                                Direction
                               > Type;
     };
@@ -186,23 +175,24 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     template<typename BinaryOp> struct ReduxReturnType
     {
       typedef PartialReduxExpr<ExpressionType,
-                               internal::member_redux<BinaryOp,typename internal::traits<ExpressionType>::Scalar>,
+                               internal::member_redux<BinaryOp,Scalar>,
                                Direction
                               > Type;
     };
 
     enum {
-      IsVertical   = (Direction==Vertical) ? 1 : 0,
-      IsHorizontal = (Direction==Horizontal) ? 1 : 0
+      isVertical   = (Direction==Vertical) ? 1 : 0,
+      isHorizontal = (Direction==Horizontal) ? 1 : 0
     };
 
   protected:
 
-    /** \internal
-      * \returns the i-th subvector according to the \c Direction */
-    typedef typename internal::conditional<Direction==Vertical,
+    typedef typename internal::conditional<isVertical,
                                typename ExpressionType::ColXpr,
                                typename ExpressionType::RowXpr>::type SubVector;
+    /** \internal
+      * \returns the i-th subvector according to the \c Direction */
+    EIGEN_DEVICE_FUNC
     SubVector subVector(Index i)
     {
       return SubVector(m_matrix.derived(),i);
@@ -210,58 +200,62 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
 
     /** \internal
       * \returns the number of subvectors in the direction \c Direction */
+    EIGEN_DEVICE_FUNC
     Index subVectors() const
-    { return Direction==Vertical?m_matrix.cols():m_matrix.rows(); }
+    { return isVertical?m_matrix.cols():m_matrix.rows(); }
 
     template<typename OtherDerived> struct ExtendedType {
       typedef Replicate<OtherDerived,
-                        Direction==Vertical   ? 1 : ExpressionType::RowsAtCompileTime,
-                        Direction==Horizontal ? 1 : ExpressionType::ColsAtCompileTime> Type;
+                        isVertical   ? 1 : ExpressionType::RowsAtCompileTime,
+                        isHorizontal ? 1 : ExpressionType::ColsAtCompileTime> Type;
     };
 
     /** \internal
       * Replicates a vector to match the size of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     typename ExtendedType<OtherDerived>::Type
     extendedTo(const DenseBase<OtherDerived>& other) const
     {
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Vertical, OtherDerived::MaxColsAtCompileTime==1),
+      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isVertical, OtherDerived::MaxColsAtCompileTime==1),
                           YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED)
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Horizontal, OtherDerived::MaxRowsAtCompileTime==1),
+      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isHorizontal, OtherDerived::MaxRowsAtCompileTime==1),
                           YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED)
       return typename ExtendedType<OtherDerived>::Type
                       (other.derived(),
-                       Direction==Vertical   ? 1 : m_matrix.rows(),
-                       Direction==Horizontal ? 1 : m_matrix.cols());
+                       isVertical   ? 1 : m_matrix.rows(),
+                       isHorizontal ? 1 : m_matrix.cols());
     }
-    
+
     template<typename OtherDerived> struct OppositeExtendedType {
       typedef Replicate<OtherDerived,
-                        Direction==Horizontal ? 1 : ExpressionType::RowsAtCompileTime,
-                        Direction==Vertical   ? 1 : ExpressionType::ColsAtCompileTime> Type;
+                        isHorizontal ? 1 : ExpressionType::RowsAtCompileTime,
+                        isVertical   ? 1 : ExpressionType::ColsAtCompileTime> Type;
     };
 
     /** \internal
       * Replicates a vector in the opposite direction to match the size of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     typename OppositeExtendedType<OtherDerived>::Type
     extendedToOpposite(const DenseBase<OtherDerived>& other) const
     {
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Horizontal, OtherDerived::MaxColsAtCompileTime==1),
+      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isHorizontal, OtherDerived::MaxColsAtCompileTime==1),
                           YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED)
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Vertical, OtherDerived::MaxRowsAtCompileTime==1),
+      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isVertical, OtherDerived::MaxRowsAtCompileTime==1),
                           YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED)
       return typename OppositeExtendedType<OtherDerived>::Type
                       (other.derived(),
-                       Direction==Horizontal  ? 1 : m_matrix.rows(),
-                       Direction==Vertical    ? 1 : m_matrix.cols());
+                       isHorizontal  ? 1 : m_matrix.rows(),
+                       isVertical    ? 1 : m_matrix.cols());
     }
 
   public:
-
-    inline VectorwiseOp(ExpressionType& matrix) : m_matrix(matrix) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline VectorwiseOp(ExpressionType& matrix) : m_matrix(matrix) {}
 
     /** \internal */
+    EIGEN_DEVICE_FUNC
     inline const ExpressionType& _expression() const { return m_matrix; }
 
     /** \returns a row or column vector expression of \c *this reduxed by \a func
@@ -272,80 +266,126 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       * \sa class VectorwiseOp, DenseBase::colwise(), DenseBase::rowwise()
       */
     template<typename BinaryOp>
+    EIGEN_DEVICE_FUNC
     const typename ReduxReturnType<BinaryOp>::Type
     redux(const BinaryOp& func = BinaryOp()) const
-    { return typename ReduxReturnType<BinaryOp>::Type(_expression(), func); }
+    { return typename ReduxReturnType<BinaryOp>::Type(_expression(), internal::member_redux<BinaryOp,Scalar>(func)); }
+
+    typedef typename ReturnType<internal::member_minCoeff>::Type MinCoeffReturnType;
+    typedef typename ReturnType<internal::member_maxCoeff>::Type MaxCoeffReturnType;
+    typedef typename ReturnType<internal::member_squaredNorm,RealScalar>::Type SquaredNormReturnType;
+    typedef typename ReturnType<internal::member_norm,RealScalar>::Type NormReturnType;
+    typedef typename ReturnType<internal::member_blueNorm,RealScalar>::Type BlueNormReturnType;
+    typedef typename ReturnType<internal::member_stableNorm,RealScalar>::Type StableNormReturnType;
+    typedef typename ReturnType<internal::member_hypotNorm,RealScalar>::Type HypotNormReturnType;
+    typedef typename ReturnType<internal::member_sum>::Type SumReturnType;
+    typedef typename ReturnType<internal::member_mean>::Type MeanReturnType;
+    typedef typename ReturnType<internal::member_all>::Type AllReturnType;
+    typedef typename ReturnType<internal::member_any>::Type AnyReturnType;
+    typedef PartialReduxExpr<ExpressionType, internal::member_count<Index>, Direction> CountReturnType;
+    typedef typename ReturnType<internal::member_prod>::Type ProdReturnType;
+    typedef Reverse<const ExpressionType, Direction> ConstReverseReturnType;
+    typedef Reverse<ExpressionType, Direction> ReverseReturnType;
+
+    template<int p> struct LpNormReturnType {
+      typedef PartialReduxExpr<ExpressionType, internal::member_lpnorm<p,RealScalar>,Direction> Type;
+    };
 
     /** \returns a row (or column) vector expression of the smallest coefficient
       * of each column (or row) of the referenced expression.
-      * 
+      *
       * \warning the result is undefined if \c *this contains NaN.
       *
       * Example: \include PartialRedux_minCoeff.cpp
       * Output: \verbinclude PartialRedux_minCoeff.out
       *
       * \sa DenseBase::minCoeff() */
-    const typename ReturnType<internal::member_minCoeff>::Type minCoeff() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const MinCoeffReturnType minCoeff() const
+    { return MinCoeffReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the largest coefficient
       * of each column (or row) of the referenced expression.
-      * 
+      *
       * \warning the result is undefined if \c *this contains NaN.
       *
       * Example: \include PartialRedux_maxCoeff.cpp
       * Output: \verbinclude PartialRedux_maxCoeff.out
       *
       * \sa DenseBase::maxCoeff() */
-    const typename ReturnType<internal::member_maxCoeff>::Type maxCoeff() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const MaxCoeffReturnType maxCoeff() const
+    { return MaxCoeffReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the squared norm
       * of each column (or row) of the referenced expression.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * Example: \include PartialRedux_squaredNorm.cpp
       * Output: \verbinclude PartialRedux_squaredNorm.out
       *
       * \sa DenseBase::squaredNorm() */
-    const typename ReturnType<internal::member_squaredNorm,RealScalar>::Type squaredNorm() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const SquaredNormReturnType squaredNorm() const
+    { return SquaredNormReturnType(_expression()); }
+
+    /** \returns a row (or column) vector expression of the norm
+      * of each column (or row) of the referenced expression.
+      * This is a vector with real entries, even if the original matrix has complex entries.
+      *
+      * Example: \include PartialRedux_norm.cpp
+      * Output: \verbinclude PartialRedux_norm.out
+      *
+      * \sa DenseBase::norm() */
+    EIGEN_DEVICE_FUNC
+    const NormReturnType norm() const
+    { return NormReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * Example: \include PartialRedux_norm.cpp
       * Output: \verbinclude PartialRedux_norm.out
       *
       * \sa DenseBase::norm() */
-    const typename ReturnType<internal::member_norm,RealScalar>::Type norm() const
-    { return _expression(); }
+    template<int p>
+    EIGEN_DEVICE_FUNC
+    const typename LpNormReturnType<p>::Type lpNorm() const
+    { return typename LpNormReturnType<p>::Type(_expression()); }
 
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression, using
-      * blue's algorithm.
+      * Blue's algorithm.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * \sa DenseBase::blueNorm() */
-    const typename ReturnType<internal::member_blueNorm,RealScalar>::Type blueNorm() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const BlueNormReturnType blueNorm() const
+    { return BlueNormReturnType(_expression()); }
 
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression, avoiding
       * underflow and overflow.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * \sa DenseBase::stableNorm() */
-    const typename ReturnType<internal::member_stableNorm,RealScalar>::Type stableNorm() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const StableNormReturnType stableNorm() const
+    { return StableNormReturnType(_expression()); }
 
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression, avoiding
       * underflow and overflow using a concatenation of hypot() calls.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * \sa DenseBase::hypotNorm() */
-    const typename ReturnType<internal::member_hypotNorm,RealScalar>::Type hypotNorm() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const HypotNormReturnType hypotNorm() const
+    { return HypotNormReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the sum
       * of each column (or row) of the referenced expression.
@@ -354,39 +394,48 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       * Output: \verbinclude PartialRedux_sum.out
       *
       * \sa DenseBase::sum() */
-    const typename ReturnType<internal::member_sum>::Type sum() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const SumReturnType sum() const
+    { return SumReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the mean
     * of each column (or row) of the referenced expression.
     *
     * \sa DenseBase::mean() */
-    const typename ReturnType<internal::member_mean>::Type mean() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const MeanReturnType mean() const
+    { return MeanReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression representing
       * whether \b all coefficients of each respective column (or row) are \c true.
+      * This expression can be assigned to a vector with entries of type \c bool.
       *
       * \sa DenseBase::all() */
-    const typename ReturnType<internal::member_all>::Type all() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const AllReturnType all() const
+    { return AllReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression representing
       * whether \b at \b least one coefficient of each respective column (or row) is \c true.
+      * This expression can be assigned to a vector with entries of type \c bool.
       *
       * \sa DenseBase::any() */
-    const typename ReturnType<internal::member_any>::Type any() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const AnyReturnType any() const
+    { return AnyReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression representing
       * the number of \c true coefficients of each respective column (or row).
+      * This expression can be assigned to a vector whose entries have the same type as is used to
+      * index entries of the original matrix; for dense matrices, this is \c std::ptrdiff_t .
       *
       * Example: \include PartialRedux_count.cpp
       * Output: \verbinclude PartialRedux_count.out
       *
       * \sa DenseBase::count() */
-    const PartialReduxExpr<ExpressionType, internal::member_count<Index>, Direction> count() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const CountReturnType count() const
+    { return CountReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the product
       * of each column (or row) of the referenced expression.
@@ -395,8 +444,9 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       * Output: \verbinclude PartialRedux_prod.out
       *
       * \sa DenseBase::prod() */
-    const typename ReturnType<internal::member_prod>::Type prod() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const ProdReturnType prod() const
+    { return ProdReturnType(_expression()); }
 
 
     /** \returns a matrix expression
@@ -406,10 +456,20 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       * Output: \verbinclude Vectorwise_reverse.out
       *
       * \sa DenseBase::reverse() */
-    const Reverse<ExpressionType, Direction> reverse() const
-    { return Reverse<ExpressionType, Direction>( _expression() ); }
+    EIGEN_DEVICE_FUNC
+    const ConstReverseReturnType reverse() const
+    { return ConstReverseReturnType( _expression() ); }
 
-    typedef Replicate<ExpressionType,Direction==Vertical?Dynamic:1,Direction==Horizontal?Dynamic:1> ReplicateReturnType;
+    /** \returns a writable matrix expression
+      * where each column (or row) are reversed.
+      *
+      * \sa reverse() const */
+    EIGEN_DEVICE_FUNC
+    ReverseReturnType reverse()
+    { return ReverseReturnType( _expression() ); }
+
+    typedef Replicate<ExpressionType,(isVertical?Dynamic:1),(isHorizontal?Dynamic:1)> ReplicateReturnType;
+    EIGEN_DEVICE_FUNC
     const ReplicateReturnType replicate(Index factor) const;
 
     /**
@@ -421,17 +481,20 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       * \sa VectorwiseOp::replicate(Index), DenseBase::replicate(), class Replicate
       */
     // NOTE implemented here because of sunstudio's compilation errors
-    template<int Factor> const Replicate<ExpressionType,(IsVertical?Factor:1),(IsHorizontal?Factor:1)>
+    // isVertical*Factor+isHorizontal instead of (isVertical?Factor:1) to handle CUDA bug with ternary operator
+    template<int Factor> const Replicate<ExpressionType,isVertical*Factor+isHorizontal,isHorizontal*Factor+isVertical>
+    EIGEN_DEVICE_FUNC
     replicate(Index factor = Factor) const
     {
-      return Replicate<ExpressionType,Direction==Vertical?Factor:1,Direction==Horizontal?Factor:1>
-          (_expression(),Direction==Vertical?factor:1,Direction==Horizontal?factor:1);
+      return Replicate<ExpressionType,(isVertical?Factor:1),(isHorizontal?Factor:1)>
+          (_expression(),isVertical?factor:1,isHorizontal?factor:1);
     }
 
 /////////// Artithmetic operators ///////////
 
     /** Copies the vector \a other to each subvector of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -442,6 +505,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
 
     /** Adds the vector \a other to each subvector of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator+=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -451,6 +515,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
 
     /** Substracts the vector \a other to each subvector of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator-=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -460,6 +525,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
 
     /** Multiples each subvector of \c *this by the vector \a other */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator*=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -471,6 +537,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
 
     /** Divides each subvector of \c *this by the vector \a other */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator/=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -481,8 +548,8 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     }
 
     /** Returns the expression of the sum of the vector \a other to each subvector of \c *this */
-    template<typename OtherDerived> EIGEN_STRONG_INLINE
-    CwiseBinaryOp<internal::scalar_sum_op<Scalar>,
+    template<typename OtherDerived> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
+    CwiseBinaryOp<internal::scalar_sum_op<Scalar,typename OtherDerived::Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename ExtendedType<OtherDerived>::Type>
     operator+(const DenseBase<OtherDerived>& other) const
@@ -494,7 +561,8 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
 
     /** Returns the expression of the difference between each subvector of \c *this and the vector \a other */
     template<typename OtherDerived>
-    CwiseBinaryOp<internal::scalar_difference_op<Scalar>,
+    EIGEN_DEVICE_FUNC
+    CwiseBinaryOp<internal::scalar_difference_op<Scalar,typename OtherDerived::Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename ExtendedType<OtherDerived>::Type>
     operator-(const DenseBase<OtherDerived>& other) const
@@ -506,10 +574,11 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
 
     /** Returns the expression where each subvector is the product of the vector \a other
       * by the corresponding subvector of \c *this */
-    template<typename OtherDerived> EIGEN_STRONG_INLINE
+    template<typename OtherDerived> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
     CwiseBinaryOp<internal::scalar_product_op<Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename ExtendedType<OtherDerived>::Type>
+    EIGEN_DEVICE_FUNC
     operator*(const DenseBase<OtherDerived>& other) const
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -521,6 +590,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     /** Returns the expression where each subvector is the quotient of the corresponding
       * subvector of \c *this by the vector \a other */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     CwiseBinaryOp<internal::scalar_quotient_op<Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename ExtendedType<OtherDerived>::Type>
@@ -531,32 +601,36 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
       return m_matrix / extendedTo(other.derived());
     }
-    
-    /** \returns an expression where each column of row of the referenced matrix are normalized.
+
+    /** \returns an expression where each column (or row) of the referenced matrix are normalized.
       * The referenced matrix is \b not modified.
       * \sa MatrixBase::normalized(), normalize()
       */
+    EIGEN_DEVICE_FUNC
     CwiseBinaryOp<internal::scalar_quotient_op<Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename OppositeExtendedType<typename ReturnType<internal::member_norm,RealScalar>::Type>::Type>
     normalized() const { return m_matrix.cwiseQuotient(extendedToOpposite(this->norm())); }
-    
-    
+
+
     /** Normalize in-place each row or columns of the referenced matrix.
       * \sa MatrixBase::normalize(), normalized()
       */
-    void normalize() {
+    EIGEN_DEVICE_FUNC void normalize() {
       m_matrix = this->normalized();
     }
 
+    EIGEN_DEVICE_FUNC inline void reverseInPlace();
+
 /////////// Geometry module ///////////
 
-    #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
-    Homogeneous<ExpressionType,Direction> homogeneous() const;
-    #endif
+    typedef Homogeneous<ExpressionType,Direction> HomogeneousReturnType;
+    EIGEN_DEVICE_FUNC
+    HomogeneousReturnType homogeneous() const;
 
     typedef typename ExpressionType::PlainObject CrossReturnType;
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     const CrossReturnType cross(const MatrixBase<OtherDerived>& other) const;
 
     enum {
@@ -581,25 +655,15 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
                   Direction==Horizontal ? HNormalized_SizeMinusOne : 1> >
             HNormalizedReturnType;
 
+    EIGEN_DEVICE_FUNC
     const HNormalizedReturnType hnormalized() const;
 
   protected:
     ExpressionTypeNested m_matrix;
 };
 
-/** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
-  *
-  * Example: \include MatrixBase_colwise.cpp
-  * Output: \verbinclude MatrixBase_colwise.out
-  *
-  * \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
-  */
-template<typename Derived>
-inline const typename DenseBase<Derived>::ConstColwiseReturnType
-DenseBase<Derived>::colwise() const
-{
-  return derived();
-}
+//const colwise moved to DenseBase.h due to CUDA compiler bug
+
 
 /** \returns a writable VectorwiseOp wrapper of *this providing additional partial reduction operations
   *
@@ -609,22 +673,11 @@ template<typename Derived>
 inline typename DenseBase<Derived>::ColwiseReturnType
 DenseBase<Derived>::colwise()
 {
-  return derived();
+  return ColwiseReturnType(derived());
 }
 
-/** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
-  *
-  * Example: \include MatrixBase_rowwise.cpp
-  * Output: \verbinclude MatrixBase_rowwise.out
-  *
-  * \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
-  */
-template<typename Derived>
-inline const typename DenseBase<Derived>::ConstRowwiseReturnType
-DenseBase<Derived>::rowwise() const
-{
-  return derived();
-}
+//const rowwise moved to DenseBase.h due to CUDA compiler bug
+
 
 /** \returns a writable VectorwiseOp wrapper of *this providing additional partial reduction operations
   *
@@ -634,7 +687,7 @@ template<typename Derived>
 inline typename DenseBase<Derived>::RowwiseReturnType
 DenseBase<Derived>::rowwise()
 {
-  return derived();
+  return RowwiseReturnType(derived());
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/Visitor.h b/extern/Eigen3/Eigen/src/Core/Visitor.h
index 64867b7a2cb..54c1883d98f 100644
--- a/extern/Eigen3/Eigen/src/Core/Visitor.h
+++ b/extern/Eigen3/Eigen/src/Core/Visitor.h
@@ -22,6 +22,7 @@ struct visitor_impl
     row = (UnrollCount-1) % Derived::RowsAtCompileTime
   };
 
+  EIGEN_DEVICE_FUNC
   static inline void run(const Derived &mat, Visitor& visitor)
   {
     visitor_impl<Visitor, Derived, UnrollCount-1>::run(mat, visitor);
@@ -32,6 +33,7 @@ struct visitor_impl
 template<typename Visitor, typename Derived>
 struct visitor_impl<Visitor, Derived, 1>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(const Derived &mat, Visitor& visitor)
   {
     return visitor.init(mat.coeff(0, 0), 0, 0);
@@ -41,7 +43,7 @@ struct visitor_impl<Visitor, Derived, 1>
 template<typename Visitor, typename Derived>
 struct visitor_impl<Visitor, Derived, Dynamic>
 {
-  typedef typename Derived::Index Index;
+  EIGEN_DEVICE_FUNC
   static inline void run(const Derived& mat, Visitor& visitor)
   {
     visitor.init(mat.coeff(0,0), 0, 0);
@@ -53,6 +55,33 @@ struct visitor_impl<Visitor, Derived, Dynamic>
   }
 };
 
+// evaluator adaptor
+template<typename XprType>
+class visitor_evaluator
+{
+public:
+  EIGEN_DEVICE_FUNC
+  explicit visitor_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
+  
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  
+  enum {
+    RowsAtCompileTime = XprType::RowsAtCompileTime,
+    CoeffReadCost = internal::evaluator<XprType>::CoeffReadCost
+  };
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
+  EIGEN_DEVICE_FUNC Index size() const { return m_xpr.size(); }
+
+  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  { return m_evaluator.coeff(row, col); }
+  
+protected:
+  internal::evaluator<XprType> m_evaluator;
+  const XprType &m_xpr;
+};
 } // end namespace internal
 
 /** Applies the visitor \a visitor to the whole coefficients of the matrix or vector.
@@ -74,16 +103,17 @@ struct visitor_impl<Visitor, Derived, Dynamic>
   */
 template<typename Derived>
 template<typename Visitor>
+EIGEN_DEVICE_FUNC
 void DenseBase<Derived>::visit(Visitor& visitor) const
 {
-  enum { unroll = SizeAtCompileTime != Dynamic
-                   && CoeffReadCost != Dynamic
-                   && (SizeAtCompileTime == 1 || internal::functor_traits<Visitor>::Cost != Dynamic)
-                   && SizeAtCompileTime * CoeffReadCost + (SizeAtCompileTime-1) * internal::functor_traits<Visitor>::Cost
-                      <= EIGEN_UNROLLING_LIMIT };
-  return internal::visitor_impl<Visitor, Derived,
-      unroll ? int(SizeAtCompileTime) : Dynamic
-    >::run(derived(), visitor);
+  typedef typename internal::visitor_evaluator<Derived> ThisEvaluator;
+  ThisEvaluator thisEval(derived());
+  
+  enum {
+    unroll =  SizeAtCompileTime != Dynamic
+           && SizeAtCompileTime * ThisEvaluator::CoeffReadCost + (SizeAtCompileTime-1) * internal::functor_traits<Visitor>::Cost <= EIGEN_UNROLLING_LIMIT
+  };
+  return internal::visitor_impl<Visitor, ThisEvaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(thisEval, visitor);
 }
 
 namespace internal {
@@ -94,10 +124,10 @@ namespace internal {
 template <typename Derived>
 struct coeff_visitor
 {
-  typedef typename Derived::Index Index;
   typedef typename Derived::Scalar Scalar;
   Index row, col;
   Scalar res;
+  EIGEN_DEVICE_FUNC
   inline void init(const Scalar& value, Index i, Index j)
   {
     res = value;
@@ -114,8 +144,8 @@ struct coeff_visitor
 template <typename Derived>
 struct min_coeff_visitor : coeff_visitor<Derived>
 {
-  typedef typename Derived::Index Index;
   typedef typename Derived::Scalar Scalar;
+  EIGEN_DEVICE_FUNC
   void operator() (const Scalar& value, Index i, Index j)
   {
     if(value < this->res)
@@ -142,8 +172,8 @@ struct functor_traits<min_coeff_visitor<Scalar> > {
 template <typename Derived>
 struct max_coeff_visitor : coeff_visitor<Derived>
 {
-  typedef typename Derived::Index Index;
-  typedef typename Derived::Scalar Scalar;
+  typedef typename Derived::Scalar Scalar; 
+  EIGEN_DEVICE_FUNC
   void operator() (const Scalar& value, Index i, Index j)
   {
     if(value > this->res)
@@ -164,13 +194,15 @@ struct functor_traits<max_coeff_visitor<Scalar> > {
 
 } // end namespace internal
 
-/** \returns the minimum of all coefficients of *this and puts in *row and *col its location.
+/** \fn DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
+  * \returns the minimum of all coefficients of *this and puts in *row and *col its location.
   * \warning the result is undefined if \c *this contains NaN.
   *
-  * \sa DenseBase::minCoeff(Index*), DenseBase::maxCoeff(Index*,Index*), DenseBase::visitor(), DenseBase::minCoeff()
+  * \sa DenseBase::minCoeff(Index*), DenseBase::maxCoeff(Index*,Index*), DenseBase::visit(), DenseBase::minCoeff()
   */
 template<typename Derived>
 template<typename IndexType>
+EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
 {
@@ -184,27 +216,30 @@ DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
 /** \returns the minimum of all coefficients of *this and puts in *index its location.
   * \warning the result is undefined if \c *this contains NaN. 
   *
-  * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::visitor(), DenseBase::minCoeff()
+  * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::minCoeff()
   */
 template<typename Derived>
 template<typename IndexType>
+EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff(IndexType* index) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   internal::min_coeff_visitor<Derived> minVisitor;
   this->visit(minVisitor);
-  *index = (RowsAtCompileTime==1) ? minVisitor.col : minVisitor.row;
+  *index = IndexType((RowsAtCompileTime==1) ? minVisitor.col : minVisitor.row);
   return minVisitor.res;
 }
 
-/** \returns the maximum of all coefficients of *this and puts in *row and *col its location.
+/** \fn DenseBase<Derived>::maxCoeff(IndexType* rowId, IndexType* colId) const
+  * \returns the maximum of all coefficients of *this and puts in *row and *col its location.
   * \warning the result is undefined if \c *this contains NaN. 
   *
-  * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visitor(), DenseBase::maxCoeff()
+  * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::maxCoeff()
   */
 template<typename Derived>
 template<typename IndexType>
+EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff(IndexType* rowPtr, IndexType* colPtr) const
 {
@@ -222,6 +257,7 @@ DenseBase<Derived>::maxCoeff(IndexType* rowPtr, IndexType* colPtr) const
   */
 template<typename Derived>
 template<typename IndexType>
+EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff(IndexType* index) const
 {
diff --git a/extern/Eigen3/Eigen/src/Core/arch/AVX/Complex.h b/extern/Eigen3/Eigen/src/Core/arch/AVX/Complex.h
new file mode 100644
index 00000000000..7fa61969dc1
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/AVX/Complex.h
@@ -0,0 +1,451 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner (benoit.steiner.goog@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_COMPLEX_AVX_H
+#define EIGEN_COMPLEX_AVX_H
+
+namespace Eigen {
+
+namespace internal {
+
+//---------- float ----------
+struct Packet4cf
+{
+  EIGEN_STRONG_INLINE Packet4cf() {}
+  EIGEN_STRONG_INLINE explicit Packet4cf(const __m256& a) : v(a) {}
+  __m256  v;
+};
+
+template<> struct packet_traits<std::complex<float> >  : default_packet_traits
+{
+  typedef Packet4cf type;
+  typedef Packet2cf half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 4,
+    HasHalfPacket = 1,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet4cf> { typedef std::complex<float> type; enum {size=4, alignment=Aligned32}; typedef Packet2cf half; };
+
+template<> EIGEN_STRONG_INLINE Packet4cf padd<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_add_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf psub<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_sub_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pnegate(const Packet4cf& a)
+{
+  return Packet4cf(pnegate(a.v));
+}
+template<> EIGEN_STRONG_INLINE Packet4cf pconj(const Packet4cf& a)
+{
+  const __m256 mask = _mm256_castsi256_ps(_mm256_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000));
+  return Packet4cf(_mm256_xor_ps(a.v,mask));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pmul<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
+{
+  __m256 tmp1 = _mm256_mul_ps(_mm256_moveldup_ps(a.v), b.v);
+  __m256 tmp2 = _mm256_mul_ps(_mm256_movehdup_ps(a.v), _mm256_permute_ps(b.v, _MM_SHUFFLE(2,3,0,1)));
+  __m256 result = _mm256_addsub_ps(tmp1, tmp2);
+  return Packet4cf(result);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pand   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_and_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf por    <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_or_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pxor   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_xor_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(a.v,b.v)); }
+
+template<> EIGEN_STRONG_INLINE Packet4cf pload <Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet4cf(pload<Packet8f>(&numext::real_ref(*from))); }
+template<> EIGEN_STRONG_INLINE Packet4cf ploadu<Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet4cf(ploadu<Packet8f>(&numext::real_ref(*from))); }
+
+
+template<> EIGEN_STRONG_INLINE Packet4cf pset1<Packet4cf>(const std::complex<float>& from)
+{
+  return Packet4cf(_mm256_castpd_ps(_mm256_broadcast_sd((const double*)(const void*)&from)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf ploaddup<Packet4cf>(const std::complex<float>* from)
+{
+  // FIXME The following might be optimized using _mm256_movedup_pd
+  Packet2cf a = ploaddup<Packet2cf>(from);
+  Packet2cf b = ploaddup<Packet2cf>(from+1);
+  return  Packet4cf(_mm256_insertf128_ps(_mm256_castps128_ps256(a.v), b.v, 1));
+}
+
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float>* to, const Packet4cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float>* to, const Packet4cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), from.v); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet4cf pgather<std::complex<float>, Packet4cf>(const std::complex<float>* from, Index stride)
+{
+  return Packet4cf(_mm256_set_ps(std::imag(from[3*stride]), std::real(from[3*stride]),
+                                 std::imag(from[2*stride]), std::real(from[2*stride]),
+                                 std::imag(from[1*stride]), std::real(from[1*stride]),
+                                 std::imag(from[0*stride]), std::real(from[0*stride])));
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet4cf>(std::complex<float>* to, const Packet4cf& from, Index stride)
+{
+  __m128 low = _mm256_extractf128_ps(from.v, 0);
+  to[stride*0] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(low, low, 0)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(low, low, 1)));
+  to[stride*1] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(low, low, 2)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(low, low, 3)));
+
+  __m128 high = _mm256_extractf128_ps(from.v, 1);
+  to[stride*2] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(high, high, 0)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(high, high, 1)));
+  to[stride*3] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(high, high, 2)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(high, high, 3)));
+
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet4cf>(const Packet4cf& a)
+{
+  return pfirst(Packet2cf(_mm256_castps256_ps128(a.v)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf preverse(const Packet4cf& a) {
+  __m128 low  = _mm256_extractf128_ps(a.v, 0);
+  __m128 high = _mm256_extractf128_ps(a.v, 1);
+  __m128d lowd  = _mm_castps_pd(low);
+  __m128d highd = _mm_castps_pd(high);
+  low  = _mm_castpd_ps(_mm_shuffle_pd(lowd,lowd,0x1));
+  high = _mm_castpd_ps(_mm_shuffle_pd(highd,highd,0x1));
+  __m256 result = _mm256_setzero_ps();
+  result = _mm256_insertf128_ps(result, low, 1);
+  result = _mm256_insertf128_ps(result, high, 0);
+  return Packet4cf(result);
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet4cf>(const Packet4cf& a)
+{
+  return predux(padd(Packet2cf(_mm256_extractf128_ps(a.v,0)),
+                     Packet2cf(_mm256_extractf128_ps(a.v,1))));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf preduxp<Packet4cf>(const Packet4cf* vecs)
+{
+  Packet8f t0 = _mm256_shuffle_ps(vecs[0].v, vecs[0].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  Packet8f t1 = _mm256_shuffle_ps(vecs[1].v, vecs[1].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  t0 = _mm256_hadd_ps(t0,t1);
+  Packet8f t2 = _mm256_shuffle_ps(vecs[2].v, vecs[2].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  Packet8f t3 = _mm256_shuffle_ps(vecs[3].v, vecs[3].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  t2 = _mm256_hadd_ps(t2,t3);
+  
+  t1 = _mm256_permute2f128_ps(t0,t2, 0 + (2<<4));
+  t3 = _mm256_permute2f128_ps(t0,t2, 1 + (3<<4));
+
+  return Packet4cf(_mm256_add_ps(t1,t3));
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet4cf>(const Packet4cf& a)
+{
+  return predux_mul(pmul(Packet2cf(_mm256_extractf128_ps(a.v, 0)),
+                         Packet2cf(_mm256_extractf128_ps(a.v, 1))));
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet4cf>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4cf& first, const Packet4cf& second)
+  {
+    if (Offset==0) return;
+    palign_impl<Offset*2,Packet8f>::run(first.v, second.v);
+  }
+};
+
+template<> struct conj_helper<Packet4cf, Packet4cf, false,true>
+{
+  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet4cf, Packet4cf, true,false>
+{
+  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet4cf, Packet4cf, true,true>
+{
+  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet4cf,Packet8f)
+
+template<> EIGEN_STRONG_INLINE Packet4cf pdiv<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
+{
+  Packet4cf num = pmul(a, pconj(b));
+  __m256 tmp = _mm256_mul_ps(b.v, b.v);
+  __m256 tmp2    = _mm256_shuffle_ps(tmp,tmp,0xB1);
+  __m256 denom = _mm256_add_ps(tmp, tmp2);
+  return Packet4cf(_mm256_div_ps(num.v, denom));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pcplxflip<Packet4cf>(const Packet4cf& x)
+{
+  return Packet4cf(_mm256_shuffle_ps(x.v, x.v, _MM_SHUFFLE(2, 3, 0 ,1)));
+}
+
+//---------- double ----------
+struct Packet2cd
+{
+  EIGEN_STRONG_INLINE Packet2cd() {}
+  EIGEN_STRONG_INLINE explicit Packet2cd(const __m256d& a) : v(a) {}
+  __m256d  v;
+};
+
+template<> struct packet_traits<std::complex<double> >  : default_packet_traits
+{
+  typedef Packet2cd type;
+  typedef Packet1cd half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 0,
+    size = 2,
+    HasHalfPacket = 1,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet2cd> { typedef std::complex<double> type; enum {size=2, alignment=Aligned32}; typedef Packet1cd half; };
+
+template<> EIGEN_STRONG_INLINE Packet2cd padd<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_add_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd psub<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_sub_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pnegate(const Packet2cd& a) { return Packet2cd(pnegate(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pconj(const Packet2cd& a)
+{
+  const __m256d mask = _mm256_castsi256_pd(_mm256_set_epi32(0x80000000,0x0,0x0,0x0,0x80000000,0x0,0x0,0x0));
+  return Packet2cd(_mm256_xor_pd(a.v,mask));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pmul<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
+{
+  __m256d tmp1 = _mm256_shuffle_pd(a.v,a.v,0x0);
+  __m256d even = _mm256_mul_pd(tmp1, b.v);
+  __m256d tmp2 = _mm256_shuffle_pd(a.v,a.v,0xF);
+  __m256d tmp3 = _mm256_shuffle_pd(b.v,b.v,0x5);
+  __m256d odd  = _mm256_mul_pd(tmp2, tmp3);
+  return Packet2cd(_mm256_addsub_pd(even, odd));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pand   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_and_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd por    <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_or_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pxor   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_xor_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(a.v,b.v)); }
+
+template<> EIGEN_STRONG_INLINE Packet2cd pload <Packet2cd>(const std::complex<double>* from)
+{ EIGEN_DEBUG_ALIGNED_LOAD return Packet2cd(pload<Packet4d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE Packet2cd ploadu<Packet2cd>(const std::complex<double>* from)
+{ EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cd(ploadu<Packet4d>((const double*)from)); }
+
+template<> EIGEN_STRONG_INLINE Packet2cd pset1<Packet2cd>(const std::complex<double>& from)
+{
+  // in case casting to a __m128d* is really not safe, then we can still fallback to this version: (much slower though)
+//   return Packet2cd(_mm256_loadu2_m128d((const double*)&from,(const double*)&from));
+    return Packet2cd(_mm256_broadcast_pd((const __m128d*)(const void*)&from));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd ploaddup<Packet2cd>(const std::complex<double>* from) { return pset1<Packet2cd>(*from); }
+
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet2cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet2cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet2cd pgather<std::complex<double>, Packet2cd>(const std::complex<double>* from, Index stride)
+{
+  return Packet2cd(_mm256_set_pd(std::imag(from[1*stride]), std::real(from[1*stride]),
+				 std::imag(from[0*stride]), std::real(from[0*stride])));
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet2cd>(std::complex<double>* to, const Packet2cd& from, Index stride)
+{
+  __m128d low = _mm256_extractf128_pd(from.v, 0);
+  to[stride*0] = std::complex<double>(_mm_cvtsd_f64(low), _mm_cvtsd_f64(_mm_shuffle_pd(low, low, 1)));
+  __m128d high = _mm256_extractf128_pd(from.v, 1);
+  to[stride*1] = std::complex<double>(_mm_cvtsd_f64(high), _mm_cvtsd_f64(_mm_shuffle_pd(high, high, 1)));
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet2cd>(const Packet2cd& a)
+{
+  __m128d low = _mm256_extractf128_pd(a.v, 0);
+  EIGEN_ALIGN16 double res[2];
+  _mm_store_pd(res, low);
+  return std::complex<double>(res[0],res[1]);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd preverse(const Packet2cd& a) {
+  __m256d result = _mm256_permute2f128_pd(a.v, a.v, 1);
+  return Packet2cd(result);
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet2cd>(const Packet2cd& a)
+{
+  return predux(padd(Packet1cd(_mm256_extractf128_pd(a.v,0)),
+                     Packet1cd(_mm256_extractf128_pd(a.v,1))));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd preduxp<Packet2cd>(const Packet2cd* vecs)
+{
+  Packet4d t0 = _mm256_permute2f128_pd(vecs[0].v,vecs[1].v, 0 + (2<<4));
+  Packet4d t1 = _mm256_permute2f128_pd(vecs[0].v,vecs[1].v, 1 + (3<<4));
+
+  return Packet2cd(_mm256_add_pd(t0,t1));
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet2cd>(const Packet2cd& a)
+{
+  return predux(pmul(Packet1cd(_mm256_extractf128_pd(a.v,0)),
+                     Packet1cd(_mm256_extractf128_pd(a.v,1))));
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet2cd>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2cd& first, const Packet2cd& second)
+  {
+    if (Offset==0) return;
+    palign_impl<Offset*2,Packet4d>::run(first.v, second.v);
+  }
+};
+
+template<> struct conj_helper<Packet2cd, Packet2cd, false,true>
+{
+  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet2cd, Packet2cd, true,false>
+{
+  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet2cd, Packet2cd, true,true>
+{
+  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cd,Packet4d)
+
+template<> EIGEN_STRONG_INLINE Packet2cd pdiv<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
+{
+  Packet2cd num = pmul(a, pconj(b));
+  __m256d tmp = _mm256_mul_pd(b.v, b.v);
+  __m256d denom = _mm256_hadd_pd(tmp, tmp);
+  return Packet2cd(_mm256_div_pd(num.v, denom));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pcplxflip<Packet2cd>(const Packet2cd& x)
+{
+  return Packet2cd(_mm256_shuffle_pd(x.v, x.v, 0x5));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4cf,4>& kernel) {
+  __m256d P0 = _mm256_castps_pd(kernel.packet[0].v);
+  __m256d P1 = _mm256_castps_pd(kernel.packet[1].v);
+  __m256d P2 = _mm256_castps_pd(kernel.packet[2].v);
+  __m256d P3 = _mm256_castps_pd(kernel.packet[3].v);
+
+  __m256d T0 = _mm256_shuffle_pd(P0, P1, 15);
+  __m256d T1 = _mm256_shuffle_pd(P0, P1, 0);
+  __m256d T2 = _mm256_shuffle_pd(P2, P3, 15);
+  __m256d T3 = _mm256_shuffle_pd(P2, P3, 0);
+
+  kernel.packet[1].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T0, T2, 32));
+  kernel.packet[3].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T0, T2, 49));
+  kernel.packet[0].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T1, T3, 32));
+  kernel.packet[2].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T1, T3, 49));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2cd,2>& kernel) {
+  __m256d tmp = _mm256_permute2f128_pd(kernel.packet[0].v, kernel.packet[1].v, 0+(2<<4));
+  kernel.packet[1].v = _mm256_permute2f128_pd(kernel.packet[0].v, kernel.packet[1].v, 1+(3<<4));
+ kernel.packet[0].v = tmp;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pinsertfirst(const Packet4cf& a, std::complex<float> b)
+{
+  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,1|2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pinsertfirst(const Packet2cd& a, std::complex<double> b)
+{
+  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,1|2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pinsertlast(const Packet4cf& a, std::complex<float> b)
+{
+  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,(1<<7)|(1<<6)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pinsertlast(const Packet2cd& a, std::complex<double> b)
+{
+  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,(1<<3)|(1<<2)));
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_COMPLEX_AVX_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/AVX/MathFunctions.h b/extern/Eigen3/Eigen/src/Core/arch/AVX/MathFunctions.h
new file mode 100644
index 00000000000..6af67ce2d65
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/AVX/MathFunctions.h
@@ -0,0 +1,439 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Pedro Gonnet (pedro.gonnet@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_MATH_FUNCTIONS_AVX_H
+#define EIGEN_MATH_FUNCTIONS_AVX_H
+
+/* The sin, cos, exp, and log functions of this file are loosely derived from
+ * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
+ */
+
+namespace Eigen {
+
+namespace internal {
+
+inline Packet8i pshiftleft(Packet8i v, int n)
+{
+#ifdef EIGEN_VECTORIZE_AVX2
+  return _mm256_slli_epi32(v, n);
+#else
+  __m128i lo = _mm_slli_epi32(_mm256_extractf128_si256(v, 0), n);
+  __m128i hi = _mm_slli_epi32(_mm256_extractf128_si256(v, 1), n);
+  return _mm256_insertf128_si256(_mm256_castsi128_si256(lo), (hi), 1);
+#endif
+}
+
+inline Packet8f pshiftright(Packet8f v, int n)
+{
+#ifdef EIGEN_VECTORIZE_AVX2
+  return _mm256_cvtepi32_ps(_mm256_srli_epi32(_mm256_castps_si256(v), n));
+#else
+  __m128i lo = _mm_srli_epi32(_mm256_extractf128_si256(_mm256_castps_si256(v), 0), n);
+  __m128i hi = _mm_srli_epi32(_mm256_extractf128_si256(_mm256_castps_si256(v), 1), n);
+  return _mm256_cvtepi32_ps(_mm256_insertf128_si256(_mm256_castsi128_si256(lo), (hi), 1));
+#endif
+}
+
+// Sine function
+// Computes sin(x) by wrapping x to the interval [-Pi/4,3*Pi/4] and
+// evaluating interpolants in [-Pi/4,Pi/4] or [Pi/4,3*Pi/4]. The interpolants
+// are (anti-)symmetric and thus have only odd/even coefficients
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+psin<Packet8f>(const Packet8f& _x) {
+  Packet8f x = _x;
+
+  // Some useful values.
+  _EIGEN_DECLARE_CONST_Packet8i(one, 1);
+  _EIGEN_DECLARE_CONST_Packet8f(one, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(two, 2.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(one_over_four, 0.25f);
+  _EIGEN_DECLARE_CONST_Packet8f(one_over_pi, 3.183098861837907e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(neg_pi_first, -3.140625000000000e+00f);
+  _EIGEN_DECLARE_CONST_Packet8f(neg_pi_second, -9.670257568359375e-04f);
+  _EIGEN_DECLARE_CONST_Packet8f(neg_pi_third, -6.278329571784980e-07f);
+  _EIGEN_DECLARE_CONST_Packet8f(four_over_pi, 1.273239544735163e+00f);
+
+  // Map x from [-Pi/4,3*Pi/4] to z in [-1,3] and subtract the shifted period.
+  Packet8f z = pmul(x, p8f_one_over_pi);
+  Packet8f shift = _mm256_floor_ps(padd(z, p8f_one_over_four));
+  x = pmadd(shift, p8f_neg_pi_first, x);
+  x = pmadd(shift, p8f_neg_pi_second, x);
+  x = pmadd(shift, p8f_neg_pi_third, x);
+  z = pmul(x, p8f_four_over_pi);
+
+  // Make a mask for the entries that need flipping, i.e. wherever the shift
+  // is odd.
+  Packet8i shift_ints = _mm256_cvtps_epi32(shift);
+  Packet8i shift_isodd = _mm256_castps_si256(_mm256_and_ps(_mm256_castsi256_ps(shift_ints), _mm256_castsi256_ps(p8i_one)));
+  Packet8i sign_flip_mask = pshiftleft(shift_isodd, 31);
+
+  // Create a mask for which interpolant to use, i.e. if z > 1, then the mask
+  // is set to ones for that entry.
+  Packet8f ival_mask = _mm256_cmp_ps(z, p8f_one, _CMP_GT_OQ);
+
+  // Evaluate the polynomial for the interval [1,3] in z.
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_0, 9.999999724233232e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_2, -3.084242535619928e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_4, 1.584991525700324e-02f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_6, -3.188805084631342e-04f);
+  Packet8f z_minus_two = psub(z, p8f_two);
+  Packet8f z_minus_two2 = pmul(z_minus_two, z_minus_two);
+  Packet8f right = pmadd(p8f_coeff_right_6, z_minus_two2, p8f_coeff_right_4);
+  right = pmadd(right, z_minus_two2, p8f_coeff_right_2);
+  right = pmadd(right, z_minus_two2, p8f_coeff_right_0);
+
+  // Evaluate the polynomial for the interval [-1,1] in z.
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_1, 7.853981525427295e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_3, -8.074536727092352e-02f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_5, 2.489871967827018e-03f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_7, -3.587725841214251e-05f);
+  Packet8f z2 = pmul(z, z);
+  Packet8f left = pmadd(p8f_coeff_left_7, z2, p8f_coeff_left_5);
+  left = pmadd(left, z2, p8f_coeff_left_3);
+  left = pmadd(left, z2, p8f_coeff_left_1);
+  left = pmul(left, z);
+
+  // Assemble the results, i.e. select the left and right polynomials.
+  left = _mm256_andnot_ps(ival_mask, left);
+  right = _mm256_and_ps(ival_mask, right);
+  Packet8f res = _mm256_or_ps(left, right);
+
+  // Flip the sign on the odd intervals and return the result.
+  res = _mm256_xor_ps(res, _mm256_castsi256_ps(sign_flip_mask));
+  return res;
+}
+
+// Natural logarithm
+// Computes log(x) as log(2^e * m) = C*e + log(m), where the constant C =log(2)
+// and m is in the range [sqrt(1/2),sqrt(2)). In this range, the logarithm can
+// be easily approximated by a polynomial centered on m=1 for stability.
+// TODO(gonnet): Further reduce the interval allowing for lower-degree
+//               polynomial interpolants -> ... -> profit!
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+plog<Packet8f>(const Packet8f& _x) {
+  Packet8f x = _x;
+  _EIGEN_DECLARE_CONST_Packet8f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet8f(126f, 126.0f);
+
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(inv_mant_mask, ~0x7f800000);
+
+  // The smallest non denormalized float number.
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(min_norm_pos, 0x00800000);
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(minus_inf, 0xff800000);
+
+  // Polynomial coefficients.
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_SQRTHF, 0.707106781186547524f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p0, 7.0376836292E-2f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p1, -1.1514610310E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p2, 1.1676998740E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p3, -1.2420140846E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p4, +1.4249322787E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p5, -1.6668057665E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p6, +2.0000714765E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p7, -2.4999993993E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p8, +3.3333331174E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_q1, -2.12194440e-4f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_q2, 0.693359375f);
+
+  Packet8f invalid_mask = _mm256_cmp_ps(x, _mm256_setzero_ps(), _CMP_NGE_UQ); // not greater equal is true if x is NaN
+  Packet8f iszero_mask = _mm256_cmp_ps(x, _mm256_setzero_ps(), _CMP_EQ_OQ);
+
+  // Truncate input values to the minimum positive normal.
+  x = pmax(x, p8f_min_norm_pos);
+
+  Packet8f emm0 = pshiftright(x,23);
+  Packet8f e = _mm256_sub_ps(emm0, p8f_126f);
+
+  // Set the exponents to -1, i.e. x are in the range [0.5,1).
+  x = _mm256_and_ps(x, p8f_inv_mant_mask);
+  x = _mm256_or_ps(x, p8f_half);
+
+  // part2: Shift the inputs from the range [0.5,1) to [sqrt(1/2),sqrt(2))
+  // and shift by -1. The values are then centered around 0, which improves
+  // the stability of the polynomial evaluation.
+  //   if( x < SQRTHF ) {
+  //     e -= 1;
+  //     x = x + x - 1.0;
+  //   } else { x = x - 1.0; }
+  Packet8f mask = _mm256_cmp_ps(x, p8f_cephes_SQRTHF, _CMP_LT_OQ);
+  Packet8f tmp = _mm256_and_ps(x, mask);
+  x = psub(x, p8f_1);
+  e = psub(e, _mm256_and_ps(p8f_1, mask));
+  x = padd(x, tmp);
+
+  Packet8f x2 = pmul(x, x);
+  Packet8f x3 = pmul(x2, x);
+
+  // Evaluate the polynomial approximant of degree 8 in three parts, probably
+  // to improve instruction-level parallelism.
+  Packet8f y, y1, y2;
+  y = pmadd(p8f_cephes_log_p0, x, p8f_cephes_log_p1);
+  y1 = pmadd(p8f_cephes_log_p3, x, p8f_cephes_log_p4);
+  y2 = pmadd(p8f_cephes_log_p6, x, p8f_cephes_log_p7);
+  y = pmadd(y, x, p8f_cephes_log_p2);
+  y1 = pmadd(y1, x, p8f_cephes_log_p5);
+  y2 = pmadd(y2, x, p8f_cephes_log_p8);
+  y = pmadd(y, x3, y1);
+  y = pmadd(y, x3, y2);
+  y = pmul(y, x3);
+
+  // Add the logarithm of the exponent back to the result of the interpolation.
+  y1 = pmul(e, p8f_cephes_log_q1);
+  tmp = pmul(x2, p8f_half);
+  y = padd(y, y1);
+  x = psub(x, tmp);
+  y2 = pmul(e, p8f_cephes_log_q2);
+  x = padd(x, y);
+  x = padd(x, y2);
+
+  // Filter out invalid inputs, i.e. negative arg will be NAN, 0 will be -INF.
+  return _mm256_or_ps(
+      _mm256_andnot_ps(iszero_mask, _mm256_or_ps(x, invalid_mask)),
+      _mm256_and_ps(iszero_mask, p8f_minus_inf));
+}
+
+// Exponential function. Works by writing "x = m*log(2) + r" where
+// "m = floor(x/log(2)+1/2)" and "r" is the remainder. The result is then
+// "exp(x) = 2^m*exp(r)" where exp(r) is in the range [-1,1).
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+pexp<Packet8f>(const Packet8f& _x) {
+  _EIGEN_DECLARE_CONST_Packet8f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet8f(127, 127.0f);
+
+  _EIGEN_DECLARE_CONST_Packet8f(exp_hi, 88.3762626647950f);
+  _EIGEN_DECLARE_CONST_Packet8f(exp_lo, -88.3762626647949f);
+
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_LOG2EF, 1.44269504088896341f);
+
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p0, 1.9875691500E-4f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p1, 1.3981999507E-3f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p2, 8.3334519073E-3f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p3, 4.1665795894E-2f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p4, 1.6666665459E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p5, 5.0000001201E-1f);
+
+  // Clamp x.
+  Packet8f x = pmax(pmin(_x, p8f_exp_hi), p8f_exp_lo);
+
+  // Express exp(x) as exp(m*ln(2) + r), start by extracting
+  // m = floor(x/ln(2) + 0.5).
+  Packet8f m = _mm256_floor_ps(pmadd(x, p8f_cephes_LOG2EF, p8f_half));
+
+// Get r = x - m*ln(2). If no FMA instructions are available, m*ln(2) is
+// subtracted out in two parts, m*C1+m*C2 = m*ln(2), to avoid accumulating
+// truncation errors. Note that we don't use the "pmadd" function here to
+// ensure that a precision-preserving FMA instruction is used.
+#ifdef EIGEN_VECTORIZE_FMA
+  _EIGEN_DECLARE_CONST_Packet8f(nln2, -0.6931471805599453f);
+  Packet8f r = _mm256_fmadd_ps(m, p8f_nln2, x);
+#else
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_C1, 0.693359375f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_C2, -2.12194440e-4f);
+  Packet8f r = psub(x, pmul(m, p8f_cephes_exp_C1));
+  r = psub(r, pmul(m, p8f_cephes_exp_C2));
+#endif
+
+  Packet8f r2 = pmul(r, r);
+
+  // TODO(gonnet): Split into odd/even polynomials and try to exploit
+  //               instruction-level parallelism.
+  Packet8f y = p8f_cephes_exp_p0;
+  y = pmadd(y, r, p8f_cephes_exp_p1);
+  y = pmadd(y, r, p8f_cephes_exp_p2);
+  y = pmadd(y, r, p8f_cephes_exp_p3);
+  y = pmadd(y, r, p8f_cephes_exp_p4);
+  y = pmadd(y, r, p8f_cephes_exp_p5);
+  y = pmadd(y, r2, r);
+  y = padd(y, p8f_1);
+
+  // Build emm0 = 2^m.
+  Packet8i emm0 = _mm256_cvttps_epi32(padd(m, p8f_127));
+  emm0 = pshiftleft(emm0, 23);
+
+  // Return 2^m * exp(r).
+  return pmax(pmul(y, _mm256_castsi256_ps(emm0)), _x);
+}
+
+// Hyperbolic Tangent function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+ptanh<Packet8f>(const Packet8f& x) {
+  return internal::generic_fast_tanh_float(x);
+}
+
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d
+pexp<Packet4d>(const Packet4d& _x) {
+  Packet4d x = _x;
+
+  _EIGEN_DECLARE_CONST_Packet4d(1, 1.0);
+  _EIGEN_DECLARE_CONST_Packet4d(2, 2.0);
+  _EIGEN_DECLARE_CONST_Packet4d(half, 0.5);
+
+  _EIGEN_DECLARE_CONST_Packet4d(exp_hi, 709.437);
+  _EIGEN_DECLARE_CONST_Packet4d(exp_lo, -709.436139303);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_LOG2EF, 1.4426950408889634073599);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_p0, 1.26177193074810590878e-4);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_p1, 3.02994407707441961300e-2);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q0, 3.00198505138664455042e-6);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q1, 2.52448340349684104192e-3);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q2, 2.27265548208155028766e-1);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q3, 2.00000000000000000009e0);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_C1, 0.693145751953125);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_C2, 1.42860682030941723212e-6);
+  _EIGEN_DECLARE_CONST_Packet4i(1023, 1023);
+
+  Packet4d tmp, fx;
+
+  // clamp x
+  x = pmax(pmin(x, p4d_exp_hi), p4d_exp_lo);
+  // Express exp(x) as exp(g + n*log(2)).
+  fx = pmadd(p4d_cephes_LOG2EF, x, p4d_half);
+
+  // Get the integer modulus of log(2), i.e. the "n" described above.
+  fx = _mm256_floor_pd(fx);
+
+  // Get the remainder modulo log(2), i.e. the "g" described above. Subtract
+  // n*log(2) out in two steps, i.e. n*C1 + n*C2, C1+C2=log2 to get the last
+  // digits right.
+  tmp = pmul(fx, p4d_cephes_exp_C1);
+  Packet4d z = pmul(fx, p4d_cephes_exp_C2);
+  x = psub(x, tmp);
+  x = psub(x, z);
+
+  Packet4d x2 = pmul(x, x);
+
+  // Evaluate the numerator polynomial of the rational interpolant.
+  Packet4d px = p4d_cephes_exp_p0;
+  px = pmadd(px, x2, p4d_cephes_exp_p1);
+  px = pmadd(px, x2, p4d_cephes_exp_p2);
+  px = pmul(px, x);
+
+  // Evaluate the denominator polynomial of the rational interpolant.
+  Packet4d qx = p4d_cephes_exp_q0;
+  qx = pmadd(qx, x2, p4d_cephes_exp_q1);
+  qx = pmadd(qx, x2, p4d_cephes_exp_q2);
+  qx = pmadd(qx, x2, p4d_cephes_exp_q3);
+
+  // I don't really get this bit, copied from the SSE2 routines, so...
+  // TODO(gonnet): Figure out what is going on here, perhaps find a better
+  // rational interpolant?
+  x = _mm256_div_pd(px, psub(qx, px));
+  x = pmadd(p4d_2, x, p4d_1);
+
+  // Build e=2^n by constructing the exponents in a 128-bit vector and
+  // shifting them to where they belong in double-precision values.
+  __m128i emm0 = _mm256_cvtpd_epi32(fx);
+  emm0 = _mm_add_epi32(emm0, p4i_1023);
+  emm0 = _mm_shuffle_epi32(emm0, _MM_SHUFFLE(3, 1, 2, 0));
+  __m128i lo = _mm_slli_epi64(emm0, 52);
+  __m128i hi = _mm_slli_epi64(_mm_srli_epi64(emm0, 32), 52);
+  __m256i e = _mm256_insertf128_si256(_mm256_setzero_si256(), lo, 0);
+  e = _mm256_insertf128_si256(e, hi, 1);
+
+  // Construct the result 2^n * exp(g) = e * x. The max is used to catch
+  // non-finite values in the input.
+  return pmax(pmul(x, _mm256_castsi256_pd(e)), _x);
+}
+
+// Functions for sqrt.
+// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step
+// of Newton's method, at a cost of 1-2 bits of precision as opposed to the
+// exact solution. It does not handle +inf, or denormalized numbers correctly.
+// The main advantage of this approach is not just speed, but also the fact that
+// it can be inlined and pipelined with other computations, further reducing its
+// effective latency. This is similar to Quake3's fast inverse square root.
+// For detail see here: http://www.beyond3d.com/content/articles/8/
+#if EIGEN_FAST_MATH
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+psqrt<Packet8f>(const Packet8f& _x) {
+  Packet8f half = pmul(_x, pset1<Packet8f>(.5f));
+  Packet8f denormal_mask = _mm256_and_ps(
+      _mm256_cmp_ps(_x, pset1<Packet8f>((std::numeric_limits<float>::min)()),
+                    _CMP_LT_OQ),
+      _mm256_cmp_ps(_x, _mm256_setzero_ps(), _CMP_GE_OQ));
+
+  // Compute approximate reciprocal sqrt.
+  Packet8f x = _mm256_rsqrt_ps(_x);
+  // Do a single step of Newton's iteration.
+  x = pmul(x, psub(pset1<Packet8f>(1.5f), pmul(half, pmul(x,x))));
+  // Flush results for denormals to zero.
+  return _mm256_andnot_ps(denormal_mask, pmul(_x,x));
+}
+#else
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet8f psqrt<Packet8f>(const Packet8f& x) {
+  return _mm256_sqrt_ps(x);
+}
+#endif
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4d psqrt<Packet4d>(const Packet4d& x) {
+  return _mm256_sqrt_pd(x);
+}
+#if EIGEN_FAST_MATH
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet8f prsqrt<Packet8f>(const Packet8f& _x) {
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(inf, 0x7f800000);
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(nan, 0x7fc00000);
+  _EIGEN_DECLARE_CONST_Packet8f(one_point_five, 1.5f);
+  _EIGEN_DECLARE_CONST_Packet8f(minus_half, -0.5f);
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(flt_min, 0x00800000);
+
+  Packet8f neg_half = pmul(_x, p8f_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  Packet8f le_zero_mask = _mm256_cmp_ps(_x, p8f_flt_min, _CMP_LT_OQ);
+  Packet8f x = _mm256_andnot_ps(le_zero_mask, _mm256_rsqrt_ps(_x));
+
+  // Fill in NaNs and Infs for the negative/zero entries.
+  Packet8f neg_mask = _mm256_cmp_ps(_x, _mm256_setzero_ps(), _CMP_LT_OQ);
+  Packet8f zero_mask = _mm256_andnot_ps(neg_mask, le_zero_mask);
+  Packet8f infs_and_nans = _mm256_or_ps(_mm256_and_ps(neg_mask, p8f_nan),
+                                        _mm256_and_ps(zero_mask, p8f_inf));
+
+  // Do a single step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8f_one_point_five));
+
+  // Insert NaNs and Infs in all the right places.
+  return _mm256_or_ps(x, infs_and_nans);
+}
+
+#else
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet8f prsqrt<Packet8f>(const Packet8f& x) {
+  _EIGEN_DECLARE_CONST_Packet8f(one, 1.0f);
+  return _mm256_div_ps(p8f_one, _mm256_sqrt_ps(x));
+}
+#endif
+
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4d prsqrt<Packet4d>(const Packet4d& x) {
+  _EIGEN_DECLARE_CONST_Packet4d(one, 1.0);
+  return _mm256_div_pd(p4d_one, _mm256_sqrt_pd(x));
+}
+
+
+}  // end namespace internal
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_MATH_FUNCTIONS_AVX_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/AVX/PacketMath.h b/extern/Eigen3/Eigen/src/Core/arch/AVX/PacketMath.h
new file mode 100644
index 00000000000..923a124b206
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/AVX/PacketMath.h
@@ -0,0 +1,637 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner (benoit.steiner.goog@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_AVX_H
+#define EIGEN_PACKET_MATH_AVX_H
+
+namespace Eigen {
+
+namespace internal {
+
+#ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+#define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 8
+#endif
+
+#ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS (2*sizeof(void*))
+#endif
+
+#ifdef __FMA__
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#endif
+#endif
+
+typedef __m256  Packet8f;
+typedef __m256i Packet8i;
+typedef __m256d Packet4d;
+
+template<> struct is_arithmetic<__m256>  { enum { value = true }; };
+template<> struct is_arithmetic<__m256i> { enum { value = true }; };
+template<> struct is_arithmetic<__m256d> { enum { value = true }; };
+
+#define _EIGEN_DECLARE_CONST_Packet8f(NAME,X) \
+  const Packet8f p8f_##NAME = pset1<Packet8f>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet4d(NAME,X) \
+  const Packet4d p4d_##NAME = pset1<Packet4d>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(NAME,X) \
+  const Packet8f p8f_##NAME = _mm256_castsi256_ps(pset1<Packet8i>(X))
+
+#define _EIGEN_DECLARE_CONST_Packet8i(NAME,X) \
+  const Packet8i p8i_##NAME = pset1<Packet8i>(X)
+
+// Use the packet_traits defined in AVX512/PacketMath.h instead if we're going
+// to leverage AVX512 instructions.
+#ifndef EIGEN_VECTORIZE_AVX512
+template<> struct packet_traits<float>  : default_packet_traits
+{
+  typedef Packet8f type;
+  typedef Packet4f half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=8,
+    HasHalfPacket = 1,
+
+    HasDiv  = 1,
+    HasSin  = EIGEN_FAST_MATH,
+    HasCos  = 0,
+    HasLog  = 1,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasTanh  = EIGEN_FAST_MATH,
+    HasBlend = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+  };
+};
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef Packet4d type;
+  typedef Packet2d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=4,
+    HasHalfPacket = 1,
+
+    HasDiv  = 1,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasBlend = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+  };
+};
+#endif
+
+template<> struct scalar_div_cost<float,true> { enum { value = 14 }; };
+template<> struct scalar_div_cost<double,true> { enum { value = 16 }; };
+
+/* Proper support for integers is only provided by AVX2. In the meantime, we'll
+   use SSE instructions and packets to deal with integers.
+template<> struct packet_traits<int>    : default_packet_traits
+{
+  typedef Packet8i type;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=8
+  };
+};
+*/
+
+template<> struct unpacket_traits<Packet8f> { typedef float  type; typedef Packet4f half; enum {size=8, alignment=Aligned32}; };
+template<> struct unpacket_traits<Packet4d> { typedef double type; typedef Packet2d half; enum {size=4, alignment=Aligned32}; };
+template<> struct unpacket_traits<Packet8i> { typedef int    type; typedef Packet4i half; enum {size=8, alignment=Aligned32}; };
+
+template<> EIGEN_STRONG_INLINE Packet8f pset1<Packet8f>(const float&  from) { return _mm256_set1_ps(from); }
+template<> EIGEN_STRONG_INLINE Packet4d pset1<Packet4d>(const double& from) { return _mm256_set1_pd(from); }
+template<> EIGEN_STRONG_INLINE Packet8i pset1<Packet8i>(const int&    from) { return _mm256_set1_epi32(from); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pload1<Packet8f>(const float*  from) { return _mm256_broadcast_ss(from); }
+template<> EIGEN_STRONG_INLINE Packet4d pload1<Packet4d>(const double* from) { return _mm256_broadcast_sd(from); }
+
+template<> EIGEN_STRONG_INLINE Packet8f plset<Packet8f>(const float& a) { return _mm256_add_ps(_mm256_set1_ps(a), _mm256_set_ps(7.0,6.0,5.0,4.0,3.0,2.0,1.0,0.0)); }
+template<> EIGEN_STRONG_INLINE Packet4d plset<Packet4d>(const double& a) { return _mm256_add_pd(_mm256_set1_pd(a), _mm256_set_pd(3.0,2.0,1.0,0.0)); }
+
+template<> EIGEN_STRONG_INLINE Packet8f padd<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_add_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d padd<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_add_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f psub<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_sub_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d psub<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_sub_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pnegate(const Packet8f& a)
+{
+  return _mm256_sub_ps(_mm256_set1_ps(0.0),a);
+}
+template<> EIGEN_STRONG_INLINE Packet4d pnegate(const Packet4d& a)
+{
+  return _mm256_sub_pd(_mm256_set1_pd(0.0),a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pconj(const Packet8f& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet4d pconj(const Packet4d& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet8i pconj(const Packet8i& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet8f pmul<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_mul_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pmul<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_mul_pd(a,b); }
+
+
+template<> EIGEN_STRONG_INLINE Packet8f pdiv<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_div_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pdiv<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_div_pd(a,b); }
+template<> EIGEN_STRONG_INLINE Packet8i pdiv<Packet8i>(const Packet8i& /*a*/, const Packet8i& /*b*/)
+{ eigen_assert(false && "packet integer division are not supported by AVX");
+  return pset1<Packet8i>(0);
+}
+
+#ifdef __FMA__
+template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f& b, const Packet8f& c) {
+#if ( (EIGEN_COMP_GNUC_STRICT && EIGEN_COMP_GNUC<80) || (EIGEN_COMP_CLANG) )
+  // Clang stupidly generates a vfmadd213ps instruction plus some vmovaps on registers,
+  //  and even register spilling with clang>=6.0 (bug 1637).
+  // Gcc stupidly generates a vfmadd132ps instruction.
+  // So let's enforce it to generate a vfmadd231ps instruction since the most common use
+  //  case is to accumulate the result of the product.
+  Packet8f res = c;
+  __asm__("vfmadd231ps %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
+  return res;
+#else
+  return _mm256_fmadd_ps(a,b,c);
+#endif
+}
+template<> EIGEN_STRONG_INLINE Packet4d pmadd(const Packet4d& a, const Packet4d& b, const Packet4d& c) {
+#if ( (EIGEN_COMP_GNUC_STRICT && EIGEN_COMP_GNUC<80) || (EIGEN_COMP_CLANG) )
+  // see above
+  Packet4d res = c;
+  __asm__("vfmadd231pd %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
+  return res;
+#else
+  return _mm256_fmadd_pd(a,b,c);
+#endif
+}
+#endif
+
+template<> EIGEN_STRONG_INLINE Packet8f pmin<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_min_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pmin<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_min_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pmax<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_max_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pmax<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_max_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pround<Packet8f>(const Packet8f& a) { return _mm256_round_ps(a, _MM_FROUND_CUR_DIRECTION); }
+template<> EIGEN_STRONG_INLINE Packet4d pround<Packet4d>(const Packet4d& a) { return _mm256_round_pd(a, _MM_FROUND_CUR_DIRECTION); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pceil<Packet8f>(const Packet8f& a) { return _mm256_ceil_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet4d pceil<Packet4d>(const Packet4d& a) { return _mm256_ceil_pd(a); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pfloor<Packet8f>(const Packet8f& a) { return _mm256_floor_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet4d pfloor<Packet4d>(const Packet4d& a) { return _mm256_floor_pd(a); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pand<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_and_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pand<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_and_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f por<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_or_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d por<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_or_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pxor<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_xor_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pxor<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_xor_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pandnot<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_andnot_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pandnot<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_andnot_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pload<Packet8f>(const float*   from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm256_load_ps(from); }
+template<> EIGEN_STRONG_INLINE Packet4d pload<Packet4d>(const double*  from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm256_load_pd(from); }
+template<> EIGEN_STRONG_INLINE Packet8i pload<Packet8i>(const int*     from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm256_load_si256(reinterpret_cast<const __m256i*>(from)); }
+
+template<> EIGEN_STRONG_INLINE Packet8f ploadu<Packet8f>(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm256_loadu_ps(from); }
+template<> EIGEN_STRONG_INLINE Packet4d ploadu<Packet4d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm256_loadu_pd(from); }
+template<> EIGEN_STRONG_INLINE Packet8i ploadu<Packet8i>(const int* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm256_loadu_si256(reinterpret_cast<const __m256i*>(from)); }
+
+// Loads 4 floats from memory a returns the packet {a0, a0  a1, a1, a2, a2, a3, a3}
+template<> EIGEN_STRONG_INLINE Packet8f ploaddup<Packet8f>(const float* from)
+{
+  // TODO try to find a way to avoid the need of a temporary register
+//   Packet8f tmp  = _mm256_castps128_ps256(_mm_loadu_ps(from));
+//   tmp = _mm256_insertf128_ps(tmp, _mm_movehl_ps(_mm256_castps256_ps128(tmp),_mm256_castps256_ps128(tmp)), 1);
+//   return _mm256_unpacklo_ps(tmp,tmp);
+  
+  // _mm256_insertf128_ps is very slow on Haswell, thus:
+  Packet8f tmp = _mm256_broadcast_ps((const __m128*)(const void*)from);
+  // mimic an "inplace" permutation of the lower 128bits using a blend
+  tmp = _mm256_blend_ps(tmp,_mm256_castps128_ps256(_mm_permute_ps( _mm256_castps256_ps128(tmp), _MM_SHUFFLE(1,0,1,0))), 15);
+  // then we can perform a consistent permutation on the global register to get everything in shape:
+  return  _mm256_permute_ps(tmp, _MM_SHUFFLE(3,3,2,2));
+}
+// Loads 2 doubles from memory a returns the packet {a0, a0  a1, a1}
+template<> EIGEN_STRONG_INLINE Packet4d ploaddup<Packet4d>(const double* from)
+{
+  Packet4d tmp = _mm256_broadcast_pd((const __m128d*)(const void*)from);
+  return  _mm256_permute_pd(tmp, 3<<2);
+}
+
+// Loads 2 floats from memory a returns the packet {a0, a0  a0, a0, a1, a1, a1, a1}
+template<> EIGEN_STRONG_INLINE Packet8f ploadquad<Packet8f>(const float* from)
+{
+  Packet8f tmp = _mm256_castps128_ps256(_mm_broadcast_ss(from));
+  return _mm256_insertf128_ps(tmp, _mm_broadcast_ss(from+1), 1);
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet8f& from) { EIGEN_DEBUG_ALIGNED_STORE _mm256_store_ps(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<double>(double* to, const Packet4d& from) { EIGEN_DEBUG_ALIGNED_STORE _mm256_store_pd(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet8i& from) { EIGEN_DEBUG_ALIGNED_STORE _mm256_storeu_si256(reinterpret_cast<__m256i*>(to), from); }
+
+template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*   to, const Packet8f& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm256_storeu_ps(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet4d& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm256_storeu_pd(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*       to, const Packet8i& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm256_storeu_si256(reinterpret_cast<__m256i*>(to), from); }
+
+// NOTE: leverage _mm256_i32gather_ps and _mm256_i32gather_pd if AVX2 instructions are available
+// NOTE: for the record the following seems to be slower: return _mm256_i32gather_ps(from, _mm256_set1_epi32(stride), 4);
+template<> EIGEN_DEVICE_FUNC inline Packet8f pgather<float, Packet8f>(const float* from, Index stride)
+{
+  return _mm256_set_ps(from[7*stride], from[6*stride], from[5*stride], from[4*stride],
+                       from[3*stride], from[2*stride], from[1*stride], from[0*stride]);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet4d pgather<double, Packet4d>(const double* from, Index stride)
+{
+  return _mm256_set_pd(from[3*stride], from[2*stride], from[1*stride], from[0*stride]);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet8f>(float* to, const Packet8f& from, Index stride)
+{
+  __m128 low = _mm256_extractf128_ps(from, 0);
+  to[stride*0] = _mm_cvtss_f32(low);
+  to[stride*1] = _mm_cvtss_f32(_mm_shuffle_ps(low, low, 1));
+  to[stride*2] = _mm_cvtss_f32(_mm_shuffle_ps(low, low, 2));
+  to[stride*3] = _mm_cvtss_f32(_mm_shuffle_ps(low, low, 3));
+
+  __m128 high = _mm256_extractf128_ps(from, 1);
+  to[stride*4] = _mm_cvtss_f32(high);
+  to[stride*5] = _mm_cvtss_f32(_mm_shuffle_ps(high, high, 1));
+  to[stride*6] = _mm_cvtss_f32(_mm_shuffle_ps(high, high, 2));
+  to[stride*7] = _mm_cvtss_f32(_mm_shuffle_ps(high, high, 3));
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet4d>(double* to, const Packet4d& from, Index stride)
+{
+  __m128d low = _mm256_extractf128_pd(from, 0);
+  to[stride*0] = _mm_cvtsd_f64(low);
+  to[stride*1] = _mm_cvtsd_f64(_mm_shuffle_pd(low, low, 1));
+  __m128d high = _mm256_extractf128_pd(from, 1);
+  to[stride*2] = _mm_cvtsd_f64(high);
+  to[stride*3] = _mm_cvtsd_f64(_mm_shuffle_pd(high, high, 1));
+}
+
+template<> EIGEN_STRONG_INLINE void pstore1<Packet8f>(float* to, const float& a)
+{
+  Packet8f pa = pset1<Packet8f>(a);
+  pstore(to, pa);
+}
+template<> EIGEN_STRONG_INLINE void pstore1<Packet4d>(double* to, const double& a)
+{
+  Packet4d pa = pset1<Packet4d>(a);
+  pstore(to, pa);
+}
+template<> EIGEN_STRONG_INLINE void pstore1<Packet8i>(int* to, const int& a)
+{
+  Packet8i pa = pset1<Packet8i>(a);
+  pstore(to, pa);
+}
+
+#ifndef EIGEN_VECTORIZE_AVX512
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+#endif
+
+template<> EIGEN_STRONG_INLINE float  pfirst<Packet8f>(const Packet8f& a) {
+  return _mm_cvtss_f32(_mm256_castps256_ps128(a));
+}
+template<> EIGEN_STRONG_INLINE double pfirst<Packet4d>(const Packet4d& a) {
+  return _mm_cvtsd_f64(_mm256_castpd256_pd128(a));
+}
+template<> EIGEN_STRONG_INLINE int    pfirst<Packet8i>(const Packet8i& a) {
+  return _mm_cvtsi128_si32(_mm256_castsi256_si128(a));
+}
+
+
+template<> EIGEN_STRONG_INLINE Packet8f preverse(const Packet8f& a)
+{
+  __m256 tmp = _mm256_shuffle_ps(a,a,0x1b);
+  return _mm256_permute2f128_ps(tmp, tmp, 1);
+}
+template<> EIGEN_STRONG_INLINE Packet4d preverse(const Packet4d& a)
+{
+   __m256d tmp = _mm256_shuffle_pd(a,a,5);
+  return _mm256_permute2f128_pd(tmp, tmp, 1);
+  #if 0
+  // This version is unlikely to be faster as _mm256_shuffle_ps and _mm256_permute_pd
+  // exhibit the same latency/throughput, but it is here for future reference/benchmarking...
+  __m256d swap_halves = _mm256_permute2f128_pd(a,a,1);
+    return _mm256_permute_pd(swap_halves,5);
+  #endif
+}
+
+// pabs should be ok
+template<> EIGEN_STRONG_INLINE Packet8f pabs(const Packet8f& a)
+{
+  const Packet8f mask = _mm256_castsi256_ps(_mm256_setr_epi32(0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF));
+  return _mm256_and_ps(a,mask);
+}
+template<> EIGEN_STRONG_INLINE Packet4d pabs(const Packet4d& a)
+{
+  const Packet4d mask = _mm256_castsi256_pd(_mm256_setr_epi32(0xFFFFFFFF,0x7FFFFFFF,0xFFFFFFFF,0x7FFFFFFF,0xFFFFFFFF,0x7FFFFFFF,0xFFFFFFFF,0x7FFFFFFF));
+  return _mm256_and_pd(a,mask);
+}
+
+// preduxp should be ok
+// FIXME: why is this ok? why isn't the simply implementation working as expected?
+template<> EIGEN_STRONG_INLINE Packet8f preduxp<Packet8f>(const Packet8f* vecs)
+{
+    __m256 hsum1 = _mm256_hadd_ps(vecs[0], vecs[1]);
+    __m256 hsum2 = _mm256_hadd_ps(vecs[2], vecs[3]);
+    __m256 hsum3 = _mm256_hadd_ps(vecs[4], vecs[5]);
+    __m256 hsum4 = _mm256_hadd_ps(vecs[6], vecs[7]);
+
+    __m256 hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+    __m256 hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+    __m256 hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+    __m256 hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+    __m256 perm1 =  _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+    __m256 perm2 =  _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+    __m256 perm3 =  _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+    __m256 perm4 =  _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+    __m256 sum1 = _mm256_add_ps(perm1, hsum5);
+    __m256 sum2 = _mm256_add_ps(perm2, hsum6);
+    __m256 sum3 = _mm256_add_ps(perm3, hsum7);
+    __m256 sum4 = _mm256_add_ps(perm4, hsum8);
+
+    __m256 blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+    __m256 blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+    __m256 final = _mm256_blend_ps(blend1, blend2, 0xf0);
+    return final;
+}
+template<> EIGEN_STRONG_INLINE Packet4d preduxp<Packet4d>(const Packet4d* vecs)
+{
+ Packet4d tmp0, tmp1;
+
+  tmp0 = _mm256_hadd_pd(vecs[0], vecs[1]);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs[2], vecs[3]);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  return _mm256_blend_pd(tmp0, tmp1, 0xC);
+}
+
+template<> EIGEN_STRONG_INLINE float predux<Packet8f>(const Packet8f& a)
+{
+  return predux(Packet4f(_mm_add_ps(_mm256_castps256_ps128(a),_mm256_extractf128_ps(a,1))));
+}
+template<> EIGEN_STRONG_INLINE double predux<Packet4d>(const Packet4d& a)
+{
+  return predux(Packet2d(_mm_add_pd(_mm256_castpd256_pd128(a),_mm256_extractf128_pd(a,1))));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f predux_downto4<Packet8f>(const Packet8f& a)
+{
+  return _mm_add_ps(_mm256_castps256_ps128(a),_mm256_extractf128_ps(a,1));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_mul<Packet8f>(const Packet8f& a)
+{
+  Packet8f tmp;
+  tmp = _mm256_mul_ps(a, _mm256_permute2f128_ps(a,a,1));
+  tmp = _mm256_mul_ps(tmp, _mm256_shuffle_ps(tmp,tmp,_MM_SHUFFLE(1,0,3,2)));
+  return pfirst(_mm256_mul_ps(tmp, _mm256_shuffle_ps(tmp,tmp,1)));
+}
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet4d>(const Packet4d& a)
+{
+  Packet4d tmp;
+  tmp = _mm256_mul_pd(a, _mm256_permute2f128_pd(a,a,1));
+  return pfirst(_mm256_mul_pd(tmp, _mm256_shuffle_pd(tmp,tmp,1)));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_min<Packet8f>(const Packet8f& a)
+{
+  Packet8f tmp = _mm256_min_ps(a, _mm256_permute2f128_ps(a,a,1));
+  tmp = _mm256_min_ps(tmp, _mm256_shuffle_ps(tmp,tmp,_MM_SHUFFLE(1,0,3,2)));
+  return pfirst(_mm256_min_ps(tmp, _mm256_shuffle_ps(tmp,tmp,1)));
+}
+template<> EIGEN_STRONG_INLINE double predux_min<Packet4d>(const Packet4d& a)
+{
+  Packet4d tmp = _mm256_min_pd(a, _mm256_permute2f128_pd(a,a,1));
+  return pfirst(_mm256_min_pd(tmp, _mm256_shuffle_pd(tmp, tmp, 1)));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_max<Packet8f>(const Packet8f& a)
+{
+  Packet8f tmp = _mm256_max_ps(a, _mm256_permute2f128_ps(a,a,1));
+  tmp = _mm256_max_ps(tmp, _mm256_shuffle_ps(tmp,tmp,_MM_SHUFFLE(1,0,3,2)));
+  return pfirst(_mm256_max_ps(tmp, _mm256_shuffle_ps(tmp,tmp,1)));
+}
+
+template<> EIGEN_STRONG_INLINE double predux_max<Packet4d>(const Packet4d& a)
+{
+  Packet4d tmp = _mm256_max_pd(a, _mm256_permute2f128_pd(a,a,1));
+  return pfirst(_mm256_max_pd(tmp, _mm256_shuffle_pd(tmp, tmp, 1)));
+}
+
+
+template<int Offset>
+struct palign_impl<Offset,Packet8f>
+{
+  static EIGEN_STRONG_INLINE void run(Packet8f& first, const Packet8f& second)
+  {
+    if (Offset==1)
+    {
+      first = _mm256_blend_ps(first, second, 1);
+      Packet8f tmp1 = _mm256_permute_ps (first, _MM_SHUFFLE(0,3,2,1));
+      Packet8f tmp2 = _mm256_permute2f128_ps (tmp1, tmp1, 1);
+      first = _mm256_blend_ps(tmp1, tmp2, 0x88);
+    }
+    else if (Offset==2)
+    {
+      first = _mm256_blend_ps(first, second, 3);
+      Packet8f tmp1 = _mm256_permute_ps (first, _MM_SHUFFLE(1,0,3,2));
+      Packet8f tmp2 = _mm256_permute2f128_ps (tmp1, tmp1, 1);
+      first = _mm256_blend_ps(tmp1, tmp2, 0xcc);
+    }
+    else if (Offset==3)
+    {
+      first = _mm256_blend_ps(first, second, 7);
+      Packet8f tmp1 = _mm256_permute_ps (first, _MM_SHUFFLE(2,1,0,3));
+      Packet8f tmp2 = _mm256_permute2f128_ps (tmp1, tmp1, 1);
+      first = _mm256_blend_ps(tmp1, tmp2, 0xee);
+    }
+    else if (Offset==4)
+    {
+      first = _mm256_blend_ps(first, second, 15);
+      Packet8f tmp1 = _mm256_permute_ps (first, _MM_SHUFFLE(3,2,1,0));
+      Packet8f tmp2 = _mm256_permute2f128_ps (tmp1, tmp1, 1);
+      first = _mm256_permute_ps(tmp2, _MM_SHUFFLE(3,2,1,0));
+    }
+    else if (Offset==5)
+    {
+      first = _mm256_blend_ps(first, second, 31);
+      first = _mm256_permute2f128_ps(first, first, 1);
+      Packet8f tmp = _mm256_permute_ps (first, _MM_SHUFFLE(0,3,2,1));
+      first = _mm256_permute2f128_ps(tmp, tmp, 1);
+      first = _mm256_blend_ps(tmp, first, 0x88);
+    }
+    else if (Offset==6)
+    {
+      first = _mm256_blend_ps(first, second, 63);
+      first = _mm256_permute2f128_ps(first, first, 1);
+      Packet8f tmp = _mm256_permute_ps (first, _MM_SHUFFLE(1,0,3,2));
+      first = _mm256_permute2f128_ps(tmp, tmp, 1);
+      first = _mm256_blend_ps(tmp, first, 0xcc);
+    }
+    else if (Offset==7)
+    {
+      first = _mm256_blend_ps(first, second, 127);
+      first = _mm256_permute2f128_ps(first, first, 1);
+      Packet8f tmp = _mm256_permute_ps (first, _MM_SHUFFLE(2,1,0,3));
+      first = _mm256_permute2f128_ps(tmp, tmp, 1);
+      first = _mm256_blend_ps(tmp, first, 0xee);
+    }
+  }
+};
+
+template<int Offset>
+struct palign_impl<Offset,Packet4d>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4d& first, const Packet4d& second)
+  {
+    if (Offset==1)
+    {
+      first = _mm256_blend_pd(first, second, 1);
+      __m256d tmp = _mm256_permute_pd(first, 5);
+      first = _mm256_permute2f128_pd(tmp, tmp, 1);
+      first = _mm256_blend_pd(tmp, first, 0xA);
+    }
+    else if (Offset==2)
+    {
+      first = _mm256_blend_pd(first, second, 3);
+      first = _mm256_permute2f128_pd(first, first, 1);
+    }
+    else if (Offset==3)
+    {
+      first = _mm256_blend_pd(first, second, 7);
+      __m256d tmp = _mm256_permute_pd(first, 5);
+      first = _mm256_permute2f128_pd(tmp, tmp, 1);
+      first = _mm256_blend_pd(tmp, first, 5);
+    }
+  }
+};
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet8f,8>& kernel) {
+  __m256 T0 = _mm256_unpacklo_ps(kernel.packet[0], kernel.packet[1]);
+  __m256 T1 = _mm256_unpackhi_ps(kernel.packet[0], kernel.packet[1]);
+  __m256 T2 = _mm256_unpacklo_ps(kernel.packet[2], kernel.packet[3]);
+  __m256 T3 = _mm256_unpackhi_ps(kernel.packet[2], kernel.packet[3]);
+  __m256 T4 = _mm256_unpacklo_ps(kernel.packet[4], kernel.packet[5]);
+  __m256 T5 = _mm256_unpackhi_ps(kernel.packet[4], kernel.packet[5]);
+  __m256 T6 = _mm256_unpacklo_ps(kernel.packet[6], kernel.packet[7]);
+  __m256 T7 = _mm256_unpackhi_ps(kernel.packet[6], kernel.packet[7]);
+  __m256 S0 = _mm256_shuffle_ps(T0,T2,_MM_SHUFFLE(1,0,1,0));
+  __m256 S1 = _mm256_shuffle_ps(T0,T2,_MM_SHUFFLE(3,2,3,2));
+  __m256 S2 = _mm256_shuffle_ps(T1,T3,_MM_SHUFFLE(1,0,1,0));
+  __m256 S3 = _mm256_shuffle_ps(T1,T3,_MM_SHUFFLE(3,2,3,2));
+  __m256 S4 = _mm256_shuffle_ps(T4,T6,_MM_SHUFFLE(1,0,1,0));
+  __m256 S5 = _mm256_shuffle_ps(T4,T6,_MM_SHUFFLE(3,2,3,2));
+  __m256 S6 = _mm256_shuffle_ps(T5,T7,_MM_SHUFFLE(1,0,1,0));
+  __m256 S7 = _mm256_shuffle_ps(T5,T7,_MM_SHUFFLE(3,2,3,2));
+  kernel.packet[0] = _mm256_permute2f128_ps(S0, S4, 0x20);
+  kernel.packet[1] = _mm256_permute2f128_ps(S1, S5, 0x20);
+  kernel.packet[2] = _mm256_permute2f128_ps(S2, S6, 0x20);
+  kernel.packet[3] = _mm256_permute2f128_ps(S3, S7, 0x20);
+  kernel.packet[4] = _mm256_permute2f128_ps(S0, S4, 0x31);
+  kernel.packet[5] = _mm256_permute2f128_ps(S1, S5, 0x31);
+  kernel.packet[6] = _mm256_permute2f128_ps(S2, S6, 0x31);
+  kernel.packet[7] = _mm256_permute2f128_ps(S3, S7, 0x31);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet8f,4>& kernel) {
+  __m256 T0 = _mm256_unpacklo_ps(kernel.packet[0], kernel.packet[1]);
+  __m256 T1 = _mm256_unpackhi_ps(kernel.packet[0], kernel.packet[1]);
+  __m256 T2 = _mm256_unpacklo_ps(kernel.packet[2], kernel.packet[3]);
+  __m256 T3 = _mm256_unpackhi_ps(kernel.packet[2], kernel.packet[3]);
+
+  __m256 S0 = _mm256_shuffle_ps(T0,T2,_MM_SHUFFLE(1,0,1,0));
+  __m256 S1 = _mm256_shuffle_ps(T0,T2,_MM_SHUFFLE(3,2,3,2));
+  __m256 S2 = _mm256_shuffle_ps(T1,T3,_MM_SHUFFLE(1,0,1,0));
+  __m256 S3 = _mm256_shuffle_ps(T1,T3,_MM_SHUFFLE(3,2,3,2));
+
+  kernel.packet[0] = _mm256_permute2f128_ps(S0, S1, 0x20);
+  kernel.packet[1] = _mm256_permute2f128_ps(S2, S3, 0x20);
+  kernel.packet[2] = _mm256_permute2f128_ps(S0, S1, 0x31);
+  kernel.packet[3] = _mm256_permute2f128_ps(S2, S3, 0x31);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4d,4>& kernel) {
+  __m256d T0 = _mm256_shuffle_pd(kernel.packet[0], kernel.packet[1], 15);
+  __m256d T1 = _mm256_shuffle_pd(kernel.packet[0], kernel.packet[1], 0);
+  __m256d T2 = _mm256_shuffle_pd(kernel.packet[2], kernel.packet[3], 15);
+  __m256d T3 = _mm256_shuffle_pd(kernel.packet[2], kernel.packet[3], 0);
+
+  kernel.packet[1] = _mm256_permute2f128_pd(T0, T2, 32);
+  kernel.packet[3] = _mm256_permute2f128_pd(T0, T2, 49);
+  kernel.packet[0] = _mm256_permute2f128_pd(T1, T3, 32);
+  kernel.packet[2] = _mm256_permute2f128_pd(T1, T3, 49);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pblend(const Selector<8>& ifPacket, const Packet8f& thenPacket, const Packet8f& elsePacket) {
+  const __m256 zero = _mm256_setzero_ps();
+  const __m256 select = _mm256_set_ps(ifPacket.select[7], ifPacket.select[6], ifPacket.select[5], ifPacket.select[4], ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m256 false_mask = _mm256_cmp_ps(select, zero, _CMP_EQ_UQ);
+  return _mm256_blendv_ps(thenPacket, elsePacket, false_mask);
+}
+template<> EIGEN_STRONG_INLINE Packet4d pblend(const Selector<4>& ifPacket, const Packet4d& thenPacket, const Packet4d& elsePacket) {
+  const __m256d zero = _mm256_setzero_pd();
+  const __m256d select = _mm256_set_pd(ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m256d false_mask = _mm256_cmp_pd(select, zero, _CMP_EQ_UQ);
+  return _mm256_blendv_pd(thenPacket, elsePacket, false_mask);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pinsertfirst(const Packet8f& a, float b)
+{
+  return _mm256_blend_ps(a,pset1<Packet8f>(b),1);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4d pinsertfirst(const Packet4d& a, double b)
+{
+  return _mm256_blend_pd(a,pset1<Packet4d>(b),1);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pinsertlast(const Packet8f& a, float b)
+{
+  return _mm256_blend_ps(a,pset1<Packet8f>(b),(1<<7));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4d pinsertlast(const Packet4d& a, double b)
+{
+  return _mm256_blend_pd(a,pset1<Packet4d>(b),(1<<3));
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PACKET_MATH_AVX_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/AVX/TypeCasting.h b/extern/Eigen3/Eigen/src/Core/arch/AVX/TypeCasting.h
new file mode 100644
index 00000000000..83bfdc604be
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/AVX/TypeCasting.h
@@ -0,0 +1,51 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_TYPE_CASTING_AVX_H
+#define EIGEN_TYPE_CASTING_AVX_H
+
+namespace Eigen {
+
+namespace internal {
+
+// For now we use SSE to handle integers, so we can't use AVX instructions to cast
+// from int to float
+template <>
+struct type_casting_traits<float, int> {
+  enum {
+    VectorizedCast = 0,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template <>
+struct type_casting_traits<int, float> {
+  enum {
+    VectorizedCast = 0,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+
+
+template<> EIGEN_STRONG_INLINE Packet8i pcast<Packet8f, Packet8i>(const Packet8f& a) {
+  return _mm256_cvtps_epi32(a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8i, Packet8f>(const Packet8i& a) {
+  return _mm256_cvtepi32_ps(a);
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TYPE_CASTING_AVX_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/AVX512/MathFunctions.h b/extern/Eigen3/Eigen/src/Core/arch/AVX512/MathFunctions.h
new file mode 100644
index 00000000000..9c1717f76dd
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/AVX512/MathFunctions.h
@@ -0,0 +1,391 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Pedro Gonnet (pedro.gonnet@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef THIRD_PARTY_EIGEN3_EIGEN_SRC_CORE_ARCH_AVX512_MATHFUNCTIONS_H_
+#define THIRD_PARTY_EIGEN3_EIGEN_SRC_CORE_ARCH_AVX512_MATHFUNCTIONS_H_
+
+namespace Eigen {
+
+namespace internal {
+
+// Disable the code for older versions of gcc that don't support many of the required avx512 instrinsics.
+#if EIGEN_GNUC_AT_LEAST(5, 3)
+
+#define _EIGEN_DECLARE_CONST_Packet16f(NAME, X) \
+  const Packet16f p16f_##NAME = pset1<Packet16f>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(NAME, X) \
+  const Packet16f p16f_##NAME = (__m512)pset1<Packet16i>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet8d(NAME, X) \
+  const Packet8d p8d_##NAME = pset1<Packet8d>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(NAME, X) \
+  const Packet8d p8d_##NAME = _mm512_castsi512_pd(_mm512_set1_epi64(X))
+
+// Natural logarithm
+// Computes log(x) as log(2^e * m) = C*e + log(m), where the constant C =log(2)
+// and m is in the range [sqrt(1/2),sqrt(2)). In this range, the logarithm can
+// be easily approximated by a polynomial centered on m=1 for stability.
+#if defined(EIGEN_VECTORIZE_AVX512DQ)
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+plog<Packet16f>(const Packet16f& _x) {
+  Packet16f x = _x;
+  _EIGEN_DECLARE_CONST_Packet16f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet16f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet16f(126f, 126.0f);
+
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(inv_mant_mask, ~0x7f800000);
+
+  // The smallest non denormalized float number.
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(min_norm_pos, 0x00800000);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(minus_inf, 0xff800000);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(nan, 0x7fc00000);
+
+  // Polynomial coefficients.
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_SQRTHF, 0.707106781186547524f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p0, 7.0376836292E-2f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p1, -1.1514610310E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p2, 1.1676998740E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p3, -1.2420140846E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p4, +1.4249322787E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p5, -1.6668057665E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p6, +2.0000714765E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p7, -2.4999993993E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p8, +3.3333331174E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_q1, -2.12194440e-4f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_q2, 0.693359375f);
+
+  // invalid_mask is set to true when x is NaN
+  __mmask16 invalid_mask =
+      _mm512_cmp_ps_mask(x, _mm512_setzero_ps(), _CMP_NGE_UQ);
+  __mmask16 iszero_mask =
+      _mm512_cmp_ps_mask(x, _mm512_setzero_ps(), _CMP_EQ_UQ);
+
+  // Truncate input values to the minimum positive normal.
+  x = pmax(x, p16f_min_norm_pos);
+
+  // Extract the shifted exponents.
+  Packet16f emm0 = _mm512_cvtepi32_ps(_mm512_srli_epi32((__m512i)x, 23));
+  Packet16f e = _mm512_sub_ps(emm0, p16f_126f);
+
+  // Set the exponents to -1, i.e. x are in the range [0.5,1).
+  x = _mm512_and_ps(x, p16f_inv_mant_mask);
+  x = _mm512_or_ps(x, p16f_half);
+
+  // part2: Shift the inputs from the range [0.5,1) to [sqrt(1/2),sqrt(2))
+  // and shift by -1. The values are then centered around 0, which improves
+  // the stability of the polynomial evaluation.
+  //   if( x < SQRTHF ) {
+  //     e -= 1;
+  //     x = x + x - 1.0;
+  //   } else { x = x - 1.0; }
+  __mmask16 mask = _mm512_cmp_ps_mask(x, p16f_cephes_SQRTHF, _CMP_LT_OQ);
+  Packet16f tmp = _mm512_mask_blend_ps(mask, _mm512_setzero_ps(), x);
+  x = psub(x, p16f_1);
+  e = psub(e, _mm512_mask_blend_ps(mask, _mm512_setzero_ps(), p16f_1));
+  x = padd(x, tmp);
+
+  Packet16f x2 = pmul(x, x);
+  Packet16f x3 = pmul(x2, x);
+
+  // Evaluate the polynomial approximant of degree 8 in three parts, probably
+  // to improve instruction-level parallelism.
+  Packet16f y, y1, y2;
+  y = pmadd(p16f_cephes_log_p0, x, p16f_cephes_log_p1);
+  y1 = pmadd(p16f_cephes_log_p3, x, p16f_cephes_log_p4);
+  y2 = pmadd(p16f_cephes_log_p6, x, p16f_cephes_log_p7);
+  y = pmadd(y, x, p16f_cephes_log_p2);
+  y1 = pmadd(y1, x, p16f_cephes_log_p5);
+  y2 = pmadd(y2, x, p16f_cephes_log_p8);
+  y = pmadd(y, x3, y1);
+  y = pmadd(y, x3, y2);
+  y = pmul(y, x3);
+
+  // Add the logarithm of the exponent back to the result of the interpolation.
+  y1 = pmul(e, p16f_cephes_log_q1);
+  tmp = pmul(x2, p16f_half);
+  y = padd(y, y1);
+  x = psub(x, tmp);
+  y2 = pmul(e, p16f_cephes_log_q2);
+  x = padd(x, y);
+  x = padd(x, y2);
+
+  // Filter out invalid inputs, i.e. negative arg will be NAN, 0 will be -INF.
+  return _mm512_mask_blend_ps(iszero_mask,
+                              _mm512_mask_blend_ps(invalid_mask, x, p16f_nan),
+                              p16f_minus_inf);
+}
+#endif
+
+// Exponential function. Works by writing "x = m*log(2) + r" where
+// "m = floor(x/log(2)+1/2)" and "r" is the remainder. The result is then
+// "exp(x) = 2^m*exp(r)" where exp(r) is in the range [-1,1).
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+pexp<Packet16f>(const Packet16f& _x) {
+  _EIGEN_DECLARE_CONST_Packet16f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet16f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet16f(127, 127.0f);
+
+  _EIGEN_DECLARE_CONST_Packet16f(exp_hi, 88.3762626647950f);
+  _EIGEN_DECLARE_CONST_Packet16f(exp_lo, -88.3762626647949f);
+
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_LOG2EF, 1.44269504088896341f);
+
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p0, 1.9875691500E-4f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p1, 1.3981999507E-3f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p2, 8.3334519073E-3f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p3, 4.1665795894E-2f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p4, 1.6666665459E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p5, 5.0000001201E-1f);
+
+  // Clamp x.
+  Packet16f x = pmax(pmin(_x, p16f_exp_hi), p16f_exp_lo);
+
+  // Express exp(x) as exp(m*ln(2) + r), start by extracting
+  // m = floor(x/ln(2) + 0.5).
+  Packet16f m = _mm512_floor_ps(pmadd(x, p16f_cephes_LOG2EF, p16f_half));
+
+  // Get r = x - m*ln(2). Note that we can do this without losing more than one
+  // ulp precision due to the FMA instruction.
+  _EIGEN_DECLARE_CONST_Packet16f(nln2, -0.6931471805599453f);
+  Packet16f r = _mm512_fmadd_ps(m, p16f_nln2, x);
+  Packet16f r2 = pmul(r, r);
+
+  // TODO(gonnet): Split into odd/even polynomials and try to exploit
+  //               instruction-level parallelism.
+  Packet16f y = p16f_cephes_exp_p0;
+  y = pmadd(y, r, p16f_cephes_exp_p1);
+  y = pmadd(y, r, p16f_cephes_exp_p2);
+  y = pmadd(y, r, p16f_cephes_exp_p3);
+  y = pmadd(y, r, p16f_cephes_exp_p4);
+  y = pmadd(y, r, p16f_cephes_exp_p5);
+  y = pmadd(y, r2, r);
+  y = padd(y, p16f_1);
+
+  // Build emm0 = 2^m.
+  Packet16i emm0 = _mm512_cvttps_epi32(padd(m, p16f_127));
+  emm0 = _mm512_slli_epi32(emm0, 23);
+
+  // Return 2^m * exp(r).
+  return pmax(pmul(y, _mm512_castsi512_ps(emm0)), _x);
+}
+
+/*template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
+pexp<Packet8d>(const Packet8d& _x) {
+  Packet8d x = _x;
+
+  _EIGEN_DECLARE_CONST_Packet8d(1, 1.0);
+  _EIGEN_DECLARE_CONST_Packet8d(2, 2.0);
+
+  _EIGEN_DECLARE_CONST_Packet8d(exp_hi, 709.437);
+  _EIGEN_DECLARE_CONST_Packet8d(exp_lo, -709.436139303);
+
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_LOG2EF, 1.4426950408889634073599);
+
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_p0, 1.26177193074810590878e-4);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_p1, 3.02994407707441961300e-2);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_q0, 3.00198505138664455042e-6);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_q1, 2.52448340349684104192e-3);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_q2, 2.27265548208155028766e-1);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_q3, 2.00000000000000000009e0);
+
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_C1, 0.693145751953125);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_C2, 1.42860682030941723212e-6);
+
+  // clamp x
+  x = pmax(pmin(x, p8d_exp_hi), p8d_exp_lo);
+
+  // Express exp(x) as exp(g + n*log(2)).
+  const Packet8d n =
+      _mm512_mul_round_pd(p8d_cephes_LOG2EF, x, _MM_FROUND_TO_NEAREST_INT);
+
+  // Get the remainder modulo log(2), i.e. the "g" described above. Subtract
+  // n*log(2) out in two steps, i.e. n*C1 + n*C2, C1+C2=log2 to get the last
+  // digits right.
+  const Packet8d nC1 = pmul(n, p8d_cephes_exp_C1);
+  const Packet8d nC2 = pmul(n, p8d_cephes_exp_C2);
+  x = psub(x, nC1);
+  x = psub(x, nC2);
+
+  const Packet8d x2 = pmul(x, x);
+
+  // Evaluate the numerator polynomial of the rational interpolant.
+  Packet8d px = p8d_cephes_exp_p0;
+  px = pmadd(px, x2, p8d_cephes_exp_p1);
+  px = pmadd(px, x2, p8d_cephes_exp_p2);
+  px = pmul(px, x);
+
+  // Evaluate the denominator polynomial of the rational interpolant.
+  Packet8d qx = p8d_cephes_exp_q0;
+  qx = pmadd(qx, x2, p8d_cephes_exp_q1);
+  qx = pmadd(qx, x2, p8d_cephes_exp_q2);
+  qx = pmadd(qx, x2, p8d_cephes_exp_q3);
+
+  // I don't really get this bit, copied from the SSE2 routines, so...
+  // TODO(gonnet): Figure out what is going on here, perhaps find a better
+  // rational interpolant?
+  x = _mm512_div_pd(px, psub(qx, px));
+  x = pmadd(p8d_2, x, p8d_1);
+
+  // Build e=2^n.
+  const Packet8d e = _mm512_castsi512_pd(_mm512_slli_epi64(
+      _mm512_add_epi64(_mm512_cvtpd_epi64(n), _mm512_set1_epi64(1023)), 52));
+
+  // Construct the result 2^n * exp(g) = e * x. The max is used to catch
+  // non-finite values in the input.
+  return pmax(pmul(x, e), _x);
+  }*/
+
+// Functions for sqrt.
+// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step
+// of Newton's method, at a cost of 1-2 bits of precision as opposed to the
+// exact solution. The main advantage of this approach is not just speed, but
+// also the fact that it can be inlined and pipelined with other computations,
+// further reducing its effective latency.
+#if EIGEN_FAST_MATH
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+psqrt<Packet16f>(const Packet16f& _x) {
+  _EIGEN_DECLARE_CONST_Packet16f(one_point_five, 1.5f);
+  _EIGEN_DECLARE_CONST_Packet16f(minus_half, -0.5f);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(flt_min, 0x00800000);
+
+  Packet16f neg_half = pmul(_x, p16f_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  __mmask16 non_zero_mask = _mm512_cmp_ps_mask(_x, p16f_flt_min, _CMP_GE_OQ);
+  Packet16f x = _mm512_mask_blend_ps(non_zero_mask, _mm512_setzero_ps(), _mm512_rsqrt14_ps(_x));
+
+  // Do a single step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p16f_one_point_five));
+
+  // Multiply the original _x by it's reciprocal square root to extract the
+  // square root.
+  return pmul(_x, x);
+}
+
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
+psqrt<Packet8d>(const Packet8d& _x) {
+  _EIGEN_DECLARE_CONST_Packet8d(one_point_five, 1.5);
+  _EIGEN_DECLARE_CONST_Packet8d(minus_half, -0.5);
+  _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(dbl_min, 0x0010000000000000LL);
+
+  Packet8d neg_half = pmul(_x, p8d_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  __mmask8 non_zero_mask = _mm512_cmp_pd_mask(_x, p8d_dbl_min, _CMP_GE_OQ);
+  Packet8d x = _mm512_mask_blend_pd(non_zero_mask, _mm512_setzero_pd(), _mm512_rsqrt14_pd(_x));
+
+  // Do a first step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8d_one_point_five));
+
+  // Do a second step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8d_one_point_five));
+
+  // Multiply the original _x by it's reciprocal square root to extract the
+  // square root.
+  return pmul(_x, x);
+}
+#else
+template <>
+EIGEN_STRONG_INLINE Packet16f psqrt<Packet16f>(const Packet16f& x) {
+  return _mm512_sqrt_ps(x);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d psqrt<Packet8d>(const Packet8d& x) {
+  return _mm512_sqrt_pd(x);
+}
+#endif
+
+// Functions for rsqrt.
+// Almost identical to the sqrt routine, just leave out the last multiplication
+// and fill in NaN/Inf where needed. Note that this function only exists as an
+// iterative version for doubles since there is no instruction for diretly
+// computing the reciprocal square root in AVX-512.
+#ifdef EIGEN_FAST_MATH
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+prsqrt<Packet16f>(const Packet16f& _x) {
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(inf, 0x7f800000);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(nan, 0x7fc00000);
+  _EIGEN_DECLARE_CONST_Packet16f(one_point_five, 1.5f);
+  _EIGEN_DECLARE_CONST_Packet16f(minus_half, -0.5f);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(flt_min, 0x00800000);
+
+  Packet16f neg_half = pmul(_x, p16f_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  __mmask16 le_zero_mask = _mm512_cmp_ps_mask(_x, p16f_flt_min, _CMP_LT_OQ);
+  Packet16f x = _mm512_mask_blend_ps(le_zero_mask, _mm512_rsqrt14_ps(_x), _mm512_setzero_ps());
+
+  // Fill in NaNs and Infs for the negative/zero entries.
+  __mmask16 neg_mask = _mm512_cmp_ps_mask(_x, _mm512_setzero_ps(), _CMP_LT_OQ);
+  Packet16f infs_and_nans = _mm512_mask_blend_ps(
+      neg_mask, _mm512_mask_blend_ps(le_zero_mask, _mm512_setzero_ps(), p16f_inf), p16f_nan);
+
+  // Do a single step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p16f_one_point_five));
+
+  // Insert NaNs and Infs in all the right places.
+  return _mm512_mask_blend_ps(le_zero_mask, x, infs_and_nans);
+}
+
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
+prsqrt<Packet8d>(const Packet8d& _x) {
+  _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(inf, 0x7ff0000000000000LL);
+  _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(nan, 0x7ff1000000000000LL);
+  _EIGEN_DECLARE_CONST_Packet8d(one_point_five, 1.5);
+  _EIGEN_DECLARE_CONST_Packet8d(minus_half, -0.5);
+  _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(dbl_min, 0x0010000000000000LL);
+
+  Packet8d neg_half = pmul(_x, p8d_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  __mmask8 le_zero_mask = _mm512_cmp_pd_mask(_x, p8d_dbl_min, _CMP_LT_OQ);
+  Packet8d x = _mm512_mask_blend_pd(le_zero_mask, _mm512_rsqrt14_pd(_x), _mm512_setzero_pd());
+
+  // Fill in NaNs and Infs for the negative/zero entries.
+  __mmask8 neg_mask = _mm512_cmp_pd_mask(_x, _mm512_setzero_pd(), _CMP_LT_OQ);
+  Packet8d infs_and_nans = _mm512_mask_blend_pd(
+      neg_mask, _mm512_mask_blend_pd(le_zero_mask, _mm512_setzero_pd(), p8d_inf), p8d_nan);
+
+  // Do a first step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8d_one_point_five));
+
+  // Do a second step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8d_one_point_five));
+
+  // Insert NaNs and Infs in all the right places.
+  return _mm512_mask_blend_pd(le_zero_mask, x, infs_and_nans);
+}
+#elif defined(EIGEN_VECTORIZE_AVX512ER)
+template <>
+EIGEN_STRONG_INLINE Packet16f prsqrt<Packet16f>(const Packet16f& x) {
+  return _mm512_rsqrt28_ps(x);
+}
+#endif
+#endif
+
+}  // end namespace internal
+
+}  // end namespace Eigen
+
+#endif  // THIRD_PARTY_EIGEN3_EIGEN_SRC_CORE_ARCH_AVX512_MATHFUNCTIONS_H_
diff --git a/extern/Eigen3/Eigen/src/Core/arch/AVX512/PacketMath.h b/extern/Eigen3/Eigen/src/Core/arch/AVX512/PacketMath.h
new file mode 100644
index 00000000000..5adddc7aec4
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/AVX512/PacketMath.h
@@ -0,0 +1,1316 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Benoit Steiner (benoit.steiner.goog@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_AVX512_H
+#define EIGEN_PACKET_MATH_AVX512_H
+
+namespace Eigen {
+
+namespace internal {
+
+#ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+#define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 8
+#endif
+
+#ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS (2*sizeof(void*))
+#endif
+
+#ifdef __FMA__
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#endif
+#endif
+
+typedef __m512 Packet16f;
+typedef __m512i Packet16i;
+typedef __m512d Packet8d;
+
+template <>
+struct is_arithmetic<__m512> {
+  enum { value = true };
+};
+template <>
+struct is_arithmetic<__m512i> {
+  enum { value = true };
+};
+template <>
+struct is_arithmetic<__m512d> {
+  enum { value = true };
+};
+
+template<> struct packet_traits<float>  : default_packet_traits
+{
+  typedef Packet16f type;
+  typedef Packet8f half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 16,
+    HasHalfPacket = 1,
+#if EIGEN_GNUC_AT_LEAST(5, 3)
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+    HasLog = 1,
+#endif
+    HasExp = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+#endif
+    HasDiv = 1
+  };
+ };
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef Packet8d type;
+  typedef Packet4d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 8,
+    HasHalfPacket = 1,
+#if EIGEN_GNUC_AT_LEAST(5, 3)
+    HasSqrt = 1,
+    HasRsqrt = EIGEN_FAST_MATH,
+#endif
+    HasDiv = 1
+  };
+};
+
+/* TODO Implement AVX512 for integers
+template<> struct packet_traits<int>    : default_packet_traits
+{
+  typedef Packet16i type;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=8
+  };
+};
+*/
+
+template <>
+struct unpacket_traits<Packet16f> {
+  typedef float type;
+  typedef Packet8f half;
+  enum { size = 16, alignment=Aligned64 };
+};
+template <>
+struct unpacket_traits<Packet8d> {
+  typedef double type;
+  typedef Packet4d half;
+  enum { size = 8, alignment=Aligned64 };
+};
+template <>
+struct unpacket_traits<Packet16i> {
+  typedef int type;
+  typedef Packet8i half;
+  enum { size = 16, alignment=Aligned64 };
+};
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pset1<Packet16f>(const float& from) {
+  return _mm512_set1_ps(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pset1<Packet8d>(const double& from) {
+  return _mm512_set1_pd(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet16i pset1<Packet16i>(const int& from) {
+  return _mm512_set1_epi32(from);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pload1<Packet16f>(const float* from) {
+  return _mm512_broadcastss_ps(_mm_load_ps1(from));
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pload1<Packet8d>(const double* from) {
+  return _mm512_broadcastsd_pd(_mm_load_pd1(from));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f plset<Packet16f>(const float& a) {
+  return _mm512_add_ps(
+      _mm512_set1_ps(a),
+      _mm512_set_ps(15.0f, 14.0f, 13.0f, 12.0f, 11.0f, 10.0f, 9.0f, 8.0f, 7.0f, 6.0f, 5.0f,
+                    4.0f, 3.0f, 2.0f, 1.0f, 0.0f));
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d plset<Packet8d>(const double& a) {
+  return _mm512_add_pd(_mm512_set1_pd(a),
+                       _mm512_set_pd(7.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0, 0.0));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f padd<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_add_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d padd<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_add_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f psub<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_sub_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d psub<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_sub_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pnegate(const Packet16f& a) {
+  return _mm512_sub_ps(_mm512_set1_ps(0.0), a);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pnegate(const Packet8d& a) {
+  return _mm512_sub_pd(_mm512_set1_pd(0.0), a);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pconj(const Packet16f& a) {
+  return a;
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pconj(const Packet8d& a) {
+  return a;
+}
+template <>
+EIGEN_STRONG_INLINE Packet16i pconj(const Packet16i& a) {
+  return a;
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pmul<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_mul_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pmul<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_mul_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pdiv<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_div_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pdiv<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_div_pd(a, b);
+}
+
+#ifdef __FMA__
+template <>
+EIGEN_STRONG_INLINE Packet16f pmadd(const Packet16f& a, const Packet16f& b,
+                                    const Packet16f& c) {
+  return _mm512_fmadd_ps(a, b, c);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pmadd(const Packet8d& a, const Packet8d& b,
+                                   const Packet8d& c) {
+  return _mm512_fmadd_pd(a, b, c);
+}
+#endif
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pmin<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_min_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pmin<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_min_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pmax<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_max_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pmax<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_max_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pand<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_and_ps(a, b);
+#else
+  Packet16f res = _mm512_undefined_ps();
+  Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0);
+  res = _mm512_insertf32x4(res, _mm_and_ps(lane0_a, lane0_b), 0);
+
+  Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1);
+  res = _mm512_insertf32x4(res, _mm_and_ps(lane1_a, lane1_b), 1);
+
+  Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2);
+  res = _mm512_insertf32x4(res, _mm_and_ps(lane2_a, lane2_b), 2);
+
+  Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3);
+  Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3);
+  res = _mm512_insertf32x4(res, _mm_and_ps(lane3_a, lane3_b), 3);
+
+  return res;
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pand<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_and_pd(a, b);
+#else
+  Packet8d res = _mm512_undefined_pd();
+  Packet4d lane0_a = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane0_b = _mm512_extractf64x4_pd(b, 0);
+  res = _mm512_insertf64x4(res, _mm256_and_pd(lane0_a, lane0_b), 0);
+
+  Packet4d lane1_a = _mm512_extractf64x4_pd(a, 1);
+  Packet4d lane1_b = _mm512_extractf64x4_pd(b, 1);
+  res = _mm512_insertf64x4(res, _mm256_and_pd(lane1_a, lane1_b), 1);
+
+  return res;
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet16f por<Packet16f>(const Packet16f& a,
+                                             const Packet16f& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_or_ps(a, b);
+#else
+  Packet16f res = _mm512_undefined_ps();
+  Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0);
+  res = _mm512_insertf32x4(res, _mm_or_ps(lane0_a, lane0_b), 0);
+
+  Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1);
+  res = _mm512_insertf32x4(res, _mm_or_ps(lane1_a, lane1_b), 1);
+
+  Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2);
+  res = _mm512_insertf32x4(res, _mm_or_ps(lane2_a, lane2_b), 2);
+
+  Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3);
+  Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3);
+  res = _mm512_insertf32x4(res, _mm_or_ps(lane3_a, lane3_b), 3);
+
+  return res;
+#endif
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet8d por<Packet8d>(const Packet8d& a,
+                                           const Packet8d& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_or_pd(a, b);
+#else
+  Packet8d res = _mm512_undefined_pd();
+  Packet4d lane0_a = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane0_b = _mm512_extractf64x4_pd(b, 0);
+  res = _mm512_insertf64x4(res, _mm256_or_pd(lane0_a, lane0_b), 0);
+
+  Packet4d lane1_a = _mm512_extractf64x4_pd(a, 1);
+  Packet4d lane1_b = _mm512_extractf64x4_pd(b, 1);
+  res = _mm512_insertf64x4(res, _mm256_or_pd(lane1_a, lane1_b), 1);
+
+  return res;
+#endif
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pxor<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_xor_ps(a, b);
+#else
+  Packet16f res = _mm512_undefined_ps();
+  Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0);
+  res = _mm512_insertf32x4(res, _mm_xor_ps(lane0_a, lane0_b), 0);
+
+  Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1);
+  res = _mm512_insertf32x4(res, _mm_xor_ps(lane1_a, lane1_b), 1);
+
+  Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2);
+  res = _mm512_insertf32x4(res, _mm_xor_ps(lane2_a, lane2_b), 2);
+
+  Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3);
+  Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3);
+  res = _mm512_insertf32x4(res, _mm_xor_ps(lane3_a, lane3_b), 3);
+
+  return res;
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pxor<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_xor_pd(a, b);
+#else
+  Packet8d res = _mm512_undefined_pd();
+  Packet4d lane0_a = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane0_b = _mm512_extractf64x4_pd(b, 0);
+  res = _mm512_insertf64x4(res, _mm256_xor_pd(lane0_a, lane0_b), 0);
+
+  Packet4d lane1_a = _mm512_extractf64x4_pd(a, 1);
+  Packet4d lane1_b = _mm512_extractf64x4_pd(b, 1);
+  res = _mm512_insertf64x4(res, _mm256_xor_pd(lane1_a, lane1_b), 1);
+
+  return res;
+#endif
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pandnot<Packet16f>(const Packet16f& a,
+                                                 const Packet16f& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_andnot_ps(a, b);
+#else
+  Packet16f res = _mm512_undefined_ps();
+  Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0);
+  res = _mm512_insertf32x4(res, _mm_andnot_ps(lane0_a, lane0_b), 0);
+
+  Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1);
+  res = _mm512_insertf32x4(res, _mm_andnot_ps(lane1_a, lane1_b), 1);
+
+  Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2);
+  res = _mm512_insertf32x4(res, _mm_andnot_ps(lane2_a, lane2_b), 2);
+
+  Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3);
+  Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3);
+  res = _mm512_insertf32x4(res, _mm_andnot_ps(lane3_a, lane3_b), 3);
+
+  return res;
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pandnot<Packet8d>(const Packet8d& a,
+                                               const Packet8d& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_andnot_pd(a, b);
+#else
+  Packet8d res = _mm512_undefined_pd();
+  Packet4d lane0_a = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane0_b = _mm512_extractf64x4_pd(b, 0);
+  res = _mm512_insertf64x4(res, _mm256_andnot_pd(lane0_a, lane0_b), 0);
+
+  Packet4d lane1_a = _mm512_extractf64x4_pd(a, 1);
+  Packet4d lane1_b = _mm512_extractf64x4_pd(b, 1);
+  res = _mm512_insertf64x4(res, _mm256_andnot_pd(lane1_a, lane1_b), 1);
+
+  return res;
+#endif
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pload<Packet16f>(const float* from) {
+  EIGEN_DEBUG_ALIGNED_LOAD return _mm512_load_ps(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pload<Packet8d>(const double* from) {
+  EIGEN_DEBUG_ALIGNED_LOAD return _mm512_load_pd(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet16i pload<Packet16i>(const int* from) {
+  EIGEN_DEBUG_ALIGNED_LOAD return _mm512_load_si512(
+      reinterpret_cast<const __m512i*>(from));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f ploadu<Packet16f>(const float* from) {
+  EIGEN_DEBUG_UNALIGNED_LOAD return _mm512_loadu_ps(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d ploadu<Packet8d>(const double* from) {
+  EIGEN_DEBUG_UNALIGNED_LOAD return _mm512_loadu_pd(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet16i ploadu<Packet16i>(const int* from) {
+  EIGEN_DEBUG_UNALIGNED_LOAD return _mm512_loadu_si512(
+      reinterpret_cast<const __m512i*>(from));
+}
+
+// Loads 8 floats from memory a returns the packet
+// {a0, a0  a1, a1, a2, a2, a3, a3, a4, a4, a5, a5, a6, a6, a7, a7}
+template <>
+EIGEN_STRONG_INLINE Packet16f ploaddup<Packet16f>(const float* from) {
+  Packet8f lane0 = _mm256_broadcast_ps((const __m128*)(const void*)from);
+  // mimic an "inplace" permutation of the lower 128bits using a blend
+  lane0 = _mm256_blend_ps(
+      lane0, _mm256_castps128_ps256(_mm_permute_ps(
+                 _mm256_castps256_ps128(lane0), _MM_SHUFFLE(1, 0, 1, 0))),
+      15);
+  // then we can perform a consistent permutation on the global register to get
+  // everything in shape:
+  lane0 = _mm256_permute_ps(lane0, _MM_SHUFFLE(3, 3, 2, 2));
+
+  Packet8f lane1 = _mm256_broadcast_ps((const __m128*)(const void*)(from + 4));
+  // mimic an "inplace" permutation of the lower 128bits using a blend
+  lane1 = _mm256_blend_ps(
+      lane1, _mm256_castps128_ps256(_mm_permute_ps(
+                 _mm256_castps256_ps128(lane1), _MM_SHUFFLE(1, 0, 1, 0))),
+      15);
+  // then we can perform a consistent permutation on the global register to get
+  // everything in shape:
+  lane1 = _mm256_permute_ps(lane1, _MM_SHUFFLE(3, 3, 2, 2));
+
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  Packet16f res = _mm512_undefined_ps();
+  return _mm512_insertf32x8(res, lane0, 0);
+  return _mm512_insertf32x8(res, lane1, 1);
+  return res;
+#else
+  Packet16f res = _mm512_undefined_ps();
+  res = _mm512_insertf32x4(res, _mm256_extractf128_ps(lane0, 0), 0);
+  res = _mm512_insertf32x4(res, _mm256_extractf128_ps(lane0, 1), 1);
+  res = _mm512_insertf32x4(res, _mm256_extractf128_ps(lane1, 0), 2);
+  res = _mm512_insertf32x4(res, _mm256_extractf128_ps(lane1, 1), 3);
+  return res;
+#endif
+}
+// Loads 4 doubles from memory a returns the packet {a0, a0  a1, a1, a2, a2, a3,
+// a3}
+template <>
+EIGEN_STRONG_INLINE Packet8d ploaddup<Packet8d>(const double* from) {
+  Packet4d lane0 = _mm256_broadcast_pd((const __m128d*)(const void*)from);
+  lane0 = _mm256_permute_pd(lane0, 3 << 2);
+
+  Packet4d lane1 = _mm256_broadcast_pd((const __m128d*)(const void*)(from + 2));
+  lane1 = _mm256_permute_pd(lane1, 3 << 2);
+
+  Packet8d res = _mm512_undefined_pd();
+  res = _mm512_insertf64x4(res, lane0, 0);
+  return _mm512_insertf64x4(res, lane1, 1);
+}
+
+// Loads 4 floats from memory a returns the packet
+// {a0, a0  a0, a0, a1, a1, a1, a1, a2, a2, a2, a2, a3, a3, a3, a3}
+template <>
+EIGEN_STRONG_INLINE Packet16f ploadquad<Packet16f>(const float* from) {
+  Packet16f tmp = _mm512_undefined_ps();
+  tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from), 0);
+  tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from + 1), 1);
+  tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from + 2), 2);
+  tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from + 3), 3);
+  return tmp;
+}
+// Loads 2 doubles from memory a returns the packet
+// {a0, a0  a0, a0, a1, a1, a1, a1}
+template <>
+EIGEN_STRONG_INLINE Packet8d ploadquad<Packet8d>(const double* from) {
+  Packet8d tmp = _mm512_undefined_pd();
+  Packet2d tmp0 = _mm_load_pd1(from);
+  Packet2d tmp1 = _mm_load_pd1(from + 1);
+  Packet4d lane0 = _mm256_broadcastsd_pd(tmp0);
+  Packet4d lane1 = _mm256_broadcastsd_pd(tmp1);
+  tmp = _mm512_insertf64x4(tmp, lane0, 0);
+  return _mm512_insertf64x4(tmp, lane1, 1);
+}
+
+template <>
+EIGEN_STRONG_INLINE void pstore<float>(float* to, const Packet16f& from) {
+  EIGEN_DEBUG_ALIGNED_STORE _mm512_store_ps(to, from);
+}
+template <>
+EIGEN_STRONG_INLINE void pstore<double>(double* to, const Packet8d& from) {
+  EIGEN_DEBUG_ALIGNED_STORE _mm512_store_pd(to, from);
+}
+template <>
+EIGEN_STRONG_INLINE void pstore<int>(int* to, const Packet16i& from) {
+  EIGEN_DEBUG_ALIGNED_STORE _mm512_storeu_si512(reinterpret_cast<__m512i*>(to),
+                                                from);
+}
+
+template <>
+EIGEN_STRONG_INLINE void pstoreu<float>(float* to, const Packet16f& from) {
+  EIGEN_DEBUG_UNALIGNED_STORE _mm512_storeu_ps(to, from);
+}
+template <>
+EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet8d& from) {
+  EIGEN_DEBUG_UNALIGNED_STORE _mm512_storeu_pd(to, from);
+}
+template <>
+EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet16i& from) {
+  EIGEN_DEBUG_UNALIGNED_STORE _mm512_storeu_si512(
+      reinterpret_cast<__m512i*>(to), from);
+}
+
+template <>
+EIGEN_DEVICE_FUNC inline Packet16f pgather<float, Packet16f>(const float* from,
+                                                             Index stride) {
+  Packet16i stride_vector = _mm512_set1_epi32(stride);
+  Packet16i stride_multiplier =
+      _mm512_set_epi32(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+  Packet16i indices = _mm512_mullo_epi32(stride_vector, stride_multiplier);
+
+  return _mm512_i32gather_ps(indices, from, 4);
+}
+template <>
+EIGEN_DEVICE_FUNC inline Packet8d pgather<double, Packet8d>(const double* from,
+                                                            Index stride) {
+  Packet8i stride_vector = _mm256_set1_epi32(stride);
+  Packet8i stride_multiplier = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0);
+  Packet8i indices = _mm256_mullo_epi32(stride_vector, stride_multiplier);
+
+  return _mm512_i32gather_pd(indices, from, 8);
+}
+
+template <>
+EIGEN_DEVICE_FUNC inline void pscatter<float, Packet16f>(float* to,
+                                                         const Packet16f& from,
+                                                         Index stride) {
+  Packet16i stride_vector = _mm512_set1_epi32(stride);
+  Packet16i stride_multiplier =
+      _mm512_set_epi32(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+  Packet16i indices = _mm512_mullo_epi32(stride_vector, stride_multiplier);
+  _mm512_i32scatter_ps(to, indices, from, 4);
+}
+template <>
+EIGEN_DEVICE_FUNC inline void pscatter<double, Packet8d>(double* to,
+                                                         const Packet8d& from,
+                                                         Index stride) {
+  Packet8i stride_vector = _mm256_set1_epi32(stride);
+  Packet8i stride_multiplier = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0);
+  Packet8i indices = _mm256_mullo_epi32(stride_vector, stride_multiplier);
+  _mm512_i32scatter_pd(to, indices, from, 8);
+}
+
+template <>
+EIGEN_STRONG_INLINE void pstore1<Packet16f>(float* to, const float& a) {
+  Packet16f pa = pset1<Packet16f>(a);
+  pstore(to, pa);
+}
+template <>
+EIGEN_STRONG_INLINE void pstore1<Packet8d>(double* to, const double& a) {
+  Packet8d pa = pset1<Packet8d>(a);
+  pstore(to, pa);
+}
+template <>
+EIGEN_STRONG_INLINE void pstore1<Packet16i>(int* to, const int& a) {
+  Packet16i pa = pset1<Packet16i>(a);
+  pstore(to, pa);
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+
+template <>
+EIGEN_STRONG_INLINE float pfirst<Packet16f>(const Packet16f& a) {
+  return _mm_cvtss_f32(_mm512_extractf32x4_ps(a, 0));
+}
+template <>
+EIGEN_STRONG_INLINE double pfirst<Packet8d>(const Packet8d& a) {
+  return _mm_cvtsd_f64(_mm256_extractf128_pd(_mm512_extractf64x4_pd(a, 0), 0));
+}
+template <>
+EIGEN_STRONG_INLINE int pfirst<Packet16i>(const Packet16i& a) {
+  return _mm_extract_epi32(_mm512_extracti32x4_epi32(a, 0), 0);
+}
+
+template<> EIGEN_STRONG_INLINE Packet16f preverse(const Packet16f& a)
+{
+  return _mm512_permutexvar_ps(_mm512_set_epi32(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15), a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8d preverse(const Packet8d& a)
+{
+  return _mm512_permutexvar_pd(_mm512_set_epi32(0, 0, 0, 1, 0, 2, 0, 3, 0, 4, 0, 5, 0, 6, 0, 7), a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet16f pabs(const Packet16f& a)
+{
+  // _mm512_abs_ps intrinsic not found, so hack around it
+  return _mm512_castsi512_ps(_mm512_and_si512(_mm512_castps_si512(a), _mm512_set1_epi32(0x7fffffff)));
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pabs(const Packet8d& a) {
+  // _mm512_abs_ps intrinsic not found, so hack around it
+  return _mm512_castsi512_pd(_mm512_and_si512(_mm512_castpd_si512(a),
+                                   _mm512_set1_epi64(0x7fffffffffffffff)));
+}
+
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+// AVX512F does not define _mm512_extractf32x8_ps to extract _m256 from _m512
+#define EIGEN_EXTRACT_8f_FROM_16f(INPUT, OUTPUT)                           \
+  __m256 OUTPUT##_0 = _mm512_extractf32x8_ps(INPUT, 0) __m256 OUTPUT##_1 = \
+      _mm512_extractf32x8_ps(INPUT, 1)
+#else
+#define EIGEN_EXTRACT_8f_FROM_16f(INPUT, OUTPUT)                \
+  __m256 OUTPUT##_0 = _mm256_insertf128_ps(                     \
+      _mm256_castps128_ps256(_mm512_extractf32x4_ps(INPUT, 0)), \
+      _mm512_extractf32x4_ps(INPUT, 1), 1);                     \
+  __m256 OUTPUT##_1 = _mm256_insertf128_ps(                     \
+      _mm256_castps128_ps256(_mm512_extractf32x4_ps(INPUT, 2)), \
+      _mm512_extractf32x4_ps(INPUT, 3), 1);
+#endif
+
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+#define EIGEN_INSERT_8f_INTO_16f(OUTPUT, INPUTA, INPUTB) \
+  OUTPUT = _mm512_insertf32x8(OUTPUT, INPUTA, 0);        \
+  OUTPUT = _mm512_insertf32x8(OUTPUT, INPUTB, 1);
+#else
+#define EIGEN_INSERT_8f_INTO_16f(OUTPUT, INPUTA, INPUTB)                    \
+  OUTPUT = _mm512_insertf32x4(OUTPUT, _mm256_extractf128_ps(INPUTA, 0), 0); \
+  OUTPUT = _mm512_insertf32x4(OUTPUT, _mm256_extractf128_ps(INPUTA, 1), 1); \
+  OUTPUT = _mm512_insertf32x4(OUTPUT, _mm256_extractf128_ps(INPUTB, 0), 2); \
+  OUTPUT = _mm512_insertf32x4(OUTPUT, _mm256_extractf128_ps(INPUTB, 1), 3);
+#endif
+template<> EIGEN_STRONG_INLINE Packet16f preduxp<Packet16f>(const Packet16f*
+vecs)
+{
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[0], vecs0);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[1], vecs1);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[2], vecs2);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[3], vecs3);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[4], vecs4);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[5], vecs5);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[6], vecs6);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[7], vecs7);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[8], vecs8);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[9], vecs9);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[10], vecs10);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[11], vecs11);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[12], vecs12);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[13], vecs13);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[14], vecs14);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[15], vecs15);
+
+  __m256 hsum1 = _mm256_hadd_ps(vecs0_0, vecs1_0);
+  __m256 hsum2 = _mm256_hadd_ps(vecs2_0, vecs3_0);
+  __m256 hsum3 = _mm256_hadd_ps(vecs4_0, vecs5_0);
+  __m256 hsum4 = _mm256_hadd_ps(vecs6_0, vecs7_0);
+
+  __m256 hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+  __m256 hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+  __m256 hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+  __m256 hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+  __m256 perm1 = _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+  __m256 perm2 = _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+  __m256 perm3 = _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+  __m256 perm4 = _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+  __m256 sum1 = _mm256_add_ps(perm1, hsum5);
+  __m256 sum2 = _mm256_add_ps(perm2, hsum6);
+  __m256 sum3 = _mm256_add_ps(perm3, hsum7);
+  __m256 sum4 = _mm256_add_ps(perm4, hsum8);
+
+  __m256 blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+  __m256 blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+  __m256 final = _mm256_blend_ps(blend1, blend2, 0xf0);
+
+  hsum1 = _mm256_hadd_ps(vecs0_1, vecs1_1);
+  hsum2 = _mm256_hadd_ps(vecs2_1, vecs3_1);
+  hsum3 = _mm256_hadd_ps(vecs4_1, vecs5_1);
+  hsum4 = _mm256_hadd_ps(vecs6_1, vecs7_1);
+
+  hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+  hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+  hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+  hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+  perm1 = _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+  perm2 = _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+  perm3 = _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+  perm4 = _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+  sum1 = _mm256_add_ps(perm1, hsum5);
+  sum2 = _mm256_add_ps(perm2, hsum6);
+  sum3 = _mm256_add_ps(perm3, hsum7);
+  sum4 = _mm256_add_ps(perm4, hsum8);
+
+  blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+  blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+  final = padd(final, _mm256_blend_ps(blend1, blend2, 0xf0));
+
+  hsum1 = _mm256_hadd_ps(vecs8_0, vecs9_0);
+  hsum2 = _mm256_hadd_ps(vecs10_0, vecs11_0);
+  hsum3 = _mm256_hadd_ps(vecs12_0, vecs13_0);
+  hsum4 = _mm256_hadd_ps(vecs14_0, vecs15_0);
+
+  hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+  hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+  hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+  hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+  perm1 = _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+  perm2 = _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+  perm3 = _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+  perm4 = _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+  sum1 = _mm256_add_ps(perm1, hsum5);
+  sum2 = _mm256_add_ps(perm2, hsum6);
+  sum3 = _mm256_add_ps(perm3, hsum7);
+  sum4 = _mm256_add_ps(perm4, hsum8);
+
+  blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+  blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+  __m256 final_1 = _mm256_blend_ps(blend1, blend2, 0xf0);
+
+  hsum1 = _mm256_hadd_ps(vecs8_1, vecs9_1);
+  hsum2 = _mm256_hadd_ps(vecs10_1, vecs11_1);
+  hsum3 = _mm256_hadd_ps(vecs12_1, vecs13_1);
+  hsum4 = _mm256_hadd_ps(vecs14_1, vecs15_1);
+
+  hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+  hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+  hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+  hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+  perm1 = _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+  perm2 = _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+  perm3 = _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+  perm4 = _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+  sum1 = _mm256_add_ps(perm1, hsum5);
+  sum2 = _mm256_add_ps(perm2, hsum6);
+  sum3 = _mm256_add_ps(perm3, hsum7);
+  sum4 = _mm256_add_ps(perm4, hsum8);
+
+  blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+  blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+  final_1 = padd(final_1, _mm256_blend_ps(blend1, blend2, 0xf0));
+
+  __m512 final_output;
+
+  EIGEN_INSERT_8f_INTO_16f(final_output, final, final_1);
+  return final_output;
+}
+
+template<> EIGEN_STRONG_INLINE Packet8d preduxp<Packet8d>(const Packet8d* vecs)
+{
+  Packet4d vecs0_0 = _mm512_extractf64x4_pd(vecs[0], 0);
+  Packet4d vecs0_1 = _mm512_extractf64x4_pd(vecs[0], 1);
+
+  Packet4d vecs1_0 = _mm512_extractf64x4_pd(vecs[1], 0);
+  Packet4d vecs1_1 = _mm512_extractf64x4_pd(vecs[1], 1);
+
+  Packet4d vecs2_0 = _mm512_extractf64x4_pd(vecs[2], 0);
+  Packet4d vecs2_1 = _mm512_extractf64x4_pd(vecs[2], 1);
+
+  Packet4d vecs3_0 = _mm512_extractf64x4_pd(vecs[3], 0);
+  Packet4d vecs3_1 = _mm512_extractf64x4_pd(vecs[3], 1);
+
+  Packet4d vecs4_0 = _mm512_extractf64x4_pd(vecs[4], 0);
+  Packet4d vecs4_1 = _mm512_extractf64x4_pd(vecs[4], 1);
+
+  Packet4d vecs5_0 = _mm512_extractf64x4_pd(vecs[5], 0);
+  Packet4d vecs5_1 = _mm512_extractf64x4_pd(vecs[5], 1);
+
+  Packet4d vecs6_0 = _mm512_extractf64x4_pd(vecs[6], 0);
+  Packet4d vecs6_1 = _mm512_extractf64x4_pd(vecs[6], 1);
+
+  Packet4d vecs7_0 = _mm512_extractf64x4_pd(vecs[7], 0);
+  Packet4d vecs7_1 = _mm512_extractf64x4_pd(vecs[7], 1);
+
+  Packet4d tmp0, tmp1;
+
+  tmp0 = _mm256_hadd_pd(vecs0_0, vecs1_0);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs2_0, vecs3_0);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  __m256d final_0 = _mm256_blend_pd(tmp0, tmp1, 0xC);
+
+  tmp0 = _mm256_hadd_pd(vecs0_1, vecs1_1);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs2_1, vecs3_1);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  final_0 = padd(final_0, _mm256_blend_pd(tmp0, tmp1, 0xC));
+
+  tmp0 = _mm256_hadd_pd(vecs4_0, vecs5_0);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs6_0, vecs7_0);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  __m256d final_1 = _mm256_blend_pd(tmp0, tmp1, 0xC);
+
+  tmp0 = _mm256_hadd_pd(vecs4_1, vecs5_1);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs6_1, vecs7_1);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  final_1 = padd(final_1, _mm256_blend_pd(tmp0, tmp1, 0xC));
+
+  __m512d final_output = _mm512_insertf64x4(final_output, final_0, 0);
+
+  return _mm512_insertf64x4(final_output, final_1, 1);
+}
+
+template <>
+EIGEN_STRONG_INLINE float predux<Packet16f>(const Packet16f& a) {
+  //#ifdef EIGEN_VECTORIZE_AVX512DQ
+#if 0
+  Packet8f lane0 = _mm512_extractf32x8_ps(a, 0);
+  Packet8f lane1 = _mm512_extractf32x8_ps(a, 1);
+  Packet8f sum = padd(lane0, lane1);
+  Packet8f tmp0 = _mm256_hadd_ps(sum, _mm256_permute2f128_ps(a, a, 1));
+  tmp0 = _mm256_hadd_ps(tmp0, tmp0);
+  return pfirst(_mm256_hadd_ps(tmp0, tmp0));
+#else
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f sum = padd(padd(lane0, lane1), padd(lane2, lane3));
+  sum = _mm_hadd_ps(sum, sum);
+  sum = _mm_hadd_ps(sum, _mm_permute_ps(sum, 1));
+  return pfirst(sum);
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE double predux<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d sum = padd(lane0, lane1);
+  Packet4d tmp0 = _mm256_hadd_pd(sum, _mm256_permute2f128_pd(sum, sum, 1));
+  return pfirst(_mm256_hadd_pd(tmp0, tmp0));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet8f predux_downto4<Packet16f>(const Packet16f& a) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  Packet8f lane0 = _mm512_extractf32x8_ps(a, 0);
+  Packet8f lane1 = _mm512_extractf32x8_ps(a, 1);
+  return padd(lane0, lane1);
+#else
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f sum0 = padd(lane0, lane2);
+  Packet4f sum1 = padd(lane1, lane3);
+  return _mm256_insertf128_ps(_mm256_castps128_ps256(sum0), sum1, 1);
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet4d predux_downto4<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d res = padd(lane0, lane1);
+  return res;
+}
+
+template <>
+EIGEN_STRONG_INLINE float predux_mul<Packet16f>(const Packet16f& a) {
+//#ifdef EIGEN_VECTORIZE_AVX512DQ
+#if 0
+  Packet8f lane0 = _mm512_extractf32x8_ps(a, 0);
+  Packet8f lane1 = _mm512_extractf32x8_ps(a, 1);
+  Packet8f res = pmul(lane0, lane1);
+  res = pmul(res, _mm256_permute2f128_ps(res, res, 1));
+  res = pmul(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 3, 2)));
+  return pfirst(pmul(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 0, 1))));
+#else
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f res = pmul(pmul(lane0, lane1), pmul(lane2, lane3));
+  res = pmul(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 3, 2)));
+  return pfirst(pmul(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 0, 1))));
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE double predux_mul<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d res = pmul(lane0, lane1);
+  res = pmul(res, _mm256_permute2f128_pd(res, res, 1));
+  return pfirst(pmul(res, _mm256_shuffle_pd(res, res, 1)));
+}
+
+template <>
+EIGEN_STRONG_INLINE float predux_min<Packet16f>(const Packet16f& a) {
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f res = _mm_min_ps(_mm_min_ps(lane0, lane1), _mm_min_ps(lane2, lane3));
+  res = _mm_min_ps(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 3, 2)));
+  return pfirst(_mm_min_ps(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 0, 1))));
+}
+template <>
+EIGEN_STRONG_INLINE double predux_min<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d res = _mm256_min_pd(lane0, lane1);
+  res = _mm256_min_pd(res, _mm256_permute2f128_pd(res, res, 1));
+  return pfirst(_mm256_min_pd(res, _mm256_shuffle_pd(res, res, 1)));
+}
+
+template <>
+EIGEN_STRONG_INLINE float predux_max<Packet16f>(const Packet16f& a) {
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f res = _mm_max_ps(_mm_max_ps(lane0, lane1), _mm_max_ps(lane2, lane3));
+  res = _mm_max_ps(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 3, 2)));
+  return pfirst(_mm_max_ps(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 0, 1))));
+}
+template <>
+EIGEN_STRONG_INLINE double predux_max<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d res = _mm256_max_pd(lane0, lane1);
+  res = _mm256_max_pd(res, _mm256_permute2f128_pd(res, res, 1));
+  return pfirst(_mm256_max_pd(res, _mm256_shuffle_pd(res, res, 1)));
+}
+
+template <int Offset>
+struct palign_impl<Offset, Packet16f> {
+  static EIGEN_STRONG_INLINE void run(Packet16f& first,
+                                      const Packet16f& second) {
+    if (Offset != 0) {
+      __m512i first_idx = _mm512_set_epi32(
+          Offset + 15, Offset + 14, Offset + 13, Offset + 12, Offset + 11,
+          Offset + 10, Offset + 9, Offset + 8, Offset + 7, Offset + 6,
+          Offset + 5, Offset + 4, Offset + 3, Offset + 2, Offset + 1, Offset);
+
+      __m512i second_idx =
+          _mm512_set_epi32(Offset - 1, Offset - 2, Offset - 3, Offset - 4,
+                           Offset - 5, Offset - 6, Offset - 7, Offset - 8,
+                           Offset - 9, Offset - 10, Offset - 11, Offset - 12,
+                           Offset - 13, Offset - 14, Offset - 15, Offset - 16);
+
+      unsigned short mask = 0xFFFF;
+      mask <<= (16 - Offset);
+
+      first = _mm512_permutexvar_ps(first_idx, first);
+      Packet16f tmp = _mm512_permutexvar_ps(second_idx, second);
+      first = _mm512_mask_blend_ps(mask, first, tmp);
+    }
+  }
+};
+template <int Offset>
+struct palign_impl<Offset, Packet8d> {
+  static EIGEN_STRONG_INLINE void run(Packet8d& first, const Packet8d& second) {
+    if (Offset != 0) {
+      __m512i first_idx = _mm512_set_epi32(
+          0, Offset + 7, 0, Offset + 6, 0, Offset + 5, 0, Offset + 4, 0,
+          Offset + 3, 0, Offset + 2, 0, Offset + 1, 0, Offset);
+
+      __m512i second_idx = _mm512_set_epi32(
+          0, Offset - 1, 0, Offset - 2, 0, Offset - 3, 0, Offset - 4, 0,
+          Offset - 5, 0, Offset - 6, 0, Offset - 7, 0, Offset - 8);
+
+      unsigned char mask = 0xFF;
+      mask <<= (8 - Offset);
+
+      first = _mm512_permutexvar_pd(first_idx, first);
+      Packet8d tmp = _mm512_permutexvar_pd(second_idx, second);
+      first = _mm512_mask_blend_pd(mask, first, tmp);
+    }
+  }
+};
+
+
+#define PACK_OUTPUT(OUTPUT, INPUT, INDEX, STRIDE) \
+  EIGEN_INSERT_8f_INTO_16f(OUTPUT[INDEX], INPUT[INDEX], INPUT[INDEX + STRIDE]);
+
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet16f, 16>& kernel) {
+  __m512 T0 = _mm512_unpacklo_ps(kernel.packet[0], kernel.packet[1]);
+  __m512 T1 = _mm512_unpackhi_ps(kernel.packet[0], kernel.packet[1]);
+  __m512 T2 = _mm512_unpacklo_ps(kernel.packet[2], kernel.packet[3]);
+  __m512 T3 = _mm512_unpackhi_ps(kernel.packet[2], kernel.packet[3]);
+  __m512 T4 = _mm512_unpacklo_ps(kernel.packet[4], kernel.packet[5]);
+  __m512 T5 = _mm512_unpackhi_ps(kernel.packet[4], kernel.packet[5]);
+  __m512 T6 = _mm512_unpacklo_ps(kernel.packet[6], kernel.packet[7]);
+  __m512 T7 = _mm512_unpackhi_ps(kernel.packet[6], kernel.packet[7]);
+  __m512 T8 = _mm512_unpacklo_ps(kernel.packet[8], kernel.packet[9]);
+  __m512 T9 = _mm512_unpackhi_ps(kernel.packet[8], kernel.packet[9]);
+  __m512 T10 = _mm512_unpacklo_ps(kernel.packet[10], kernel.packet[11]);
+  __m512 T11 = _mm512_unpackhi_ps(kernel.packet[10], kernel.packet[11]);
+  __m512 T12 = _mm512_unpacklo_ps(kernel.packet[12], kernel.packet[13]);
+  __m512 T13 = _mm512_unpackhi_ps(kernel.packet[12], kernel.packet[13]);
+  __m512 T14 = _mm512_unpacklo_ps(kernel.packet[14], kernel.packet[15]);
+  __m512 T15 = _mm512_unpackhi_ps(kernel.packet[14], kernel.packet[15]);
+  __m512 S0 = _mm512_shuffle_ps(T0, T2, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S1 = _mm512_shuffle_ps(T0, T2, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S2 = _mm512_shuffle_ps(T1, T3, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S3 = _mm512_shuffle_ps(T1, T3, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S4 = _mm512_shuffle_ps(T4, T6, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S5 = _mm512_shuffle_ps(T4, T6, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S6 = _mm512_shuffle_ps(T5, T7, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S7 = _mm512_shuffle_ps(T5, T7, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S8 = _mm512_shuffle_ps(T8, T10, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S9 = _mm512_shuffle_ps(T8, T10, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S10 = _mm512_shuffle_ps(T9, T11, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S11 = _mm512_shuffle_ps(T9, T11, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S12 = _mm512_shuffle_ps(T12, T14, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S13 = _mm512_shuffle_ps(T12, T14, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S14 = _mm512_shuffle_ps(T13, T15, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S15 = _mm512_shuffle_ps(T13, T15, _MM_SHUFFLE(3, 2, 3, 2));
+
+  EIGEN_EXTRACT_8f_FROM_16f(S0, S0);
+  EIGEN_EXTRACT_8f_FROM_16f(S1, S1);
+  EIGEN_EXTRACT_8f_FROM_16f(S2, S2);
+  EIGEN_EXTRACT_8f_FROM_16f(S3, S3);
+  EIGEN_EXTRACT_8f_FROM_16f(S4, S4);
+  EIGEN_EXTRACT_8f_FROM_16f(S5, S5);
+  EIGEN_EXTRACT_8f_FROM_16f(S6, S6);
+  EIGEN_EXTRACT_8f_FROM_16f(S7, S7);
+  EIGEN_EXTRACT_8f_FROM_16f(S8, S8);
+  EIGEN_EXTRACT_8f_FROM_16f(S9, S9);
+  EIGEN_EXTRACT_8f_FROM_16f(S10, S10);
+  EIGEN_EXTRACT_8f_FROM_16f(S11, S11);
+  EIGEN_EXTRACT_8f_FROM_16f(S12, S12);
+  EIGEN_EXTRACT_8f_FROM_16f(S13, S13);
+  EIGEN_EXTRACT_8f_FROM_16f(S14, S14);
+  EIGEN_EXTRACT_8f_FROM_16f(S15, S15);
+
+  PacketBlock<Packet8f, 32> tmp;
+
+  tmp.packet[0] = _mm256_permute2f128_ps(S0_0, S4_0, 0x20);
+  tmp.packet[1] = _mm256_permute2f128_ps(S1_0, S5_0, 0x20);
+  tmp.packet[2] = _mm256_permute2f128_ps(S2_0, S6_0, 0x20);
+  tmp.packet[3] = _mm256_permute2f128_ps(S3_0, S7_0, 0x20);
+  tmp.packet[4] = _mm256_permute2f128_ps(S0_0, S4_0, 0x31);
+  tmp.packet[5] = _mm256_permute2f128_ps(S1_0, S5_0, 0x31);
+  tmp.packet[6] = _mm256_permute2f128_ps(S2_0, S6_0, 0x31);
+  tmp.packet[7] = _mm256_permute2f128_ps(S3_0, S7_0, 0x31);
+
+  tmp.packet[8] = _mm256_permute2f128_ps(S0_1, S4_1, 0x20);
+  tmp.packet[9] = _mm256_permute2f128_ps(S1_1, S5_1, 0x20);
+  tmp.packet[10] = _mm256_permute2f128_ps(S2_1, S6_1, 0x20);
+  tmp.packet[11] = _mm256_permute2f128_ps(S3_1, S7_1, 0x20);
+  tmp.packet[12] = _mm256_permute2f128_ps(S0_1, S4_1, 0x31);
+  tmp.packet[13] = _mm256_permute2f128_ps(S1_1, S5_1, 0x31);
+  tmp.packet[14] = _mm256_permute2f128_ps(S2_1, S6_1, 0x31);
+  tmp.packet[15] = _mm256_permute2f128_ps(S3_1, S7_1, 0x31);
+
+  // Second set of _m256 outputs
+  tmp.packet[16] = _mm256_permute2f128_ps(S8_0, S12_0, 0x20);
+  tmp.packet[17] = _mm256_permute2f128_ps(S9_0, S13_0, 0x20);
+  tmp.packet[18] = _mm256_permute2f128_ps(S10_0, S14_0, 0x20);
+  tmp.packet[19] = _mm256_permute2f128_ps(S11_0, S15_0, 0x20);
+  tmp.packet[20] = _mm256_permute2f128_ps(S8_0, S12_0, 0x31);
+  tmp.packet[21] = _mm256_permute2f128_ps(S9_0, S13_0, 0x31);
+  tmp.packet[22] = _mm256_permute2f128_ps(S10_0, S14_0, 0x31);
+  tmp.packet[23] = _mm256_permute2f128_ps(S11_0, S15_0, 0x31);
+
+  tmp.packet[24] = _mm256_permute2f128_ps(S8_1, S12_1, 0x20);
+  tmp.packet[25] = _mm256_permute2f128_ps(S9_1, S13_1, 0x20);
+  tmp.packet[26] = _mm256_permute2f128_ps(S10_1, S14_1, 0x20);
+  tmp.packet[27] = _mm256_permute2f128_ps(S11_1, S15_1, 0x20);
+  tmp.packet[28] = _mm256_permute2f128_ps(S8_1, S12_1, 0x31);
+  tmp.packet[29] = _mm256_permute2f128_ps(S9_1, S13_1, 0x31);
+  tmp.packet[30] = _mm256_permute2f128_ps(S10_1, S14_1, 0x31);
+  tmp.packet[31] = _mm256_permute2f128_ps(S11_1, S15_1, 0x31);
+
+  // Pack them into the output
+  PACK_OUTPUT(kernel.packet, tmp.packet, 0, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 1, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 2, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 3, 16);
+
+  PACK_OUTPUT(kernel.packet, tmp.packet, 4, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 5, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 6, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 7, 16);
+
+  PACK_OUTPUT(kernel.packet, tmp.packet, 8, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 9, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 10, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 11, 16);
+
+  PACK_OUTPUT(kernel.packet, tmp.packet, 12, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 13, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 14, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 15, 16);
+}
+#define PACK_OUTPUT_2(OUTPUT, INPUT, INDEX, STRIDE)         \
+  EIGEN_INSERT_8f_INTO_16f(OUTPUT[INDEX], INPUT[2 * INDEX], \
+                           INPUT[2 * INDEX + STRIDE]);
+
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet16f, 4>& kernel) {
+  __m512 T0 = _mm512_unpacklo_ps(kernel.packet[0], kernel.packet[1]);
+  __m512 T1 = _mm512_unpackhi_ps(kernel.packet[0], kernel.packet[1]);
+  __m512 T2 = _mm512_unpacklo_ps(kernel.packet[2], kernel.packet[3]);
+  __m512 T3 = _mm512_unpackhi_ps(kernel.packet[2], kernel.packet[3]);
+
+  __m512 S0 = _mm512_shuffle_ps(T0, T2, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S1 = _mm512_shuffle_ps(T0, T2, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S2 = _mm512_shuffle_ps(T1, T3, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S3 = _mm512_shuffle_ps(T1, T3, _MM_SHUFFLE(3, 2, 3, 2));
+
+  EIGEN_EXTRACT_8f_FROM_16f(S0, S0);
+  EIGEN_EXTRACT_8f_FROM_16f(S1, S1);
+  EIGEN_EXTRACT_8f_FROM_16f(S2, S2);
+  EIGEN_EXTRACT_8f_FROM_16f(S3, S3);
+
+  PacketBlock<Packet8f, 8> tmp;
+
+  tmp.packet[0] = _mm256_permute2f128_ps(S0_0, S1_0, 0x20);
+  tmp.packet[1] = _mm256_permute2f128_ps(S2_0, S3_0, 0x20);
+  tmp.packet[2] = _mm256_permute2f128_ps(S0_0, S1_0, 0x31);
+  tmp.packet[3] = _mm256_permute2f128_ps(S2_0, S3_0, 0x31);
+
+  tmp.packet[4] = _mm256_permute2f128_ps(S0_1, S1_1, 0x20);
+  tmp.packet[5] = _mm256_permute2f128_ps(S2_1, S3_1, 0x20);
+  tmp.packet[6] = _mm256_permute2f128_ps(S0_1, S1_1, 0x31);
+  tmp.packet[7] = _mm256_permute2f128_ps(S2_1, S3_1, 0x31);
+
+  PACK_OUTPUT_2(kernel.packet, tmp.packet, 0, 1);
+  PACK_OUTPUT_2(kernel.packet, tmp.packet, 1, 1);
+  PACK_OUTPUT_2(kernel.packet, tmp.packet, 2, 1);
+  PACK_OUTPUT_2(kernel.packet, tmp.packet, 3, 1);
+}
+
+#define PACK_OUTPUT_SQ_D(OUTPUT, INPUT, INDEX, STRIDE)                \
+  OUTPUT[INDEX] = _mm512_insertf64x4(OUTPUT[INDEX], INPUT[INDEX], 0); \
+  OUTPUT[INDEX] = _mm512_insertf64x4(OUTPUT[INDEX], INPUT[INDEX + STRIDE], 1);
+
+#define PACK_OUTPUT_D(OUTPUT, INPUT, INDEX, STRIDE)                         \
+  OUTPUT[INDEX] = _mm512_insertf64x4(OUTPUT[INDEX], INPUT[(2 * INDEX)], 0); \
+  OUTPUT[INDEX] =                                                           \
+      _mm512_insertf64x4(OUTPUT[INDEX], INPUT[(2 * INDEX) + STRIDE], 1);
+
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet8d, 4>& kernel) {
+  __m512d T0 = _mm512_shuffle_pd(kernel.packet[0], kernel.packet[1], 0);
+  __m512d T1 = _mm512_shuffle_pd(kernel.packet[0], kernel.packet[1], 0xff);
+  __m512d T2 = _mm512_shuffle_pd(kernel.packet[2], kernel.packet[3], 0);
+  __m512d T3 = _mm512_shuffle_pd(kernel.packet[2], kernel.packet[3], 0xff);
+
+  PacketBlock<Packet4d, 8> tmp;
+
+  tmp.packet[0] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 0),
+                                         _mm512_extractf64x4_pd(T2, 0), 0x20);
+  tmp.packet[1] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 0),
+                                         _mm512_extractf64x4_pd(T3, 0), 0x20);
+  tmp.packet[2] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 0),
+                                         _mm512_extractf64x4_pd(T2, 0), 0x31);
+  tmp.packet[3] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 0),
+                                         _mm512_extractf64x4_pd(T3, 0), 0x31);
+
+  tmp.packet[4] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 1),
+                                         _mm512_extractf64x4_pd(T2, 1), 0x20);
+  tmp.packet[5] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 1),
+                                         _mm512_extractf64x4_pd(T3, 1), 0x20);
+  tmp.packet[6] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 1),
+                                         _mm512_extractf64x4_pd(T2, 1), 0x31);
+  tmp.packet[7] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 1),
+                                         _mm512_extractf64x4_pd(T3, 1), 0x31);
+
+  PACK_OUTPUT_D(kernel.packet, tmp.packet, 0, 1);
+  PACK_OUTPUT_D(kernel.packet, tmp.packet, 1, 1);
+  PACK_OUTPUT_D(kernel.packet, tmp.packet, 2, 1);
+  PACK_OUTPUT_D(kernel.packet, tmp.packet, 3, 1);
+}
+
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet8d, 8>& kernel) {
+  __m512d T0 = _mm512_unpacklo_pd(kernel.packet[0], kernel.packet[1]);
+  __m512d T1 = _mm512_unpackhi_pd(kernel.packet[0], kernel.packet[1]);
+  __m512d T2 = _mm512_unpacklo_pd(kernel.packet[2], kernel.packet[3]);
+  __m512d T3 = _mm512_unpackhi_pd(kernel.packet[2], kernel.packet[3]);
+  __m512d T4 = _mm512_unpacklo_pd(kernel.packet[4], kernel.packet[5]);
+  __m512d T5 = _mm512_unpackhi_pd(kernel.packet[4], kernel.packet[5]);
+  __m512d T6 = _mm512_unpacklo_pd(kernel.packet[6], kernel.packet[7]);
+  __m512d T7 = _mm512_unpackhi_pd(kernel.packet[6], kernel.packet[7]);
+
+  PacketBlock<Packet4d, 16> tmp;
+
+  tmp.packet[0] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 0),
+                                         _mm512_extractf64x4_pd(T2, 0), 0x20);
+  tmp.packet[1] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 0),
+                                         _mm512_extractf64x4_pd(T3, 0), 0x20);
+  tmp.packet[2] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 0),
+                                         _mm512_extractf64x4_pd(T2, 0), 0x31);
+  tmp.packet[3] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 0),
+                                         _mm512_extractf64x4_pd(T3, 0), 0x31);
+
+  tmp.packet[4] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 1),
+                                         _mm512_extractf64x4_pd(T2, 1), 0x20);
+  tmp.packet[5] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 1),
+                                         _mm512_extractf64x4_pd(T3, 1), 0x20);
+  tmp.packet[6] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 1),
+                                         _mm512_extractf64x4_pd(T2, 1), 0x31);
+  tmp.packet[7] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 1),
+                                         _mm512_extractf64x4_pd(T3, 1), 0x31);
+
+  tmp.packet[8] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T4, 0),
+                                         _mm512_extractf64x4_pd(T6, 0), 0x20);
+  tmp.packet[9] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T5, 0),
+                                         _mm512_extractf64x4_pd(T7, 0), 0x20);
+  tmp.packet[10] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T4, 0),
+                                          _mm512_extractf64x4_pd(T6, 0), 0x31);
+  tmp.packet[11] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T5, 0),
+                                          _mm512_extractf64x4_pd(T7, 0), 0x31);
+
+  tmp.packet[12] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T4, 1),
+                                          _mm512_extractf64x4_pd(T6, 1), 0x20);
+  tmp.packet[13] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T5, 1),
+                                          _mm512_extractf64x4_pd(T7, 1), 0x20);
+  tmp.packet[14] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T4, 1),
+                                          _mm512_extractf64x4_pd(T6, 1), 0x31);
+  tmp.packet[15] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T5, 1),
+                                          _mm512_extractf64x4_pd(T7, 1), 0x31);
+
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 0, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 1, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 2, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 3, 8);
+
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 4, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 5, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 6, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 7, 8);
+}
+template <>
+EIGEN_STRONG_INLINE Packet16f pblend(const Selector<16>& /*ifPacket*/,
+                                     const Packet16f& /*thenPacket*/,
+                                     const Packet16f& /*elsePacket*/) {
+  assert(false && "To be implemented");
+  return Packet16f();
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pblend(const Selector<8>& /*ifPacket*/,
+                                    const Packet8d& /*thenPacket*/,
+                                    const Packet8d& /*elsePacket*/) {
+  assert(false && "To be implemented");
+  return Packet8d();
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PACKET_MATH_AVX512_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/AltiVec/Complex.h b/extern/Eigen3/Eigen/src/Core/arch/AltiVec/Complex.h
index 68d9a2bff8d..3e665730cfa 100644
--- a/extern/Eigen3/Eigen/src/Core/arch/AltiVec/Complex.h
+++ b/extern/Eigen3/Eigen/src/Core/arch/AltiVec/Complex.h
@@ -2,30 +2,34 @@
 // for linear algebra.
 //
 // Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2010-2016 Konstantinos Margaritis <markos@freevec.org>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
-#ifndef EIGEN_COMPLEX_ALTIVEC_H
-#define EIGEN_COMPLEX_ALTIVEC_H
+#ifndef EIGEN_COMPLEX32_ALTIVEC_H
+#define EIGEN_COMPLEX32_ALTIVEC_H
 
 namespace Eigen {
 
 namespace internal {
 
-static Packet4ui  p4ui_CONJ_XOR = vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_ZERO_);//{ 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
-static Packet16uc p16uc_COMPLEX_RE   = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 };
-static Packet16uc p16uc_COMPLEX_IM   = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 1), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 };
-static Packet16uc p16uc_COMPLEX_REV  = vec_sld(p16uc_REVERSE, p16uc_REVERSE, 8);//{ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11 };
-static Packet16uc p16uc_COMPLEX_REV2 = vec_sld(p16uc_FORWARD, p16uc_FORWARD, 8);//{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
-static Packet16uc p16uc_PSET_HI = (Packet16uc) vec_mergeh((Packet4ui) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet4ui) vec_splat((Packet4ui)p16uc_FORWARD, 1));//{ 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };
-static Packet16uc p16uc_PSET_LO = (Packet16uc) vec_mergeh((Packet4ui) vec_splat((Packet4ui)p16uc_FORWARD, 2), (Packet4ui) vec_splat((Packet4ui)p16uc_FORWARD, 3));//{ 8,9,10,11, 12,13,14,15, 8,9,10,11, 12,13,14,15 };
+static Packet4ui  p4ui_CONJ_XOR = vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_MZERO);//{ 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
+#ifdef __VSX__
+#if defined(_BIG_ENDIAN)
+static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_MZERO, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_MZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+#else
+static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_MZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2d_MZERO, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+#endif
+#endif
 
 //---------- float ----------
 struct Packet2cf
 {
-  EIGEN_STRONG_INLINE Packet2cf() {}
+  EIGEN_STRONG_INLINE explicit Packet2cf() : v(p4f_ZERO) {}
   EIGEN_STRONG_INLINE explicit Packet2cf(const Packet4f& a) : v(a) {}
   Packet4f  v;
 };
@@ -33,10 +37,12 @@ struct Packet2cf
 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
 {
   typedef Packet2cf type;
+  typedef Packet2cf half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 2,
+    HasHalfPacket = 0,
 
     HasAdd    = 1,
     HasSub    = 1,
@@ -47,65 +53,78 @@ template<> struct packet_traits<std::complex<float> >  : default_packet_traits
     HasAbs2   = 0,
     HasMin    = 0,
     HasMax    = 0,
+#ifdef __VSX__
+    HasBlend  = 1,
+#endif
     HasSetLinear = 0
   };
 };
 
-template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; };
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
 
 template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>&  from)
 {
   Packet2cf res;
-  /* On AltiVec we cannot load 64-bit registers, so wa have to take care of alignment */
-  if((ptrdiff_t(&from) % 16) == 0)
+  if((std::ptrdiff_t(&from) % 16) == 0)
     res.v = pload<Packet4f>((const float *)&from);
   else
     res.v = ploadu<Packet4f>((const float *)&from);
-  res.v = vec_perm(res.v, res.v, p16uc_PSET_HI);
+  res.v = vec_perm(res.v, res.v, p16uc_PSET64_HI);
   return res;
 }
 
-template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_add(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_sub(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pload<Packet2cf>(const std::complex<float>*        from) { return Packet2cf(pload<Packet4f>((const float *) from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>*       from) { return Packet2cf(ploadu<Packet4f>((const float*) from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>*     from) { return pset1<Packet2cf>(*from); }
+
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { pstore((float*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { pstoreu((float*)to, from.v); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  return pload<Packet2cf>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  pstore<std::complex<float> >((std::complex<float> *) af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(a.v + b.v); }
+template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(a.v - b.v); }
 template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(a.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) { return Packet2cf((Packet4f)vec_xor((Packet4ui)a.v, p4ui_CONJ_XOR)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) { return Packet2cf(pxor<Packet4f>(a.v, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR))); }
 
 template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
   Packet4f v1, v2;
 
   // Permute and multiply the real parts of a and b
-  v1 = vec_perm(a.v, a.v, p16uc_COMPLEX_RE);
+  v1 = vec_perm(a.v, a.v, p16uc_PSET32_WODD);
   // Get the imaginary parts of a
-  v2 = vec_perm(a.v, a.v, p16uc_COMPLEX_IM);
+  v2 = vec_perm(a.v, a.v, p16uc_PSET32_WEVEN);
   // multiply a_re * b 
   v1 = vec_madd(v1, b.v, p4f_ZERO);
   // multiply a_im * b and get the conjugate result
   v2 = vec_madd(v2, b.v, p4f_ZERO);
-  v2 = (Packet4f) vec_xor((Packet4ui)v2, p4ui_CONJ_XOR);
+  v2 = reinterpret_cast<Packet4f>(pxor(v2, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR)));
   // permute back to a proper order
-  v2 = vec_perm(v2, v2, p16uc_COMPLEX_REV);
+  v2 = vec_perm(v2, v2, p16uc_COMPLEX32_REV);
   
-  return Packet2cf(vec_add(v1, v2));
+  return Packet2cf(padd<Packet4f>(v1, v2));
 }
 
-template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_and(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_or(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pxor   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_xor(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_and(a.v, vec_nor(b.v,b.v))); }
+template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pxor   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v, b.v)); }
 
-template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>((const float*)from)); }
-template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>((const float*)from)); }
-
-template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>*     from)
-{
-  return pset1<Packet2cf>(*from);
-}
-
-template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
-template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
-
-template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { vec_dstt((float *)addr, DST_CTRL(2,2,32), DST_CHAN); }
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr)    { EIGEN_PPC_PREFETCH(addr); }
 
 template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
 {
@@ -118,26 +137,30 @@ template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Pack
 template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
 {
   Packet4f rev_a;
-  rev_a = vec_perm(a.v, a.v, p16uc_COMPLEX_REV2);
+  rev_a = vec_perm(a.v, a.v, p16uc_COMPLEX32_REV2);
   return Packet2cf(rev_a);
 }
 
 template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
 {
   Packet4f b;
-  b = (Packet4f) vec_sld(a.v, a.v, 8);
-  b = padd(a.v, b);
-  return pfirst(Packet2cf(b));
+  b = vec_sld(a.v, a.v, 8);
+  b = padd<Packet4f>(a.v, b);
+  return pfirst<Packet2cf>(Packet2cf(b));
 }
 
 template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
 {
   Packet4f b1, b2;
-  
-  b1 = (Packet4f) vec_sld(vecs[0].v, vecs[1].v, 8);
-  b2 = (Packet4f) vec_sld(vecs[1].v, vecs[0].v, 8);
-  b2 = (Packet4f) vec_sld(b2, b2, 8);
-  b2 = padd(b1, b2);
+#ifdef _BIG_ENDIAN  
+  b1 = vec_sld(vecs[0].v, vecs[1].v, 8);
+  b2 = vec_sld(vecs[1].v, vecs[0].v, 8);
+#else
+  b1 = vec_sld(vecs[1].v, vecs[0].v, 8);
+  b2 = vec_sld(vecs[0].v, vecs[1].v, 8);
+#endif
+  b2 = vec_sld(b2, b2, 8);
+  b2 = padd<Packet4f>(b1, b2);
 
   return Packet2cf(b2);
 }
@@ -146,10 +169,10 @@ template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const P
 {
   Packet4f b;
   Packet2cf prod;
-  b = (Packet4f) vec_sld(a.v, a.v, 8);
-  prod = pmul(a, Packet2cf(b));
+  b = vec_sld(a.v, a.v, 8);
+  prod = pmul<Packet2cf>(a, Packet2cf(b));
 
-  return pfirst(prod);
+  return pfirst<Packet2cf>(prod);
 }
 
 template<int Offset>
@@ -159,7 +182,11 @@ struct palign_impl<Offset,Packet2cf>
   {
     if (Offset==1)
     {
+#ifdef _BIG_ENDIAN
       first.v = vec_sld(first.v, second.v, 8);
+#else
+      first.v = vec_sld(second.v, first.v, 8);
+#endif
     }
   }
 };
@@ -197,21 +224,207 @@ template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
   }
 };
 
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
+
 template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
   // TODO optimize it for AltiVec
-  Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a,b);
-  Packet4f s = vec_madd(b.v, b.v, p4f_ZERO);
-  return Packet2cf(pdiv(res.v, vec_add(s,vec_perm(s, s, p16uc_COMPLEX_REV))));
+  Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a, b);
+  Packet4f s = pmul<Packet4f>(b.v, b.v);
+  return Packet2cf(pdiv(res.v, padd<Packet4f>(s, vec_perm(s, s, p16uc_COMPLEX32_REV))));
 }
 
 template<> EIGEN_STRONG_INLINE Packet2cf pcplxflip<Packet2cf>(const Packet2cf& x)
 {
-  return Packet2cf(vec_perm(x.v, x.v, p16uc_COMPLEX_REV));
+  return Packet2cf(vec_perm(x.v, x.v, p16uc_COMPLEX32_REV));
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
+{
+  Packet4f tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
+  kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
+  kernel.packet[0].v = tmp;
+}
+
+#ifdef __VSX__
+template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
+  Packet2cf result;
+  result.v = reinterpret_cast<Packet4f>(pblend<Packet2d>(ifPacket, reinterpret_cast<Packet2d>(thenPacket.v), reinterpret_cast<Packet2d>(elsePacket.v)));
+  return result;
+}
+#endif
+
+//---------- double ----------
+#ifdef __VSX__
+struct Packet1cd
+{
+  EIGEN_STRONG_INLINE Packet1cd() {}
+  EIGEN_STRONG_INLINE explicit Packet1cd(const Packet2d& a) : v(a) {}
+  Packet2d v;
+};
+
+template<> struct packet_traits<std::complex<double> >  : default_packet_traits
+{
+  typedef Packet1cd type;
+  typedef Packet1cd half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 0,
+    size = 1,
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
+
+template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from) { return Packet1cd(pload<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { return Packet1cd(ploadu<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { pstore((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { pstoreu((double*)to, from.v); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>&  from)
+{ /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index stride)
+{
+  std::complex<double> EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  return pload<Packet1cd>(af);
 }
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index stride)
+{
+  std::complex<double> EIGEN_ALIGN16 af[2];
+  pstore<std::complex<double> >(af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v + b.v); }
+template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v - b.v); }
+template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
+template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd(pxor(a.v, reinterpret_cast<Packet2d>(p2ul_CONJ_XOR2))); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  Packet2d a_re, a_im, v1, v2;
+
+  // Permute and multiply the real parts of a and b
+  a_re = vec_perm(a.v, a.v, p16uc_PSET64_HI);
+  // Get the imaginary parts of a
+  a_im = vec_perm(a.v, a.v, p16uc_PSET64_LO);
+  // multiply a_re * b
+  v1 = vec_madd(a_re, b.v, p2d_ZERO);
+  // multiply a_im * b and get the conjugate result
+  v2 = vec_madd(a_im, b.v, p2d_ZERO);
+  v2 = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(v2), reinterpret_cast<Packet4ui>(v2), 8));
+  v2 = pxor(v2, reinterpret_cast<Packet2d>(p2ul_CONJ_XOR1));
 
+  return Packet1cd(padd<Packet2d>(v1, v2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(pand(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(por(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(pxor(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(pandnot(a.v, b.v)); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>*     from)  { return pset1<Packet1cd>(*from); }
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> * addr)    { EIGEN_PPC_PREFETCH(addr); }
+
+template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
+{
+  std::complex<double> EIGEN_ALIGN16 res[2];
+  pstore<std::complex<double> >(res, a);
+
+  return res[0];
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
+template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs)        { return vecs[0]; }
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
+
+template<int Offset>
+struct palign_impl<Offset,Packet1cd>
+{
+  static EIGEN_STRONG_INLINE void run(Packet1cd& /*first*/, const Packet1cd& /*second*/)
+  {
+    // FIXME is it sure we never have to align a Packet1cd?
+    // Even though a std::complex<double> has 16 bytes, it is not necessarily aligned on a 16 bytes boundary...
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,false>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
+
+template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  // TODO optimize it for AltiVec
+  Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
+  Packet2d s = pmul<Packet2d>(b.v, b.v);
+  return Packet1cd(pdiv(res.v, padd<Packet2d>(s, vec_perm(s, s, p16uc_REVERSE64))));
+}
+
+EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
+{
+  return Packet1cd(preverse(Packet2d(x.v)));
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
+{
+  Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
+  kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
+  kernel.packet[0].v = tmp;
+}
+#endif // __VSX__
 } // end namespace internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_COMPLEX_ALTIVEC_H
+#endif // EIGEN_COMPLEX32_ALTIVEC_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/AltiVec/MathFunctions.h b/extern/Eigen3/Eigen/src/Core/arch/AltiVec/MathFunctions.h
new file mode 100644
index 00000000000..c5e4bede74d
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/AltiVec/MathFunctions.h
@@ -0,0 +1,322 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2007 Julien Pommier
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+/* The sin, cos, exp, and log functions of this file come from
+ * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
+ */
+
+#ifndef EIGEN_MATH_FUNCTIONS_ALTIVEC_H
+#define EIGEN_MATH_FUNCTIONS_ALTIVEC_H
+
+namespace Eigen {
+
+namespace internal {
+
+static _EIGEN_DECLARE_CONST_Packet4f(1 , 1.0f);
+static _EIGEN_DECLARE_CONST_Packet4f(half, 0.5f);
+static _EIGEN_DECLARE_CONST_Packet4i(0x7f, 0x7f);
+static _EIGEN_DECLARE_CONST_Packet4i(23, 23);
+
+static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(inv_mant_mask, ~0x7f800000);
+
+/* the smallest non denormalized float number */
+static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(min_norm_pos,  0x00800000);
+static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(minus_inf,     0xff800000); // -1.f/0.f
+static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(minus_nan,     0xffffffff);
+  
+/* natural logarithm computed for 4 simultaneous float
+  return NaN for x <= 0
+*/
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_SQRTHF, 0.707106781186547524f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p0, 7.0376836292E-2f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p1, - 1.1514610310E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p2, 1.1676998740E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p3, - 1.2420140846E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p4, + 1.4249322787E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p5, - 1.6668057665E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p6, + 2.0000714765E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p7, - 2.4999993993E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p8, + 3.3333331174E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_q1, -2.12194440e-4f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_q2, 0.693359375f);
+
+static _EIGEN_DECLARE_CONST_Packet4f(exp_hi,  88.3762626647950f);
+static _EIGEN_DECLARE_CONST_Packet4f(exp_lo, -88.3762626647949f);
+
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_LOG2EF, 1.44269504088896341f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C1, 0.693359375f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C2, -2.12194440e-4f);
+
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p0, 1.9875691500E-4f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p1, 1.3981999507E-3f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p2, 8.3334519073E-3f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p3, 4.1665795894E-2f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p4, 1.6666665459E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p5, 5.0000001201E-1f);
+
+#ifdef __VSX__
+static _EIGEN_DECLARE_CONST_Packet2d(1 , 1.0);
+static _EIGEN_DECLARE_CONST_Packet2d(2 , 2.0);
+static _EIGEN_DECLARE_CONST_Packet2d(half, 0.5);
+
+static _EIGEN_DECLARE_CONST_Packet2d(exp_hi,  709.437);
+static _EIGEN_DECLARE_CONST_Packet2d(exp_lo, -709.436139303);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_LOG2EF, 1.4426950408889634073599);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p0, 1.26177193074810590878e-4);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p1, 3.02994407707441961300e-2);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q0, 3.00198505138664455042e-6);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q1, 2.52448340349684104192e-3);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q2, 2.27265548208155028766e-1);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q3, 2.00000000000000000009e0);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C1, 0.693145751953125);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C2, 1.42860682030941723212e-6);
+
+#ifdef __POWER8_VECTOR__
+static Packet2l p2l_1023 = { 1023, 1023 };
+static Packet2ul p2ul_52 = { 52, 52 };
+#endif
+
+#endif
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f plog<Packet4f>(const Packet4f& _x)
+{
+  Packet4f x = _x;
+
+  Packet4i emm0;
+
+  /* isvalid_mask is 0 if x < 0 or x is NaN. */
+  Packet4ui isvalid_mask = reinterpret_cast<Packet4ui>(vec_cmpge(x, p4f_ZERO));
+  Packet4ui iszero_mask = reinterpret_cast<Packet4ui>(vec_cmpeq(x, p4f_ZERO));
+
+  x = pmax(x, p4f_min_norm_pos);  /* cut off denormalized stuff */
+  emm0 = vec_sr(reinterpret_cast<Packet4i>(x),
+                reinterpret_cast<Packet4ui>(p4i_23));
+
+  /* keep only the fractional part */
+  x = pand(x, p4f_inv_mant_mask);
+  x = por(x, p4f_half);
+
+  emm0 = psub(emm0, p4i_0x7f);
+  Packet4f e = padd(vec_ctf(emm0, 0), p4f_1);
+
+  /* part2:
+     if( x < SQRTHF ) {
+       e -= 1;
+       x = x + x - 1.0;
+     } else { x = x - 1.0; }
+  */
+  Packet4f mask = reinterpret_cast<Packet4f>(vec_cmplt(x, p4f_cephes_SQRTHF));
+  Packet4f tmp = pand(x, mask);
+  x = psub(x, p4f_1);
+  e = psub(e, pand(p4f_1, mask));
+  x = padd(x, tmp);
+
+  Packet4f x2 = pmul(x,x);
+  Packet4f x3 = pmul(x2,x);
+
+  Packet4f y, y1, y2;
+  y  = pmadd(p4f_cephes_log_p0, x, p4f_cephes_log_p1);
+  y1 = pmadd(p4f_cephes_log_p3, x, p4f_cephes_log_p4);
+  y2 = pmadd(p4f_cephes_log_p6, x, p4f_cephes_log_p7);
+  y  = pmadd(y , x, p4f_cephes_log_p2);
+  y1 = pmadd(y1, x, p4f_cephes_log_p5);
+  y2 = pmadd(y2, x, p4f_cephes_log_p8);
+  y = pmadd(y, x3, y1);
+  y = pmadd(y, x3, y2);
+  y = pmul(y, x3);
+
+  y1 = pmul(e, p4f_cephes_log_q1);
+  tmp = pmul(x2, p4f_half);
+  y = padd(y, y1);
+  x = psub(x, tmp);
+  y2 = pmul(e, p4f_cephes_log_q2);
+  x = padd(x, y);
+  x = padd(x, y2);
+  // negative arg will be NAN, 0 will be -INF
+  x = vec_sel(x, p4f_minus_inf, iszero_mask);
+  x = vec_sel(p4f_minus_nan, x, isvalid_mask);
+  return x;
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f pexp<Packet4f>(const Packet4f& _x)
+{
+  Packet4f x = _x;
+
+  Packet4f tmp, fx;
+  Packet4i emm0;
+
+  // clamp x
+  x = pmax(pmin(x, p4f_exp_hi), p4f_exp_lo);
+
+  // express exp(x) as exp(g + n*log(2))
+  fx = pmadd(x, p4f_cephes_LOG2EF, p4f_half);
+
+  fx = pfloor(fx);
+
+  tmp = pmul(fx, p4f_cephes_exp_C1);
+  Packet4f z = pmul(fx, p4f_cephes_exp_C2);
+  x = psub(x, tmp);
+  x = psub(x, z);
+
+  z = pmul(x,x);
+
+  Packet4f y = p4f_cephes_exp_p0;
+  y = pmadd(y, x, p4f_cephes_exp_p1);
+  y = pmadd(y, x, p4f_cephes_exp_p2);
+  y = pmadd(y, x, p4f_cephes_exp_p3);
+  y = pmadd(y, x, p4f_cephes_exp_p4);
+  y = pmadd(y, x, p4f_cephes_exp_p5);
+  y = pmadd(y, z, x);
+  y = padd(y, p4f_1);
+
+  // build 2^n
+  emm0 = vec_cts(fx, 0);
+  emm0 = vec_add(emm0, p4i_0x7f);
+  emm0 = vec_sl(emm0, reinterpret_cast<Packet4ui>(p4i_23));
+
+  // Altivec's max & min operators just drop silent NaNs. Check NaNs in 
+  // inputs and return them unmodified.
+  Packet4ui isnumber_mask = reinterpret_cast<Packet4ui>(vec_cmpeq(_x, _x));
+  return vec_sel(_x, pmax(pmul(y, reinterpret_cast<Packet4f>(emm0)), _x),
+                 isnumber_mask);
+}
+
+#ifndef EIGEN_COMP_CLANG
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f prsqrt<Packet4f>(const Packet4f& x)
+{
+  return  vec_rsqrt(x);
+}
+#endif
+
+#ifdef __VSX__
+#ifndef EIGEN_COMP_CLANG
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d prsqrt<Packet2d>(const Packet2d& x)
+{
+  return  vec_rsqrt(x);
+}
+#endif
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f psqrt<Packet4f>(const Packet4f& x)
+{
+  return  vec_sqrt(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d psqrt<Packet2d>(const Packet2d& x)
+{
+  return  vec_sqrt(x);
+}
+
+// VSX support varies between different compilers and even different
+// versions of the same compiler.  For gcc version >= 4.9.3, we can use
+// vec_cts to efficiently convert Packet2d to Packet2l.  Otherwise, use
+// a slow version that works with older compilers. 
+// Update: apparently vec_cts/vec_ctf intrinsics for 64-bit doubles
+// are buggy, https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70963
+static inline Packet2l ConvertToPacket2l(const Packet2d& x) {
+#if EIGEN_GNUC_AT_LEAST(5, 4) || \
+    (EIGEN_GNUC_AT(6, 1) && __GNUC_PATCHLEVEL__ >= 1)
+  return vec_cts(x, 0);    // TODO: check clang version.
+#else
+  double tmp[2];
+  memcpy(tmp, &x, sizeof(tmp));
+  Packet2l l = { static_cast<long long>(tmp[0]),
+                 static_cast<long long>(tmp[1]) };
+  return l;
+#endif
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d pexp<Packet2d>(const Packet2d& _x)
+{
+  Packet2d x = _x;
+
+  Packet2d tmp, fx;
+  Packet2l emm0;
+
+  // clamp x
+  x = pmax(pmin(x, p2d_exp_hi), p2d_exp_lo);
+
+  /* express exp(x) as exp(g + n*log(2)) */
+  fx = pmadd(x, p2d_cephes_LOG2EF, p2d_half);
+
+  fx = pfloor(fx);
+
+  tmp = pmul(fx, p2d_cephes_exp_C1);
+  Packet2d z = pmul(fx, p2d_cephes_exp_C2);
+  x = psub(x, tmp);
+  x = psub(x, z);
+
+  Packet2d x2 = pmul(x,x);
+
+  Packet2d px = p2d_cephes_exp_p0;
+  px = pmadd(px, x2, p2d_cephes_exp_p1);
+  px = pmadd(px, x2, p2d_cephes_exp_p2);
+  px = pmul (px, x);
+
+  Packet2d qx = p2d_cephes_exp_q0;
+  qx = pmadd(qx, x2, p2d_cephes_exp_q1);
+  qx = pmadd(qx, x2, p2d_cephes_exp_q2);
+  qx = pmadd(qx, x2, p2d_cephes_exp_q3);
+
+  x = pdiv(px,psub(qx,px));
+  x = pmadd(p2d_2,x,p2d_1);
+
+  // build 2^n
+  emm0 = ConvertToPacket2l(fx);
+
+#ifdef __POWER8_VECTOR__ 
+  emm0 = vec_add(emm0, p2l_1023);
+  emm0 = vec_sl(emm0, p2ul_52);
+#else
+  // Code is a bit complex for POWER7.  There is actually a
+  // vec_xxsldi intrinsic but it is not supported by some gcc versions.
+  // So we shift (52-32) bits and do a word swap with zeros.
+  _EIGEN_DECLARE_CONST_Packet4i(1023, 1023);
+  _EIGEN_DECLARE_CONST_Packet4i(20, 20);    // 52 - 32
+
+  Packet4i emm04i = reinterpret_cast<Packet4i>(emm0);
+  emm04i = vec_add(emm04i, p4i_1023);
+  emm04i = vec_sl(emm04i, reinterpret_cast<Packet4ui>(p4i_20));
+  static const Packet16uc perm = {
+    0x14, 0x15, 0x16, 0x17, 0x00, 0x01, 0x02, 0x03, 
+    0x1c, 0x1d, 0x1e, 0x1f, 0x08, 0x09, 0x0a, 0x0b };
+#ifdef  _BIG_ENDIAN
+  emm0 = reinterpret_cast<Packet2l>(vec_perm(p4i_ZERO, emm04i, perm));
+#else
+  emm0 = reinterpret_cast<Packet2l>(vec_perm(emm04i, p4i_ZERO, perm));
+#endif
+
+#endif
+
+  // Altivec's max & min operators just drop silent NaNs. Check NaNs in 
+  // inputs and return them unmodified.
+  Packet2ul isnumber_mask = reinterpret_cast<Packet2ul>(vec_cmpeq(_x, _x));
+  return vec_sel(_x, pmax(pmul(x, reinterpret_cast<Packet2d>(emm0)), _x),
+                 isnumber_mask);
+}
+#endif
+
+}  // end namespace internal
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_MATH_FUNCTIONS_ALTIVEC_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/AltiVec/PacketMath.h b/extern/Eigen3/Eigen/src/Core/arch/AltiVec/PacketMath.h
index e4089962dea..08a27d15302 100644
--- a/extern/Eigen3/Eigen/src/Core/arch/AltiVec/PacketMath.h
+++ b/extern/Eigen3/Eigen/src/Core/arch/AltiVec/PacketMath.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Konstantinos Margaritis <markos@codex.gr>
+// Copyright (C) 2008-2016 Konstantinos Margaritis <markos@freevec.org>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -18,13 +18,17 @@ namespace internal {
 #define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 4
 #endif
 
-#ifndef EIGEN_HAS_FUSE_CJMADD
-#define EIGEN_HAS_FUSE_CJMADD 1
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#endif
+
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
 #endif
 
 // NOTE Altivec has 32 registers, but Eigen only accepts a value of 8 or 16
 #ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
-#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 16
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS  32
 #endif
 
 typedef __vector float          Packet4f;
@@ -38,7 +42,7 @@ typedef __vector unsigned char  Packet16uc;
 // and it doesn't really work to declare them global, so we define macros instead
 
 #define _EIGEN_DECLARE_CONST_FAST_Packet4f(NAME,X) \
-  Packet4f p4f_##NAME = (Packet4f) vec_splat_s32(X)
+  Packet4f p4f_##NAME = reinterpret_cast<Packet4f>(vec_splat_s32(X))
 
 #define _EIGEN_DECLARE_CONST_FAST_Packet4i(NAME,X) \
   Packet4i p4i_##NAME = vec_splat_s32(X)
@@ -46,60 +50,158 @@ typedef __vector unsigned char  Packet16uc;
 #define _EIGEN_DECLARE_CONST_Packet4f(NAME,X) \
   Packet4f p4f_##NAME = pset1<Packet4f>(X)
 
-#define _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(NAME,X) \
-  Packet4f p4f_##NAME = vreinterpretq_f32_u32(pset1<int>(X))
-
 #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
   Packet4i p4i_##NAME = pset1<Packet4i>(X)
 
+#define _EIGEN_DECLARE_CONST_Packet2d(NAME,X) \
+  Packet2d p2d_##NAME = pset1<Packet2d>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet2l(NAME,X) \
+  Packet2l p2l_##NAME = pset1<Packet2l>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(NAME,X) \
+  const Packet4f p4f_##NAME = reinterpret_cast<Packet4f>(pset1<Packet4i>(X))
+
 #define DST_CHAN 1
 #define DST_CTRL(size, count, stride) (((size) << 24) | ((count) << 16) | (stride))
 
+
+// These constants are endian-agnostic
+static _EIGEN_DECLARE_CONST_FAST_Packet4f(ZERO, 0); //{ 0.0, 0.0, 0.0, 0.0}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0); //{ 0, 0, 0, 0,}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(ONE,1); //{ 1, 1, 1, 1}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS16,-16); //{ -16, -16, -16, -16}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS1,-1); //{ -1, -1, -1, -1}
+static Packet4f p4f_MZERO = (Packet4f) vec_sl((Packet4ui)p4i_MINUS1, (Packet4ui)p4i_MINUS1); //{ 0x80000000, 0x80000000, 0x80000000, 0x80000000}
+#ifndef __VSX__
+static Packet4f p4f_ONE = vec_ctf(p4i_ONE, 0); //{ 1.0, 1.0, 1.0, 1.0}
+#endif
+
+static Packet4f p4f_COUNTDOWN = { 0.0, 1.0, 2.0, 3.0 };
+static Packet4i p4i_COUNTDOWN = { 0, 1, 2, 3 };
+
+static Packet16uc p16uc_REVERSE32 = { 12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3 };
+static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 };
+
+// Mask alignment
+#ifdef __PPC64__
+#define _EIGEN_MASK_ALIGNMENT	0xfffffffffffffff0
+#else
+#define _EIGEN_MASK_ALIGNMENT	0xfffffff0
+#endif
+
+#define _EIGEN_ALIGNED_PTR(x)	((std::ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT)
+
+// Handle endianness properly while loading constants
 // Define global static constants:
-static Packet4f p4f_COUNTDOWN = { 3.0, 2.0, 1.0, 0.0 };
-static Packet4i p4i_COUNTDOWN = { 3, 2, 1, 0 };
-static Packet16uc p16uc_REVERSE = {12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3};
+#ifdef _BIG_ENDIAN
 static Packet16uc p16uc_FORWARD = vec_lvsl(0, (float*)0);
-static Packet16uc p16uc_DUPLICATE = {0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7};
-
-static _EIGEN_DECLARE_CONST_FAST_Packet4f(ZERO, 0);
-static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0);
-static _EIGEN_DECLARE_CONST_FAST_Packet4i(ONE,1);
-static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS16,-16);
-static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS1,-1);
-static Packet4f p4f_ONE = vec_ctf(p4i_ONE, 0);
-static Packet4f p4f_ZERO_ = (Packet4f) vec_sl((Packet4ui)p4i_MINUS1, (Packet4ui)p4i_MINUS1);
+#ifdef __VSX__
+static Packet16uc p16uc_REVERSE64 = { 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+#endif
+static Packet16uc p16uc_PSET32_WODD   = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 };
+static Packet16uc p16uc_PSET32_WEVEN  = vec_sld(p16uc_DUPLICATE32_HI, (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 };
+static Packet16uc p16uc_HALF64_0_16 = vec_sld((Packet16uc)p4i_ZERO, vec_splat((Packet16uc) vec_abs(p4i_MINUS16), 3), 8);      //{ 0,0,0,0, 0,0,0,0, 16,16,16,16, 16,16,16,16};
+#else
+static Packet16uc p16uc_FORWARD = p16uc_REVERSE32;
+static Packet16uc p16uc_REVERSE64 = { 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+static Packet16uc p16uc_PSET32_WODD = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 1), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 };
+static Packet16uc p16uc_PSET32_WEVEN = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 };
+static Packet16uc p16uc_HALF64_0_16 = vec_sld(vec_splat((Packet16uc) vec_abs(p4i_MINUS16), 0), (Packet16uc)p4i_ZERO, 8);      //{ 0,0,0,0, 0,0,0,0, 16,16,16,16, 16,16,16,16};
+#endif // _BIG_ENDIAN
+
+static Packet16uc p16uc_PSET64_HI = (Packet16uc) vec_mergeh((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };
+static Packet16uc p16uc_PSET64_LO = (Packet16uc) vec_mergel((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 8,9,10,11, 12,13,14,15, 8,9,10,11, 12,13,14,15 };
+static Packet16uc p16uc_TRANSPOSE64_HI = p16uc_PSET64_HI + p16uc_HALF64_0_16;                                         //{ 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23};
+static Packet16uc p16uc_TRANSPOSE64_LO = p16uc_PSET64_LO + p16uc_HALF64_0_16;                                         //{ 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31};
+
+static Packet16uc p16uc_COMPLEX32_REV = vec_sld(p16uc_REVERSE32, p16uc_REVERSE32, 8);                                         //{ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11 };
+
+#ifdef _BIG_ENDIAN
+static Packet16uc p16uc_COMPLEX32_REV2 = vec_sld(p16uc_FORWARD, p16uc_FORWARD, 8);                                            //{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+#else
+static Packet16uc p16uc_COMPLEX32_REV2 = vec_sld(p16uc_PSET64_HI, p16uc_PSET64_LO, 8);                                            //{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+#endif // _BIG_ENDIAN
+
+#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC
+  #define EIGEN_PPC_PREFETCH(ADDR) __builtin_prefetch(ADDR);
+#else
+  #define EIGEN_PPC_PREFETCH(ADDR) asm( "   dcbt [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
+#endif
 
 template<> struct packet_traits<float>  : default_packet_traits
 {
   typedef Packet4f type;
+  typedef Packet4f half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size=4,
-
-    // FIXME check the Has*
+    HasHalfPacket = 1,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasMin  = 1,
+    HasMax  = 1,
+    HasAbs  = 1,
     HasSin  = 0,
     HasCos  = 0,
     HasLog  = 0,
-    HasExp  = 0,
-    HasSqrt = 0
+    HasExp  = 1,
+#ifdef __VSX__
+    HasSqrt = 1,
+#if !EIGEN_COMP_CLANG
+    HasRsqrt = 1,
+#else
+    HasRsqrt = 0,
+#endif
+#else
+    HasSqrt = 0,
+    HasRsqrt = 0,
+#endif
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1,
+    HasNegate = 1,
+    HasBlend = 1
   };
 };
 template<> struct packet_traits<int>    : default_packet_traits
 {
   typedef Packet4i type;
+  typedef Packet4i half;
   enum {
-    // FIXME check the Has*
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=4
+    size = 4,
+    HasHalfPacket = 0,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 0,
+    HasBlend = 1
   };
 };
 
-template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4}; };
-template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4}; };
-/*
+
+template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
+template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
+
+inline std::ostream & operator <<(std::ostream & s, const Packet16uc & v)
+{
+  union {
+    Packet16uc   v;
+    unsigned char n[16];
+  } vt;
+  vt.v = v;
+  for (int i=0; i< 16; i++)
+    s << (int)vt.n[i] << ", ";
+  return s;
+}
+
 inline std::ostream & operator <<(std::ostream & s, const Packet4f & v)
 {
   union {
@@ -133,89 +235,136 @@ inline std::ostream & operator <<(std::ostream & s, const Packet4ui & v)
   return s;
 }
 
-inline std::ostream & operator <<(std::ostream & s, const Packetbi & v)
+// Need to define them first or we get specialization after instantiation errors
+template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from)
 {
-  union {
-    Packet4bi v;
-    unsigned int n[4];
-  } vt;
-  vt.v = v;
-  s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3];
-  return s;
-}
-*/
-template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) {
-  // Taken from http://developer.apple.com/hardwaredrivers/ve/alignment.html
-  float EIGEN_ALIGN16 af[4];
-  af[0] = from;
-  Packet4f vc = vec_ld(0, af);
-  vc = vec_splat(vc, 0);
-  return vc;
+  EIGEN_DEBUG_ALIGNED_LOAD
+#ifdef __VSX__
+  return vec_vsx_ld(0, from);
+#else
+  return vec_ld(0, from);
+#endif
 }
 
-template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)   {
-  int EIGEN_ALIGN16 ai[4];
-  ai[0] = from;
-  Packet4i vc = vec_ld(0, ai);
-  vc = vec_splat(vc, 0);
-  return vc;
+template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from)
+{
+  EIGEN_DEBUG_ALIGNED_LOAD
+#ifdef __VSX__
+  return vec_vsx_ld(0, from);
+#else
+  return vec_ld(0, from);
+#endif
 }
 
-template<> EIGEN_STRONG_INLINE Packet4f plset<float>(const float& a) { return vec_add(pset1<Packet4f>(a), p4f_COUNTDOWN); }
-template<> EIGEN_STRONG_INLINE Packet4i plset<int>(const int& a)     { return vec_add(pset1<Packet4i>(a), p4i_COUNTDOWN); }
-
-template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_add(a,b); }
-template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_add(a,b); }
+template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+#ifdef __VSX__
+  vec_vsx_st(from, 0, to);
+#else
+  vec_st(from, 0, to);
+#endif
+}
 
-template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_sub(a,b); }
-template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_sub(a,b); }
+template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+#ifdef __VSX__
+  vec_vsx_st(from, 0, to);
+#else
+  vec_st(from, 0, to);
+#endif
+}
 
-template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a) { return psub<Packet4f>(p4f_ZERO, a); }
-template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return psub<Packet4i>(p4i_ZERO, a); }
+template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) {
+  Packet4f v = {from, from, from, from};
+  return v;
+}
 
-template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; }
-template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)   {
+  Packet4i v = {from, from, from, from};
+  return v;
+}
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4f>(const float *a,
+                      Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
+{
+  a3 = pload<Packet4f>(a);
+  a0 = vec_splat(a3, 0);
+  a1 = vec_splat(a3, 1);
+  a2 = vec_splat(a3, 2);
+  a3 = vec_splat(a3, 3);
+}
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4i>(const int *a,
+                      Packet4i& a0, Packet4i& a1, Packet4i& a2, Packet4i& a3)
+{
+  a3 = pload<Packet4i>(a);
+  a0 = vec_splat(a3, 0);
+  a1 = vec_splat(a3, 1);
+  a2 = vec_splat(a3, 2);
+  a3 = vec_splat(a3, 3);
+}
 
-template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_madd(a,b,p4f_ZERO); }
-/* Commented out: it's actually slower than processing it scalar
- *
-template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b)
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
 {
-  // Detailed in: http://freevec.org/content/32bit_signed_integer_multiplication_altivec
-  //Set up constants, variables
-  Packet4i a1, b1, bswap, low_prod, high_prod, prod, prod_, v1sel;
+  float EIGEN_ALIGN16 af[4];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  af[2] = from[2*stride];
+  af[3] = from[3*stride];
+ return pload<Packet4f>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  ai[0] = from[0*stride];
+  ai[1] = from[1*stride];
+  ai[2] = from[2*stride];
+  ai[3] = from[3*stride];
+ return pload<Packet4i>(ai);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
+{
+  float EIGEN_ALIGN16 af[4];
+  pstore<float>(af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+  to[2*stride] = af[2];
+  to[3*stride] = af[3];
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  pstore<int>((int *)ai, from);
+  to[0*stride] = ai[0];
+  to[1*stride] = ai[1];
+  to[2*stride] = ai[2];
+  to[3*stride] = ai[3];
+}
 
-  // Get the absolute values
-  a1  = vec_abs(a);
-  b1  = vec_abs(b);
+template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a) { return pset1<Packet4f>(a) + p4f_COUNTDOWN; }
+template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a)   { return pset1<Packet4i>(a) + p4i_COUNTDOWN; }
 
-  // Get the signs using xor
-  Packet4bi sgn = (Packet4bi) vec_cmplt(vec_xor(a, b), p4i_ZERO);
+template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) { return a + b; }
+template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return a + b; }
 
-  // Do the multiplication for the asbolute values.
-  bswap = (Packet4i) vec_rl((Packet4ui) b1, (Packet4ui) p4i_MINUS16 );
-  low_prod = vec_mulo((Packet8i) a1, (Packet8i)b1);
-  high_prod = vec_msum((Packet8i) a1, (Packet8i) bswap, p4i_ZERO);
-  high_prod = (Packet4i) vec_sl((Packet4ui) high_prod, (Packet4ui) p4i_MINUS16);
-  prod = vec_add( low_prod, high_prod );
+template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b) { return a - b; }
+template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return a - b; }
 
-  // NOR the product and select only the negative elements according to the sign mask
-  prod_ = vec_nor(prod, prod);
-  prod_ = vec_sel(p4i_ZERO, prod_, sgn);
+template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a) { return p4f_ZERO - a; }
+template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return p4i_ZERO - a; }
 
-  // Add 1 to the result to get the negative numbers
-  v1sel = vec_sel(p4i_ZERO, p4i_ONE, sgn);
-  prod_ = vec_add(prod_, v1sel);
+template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; }
 
-  // Merge the results back to the final vector.
-  prod = vec_sel(prod, prod_, sgn);
+template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_madd(a,b, p4f_MZERO); }
+template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b) { return a * b; }
 
-  return prod;
-}
-*/
 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b)
 {
-  Packet4f t, y_0, y_1, res;
+#ifndef __VSX__  // VSX actually provides a div instruction
+  Packet4f t, y_0, y_1;
 
   // Altivec does not offer a divide instruction, we have to do a reciprocal approximation
   y_0 = vec_re(b);
@@ -224,8 +373,10 @@ template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const
   t   = vec_nmsub(y_0, b, p4f_ONE);
   y_1 = vec_madd(y_0, t, y_0);
 
-  res = vec_madd(a, y_1, p4f_ZERO);
-  return res;
+  return vec_madd(a, y_1, p4f_MZERO);
+#else
+  return vec_div(a, b);
+#endif
 }
 
 template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& /*a*/, const Packet4i& /*b*/)
@@ -234,16 +385,33 @@ template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& /*a*/, co
 }
 
 // for some weird raisons, it has to be overloaded for packet of integers
-template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return vec_madd(a, b, c); }
-template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return padd(pmul(a,b), c); }
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return vec_madd(a,b,c); }
+template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return a*b + c; }
 
-template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_min(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  #ifdef __VSX__
+  Packet4f ret;
+  __asm__ ("xvcmpgesp %x0,%x1,%x2\n\txxsel %x0,%x1,%x2,%x0" : "=&wa" (ret) : "wa" (a), "wa" (b));
+  return ret;
+  #else
+  return vec_min(a, b);
+  #endif
+}
 template<> EIGEN_STRONG_INLINE Packet4i pmin<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_min(a, b); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_max(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  #ifdef __VSX__
+  Packet4f ret;
+  __asm__ ("xvcmpgtsp %x0,%x2,%x1\n\txxsel %x0,%x1,%x2,%x0" : "=&wa" (ret) : "wa" (a), "wa" (b));
+  return ret;
+  #else
+  return vec_max(a, b);
+  #endif
+}
 template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_max(a, b); }
 
-// Logical Operations are not supported for float, so we have to reinterpret casts using NEON intrinsics
 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_and(a, b); }
 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, b); }
 
@@ -256,13 +424,14 @@ template<> EIGEN_STRONG_INLINE Packet4i pxor<Packet4i>(const Packet4i& a, const
 template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_and(a, vec_nor(b, b)); }
 template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, vec_nor(b, b)); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from) { EIGEN_DEBUG_ALIGNED_LOAD return vec_ld(0, from); }
-template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from) { EIGEN_DEBUG_ALIGNED_LOAD return vec_ld(0, from); }
+template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a) { return vec_round(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const  Packet4f& a) { return vec_ceil(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a) { return vec_floor(a); }
 
+#ifdef _BIG_ENDIAN
 template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
 {
   EIGEN_DEBUG_ALIGNED_LOAD
-  // Taken from http://developer.apple.com/hardwaredrivers/ve/alignment.html
   Packet16uc MSQ, LSQ;
   Packet16uc mask;
   MSQ = vec_ld(0, (unsigned char *)from);          // most significant quadword
@@ -282,25 +451,36 @@ template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
   mask = vec_lvsl(0, from);                        // create the permute mask
   return (Packet4i) vec_perm(MSQ, LSQ, mask);    // align the data
 }
+#else
+// We also need ot redefine little endian loading of Packet4i/Packet4f using VSX
+template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
+{
+  EIGEN_DEBUG_UNALIGNED_LOAD
+  return (Packet4i) vec_vsx_ld((long)from & 15, (const int*) _EIGEN_ALIGNED_PTR(from));
+}
+template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
+{
+  EIGEN_DEBUG_UNALIGNED_LOAD
+  return (Packet4f) vec_vsx_ld((long)from & 15, (const float*) _EIGEN_ALIGNED_PTR(from));
+}
+#endif
 
 template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*   from)
 {
   Packet4f p;
-  if((ptrdiff_t(&from) % 16) == 0)  p = pload<Packet4f>(from);
-  else                              p = ploadu<Packet4f>(from);
-  return vec_perm(p, p, p16uc_DUPLICATE);
+  if((std::ptrdiff_t(from) % 16) == 0)  p = pload<Packet4f>(from);
+  else                                  p = ploadu<Packet4f>(from);
+  return vec_perm(p, p, p16uc_DUPLICATE32_HI);
 }
 template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
 {
   Packet4i p;
-  if((ptrdiff_t(&from) % 16) == 0)  p = pload<Packet4i>(from);
-  else                              p = ploadu<Packet4i>(from);
-  return vec_perm(p, p, p16uc_DUPLICATE);
+  if((std::ptrdiff_t(from) % 16) == 0)  p = pload<Packet4i>(from);
+  else                                  p = ploadu<Packet4i>(from);
+  return vec_perm(p, p, p16uc_DUPLICATE32_HI);
 }
 
-template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE vec_st(from, 0, to); }
-template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE vec_st(from, 0, to); }
-
+#ifdef _BIG_ENDIAN
 template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*  to, const Packet4f& from)
 {
   EIGEN_DEBUG_UNALIGNED_STORE
@@ -337,15 +517,33 @@ template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*      to, const Packet4i& f
   vec_st( LSQ, 15, (unsigned char *)to );                   // Store the LSQ part first
   vec_st( MSQ, 0, (unsigned char *)to );                    // Store the MSQ part
 }
+#else
+// We also need ot redefine little endian loading of Packet4i/Packet4f using VSX
+template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*       to, const Packet4i& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+  vec_vsx_st(from, (long)to & 15, (int*) _EIGEN_ALIGNED_PTR(to));
+}
+template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*   to, const Packet4f& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+  vec_vsx_st(from, (long)to & 15, (float*) _EIGEN_ALIGNED_PTR(to));
+}
+#endif
 
-template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { vec_dstt(addr, DST_CTRL(2,2,32), DST_CHAN); }
-template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*     addr) { vec_dstt(addr, DST_CTRL(2,2,32), DST_CHAN); }
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr)    { EIGEN_PPC_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*     addr)    { EIGEN_PPC_PREFETCH(addr); }
 
-template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x[4]; vec_st(a, 0, x); return x[0]; }
-template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int   EIGEN_ALIGN16 x[4]; vec_st(a, 0, x); return x[0]; }
+template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x; vec_ste(a, 0, &x); return x; }
+template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int   EIGEN_ALIGN16 x; vec_ste(a, 0, &x); return x; }
 
-template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a) { return (Packet4f)vec_perm((Packet16uc)a,(Packet16uc)a, p16uc_REVERSE); }
-template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a) { return (Packet4i)vec_perm((Packet16uc)a,(Packet16uc)a, p16uc_REVERSE); }
+template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a)
+{
+  return reinterpret_cast<Packet4f>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32));
+}
+template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a)
+{
+  return reinterpret_cast<Packet4i>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32)); }
 
 template<> EIGEN_STRONG_INLINE Packet4f pabs(const Packet4f& a) { return vec_abs(a); }
 template<> EIGEN_STRONG_INLINE Packet4i pabs(const Packet4i& a) { return vec_abs(a); }
@@ -353,10 +551,10 @@ template<> EIGEN_STRONG_INLINE Packet4i pabs(const Packet4i& a) { return vec_abs
 template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
 {
   Packet4f b, sum;
-  b   = (Packet4f) vec_sld(a, a, 8);
-  sum = vec_add(a, b);
-  b   = (Packet4f) vec_sld(sum, sum, 4);
-  sum = vec_add(sum, b);
+  b   = vec_sld(a, a, 8);
+  sum = a + b;
+  b   = vec_sld(sum, sum, 4);
+  sum += b;
   return pfirst(sum);
 }
 
@@ -379,11 +577,11 @@ template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
 
   // Now do the summation:
   // Lines 0+1
-  sum[0] = vec_add(sum[0], sum[1]);
+  sum[0] = sum[0] + sum[1];
   // Lines 2+3
-  sum[1] = vec_add(sum[2], sum[3]);
+  sum[1] = sum[2] + sum[3];
   // Add the results
-  sum[0] = vec_add(sum[0], sum[1]);
+  sum[0] = sum[0] + sum[1];
 
   return sum[0];
 }
@@ -392,7 +590,11 @@ template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
 {
   Packet4i sum;
   sum = vec_sums(a, p4i_ZERO);
+#ifdef _BIG_ENDIAN
   sum = vec_sld(sum, p4i_ZERO, 12);
+#else
+  sum = vec_sld(p4i_ZERO, sum, 4);
+#endif
   return pfirst(sum);
 }
 
@@ -415,11 +617,11 @@ template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
 
   // Now do the summation:
   // Lines 0+1
-  sum[0] = vec_add(sum[0], sum[1]);
+  sum[0] = sum[0] + sum[1];
   // Lines 2+3
-  sum[1] = vec_add(sum[2], sum[3]);
+  sum[1] = sum[2] + sum[3];
   // Add the results
-  sum[0] = vec_add(sum[0], sum[1]);
+  sum[0] = sum[0] + sum[1];
 
   return sum[0];
 }
@@ -429,8 +631,8 @@ template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
 template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
 {
   Packet4f prod;
-  prod = pmul(a, (Packet4f)vec_sld(a, a, 8));
-  return pfirst(pmul(prod, (Packet4f)vec_sld(prod, prod, 4)));
+  prod = pmul(a, vec_sld(a, a, 8));
+  return pfirst(pmul(prod, vec_sld(prod, prod, 4)));
 }
 
 template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
@@ -479,8 +681,25 @@ struct palign_impl<Offset,Packet4f>
 {
   static EIGEN_STRONG_INLINE void run(Packet4f& first, const Packet4f& second)
   {
-    if (Offset!=0)
-      first = vec_sld(first, second, Offset*4);
+#ifdef _BIG_ENDIAN
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(first, second, 4); break;
+    case 2:
+      first = vec_sld(first, second, 8); break;
+    case 3:
+      first = vec_sld(first, second, 12); break;
+    }
+#else
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(second, first, 12); break;
+    case 2:
+      first = vec_sld(second, first, 8); break;
+    case 3:
+      first = vec_sld(second, first, 4); break;
+    }
+#endif
   }
 };
 
@@ -489,11 +708,352 @@ struct palign_impl<Offset,Packet4i>
 {
   static EIGEN_STRONG_INLINE void run(Packet4i& first, const Packet4i& second)
   {
-    if (Offset!=0)
-      first = vec_sld(first, second, Offset*4);
+#ifdef _BIG_ENDIAN
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(first, second, 4); break;
+    case 2:
+      first = vec_sld(first, second, 8); break;
+    case 3:
+      first = vec_sld(first, second, 12); break;
+    }
+#else
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(second, first, 12); break;
+    case 2:
+      first = vec_sld(second, first, 8); break;
+    case 3:
+      first = vec_sld(second, first, 4); break;
+    }
+#endif
+  }
+};
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4f,4>& kernel) {
+  Packet4f t0, t1, t2, t3;
+  t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
+  t1 = vec_mergel(kernel.packet[0], kernel.packet[2]);
+  t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]);
+  t3 = vec_mergel(kernel.packet[1], kernel.packet[3]);
+  kernel.packet[0] = vec_mergeh(t0, t2);
+  kernel.packet[1] = vec_mergel(t0, t2);
+  kernel.packet[2] = vec_mergeh(t1, t3);
+  kernel.packet[3] = vec_mergel(t1, t3);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4i,4>& kernel) {
+  Packet4i t0, t1, t2, t3;
+  t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
+  t1 = vec_mergel(kernel.packet[0], kernel.packet[2]);
+  t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]);
+  t3 = vec_mergel(kernel.packet[1], kernel.packet[3]);
+  kernel.packet[0] = vec_mergeh(t0, t2);
+  kernel.packet[1] = vec_mergel(t0, t2);
+  kernel.packet[2] = vec_mergeh(t1, t3);
+  kernel.packet[3] = vec_mergel(t1, t3);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pblend(const Selector<4>& ifPacket, const Packet4i& thenPacket, const Packet4i& elsePacket) {
+  Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] };
+  Packet4ui mask = reinterpret_cast<Packet4ui>(vec_cmpeq(reinterpret_cast<Packet4ui>(select), reinterpret_cast<Packet4ui>(p4i_ONE)));
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) {
+  Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] };
+  Packet4ui mask = reinterpret_cast<Packet4ui>(vec_cmpeq(reinterpret_cast<Packet4ui>(select), reinterpret_cast<Packet4ui>(p4i_ONE)));
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+
+
+//---------- double ----------
+#ifdef __VSX__
+typedef __vector double              Packet2d;
+typedef __vector unsigned long long  Packet2ul;
+typedef __vector long long           Packet2l;
+#if EIGEN_COMP_CLANG
+typedef Packet2ul                    Packet2bl;
+#else
+typedef __vector __bool long         Packet2bl;
+#endif
+
+static Packet2l  p2l_ONE  = { 1, 1 };
+static Packet2l  p2l_ZERO = reinterpret_cast<Packet2l>(p4i_ZERO);
+static Packet2d  p2d_ONE  = { 1.0, 1.0 };
+static Packet2d  p2d_ZERO = reinterpret_cast<Packet2d>(p4f_ZERO);
+static Packet2d  p2d_MZERO = { -0.0, -0.0 };
+
+#ifdef _BIG_ENDIAN
+static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(p2d_ZERO), reinterpret_cast<Packet4f>(p2d_ONE), 8));
+#else
+static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(p2d_ONE), reinterpret_cast<Packet4f>(p2d_ZERO), 8));
+#endif
+
+template<int index> Packet2d vec_splat_dbl(Packet2d& a);
+
+template<> EIGEN_STRONG_INLINE Packet2d vec_splat_dbl<0>(Packet2d& a)
+{
+  return reinterpret_cast<Packet2d>(vec_perm(a, a, p16uc_PSET64_HI));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d vec_splat_dbl<1>(Packet2d& a)
+{
+  return reinterpret_cast<Packet2d>(vec_perm(a, a, p16uc_PSET64_LO));
+}
+
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef Packet2d type;
+  typedef Packet2d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=2,
+    HasHalfPacket = 1,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasMin  = 1,
+    HasMax  = 1,
+    HasAbs  = 1,
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 0,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1,
+    HasNegate = 1,
+    HasBlend = 1
+  };
+};
+
+template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2l & v)
+{
+  union {
+    Packet2l   v;
+    int64_t n[2];
+  } vt;
+  vt.v = v;
+  s << vt.n[0] << ", " << vt.n[1];
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2d & v)
+{
+  union {
+    Packet2d   v;
+    double n[2];
+  } vt;
+  vt.v = v;
+  s << vt.n[0] << ", " << vt.n[1];
+  return s;
+}
+
+// Need to define them first or we get specialization after instantiation errors
+template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from)
+{
+  EIGEN_DEBUG_ALIGNED_LOAD
+#ifdef __VSX__
+  return vec_vsx_ld(0, from);
+#else
+  return vec_ld(0, from);
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<double>(double*   to, const Packet2d& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+#ifdef __VSX__
+  vec_vsx_st(from, 0, to);
+#else
+  vec_st(from, 0, to);
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double&  from) {
+  Packet2d v = {from, from};
+  return v;
+}
+
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet2d>(const double *a,
+                      Packet2d& a0, Packet2d& a1, Packet2d& a2, Packet2d& a3)
+{
+  a1 = pload<Packet2d>(a);
+  a0 = vec_splat_dbl<0>(a1);
+  a1 = vec_splat_dbl<1>(a1);
+  a3 = pload<Packet2d>(a+2);
+  a2 = vec_splat_dbl<0>(a3);
+  a3 = vec_splat_dbl<1>(a3);
+}
+
+template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride)
+{
+  double EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+ return pload<Packet2d>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride)
+{
+  double EIGEN_ALIGN16 af[2];
+  pstore<double>(af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return pset1<Packet2d>(a) + p2d_COUNTDOWN; }
+
+template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return a + b; }
+
+template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return a - b; }
+
+template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return p2d_ZERO - a; }
+
+template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_madd(a,b,p2d_MZERO); }
+template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_div(a,b); }
+
+// for some weird raisons, it has to be overloaded for packet of integers
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vec_madd(a, b, c); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  Packet2d ret;
+  __asm__ ("xvcmpgedp %x0,%x1,%x2\n\txxsel %x0,%x1,%x2,%x0" : "=&wa" (ret) : "wa" (a), "wa" (b));
+  return ret;
+ }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmax<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  Packet2d ret;
+  __asm__ ("xvcmpgtdp %x0,%x2,%x1\n\txxsel %x0,%x1,%x2,%x0" : "=&wa" (ret) : "wa" (a), "wa" (b));
+  return ret;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d por<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_or(a, b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_xor(a, b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, vec_nor(b, b)); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return vec_round(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const  Packet2d& a) { return vec_ceil(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return vec_floor(a); }
+
+template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
+{
+  EIGEN_DEBUG_ALIGNED_LOAD
+  return (Packet2d) vec_vsx_ld((long)from & 15, (const double*) _EIGEN_ALIGNED_PTR(from));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
+{
+  Packet2d p;
+  if((std::ptrdiff_t(from) % 16) == 0)  p = pload<Packet2d>(from);
+  else                                  p = ploadu<Packet2d>(from);
+  return vec_splat_dbl<0>(p);
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double*  to, const Packet2d& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+  vec_vsx_st((Packet4f)from, (long)to & 15, (float*) _EIGEN_ALIGNED_PTR(to));
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_PPC_PREFETCH(addr); }
+
+template<> EIGEN_STRONG_INLINE double  pfirst<Packet2d>(const Packet2d& a) { double EIGEN_ALIGN16 x[2]; pstore<double>(x, a); return x[0]; }
+
+template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a)
+{
+  return reinterpret_cast<Packet2d>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE64));
+}
+template<> EIGEN_STRONG_INLINE Packet2d pabs(const Packet2d& a) { return vec_abs(a); }
+
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
+{
+  Packet2d b, sum;
+  b   = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(a), reinterpret_cast<Packet4f>(a), 8));
+  sum = a + b;
+  return pfirst<Packet2d>(sum);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
+{
+  Packet2d v[2], sum;
+  v[0] = vecs[0] + reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(vecs[0]), reinterpret_cast<Packet4f>(vecs[0]), 8));
+  v[1] = vecs[1] + reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(vecs[1]), reinterpret_cast<Packet4f>(vecs[1]), 8));
+
+#ifdef _BIG_ENDIAN
+  sum = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(v[0]), reinterpret_cast<Packet4f>(v[1]), 8));
+#else
+  sum = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(v[1]), reinterpret_cast<Packet4f>(v[0]), 8));
+#endif
+
+  return sum;
+}
+// Other reduction functions:
+// mul
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmul(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(a), reinterpret_cast<Packet4ui>(a), 8))));
+}
+
+// min
+template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmin(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(a), reinterpret_cast<Packet4ui>(a), 8))));
+}
+
+// max
+template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmax(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(a), reinterpret_cast<Packet4ui>(a), 8))));
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet2d>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2d& first, const Packet2d& second)
+  {
+    if (Offset == 1)
+#ifdef _BIG_ENDIAN
+      first = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(first), reinterpret_cast<Packet4ui>(second), 8));
+#else
+      first = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(second), reinterpret_cast<Packet4ui>(first), 8));
+#endif
   }
 };
 
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2d,2>& kernel) {
+  Packet2d t0, t1;
+  t0 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_HI);
+  t1 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_LO);
+  kernel.packet[0] = t0;
+  kernel.packet[1] = t1;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, const Packet2d& thenPacket, const Packet2d& elsePacket) {
+  Packet2l select = { ifPacket.select[0], ifPacket.select[1] };
+  Packet2bl mask = reinterpret_cast<Packet2bl>( vec_cmpeq(reinterpret_cast<Packet2d>(select), reinterpret_cast<Packet2d>(p2l_ONE)) );
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+#endif // __VSX__
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/arch/CUDA/Complex.h b/extern/Eigen3/Eigen/src/Core/arch/CUDA/Complex.h
new file mode 100644
index 00000000000..9c25365090b
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/CUDA/Complex.h
@@ -0,0 +1,103 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_COMPLEX_CUDA_H
+#define EIGEN_COMPLEX_CUDA_H
+
+// clang-format off
+
+namespace Eigen {
+
+namespace internal {
+
+#if defined(__CUDACC__) && defined(EIGEN_USE_GPU)
+
+// Many std::complex methods such as operator+, operator-, operator* and
+// operator/ are not constexpr. Due to this, clang does not treat them as device
+// functions and thus Eigen functors making use of these operators fail to
+// compile. Here, we manually specialize these functors for complex types when
+// building for CUDA to avoid non-constexpr methods.
+
+// Sum
+template<typename T> struct scalar_sum_op<const std::complex<T>, const std::complex<T> > : binary_op_base<const std::complex<T>, const std::complex<T> > {
+  typedef typename std::complex<T> result_type;
+
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sum_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::complex<T> operator() (const std::complex<T>& a, const std::complex<T>& b) const {
+    return std::complex<T>(numext::real(a) + numext::real(b),
+                           numext::imag(a) + numext::imag(b));
+  }
+};
+
+template<typename T> struct scalar_sum_op<std::complex<T>, std::complex<T> > : scalar_sum_op<const std::complex<T>, const std::complex<T> > {};
+
+
+// Difference
+template<typename T> struct scalar_difference_op<const std::complex<T>, const std::complex<T> >  : binary_op_base<const std::complex<T>, const std::complex<T> > {
+  typedef typename std::complex<T> result_type;
+
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_difference_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::complex<T> operator() (const std::complex<T>& a, const std::complex<T>& b) const {
+    return std::complex<T>(numext::real(a) - numext::real(b),
+                           numext::imag(a) - numext::imag(b));
+  }
+};
+
+template<typename T> struct scalar_difference_op<std::complex<T>, std::complex<T> > : scalar_difference_op<const std::complex<T>, const std::complex<T> > {};
+
+
+// Product
+template<typename T> struct scalar_product_op<const std::complex<T>, const std::complex<T> >  : binary_op_base<const std::complex<T>, const std::complex<T> > {
+  enum {
+    Vectorizable = packet_traits<std::complex<T>>::HasMul
+  };
+  typedef typename std::complex<T> result_type;
+
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_product_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::complex<T> operator() (const std::complex<T>& a, const std::complex<T>& b) const {
+    const T a_real = numext::real(a);
+    const T a_imag = numext::imag(a);
+    const T b_real = numext::real(b);
+    const T b_imag = numext::imag(b);
+    return std::complex<T>(a_real * b_real - a_imag * b_imag,
+                           a_real * b_imag + a_imag * b_real);
+  }
+};
+
+template<typename T> struct scalar_product_op<std::complex<T>, std::complex<T> > : scalar_product_op<const std::complex<T>, const std::complex<T> > {};
+
+
+// Quotient
+template<typename T> struct scalar_quotient_op<const std::complex<T>, const std::complex<T> > : binary_op_base<const std::complex<T>, const std::complex<T> > {
+  enum {
+    Vectorizable = packet_traits<std::complex<T>>::HasDiv
+  };
+  typedef typename std::complex<T> result_type;
+
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_quotient_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::complex<T> operator() (const std::complex<T>& a, const std::complex<T>& b) const {
+    const T a_real = numext::real(a);
+    const T a_imag = numext::imag(a);
+    const T b_real = numext::real(b);
+    const T b_imag = numext::imag(b);
+    const T norm = T(1) / (b_real * b_real + b_imag * b_imag);
+    return std::complex<T>((a_real * b_real + a_imag * b_imag) * norm,
+                           (a_imag * b_real - a_real * b_imag) * norm);
+  }
+};
+
+template<typename T> struct scalar_quotient_op<std::complex<T>, std::complex<T> > : scalar_quotient_op<const std::complex<T>, const std::complex<T> > {};
+
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_COMPLEX_CUDA_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/CUDA/Half.h b/extern/Eigen3/Eigen/src/Core/arch/CUDA/Half.h
new file mode 100644
index 00000000000..755e6209d1e
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/CUDA/Half.h
@@ -0,0 +1,674 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+//
+// The conversion routines are Copyright (c) Fabian Giesen, 2016.
+// The original license follows:
+//
+// Copyright (c) Fabian Giesen, 2016
+// All rights reserved.
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted.
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+
+// Standard 16-bit float type, mostly useful for GPUs. Defines a new
+// type Eigen::half (inheriting from CUDA's __half struct) with
+// operator overloads such that it behaves basically as an arithmetic
+// type. It will be quite slow on CPUs (so it is recommended to stay
+// in float32_bits for CPUs, except for simple parameter conversions, I/O
+// to disk and the likes), but fast on GPUs.
+
+
+#ifndef EIGEN_HALF_CUDA_H
+#define EIGEN_HALF_CUDA_H
+
+#if __cplusplus > 199711L
+#define EIGEN_EXPLICIT_CAST(tgt_type) explicit operator tgt_type()
+#else
+#define EIGEN_EXPLICIT_CAST(tgt_type) operator tgt_type()
+#endif
+
+
+namespace Eigen {
+
+struct half;
+
+namespace half_impl {
+
+#if !defined(EIGEN_HAS_CUDA_FP16)
+// Make our own __half_raw definition that is similar to CUDA's.
+struct __half_raw {
+  EIGEN_DEVICE_FUNC __half_raw() : x(0) {}
+  explicit EIGEN_DEVICE_FUNC __half_raw(unsigned short raw) : x(raw) {}
+  unsigned short x;
+};
+#elif defined(EIGEN_CUDACC_VER) && EIGEN_CUDACC_VER < 90000
+// In CUDA < 9.0, __half is the equivalent of CUDA 9's __half_raw
+typedef __half __half_raw;
+#endif
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw raw_uint16_to_half(unsigned short x);
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw float_to_half_rtne(float ff);
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC float half_to_float(__half_raw h);
+
+struct half_base : public __half_raw {
+  EIGEN_DEVICE_FUNC half_base() {}
+  EIGEN_DEVICE_FUNC half_base(const half_base& h) : __half_raw(h) {}
+  EIGEN_DEVICE_FUNC half_base(const __half_raw& h) : __half_raw(h) {}
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDACC_VER) && EIGEN_CUDACC_VER >= 90000
+  EIGEN_DEVICE_FUNC half_base(const __half& h) : __half_raw(*(__half_raw*)&h) {}
+#endif
+};
+
+} // namespace half_impl
+
+// Class definition.
+struct half : public half_impl::half_base {
+  #if !defined(EIGEN_HAS_CUDA_FP16) || (defined(EIGEN_CUDACC_VER) && EIGEN_CUDACC_VER < 90000)
+    typedef half_impl::__half_raw __half_raw;
+  #endif
+
+  EIGEN_DEVICE_FUNC half() {}
+
+  EIGEN_DEVICE_FUNC half(const __half_raw& h) : half_impl::half_base(h) {}
+  EIGEN_DEVICE_FUNC half(const half& h) : half_impl::half_base(h) {}
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDACC_VER) && EIGEN_CUDACC_VER >= 90000
+  EIGEN_DEVICE_FUNC half(const __half& h) : half_impl::half_base(h) {}
+#endif
+
+  explicit EIGEN_DEVICE_FUNC half(bool b)
+      : half_impl::half_base(half_impl::raw_uint16_to_half(b ? 0x3c00 : 0)) {}
+  template<class T>
+  explicit EIGEN_DEVICE_FUNC half(const T& val)
+      : half_impl::half_base(half_impl::float_to_half_rtne(static_cast<float>(val))) {}
+  explicit EIGEN_DEVICE_FUNC half(float f)
+      : half_impl::half_base(half_impl::float_to_half_rtne(f)) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(bool) const {
+    // +0.0 and -0.0 become false, everything else becomes true.
+    return (x & 0x7fff) != 0;
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(signed char) const {
+    return static_cast<signed char>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned char) const {
+    return static_cast<unsigned char>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(short) const {
+    return static_cast<short>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned short) const {
+    return static_cast<unsigned short>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(int) const {
+    return static_cast<int>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned int) const {
+    return static_cast<unsigned int>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(long) const {
+    return static_cast<long>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned long) const {
+    return static_cast<unsigned long>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(long long) const {
+    return static_cast<long long>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned long long) const {
+    return static_cast<unsigned long long>(half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(float) const {
+    return half_impl::half_to_float(*this);
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(double) const {
+    return static_cast<double>(half_impl::half_to_float(*this));
+  }
+
+  EIGEN_DEVICE_FUNC half& operator=(const half& other) {
+    x = other.x;
+    return *this;
+  }
+};
+
+} // end namespace Eigen
+
+namespace std {
+template<>
+struct numeric_limits<Eigen::half> {
+  static const bool is_specialized = true;
+  static const bool is_signed = true;
+  static const bool is_integer = false;
+  static const bool is_exact = false;
+  static const bool has_infinity = true;
+  static const bool has_quiet_NaN = true;
+  static const bool has_signaling_NaN = true;
+  static const float_denorm_style has_denorm = denorm_present;
+  static const bool has_denorm_loss = false;
+  static const std::float_round_style round_style = std::round_to_nearest;
+  static const bool is_iec559 = false;
+  static const bool is_bounded = false;
+  static const bool is_modulo = false;
+  static const int digits = 11;
+  static const int digits10 = 3;      // according to http://half.sourceforge.net/structstd_1_1numeric__limits_3_01half__float_1_1half_01_4.html
+  static const int max_digits10 = 5;  // according to http://half.sourceforge.net/structstd_1_1numeric__limits_3_01half__float_1_1half_01_4.html
+  static const int radix = 2;
+  static const int min_exponent = -13;
+  static const int min_exponent10 = -4;
+  static const int max_exponent = 16;
+  static const int max_exponent10 = 4;
+  static const bool traps = true;
+  static const bool tinyness_before = false;
+
+  static Eigen::half (min)() { return Eigen::half_impl::raw_uint16_to_half(0x400); }
+  static Eigen::half lowest() { return Eigen::half_impl::raw_uint16_to_half(0xfbff); }
+  static Eigen::half (max)() { return Eigen::half_impl::raw_uint16_to_half(0x7bff); }
+  static Eigen::half epsilon() { return Eigen::half_impl::raw_uint16_to_half(0x0800); }
+  static Eigen::half round_error() { return Eigen::half(0.5); }
+  static Eigen::half infinity() { return Eigen::half_impl::raw_uint16_to_half(0x7c00); }
+  static Eigen::half quiet_NaN() { return Eigen::half_impl::raw_uint16_to_half(0x7e00); }
+  static Eigen::half signaling_NaN() { return Eigen::half_impl::raw_uint16_to_half(0x7e00); }
+  static Eigen::half denorm_min() { return Eigen::half_impl::raw_uint16_to_half(0x1); }
+};
+
+// If std::numeric_limits<T> is specialized, should also specialize
+// std::numeric_limits<const T>, std::numeric_limits<volatile T>, and
+// std::numeric_limits<const volatile T>
+// https://stackoverflow.com/a/16519653/
+template<>
+struct numeric_limits<const Eigen::half> : numeric_limits<Eigen::half> {};
+template<>
+struct numeric_limits<volatile Eigen::half> : numeric_limits<Eigen::half> {};
+template<>
+struct numeric_limits<const volatile Eigen::half> : numeric_limits<Eigen::half> {};
+} // end namespace std
+
+namespace Eigen {
+
+namespace half_impl {
+
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+
+// Intrinsics for native fp16 support. Note that on current hardware,
+// these are no faster than float32_bits arithmetic (you need to use the half2
+// versions to get the ALU speed increased), but you do save the
+// conversion steps back and forth.
+
+EIGEN_STRONG_INLINE __device__ half operator + (const half& a, const half& b) {
+  return __hadd(a, b);
+}
+EIGEN_STRONG_INLINE __device__ half operator * (const half& a, const half& b) {
+  return __hmul(a, b);
+}
+EIGEN_STRONG_INLINE __device__ half operator - (const half& a, const half& b) {
+  return __hsub(a, b);
+}
+EIGEN_STRONG_INLINE __device__ half operator / (const half& a, const half& b) {
+  float num = __half2float(a);
+  float denom = __half2float(b);
+  return __float2half(num / denom);
+}
+EIGEN_STRONG_INLINE __device__ half operator - (const half& a) {
+  return __hneg(a);
+}
+EIGEN_STRONG_INLINE __device__ half& operator += (half& a, const half& b) {
+  a = a + b;
+  return a;
+}
+EIGEN_STRONG_INLINE __device__ half& operator *= (half& a, const half& b) {
+  a = a * b;
+  return a;
+}
+EIGEN_STRONG_INLINE __device__ half& operator -= (half& a, const half& b) {
+  a = a - b;
+  return a;
+}
+EIGEN_STRONG_INLINE __device__ half& operator /= (half& a, const half& b) {
+  a = a / b;
+  return a;
+}
+EIGEN_STRONG_INLINE __device__ bool operator == (const half& a, const half& b) {
+  return __heq(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator != (const half& a, const half& b) {
+  return __hne(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator < (const half& a, const half& b) {
+  return __hlt(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator <= (const half& a, const half& b) {
+  return __hle(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator > (const half& a, const half& b) {
+  return __hgt(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator >= (const half& a, const half& b) {
+  return __hge(a, b);
+}
+
+#else  // Emulate support for half floats
+
+// Definitions for CPUs and older CUDA, mostly working through conversion
+// to/from float32_bits.
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator + (const half& a, const half& b) {
+  return half(float(a) + float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator * (const half& a, const half& b) {
+  return half(float(a) * float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a, const half& b) {
+  return half(float(a) - float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator / (const half& a, const half& b) {
+  return half(float(a) / float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a) {
+  half result;
+  result.x = a.x ^ 0x8000;
+  return result;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator += (half& a, const half& b) {
+  a = half(float(a) + float(b));
+  return a;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator *= (half& a, const half& b) {
+  a = half(float(a) * float(b));
+  return a;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator -= (half& a, const half& b) {
+  a = half(float(a) - float(b));
+  return a;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator /= (half& a, const half& b) {
+  a = half(float(a) / float(b));
+  return a;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator == (const half& a, const half& b) {
+  return numext::equal_strict(float(a),float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator != (const half& a, const half& b) {
+  return numext::not_equal_strict(float(a), float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator < (const half& a, const half& b) {
+  return float(a) < float(b);
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator <= (const half& a, const half& b) {
+  return float(a) <= float(b);
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator > (const half& a, const half& b) {
+  return float(a) > float(b);
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator >= (const half& a, const half& b) {
+  return float(a) >= float(b);
+}
+
+#endif  // Emulate support for half floats
+
+// Division by an index. Do it in full float precision to avoid accuracy
+// issues in converting the denominator to half.
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator / (const half& a, Index b) {
+  return half(static_cast<float>(a) / static_cast<float>(b));
+}
+
+// Conversion routines, including fallbacks for the host or older CUDA.
+// Note that newer Intel CPUs (Haswell or newer) have vectorized versions of
+// these in hardware. If we need more performance on older/other CPUs, they are
+// also possible to vectorize directly.
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw raw_uint16_to_half(unsigned short x) {
+  __half_raw h;
+  h.x = x;
+  return h;
+}
+
+union float32_bits {
+  unsigned int u;
+  float f;
+};
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw float_to_half_rtne(float ff) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300
+  __half tmp_ff = __float2half(ff);
+  return *(__half_raw*)&tmp_ff;
+
+#elif defined(EIGEN_HAS_FP16_C)
+  __half_raw h;
+  h.x = _cvtss_sh(ff, 0);
+  return h;
+
+#else
+  float32_bits f; f.f = ff;
+
+  const float32_bits f32infty = { 255 << 23 };
+  const float32_bits f16max = { (127 + 16) << 23 };
+  const float32_bits denorm_magic = { ((127 - 15) + (23 - 10) + 1) << 23 };
+  unsigned int sign_mask = 0x80000000u;
+  __half_raw o;
+  o.x = static_cast<unsigned short>(0x0u);
+
+  unsigned int sign = f.u & sign_mask;
+  f.u ^= sign;
+
+  // NOTE all the integer compares in this function can be safely
+  // compiled into signed compares since all operands are below
+  // 0x80000000. Important if you want fast straight SSE2 code
+  // (since there's no unsigned PCMPGTD).
+
+  if (f.u >= f16max.u) {  // result is Inf or NaN (all exponent bits set)
+    o.x = (f.u > f32infty.u) ? 0x7e00 : 0x7c00; // NaN->qNaN and Inf->Inf
+  } else {  // (De)normalized number or zero
+    if (f.u < (113 << 23)) {  // resulting FP16 is subnormal or zero
+      // use a magic value to align our 10 mantissa bits at the bottom of
+      // the float. as long as FP addition is round-to-nearest-even this
+      // just works.
+      f.f += denorm_magic.f;
+
+      // and one integer subtract of the bias later, we have our final float!
+      o.x = static_cast<unsigned short>(f.u - denorm_magic.u);
+    } else {
+      unsigned int mant_odd = (f.u >> 13) & 1; // resulting mantissa is odd
+
+      // update exponent, rounding bias part 1
+      f.u += ((unsigned int)(15 - 127) << 23) + 0xfff;
+      // rounding bias part 2
+      f.u += mant_odd;
+      // take the bits!
+      o.x = static_cast<unsigned short>(f.u >> 13);
+    }
+  }
+
+  o.x |= static_cast<unsigned short>(sign >> 16);
+  return o;
+#endif
+}
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC float half_to_float(__half_raw h) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300
+  return __half2float(h);
+
+#elif defined(EIGEN_HAS_FP16_C)
+  return _cvtsh_ss(h.x);
+
+#else
+  const float32_bits magic = { 113 << 23 };
+  const unsigned int shifted_exp = 0x7c00 << 13; // exponent mask after shift
+  float32_bits o;
+
+  o.u = (h.x & 0x7fff) << 13;             // exponent/mantissa bits
+  unsigned int exp = shifted_exp & o.u;   // just the exponent
+  o.u += (127 - 15) << 23;                // exponent adjust
+
+  // handle exponent special cases
+  if (exp == shifted_exp) {     // Inf/NaN?
+    o.u += (128 - 16) << 23;    // extra exp adjust
+  } else if (exp == 0) {        // Zero/Denormal?
+    o.u += 1 << 23;             // extra exp adjust
+    o.f -= magic.f;             // renormalize
+  }
+
+  o.u |= (h.x & 0x8000) << 16;    // sign bit
+  return o.f;
+#endif
+}
+
+// --- standard functions ---
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isinf)(const half& a) {
+  return (a.x & 0x7fff) == 0x7c00;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isnan)(const half& a) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return __hisnan(a);
+#else
+  return (a.x & 0x7fff) > 0x7c00;
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isfinite)(const half& a) {
+  return !(isinf EIGEN_NOT_A_MACRO (a)) && !(isnan EIGEN_NOT_A_MACRO (a));
+}
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half abs(const half& a) {
+  half result;
+  result.x = a.x & 0x7FFF;
+  return result;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half exp(const half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
+  return half(hexp(a));
+#else
+   return half(::expf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log(const half& a) {
+#if defined(EIGEN_HAS_CUDA_FP16) && EIGEN_CUDACC_VER >= 80000 && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return half(::hlog(a));
+#else
+  return half(::logf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log1p(const half& a) {
+  return half(numext::log1p(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log10(const half& a) {
+  return half(::log10f(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half sqrt(const half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
+  return half(hsqrt(a));
+#else
+    return half(::sqrtf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half pow(const half& a, const half& b) {
+  return half(::powf(float(a), float(b)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half sin(const half& a) {
+  return half(::sinf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half cos(const half& a) {
+  return half(::cosf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half tan(const half& a) {
+  return half(::tanf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half tanh(const half& a) {
+  return half(::tanhf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half floor(const half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300
+  return half(hfloor(a));
+#else
+  return half(::floorf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half ceil(const half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300
+  return half(hceil(a));
+#else
+  return half(::ceilf(float(a)));
+#endif
+}
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half (min)(const half& a, const half& b) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return __hlt(b, a) ? b : a;
+#else
+  const float f1 = static_cast<float>(a);
+  const float f2 = static_cast<float>(b);
+  return f2 < f1 ? b : a;
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half (max)(const half& a, const half& b) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return __hlt(a, b) ? b : a;
+#else
+  const float f1 = static_cast<float>(a);
+  const float f2 = static_cast<float>(b);
+  return f1 < f2 ? b : a;
+#endif
+}
+
+EIGEN_ALWAYS_INLINE std::ostream& operator << (std::ostream& os, const half& v) {
+  os << static_cast<float>(v);
+  return os;
+}
+
+} // end namespace half_impl
+
+// import Eigen::half_impl::half into Eigen namespace
+// using half_impl::half;
+
+namespace internal {
+
+template<>
+struct random_default_impl<half, false, false>
+{
+  static inline half run(const half& x, const half& y)
+  {
+    return x + (y-x) * half(float(std::rand()) / float(RAND_MAX));
+  }
+  static inline half run()
+  {
+    return run(half(-1.f), half(1.f));
+  }
+};
+
+template<> struct is_arithmetic<half> { enum { value = true }; };
+
+} // end namespace internal
+
+template<> struct NumTraits<Eigen::half>
+    : GenericNumTraits<Eigen::half>
+{
+  enum {
+    IsSigned = true,
+    IsInteger = false,
+    IsComplex = false,
+    RequireInitialization = false
+  };
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half epsilon() {
+    return half_impl::raw_uint16_to_half(0x0800);
+  }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half dummy_precision() { return Eigen::half(1e-2f); }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half highest() {
+    return half_impl::raw_uint16_to_half(0x7bff);
+  }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half lowest() {
+    return half_impl::raw_uint16_to_half(0xfbff);
+  }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half infinity() {
+    return half_impl::raw_uint16_to_half(0x7c00);
+  }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half quiet_NaN() {
+    return half_impl::raw_uint16_to_half(0x7c01);
+  }
+};
+
+} // end namespace Eigen
+
+// C-like standard mathematical functions and trancendentals.
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half fabsh(const Eigen::half& a) {
+  Eigen::half result;
+  result.x = a.x & 0x7FFF;
+  return result;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half exph(const Eigen::half& a) {
+  return Eigen::half(::expf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half logh(const Eigen::half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return Eigen::half(::hlog(a));
+#else
+  return Eigen::half(::logf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half sqrth(const Eigen::half& a) {
+  return Eigen::half(::sqrtf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half powh(const Eigen::half& a, const Eigen::half& b) {
+  return Eigen::half(::powf(float(a), float(b)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half floorh(const Eigen::half& a) {
+  return Eigen::half(::floorf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half ceilh(const Eigen::half& a) {
+  return Eigen::half(::ceilf(float(a)));
+}
+
+namespace std {
+
+#if __cplusplus > 199711L
+template <>
+struct hash<Eigen::half> {
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::size_t operator()(const Eigen::half& a) const {
+    return static_cast<std::size_t>(a.x);
+  }
+};
+#endif
+
+} // end namespace std
+
+
+// Add the missing shfl_xor intrinsic
+#if defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300
+__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_xor(Eigen::half var, int laneMask, int width=warpSize) {
+  #if EIGEN_CUDACC_VER < 90000
+  return static_cast<Eigen::half>(__shfl_xor(static_cast<float>(var), laneMask, width));
+  #else
+  return static_cast<Eigen::half>(__shfl_xor_sync(0xFFFFFFFF, static_cast<float>(var), laneMask, width));
+  #endif
+}
+#endif
+
+// ldg() has an overload for __half_raw, but we also need one for Eigen::half.
+#if defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 350
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half __ldg(const Eigen::half* ptr) {
+  return Eigen::half_impl::raw_uint16_to_half(
+      __ldg(reinterpret_cast<const unsigned short*>(ptr)));
+}
+#endif
+
+
+#if defined(EIGEN_CUDA_ARCH)
+namespace Eigen {
+namespace numext {
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+bool (isnan)(const Eigen::half& h) {
+  return (half_impl::isnan)(h);
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+bool (isinf)(const Eigen::half& h) {
+  return (half_impl::isinf)(h);
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+bool (isfinite)(const Eigen::half& h) {
+  return (half_impl::isfinite)(h);
+}
+
+} // namespace Eigen
+}  // namespace numext
+#endif
+
+#endif // EIGEN_HALF_CUDA_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/CUDA/MathFunctions.h b/extern/Eigen3/Eigen/src/Core/arch/CUDA/MathFunctions.h
new file mode 100644
index 00000000000..0348b41db05
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/CUDA/MathFunctions.h
@@ -0,0 +1,91 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_MATH_FUNCTIONS_CUDA_H
+#define EIGEN_MATH_FUNCTIONS_CUDA_H
+
+namespace Eigen {
+
+namespace internal {
+
+// Make sure this is only available when targeting a GPU: we don't want to
+// introduce conflicts between these packet_traits definitions and the ones
+// we'll use on the host side (SSE, AVX, ...)
+#if defined(__CUDACC__) && defined(EIGEN_USE_GPU)
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 plog<float4>(const float4& a)
+{
+  return make_float4(logf(a.x), logf(a.y), logf(a.z), logf(a.w));
+}
+
+template<>  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 plog<double2>(const double2& a)
+{
+  using ::log;
+  return make_double2(log(a.x), log(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 plog1p<float4>(const float4& a)
+{
+  return make_float4(log1pf(a.x), log1pf(a.y), log1pf(a.z), log1pf(a.w));
+}
+
+template<>  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 plog1p<double2>(const double2& a)
+{
+  return make_double2(log1p(a.x), log1p(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 pexp<float4>(const float4& a)
+{
+  return make_float4(expf(a.x), expf(a.y), expf(a.z), expf(a.w));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 pexp<double2>(const double2& a)
+{
+  using ::exp;
+  return make_double2(exp(a.x), exp(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 psqrt<float4>(const float4& a)
+{
+  return make_float4(sqrtf(a.x), sqrtf(a.y), sqrtf(a.z), sqrtf(a.w));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 psqrt<double2>(const double2& a)
+{
+  using ::sqrt;
+  return make_double2(sqrt(a.x), sqrt(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 prsqrt<float4>(const float4& a)
+{
+  return make_float4(rsqrtf(a.x), rsqrtf(a.y), rsqrtf(a.z), rsqrtf(a.w));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 prsqrt<double2>(const double2& a)
+{
+  return make_double2(rsqrt(a.x), rsqrt(a.y));
+}
+
+
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_MATH_FUNCTIONS_CUDA_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/CUDA/PacketMath.h b/extern/Eigen3/Eigen/src/Core/arch/CUDA/PacketMath.h
new file mode 100644
index 00000000000..4dda63188d2
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/CUDA/PacketMath.h
@@ -0,0 +1,333 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_CUDA_H
+#define EIGEN_PACKET_MATH_CUDA_H
+
+namespace Eigen {
+
+namespace internal {
+
+// Make sure this is only available when targeting a GPU: we don't want to
+// introduce conflicts between these packet_traits definitions and the ones
+// we'll use on the host side (SSE, AVX, ...)
+#if defined(__CUDACC__) && defined(EIGEN_USE_GPU)
+template<> struct is_arithmetic<float4>  { enum { value = true }; };
+template<> struct is_arithmetic<double2> { enum { value = true }; };
+
+template<> struct packet_traits<float> : default_packet_traits
+{
+  typedef float4 type;
+  typedef float4 half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=4,
+    HasHalfPacket = 0,
+
+    HasDiv  = 1,
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 1,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasLGamma = 1,
+    HasDiGamma = 1,
+    HasZeta = 1,
+    HasPolygamma = 1,
+    HasErf = 1,
+    HasErfc = 1,
+    HasIGamma = 1,
+    HasIGammac = 1,
+    HasBetaInc = 1,
+
+    HasBlend = 0,
+  };
+};
+
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef double2 type;
+  typedef double2 half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=2,
+    HasHalfPacket = 0,
+
+    HasDiv  = 1,
+    HasLog  = 1,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasLGamma = 1,
+    HasDiGamma = 1,
+    HasZeta = 1,
+    HasPolygamma = 1,
+    HasErf = 1,
+    HasErfc = 1,
+    HasIGamma = 1,
+    HasIGammac = 1,
+    HasBetaInc = 1,
+
+    HasBlend = 0,
+  };
+};
+
+
+template<> struct unpacket_traits<float4>  { typedef float  type; enum {size=4, alignment=Aligned16}; typedef float4 half; };
+template<> struct unpacket_traits<double2> { typedef double type; enum {size=2, alignment=Aligned16}; typedef double2 half; };
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pset1<float4>(const float&  from) {
+  return make_float4(from, from, from, from);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pset1<double2>(const double& from) {
+  return make_double2(from, from);
+}
+
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 plset<float4>(const float& a) {
+  return make_float4(a, a+1, a+2, a+3);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 plset<double2>(const double& a) {
+  return make_double2(a, a+1);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 padd<float4>(const float4& a, const float4& b) {
+  return make_float4(a.x+b.x, a.y+b.y, a.z+b.z, a.w+b.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 padd<double2>(const double2& a, const double2& b) {
+  return make_double2(a.x+b.x, a.y+b.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 psub<float4>(const float4& a, const float4& b) {
+  return make_float4(a.x-b.x, a.y-b.y, a.z-b.z, a.w-b.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 psub<double2>(const double2& a, const double2& b) {
+  return make_double2(a.x-b.x, a.y-b.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pnegate(const float4& a) {
+  return make_float4(-a.x, -a.y, -a.z, -a.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pnegate(const double2& a) {
+  return make_double2(-a.x, -a.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pconj(const float4& a) { return a; }
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pconj(const double2& a) { return a; }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pmul<float4>(const float4& a, const float4& b) {
+  return make_float4(a.x*b.x, a.y*b.y, a.z*b.z, a.w*b.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pmul<double2>(const double2& a, const double2& b) {
+  return make_double2(a.x*b.x, a.y*b.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pdiv<float4>(const float4& a, const float4& b) {
+  return make_float4(a.x/b.x, a.y/b.y, a.z/b.z, a.w/b.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pdiv<double2>(const double2& a, const double2& b) {
+  return make_double2(a.x/b.x, a.y/b.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pmin<float4>(const float4& a, const float4& b) {
+  return make_float4(fminf(a.x, b.x), fminf(a.y, b.y), fminf(a.z, b.z), fminf(a.w, b.w));
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pmin<double2>(const double2& a, const double2& b) {
+  return make_double2(fmin(a.x, b.x), fmin(a.y, b.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pmax<float4>(const float4& a, const float4& b) {
+  return make_float4(fmaxf(a.x, b.x), fmaxf(a.y, b.y), fmaxf(a.z, b.z), fmaxf(a.w, b.w));
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pmax<double2>(const double2& a, const double2& b) {
+  return make_double2(fmax(a.x, b.x), fmax(a.y, b.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pload<float4>(const float* from) {
+  return *reinterpret_cast<const float4*>(from);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pload<double2>(const double* from) {
+  return *reinterpret_cast<const double2*>(from);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 ploadu<float4>(const float* from) {
+  return make_float4(from[0], from[1], from[2], from[3]);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 ploadu<double2>(const double* from) {
+  return make_double2(from[0], from[1]);
+}
+
+template<> EIGEN_STRONG_INLINE float4 ploaddup<float4>(const float*   from) {
+  return make_float4(from[0], from[0], from[1], from[1]);
+}
+template<> EIGEN_STRONG_INLINE double2 ploaddup<double2>(const double*  from) {
+  return make_double2(from[0], from[0]);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstore<float>(float*   to, const float4& from) {
+  *reinterpret_cast<float4*>(to) = from;
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstore<double>(double* to, const double2& from) {
+  *reinterpret_cast<double2*>(to) = from;
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstoreu<float>(float*  to, const float4& from) {
+  to[0] = from.x;
+  to[1] = from.y;
+  to[2] = from.z;
+  to[3] = from.w;
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const double2& from) {
+  to[0] = from.x;
+  to[1] = from.y;
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE float4 ploadt_ro<float4, Aligned>(const float* from) {
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
+  return __ldg((const float4*)from);
+#else
+  return make_float4(from[0], from[1], from[2], from[3]);
+#endif
+}
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE double2 ploadt_ro<double2, Aligned>(const double* from) {
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
+  return __ldg((const double2*)from);
+#else
+  return make_double2(from[0], from[1]);
+#endif
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE float4 ploadt_ro<float4, Unaligned>(const float* from) {
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
+  return make_float4(__ldg(from+0), __ldg(from+1), __ldg(from+2), __ldg(from+3));
+#else
+  return make_float4(from[0], from[1], from[2], from[3]);
+#endif
+}
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE double2 ploadt_ro<double2, Unaligned>(const double* from) {
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
+  return make_double2(__ldg(from+0), __ldg(from+1));
+#else
+  return make_double2(from[0], from[1]);
+#endif
+}
+
+template<> EIGEN_DEVICE_FUNC inline float4 pgather<float, float4>(const float* from, Index stride) {
+  return make_float4(from[0*stride], from[1*stride], from[2*stride], from[3*stride]);
+}
+
+template<> EIGEN_DEVICE_FUNC inline double2 pgather<double, double2>(const double* from, Index stride) {
+  return make_double2(from[0*stride], from[1*stride]);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, float4>(float* to, const float4& from, Index stride) {
+  to[stride*0] = from.x;
+  to[stride*1] = from.y;
+  to[stride*2] = from.z;
+  to[stride*3] = from.w;
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, double2>(double* to, const double2& from, Index stride) {
+  to[stride*0] = from.x;
+  to[stride*1] = from.y;
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  pfirst<float4>(const float4& a) {
+  return a.x;
+}
+template<> EIGEN_DEVICE_FUNC inline double pfirst<double2>(const double2& a) {
+  return a.x;
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  predux<float4>(const float4& a) {
+  return a.x + a.y + a.z + a.w;
+}
+template<> EIGEN_DEVICE_FUNC inline double predux<double2>(const double2& a) {
+  return a.x + a.y;
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  predux_max<float4>(const float4& a) {
+  return fmaxf(fmaxf(a.x, a.y), fmaxf(a.z, a.w));
+}
+template<> EIGEN_DEVICE_FUNC inline double predux_max<double2>(const double2& a) {
+  return fmax(a.x, a.y);
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  predux_min<float4>(const float4& a) {
+  return fminf(fminf(a.x, a.y), fminf(a.z, a.w));
+}
+template<> EIGEN_DEVICE_FUNC inline double predux_min<double2>(const double2& a) {
+  return fmin(a.x, a.y);
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  predux_mul<float4>(const float4& a) {
+  return a.x * a.y * a.z * a.w;
+}
+template<> EIGEN_DEVICE_FUNC inline double predux_mul<double2>(const double2& a) {
+  return a.x * a.y;
+}
+
+template<> EIGEN_DEVICE_FUNC inline float4  pabs<float4>(const float4& a) {
+  return make_float4(fabsf(a.x), fabsf(a.y), fabsf(a.z), fabsf(a.w));
+}
+template<> EIGEN_DEVICE_FUNC inline double2 pabs<double2>(const double2& a) {
+  return make_double2(fabs(a.x), fabs(a.y));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<float4,4>& kernel) {
+  float tmp = kernel.packet[0].y;
+  kernel.packet[0].y = kernel.packet[1].x;
+  kernel.packet[1].x = tmp;
+
+  tmp = kernel.packet[0].z;
+  kernel.packet[0].z = kernel.packet[2].x;
+  kernel.packet[2].x = tmp;
+
+  tmp = kernel.packet[0].w;
+  kernel.packet[0].w = kernel.packet[3].x;
+  kernel.packet[3].x = tmp;
+
+  tmp = kernel.packet[1].z;
+  kernel.packet[1].z = kernel.packet[2].y;
+  kernel.packet[2].y = tmp;
+
+  tmp = kernel.packet[1].w;
+  kernel.packet[1].w = kernel.packet[3].y;
+  kernel.packet[3].y = tmp;
+
+  tmp = kernel.packet[2].w;
+  kernel.packet[2].w = kernel.packet[3].z;
+  kernel.packet[3].z = tmp;
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<double2,2>& kernel) {
+  double tmp = kernel.packet[0].y;
+  kernel.packet[0].y = kernel.packet[1].x;
+  kernel.packet[1].x = tmp;
+}
+
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+
+#endif // EIGEN_PACKET_MATH_CUDA_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/CUDA/PacketMathHalf.h b/extern/Eigen3/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
new file mode 100644
index 00000000000..c66d38469f8
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
@@ -0,0 +1,1124 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_HALF_CUDA_H
+#define EIGEN_PACKET_MATH_HALF_CUDA_H
+
+
+namespace Eigen {
+namespace internal {
+
+// Most of the following operations require arch >= 3.0
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDACC__) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+
+template<> struct is_arithmetic<half2> { enum { value = true }; };
+
+template<> struct packet_traits<Eigen::half> : default_packet_traits
+{
+  typedef half2 type;
+  typedef half2 half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=2,
+    HasHalfPacket = 0,
+    HasAdd    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasSqrt   = 1,
+    HasRsqrt  = 1,
+    HasExp    = 1,
+    HasLog    = 1,
+    HasLog1p  = 1
+  };
+};
+
+template<> struct unpacket_traits<half2> { typedef Eigen::half type; enum {size=2, alignment=Aligned16}; typedef half2 half; };
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pset1<half2>(const Eigen::half& from) {
+  return __half2half2(from);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pload<half2>(const Eigen::half* from) {
+  return *reinterpret_cast<const half2*>(from);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 ploadu<half2>(const Eigen::half* from) {
+  return __halves2half2(from[0], from[1]);
+}
+
+template<> EIGEN_STRONG_INLINE half2 ploaddup<half2>(const Eigen::half*  from) {
+  return __halves2half2(from[0], from[0]);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE void pstore<Eigen::half>(Eigen::half* to, const half2& from) {
+  *reinterpret_cast<half2*>(to) = from;
+}
+
+template<> __device__ EIGEN_STRONG_INLINE void pstoreu<Eigen::half>(Eigen::half* to, const half2& from) {
+  to[0] = __low2half(from);
+  to[1] = __high2half(from);
+}
+
+template<>
+ __device__ EIGEN_ALWAYS_INLINE half2 ploadt_ro<half2, Aligned>(const Eigen::half* from) {
+#if __CUDA_ARCH__ >= 350
+   return __ldg((const half2*)from);
+#else
+  return __halves2half2(*(from+0), *(from+1));
+#endif
+}
+
+template<>
+__device__ EIGEN_ALWAYS_INLINE half2 ploadt_ro<half2, Unaligned>(const Eigen::half* from) {
+#if __CUDA_ARCH__ >= 350
+   return __halves2half2(__ldg(from+0), __ldg(from+1));
+#else
+  return __halves2half2(*(from+0), *(from+1));
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pgather<Eigen::half, half2>(const Eigen::half* from, Index stride) {
+  return __halves2half2(from[0*stride], from[1*stride]);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE void pscatter<Eigen::half, half2>(Eigen::half* to, const half2& from, Index stride) {
+  to[stride*0] = __low2half(from);
+  to[stride*1] = __high2half(from);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half pfirst<half2>(const half2& a) {
+  return __low2half(a);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pabs<half2>(const half2& a) {
+  half2 result;
+  unsigned temp = *(reinterpret_cast<const unsigned*>(&(a)));
+  *(reinterpret_cast<unsigned*>(&(result))) = temp & 0x7FFF7FFF;
+  return result;
+}
+
+
+__device__ EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<half2,2>& kernel) {
+  __half a1 = __low2half(kernel.packet[0]);
+  __half a2 = __high2half(kernel.packet[0]);
+  __half b1 = __low2half(kernel.packet[1]);
+  __half b2 = __high2half(kernel.packet[1]);
+  kernel.packet[0] = __halves2half2(a1, b1);
+  kernel.packet[1] = __halves2half2(a2, b2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 plset<half2>(const Eigen::half& a) {
+#if __CUDA_ARCH__ >= 530
+  return __halves2half2(a, __hadd(a, __float2half(1.0f)));
+#else
+  float f = __half2float(a) + 1.0f;
+  return __halves2half2(a, __float2half(f));
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 padd<half2>(const half2& a, const half2& b) {
+#if __CUDA_ARCH__ >= 530
+  return __hadd2(a, b);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float r1 = a1 + b1;
+  float r2 = a2 + b2;
+  return __floats2half2_rn(r1, r2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 psub<half2>(const half2& a, const half2& b) {
+#if __CUDA_ARCH__ >= 530
+  return __hsub2(a, b);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float r1 = a1 - b1;
+  float r2 = a2 - b2;
+  return __floats2half2_rn(r1, r2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pnegate(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  return __hneg2(a);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return __floats2half2_rn(-a1, -a2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pconj(const half2& a) { return a; }
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pmul<half2>(const half2& a, const half2& b) {
+#if __CUDA_ARCH__ >= 530
+  return __hmul2(a, b);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float r1 = a1 * b1;
+  float r2 = a2 * b2;
+  return __floats2half2_rn(r1, r2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pmadd<half2>(const half2& a, const half2& b, const half2& c) {
+#if __CUDA_ARCH__ >= 530
+   return __hfma2(a, b, c);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float c1 = __low2float(c);
+  float c2 = __high2float(c);
+  float r1 = a1 * b1 + c1;
+  float r2 = a2 * b2 + c2;
+  return __floats2half2_rn(r1, r2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pdiv<half2>(const half2& a, const half2& b) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float r1 = a1 / b1;
+  float r2 = a2 / b2;
+  return __floats2half2_rn(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pmin<half2>(const half2& a, const half2& b) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  __half r1 = a1 < b1 ? __low2half(a) : __low2half(b);
+  __half r2 = a2 < b2 ? __high2half(a) : __high2half(b);
+  return __halves2half2(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pmax<half2>(const half2& a, const half2& b) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  __half r1 = a1 > b1 ? __low2half(a) : __low2half(b);
+  __half r2 = a2 > b2 ? __high2half(a) : __high2half(b);
+  return __halves2half2(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half predux<half2>(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  return __hadd(__low2half(a), __high2half(a));
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return Eigen::half(half_impl::raw_uint16_to_half(__float2half_rn(a1 + a2)));
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half predux_max<half2>(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  __half first = __low2half(a);
+  __half second = __high2half(a);
+  return __hgt(first, second) ? first : second;
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return a1 > a2 ? __low2half(a) : __high2half(a);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half predux_min<half2>(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  __half first = __low2half(a);
+  __half second = __high2half(a);
+  return __hlt(first, second) ? first : second;
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return a1 < a2 ? __low2half(a) : __high2half(a);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half predux_mul<half2>(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  return __hmul(__low2half(a), __high2half(a));
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return Eigen::half(half_impl::raw_uint16_to_half(__float2half_rn(a1 * a2)));
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 plog1p<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = log1pf(a1);
+  float r2 = log1pf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
+
+template<>  __device__ EIGEN_STRONG_INLINE
+half2 plog<half2>(const half2& a) {
+  return h2log(a);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE
+half2 pexp<half2>(const half2& a) {
+  return h2exp(a);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE
+half2 psqrt<half2>(const half2& a) {
+  return h2sqrt(a);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE
+half2 prsqrt<half2>(const half2& a) {
+  return h2rsqrt(a);
+}
+
+#else
+
+template<> __device__ EIGEN_STRONG_INLINE half2 plog<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = logf(a1);
+  float r2 = logf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pexp<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = expf(a1);
+  float r2 = expf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 psqrt<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = sqrtf(a1);
+  float r2 = sqrtf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 prsqrt<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = rsqrtf(a1);
+  float r2 = rsqrtf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+#endif
+
+#elif defined EIGEN_VECTORIZE_AVX512
+
+typedef struct {
+  __m256i x;
+} Packet16h;
+
+
+template<> struct is_arithmetic<Packet16h> { enum { value = true }; };
+
+template <>
+struct packet_traits<half> : default_packet_traits {
+  typedef Packet16h type;
+  // There is no half-size packet for Packet16h.
+  typedef Packet16h half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 16,
+    HasHalfPacket = 0,
+    HasAdd    = 0,
+    HasSub    = 0,
+    HasMul    = 0,
+    HasNegate = 0,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasConj   = 0,
+    HasSetLinear = 0,
+    HasDiv = 0,
+    HasSqrt = 0,
+    HasRsqrt = 0,
+    HasExp = 0,
+    HasLog = 0,
+    HasBlend = 0
+  };
+};
+
+
+template<> struct unpacket_traits<Packet16h> { typedef Eigen::half type; enum {size=16, alignment=Aligned32}; typedef Packet16h half; };
+
+template<> EIGEN_STRONG_INLINE Packet16h pset1<Packet16h>(const Eigen::half& from) {
+  Packet16h result;
+  result.x = _mm256_set1_epi16(from.x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half pfirst<Packet16h>(const Packet16h& from) {
+  return half_impl::raw_uint16_to_half(static_cast<unsigned short>(_mm256_extract_epi16(from.x, 0)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h pload<Packet16h>(const Eigen::half* from) {
+  Packet16h result;
+  result.x = _mm256_load_si256(reinterpret_cast<const __m256i*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h ploadu<Packet16h>(const Eigen::half* from) {
+  Packet16h result;
+  result.x = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<half>(Eigen::half* to, const Packet16h& from) {
+  _mm256_store_si256((__m256i*)to, from.x);
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<half>(Eigen::half* to, const Packet16h& from) {
+  _mm256_storeu_si256((__m256i*)to, from.x);
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h
+ploadquad(const Eigen::half* from) {
+  Packet16h result;
+  unsigned short a = from[0].x;
+  unsigned short b = from[1].x;
+  unsigned short c = from[2].x;
+  unsigned short d = from[3].x;
+  result.x = _mm256_set_epi16(d, d, d, d, c, c, c, c, b, b, b, b, a, a, a, a);
+  return result;
+}
+
+EIGEN_STRONG_INLINE Packet16f half2float(const Packet16h& a) {
+#ifdef EIGEN_HAS_FP16_C
+  return _mm512_cvtph_ps(a.x);
+#else
+  EIGEN_ALIGN64 half aux[16];
+  pstore(aux, a);
+  float f0(aux[0]);
+  float f1(aux[1]);
+  float f2(aux[2]);
+  float f3(aux[3]);
+  float f4(aux[4]);
+  float f5(aux[5]);
+  float f6(aux[6]);
+  float f7(aux[7]);
+  float f8(aux[8]);
+  float f9(aux[9]);
+  float fa(aux[10]);
+  float fb(aux[11]);
+  float fc(aux[12]);
+  float fd(aux[13]);
+  float fe(aux[14]);
+  float ff(aux[15]);
+
+  return _mm512_set_ps(
+      ff, fe, fd, fc, fb, fa, f9, f8, f7, f6, f5, f4, f3, f2, f1, f0);
+#endif
+}
+
+EIGEN_STRONG_INLINE Packet16h float2half(const Packet16f& a) {
+#ifdef EIGEN_HAS_FP16_C
+  Packet16h result;
+  result.x = _mm512_cvtps_ph(a, _MM_FROUND_TO_NEAREST_INT|_MM_FROUND_NO_EXC);
+  return result;
+#else
+  EIGEN_ALIGN64 float aux[16];
+  pstore(aux, a);
+  half h0(aux[0]);
+  half h1(aux[1]);
+  half h2(aux[2]);
+  half h3(aux[3]);
+  half h4(aux[4]);
+  half h5(aux[5]);
+  half h6(aux[6]);
+  half h7(aux[7]);
+  half h8(aux[8]);
+  half h9(aux[9]);
+  half ha(aux[10]);
+  half hb(aux[11]);
+  half hc(aux[12]);
+  half hd(aux[13]);
+  half he(aux[14]);
+  half hf(aux[15]);
+
+  Packet16h result;
+  result.x = _mm256_set_epi16(
+      hf.x, he.x, hd.x, hc.x, hb.x, ha.x, h9.x, h8.x,
+      h7.x, h6.x, h5.x, h4.x, h3.x, h2.x, h1.x, h0.x);
+  return result;
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h padd<Packet16h>(const Packet16h& a, const Packet16h& b) {
+  Packet16f af = half2float(a);
+  Packet16f bf = half2float(b);
+  Packet16f rf = padd(af, bf);
+  return float2half(rf);
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h pmul<Packet16h>(const Packet16h& a, const Packet16h& b) {
+  Packet16f af = half2float(a);
+  Packet16f bf = half2float(b);
+  Packet16f rf = pmul(af, bf);
+  return float2half(rf);
+}
+
+template<> EIGEN_STRONG_INLINE half predux<Packet16h>(const Packet16h& from) {
+  Packet16f from_float = half2float(from);
+  return half(predux(from_float));
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h pgather<Eigen::half, Packet16h>(const Eigen::half* from, Index stride)
+{
+  Packet16h result;
+  result.x = _mm256_set_epi16(
+      from[15*stride].x, from[14*stride].x, from[13*stride].x, from[12*stride].x,
+      from[11*stride].x, from[10*stride].x, from[9*stride].x, from[8*stride].x,
+      from[7*stride].x, from[6*stride].x, from[5*stride].x, from[4*stride].x,
+      from[3*stride].x, from[2*stride].x, from[1*stride].x, from[0*stride].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pscatter<half, Packet16h>(half* to, const Packet16h& from, Index stride)
+{
+  EIGEN_ALIGN64 half aux[16];
+  pstore(aux, from);
+  to[stride*0].x = aux[0].x;
+  to[stride*1].x = aux[1].x;
+  to[stride*2].x = aux[2].x;
+  to[stride*3].x = aux[3].x;
+  to[stride*4].x = aux[4].x;
+  to[stride*5].x = aux[5].x;
+  to[stride*6].x = aux[6].x;
+  to[stride*7].x = aux[7].x;
+  to[stride*8].x = aux[8].x;
+  to[stride*9].x = aux[9].x;
+  to[stride*10].x = aux[10].x;
+  to[stride*11].x = aux[11].x;
+  to[stride*12].x = aux[12].x;
+  to[stride*13].x = aux[13].x;
+  to[stride*14].x = aux[14].x;
+  to[stride*15].x = aux[15].x;
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet16h,16>& kernel) {
+  __m256i a = kernel.packet[0].x;
+  __m256i b = kernel.packet[1].x;
+  __m256i c = kernel.packet[2].x;
+  __m256i d = kernel.packet[3].x;
+  __m256i e = kernel.packet[4].x;
+  __m256i f = kernel.packet[5].x;
+  __m256i g = kernel.packet[6].x;
+  __m256i h = kernel.packet[7].x;
+  __m256i i = kernel.packet[8].x;
+  __m256i j = kernel.packet[9].x;
+  __m256i k = kernel.packet[10].x;
+  __m256i l = kernel.packet[11].x;
+  __m256i m = kernel.packet[12].x;
+  __m256i n = kernel.packet[13].x;
+  __m256i o = kernel.packet[14].x;
+  __m256i p = kernel.packet[15].x;
+
+  __m256i ab_07 = _mm256_unpacklo_epi16(a, b);
+  __m256i cd_07 = _mm256_unpacklo_epi16(c, d);
+  __m256i ef_07 = _mm256_unpacklo_epi16(e, f);
+  __m256i gh_07 = _mm256_unpacklo_epi16(g, h);
+  __m256i ij_07 = _mm256_unpacklo_epi16(i, j);
+  __m256i kl_07 = _mm256_unpacklo_epi16(k, l);
+  __m256i mn_07 = _mm256_unpacklo_epi16(m, n);
+  __m256i op_07 = _mm256_unpacklo_epi16(o, p);
+
+  __m256i ab_8f = _mm256_unpackhi_epi16(a, b);
+  __m256i cd_8f = _mm256_unpackhi_epi16(c, d);
+  __m256i ef_8f = _mm256_unpackhi_epi16(e, f);
+  __m256i gh_8f = _mm256_unpackhi_epi16(g, h);
+  __m256i ij_8f = _mm256_unpackhi_epi16(i, j);
+  __m256i kl_8f = _mm256_unpackhi_epi16(k, l);
+  __m256i mn_8f = _mm256_unpackhi_epi16(m, n);
+  __m256i op_8f = _mm256_unpackhi_epi16(o, p);
+
+  __m256i abcd_03 = _mm256_unpacklo_epi32(ab_07, cd_07);
+  __m256i abcd_47 = _mm256_unpackhi_epi32(ab_07, cd_07);
+  __m256i efgh_03 = _mm256_unpacklo_epi32(ef_07, gh_07);
+  __m256i efgh_47 = _mm256_unpackhi_epi32(ef_07, gh_07);
+  __m256i ijkl_03 = _mm256_unpacklo_epi32(ij_07, kl_07);
+  __m256i ijkl_47 = _mm256_unpackhi_epi32(ij_07, kl_07);
+  __m256i mnop_03 = _mm256_unpacklo_epi32(mn_07, op_07);
+  __m256i mnop_47 = _mm256_unpackhi_epi32(mn_07, op_07);
+
+  __m256i abcd_8b = _mm256_unpacklo_epi32(ab_8f, cd_8f);
+  __m256i abcd_cf = _mm256_unpackhi_epi32(ab_8f, cd_8f);
+  __m256i efgh_8b = _mm256_unpacklo_epi32(ef_8f, gh_8f);
+  __m256i efgh_cf = _mm256_unpackhi_epi32(ef_8f, gh_8f);
+  __m256i ijkl_8b = _mm256_unpacklo_epi32(ij_8f, kl_8f);
+  __m256i ijkl_cf = _mm256_unpackhi_epi32(ij_8f, kl_8f);
+  __m256i mnop_8b = _mm256_unpacklo_epi32(mn_8f, op_8f);
+  __m256i mnop_cf = _mm256_unpackhi_epi32(mn_8f, op_8f);
+
+  __m256i abcdefgh_01 = _mm256_unpacklo_epi64(abcd_03, efgh_03);
+  __m256i abcdefgh_23 = _mm256_unpackhi_epi64(abcd_03, efgh_03);
+  __m256i ijklmnop_01 = _mm256_unpacklo_epi64(ijkl_03, mnop_03);
+  __m256i ijklmnop_23 = _mm256_unpackhi_epi64(ijkl_03, mnop_03);
+  __m256i abcdefgh_45 = _mm256_unpacklo_epi64(abcd_47, efgh_47);
+  __m256i abcdefgh_67 = _mm256_unpackhi_epi64(abcd_47, efgh_47);
+  __m256i ijklmnop_45 = _mm256_unpacklo_epi64(ijkl_47, mnop_47);
+  __m256i ijklmnop_67 = _mm256_unpackhi_epi64(ijkl_47, mnop_47);
+  __m256i abcdefgh_89 = _mm256_unpacklo_epi64(abcd_8b, efgh_8b);
+  __m256i abcdefgh_ab = _mm256_unpackhi_epi64(abcd_8b, efgh_8b);
+  __m256i ijklmnop_89 = _mm256_unpacklo_epi64(ijkl_8b, mnop_8b);
+  __m256i ijklmnop_ab = _mm256_unpackhi_epi64(ijkl_8b, mnop_8b);
+  __m256i abcdefgh_cd = _mm256_unpacklo_epi64(abcd_cf, efgh_cf);
+  __m256i abcdefgh_ef = _mm256_unpackhi_epi64(abcd_cf, efgh_cf);
+  __m256i ijklmnop_cd = _mm256_unpacklo_epi64(ijkl_cf, mnop_cf);
+  __m256i ijklmnop_ef = _mm256_unpackhi_epi64(ijkl_cf, mnop_cf);
+
+  // NOTE: no unpacklo/hi instr in this case, so using permute instr.
+  __m256i a_p_0 = _mm256_permute2x128_si256(abcdefgh_01, ijklmnop_01, 0x20);
+  __m256i a_p_1 = _mm256_permute2x128_si256(abcdefgh_01, ijklmnop_01, 0x31);
+  __m256i a_p_2 = _mm256_permute2x128_si256(abcdefgh_23, ijklmnop_23, 0x20);
+  __m256i a_p_3 = _mm256_permute2x128_si256(abcdefgh_23, ijklmnop_23, 0x31);
+  __m256i a_p_4 = _mm256_permute2x128_si256(abcdefgh_45, ijklmnop_45, 0x20);
+  __m256i a_p_5 = _mm256_permute2x128_si256(abcdefgh_45, ijklmnop_45, 0x31);
+  __m256i a_p_6 = _mm256_permute2x128_si256(abcdefgh_67, ijklmnop_67, 0x20);
+  __m256i a_p_7 = _mm256_permute2x128_si256(abcdefgh_67, ijklmnop_67, 0x31);
+  __m256i a_p_8 = _mm256_permute2x128_si256(abcdefgh_89, ijklmnop_89, 0x20);
+  __m256i a_p_9 = _mm256_permute2x128_si256(abcdefgh_89, ijklmnop_89, 0x31);
+  __m256i a_p_a = _mm256_permute2x128_si256(abcdefgh_ab, ijklmnop_ab, 0x20);
+  __m256i a_p_b = _mm256_permute2x128_si256(abcdefgh_ab, ijklmnop_ab, 0x31);
+  __m256i a_p_c = _mm256_permute2x128_si256(abcdefgh_cd, ijklmnop_cd, 0x20);
+  __m256i a_p_d = _mm256_permute2x128_si256(abcdefgh_cd, ijklmnop_cd, 0x31);
+  __m256i a_p_e = _mm256_permute2x128_si256(abcdefgh_ef, ijklmnop_ef, 0x20);
+  __m256i a_p_f = _mm256_permute2x128_si256(abcdefgh_ef, ijklmnop_ef, 0x31);
+
+  kernel.packet[0].x = a_p_0;
+  kernel.packet[1].x = a_p_1;
+  kernel.packet[2].x = a_p_2;
+  kernel.packet[3].x = a_p_3;
+  kernel.packet[4].x = a_p_4;
+  kernel.packet[5].x = a_p_5;
+  kernel.packet[6].x = a_p_6;
+  kernel.packet[7].x = a_p_7;
+  kernel.packet[8].x = a_p_8;
+  kernel.packet[9].x = a_p_9;
+  kernel.packet[10].x = a_p_a;
+  kernel.packet[11].x = a_p_b;
+  kernel.packet[12].x = a_p_c;
+  kernel.packet[13].x = a_p_d;
+  kernel.packet[14].x = a_p_e;
+  kernel.packet[15].x = a_p_f;
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet16h,8>& kernel) {
+  EIGEN_ALIGN64 half in[8][16];
+  pstore<half>(in[0], kernel.packet[0]);
+  pstore<half>(in[1], kernel.packet[1]);
+  pstore<half>(in[2], kernel.packet[2]);
+  pstore<half>(in[3], kernel.packet[3]);
+  pstore<half>(in[4], kernel.packet[4]);
+  pstore<half>(in[5], kernel.packet[5]);
+  pstore<half>(in[6], kernel.packet[6]);
+  pstore<half>(in[7], kernel.packet[7]);
+
+  EIGEN_ALIGN64 half out[8][16];
+
+  for (int i = 0; i < 8; ++i) {
+    for (int j = 0; j < 8; ++j) {
+      out[i][j] = in[j][2*i];
+    }
+    for (int j = 0; j < 8; ++j) {
+      out[i][j+8] = in[j][2*i+1];
+    }
+  }
+
+  kernel.packet[0] = pload<Packet16h>(out[0]);
+  kernel.packet[1] = pload<Packet16h>(out[1]);
+  kernel.packet[2] = pload<Packet16h>(out[2]);
+  kernel.packet[3] = pload<Packet16h>(out[3]);
+  kernel.packet[4] = pload<Packet16h>(out[4]);
+  kernel.packet[5] = pload<Packet16h>(out[5]);
+  kernel.packet[6] = pload<Packet16h>(out[6]);
+  kernel.packet[7] = pload<Packet16h>(out[7]);
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet16h,4>& kernel) {
+  EIGEN_ALIGN64 half in[4][16];
+  pstore<half>(in[0], kernel.packet[0]);
+  pstore<half>(in[1], kernel.packet[1]);
+  pstore<half>(in[2], kernel.packet[2]);
+  pstore<half>(in[3], kernel.packet[3]);
+
+  EIGEN_ALIGN64 half out[4][16];
+
+  for (int i = 0; i < 4; ++i) {
+    for (int j = 0; j < 4; ++j) {
+      out[i][j] = in[j][4*i];
+    }
+    for (int j = 0; j < 4; ++j) {
+      out[i][j+4] = in[j][4*i+1];
+    }
+    for (int j = 0; j < 4; ++j) {
+      out[i][j+8] = in[j][4*i+2];
+    }
+    for (int j = 0; j < 4; ++j) {
+      out[i][j+12] = in[j][4*i+3];
+    }
+  }
+
+  kernel.packet[0] = pload<Packet16h>(out[0]);
+  kernel.packet[1] = pload<Packet16h>(out[1]);
+  kernel.packet[2] = pload<Packet16h>(out[2]);
+  kernel.packet[3] = pload<Packet16h>(out[3]);
+}
+
+
+#elif defined EIGEN_VECTORIZE_AVX
+
+typedef struct {
+  __m128i x;
+} Packet8h;
+
+
+template<> struct is_arithmetic<Packet8h> { enum { value = true }; };
+
+template <>
+struct packet_traits<Eigen::half> : default_packet_traits {
+  typedef Packet8h type;
+  // There is no half-size packet for Packet8h.
+  typedef Packet8h half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 8,
+    HasHalfPacket = 0,
+    HasAdd    = 0,
+    HasSub    = 0,
+    HasMul    = 0,
+    HasNegate = 0,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasConj   = 0,
+    HasSetLinear = 0,
+    HasDiv = 0,
+    HasSqrt = 0,
+    HasRsqrt = 0,
+    HasExp = 0,
+    HasLog = 0,
+    HasBlend = 0
+  };
+};
+
+
+template<> struct unpacket_traits<Packet8h> { typedef Eigen::half type; enum {size=8, alignment=Aligned16}; typedef Packet8h half; };
+
+template<> EIGEN_STRONG_INLINE Packet8h pset1<Packet8h>(const Eigen::half& from) {
+  Packet8h result;
+  result.x = _mm_set1_epi16(from.x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half pfirst<Packet8h>(const Packet8h& from) {
+  return half_impl::raw_uint16_to_half(static_cast<unsigned short>(_mm_extract_epi16(from.x, 0)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h pload<Packet8h>(const Eigen::half* from) {
+  Packet8h result;
+  result.x = _mm_load_si128(reinterpret_cast<const __m128i*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h ploadu<Packet8h>(const Eigen::half* from) {
+  Packet8h result;
+  result.x = _mm_loadu_si128(reinterpret_cast<const __m128i*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<Eigen::half>(Eigen::half* to, const Packet8h& from) {
+  _mm_store_si128(reinterpret_cast<__m128i*>(to), from.x);
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<Eigen::half>(Eigen::half* to, const Packet8h& from) {
+  _mm_storeu_si128(reinterpret_cast<__m128i*>(to), from.x);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h
+ploadquad<Packet8h>(const Eigen::half* from) {
+  Packet8h result;
+  unsigned short a = from[0].x;
+  unsigned short b = from[1].x;
+  result.x = _mm_set_epi16(b, b, b, b, a, a, a, a);
+  return result;
+}
+
+EIGEN_STRONG_INLINE Packet8f half2float(const Packet8h& a) {
+#ifdef EIGEN_HAS_FP16_C
+  return _mm256_cvtph_ps(a.x);
+#else
+  EIGEN_ALIGN32 Eigen::half aux[8];
+  pstore(aux, a);
+  float f0(aux[0]);
+  float f1(aux[1]);
+  float f2(aux[2]);
+  float f3(aux[3]);
+  float f4(aux[4]);
+  float f5(aux[5]);
+  float f6(aux[6]);
+  float f7(aux[7]);
+
+  return _mm256_set_ps(f7, f6, f5, f4, f3, f2, f1, f0);
+#endif
+}
+
+EIGEN_STRONG_INLINE Packet8h float2half(const Packet8f& a) {
+#ifdef EIGEN_HAS_FP16_C
+  Packet8h result;
+  result.x = _mm256_cvtps_ph(a, _MM_FROUND_TO_NEAREST_INT|_MM_FROUND_NO_EXC);
+  return result;
+#else
+  EIGEN_ALIGN32 float aux[8];
+  pstore(aux, a);
+  Eigen::half h0(aux[0]);
+  Eigen::half h1(aux[1]);
+  Eigen::half h2(aux[2]);
+  Eigen::half h3(aux[3]);
+  Eigen::half h4(aux[4]);
+  Eigen::half h5(aux[5]);
+  Eigen::half h6(aux[6]);
+  Eigen::half h7(aux[7]);
+
+  Packet8h result;
+  result.x = _mm_set_epi16(h7.x, h6.x, h5.x, h4.x, h3.x, h2.x, h1.x, h0.x);
+  return result;
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h pconj(const Packet8h& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet8h padd<Packet8h>(const Packet8h& a, const Packet8h& b) {
+  Packet8f af = half2float(a);
+  Packet8f bf = half2float(b);
+  Packet8f rf = padd(af, bf);
+  return float2half(rf);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h pmul<Packet8h>(const Packet8h& a, const Packet8h& b) {
+  Packet8f af = half2float(a);
+  Packet8f bf = half2float(b);
+  Packet8f rf = pmul(af, bf);
+  return float2half(rf);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h pgather<Eigen::half, Packet8h>(const Eigen::half* from, Index stride)
+{
+  Packet8h result;
+  result.x = _mm_set_epi16(from[7*stride].x, from[6*stride].x, from[5*stride].x, from[4*stride].x, from[3*stride].x, from[2*stride].x, from[1*stride].x, from[0*stride].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pscatter<Eigen::half, Packet8h>(Eigen::half* to, const Packet8h& from, Index stride)
+{
+  EIGEN_ALIGN32 Eigen::half aux[8];
+  pstore(aux, from);
+  to[stride*0].x = aux[0].x;
+  to[stride*1].x = aux[1].x;
+  to[stride*2].x = aux[2].x;
+  to[stride*3].x = aux[3].x;
+  to[stride*4].x = aux[4].x;
+  to[stride*5].x = aux[5].x;
+  to[stride*6].x = aux[6].x;
+  to[stride*7].x = aux[7].x;
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half predux<Packet8h>(const Packet8h& a) {
+  Packet8f af = half2float(a);
+  float reduced = predux<Packet8f>(af);
+  return Eigen::half(reduced);
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half predux_max<Packet8h>(const Packet8h& a) {
+  Packet8f af = half2float(a);
+  float reduced = predux_max<Packet8f>(af);
+  return Eigen::half(reduced);
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half predux_min<Packet8h>(const Packet8h& a) {
+  Packet8f af = half2float(a);
+  float reduced = predux_min<Packet8f>(af);
+  return Eigen::half(reduced);
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half predux_mul<Packet8h>(const Packet8h& a) {
+  Packet8f af = half2float(a);
+  float reduced = predux_mul<Packet8f>(af);
+  return Eigen::half(reduced);
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet8h,8>& kernel) {
+  __m128i a = kernel.packet[0].x;
+  __m128i b = kernel.packet[1].x;
+  __m128i c = kernel.packet[2].x;
+  __m128i d = kernel.packet[3].x;
+  __m128i e = kernel.packet[4].x;
+  __m128i f = kernel.packet[5].x;
+  __m128i g = kernel.packet[6].x;
+  __m128i h = kernel.packet[7].x;
+
+  __m128i a03b03 = _mm_unpacklo_epi16(a, b);
+  __m128i c03d03 = _mm_unpacklo_epi16(c, d);
+  __m128i e03f03 = _mm_unpacklo_epi16(e, f);
+  __m128i g03h03 = _mm_unpacklo_epi16(g, h);
+  __m128i a47b47 = _mm_unpackhi_epi16(a, b);
+  __m128i c47d47 = _mm_unpackhi_epi16(c, d);
+  __m128i e47f47 = _mm_unpackhi_epi16(e, f);
+  __m128i g47h47 = _mm_unpackhi_epi16(g, h);
+
+  __m128i a01b01c01d01 = _mm_unpacklo_epi32(a03b03, c03d03);
+  __m128i a23b23c23d23 = _mm_unpackhi_epi32(a03b03, c03d03);
+  __m128i e01f01g01h01 = _mm_unpacklo_epi32(e03f03, g03h03);
+  __m128i e23f23g23h23 = _mm_unpackhi_epi32(e03f03, g03h03);
+  __m128i a45b45c45d45 = _mm_unpacklo_epi32(a47b47, c47d47);
+  __m128i a67b67c67d67 = _mm_unpackhi_epi32(a47b47, c47d47);
+  __m128i e45f45g45h45 = _mm_unpacklo_epi32(e47f47, g47h47);
+  __m128i e67f67g67h67 = _mm_unpackhi_epi32(e47f47, g47h47);
+
+  __m128i a0b0c0d0e0f0g0h0 = _mm_unpacklo_epi64(a01b01c01d01, e01f01g01h01);
+  __m128i a1b1c1d1e1f1g1h1 = _mm_unpackhi_epi64(a01b01c01d01, e01f01g01h01);
+  __m128i a2b2c2d2e2f2g2h2 = _mm_unpacklo_epi64(a23b23c23d23, e23f23g23h23);
+  __m128i a3b3c3d3e3f3g3h3 = _mm_unpackhi_epi64(a23b23c23d23, e23f23g23h23);
+  __m128i a4b4c4d4e4f4g4h4 = _mm_unpacklo_epi64(a45b45c45d45, e45f45g45h45);
+  __m128i a5b5c5d5e5f5g5h5 = _mm_unpackhi_epi64(a45b45c45d45, e45f45g45h45);
+  __m128i a6b6c6d6e6f6g6h6 = _mm_unpacklo_epi64(a67b67c67d67, e67f67g67h67);
+  __m128i a7b7c7d7e7f7g7h7 = _mm_unpackhi_epi64(a67b67c67d67, e67f67g67h67);
+
+  kernel.packet[0].x = a0b0c0d0e0f0g0h0;
+  kernel.packet[1].x = a1b1c1d1e1f1g1h1;
+  kernel.packet[2].x = a2b2c2d2e2f2g2h2;
+  kernel.packet[3].x = a3b3c3d3e3f3g3h3;
+  kernel.packet[4].x = a4b4c4d4e4f4g4h4;
+  kernel.packet[5].x = a5b5c5d5e5f5g5h5;
+  kernel.packet[6].x = a6b6c6d6e6f6g6h6;
+  kernel.packet[7].x = a7b7c7d7e7f7g7h7;
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet8h,4>& kernel) {
+  EIGEN_ALIGN32 Eigen::half in[4][8];
+  pstore<Eigen::half>(in[0], kernel.packet[0]);
+  pstore<Eigen::half>(in[1], kernel.packet[1]);
+  pstore<Eigen::half>(in[2], kernel.packet[2]);
+  pstore<Eigen::half>(in[3], kernel.packet[3]);
+
+  EIGEN_ALIGN32 Eigen::half out[4][8];
+
+  for (int i = 0; i < 4; ++i) {
+    for (int j = 0; j < 4; ++j) {
+      out[i][j] = in[j][2*i];
+    }
+    for (int j = 0; j < 4; ++j) {
+      out[i][j+4] = in[j][2*i+1];
+    }
+  }
+
+  kernel.packet[0] = pload<Packet8h>(out[0]);
+  kernel.packet[1] = pload<Packet8h>(out[1]);
+  kernel.packet[2] = pload<Packet8h>(out[2]);
+  kernel.packet[3] = pload<Packet8h>(out[3]);
+}
+
+
+// Disable the following code since it's broken on too many platforms / compilers.
+//#elif defined(EIGEN_VECTORIZE_SSE) && (!EIGEN_ARCH_x86_64) && (!EIGEN_COMP_MSVC)
+#elif 0
+
+typedef struct {
+  __m64 x;
+} Packet4h;
+
+
+template<> struct is_arithmetic<Packet4h> { enum { value = true }; };
+
+template <>
+struct packet_traits<Eigen::half> : default_packet_traits {
+  typedef Packet4h type;
+  // There is no half-size packet for Packet4h.
+  typedef Packet4h half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 4,
+    HasHalfPacket = 0,
+    HasAdd    = 0,
+    HasSub    = 0,
+    HasMul    = 0,
+    HasNegate = 0,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasConj   = 0,
+    HasSetLinear = 0,
+    HasDiv = 0,
+    HasSqrt = 0,
+    HasRsqrt = 0,
+    HasExp = 0,
+    HasLog = 0,
+    HasBlend = 0
+  };
+};
+
+
+template<> struct unpacket_traits<Packet4h> { typedef Eigen::half type; enum {size=4, alignment=Aligned16}; typedef Packet4h half; };
+
+template<> EIGEN_STRONG_INLINE Packet4h pset1<Packet4h>(const Eigen::half& from) {
+  Packet4h result;
+  result.x = _mm_set1_pi16(from.x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half pfirst<Packet4h>(const Packet4h& from) {
+  return half_impl::raw_uint16_to_half(static_cast<unsigned short>(_mm_cvtsi64_si32(from.x)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h pconj(const Packet4h& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet4h padd<Packet4h>(const Packet4h& a, const Packet4h& b) {
+  __int64_t a64 = _mm_cvtm64_si64(a.x);
+  __int64_t b64 = _mm_cvtm64_si64(b.x);
+
+  Eigen::half h[4];
+
+  Eigen::half ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64));
+  Eigen::half hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64));
+  h[0] = ha + hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 16));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 16));
+  h[1] = ha + hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 32));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 32));
+  h[2] = ha + hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 48));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 48));
+  h[3] = ha + hb;
+  Packet4h result;
+  result.x = _mm_set_pi16(h[3].x, h[2].x, h[1].x, h[0].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h pmul<Packet4h>(const Packet4h& a, const Packet4h& b) {
+  __int64_t a64 = _mm_cvtm64_si64(a.x);
+  __int64_t b64 = _mm_cvtm64_si64(b.x);
+
+  Eigen::half h[4];
+
+  Eigen::half ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64));
+  Eigen::half hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64));
+  h[0] = ha * hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 16));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 16));
+  h[1] = ha * hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 32));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 32));
+  h[2] = ha * hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 48));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 48));
+  h[3] = ha * hb;
+  Packet4h result;
+  result.x = _mm_set_pi16(h[3].x, h[2].x, h[1].x, h[0].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h pload<Packet4h>(const Eigen::half* from) {
+  Packet4h result;
+  result.x = _mm_cvtsi64_m64(*reinterpret_cast<const __int64_t*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h ploadu<Packet4h>(const Eigen::half* from) {
+  Packet4h result;
+  result.x = _mm_cvtsi64_m64(*reinterpret_cast<const __int64_t*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<Eigen::half>(Eigen::half* to, const Packet4h& from) {
+  __int64_t r = _mm_cvtm64_si64(from.x);
+  *(reinterpret_cast<__int64_t*>(to)) = r;
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<Eigen::half>(Eigen::half* to, const Packet4h& from) {
+  __int64_t r = _mm_cvtm64_si64(from.x);
+  *(reinterpret_cast<__int64_t*>(to)) = r;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h
+ploadquad<Packet4h>(const Eigen::half* from) {
+  return pset1<Packet4h>(*from);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h pgather<Eigen::half, Packet4h>(const Eigen::half* from, Index stride)
+{
+  Packet4h result;
+  result.x = _mm_set_pi16(from[3*stride].x, from[2*stride].x, from[1*stride].x, from[0*stride].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pscatter<Eigen::half, Packet4h>(Eigen::half* to, const Packet4h& from, Index stride)
+{
+  __int64_t a = _mm_cvtm64_si64(from.x);
+  to[stride*0].x = static_cast<unsigned short>(a);
+  to[stride*1].x = static_cast<unsigned short>(a >> 16);
+  to[stride*2].x = static_cast<unsigned short>(a >> 32);
+  to[stride*3].x = static_cast<unsigned short>(a >> 48);
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet4h,4>& kernel) {
+  __m64 T0 = _mm_unpacklo_pi16(kernel.packet[0].x, kernel.packet[1].x);
+  __m64 T1 = _mm_unpacklo_pi16(kernel.packet[2].x, kernel.packet[3].x);
+  __m64 T2 = _mm_unpackhi_pi16(kernel.packet[0].x, kernel.packet[1].x);
+  __m64 T3 = _mm_unpackhi_pi16(kernel.packet[2].x, kernel.packet[3].x);
+
+  kernel.packet[0].x = _mm_unpacklo_pi32(T0, T1);
+  kernel.packet[1].x = _mm_unpackhi_pi32(T0, T1);
+  kernel.packet[2].x = _mm_unpacklo_pi32(T2, T3);
+  kernel.packet[3].x = _mm_unpackhi_pi32(T2, T3);
+}
+
+#endif
+
+}
+}
+
+#endif // EIGEN_PACKET_MATH_HALF_CUDA_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/CUDA/TypeCasting.h b/extern/Eigen3/Eigen/src/Core/arch/CUDA/TypeCasting.h
new file mode 100644
index 00000000000..aa5fbce8eac
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/CUDA/TypeCasting.h
@@ -0,0 +1,212 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_TYPE_CASTING_CUDA_H
+#define EIGEN_TYPE_CASTING_CUDA_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<>
+struct scalar_cast_op<float, Eigen::half> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
+  typedef Eigen::half result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half operator() (const float& a) const {
+    #if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+      return __float2half(a);
+    #else
+      return Eigen::half(a);
+    #endif
+  }
+};
+
+template<>
+struct functor_traits<scalar_cast_op<float, Eigen::half> >
+{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
+
+
+template<>
+struct scalar_cast_op<int, Eigen::half> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
+  typedef Eigen::half result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half operator() (const int& a) const {
+    #if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+      return __float2half(static_cast<float>(a));
+    #else
+      return Eigen::half(static_cast<float>(a));
+    #endif
+  }
+};
+
+template<>
+struct functor_traits<scalar_cast_op<int, Eigen::half> >
+{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
+
+
+template<>
+struct scalar_cast_op<Eigen::half, float> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
+  typedef float result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float operator() (const Eigen::half& a) const {
+    #if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+      return __half2float(a);
+    #else
+      return static_cast<float>(a);
+    #endif
+  }
+};
+
+template<>
+struct functor_traits<scalar_cast_op<Eigen::half, float> >
+{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
+
+
+
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+
+template <>
+struct type_casting_traits<Eigen::half, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 2,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pcast<half2, float4>(const half2& a, const half2& b) {
+  float2 r1 = __half22float2(a);
+  float2 r2 = __half22float2(b);
+  return make_float4(r1.x, r1.y, r2.x, r2.y);
+}
+
+template <>
+struct type_casting_traits<float, Eigen::half> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 2
+  };
+};
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pcast<float4, half2>(const float4& a) {
+  // Simply discard the second half of the input
+  return __floats2half2_rn(a.x, a.y);
+}
+
+#elif defined EIGEN_VECTORIZE_AVX512
+template <>
+struct type_casting_traits<half, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet16f pcast<Packet16h, Packet16f>(const Packet16h& a) {
+  return half2float(a);
+}
+
+template <>
+struct type_casting_traits<float, half> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet16h pcast<Packet16f, Packet16h>(const Packet16f& a) {
+  return float2half(a);
+}
+
+#elif defined EIGEN_VECTORIZE_AVX
+
+template <>
+struct type_casting_traits<Eigen::half, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8h, Packet8f>(const Packet8h& a) {
+  return half2float(a);
+}
+
+template <>
+struct type_casting_traits<float, Eigen::half> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet8h pcast<Packet8f, Packet8h>(const Packet8f& a) {
+  return float2half(a);
+}
+
+// Disable the following code since it's broken on too many platforms / compilers.
+//#elif defined(EIGEN_VECTORIZE_SSE) && (!EIGEN_ARCH_x86_64) && (!EIGEN_COMP_MSVC)
+#elif 0
+
+template <>
+struct type_casting_traits<Eigen::half, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet4h, Packet4f>(const Packet4h& a) {
+  __int64_t a64 = _mm_cvtm64_si64(a.x);
+  Eigen::half h = raw_uint16_to_half(static_cast<unsigned short>(a64));
+  float f1 = static_cast<float>(h);
+  h = raw_uint16_to_half(static_cast<unsigned short>(a64 >> 16));
+  float f2 = static_cast<float>(h);
+  h = raw_uint16_to_half(static_cast<unsigned short>(a64 >> 32));
+  float f3 = static_cast<float>(h);
+  h = raw_uint16_to_half(static_cast<unsigned short>(a64 >> 48));
+  float f4 = static_cast<float>(h);
+  return _mm_set_ps(f4, f3, f2, f1);
+}
+
+template <>
+struct type_casting_traits<float, Eigen::half> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet4h pcast<Packet4f, Packet4h>(const Packet4f& a) {
+  EIGEN_ALIGN16 float aux[4];
+  pstore(aux, a);
+  Eigen::half h0(aux[0]);
+  Eigen::half h1(aux[1]);
+  Eigen::half h2(aux[2]);
+  Eigen::half h3(aux[3]);
+
+  Packet4h result;
+  result.x = _mm_set_pi16(h3.x, h2.x, h1.x, h0.x);
+  return result;
+}
+
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TYPE_CASTING_CUDA_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/Default/ConjHelper.h b/extern/Eigen3/Eigen/src/Core/arch/Default/ConjHelper.h
new file mode 100644
index 00000000000..4cfe34e0526
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/Default/ConjHelper.h
@@ -0,0 +1,29 @@
+
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2017 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ARCH_CONJ_HELPER_H
+#define EIGEN_ARCH_CONJ_HELPER_H
+
+#define EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(PACKET_CPLX, PACKET_REAL)                                                          \
+  template<> struct conj_helper<PACKET_REAL, PACKET_CPLX, false,false> {                                          \
+    EIGEN_STRONG_INLINE PACKET_CPLX pmadd(const PACKET_REAL& x, const PACKET_CPLX& y, const PACKET_CPLX& c) const \
+    { return padd(c, pmul(x,y)); }                                                                                \
+    EIGEN_STRONG_INLINE PACKET_CPLX pmul(const PACKET_REAL& x, const PACKET_CPLX& y) const                        \
+    { return PACKET_CPLX(Eigen::internal::pmul<PACKET_REAL>(x, y.v)); }                                           \
+  };                                                                                                              \
+                                                                                                                  \
+  template<> struct conj_helper<PACKET_CPLX, PACKET_REAL, false,false> {                                          \
+    EIGEN_STRONG_INLINE PACKET_CPLX pmadd(const PACKET_CPLX& x, const PACKET_REAL& y, const PACKET_CPLX& c) const \
+    { return padd(c, pmul(x,y)); }                                                                                \
+    EIGEN_STRONG_INLINE PACKET_CPLX pmul(const PACKET_CPLX& x, const PACKET_REAL& y) const                        \
+    { return PACKET_CPLX(Eigen::internal::pmul<PACKET_REAL>(x.v, y)); }                                           \
+  };
+
+#endif // EIGEN_ARCH_CONJ_HELPER_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/NEON/Complex.h b/extern/Eigen3/Eigen/src/Core/arch/NEON/Complex.h
index 8d9255eef6a..306a309beb2 100644
--- a/extern/Eigen3/Eigen/src/Core/arch/NEON/Complex.h
+++ b/extern/Eigen3/Eigen/src/Core/arch/NEON/Complex.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2010 Konstantinos Margaritis <markos@freevec.org>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -14,8 +15,21 @@ namespace Eigen {
 
 namespace internal {
 
-static uint32x4_t p4ui_CONJ_XOR = EIGEN_INIT_NEON_PACKET4(0x00000000, 0x80000000, 0x00000000, 0x80000000);
-static uint32x2_t p2ui_CONJ_XOR = EIGEN_INIT_NEON_PACKET2(0x00000000, 0x80000000);
+inline uint32x4_t p4ui_CONJ_XOR() {
+// See bug 1325, clang fails to call vld1q_u64.
+#if EIGEN_COMP_CLANG
+  uint32x4_t ret = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
+  return ret;
+#else
+  static const uint32_t conj_XOR_DATA[] = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
+  return vld1q_u32( conj_XOR_DATA );
+#endif
+}
+
+inline uint32x2_t p2ui_CONJ_XOR() {
+  static const uint32_t conj_XOR_DATA[] = { 0x00000000, 0x80000000 };
+  return vld1_u32( conj_XOR_DATA );
+}
 
 //---------- float ----------
 struct Packet2cf
@@ -28,10 +42,12 @@ struct Packet2cf
 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
 {
   typedef Packet2cf type;
+  typedef Packet2cf half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 2,
+    HasHalfPacket = 0,
 
     HasAdd    = 1,
     HasSub    = 1,
@@ -46,12 +62,12 @@ template<> struct packet_traits<std::complex<float> >  : default_packet_traits
   };
 };
 
-template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; };
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
 
 template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>&  from)
 {
   float32x2_t r64;
-  r64 = vld1_f32((float *)&from);
+  r64 = vld1_f32((const float *)&from);
 
   return Packet2cf(vcombine_f32(r64, r64));
 }
@@ -62,7 +78,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Pa
 template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
 {
   Packet4ui b = vreinterpretq_u32_f32(a.v);
-  return Packet2cf(vreinterpretq_f32_u32(veorq_u32(b, p4ui_CONJ_XOR)));
+  return Packet2cf(vreinterpretq_f32_u32(veorq_u32(b, p4ui_CONJ_XOR())));
 }
 
 template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
@@ -71,14 +87,14 @@ template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, con
 
   // Get the real values of a | a1_re | a1_re | a2_re | a2_re |
   v1 = vcombine_f32(vdup_lane_f32(vget_low_f32(a.v), 0), vdup_lane_f32(vget_high_f32(a.v), 0));
-  // Get the real values of a | a1_im | a1_im | a2_im | a2_im |
+  // Get the imag values of a | a1_im | a1_im | a2_im | a2_im |
   v2 = vcombine_f32(vdup_lane_f32(vget_low_f32(a.v), 1), vdup_lane_f32(vget_high_f32(a.v), 1));
   // Multiply the real a with b
   v1 = vmulq_f32(v1, b.v);
   // Multiply the imag a with b
   v2 = vmulq_f32(v2, b.v);
   // Conjugate v2 
-  v2 = vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(v2), p4ui_CONJ_XOR));
+  v2 = vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(v2), p4ui_CONJ_XOR()));
   // Swap real/imag elements in v2.
   v2 = vrev64q_f32(v2);
   // Add and return the result
@@ -87,7 +103,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, con
 
 template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
-  return Packet2cf(vreinterpretq_f32_u32(vorrq_u32(vreinterpretq_u32_f32(a.v),vreinterpretq_u32_f32(b.v))));
+  return Packet2cf(vreinterpretq_f32_u32(vandq_u32(vreinterpretq_u32_f32(a.v),vreinterpretq_u32_f32(b.v))));
 }
 template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
@@ -110,7 +126,23 @@ template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<
 template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
 
-template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { EIGEN_ARM_PREFETCH((float *)addr); }
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
+{
+  Packet4f res = pset1<Packet4f>(0.f);
+  res = vsetq_lane_f32(std::real(from[0*stride]), res, 0);
+  res = vsetq_lane_f32(std::imag(from[0*stride]), res, 1);
+  res = vsetq_lane_f32(std::real(from[1*stride]), res, 2);
+  res = vsetq_lane_f32(std::imag(from[1*stride]), res, 3);
+  return Packet2cf(res);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
+{
+  to[stride*0] = std::complex<float>(vgetq_lane_f32(from.v, 0), vgetq_lane_f32(from.v, 1));
+  to[stride*1] = std::complex<float>(vgetq_lane_f32(from.v, 2), vgetq_lane_f32(from.v, 3));
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { EIGEN_ARM_PREFETCH((const float *)addr); }
 
 template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
 {
@@ -177,7 +209,7 @@ template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const P
   // Multiply the imag a with b
   v2 = vmul_f32(v2, a2);
   // Conjugate v2 
-  v2 = vreinterpret_f32_u32(veor_u32(vreinterpret_u32_f32(v2), p2ui_CONJ_XOR));
+  v2 = vreinterpret_f32_u32(veor_u32(vreinterpret_u32_f32(v2), p2ui_CONJ_XOR()));
   // Swap real/imag elements in v2.
   v2 = vrev64_f32(v2);
   // Add v1, v2
@@ -233,9 +265,11 @@ template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
   }
 };
 
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
+
 template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
-  // TODO optimize it for AltiVec
+  // TODO optimize it for NEON
   Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a,b);
   Packet4f s, rev_s;
 
@@ -243,8 +277,211 @@ template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, con
   s = vmulq_f32(b.v, b.v);
   rev_s = vrev64q_f32(s);
 
-  return Packet2cf(pdiv(res.v, vaddq_f32(s,rev_s)));
+  return Packet2cf(pdiv<Packet4f>(res.v, vaddq_f32(s,rev_s)));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2cf,2>& kernel) {
+  Packet4f tmp = vcombine_f32(vget_high_f32(kernel.packet[0].v), vget_high_f32(kernel.packet[1].v));
+  kernel.packet[0].v = vcombine_f32(vget_low_f32(kernel.packet[0].v), vget_low_f32(kernel.packet[1].v));
+  kernel.packet[1].v = tmp;
+}
+
+//---------- double ----------
+#if EIGEN_ARCH_ARM64 && !EIGEN_APPLE_DOUBLE_NEON_BUG
+
+// See bug 1325, clang fails to call vld1q_u64.
+#if EIGEN_COMP_CLANG
+  static uint64x2_t p2ul_CONJ_XOR = {0x0, 0x8000000000000000};
+#else
+  const uint64_t  p2ul_conj_XOR_DATA[] = { 0x0, 0x8000000000000000 };
+  static uint64x2_t p2ul_CONJ_XOR = vld1q_u64( p2ul_conj_XOR_DATA );
+#endif
+
+struct Packet1cd
+{
+  EIGEN_STRONG_INLINE Packet1cd() {}
+  EIGEN_STRONG_INLINE explicit Packet1cd(const Packet2d& a) : v(a) {}
+  Packet2d v;
+};
+
+template<> struct packet_traits<std::complex<double> >  : default_packet_traits
+{
+  typedef Packet1cd type;
+  typedef Packet1cd half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 0,
+    size = 1,
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
+
+template<> EIGEN_STRONG_INLINE Packet1cd pload<Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(pload<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ploadu<Packet2d>((const double*)from)); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>&  from)
+{ /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(padd<Packet2d>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(psub<Packet2d>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate<Packet2d>(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd(vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(a.v), p2ul_CONJ_XOR))); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  Packet2d v1, v2;
+
+  // Get the real values of a 
+  v1 = vdupq_lane_f64(vget_low_f64(a.v), 0);
+  // Get the imag values of a
+  v2 = vdupq_lane_f64(vget_high_f64(a.v), 0);
+  // Multiply the real a with b
+  v1 = vmulq_f64(v1, b.v);
+  // Multiply the imag a with b
+  v2 = vmulq_f64(v2, b.v);
+  // Conjugate v2 
+  v2 = vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(v2), p2ul_CONJ_XOR));
+  // Swap real/imag elements in v2.
+  v2 = preverse<Packet2d>(v2);
+  // Add and return the result
+  return Packet1cd(vaddq_f64(v1, v2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  return Packet1cd(vreinterpretq_f64_u64(vandq_u64(vreinterpretq_u64_f64(a.v),vreinterpretq_u64_f64(b.v))));
+}
+template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  return Packet1cd(vreinterpretq_f64_u64(vorrq_u64(vreinterpretq_u64_f64(a.v),vreinterpretq_u64_f64(b.v))));
+}
+template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  return Packet1cd(vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(a.v),vreinterpretq_u64_f64(b.v))));
+}
+template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  return Packet1cd(vreinterpretq_f64_u64(vbicq_u64(vreinterpretq_u64_f64(a.v),vreinterpretq_u64_f64(b.v))));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>* from) { return pset1<Packet1cd>(*from); }
+
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { EIGEN_ARM_PREFETCH((const double *)addr); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index stride)
+{
+  Packet2d res = pset1<Packet2d>(0.0);
+  res = vsetq_lane_f64(std::real(from[0*stride]), res, 0);
+  res = vsetq_lane_f64(std::imag(from[0*stride]), res, 1);
+  return Packet1cd(res);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index stride)
+{
+  to[stride*0] = std::complex<double>(vgetq_lane_f64(from.v, 0), vgetq_lane_f64(from.v, 1));
+}
+
+
+template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
+{
+  std::complex<double> EIGEN_ALIGN16 res;
+  pstore<std::complex<double> >(&res, a);
+
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs) { return vecs[0]; }
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
+
+template<int Offset>
+struct palign_impl<Offset,Packet1cd>
+{
+  static EIGEN_STRONG_INLINE void run(Packet1cd& /*first*/, const Packet1cd& /*second*/)
+  {
+    // FIXME is it sure we never have to align a Packet1cd?
+    // Even though a std::complex<double> has 16 bytes, it is not necessarily aligned on a 16 bytes boundary...
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,false>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
+
+template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  // TODO optimize it for NEON
+  Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
+  Packet2d s = pmul<Packet2d>(b.v, b.v);
+  Packet2d rev_s = preverse<Packet2d>(s);
+
+  return Packet1cd(pdiv(res.v, padd<Packet2d>(s,rev_s)));
+}
+
+EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
+{
+  return Packet1cd(preverse(Packet2d(x.v)));
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
+{
+  Packet2d tmp = vcombine_f64(vget_high_f64(kernel.packet[0].v), vget_high_f64(kernel.packet[1].v));
+  kernel.packet[0].v = vcombine_f64(vget_low_f64(kernel.packet[0].v), vget_low_f64(kernel.packet[1].v));
+  kernel.packet[1].v = tmp;
 }
+#endif // EIGEN_ARCH_ARM64
 
 } // end namespace internal
 
diff --git a/extern/Eigen3/Eigen/src/Core/arch/NEON/MathFunctions.h b/extern/Eigen3/Eigen/src/Core/arch/NEON/MathFunctions.h
new file mode 100644
index 00000000000..6bb05bb922a
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/NEON/MathFunctions.h
@@ -0,0 +1,91 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+/* The sin, cos, exp, and log functions of this file come from
+ * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
+ */
+
+#ifndef EIGEN_MATH_FUNCTIONS_NEON_H
+#define EIGEN_MATH_FUNCTIONS_NEON_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f pexp<Packet4f>(const Packet4f& _x)
+{
+  Packet4f x = _x;
+  Packet4f tmp, fx;
+
+  _EIGEN_DECLARE_CONST_Packet4f(1 , 1.0f);
+  _EIGEN_DECLARE_CONST_Packet4f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet4i(0x7f, 0x7f);
+  _EIGEN_DECLARE_CONST_Packet4f(exp_hi,  88.3762626647950f);
+  _EIGEN_DECLARE_CONST_Packet4f(exp_lo, -88.3762626647949f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_LOG2EF, 1.44269504088896341f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C1, 0.693359375f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C2, -2.12194440e-4f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p0, 1.9875691500E-4f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p1, 1.3981999507E-3f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p2, 8.3334519073E-3f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p3, 4.1665795894E-2f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p4, 1.6666665459E-1f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p5, 5.0000001201E-1f);
+
+  x = vminq_f32(x, p4f_exp_hi);
+  x = vmaxq_f32(x, p4f_exp_lo);
+
+  /* express exp(x) as exp(g + n*log(2)) */
+  fx = vmlaq_f32(p4f_half, x, p4f_cephes_LOG2EF);
+
+  /* perform a floorf */
+  tmp = vcvtq_f32_s32(vcvtq_s32_f32(fx));
+
+  /* if greater, substract 1 */
+  Packet4ui mask = vcgtq_f32(tmp, fx);
+  mask = vandq_u32(mask, vreinterpretq_u32_f32(p4f_1));
+
+  fx = vsubq_f32(tmp, vreinterpretq_f32_u32(mask));
+
+  tmp = vmulq_f32(fx, p4f_cephes_exp_C1);
+  Packet4f z = vmulq_f32(fx, p4f_cephes_exp_C2);
+  x = vsubq_f32(x, tmp);
+  x = vsubq_f32(x, z);
+
+  Packet4f y = vmulq_f32(p4f_cephes_exp_p0, x);
+  z = vmulq_f32(x, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p1);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p2);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p3);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p4);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p5);
+
+  y = vmulq_f32(y, z);
+  y = vaddq_f32(y, x);
+  y = vaddq_f32(y, p4f_1);
+
+  /* build 2^n */
+  int32x4_t mm;
+  mm = vcvtq_s32_f32(fx);
+  mm = vaddq_s32(mm, p4i_0x7f);
+  mm = vshlq_n_s32(mm, 23);
+  Packet4f pow2n = vreinterpretq_f32_s32(mm);
+
+  y = vmulq_f32(y, pow2n);
+  return y;
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_MATH_FUNCTIONS_NEON_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h b/extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h
index d49670e0410..3d5ed0d240c 100644
--- a/extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h
+++ b/extern/Eigen3/Eigen/src/Core/arch/NEON/PacketMath.h
@@ -2,7 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
-// Copyright (C) 2010 Konstantinos Margaritis <markos@codex.gr>
+// Copyright (C) 2010 Konstantinos Margaritis <markos@freevec.org>
 // Heavily based on Gael's SSE version.
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -20,43 +20,79 @@ namespace internal {
 #define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 8
 #endif
 
-// FIXME NEON has 16 quad registers, but since the current register allocator
-// is so bad, it is much better to reduce it to 8
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#endif
+
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#endif
+
 #ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
-#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 8
+#if EIGEN_ARCH_ARM64
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 32
+#else
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 16 
 #endif
+#endif
+
+#if EIGEN_COMP_MSVC
+
+// In MSVC's arm_neon.h header file, all NEON vector types
+// are aliases to the same underlying type __n128.
+// We thus have to wrap them to make them different C++ types.
+// (See also bug 1428)
 
+template<typename T,int unique_id>
+struct eigen_packet_wrapper
+{
+  operator T&() { return m_val; }
+  operator const T&() const { return m_val; }
+  eigen_packet_wrapper() {}
+  eigen_packet_wrapper(const T &v) : m_val(v) {}
+  eigen_packet_wrapper& operator=(const T &v) {
+    m_val = v;
+    return *this;
+  }
+
+  T m_val;
+};
+typedef eigen_packet_wrapper<float32x2_t,0> Packet2f;
+typedef eigen_packet_wrapper<float32x4_t,1> Packet4f;
+typedef eigen_packet_wrapper<int32x4_t  ,2> Packet4i;
+typedef eigen_packet_wrapper<int32x2_t  ,3> Packet2i;
+typedef eigen_packet_wrapper<uint32x4_t ,4> Packet4ui;
+
+#else
+
+typedef float32x2_t Packet2f;
 typedef float32x4_t Packet4f;
 typedef int32x4_t   Packet4i;
+typedef int32x2_t   Packet2i;
 typedef uint32x4_t  Packet4ui;
 
+#endif // EIGEN_COMP_MSVC
+
 #define _EIGEN_DECLARE_CONST_Packet4f(NAME,X) \
   const Packet4f p4f_##NAME = pset1<Packet4f>(X)
 
 #define _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(NAME,X) \
-  const Packet4f p4f_##NAME = vreinterpretq_f32_u32(pset1<int>(X))
+  const Packet4f p4f_##NAME = vreinterpretq_f32_u32(pset1<int32_t>(X))
 
 #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
   const Packet4i p4i_##NAME = pset1<Packet4i>(X)
 
-#if defined(__llvm__) && !defined(__clang__)
-  //Special treatment for Apple's llvm-gcc, its NEON packet types are unions
-  #define EIGEN_INIT_NEON_PACKET2(X, Y)       {{X, Y}}
-  #define EIGEN_INIT_NEON_PACKET4(X, Y, Z, W) {{X, Y, Z, W}}
-#else
-  //Default initializer for packets
-  #define EIGEN_INIT_NEON_PACKET2(X, Y)       {X, Y}
-  #define EIGEN_INIT_NEON_PACKET4(X, Y, Z, W) {X, Y, Z, W}
-#endif
-
-// arm64 does have the pld instruction. If available, let's trust the __builtin_prefetch built-in function
-// which available on LLVM and GCC (at least)
-#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || defined(__GNUC__)
+#if EIGEN_ARCH_ARM64
+  // __builtin_prefetch tends to do nothing on ARM64 compilers because the
+  // prefetch instructions there are too detailed for __builtin_prefetch to map
+  // meaningfully to them.
+  #define EIGEN_ARM_PREFETCH(ADDR)  __asm__ __volatile__("prfm pldl1keep, [%[addr]]\n" ::[addr] "r"(ADDR) : );
+#elif EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC
   #define EIGEN_ARM_PREFETCH(ADDR) __builtin_prefetch(ADDR);
 #elif defined __pld
   #define EIGEN_ARM_PREFETCH(ADDR) __pld(ADDR)
-#elif !defined(__aarch64__)
-  #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ( "   pld [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
+#elif EIGEN_ARCH_ARM32
+  #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ("pld [%[addr]]\n" :: [addr] "r" (ADDR) : );
 #else
   // by default no explicit prefetching
   #define EIGEN_ARM_PREFETCH(ADDR)
@@ -65,53 +101,60 @@ typedef uint32x4_t  Packet4ui;
 template<> struct packet_traits<float>  : default_packet_traits
 {
   typedef Packet4f type;
+  typedef Packet4f half; // Packet2f intrinsics not implemented yet
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 4,
+    HasHalfPacket=0, // Packet2f intrinsics not implemented yet
    
     HasDiv  = 1,
     // FIXME check the Has*
     HasSin  = 0,
     HasCos  = 0,
     HasLog  = 0,
-    HasExp  = 0,
+    HasExp  = 1,
     HasSqrt = 0
   };
 };
-template<> struct packet_traits<int>    : default_packet_traits
+template<> struct packet_traits<int32_t>    : default_packet_traits
 {
   typedef Packet4i type;
+  typedef Packet4i half; // Packet2i intrinsics not implemented yet
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=4
+    size=4,
+    HasHalfPacket=0 // Packet2i intrinsics not implemented yet
     // FIXME check the Has*
   };
 };
 
-#if EIGEN_GNUC_AT_MOST(4,4) && !defined(__llvm__)
+#if EIGEN_GNUC_AT_MOST(4,4) && !EIGEN_COMP_LLVM
 // workaround gcc 4.2, 4.3 and 4.4 compilatin issue
 EIGEN_STRONG_INLINE float32x4_t vld1q_f32(const float* x) { return ::vld1q_f32((const float32_t*)x); }
 EIGEN_STRONG_INLINE float32x2_t vld1_f32 (const float* x) { return ::vld1_f32 ((const float32_t*)x); }
+EIGEN_STRONG_INLINE float32x2_t vld1_dup_f32 (const float* x) { return ::vld1_dup_f32 ((const float32_t*)x); }
 EIGEN_STRONG_INLINE void        vst1q_f32(float* to, float32x4_t from) { ::vst1q_f32((float32_t*)to,from); }
 EIGEN_STRONG_INLINE void        vst1_f32 (float* to, float32x2_t from) { ::vst1_f32 ((float32_t*)to,from); }
 #endif
 
-template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4}; };
-template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4}; };
+template<> struct unpacket_traits<Packet4f> { typedef float   type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
+template<> struct unpacket_traits<Packet4i> { typedef int32_t type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
 
 template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) { return vdupq_n_f32(from); }
-template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)   { return vdupq_n_s32(from); }
+template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int32_t&    from)   { return vdupq_n_s32(from); }
 
-template<> EIGEN_STRONG_INLINE Packet4f plset<float>(const float& a)
+template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a)
 {
-  Packet4f countdown = EIGEN_INIT_NEON_PACKET4(0, 1, 2, 3);
+  const float f[] = {0, 1, 2, 3};
+  Packet4f countdown = vld1q_f32(f);
   return vaddq_f32(pset1<Packet4f>(a), countdown);
 }
-template<> EIGEN_STRONG_INLINE Packet4i plset<int>(const int& a)
+template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int32_t& a)
 {
-  Packet4i countdown = EIGEN_INIT_NEON_PACKET4(0, 1, 2, 3);
+  const int32_t i[] = {0, 1, 2, 3};
+  Packet4i countdown = vld1q_s32(i);
   return vaddq_s32(pset1<Packet4i>(a), countdown);
 }
 
@@ -132,6 +175,9 @@ template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const
 
 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b)
 {
+#if EIGEN_ARCH_ARM64
+  return vdivq_f32(a,b);
+#else
   Packet4f inv, restep, div;
 
   // NEON does not offer a divide instruction, we have to do a reciprocal approximation
@@ -150,14 +196,51 @@ template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const
   div = vmulq_f32(a, inv);
 
   return div;
+#endif
 }
+
 template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& /*a*/, const Packet4i& /*b*/)
 { eigen_assert(false && "packet integer division are not supported by NEON");
   return pset1<Packet4i>(0);
 }
 
-// for some weird raisons, it has to be overloaded for packet of integers
-template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return vmlaq_f32(c,a,b); }
+// Clang/ARM wrongly advertises __ARM_FEATURE_FMA even when it's not available,
+// then implements a slow software scalar fallback calling fmaf()!
+// Filed LLVM bug:
+//     https://llvm.org/bugs/show_bug.cgi?id=27216
+#if (defined __ARM_FEATURE_FMA) && !(EIGEN_COMP_CLANG && EIGEN_ARCH_ARM)
+// See bug 936.
+// FMA is available on VFPv4 i.e. when compiling with -mfpu=neon-vfpv4.
+// FMA is a true fused multiply-add i.e. only 1 rounding at the end, no intermediate rounding.
+// MLA is not fused i.e. does 2 roundings.
+// In addition to giving better accuracy, FMA also gives better performance here on a Krait (Nexus 4):
+// MLA: 10 GFlop/s ; FMA: 12 GFlops/s.
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return vfmaq_f32(c,a,b); }
+#else
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) {
+#if EIGEN_COMP_CLANG && EIGEN_ARCH_ARM
+  // Clang/ARM will replace VMLA by VMUL+VADD at least for some values of -mcpu,
+  // at least -mcpu=cortex-a8 and -mcpu=cortex-a7. Since the former is the default on
+  // -march=armv7-a, that is a very common case.
+  // See e.g. this thread:
+  //     http://lists.llvm.org/pipermail/llvm-dev/2013-December/068806.html
+  // Filed LLVM bug:
+  //     https://llvm.org/bugs/show_bug.cgi?id=27219
+  Packet4f r = c;
+  asm volatile(
+    "vmla.f32 %q[r], %q[a], %q[b]"
+    : [r] "+w" (r)
+    : [a] "w" (a),
+      [b] "w" (b)
+    : );
+  return r;
+#else
+  return vmlaq_f32(c,a,b);
+#endif
+}
+#endif
+
+// No FMA instruction for int, so use MLA unconditionally.
 template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return vmlaq_s32(c,a,b); }
 
 template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) { return vminq_f32(a,b); }
@@ -191,20 +274,20 @@ template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, con
 }
 template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return vbicq_s32(a,b); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_f32(from); }
-template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*   from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_s32(from); }
+template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float*    from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_f32(from); }
+template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int32_t*  from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_s32(from); }
 
-template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_f32(from); }
-template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)   { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_s32(from); }
+template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float*   from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_f32(from); }
+template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int32_t* from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_s32(from); }
 
-template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*   from)
+template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from)
 {
   float32x2_t lo, hi;
   lo = vld1_dup_f32(from);
   hi = vld1_dup_f32(from+1);
   return vcombine_f32(lo, hi);
 }
-template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
+template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int32_t* from)
 {
   int32x2_t lo, hi;
   lo = vld1_dup_s32(from);
@@ -212,18 +295,52 @@ template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
   return vcombine_s32(lo, hi);
 }
 
-template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_f32(to, from); }
-template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_s32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<float>  (float*    to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_f32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<int32_t>(int32_t*  to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_s32(to, from); }
 
-template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*  to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f32(to, from); }
-template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*      to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_s32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<float>  (float*   to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<int32_t>(int32_t* to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_s32(to, from); }
 
-template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { EIGEN_ARM_PREFETCH(addr); }
-template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*     addr) { EIGEN_ARM_PREFETCH(addr); }
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
+{
+  Packet4f res = pset1<Packet4f>(0.f);
+  res = vsetq_lane_f32(from[0*stride], res, 0);
+  res = vsetq_lane_f32(from[1*stride], res, 1);
+  res = vsetq_lane_f32(from[2*stride], res, 2);
+  res = vsetq_lane_f32(from[3*stride], res, 3);
+  return res;
+}
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int32_t, Packet4i>(const int32_t* from, Index stride)
+{
+  Packet4i res = pset1<Packet4i>(0);
+  res = vsetq_lane_s32(from[0*stride], res, 0);
+  res = vsetq_lane_s32(from[1*stride], res, 1);
+  res = vsetq_lane_s32(from[2*stride], res, 2);
+  res = vsetq_lane_s32(from[3*stride], res, 3);
+  return res;
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
+{
+  to[stride*0] = vgetq_lane_f32(from, 0);
+  to[stride*1] = vgetq_lane_f32(from, 1);
+  to[stride*2] = vgetq_lane_f32(from, 2);
+  to[stride*3] = vgetq_lane_f32(from, 3);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int32_t, Packet4i>(int32_t* to, const Packet4i& from, Index stride)
+{
+  to[stride*0] = vgetq_lane_s32(from, 0);
+  to[stride*1] = vgetq_lane_s32(from, 1);
+  to[stride*2] = vgetq_lane_s32(from, 2);
+  to[stride*3] = vgetq_lane_s32(from, 3);
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<float>  (const float*    addr) { EIGEN_ARM_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<int32_t>(const int32_t*  addr) { EIGEN_ARM_PREFETCH(addr); }
 
 // FIXME only store the 2 first elements ?
-template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x[4]; vst1q_f32(x, a); return x[0]; }
-template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int   EIGEN_ALIGN16 x[4]; vst1q_s32(x, a); return x[0]; }
+template<> EIGEN_STRONG_INLINE float   pfirst<Packet4f>(const Packet4f& a) { float   EIGEN_ALIGN16 x[4]; vst1q_f32(x, a); return x[0]; }
+template<> EIGEN_STRONG_INLINE int32_t pfirst<Packet4i>(const Packet4i& a) { int32_t EIGEN_ALIGN16 x[4]; vst1q_s32(x, a); return x[0]; }
 
 template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a) {
   float32x2_t a_lo, a_hi;
@@ -243,6 +360,7 @@ template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a) {
   a_hi = vget_high_s32(a_r64);
   return vcombine_s32(a_hi, a_lo);
 }
+
 template<> EIGEN_STRONG_INLINE Packet4f pabs(const Packet4f& a) { return vabsq_f32(a); }
 template<> EIGEN_STRONG_INLINE Packet4i pabs(const Packet4i& a) { return vabsq_s32(a); }
 
@@ -277,7 +395,7 @@ template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
   return sum;
 }
 
-template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, sum;
 
@@ -324,7 +442,7 @@ template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
 
   return vget_lane_f32(prod, 0);
 }
-template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux_mul<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, prod;
 
@@ -352,7 +470,7 @@ template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a)
   return vget_lane_f32(min, 0);
 }
 
-template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux_min<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, min;
 
@@ -377,7 +495,7 @@ template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a)
   return vget_lane_f32(max, 0);
 }
 
-template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux_max<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, max;
 
@@ -410,9 +528,231 @@ PALIGN_NEON(0,Packet4i,vextq_s32)
 PALIGN_NEON(1,Packet4i,vextq_s32)
 PALIGN_NEON(2,Packet4i,vextq_s32)
 PALIGN_NEON(3,Packet4i,vextq_s32)
-    
+
 #undef PALIGN_NEON
 
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4f,4>& kernel) {
+  float32x4x2_t tmp1 = vzipq_f32(kernel.packet[0], kernel.packet[1]);
+  float32x4x2_t tmp2 = vzipq_f32(kernel.packet[2], kernel.packet[3]);
+
+  kernel.packet[0] = vcombine_f32(vget_low_f32(tmp1.val[0]), vget_low_f32(tmp2.val[0]));
+  kernel.packet[1] = vcombine_f32(vget_high_f32(tmp1.val[0]), vget_high_f32(tmp2.val[0]));
+  kernel.packet[2] = vcombine_f32(vget_low_f32(tmp1.val[1]), vget_low_f32(tmp2.val[1]));
+  kernel.packet[3] = vcombine_f32(vget_high_f32(tmp1.val[1]), vget_high_f32(tmp2.val[1]));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4i,4>& kernel) {
+  int32x4x2_t tmp1 = vzipq_s32(kernel.packet[0], kernel.packet[1]);
+  int32x4x2_t tmp2 = vzipq_s32(kernel.packet[2], kernel.packet[3]);
+  kernel.packet[0] = vcombine_s32(vget_low_s32(tmp1.val[0]), vget_low_s32(tmp2.val[0]));
+  kernel.packet[1] = vcombine_s32(vget_high_s32(tmp1.val[0]), vget_high_s32(tmp2.val[0]));
+  kernel.packet[2] = vcombine_s32(vget_low_s32(tmp1.val[1]), vget_low_s32(tmp2.val[1]));
+  kernel.packet[3] = vcombine_s32(vget_high_s32(tmp1.val[1]), vget_high_s32(tmp2.val[1]));
+}
+
+//---------- double ----------
+
+// Clang 3.5 in the iOS toolchain has an ICE triggered by NEON intrisics for double.
+// Confirmed at least with __apple_build_version__ = 6000054.
+#ifdef __apple_build_version__
+// Let's hope that by the time __apple_build_version__ hits the 601* range, the bug will be fixed.
+// https://gist.github.com/yamaya/2924292 suggests that the 3 first digits are only updated with
+// major toolchain updates.
+#define EIGEN_APPLE_DOUBLE_NEON_BUG (__apple_build_version__ < 6010000)
+#else
+#define EIGEN_APPLE_DOUBLE_NEON_BUG 0
+#endif
+
+#if EIGEN_ARCH_ARM64 && !EIGEN_APPLE_DOUBLE_NEON_BUG
+
+// Bug 907: workaround missing declarations of the following two functions in the ADK
+// Defining these functions as templates ensures that if these intrinsics are
+// already defined in arm_neon.h, then our workaround doesn't cause a conflict
+// and has lower priority in overload resolution.
+template <typename T>
+uint64x2_t vreinterpretq_u64_f64(T a)
+{
+  return (uint64x2_t) a;
+}
+
+template <typename T>
+float64x2_t vreinterpretq_f64_u64(T a)
+{
+  return (float64x2_t) a;
+}
+
+typedef float64x2_t Packet2d;
+typedef float64x1_t Packet1d;
+
+template<> struct packet_traits<double>  : default_packet_traits
+{
+  typedef Packet2d type;
+  typedef Packet2d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 2,
+    HasHalfPacket=0,
+   
+    HasDiv  = 1,
+    // FIXME check the Has*
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 0,
+    HasExp  = 0,
+    HasSqrt = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet2d> { typedef double  type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
+
+template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double&  from) { return vdupq_n_f64(from); }
+
+template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a)
+{
+  const double countdown_raw[] = {0.0,1.0};
+  const Packet2d countdown = vld1q_f64(countdown_raw);
+  return vaddq_f64(pset1<Packet2d>(a), countdown);
+}
+template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return vaddq_f64(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return vsubq_f64(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return vnegq_f64(a); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return vmulq_f64(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return vdivq_f64(a,b); }
+
+#ifdef __ARM_FEATURE_FMA
+// See bug 936. See above comment about FMA for float.
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vfmaq_f64(c,a,b); }
+#else
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vmlaq_f64(c,a,b); }
+#endif
+
+template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b) { return vminq_f64(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmax<Packet2d>(const Packet2d& a, const Packet2d& b) { return vmaxq_f64(a,b); }
+
+// Logical Operations are not supported for float, so we have to reinterpret casts using NEON intrinsics
+template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  return vreinterpretq_f64_u64(vandq_u64(vreinterpretq_u64_f64(a),vreinterpretq_u64_f64(b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d por<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  return vreinterpretq_f64_u64(vorrq_u64(vreinterpretq_u64_f64(a),vreinterpretq_u64_f64(b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  return vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(a),vreinterpretq_u64_f64(b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  return vreinterpretq_f64_u64(vbicq_u64(vreinterpretq_u64_f64(a),vreinterpretq_u64_f64(b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_f64(from); }
+
+template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_f64(from); }
+
+template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
+{
+  return vld1q_dup_f64(from);
+}
+template<> EIGEN_STRONG_INLINE void pstore<double>(double*   to, const Packet2d& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_f64(to, from); }
+
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double*  to, const Packet2d& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f64(to, from); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride)
+{
+  Packet2d res = pset1<Packet2d>(0.0);
+  res = vsetq_lane_f64(from[0*stride], res, 0);
+  res = vsetq_lane_f64(from[1*stride], res, 1);
+  return res;
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride)
+{
+  to[stride*0] = vgetq_lane_f64(from, 0);
+  to[stride*1] = vgetq_lane_f64(from, 1);
+}
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_ARM_PREFETCH(addr); }
+
+// FIXME only store the 2 first elements ?
+template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { return vgetq_lane_f64(a, 0); }
+
+template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a) { return vcombine_f64(vget_high_f64(a), vget_low_f64(a)); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pabs(const Packet2d& a) { return vabsq_f64(a); }
+
+#if EIGEN_COMP_CLANG && defined(__apple_build_version__)
+// workaround ICE, see bug 907
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) { return (vget_low_f64(a) + vget_high_f64(a))[0]; }
+#else
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) { return vget_lane_f64(vget_low_f64(a) + vget_high_f64(a), 0); }
+#endif
+
+template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
+{
+  float64x2_t trn1, trn2;
+
+  // NEON zip performs interleaving of the supplied vectors.
+  // We perform two interleaves in a row to acquire the transposed vector
+  trn1 = vzip1q_f64(vecs[0], vecs[1]);
+  trn2 = vzip2q_f64(vecs[0], vecs[1]);
+
+  // Do the addition of the resulting vectors
+  return vaddq_f64(trn1, trn2);
+}
+// Other reduction functions:
+// mul
+#if EIGEN_COMP_CLANG && defined(__apple_build_version__)
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a) { return (vget_low_f64(a) * vget_high_f64(a))[0]; }
+#else
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a) { return vget_lane_f64(vget_low_f64(a) * vget_high_f64(a), 0); }
+#endif
+
+// min
+template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a) { return vgetq_lane_f64(vpminq_f64(a, a), 0); }
+
+// max
+template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a) { return vgetq_lane_f64(vpmaxq_f64(a, a), 0); }
+
+// this PALIGN_NEON business is to work around a bug in LLVM Clang 3.0 causing incorrect compilation errors,
+// see bug 347 and this LLVM bug: http://llvm.org/bugs/show_bug.cgi?id=11074
+#define PALIGN_NEON(Offset,Type,Command) \
+template<>\
+struct palign_impl<Offset,Type>\
+{\
+    EIGEN_STRONG_INLINE static void run(Type& first, const Type& second)\
+    {\
+        if (Offset!=0)\
+            first = Command(first, second, Offset);\
+    }\
+};\
+
+PALIGN_NEON(0,Packet2d,vextq_f64)
+PALIGN_NEON(1,Packet2d,vextq_f64)
+#undef PALIGN_NEON
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2d,2>& kernel) {
+  float64x2_t trn1 = vzip1q_f64(kernel.packet[0], kernel.packet[1]);
+  float64x2_t trn2 = vzip2q_f64(kernel.packet[0], kernel.packet[1]);
+
+  kernel.packet[0] = trn1;
+  kernel.packet[1] = trn2;
+}
+#endif // EIGEN_ARCH_ARM64 
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/arch/SSE/Complex.h b/extern/Eigen3/Eigen/src/Core/arch/SSE/Complex.h
index 91bba5e38fe..d075043ce1b 100644
--- a/extern/Eigen3/Eigen/src/Core/arch/SSE/Complex.h
+++ b/extern/Eigen3/Eigen/src/Core/arch/SSE/Complex.h
@@ -22,13 +22,18 @@ struct Packet2cf
   __m128  v;
 };
 
+// Use the packet_traits defined in AVX/PacketMath.h instead if we're going
+// to leverage AVX instructions.
+#ifndef EIGEN_VECTORIZE_AVX
 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
 {
   typedef Packet2cf type;
+  typedef Packet2cf half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 2,
+    HasHalfPacket = 0,
 
     HasAdd    = 1,
     HasSub    = 1,
@@ -39,11 +44,13 @@ template<> struct packet_traits<std::complex<float> >  : default_packet_traits
     HasAbs2   = 0,
     HasMin    = 0,
     HasMax    = 0,
-    HasSetLinear = 0
+    HasSetLinear = 0,
+    HasBlend = 1
   };
 };
+#endif
 
-template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; };
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
 
 template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_add_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_sub_ps(a.v,b.v)); }
@@ -60,7 +67,6 @@ template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
 
 template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
-  // TODO optimize it for SSE3 and 4
   #ifdef EIGEN_VECTORIZE_SSE3
   return Packet2cf(_mm_addsub_ps(_mm_mul_ps(_mm_moveldup_ps(a.v), b.v),
                                  _mm_mul_ps(_mm_movehdup_ps(a.v),
@@ -104,10 +110,25 @@ template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<flo
 
 template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>* from) { return pset1<Packet2cf>(*from); }
 
-template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), from.v); }
-template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), from.v); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), Packet4f(from.v)); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), Packet4f(from.v)); }
 
-template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
+{
+  return Packet2cf(_mm_set_ps(std::imag(from[1*stride]), std::real(from[1*stride]),
+                              std::imag(from[0*stride]), std::real(from[0*stride])));
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
+{
+  to[stride*0] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 0)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 1)));
+  to[stride*1] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 2)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 3)));
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
 
 template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
 {
@@ -124,7 +145,7 @@ template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Pack
   #endif
 }
 
-template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a) { return Packet2cf(_mm_castpd_ps(preverse(_mm_castps_pd(a.v)))); }
+template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a) { return Packet2cf(_mm_castpd_ps(preverse(Packet2d(_mm_castps_pd(a.v))))); }
 
 template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
 {
@@ -208,23 +229,7 @@ template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
   }
 };
 
-template<> struct conj_helper<Packet4f, Packet2cf, false,false>
-{
-  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet4f& x, const Packet2cf& y, const Packet2cf& c) const
-  { return padd(c, pmul(x,y)); }
-
-  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet4f& x, const Packet2cf& y) const
-  { return Packet2cf(Eigen::internal::pmul(x, y.v)); }
-};
-
-template<> struct conj_helper<Packet2cf, Packet4f, false,false>
-{
-  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet4f& y, const Packet2cf& c) const
-  { return padd(c, pmul(x,y)); }
-
-  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& x, const Packet4f& y) const
-  { return Packet2cf(Eigen::internal::pmul(x.v, y)); }
-};
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
 
 template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
@@ -234,7 +239,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, con
   return Packet2cf(_mm_div_ps(res.v,_mm_add_ps(s,_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(s), 0xb1)))));
 }
 
-EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
+EIGEN_STRONG_INLINE Packet2cf pcplxflip/* <Packet2cf> */(const Packet2cf& x)
 {
   return Packet2cf(vec4f_swizzle1(x.v, 1, 0, 3, 2));
 }
@@ -248,13 +253,18 @@ struct Packet1cd
   __m128d  v;
 };
 
+// Use the packet_traits defined in AVX/PacketMath.h instead if we're going
+// to leverage AVX instructions.
+#ifndef EIGEN_VECTORIZE_AVX
 template<> struct packet_traits<std::complex<double> >  : default_packet_traits
 {
   typedef Packet1cd type;
+  typedef Packet1cd half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 0,
     size = 1,
+    HasHalfPacket = 0,
 
     HasAdd    = 1,
     HasSub    = 1,
@@ -268,12 +278,13 @@ template<> struct packet_traits<std::complex<double> >  : default_packet_traits
     HasSetLinear = 0
   };
 };
+#endif
 
-template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1}; };
+template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
 
 template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_add_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_sub_pd(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
 template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a)
 {
   const __m128d mask = _mm_castsi128_pd(_mm_set_epi32(0x80000000,0x0,0x0,0x0));
@@ -282,9 +293,8 @@ template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a)
 
 template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
 {
-  // TODO optimize it for SSE3 and 4
   #ifdef EIGEN_VECTORIZE_SSE3
-  return Packet1cd(_mm_addsub_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 0, 0), b.v),
+  return Packet1cd(_mm_addsub_pd(_mm_mul_pd(_mm_movedup_pd(a.v), b.v),
                                  _mm_mul_pd(vec2d_swizzle1(a.v, 1, 1),
                                             vec2d_swizzle1(b.v, 1, 0))));
   #else
@@ -311,10 +321,10 @@ template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<dou
 template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>* from) { return pset1<Packet1cd>(*from); }
 
 // FIXME force unaligned store, this is a temporary fix
-template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
-template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, Packet2d(from.v)); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, Packet2d(from.v)); }
 
-template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
 
 template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
 {
@@ -404,23 +414,7 @@ template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
   }
 };
 
-template<> struct conj_helper<Packet2d, Packet1cd, false,false>
-{
-  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet2d& x, const Packet1cd& y, const Packet1cd& c) const
-  { return padd(c, pmul(x,y)); }
-
-  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet2d& x, const Packet1cd& y) const
-  { return Packet1cd(Eigen::internal::pmul(x, y.v)); }
-};
-
-template<> struct conj_helper<Packet1cd, Packet2d, false,false>
-{
-  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet2d& y, const Packet1cd& c) const
-  { return padd(c, pmul(x,y)); }
-
-  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& x, const Packet2d& y) const
-  { return Packet1cd(Eigen::internal::pmul(x.v, y)); }
-};
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
 
 template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
 {
@@ -430,9 +424,44 @@ template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, con
   return Packet1cd(_mm_div_pd(res.v, _mm_add_pd(s,_mm_shuffle_pd(s, s, 0x1))));
 }
 
-EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
+EIGEN_STRONG_INLINE Packet1cd pcplxflip/* <Packet1cd> */(const Packet1cd& x)
+{
+  return Packet1cd(preverse(Packet2d(x.v)));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2cf,2>& kernel) {
+  __m128d w1 = _mm_castps_pd(kernel.packet[0].v);
+  __m128d w2 = _mm_castps_pd(kernel.packet[1].v);
+
+  __m128 tmp = _mm_castpd_ps(_mm_unpackhi_pd(w1, w2));
+  kernel.packet[0].v = _mm_castpd_ps(_mm_unpacklo_pd(w1, w2));
+  kernel.packet[1].v = tmp;
+}
+
+template<>  EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
+  __m128d result = pblend<Packet2d>(ifPacket, _mm_castps_pd(thenPacket.v), _mm_castps_pd(elsePacket.v));
+  return Packet2cf(_mm_castpd_ps(result));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pinsertfirst(const Packet2cf& a, std::complex<float> b)
+{
+  return Packet2cf(_mm_loadl_pi(a.v, reinterpret_cast<const __m64*>(&b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pinsertfirst(const Packet1cd&, std::complex<double> b)
+{
+  return pset1<Packet1cd>(b);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pinsertlast(const Packet2cf& a, std::complex<float> b)
+{
+  return Packet2cf(_mm_loadh_pi(a.v, reinterpret_cast<const __m64*>(&b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pinsertlast(const Packet1cd&, std::complex<double> b)
 {
-  return Packet1cd(preverse(x.v));
+  return pset1<Packet1cd>(b);
 }
 
 } // end namespace internal
diff --git a/extern/Eigen3/Eigen/src/Core/arch/SSE/MathFunctions.h b/extern/Eigen3/Eigen/src/Core/arch/SSE/MathFunctions.h
index 2b07168ae4f..7b5f948e119 100644
--- a/extern/Eigen3/Eigen/src/Core/arch/SSE/MathFunctions.h
+++ b/extern/Eigen3/Eigen/src/Core/arch/SSE/MathFunctions.h
@@ -32,7 +32,7 @@ Packet4f plog<Packet4f>(const Packet4f& _x)
   /* the smallest non denormalized float number */
   _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(min_norm_pos,  0x00800000);
   _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(minus_inf,     0xff800000);//-1.f/0.f);
-  
+
   /* natural logarithm computed for 4 simultaneous float
     return NaN for x <= 0
   */
@@ -63,7 +63,7 @@ Packet4f plog<Packet4f>(const Packet4f& _x)
   x = _mm_or_ps(x, p4f_half);
 
   emm0 = _mm_sub_epi32(emm0, p4i_0x7f);
-  Packet4f e = padd(_mm_cvtepi32_ps(emm0), p4f_1);
+  Packet4f e = padd(Packet4f(_mm_cvtepi32_ps(emm0)), p4f_1);
 
   /* part2:
      if( x < SQRTHF ) {
@@ -72,9 +72,9 @@ Packet4f plog<Packet4f>(const Packet4f& _x)
      } else { x = x - 1.0; }
   */
   Packet4f mask = _mm_cmplt_ps(x, p4f_cephes_SQRTHF);
-  Packet4f tmp = _mm_and_ps(x, mask);
+  Packet4f tmp = pand(x, mask);
   x = psub(x, p4f_1);
-  e = psub(e, _mm_and_ps(p4f_1, mask));
+  e = psub(e, pand(p4f_1, mask));
   x = padd(x, tmp);
 
   Packet4f x2 = pmul(x,x);
@@ -444,32 +444,119 @@ Packet4f pcos<Packet4f>(const Packet4f& _x)
 
 #if EIGEN_FAST_MATH
 
-// This is based on Quake3's fast inverse square root.
+// Functions for sqrt.
+// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step
+// of Newton's method, at a cost of 1-2 bits of precision as opposed to the
+// exact solution. It does not handle +inf, or denormalized numbers correctly.
+// The main advantage of this approach is not just speed, but also the fact that
+// it can be inlined and pipelined with other computations, further reducing its
+// effective latency. This is similar to Quake3's fast inverse square root.
 // For detail see here: http://www.beyond3d.com/content/articles/8/
-// It lacks 1 (or 2 bits in some rare cases) of precision, and does not handle negative, +inf, or denormalized numbers correctly.
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
 Packet4f psqrt<Packet4f>(const Packet4f& _x)
 {
   Packet4f half = pmul(_x, pset1<Packet4f>(.5f));
+  Packet4f denormal_mask = _mm_and_ps(
+      _mm_cmpge_ps(_x, _mm_setzero_ps()),
+      _mm_cmplt_ps(_x, pset1<Packet4f>((std::numeric_limits<float>::min)())));
 
-  /* select only the inverse sqrt of non-zero inputs */
-  Packet4f non_zero_mask = _mm_cmpge_ps(_x, pset1<Packet4f>((std::numeric_limits<float>::min)()));
-  Packet4f x = _mm_and_ps(non_zero_mask, _mm_rsqrt_ps(_x));
-
+  // Compute approximate reciprocal sqrt.
+  Packet4f x = _mm_rsqrt_ps(_x);
+  // Do a single step of Newton's iteration.
   x = pmul(x, psub(pset1<Packet4f>(1.5f), pmul(half, pmul(x,x))));
-  return pmul(_x,x);
+  // Flush results for denormals to zero.
+  return _mm_andnot_ps(denormal_mask, pmul(_x,x));
 }
 
 #else
 
-template<> EIGEN_STRONG_INLINE Packet4f psqrt<Packet4f>(const Packet4f& x) { return _mm_sqrt_ps(x); }
+template<>EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f psqrt<Packet4f>(const Packet4f& x) { return _mm_sqrt_ps(x); }
+
+#endif
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d psqrt<Packet2d>(const Packet2d& x) { return _mm_sqrt_pd(x); }
+
+#if EIGEN_FAST_MATH
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f prsqrt<Packet4f>(const Packet4f& _x) {
+  _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(inf, 0x7f800000);
+  _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(nan, 0x7fc00000);
+  _EIGEN_DECLARE_CONST_Packet4f(one_point_five, 1.5f);
+  _EIGEN_DECLARE_CONST_Packet4f(minus_half, -0.5f);
+  _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(flt_min, 0x00800000);
+
+  Packet4f neg_half = pmul(_x, p4f_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  Packet4f le_zero_mask = _mm_cmple_ps(_x, p4f_flt_min);
+  Packet4f x = _mm_andnot_ps(le_zero_mask, _mm_rsqrt_ps(_x));
+
+  // Fill in NaNs and Infs for the negative/zero entries.
+  Packet4f neg_mask = _mm_cmplt_ps(_x, _mm_setzero_ps());
+  Packet4f zero_mask = _mm_andnot_ps(neg_mask, le_zero_mask);
+  Packet4f infs_and_nans = _mm_or_ps(_mm_and_ps(neg_mask, p4f_nan),
+                                     _mm_and_ps(zero_mask, p4f_inf));
+
+  // Do a single step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p4f_one_point_five));
+
+  // Insert NaNs and Infs in all the right places.
+  return _mm_or_ps(x, infs_and_nans);
+}
+
+#else
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f prsqrt<Packet4f>(const Packet4f& x) {
+  // Unfortunately we can't use the much faster mm_rqsrt_ps since it only provides an approximation.
+  return _mm_div_ps(pset1<Packet4f>(1.0f), _mm_sqrt_ps(x));
+}
 
 #endif
 
-template<> EIGEN_STRONG_INLINE Packet2d psqrt<Packet2d>(const Packet2d& x) { return _mm_sqrt_pd(x); }
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d prsqrt<Packet2d>(const Packet2d& x) {
+  // Unfortunately we can't use the much faster mm_rqsrt_pd since it only provides an approximation.
+  return _mm_div_pd(pset1<Packet2d>(1.0), _mm_sqrt_pd(x));
+}
+
+// Hyperbolic Tangent function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+ptanh<Packet4f>(const Packet4f& x) {
+  return internal::generic_fast_tanh_float(x);
+}
 
 } // end namespace internal
 
+namespace numext {
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float sqrt(const float &x)
+{
+  return internal::pfirst(internal::Packet4f(_mm_sqrt_ss(_mm_set_ss(x))));
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double sqrt(const double &x)
+{
+#if EIGEN_COMP_GNUC_STRICT
+  // This works around a GCC bug generating poor code for _mm_sqrt_pd
+  // See https://bitbucket.org/eigen/eigen/commits/14f468dba4d350d7c19c9b93072e19f7b3df563b
+  return internal::pfirst(internal::Packet2d(__builtin_ia32_sqrtsd(_mm_set_sd(x))));
+#else
+  return internal::pfirst(internal::Packet2d(_mm_sqrt_pd(_mm_set_sd(x))));
+#endif
+}
+
+} // end namespace numex
+
 } // end namespace Eigen
 
 #endif // EIGEN_MATH_FUNCTIONS_SSE_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/SSE/PacketMath.h b/extern/Eigen3/Eigen/src/Core/arch/SSE/PacketMath.h
index cb535e1b78f..60e2517e4bd 100644
--- a/extern/Eigen3/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/extern/Eigen3/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -22,9 +22,40 @@ namespace internal {
 #define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS (2*sizeof(void*))
 #endif
 
+#ifdef __FMA__
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD 1
+#endif
+#endif
+
+#if ((defined EIGEN_VECTORIZE_AVX) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_MINGW) && (__GXX_ABI_VERSION < 1004)) || EIGEN_OS_QNX
+// With GCC's default ABI version, a __m128 or __m256 are the same types and therefore we cannot
+// have overloads for both types without linking error.
+// One solution is to increase ABI version using -fabi-version=4 (or greater).
+// Otherwise, we workaround this inconvenience by wrapping 128bit types into the following helper
+// structure:
+template<typename T>
+struct eigen_packet_wrapper
+{
+  EIGEN_ALWAYS_INLINE operator T&() { return m_val; }
+  EIGEN_ALWAYS_INLINE operator const T&() const { return m_val; }
+  EIGEN_ALWAYS_INLINE eigen_packet_wrapper() {}
+  EIGEN_ALWAYS_INLINE eigen_packet_wrapper(const T &v) : m_val(v) {}
+  EIGEN_ALWAYS_INLINE eigen_packet_wrapper& operator=(const T &v) {
+    m_val = v;
+    return *this;
+  }
+  
+  T m_val;
+};
+typedef eigen_packet_wrapper<__m128>  Packet4f;
+typedef eigen_packet_wrapper<__m128i> Packet4i;
+typedef eigen_packet_wrapper<__m128d> Packet2d;
+#else
 typedef __m128  Packet4f;
 typedef __m128i Packet4i;
 typedef __m128d Packet2d;
+#endif
 
 template<> struct is_arithmetic<__m128>  { enum { value = true }; };
 template<> struct is_arithmetic<__m128i> { enum { value = true }; };
@@ -58,51 +89,85 @@ template<> struct is_arithmetic<__m128d> { enum { value = true }; };
   const Packet4i p4i_##NAME = pset1<Packet4i>(X)
 
 
+// Use the packet_traits defined in AVX/PacketMath.h instead if we're going
+// to leverage AVX instructions.
+#ifndef EIGEN_VECTORIZE_AVX
 template<> struct packet_traits<float>  : default_packet_traits
 {
   typedef Packet4f type;
+  typedef Packet4f half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size=4,
+    HasHalfPacket = 0,
 
     HasDiv  = 1,
     HasSin  = EIGEN_FAST_MATH,
     HasCos  = EIGEN_FAST_MATH,
     HasLog  = 1,
     HasExp  = 1,
-    HasSqrt = 1
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasTanh  = EIGEN_FAST_MATH,
+    HasBlend = 1
+
+#ifdef EIGEN_VECTORIZE_SSE4_1
+    ,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+#endif
   };
 };
 template<> struct packet_traits<double> : default_packet_traits
 {
   typedef Packet2d type;
+  typedef Packet2d half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size=2,
+    HasHalfPacket = 0,
 
     HasDiv  = 1,
     HasExp  = 1,
-    HasSqrt = 1
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasBlend = 1
+
+#ifdef EIGEN_VECTORIZE_SSE4_1
+    ,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+#endif
   };
 };
+#endif
 template<> struct packet_traits<int>    : default_packet_traits
 {
   typedef Packet4i type;
+  typedef Packet4i half;
   enum {
-    // FIXME check the Has*
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=4
+    size=4,
+
+    HasBlend = 1
   };
 };
 
-template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4}; };
-template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2}; };
-template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4}; };
+template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
+template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
+template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
 
-#if defined(_MSC_VER) && (_MSC_VER==1500)
+#ifndef EIGEN_VECTORIZE_AVX
+template<> struct scalar_div_cost<float,true> { enum { value = 7 }; };
+template<> struct scalar_div_cost<double,true> { enum { value = 8 }; };
+#endif
+
+#if EIGEN_COMP_MSVC==1500
 // Workaround MSVC 9 internal compiler error.
 // TODO: It has been detected with win64 builds (amd64), so let's check whether it also happens in 32bits+SSE mode
 // TODO: let's check whether there does not exist a better fix, like adding a pset0() function. (it crashed on pset1(0)).
@@ -110,14 +175,25 @@ template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) { re
 template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) { return _mm_set_pd(from,from); }
 template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from) { return _mm_set_epi32(from,from,from,from); }
 #else
-template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) { return _mm_set1_ps(from); }
+template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) { return _mm_set_ps1(from); }
 template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) { return _mm_set1_pd(from); }
 template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from) { return _mm_set1_epi32(from); }
 #endif
 
-template<> EIGEN_STRONG_INLINE Packet4f plset<float>(const float& a) { return _mm_add_ps(pset1<Packet4f>(a), _mm_set_ps(3,2,1,0)); }
-template<> EIGEN_STRONG_INLINE Packet2d plset<double>(const double& a) { return _mm_add_pd(pset1<Packet2d>(a),_mm_set_pd(1,0)); }
-template<> EIGEN_STRONG_INLINE Packet4i plset<int>(const int& a) { return _mm_add_epi32(pset1<Packet4i>(a),_mm_set_epi32(3,2,1,0)); }
+// GCC generates a shufps instruction for _mm_set1_ps/_mm_load1_ps instead of the more efficient pshufd instruction.
+// However, using inrinsics for pset1 makes gcc to generate crappy code in some cases (see bug 203)
+// Using inline assembly is also not an option because then gcc fails to reorder properly the instructions.
+// Therefore, we introduced the pload1 functions to be used in product kernels for which bug 203 does not apply.
+// Also note that with AVX, we want it to generate a vbroadcastss.
+#if EIGEN_COMP_GNUC_STRICT && (!defined __AVX__)
+template<> EIGEN_STRONG_INLINE Packet4f pload1<Packet4f>(const float *from) {
+  return vec4f_swizzle1(_mm_load_ss(from),0,0,0,0);
+}
+#endif
+  
+template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a) { return _mm_add_ps(pset1<Packet4f>(a), _mm_set_ps(3,2,1,0)); }
+template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return _mm_add_pd(pset1<Packet2d>(a),_mm_set_pd(1,0)); }
+template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a) { return _mm_add_epi32(pset1<Packet4i>(a),_mm_set_epi32(3,2,1,0)); }
 
 template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_add_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_add_pd(a,b); }
@@ -139,7 +215,7 @@ template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a)
 }
 template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a)
 {
-  return psub(_mm_setr_epi32(0,0,0,0), a);
+  return psub(Packet4i(_mm_setr_epi32(0,0,0,0)), a);
 }
 
 template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; }
@@ -166,13 +242,13 @@ template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const
 
 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_div_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_div_pd(a,b); }
-template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& /*a*/, const Packet4i& /*b*/)
-{ eigen_assert(false && "packet integer division are not supported by SSE");
-  return pset1<Packet4i>(0);
-}
 
 // for some weird raisons, it has to be overloaded for packet of integers
 template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return padd(pmul(a,b), c); }
+#ifdef __FMA__
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return _mm_fmadd_ps(a,b,c); }
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return _mm_fmadd_pd(a,b,c); }
+#endif
 
 template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_min_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_min_pd(a,b); }
@@ -200,6 +276,17 @@ template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const
 #endif
 }
 
+#ifdef EIGEN_VECTORIZE_SSE4_1
+template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a) { return _mm_round_ps(a, 0); }
+template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return _mm_round_pd(a, 0); }
+
+template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const Packet4f& a) { return _mm_ceil_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const Packet2d& a) { return _mm_ceil_pd(a); }
+
+template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a) { return _mm_floor_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return _mm_floor_pd(a); }
+#endif
+
 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_and_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_and_pd(a,b); }
 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return _mm_and_si128(a,b); }
@@ -218,16 +305,14 @@ template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, con
 
 template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float*   from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_ps(from); }
 template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double*  from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_pd(from); }
-template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_si128(reinterpret_cast<const Packet4i*>(from)); }
+template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_si128(reinterpret_cast<const __m128i*>(from)); }
 
-#if defined(_MSC_VER)
+#if EIGEN_COMP_MSVC
   template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float*  from) {
     EIGEN_DEBUG_UNALIGNED_LOAD
-    #if (_MSC_VER==1600)
+    #if (EIGEN_COMP_MSVC==1600)
     // NOTE Some version of MSVC10 generates bad code when using _mm_loadu_ps
     // (i.e., it does not generate an unaligned load!!
-    // TODO On most architectures this version should also be faster than a single _mm_loadu_ps
-    // so we could also enable it for MSVC08 but first we have to make this later does not generate crap when doing so...
     __m128 res = _mm_loadl_pi(_mm_set1_ps(0.0f), (const __m64*)(from));
     res = _mm_loadh_pi(res, (const __m64*)(from+2));
     return res;
@@ -235,63 +320,27 @@ template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from) { E
     return _mm_loadu_ps(from);
     #endif
   }
-  template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_pd(from); }
-  template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int*    from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_si128(reinterpret_cast<const Packet4i*>(from)); }
 #else
-// Fast unaligned loads. Note that here we cannot directly use intrinsics: this would
-// require pointer casting to incompatible pointer types and leads to invalid code
-// because of the strict aliasing rule. The "dummy" stuff are required to enforce
-// a correct instruction dependency.
-// TODO: do the same for MSVC (ICC is compatible)
 // NOTE: with the code below, MSVC's compiler crashes!
 
-#if defined(__GNUC__) && defined(__i386__)
-  // bug 195: gcc/i386 emits weird x87 fldl/fstpl instructions for _mm_load_sd
-  #define EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS 1
-#elif defined(__clang__)
-  // bug 201: Segfaults in __mm_loadh_pd with clang 2.8
-  #define EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS 1
-#else
-  #define EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS 0
-#endif
-
 template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
 {
   EIGEN_DEBUG_UNALIGNED_LOAD
-#if EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS
   return _mm_loadu_ps(from);
-#else
-  __m128d res;
-  res =  _mm_load_sd((const double*)(from)) ;
-  res =  _mm_loadh_pd(res, (const double*)(from+2)) ;
-  return _mm_castpd_ps(res);
-#endif
 }
+#endif
+
 template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
 {
   EIGEN_DEBUG_UNALIGNED_LOAD
-#if EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS
   return _mm_loadu_pd(from);
-#else
-  __m128d res;
-  res = _mm_load_sd(from) ;
-  res = _mm_loadh_pd(res,from+1);
-  return res;
-#endif
 }
 template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
 {
   EIGEN_DEBUG_UNALIGNED_LOAD
-#if EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS
-  return _mm_loadu_si128(reinterpret_cast<const Packet4i*>(from));
-#else
-  __m128d res;
-  res =  _mm_load_sd((const double*)(from)) ;
-  res =  _mm_loadh_pd(res, (const double*)(from+2)) ;
-  return _mm_castpd_si128(res);
-#endif
+  return _mm_loadu_si128(reinterpret_cast<const __m128i*>(from));
 }
-#endif
+
 
 template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*   from)
 {
@@ -302,46 +351,83 @@ template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*  from)
 template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
 {
   Packet4i tmp;
-  tmp = _mm_loadl_epi64(reinterpret_cast<const Packet4i*>(from));
+  tmp = _mm_loadl_epi64(reinterpret_cast<const __m128i*>(from));
   return vec4i_swizzle1(tmp, 0, 0, 1, 1);
 }
 
 template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_ps(to, from); }
 template<> EIGEN_STRONG_INLINE void pstore<double>(double* to, const Packet2d& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_pd(to, from); }
-template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_si128(reinterpret_cast<Packet4i*>(to), from); }
+template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_si128(reinterpret_cast<__m128i*>(to), from); }
+
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet2d& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_storeu_pd(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_storeu_ps(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_storeu_si128(reinterpret_cast<__m128i*>(to), from); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
+{
+ return _mm_set_ps(from[3*stride], from[2*stride], from[1*stride], from[0*stride]);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride)
+{
+ return _mm_set_pd(from[1*stride], from[0*stride]);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride)
+{
+ return _mm_set_epi32(from[3*stride], from[2*stride], from[1*stride], from[0*stride]);
+ }
 
-template<> EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet2d& from) {
-  EIGEN_DEBUG_UNALIGNED_STORE
-  _mm_storel_pd((to), from);
-  _mm_storeh_pd((to+1), from);
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
+{
+  to[stride*0] = _mm_cvtss_f32(from);
+  to[stride*1] = _mm_cvtss_f32(_mm_shuffle_ps(from, from, 1));
+  to[stride*2] = _mm_cvtss_f32(_mm_shuffle_ps(from, from, 2));
+  to[stride*3] = _mm_cvtss_f32(_mm_shuffle_ps(from, from, 3));
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride)
+{
+  to[stride*0] = _mm_cvtsd_f64(from);
+  to[stride*1] = _mm_cvtsd_f64(_mm_shuffle_pd(from, from, 1));
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride)
+{
+  to[stride*0] = _mm_cvtsi128_si32(from);
+  to[stride*1] = _mm_cvtsi128_si32(_mm_shuffle_epi32(from, 1));
+  to[stride*2] = _mm_cvtsi128_si32(_mm_shuffle_epi32(from, 2));
+  to[stride*3] = _mm_cvtsi128_si32(_mm_shuffle_epi32(from, 3));
 }
-template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*  to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(reinterpret_cast<double*>(to), _mm_castps_pd(from)); }
-template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*      to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(reinterpret_cast<double*>(to), _mm_castsi128_pd(from)); }
 
 // some compilers might be tempted to perform multiple moves instead of using a vector path.
 template<> EIGEN_STRONG_INLINE void pstore1<Packet4f>(float* to, const float& a)
 {
   Packet4f pa = _mm_set_ss(a);
-  pstore(to, vec4f_swizzle1(pa,0,0,0,0));
+  pstore(to, Packet4f(vec4f_swizzle1(pa,0,0,0,0)));
 }
 // some compilers might be tempted to perform multiple moves instead of using a vector path.
 template<> EIGEN_STRONG_INLINE void pstore1<Packet2d>(double* to, const double& a)
 {
   Packet2d pa = _mm_set_sd(a);
-  pstore(to, vec2d_swizzle1(pa,0,0));
+  pstore(to, Packet2d(vec2d_swizzle1(pa,0,0)));
 }
 
-template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
-template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
-template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
+#if EIGEN_COMP_PGI
+typedef const void * SsePrefetchPtrType;
+#else
+typedef const char * SsePrefetchPtrType;
+#endif
+
+#ifndef EIGEN_VECTORIZE_AVX
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+#endif
 
-#if defined(_MSC_VER) && defined(_WIN64) && !defined(__INTEL_COMPILER) && !defined(__clang__)
+#if EIGEN_COMP_MSVC_STRICT && EIGEN_OS_WIN64
 // The temporary variable fixes an internal compilation error in vs <= 2008 and a wrong-result bug in vs 2010
 // Direct of the struct members fixed bug #62.
 template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { return a.m128_f32[0]; }
 template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { return a.m128d_f64[0]; }
 template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int x = _mm_cvtsi128_si32(a); return x; }
-#elif defined(_MSC_VER) && !defined(__INTEL_COMPILER)
+#elif EIGEN_COMP_MSVC_STRICT
 // The temporary variable fixes an internal compilation error in vs <= 2008 and a wrong-result bug in vs 2010
 template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float x = _mm_cvtss_f32(a); return x; }
 template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { double x = _mm_cvtsd_f64(a); return x; }
@@ -359,7 +445,6 @@ template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a)
 template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a)
 { return _mm_shuffle_epi32(a,0x1B); }
 
-
 template<> EIGEN_STRONG_INLINE Packet4f pabs(const Packet4f& a)
 {
   const Packet4f mask = _mm_castsi128_ps(_mm_setr_epi32(0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF));
@@ -380,6 +465,38 @@ template<> EIGEN_STRONG_INLINE Packet4i pabs(const Packet4i& a)
   #endif
 }
 
+// with AVX, the default implementations based on pload1 are faster
+#ifndef __AVX__
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4f>(const float *a,
+                      Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
+{
+  a3 = pload<Packet4f>(a);
+  a0 = vec4f_swizzle1(a3, 0,0,0,0);
+  a1 = vec4f_swizzle1(a3, 1,1,1,1);
+  a2 = vec4f_swizzle1(a3, 2,2,2,2);
+  a3 = vec4f_swizzle1(a3, 3,3,3,3);
+}
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet2d>(const double *a,
+                      Packet2d& a0, Packet2d& a1, Packet2d& a2, Packet2d& a3)
+{
+#ifdef EIGEN_VECTORIZE_SSE3
+  a0 = _mm_loaddup_pd(a+0);
+  a1 = _mm_loaddup_pd(a+1);
+  a2 = _mm_loaddup_pd(a+2);
+  a3 = _mm_loaddup_pd(a+3);
+#else
+  a1 = pload<Packet2d>(a);
+  a0 = vec2d_swizzle1(a1, 0,0);
+  a1 = vec2d_swizzle1(a1, 1,1);
+  a3 = pload<Packet2d>(a+2);
+  a2 = vec2d_swizzle1(a3, 0,0);
+  a3 = vec2d_swizzle1(a3, 1,1);
+#endif
+}
+#endif
+
 EIGEN_STRONG_INLINE void punpackp(Packet4f* vecs)
 {
   vecs[1] = _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(vecs[0]), 0x55));
@@ -389,47 +506,17 @@ EIGEN_STRONG_INLINE void punpackp(Packet4f* vecs)
 }
 
 #ifdef EIGEN_VECTORIZE_SSE3
-// TODO implement SSE2 versions as well as integer versions
 template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
 {
   return _mm_hadd_ps(_mm_hadd_ps(vecs[0], vecs[1]),_mm_hadd_ps(vecs[2], vecs[3]));
 }
+
 template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
 {
   return _mm_hadd_pd(vecs[0], vecs[1]);
 }
-// SSSE3 version:
-// EIGEN_STRONG_INLINE Packet4i preduxp(const Packet4i* vecs)
-// {
-//   return _mm_hadd_epi32(_mm_hadd_epi32(vecs[0], vecs[1]),_mm_hadd_epi32(vecs[2], vecs[3]));
-// }
-
-template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
-{
-  Packet4f tmp0 = _mm_hadd_ps(a,a);
-  return pfirst(_mm_hadd_ps(tmp0, tmp0));
-}
 
-template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) { return pfirst(_mm_hadd_pd(a, a)); }
-
-// SSSE3 version:
-// EIGEN_STRONG_INLINE float predux(const Packet4i& a)
-// {
-//   Packet4i tmp0 = _mm_hadd_epi32(a,a);
-//   return pfirst(_mm_hadd_epi32(tmp0, tmp0));
-// }
 #else
-// SSE2 versions
-template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
-{
-  Packet4f tmp = _mm_add_ps(a, _mm_movehl_ps(a,a));
-  return pfirst(_mm_add_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
-}
-template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
-{
-  return pfirst(_mm_add_sd(a, _mm_unpackhi_pd(a,a)));
-}
-
 template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
 {
   Packet4f tmp0, tmp1, tmp2;
@@ -450,10 +537,45 @@ template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
 }
 #endif  // SSE3
 
+template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
+{
+  // Disable SSE3 _mm_hadd_pd that is extremely slow on all existing Intel's architectures
+  // (from Nehalem to Haswell)
+// #ifdef EIGEN_VECTORIZE_SSE3
+//   Packet4f tmp = _mm_add_ps(a, vec4f_swizzle1(a,2,3,2,3));
+//   return pfirst<Packet4f>(_mm_hadd_ps(tmp, tmp));
+// #else
+  Packet4f tmp = _mm_add_ps(a, _mm_movehl_ps(a,a));
+  return pfirst<Packet4f>(_mm_add_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
+// #endif
+}
+
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
+{
+  // Disable SSE3 _mm_hadd_pd that is extremely slow on all existing Intel's architectures
+  // (from Nehalem to Haswell)
+// #ifdef EIGEN_VECTORIZE_SSE3
+//   return pfirst<Packet2d>(_mm_hadd_pd(a, a));
+// #else
+  return pfirst<Packet2d>(_mm_add_sd(a, _mm_unpackhi_pd(a,a)));
+// #endif
+}
+
+#ifdef EIGEN_VECTORIZE_SSSE3
+template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
+{
+  return _mm_hadd_epi32(_mm_hadd_epi32(vecs[0], vecs[1]),_mm_hadd_epi32(vecs[2], vecs[3]));
+}
+template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
+{
+  Packet4i tmp0 = _mm_hadd_epi32(a,a);
+  return pfirst<Packet4i>(_mm_hadd_epi32(tmp0,tmp0));
+}
+#else
 template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
 {
   Packet4i tmp = _mm_add_epi32(a, _mm_unpackhi_epi64(a,a));
-  return pfirst(tmp) + pfirst(_mm_shuffle_epi32(tmp, 1));
+  return pfirst(tmp) + pfirst<Packet4i>(_mm_shuffle_epi32(tmp, 1));
 }
 
 template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
@@ -469,18 +591,18 @@ template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
   tmp0 = _mm_unpackhi_epi64(tmp0, tmp1);
   return _mm_add_epi32(tmp0, tmp2);
 }
-
+#endif
 // Other reduction functions:
 
 // mul
 template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
 {
   Packet4f tmp = _mm_mul_ps(a, _mm_movehl_ps(a,a));
-  return pfirst(_mm_mul_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
+  return pfirst<Packet4f>(_mm_mul_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
 }
 template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a)
 {
-  return pfirst(_mm_mul_sd(a, _mm_unpackhi_pd(a,a)));
+  return pfirst<Packet2d>(_mm_mul_sd(a, _mm_unpackhi_pd(a,a)));
 }
 template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
 {
@@ -496,14 +618,18 @@ template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
 template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a)
 {
   Packet4f tmp = _mm_min_ps(a, _mm_movehl_ps(a,a));
-  return pfirst(_mm_min_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
+  return pfirst<Packet4f>(_mm_min_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
 }
 template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a)
 {
-  return pfirst(_mm_min_sd(a, _mm_unpackhi_pd(a,a)));
+  return pfirst<Packet2d>(_mm_min_sd(a, _mm_unpackhi_pd(a,a)));
 }
 template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
 {
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  Packet4i tmp = _mm_min_epi32(a, _mm_shuffle_epi32(a, _MM_SHUFFLE(0,0,3,2)));
+  return pfirst<Packet4i>(_mm_min_epi32(tmp,_mm_shuffle_epi32(tmp, 1)));
+#else
   // after some experiments, it is seems this is the fastest way to implement it
   // for GCC (eg., it does not like using std::min after the pstore !!)
   EIGEN_ALIGN16 int aux[4];
@@ -511,20 +637,25 @@ template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
   int aux0 = aux[0]<aux[1] ? aux[0] : aux[1];
   int aux2 = aux[2]<aux[3] ? aux[2] : aux[3];
   return aux0<aux2 ? aux0 : aux2;
+#endif // EIGEN_VECTORIZE_SSE4_1
 }
 
 // max
 template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a)
 {
   Packet4f tmp = _mm_max_ps(a, _mm_movehl_ps(a,a));
-  return pfirst(_mm_max_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
+  return pfirst<Packet4f>(_mm_max_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
 }
 template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a)
 {
-  return pfirst(_mm_max_sd(a, _mm_unpackhi_pd(a,a)));
+  return pfirst<Packet2d>(_mm_max_sd(a, _mm_unpackhi_pd(a,a)));
 }
 template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
 {
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  Packet4i tmp = _mm_max_epi32(a, _mm_shuffle_epi32(a, _MM_SHUFFLE(0,0,3,2)));
+  return pfirst<Packet4i>(_mm_max_epi32(tmp,_mm_shuffle_epi32(tmp, 1)));
+#else
   // after some experiments, it is seems this is the fastest way to implement it
   // for GCC (eg., it does not like using std::min after the pstore !!)
   EIGEN_ALIGN16 int aux[4];
@@ -532,9 +663,10 @@ template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
   int aux0 = aux[0]>aux[1] ? aux[0] : aux[1];
   int aux2 = aux[2]>aux[3] ? aux[2] : aux[3];
   return aux0>aux2 ? aux0 : aux2;
+#endif // EIGEN_VECTORIZE_SSE4_1
 }
 
-#if (defined __GNUC__)
+#if EIGEN_COMP_GNUC
 // template <> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f&  a, const Packet4f&  b, const Packet4f&  c)
 // {
 //   Packet4f res = b;
@@ -642,8 +774,122 @@ struct palign_impl<Offset,Packet2d>
 };
 #endif
 
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4f,4>& kernel) {
+  _MM_TRANSPOSE4_PS(kernel.packet[0], kernel.packet[1], kernel.packet[2], kernel.packet[3]);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2d,2>& kernel) {
+  __m128d tmp = _mm_unpackhi_pd(kernel.packet[0], kernel.packet[1]);
+  kernel.packet[0] = _mm_unpacklo_pd(kernel.packet[0], kernel.packet[1]);
+  kernel.packet[1] = tmp;
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4i,4>& kernel) {
+  __m128i T0 = _mm_unpacklo_epi32(kernel.packet[0], kernel.packet[1]);
+  __m128i T1 = _mm_unpacklo_epi32(kernel.packet[2], kernel.packet[3]);
+  __m128i T2 = _mm_unpackhi_epi32(kernel.packet[0], kernel.packet[1]);
+  __m128i T3 = _mm_unpackhi_epi32(kernel.packet[2], kernel.packet[3]);
+
+  kernel.packet[0] = _mm_unpacklo_epi64(T0, T1);
+  kernel.packet[1] = _mm_unpackhi_epi64(T0, T1);
+  kernel.packet[2] = _mm_unpacklo_epi64(T2, T3);
+  kernel.packet[3] = _mm_unpackhi_epi64(T2, T3);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pblend(const Selector<4>& ifPacket, const Packet4i& thenPacket, const Packet4i& elsePacket) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i select = _mm_set_epi32(ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m128i false_mask = _mm_cmpeq_epi32(select, zero);
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blendv_epi8(thenPacket, elsePacket, false_mask);
+#else
+  return _mm_or_si128(_mm_andnot_si128(false_mask, thenPacket), _mm_and_si128(false_mask, elsePacket));
+#endif
+}
+template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) {
+  const __m128 zero = _mm_setzero_ps();
+  const __m128 select = _mm_set_ps(ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m128 false_mask = _mm_cmpeq_ps(select, zero);
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blendv_ps(thenPacket, elsePacket, false_mask);
+#else
+  return _mm_or_ps(_mm_andnot_ps(false_mask, thenPacket), _mm_and_ps(false_mask, elsePacket));
+#endif
+}
+template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, const Packet2d& thenPacket, const Packet2d& elsePacket) {
+  const __m128d zero = _mm_setzero_pd();
+  const __m128d select = _mm_set_pd(ifPacket.select[1], ifPacket.select[0]);
+  __m128d false_mask = _mm_cmpeq_pd(select, zero);
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blendv_pd(thenPacket, elsePacket, false_mask);
+#else
+  return _mm_or_pd(_mm_andnot_pd(false_mask, thenPacket), _mm_and_pd(false_mask, elsePacket));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pinsertfirst(const Packet4f& a, float b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_ps(a,pset1<Packet4f>(b),1);
+#else
+  return _mm_move_ss(a, _mm_load_ss(&b));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pinsertfirst(const Packet2d& a, double b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_pd(a,pset1<Packet2d>(b),1);
+#else
+  return _mm_move_sd(a, _mm_load_sd(&b));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pinsertlast(const Packet4f& a, float b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_ps(a,pset1<Packet4f>(b),(1<<3));
+#else
+  const Packet4f mask = _mm_castsi128_ps(_mm_setr_epi32(0x0,0x0,0x0,0xFFFFFFFF));
+  return _mm_or_ps(_mm_andnot_ps(mask, a), _mm_and_ps(mask, pset1<Packet4f>(b)));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pinsertlast(const Packet2d& a, double b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_pd(a,pset1<Packet2d>(b),(1<<1));
+#else
+  const Packet2d mask = _mm_castsi128_pd(_mm_setr_epi32(0x0,0x0,0xFFFFFFFF,0xFFFFFFFF));
+  return _mm_or_pd(_mm_andnot_pd(mask, a), _mm_and_pd(mask, pset1<Packet2d>(b)));
+#endif
+}
+
+// Scalar path for pmadd with FMA to ensure consistency with vectorized path.
+#ifdef __FMA__
+template<> EIGEN_STRONG_INLINE float pmadd(const float& a, const float& b, const float& c) {
+  return ::fmaf(a,b,c);
+}
+template<> EIGEN_STRONG_INLINE double pmadd(const double& a, const double& b, const double& c) {
+  return ::fma(a,b,c);
+}
+#endif
+
 } // end namespace internal
 
 } // end namespace Eigen
 
+#if EIGEN_COMP_PGI
+// PGI++ does not define the following intrinsics in C++ mode.
+static inline __m128  _mm_castpd_ps   (__m128d x) { return reinterpret_cast<__m128&>(x);  }
+static inline __m128i _mm_castpd_si128(__m128d x) { return reinterpret_cast<__m128i&>(x); }
+static inline __m128d _mm_castps_pd   (__m128  x) { return reinterpret_cast<__m128d&>(x); }
+static inline __m128i _mm_castps_si128(__m128  x) { return reinterpret_cast<__m128i&>(x); }
+static inline __m128  _mm_castsi128_ps(__m128i x) { return reinterpret_cast<__m128&>(x);  }
+static inline __m128d _mm_castsi128_pd(__m128i x) { return reinterpret_cast<__m128d&>(x); }
+#endif
+
 #endif // EIGEN_PACKET_MATH_SSE_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/SSE/TypeCasting.h b/extern/Eigen3/Eigen/src/Core/arch/SSE/TypeCasting.h
new file mode 100644
index 00000000000..c6ca8c716c0
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/SSE/TypeCasting.h
@@ -0,0 +1,77 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_TYPE_CASTING_SSE_H
+#define EIGEN_TYPE_CASTING_SSE_H
+
+namespace Eigen {
+
+namespace internal {
+
+#ifndef EIGEN_VECTORIZE_AVX
+template <>
+struct type_casting_traits<float, int> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template <>
+struct type_casting_traits<int, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template <>
+struct type_casting_traits<double, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 2,
+    TgtCoeffRatio = 1
+  };
+};
+
+template <>
+struct type_casting_traits<float, double> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 2
+  };
+};
+#endif
+
+template<> EIGEN_STRONG_INLINE Packet4i pcast<Packet4f, Packet4i>(const Packet4f& a) {
+  return _mm_cvttps_epi32(a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet4i, Packet4f>(const Packet4i& a) {
+  return _mm_cvtepi32_ps(a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet2d, Packet4f>(const Packet2d& a, const Packet2d& b) {
+  return _mm_shuffle_ps(_mm_cvtpd_ps(a), _mm_cvtpd_ps(b), (1 << 2) | (1 << 6));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pcast<Packet4f, Packet2d>(const Packet4f& a) {
+  // Simply discard the second half of the input
+  return _mm_cvtps_pd(a);
+}
+
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TYPE_CASTING_SSE_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/ZVector/Complex.h b/extern/Eigen3/Eigen/src/Core/arch/ZVector/Complex.h
new file mode 100644
index 00000000000..1bfb73397d6
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/ZVector/Complex.h
@@ -0,0 +1,397 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_COMPLEX32_ALTIVEC_H
+#define EIGEN_COMPLEX32_ALTIVEC_H
+
+namespace Eigen {
+
+namespace internal {
+
+static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_ZERO_, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_ZERO_, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+
+struct Packet1cd
+{
+  EIGEN_STRONG_INLINE Packet1cd() {}
+  EIGEN_STRONG_INLINE explicit Packet1cd(const Packet2d& a) : v(a) {}
+  Packet2d v;
+};
+
+struct Packet2cf
+{
+  EIGEN_STRONG_INLINE Packet2cf() {}
+  EIGEN_STRONG_INLINE explicit Packet2cf(const Packet4f& a) : v(a) {}
+  union {
+    Packet4f v;
+    Packet1cd cd[2];
+  };
+};
+
+template<> struct packet_traits<std::complex<float> >  : default_packet_traits
+{
+  typedef Packet2cf type;
+  typedef Packet2cf half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 2,
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasBlend  = 1,
+    HasSetLinear = 0
+  };
+};
+
+
+template<> struct packet_traits<std::complex<double> >  : default_packet_traits
+{
+  typedef Packet1cd type;
+  typedef Packet1cd half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 1,
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float>  type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
+template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
+
+/* Forward declaration */
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel);
+
+template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from)  { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>((const float*)from)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(pload<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from)  { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>((const float*)from)); }
+template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ploadu<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *     to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *     to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>&  from)
+{ /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
+
+template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>&  from)
+{
+  Packet2cf res;
+  res.cd[0] = Packet1cd(vec_ld2f((const float *)&from));
+  res.cd[1] = res.cd[0];
+  return res;
+}
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  return pload<Packet2cf>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index stride EIGEN_UNUSED)
+{
+  return pload<Packet1cd>(from);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  pstore<std::complex<float> >((std::complex<float> *) af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index stride EIGEN_UNUSED)
+{
+  pstore<std::complex<double> >(to, from);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(padd<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v + b.v); }
+template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(psub<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v - b.v); }
+template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
+template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(Packet4f(a.v))); }
+template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd((Packet2d)vec_xor((Packet2d)a.v, (Packet2d)p2ul_CONJ_XOR2)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
+{
+  Packet2cf res;
+  res.v.v4f[0] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0]))).v;
+  res.v.v4f[1] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1]))).v;
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  Packet2d a_re, a_im, v1, v2;
+
+  // Permute and multiply the real parts of a and b
+  a_re = vec_perm(a.v, a.v, p16uc_PSET64_HI);
+  // Get the imaginary parts of a
+  a_im = vec_perm(a.v, a.v, p16uc_PSET64_LO);
+  // multiply a_re * b
+  v1 = vec_madd(a_re, b.v, p2d_ZERO);
+  // multiply a_im * b and get the conjugate result
+  v2 = vec_madd(a_im, b.v, p2d_ZERO);
+  v2 = (Packet2d) vec_sld((Packet4ui)v2, (Packet4ui)v2, 8);
+  v2 = (Packet2d) vec_xor((Packet2d)v2, (Packet2d) p2ul_CONJ_XOR1);
+
+  return Packet1cd(v1 + v2);
+}
+template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+{
+  Packet2cf res;
+  res.v.v4f[0] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[0]))).v;
+  res.v.v4f[1] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[1]))).v;
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_or(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_xor(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pxor   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v, vec_nor(b.v,b.v))); }
+template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v,b.v)); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>*     from) {  return pset1<Packet1cd>(*from); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>*      from) {  return pset1<Packet2cf>(*from); }
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *     addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+
+template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
+{
+  std::complex<double> EIGEN_ALIGN16 res;
+  pstore<std::complex<double> >(&res, a);
+
+  return res;
+}
+template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
+{
+  std::complex<float> EIGEN_ALIGN16 res[2];
+  pstore<std::complex<float> >(res, a);
+
+  return res[0];
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
+{
+  Packet2cf res;
+  res.cd[0] = a.cd[1];
+  res.cd[1] = a.cd[0];
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a)
+{
+  return pfirst(a);
+}
+template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
+{
+  std::complex<float> res;
+  Packet1cd b = padd<Packet1cd>(a.cd[0], a.cd[1]);
+  vec_st2f(b.v, (float*)&res);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs)
+{
+  return vecs[0];
+}
+template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
+{
+  PacketBlock<Packet2cf,2> transpose;
+  transpose.packet[0] = vecs[0];
+  transpose.packet[1] = vecs[1];
+  ptranspose(transpose);
+
+  return padd<Packet2cf>(transpose.packet[0], transpose.packet[1]);
+} 
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a)
+{
+  return pfirst(a);
+}
+template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
+{
+  std::complex<float> res;
+  Packet1cd b = pmul<Packet1cd>(a.cd[0], a.cd[1]);
+  vec_st2f(b.v, (float*)&res);
+  return res;
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet1cd>
+{
+  static EIGEN_STRONG_INLINE void run(Packet1cd& /*first*/, const Packet1cd& /*second*/)
+  {
+    // FIXME is it sure we never have to align a Packet1cd?
+    // Even though a std::complex<double> has 16 bytes, it is not necessarily aligned on a 16 bytes boundary...
+  }
+};
+
+template<int Offset>
+struct palign_impl<Offset,Packet2cf>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2cf& first, const Packet2cf& second)
+  {
+    if (Offset == 1) {
+      first.cd[0] = first.cd[1];
+      first.cd[1] = second.cd[0];
+    }
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,false>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+template<> struct conj_helper<Packet2cf, Packet2cf, false,true>
+{
+  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet2cf, Packet2cf, true,false>
+{
+  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
+{
+  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
+
+template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  // TODO optimize it for AltiVec
+  Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
+  Packet2d s = vec_madd(b.v, b.v, p2d_ZERO_);
+  return Packet1cd(pdiv(res.v, s + vec_perm(s, s, p16uc_REVERSE64)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+{
+  // TODO optimize it for AltiVec
+  Packet2cf res;
+  res.cd[0] = pdiv<Packet1cd>(a.cd[0], b.cd[0]);
+  res.cd[1] = pdiv<Packet1cd>(a.cd[1], b.cd[1]);
+  return res;
+}
+
+EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
+{
+  return Packet1cd(preverse(Packet2d(x.v)));
+}
+
+EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
+{
+  Packet2cf res;
+  res.cd[0] = pcplxflip(x.cd[0]);
+  res.cd[1] = pcplxflip(x.cd[1]);
+  return res;
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
+{
+  Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
+  kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
+  kernel.packet[0].v = tmp;
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
+{
+  Packet1cd tmp = kernel.packet[0].cd[1];
+  kernel.packet[0].cd[1] = kernel.packet[1].cd[0];
+  kernel.packet[1].cd[0] = tmp;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
+  Packet2cf result;
+  const Selector<4> ifPacket4 = { ifPacket.select[0], ifPacket.select[0], ifPacket.select[1], ifPacket.select[1] };
+  result.v = pblend<Packet4f>(ifPacket4, thenPacket.v, elsePacket.v);
+  return result;
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_COMPLEX32_ALTIVEC_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/ZVector/MathFunctions.h b/extern/Eigen3/Eigen/src/Core/arch/ZVector/MathFunctions.h
new file mode 100644
index 00000000000..5c7aa725678
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/ZVector/MathFunctions.h
@@ -0,0 +1,137 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2007 Julien Pommier
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+/* The sin, cos, exp, and log functions of this file come from
+ * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
+ */
+
+#ifndef EIGEN_MATH_FUNCTIONS_ALTIVEC_H
+#define EIGEN_MATH_FUNCTIONS_ALTIVEC_H
+
+namespace Eigen {
+
+namespace internal {
+
+static _EIGEN_DECLARE_CONST_Packet2d(1 , 1.0);
+static _EIGEN_DECLARE_CONST_Packet2d(2 , 2.0);
+static _EIGEN_DECLARE_CONST_Packet2d(half, 0.5);
+
+static _EIGEN_DECLARE_CONST_Packet2d(exp_hi,  709.437);
+static _EIGEN_DECLARE_CONST_Packet2d(exp_lo, -709.436139303);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_LOG2EF, 1.4426950408889634073599);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p0, 1.26177193074810590878e-4);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p1, 3.02994407707441961300e-2);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q0, 3.00198505138664455042e-6);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q1, 2.52448340349684104192e-3);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q2, 2.27265548208155028766e-1);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q3, 2.00000000000000000009e0);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C1, 0.693145751953125);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C2, 1.42860682030941723212e-6);
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d pexp<Packet2d>(const Packet2d& _x)
+{
+  Packet2d x = _x;
+
+  Packet2d tmp, fx;
+  Packet2l emm0;
+
+  // clamp x
+  x = pmax(pmin(x, p2d_exp_hi), p2d_exp_lo);
+  /* express exp(x) as exp(g + n*log(2)) */
+  fx = pmadd(p2d_cephes_LOG2EF, x, p2d_half);
+
+  fx = vec_floor(fx);
+
+  tmp = pmul(fx, p2d_cephes_exp_C1);
+  Packet2d z = pmul(fx, p2d_cephes_exp_C2);
+  x = psub(x, tmp);
+  x = psub(x, z);
+
+  Packet2d x2 = pmul(x,x);
+
+  Packet2d px = p2d_cephes_exp_p0;
+  px = pmadd(px, x2, p2d_cephes_exp_p1);
+  px = pmadd(px, x2, p2d_cephes_exp_p2);
+  px = pmul (px, x);
+
+  Packet2d qx = p2d_cephes_exp_q0;
+  qx = pmadd(qx, x2, p2d_cephes_exp_q1);
+  qx = pmadd(qx, x2, p2d_cephes_exp_q2);
+  qx = pmadd(qx, x2, p2d_cephes_exp_q3);
+
+  x = pdiv(px,psub(qx,px));
+  x = pmadd(p2d_2,x,p2d_1);
+
+  // build 2^n
+  emm0 = vec_ctsl(fx, 0);
+
+  static const Packet2l p2l_1023 = { 1023, 1023 };
+  static const Packet2ul p2ul_52 = { 52, 52 };
+
+  emm0 = emm0 + p2l_1023;
+  emm0 = emm0 << reinterpret_cast<Packet2l>(p2ul_52);
+
+  // Altivec's max & min operators just drop silent NaNs. Check NaNs in 
+  // inputs and return them unmodified.
+  Packet2ul isnumber_mask = reinterpret_cast<Packet2ul>(vec_cmpeq(_x, _x));
+  return vec_sel(_x, pmax(pmul(x, reinterpret_cast<Packet2d>(emm0)), _x),
+                 isnumber_mask);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f pexp<Packet4f>(const Packet4f& x)
+{
+  Packet4f res;
+  res.v4f[0] = pexp<Packet2d>(x.v4f[0]);
+  res.v4f[1] = pexp<Packet2d>(x.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d psqrt<Packet2d>(const Packet2d& x)
+{
+  return  __builtin_s390_vfsqdb(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f psqrt<Packet4f>(const Packet4f& x)
+{
+  Packet4f res;
+  res.v4f[0] = psqrt<Packet2d>(x.v4f[0]);
+  res.v4f[1] = psqrt<Packet2d>(x.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d prsqrt<Packet2d>(const Packet2d& x) {
+  // Unfortunately we can't use the much faster mm_rqsrt_pd since it only provides an approximation.
+  return pset1<Packet2d>(1.0) / psqrt<Packet2d>(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f prsqrt<Packet4f>(const Packet4f& x) {
+  Packet4f res;
+  res.v4f[0] = prsqrt<Packet2d>(x.v4f[0]);
+  res.v4f[1] = prsqrt<Packet2d>(x.v4f[1]);
+  return res;
+}
+
+}  // end namespace internal
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_MATH_FUNCTIONS_ALTIVEC_H
diff --git a/extern/Eigen3/Eigen/src/Core/arch/ZVector/PacketMath.h b/extern/Eigen3/Eigen/src/Core/arch/ZVector/PacketMath.h
new file mode 100755
index 00000000000..57b01fc634d
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/arch/ZVector/PacketMath.h
@@ -0,0 +1,945 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_ZVECTOR_H
+#define EIGEN_PACKET_MATH_ZVECTOR_H
+
+#include <stdint.h>
+
+namespace Eigen {
+
+namespace internal {
+
+#ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+#define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 4
+#endif
+
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#endif
+
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#endif
+
+#ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS  16
+#endif
+
+typedef __vector int                 Packet4i;
+typedef __vector unsigned int        Packet4ui;
+typedef __vector __bool int          Packet4bi;
+typedef __vector short int           Packet8i;
+typedef __vector unsigned char       Packet16uc;
+typedef __vector double              Packet2d;
+typedef __vector unsigned long long  Packet2ul;
+typedef __vector long long           Packet2l;
+
+typedef struct {
+	Packet2d  v4f[2];
+} Packet4f;
+
+typedef union {
+  int32_t   i[4];
+  uint32_t ui[4];
+  int64_t   l[2];
+  uint64_t ul[2];
+  double    d[2];
+  Packet4i  v4i;
+  Packet4ui v4ui;
+  Packet2l  v2l;
+  Packet2ul v2ul;
+  Packet2d  v2d;
+} Packet;
+
+// We don't want to write the same code all the time, but we need to reuse the constants
+// and it doesn't really work to declare them global, so we define macros instead
+
+#define _EIGEN_DECLARE_CONST_FAST_Packet4i(NAME,X) \
+  Packet4i p4i_##NAME = reinterpret_cast<Packet4i>(vec_splat_s32(X))
+
+#define _EIGEN_DECLARE_CONST_FAST_Packet2d(NAME,X) \
+  Packet2d p2d_##NAME = reinterpret_cast<Packet2d>(vec_splat_s64(X))
+
+#define _EIGEN_DECLARE_CONST_FAST_Packet2l(NAME,X) \
+  Packet2l p2l_##NAME = reinterpret_cast<Packet2l>(vec_splat_s64(X))
+
+#define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
+  Packet4i p4i_##NAME = pset1<Packet4i>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet2d(NAME,X) \
+  Packet2d p2d_##NAME = pset1<Packet2d>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet2l(NAME,X) \
+  Packet2l p2l_##NAME = pset1<Packet2l>(X)
+
+// These constants are endian-agnostic
+//static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0); //{ 0, 0, 0, 0,}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(ONE, 1); //{ 1, 1, 1, 1}
+
+static _EIGEN_DECLARE_CONST_FAST_Packet2d(ZERO, 0);
+static _EIGEN_DECLARE_CONST_FAST_Packet2l(ZERO, 0);
+static _EIGEN_DECLARE_CONST_FAST_Packet2l(ONE, 1);
+
+static Packet2d p2d_ONE = { 1.0, 1.0 }; 
+static Packet2d p2d_ZERO_ = { -0.0, -0.0 };
+
+static Packet4i p4i_COUNTDOWN = { 0, 1, 2, 3 };
+static Packet4f p4f_COUNTDOWN = { 0.0, 1.0, 2.0, 3.0 };
+static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet16uc>(p2d_ZERO), reinterpret_cast<Packet16uc>(p2d_ONE), 8));
+
+static Packet16uc p16uc_PSET64_HI = { 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };
+static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 };
+
+// Mask alignment
+#define _EIGEN_MASK_ALIGNMENT	0xfffffffffffffff0
+
+#define _EIGEN_ALIGNED_PTR(x)	((std::ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT)
+
+// Handle endianness properly while loading constants
+// Define global static constants:
+
+static Packet16uc p16uc_FORWARD =   { 0,1,2,3, 4,5,6,7, 8,9,10,11, 12,13,14,15 };
+static Packet16uc p16uc_REVERSE32 = { 12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3 };
+static Packet16uc p16uc_REVERSE64 = { 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+
+static Packet16uc p16uc_PSET32_WODD   = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 };
+static Packet16uc p16uc_PSET32_WEVEN  = vec_sld(p16uc_DUPLICATE32_HI, (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 };
+/*static Packet16uc p16uc_HALF64_0_16 = vec_sld((Packet16uc)p4i_ZERO, vec_splat((Packet16uc) vec_abs(p4i_MINUS16), 3), 8);      //{ 0,0,0,0, 0,0,0,0, 16,16,16,16, 16,16,16,16};
+
+static Packet16uc p16uc_PSET64_HI = (Packet16uc) vec_mergeh((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };*/
+static Packet16uc p16uc_PSET64_LO = (Packet16uc) vec_mergel((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 8,9,10,11, 12,13,14,15, 8,9,10,11, 12,13,14,15 };
+/*static Packet16uc p16uc_TRANSPOSE64_HI = vec_add(p16uc_PSET64_HI, p16uc_HALF64_0_16);                                         //{ 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23};
+static Packet16uc p16uc_TRANSPOSE64_LO = vec_add(p16uc_PSET64_LO, p16uc_HALF64_0_16);                                         //{ 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31};*/
+static Packet16uc p16uc_TRANSPOSE64_HI = { 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23};
+static Packet16uc p16uc_TRANSPOSE64_LO = { 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31};
+
+//static Packet16uc p16uc_COMPLEX32_REV = vec_sld(p16uc_REVERSE32, p16uc_REVERSE32, 8);                                         //{ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11 };
+
+//static Packet16uc p16uc_COMPLEX32_REV2 = vec_sld(p16uc_FORWARD, p16uc_FORWARD, 8);                                            //{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+
+
+#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC
+  #define EIGEN_ZVECTOR_PREFETCH(ADDR) __builtin_prefetch(ADDR);
+#else
+  #define EIGEN_ZVECTOR_PREFETCH(ADDR) asm( "   pfd [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
+#endif
+
+template<> struct packet_traits<int>    : default_packet_traits
+{
+  typedef Packet4i type;
+  typedef Packet4i half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 4,
+    HasHalfPacket = 0,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasBlend = 1
+  };
+};
+
+template<> struct packet_traits<float> : default_packet_traits
+{
+  typedef Packet4f type;
+  typedef Packet4f half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=4,
+    HasHalfPacket = 0,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasMin  = 1,
+    HasMax  = 1,
+    HasAbs  = 1,
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 0,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1,
+    HasNegate = 1,
+    HasBlend = 1
+  };
+};
+
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef Packet2d type;
+  typedef Packet2d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=2,
+    HasHalfPacket = 1,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasMin  = 1,
+    HasMax  = 1,
+    HasAbs  = 1,
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 0,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1,
+    HasNegate = 1,
+    HasBlend = 1
+  };
+};
+
+template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
+template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
+template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
+
+/* Forward declaration */
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet4f,4>& kernel);
+ 
+inline std::ostream & operator <<(std::ostream & s, const Packet4i & v)
+{
+  Packet vt;
+  vt.v4i = v;
+  s << vt.i[0] << ", " << vt.i[1] << ", " << vt.i[2] << ", " << vt.i[3];
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet4ui & v)
+{
+  Packet vt;
+  vt.v4ui = v;
+  s << vt.ui[0] << ", " << vt.ui[1] << ", " << vt.ui[2] << ", " << vt.ui[3];
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2l & v)
+{
+  Packet vt;
+  vt.v2l = v;
+  s << vt.l[0] << ", " << vt.l[1];
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2ul & v)
+{
+  Packet vt;
+  vt.v2ul = v;
+  s << vt.ul[0] << ", " << vt.ul[1] ;
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2d & v)
+{
+  Packet vt;
+  vt.v2d = v;
+  s << vt.d[0] << ", " << vt.d[1];
+  return s;
+}
+
+/* Helper function to simulate a vec_splat_packet4f
+ */
+template<int element> EIGEN_STRONG_INLINE Packet4f vec_splat_packet4f(const Packet4f&   from)
+{
+  Packet4f splat;
+  switch (element) {
+  case 0:
+    splat.v4f[0] = vec_splat(from.v4f[0], 0);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  case 1:
+    splat.v4f[0] = vec_splat(from.v4f[0], 1);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  case 2:
+    splat.v4f[0] = vec_splat(from.v4f[1], 0);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  case 3:
+    splat.v4f[0] = vec_splat(from.v4f[1], 1);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  }
+  return splat;
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet4i>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4i& first, const Packet4i& second)
+  {
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(first, second, 4); break;
+    case 2:
+      first = vec_sld(first, second, 8); break;
+    case 3:
+      first = vec_sld(first, second, 12); break;
+    }
+  }
+};
+
+/* This is a tricky one, we have to translate float alignment to vector elements of sizeof double
+ */
+template<int Offset>
+struct palign_impl<Offset,Packet4f>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4f& first, const Packet4f& second)
+  {
+    switch (Offset % 4) {
+    case 1:
+      first.v4f[0] = vec_sld(first.v4f[0], first.v4f[1], 8);
+      first.v4f[1] = vec_sld(first.v4f[1], second.v4f[0], 8);
+      break;
+    case 2:
+      first.v4f[0] = first.v4f[1];
+      first.v4f[1] = second.v4f[0];
+      break;
+    case 3:
+      first.v4f[0] = vec_sld(first.v4f[1],  second.v4f[0], 8);
+      first.v4f[1] = vec_sld(second.v4f[0], second.v4f[1], 8);
+      break;
+    }
+  }
+};
+
+
+template<int Offset>
+struct palign_impl<Offset,Packet2d>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2d& first, const Packet2d& second)
+  {
+    if (Offset == 1)
+      first = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(first), reinterpret_cast<Packet4i>(second), 8));
+  }
+};
+
+template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_LOAD
+  Packet *vfrom;
+  vfrom = (Packet *) from;
+  return vfrom->v4i;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float*   from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_LOAD
+  Packet4f vfrom;
+  vfrom.v4f[0] = vec_ld2f(&from[0]);
+  vfrom.v4f[1] = vec_ld2f(&from[2]);
+  return vfrom;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_LOAD
+  Packet *vfrom;
+  vfrom = (Packet *) from;
+  return vfrom->v2d;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_STORE
+  Packet *vto;
+  vto = (Packet *) to;
+  vto->v4i = from;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_STORE
+  vec_st2f(from.v4f[0], &to[0]);
+  vec_st2f(from.v4f[1], &to[2]);
+}
+
+
+template<> EIGEN_STRONG_INLINE void pstore<double>(double*   to, const Packet2d& from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_STORE
+  Packet *vto;
+  vto = (Packet *) to;
+  vto->v2d = from;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)
+{
+  return vec_splats(from);
+}
+template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) {
+  return vec_splats(from);
+}
+template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&    from)
+{
+  Packet4f to;
+  to.v4f[0] = pset1<Packet2d>(static_cast<const double&>(from));
+  to.v4f[1] = to.v4f[0];
+  return to;
+}
+
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4i>(const int *a,
+                      Packet4i& a0, Packet4i& a1, Packet4i& a2, Packet4i& a3)
+{
+  a3 = pload<Packet4i>(a);
+  a0 = vec_splat(a3, 0);
+  a1 = vec_splat(a3, 1);
+  a2 = vec_splat(a3, 2);
+  a3 = vec_splat(a3, 3);
+}
+
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4f>(const float *a,
+                      Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
+{
+  a3 = pload<Packet4f>(a);
+  a0 = vec_splat_packet4f<0>(a3);
+  a1 = vec_splat_packet4f<1>(a3);
+  a2 = vec_splat_packet4f<2>(a3);
+  a3 = vec_splat_packet4f<3>(a3);
+}
+
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet2d>(const double *a,
+                      Packet2d& a0, Packet2d& a1, Packet2d& a2, Packet2d& a3)
+{
+  a1 = pload<Packet2d>(a);
+  a0 = vec_splat(a1, 0);
+  a1 = vec_splat(a1, 1);
+  a3 = pload<Packet2d>(a+2);
+  a2 = vec_splat(a3, 0);
+  a3 = vec_splat(a3, 1);
+}
+
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  ai[0] = from[0*stride];
+  ai[1] = from[1*stride];
+  ai[2] = from[2*stride];
+  ai[3] = from[3*stride];
+ return pload<Packet4i>(ai);
+}
+
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
+{
+  float EIGEN_ALIGN16 ai[4];
+  ai[0] = from[0*stride];
+  ai[1] = from[1*stride];
+  ai[2] = from[2*stride];
+  ai[3] = from[3*stride];
+ return pload<Packet4f>(ai);
+}
+
+template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride)
+{
+  double EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+ return pload<Packet2d>(af);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  pstore<int>((int *)ai, from);
+  to[0*stride] = ai[0];
+  to[1*stride] = ai[1];
+  to[2*stride] = ai[2];
+  to[3*stride] = ai[3];
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
+{
+  float EIGEN_ALIGN16 ai[4];
+  pstore<float>((float *)ai, from);
+  to[0*stride] = ai[0];
+  to[1*stride] = ai[1];
+  to[2*stride] = ai[2];
+  to[3*stride] = ai[3];
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride)
+{
+  double EIGEN_ALIGN16 af[2];
+  pstore<double>(af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a + b); }
+template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] + b.v4f[0];
+  c.v4f[1] = a.v4f[1] + b.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a + b); }
+
+template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a - b); }
+template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] - b.v4f[0];
+  c.v4f[1] = a.v4f[1] - b.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a - b); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a * b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] * b.v4f[0];
+  c.v4f[1] = a.v4f[1] * b.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a * b); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a / b); }
+template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] / b.v4f[0];
+  c.v4f[1] = a.v4f[1] / b.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a / b); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return (-a); }
+template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a)
+{
+  Packet4f c;
+  c.v4f[0] = -a.v4f[0];
+  c.v4f[1] = -a.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return (-a); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return padd<Packet4i>(pmul<Packet4i>(a, b), c); }
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c)
+{
+  Packet4f res;
+  res.v4f[0] = vec_madd(a.v4f[0], b.v4f[0], c.v4f[0]);
+  res.v4f[1] = vec_madd(a.v4f[1], b.v4f[1], c.v4f[1]);
+  return res;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vec_madd(a, b, c); }
+
+template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a)    { return padd<Packet4i>(pset1<Packet4i>(a), p4i_COUNTDOWN); }
+template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a)  { return padd<Packet4f>(pset1<Packet4f>(a), p4f_COUNTDOWN); }
+template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return padd<Packet2d>(pset1<Packet2d>(a), p2d_COUNTDOWN); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pmin<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_min(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_min(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pmin(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pmin(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_max(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d pmax<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_max(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pmax(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pmax(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pand(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pand(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i por<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_or(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d por<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_or(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f por<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pand(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pand(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pxor<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_xor(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_xor(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pxor<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pand(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pand(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return pand<Packet4i>(a, vec_nor(b, b)); }
+template<> EIGEN_STRONG_INLINE Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, vec_nor(b, b)); }
+template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pandnot(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pandnot(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = vec_round(a.v4f[0]);
+  res.v4f[1] = vec_round(a.v4f[1]);
+  return res;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return vec_round(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const  Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = vec_ceil(a.v4f[0]);
+  res.v4f[1] = vec_ceil(a.v4f[1]);
+  return res;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const  Packet2d& a) { return vec_ceil(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = vec_floor(a.v4f[0]);
+  res.v4f[1] = vec_floor(a.v4f[1]);
+  return res;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return vec_floor(a); }
+
+template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int*       from) { return pload<Packet4i>(from); }
+template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float*     from) { return pload<Packet4f>(from); }
+template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double*    from) { return pload<Packet2d>(from); }
+
+
+template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
+{
+  Packet4i p = pload<Packet4i>(from);
+  return vec_perm(p, p, p16uc_DUPLICATE32_HI);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*    from)
+{
+  Packet4f p = pload<Packet4f>(from);
+  p.v4f[1] = vec_splat(p.v4f[0], 1);
+  p.v4f[0] = vec_splat(p.v4f[0], 0);
+  return p;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
+{
+  Packet2d p = pload<Packet2d>(from);
+  return vec_perm(p, p, p16uc_PSET64_HI);
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*        to, const Packet4i& from) { pstore<int>(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*    to, const Packet4f& from) { pstore<float>(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double*  to, const Packet2d& from) { pstore<double>(to, from); }
+
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+
+template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int    EIGEN_ALIGN16 x[4]; pstore(x, a); return x[0]; }
+template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float  EIGEN_ALIGN16 x[2]; vec_st2f(a.v4f[0], &x[0]); return x[0]; }
+template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { double EIGEN_ALIGN16 x[2]; pstore(x, a); return x[0]; }
+
+template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a)
+{
+  return reinterpret_cast<Packet4i>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a)
+{
+  return reinterpret_cast<Packet2d>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE64));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a)
+{
+  Packet4f rev;
+  rev.v4f[0] = preverse<Packet2d>(a.v4f[1]);
+  rev.v4f[1] = preverse<Packet2d>(a.v4f[0]);
+  return rev;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pabs<Packet4i>(const Packet4i& a) { return vec_abs(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pabs<Packet2d>(const Packet2d& a) { return vec_abs(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pabs<Packet4f>(const Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = pabs(a.v4f[0]);
+  res.v4f[1] = pabs(a.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
+{
+  Packet4i b, sum;
+  b   = vec_sld(a, a, 8);
+  sum = padd<Packet4i>(a, b);
+  b   = vec_sld(sum, sum, 4);
+  sum = padd<Packet4i>(sum, b);
+  return pfirst(sum);
+}
+
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
+{
+  Packet2d b, sum;
+  b   = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8));
+  sum = padd<Packet2d>(a, b);
+  return pfirst(sum);
+}
+template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
+{
+  Packet2d sum;
+  sum = padd<Packet2d>(a.v4f[0], a.v4f[1]);
+  double first = predux<Packet2d>(sum);
+  return static_cast<float>(first);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
+{
+  Packet4i v[4], sum[4];
+
+  // It's easier and faster to transpose then add as columns
+  // Check: http://www.freevec.org/function/matrix_4x4_transpose_floats for explanation
+  // Do the transpose, first set of moves
+  v[0] = vec_mergeh(vecs[0], vecs[2]);
+  v[1] = vec_mergel(vecs[0], vecs[2]);
+  v[2] = vec_mergeh(vecs[1], vecs[3]);
+  v[3] = vec_mergel(vecs[1], vecs[3]);
+  // Get the resulting vectors
+  sum[0] = vec_mergeh(v[0], v[2]);
+  sum[1] = vec_mergel(v[0], v[2]);
+  sum[2] = vec_mergeh(v[1], v[3]);
+  sum[3] = vec_mergel(v[1], v[3]);
+
+  // Now do the summation:
+  // Lines 0+1
+  sum[0] = padd<Packet4i>(sum[0], sum[1]);
+  // Lines 2+3
+  sum[1] = padd<Packet4i>(sum[2], sum[3]);
+  // Add the results
+  sum[0] = padd<Packet4i>(sum[0], sum[1]);
+
+  return sum[0];
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
+{
+  Packet2d v[2], sum;
+  v[0] = padd<Packet2d>(vecs[0], reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(vecs[0]), reinterpret_cast<Packet4ui>(vecs[0]), 8)));
+  v[1] = padd<Packet2d>(vecs[1], reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(vecs[1]), reinterpret_cast<Packet4ui>(vecs[1]), 8)));
+ 
+  sum = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(v[0]), reinterpret_cast<Packet4ui>(v[1]), 8));
+
+  return sum;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
+{
+  PacketBlock<Packet4f,4> transpose;
+  transpose.packet[0] = vecs[0];
+  transpose.packet[1] = vecs[1];
+  transpose.packet[2] = vecs[2];
+  transpose.packet[3] = vecs[3];
+  ptranspose(transpose);
+
+  Packet4f sum = padd(transpose.packet[0], transpose.packet[1]);
+  sum = padd(sum, transpose.packet[2]);
+  sum = padd(sum, transpose.packet[3]);
+  return sum;
+}
+
+// Other reduction functions:
+// mul
+template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
+{
+  EIGEN_ALIGN16 int aux[4];
+  pstore(aux, a);
+  return aux[0] * aux[1] * aux[2] * aux[3];
+}
+
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmul(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8))));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
+{
+  // Return predux_mul<Packet2d> of the subvectors product
+  return static_cast<float>(pfirst(predux_mul(pmul(a.v4f[0], a.v4f[1]))));
+}
+
+// min
+template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
+{
+  Packet4i b, res;
+  b   = pmin<Packet4i>(a, vec_sld(a, a, 8));
+  res = pmin<Packet4i>(b, vec_sld(b, b, 4));
+  return pfirst(res);
+}
+
+template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmin<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8))));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a)
+{
+  Packet2d b, res;
+  b   = pmin<Packet2d>(a.v4f[0], a.v4f[1]);
+  res = pmin<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8)));
+  return static_cast<float>(pfirst(res));
+}
+
+// max
+template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
+{
+  Packet4i b, res;
+  b = pmax<Packet4i>(a, vec_sld(a, a, 8));
+  res = pmax<Packet4i>(b, vec_sld(b, b, 4));
+  return pfirst(res);
+}
+
+// max
+template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmax<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8))));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a)
+{
+  Packet2d b, res;
+  b   = pmax<Packet2d>(a.v4f[0], a.v4f[1]);
+  res = pmax<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8)));
+  return static_cast<float>(pfirst(res));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4i,4>& kernel) {
+  Packet4i t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
+  Packet4i t1 = vec_mergel(kernel.packet[0], kernel.packet[2]);
+  Packet4i t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]);
+  Packet4i t3 = vec_mergel(kernel.packet[1], kernel.packet[3]);
+  kernel.packet[0] = vec_mergeh(t0, t2);
+  kernel.packet[1] = vec_mergel(t0, t2);
+  kernel.packet[2] = vec_mergeh(t1, t3);
+  kernel.packet[3] = vec_mergel(t1, t3);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2d,2>& kernel) {
+  Packet2d t0 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_HI);
+  Packet2d t1 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_LO);
+  kernel.packet[0] = t0;
+  kernel.packet[1] = t1;
+}
+
+/* Split the Packet4f PacketBlock into 4 Packet2d PacketBlocks and transpose each one
+ */
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4f,4>& kernel) {
+  PacketBlock<Packet2d,2> t0,t1,t2,t3;
+  // copy top-left 2x2 Packet2d block
+  t0.packet[0] = kernel.packet[0].v4f[0];
+  t0.packet[1] = kernel.packet[1].v4f[0];
+
+  // copy top-right 2x2 Packet2d block
+  t1.packet[0] = kernel.packet[0].v4f[1];
+  t1.packet[1] = kernel.packet[1].v4f[1];
+
+  // copy bottom-left 2x2 Packet2d block
+  t2.packet[0] = kernel.packet[2].v4f[0];
+  t2.packet[1] = kernel.packet[3].v4f[0];
+
+  // copy bottom-right 2x2 Packet2d block
+  t3.packet[0] = kernel.packet[2].v4f[1];
+  t3.packet[1] = kernel.packet[3].v4f[1];
+
+  // Transpose all 2x2 blocks
+  ptranspose(t0);
+  ptranspose(t1);
+  ptranspose(t2);
+  ptranspose(t3);
+
+  // Copy back transposed blocks, but exchange t1 and t2 due to transposition
+  kernel.packet[0].v4f[0] = t0.packet[0];
+  kernel.packet[0].v4f[1] = t2.packet[0];
+  kernel.packet[1].v4f[0] = t0.packet[1];
+  kernel.packet[1].v4f[1] = t2.packet[1];
+  kernel.packet[2].v4f[0] = t1.packet[0];
+  kernel.packet[2].v4f[1] = t3.packet[0];
+  kernel.packet[3].v4f[0] = t1.packet[1];
+  kernel.packet[3].v4f[1] = t3.packet[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pblend(const Selector<4>& ifPacket, const Packet4i& thenPacket, const Packet4i& elsePacket) {
+  Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] };
+  Packet4ui mask = vec_cmpeq(select, reinterpret_cast<Packet4ui>(p4i_ONE));
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) {
+  Packet2ul select_hi = { ifPacket.select[0], ifPacket.select[1] };
+  Packet2ul select_lo = { ifPacket.select[2], ifPacket.select[3] };
+  Packet2ul mask_hi = vec_cmpeq(select_hi, reinterpret_cast<Packet2ul>(p2l_ONE));
+  Packet2ul mask_lo = vec_cmpeq(select_lo, reinterpret_cast<Packet2ul>(p2l_ONE));
+  Packet4f result;
+  result.v4f[0] = vec_sel(elsePacket.v4f[0], thenPacket.v4f[0], mask_hi);
+  result.v4f[1] = vec_sel(elsePacket.v4f[1], thenPacket.v4f[1], mask_lo);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, const Packet2d& thenPacket, const Packet2d& elsePacket) {
+  Packet2ul select = { ifPacket.select[0], ifPacket.select[1] };
+  Packet2ul mask = vec_cmpeq(select, reinterpret_cast<Packet2ul>(p2l_ONE));
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PACKET_MATH_ZVECTOR_H
diff --git a/extern/Eigen3/Eigen/src/Core/functors/AssignmentFunctors.h b/extern/Eigen3/Eigen/src/Core/functors/AssignmentFunctors.h
new file mode 100644
index 00000000000..4153b877cff
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/functors/AssignmentFunctors.h
@@ -0,0 +1,168 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ASSIGNMENT_FUNCTORS_H
+#define EIGEN_ASSIGNMENT_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+  
+/** \internal
+  * \brief Template functor for scalar/packet assignment
+  *
+  */
+template<typename DstScalar,typename SrcScalar> struct assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a = b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,b); }
+};
+
+// Empty overload for void type (used by PermutationMatrix)
+template<typename DstScalar> struct assign_op<DstScalar,void> {};
+
+template<typename DstScalar,typename SrcScalar>
+struct functor_traits<assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::Vectorizable && packet_traits<SrcScalar>::Vectorizable
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with addition
+  *
+  */
+template<typename DstScalar,typename SrcScalar> struct add_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(add_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a += b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,internal::padd(internal::ploadt<Packet,Alignment>(a),b)); }
+};
+template<typename DstScalar,typename SrcScalar>
+struct functor_traits<add_assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost + NumTraits<DstScalar>::AddCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::HasAdd
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with subtraction
+  *
+  */
+template<typename DstScalar,typename SrcScalar> struct sub_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(sub_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a -= b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,internal::psub(internal::ploadt<Packet,Alignment>(a),b)); }
+};
+template<typename DstScalar,typename SrcScalar>
+struct functor_traits<sub_assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost + NumTraits<DstScalar>::AddCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::HasSub
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with multiplication
+  *
+  */
+template<typename DstScalar, typename SrcScalar=DstScalar>
+struct mul_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(mul_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a *= b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,internal::pmul(internal::ploadt<Packet,Alignment>(a),b)); }
+};
+template<typename DstScalar, typename SrcScalar>
+struct functor_traits<mul_assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost + NumTraits<DstScalar>::MulCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::HasMul
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with diviving
+  *
+  */
+template<typename DstScalar, typename SrcScalar=DstScalar> struct div_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(div_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a /= b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,internal::pdiv(internal::ploadt<Packet,Alignment>(a),b)); }
+};
+template<typename DstScalar, typename SrcScalar>
+struct functor_traits<div_assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost + NumTraits<DstScalar>::MulCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::HasDiv
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with swapping
+  *
+  * It works as follow. For a non-vectorized evaluation loop, we have:
+  *   for(i) func(A.coeffRef(i), B.coeff(i));
+  * where B is a SwapWrapper expression. The trick is to make SwapWrapper::coeff behaves like a non-const coeffRef.
+  * Actually, SwapWrapper might not even be needed since even if B is a plain expression, since it has to be writable
+  * B.coeff already returns a const reference to the underlying scalar value.
+  * 
+  * The case of a vectorized loop is more tricky:
+  *   for(i,j) func.assignPacket<A_Align>(&A.coeffRef(i,j), B.packet<B_Align>(i,j));
+  * Here, B must be a SwapWrapper whose packet function actually returns a proxy object holding a Scalar*,
+  * the actual alignment and Packet type.
+  *
+  */
+template<typename Scalar> struct swap_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(swap_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Scalar& a, const Scalar& b) const
+  {
+#ifdef __CUDACC__
+    // FIXME is there some kind of cuda::swap?
+    Scalar t=b; const_cast<Scalar&>(b)=a; a=t;
+#else
+    using std::swap;
+    swap(a,const_cast<Scalar&>(b));
+#endif
+  }
+};
+template<typename Scalar>
+struct functor_traits<swap_assign_op<Scalar> > {
+  enum {
+    Cost = 3 * NumTraits<Scalar>::ReadCost,
+    PacketAccess = packet_traits<Scalar>::Vectorizable
+  };
+};
+
+} // namespace internal
+
+} // namespace Eigen
+
+#endif // EIGEN_ASSIGNMENT_FUNCTORS_H
diff --git a/extern/Eigen3/Eigen/src/Core/functors/BinaryFunctors.h b/extern/Eigen3/Eigen/src/Core/functors/BinaryFunctors.h
new file mode 100644
index 00000000000..3eae6b8cada
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/functors/BinaryFunctors.h
@@ -0,0 +1,475 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_BINARY_FUNCTORS_H
+#define EIGEN_BINARY_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+//---------- associative binary functors ----------
+
+template<typename Arg1, typename Arg2>
+struct binary_op_base
+{
+  typedef Arg1 first_argument_type;
+  typedef Arg2 second_argument_type;
+};
+
+/** \internal
+  * \brief Template functor to compute the sum of two scalars
+  *
+  * \sa class CwiseBinaryOp, MatrixBase::operator+, class VectorwiseOp, DenseBase::sum()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_sum_op : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_sum_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sum_op)
+#else
+  scalar_sum_op() {
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a + b; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::padd(a,b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
+  { return internal::predux(a); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_sum_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2, // rough estimate!
+    PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasAdd && packet_traits<RhsScalar>::HasAdd
+    // TODO vectorize mixed sum
+  };
+};
+
+/** \internal
+  * \brief Template specialization to deprecate the summation of boolean expressions.
+  * This is required to solve Bug 426.
+  * \sa DenseBase::count(), DenseBase::any(), ArrayBase::cast(), MatrixBase::cast()
+  */
+template<> struct scalar_sum_op<bool,bool> : scalar_sum_op<int,int> {
+  EIGEN_DEPRECATED
+  scalar_sum_op() {}
+};
+
+
+/** \internal
+  * \brief Template functor to compute the product of two scalars
+  *
+  * \sa class CwiseBinaryOp, Cwise::operator*(), class VectorwiseOp, MatrixBase::redux()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_product_op  : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_product_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_product_op)
+#else
+  scalar_product_op() {
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a * b; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pmul(a,b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
+  { return internal::predux_mul(a); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_product_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::MulCost + NumTraits<RhsScalar>::MulCost)/2, // rough estimate!
+    PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasMul && packet_traits<RhsScalar>::HasMul
+    // TODO vectorize mixed product
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the conjugate product of two scalars
+  *
+  * This is a short cut for conj(x) * y which is needed for optimization purpose; in Eigen2 support mode, this becomes x * conj(y)
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_conj_product_op  : binary_op_base<LhsScalar,RhsScalar>
+{
+
+  enum {
+    Conj = NumTraits<LhsScalar>::IsComplex
+  };
+  
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_conj_product_op>::ReturnType result_type;
+  
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_conj_product_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const
+  { return conj_helper<LhsScalar,RhsScalar,Conj,false>().pmul(a,b); }
+  
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return conj_helper<Packet,Packet,Conj,false>().pmul(a,b); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_conj_product_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = NumTraits<LhsScalar>::MulCost,
+    PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMul
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the min of two scalars
+  *
+  * \sa class CwiseBinaryOp, MatrixBase::cwiseMin, class VectorwiseOp, MatrixBase::minCoeff()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_min_op : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_min_op>::ReturnType result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_min_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return numext::mini(a, b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pmin(a,b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
+  { return internal::predux_min(a); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_min_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
+    PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMin
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the max of two scalars
+  *
+  * \sa class CwiseBinaryOp, MatrixBase::cwiseMax, class VectorwiseOp, MatrixBase::maxCoeff()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_max_op  : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_max_op>::ReturnType result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_max_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return numext::maxi(a, b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pmax(a,b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
+  { return internal::predux_max(a); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_max_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
+    PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMax
+  };
+};
+
+/** \internal
+  * \brief Template functors for comparison of two scalars
+  * \todo Implement packet-comparisons
+  */
+template<typename LhsScalar, typename RhsScalar, ComparisonName cmp> struct scalar_cmp_op;
+
+template<typename LhsScalar, typename RhsScalar, ComparisonName cmp>
+struct functor_traits<scalar_cmp_op<LhsScalar,RhsScalar, cmp> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
+    PacketAccess = false
+  };
+};
+
+template<ComparisonName Cmp, typename LhsScalar, typename RhsScalar>
+struct result_of<scalar_cmp_op<LhsScalar, RhsScalar, Cmp>(LhsScalar,RhsScalar)> {
+  typedef bool type;
+};
+
+
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_EQ> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a==b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_LT> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a<b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_LE> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a<=b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_GT> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a>b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_GE> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a>=b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_UNORD> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return !(a<=b || b<=a);}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_NEQ> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a!=b;}
+};
+
+
+/** \internal
+  * \brief Template functor to compute the hypot of two \b positive \b and \b real scalars
+  *
+  * \sa MatrixBase::stableNorm(), class Redux
+  */
+template<typename Scalar>
+struct scalar_hypot_op<Scalar,Scalar> : binary_op_base<Scalar,Scalar>
+{
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_hypot_op)
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar &x, const Scalar &y) const
+  {
+    // This functor is used by hypotNorm only for which it is faster to first apply abs
+    // on all coefficients prior to reduction through hypot.
+    // This way we avoid calling abs on positive and real entries, and this also permits
+    // to seamlessly handle complexes. Otherwise we would have to handle both real and complexes
+    // through the same functor...
+    return internal::positive_real_hypot(x,y);
+  }
+};
+template<typename Scalar>
+struct functor_traits<scalar_hypot_op<Scalar,Scalar> > {
+  enum
+  {
+    Cost = 3 * NumTraits<Scalar>::AddCost +
+           2 * NumTraits<Scalar>::MulCost +
+           2 * scalar_div_cost<Scalar,false>::value,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the pow of two scalars
+  */
+template<typename Scalar, typename Exponent>
+struct scalar_pow_op  : binary_op_base<Scalar,Exponent>
+{
+  typedef typename ScalarBinaryOpTraits<Scalar,Exponent,scalar_pow_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_pow_op)
+#else
+  scalar_pow_op() {
+    typedef Scalar LhsScalar;
+    typedef Exponent RhsScalar;
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC
+  inline result_type operator() (const Scalar& a, const Exponent& b) const { return numext::pow(a, b); }
+};
+template<typename Scalar, typename Exponent>
+struct functor_traits<scalar_pow_op<Scalar,Exponent> > {
+  enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false };
+};
+
+
+
+//---------- non associative binary functors ----------
+
+/** \internal
+  * \brief Template functor to compute the difference of two scalars
+  *
+  * \sa class CwiseBinaryOp, MatrixBase::operator-
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_difference_op : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_difference_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_difference_op)
+#else
+  scalar_difference_op() {
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a - b; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::psub(a,b); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_difference_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
+    PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasSub && packet_traits<RhsScalar>::HasSub
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the quotient of two scalars
+  *
+  * \sa class CwiseBinaryOp, Cwise::operator/()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_quotient_op  : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_quotient_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_quotient_op)
+#else
+  scalar_quotient_op() {
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a / b; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pdiv(a,b); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_quotient_op<LhsScalar,RhsScalar> > {
+  typedef typename scalar_quotient_op<LhsScalar,RhsScalar>::result_type result_type;
+  enum {
+    PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasDiv && packet_traits<RhsScalar>::HasDiv,
+    Cost = scalar_div_cost<result_type,PacketAccess>::value
+  };
+};
+
+
+
+/** \internal
+  * \brief Template functor to compute the and of two booleans
+  *
+  * \sa class CwiseBinaryOp, ArrayBase::operator&&
+  */
+struct scalar_boolean_and_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_and_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a && b; }
+};
+template<> struct functor_traits<scalar_boolean_and_op> {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the or of two booleans
+  *
+  * \sa class CwiseBinaryOp, ArrayBase::operator||
+  */
+struct scalar_boolean_or_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_or_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a || b; }
+};
+template<> struct functor_traits<scalar_boolean_or_op> {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+ * \brief Template functor to compute the xor of two booleans
+ *
+ * \sa class CwiseBinaryOp, ArrayBase::operator^
+ */
+struct scalar_boolean_xor_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_xor_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a ^ b; }
+};
+template<> struct functor_traits<scalar_boolean_xor_op> {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+
+
+//---------- binary functors bound to a constant, thus appearing as a unary functor ----------
+
+// The following two classes permits to turn any binary functor into a unary one with one argument bound to a constant value.
+// They are analogues to std::binder1st/binder2nd but with the following differences:
+//  - they are compatible with packetOp
+//  - they are portable across C++ versions (the std::binder* are deprecated in C++11)
+template<typename BinaryOp> struct bind1st_op : BinaryOp {
+
+  typedef typename BinaryOp::first_argument_type  first_argument_type;
+  typedef typename BinaryOp::second_argument_type second_argument_type;
+  typedef typename BinaryOp::result_type          result_type;
+
+  bind1st_op(const first_argument_type &val) : m_value(val) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const second_argument_type& b) const { return BinaryOp::operator()(m_value,b); }
+
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& b) const
+  { return BinaryOp::packetOp(internal::pset1<Packet>(m_value), b); }
+
+  first_argument_type m_value;
+};
+template<typename BinaryOp> struct functor_traits<bind1st_op<BinaryOp> > : functor_traits<BinaryOp> {};
+
+
+template<typename BinaryOp> struct bind2nd_op : BinaryOp {
+
+  typedef typename BinaryOp::first_argument_type  first_argument_type;
+  typedef typename BinaryOp::second_argument_type second_argument_type;
+  typedef typename BinaryOp::result_type          result_type;
+
+  bind2nd_op(const second_argument_type &val) : m_value(val) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const first_argument_type& a) const { return BinaryOp::operator()(a,m_value); }
+
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return BinaryOp::packetOp(a,internal::pset1<Packet>(m_value)); }
+
+  second_argument_type m_value;
+};
+template<typename BinaryOp> struct functor_traits<bind2nd_op<BinaryOp> > : functor_traits<BinaryOp> {};
+
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_BINARY_FUNCTORS_H
diff --git a/extern/Eigen3/Eigen/src/Core/functors/NullaryFunctors.h b/extern/Eigen3/Eigen/src/Core/functors/NullaryFunctors.h
new file mode 100644
index 00000000000..b03be0269c9
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/functors/NullaryFunctors.h
@@ -0,0 +1,188 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_NULLARY_FUNCTORS_H
+#define EIGEN_NULLARY_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<typename Scalar>
+struct scalar_constant_op {
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE scalar_constant_op(const scalar_constant_op& other) : m_other(other.m_other) { }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE scalar_constant_op(const Scalar& other) : m_other(other) { }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() () const { return m_other; }
+  template<typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const PacketType packetOp() const { return internal::pset1<PacketType>(m_other); }
+  const Scalar m_other;
+};
+template<typename Scalar>
+struct functor_traits<scalar_constant_op<Scalar> >
+{ enum { Cost = 0 /* as the constant value should be loaded in register only once for the whole expression */,
+         PacketAccess = packet_traits<Scalar>::Vectorizable, IsRepeatable = true }; };
+
+template<typename Scalar> struct scalar_identity_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_identity_op)
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (IndexType row, IndexType col) const { return row==col ? Scalar(1) : Scalar(0); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_identity_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = false, IsRepeatable = true }; };
+
+template <typename Scalar, typename Packet, bool IsInteger> struct linspaced_op_impl;
+
+template <typename Scalar, typename Packet>
+struct linspaced_op_impl<Scalar,Packet,/*IsInteger*/false>
+{
+  linspaced_op_impl(const Scalar& low, const Scalar& high, Index num_steps) :
+    m_low(low), m_high(high), m_size1(num_steps==1 ? 1 : num_steps-1), m_step(num_steps==1 ? Scalar() : (high-low)/Scalar(num_steps-1)),
+    m_flip(numext::abs(high)<numext::abs(low))
+  {}
+
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (IndexType i) const {
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    if(m_flip)
+      return (i==0)? m_low : (m_high - RealScalar(m_size1-i)*m_step);
+    else
+      return (i==m_size1)? m_high : (m_low + RealScalar(i)*m_step);
+  }
+
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(IndexType i) const
+  {
+    // Principle:
+    // [low, ..., low] + ( [step, ..., step] * ( [i, ..., i] + [0, ..., size] ) )
+    if(m_flip)
+    {
+      Packet pi = plset<Packet>(Scalar(i-m_size1));
+      Packet res = padd(pset1<Packet>(m_high), pmul(pset1<Packet>(m_step), pi));
+      if(i==0)
+        res = pinsertfirst(res, m_low);
+      return res;
+    }
+    else
+    {
+      Packet pi = plset<Packet>(Scalar(i));
+      Packet res = padd(pset1<Packet>(m_low), pmul(pset1<Packet>(m_step), pi));
+      if(i==m_size1-unpacket_traits<Packet>::size+1)
+        res = pinsertlast(res, m_high);
+      return res;
+    }
+  }
+
+  const Scalar m_low;
+  const Scalar m_high;
+  const Index m_size1;
+  const Scalar m_step;
+  const bool m_flip;
+};
+
+template <typename Scalar, typename Packet>
+struct linspaced_op_impl<Scalar,Packet,/*IsInteger*/true>
+{
+  linspaced_op_impl(const Scalar& low, const Scalar& high, Index num_steps) :
+    m_low(low),
+    m_multiplier((high-low)/convert_index<Scalar>(num_steps<=1 ? 1 : num_steps-1)),
+    m_divisor(convert_index<Scalar>((high>=low?num_steps:-num_steps)+(high-low))/((numext::abs(high-low)+1)==0?1:(numext::abs(high-low)+1))),
+    m_use_divisor(num_steps>1 && (numext::abs(high-low)+1)<num_steps)
+  {}
+
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar operator() (IndexType i) const
+  {
+    if(m_use_divisor) return m_low + convert_index<Scalar>(i)/m_divisor;
+    else              return m_low + convert_index<Scalar>(i)*m_multiplier;
+  }
+
+  const Scalar m_low;
+  const Scalar m_multiplier;
+  const Scalar m_divisor;
+  const bool m_use_divisor;
+};
+
+// ----- Linspace functor ----------------------------------------------------------------
+
+// Forward declaration (we default to random access which does not really give
+// us a speed gain when using packet access but it allows to use the functor in
+// nested expressions).
+template <typename Scalar, typename PacketType> struct linspaced_op;
+template <typename Scalar, typename PacketType> struct functor_traits< linspaced_op<Scalar,PacketType> >
+{
+  enum
+  {
+    Cost = 1,
+    PacketAccess =   (!NumTraits<Scalar>::IsInteger) && packet_traits<Scalar>::HasSetLinear && packet_traits<Scalar>::HasBlend,
+                  /*&& ((!NumTraits<Scalar>::IsInteger) || packet_traits<Scalar>::HasDiv),*/ // <- vectorization for integer is currently disabled
+    IsRepeatable = true
+  };
+};
+template <typename Scalar, typename PacketType> struct linspaced_op
+{
+  linspaced_op(const Scalar& low, const Scalar& high, Index num_steps)
+    : impl((num_steps==1 ? high : low),high,num_steps)
+  {}
+
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (IndexType i) const { return impl(i); }
+
+  template<typename Packet,typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(IndexType i) const { return impl.packetOp(i); }
+
+  // This proxy object handles the actual required temporaries and the different
+  // implementations (integer vs. floating point).
+  const linspaced_op_impl<Scalar,PacketType,NumTraits<Scalar>::IsInteger> impl;
+};
+
+// Linear access is automatically determined from the operator() prototypes available for the given functor.
+// If it exposes an operator()(i,j), then we assume the i and j coefficients are required independently
+// and linear access is not possible. In all other cases, linear access is enabled.
+// Users should not have to deal with this structure.
+template<typename Functor> struct functor_has_linear_access { enum { ret = !has_binary_operator<Functor>::value }; };
+
+// For unreliable compilers, let's specialize the has_*ary_operator
+// helpers so that at least built-in nullary functors work fine.
+#if !( (EIGEN_COMP_MSVC>1600) || (EIGEN_GNUC_AT_LEAST(4,8)) || (EIGEN_COMP_ICC>=1600))
+template<typename Scalar,typename IndexType>
+struct has_nullary_operator<scalar_constant_op<Scalar>,IndexType> { enum { value = 1}; };
+template<typename Scalar,typename IndexType>
+struct has_unary_operator<scalar_constant_op<Scalar>,IndexType> { enum { value = 0}; };
+template<typename Scalar,typename IndexType>
+struct has_binary_operator<scalar_constant_op<Scalar>,IndexType> { enum { value = 0}; };
+
+template<typename Scalar,typename IndexType>
+struct has_nullary_operator<scalar_identity_op<Scalar>,IndexType> { enum { value = 0}; };
+template<typename Scalar,typename IndexType>
+struct has_unary_operator<scalar_identity_op<Scalar>,IndexType> { enum { value = 0}; };
+template<typename Scalar,typename IndexType>
+struct has_binary_operator<scalar_identity_op<Scalar>,IndexType> { enum { value = 1}; };
+
+template<typename Scalar, typename PacketType,typename IndexType>
+struct has_nullary_operator<linspaced_op<Scalar,PacketType>,IndexType> { enum { value = 0}; };
+template<typename Scalar, typename PacketType,typename IndexType>
+struct has_unary_operator<linspaced_op<Scalar,PacketType>,IndexType> { enum { value = 1}; };
+template<typename Scalar, typename PacketType,typename IndexType>
+struct has_binary_operator<linspaced_op<Scalar,PacketType>,IndexType> { enum { value = 0}; };
+
+template<typename Scalar,typename IndexType>
+struct has_nullary_operator<scalar_random_op<Scalar>,IndexType> { enum { value = 1}; };
+template<typename Scalar,typename IndexType>
+struct has_unary_operator<scalar_random_op<Scalar>,IndexType> { enum { value = 0}; };
+template<typename Scalar,typename IndexType>
+struct has_binary_operator<scalar_random_op<Scalar>,IndexType> { enum { value = 0}; };
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_NULLARY_FUNCTORS_H
diff --git a/extern/Eigen3/Eigen/src/Core/functors/StlFunctors.h b/extern/Eigen3/Eigen/src/Core/functors/StlFunctors.h
new file mode 100644
index 00000000000..9c1d75850be
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/functors/StlFunctors.h
@@ -0,0 +1,136 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_STL_FUNCTORS_H
+#define EIGEN_STL_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+// default functor traits for STL functors:
+
+template<typename T>
+struct functor_traits<std::multiplies<T> >
+{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::divides<T> >
+{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::plus<T> >
+{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::minus<T> >
+{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::negate<T> >
+{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::logical_or<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::logical_and<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::logical_not<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::greater<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::less<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::greater_equal<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::less_equal<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::equal_to<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::not_equal_to<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+#if (__cplusplus < 201103L) && (EIGEN_COMP_MSVC <= 1900)
+// std::binder* are deprecated since c++11 and will be removed in c++17
+template<typename T>
+struct functor_traits<std::binder2nd<T> >
+{ enum { Cost = functor_traits<T>::Cost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::binder1st<T> >
+{ enum { Cost = functor_traits<T>::Cost, PacketAccess = false }; };
+#endif
+
+#if (__cplusplus < 201703L) && (EIGEN_COMP_MSVC < 1910)
+// std::unary_negate is deprecated since c++17 and will be removed in c++20
+template<typename T>
+struct functor_traits<std::unary_negate<T> >
+{ enum { Cost = 1 + functor_traits<T>::Cost, PacketAccess = false }; };
+
+// std::binary_negate is deprecated since c++17 and will be removed in c++20
+template<typename T>
+struct functor_traits<std::binary_negate<T> >
+{ enum { Cost = 1 + functor_traits<T>::Cost, PacketAccess = false }; };
+#endif
+
+#ifdef EIGEN_STDEXT_SUPPORT
+
+template<typename T0,typename T1>
+struct functor_traits<std::project1st<T0,T1> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::project2nd<T0,T1> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::select2nd<std::pair<T0,T1> > >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::select1st<std::pair<T0,T1> > >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::unary_compose<T0,T1> >
+{ enum { Cost = functor_traits<T0>::Cost + functor_traits<T1>::Cost, PacketAccess = false }; };
+
+template<typename T0,typename T1,typename T2>
+struct functor_traits<std::binary_compose<T0,T1,T2> >
+{ enum { Cost = functor_traits<T0>::Cost + functor_traits<T1>::Cost + functor_traits<T2>::Cost, PacketAccess = false }; };
+
+#endif // EIGEN_STDEXT_SUPPORT
+
+// allow to add new functors and specializations of functor_traits from outside Eigen.
+// this macro is really needed because functor_traits must be specialized after it is declared but before it is used...
+#ifdef EIGEN_FUNCTORS_PLUGIN
+#include EIGEN_FUNCTORS_PLUGIN
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_STL_FUNCTORS_H
diff --git a/extern/Eigen3/Eigen/src/Core/functors/TernaryFunctors.h b/extern/Eigen3/Eigen/src/Core/functors/TernaryFunctors.h
new file mode 100644
index 00000000000..b254e96c6ad
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/functors/TernaryFunctors.h
@@ -0,0 +1,25 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Eugene Brevdo <ebrevdo@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_TERNARY_FUNCTORS_H
+#define EIGEN_TERNARY_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+//---------- associative ternary functors ----------
+
+
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TERNARY_FUNCTORS_H
diff --git a/extern/Eigen3/Eigen/src/Core/functors/UnaryFunctors.h b/extern/Eigen3/Eigen/src/Core/functors/UnaryFunctors.h
new file mode 100644
index 00000000000..2e6a00ffd1b
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/functors/UnaryFunctors.h
@@ -0,0 +1,792 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_UNARY_FUNCTORS_H
+#define EIGEN_UNARY_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+/** \internal
+  * \brief Template functor to compute the opposite of a scalar
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::operator-
+  */
+template<typename Scalar> struct scalar_opposite_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_opposite_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return -a; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::pnegate(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_opposite_op<Scalar> >
+{ enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = packet_traits<Scalar>::HasNegate };
+};
+
+/** \internal
+  * \brief Template functor to compute the absolute value of a scalar
+  *
+  * \sa class CwiseUnaryOp, Cwise::abs
+  */
+template<typename Scalar> struct scalar_abs_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_abs_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return numext::abs(a); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::pabs(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_abs_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = packet_traits<Scalar>::HasAbs
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the score of a scalar, to chose a pivot
+  *
+  * \sa class CwiseUnaryOp
+  */
+template<typename Scalar> struct scalar_score_coeff_op : scalar_abs_op<Scalar>
+{
+  typedef void Score_is_abs;
+};
+template<typename Scalar>
+struct functor_traits<scalar_score_coeff_op<Scalar> > : functor_traits<scalar_abs_op<Scalar> > {};
+
+/* Avoid recomputing abs when we know the score and they are the same. Not a true Eigen functor.  */
+template<typename Scalar, typename=void> struct abs_knowing_score
+{
+  EIGEN_EMPTY_STRUCT_CTOR(abs_knowing_score)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  template<typename Score>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a, const Score&) const { return numext::abs(a); }
+};
+template<typename Scalar> struct abs_knowing_score<Scalar, typename scalar_score_coeff_op<Scalar>::Score_is_abs>
+{
+  EIGEN_EMPTY_STRUCT_CTOR(abs_knowing_score)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  template<typename Scal>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const Scal&, const result_type& a) const { return a; }
+};
+
+/** \internal
+  * \brief Template functor to compute the squared absolute value of a scalar
+  *
+  * \sa class CwiseUnaryOp, Cwise::abs2
+  */
+template<typename Scalar> struct scalar_abs2_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_abs2_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return numext::abs2(a); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::pmul(a,a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_abs2_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasAbs2 }; };
+
+/** \internal
+  * \brief Template functor to compute the conjugate of a complex value
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::conjugate()
+  */
+template<typename Scalar> struct scalar_conjugate_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_conjugate_op)
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { using numext::conj; return conj(a); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const { return internal::pconj(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_conjugate_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::IsComplex ? NumTraits<Scalar>::AddCost : 0,
+    PacketAccess = packet_traits<Scalar>::HasConj
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the phase angle of a complex
+  *
+  * \sa class CwiseUnaryOp, Cwise::arg
+  */
+template<typename Scalar> struct scalar_arg_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_arg_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { using numext::arg; return arg(a); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::parg(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_arg_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::IsComplex ? 5 * NumTraits<Scalar>::MulCost : NumTraits<Scalar>::AddCost,
+    PacketAccess = packet_traits<Scalar>::HasArg
+  };
+};
+/** \internal
+  * \brief Template functor to cast a scalar to another type
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::cast()
+  */
+template<typename Scalar, typename NewType>
+struct scalar_cast_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
+  typedef NewType result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const NewType operator() (const Scalar& a) const { return cast<Scalar, NewType>(a); }
+};
+template<typename Scalar, typename NewType>
+struct functor_traits<scalar_cast_op<Scalar,NewType> >
+{ enum { Cost = is_same<Scalar, NewType>::value ? 0 : NumTraits<NewType>::AddCost, PacketAccess = false }; };
+
+/** \internal
+  * \brief Template functor to extract the real part of a complex
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::real()
+  */
+template<typename Scalar>
+struct scalar_real_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_real_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::real(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_real_op<Scalar> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+/** \internal
+  * \brief Template functor to extract the imaginary part of a complex
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::imag()
+  */
+template<typename Scalar>
+struct scalar_imag_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::imag(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_imag_op<Scalar> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+/** \internal
+  * \brief Template functor to extract the real part of a complex as a reference
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::real()
+  */
+template<typename Scalar>
+struct scalar_real_ref_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_real_ref_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::real_ref(*const_cast<Scalar*>(&a)); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_real_ref_op<Scalar> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+/** \internal
+  * \brief Template functor to extract the imaginary part of a complex as a reference
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::imag()
+  */
+template<typename Scalar>
+struct scalar_imag_ref_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_ref_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::imag_ref(*const_cast<Scalar*>(&a)); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_imag_ref_op<Scalar> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+/** \internal
+  *
+  * \brief Template functor to compute the exponential of a scalar
+  *
+  * \sa class CwiseUnaryOp, Cwise::exp()
+  */
+template<typename Scalar> struct scalar_exp_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_exp_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::exp(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pexp(a); }
+};
+template <typename Scalar>
+struct functor_traits<scalar_exp_op<Scalar> > {
+  enum {
+    PacketAccess = packet_traits<Scalar>::HasExp,
+    // The following numbers are based on the AVX implementation.
+#ifdef EIGEN_VECTORIZE_FMA
+    // Haswell can issue 2 add/mul/madd per cycle.
+    Cost =
+    (sizeof(Scalar) == 4
+     // float: 8 pmadd, 4 pmul, 2 padd/psub, 6 other
+     ? (8 * NumTraits<Scalar>::AddCost + 6 * NumTraits<Scalar>::MulCost)
+     // double: 7 pmadd, 5 pmul, 3 padd/psub, 1 div,  13 other
+     : (14 * NumTraits<Scalar>::AddCost +
+        6 * NumTraits<Scalar>::MulCost +
+        scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value))
+#else
+    Cost =
+    (sizeof(Scalar) == 4
+     // float: 7 pmadd, 6 pmul, 4 padd/psub, 10 other
+     ? (21 * NumTraits<Scalar>::AddCost + 13 * NumTraits<Scalar>::MulCost)
+     // double: 7 pmadd, 5 pmul, 3 padd/psub, 1 div,  13 other
+     : (23 * NumTraits<Scalar>::AddCost +
+        12 * NumTraits<Scalar>::MulCost +
+        scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value))
+#endif
+  };
+};
+
+/** \internal
+  *
+  * \brief Template functor to compute the logarithm of a scalar
+  *
+  * \sa class CwiseUnaryOp, ArrayBase::log()
+  */
+template<typename Scalar> struct scalar_log_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_log_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::log(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::plog(a); }
+};
+template <typename Scalar>
+struct functor_traits<scalar_log_op<Scalar> > {
+  enum {
+    PacketAccess = packet_traits<Scalar>::HasLog,
+    Cost =
+    (PacketAccess
+     // The following numbers are based on the AVX implementation.
+#ifdef EIGEN_VECTORIZE_FMA
+     // 8 pmadd, 6 pmul, 8 padd/psub, 16 other, can issue 2 add/mul/madd per cycle.
+     ? (20 * NumTraits<Scalar>::AddCost + 7 * NumTraits<Scalar>::MulCost)
+#else
+     // 8 pmadd, 6 pmul, 8 padd/psub, 20 other
+     ? (36 * NumTraits<Scalar>::AddCost + 14 * NumTraits<Scalar>::MulCost)
+#endif
+     // Measured cost of std::log.
+     : sizeof(Scalar)==4 ? 40 : 85)
+  };
+};
+
+/** \internal
+  *
+  * \brief Template functor to compute the logarithm of 1 plus a scalar value
+  *
+  * \sa class CwiseUnaryOp, ArrayBase::log1p()
+  */
+template<typename Scalar> struct scalar_log1p_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_log1p_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::log1p(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::plog1p(a); }
+};
+template <typename Scalar>
+struct functor_traits<scalar_log1p_op<Scalar> > {
+  enum {
+    PacketAccess = packet_traits<Scalar>::HasLog1p,
+    Cost = functor_traits<scalar_log_op<Scalar> >::Cost // TODO measure cost of log1p
+  };
+};
+
+/** \internal
+  *
+  * \brief Template functor to compute the base-10 logarithm of a scalar
+  *
+  * \sa class CwiseUnaryOp, Cwise::log10()
+  */
+template<typename Scalar> struct scalar_log10_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_log10_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { EIGEN_USING_STD_MATH(log10) return log10(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::plog10(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_log10_op<Scalar> >
+{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasLog10 }; };
+
+/** \internal
+  * \brief Template functor to compute the square root of a scalar
+  * \sa class CwiseUnaryOp, Cwise::sqrt()
+  */
+template<typename Scalar> struct scalar_sqrt_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sqrt_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::sqrt(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psqrt(a); }
+};
+template <typename Scalar>
+struct functor_traits<scalar_sqrt_op<Scalar> > {
+  enum {
+#if EIGEN_FAST_MATH
+    // The following numbers are based on the AVX implementation.
+    Cost = (sizeof(Scalar) == 8 ? 28
+                                // 4 pmul, 1 pmadd, 3 other
+                                : (3 * NumTraits<Scalar>::AddCost +
+                                   5 * NumTraits<Scalar>::MulCost)),
+#else
+    // The following numbers are based on min VSQRT throughput on Haswell.
+    Cost = (sizeof(Scalar) == 8 ? 28 : 14),
+#endif
+    PacketAccess = packet_traits<Scalar>::HasSqrt
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the reciprocal square root of a scalar
+  * \sa class CwiseUnaryOp, Cwise::rsqrt()
+  */
+template<typename Scalar> struct scalar_rsqrt_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_rsqrt_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return Scalar(1)/numext::sqrt(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::prsqrt(a); }
+};
+
+template<typename Scalar>
+struct functor_traits<scalar_rsqrt_op<Scalar> >
+{ enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasRsqrt
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the cosine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::cos()
+  */
+template<typename Scalar> struct scalar_cos_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cos_op)
+  EIGEN_DEVICE_FUNC inline Scalar operator() (const Scalar& a) const { return numext::cos(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pcos(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_cos_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasCos
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the sine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::sin()
+  */
+template<typename Scalar> struct scalar_sin_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sin_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::sin(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psin(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_sin_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasSin
+  };
+};
+
+
+/** \internal
+  * \brief Template functor to compute the tan of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::tan()
+  */
+template<typename Scalar> struct scalar_tan_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_tan_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::tan(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::ptan(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_tan_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasTan
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the arc cosine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::acos()
+  */
+template<typename Scalar> struct scalar_acos_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_acos_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::acos(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pacos(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_acos_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasACos
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the arc sine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::asin()
+  */
+template<typename Scalar> struct scalar_asin_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_asin_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::asin(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pasin(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_asin_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasASin
+  };
+};
+
+
+/** \internal
+  * \brief Template functor to compute the atan of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::atan()
+  */
+template<typename Scalar> struct scalar_atan_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_atan_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::atan(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::patan(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_atan_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasATan
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the tanh of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::tanh()
+  */
+template <typename Scalar>
+struct scalar_tanh_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_tanh_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator()(const Scalar& a) const { return numext::tanh(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& x) const { return ptanh(x); }
+};
+
+template <typename Scalar>
+struct functor_traits<scalar_tanh_op<Scalar> > {
+  enum {
+    PacketAccess = packet_traits<Scalar>::HasTanh,
+    Cost = ( (EIGEN_FAST_MATH && is_same<Scalar,float>::value)
+// The following numbers are based on the AVX implementation,
+#ifdef EIGEN_VECTORIZE_FMA
+                // Haswell can issue 2 add/mul/madd per cycle.
+                // 9 pmadd, 2 pmul, 1 div, 2 other
+                ? (2 * NumTraits<Scalar>::AddCost +
+                   6 * NumTraits<Scalar>::MulCost +
+                   scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value)
+#else
+                ? (11 * NumTraits<Scalar>::AddCost +
+                   11 * NumTraits<Scalar>::MulCost +
+                   scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value)
+#endif
+                // This number assumes a naive implementation of tanh
+                : (6 * NumTraits<Scalar>::AddCost +
+                   3 * NumTraits<Scalar>::MulCost +
+                   2 * scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value +
+                   functor_traits<scalar_exp_op<Scalar> >::Cost))
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the sinh of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::sinh()
+  */
+template<typename Scalar> struct scalar_sinh_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sinh_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::sinh(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psinh(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_sinh_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasSinh
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the cosh of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::cosh()
+  */
+template<typename Scalar> struct scalar_cosh_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cosh_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::cosh(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pcosh(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_cosh_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasCosh
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the inverse of a scalar
+  * \sa class CwiseUnaryOp, Cwise::inverse()
+  */
+template<typename Scalar>
+struct scalar_inverse_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_inverse_op)
+  EIGEN_DEVICE_FUNC inline Scalar operator() (const Scalar& a) const { return Scalar(1)/a; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC inline const Packet packetOp(const Packet& a) const
+  { return internal::pdiv(pset1<Packet>(Scalar(1)),a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_inverse_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasDiv }; };
+
+/** \internal
+  * \brief Template functor to compute the square of a scalar
+  * \sa class CwiseUnaryOp, Cwise::square()
+  */
+template<typename Scalar>
+struct scalar_square_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_square_op)
+  EIGEN_DEVICE_FUNC inline Scalar operator() (const Scalar& a) const { return a*a; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC inline const Packet packetOp(const Packet& a) const
+  { return internal::pmul(a,a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_square_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
+
+/** \internal
+  * \brief Template functor to compute the cube of a scalar
+  * \sa class CwiseUnaryOp, Cwise::cube()
+  */
+template<typename Scalar>
+struct scalar_cube_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cube_op)
+  EIGEN_DEVICE_FUNC inline Scalar operator() (const Scalar& a) const { return a*a*a; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC inline const Packet packetOp(const Packet& a) const
+  { return internal::pmul(a,pmul(a,a)); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_cube_op<Scalar> >
+{ enum { Cost = 2*NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
+
+/** \internal
+  * \brief Template functor to compute the rounded value of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::round()
+  */
+template<typename Scalar> struct scalar_round_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_round_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return numext::round(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pround(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_round_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasRound
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the floor of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::floor()
+  */
+template<typename Scalar> struct scalar_floor_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_floor_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return numext::floor(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pfloor(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_floor_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasFloor
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the ceil of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::ceil()
+  */
+template<typename Scalar> struct scalar_ceil_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_ceil_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return numext::ceil(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pceil(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_ceil_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasCeil
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute whether a scalar is NaN
+  * \sa class CwiseUnaryOp, ArrayBase::isnan()
+  */
+template<typename Scalar> struct scalar_isnan_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_isnan_op)
+  typedef bool result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return (numext::isnan)(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_isnan_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to check whether a scalar is +/-inf
+  * \sa class CwiseUnaryOp, ArrayBase::isinf()
+  */
+template<typename Scalar> struct scalar_isinf_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_isinf_op)
+  typedef bool result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return (numext::isinf)(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_isinf_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to check whether a scalar has a finite value
+  * \sa class CwiseUnaryOp, ArrayBase::isfinite()
+  */
+template<typename Scalar> struct scalar_isfinite_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_isfinite_op)
+  typedef bool result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return (numext::isfinite)(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_isfinite_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the logical not of a boolean
+  *
+  * \sa class CwiseUnaryOp, ArrayBase::operator!
+  */
+template<typename Scalar> struct scalar_boolean_not_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_not_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a) const { return !a; }
+};
+template<typename Scalar>
+struct functor_traits<scalar_boolean_not_op<Scalar> > {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the signum of a scalar
+  * \sa class CwiseUnaryOp, Cwise::sign()
+  */
+template<typename Scalar,bool iscpx=(NumTraits<Scalar>::IsComplex!=0) > struct scalar_sign_op;
+template<typename Scalar>
+struct scalar_sign_op<Scalar,false> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sign_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const
+  {
+      return Scalar( (a>Scalar(0)) - (a<Scalar(0)) );
+  }
+  //TODO
+  //template <typename Packet>
+  //EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psign(a); }
+};
+template<typename Scalar>
+struct scalar_sign_op<Scalar,true> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sign_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const
+  {
+    typedef typename NumTraits<Scalar>::Real real_type;
+    real_type aa = numext::abs(a);
+    if (aa==real_type(0))
+      return Scalar(0);
+    aa = real_type(1)/aa;
+    return Scalar(real(a)*aa, imag(a)*aa );
+  }
+  //TODO
+  //template <typename Packet>
+  //EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psign(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_sign_op<Scalar> >
+{ enum {
+    Cost = 
+        NumTraits<Scalar>::IsComplex
+        ? ( 8*NumTraits<Scalar>::MulCost  ) // roughly
+        : ( 3*NumTraits<Scalar>::AddCost),
+    PacketAccess = packet_traits<Scalar>::HasSign
+  };
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_FUNCTORS_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/GeneralBlockPanelKernel.h b/extern/Eigen3/Eigen/src/Core/products/GeneralBlockPanelKernel.h
index bcdca5b0d23..e3980f6ffd4 100644
--- a/extern/Eigen3/Eigen/src/Core/products/GeneralBlockPanelKernel.h
+++ b/extern/Eigen3/Eigen/src/Core/products/GeneralBlockPanelKernel.h
@@ -10,8 +10,9 @@
 #ifndef EIGEN_GENERAL_BLOCK_PANEL_H
 #define EIGEN_GENERAL_BLOCK_PANEL_H
 
-namespace Eigen { 
-  
+
+namespace Eigen {
+
 namespace internal {
 
 template<typename _LhsScalar, typename _RhsScalar, bool _ConjLhs=false, bool _ConjRhs=false>
@@ -24,29 +25,51 @@ inline std::ptrdiff_t manage_caching_sizes_helper(std::ptrdiff_t a, std::ptrdiff
   return a<=0 ? b : a;
 }
 
+#if EIGEN_ARCH_i386_OR_x86_64
+const std::ptrdiff_t defaultL1CacheSize = 32*1024;
+const std::ptrdiff_t defaultL2CacheSize = 256*1024;
+const std::ptrdiff_t defaultL3CacheSize = 2*1024*1024;
+#else
+const std::ptrdiff_t defaultL1CacheSize = 16*1024;
+const std::ptrdiff_t defaultL2CacheSize = 512*1024;
+const std::ptrdiff_t defaultL3CacheSize = 512*1024;
+#endif
+
 /** \internal */
-inline void manage_caching_sizes(Action action, std::ptrdiff_t* l1=0, std::ptrdiff_t* l2=0)
-{
-  static std::ptrdiff_t m_l1CacheSize = 0;
-  static std::ptrdiff_t m_l2CacheSize = 0;
-  if(m_l2CacheSize==0)
-  {
-    m_l1CacheSize = manage_caching_sizes_helper(queryL1CacheSize(),8 * 1024);
-    m_l2CacheSize = manage_caching_sizes_helper(queryTopLevelCacheSize(),1*1024*1024);
+struct CacheSizes {
+  CacheSizes(): m_l1(-1),m_l2(-1),m_l3(-1) {
+    int l1CacheSize, l2CacheSize, l3CacheSize;
+    queryCacheSizes(l1CacheSize, l2CacheSize, l3CacheSize);
+    m_l1 = manage_caching_sizes_helper(l1CacheSize, defaultL1CacheSize);
+    m_l2 = manage_caching_sizes_helper(l2CacheSize, defaultL2CacheSize);
+    m_l3 = manage_caching_sizes_helper(l3CacheSize, defaultL3CacheSize);
   }
-  
+
+  std::ptrdiff_t m_l1;
+  std::ptrdiff_t m_l2;
+  std::ptrdiff_t m_l3;
+};
+
+
+/** \internal */
+inline void manage_caching_sizes(Action action, std::ptrdiff_t* l1, std::ptrdiff_t* l2, std::ptrdiff_t* l3)
+{
+  static CacheSizes m_cacheSizes;
+
   if(action==SetAction)
   {
     // set the cpu cache size and cache all block sizes from a global cache size in byte
     eigen_internal_assert(l1!=0 && l2!=0);
-    m_l1CacheSize = *l1;
-    m_l2CacheSize = *l2;
+    m_cacheSizes.m_l1 = *l1;
+    m_cacheSizes.m_l2 = *l2;
+    m_cacheSizes.m_l3 = *l3;
   }
   else if(action==GetAction)
   {
     eigen_internal_assert(l1!=0 && l2!=0);
-    *l1 = m_l1CacheSize;
-    *l2 = m_l2CacheSize;
+    *l1 = m_cacheSizes.m_l1;
+    *l2 = m_cacheSizes.m_l2;
+    *l3 = m_cacheSizes.m_l3;
   }
   else
   {
@@ -54,6 +77,206 @@ inline void manage_caching_sizes(Action action, std::ptrdiff_t* l1=0, std::ptrdi
   }
 }
 
+/* Helper for computeProductBlockingSizes.
+ *
+ * Given a m x k times k x n matrix product of scalar types \c LhsScalar and \c RhsScalar,
+ * this function computes the blocking size parameters along the respective dimensions
+ * for matrix products and related algorithms. The blocking sizes depends on various
+ * parameters:
+ * - the L1 and L2 cache sizes,
+ * - the register level blocking sizes defined by gebp_traits,
+ * - the number of scalars that fit into a packet (when vectorization is enabled).
+ *
+ * \sa setCpuCacheSizes */
+
+template<typename LhsScalar, typename RhsScalar, int KcFactor, typename Index>
+void evaluateProductBlockingSizesHeuristic(Index& k, Index& m, Index& n, Index num_threads = 1)
+{
+  typedef gebp_traits<LhsScalar,RhsScalar> Traits;
+
+  // Explanations:
+  // Let's recall that the product algorithms form mc x kc vertical panels A' on the lhs and
+  // kc x nc blocks B' on the rhs. B' has to fit into L2/L3 cache. Moreover, A' is processed
+  // per mr x kc horizontal small panels where mr is the blocking size along the m dimension
+  // at the register level. This small horizontal panel has to stay within L1 cache.
+  std::ptrdiff_t l1, l2, l3;
+  manage_caching_sizes(GetAction, &l1, &l2, &l3);
+
+  if (num_threads > 1) {
+    typedef typename Traits::ResScalar ResScalar;
+    enum {
+      kdiv = KcFactor * (Traits::mr * sizeof(LhsScalar) + Traits::nr * sizeof(RhsScalar)),
+      ksub = Traits::mr * Traits::nr * sizeof(ResScalar),
+      kr = 8,
+      mr = Traits::mr,
+      nr = Traits::nr
+    };
+    // Increasing k gives us more time to prefetch the content of the "C"
+    // registers. However once the latency is hidden there is no point in
+    // increasing the value of k, so we'll cap it at 320 (value determined
+    // experimentally).
+    const Index k_cache = (numext::mini<Index>)((l1-ksub)/kdiv, 320);
+    if (k_cache < k) {
+      k = k_cache - (k_cache % kr);
+      eigen_internal_assert(k > 0);
+    }
+
+    const Index n_cache = (l2-l1) / (nr * sizeof(RhsScalar) * k);
+    const Index n_per_thread = numext::div_ceil(n, num_threads);
+    if (n_cache <= n_per_thread) {
+      // Don't exceed the capacity of the l2 cache.
+      eigen_internal_assert(n_cache >= static_cast<Index>(nr));
+      n = n_cache - (n_cache % nr);
+      eigen_internal_assert(n > 0);
+    } else {
+      n = (numext::mini<Index>)(n, (n_per_thread + nr - 1) - ((n_per_thread + nr - 1) % nr));
+    }
+
+    if (l3 > l2) {
+      // l3 is shared between all cores, so we'll give each thread its own chunk of l3.
+      const Index m_cache = (l3-l2) / (sizeof(LhsScalar) * k * num_threads);
+      const Index m_per_thread = numext::div_ceil(m, num_threads);
+      if(m_cache < m_per_thread && m_cache >= static_cast<Index>(mr)) {
+        m = m_cache - (m_cache % mr);
+        eigen_internal_assert(m > 0);
+      } else {
+        m = (numext::mini<Index>)(m, (m_per_thread + mr - 1) - ((m_per_thread + mr - 1) % mr));
+      }
+    }
+  }
+  else {
+    // In unit tests we do not want to use extra large matrices,
+    // so we reduce the cache size to check the blocking strategy is not flawed
+#ifdef EIGEN_DEBUG_SMALL_PRODUCT_BLOCKS
+    l1 = 9*1024;
+    l2 = 32*1024;
+    l3 = 512*1024;
+#endif
+
+    // Early return for small problems because the computation below are time consuming for small problems.
+    // Perhaps it would make more sense to consider k*n*m??
+    // Note that for very tiny problem, this function should be bypassed anyway
+    // because we use the coefficient-based implementation for them.
+    if((numext::maxi)(k,(numext::maxi)(m,n))<48)
+      return;
+
+    typedef typename Traits::ResScalar ResScalar;
+    enum {
+      k_peeling = 8,
+      k_div = KcFactor * (Traits::mr * sizeof(LhsScalar) + Traits::nr * sizeof(RhsScalar)),
+      k_sub = Traits::mr * Traits::nr * sizeof(ResScalar)
+    };
+
+    // ---- 1st level of blocking on L1, yields kc ----
+
+    // Blocking on the third dimension (i.e., k) is chosen so that an horizontal panel
+    // of size mr x kc of the lhs plus a vertical panel of kc x nr of the rhs both fits within L1 cache.
+    // We also include a register-level block of the result (mx x nr).
+    // (In an ideal world only the lhs panel would stay in L1)
+    // Moreover, kc has to be a multiple of 8 to be compatible with loop peeling, leading to a maximum blocking size of:
+    const Index max_kc = numext::maxi<Index>(((l1-k_sub)/k_div) & (~(k_peeling-1)),1);
+    const Index old_k = k;
+    if(k>max_kc)
+    {
+      // We are really blocking on the third dimension:
+      // -> reduce blocking size to make sure the last block is as large as possible
+      //    while keeping the same number of sweeps over the result.
+      k = (k%max_kc)==0 ? max_kc
+                        : max_kc - k_peeling * ((max_kc-1-(k%max_kc))/(k_peeling*(k/max_kc+1)));
+
+      eigen_internal_assert(((old_k/k) == (old_k/max_kc)) && "the number of sweeps has to remain the same");
+    }
+
+    // ---- 2nd level of blocking on max(L2,L3), yields nc ----
+
+    // TODO find a reliable way to get the actual amount of cache per core to use for 2nd level blocking, that is:
+    //      actual_l2 = max(l2, l3/nb_core_sharing_l3)
+    // The number below is quite conservative: it is better to underestimate the cache size rather than overestimating it)
+    // For instance, it corresponds to 6MB of L3 shared among 4 cores.
+    #ifdef EIGEN_DEBUG_SMALL_PRODUCT_BLOCKS
+    const Index actual_l2 = l3;
+    #else
+    const Index actual_l2 = 1572864; // == 1.5 MB
+    #endif
+
+    // Here, nc is chosen such that a block of kc x nc of the rhs fit within half of L2.
+    // The second half is implicitly reserved to access the result and lhs coefficients.
+    // When k<max_kc, then nc can arbitrarily growth. In practice, it seems to be fruitful
+    // to limit this growth: we bound nc to growth by a factor x1.5.
+    // However, if the entire lhs block fit within L1, then we are not going to block on the rows at all,
+    // and it becomes fruitful to keep the packed rhs blocks in L1 if there is enough remaining space.
+    Index max_nc;
+    const Index lhs_bytes = m * k * sizeof(LhsScalar);
+    const Index remaining_l1 = l1- k_sub - lhs_bytes;
+    if(remaining_l1 >= Index(Traits::nr*sizeof(RhsScalar))*k)
+    {
+      // L1 blocking
+      max_nc = remaining_l1 / (k*sizeof(RhsScalar));
+    }
+    else
+    {
+      // L2 blocking
+      max_nc = (3*actual_l2)/(2*2*max_kc*sizeof(RhsScalar));
+    }
+    // WARNING Below, we assume that Traits::nr is a power of two.
+    Index nc = numext::mini<Index>(actual_l2/(2*k*sizeof(RhsScalar)), max_nc) & (~(Traits::nr-1));
+    if(n>nc)
+    {
+      // We are really blocking over the columns:
+      // -> reduce blocking size to make sure the last block is as large as possible
+      //    while keeping the same number of sweeps over the packed lhs.
+      //    Here we allow one more sweep if this gives us a perfect match, thus the commented "-1"
+      n = (n%nc)==0 ? nc
+                    : (nc - Traits::nr * ((nc/*-1*/-(n%nc))/(Traits::nr*(n/nc+1))));
+    }
+    else if(old_k==k)
+    {
+      // So far, no blocking at all, i.e., kc==k, and nc==n.
+      // In this case, let's perform a blocking over the rows such that the packed lhs data is kept in cache L1/L2
+      // TODO: part of this blocking strategy is now implemented within the kernel itself, so the L1-based heuristic here should be obsolete.
+      Index problem_size = k*n*sizeof(LhsScalar);
+      Index actual_lm = actual_l2;
+      Index max_mc = m;
+      if(problem_size<=1024)
+      {
+        // problem is small enough to keep in L1
+        // Let's choose m such that lhs's block fit in 1/3 of L1
+        actual_lm = l1;
+      }
+      else if(l3!=0 && problem_size<=32768)
+      {
+        // we have both L2 and L3, and problem is small enough to be kept in L2
+        // Let's choose m such that lhs's block fit in 1/3 of L2
+        actual_lm = l2;
+        max_mc = (numext::mini<Index>)(576,max_mc);
+      }
+      Index mc = (numext::mini<Index>)(actual_lm/(3*k*sizeof(LhsScalar)), max_mc);
+      if (mc > Traits::mr) mc -= mc % Traits::mr;
+      else if (mc==0) return;
+      m = (m%mc)==0 ? mc
+                    : (mc - Traits::mr * ((mc/*-1*/-(m%mc))/(Traits::mr*(m/mc+1))));
+    }
+  }
+}
+
+template <typename Index>
+inline bool useSpecificBlockingSizes(Index& k, Index& m, Index& n)
+{
+#ifdef EIGEN_TEST_SPECIFIC_BLOCKING_SIZES
+  if (EIGEN_TEST_SPECIFIC_BLOCKING_SIZES) {
+    k = numext::mini<Index>(k, EIGEN_TEST_SPECIFIC_BLOCKING_SIZE_K);
+    m = numext::mini<Index>(m, EIGEN_TEST_SPECIFIC_BLOCKING_SIZE_M);
+    n = numext::mini<Index>(n, EIGEN_TEST_SPECIFIC_BLOCKING_SIZE_N);
+    return true;
+  }
+#else
+  EIGEN_UNUSED_VARIABLE(k)
+  EIGEN_UNUSED_VARIABLE(m)
+  EIGEN_UNUSED_VARIABLE(n)
+#endif
+  return false;
+}
+
 /** \brief Computes the blocking parameters for a m x k times k x n matrix product
   *
   * \param[in,out] k Input: the third dimension of the product. Output: the blocking size along the same dimension.
@@ -62,48 +285,30 @@ inline void manage_caching_sizes(Action action, std::ptrdiff_t* l1=0, std::ptrdi
   *
   * Given a m x k times k x n matrix product of scalar types \c LhsScalar and \c RhsScalar,
   * this function computes the blocking size parameters along the respective dimensions
-  * for matrix products and related algorithms. The blocking sizes depends on various
-  * parameters:
-  * - the L1 and L2 cache sizes,
-  * - the register level blocking sizes defined by gebp_traits,
-  * - the number of scalars that fit into a packet (when vectorization is enabled).
+  * for matrix products and related algorithms.
+  *
+  * The blocking size parameters may be evaluated:
+  *   - either by a heuristic based on cache sizes;
+  *   - or using fixed prescribed values (for testing purposes).
   *
   * \sa setCpuCacheSizes */
-template<typename LhsScalar, typename RhsScalar, int KcFactor, typename SizeType>
-void computeProductBlockingSizes(SizeType& k, SizeType& m, SizeType& n)
-{
-  EIGEN_UNUSED_VARIABLE(n);
-  // Explanations:
-  // Let's recall the product algorithms form kc x nc horizontal panels B' on the rhs and
-  // mc x kc blocks A' on the lhs. A' has to fit into L2 cache. Moreover, B' is processed
-  // per kc x nr vertical small panels where nr is the blocking size along the n dimension
-  // at the register level. For vectorization purpose, these small vertical panels are unpacked,
-  // e.g., each coefficient is replicated to fit a packet. This small vertical panel has to
-  // stay in L1 cache.
-  std::ptrdiff_t l1, l2;
 
-  typedef gebp_traits<LhsScalar,RhsScalar> Traits;
-  enum {
-    kdiv = KcFactor * 2 * Traits::nr
-         * Traits::RhsProgress * sizeof(RhsScalar),
-    mr = gebp_traits<LhsScalar,RhsScalar>::mr,
-    mr_mask = (0xffffffff/mr)*mr
-  };
-
-  manage_caching_sizes(GetAction, &l1, &l2);
-  k = std::min<SizeType>(k, l1/kdiv);
-  SizeType _m = k>0 ? l2/(4 * sizeof(LhsScalar) * k) : 0;
-  if(_m<m) m = _m & mr_mask;
+template<typename LhsScalar, typename RhsScalar, int KcFactor, typename Index>
+void computeProductBlockingSizes(Index& k, Index& m, Index& n, Index num_threads = 1)
+{
+  if (!useSpecificBlockingSizes(k, m, n)) {
+    evaluateProductBlockingSizesHeuristic<LhsScalar, RhsScalar, KcFactor, Index>(k, m, n, num_threads);
+  }
 }
 
-template<typename LhsScalar, typename RhsScalar, typename SizeType>
-inline void computeProductBlockingSizes(SizeType& k, SizeType& m, SizeType& n)
+template<typename LhsScalar, typename RhsScalar, typename Index>
+inline void computeProductBlockingSizes(Index& k, Index& m, Index& n, Index num_threads = 1)
 {
-  computeProductBlockingSizes<LhsScalar,RhsScalar,1>(k, m, n);
+  computeProductBlockingSizes<LhsScalar,RhsScalar,1,Index>(k, m, n, num_threads);
 }
 
-#ifdef EIGEN_HAS_FUSE_CJMADD
-  #define MADD(CJ,A,B,C,T)  C = CJ.pmadd(A,B,C);
+#ifdef EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+  #define CJMADD(CJ,A,B,C,T)  C = CJ.pmadd(A,B,C);
 #else
 
   // FIXME (a bit overkill maybe ?)
@@ -128,8 +333,8 @@ inline void computeProductBlockingSizes(SizeType& k, SizeType& m, SizeType& n)
     gebp_madd_selector<CJ,A,B,C,T>::run(cj,a,b,c,t);
   }
 
-  #define MADD(CJ,A,B,C,T)  gebp_madd(CJ,A,B,C,T);
-//   #define MADD(CJ,A,B,C,T)  T = B; T = CJ.pmul(A,T); C = padd(C,T);
+  #define CJMADD(CJ,A,B,C,T)  gebp_madd(CJ,A,B,C,T);
+//   #define CJMADD(CJ,A,B,C,T)  T = B; T = CJ.pmul(A,T); C = padd(C,T);
 #endif
 
 /* Vectorization logic
@@ -148,7 +353,7 @@ class gebp_traits
 public:
   typedef _LhsScalar LhsScalar;
   typedef _RhsScalar RhsScalar;
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 
   enum {
     ConjLhs = _ConjLhs,
@@ -160,16 +365,22 @@ public:
     
     NumberOfRegisters = EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS,
 
-    // register block size along the N direction (must be either 2 or 4)
-    nr = NumberOfRegisters/4,
+    // register block size along the N direction must be 1 or 4
+    nr = 4,
 
     // register block size along the M direction (currently, this one cannot be modified)
-    mr = 2 * LhsPacketSize,
+    default_mr = (EIGEN_PLAIN_ENUM_MIN(16,NumberOfRegisters)/2/nr)*LhsPacketSize,
+#if defined(EIGEN_HAS_SINGLE_INSTRUCTION_MADD) && !defined(EIGEN_VECTORIZE_ALTIVEC) && !defined(EIGEN_VECTORIZE_VSX)
+    // we assume 16 registers
+    // See bug 992, if the scalar type is not vectorizable but that EIGEN_HAS_SINGLE_INSTRUCTION_MADD is defined,
+    // then using 3*LhsPacketSize triggers non-implemented paths in syrk.
+    mr = Vectorizable ? 3*LhsPacketSize : default_mr,
+#else
+    mr = default_mr,
+#endif
     
-    WorkSpaceFactor = nr * RhsPacketSize,
-
     LhsProgress = LhsPacketSize,
-    RhsProgress = RhsPacketSize
+    RhsProgress = 1
   };
 
   typedef typename packet_traits<LhsScalar>::type  _LhsPacket;
@@ -186,36 +397,67 @@ public:
   {
     p = pset1<ResPacket>(ResScalar(0));
   }
-
-  EIGEN_STRONG_INLINE void unpackRhs(DenseIndex n, const RhsScalar* rhs, RhsScalar* b)
+  
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1, RhsPacket& b2, RhsPacket& b3)
+  {
+    pbroadcast4(b, b0, b1, b2, b3);
+  }
+  
+//   EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1)
+//   {
+//     pbroadcast2(b, b0, b1);
+//   }
+  
+  template<typename RhsPacketType>
+  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacketType& dest) const
+  {
+    dest = pset1<RhsPacketType>(*b);
+  }
+  
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, RhsPacket& dest) const
   {
-    for(DenseIndex k=0; k<n; k++)
-      pstore1<RhsPacket>(&b[k*RhsPacketSize], rhs[k]);
+    dest = ploadquad<RhsPacket>(b);
   }
 
-  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacket& dest) const
+  template<typename LhsPacketType>
+  EIGEN_STRONG_INLINE void loadLhs(const LhsScalar* a, LhsPacketType& dest) const
   {
-    dest = pload<RhsPacket>(b);
+    dest = pload<LhsPacketType>(a);
   }
 
-  EIGEN_STRONG_INLINE void loadLhs(const LhsScalar* a, LhsPacket& dest) const
+  template<typename LhsPacketType>
+  EIGEN_STRONG_INLINE void loadLhsUnaligned(const LhsScalar* a, LhsPacketType& dest) const
   {
-    dest = pload<LhsPacket>(a);
+    dest = ploadu<LhsPacketType>(a);
   }
 
-  EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, AccPacket& c, AccPacket& tmp) const
+  template<typename LhsPacketType, typename RhsPacketType, typename AccPacketType>
+  EIGEN_STRONG_INLINE void madd(const LhsPacketType& a, const RhsPacketType& b, AccPacketType& c, AccPacketType& tmp) const
   {
-    tmp = b; tmp = pmul(a,tmp); c = padd(c,tmp);
+    conj_helper<LhsPacketType,RhsPacketType,ConjLhs,ConjRhs> cj;
+    // It would be a lot cleaner to call pmadd all the time. Unfortunately if we
+    // let gcc allocate the register in which to store the result of the pmul
+    // (in the case where there is no FMA) gcc fails to figure out how to avoid
+    // spilling register.
+#ifdef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+    EIGEN_UNUSED_VARIABLE(tmp);
+    c = cj.pmadd(a,b,c);
+#else
+    tmp = b; tmp = cj.pmul(a,tmp); c = padd(c,tmp);
+#endif
   }
 
   EIGEN_STRONG_INLINE void acc(const AccPacket& c, const ResPacket& alpha, ResPacket& r) const
   {
     r = pmadd(c,alpha,r);
   }
+  
+  template<typename ResPacketHalf>
+  EIGEN_STRONG_INLINE void acc(const ResPacketHalf& c, const ResPacketHalf& alpha, ResPacketHalf& r) const
+  {
+    r = pmadd(c,alpha,r);
+  }
 
-protected:
-//   conj_helper<LhsScalar,RhsScalar,ConjLhs,ConjRhs> cj;
-//   conj_helper<LhsPacket,RhsPacket,ConjLhs,ConjRhs> pcj;
 };
 
 template<typename RealScalar, bool _ConjLhs>
@@ -224,7 +466,7 @@ class gebp_traits<std::complex<RealScalar>, RealScalar, _ConjLhs, false>
 public:
   typedef std::complex<RealScalar> LhsScalar;
   typedef RealScalar RhsScalar;
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 
   enum {
     ConjLhs = _ConjLhs,
@@ -235,12 +477,16 @@ public:
     ResPacketSize = Vectorizable ? packet_traits<ResScalar>::size : 1,
     
     NumberOfRegisters = EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS,
-    nr = NumberOfRegisters/4,
-    mr = 2 * LhsPacketSize,
-    WorkSpaceFactor = nr*RhsPacketSize,
+    nr = 4,
+#if defined(EIGEN_HAS_SINGLE_INSTRUCTION_MADD) && !defined(EIGEN_VECTORIZE_ALTIVEC) && !defined(EIGEN_VECTORIZE_VSX)
+    // we assume 16 registers
+    mr = 3*LhsPacketSize,
+#else
+    mr = (EIGEN_PLAIN_ENUM_MIN(16,NumberOfRegisters)/2/nr)*LhsPacketSize,
+#endif
 
     LhsProgress = LhsPacketSize,
-    RhsProgress = RhsPacketSize
+    RhsProgress = 1
   };
 
   typedef typename packet_traits<LhsScalar>::type  _LhsPacket;
@@ -258,15 +504,14 @@ public:
     p = pset1<ResPacket>(ResScalar(0));
   }
 
-  EIGEN_STRONG_INLINE void unpackRhs(DenseIndex n, const RhsScalar* rhs, RhsScalar* b)
+  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacket& dest) const
   {
-    for(DenseIndex k=0; k<n; k++)
-      pstore1<RhsPacket>(&b[k*RhsPacketSize], rhs[k]);
+    dest = pset1<RhsPacket>(*b);
   }
-
-  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacket& dest) const
+  
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, RhsPacket& dest) const
   {
-    dest = pload<RhsPacket>(b);
+    dest = pset1<RhsPacket>(*b);
   }
 
   EIGEN_STRONG_INLINE void loadLhs(const LhsScalar* a, LhsPacket& dest) const
@@ -274,6 +519,21 @@ public:
     dest = pload<LhsPacket>(a);
   }
 
+  EIGEN_STRONG_INLINE void loadLhsUnaligned(const LhsScalar* a, LhsPacket& dest) const
+  {
+    dest = ploadu<LhsPacket>(a);
+  }
+
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1, RhsPacket& b2, RhsPacket& b3)
+  {
+    pbroadcast4(b, b0, b1, b2, b3);
+  }
+  
+//   EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1)
+//   {
+//     pbroadcast2(b, b0, b1);
+//   }
+
   EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, AccPacket& c, RhsPacket& tmp) const
   {
     madd_impl(a, b, c, tmp, typename conditional<Vectorizable,true_type,false_type>::type());
@@ -281,7 +541,12 @@ public:
 
   EIGEN_STRONG_INLINE void madd_impl(const LhsPacket& a, const RhsPacket& b, AccPacket& c, RhsPacket& tmp, const true_type&) const
   {
+#ifdef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+    EIGEN_UNUSED_VARIABLE(tmp);
+    c.v = pmadd(a.v,b,c.v);
+#else
     tmp = b; tmp = pmul(a.v,tmp); c.v = padd(c.v,tmp);
+#endif
   }
 
   EIGEN_STRONG_INLINE void madd_impl(const LhsScalar& a, const RhsScalar& b, ResScalar& c, RhsScalar& /*tmp*/, const false_type&) const
@@ -298,6 +563,38 @@ protected:
   conj_helper<ResPacket,ResPacket,ConjLhs,false> cj;
 };
 
+template<typename Packet>
+struct DoublePacket
+{
+  Packet first;
+  Packet second;
+};
+
+template<typename Packet>
+DoublePacket<Packet> padd(const DoublePacket<Packet> &a, const DoublePacket<Packet> &b)
+{
+  DoublePacket<Packet> res;
+  res.first  = padd(a.first, b.first);
+  res.second = padd(a.second,b.second);
+  return res;
+}
+
+template<typename Packet>
+const DoublePacket<Packet>& predux_downto4(const DoublePacket<Packet> &a)
+{
+  return a;
+}
+
+template<typename Packet> struct unpacket_traits<DoublePacket<Packet> > { typedef DoublePacket<Packet> half; };
+// template<typename Packet>
+// DoublePacket<Packet> pmadd(const DoublePacket<Packet> &a, const DoublePacket<Packet> &b)
+// {
+//   DoublePacket<Packet> res;
+//   res.first  = padd(a.first, b.first);
+//   res.second = padd(a.second,b.second);
+//   return res;
+// }
+
 template<typename RealScalar, bool _ConjLhs, bool _ConjRhs>
 class gebp_traits<std::complex<RealScalar>, std::complex<RealScalar>, _ConjLhs, _ConjRhs >
 {
@@ -314,60 +611,80 @@ public:
                 && packet_traits<Scalar>::Vectorizable,
     RealPacketSize  = Vectorizable ? packet_traits<RealScalar>::size : 1,
     ResPacketSize   = Vectorizable ? packet_traits<ResScalar>::size : 1,
-    
-    nr = 2,
-    mr = 2 * ResPacketSize,
-    WorkSpaceFactor = Vectorizable ? 2*nr*RealPacketSize : nr,
+    LhsPacketSize = Vectorizable ? packet_traits<LhsScalar>::size : 1,
+    RhsPacketSize = Vectorizable ? packet_traits<RhsScalar>::size : 1,
+
+    // FIXME: should depend on NumberOfRegisters
+    nr = 4,
+    mr = ResPacketSize,
 
     LhsProgress = ResPacketSize,
-    RhsProgress = Vectorizable ? 2*ResPacketSize : 1
+    RhsProgress = 1
   };
   
   typedef typename packet_traits<RealScalar>::type RealPacket;
   typedef typename packet_traits<Scalar>::type     ScalarPacket;
-  struct DoublePacket
-  {
-    RealPacket first;
-    RealPacket second;
-  };
+  typedef DoublePacket<RealPacket> DoublePacketType;
 
   typedef typename conditional<Vectorizable,RealPacket,  Scalar>::type LhsPacket;
-  typedef typename conditional<Vectorizable,DoublePacket,Scalar>::type RhsPacket;
+  typedef typename conditional<Vectorizable,DoublePacketType,Scalar>::type RhsPacket;
   typedef typename conditional<Vectorizable,ScalarPacket,Scalar>::type ResPacket;
-  typedef typename conditional<Vectorizable,DoublePacket,Scalar>::type AccPacket;
+  typedef typename conditional<Vectorizable,DoublePacketType,Scalar>::type AccPacket;
   
   EIGEN_STRONG_INLINE void initAcc(Scalar& p) { p = Scalar(0); }
 
-  EIGEN_STRONG_INLINE void initAcc(DoublePacket& p)
+  EIGEN_STRONG_INLINE void initAcc(DoublePacketType& p)
   {
     p.first   = pset1<RealPacket>(RealScalar(0));
     p.second  = pset1<RealPacket>(RealScalar(0));
   }
 
-  /* Unpack the rhs coeff such that each complex coefficient is spread into
-   * two packects containing respectively the real and imaginary coefficient
-   * duplicated as many time as needed: (x+iy) => [x, ..., x] [y, ..., y]
-   */
-  EIGEN_STRONG_INLINE void unpackRhs(DenseIndex n, const Scalar* rhs, Scalar* b)
+  // Scalar path
+  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, ResPacket& dest) const
   {
-    for(DenseIndex k=0; k<n; k++)
-    {
-      if(Vectorizable)
-      {
-        pstore1<RealPacket>((RealScalar*)&b[k*ResPacketSize*2+0],             real(rhs[k]));
-        pstore1<RealPacket>((RealScalar*)&b[k*ResPacketSize*2+ResPacketSize], imag(rhs[k]));
-      }
-      else
-        b[k] = rhs[k];
-    }
+    dest = pset1<ResPacket>(*b);
   }
 
-  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, ResPacket& dest) const { dest = *b; }
-
-  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, DoublePacket& dest) const
+  // Vectorized path
+  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, DoublePacketType& dest) const
+  {
+    dest.first  = pset1<RealPacket>(real(*b));
+    dest.second = pset1<RealPacket>(imag(*b));
+  }
+  
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, ResPacket& dest) const
+  {
+    loadRhs(b,dest);
+  }
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, DoublePacketType& dest) const
+  {
+    eigen_internal_assert(unpacket_traits<ScalarPacket>::size<=4);
+    loadRhs(b,dest);
+  }
+  
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1, RhsPacket& b2, RhsPacket& b3)
   {
-    dest.first  = pload<RealPacket>((const RealScalar*)b);
-    dest.second = pload<RealPacket>((const RealScalar*)(b+ResPacketSize));
+    // FIXME not sure that's the best way to implement it!
+    loadRhs(b+0, b0);
+    loadRhs(b+1, b1);
+    loadRhs(b+2, b2);
+    loadRhs(b+3, b3);
+  }
+  
+  // Vectorized path
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, DoublePacketType& b0, DoublePacketType& b1)
+  {
+    // FIXME not sure that's the best way to implement it!
+    loadRhs(b+0, b0);
+    loadRhs(b+1, b1);
+  }
+  
+  // Scalar path
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsScalar& b0, RhsScalar& b1)
+  {
+    // FIXME not sure that's the best way to implement it!
+    loadRhs(b+0, b0);
+    loadRhs(b+1, b1);
   }
 
   // nothing special here
@@ -376,7 +693,12 @@ public:
     dest = pload<LhsPacket>((const typename unpacket_traits<LhsPacket>::type*)(a));
   }
 
-  EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, DoublePacket& c, RhsPacket& /*tmp*/) const
+  EIGEN_STRONG_INLINE void loadLhsUnaligned(const LhsScalar* a, LhsPacket& dest) const
+  {
+    dest = ploadu<LhsPacket>((const typename unpacket_traits<LhsPacket>::type*)(a));
+  }
+
+  EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, DoublePacketType& c, RhsPacket& /*tmp*/) const
   {
     c.first   = padd(pmul(a,b.first), c.first);
     c.second  = padd(pmul(a,b.second),c.second);
@@ -389,7 +711,7 @@ public:
   
   EIGEN_STRONG_INLINE void acc(const Scalar& c, const Scalar& alpha, Scalar& r) const { r += alpha * c; }
   
-  EIGEN_STRONG_INLINE void acc(const DoublePacket& c, const ResPacket& alpha, ResPacket& r) const
+  EIGEN_STRONG_INLINE void acc(const DoublePacketType& c, const ResPacket& alpha, ResPacket& r) const
   {
     // assemble c
     ResPacket tmp;
@@ -440,12 +762,12 @@ public:
     ResPacketSize = Vectorizable ? packet_traits<ResScalar>::size : 1,
     
     NumberOfRegisters = EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS,
+    // FIXME: should depend on NumberOfRegisters
     nr = 4,
-    mr = 2*ResPacketSize,
-    WorkSpaceFactor = nr*RhsPacketSize,
+    mr = (EIGEN_PLAIN_ENUM_MIN(16,NumberOfRegisters)/2/nr)*ResPacketSize,
 
     LhsProgress = ResPacketSize,
-    RhsProgress = ResPacketSize
+    RhsProgress = 1
   };
 
   typedef typename packet_traits<LhsScalar>::type  _LhsPacket;
@@ -463,21 +785,38 @@ public:
     p = pset1<ResPacket>(ResScalar(0));
   }
 
-  EIGEN_STRONG_INLINE void unpackRhs(DenseIndex n, const RhsScalar* rhs, RhsScalar* b)
+  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacket& dest) const
   {
-    for(DenseIndex k=0; k<n; k++)
-      pstore1<RhsPacket>(&b[k*RhsPacketSize], rhs[k]);
+    dest = pset1<RhsPacket>(*b);
   }
-
-  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacket& dest) const
+  
+  void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1, RhsPacket& b2, RhsPacket& b3)
   {
-    dest = pload<RhsPacket>(b);
+    pbroadcast4(b, b0, b1, b2, b3);
   }
+  
+//   EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1)
+//   {
+//     // FIXME not sure that's the best way to implement it!
+//     b0 = pload1<RhsPacket>(b+0);
+//     b1 = pload1<RhsPacket>(b+1);
+//   }
 
   EIGEN_STRONG_INLINE void loadLhs(const LhsScalar* a, LhsPacket& dest) const
   {
     dest = ploaddup<LhsPacket>(a);
   }
+  
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, RhsPacket& dest) const
+  {
+    eigen_internal_assert(unpacket_traits<RhsPacket>::size<=4);
+    loadRhs(b,dest);
+  }
+
+  EIGEN_STRONG_INLINE void loadLhsUnaligned(const LhsScalar* a, LhsPacket& dest) const
+  {
+    dest = ploaddup<LhsPacket>(a);
+  }
 
   EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, AccPacket& c, RhsPacket& tmp) const
   {
@@ -486,7 +825,13 @@ public:
 
   EIGEN_STRONG_INLINE void madd_impl(const LhsPacket& a, const RhsPacket& b, AccPacket& c, RhsPacket& tmp, const true_type&) const
   {
+#ifdef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+    EIGEN_UNUSED_VARIABLE(tmp);
+    c.v = pmadd(a,b.v,c.v);
+#else
     tmp = b; tmp.v = pmul(a,tmp.v); c = padd(c,tmp);
+#endif
+    
   }
 
   EIGEN_STRONG_INLINE void madd_impl(const LhsScalar& a, const RhsScalar& b, ResScalar& c, RhsScalar& /*tmp*/, const false_type&) const
@@ -510,7 +855,7 @@ protected:
  *  |real |cplx | no vectorization yet, would require to pack A with duplication
  *  |cplx |real | easy vectorization
  */
-template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
+template<typename LhsScalar, typename RhsScalar, typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
 struct gebp_kernel
 {
   typedef gebp_traits<LhsScalar,RhsScalar,ConjugateLhs,ConjugateRhs> Traits;
@@ -520,6 +865,15 @@ struct gebp_kernel
   typedef typename Traits::ResPacket ResPacket;
   typedef typename Traits::AccPacket AccPacket;
 
+  typedef gebp_traits<RhsScalar,LhsScalar,ConjugateRhs,ConjugateLhs> SwappedTraits;
+  typedef typename SwappedTraits::ResScalar SResScalar;
+  typedef typename SwappedTraits::LhsPacket SLhsPacket;
+  typedef typename SwappedTraits::RhsPacket SRhsPacket;
+  typedef typename SwappedTraits::ResPacket SResPacket;
+  typedef typename SwappedTraits::AccPacket SAccPacket;
+
+  typedef typename DataMapper::LinearMapper LinearMapper;
+
   enum {
     Vectorizable  = Traits::Vectorizable,
     LhsProgress   = Traits::LhsProgress,
@@ -528,571 +882,795 @@ struct gebp_kernel
   };
 
   EIGEN_DONT_INLINE
-  void operator()(ResScalar* res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index rows, Index depth, Index cols, ResScalar alpha,
-                  Index strideA=-1, Index strideB=-1, Index offsetA=0, Index offsetB=0, RhsScalar* unpackedB=0);
+  void operator()(const DataMapper& res, const LhsScalar* blockA, const RhsScalar* blockB,
+                  Index rows, Index depth, Index cols, ResScalar alpha,
+                  Index strideA=-1, Index strideB=-1, Index offsetA=0, Index offsetB=0);
 };
 
-template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
+template<typename LhsScalar, typename RhsScalar, typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
 EIGEN_DONT_INLINE
-void gebp_kernel<LhsScalar,RhsScalar,Index,mr,nr,ConjugateLhs,ConjugateRhs>
-  ::operator()(ResScalar* res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index rows, Index depth, Index cols, ResScalar alpha,
-               Index strideA, Index strideB, Index offsetA, Index offsetB, RhsScalar* unpackedB)
+void gebp_kernel<LhsScalar,RhsScalar,Index,DataMapper,mr,nr,ConjugateLhs,ConjugateRhs>
+  ::operator()(const DataMapper& res, const LhsScalar* blockA, const RhsScalar* blockB,
+               Index rows, Index depth, Index cols, ResScalar alpha,
+               Index strideA, Index strideB, Index offsetA, Index offsetB)
   {
     Traits traits;
+    SwappedTraits straits;
     
     if(strideA==-1) strideA = depth;
     if(strideB==-1) strideB = depth;
     conj_helper<LhsScalar,RhsScalar,ConjugateLhs,ConjugateRhs> cj;
-//     conj_helper<LhsPacket,RhsPacket,ConjugateLhs,ConjugateRhs> pcj;
-    Index packet_cols = (cols/nr) * nr;
-    const Index peeled_mc = (rows/mr)*mr;
-    // FIXME:
-    const Index peeled_mc2 = peeled_mc + (rows-peeled_mc >= LhsProgress ? LhsProgress : 0);
-    const Index peeled_kc = (depth/4)*4;
-
-    if(unpackedB==0)
-      unpackedB = const_cast<RhsScalar*>(blockB - strideB * nr * RhsProgress);
-
-    // loops on each micro vertical panel of rhs (depth x nr)
-    for(Index j2=0; j2<packet_cols; j2+=nr)
+    Index packet_cols4 = nr>=4 ? (cols/4) * 4 : 0;
+    const Index peeled_mc3 = mr>=3*Traits::LhsProgress ? (rows/(3*LhsProgress))*(3*LhsProgress) : 0;
+    const Index peeled_mc2 = mr>=2*Traits::LhsProgress ? peeled_mc3+((rows-peeled_mc3)/(2*LhsProgress))*(2*LhsProgress) : 0;
+    const Index peeled_mc1 = mr>=1*Traits::LhsProgress ? (rows/(1*LhsProgress))*(1*LhsProgress) : 0;
+    enum { pk = 8 }; // NOTE Such a large peeling factor is important for large matrices (~ +5% when >1000 on Haswell)
+    const Index peeled_kc  = depth & ~(pk-1);
+    const Index prefetch_res_offset = 32/sizeof(ResScalar);    
+//     const Index depth2     = depth & ~1;
+
+    //---------- Process 3 * LhsProgress rows at once ----------
+    // This corresponds to 3*LhsProgress x nr register blocks.
+    // Usually, make sense only with FMA
+    if(mr>=3*Traits::LhsProgress)
     {
-      traits.unpackRhs(depth*nr,&blockB[j2*strideB+offsetB*nr],unpackedB); 
-
-      // loops on each largest micro horizontal panel of lhs (mr x depth)
-      // => we select a mr x nr micro block of res which is entirely
-      //    stored into mr/packet_size x nr registers.
-      for(Index i=0; i<peeled_mc; i+=mr)
+      // Here, the general idea is to loop on each largest micro horizontal panel of the lhs (3*Traits::LhsProgress x depth)
+      // and on each largest micro vertical panel of the rhs (depth * nr).
+      // Blocking sizes, i.e., 'depth' has been computed so that the micro horizontal panel of the lhs fit in L1.
+      // However, if depth is too small, we can extend the number of rows of these horizontal panels.
+      // This actual number of rows is computed as follow:
+      const Index l1 = defaultL1CacheSize; // in Bytes, TODO, l1 should be passed to this function.
+      // The max(1, ...) here is needed because we may be using blocking params larger than what our known l1 cache size
+      // suggests we should be using: either because our known l1 cache size is inaccurate (e.g. on Android, we can only guess),
+      // or because we are testing specific blocking sizes.
+      const Index actual_panel_rows = (3*LhsProgress) * std::max<Index>(1,( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 3*LhsProgress) ));
+      for(Index i1=0; i1<peeled_mc3; i1+=actual_panel_rows)
       {
-        const LhsScalar* blA = &blockA[i*strideA+offsetA*mr];
-        prefetch(&blA[0]);
-
-        // gets res block as register
-        AccPacket C0, C1, C2, C3, C4, C5, C6, C7;
-                  traits.initAcc(C0);
-                  traits.initAcc(C1);
-        if(nr==4) traits.initAcc(C2);
-        if(nr==4) traits.initAcc(C3);
-                  traits.initAcc(C4);
-                  traits.initAcc(C5);
-        if(nr==4) traits.initAcc(C6);
-        if(nr==4) traits.initAcc(C7);
-
-        ResScalar* r0 = &res[(j2+0)*resStride + i];
-        ResScalar* r1 = r0 + resStride;
-        ResScalar* r2 = r1 + resStride;
-        ResScalar* r3 = r2 + resStride;
-
-        prefetch(r0+16);
-        prefetch(r1+16);
-        prefetch(r2+16);
-        prefetch(r3+16);
-
-        // performs "inner" product
-        // TODO let's check wether the folowing peeled loop could not be
-        //      optimized via optimal prefetching from one loop to the other
-        const RhsScalar* blB = unpackedB;
-        for(Index k=0; k<peeled_kc; k+=4)
+        const Index actual_panel_end = (std::min)(i1+actual_panel_rows, peeled_mc3);
+        for(Index j2=0; j2<packet_cols4; j2+=nr)
         {
-          if(nr==2)
-          {
-            LhsPacket A0, A1;
-            RhsPacket B_0;
-            RhsPacket T0;
-            
-EIGEN_ASM_COMMENT("mybegin2");
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadLhs(&blA[1*LhsProgress], A1);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
-
-            traits.loadLhs(&blA[2*LhsProgress], A0);
-            traits.loadLhs(&blA[3*LhsProgress], A1);
-            traits.loadRhs(&blB[2*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[3*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
-
-            traits.loadLhs(&blA[4*LhsProgress], A0);
-            traits.loadLhs(&blA[5*LhsProgress], A1);
-            traits.loadRhs(&blB[4*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[5*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
-
-            traits.loadLhs(&blA[6*LhsProgress], A0);
-            traits.loadLhs(&blA[7*LhsProgress], A1);
-            traits.loadRhs(&blB[6*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[7*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
-EIGEN_ASM_COMMENT("myend");
-          }
-          else
+          for(Index i=i1; i<actual_panel_end; i+=3*LhsProgress)
           {
-EIGEN_ASM_COMMENT("mybegin4");
-            LhsPacket A0, A1;
-            RhsPacket B_0, B1, B2, B3;
-            RhsPacket T0;
-            
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadLhs(&blA[1*LhsProgress], A1);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-
-            traits.madd(A0,B_0,C0,T0);
-            traits.loadRhs(&blB[2*RhsProgress], B2);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[3*RhsProgress], B3);
-            traits.loadRhs(&blB[4*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.loadRhs(&blB[5*RhsProgress], B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.loadRhs(&blB[6*RhsProgress], B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.loadLhs(&blA[2*LhsProgress], A0);
-            traits.madd(A1,B3,C7,B3);
-            traits.loadLhs(&blA[3*LhsProgress], A1);
-            traits.loadRhs(&blB[7*RhsProgress], B3);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[8*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.loadRhs(&blB[9*RhsProgress], B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.loadRhs(&blB[10*RhsProgress], B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.loadLhs(&blA[4*LhsProgress], A0);
-            traits.madd(A1,B3,C7,B3);
-            traits.loadLhs(&blA[5*LhsProgress], A1);
-            traits.loadRhs(&blB[11*RhsProgress], B3);
-
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[12*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.loadRhs(&blB[13*RhsProgress], B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.loadRhs(&blB[14*RhsProgress], B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.loadLhs(&blA[6*LhsProgress], A0);
-            traits.madd(A1,B3,C7,B3);
-            traits.loadLhs(&blA[7*LhsProgress], A1);
-            traits.loadRhs(&blB[15*RhsProgress], B3);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.madd(A1,B3,C7,B3);
-          }
+          
+          // We selected a 3*Traits::LhsProgress x nr micro block of res which is entirely
+          // stored into 3 x nr registers.
+          
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(3*LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0, C1, C2,  C3,
+                    C4, C5, C6,  C7,
+                    C8, C9, C10, C11;
+          traits.initAcc(C0);  traits.initAcc(C1);  traits.initAcc(C2);  traits.initAcc(C3);
+          traits.initAcc(C4);  traits.initAcc(C5);  traits.initAcc(C6);  traits.initAcc(C7);
+          traits.initAcc(C8);  traits.initAcc(C9);  traits.initAcc(C10); traits.initAcc(C11);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2 + 0);
+          LinearMapper r1 = res.getLinearMapper(i, j2 + 1);
+          LinearMapper r2 = res.getLinearMapper(i, j2 + 2);
+          LinearMapper r3 = res.getLinearMapper(i, j2 + 3);
+
+          r0.prefetch(0);
+          r1.prefetch(0);
+          r2.prefetch(0);
+          r3.prefetch(0);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
+          prefetch(&blB[0]);
+          LhsPacket A0, A1;
 
-          blB += 4*nr*RhsProgress;
-          blA += 4*mr;
-        }
-        // process remaining peeled loop
-        for(Index k=peeled_kc; k<depth; k++)
-        {
-          if(nr==2)
+          for(Index k=0; k<peeled_kc; k+=pk)
           {
-            LhsPacket A0, A1;
-            RhsPacket B_0;
-            RhsPacket T0;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadLhs(&blA[1*LhsProgress], A1);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 3pX4");
+            RhsPacket B_0, T0;
+            LhsPacket A2;
+
+#define EIGEN_GEBP_ONESTEP(K) \
+            do { \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 3pX4"); \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              internal::prefetch(blA+(3*K+16)*LhsProgress); \
+              if (EIGEN_ARCH_ARM) { internal::prefetch(blB+(4*K+16)*RhsProgress); } /* Bug 953 */ \
+              traits.loadLhs(&blA[(0+3*K)*LhsProgress], A0);  \
+              traits.loadLhs(&blA[(1+3*K)*LhsProgress], A1);  \
+              traits.loadLhs(&blA[(2+3*K)*LhsProgress], A2);  \
+              traits.loadRhs(blB + (0+4*K)*Traits::RhsProgress, B_0); \
+              traits.madd(A0, B_0, C0, T0); \
+              traits.madd(A1, B_0, C4, T0); \
+              traits.madd(A2, B_0, C8, B_0); \
+              traits.loadRhs(blB + (1+4*K)*Traits::RhsProgress, B_0); \
+              traits.madd(A0, B_0, C1, T0); \
+              traits.madd(A1, B_0, C5, T0); \
+              traits.madd(A2, B_0, C9, B_0); \
+              traits.loadRhs(blB + (2+4*K)*Traits::RhsProgress, B_0); \
+              traits.madd(A0, B_0, C2,  T0); \
+              traits.madd(A1, B_0, C6,  T0); \
+              traits.madd(A2, B_0, C10, B_0); \
+              traits.loadRhs(blB + (3+4*K)*Traits::RhsProgress, B_0); \
+              traits.madd(A0, B_0, C3 , T0); \
+              traits.madd(A1, B_0, C7,  T0); \
+              traits.madd(A2, B_0, C11, B_0); \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 3pX4"); \
+            } while(false)
+
+            internal::prefetch(blB);
+            EIGEN_GEBP_ONESTEP(0);
+            EIGEN_GEBP_ONESTEP(1);
+            EIGEN_GEBP_ONESTEP(2);
+            EIGEN_GEBP_ONESTEP(3);
+            EIGEN_GEBP_ONESTEP(4);
+            EIGEN_GEBP_ONESTEP(5);
+            EIGEN_GEBP_ONESTEP(6);
+            EIGEN_GEBP_ONESTEP(7);
+
+            blB += pk*4*RhsProgress;
+            blA += pk*3*Traits::LhsProgress;
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 3pX4");
           }
-          else
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
           {
-            LhsPacket A0, A1;
-            RhsPacket B_0, B1, B2, B3;
-            RhsPacket T0;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadLhs(&blA[1*LhsProgress], A1);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-
-            traits.madd(A0,B_0,C0,T0);
-            traits.loadRhs(&blB[2*RhsProgress], B2);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[3*RhsProgress], B3);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.madd(A1,B3,C7,B3);
+            RhsPacket B_0, T0;
+            LhsPacket A2;
+            EIGEN_GEBP_ONESTEP(0);
+            blB += 4*RhsProgress;
+            blA += 3*Traits::LhsProgress;
           }
 
-          blB += nr*RhsProgress;
-          blA += mr;
-        }
+#undef EIGEN_GEBP_ONESTEP
 
-        if(nr==4)
-        {
-          ResPacket R0, R1, R2, R3, R4, R5, R6;
+          ResPacket R0, R1, R2;
           ResPacket alphav = pset1<ResPacket>(alpha);
 
-          R0 = ploadu<ResPacket>(r0);
-          R1 = ploadu<ResPacket>(r1);
-          R2 = ploadu<ResPacket>(r2);
-          R3 = ploadu<ResPacket>(r3);
-          R4 = ploadu<ResPacket>(r0 + ResPacketSize);
-          R5 = ploadu<ResPacket>(r1 + ResPacketSize);
-          R6 = ploadu<ResPacket>(r2 + ResPacketSize);
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r0.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r0.loadPacket(2 * Traits::ResPacketSize);
           traits.acc(C0, alphav, R0);
-          pstoreu(r0, R0);
-          R0 = ploadu<ResPacket>(r3 + ResPacketSize);
-
-          traits.acc(C1, alphav, R1);
-          traits.acc(C2, alphav, R2);
-          traits.acc(C3, alphav, R3);
-          traits.acc(C4, alphav, R4);
-          traits.acc(C5, alphav, R5);
-          traits.acc(C6, alphav, R6);
-          traits.acc(C7, alphav, R0);
-          
-          pstoreu(r1, R1);
-          pstoreu(r2, R2);
-          pstoreu(r3, R3);
-          pstoreu(r0 + ResPacketSize, R4);
-          pstoreu(r1 + ResPacketSize, R5);
-          pstoreu(r2 + ResPacketSize, R6);
-          pstoreu(r3 + ResPacketSize, R0);
+          traits.acc(C4, alphav, R1);
+          traits.acc(C8, alphav, R2);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r0.storePacket(1 * Traits::ResPacketSize, R1);
+          r0.storePacket(2 * Traits::ResPacketSize, R2);
+
+          R0 = r1.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r1.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r1.loadPacket(2 * Traits::ResPacketSize);
+          traits.acc(C1, alphav, R0);
+          traits.acc(C5, alphav, R1);
+          traits.acc(C9, alphav, R2);
+          r1.storePacket(0 * Traits::ResPacketSize, R0);
+          r1.storePacket(1 * Traits::ResPacketSize, R1);
+          r1.storePacket(2 * Traits::ResPacketSize, R2);
+
+          R0 = r2.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r2.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r2.loadPacket(2 * Traits::ResPacketSize);
+          traits.acc(C2, alphav, R0);
+          traits.acc(C6, alphav, R1);
+          traits.acc(C10, alphav, R2);
+          r2.storePacket(0 * Traits::ResPacketSize, R0);
+          r2.storePacket(1 * Traits::ResPacketSize, R1);
+          r2.storePacket(2 * Traits::ResPacketSize, R2);
+
+          R0 = r3.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r3.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r3.loadPacket(2 * Traits::ResPacketSize);
+          traits.acc(C3, alphav, R0);
+          traits.acc(C7, alphav, R1);
+          traits.acc(C11, alphav, R2);
+          r3.storePacket(0 * Traits::ResPacketSize, R0);
+          r3.storePacket(1 * Traits::ResPacketSize, R1);
+          r3.storePacket(2 * Traits::ResPacketSize, R2);          
+          }
         }
-        else
+
+        // Deal with remaining columns of the rhs
+        for(Index j2=packet_cols4; j2<cols; j2++)
         {
-          ResPacket R0, R1, R4;
+          for(Index i=i1; i<actual_panel_end; i+=3*LhsProgress)
+          {
+          // One column at a time
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(3*Traits::LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0, C4, C8;
+          traits.initAcc(C0);
+          traits.initAcc(C4);
+          traits.initAcc(C8);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2);
+          r0.prefetch(0);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB];
+          LhsPacket A0, A1, A2;
+          
+          for(Index k=0; k<peeled_kc; k+=pk)
+          {
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 3pX1");
+            RhsPacket B_0;
+#define EIGEN_GEBGP_ONESTEP(K) \
+            do { \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 3pX1"); \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              traits.loadLhs(&blA[(0+3*K)*LhsProgress], A0);  \
+              traits.loadLhs(&blA[(1+3*K)*LhsProgress], A1);  \
+              traits.loadLhs(&blA[(2+3*K)*LhsProgress], A2);  \
+              traits.loadRhs(&blB[(0+K)*RhsProgress], B_0);   \
+              traits.madd(A0, B_0, C0, B_0); \
+              traits.madd(A1, B_0, C4, B_0); \
+              traits.madd(A2, B_0, C8, B_0); \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 3pX1"); \
+            } while(false)
+        
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
+
+            blB += pk*RhsProgress;
+            blA += pk*3*Traits::LhsProgress;
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 3pX1");
+          }
+
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
+          {
+            RhsPacket B_0;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += RhsProgress;
+            blA += 3*Traits::LhsProgress;
+          }
+#undef EIGEN_GEBGP_ONESTEP
+          ResPacket R0, R1, R2;
           ResPacket alphav = pset1<ResPacket>(alpha);
 
-          R0 = ploadu<ResPacket>(r0);
-          R1 = ploadu<ResPacket>(r1);
-          R4 = ploadu<ResPacket>(r0 + ResPacketSize);
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r0.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r0.loadPacket(2 * Traits::ResPacketSize);
           traits.acc(C0, alphav, R0);
-          pstoreu(r0, R0);
-          R0 = ploadu<ResPacket>(r1 + ResPacketSize);
-          traits.acc(C1, alphav, R1);
-          traits.acc(C4, alphav, R4);
-          traits.acc(C5, alphav, R0);
-          pstoreu(r1, R1);
-          pstoreu(r0 + ResPacketSize, R4);
-          pstoreu(r1 + ResPacketSize, R0);
+          traits.acc(C4, alphav, R1);
+          traits.acc(C8, alphav, R2);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r0.storePacket(1 * Traits::ResPacketSize, R1);
+          r0.storePacket(2 * Traits::ResPacketSize, R2);          
+          }
         }
-        
       }
-      
-      if(rows-peeled_mc>=LhsProgress)
+    }
+
+    //---------- Process 2 * LhsProgress rows at once ----------
+    if(mr>=2*Traits::LhsProgress)
+    {
+      const Index l1 = defaultL1CacheSize; // in Bytes, TODO, l1 should be passed to this function.
+      // The max(1, ...) here is needed because we may be using blocking params larger than what our known l1 cache size
+      // suggests we should be using: either because our known l1 cache size is inaccurate (e.g. on Android, we can only guess),
+      // or because we are testing specific blocking sizes.
+      Index actual_panel_rows = (2*LhsProgress) * std::max<Index>(1,( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 2*LhsProgress) ));
+
+      for(Index i1=peeled_mc3; i1<peeled_mc2; i1+=actual_panel_rows)
       {
-        Index i = peeled_mc;
-        const LhsScalar* blA = &blockA[i*strideA+offsetA*LhsProgress];
-        prefetch(&blA[0]);
-
-        // gets res block as register
-        AccPacket C0, C1, C2, C3;
-                  traits.initAcc(C0);
-                  traits.initAcc(C1);
-        if(nr==4) traits.initAcc(C2);
-        if(nr==4) traits.initAcc(C3);
-
-        // performs "inner" product
-        const RhsScalar* blB = unpackedB;
-        for(Index k=0; k<peeled_kc; k+=4)
+        Index actual_panel_end = (std::min)(i1+actual_panel_rows, peeled_mc2);
+        for(Index j2=0; j2<packet_cols4; j2+=nr)
         {
-          if(nr==2)
+          for(Index i=i1; i<actual_panel_end; i+=2*LhsProgress)
           {
-            LhsPacket A0;
-            RhsPacket B_0, B1;
+          
+          // We selected a 2*Traits::LhsProgress x nr micro block of res which is entirely
+          // stored into 2 x nr registers.
+          
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(2*Traits::LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0, C1, C2, C3,
+                    C4, C5, C6, C7;
+          traits.initAcc(C0); traits.initAcc(C1); traits.initAcc(C2); traits.initAcc(C3);
+          traits.initAcc(C4); traits.initAcc(C5); traits.initAcc(C6); traits.initAcc(C7);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2 + 0);
+          LinearMapper r1 = res.getLinearMapper(i, j2 + 1);
+          LinearMapper r2 = res.getLinearMapper(i, j2 + 2);
+          LinearMapper r3 = res.getLinearMapper(i, j2 + 3);
+
+          r0.prefetch(prefetch_res_offset);
+          r1.prefetch(prefetch_res_offset);
+          r2.prefetch(prefetch_res_offset);
+          r3.prefetch(prefetch_res_offset);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
+          prefetch(&blB[0]);
+          LhsPacket A0, A1;
 
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[2*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadLhs(&blA[1*LhsProgress], A0);
-            traits.loadRhs(&blB[3*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[4*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadLhs(&blA[2*LhsProgress], A0);
-            traits.loadRhs(&blB[5*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[6*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadLhs(&blA[3*LhsProgress], A0);
-            traits.loadRhs(&blB[7*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.madd(A0,B1,C1,B1);
-          }
-          else
+          for(Index k=0; k<peeled_kc; k+=pk)
           {
-            LhsPacket A0;
-            RhsPacket B_0, B1, B2, B3;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[2*RhsProgress], B2);
-            traits.loadRhs(&blB[3*RhsProgress], B3);
-            traits.loadRhs(&blB[4*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadRhs(&blB[5*RhsProgress], B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.loadRhs(&blB[6*RhsProgress], B2);
-            traits.madd(A0,B3,C3,B3);
-            traits.loadLhs(&blA[1*LhsProgress], A0);
-            traits.loadRhs(&blB[7*RhsProgress], B3);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[8*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadRhs(&blB[9*RhsProgress], B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.loadRhs(&blB[10*RhsProgress], B2);
-            traits.madd(A0,B3,C3,B3);
-            traits.loadLhs(&blA[2*LhsProgress], A0);
-            traits.loadRhs(&blB[11*RhsProgress], B3);
-
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[12*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadRhs(&blB[13*RhsProgress], B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.loadRhs(&blB[14*RhsProgress], B2);
-            traits.madd(A0,B3,C3,B3);
-
-            traits.loadLhs(&blA[3*LhsProgress], A0);
-            traits.loadRhs(&blB[15*RhsProgress], B3);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.madd(A0,B3,C3,B3);
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 2pX4");
+            RhsPacket B_0, B1, B2, B3, T0;
+
+          // NOTE: the begin/end asm comments below work around bug 935!
+          // but they are not enough for gcc>=6 without FMA (bug 1637)
+          #if EIGEN_GNUC_AT_LEAST(6,0) && defined(EIGEN_VECTORIZE_SSE)
+            #define EIGEN_GEBP_2PX4_SPILLING_WORKAROUND __asm__  ("" : [a0] "+x,m" (A0),[a1] "+x,m" (A1));
+          #else
+            #define EIGEN_GEBP_2PX4_SPILLING_WORKAROUND
+          #endif
+          #define EIGEN_GEBGP_ONESTEP(K) \
+            do {                                                                \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 2pX4");        \
+              traits.loadLhs(&blA[(0+2*K)*LhsProgress], A0);                    \
+              traits.loadLhs(&blA[(1+2*K)*LhsProgress], A1);                    \
+              traits.broadcastRhs(&blB[(0+4*K)*RhsProgress], B_0, B1, B2, B3);  \
+              traits.madd(A0, B_0, C0, T0);                                     \
+              traits.madd(A1, B_0, C4, B_0);                                    \
+              traits.madd(A0, B1,  C1, T0);                                     \
+              traits.madd(A1, B1,  C5, B1);                                     \
+              traits.madd(A0, B2,  C2, T0);                                     \
+              traits.madd(A1, B2,  C6, B2);                                     \
+              traits.madd(A0, B3,  C3, T0);                                     \
+              traits.madd(A1, B3,  C7, B3);                                     \
+              EIGEN_GEBP_2PX4_SPILLING_WORKAROUND                               \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 2pX4");          \
+            } while(false)
+            
+            internal::prefetch(blB+(48+0));
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            internal::prefetch(blB+(48+16));
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
+
+            blB += pk*4*RhsProgress;
+            blA += pk*(2*Traits::LhsProgress);
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 2pX4");
           }
-
-          blB += nr*4*RhsProgress;
-          blA += 4*LhsProgress;
-        }
-        // process remaining peeled loop
-        for(Index k=peeled_kc; k<depth; k++)
-        {
-          if(nr==2)
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
           {
-            LhsPacket A0;
-            RhsPacket B_0, B1;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.madd(A0,B1,C1,B1);
+            RhsPacket B_0, B1, B2, B3, T0;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += 4*RhsProgress;
+            blA += 2*Traits::LhsProgress;
           }
-          else
-          {
-            LhsPacket A0;
-            RhsPacket B_0, B1, B2, B3;
+#undef EIGEN_GEBGP_ONESTEP
 
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-            traits.loadRhs(&blB[2*RhsProgress], B2);
-            traits.loadRhs(&blB[3*RhsProgress], B3);
+          ResPacket R0, R1, R2, R3;
+          ResPacket alphav = pset1<ResPacket>(alpha);
 
-            traits.madd(A0,B_0,C0,B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.madd(A0,B3,C3,B3);
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r0.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r1.loadPacket(0 * Traits::ResPacketSize);
+          R3 = r1.loadPacket(1 * Traits::ResPacketSize);
+          traits.acc(C0, alphav, R0);
+          traits.acc(C4, alphav, R1);
+          traits.acc(C1, alphav, R2);
+          traits.acc(C5, alphav, R3);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r0.storePacket(1 * Traits::ResPacketSize, R1);
+          r1.storePacket(0 * Traits::ResPacketSize, R2);
+          r1.storePacket(1 * Traits::ResPacketSize, R3);
+
+          R0 = r2.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r2.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r3.loadPacket(0 * Traits::ResPacketSize);
+          R3 = r3.loadPacket(1 * Traits::ResPacketSize);
+          traits.acc(C2,  alphav, R0);
+          traits.acc(C6,  alphav, R1);
+          traits.acc(C3,  alphav, R2);
+          traits.acc(C7,  alphav, R3);
+          r2.storePacket(0 * Traits::ResPacketSize, R0);
+          r2.storePacket(1 * Traits::ResPacketSize, R1);
+          r3.storePacket(0 * Traits::ResPacketSize, R2);
+          r3.storePacket(1 * Traits::ResPacketSize, R3);
           }
-
-          blB += nr*RhsProgress;
-          blA += LhsProgress;
         }
+      
+        // Deal with remaining columns of the rhs
+        for(Index j2=packet_cols4; j2<cols; j2++)
+        {
+          for(Index i=i1; i<actual_panel_end; i+=2*LhsProgress)
+          {
+          // One column at a time
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(2*Traits::LhsProgress)];
+          prefetch(&blA[0]);
 
-        ResPacket R0, R1, R2, R3;
-        ResPacket alphav = pset1<ResPacket>(alpha);
-
-        ResScalar* r0 = &res[(j2+0)*resStride + i];
-        ResScalar* r1 = r0 + resStride;
-        ResScalar* r2 = r1 + resStride;
-        ResScalar* r3 = r2 + resStride;
+          // gets res block as register
+          AccPacket C0, C4;
+          traits.initAcc(C0);
+          traits.initAcc(C4);
 
-                  R0 = ploadu<ResPacket>(r0);
-                  R1 = ploadu<ResPacket>(r1);
-        if(nr==4) R2 = ploadu<ResPacket>(r2);
-        if(nr==4) R3 = ploadu<ResPacket>(r3);
+          LinearMapper r0 = res.getLinearMapper(i, j2);
+          r0.prefetch(prefetch_res_offset);
 
-                  traits.acc(C0, alphav, R0);
-                  traits.acc(C1, alphav, R1);
-        if(nr==4) traits.acc(C2, alphav, R2);
-        if(nr==4) traits.acc(C3, alphav, R3);
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB];
+          LhsPacket A0, A1;
 
-                  pstoreu(r0, R0);
-                  pstoreu(r1, R1);
-        if(nr==4) pstoreu(r2, R2);
-        if(nr==4) pstoreu(r3, R3);
-      }
-      for(Index i=peeled_mc2; i<rows; i++)
-      {
-        const LhsScalar* blA = &blockA[i*strideA+offsetA];
-        prefetch(&blA[0]);
-
-        // gets a 1 x nr res block as registers
-        ResScalar C0(0), C1(0), C2(0), C3(0);
-        // TODO directly use blockB ???
-        const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
-        for(Index k=0; k<depth; k++)
-        {
-          if(nr==2)
+          for(Index k=0; k<peeled_kc; k+=pk)
           {
-            LhsScalar A0;
-            RhsScalar B_0, B1;
-
-            A0 = blA[k];
-            B_0 = blB[0];
-            B1 = blB[1];
-            MADD(cj,A0,B_0,C0,B_0);
-            MADD(cj,A0,B1,C1,B1);
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 2pX1");
+            RhsPacket B_0, B1;
+        
+#define EIGEN_GEBGP_ONESTEP(K) \
+            do {                                                                  \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 2pX1");          \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              traits.loadLhs(&blA[(0+2*K)*LhsProgress], A0);                      \
+              traits.loadLhs(&blA[(1+2*K)*LhsProgress], A1);                      \
+              traits.loadRhs(&blB[(0+K)*RhsProgress], B_0);                       \
+              traits.madd(A0, B_0, C0, B1);                                       \
+              traits.madd(A1, B_0, C4, B_0);                                      \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 2pX1");            \
+            } while(false)
+        
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
+
+            blB += pk*RhsProgress;
+            blA += pk*2*Traits::LhsProgress;
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 2pX1");
           }
-          else
+
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
           {
-            LhsScalar A0;
-            RhsScalar B_0, B1, B2, B3;
-
-            A0 = blA[k];
-            B_0 = blB[0];
-            B1 = blB[1];
-            B2 = blB[2];
-            B3 = blB[3];
-
-            MADD(cj,A0,B_0,C0,B_0);
-            MADD(cj,A0,B1,C1,B1);
-            MADD(cj,A0,B2,C2,B2);
-            MADD(cj,A0,B3,C3,B3);
+            RhsPacket B_0, B1;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += RhsProgress;
+            blA += 2*Traits::LhsProgress;
           }
+#undef EIGEN_GEBGP_ONESTEP
+          ResPacket R0, R1;
+          ResPacket alphav = pset1<ResPacket>(alpha);
 
-          blB += nr;
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r0.loadPacket(1 * Traits::ResPacketSize);
+          traits.acc(C0, alphav, R0);
+          traits.acc(C4, alphav, R1);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r0.storePacket(1 * Traits::ResPacketSize, R1);
+          }
         }
-                  res[(j2+0)*resStride + i] += alpha*C0;
-                  res[(j2+1)*resStride + i] += alpha*C1;
-        if(nr==4) res[(j2+2)*resStride + i] += alpha*C2;
-        if(nr==4) res[(j2+3)*resStride + i] += alpha*C3;
       }
     }
-    // process remaining rhs/res columns one at a time
-    // => do the same but with nr==1
-    for(Index j2=packet_cols; j2<cols; j2++)
+    //---------- Process 1 * LhsProgress rows at once ----------
+    if(mr>=1*Traits::LhsProgress)
     {
-      // unpack B
-      traits.unpackRhs(depth, &blockB[j2*strideB+offsetB], unpackedB);
-
-      for(Index i=0; i<peeled_mc; i+=mr)
+      // loops on each largest micro horizontal panel of lhs (1*LhsProgress x depth)
+      for(Index i=peeled_mc2; i<peeled_mc1; i+=1*LhsProgress)
       {
-        const LhsScalar* blA = &blockA[i*strideA+offsetA*mr];
-        prefetch(&blA[0]);
+        // loops on each largest micro vertical panel of rhs (depth * nr)
+        for(Index j2=0; j2<packet_cols4; j2+=nr)
+        {
+          // We select a 1*Traits::LhsProgress x nr micro block of res which is entirely
+          // stored into 1 x nr registers.
+          
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(1*Traits::LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0, C1, C2, C3;
+          traits.initAcc(C0);
+          traits.initAcc(C1);
+          traits.initAcc(C2);
+          traits.initAcc(C3);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2 + 0);
+          LinearMapper r1 = res.getLinearMapper(i, j2 + 1);
+          LinearMapper r2 = res.getLinearMapper(i, j2 + 2);
+          LinearMapper r3 = res.getLinearMapper(i, j2 + 3);
+
+          r0.prefetch(prefetch_res_offset);
+          r1.prefetch(prefetch_res_offset);
+          r2.prefetch(prefetch_res_offset);
+          r3.prefetch(prefetch_res_offset);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
+          prefetch(&blB[0]);
+          LhsPacket A0;
+
+          for(Index k=0; k<peeled_kc; k+=pk)
+          {
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 1pX4");
+            RhsPacket B_0, B1, B2, B3;
+               
+#define EIGEN_GEBGP_ONESTEP(K) \
+            do {                                                                \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 1pX4");        \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              traits.loadLhs(&blA[(0+1*K)*LhsProgress], A0);                    \
+              traits.broadcastRhs(&blB[(0+4*K)*RhsProgress], B_0, B1, B2, B3);  \
+              traits.madd(A0, B_0, C0, B_0);                                    \
+              traits.madd(A0, B1,  C1, B1);                                     \
+              traits.madd(A0, B2,  C2, B2);                                     \
+              traits.madd(A0, B3,  C3, B3);                                     \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 1pX4");          \
+            } while(false)
+            
+            internal::prefetch(blB+(48+0));
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            internal::prefetch(blB+(48+16));
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
+
+            blB += pk*4*RhsProgress;
+            blA += pk*1*LhsProgress;
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 1pX4");
+          }
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
+          {
+            RhsPacket B_0, B1, B2, B3;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += 4*RhsProgress;
+            blA += 1*LhsProgress;
+          }
+#undef EIGEN_GEBGP_ONESTEP
 
-        // TODO move the res loads to the stores
+          ResPacket R0, R1;
+          ResPacket alphav = pset1<ResPacket>(alpha);
 
-        // get res block as registers
-        AccPacket C0, C4;
-        traits.initAcc(C0);
-        traits.initAcc(C4);
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r1.loadPacket(0 * Traits::ResPacketSize);
+          traits.acc(C0, alphav, R0);
+          traits.acc(C1,  alphav, R1);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r1.storePacket(0 * Traits::ResPacketSize, R1);
+
+          R0 = r2.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r3.loadPacket(0 * Traits::ResPacketSize);
+          traits.acc(C2,  alphav, R0);
+          traits.acc(C3,  alphav, R1);
+          r2.storePacket(0 * Traits::ResPacketSize, R0);
+          r3.storePacket(0 * Traits::ResPacketSize, R1);
+        }
 
-        const RhsScalar* blB = unpackedB;
-        for(Index k=0; k<depth; k++)
+        // Deal with remaining columns of the rhs
+        for(Index j2=packet_cols4; j2<cols; j2++)
         {
-          LhsPacket A0, A1;
-          RhsPacket B_0;
-          RhsPacket T0;
-
-          traits.loadLhs(&blA[0*LhsProgress], A0);
-          traits.loadLhs(&blA[1*LhsProgress], A1);
-          traits.loadRhs(&blB[0*RhsProgress], B_0);
-          traits.madd(A0,B_0,C0,T0);
-          traits.madd(A1,B_0,C4,B_0);
+          // One column at a time
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(1*Traits::LhsProgress)];
+          prefetch(&blA[0]);
 
-          blB += RhsProgress;
-          blA += 2*LhsProgress;
-        }
-        ResPacket R0, R4;
-        ResPacket alphav = pset1<ResPacket>(alpha);
+          // gets res block as register
+          AccPacket C0;
+          traits.initAcc(C0);
 
-        ResScalar* r0 = &res[(j2+0)*resStride + i];
+          LinearMapper r0 = res.getLinearMapper(i, j2);
 
-        R0 = ploadu<ResPacket>(r0);
-        R4 = ploadu<ResPacket>(r0+ResPacketSize);
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB];
+          LhsPacket A0;
 
-        traits.acc(C0, alphav, R0);
-        traits.acc(C4, alphav, R4);
+          for(Index k=0; k<peeled_kc; k+=pk)
+          {
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 1pX1");
+            RhsPacket B_0;
+        
+#define EIGEN_GEBGP_ONESTEP(K) \
+            do {                                                                \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 1pX1");        \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              traits.loadLhs(&blA[(0+1*K)*LhsProgress], A0);                    \
+              traits.loadRhs(&blB[(0+K)*RhsProgress], B_0);                     \
+              traits.madd(A0, B_0, C0, B_0);                                    \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 1pX1");          \
+            } while(false);
+
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
+
+            blB += pk*RhsProgress;
+            blA += pk*1*Traits::LhsProgress;
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 1pX1");
+          }
 
-        pstoreu(r0,               R0);
-        pstoreu(r0+ResPacketSize, R4);
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
+          {
+            RhsPacket B_0;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += RhsProgress;
+            blA += 1*Traits::LhsProgress;
+          }
+#undef EIGEN_GEBGP_ONESTEP
+          ResPacket R0;
+          ResPacket alphav = pset1<ResPacket>(alpha);
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          traits.acc(C0, alphav, R0);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+        }
       }
-      if(rows-peeled_mc>=LhsProgress)
+    }
+    //---------- Process remaining rows, 1 at once ----------
+    if(peeled_mc1<rows)
+    {
+      // loop on each panel of the rhs
+      for(Index j2=0; j2<packet_cols4; j2+=nr)
       {
-        Index i = peeled_mc;
-        const LhsScalar* blA = &blockA[i*strideA+offsetA*LhsProgress];
-        prefetch(&blA[0]);
-
-        AccPacket C0;
-        traits.initAcc(C0);
-
-        const RhsScalar* blB = unpackedB;
-        for(Index k=0; k<depth; k++)
+        // loop on each row of the lhs (1*LhsProgress x depth)
+        for(Index i=peeled_mc1; i<rows; i+=1)
         {
-          LhsPacket A0;
-          RhsPacket B_0;
-          traits.loadLhs(blA, A0);
-          traits.loadRhs(blB, B_0);
-          traits.madd(A0, B_0, C0, B_0);
-          blB += RhsProgress;
-          blA += LhsProgress;
+          const LhsScalar* blA = &blockA[i*strideA+offsetA];
+          prefetch(&blA[0]);
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
+
+          // The following piece of code wont work for 512 bit registers
+          // Moreover, if LhsProgress==8 it assumes that there is a half packet of the same size
+          // as nr (which is currently 4) for the return type.
+          const int SResPacketHalfSize = unpacket_traits<typename unpacket_traits<SResPacket>::half>::size;
+          if ((SwappedTraits::LhsProgress % 4) == 0 &&
+              (SwappedTraits::LhsProgress <= 8) &&
+              (SwappedTraits::LhsProgress!=8 || SResPacketHalfSize==nr))
+          {
+            SAccPacket C0, C1, C2, C3;
+            straits.initAcc(C0);
+            straits.initAcc(C1);
+            straits.initAcc(C2);
+            straits.initAcc(C3);
+
+            const Index spk   = (std::max)(1,SwappedTraits::LhsProgress/4);
+            const Index endk  = (depth/spk)*spk;
+            const Index endk4 = (depth/(spk*4))*(spk*4);
+
+            Index k=0;
+            for(; k<endk4; k+=4*spk)
+            {
+              SLhsPacket A0,A1;
+              SRhsPacket B_0,B_1;
+
+              straits.loadLhsUnaligned(blB+0*SwappedTraits::LhsProgress, A0);
+              straits.loadLhsUnaligned(blB+1*SwappedTraits::LhsProgress, A1);
+
+              straits.loadRhsQuad(blA+0*spk, B_0);
+              straits.loadRhsQuad(blA+1*spk, B_1);
+              straits.madd(A0,B_0,C0,B_0);
+              straits.madd(A1,B_1,C1,B_1);
+
+              straits.loadLhsUnaligned(blB+2*SwappedTraits::LhsProgress, A0);
+              straits.loadLhsUnaligned(blB+3*SwappedTraits::LhsProgress, A1);
+              straits.loadRhsQuad(blA+2*spk, B_0);
+              straits.loadRhsQuad(blA+3*spk, B_1);
+              straits.madd(A0,B_0,C2,B_0);
+              straits.madd(A1,B_1,C3,B_1);
+
+              blB += 4*SwappedTraits::LhsProgress;
+              blA += 4*spk;
+            }
+            C0 = padd(padd(C0,C1),padd(C2,C3));
+            for(; k<endk; k+=spk)
+            {
+              SLhsPacket A0;
+              SRhsPacket B_0;
+
+              straits.loadLhsUnaligned(blB, A0);
+              straits.loadRhsQuad(blA, B_0);
+              straits.madd(A0,B_0,C0,B_0);
+
+              blB += SwappedTraits::LhsProgress;
+              blA += spk;
+            }
+            if(SwappedTraits::LhsProgress==8)
+            {
+              // Special case where we have to first reduce the accumulation register C0
+              typedef typename conditional<SwappedTraits::LhsProgress>=8,typename unpacket_traits<SResPacket>::half,SResPacket>::type SResPacketHalf;
+              typedef typename conditional<SwappedTraits::LhsProgress>=8,typename unpacket_traits<SLhsPacket>::half,SLhsPacket>::type SLhsPacketHalf;
+              typedef typename conditional<SwappedTraits::LhsProgress>=8,typename unpacket_traits<SLhsPacket>::half,SRhsPacket>::type SRhsPacketHalf;
+              typedef typename conditional<SwappedTraits::LhsProgress>=8,typename unpacket_traits<SAccPacket>::half,SAccPacket>::type SAccPacketHalf;
+
+              SResPacketHalf R = res.template gatherPacket<SResPacketHalf>(i, j2);
+              SResPacketHalf alphav = pset1<SResPacketHalf>(alpha);
+
+              if(depth-endk>0)
+              {
+                // We have to handle the last row of the rhs which corresponds to a half-packet
+                SLhsPacketHalf a0;
+                SRhsPacketHalf b0;
+                straits.loadLhsUnaligned(blB, a0);
+                straits.loadRhs(blA, b0);
+                SAccPacketHalf c0 = predux_downto4(C0);
+                straits.madd(a0,b0,c0,b0);
+                straits.acc(c0, alphav, R);
+              }
+              else
+              {
+                straits.acc(predux_downto4(C0), alphav, R);
+              }
+              res.scatterPacket(i, j2, R);
+            }
+            else
+            {
+              SResPacket R = res.template gatherPacket<SResPacket>(i, j2);
+              SResPacket alphav = pset1<SResPacket>(alpha);
+              straits.acc(C0, alphav, R);
+              res.scatterPacket(i, j2, R);
+            }
+          }
+          else // scalar path
+          {
+            // get a 1 x 4 res block as registers
+            ResScalar C0(0), C1(0), C2(0), C3(0);
+
+            for(Index k=0; k<depth; k++)
+            {
+              LhsScalar A0;
+              RhsScalar B_0, B_1;
+
+              A0 = blA[k];
+
+              B_0 = blB[0];
+              B_1 = blB[1];
+              CJMADD(cj,A0,B_0,C0,  B_0);
+              CJMADD(cj,A0,B_1,C1,  B_1);
+              
+              B_0 = blB[2];
+              B_1 = blB[3];
+              CJMADD(cj,A0,B_0,C2,  B_0);
+              CJMADD(cj,A0,B_1,C3,  B_1);
+              
+              blB += 4;
+            }
+            res(i, j2 + 0) += alpha * C0;
+            res(i, j2 + 1) += alpha * C1;
+            res(i, j2 + 2) += alpha * C2;
+            res(i, j2 + 3) += alpha * C3;
+          }
         }
-
-        ResPacket alphav = pset1<ResPacket>(alpha);
-        ResPacket R0 = ploadu<ResPacket>(&res[(j2+0)*resStride + i]);
-        traits.acc(C0, alphav, R0);
-        pstoreu(&res[(j2+0)*resStride + i], R0);
       }
-      for(Index i=peeled_mc2; i<rows; i++)
+      // remaining columns
+      for(Index j2=packet_cols4; j2<cols; j2++)
       {
-        const LhsScalar* blA = &blockA[i*strideA+offsetA];
-        prefetch(&blA[0]);
-
-        // gets a 1 x 1 res block as registers
-        ResScalar C0(0);
-        // FIXME directly use blockB ??
-        const RhsScalar* blB = &blockB[j2*strideB+offsetB];
-        for(Index k=0; k<depth; k++)
+        // loop on each row of the lhs (1*LhsProgress x depth)
+        for(Index i=peeled_mc1; i<rows; i+=1)
         {
-          LhsScalar A0 = blA[k];
-          RhsScalar B_0 = blB[k];
-          MADD(cj, A0, B_0, C0, B_0);
+          const LhsScalar* blA = &blockA[i*strideA+offsetA];
+          prefetch(&blA[0]);
+          // gets a 1 x 1 res block as registers
+          ResScalar C0(0);
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB];
+          for(Index k=0; k<depth; k++)
+          {
+            LhsScalar A0 = blA[k];
+            RhsScalar B_0 = blB[k];
+            CJMADD(cj, A0, B_0, C0, B_0);
+          }
+          res(i, j2) += alpha * C0;
         }
-        res[(j2+0)*resStride + i] += alpha*C0;
       }
     }
   }
@@ -1114,81 +1692,193 @@ EIGEN_ASM_COMMENT("mybegin4");
 //
 //  32 33 34 35 ...
 //  36 36 38 39 ...
-template<typename Scalar, typename Index, int Pack1, int Pack2, int StorageOrder, bool Conjugate, bool PanelMode>
-struct gemm_pack_lhs
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, bool Conjugate, bool PanelMode>
+struct gemm_pack_lhs<Scalar, Index, DataMapper, Pack1, Pack2, ColMajor, Conjugate, PanelMode>
 {
-  EIGEN_DONT_INLINE void operator()(Scalar* blockA, const Scalar* EIGEN_RESTRICT _lhs, Index lhsStride, Index depth, Index rows, Index stride=0, Index offset=0);
+  typedef typename DataMapper::LinearMapper LinearMapper;
+  EIGEN_DONT_INLINE void operator()(Scalar* blockA, const DataMapper& lhs, Index depth, Index rows, Index stride=0, Index offset=0);
 };
 
-template<typename Scalar, typename Index, int Pack1, int Pack2, int StorageOrder, bool Conjugate, bool PanelMode>
-EIGEN_DONT_INLINE void gemm_pack_lhs<Scalar, Index, Pack1, Pack2, StorageOrder, Conjugate, PanelMode>
-  ::operator()(Scalar* blockA, const Scalar* EIGEN_RESTRICT _lhs, Index lhsStride, Index depth, Index rows, Index stride, Index offset)
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, bool Conjugate, bool PanelMode>
+EIGEN_DONT_INLINE void gemm_pack_lhs<Scalar, Index, DataMapper, Pack1, Pack2, ColMajor, Conjugate, PanelMode>
+  ::operator()(Scalar* blockA, const DataMapper& lhs, Index depth, Index rows, Index stride, Index offset)
 {
   typedef typename packet_traits<Scalar>::type Packet;
   enum { PacketSize = packet_traits<Scalar>::size };
 
   EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK LHS");
-  EIGEN_UNUSED_VARIABLE(stride)
-  EIGEN_UNUSED_VARIABLE(offset)
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
   eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
-  eigen_assert( (StorageOrder==RowMajor) || ((Pack1%PacketSize)==0 && Pack1<=4*PacketSize) );
+  eigen_assert( ((Pack1%PacketSize)==0 && Pack1<=4*PacketSize) || (Pack1<=4) );
   conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
-  const_blas_data_mapper<Scalar, Index, StorageOrder> lhs(_lhs,lhsStride);
   Index count = 0;
-  Index peeled_mc = (rows/Pack1)*Pack1;
-  for(Index i=0; i<peeled_mc; i+=Pack1)
+
+  const Index peeled_mc3 = Pack1>=3*PacketSize ? (rows/(3*PacketSize))*(3*PacketSize) : 0;
+  const Index peeled_mc2 = Pack1>=2*PacketSize ? peeled_mc3+((rows-peeled_mc3)/(2*PacketSize))*(2*PacketSize) : 0;
+  const Index peeled_mc1 = Pack1>=1*PacketSize ? (rows/(1*PacketSize))*(1*PacketSize) : 0;
+  const Index peeled_mc0 = Pack2>=1*PacketSize ? peeled_mc1
+                         : Pack2>1             ? (rows/Pack2)*Pack2 : 0;
+
+  Index i=0;
+
+  // Pack 3 packets
+  if(Pack1>=3*PacketSize)
   {
-    if(PanelMode) count += Pack1 * offset;
+    for(; i<peeled_mc3; i+=3*PacketSize)
+    {
+      if(PanelMode) count += (3*PacketSize) * offset;
 
-    if(StorageOrder==ColMajor)
+      for(Index k=0; k<depth; k++)
+      {
+        Packet A, B, C;
+        A = lhs.loadPacket(i+0*PacketSize, k);
+        B = lhs.loadPacket(i+1*PacketSize, k);
+        C = lhs.loadPacket(i+2*PacketSize, k);
+        pstore(blockA+count, cj.pconj(A)); count+=PacketSize;
+        pstore(blockA+count, cj.pconj(B)); count+=PacketSize;
+        pstore(blockA+count, cj.pconj(C)); count+=PacketSize;
+      }
+      if(PanelMode) count += (3*PacketSize) * (stride-offset-depth);
+    }
+  }
+  // Pack 2 packets
+  if(Pack1>=2*PacketSize)
+  {
+    for(; i<peeled_mc2; i+=2*PacketSize)
     {
+      if(PanelMode) count += (2*PacketSize) * offset;
+
       for(Index k=0; k<depth; k++)
       {
-        Packet A, B, C, D;
-        if(Pack1>=1*PacketSize) A = ploadu<Packet>(&lhs(i+0*PacketSize, k));
-        if(Pack1>=2*PacketSize) B = ploadu<Packet>(&lhs(i+1*PacketSize, k));
-        if(Pack1>=3*PacketSize) C = ploadu<Packet>(&lhs(i+2*PacketSize, k));
-        if(Pack1>=4*PacketSize) D = ploadu<Packet>(&lhs(i+3*PacketSize, k));
-        if(Pack1>=1*PacketSize) { pstore(blockA+count, cj.pconj(A)); count+=PacketSize; }
-        if(Pack1>=2*PacketSize) { pstore(blockA+count, cj.pconj(B)); count+=PacketSize; }
-        if(Pack1>=3*PacketSize) { pstore(blockA+count, cj.pconj(C)); count+=PacketSize; }
-        if(Pack1>=4*PacketSize) { pstore(blockA+count, cj.pconj(D)); count+=PacketSize; }
+        Packet A, B;
+        A = lhs.loadPacket(i+0*PacketSize, k);
+        B = lhs.loadPacket(i+1*PacketSize, k);
+        pstore(blockA+count, cj.pconj(A)); count+=PacketSize;
+        pstore(blockA+count, cj.pconj(B)); count+=PacketSize;
       }
+      if(PanelMode) count += (2*PacketSize) * (stride-offset-depth);
     }
-    else
+  }
+  // Pack 1 packets
+  if(Pack1>=1*PacketSize)
+  {
+    for(; i<peeled_mc1; i+=1*PacketSize)
     {
+      if(PanelMode) count += (1*PacketSize) * offset;
+
+      for(Index k=0; k<depth; k++)
+      {
+        Packet A;
+        A = lhs.loadPacket(i+0*PacketSize, k);
+        pstore(blockA+count, cj.pconj(A));
+        count+=PacketSize;
+      }
+      if(PanelMode) count += (1*PacketSize) * (stride-offset-depth);
+    }
+  }
+  // Pack scalars
+  if(Pack2<PacketSize && Pack2>1)
+  {
+    for(; i<peeled_mc0; i+=Pack2)
+    {
+      if(PanelMode) count += Pack2 * offset;
+
       for(Index k=0; k<depth; k++)
+        for(Index w=0; w<Pack2; w++)
+          blockA[count++] = cj(lhs(i+w, k));
+
+      if(PanelMode) count += Pack2 * (stride-offset-depth);
+    }
+  }
+  for(; i<rows; i++)
+  {
+    if(PanelMode) count += offset;
+    for(Index k=0; k<depth; k++)
+      blockA[count++] = cj(lhs(i, k));
+    if(PanelMode) count += (stride-offset-depth);
+  }
+}
+
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, bool Conjugate, bool PanelMode>
+struct gemm_pack_lhs<Scalar, Index, DataMapper, Pack1, Pack2, RowMajor, Conjugate, PanelMode>
+{
+  typedef typename DataMapper::LinearMapper LinearMapper;
+  EIGEN_DONT_INLINE void operator()(Scalar* blockA, const DataMapper& lhs, Index depth, Index rows, Index stride=0, Index offset=0);
+};
+
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, bool Conjugate, bool PanelMode>
+EIGEN_DONT_INLINE void gemm_pack_lhs<Scalar, Index, DataMapper, Pack1, Pack2, RowMajor, Conjugate, PanelMode>
+  ::operator()(Scalar* blockA, const DataMapper& lhs, Index depth, Index rows, Index stride, Index offset)
+{
+  typedef typename packet_traits<Scalar>::type Packet;
+  enum { PacketSize = packet_traits<Scalar>::size };
+
+  EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK LHS");
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
+  eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
+  conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
+  Index count = 0;
+
+//   const Index peeled_mc3 = Pack1>=3*PacketSize ? (rows/(3*PacketSize))*(3*PacketSize) : 0;
+//   const Index peeled_mc2 = Pack1>=2*PacketSize ? peeled_mc3+((rows-peeled_mc3)/(2*PacketSize))*(2*PacketSize) : 0;
+//   const Index peeled_mc1 = Pack1>=1*PacketSize ? (rows/(1*PacketSize))*(1*PacketSize) : 0;
+
+  int pack = Pack1;
+  Index i = 0;
+  while(pack>0)
+  {
+    Index remaining_rows = rows-i;
+    Index peeled_mc = i+(remaining_rows/pack)*pack;
+    for(; i<peeled_mc; i+=pack)
+    {
+      if(PanelMode) count += pack * offset;
+
+      const Index peeled_k = (depth/PacketSize)*PacketSize;
+      Index k=0;
+      if(pack>=PacketSize)
+      {
+        for(; k<peeled_k; k+=PacketSize)
+        {
+          for (Index m = 0; m < pack; m += PacketSize)
+          {
+            PacketBlock<Packet> kernel;
+            for (int p = 0; p < PacketSize; ++p) kernel.packet[p] = lhs.loadPacket(i+p+m, k);
+            ptranspose(kernel);
+            for (int p = 0; p < PacketSize; ++p) pstore(blockA+count+m+(pack)*p, cj.pconj(kernel.packet[p]));
+          }
+          count += PacketSize*pack;
+        }
+      }
+      for(; k<depth; k++)
       {
-        // TODO add a vectorized transpose here
         Index w=0;
-        for(; w<Pack1-3; w+=4)
+        for(; w<pack-3; w+=4)
         {
           Scalar a(cj(lhs(i+w+0, k))),
-                  b(cj(lhs(i+w+1, k))),
-                  c(cj(lhs(i+w+2, k))),
-                  d(cj(lhs(i+w+3, k)));
+                 b(cj(lhs(i+w+1, k))),
+                 c(cj(lhs(i+w+2, k))),
+                 d(cj(lhs(i+w+3, k)));
           blockA[count++] = a;
           blockA[count++] = b;
           blockA[count++] = c;
           blockA[count++] = d;
         }
-        if(Pack1%4)
-          for(;w<Pack1;++w)
+        if(pack%4)
+          for(;w<pack;++w)
             blockA[count++] = cj(lhs(i+w, k));
       }
+
+      if(PanelMode) count += pack * (stride-offset-depth);
     }
-    if(PanelMode) count += Pack1 * (stride-offset-depth);
-  }
-  if(rows-peeled_mc>=Pack2)
-  {
-    if(PanelMode) count += Pack2*offset;
-    for(Index k=0; k<depth; k++)
-      for(Index w=0; w<Pack2; w++)
-        blockA[count++] = cj(lhs(peeled_mc+w, k));
-    if(PanelMode) count += Pack2 * (stride-offset-depth);
-    peeled_mc += Pack2;
+
+    pack -= PacketSize;
+    if(pack<Pack2 && (pack+PacketSize)!=Pack2)
+      pack = Pack2;
   }
-  for(Index i=peeled_mc; i<rows; i++)
+
+  for(; i<rows; i++)
   {
     if(PanelMode) count += offset;
     for(Index k=0; k<depth; k++)
@@ -1204,53 +1894,123 @@ EIGEN_DONT_INLINE void gemm_pack_lhs<Scalar, Index, Pack1, Pack2, StorageOrder,
 //  4  5  6  7   16 17 18 19   25 28
 //  8  9 10 11   20 21 22 23   26 29
 //  .  .  .  .    .  .  .  .    .  .
-template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
-struct gemm_pack_rhs<Scalar, Index, nr, ColMajor, Conjugate, PanelMode>
+template<typename Scalar, typename Index, typename DataMapper, int nr, bool Conjugate, bool PanelMode>
+struct gemm_pack_rhs<Scalar, Index, DataMapper, nr, ColMajor, Conjugate, PanelMode>
 {
   typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename DataMapper::LinearMapper LinearMapper;
   enum { PacketSize = packet_traits<Scalar>::size };
-  EIGEN_DONT_INLINE void operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride=0, Index offset=0);
+  EIGEN_DONT_INLINE void operator()(Scalar* blockB, const DataMapper& rhs, Index depth, Index cols, Index stride=0, Index offset=0);
 };
 
-template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
-EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, ColMajor, Conjugate, PanelMode>
-  ::operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride, Index offset)
+template<typename Scalar, typename Index, typename DataMapper, int nr, bool Conjugate, bool PanelMode>
+EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, DataMapper, nr, ColMajor, Conjugate, PanelMode>
+  ::operator()(Scalar* blockB, const DataMapper& rhs, Index depth, Index cols, Index stride, Index offset)
 {
   EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK RHS COLMAJOR");
-  EIGEN_UNUSED_VARIABLE(stride)
-  EIGEN_UNUSED_VARIABLE(offset)
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
   eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
   conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
-  Index packet_cols = (cols/nr) * nr;
+  Index packet_cols8 = nr>=8 ? (cols/8) * 8 : 0;
+  Index packet_cols4 = nr>=4 ? (cols/4) * 4 : 0;
   Index count = 0;
-  for(Index j2=0; j2<packet_cols; j2+=nr)
+  const Index peeled_k = (depth/PacketSize)*PacketSize;
+//   if(nr>=8)
+//   {
+//     for(Index j2=0; j2<packet_cols8; j2+=8)
+//     {
+//       // skip what we have before
+//       if(PanelMode) count += 8 * offset;
+//       const Scalar* b0 = &rhs[(j2+0)*rhsStride];
+//       const Scalar* b1 = &rhs[(j2+1)*rhsStride];
+//       const Scalar* b2 = &rhs[(j2+2)*rhsStride];
+//       const Scalar* b3 = &rhs[(j2+3)*rhsStride];
+//       const Scalar* b4 = &rhs[(j2+4)*rhsStride];
+//       const Scalar* b5 = &rhs[(j2+5)*rhsStride];
+//       const Scalar* b6 = &rhs[(j2+6)*rhsStride];
+//       const Scalar* b7 = &rhs[(j2+7)*rhsStride];
+//       Index k=0;
+//       if(PacketSize==8) // TODO enbale vectorized transposition for PacketSize==4
+//       {
+//         for(; k<peeled_k; k+=PacketSize) {
+//           PacketBlock<Packet> kernel;
+//           for (int p = 0; p < PacketSize; ++p) {
+//             kernel.packet[p] = ploadu<Packet>(&rhs[(j2+p)*rhsStride+k]);
+//           }
+//           ptranspose(kernel);
+//           for (int p = 0; p < PacketSize; ++p) {
+//             pstoreu(blockB+count, cj.pconj(kernel.packet[p]));
+//             count+=PacketSize;
+//           }
+//         }
+//       }
+//       for(; k<depth; k++)
+//       {
+//         blockB[count+0] = cj(b0[k]);
+//         blockB[count+1] = cj(b1[k]);
+//         blockB[count+2] = cj(b2[k]);
+//         blockB[count+3] = cj(b3[k]);
+//         blockB[count+4] = cj(b4[k]);
+//         blockB[count+5] = cj(b5[k]);
+//         blockB[count+6] = cj(b6[k]);
+//         blockB[count+7] = cj(b7[k]);
+//         count += 8;
+//       }
+//       // skip what we have after
+//       if(PanelMode) count += 8 * (stride-offset-depth);
+//     }
+//   }
+
+  if(nr>=4)
   {
-    // skip what we have before
-    if(PanelMode) count += nr * offset;
-    const Scalar* b0 = &rhs[(j2+0)*rhsStride];
-    const Scalar* b1 = &rhs[(j2+1)*rhsStride];
-    const Scalar* b2 = &rhs[(j2+2)*rhsStride];
-    const Scalar* b3 = &rhs[(j2+3)*rhsStride];
-    for(Index k=0; k<depth; k++)
+    for(Index j2=packet_cols8; j2<packet_cols4; j2+=4)
     {
-                blockB[count+0] = cj(b0[k]);
-                blockB[count+1] = cj(b1[k]);
-      if(nr==4) blockB[count+2] = cj(b2[k]);
-      if(nr==4) blockB[count+3] = cj(b3[k]);
-      count += nr;
+      // skip what we have before
+      if(PanelMode) count += 4 * offset;
+      const LinearMapper dm0 = rhs.getLinearMapper(0, j2 + 0);
+      const LinearMapper dm1 = rhs.getLinearMapper(0, j2 + 1);
+      const LinearMapper dm2 = rhs.getLinearMapper(0, j2 + 2);
+      const LinearMapper dm3 = rhs.getLinearMapper(0, j2 + 3);
+
+      Index k=0;
+      if((PacketSize%4)==0) // TODO enable vectorized transposition for PacketSize==2 ??
+      {
+        for(; k<peeled_k; k+=PacketSize) {
+          PacketBlock<Packet,(PacketSize%4)==0?4:PacketSize> kernel;
+          kernel.packet[0] = dm0.loadPacket(k);
+          kernel.packet[1%PacketSize] = dm1.loadPacket(k);
+          kernel.packet[2%PacketSize] = dm2.loadPacket(k);
+          kernel.packet[3%PacketSize] = dm3.loadPacket(k);
+          ptranspose(kernel);
+          pstoreu(blockB+count+0*PacketSize, cj.pconj(kernel.packet[0]));
+          pstoreu(blockB+count+1*PacketSize, cj.pconj(kernel.packet[1%PacketSize]));
+          pstoreu(blockB+count+2*PacketSize, cj.pconj(kernel.packet[2%PacketSize]));
+          pstoreu(blockB+count+3*PacketSize, cj.pconj(kernel.packet[3%PacketSize]));
+          count+=4*PacketSize;
+        }
+      }
+      for(; k<depth; k++)
+      {
+        blockB[count+0] = cj(dm0(k));
+        blockB[count+1] = cj(dm1(k));
+        blockB[count+2] = cj(dm2(k));
+        blockB[count+3] = cj(dm3(k));
+        count += 4;
+      }
+      // skip what we have after
+      if(PanelMode) count += 4 * (stride-offset-depth);
     }
-    // skip what we have after
-    if(PanelMode) count += nr * (stride-offset-depth);
   }
 
   // copy the remaining columns one at a time (nr==1)
-  for(Index j2=packet_cols; j2<cols; ++j2)
+  for(Index j2=packet_cols4; j2<cols; ++j2)
   {
     if(PanelMode) count += offset;
-    const Scalar* b0 = &rhs[(j2+0)*rhsStride];
+    const LinearMapper dm0 = rhs.getLinearMapper(0, j2);
     for(Index k=0; k<depth; k++)
     {
-      blockB[count] = cj(b0[k]);
+      blockB[count] = cj(dm0(k));
       count += 1;
     }
     if(PanelMode) count += (stride-offset-depth);
@@ -1258,48 +2018,93 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, ColMajor, Conjugate, Pan
 }
 
 // this version is optimized for row major matrices
-template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
-struct gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, PanelMode>
+template<typename Scalar, typename Index, typename DataMapper, int nr, bool Conjugate, bool PanelMode>
+struct gemm_pack_rhs<Scalar, Index, DataMapper, nr, RowMajor, Conjugate, PanelMode>
 {
+  typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename DataMapper::LinearMapper LinearMapper;
   enum { PacketSize = packet_traits<Scalar>::size };
-  EIGEN_DONT_INLINE void operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride=0, Index offset=0);
+  EIGEN_DONT_INLINE void operator()(Scalar* blockB, const DataMapper& rhs, Index depth, Index cols, Index stride=0, Index offset=0);
 };
 
-template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
-EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, PanelMode>
-  ::operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride, Index offset)
+template<typename Scalar, typename Index, typename DataMapper, int nr, bool Conjugate, bool PanelMode>
+EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, DataMapper, nr, RowMajor, Conjugate, PanelMode>
+  ::operator()(Scalar* blockB, const DataMapper& rhs, Index depth, Index cols, Index stride, Index offset)
 {
   EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK RHS ROWMAJOR");
-  EIGEN_UNUSED_VARIABLE(stride)
-  EIGEN_UNUSED_VARIABLE(offset)
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
   eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
   conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
-  Index packet_cols = (cols/nr) * nr;
+  Index packet_cols8 = nr>=8 ? (cols/8) * 8 : 0;
+  Index packet_cols4 = nr>=4 ? (cols/4) * 4 : 0;
   Index count = 0;
-  for(Index j2=0; j2<packet_cols; j2+=nr)
+
+//   if(nr>=8)
+//   {
+//     for(Index j2=0; j2<packet_cols8; j2+=8)
+//     {
+//       // skip what we have before
+//       if(PanelMode) count += 8 * offset;
+//       for(Index k=0; k<depth; k++)
+//       {
+//         if (PacketSize==8) {
+//           Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]);
+//           pstoreu(blockB+count, cj.pconj(A));
+//         } else if (PacketSize==4) {
+//           Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]);
+//           Packet B = ploadu<Packet>(&rhs[k*rhsStride + j2 + PacketSize]);
+//           pstoreu(blockB+count, cj.pconj(A));
+//           pstoreu(blockB+count+PacketSize, cj.pconj(B));
+//         } else {
+//           const Scalar* b0 = &rhs[k*rhsStride + j2];
+//           blockB[count+0] = cj(b0[0]);
+//           blockB[count+1] = cj(b0[1]);
+//           blockB[count+2] = cj(b0[2]);
+//           blockB[count+3] = cj(b0[3]);
+//           blockB[count+4] = cj(b0[4]);
+//           blockB[count+5] = cj(b0[5]);
+//           blockB[count+6] = cj(b0[6]);
+//           blockB[count+7] = cj(b0[7]);
+//         }
+//         count += 8;
+//       }
+//       // skip what we have after
+//       if(PanelMode) count += 8 * (stride-offset-depth);
+//     }
+//   }
+  if(nr>=4)
   {
-    // skip what we have before
-    if(PanelMode) count += nr * offset;
-    for(Index k=0; k<depth; k++)
+    for(Index j2=packet_cols8; j2<packet_cols4; j2+=4)
     {
-      const Scalar* b0 = &rhs[k*rhsStride + j2];
-                blockB[count+0] = cj(b0[0]);
-                blockB[count+1] = cj(b0[1]);
-      if(nr==4) blockB[count+2] = cj(b0[2]);
-      if(nr==4) blockB[count+3] = cj(b0[3]);
-      count += nr;
+      // skip what we have before
+      if(PanelMode) count += 4 * offset;
+      for(Index k=0; k<depth; k++)
+      {
+        if (PacketSize==4) {
+          Packet A = rhs.loadPacket(k, j2);
+          pstoreu(blockB+count, cj.pconj(A));
+          count += PacketSize;
+        } else {
+          const LinearMapper dm0 = rhs.getLinearMapper(k, j2);
+          blockB[count+0] = cj(dm0(0));
+          blockB[count+1] = cj(dm0(1));
+          blockB[count+2] = cj(dm0(2));
+          blockB[count+3] = cj(dm0(3));
+          count += 4;
+        }
+      }
+      // skip what we have after
+      if(PanelMode) count += 4 * (stride-offset-depth);
     }
-    // skip what we have after
-    if(PanelMode) count += nr * (stride-offset-depth);
   }
   // copy the remaining columns one at a time (nr==1)
-  for(Index j2=packet_cols; j2<cols; ++j2)
+  for(Index j2=packet_cols4; j2<cols; ++j2)
   {
     if(PanelMode) count += offset;
-    const Scalar* b0 = &rhs[j2];
     for(Index k=0; k<depth; k++)
     {
-      blockB[count] = cj(b0[k*rhsStride]);
+      blockB[count] = cj(rhs(k, j2));
       count += 1;
     }
     if(PanelMode) count += stride-offset-depth;
@@ -1312,8 +2117,8 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, Pan
   * \sa setCpuCacheSize */
 inline std::ptrdiff_t l1CacheSize()
 {
-  std::ptrdiff_t l1, l2;
-  internal::manage_caching_sizes(GetAction, &l1, &l2);
+  std::ptrdiff_t l1, l2, l3;
+  internal::manage_caching_sizes(GetAction, &l1, &l2, &l3);
   return l1;
 }
 
@@ -1321,19 +2126,29 @@ inline std::ptrdiff_t l1CacheSize()
   * \sa setCpuCacheSize */
 inline std::ptrdiff_t l2CacheSize()
 {
-  std::ptrdiff_t l1, l2;
-  internal::manage_caching_sizes(GetAction, &l1, &l2);
+  std::ptrdiff_t l1, l2, l3;
+  internal::manage_caching_sizes(GetAction, &l1, &l2, &l3);
   return l2;
 }
 
+/** \returns the currently set level 3 cpu cache size (in bytes) used to estimate the ideal blocking size paramete\
+rs.                                                                                                                
+* \sa setCpuCacheSize */
+inline std::ptrdiff_t l3CacheSize()
+{
+  std::ptrdiff_t l1, l2, l3;
+  internal::manage_caching_sizes(GetAction, &l1, &l2, &l3);
+  return l3;
+}
+
 /** Set the cpu L1 and L2 cache sizes (in bytes).
   * These values are use to adjust the size of the blocks
   * for the algorithms working per blocks.
   *
   * \sa computeProductBlockingSizes */
-inline void setCpuCacheSizes(std::ptrdiff_t l1, std::ptrdiff_t l2)
+inline void setCpuCacheSizes(std::ptrdiff_t l1, std::ptrdiff_t l2, std::ptrdiff_t l3)
 {
-  internal::manage_caching_sizes(SetAction, &l1, &l2);
+  internal::manage_caching_sizes(SetAction, &l1, &l2, &l3);
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrix.h b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrix.h
index a36c7c7a61c..6440e1d09c7 100644
--- a/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrix.h
+++ b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrix.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_GENERAL_MATRIX_MATRIX_H
 #define EIGEN_GENERAL_MATRIX_MATRIX_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -23,7 +23,9 @@ template<
   typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs>
 struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef gebp_traits<RhsScalar,LhsScalar> Traits;
+
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   static EIGEN_STRONG_INLINE void run(
     Index rows, Index cols, Index depth,
     const LhsScalar* lhs, Index lhsStride,
@@ -51,42 +53,44 @@ template<
 struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor>
 {
 
-typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+typedef gebp_traits<LhsScalar,RhsScalar> Traits;
+
+typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 static void run(Index rows, Index cols, Index depth,
   const LhsScalar* _lhs, Index lhsStride,
   const RhsScalar* _rhs, Index rhsStride,
-  ResScalar* res, Index resStride,
+  ResScalar* _res, Index resStride,
   ResScalar alpha,
   level3_blocking<LhsScalar,RhsScalar>& blocking,
   GemmParallelInfo<Index>* info = 0)
 {
-  const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-  const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
-
-  typedef gebp_traits<LhsScalar,RhsScalar> Traits;
+  typedef const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> LhsMapper;
+  typedef const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> RhsMapper;
+  typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+  LhsMapper lhs(_lhs,lhsStride);
+  RhsMapper rhs(_rhs,rhsStride);
+  ResMapper res(_res, resStride);
 
   Index kc = blocking.kc();                   // cache block size along the K direction
   Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
-  //Index nc = blocking.nc(); // cache block size along the N direction
+  Index nc = (std::min)(cols,blocking.nc());  // cache block size along the N direction
 
-  gemm_pack_lhs<LhsScalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-  gemm_pack_rhs<RhsScalar, Index, Traits::nr, RhsStorageOrder> pack_rhs;
-  gebp_kernel<LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
+  gemm_pack_lhs<LhsScalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+  gemm_pack_rhs<RhsScalar, Index, RhsMapper, Traits::nr, RhsStorageOrder> pack_rhs;
+  gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
 
 #ifdef EIGEN_HAS_OPENMP
   if(info)
   {
     // this is the parallel version!
-    Index tid = omp_get_thread_num();
-    Index threads = omp_get_num_threads();
-    
-    std::size_t sizeA = kc*mc;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
-    ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, 0);
-    ei_declare_aligned_stack_constructed_variable(RhsScalar, w, sizeW, 0);
-    
-    RhsScalar* blockB = blocking.blockB();
-    eigen_internal_assert(blockB!=0);
+    int tid = omp_get_thread_num();
+    int threads = omp_get_num_threads();
+
+    LhsScalar* blockA = blocking.blockA();
+    eigen_internal_assert(blockA!=0);
+
+    std::size_t sizeB = kc*nc;
+    ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, 0);
 
     // For each horizontal panel of the rhs, and corresponding vertical panel of the lhs...
     for(Index k=0; k<depth; k+=kc)
@@ -94,60 +98,56 @@ static void run(Index rows, Index cols, Index depth,
       const Index actual_kc = (std::min)(k+kc,depth)-k; // => rows of B', and cols of the A'
 
       // In order to reduce the chance that a thread has to wait for the other,
-      // let's start by packing A'.
-      pack_lhs(blockA, &lhs(0,k), lhsStride, actual_kc, mc);
+      // let's start by packing B'.
+      pack_rhs(blockB, rhs.getSubMapper(k,0), actual_kc, nc);
 
-      // Pack B_k to B' in a parallel fashion:
-      // each thread packs the sub block B_k,j to B'_j where j is the thread id.
+      // Pack A_k to A' in a parallel fashion:
+      // each thread packs the sub block A_k,i to A'_i where i is the thread id.
 
-      // However, before copying to B'_j, we have to make sure that no other thread is still using it,
+      // However, before copying to A'_i, we have to make sure that no other thread is still using it,
       // i.e., we test that info[tid].users equals 0.
       // Then, we set info[tid].users to the number of threads to mark that all other threads are going to use it.
       while(info[tid].users!=0) {}
       info[tid].users += threads;
 
-      pack_rhs(blockB+info[tid].rhs_start*actual_kc, &rhs(k,info[tid].rhs_start), rhsStride, actual_kc, info[tid].rhs_length);
+      pack_lhs(blockA+info[tid].lhs_start*actual_kc, lhs.getSubMapper(info[tid].lhs_start,k), actual_kc, info[tid].lhs_length);
 
-      // Notify the other threads that the part B'_j is ready to go.
+      // Notify the other threads that the part A'_i is ready to go.
       info[tid].sync = k;
 
-      // Computes C_i += A' * B' per B'_j
-      for(Index shift=0; shift<threads; ++shift)
+      // Computes C_i += A' * B' per A'_i
+      for(int shift=0; shift<threads; ++shift)
       {
-        Index j = (tid+shift)%threads;
+        int i = (tid+shift)%threads;
 
-        // At this point we have to make sure that B'_j has been updated by the thread j,
+        // At this point we have to make sure that A'_i has been updated by the thread i,
         // we use testAndSetOrdered to mimic a volatile access.
         // However, no need to wait for the B' part which has been updated by the current thread!
-        if(shift>0)
-          while(info[j].sync!=k) {}
+        if (shift>0) {
+          while(info[i].sync!=k) {
+          }
+        }
 
-        gebp(res+info[j].rhs_start*resStride, resStride, blockA, blockB+info[j].rhs_start*actual_kc, mc, actual_kc, info[j].rhs_length, alpha, -1,-1,0,0, w);
+        gebp(res.getSubMapper(info[i].lhs_start, 0), blockA+info[i].lhs_start*actual_kc, blockB, info[i].lhs_length, actual_kc, nc, alpha);
       }
 
-      // Then keep going as usual with the remaining A'
-      for(Index i=mc; i<rows; i+=mc)
+      // Then keep going as usual with the remaining B'
+      for(Index j=nc; j<cols; j+=nc)
       {
-        const Index actual_mc = (std::min)(i+mc,rows)-i;
+        const Index actual_nc = (std::min)(j+nc,cols)-j;
 
-        // pack A_i,k to A'
-        pack_lhs(blockA, &lhs(i,k), lhsStride, actual_kc, actual_mc);
+        // pack B_k,j to B'
+        pack_rhs(blockB, rhs.getSubMapper(k,j), actual_kc, actual_nc);
 
-        // C_i += A' * B'
-        gebp(res+i, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha, -1,-1,0,0, w);
+        // C_j += A' * B'
+        gebp(res.getSubMapper(0, j), blockA, blockB, rows, actual_kc, actual_nc, alpha);
       }
 
-      // Release all the sub blocks B'_j of B' for the current thread,
+      // Release all the sub blocks A'_i of A' for the current thread,
       // i.e., we simply decrement the number of users by 1
-      for(Index j=0; j<threads; ++j)
-      {
+      for(Index i=0; i<threads; ++i)
         #pragma omp atomic
-#if defined(_MSC_VER) && _MSC_VER >= 1900
-		(info[j].users) -= 1;
-#else
-        --(info[j].users);
-#endif
-      }
+        info[i].users -= 1;
     }
   }
   else
@@ -157,38 +157,42 @@ static void run(Index rows, Index cols, Index depth,
 
     // this is the sequential version!
     std::size_t sizeA = kc*mc;
-    std::size_t sizeB = kc*cols;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
+    std::size_t sizeB = kc*nc;
 
     ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(RhsScalar, blockW, sizeW, blocking.blockW());
+
+    const bool pack_rhs_once = mc!=rows && kc==depth && nc==cols;
 
     // For each horizontal panel of the rhs, and corresponding panel of the lhs...
-    // (==GEMM_VAR1)
-    for(Index k2=0; k2<depth; k2+=kc)
+    for(Index i2=0; i2<rows; i2+=mc)
     {
-      const Index actual_kc = (std::min)(k2+kc,depth)-k2;
-
-      // OK, here we have selected one horizontal panel of rhs and one vertical panel of lhs.
-      // => Pack rhs's panel into a sequential chunk of memory (L2 caching)
-      // Note that this panel will be read as many times as the number of blocks in the lhs's
-      // vertical panel which is, in practice, a very low number.
-      pack_rhs(blockB, &rhs(k2,0), rhsStride, actual_kc, cols);
+      const Index actual_mc = (std::min)(i2+mc,rows)-i2;
 
-      // For each mc x kc block of the lhs's vertical panel...
-      // (==GEPP_VAR1)
-      for(Index i2=0; i2<rows; i2+=mc)
+      for(Index k2=0; k2<depth; k2+=kc)
       {
-        const Index actual_mc = (std::min)(i2+mc,rows)-i2;
-
-        // We pack the lhs's block into a sequential chunk of memory (L1 caching)
-        // Note that this block will be read a very high number of times, which is equal to the number of
-        // micro vertical panel of the large rhs's panel (e.g., cols/4 times).
-        pack_lhs(blockA, &lhs(i2,k2), lhsStride, actual_kc, actual_mc);
-
-        // Everything is packed, we can now call the block * panel kernel:
-        gebp(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha, -1, -1, 0, 0, blockW);
+        const Index actual_kc = (std::min)(k2+kc,depth)-k2;
+
+        // OK, here we have selected one horizontal panel of rhs and one vertical panel of lhs.
+        // => Pack lhs's panel into a sequential chunk of memory (L2/L3 caching)
+        // Note that this panel will be read as many times as the number of blocks in the rhs's
+        // horizontal panel which is, in practice, a very low number.
+        pack_lhs(blockA, lhs.getSubMapper(i2,k2), actual_kc, actual_mc);
+
+        // For each kc x nc block of the rhs's horizontal panel...
+        for(Index j2=0; j2<cols; j2+=nc)
+        {
+          const Index actual_nc = (std::min)(j2+nc,cols)-j2;
+
+          // We pack the rhs's block into a sequential chunk of memory (L2 caching)
+          // Note that this block will be read a very high number of times, which is equal to the number of
+          // micro horizontal panel of the large rhs's panel (e.g., rows/12 times).
+          if((!pack_rhs_once) || i2==0)
+            pack_rhs(blockB, rhs.getSubMapper(k2,j2), actual_kc, actual_nc);
+
+          // Everything is packed, we can now call the panel * block kernel:
+          gebp(res.getSubMapper(i2, j2), blockA, blockB, actual_mc, actual_kc, actual_nc, alpha);
+        }
       }
     }
   }
@@ -197,26 +201,21 @@ static void run(Index rows, Index cols, Index depth,
 };
 
 /*********************************************************************************
-*  Specialization of GeneralProduct<> for "large" GEMM, i.e.,
+*  Specialization of generic_product_impl for "large" GEMM, i.e.,
 *  implementation of the high level wrapper to general_matrix_matrix_product
 **********************************************************************************/
 
-template<typename Lhs, typename Rhs>
-struct traits<GeneralProduct<Lhs,Rhs,GemmProduct> >
- : traits<ProductBase<GeneralProduct<Lhs,Rhs,GemmProduct>, Lhs, Rhs> >
-{};
-
 template<typename Scalar, typename Index, typename Gemm, typename Lhs, typename Rhs, typename Dest, typename BlockingType>
 struct gemm_functor
 {
-  gemm_functor(const Lhs& lhs, const Rhs& rhs, Dest& dest, const Scalar& actualAlpha,
-                  BlockingType& blocking)
+  gemm_functor(const Lhs& lhs, const Rhs& rhs, Dest& dest, const Scalar& actualAlpha, BlockingType& blocking)
     : m_lhs(lhs), m_rhs(rhs), m_dest(dest), m_actualAlpha(actualAlpha), m_blocking(blocking)
   {}
 
-  void initParallelSession() const
+  void initParallelSession(Index num_threads) const
   {
-    m_blocking.allocateB();
+    m_blocking.initParallel(m_lhs.rows(), m_rhs.cols(), m_lhs.cols(), num_threads);
+    m_blocking.allocateA();
   }
 
   void operator() (Index row, Index rows, Index col=0, Index cols=-1, GemmParallelInfo<Index>* info=0) const
@@ -225,12 +224,14 @@ struct gemm_functor
       cols = m_rhs.cols();
 
     Gemm::run(rows, cols, m_lhs.cols(),
-              /*(const Scalar*)*/&m_lhs.coeffRef(row,0), m_lhs.outerStride(),
-              /*(const Scalar*)*/&m_rhs.coeffRef(0,col), m_rhs.outerStride(),
+              &m_lhs.coeffRef(row,0), m_lhs.outerStride(),
+              &m_rhs.coeffRef(0,col), m_rhs.outerStride(),
               (Scalar*)&(m_dest.coeffRef(row,col)), m_dest.outerStride(),
               m_actualAlpha, m_blocking, info);
   }
 
+  typedef typename Gemm::Traits Traits;
+
   protected:
     const Lhs& m_lhs;
     const Rhs& m_rhs;
@@ -251,29 +252,27 @@ class level3_blocking
   protected:
     LhsScalar* m_blockA;
     RhsScalar* m_blockB;
-    RhsScalar* m_blockW;
 
-    DenseIndex m_mc;
-    DenseIndex m_nc;
-    DenseIndex m_kc;
+    Index m_mc;
+    Index m_nc;
+    Index m_kc;
 
   public:
 
     level3_blocking()
-      : m_blockA(0), m_blockB(0), m_blockW(0), m_mc(0), m_nc(0), m_kc(0)
+      : m_blockA(0), m_blockB(0), m_mc(0), m_nc(0), m_kc(0)
     {}
 
-    inline DenseIndex mc() const { return m_mc; }
-    inline DenseIndex nc() const { return m_nc; }
-    inline DenseIndex kc() const { return m_kc; }
+    inline Index mc() const { return m_mc; }
+    inline Index nc() const { return m_nc; }
+    inline Index kc() const { return m_kc; }
 
     inline LhsScalar* blockA() { return m_blockA; }
     inline RhsScalar* blockB() { return m_blockB; }
-    inline RhsScalar* blockW() { return m_blockW; }
 };
 
 template<int StorageOrder, typename _LhsScalar, typename _RhsScalar, int MaxRows, int MaxCols, int MaxDepth, int KcFactor>
-class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, MaxDepth, KcFactor, true>
+class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, MaxDepth, KcFactor, true /* == FiniteAtCompileTime */>
   : public level3_blocking<
       typename conditional<StorageOrder==RowMajor,_RhsScalar,_LhsScalar>::type,
       typename conditional<StorageOrder==RowMajor,_LhsScalar,_RhsScalar>::type>
@@ -288,29 +287,38 @@ class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, M
     typedef gebp_traits<LhsScalar,RhsScalar> Traits;
     enum {
       SizeA = ActualRows * MaxDepth,
-      SizeB = ActualCols * MaxDepth,
-      SizeW = MaxDepth * Traits::WorkSpaceFactor
+      SizeB = ActualCols * MaxDepth
     };
 
-    EIGEN_ALIGN16 LhsScalar m_staticA[SizeA];
-    EIGEN_ALIGN16 RhsScalar m_staticB[SizeB];
-    EIGEN_ALIGN16 RhsScalar m_staticW[SizeW];
+#if EIGEN_MAX_STATIC_ALIGN_BYTES >= EIGEN_DEFAULT_ALIGN_BYTES
+    EIGEN_ALIGN_MAX LhsScalar m_staticA[SizeA];
+    EIGEN_ALIGN_MAX RhsScalar m_staticB[SizeB];
+#else
+    EIGEN_ALIGN_MAX char m_staticA[SizeA * sizeof(LhsScalar) + EIGEN_DEFAULT_ALIGN_BYTES-1];
+    EIGEN_ALIGN_MAX char m_staticB[SizeB * sizeof(RhsScalar) + EIGEN_DEFAULT_ALIGN_BYTES-1];
+#endif
 
   public:
 
-    gemm_blocking_space(DenseIndex /*rows*/, DenseIndex /*cols*/, DenseIndex /*depth*/)
+    gemm_blocking_space(Index /*rows*/, Index /*cols*/, Index /*depth*/, Index /*num_threads*/, bool /*full_rows = false*/)
     {
       this->m_mc = ActualRows;
       this->m_nc = ActualCols;
       this->m_kc = MaxDepth;
+#if EIGEN_MAX_STATIC_ALIGN_BYTES >= EIGEN_DEFAULT_ALIGN_BYTES
       this->m_blockA = m_staticA;
       this->m_blockB = m_staticB;
-      this->m_blockW = m_staticW;
+#else
+      this->m_blockA = reinterpret_cast<LhsScalar*>((internal::UIntPtr(m_staticA) + (EIGEN_DEFAULT_ALIGN_BYTES-1)) & ~std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1));
+      this->m_blockB = reinterpret_cast<RhsScalar*>((internal::UIntPtr(m_staticB) + (EIGEN_DEFAULT_ALIGN_BYTES-1)) & ~std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1));
+#endif
     }
 
+    void initParallel(Index, Index, Index, Index)
+    {}
+
     inline void allocateA() {}
     inline void allocateB() {}
-    inline void allocateW() {}
     inline void allocateAll() {}
 };
 
@@ -327,22 +335,42 @@ class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, M
     typedef typename conditional<Transpose,_LhsScalar,_RhsScalar>::type RhsScalar;
     typedef gebp_traits<LhsScalar,RhsScalar> Traits;
 
-    DenseIndex m_sizeA;
-    DenseIndex m_sizeB;
-    DenseIndex m_sizeW;
+    Index m_sizeA;
+    Index m_sizeB;
 
   public:
 
-    gemm_blocking_space(DenseIndex rows, DenseIndex cols, DenseIndex depth)
+    gemm_blocking_space(Index rows, Index cols, Index depth, Index num_threads, bool l3_blocking)
     {
       this->m_mc = Transpose ? cols : rows;
       this->m_nc = Transpose ? rows : cols;
       this->m_kc = depth;
 
-      computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, this->m_mc, this->m_nc);
+      if(l3_blocking)
+      {
+        computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, this->m_mc, this->m_nc, num_threads);
+      }
+      else  // no l3 blocking
+      {
+        Index n = this->m_nc;
+        computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, this->m_mc, n, num_threads);
+      }
+
+      m_sizeA = this->m_mc * this->m_kc;
+      m_sizeB = this->m_kc * this->m_nc;
+    }
+
+    void initParallel(Index rows, Index cols, Index depth, Index num_threads)
+    {
+      this->m_mc = Transpose ? cols : rows;
+      this->m_nc = Transpose ? rows : cols;
+      this->m_kc = depth;
+
+      eigen_internal_assert(this->m_blockA==0 && this->m_blockB==0);
+      Index m = this->m_mc;
+      computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, m, this->m_nc, num_threads);
       m_sizeA = this->m_mc * this->m_kc;
       m_sizeB = this->m_kc * this->m_nc;
-      m_sizeW = this->m_kc*Traits::WorkSpaceFactor;
     }
 
     void allocateA()
@@ -357,77 +385,108 @@ class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, M
         this->m_blockB = aligned_new<RhsScalar>(m_sizeB);
     }
 
-    void allocateW()
-    {
-      if(this->m_blockW==0)
-        this->m_blockW = aligned_new<RhsScalar>(m_sizeW);
-    }
-
     void allocateAll()
     {
       allocateA();
       allocateB();
-      allocateW();
     }
 
     ~gemm_blocking_space()
     {
       aligned_delete(this->m_blockA, m_sizeA);
       aligned_delete(this->m_blockB, m_sizeB);
-      aligned_delete(this->m_blockW, m_sizeW);
     }
 };
 
 } // end namespace internal
 
+namespace internal {
+
 template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, GemmProduct>
-  : public ProductBase<GeneralProduct<Lhs,Rhs,GemmProduct>, Lhs, Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct> >
 {
-    enum {
-      MaxDepthAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(Lhs::MaxColsAtCompileTime,Rhs::MaxRowsAtCompileTime)
-    };
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
-    
-    typedef typename  Lhs::Scalar LhsScalar;
-    typedef typename  Rhs::Scalar RhsScalar;
-    typedef           Scalar      ResScalar;
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  typedef typename Lhs::Scalar LhsScalar;
+  typedef typename Rhs::Scalar RhsScalar;
 
-    GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {
-      typedef internal::scalar_product_op<LhsScalar,RhsScalar> BinOp;
-      EIGEN_CHECK_BINARY_COMPATIBILIY(BinOp,LhsScalar,RhsScalar);
-    }
+  typedef internal::blas_traits<Lhs> LhsBlasTraits;
+  typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+  typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned;
 
-    template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
-    {
-      eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
+  typedef internal::blas_traits<Rhs> RhsBlasTraits;
+  typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+  typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
 
-      typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
-      typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
+  enum {
+    MaxDepthAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(Lhs::MaxColsAtCompileTime,Rhs::MaxRowsAtCompileTime)
+  };
 
-      Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
-                                 * RhsBlasTraits::extractScalarFactor(m_rhs);
+  typedef generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode> lazyproduct;
 
-      typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,LhsScalar,RhsScalar,
-              Dest::MaxRowsAtCompileTime,Dest::MaxColsAtCompileTime,MaxDepthAtCompileTime> BlockingType;
+  template<typename Dst>
+  static void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    if((rhs.rows()+dst.rows()+dst.cols())<20 && rhs.rows()>0)
+      lazyproduct::evalTo(dst, lhs, rhs);
+    else
+    {
+      dst.setZero();
+      scaleAndAddTo(dst, lhs, rhs, Scalar(1));
+    }
+  }
 
-      typedef internal::gemm_functor<
-        Scalar, Index,
-        internal::general_matrix_matrix_product<
-          Index,
-          LhsScalar, (_ActualLhsType::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(LhsBlasTraits::NeedToConjugate),
-          RhsScalar, (_ActualRhsType::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(RhsBlasTraits::NeedToConjugate),
-          (Dest::Flags&RowMajorBit) ? RowMajor : ColMajor>,
-        _ActualLhsType, _ActualRhsType, Dest, BlockingType> GemmFunctor;
+  template<typename Dst>
+  static void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    if((rhs.rows()+dst.rows()+dst.cols())<20 && rhs.rows()>0)
+      lazyproduct::addTo(dst, lhs, rhs);
+    else
+      scaleAndAddTo(dst,lhs, rhs, Scalar(1));
+  }
 
-      BlockingType blocking(dst.rows(), dst.cols(), lhs.cols());
+  template<typename Dst>
+  static void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    if((rhs.rows()+dst.rows()+dst.cols())<20 && rhs.rows()>0)
+      lazyproduct::subTo(dst, lhs, rhs);
+    else
+      scaleAndAddTo(dst, lhs, rhs, Scalar(-1));
+  }
 
-      internal::parallelize_gemm<(Dest::MaxRowsAtCompileTime>32 || Dest::MaxRowsAtCompileTime==Dynamic)>(GemmFunctor(lhs, rhs, dst, actualAlpha, blocking), this->rows(), this->cols(), Dest::Flags&RowMajorBit);
-    }
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& a_lhs, const Rhs& a_rhs, const Scalar& alpha)
+  {
+    eigen_assert(dst.rows()==a_lhs.rows() && dst.cols()==a_rhs.cols());
+    if(a_lhs.cols()==0 || a_lhs.rows()==0 || a_rhs.cols()==0)
+      return;
+
+    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
+
+    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
+                               * RhsBlasTraits::extractScalarFactor(a_rhs);
+
+    typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,LhsScalar,RhsScalar,
+            Dest::MaxRowsAtCompileTime,Dest::MaxColsAtCompileTime,MaxDepthAtCompileTime> BlockingType;
+
+    typedef internal::gemm_functor<
+      Scalar, Index,
+      internal::general_matrix_matrix_product<
+        Index,
+        LhsScalar, (ActualLhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(LhsBlasTraits::NeedToConjugate),
+        RhsScalar, (ActualRhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(RhsBlasTraits::NeedToConjugate),
+        (Dest::Flags&RowMajorBit) ? RowMajor : ColMajor>,
+      ActualLhsTypeCleaned, ActualRhsTypeCleaned, Dest, BlockingType> GemmFunctor;
+
+    BlockingType blocking(dst.rows(), dst.cols(), lhs.cols(), 1, true);
+    internal::parallelize_gemm<(Dest::MaxRowsAtCompileTime>32 || Dest::MaxRowsAtCompileTime==Dynamic)>
+        (GemmFunctor(lhs, rhs, dst, actualAlpha, blocking), a_lhs.rows(), a_rhs.cols(), a_lhs.cols(), Dest::Flags&RowMajorBit);
+  }
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_GENERAL_MATRIX_MATRIX_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h
index 5c37639091c..e844e37d16b 100644
--- a/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h
+++ b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h
@@ -20,7 +20,7 @@ namespace internal {
 /**********************************************************************
 * This file implements a general A * B product while
 * evaluating only one triangular part of the product.
-* This is more general version of self adjoint product (C += A A^T)
+* This is a more general version of self adjoint product (C += A A^T)
 * as the level 3 SYRK Blas routine.
 **********************************************************************/
 
@@ -40,15 +40,16 @@ template <typename Index, typename LhsScalar, int LhsStorageOrder, bool Conjugat
                           typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, int  UpLo, int Version>
 struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor,UpLo,Version>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* lhs, Index lhsStride,
-                                      const RhsScalar* rhs, Index rhsStride, ResScalar* res, Index resStride, const ResScalar& alpha)
+                                      const RhsScalar* rhs, Index rhsStride, ResScalar* res, Index resStride,
+                                      const ResScalar& alpha, level3_blocking<RhsScalar,LhsScalar>& blocking)
   {
     general_matrix_matrix_triangular_product<Index,
         RhsScalar, RhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateRhs,
         LhsScalar, LhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateLhs,
         ColMajor, UpLo==Lower?Upper:Lower>
-      ::run(size,depth,rhs,rhsStride,lhs,lhsStride,res,resStride,alpha);
+      ::run(size,depth,rhs,rhsStride,lhs,lhsStride,res,resStride,alpha,blocking);
   }
 };
 
@@ -56,32 +57,36 @@ template <typename Index, typename LhsScalar, int LhsStorageOrder, bool Conjugat
                           typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, int  UpLo, int Version>
 struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor,UpLo,Version>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* _lhs, Index lhsStride,
-                                      const RhsScalar* _rhs, Index rhsStride, ResScalar* res, Index resStride, const ResScalar& alpha)
+                                      const RhsScalar* _rhs, Index rhsStride, ResScalar* _res, Index resStride,
+                                      const ResScalar& alpha, level3_blocking<LhsScalar,RhsScalar>& blocking)
   {
-    const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-    const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
-
     typedef gebp_traits<LhsScalar,RhsScalar> Traits;
 
-    Index kc = depth; // cache block size along the K direction
-    Index mc = size;  // cache block size along the M direction
-    Index nc = size;  // cache block size along the N direction
-    computeProductBlockingSizes<LhsScalar,RhsScalar>(kc, mc, nc);
+    typedef const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> RhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    RhsMapper rhs(_rhs,rhsStride);
+    ResMapper res(_res, resStride);
+
+    Index kc = blocking.kc();
+    Index mc = (std::min)(size,blocking.mc());
+
     // !!! mc must be a multiple of nr:
     if(mc > Traits::nr)
       mc = (mc/Traits::nr)*Traits::nr;
 
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
-    std::size_t sizeB = sizeW + kc*size;
-    ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, kc*mc, 0);
-    ei_declare_aligned_stack_constructed_variable(RhsScalar, allocatedBlockB, sizeB, 0);
-    RhsScalar* blockB = allocatedBlockB + sizeW;
-    
-    gemm_pack_lhs<LhsScalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-    gemm_pack_rhs<RhsScalar, Index, Traits::nr, RhsStorageOrder> pack_rhs;
-    gebp_kernel <LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
+    std::size_t sizeA = kc*mc;
+    std::size_t sizeB = kc*size;
+
+    ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, blocking.blockA());
+    ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, blocking.blockB());
+
+    gemm_pack_lhs<LhsScalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+    gemm_pack_rhs<RhsScalar, Index, RhsMapper, Traits::nr, RhsStorageOrder> pack_rhs;
+    gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
     tribb_kernel<LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs, UpLo> sybb;
 
     for(Index k2=0; k2<depth; k2+=kc)
@@ -89,29 +94,30 @@ struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,
       const Index actual_kc = (std::min)(k2+kc,depth)-k2;
 
       // note that the actual rhs is the transpose/adjoint of mat
-      pack_rhs(blockB, &rhs(k2,0), rhsStride, actual_kc, size);
+      pack_rhs(blockB, rhs.getSubMapper(k2,0), actual_kc, size);
 
       for(Index i2=0; i2<size; i2+=mc)
       {
         const Index actual_mc = (std::min)(i2+mc,size)-i2;
 
-        pack_lhs(blockA, &lhs(i2, k2), lhsStride, actual_kc, actual_mc);
+        pack_lhs(blockA, lhs.getSubMapper(i2, k2), actual_kc, actual_mc);
 
         // the selected actual_mc * size panel of res is split into three different part:
         //  1 - before the diagonal => processed with gebp or skipped
         //  2 - the actual_mc x actual_mc symmetric block => processed with a special kernel
         //  3 - after the diagonal => processed with gebp or skipped
         if (UpLo==Lower)
-          gebp(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, (std::min)(size,i2), alpha,
-               -1, -1, 0, 0, allocatedBlockB);
+          gebp(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc,
+               (std::min)(size,i2), alpha, -1, -1, 0, 0);
+
 
-        sybb(res+resStride*i2 + i2, resStride, blockA, blockB + actual_kc*i2, actual_mc, actual_kc, alpha, allocatedBlockB);
+        sybb(_res+resStride*i2 + i2, resStride, blockA, blockB + actual_kc*i2, actual_mc, actual_kc, alpha);
 
         if (UpLo==Upper)
         {
           Index j2 = i2+actual_mc;
-          gebp(res+resStride*j2+i2, resStride, blockA, blockB+actual_kc*j2, actual_mc, actual_kc, (std::max)(Index(0), size-j2), alpha,
-               -1, -1, 0, 0, allocatedBlockB);
+          gebp(res.getSubMapper(i2, j2), blockA, blockB+actual_kc*j2, actual_mc,
+               actual_kc, (std::max)(Index(0), size-j2), alpha, -1, -1, 0, 0);
         }
       }
     }
@@ -132,14 +138,17 @@ struct tribb_kernel
 {
   typedef gebp_traits<LhsScalar,RhsScalar,ConjLhs,ConjRhs> Traits;
   typedef typename Traits::ResScalar ResScalar;
-  
+
   enum {
-    BlockSize  = EIGEN_PLAIN_ENUM_MAX(mr,nr)
+    BlockSize  = meta_least_common_multiple<EIGEN_PLAIN_ENUM_MAX(mr,nr),EIGEN_PLAIN_ENUM_MIN(mr,nr)>::ret
   };
-  void operator()(ResScalar* res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index size, Index depth, const ResScalar& alpha, RhsScalar* workspace)
+  void operator()(ResScalar* _res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index size, Index depth, const ResScalar& alpha)
   {
-    gebp_kernel<LhsScalar, RhsScalar, Index, mr, nr, ConjLhs, ConjRhs> gebp_kernel;
-    Matrix<ResScalar,BlockSize,BlockSize,ColMajor> buffer;
+    typedef blas_data_mapper<ResScalar, Index, ColMajor> ResMapper;
+    ResMapper res(_res, resStride);
+    gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, mr, nr, ConjLhs, ConjRhs> gebp_kernel;
+
+    Matrix<ResScalar,BlockSize,BlockSize,ColMajor> buffer((internal::constructor_without_unaligned_array_assert()));
 
     // let's process the block per panel of actual_mc x BlockSize,
     // again, each is split into three parts, etc.
@@ -149,20 +158,20 @@ struct tribb_kernel
       const RhsScalar* actual_b = blockB+j*depth;
 
       if(UpLo==Upper)
-        gebp_kernel(res+j*resStride, resStride, blockA, actual_b, j, depth, actualBlockSize, alpha,
-                    -1, -1, 0, 0, workspace);
+        gebp_kernel(res.getSubMapper(0, j), blockA, actual_b, j, depth, actualBlockSize, alpha,
+                    -1, -1, 0, 0);
 
       // selfadjoint micro block
       {
         Index i = j;
         buffer.setZero();
         // 1 - apply the kernel on the temporary buffer
-        gebp_kernel(buffer.data(), BlockSize, blockA+depth*i, actual_b, actualBlockSize, depth, actualBlockSize, alpha,
-                    -1, -1, 0, 0, workspace);
+        gebp_kernel(ResMapper(buffer.data(), BlockSize), blockA+depth*i, actual_b, actualBlockSize, depth, actualBlockSize, alpha,
+                    -1, -1, 0, 0);
         // 2 - triangular accumulation
         for(Index j1=0; j1<actualBlockSize; ++j1)
         {
-          ResScalar* r = res + (j+j1)*resStride + i;
+          ResScalar* r = &res(i, j + j1);
           for(Index i1=UpLo==Lower ? j1 : 0;
               UpLo==Lower ? i1<actualBlockSize : i1<=j1; ++i1)
             r[i1] += buffer(i1,j1);
@@ -172,8 +181,8 @@ struct tribb_kernel
       if(UpLo==Lower)
       {
         Index i = j+actualBlockSize;
-        gebp_kernel(res+j*resStride+i, resStride, blockA+depth*i, actual_b, size-i, depth, actualBlockSize, alpha,
-                    -1, -1, 0, 0, workspace);
+        gebp_kernel(res.getSubMapper(i, j), blockA+depth*i, actual_b, size-i, 
+                    depth, actualBlockSize, alpha, -1, -1, 0, 0);
       }
     }
   }
@@ -190,10 +199,9 @@ struct general_product_to_triangular_selector;
 template<typename MatrixType, typename ProductType, int UpLo>
 struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,true>
 {
-  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha)
+  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha, bool beta)
   {
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
     
     typedef typename internal::remove_all<typename ProductType::LhsNested>::type Lhs;
     typedef internal::blas_traits<Lhs> LhsBlasTraits;
@@ -209,6 +217,9 @@ struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,true>
 
     Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
 
+    if(!beta)
+      mat.template triangularView<UpLo>().setZero();
+
     enum {
       StorageOrder = (internal::traits<MatrixType>::Flags&RowMajorBit) ? RowMajor : ColMajor,
       UseLhsDirectly = _ActualLhs::InnerStrideAtCompileTime==1,
@@ -236,10 +247,8 @@ struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,true>
 template<typename MatrixType, typename ProductType, int UpLo>
 struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
 {
-  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha)
+  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha, bool beta)
   {
-    typedef typename MatrixType::Index Index;
-    
     typedef typename internal::remove_all<typename ProductType::LhsNested>::type Lhs;
     typedef internal::blas_traits<Lhs> LhsBlasTraits;
     typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhs;
@@ -254,23 +263,47 @@ struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
 
     typename ProductType::Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
 
+    if(!beta)
+      mat.template triangularView<UpLo>().setZero();
+
+    enum {
+      IsRowMajor = (internal::traits<MatrixType>::Flags&RowMajorBit) ? 1 : 0,
+      LhsIsRowMajor = _ActualLhs::Flags&RowMajorBit ? 1 : 0,
+      RhsIsRowMajor = _ActualRhs::Flags&RowMajorBit ? 1 : 0,
+      SkipDiag = (UpLo&(UnitDiag|ZeroDiag))!=0
+    };
+
+    Index size = mat.cols();
+    if(SkipDiag)
+      size--;
+    Index depth = actualLhs.cols();
+
+    typedef internal::gemm_blocking_space<IsRowMajor ? RowMajor : ColMajor,typename Lhs::Scalar,typename Rhs::Scalar,
+          MatrixType::MaxColsAtCompileTime, MatrixType::MaxColsAtCompileTime, _ActualRhs::MaxColsAtCompileTime> BlockingType;
+
+    BlockingType blocking(size, size, depth, 1, false);
+
     internal::general_matrix_matrix_triangular_product<Index,
-      typename Lhs::Scalar, _ActualLhs::Flags&RowMajorBit ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
-      typename Rhs::Scalar, _ActualRhs::Flags&RowMajorBit ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
-      MatrixType::Flags&RowMajorBit ? RowMajor : ColMajor, UpLo>
-      ::run(mat.cols(), actualLhs.cols(),
-            &actualLhs.coeffRef(0,0), actualLhs.outerStride(), &actualRhs.coeffRef(0,0), actualRhs.outerStride(),
-            mat.data(), mat.outerStride(), actualAlpha);
+      typename Lhs::Scalar, LhsIsRowMajor ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
+      typename Rhs::Scalar, RhsIsRowMajor ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
+      IsRowMajor ? RowMajor : ColMajor, UpLo&(Lower|Upper)>
+      ::run(size, depth,
+            &actualLhs.coeffRef(SkipDiag&&(UpLo&Lower)==Lower ? 1 : 0,0), actualLhs.outerStride(),
+            &actualRhs.coeffRef(0,SkipDiag&&(UpLo&Upper)==Upper ? 1 : 0), actualRhs.outerStride(),
+            mat.data() + (SkipDiag ? (bool(IsRowMajor) != ((UpLo&Lower)==Lower) ? 1 : mat.outerStride() ) : 0), mat.outerStride(), actualAlpha, blocking);
   }
 };
 
 template<typename MatrixType, unsigned int UpLo>
-template<typename ProductDerived, typename _Lhs, typename _Rhs>
-TriangularView<MatrixType,UpLo>& TriangularView<MatrixType,UpLo>::assignProduct(const ProductBase<ProductDerived, _Lhs,_Rhs>& prod, const Scalar& alpha)
+template<typename ProductType>
+TriangularView<MatrixType,UpLo>& TriangularViewImpl<MatrixType,UpLo,Dense>::_assignProduct(const ProductType& prod, const Scalar& alpha, bool beta)
 {
-  general_product_to_triangular_selector<MatrixType, ProductDerived, UpLo, (_Lhs::ColsAtCompileTime==1) || (_Rhs::RowsAtCompileTime==1)>::run(m_matrix.const_cast_derived(), prod.derived(), alpha);
+  EIGEN_STATIC_ASSERT((UpLo&UnitDiag)==0, WRITING_TO_TRIANGULAR_PART_WITH_UNIT_DIAGONAL_IS_NOT_SUPPORTED);
+  eigen_assert(derived().nestedExpression().rows() == prod.rows() && derived().cols() == prod.cols());
+  
+  general_product_to_triangular_selector<MatrixType, ProductType, UpLo, internal::traits<ProductType>::InnerSize==1>::run(derived().nestedExpression().const_cast_derived(), prod, alpha, beta);
   
-  return *this;
+  return derived();
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h
new file mode 100644
index 00000000000..f6f9ebecae7
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h
@@ -0,0 +1,145 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   Level 3 BLAS SYRK/HERK implementation.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
+#define EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
+
+namespace Eigen {
+
+namespace internal {
+
+template <typename Index, typename Scalar, int AStorageOrder, bool ConjugateA, int ResStorageOrder, int  UpLo>
+struct general_matrix_matrix_rankupdate :
+       general_matrix_matrix_triangular_product<
+         Index,Scalar,AStorageOrder,ConjugateA,Scalar,AStorageOrder,ConjugateA,ResStorageOrder,UpLo,BuiltIn> {};
+
+
+// try to go to BLAS specialization
+#define EIGEN_BLAS_RANKUPDATE_SPECIALIZE(Scalar) \
+template <typename Index, int LhsStorageOrder, bool ConjugateLhs, \
+                          int RhsStorageOrder, bool ConjugateRhs, int  UpLo> \
+struct general_matrix_matrix_triangular_product<Index,Scalar,LhsStorageOrder,ConjugateLhs, \
+               Scalar,RhsStorageOrder,ConjugateRhs,ColMajor,UpLo,Specialized> { \
+  static EIGEN_STRONG_INLINE void run(Index size, Index depth,const Scalar* lhs, Index lhsStride, \
+                          const Scalar* rhs, Index rhsStride, Scalar* res, Index resStride, Scalar alpha, level3_blocking<Scalar, Scalar>& blocking) \
+  { \
+    if ( lhs==rhs && ((UpLo&(Lower|Upper))==UpLo) ) { \
+      general_matrix_matrix_rankupdate<Index,Scalar,LhsStorageOrder,ConjugateLhs,ColMajor,UpLo> \
+      ::run(size,depth,lhs,lhsStride,rhs,rhsStride,res,resStride,alpha,blocking); \
+    } else { \
+      general_matrix_matrix_triangular_product<Index, \
+        Scalar, LhsStorageOrder, ConjugateLhs, \
+        Scalar, RhsStorageOrder, ConjugateRhs, \
+        ColMajor, UpLo, BuiltIn> \
+      ::run(size,depth,lhs,lhsStride,rhs,rhsStride,res,resStride,alpha,blocking); \
+    } \
+  } \
+};
+
+EIGEN_BLAS_RANKUPDATE_SPECIALIZE(double)
+EIGEN_BLAS_RANKUPDATE_SPECIALIZE(float)
+// TODO handle complex cases
+// EIGEN_BLAS_RANKUPDATE_SPECIALIZE(dcomplex)
+// EIGEN_BLAS_RANKUPDATE_SPECIALIZE(scomplex)
+
+// SYRK for float/double
+#define EIGEN_BLAS_RANKUPDATE_R(EIGTYPE, BLASTYPE, BLASFUNC) \
+template <typename Index, int AStorageOrder, bool ConjugateA, int  UpLo> \
+struct general_matrix_matrix_rankupdate<Index,EIGTYPE,AStorageOrder,ConjugateA,ColMajor,UpLo> { \
+  enum { \
+    IsLower = (UpLo&Lower) == Lower, \
+    LowUp = IsLower ? Lower : Upper, \
+    conjA = ((AStorageOrder==ColMajor) && ConjugateA) ? 1 : 0 \
+  }; \
+  static EIGEN_STRONG_INLINE void run(Index size, Index depth,const EIGTYPE* lhs, Index lhsStride, \
+                          const EIGTYPE* /*rhs*/, Index /*rhsStride*/, EIGTYPE* res, Index resStride, EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
+  { \
+  /* typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs;*/ \
+\
+   BlasIndex lda=convert_index<BlasIndex>(lhsStride), ldc=convert_index<BlasIndex>(resStride), n=convert_index<BlasIndex>(size), k=convert_index<BlasIndex>(depth); \
+   char uplo=((IsLower) ? 'L' : 'U'), trans=((AStorageOrder==RowMajor) ? 'T':'N'); \
+   EIGTYPE beta(1); \
+   BLASFUNC(&uplo, &trans, &n, &k, (const BLASTYPE*)&numext::real_ref(alpha), lhs, &lda, (const BLASTYPE*)&numext::real_ref(beta), res, &ldc); \
+  } \
+};
+
+// HERK for complex data
+#define EIGEN_BLAS_RANKUPDATE_C(EIGTYPE, BLASTYPE, RTYPE, BLASFUNC) \
+template <typename Index, int AStorageOrder, bool ConjugateA, int  UpLo> \
+struct general_matrix_matrix_rankupdate<Index,EIGTYPE,AStorageOrder,ConjugateA,ColMajor,UpLo> { \
+  enum { \
+    IsLower = (UpLo&Lower) == Lower, \
+    LowUp = IsLower ? Lower : Upper, \
+    conjA = (((AStorageOrder==ColMajor) && ConjugateA) || ((AStorageOrder==RowMajor) && !ConjugateA)) ? 1 : 0 \
+  }; \
+  static EIGEN_STRONG_INLINE void run(Index size, Index depth,const EIGTYPE* lhs, Index lhsStride, \
+                          const EIGTYPE* /*rhs*/, Index /*rhsStride*/, EIGTYPE* res, Index resStride, EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
+  { \
+   typedef Matrix<EIGTYPE, Dynamic, Dynamic, AStorageOrder> MatrixType; \
+\
+   BlasIndex lda=convert_index<BlasIndex>(lhsStride), ldc=convert_index<BlasIndex>(resStride), n=convert_index<BlasIndex>(size), k=convert_index<BlasIndex>(depth); \
+   char uplo=((IsLower) ? 'L' : 'U'), trans=((AStorageOrder==RowMajor) ? 'C':'N'); \
+   RTYPE alpha_, beta_; \
+   const EIGTYPE* a_ptr; \
+\
+   alpha_ = alpha.real(); \
+   beta_ = 1.0; \
+/* Copy with conjugation in some cases*/ \
+   MatrixType a; \
+   if (conjA) { \
+     Map<const MatrixType, 0, OuterStride<> > mapA(lhs,n,k,OuterStride<>(lhsStride)); \
+     a = mapA.conjugate(); \
+     lda = a.outerStride(); \
+     a_ptr = a.data(); \
+   } else a_ptr=lhs; \
+   BLASFUNC(&uplo, &trans, &n, &k, &alpha_, (BLASTYPE*)a_ptr, &lda, &beta_, (BLASTYPE*)res, &ldc); \
+  } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_RANKUPDATE_R(double, double, dsyrk)
+EIGEN_BLAS_RANKUPDATE_R(float,  float,  ssyrk)
+#else
+EIGEN_BLAS_RANKUPDATE_R(double, double, dsyrk_)
+EIGEN_BLAS_RANKUPDATE_R(float,  float,  ssyrk_)
+#endif
+
+// TODO hanlde complex cases
+// EIGEN_BLAS_RANKUPDATE_C(dcomplex, double, double, zherk_)
+// EIGEN_BLAS_RANKUPDATE_C(scomplex, float,  float, cherk_)
+
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h
new file mode 100644
index 00000000000..b0f6b0d5b9e
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h
@@ -0,0 +1,122 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   General matrix-matrix product functionality based on ?GEMM.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_GENERAL_MATRIX_MATRIX_BLAS_H
+#define EIGEN_GENERAL_MATRIX_MATRIX_BLAS_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/**********************************************************************
+* This file implements general matrix-matrix multiplication using BLAS
+* gemm function via partial specialization of
+* general_matrix_matrix_product::run(..) method for float, double,
+* std::complex<float> and std::complex<double> types
+**********************************************************************/
+
+// gemm specialization
+
+#define GEMM_SPECIALIZATION(EIGTYPE, EIGPREFIX, BLASTYPE, BLASFUNC) \
+template< \
+  typename Index, \
+  int LhsStorageOrder, bool ConjugateLhs, \
+  int RhsStorageOrder, bool ConjugateRhs> \
+struct general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor> \
+{ \
+typedef gebp_traits<EIGTYPE,EIGTYPE> Traits; \
+\
+static void run(Index rows, Index cols, Index depth, \
+  const EIGTYPE* _lhs, Index lhsStride, \
+  const EIGTYPE* _rhs, Index rhsStride, \
+  EIGTYPE* res, Index resStride, \
+  EIGTYPE alpha, \
+  level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/, \
+  GemmParallelInfo<Index>* /*info = 0*/) \
+{ \
+  using std::conj; \
+\
+  char transa, transb; \
+  BlasIndex m, n, k, lda, ldb, ldc; \
+  const EIGTYPE *a, *b; \
+  EIGTYPE beta(1); \
+  MatrixX##EIGPREFIX a_tmp, b_tmp; \
+\
+/* Set transpose options */ \
+  transa = (LhsStorageOrder==RowMajor) ? ((ConjugateLhs) ? 'C' : 'T') : 'N'; \
+  transb = (RhsStorageOrder==RowMajor) ? ((ConjugateRhs) ? 'C' : 'T') : 'N'; \
+\
+/* Set m, n, k */ \
+  m = convert_index<BlasIndex>(rows);  \
+  n = convert_index<BlasIndex>(cols);  \
+  k = convert_index<BlasIndex>(depth); \
+\
+/* Set lda, ldb, ldc */ \
+  lda = convert_index<BlasIndex>(lhsStride); \
+  ldb = convert_index<BlasIndex>(rhsStride); \
+  ldc = convert_index<BlasIndex>(resStride); \
+\
+/* Set a, b, c */ \
+  if ((LhsStorageOrder==ColMajor) && (ConjugateLhs)) { \
+    Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > lhs(_lhs,m,k,OuterStride<>(lhsStride)); \
+    a_tmp = lhs.conjugate(); \
+    a = a_tmp.data(); \
+    lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
+  } else a = _lhs; \
+\
+  if ((RhsStorageOrder==ColMajor) && (ConjugateRhs)) { \
+    Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > rhs(_rhs,k,n,OuterStride<>(rhsStride)); \
+    b_tmp = rhs.conjugate(); \
+    b = b_tmp.data(); \
+    ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
+  } else b = _rhs; \
+\
+  BLASFUNC(&transa, &transb, &m, &n, &k, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
+}};
+
+#ifdef EIGEN_USE_MKL
+GEMM_SPECIALIZATION(double,   d,  double, dgemm)
+GEMM_SPECIALIZATION(float,    f,  float,  sgemm)
+GEMM_SPECIALIZATION(dcomplex, cd, MKL_Complex16, zgemm)
+GEMM_SPECIALIZATION(scomplex, cf, MKL_Complex8,  cgemm)
+#else
+GEMM_SPECIALIZATION(double,   d,  double, dgemm_)
+GEMM_SPECIALIZATION(float,    f,  float,  sgemm_)
+GEMM_SPECIALIZATION(dcomplex, cd, double, zgemm_)
+GEMM_SPECIALIZATION(scomplex, cf, float,  cgemm_)
+#endif
+
+} // end namespase internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_GENERAL_MATRIX_MATRIX_BLAS_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixVector.h b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixVector.h
index 09387703efa..a597c1f4ee6 100644
--- a/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixVector.h
+++ b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixVector.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_GENERAL_MATRIX_VECTOR_H
 #define EIGEN_GENERAL_MATRIX_VECTOR_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -26,11 +26,39 @@ namespace internal {
  *  |real |cplx |real | alpha is converted to a cplx when calling the run function, no vectorization
  *  |cplx |real |cplx | invalid, the caller has to do tmp: = A * B; C += alpha*tmp
  *  |cplx |real |real | optimal case, vectorization possible via real-cplx mul
+ *
+ * Accesses to the matrix coefficients follow the following logic:
+ *
+ * - if all columns have the same alignment then
+ *   - if the columns have the same alignment as the result vector, then easy! (-> AllAligned case)
+ *   - otherwise perform unaligned loads only (-> NoneAligned case)
+ * - otherwise
+ *   - if even columns have the same alignment then
+ *     // odd columns are guaranteed to have the same alignment too
+ *     - if even or odd columns have the same alignment as the result, then
+ *       // for a register size of 2 scalars, this is guarantee to be the case (e.g., SSE with double)
+ *       - perform half aligned and half unaligned loads (-> EvenAligned case)
+ *     - otherwise perform unaligned loads only (-> NoneAligned case)
+ *   - otherwise, if the register size is 4 scalars (e.g., SSE with float) then
+ *     - one over 4 consecutive columns is guaranteed to be aligned with the result vector,
+ *       perform simple aligned loads for this column and aligned loads plus re-alignment for the other. (-> FirstAligned case)
+ *       // this re-alignment is done by the palign function implemented for SSE in Eigen/src/Core/arch/SSE/PacketMath.h
+ *   - otherwise,
+ *     // if we get here, this means the register size is greater than 4 (e.g., AVX with floats),
+ *     // we currently fall back to the NoneAligned case
+ *
+ * The same reasoning apply for the transposed case.
+ *
+ * The last case (PacketSize>4) could probably be improved by generalizing the FirstAligned case, but since we do not support AVX yet...
+ * One might also wonder why in the EvenAligned case we perform unaligned loads instead of using the aligned-loads plus re-alignment
+ * strategy as in the FirstAligned case. The reason is that we observed that unaligned loads on a 8 byte boundary are not too slow
+ * compared to unaligned loads on a 4 byte boundary.
+ *
  */
-template<typename Index, typename LhsScalar, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version>
-struct general_matrix_vector_product<Index,LhsScalar,ColMajor,ConjugateLhs,RhsScalar,ConjugateRhs,Version>
+template<typename Index, typename LhsScalar, typename LhsMapper, bool ConjugateLhs, typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version>
+struct general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,ConjugateLhs,RhsScalar,RhsMapper,ConjugateRhs,Version>
 {
-typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 
 enum {
   Vectorizable = packet_traits<LhsScalar>::Vectorizable && packet_traits<RhsScalar>::Vectorizable
@@ -50,31 +78,35 @@ typedef typename conditional<Vectorizable,_ResPacket,ResScalar>::type ResPacket;
 
 EIGEN_DONT_INLINE static void run(
   Index rows, Index cols,
-  const LhsScalar* lhs, Index lhsStride,
-  const RhsScalar* rhs, Index rhsIncr,
-  ResScalar* res, Index resIncr, RhsScalar alpha);
+  const LhsMapper& lhs,
+  const RhsMapper& rhs,
+        ResScalar* res, Index resIncr,
+  RhsScalar alpha);
 };
 
-template<typename Index, typename LhsScalar, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version>
-EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,ConjugateLhs,RhsScalar,ConjugateRhs,Version>::run(
+template<typename Index, typename LhsScalar, typename LhsMapper, bool ConjugateLhs, typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version>
+EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,ConjugateLhs,RhsScalar,RhsMapper,ConjugateRhs,Version>::run(
   Index rows, Index cols,
-  const LhsScalar* lhs, Index lhsStride,
-  const RhsScalar* rhs, Index rhsIncr,
-  ResScalar* res, Index resIncr, RhsScalar alpha)
+  const LhsMapper& lhs,
+  const RhsMapper& rhs,
+        ResScalar* res, Index resIncr,
+  RhsScalar alpha)
 {
-  EIGEN_UNUSED_VARIABLE(resIncr)
+  EIGEN_UNUSED_VARIABLE(resIncr);
   eigen_internal_assert(resIncr==1);
   #ifdef _EIGEN_ACCUMULATE_PACKETS
   #error _EIGEN_ACCUMULATE_PACKETS has already been defined
   #endif
-  #define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2) \
+  #define _EIGEN_ACCUMULATE_PACKETS(Alignment0,Alignment13,Alignment2) \
     pstore(&res[j], \
       padd(pload<ResPacket>(&res[j]), \
         padd( \
-          padd(pcj.pmul(EIGEN_CAT(ploa , A0)<LhsPacket>(&lhs0[j]),    ptmp0), \
-                  pcj.pmul(EIGEN_CAT(ploa , A13)<LhsPacket>(&lhs1[j]),   ptmp1)), \
-          padd(pcj.pmul(EIGEN_CAT(ploa , A2)<LhsPacket>(&lhs2[j]),    ptmp2), \
-                  pcj.pmul(EIGEN_CAT(ploa , A13)<LhsPacket>(&lhs3[j]),   ptmp3)) )))
+      padd(pcj.pmul(lhs0.template load<LhsPacket, Alignment0>(j),    ptmp0), \
+      pcj.pmul(lhs1.template load<LhsPacket, Alignment13>(j),   ptmp1)),   \
+      padd(pcj.pmul(lhs2.template load<LhsPacket, Alignment2>(j),    ptmp2), \
+      pcj.pmul(lhs3.template load<LhsPacket, Alignment13>(j),   ptmp3)) )))
+
+  typedef typename LhsMapper::VectorMapper LhsScalars;
 
   conj_helper<LhsScalar,RhsScalar,ConjugateLhs,ConjugateRhs> cj;
   conj_helper<LhsPacket,RhsPacket,ConjugateLhs,ConjugateRhs> pcj;
@@ -88,10 +120,12 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
   const Index ResPacketAlignedMask = ResPacketSize-1;
 //  const Index PeelAlignedMask = ResPacketSize*peels-1;
   const Index size = rows;
-  
+
+  const Index lhsStride = lhs.stride();
+
   // How many coeffs of the result do we have to skip to be aligned.
   // Here we assume data are at least aligned on the base scalar type.
-  Index alignedStart = internal::first_aligned(res,size);
+  Index alignedStart = internal::first_default_aligned(res,size);
   Index alignedSize = ResPacketSize>1 ? alignedStart + ((size-alignedStart) & ~ResPacketAlignedMask) : 0;
   const Index peeledSize = alignedSize - RhsPacketSize*peels - RhsPacketSize + 1;
 
@@ -101,19 +135,26 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
                        : FirstAligned;
 
   // we cannot assume the first element is aligned because of sub-matrices
-  const Index lhsAlignmentOffset = internal::first_aligned(lhs,size);
+  const Index lhsAlignmentOffset = lhs.firstAligned(size);
 
   // find how many columns do we have to skip to be aligned with the result (if possible)
   Index skipColumns = 0;
   // if the data cannot be aligned (TODO add some compile time tests when possible, e.g. for floats)
-  if( (size_t(lhs)%sizeof(LhsScalar)) || (size_t(res)%sizeof(ResScalar)) )
+  if( (lhsAlignmentOffset < 0) || (lhsAlignmentOffset == size) || (UIntPtr(res)%sizeof(ResScalar)) )
   {
     alignedSize = 0;
     alignedStart = 0;
+    alignmentPattern = NoneAligned;
+  }
+  else if(LhsPacketSize > 4)
+  {
+    // TODO: extend the code to support aligned loads whenever possible when LhsPacketSize > 4.
+    // Currently, it seems to be better to perform unaligned loads anyway
+    alignmentPattern = NoneAligned;
   }
   else if (LhsPacketSize>1)
   {
-    eigen_internal_assert(size_t(lhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0 || size<LhsPacketSize);
+  //    eigen_internal_assert(size_t(firstLhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0 || size<LhsPacketSize);
 
     while (skipColumns<LhsPacketSize &&
           alignedStart != ((lhsAlignmentOffset + alignmentStep*skipColumns)%LhsPacketSize))
@@ -130,10 +171,10 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
       // note that the skiped columns are processed later.
     }
 
-    eigen_internal_assert(  (alignmentPattern==NoneAligned)
+    /*    eigen_internal_assert(  (alignmentPattern==NoneAligned)
                       || (skipColumns + columnsAtOnce >= cols)
                       || LhsPacketSize > size
-                      || (size_t(lhs+alignedStart+lhsStride*skipColumns)%sizeof(LhsPacket))==0);
+                      || (size_t(firstLhs+alignedStart+lhsStride*skipColumns)%sizeof(LhsPacket))==0);*/
   }
   else if(Vectorizable)
   {
@@ -142,20 +183,20 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
     alignmentPattern = AllAligned;
   }
 
-  Index offset1 = (FirstAligned && alignmentStep==1?3:1);
-  Index offset3 = (FirstAligned && alignmentStep==1?1:3);
+  const Index offset1 = (alignmentPattern==FirstAligned && alignmentStep==1)?3:1;
+  const Index offset3 = (alignmentPattern==FirstAligned && alignmentStep==1)?1:3;
 
   Index columnBound = ((cols-skipColumns)/columnsAtOnce)*columnsAtOnce + skipColumns;
   for (Index i=skipColumns; i<columnBound; i+=columnsAtOnce)
   {
-    RhsPacket ptmp0 = pset1<RhsPacket>(alpha*rhs[i*rhsIncr]),
-              ptmp1 = pset1<RhsPacket>(alpha*rhs[(i+offset1)*rhsIncr]),
-              ptmp2 = pset1<RhsPacket>(alpha*rhs[(i+2)*rhsIncr]),
-              ptmp3 = pset1<RhsPacket>(alpha*rhs[(i+offset3)*rhsIncr]);
+    RhsPacket ptmp0 = pset1<RhsPacket>(alpha*rhs(i, 0)),
+              ptmp1 = pset1<RhsPacket>(alpha*rhs(i+offset1, 0)),
+              ptmp2 = pset1<RhsPacket>(alpha*rhs(i+2, 0)),
+              ptmp3 = pset1<RhsPacket>(alpha*rhs(i+offset3, 0));
 
     // this helps a lot generating better binary code
-    const LhsScalar *lhs0 = lhs + i*lhsStride,     *lhs1 = lhs + (i+offset1)*lhsStride,
-                    *lhs2 = lhs + (i+2)*lhsStride, *lhs3 = lhs + (i+offset3)*lhsStride;
+    const LhsScalars lhs0 = lhs.getVectorMapper(0, i+0),   lhs1 = lhs.getVectorMapper(0, i+offset1),
+                     lhs2 = lhs.getVectorMapper(0, i+2),   lhs3 = lhs.getVectorMapper(0, i+offset3);
 
     if (Vectorizable)
     {
@@ -163,10 +204,10 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
       // process initial unaligned coeffs
       for (Index j=0; j<alignedStart; ++j)
       {
-        res[j] = cj.pmadd(lhs0[j], pfirst(ptmp0), res[j]);
-        res[j] = cj.pmadd(lhs1[j], pfirst(ptmp1), res[j]);
-        res[j] = cj.pmadd(lhs2[j], pfirst(ptmp2), res[j]);
-        res[j] = cj.pmadd(lhs3[j], pfirst(ptmp3), res[j]);
+        res[j] = cj.pmadd(lhs0(j), pfirst(ptmp0), res[j]);
+        res[j] = cj.pmadd(lhs1(j), pfirst(ptmp1), res[j]);
+        res[j] = cj.pmadd(lhs2(j), pfirst(ptmp2), res[j]);
+        res[j] = cj.pmadd(lhs3(j), pfirst(ptmp3), res[j]);
       }
 
       if (alignedSize>alignedStart)
@@ -175,11 +216,11 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
         {
           case AllAligned:
             for (Index j = alignedStart; j<alignedSize; j+=ResPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,d,d);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Aligned,Aligned);
             break;
           case EvenAligned:
             for (Index j = alignedStart; j<alignedSize; j+=ResPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,d);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Unaligned,Aligned);
             break;
           case FirstAligned:
           {
@@ -189,28 +230,28 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
               LhsPacket A00, A01, A02, A03, A10, A11, A12, A13;
               ResPacket T0, T1;
 
-              A01 = pload<LhsPacket>(&lhs1[alignedStart-1]);
-              A02 = pload<LhsPacket>(&lhs2[alignedStart-2]);
-              A03 = pload<LhsPacket>(&lhs3[alignedStart-3]);
+              A01 = lhs1.template load<LhsPacket, Aligned>(alignedStart-1);
+              A02 = lhs2.template load<LhsPacket, Aligned>(alignedStart-2);
+              A03 = lhs3.template load<LhsPacket, Aligned>(alignedStart-3);
 
               for (; j<peeledSize; j+=peels*ResPacketSize)
               {
-                A11 = pload<LhsPacket>(&lhs1[j-1+LhsPacketSize]);  palign<1>(A01,A11);
-                A12 = pload<LhsPacket>(&lhs2[j-2+LhsPacketSize]);  palign<2>(A02,A12);
-                A13 = pload<LhsPacket>(&lhs3[j-3+LhsPacketSize]);  palign<3>(A03,A13);
+                A11 = lhs1.template load<LhsPacket, Aligned>(j-1+LhsPacketSize);  palign<1>(A01,A11);
+                A12 = lhs2.template load<LhsPacket, Aligned>(j-2+LhsPacketSize);  palign<2>(A02,A12);
+                A13 = lhs3.template load<LhsPacket, Aligned>(j-3+LhsPacketSize);  palign<3>(A03,A13);
 
-                A00 = pload<LhsPacket>(&lhs0[j]);
-                A10 = pload<LhsPacket>(&lhs0[j+LhsPacketSize]);
+                A00 = lhs0.template load<LhsPacket, Aligned>(j);
+                A10 = lhs0.template load<LhsPacket, Aligned>(j+LhsPacketSize);
                 T0  = pcj.pmadd(A00, ptmp0, pload<ResPacket>(&res[j]));
                 T1  = pcj.pmadd(A10, ptmp0, pload<ResPacket>(&res[j+ResPacketSize]));
 
                 T0  = pcj.pmadd(A01, ptmp1, T0);
-                A01 = pload<LhsPacket>(&lhs1[j-1+2*LhsPacketSize]);  palign<1>(A11,A01);
+                A01 = lhs1.template load<LhsPacket, Aligned>(j-1+2*LhsPacketSize);  palign<1>(A11,A01);
                 T0  = pcj.pmadd(A02, ptmp2, T0);
-                A02 = pload<LhsPacket>(&lhs2[j-2+2*LhsPacketSize]);  palign<2>(A12,A02);
+                A02 = lhs2.template load<LhsPacket, Aligned>(j-2+2*LhsPacketSize);  palign<2>(A12,A02);
                 T0  = pcj.pmadd(A03, ptmp3, T0);
                 pstore(&res[j],T0);
-                A03 = pload<LhsPacket>(&lhs3[j-3+2*LhsPacketSize]);  palign<3>(A13,A03);
+                A03 = lhs3.template load<LhsPacket, Aligned>(j-3+2*LhsPacketSize);  palign<3>(A13,A03);
                 T1  = pcj.pmadd(A11, ptmp1, T1);
                 T1  = pcj.pmadd(A12, ptmp2, T1);
                 T1  = pcj.pmadd(A13, ptmp3, T1);
@@ -218,12 +259,12 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
               }
             }
             for (; j<alignedSize; j+=ResPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,du);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Unaligned,Unaligned);
             break;
           }
           default:
             for (Index j = alignedStart; j<alignedSize; j+=ResPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(du,du,du);
+              _EIGEN_ACCUMULATE_PACKETS(Unaligned,Unaligned,Unaligned);
             break;
         }
       }
@@ -232,10 +273,10 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
     /* process remaining coeffs (or all if there is no explicit vectorization) */
     for (Index j=alignedSize; j<size; ++j)
     {
-      res[j] = cj.pmadd(lhs0[j], pfirst(ptmp0), res[j]);
-      res[j] = cj.pmadd(lhs1[j], pfirst(ptmp1), res[j]);
-      res[j] = cj.pmadd(lhs2[j], pfirst(ptmp2), res[j]);
-      res[j] = cj.pmadd(lhs3[j], pfirst(ptmp3), res[j]);
+      res[j] = cj.pmadd(lhs0(j), pfirst(ptmp0), res[j]);
+      res[j] = cj.pmadd(lhs1(j), pfirst(ptmp1), res[j]);
+      res[j] = cj.pmadd(lhs2(j), pfirst(ptmp2), res[j]);
+      res[j] = cj.pmadd(lhs3(j), pfirst(ptmp3), res[j]);
     }
   }
 
@@ -246,27 +287,27 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
   {
     for (Index k=start; k<end; ++k)
     {
-      RhsPacket ptmp0 = pset1<RhsPacket>(alpha*rhs[k*rhsIncr]);
-      const LhsScalar* lhs0 = lhs + k*lhsStride;
+      RhsPacket ptmp0 = pset1<RhsPacket>(alpha*rhs(k, 0));
+      const LhsScalars lhs0 = lhs.getVectorMapper(0, k);
 
       if (Vectorizable)
       {
         /* explicit vectorization */
         // process first unaligned result's coeffs
         for (Index j=0; j<alignedStart; ++j)
-          res[j] += cj.pmul(lhs0[j], pfirst(ptmp0));
+          res[j] += cj.pmul(lhs0(j), pfirst(ptmp0));
         // process aligned result's coeffs
-        if ((size_t(lhs0+alignedStart)%sizeof(LhsPacket))==0)
+        if (lhs0.template aligned<LhsPacket>(alignedStart))
           for (Index i = alignedStart;i<alignedSize;i+=ResPacketSize)
-            pstore(&res[i], pcj.pmadd(pload<LhsPacket>(&lhs0[i]), ptmp0, pload<ResPacket>(&res[i])));
+            pstore(&res[i], pcj.pmadd(lhs0.template load<LhsPacket, Aligned>(i), ptmp0, pload<ResPacket>(&res[i])));
         else
           for (Index i = alignedStart;i<alignedSize;i+=ResPacketSize)
-            pstore(&res[i], pcj.pmadd(ploadu<LhsPacket>(&lhs0[i]), ptmp0, pload<ResPacket>(&res[i])));
+            pstore(&res[i], pcj.pmadd(lhs0.template load<LhsPacket, Unaligned>(i), ptmp0, pload<ResPacket>(&res[i])));
       }
 
       // process remaining scalars (or all if no explicit vectorization)
       for (Index i=alignedSize; i<size; ++i)
-        res[i] += cj.pmul(lhs0[i], pfirst(ptmp0));
+        res[i] += cj.pmul(lhs0(i), pfirst(ptmp0));
     }
     if (skipColumns)
     {
@@ -290,10 +331,10 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,Co
  *  - alpha is always a complex (or converted to a complex)
  *  - no vectorization
  */
-template<typename Index, typename LhsScalar, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version>
-struct general_matrix_vector_product<Index,LhsScalar,RowMajor,ConjugateLhs,RhsScalar,ConjugateRhs,Version>
+template<typename Index, typename LhsScalar, typename LhsMapper, bool ConjugateLhs, typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version>
+struct general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,ConjugateLhs,RhsScalar,RhsMapper,ConjugateRhs,Version>
 {
-typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 
 enum {
   Vectorizable = packet_traits<LhsScalar>::Vectorizable && packet_traits<RhsScalar>::Vectorizable
@@ -310,73 +351,84 @@ typedef typename packet_traits<ResScalar>::type  _ResPacket;
 typedef typename conditional<Vectorizable,_LhsPacket,LhsScalar>::type LhsPacket;
 typedef typename conditional<Vectorizable,_RhsPacket,RhsScalar>::type RhsPacket;
 typedef typename conditional<Vectorizable,_ResPacket,ResScalar>::type ResPacket;
-  
+
 EIGEN_DONT_INLINE static void run(
   Index rows, Index cols,
-  const LhsScalar* lhs, Index lhsStride,
-  const RhsScalar* rhs, Index rhsIncr,
-  ResScalar* res, Index resIncr,
+  const LhsMapper& lhs,
+  const RhsMapper& rhs,
+        ResScalar* res, Index resIncr,
   ResScalar alpha);
 };
 
-template<typename Index, typename LhsScalar, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version>
-EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,ConjugateLhs,RhsScalar,ConjugateRhs,Version>::run(
+template<typename Index, typename LhsScalar, typename LhsMapper, bool ConjugateLhs, typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version>
+EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,ConjugateLhs,RhsScalar,RhsMapper,ConjugateRhs,Version>::run(
   Index rows, Index cols,
-  const LhsScalar* lhs, Index lhsStride,
-  const RhsScalar* rhs, Index rhsIncr,
+  const LhsMapper& lhs,
+  const RhsMapper& rhs,
   ResScalar* res, Index resIncr,
   ResScalar alpha)
 {
-  EIGEN_UNUSED_VARIABLE(rhsIncr);
-  eigen_internal_assert(rhsIncr==1);
+  eigen_internal_assert(rhs.stride()==1);
+
   #ifdef _EIGEN_ACCUMULATE_PACKETS
   #error _EIGEN_ACCUMULATE_PACKETS has already been defined
   #endif
 
-  #define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2) {\
-    RhsPacket b = pload<RhsPacket>(&rhs[j]); \
-    ptmp0 = pcj.pmadd(EIGEN_CAT(ploa,A0) <LhsPacket>(&lhs0[j]), b, ptmp0); \
-    ptmp1 = pcj.pmadd(EIGEN_CAT(ploa,A13)<LhsPacket>(&lhs1[j]), b, ptmp1); \
-    ptmp2 = pcj.pmadd(EIGEN_CAT(ploa,A2) <LhsPacket>(&lhs2[j]), b, ptmp2); \
-    ptmp3 = pcj.pmadd(EIGEN_CAT(ploa,A13)<LhsPacket>(&lhs3[j]), b, ptmp3); }
+  #define _EIGEN_ACCUMULATE_PACKETS(Alignment0,Alignment13,Alignment2) {\
+    RhsPacket b = rhs.getVectorMapper(j, 0).template load<RhsPacket, Aligned>(0);  \
+    ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Alignment0>(j), b, ptmp0); \
+    ptmp1 = pcj.pmadd(lhs1.template load<LhsPacket, Alignment13>(j), b, ptmp1); \
+    ptmp2 = pcj.pmadd(lhs2.template load<LhsPacket, Alignment2>(j), b, ptmp2); \
+    ptmp3 = pcj.pmadd(lhs3.template load<LhsPacket, Alignment13>(j), b, ptmp3); }
 
   conj_helper<LhsScalar,RhsScalar,ConjugateLhs,ConjugateRhs> cj;
   conj_helper<LhsPacket,RhsPacket,ConjugateLhs,ConjugateRhs> pcj;
 
+  typedef typename LhsMapper::VectorMapper LhsScalars;
+
   enum { AllAligned=0, EvenAligned=1, FirstAligned=2, NoneAligned=3 };
   const Index rowsAtOnce = 4;
   const Index peels = 2;
   const Index RhsPacketAlignedMask = RhsPacketSize-1;
   const Index LhsPacketAlignedMask = LhsPacketSize-1;
-//   const Index PeelAlignedMask = RhsPacketSize*peels-1;
   const Index depth = cols;
+  const Index lhsStride = lhs.stride();
 
   // How many coeffs of the result do we have to skip to be aligned.
   // Here we assume data are at least aligned on the base scalar type
   // if that's not the case then vectorization is discarded, see below.
-  Index alignedStart = internal::first_aligned(rhs, depth);
+  Index alignedStart = rhs.firstAligned(depth);
   Index alignedSize = RhsPacketSize>1 ? alignedStart + ((depth-alignedStart) & ~RhsPacketAlignedMask) : 0;
   const Index peeledSize = alignedSize - RhsPacketSize*peels - RhsPacketSize + 1;
 
   const Index alignmentStep = LhsPacketSize>1 ? (LhsPacketSize - lhsStride % LhsPacketSize) & LhsPacketAlignedMask : 0;
   Index alignmentPattern = alignmentStep==0 ? AllAligned
-                         : alignmentStep==(LhsPacketSize/2) ? EvenAligned
-                         : FirstAligned;
+                           : alignmentStep==(LhsPacketSize/2) ? EvenAligned
+                           : FirstAligned;
 
   // we cannot assume the first element is aligned because of sub-matrices
-  const Index lhsAlignmentOffset = internal::first_aligned(lhs,depth);
+  const Index lhsAlignmentOffset = lhs.firstAligned(depth);
+  const Index rhsAlignmentOffset = rhs.firstAligned(rows);
 
   // find how many rows do we have to skip to be aligned with rhs (if possible)
   Index skipRows = 0;
   // if the data cannot be aligned (TODO add some compile time tests when possible, e.g. for floats)
-  if( (sizeof(LhsScalar)!=sizeof(RhsScalar)) || (size_t(lhs)%sizeof(LhsScalar)) || (size_t(rhs)%sizeof(RhsScalar)) )
+  if( (sizeof(LhsScalar)!=sizeof(RhsScalar)) ||
+      (lhsAlignmentOffset < 0) || (lhsAlignmentOffset == depth) ||
+      (rhsAlignmentOffset < 0) || (rhsAlignmentOffset == rows) )
   {
     alignedSize = 0;
     alignedStart = 0;
+    alignmentPattern = NoneAligned;
+  }
+  else if(LhsPacketSize > 4)
+  {
+    // TODO: extend the code to support aligned loads whenever possible when LhsPacketSize > 4.
+    alignmentPattern = NoneAligned;
   }
   else if (LhsPacketSize>1)
   {
-    eigen_internal_assert(size_t(lhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0  || depth<LhsPacketSize);
+  //    eigen_internal_assert(size_t(firstLhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0  || depth<LhsPacketSize);
 
     while (skipRows<LhsPacketSize &&
            alignedStart != ((lhsAlignmentOffset + alignmentStep*skipRows)%LhsPacketSize))
@@ -392,11 +444,11 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co
       skipRows = (std::min)(skipRows,Index(rows));
       // note that the skiped columns are processed later.
     }
-    eigen_internal_assert(  alignmentPattern==NoneAligned
+    /*    eigen_internal_assert(  alignmentPattern==NoneAligned
                       || LhsPacketSize==1
                       || (skipRows + rowsAtOnce >= rows)
                       || LhsPacketSize > depth
-                      || (size_t(lhs+alignedStart+lhsStride*skipRows)%sizeof(LhsPacket))==0);
+                      || (size_t(firstLhs+alignedStart+lhsStride*skipRows)%sizeof(LhsPacket))==0);*/
   }
   else if(Vectorizable)
   {
@@ -405,18 +457,19 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co
     alignmentPattern = AllAligned;
   }
 
-  Index offset1 = (FirstAligned && alignmentStep==1?3:1);
-  Index offset3 = (FirstAligned && alignmentStep==1?1:3);
+  const Index offset1 = (alignmentPattern==FirstAligned && alignmentStep==1)?3:1;
+  const Index offset3 = (alignmentPattern==FirstAligned && alignmentStep==1)?1:3;
 
   Index rowBound = ((rows-skipRows)/rowsAtOnce)*rowsAtOnce + skipRows;
   for (Index i=skipRows; i<rowBound; i+=rowsAtOnce)
   {
-    EIGEN_ALIGN16 ResScalar tmp0 = ResScalar(0);
+    // FIXME: what is the purpose of this EIGEN_ALIGN_DEFAULT ??
+    EIGEN_ALIGN_MAX ResScalar tmp0 = ResScalar(0);
     ResScalar tmp1 = ResScalar(0), tmp2 = ResScalar(0), tmp3 = ResScalar(0);
 
     // this helps the compiler generating good binary code
-    const LhsScalar *lhs0 = lhs + i*lhsStride,     *lhs1 = lhs + (i+offset1)*lhsStride,
-                    *lhs2 = lhs + (i+2)*lhsStride, *lhs3 = lhs + (i+offset3)*lhsStride;
+    const LhsScalars lhs0 = lhs.getVectorMapper(i+0, 0),    lhs1 = lhs.getVectorMapper(i+offset1, 0),
+                     lhs2 = lhs.getVectorMapper(i+2, 0),    lhs3 = lhs.getVectorMapper(i+offset3, 0);
 
     if (Vectorizable)
     {
@@ -428,9 +481,9 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co
       // FIXME this loop get vectorized by the compiler !
       for (Index j=0; j<alignedStart; ++j)
       {
-        RhsScalar b = rhs[j];
-        tmp0 += cj.pmul(lhs0[j],b); tmp1 += cj.pmul(lhs1[j],b);
-        tmp2 += cj.pmul(lhs2[j],b); tmp3 += cj.pmul(lhs3[j],b);
+        RhsScalar b = rhs(j, 0);
+        tmp0 += cj.pmul(lhs0(j),b); tmp1 += cj.pmul(lhs1(j),b);
+        tmp2 += cj.pmul(lhs2(j),b); tmp3 += cj.pmul(lhs3(j),b);
       }
 
       if (alignedSize>alignedStart)
@@ -439,11 +492,11 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co
         {
           case AllAligned:
             for (Index j = alignedStart; j<alignedSize; j+=RhsPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,d,d);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Aligned,Aligned);
             break;
           case EvenAligned:
             for (Index j = alignedStart; j<alignedSize; j+=RhsPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,d);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Unaligned,Aligned);
             break;
           case FirstAligned:
           {
@@ -457,39 +510,39 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co
                * than basic unaligned loads.
                */
               LhsPacket A01, A02, A03, A11, A12, A13;
-              A01 = pload<LhsPacket>(&lhs1[alignedStart-1]);
-              A02 = pload<LhsPacket>(&lhs2[alignedStart-2]);
-              A03 = pload<LhsPacket>(&lhs3[alignedStart-3]);
+              A01 = lhs1.template load<LhsPacket, Aligned>(alignedStart-1);
+              A02 = lhs2.template load<LhsPacket, Aligned>(alignedStart-2);
+              A03 = lhs3.template load<LhsPacket, Aligned>(alignedStart-3);
 
               for (; j<peeledSize; j+=peels*RhsPacketSize)
               {
-                RhsPacket b = pload<RhsPacket>(&rhs[j]);
-                A11 = pload<LhsPacket>(&lhs1[j-1+LhsPacketSize]);  palign<1>(A01,A11);
-                A12 = pload<LhsPacket>(&lhs2[j-2+LhsPacketSize]);  palign<2>(A02,A12);
-                A13 = pload<LhsPacket>(&lhs3[j-3+LhsPacketSize]);  palign<3>(A03,A13);
+                RhsPacket b = rhs.getVectorMapper(j, 0).template load<RhsPacket, Aligned>(0);
+                A11 = lhs1.template load<LhsPacket, Aligned>(j-1+LhsPacketSize);  palign<1>(A01,A11);
+                A12 = lhs2.template load<LhsPacket, Aligned>(j-2+LhsPacketSize);  palign<2>(A02,A12);
+                A13 = lhs3.template load<LhsPacket, Aligned>(j-3+LhsPacketSize);  palign<3>(A03,A13);
 
-                ptmp0 = pcj.pmadd(pload<LhsPacket>(&lhs0[j]), b, ptmp0);
+                ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Aligned>(j), b, ptmp0);
                 ptmp1 = pcj.pmadd(A01, b, ptmp1);
-                A01 = pload<LhsPacket>(&lhs1[j-1+2*LhsPacketSize]);  palign<1>(A11,A01);
+                A01 = lhs1.template load<LhsPacket, Aligned>(j-1+2*LhsPacketSize);  palign<1>(A11,A01);
                 ptmp2 = pcj.pmadd(A02, b, ptmp2);
-                A02 = pload<LhsPacket>(&lhs2[j-2+2*LhsPacketSize]);  palign<2>(A12,A02);
+                A02 = lhs2.template load<LhsPacket, Aligned>(j-2+2*LhsPacketSize);  palign<2>(A12,A02);
                 ptmp3 = pcj.pmadd(A03, b, ptmp3);
-                A03 = pload<LhsPacket>(&lhs3[j-3+2*LhsPacketSize]);  palign<3>(A13,A03);
+                A03 = lhs3.template load<LhsPacket, Aligned>(j-3+2*LhsPacketSize);  palign<3>(A13,A03);
 
-                b = pload<RhsPacket>(&rhs[j+RhsPacketSize]);
-                ptmp0 = pcj.pmadd(pload<LhsPacket>(&lhs0[j+LhsPacketSize]), b, ptmp0);
+                b = rhs.getVectorMapper(j+RhsPacketSize, 0).template load<RhsPacket, Aligned>(0);
+                ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Aligned>(j+LhsPacketSize), b, ptmp0);
                 ptmp1 = pcj.pmadd(A11, b, ptmp1);
                 ptmp2 = pcj.pmadd(A12, b, ptmp2);
                 ptmp3 = pcj.pmadd(A13, b, ptmp3);
               }
             }
             for (; j<alignedSize; j+=RhsPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,du);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Unaligned,Unaligned);
             break;
           }
           default:
             for (Index j = alignedStart; j<alignedSize; j+=RhsPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(du,du,du);
+              _EIGEN_ACCUMULATE_PACKETS(Unaligned,Unaligned,Unaligned);
             break;
         }
         tmp0 += predux(ptmp0);
@@ -503,9 +556,9 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co
     // FIXME this loop get vectorized by the compiler !
     for (Index j=alignedSize; j<depth; ++j)
     {
-      RhsScalar b = rhs[j];
-      tmp0 += cj.pmul(lhs0[j],b); tmp1 += cj.pmul(lhs1[j],b);
-      tmp2 += cj.pmul(lhs2[j],b); tmp3 += cj.pmul(lhs3[j],b);
+      RhsScalar b = rhs(j, 0);
+      tmp0 += cj.pmul(lhs0(j),b); tmp1 += cj.pmul(lhs1(j),b);
+      tmp2 += cj.pmul(lhs2(j),b); tmp3 += cj.pmul(lhs3(j),b);
     }
     res[i*resIncr]            += alpha*tmp0;
     res[(i+offset1)*resIncr]  += alpha*tmp1;
@@ -520,30 +573,30 @@ EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,Co
   {
     for (Index i=start; i<end; ++i)
     {
-      EIGEN_ALIGN16 ResScalar tmp0 = ResScalar(0);
+      EIGEN_ALIGN_MAX ResScalar tmp0 = ResScalar(0);
       ResPacket ptmp0 = pset1<ResPacket>(tmp0);
-      const LhsScalar* lhs0 = lhs + i*lhsStride;
+      const LhsScalars lhs0 = lhs.getVectorMapper(i, 0);
       // process first unaligned result's coeffs
       // FIXME this loop get vectorized by the compiler !
       for (Index j=0; j<alignedStart; ++j)
-        tmp0 += cj.pmul(lhs0[j], rhs[j]);
+        tmp0 += cj.pmul(lhs0(j), rhs(j, 0));
 
       if (alignedSize>alignedStart)
       {
         // process aligned rhs coeffs
-        if ((size_t(lhs0+alignedStart)%sizeof(LhsPacket))==0)
+        if (lhs0.template aligned<LhsPacket>(alignedStart))
           for (Index j = alignedStart;j<alignedSize;j+=RhsPacketSize)
-            ptmp0 = pcj.pmadd(pload<LhsPacket>(&lhs0[j]), pload<RhsPacket>(&rhs[j]), ptmp0);
+            ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Aligned>(j), rhs.getVectorMapper(j, 0).template load<RhsPacket, Aligned>(0), ptmp0);
         else
           for (Index j = alignedStart;j<alignedSize;j+=RhsPacketSize)
-            ptmp0 = pcj.pmadd(ploadu<LhsPacket>(&lhs0[j]), pload<RhsPacket>(&rhs[j]), ptmp0);
+            ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Unaligned>(j), rhs.getVectorMapper(j, 0).template load<RhsPacket, Aligned>(0), ptmp0);
         tmp0 += predux(ptmp0);
       }
 
       // process remaining scalars
       // FIXME this loop get vectorized by the compiler !
       for (Index j=alignedSize; j<depth; ++j)
-        tmp0 += cj.pmul(lhs0[j], rhs[j]);
+        tmp0 += cj.pmul(lhs0(j), rhs(j, 0));
       res[i*resIncr] += alpha*tmp0;
     }
     if (skipRows)
diff --git a/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixVector_BLAS.h b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixVector_BLAS.h
new file mode 100644
index 00000000000..6e36c2b3c32
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/products/GeneralMatrixVector_BLAS.h
@@ -0,0 +1,136 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   General matrix-vector product functionality based on ?GEMV.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_GENERAL_MATRIX_VECTOR_BLAS_H
+#define EIGEN_GENERAL_MATRIX_VECTOR_BLAS_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/**********************************************************************
+* This file implements general matrix-vector multiplication using BLAS
+* gemv function via partial specialization of
+* general_matrix_vector_product::run(..) method for float, double,
+* std::complex<float> and std::complex<double> types
+**********************************************************************/
+
+// gemv specialization
+
+template<typename Index, typename LhsScalar, int StorageOrder, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs>
+struct general_matrix_vector_product_gemv;
+
+#define EIGEN_BLAS_GEMV_SPECIALIZE(Scalar) \
+template<typename Index, bool ConjugateLhs, bool ConjugateRhs> \
+struct general_matrix_vector_product<Index,Scalar,const_blas_data_mapper<Scalar,Index,ColMajor>,ColMajor,ConjugateLhs,Scalar,const_blas_data_mapper<Scalar,Index,RowMajor>,ConjugateRhs,Specialized> { \
+static void run( \
+  Index rows, Index cols, \
+  const const_blas_data_mapper<Scalar,Index,ColMajor> &lhs, \
+  const const_blas_data_mapper<Scalar,Index,RowMajor> &rhs, \
+  Scalar* res, Index resIncr, Scalar alpha) \
+{ \
+  if (ConjugateLhs) { \
+    general_matrix_vector_product<Index,Scalar,const_blas_data_mapper<Scalar,Index,ColMajor>,ColMajor,ConjugateLhs,Scalar,const_blas_data_mapper<Scalar,Index,RowMajor>,ConjugateRhs,BuiltIn>::run( \
+      rows, cols, lhs, rhs, res, resIncr, alpha); \
+  } else { \
+    general_matrix_vector_product_gemv<Index,Scalar,ColMajor,ConjugateLhs,Scalar,ConjugateRhs>::run( \
+      rows, cols, lhs.data(), lhs.stride(), rhs.data(), rhs.stride(), res, resIncr, alpha); \
+  } \
+} \
+}; \
+template<typename Index, bool ConjugateLhs, bool ConjugateRhs> \
+struct general_matrix_vector_product<Index,Scalar,const_blas_data_mapper<Scalar,Index,RowMajor>,RowMajor,ConjugateLhs,Scalar,const_blas_data_mapper<Scalar,Index,ColMajor>,ConjugateRhs,Specialized> { \
+static void run( \
+  Index rows, Index cols, \
+  const const_blas_data_mapper<Scalar,Index,RowMajor> &lhs, \
+  const const_blas_data_mapper<Scalar,Index,ColMajor> &rhs, \
+  Scalar* res, Index resIncr, Scalar alpha) \
+{ \
+    general_matrix_vector_product_gemv<Index,Scalar,RowMajor,ConjugateLhs,Scalar,ConjugateRhs>::run( \
+      rows, cols, lhs.data(), lhs.stride(), rhs.data(), rhs.stride(), res, resIncr, alpha); \
+} \
+}; \
+
+EIGEN_BLAS_GEMV_SPECIALIZE(double)
+EIGEN_BLAS_GEMV_SPECIALIZE(float)
+EIGEN_BLAS_GEMV_SPECIALIZE(dcomplex)
+EIGEN_BLAS_GEMV_SPECIALIZE(scomplex)
+
+#define EIGEN_BLAS_GEMV_SPECIALIZATION(EIGTYPE,BLASTYPE,BLASFUNC) \
+template<typename Index, int LhsStorageOrder, bool ConjugateLhs, bool ConjugateRhs> \
+struct general_matrix_vector_product_gemv<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,ConjugateRhs> \
+{ \
+typedef Matrix<EIGTYPE,Dynamic,1,ColMajor> GEMVVector;\
+\
+static void run( \
+  Index rows, Index cols, \
+  const EIGTYPE* lhs, Index lhsStride, \
+  const EIGTYPE* rhs, Index rhsIncr, \
+  EIGTYPE* res, Index resIncr, EIGTYPE alpha) \
+{ \
+  BlasIndex m=convert_index<BlasIndex>(rows), n=convert_index<BlasIndex>(cols), \
+            lda=convert_index<BlasIndex>(lhsStride), incx=convert_index<BlasIndex>(rhsIncr), incy=convert_index<BlasIndex>(resIncr); \
+  const EIGTYPE beta(1); \
+  const EIGTYPE *x_ptr; \
+  char trans=(LhsStorageOrder==ColMajor) ? 'N' : (ConjugateLhs) ? 'C' : 'T'; \
+  if (LhsStorageOrder==RowMajor) { \
+    m = convert_index<BlasIndex>(cols); \
+    n = convert_index<BlasIndex>(rows); \
+  }\
+  GEMVVector x_tmp; \
+  if (ConjugateRhs) { \
+    Map<const GEMVVector, 0, InnerStride<> > map_x(rhs,cols,1,InnerStride<>(incx)); \
+    x_tmp=map_x.conjugate(); \
+    x_ptr=x_tmp.data(); \
+    incx=1; \
+  } else x_ptr=rhs; \
+  BLASFUNC(&trans, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)lhs, &lda, (const BLASTYPE*)x_ptr, &incx, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &incy); \
+}\
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_GEMV_SPECIALIZATION(double,   double, dgemv)
+EIGEN_BLAS_GEMV_SPECIALIZATION(float,    float,  sgemv)
+EIGEN_BLAS_GEMV_SPECIALIZATION(dcomplex, MKL_Complex16, zgemv)
+EIGEN_BLAS_GEMV_SPECIALIZATION(scomplex, MKL_Complex8 , cgemv)
+#else
+EIGEN_BLAS_GEMV_SPECIALIZATION(double,   double, dgemv_)
+EIGEN_BLAS_GEMV_SPECIALIZATION(float,    float,  sgemv_)
+EIGEN_BLAS_GEMV_SPECIALIZATION(dcomplex, double, zgemv_)
+EIGEN_BLAS_GEMV_SPECIALIZATION(scomplex, float,  cgemv_)
+#endif
+
+} // end namespase internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_GENERAL_MATRIX_VECTOR_BLAS_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/Parallelizer.h b/extern/Eigen3/Eigen/src/Core/products/Parallelizer.h
index 6937ee33284..c2f084c82cc 100644
--- a/extern/Eigen3/Eigen/src/Core/products/Parallelizer.h
+++ b/extern/Eigen3/Eigen/src/Core/products/Parallelizer.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_PARALLELIZER_H
 #define EIGEN_PARALLELIZER_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -49,8 +49,8 @@ inline void initParallel()
 {
   int nbt;
   internal::manage_multi_threading(GetAction, &nbt);
-  std::ptrdiff_t l1, l2;
-  internal::manage_caching_sizes(GetAction, &l1, &l2);
+  std::ptrdiff_t l1, l2, l3;
+  internal::manage_caching_sizes(GetAction, &l1, &l2, &l3);
 }
 
 /** \returns the max number of threads reserved for Eigen
@@ -73,17 +73,17 @@ namespace internal {
 
 template<typename Index> struct GemmParallelInfo
 {
-  GemmParallelInfo() : sync(-1), users(0), rhs_start(0), rhs_length(0) {}
+  GemmParallelInfo() : sync(-1), users(0), lhs_start(0), lhs_length(0) {}
 
-  int volatile sync;
+  Index volatile sync;
   int volatile users;
 
-  Index rhs_start;
-  Index rhs_length;
+  Index lhs_start;
+  Index lhs_length;
 };
 
 template<bool Condition, typename Functor, typename Index>
-void parallelize_gemm(const Functor& func, Index rows, Index cols, bool transpose)
+void parallelize_gemm(const Functor& func, Index rows, Index cols, Index depth, bool transpose)
 {
   // TODO when EIGEN_USE_BLAS is defined,
   // we should still enable OMP for other scalar types
@@ -92,6 +92,7 @@ void parallelize_gemm(const Functor& func, Index rows, Index cols, bool transpos
   // the matrix product when multithreading is enabled. This is a temporary
   // fix to support row-major destination matrices. This whole
   // parallelizer mechanism has to be redisigned anyway.
+  EIGEN_UNUSED_VARIABLE(depth);
   EIGEN_UNUSED_VARIABLE(transpose);
   func(0,rows, 0,cols);
 #else
@@ -102,56 +103,56 @@ void parallelize_gemm(const Functor& func, Index rows, Index cols, bool transpos
   // - we are not already in a parallel code
   // - the sizes are large enough
 
-  // 1- are we already in a parallel session?
-  // FIXME omp_get_num_threads()>1 only works for openmp, what if the user does not use openmp?
-  if((!Condition) || (omp_get_num_threads()>1))
-    return func(0,rows, 0,cols);
+  // compute the maximal number of threads from the size of the product:
+  // This first heuristic takes into account that the product kernel is fully optimized when working with nr columns at once.
+  Index size = transpose ? rows : cols;
+  Index pb_max_threads = std::max<Index>(1,size / Functor::Traits::nr);
 
-  Index size = transpose ? cols : rows;
+  // compute the maximal number of threads from the total amount of work:
+  double work = static_cast<double>(rows) * static_cast<double>(cols) *
+      static_cast<double>(depth);
+  double kMinTaskSize = 50000;  // FIXME improve this heuristic.
+  pb_max_threads = std::max<Index>(1, std::min<Index>(pb_max_threads, work / kMinTaskSize));
 
-  // 2- compute the maximal number of threads from the size of the product:
-  // FIXME this has to be fine tuned
-  Index max_threads = std::max<Index>(1,size / 32);
+  // compute the number of threads we are going to use
+  Index threads = std::min<Index>(nbThreads(), pb_max_threads);
 
-  // 3 - compute the number of threads we are going to use
-  Index threads = std::min<Index>(nbThreads(), max_threads);
-
-  if(threads==1)
+  // if multi-threading is explicitely disabled, not useful, or if we already are in a parallel session,
+  // then abort multi-threading
+  // FIXME omp_get_num_threads()>1 only works for openmp, what if the user does not use openmp?
+  if((!Condition) || (threads==1) || (omp_get_num_threads()>1))
     return func(0,rows, 0,cols);
 
   Eigen::initParallel();
-  func.initParallelSession();
+  func.initParallelSession(threads);
 
   if(transpose)
     std::swap(rows,cols);
 
-  GemmParallelInfo<Index>* info = new GemmParallelInfo<Index>[threads];
+  ei_declare_aligned_stack_constructed_variable(GemmParallelInfo<Index>,info,threads,0);
 
   #pragma omp parallel num_threads(threads)
   {
     Index i = omp_get_thread_num();
     // Note that the actual number of threads might be lower than the number of request ones.
     Index actual_threads = omp_get_num_threads();
-    
+
     Index blockCols = (cols / actual_threads) & ~Index(0x3);
-    Index blockRows = (rows / actual_threads) & ~Index(0x7);
-    
+    Index blockRows = (rows / actual_threads);
+    blockRows = (blockRows/Functor::Traits::mr)*Functor::Traits::mr;
+
     Index r0 = i*blockRows;
     Index actualBlockRows = (i+1==actual_threads) ? rows-r0 : blockRows;
 
     Index c0 = i*blockCols;
     Index actualBlockCols = (i+1==actual_threads) ? cols-c0 : blockCols;
 
-    info[i].rhs_start = c0;
-    info[i].rhs_length = actualBlockCols;
+    info[i].lhs_start = r0;
+    info[i].lhs_length = actualBlockRows;
 
-    if(transpose)
-      func(0, cols, r0, actualBlockRows, info);
-    else
-      func(r0, actualBlockRows, 0,cols, info);
+    if(transpose) func(c0, actualBlockCols, 0, rows, info);
+    else          func(0, rows, c0, actualBlockCols, info);
   }
-
-  delete[] info;
 #endif
 }
 
diff --git a/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixMatrix.h b/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixMatrix.h
index 99cf9e0ae27..da6f82abcd7 100644
--- a/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixMatrix.h
+++ b/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixMatrix.h
@@ -15,7 +15,7 @@ namespace Eigen {
 namespace internal {
 
 // pack a selfadjoint block diagonal for use with the gebp_kernel
-template<typename Scalar, typename Index, int Pack1, int Pack2, int StorageOrder>
+template<typename Scalar, typename Index, int Pack1, int Pack2_dummy, int StorageOrder>
 struct symm_pack_lhs
 {
   template<int BlockRows> inline
@@ -45,25 +45,32 @@ struct symm_pack_lhs
   }
   void operator()(Scalar* blockA, const Scalar* _lhs, Index lhsStride, Index cols, Index rows)
   {
+    enum { PacketSize = packet_traits<Scalar>::size };
     const_blas_data_mapper<Scalar,Index,StorageOrder> lhs(_lhs,lhsStride);
     Index count = 0;
-    Index peeled_mc = (rows/Pack1)*Pack1;
-    for(Index i=0; i<peeled_mc; i+=Pack1)
-    {
-      pack<Pack1>(blockA, lhs, cols, i, count);
-    }
-
-    if(rows-peeled_mc>=Pack2)
-    {
-      pack<Pack2>(blockA, lhs, cols, peeled_mc, count);
-      peeled_mc += Pack2;
-    }
+    //Index peeled_mc3 = (rows/Pack1)*Pack1;
+    
+    const Index peeled_mc3 = Pack1>=3*PacketSize ? (rows/(3*PacketSize))*(3*PacketSize) : 0;
+    const Index peeled_mc2 = Pack1>=2*PacketSize ? peeled_mc3+((rows-peeled_mc3)/(2*PacketSize))*(2*PacketSize) : 0;
+    const Index peeled_mc1 = Pack1>=1*PacketSize ? (rows/(1*PacketSize))*(1*PacketSize) : 0;
+    
+    if(Pack1>=3*PacketSize)
+      for(Index i=0; i<peeled_mc3; i+=3*PacketSize)
+        pack<3*PacketSize>(blockA, lhs, cols, i, count);
+    
+    if(Pack1>=2*PacketSize)
+      for(Index i=peeled_mc3; i<peeled_mc2; i+=2*PacketSize)
+        pack<2*PacketSize>(blockA, lhs, cols, i, count);
+    
+    if(Pack1>=1*PacketSize)
+      for(Index i=peeled_mc2; i<peeled_mc1; i+=1*PacketSize)
+        pack<1*PacketSize>(blockA, lhs, cols, i, count);
 
     // do the same with mr==1
-    for(Index i=peeled_mc; i<rows; i++)
+    for(Index i=peeled_mc1; i<rows; i++)
     {
       for(Index k=0; k<i; k++)
-        blockA[count++] = lhs(i, k);              // normal
+        blockA[count++] = lhs(i, k);                   // normal
 
       blockA[count++] = numext::real(lhs(i, i));       // real (diagonal)
 
@@ -82,7 +89,8 @@ struct symm_pack_rhs
     Index end_k = k2 + rows;
     Index count = 0;
     const_blas_data_mapper<Scalar,Index,StorageOrder> rhs(_rhs,rhsStride);
-    Index packet_cols = (cols/nr)*nr;
+    Index packet_cols8 = nr>=8 ? (cols/8) * 8 : 0;
+    Index packet_cols4 = nr>=4 ? (cols/4) * 4 : 0;
 
     // first part: normal case
     for(Index j2=0; j2<k2; j2+=nr)
@@ -91,79 +99,151 @@ struct symm_pack_rhs
       {
         blockB[count+0] = rhs(k,j2+0);
         blockB[count+1] = rhs(k,j2+1);
-        if (nr==4)
+        if (nr>=4)
         {
           blockB[count+2] = rhs(k,j2+2);
           blockB[count+3] = rhs(k,j2+3);
         }
+        if (nr>=8)
+        {
+          blockB[count+4] = rhs(k,j2+4);
+          blockB[count+5] = rhs(k,j2+5);
+          blockB[count+6] = rhs(k,j2+6);
+          blockB[count+7] = rhs(k,j2+7);
+        }
         count += nr;
       }
     }
 
     // second part: diagonal block
-    for(Index j2=k2; j2<(std::min)(k2+rows,packet_cols); j2+=nr)
+    Index end8 = nr>=8 ? (std::min)(k2+rows,packet_cols8) : k2;
+    if(nr>=8)
     {
-      // again we can split vertically in three different parts (transpose, symmetric, normal)
-      // transpose
-      for(Index k=k2; k<j2; k++)
+      for(Index j2=k2; j2<end8; j2+=8)
       {
-        blockB[count+0] = numext::conj(rhs(j2+0,k));
-        blockB[count+1] = numext::conj(rhs(j2+1,k));
-        if (nr==4)
+        // again we can split vertically in three different parts (transpose, symmetric, normal)
+        // transpose
+        for(Index k=k2; k<j2; k++)
         {
+          blockB[count+0] = numext::conj(rhs(j2+0,k));
+          blockB[count+1] = numext::conj(rhs(j2+1,k));
           blockB[count+2] = numext::conj(rhs(j2+2,k));
           blockB[count+3] = numext::conj(rhs(j2+3,k));
+          blockB[count+4] = numext::conj(rhs(j2+4,k));
+          blockB[count+5] = numext::conj(rhs(j2+5,k));
+          blockB[count+6] = numext::conj(rhs(j2+6,k));
+          blockB[count+7] = numext::conj(rhs(j2+7,k));
+          count += 8;
         }
-        count += nr;
-      }
-      // symmetric
-      Index h = 0;
-      for(Index k=j2; k<j2+nr; k++)
-      {
-        // normal
-        for (Index w=0 ; w<h; ++w)
-          blockB[count+w] = rhs(k,j2+w);
+        // symmetric
+        Index h = 0;
+        for(Index k=j2; k<j2+8; k++)
+        {
+          // normal
+          for (Index w=0 ; w<h; ++w)
+            blockB[count+w] = rhs(k,j2+w);
 
-        blockB[count+h] = numext::real(rhs(k,k));
+          blockB[count+h] = numext::real(rhs(k,k));
 
-        // transpose
-        for (Index w=h+1 ; w<nr; ++w)
-          blockB[count+w] = numext::conj(rhs(j2+w,k));
-        count += nr;
-        ++h;
+          // transpose
+          for (Index w=h+1 ; w<8; ++w)
+            blockB[count+w] = numext::conj(rhs(j2+w,k));
+          count += 8;
+          ++h;
+        }
+        // normal
+        for(Index k=j2+8; k<end_k; k++)
+        {
+          blockB[count+0] = rhs(k,j2+0);
+          blockB[count+1] = rhs(k,j2+1);
+          blockB[count+2] = rhs(k,j2+2);
+          blockB[count+3] = rhs(k,j2+3);
+          blockB[count+4] = rhs(k,j2+4);
+          blockB[count+5] = rhs(k,j2+5);
+          blockB[count+6] = rhs(k,j2+6);
+          blockB[count+7] = rhs(k,j2+7);
+          count += 8;
+        }
       }
-      // normal
-      for(Index k=j2+nr; k<end_k; k++)
+    }
+    if(nr>=4)
+    {
+      for(Index j2=end8; j2<(std::min)(k2+rows,packet_cols4); j2+=4)
       {
-        blockB[count+0] = rhs(k,j2+0);
-        blockB[count+1] = rhs(k,j2+1);
-        if (nr==4)
+        // again we can split vertically in three different parts (transpose, symmetric, normal)
+        // transpose
+        for(Index k=k2; k<j2; k++)
+        {
+          blockB[count+0] = numext::conj(rhs(j2+0,k));
+          blockB[count+1] = numext::conj(rhs(j2+1,k));
+          blockB[count+2] = numext::conj(rhs(j2+2,k));
+          blockB[count+3] = numext::conj(rhs(j2+3,k));
+          count += 4;
+        }
+        // symmetric
+        Index h = 0;
+        for(Index k=j2; k<j2+4; k++)
         {
+          // normal
+          for (Index w=0 ; w<h; ++w)
+            blockB[count+w] = rhs(k,j2+w);
+
+          blockB[count+h] = numext::real(rhs(k,k));
+
+          // transpose
+          for (Index w=h+1 ; w<4; ++w)
+            blockB[count+w] = numext::conj(rhs(j2+w,k));
+          count += 4;
+          ++h;
+        }
+        // normal
+        for(Index k=j2+4; k<end_k; k++)
+        {
+          blockB[count+0] = rhs(k,j2+0);
+          blockB[count+1] = rhs(k,j2+1);
           blockB[count+2] = rhs(k,j2+2);
           blockB[count+3] = rhs(k,j2+3);
+          count += 4;
         }
-        count += nr;
       }
     }
 
     // third part: transposed
-    for(Index j2=k2+rows; j2<packet_cols; j2+=nr)
+    if(nr>=8)
     {
-      for(Index k=k2; k<end_k; k++)
+      for(Index j2=k2+rows; j2<packet_cols8; j2+=8)
       {
-        blockB[count+0] = numext::conj(rhs(j2+0,k));
-        blockB[count+1] = numext::conj(rhs(j2+1,k));
-        if (nr==4)
+        for(Index k=k2; k<end_k; k++)
         {
+          blockB[count+0] = numext::conj(rhs(j2+0,k));
+          blockB[count+1] = numext::conj(rhs(j2+1,k));
           blockB[count+2] = numext::conj(rhs(j2+2,k));
           blockB[count+3] = numext::conj(rhs(j2+3,k));
+          blockB[count+4] = numext::conj(rhs(j2+4,k));
+          blockB[count+5] = numext::conj(rhs(j2+5,k));
+          blockB[count+6] = numext::conj(rhs(j2+6,k));
+          blockB[count+7] = numext::conj(rhs(j2+7,k));
+          count += 8;
+        }
+      }
+    }
+    if(nr>=4)
+    {
+      for(Index j2=(std::max)(packet_cols8,k2+rows); j2<packet_cols4; j2+=4)
+      {
+        for(Index k=k2; k<end_k; k++)
+        {
+          blockB[count+0] = numext::conj(rhs(j2+0,k));
+          blockB[count+1] = numext::conj(rhs(j2+1,k));
+          blockB[count+2] = numext::conj(rhs(j2+2,k));
+          blockB[count+3] = numext::conj(rhs(j2+3,k));
+          count += 4;
         }
-        count += nr;
       }
     }
 
     // copy the remaining columns one at a time (=> the same with nr==1)
-    for(Index j2=packet_cols; j2<cols; ++j2)
+    for(Index j2=packet_cols4; j2<cols; ++j2)
     {
       // transpose
       Index half = (std::min)(end_k,j2);
@@ -211,7 +291,7 @@ struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,LhsSelfAdjoint,Co
     const Scalar* lhs, Index lhsStride,
     const Scalar* rhs, Index rhsStride,
     Scalar* res,       Index resStride,
-    const Scalar& alpha)
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
     product_selfadjoint_matrix<Scalar, Index,
       EIGEN_LOGICAL_XOR(RhsSelfAdjoint,RhsStorageOrder==RowMajor) ? ColMajor : RowMajor,
@@ -219,7 +299,7 @@ struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,LhsSelfAdjoint,Co
       EIGEN_LOGICAL_XOR(LhsSelfAdjoint,LhsStorageOrder==RowMajor) ? ColMajor : RowMajor,
       LhsSelfAdjoint, NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(LhsSelfAdjoint,ConjugateLhs),
       ColMajor>
-      ::run(cols, rows,  rhs, rhsStride,  lhs, lhsStride,  res, resStride,  alpha);
+      ::run(cols, rows,  rhs, rhsStride,  lhs, lhsStride,  res, resStride,  alpha, blocking);
   }
 };
 
@@ -234,7 +314,7 @@ struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,true,ConjugateLhs
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
     Scalar* res,        Index resStride,
-    const Scalar& alpha);
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
 };
 
 template <typename Scalar, typename Index,
@@ -244,33 +324,35 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,t
     Index rows, Index cols,
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
-    Scalar* res,        Index resStride,
-    const Scalar& alpha)
+    Scalar* _res,        Index resStride,
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
     Index size = rows;
 
-    const_blas_data_mapper<Scalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-    const_blas_data_mapper<Scalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
-
     typedef gebp_traits<Scalar,Scalar> Traits;
 
-    Index kc = size;  // cache block size along the K direction
-    Index mc = rows;  // cache block size along the M direction
-    Index nc = cols;  // cache block size along the N direction
-    computeProductBlockingSizes<Scalar,Scalar>(kc, mc, nc);
-    // kc must smaller than mc
+    typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
+    typedef const_blas_data_mapper<Scalar, Index, (LhsStorageOrder == RowMajor) ? ColMajor : RowMajor> LhsTransposeMapper;
+    typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    LhsTransposeMapper lhs_transpose(_lhs,lhsStride);
+    RhsMapper rhs(_rhs,rhsStride);
+    ResMapper res(_res, resStride);
+
+    Index kc = blocking.kc();                   // cache block size along the K direction
+    Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
+    // kc must be smaller than mc
     kc = (std::min)(kc,mc);
+    std::size_t sizeA = kc*mc;
+    std::size_t sizeB = kc*cols;
+    ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
+    ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
 
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
-    std::size_t sizeB = sizeW + kc*cols;
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockA, kc*mc, 0);
-    ei_declare_aligned_stack_constructed_variable(Scalar, allocatedBlockB, sizeB, 0);
-    Scalar* blockB = allocatedBlockB + sizeW;
-
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
+    gebp_kernel<Scalar, Scalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
     symm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder==RowMajor?ColMajor:RowMajor, true> pack_lhs_transposed;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr,RhsStorageOrder> pack_rhs;
+    gemm_pack_lhs<Scalar, Index, LhsTransposeMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder==RowMajor?ColMajor:RowMajor, true> pack_lhs_transposed;
 
     for(Index k2=0; k2<size; k2+=kc)
     {
@@ -279,7 +361,7 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,t
       // we have selected one row panel of rhs and one column panel of lhs
       // pack rhs's panel into a sequential chunk of memory
       // and expand each coeff to a constant packet for further reuse
-      pack_rhs(blockB, &rhs(k2,0), rhsStride, actual_kc, cols);
+      pack_rhs(blockB, rhs.getSubMapper(k2,0), actual_kc, cols);
 
       // the select lhs's panel has to be split in three different parts:
       //  1 - the transposed panel above the diagonal block => transposed packed copy
@@ -289,9 +371,9 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,t
       {
         const Index actual_mc = (std::min)(i2+mc,k2)-i2;
         // transposed packed copy
-        pack_lhs_transposed(blockA, &lhs(k2, i2), lhsStride, actual_kc, actual_mc);
+        pack_lhs_transposed(blockA, lhs_transpose.getSubMapper(i2, k2), actual_kc, actual_mc);
 
-        gebp_kernel(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha);
+        gebp_kernel(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc, cols, alpha);
       }
       // the block diagonal
       {
@@ -299,16 +381,16 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,t
         // symmetric packed copy
         pack_lhs(blockA, &lhs(k2,k2), lhsStride, actual_kc, actual_mc);
 
-        gebp_kernel(res+k2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha);
+        gebp_kernel(res.getSubMapper(k2, 0), blockA, blockB, actual_mc, actual_kc, cols, alpha);
       }
 
       for(Index i2=k2+kc; i2<size; i2+=mc)
       {
         const Index actual_mc = (std::min)(i2+mc,size)-i2;
-        gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder,false>()
-          (blockA, &lhs(i2, k2), lhsStride, actual_kc, actual_mc);
+        gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder,false>()
+          (blockA, lhs.getSubMapper(i2, k2), actual_kc, actual_mc);
 
-        gebp_kernel(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha);
+        gebp_kernel(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc, cols, alpha);
       }
     }
   }
@@ -325,7 +407,7 @@ struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,false,ConjugateLh
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
     Scalar* res,        Index resStride,
-    const Scalar& alpha);
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
 };
 
 template <typename Scalar, typename Index,
@@ -335,27 +417,27 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,f
     Index rows, Index cols,
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
-    Scalar* res,        Index resStride,
-    const Scalar& alpha)
+    Scalar* _res,        Index resStride,
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
     Index size = cols;
 
-    const_blas_data_mapper<Scalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-
     typedef gebp_traits<Scalar,Scalar> Traits;
 
-    Index kc = size; // cache block size along the K direction
-    Index mc = rows;  // cache block size along the M direction
-    Index nc = cols;  // cache block size along the N direction
-    computeProductBlockingSizes<Scalar,Scalar>(kc, mc, nc);
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
-    std::size_t sizeB = sizeW + kc*cols;
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockA, kc*mc, 0);
-    ei_declare_aligned_stack_constructed_variable(Scalar, allocatedBlockB, sizeB, 0);
-    Scalar* blockB = allocatedBlockB + sizeW;
-
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+    typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    ResMapper res(_res,resStride);
+
+    Index kc = blocking.kc();                   // cache block size along the K direction
+    Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
+    std::size_t sizeA = kc*mc;
+    std::size_t sizeB = kc*cols;
+    ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
+    ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
+
+    gebp_kernel<Scalar, Scalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
+    gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
     symm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
 
     for(Index k2=0; k2<size; k2+=kc)
@@ -368,9 +450,9 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,f
       for(Index i2=0; i2<rows; i2+=mc)
       {
         const Index actual_mc = (std::min)(i2+mc,rows)-i2;
-        pack_lhs(blockA, &lhs(i2, k2), lhsStride, actual_kc, actual_mc);
+        pack_lhs(blockA, lhs.getSubMapper(i2, k2), actual_kc, actual_mc);
 
-        gebp_kernel(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha);
+        gebp_kernel(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc, cols, alpha);
       }
     }
   }
@@ -382,55 +464,58 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,f
 ***************************************************************************/
 
 namespace internal {
+  
 template<typename Lhs, int LhsMode, typename Rhs, int RhsMode>
-struct traits<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false> >
-  : traits<ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>, Lhs, Rhs> >
-{};
-}
-
-template<typename Lhs, int LhsMode, typename Rhs, int RhsMode>
-struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>
-  : public ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>, Lhs, Rhs >
+struct selfadjoint_product_impl<Lhs,LhsMode,false,Rhs,RhsMode,false>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(SelfadjointProductMatrix)
-
-  SelfadjointProductMatrix(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  typedef internal::blas_traits<Lhs> LhsBlasTraits;
+  typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+  typedef internal::blas_traits<Rhs> RhsBlasTraits;
+  typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+  
   enum {
     LhsIsUpper = (LhsMode&(Upper|Lower))==Upper,
     LhsIsSelfAdjoint = (LhsMode&SelfAdjoint)==SelfAdjoint,
     RhsIsUpper = (RhsMode&(Upper|Lower))==Upper,
     RhsIsSelfAdjoint = (RhsMode&SelfAdjoint)==SelfAdjoint
   };
-
-  template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
+  
+  template<typename Dest>
+  static void run(Dest &dst, const Lhs &a_lhs, const Rhs &a_rhs, const Scalar& alpha)
   {
-    eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
+    eigen_assert(dst.rows()==a_lhs.rows() && dst.cols()==a_rhs.cols());
 
-    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
-    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
+    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
 
-    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
-                               * RhsBlasTraits::extractScalarFactor(m_rhs);
+    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
+                               * RhsBlasTraits::extractScalarFactor(a_rhs);
+
+    typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,Scalar,Scalar,
+              Lhs::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime, Lhs::MaxColsAtCompileTime,1> BlockingType;
+
+    BlockingType blocking(lhs.rows(), rhs.cols(), lhs.cols(), 1, false);
 
     internal::product_selfadjoint_matrix<Scalar, Index,
-      EIGEN_LOGICAL_XOR(LhsIsUpper,
-                        internal::traits<Lhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, LhsIsSelfAdjoint,
+      EIGEN_LOGICAL_XOR(LhsIsUpper,internal::traits<Lhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, LhsIsSelfAdjoint,
       NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(LhsIsUpper,bool(LhsBlasTraits::NeedToConjugate)),
-      EIGEN_LOGICAL_XOR(RhsIsUpper,
-                        internal::traits<Rhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, RhsIsSelfAdjoint,
+      EIGEN_LOGICAL_XOR(RhsIsUpper,internal::traits<Rhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, RhsIsSelfAdjoint,
       NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(RhsIsUpper,bool(RhsBlasTraits::NeedToConjugate)),
       internal::traits<Dest>::Flags&RowMajorBit  ? RowMajor : ColMajor>
       ::run(
         lhs.rows(), rhs.cols(),                 // sizes
-        &lhs.coeffRef(0,0),    lhs.outerStride(),  // lhs info
-        &rhs.coeffRef(0,0),    rhs.outerStride(),  // rhs info
+        &lhs.coeffRef(0,0), lhs.outerStride(),  // lhs info
+        &rhs.coeffRef(0,0), rhs.outerStride(),  // rhs info
         &dst.coeffRef(0,0), dst.outerStride(),  // result info
-        actualAlpha                             // alpha
+        actualAlpha, blocking                   // alpha
       );
   }
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SELFADJOINT_MATRIX_MATRIX_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixMatrix_BLAS.h b/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixMatrix_BLAS.h
new file mode 100644
index 00000000000..9a5318507af
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixMatrix_BLAS.h
@@ -0,0 +1,287 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   Self adjoint matrix * matrix product functionality based on ?SYMM/?HEMM.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_SELFADJOINT_MATRIX_MATRIX_BLAS_H
+#define EIGEN_SELFADJOINT_MATRIX_MATRIX_BLAS_H
+
+namespace Eigen { 
+
+namespace internal {
+
+
+/* Optimized selfadjoint matrix * matrix (?SYMM/?HEMM) product */
+
+#define EIGEN_BLAS_SYMM_L(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
+template <typename Index, \
+          int LhsStorageOrder, bool ConjugateLhs, \
+          int RhsStorageOrder, bool ConjugateRhs> \
+struct product_selfadjoint_matrix<EIGTYPE,Index,LhsStorageOrder,true,ConjugateLhs,RhsStorageOrder,false,ConjugateRhs,ColMajor> \
+{\
+\
+  static void run( \
+    Index rows, Index cols, \
+    const EIGTYPE* _lhs, Index lhsStride, \
+    const EIGTYPE* _rhs, Index rhsStride, \
+    EIGTYPE* res,        Index resStride, \
+    EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
+  { \
+    char side='L', uplo='L'; \
+    BlasIndex m, n, lda, ldb, ldc; \
+    const EIGTYPE *a, *b; \
+    EIGTYPE beta(1); \
+    MatrixX##EIGPREFIX b_tmp; \
+\
+/* Set transpose options */ \
+/* Set m, n, k */ \
+    m = convert_index<BlasIndex>(rows);  \
+    n = convert_index<BlasIndex>(cols);  \
+\
+/* Set lda, ldb, ldc */ \
+    lda = convert_index<BlasIndex>(lhsStride); \
+    ldb = convert_index<BlasIndex>(rhsStride); \
+    ldc = convert_index<BlasIndex>(resStride); \
+\
+/* Set a, b, c */ \
+    if (LhsStorageOrder==RowMajor) uplo='U'; \
+    a = _lhs; \
+\
+    if (RhsStorageOrder==RowMajor) { \
+      Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > rhs(_rhs,n,m,OuterStride<>(rhsStride)); \
+      b_tmp = rhs.adjoint(); \
+      b = b_tmp.data(); \
+      ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
+    } else b = _rhs; \
+\
+    BLASFUNC(&side, &uplo, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
+\
+  } \
+};
+
+
+#define EIGEN_BLAS_HEMM_L(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
+template <typename Index, \
+          int LhsStorageOrder, bool ConjugateLhs, \
+          int RhsStorageOrder, bool ConjugateRhs> \
+struct product_selfadjoint_matrix<EIGTYPE,Index,LhsStorageOrder,true,ConjugateLhs,RhsStorageOrder,false,ConjugateRhs,ColMajor> \
+{\
+  static void run( \
+    Index rows, Index cols, \
+    const EIGTYPE* _lhs, Index lhsStride, \
+    const EIGTYPE* _rhs, Index rhsStride, \
+    EIGTYPE* res,        Index resStride, \
+    EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
+  { \
+    char side='L', uplo='L'; \
+    BlasIndex m, n, lda, ldb, ldc; \
+    const EIGTYPE *a, *b; \
+    EIGTYPE beta(1); \
+    MatrixX##EIGPREFIX b_tmp; \
+    Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder> a_tmp; \
+\
+/* Set transpose options */ \
+/* Set m, n, k */ \
+    m = convert_index<BlasIndex>(rows); \
+    n = convert_index<BlasIndex>(cols); \
+\
+/* Set lda, ldb, ldc */ \
+    lda = convert_index<BlasIndex>(lhsStride); \
+    ldb = convert_index<BlasIndex>(rhsStride); \
+    ldc = convert_index<BlasIndex>(resStride); \
+\
+/* Set a, b, c */ \
+    if (((LhsStorageOrder==ColMajor) && ConjugateLhs) || ((LhsStorageOrder==RowMajor) && (!ConjugateLhs))) { \
+      Map<const Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder>, 0, OuterStride<> > lhs(_lhs,m,m,OuterStride<>(lhsStride)); \
+      a_tmp = lhs.conjugate(); \
+      a = a_tmp.data(); \
+      lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
+    } else a = _lhs; \
+    if (LhsStorageOrder==RowMajor) uplo='U'; \
+\
+    if (RhsStorageOrder==ColMajor && (!ConjugateRhs)) { \
+       b = _rhs; } \
+    else { \
+      if (RhsStorageOrder==ColMajor && ConjugateRhs) { \
+        Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > rhs(_rhs,m,n,OuterStride<>(rhsStride)); \
+        b_tmp = rhs.conjugate(); \
+      } else \
+      if (ConjugateRhs) { \
+        Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > rhs(_rhs,n,m,OuterStride<>(rhsStride)); \
+        b_tmp = rhs.adjoint(); \
+      } else { \
+        Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > rhs(_rhs,n,m,OuterStride<>(rhsStride)); \
+        b_tmp = rhs.transpose(); \
+      } \
+      b = b_tmp.data(); \
+      ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
+    } \
+\
+    BLASFUNC(&side, &uplo, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
+\
+  } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_SYMM_L(double, double, d, dsymm)
+EIGEN_BLAS_SYMM_L(float, float, f, ssymm)
+EIGEN_BLAS_HEMM_L(dcomplex, MKL_Complex16, cd, zhemm)
+EIGEN_BLAS_HEMM_L(scomplex, MKL_Complex8, cf, chemm)
+#else
+EIGEN_BLAS_SYMM_L(double, double, d, dsymm_)
+EIGEN_BLAS_SYMM_L(float, float, f, ssymm_)
+EIGEN_BLAS_HEMM_L(dcomplex, double, cd, zhemm_)
+EIGEN_BLAS_HEMM_L(scomplex, float, cf, chemm_)
+#endif
+
+/* Optimized matrix * selfadjoint matrix (?SYMM/?HEMM) product */
+
+#define EIGEN_BLAS_SYMM_R(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
+template <typename Index, \
+          int LhsStorageOrder, bool ConjugateLhs, \
+          int RhsStorageOrder, bool ConjugateRhs> \
+struct product_selfadjoint_matrix<EIGTYPE,Index,LhsStorageOrder,false,ConjugateLhs,RhsStorageOrder,true,ConjugateRhs,ColMajor> \
+{\
+\
+  static void run( \
+    Index rows, Index cols, \
+    const EIGTYPE* _lhs, Index lhsStride, \
+    const EIGTYPE* _rhs, Index rhsStride, \
+    EIGTYPE* res,        Index resStride, \
+    EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
+  { \
+    char side='R', uplo='L'; \
+    BlasIndex m, n, lda, ldb, ldc; \
+    const EIGTYPE *a, *b; \
+    EIGTYPE beta(1); \
+    MatrixX##EIGPREFIX b_tmp; \
+\
+/* Set m, n, k */ \
+    m = convert_index<BlasIndex>(rows);  \
+    n = convert_index<BlasIndex>(cols);  \
+\
+/* Set lda, ldb, ldc */ \
+    lda = convert_index<BlasIndex>(rhsStride); \
+    ldb = convert_index<BlasIndex>(lhsStride); \
+    ldc = convert_index<BlasIndex>(resStride); \
+\
+/* Set a, b, c */ \
+    if (RhsStorageOrder==RowMajor) uplo='U'; \
+    a = _rhs; \
+\
+    if (LhsStorageOrder==RowMajor) { \
+      Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > lhs(_lhs,n,m,OuterStride<>(rhsStride)); \
+      b_tmp = lhs.adjoint(); \
+      b = b_tmp.data(); \
+      ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
+    } else b = _lhs; \
+\
+    BLASFUNC(&side, &uplo, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
+\
+  } \
+};
+
+
+#define EIGEN_BLAS_HEMM_R(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
+template <typename Index, \
+          int LhsStorageOrder, bool ConjugateLhs, \
+          int RhsStorageOrder, bool ConjugateRhs> \
+struct product_selfadjoint_matrix<EIGTYPE,Index,LhsStorageOrder,false,ConjugateLhs,RhsStorageOrder,true,ConjugateRhs,ColMajor> \
+{\
+  static void run( \
+    Index rows, Index cols, \
+    const EIGTYPE* _lhs, Index lhsStride, \
+    const EIGTYPE* _rhs, Index rhsStride, \
+    EIGTYPE* res,        Index resStride, \
+    EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
+  { \
+    char side='R', uplo='L'; \
+    BlasIndex m, n, lda, ldb, ldc; \
+    const EIGTYPE *a, *b; \
+    EIGTYPE beta(1); \
+    MatrixX##EIGPREFIX b_tmp; \
+    Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> a_tmp; \
+\
+/* Set m, n, k */ \
+    m = convert_index<BlasIndex>(rows); \
+    n = convert_index<BlasIndex>(cols); \
+\
+/* Set lda, ldb, ldc */ \
+    lda = convert_index<BlasIndex>(rhsStride); \
+    ldb = convert_index<BlasIndex>(lhsStride); \
+    ldc = convert_index<BlasIndex>(resStride); \
+\
+/* Set a, b, c */ \
+    if (((RhsStorageOrder==ColMajor) && ConjugateRhs) || ((RhsStorageOrder==RowMajor) && (!ConjugateRhs))) { \
+      Map<const Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder>, 0, OuterStride<> > rhs(_rhs,n,n,OuterStride<>(rhsStride)); \
+      a_tmp = rhs.conjugate(); \
+      a = a_tmp.data(); \
+      lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
+    } else a = _rhs; \
+    if (RhsStorageOrder==RowMajor) uplo='U'; \
+\
+    if (LhsStorageOrder==ColMajor && (!ConjugateLhs)) { \
+       b = _lhs; } \
+    else { \
+      if (LhsStorageOrder==ColMajor && ConjugateLhs) { \
+        Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > lhs(_lhs,m,n,OuterStride<>(lhsStride)); \
+        b_tmp = lhs.conjugate(); \
+      } else \
+      if (ConjugateLhs) { \
+        Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > lhs(_lhs,n,m,OuterStride<>(lhsStride)); \
+        b_tmp = lhs.adjoint(); \
+      } else { \
+        Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > lhs(_lhs,n,m,OuterStride<>(lhsStride)); \
+        b_tmp = lhs.transpose(); \
+      } \
+      b = b_tmp.data(); \
+      ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
+    } \
+\
+    BLASFUNC(&side, &uplo, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
+  } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_SYMM_R(double, double, d, dsymm)
+EIGEN_BLAS_SYMM_R(float, float, f, ssymm)
+EIGEN_BLAS_HEMM_R(dcomplex, MKL_Complex16, cd, zhemm)
+EIGEN_BLAS_HEMM_R(scomplex, MKL_Complex8, cf, chemm)
+#else
+EIGEN_BLAS_SYMM_R(double, double, d, dsymm_)
+EIGEN_BLAS_SYMM_R(float, float, f, ssymm_)
+EIGEN_BLAS_HEMM_R(dcomplex, double, cd, zhemm_)
+EIGEN_BLAS_HEMM_R(scomplex, float, cf, chemm_)
+#endif
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SELFADJOINT_MATRIX_MATRIX_BLAS_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixVector.h b/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixVector.h
index f698f67f9b0..3fd180e6c0b 100644
--- a/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixVector.h
+++ b/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixVector.h
@@ -30,7 +30,7 @@ struct selfadjoint_matrix_vector_product
 static EIGEN_DONT_INLINE void run(
   Index size,
   const Scalar*  lhs, Index lhsStride,
-  const Scalar* _rhs, Index rhsIncr,
+  const Scalar*  rhs,
   Scalar* res,
   Scalar alpha);
 };
@@ -39,11 +39,12 @@ template<typename Scalar, typename Index, int StorageOrder, int UpLo, bool Conju
 EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs,Version>::run(
   Index size,
   const Scalar*  lhs, Index lhsStride,
-  const Scalar* _rhs, Index rhsIncr,
+  const Scalar*  rhs,
   Scalar* res,
   Scalar alpha)
 {
   typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename NumTraits<Scalar>::Real RealScalar;
   const Index PacketSize = sizeof(Packet)/sizeof(Scalar);
 
   enum {
@@ -54,23 +55,13 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
 
   conj_helper<Scalar,Scalar,NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(ConjugateLhs,  IsRowMajor), ConjugateRhs> cj0;
   conj_helper<Scalar,Scalar,NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(ConjugateLhs, !IsRowMajor), ConjugateRhs> cj1;
-  conj_helper<Scalar,Scalar,NumTraits<Scalar>::IsComplex, ConjugateRhs> cjd;
+  conj_helper<RealScalar,Scalar,false, ConjugateRhs> cjd;
 
   conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(ConjugateLhs,  IsRowMajor), ConjugateRhs> pcj0;
   conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(ConjugateLhs, !IsRowMajor), ConjugateRhs> pcj1;
 
   Scalar cjAlpha = ConjugateRhs ? numext::conj(alpha) : alpha;
 
-  // FIXME this copy is now handled outside product_selfadjoint_vector, so it could probably be removed.
-  // if the rhs is not sequentially stored in memory we copy it to a temporary buffer,
-  // this is because we need to extract packets
-  ei_declare_aligned_stack_constructed_variable(Scalar,rhs,size,rhsIncr==1 ? const_cast<Scalar*>(_rhs) : 0);  
-  if (rhsIncr!=1)
-  {
-    const Scalar* it = _rhs;
-    for (Index i=0; i<size; ++i, it+=rhsIncr)
-      rhs[i] = *it;
-  }
 
   Index bound = (std::max)(Index(0),size-8) & 0xfffffffe;
   if (FirstTriangular)
@@ -92,12 +83,11 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
     Scalar t3(0);
     Packet ptmp3 = pset1<Packet>(t3);
 
-    size_t starti = FirstTriangular ? 0 : j+2;
-    size_t endi   = FirstTriangular ? j : size;
-    size_t alignedStart = (starti) + internal::first_aligned(&res[starti], endi-starti);
-    size_t alignedEnd = alignedStart + ((endi-alignedStart)/(PacketSize))*(PacketSize);
+    Index starti = FirstTriangular ? 0 : j+2;
+    Index endi   = FirstTriangular ? j : size;
+    Index alignedStart = (starti) + internal::first_default_aligned(&res[starti], endi-starti);
+    Index alignedEnd = alignedStart + ((endi-alignedStart)/(PacketSize))*(PacketSize);
 
-    // TODO make sure this product is a real * complex and that the rhs is properly conjugated if needed
     res[j]   += cjd.pmul(numext::real(A0[j]), t0);
     res[j+1] += cjd.pmul(numext::real(A1[j+1]), t1);
     if(FirstTriangular)
@@ -111,11 +101,11 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
       t2 += cj1.pmul(A0[j+1], rhs[j+1]);
     }
 
-    for (size_t i=starti; i<alignedStart; ++i)
+    for (Index i=starti; i<alignedStart; ++i)
     {
-      res[i] += t0 * A0[i] + t1 * A1[i];
-      t2 += numext::conj(A0[i]) * rhs[i];
-      t3 += numext::conj(A1[i]) * rhs[i];
+      res[i] += cj0.pmul(A0[i], t0) + cj0.pmul(A1[i],t1);
+      t2 += cj1.pmul(A0[i], rhs[i]);
+      t3 += cj1.pmul(A1[i], rhs[i]);
     }
     // Yes this an optimization for gcc 4.3 and 4.4 (=> huge speed up)
     // gcc 4.2 does this optimization automatically.
@@ -123,7 +113,7 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
     const Scalar* EIGEN_RESTRICT a1It  = A1  + alignedStart;
     const Scalar* EIGEN_RESTRICT rhsIt = rhs + alignedStart;
           Scalar* EIGEN_RESTRICT resIt = res + alignedStart;
-    for (size_t i=alignedStart; i<alignedEnd; i+=PacketSize)
+    for (Index i=alignedStart; i<alignedEnd; i+=PacketSize)
     {
       Packet A0i = ploadu<Packet>(a0It);  a0It  += PacketSize;
       Packet A1i = ploadu<Packet>(a1It);  a1It  += PacketSize;
@@ -135,7 +125,7 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
       ptmp3 = pcj1.pmadd(A1i,  Bi, ptmp3);
       pstore(resIt,Xi); resIt += PacketSize;
     }
-    for (size_t i=alignedEnd; i<endi; i++)
+    for (Index i=alignedEnd; i<endi; i++)
     {
       res[i] += cj0.pmul(A0[i], t0) + cj0.pmul(A1[i],t1);
       t2 += cj1.pmul(A0[i], rhs[i]);
@@ -151,7 +141,6 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
 
     Scalar t1 = cjAlpha * rhs[j];
     Scalar t2(0);
-    // TODO make sure this product is a real * complex and that the rhs is properly conjugated if needed
     res[j] += cjd.pmul(numext::real(A0[j]), t1);
     for (Index i=FirstTriangular ? 0 : j+1; i<(FirstTriangular ? j : size); i++)
     {
@@ -169,45 +158,44 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
 ***************************************************************************/
 
 namespace internal {
-template<typename Lhs, int LhsMode, typename Rhs>
-struct traits<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true> >
-  : traits<ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>, Lhs, Rhs> >
-{};
-}
 
 template<typename Lhs, int LhsMode, typename Rhs>
-struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>
-  : public ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>, Lhs, Rhs >
+struct selfadjoint_product_impl<Lhs,LhsMode,false,Rhs,0,true>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(SelfadjointProductMatrix)
-
-  enum {
-    LhsUpLo = LhsMode&(Upper|Lower)
-  };
-
-  SelfadjointProductMatrix(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  typedef internal::blas_traits<Lhs> LhsBlasTraits;
+  typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+  typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned;
+  
+  typedef internal::blas_traits<Rhs> RhsBlasTraits;
+  typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+  typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
+
+  enum { LhsUpLo = LhsMode&(Upper|Lower) };
+
+  template<typename Dest>
+  static void run(Dest& dest, const Lhs &a_lhs, const Rhs &a_rhs, const Scalar& alpha)
   {
     typedef typename Dest::Scalar ResScalar;
-    typedef typename Base::RhsScalar RhsScalar;
-    typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
+    typedef typename Rhs::Scalar RhsScalar;
+    typedef Map<Matrix<ResScalar,Dynamic,1>, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits<ResScalar>::size)> MappedDest;
     
-    eigen_assert(dest.rows()==m_lhs.rows() && dest.cols()==m_rhs.cols());
+    eigen_assert(dest.rows()==a_lhs.rows() && dest.cols()==a_rhs.cols());
 
-    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
-    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
+    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
 
-    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
-                               * RhsBlasTraits::extractScalarFactor(m_rhs);
+    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
+                               * RhsBlasTraits::extractScalarFactor(a_rhs);
 
     enum {
       EvalToDest = (Dest::InnerStrideAtCompileTime==1),
-      UseRhs = (_ActualRhsType::InnerStrideAtCompileTime==1)
+      UseRhs = (ActualRhsTypeCleaned::InnerStrideAtCompileTime==1)
     };
     
     internal::gemv_static_vector_if<ResScalar,Dest::SizeAtCompileTime,Dest::MaxSizeAtCompileTime,!EvalToDest> static_dest;
-    internal::gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!UseRhs> static_rhs;
+    internal::gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!UseRhs> static_rhs;
 
     ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
                                                   EvalToDest ? dest.data() : static_dest.data());
@@ -218,7 +206,7 @@ struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>
     if(!EvalToDest)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = dest.size();
+      Index size = dest.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
       MappedDest(actualDestPtr, dest.size()) = dest;
@@ -227,18 +215,19 @@ struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>
     if(!UseRhs)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = rhs.size();
+      Index size = rhs.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
-      Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, rhs.size()) = rhs;
+      Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, rhs.size()) = rhs;
     }
       
       
-    internal::selfadjoint_matrix_vector_product<Scalar, Index, (internal::traits<_ActualLhsType>::Flags&RowMajorBit) ? RowMajor : ColMajor, int(LhsUpLo), bool(LhsBlasTraits::NeedToConjugate), bool(RhsBlasTraits::NeedToConjugate)>::run
+    internal::selfadjoint_matrix_vector_product<Scalar, Index, (internal::traits<ActualLhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor,
+                                                int(LhsUpLo), bool(LhsBlasTraits::NeedToConjugate), bool(RhsBlasTraits::NeedToConjugate)>::run
       (
         lhs.rows(),                             // size
         &lhs.coeffRef(0,0),  lhs.outerStride(), // lhs info
-        actualRhsPtr, 1,                        // rhs info
+        actualRhsPtr,                           // rhs info
         actualDestPtr,                          // result info
         actualAlpha                             // scale factor
       );
@@ -248,34 +237,24 @@ struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>
   }
 };
 
-namespace internal {
-template<typename Lhs, typename Rhs, int RhsMode>
-struct traits<SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false> >
-  : traits<ProductBase<SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false>, Lhs, Rhs> >
-{};
-}
-
 template<typename Lhs, typename Rhs, int RhsMode>
-struct SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false>
-  : public ProductBase<SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false>, Lhs, Rhs >
+struct selfadjoint_product_impl<Lhs,0,true,Rhs,RhsMode,false>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(SelfadjointProductMatrix)
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  enum { RhsUpLo = RhsMode&(Upper|Lower)  };
 
-  enum {
-    RhsUpLo = RhsMode&(Upper|Lower)
-  };
-
-  SelfadjointProductMatrix(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+  template<typename Dest>
+  static void run(Dest& dest, const Lhs &a_lhs, const Rhs &a_rhs, const Scalar& alpha)
   {
     // let's simply transpose the product
     Transpose<Dest> destT(dest);
-    SelfadjointProductMatrix<Transpose<const Rhs>, int(RhsUpLo)==Upper ? Lower : Upper, false,
-                             Transpose<const Lhs>, 0, true>(m_rhs.transpose(), m_lhs.transpose()).scaleAndAddTo(destT, alpha);
+    selfadjoint_product_impl<Transpose<const Rhs>, int(RhsUpLo)==Upper ? Lower : Upper, false,
+                             Transpose<const Lhs>, 0, true>::run(destT, a_rhs.transpose(), a_lhs.transpose(), alpha);
   }
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SELFADJOINT_MATRIX_VECTOR_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixVector_BLAS.h b/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixVector_BLAS.h
new file mode 100644
index 00000000000..1238345e3fa
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/products/SelfadjointMatrixVector_BLAS.h
@@ -0,0 +1,118 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   Selfadjoint matrix-vector product functionality based on ?SYMV/HEMV.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_SELFADJOINT_MATRIX_VECTOR_BLAS_H
+#define EIGEN_SELFADJOINT_MATRIX_VECTOR_BLAS_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/**********************************************************************
+* This file implements selfadjoint matrix-vector multiplication using BLAS
+**********************************************************************/
+
+// symv/hemv specialization
+
+template<typename Scalar, typename Index, int StorageOrder, int UpLo, bool ConjugateLhs, bool ConjugateRhs>
+struct selfadjoint_matrix_vector_product_symv :
+  selfadjoint_matrix_vector_product<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs,BuiltIn> {};
+
+#define EIGEN_BLAS_SYMV_SPECIALIZE(Scalar) \
+template<typename Index, int StorageOrder, int UpLo, bool ConjugateLhs, bool ConjugateRhs> \
+struct selfadjoint_matrix_vector_product<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs,Specialized> { \
+static void run( \
+  Index size, const Scalar*  lhs, Index lhsStride, \
+  const Scalar* _rhs, Scalar* res, Scalar alpha) { \
+    enum {\
+      IsColMajor = StorageOrder==ColMajor \
+    }; \
+    if (IsColMajor == ConjugateLhs) {\
+      selfadjoint_matrix_vector_product<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs,BuiltIn>::run( \
+        size, lhs, lhsStride, _rhs, res, alpha);  \
+    } else {\
+      selfadjoint_matrix_vector_product_symv<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs>::run( \
+        size, lhs, lhsStride, _rhs, res, alpha);  \
+    }\
+  } \
+}; \
+
+EIGEN_BLAS_SYMV_SPECIALIZE(double)
+EIGEN_BLAS_SYMV_SPECIALIZE(float)
+EIGEN_BLAS_SYMV_SPECIALIZE(dcomplex)
+EIGEN_BLAS_SYMV_SPECIALIZE(scomplex)
+
+#define EIGEN_BLAS_SYMV_SPECIALIZATION(EIGTYPE,BLASTYPE,BLASFUNC) \
+template<typename Index, int StorageOrder, int UpLo, bool ConjugateLhs, bool ConjugateRhs> \
+struct selfadjoint_matrix_vector_product_symv<EIGTYPE,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs> \
+{ \
+typedef Matrix<EIGTYPE,Dynamic,1,ColMajor> SYMVVector;\
+\
+static void run( \
+Index size, const EIGTYPE*  lhs, Index lhsStride, \
+const EIGTYPE* _rhs, EIGTYPE* res, EIGTYPE alpha) \
+{ \
+  enum {\
+    IsRowMajor = StorageOrder==RowMajor ? 1 : 0, \
+    IsLower = UpLo == Lower ? 1 : 0 \
+  }; \
+  BlasIndex n=convert_index<BlasIndex>(size), lda=convert_index<BlasIndex>(lhsStride), incx=1, incy=1; \
+  EIGTYPE beta(1); \
+  const EIGTYPE *x_ptr; \
+  char uplo=(IsRowMajor) ? (IsLower ? 'U' : 'L') : (IsLower ? 'L' : 'U'); \
+  SYMVVector x_tmp; \
+  if (ConjugateRhs) { \
+    Map<const SYMVVector, 0 > map_x(_rhs,size,1); \
+    x_tmp=map_x.conjugate(); \
+    x_ptr=x_tmp.data(); \
+  } else x_ptr=_rhs; \
+  BLASFUNC(&uplo, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)lhs, &lda, (const BLASTYPE*)x_ptr, &incx, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &incy); \
+}\
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_SYMV_SPECIALIZATION(double,   double, dsymv)
+EIGEN_BLAS_SYMV_SPECIALIZATION(float,    float,  ssymv)
+EIGEN_BLAS_SYMV_SPECIALIZATION(dcomplex, MKL_Complex16, zhemv)
+EIGEN_BLAS_SYMV_SPECIALIZATION(scomplex, MKL_Complex8,  chemv)
+#else
+EIGEN_BLAS_SYMV_SPECIALIZATION(double,   double, dsymv_)
+EIGEN_BLAS_SYMV_SPECIALIZATION(float,    float,  ssymv_)
+EIGEN_BLAS_SYMV_SPECIALIZATION(dcomplex, double, zhemv_)
+EIGEN_BLAS_SYMV_SPECIALIZATION(scomplex, float,  chemv_)
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SELFADJOINT_MATRIX_VECTOR_BLAS_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/SelfadjointProduct.h b/extern/Eigen3/Eigen/src/Core/products/SelfadjointProduct.h
index 6ca4ae6c0fe..f038d686f5d 100644
--- a/extern/Eigen3/Eigen/src/Core/products/SelfadjointProduct.h
+++ b/extern/Eigen3/Eigen/src/Core/products/SelfadjointProduct.h
@@ -53,7 +53,6 @@ struct selfadjoint_product_selector<MatrixType,OtherType,UpLo,true>
   static void run(MatrixType& mat, const OtherType& other, const typename MatrixType::Scalar& alpha)
   {
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
     typedef internal::blas_traits<OtherType> OtherBlasTraits;
     typedef typename OtherBlasTraits::DirectLinearAccessType ActualOtherType;
     typedef typename internal::remove_all<ActualOtherType>::type _ActualOtherType;
@@ -86,7 +85,6 @@ struct selfadjoint_product_selector<MatrixType,OtherType,UpLo,false>
   static void run(MatrixType& mat, const OtherType& other, const typename MatrixType::Scalar& alpha)
   {
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
     typedef internal::blas_traits<OtherType> OtherBlasTraits;
     typedef typename OtherBlasTraits::DirectLinearAccessType ActualOtherType;
     typedef typename internal::remove_all<ActualOtherType>::type _ActualOtherType;
@@ -94,15 +92,27 @@ struct selfadjoint_product_selector<MatrixType,OtherType,UpLo,false>
 
     Scalar actualAlpha = alpha * OtherBlasTraits::extractScalarFactor(other.derived());
 
-    enum { IsRowMajor = (internal::traits<MatrixType>::Flags&RowMajorBit) ? 1 : 0 };
+    enum {
+      IsRowMajor = (internal::traits<MatrixType>::Flags&RowMajorBit) ? 1 : 0,
+      OtherIsRowMajor = _ActualOtherType::Flags&RowMajorBit ? 1 : 0
+    };
+
+    Index size = mat.cols();
+    Index depth = actualOther.cols();
+
+    typedef internal::gemm_blocking_space<IsRowMajor ? RowMajor : ColMajor,Scalar,Scalar,
+              MatrixType::MaxColsAtCompileTime, MatrixType::MaxColsAtCompileTime, _ActualOtherType::MaxColsAtCompileTime> BlockingType;
+
+    BlockingType blocking(size, size, depth, 1, false);
+
 
     internal::general_matrix_matrix_triangular_product<Index,
-      Scalar, _ActualOtherType::Flags&RowMajorBit ? RowMajor : ColMajor,   OtherBlasTraits::NeedToConjugate  && NumTraits<Scalar>::IsComplex,
-      Scalar, _ActualOtherType::Flags&RowMajorBit ? ColMajor : RowMajor, (!OtherBlasTraits::NeedToConjugate) && NumTraits<Scalar>::IsComplex,
-      MatrixType::Flags&RowMajorBit ? RowMajor : ColMajor, UpLo>
-      ::run(mat.cols(), actualOther.cols(),
+      Scalar, OtherIsRowMajor ? RowMajor : ColMajor,   OtherBlasTraits::NeedToConjugate  && NumTraits<Scalar>::IsComplex,
+      Scalar, OtherIsRowMajor ? ColMajor : RowMajor, (!OtherBlasTraits::NeedToConjugate) && NumTraits<Scalar>::IsComplex,
+      IsRowMajor ? RowMajor : ColMajor, UpLo>
+      ::run(size, depth,
             &actualOther.coeffRef(0,0), actualOther.outerStride(), &actualOther.coeffRef(0,0), actualOther.outerStride(),
-            mat.data(), mat.outerStride(), actualAlpha);
+            mat.data(), mat.outerStride(), actualAlpha, blocking);
   }
 };
 
diff --git a/extern/Eigen3/Eigen/src/Core/products/SelfadjointRank2Update.h b/extern/Eigen3/Eigen/src/Core/products/SelfadjointRank2Update.h
index 8594a97cea3..2ae3641111c 100644
--- a/extern/Eigen3/Eigen/src/Core/products/SelfadjointRank2Update.h
+++ b/extern/Eigen3/Eigen/src/Core/products/SelfadjointRank2Update.h
@@ -79,11 +79,11 @@ SelfAdjointView<MatrixType,UpLo>& SelfAdjointView<MatrixType,UpLo>
   if (IsRowMajor)
     actualAlpha = numext::conj(actualAlpha);
 
-  internal::selfadjoint_rank2_update_selector<Scalar, Index,
-    typename internal::remove_all<typename internal::conj_expr_if<IsRowMajor ^ UBlasTraits::NeedToConjugate,_ActualUType>::type>::type,
-    typename internal::remove_all<typename internal::conj_expr_if<IsRowMajor ^ VBlasTraits::NeedToConjugate,_ActualVType>::type>::type,
+  typedef typename internal::remove_all<typename internal::conj_expr_if<IsRowMajor ^ UBlasTraits::NeedToConjugate,_ActualUType>::type>::type UType;
+  typedef typename internal::remove_all<typename internal::conj_expr_if<IsRowMajor ^ VBlasTraits::NeedToConjugate,_ActualVType>::type>::type VType;
+  internal::selfadjoint_rank2_update_selector<Scalar, Index, UType, VType,
     (IsRowMajor ? int(UpLo==Upper ? Lower : Upper) : UpLo)>
-    ::run(_expression().const_cast_derived().data(),_expression().outerStride(),actualU,actualV,actualAlpha);
+    ::run(_expression().const_cast_derived().data(),_expression().outerStride(),UType(actualU),VType(actualV),actualAlpha);
 
   return *this;
 }
diff --git a/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixMatrix.h b/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixMatrix.h
index 8110507b50b..f784507e77f 100644
--- a/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixMatrix.h
+++ b/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixMatrix.h
@@ -108,7 +108,7 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
     Index _rows, Index _cols, Index _depth,
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
-    Scalar* res,        Index resStride,
+    Scalar* _res,        Index resStride,
     const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
     // strip zeros
@@ -117,30 +117,42 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
     Index depth     = IsLower ? diagSize : _depth;
     Index cols      = _cols;
     
-    const_blas_data_mapper<Scalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-    const_blas_data_mapper<Scalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
+    typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
+    typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    RhsMapper rhs(_rhs,rhsStride);
+    ResMapper res(_res, resStride);
 
     Index kc = blocking.kc();                   // cache block size along the K direction
     Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
+    // The small panel size must not be larger than blocking size.
+    // Usually this should never be the case because SmallPanelWidth^2 is very small
+    // compared to L2 cache size, but let's be safe:
+    Index panelWidth = (std::min)(Index(SmallPanelWidth),(std::min)(kc,mc));
 
     std::size_t sizeA = kc*mc;
     std::size_t sizeB = kc*cols;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
 
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockW, sizeW, blocking.blockW());
 
-    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,LhsStorageOrder> triangularBuffer;
+    // To work around an "error: member reference base type 'Matrix<...>
+    // (Eigen::internal::constructor_without_unaligned_array_assert (*)())' is
+    // not a structure or union" compilation error in nvcc (tested V8.0.61),
+    // create a dummy internal::constructor_without_unaligned_array_assert
+    // object to pass to the Matrix constructor.
+    internal::constructor_without_unaligned_array_assert a;
+    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,LhsStorageOrder> triangularBuffer(a);
     triangularBuffer.setZero();
     if((Mode&ZeroDiag)==ZeroDiag)
       triangularBuffer.diagonal().setZero();
     else
       triangularBuffer.diagonal().setOnes();
 
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
+    gebp_kernel<Scalar, Scalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
+    gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr,RhsStorageOrder> pack_rhs;
 
     for(Index k2=IsLower ? depth : 0;
         IsLower ? k2>0 : k2<depth;
@@ -156,7 +168,7 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
         k2 = k2+actual_kc-kc;
       }
 
-      pack_rhs(blockB, &rhs(actual_k2,0), rhsStride, actual_kc, cols);
+      pack_rhs(blockB, rhs.getSubMapper(actual_k2,0), actual_kc, cols);
 
       // the selected lhs's panel has to be split in three different parts:
       //  1 - the part which is zero => skip it
@@ -167,9 +179,9 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
       if(IsLower || actual_k2<rows)
       {
         // for each small vertical panels of lhs
-        for (Index k1=0; k1<actual_kc; k1+=SmallPanelWidth)
+        for (Index k1=0; k1<actual_kc; k1+=panelWidth)
         {
-          Index actualPanelWidth = std::min<Index>(actual_kc-k1, SmallPanelWidth);
+          Index actualPanelWidth = std::min<Index>(actual_kc-k1, panelWidth);
           Index lengthTarget = IsLower ? actual_kc-k1-actualPanelWidth : k1;
           Index startBlock   = actual_k2+k1;
           Index blockBOffset = k1;
@@ -184,20 +196,22 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
             for (Index i=IsLower ? k+1 : 0; IsLower ? i<actualPanelWidth : i<k; ++i)
               triangularBuffer.coeffRef(i,k) = lhs(startBlock+i,startBlock+k);
           }
-          pack_lhs(blockA, triangularBuffer.data(), triangularBuffer.outerStride(), actualPanelWidth, actualPanelWidth);
+          pack_lhs(blockA, LhsMapper(triangularBuffer.data(), triangularBuffer.outerStride()), actualPanelWidth, actualPanelWidth);
 
-          gebp_kernel(res+startBlock, resStride, blockA, blockB, actualPanelWidth, actualPanelWidth, cols, alpha,
-                      actualPanelWidth, actual_kc, 0, blockBOffset, blockW);
+          gebp_kernel(res.getSubMapper(startBlock, 0), blockA, blockB,
+                      actualPanelWidth, actualPanelWidth, cols, alpha,
+                      actualPanelWidth, actual_kc, 0, blockBOffset);
 
           // GEBP with remaining micro panel
           if (lengthTarget>0)
           {
             Index startTarget  = IsLower ? actual_k2+k1+actualPanelWidth : actual_k2;
 
-            pack_lhs(blockA, &lhs(startTarget,startBlock), lhsStride, actualPanelWidth, lengthTarget);
+            pack_lhs(blockA, lhs.getSubMapper(startTarget,startBlock), actualPanelWidth, lengthTarget);
 
-            gebp_kernel(res+startTarget, resStride, blockA, blockB, lengthTarget, actualPanelWidth, cols, alpha,
-                        actualPanelWidth, actual_kc, 0, blockBOffset, blockW);
+            gebp_kernel(res.getSubMapper(startTarget, 0), blockA, blockB,
+                        lengthTarget, actualPanelWidth, cols, alpha,
+                        actualPanelWidth, actual_kc, 0, blockBOffset);
           }
         }
       }
@@ -208,10 +222,11 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
         for(Index i2=start; i2<end; i2+=mc)
         {
           const Index actual_mc = (std::min)(i2+mc,end)-i2;
-          gemm_pack_lhs<Scalar, Index, Traits::mr,Traits::LhsProgress, LhsStorageOrder,false>()
-            (blockA, &lhs(i2, actual_k2), lhsStride, actual_kc, actual_mc);
+          gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr,Traits::LhsProgress, LhsStorageOrder,false>()
+            (blockA, lhs.getSubMapper(i2, actual_k2), actual_kc, actual_mc);
 
-          gebp_kernel(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha, -1, -1, 0, 0, blockW);
+          gebp_kernel(res.getSubMapper(i2, 0), blockA, blockB, actual_mc,
+                      actual_kc, cols, alpha, -1, -1, 0, 0);
         }
       }
     }
@@ -249,40 +264,44 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
     Index _rows, Index _cols, Index _depth,
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
-    Scalar* res,        Index resStride,
+    Scalar* _res,        Index resStride,
     const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
+    const Index PacketBytes = packet_traits<Scalar>::size*sizeof(Scalar);
     // strip zeros
     Index diagSize  = (std::min)(_cols,_depth);
     Index rows      = _rows;
     Index depth     = IsLower ? _depth : diagSize;
     Index cols      = IsLower ? diagSize : _cols;
     
-    const_blas_data_mapper<Scalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-    const_blas_data_mapper<Scalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
+    typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
+    typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    RhsMapper rhs(_rhs,rhsStride);
+    ResMapper res(_res, resStride);
 
     Index kc = blocking.kc();                   // cache block size along the K direction
     Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
 
     std::size_t sizeA = kc*mc;
-    std::size_t sizeB = kc*cols;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
+    std::size_t sizeB = kc*cols+EIGEN_MAX_ALIGN_BYTES/sizeof(Scalar);
 
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockW, sizeW, blocking.blockW());
 
-    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,RhsStorageOrder> triangularBuffer;
+    internal::constructor_without_unaligned_array_assert a;
+    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,RhsStorageOrder> triangularBuffer(a);
     triangularBuffer.setZero();
     if((Mode&ZeroDiag)==ZeroDiag)
       triangularBuffer.diagonal().setZero();
     else
       triangularBuffer.diagonal().setOnes();
 
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder,false,true> pack_rhs_panel;
+    gebp_kernel<Scalar, Scalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
+    gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr,RhsStorageOrder> pack_rhs;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr,RhsStorageOrder,false,true> pack_rhs_panel;
 
     for(Index k2=IsLower ? 0 : depth;
         IsLower ? k2<depth  : k2>0;
@@ -304,8 +323,9 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
       Index ts = (IsLower && actual_k2>=cols) ? 0 : actual_kc;
 
       Scalar* geb = blockB+ts*ts;
+      geb = geb + internal::first_aligned<PacketBytes>(geb,PacketBytes/sizeof(Scalar));
 
-      pack_rhs(geb, &rhs(actual_k2,IsLower ? 0 : k2), rhsStride, actual_kc, rs);
+      pack_rhs(geb, rhs.getSubMapper(actual_k2,IsLower ? 0 : k2), actual_kc, rs);
 
       // pack the triangular part of the rhs padding the unrolled blocks with zeros
       if(ts>0)
@@ -318,7 +338,7 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
           Index panelLength = IsLower ? actual_kc-j2-actualPanelWidth : j2;
           // general part
           pack_rhs_panel(blockB+j2*actual_kc,
-                         &rhs(actual_k2+panelOffset, actual_j2), rhsStride,
+                         rhs.getSubMapper(actual_k2+panelOffset, actual_j2),
                          panelLength, actualPanelWidth,
                          actual_kc, panelOffset);
 
@@ -332,7 +352,7 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
           }
 
           pack_rhs_panel(blockB+j2*actual_kc,
-                         triangularBuffer.data(), triangularBuffer.outerStride(),
+                         RhsMapper(triangularBuffer.data(), triangularBuffer.outerStride()),
                          actualPanelWidth, actualPanelWidth,
                          actual_kc, j2);
         }
@@ -341,7 +361,7 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
       for (Index i2=0; i2<rows; i2+=mc)
       {
         const Index actual_mc = (std::min)(mc,rows-i2);
-        pack_lhs(blockA, &lhs(i2, actual_k2), lhsStride, actual_kc, actual_mc);
+        pack_lhs(blockA, lhs.getSubMapper(i2, actual_k2), actual_kc, actual_mc);
 
         // triangular kernel
         if(ts>0)
@@ -352,19 +372,18 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
             Index panelLength = IsLower ? actual_kc-j2 : j2+actualPanelWidth;
             Index blockOffset = IsLower ? j2 : 0;
 
-            gebp_kernel(res+i2+(actual_k2+j2)*resStride, resStride,
+            gebp_kernel(res.getSubMapper(i2, actual_k2 + j2),
                         blockA, blockB+j2*actual_kc,
                         actual_mc, panelLength, actualPanelWidth,
                         alpha,
                         actual_kc, actual_kc,  // strides
-                        blockOffset, blockOffset,// offsets
-                        blockW); // workspace
+                        blockOffset, blockOffset);// offsets
           }
         }
-        gebp_kernel(res+i2+(IsLower ? 0 : k2)*resStride, resStride,
+        gebp_kernel(res.getSubMapper(i2, IsLower ? 0 : k2),
                     blockA, geb, actual_mc, actual_kc, rs,
                     alpha,
-                    -1, -1, 0, 0, blockW);
+                    -1, -1, 0, 0);
       }
     }
   }
@@ -373,28 +392,31 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
 * Wrapper to product_triangular_matrix_matrix
 ***************************************************************************/
 
-template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
-struct traits<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false> >
-  : traits<ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>, Lhs, Rhs> >
-{};
-
 } // end namespace internal
 
+namespace internal {
 template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
-struct TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
-  : public ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>, Lhs, Rhs >
+struct triangular_product_impl<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(TriangularProduct)
-
-  TriangularProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
+  template<typename Dest> static void run(Dest& dst, const Lhs &a_lhs, const Rhs &a_rhs, const typename Dest::Scalar& alpha)
   {
-    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
-    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
+    typedef typename Lhs::Scalar  LhsScalar;
+    typedef typename Rhs::Scalar  RhsScalar;
+    typedef typename Dest::Scalar Scalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
+    
+    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
 
-    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
-                               * RhsBlasTraits::extractScalarFactor(m_rhs);
+    LhsScalar lhs_alpha = LhsBlasTraits::extractScalarFactor(a_lhs);
+    RhsScalar rhs_alpha = RhsBlasTraits::extractScalarFactor(a_rhs);
+    Scalar actualAlpha = alpha * lhs_alpha * rhs_alpha;
 
     typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,Scalar,Scalar,
               Lhs::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime, Lhs::MaxColsAtCompileTime,4> BlockingType;
@@ -405,23 +427,40 @@ struct TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
     Index stripedDepth = LhsIsTriangular ? ((!IsLower) ? lhs.cols() : (std::min)(lhs.cols(),lhs.rows()))
                                          : ((IsLower)  ? rhs.rows() : (std::min)(rhs.rows(),rhs.cols()));
 
-    BlockingType blocking(stripedRows, stripedCols, stripedDepth);
+    BlockingType blocking(stripedRows, stripedCols, stripedDepth, 1, false);
 
     internal::product_triangular_matrix_matrix<Scalar, Index,
       Mode, LhsIsTriangular,
-      (internal::traits<_ActualLhsType>::Flags&RowMajorBit) ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
-      (internal::traits<_ActualRhsType>::Flags&RowMajorBit) ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
+      (internal::traits<ActualLhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
+      (internal::traits<ActualRhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
       (internal::traits<Dest          >::Flags&RowMajorBit) ? RowMajor : ColMajor>
       ::run(
         stripedRows, stripedCols, stripedDepth,   // sizes
-        &lhs.coeffRef(0,0),    lhs.outerStride(), // lhs info
-        &rhs.coeffRef(0,0),    rhs.outerStride(), // rhs info
+        &lhs.coeffRef(0,0), lhs.outerStride(),    // lhs info
+        &rhs.coeffRef(0,0), rhs.outerStride(),    // rhs info
         &dst.coeffRef(0,0), dst.outerStride(),    // result info
         actualAlpha, blocking
       );
+
+    // Apply correction if the diagonal is unit and a scalar factor was nested:
+    if ((Mode&UnitDiag)==UnitDiag)
+    {
+      if (LhsIsTriangular && lhs_alpha!=LhsScalar(1))
+      {
+        Index diagSize = (std::min)(lhs.rows(),lhs.cols());
+        dst.topRows(diagSize) -= ((lhs_alpha-LhsScalar(1))*a_rhs).topRows(diagSize);
+      }
+      else if ((!LhsIsTriangular) && rhs_alpha!=RhsScalar(1))
+      {
+        Index diagSize = (std::min)(rhs.rows(),rhs.cols());
+        dst.leftCols(diagSize) -= (rhs_alpha-RhsScalar(1))*a_lhs.leftCols(diagSize);
+      }
+    }
   }
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_TRIANGULAR_MATRIX_MATRIX_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixMatrix_BLAS.h b/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixMatrix_BLAS.h
new file mode 100644
index 00000000000..a25197ab017
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixMatrix_BLAS.h
@@ -0,0 +1,315 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   Triangular matrix * matrix product functionality based on ?TRMM.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_TRIANGULAR_MATRIX_MATRIX_BLAS_H
+#define EIGEN_TRIANGULAR_MATRIX_MATRIX_BLAS_H
+
+namespace Eigen { 
+
+namespace internal {
+
+
+template <typename Scalar, typename Index,
+          int Mode, bool LhsIsTriangular,
+          int LhsStorageOrder, bool ConjugateLhs,
+          int RhsStorageOrder, bool ConjugateRhs,
+          int ResStorageOrder>
+struct product_triangular_matrix_matrix_trmm :
+       product_triangular_matrix_matrix<Scalar,Index,Mode,
+          LhsIsTriangular,LhsStorageOrder,ConjugateLhs,
+          RhsStorageOrder, ConjugateRhs, ResStorageOrder, BuiltIn> {};
+
+
+// try to go to BLAS specialization
+#define EIGEN_BLAS_TRMM_SPECIALIZE(Scalar, LhsIsTriangular) \
+template <typename Index, int Mode, \
+          int LhsStorageOrder, bool ConjugateLhs, \
+          int RhsStorageOrder, bool ConjugateRhs> \
+struct product_triangular_matrix_matrix<Scalar,Index, Mode, LhsIsTriangular, \
+           LhsStorageOrder,ConjugateLhs, RhsStorageOrder,ConjugateRhs,ColMajor,Specialized> { \
+  static inline void run(Index _rows, Index _cols, Index _depth, const Scalar* _lhs, Index lhsStride,\
+    const Scalar* _rhs, Index rhsStride, Scalar* res, Index resStride, Scalar alpha, level3_blocking<Scalar,Scalar>& blocking) { \
+      product_triangular_matrix_matrix_trmm<Scalar,Index,Mode, \
+        LhsIsTriangular,LhsStorageOrder,ConjugateLhs, \
+        RhsStorageOrder, ConjugateRhs, ColMajor>::run( \
+        _rows, _cols, _depth, _lhs, lhsStride, _rhs, rhsStride, res, resStride, alpha, blocking); \
+  } \
+};
+
+EIGEN_BLAS_TRMM_SPECIALIZE(double, true)
+EIGEN_BLAS_TRMM_SPECIALIZE(double, false)
+EIGEN_BLAS_TRMM_SPECIALIZE(dcomplex, true)
+EIGEN_BLAS_TRMM_SPECIALIZE(dcomplex, false)
+EIGEN_BLAS_TRMM_SPECIALIZE(float, true)
+EIGEN_BLAS_TRMM_SPECIALIZE(float, false)
+EIGEN_BLAS_TRMM_SPECIALIZE(scomplex, true)
+EIGEN_BLAS_TRMM_SPECIALIZE(scomplex, false)
+
+// implements col-major += alpha * op(triangular) * op(general)
+#define EIGEN_BLAS_TRMM_L(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
+template <typename Index, int Mode, \
+          int LhsStorageOrder, bool ConjugateLhs, \
+          int RhsStorageOrder, bool ConjugateRhs> \
+struct product_triangular_matrix_matrix_trmm<EIGTYPE,Index,Mode,true, \
+         LhsStorageOrder,ConjugateLhs,RhsStorageOrder,ConjugateRhs,ColMajor> \
+{ \
+  enum { \
+    IsLower = (Mode&Lower) == Lower, \
+    SetDiag = (Mode&(ZeroDiag|UnitDiag)) ? 0 : 1, \
+    IsUnitDiag  = (Mode&UnitDiag) ? 1 : 0, \
+    IsZeroDiag  = (Mode&ZeroDiag) ? 1 : 0, \
+    LowUp = IsLower ? Lower : Upper, \
+    conjA = ((LhsStorageOrder==ColMajor) && ConjugateLhs) ? 1 : 0 \
+  }; \
+\
+  static void run( \
+    Index _rows, Index _cols, Index _depth, \
+    const EIGTYPE* _lhs, Index lhsStride, \
+    const EIGTYPE* _rhs, Index rhsStride, \
+    EIGTYPE* res,        Index resStride, \
+    EIGTYPE alpha, level3_blocking<EIGTYPE,EIGTYPE>& blocking) \
+  { \
+   Index diagSize  = (std::min)(_rows,_depth); \
+   Index rows      = IsLower ? _rows : diagSize; \
+   Index depth     = IsLower ? diagSize : _depth; \
+   Index cols      = _cols; \
+\
+   typedef Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder> MatrixLhs; \
+   typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs; \
+\
+/* Non-square case - doesn't fit to BLAS ?TRMM. Fall to default triangular product or call BLAS ?GEMM*/ \
+   if (rows != depth) { \
+\
+     /* FIXME handle mkl_domain_get_max_threads */ \
+     /*int nthr = mkl_domain_get_max_threads(EIGEN_BLAS_DOMAIN_BLAS);*/ int nthr = 1;\
+\
+     if (((nthr==1) && (((std::max)(rows,depth)-diagSize)/(double)diagSize < 0.5))) { \
+     /* Most likely no benefit to call TRMM or GEMM from BLAS */ \
+       product_triangular_matrix_matrix<EIGTYPE,Index,Mode,true, \
+       LhsStorageOrder,ConjugateLhs, RhsStorageOrder, ConjugateRhs, ColMajor, BuiltIn>::run( \
+           _rows, _cols, _depth, _lhs, lhsStride, _rhs, rhsStride, res, resStride, alpha, blocking); \
+     /*std::cout << "TRMM_L: A is not square! Go to Eigen TRMM implementation!\n";*/ \
+     } else { \
+     /* Make sense to call GEMM */ \
+       Map<const MatrixLhs, 0, OuterStride<> > lhsMap(_lhs,rows,depth,OuterStride<>(lhsStride)); \
+       MatrixLhs aa_tmp=lhsMap.template triangularView<Mode>(); \
+       BlasIndex aStride = convert_index<BlasIndex>(aa_tmp.outerStride()); \
+       gemm_blocking_space<ColMajor,EIGTYPE,EIGTYPE,Dynamic,Dynamic,Dynamic> gemm_blocking(_rows,_cols,_depth, 1, true); \
+       general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor>::run( \
+       rows, cols, depth, aa_tmp.data(), aStride, _rhs, rhsStride, res, resStride, alpha, gemm_blocking, 0); \
+\
+     /*std::cout << "TRMM_L: A is not square! Go to BLAS GEMM implementation! " << nthr<<" \n";*/ \
+     } \
+     return; \
+   } \
+   char side = 'L', transa, uplo, diag = 'N'; \
+   EIGTYPE *b; \
+   const EIGTYPE *a; \
+   BlasIndex m, n, lda, ldb; \
+\
+/* Set m, n */ \
+   m = convert_index<BlasIndex>(diagSize); \
+   n = convert_index<BlasIndex>(cols); \
+\
+/* Set trans */ \
+   transa = (LhsStorageOrder==RowMajor) ? ((ConjugateLhs) ? 'C' : 'T') : 'N'; \
+\
+/* Set b, ldb */ \
+   Map<const MatrixRhs, 0, OuterStride<> > rhs(_rhs,depth,cols,OuterStride<>(rhsStride)); \
+   MatrixX##EIGPREFIX b_tmp; \
+\
+   if (ConjugateRhs) b_tmp = rhs.conjugate(); else b_tmp = rhs; \
+   b = b_tmp.data(); \
+   ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
+\
+/* Set uplo */ \
+   uplo = IsLower ? 'L' : 'U'; \
+   if (LhsStorageOrder==RowMajor) uplo = (uplo == 'L') ? 'U' : 'L'; \
+/* Set a, lda */ \
+   Map<const MatrixLhs, 0, OuterStride<> > lhs(_lhs,rows,depth,OuterStride<>(lhsStride)); \
+   MatrixLhs a_tmp; \
+\
+   if ((conjA!=0) || (SetDiag==0)) { \
+     if (conjA) a_tmp = lhs.conjugate(); else a_tmp = lhs; \
+     if (IsZeroDiag) \
+       a_tmp.diagonal().setZero(); \
+     else if (IsUnitDiag) \
+       a_tmp.diagonal().setOnes();\
+     a = a_tmp.data(); \
+     lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
+   } else { \
+     a = _lhs; \
+     lda = convert_index<BlasIndex>(lhsStride); \
+   } \
+   /*std::cout << "TRMM_L: A is square! Go to BLAS TRMM implementation! \n";*/ \
+/* call ?trmm*/ \
+   BLASFUNC(&side, &uplo, &transa, &diag, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (BLASTYPE*)b, &ldb); \
+\
+/* Add op(a_triangular)*b into res*/ \
+   Map<MatrixX##EIGPREFIX, 0, OuterStride<> > res_tmp(res,rows,cols,OuterStride<>(resStride)); \
+   res_tmp=res_tmp+b_tmp; \
+  } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRMM_L(double, double, d, dtrmm)
+EIGEN_BLAS_TRMM_L(dcomplex, MKL_Complex16, cd, ztrmm)
+EIGEN_BLAS_TRMM_L(float, float, f, strmm)
+EIGEN_BLAS_TRMM_L(scomplex, MKL_Complex8, cf, ctrmm)
+#else
+EIGEN_BLAS_TRMM_L(double, double, d, dtrmm_)
+EIGEN_BLAS_TRMM_L(dcomplex, double, cd, ztrmm_)
+EIGEN_BLAS_TRMM_L(float, float, f, strmm_)
+EIGEN_BLAS_TRMM_L(scomplex, float, cf, ctrmm_)
+#endif
+
+// implements col-major += alpha * op(general) * op(triangular)
+#define EIGEN_BLAS_TRMM_R(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
+template <typename Index, int Mode, \
+          int LhsStorageOrder, bool ConjugateLhs, \
+          int RhsStorageOrder, bool ConjugateRhs> \
+struct product_triangular_matrix_matrix_trmm<EIGTYPE,Index,Mode,false, \
+         LhsStorageOrder,ConjugateLhs,RhsStorageOrder,ConjugateRhs,ColMajor> \
+{ \
+  enum { \
+    IsLower = (Mode&Lower) == Lower, \
+    SetDiag = (Mode&(ZeroDiag|UnitDiag)) ? 0 : 1, \
+    IsUnitDiag  = (Mode&UnitDiag) ? 1 : 0, \
+    IsZeroDiag  = (Mode&ZeroDiag) ? 1 : 0, \
+    LowUp = IsLower ? Lower : Upper, \
+    conjA = ((RhsStorageOrder==ColMajor) && ConjugateRhs) ? 1 : 0 \
+  }; \
+\
+  static void run( \
+    Index _rows, Index _cols, Index _depth, \
+    const EIGTYPE* _lhs, Index lhsStride, \
+    const EIGTYPE* _rhs, Index rhsStride, \
+    EIGTYPE* res,        Index resStride, \
+    EIGTYPE alpha, level3_blocking<EIGTYPE,EIGTYPE>& blocking) \
+  { \
+   Index diagSize  = (std::min)(_cols,_depth); \
+   Index rows      = _rows; \
+   Index depth     = IsLower ? _depth : diagSize; \
+   Index cols      = IsLower ? diagSize : _cols; \
+\
+   typedef Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder> MatrixLhs; \
+   typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs; \
+\
+/* Non-square case - doesn't fit to BLAS ?TRMM. Fall to default triangular product or call BLAS ?GEMM*/ \
+   if (cols != depth) { \
+\
+     int nthr = 1 /*mkl_domain_get_max_threads(EIGEN_BLAS_DOMAIN_BLAS)*/; \
+\
+     if ((nthr==1) && (((std::max)(cols,depth)-diagSize)/(double)diagSize < 0.5)) { \
+     /* Most likely no benefit to call TRMM or GEMM from BLAS*/ \
+       product_triangular_matrix_matrix<EIGTYPE,Index,Mode,false, \
+       LhsStorageOrder,ConjugateLhs, RhsStorageOrder, ConjugateRhs, ColMajor, BuiltIn>::run( \
+           _rows, _cols, _depth, _lhs, lhsStride, _rhs, rhsStride, res, resStride, alpha, blocking); \
+       /*std::cout << "TRMM_R: A is not square! Go to Eigen TRMM implementation!\n";*/ \
+     } else { \
+     /* Make sense to call GEMM */ \
+       Map<const MatrixRhs, 0, OuterStride<> > rhsMap(_rhs,depth,cols, OuterStride<>(rhsStride)); \
+       MatrixRhs aa_tmp=rhsMap.template triangularView<Mode>(); \
+       BlasIndex aStride = convert_index<BlasIndex>(aa_tmp.outerStride()); \
+       gemm_blocking_space<ColMajor,EIGTYPE,EIGTYPE,Dynamic,Dynamic,Dynamic> gemm_blocking(_rows,_cols,_depth, 1, true); \
+       general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor>::run( \
+       rows, cols, depth, _lhs, lhsStride, aa_tmp.data(), aStride, res, resStride, alpha, gemm_blocking, 0); \
+\
+     /*std::cout << "TRMM_R: A is not square! Go to BLAS GEMM implementation! " << nthr<<" \n";*/ \
+     } \
+     return; \
+   } \
+   char side = 'R', transa, uplo, diag = 'N'; \
+   EIGTYPE *b; \
+   const EIGTYPE *a; \
+   BlasIndex m, n, lda, ldb; \
+\
+/* Set m, n */ \
+   m = convert_index<BlasIndex>(rows); \
+   n = convert_index<BlasIndex>(diagSize); \
+\
+/* Set trans */ \
+   transa = (RhsStorageOrder==RowMajor) ? ((ConjugateRhs) ? 'C' : 'T') : 'N'; \
+\
+/* Set b, ldb */ \
+   Map<const MatrixLhs, 0, OuterStride<> > lhs(_lhs,rows,depth,OuterStride<>(lhsStride)); \
+   MatrixX##EIGPREFIX b_tmp; \
+\
+   if (ConjugateLhs) b_tmp = lhs.conjugate(); else b_tmp = lhs; \
+   b = b_tmp.data(); \
+   ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
+\
+/* Set uplo */ \
+   uplo = IsLower ? 'L' : 'U'; \
+   if (RhsStorageOrder==RowMajor) uplo = (uplo == 'L') ? 'U' : 'L'; \
+/* Set a, lda */ \
+   Map<const MatrixRhs, 0, OuterStride<> > rhs(_rhs,depth,cols, OuterStride<>(rhsStride)); \
+   MatrixRhs a_tmp; \
+\
+   if ((conjA!=0) || (SetDiag==0)) { \
+     if (conjA) a_tmp = rhs.conjugate(); else a_tmp = rhs; \
+     if (IsZeroDiag) \
+       a_tmp.diagonal().setZero(); \
+     else if (IsUnitDiag) \
+       a_tmp.diagonal().setOnes();\
+     a = a_tmp.data(); \
+     lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
+   } else { \
+     a = _rhs; \
+     lda = convert_index<BlasIndex>(rhsStride); \
+   } \
+   /*std::cout << "TRMM_R: A is square! Go to BLAS TRMM implementation! \n";*/ \
+/* call ?trmm*/ \
+   BLASFUNC(&side, &uplo, &transa, &diag, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (BLASTYPE*)b, &ldb); \
+\
+/* Add op(a_triangular)*b into res*/ \
+   Map<MatrixX##EIGPREFIX, 0, OuterStride<> > res_tmp(res,rows,cols,OuterStride<>(resStride)); \
+   res_tmp=res_tmp+b_tmp; \
+  } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRMM_R(double, double, d, dtrmm)
+EIGEN_BLAS_TRMM_R(dcomplex, MKL_Complex16, cd, ztrmm)
+EIGEN_BLAS_TRMM_R(float, float, f, strmm)
+EIGEN_BLAS_TRMM_R(scomplex, MKL_Complex8, cf, ctrmm)
+#else
+EIGEN_BLAS_TRMM_R(double, double, d, dtrmm_)
+EIGEN_BLAS_TRMM_R(dcomplex, double, cd, ztrmm_)
+EIGEN_BLAS_TRMM_R(float, float, f, strmm_)
+EIGEN_BLAS_TRMM_R(scomplex, float, cf, ctrmm_)
+#endif
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TRIANGULAR_MATRIX_MATRIX_BLAS_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixVector.h b/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixVector.h
index 6117d5a8262..76bfa159ced 100644
--- a/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixVector.h
+++ b/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixVector.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_TRIANGULARMATRIXVECTOR_H
 #define EIGEN_TRIANGULARMATRIXVECTOR_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -20,20 +20,20 @@ struct triangular_matrix_vector_product;
 template<typename Index, int Mode, typename LhsScalar, bool ConjLhs, typename RhsScalar, bool ConjRhs, int Version>
 struct triangular_matrix_vector_product<Index,Mode,LhsScalar,ConjLhs,RhsScalar,ConjRhs,ColMajor,Version>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   enum {
     IsLower = ((Mode&Lower)==Lower),
     HasUnitDiag = (Mode & UnitDiag)==UnitDiag,
     HasZeroDiag = (Mode & ZeroDiag)==ZeroDiag
   };
   static EIGEN_DONT_INLINE  void run(Index _rows, Index _cols, const LhsScalar* _lhs, Index lhsStride,
-                                     const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, const ResScalar& alpha);
+                                     const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, const RhsScalar& alpha);
 };
 
 template<typename Index, int Mode, typename LhsScalar, bool ConjLhs, typename RhsScalar, bool ConjRhs, int Version>
 EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,ConjLhs,RhsScalar,ConjRhs,ColMajor,Version>
   ::run(Index _rows, Index _cols, const LhsScalar* _lhs, Index lhsStride,
-        const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, const ResScalar& alpha)
+        const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, const RhsScalar& alpha)
   {
     static const Index PanelWidth = EIGEN_TUNE_TRIANGULAR_PANEL_WIDTH;
     Index size = (std::min)(_rows,_cols);
@@ -43,7 +43,7 @@ EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,Con
     typedef Map<const Matrix<LhsScalar,Dynamic,Dynamic,ColMajor>, 0, OuterStride<> > LhsMap;
     const LhsMap lhs(_lhs,rows,cols,OuterStride<>(lhsStride));
     typename conj_expr_if<ConjLhs,LhsMap>::type cjLhs(lhs);
-    
+
     typedef Map<const Matrix<RhsScalar,Dynamic,1>, 0, InnerStride<> > RhsMap;
     const RhsMap rhs(_rhs,cols,InnerStride<>(rhsIncr));
     typename conj_expr_if<ConjRhs,RhsMap>::type cjRhs(rhs);
@@ -51,6 +51,9 @@ EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,Con
     typedef Map<Matrix<ResScalar,Dynamic,1> > ResMap;
     ResMap res(_res,rows);
 
+    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
+
     for (Index pi=0; pi<size; pi+=PanelWidth)
     {
       Index actualPanelWidth = (std::min)(PanelWidth, size-pi);
@@ -68,19 +71,19 @@ EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,Con
       if (r>0)
       {
         Index s = IsLower ? pi+actualPanelWidth : 0;
-        general_matrix_vector_product<Index,LhsScalar,ColMajor,ConjLhs,RhsScalar,ConjRhs,BuiltIn>::run(
+        general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,ConjLhs,RhsScalar,RhsMapper,ConjRhs,BuiltIn>::run(
             r, actualPanelWidth,
-            &lhs.coeffRef(s,pi), lhsStride,
-            &rhs.coeffRef(pi), rhsIncr,
+            LhsMapper(&lhs.coeffRef(s,pi), lhsStride),
+            RhsMapper(&rhs.coeffRef(pi), rhsIncr),
             &res.coeffRef(s), resIncr, alpha);
       }
     }
     if((!IsLower) && cols>size)
     {
-      general_matrix_vector_product<Index,LhsScalar,ColMajor,ConjLhs,RhsScalar,ConjRhs>::run(
+      general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,ConjLhs,RhsScalar,RhsMapper,ConjRhs>::run(
           rows, cols-size,
-          &lhs.coeffRef(0,size), lhsStride,
-          &rhs.coeffRef(size), rhsIncr,
+          LhsMapper(&lhs.coeffRef(0,size), lhsStride),
+          RhsMapper(&rhs.coeffRef(size), rhsIncr),
           _res, resIncr, alpha);
     }
   }
@@ -88,7 +91,7 @@ EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,Con
 template<typename Index, int Mode, typename LhsScalar, bool ConjLhs, typename RhsScalar, bool ConjRhs,int Version>
 struct triangular_matrix_vector_product<Index,Mode,LhsScalar,ConjLhs,RhsScalar,ConjRhs,RowMajor,Version>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   enum {
     IsLower = ((Mode&Lower)==Lower),
     HasUnitDiag = (Mode & UnitDiag)==UnitDiag,
@@ -118,7 +121,10 @@ EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,Con
 
     typedef Map<Matrix<ResScalar,Dynamic,1>, 0, InnerStride<> > ResMap;
     ResMap res(_res,rows,InnerStride<>(resIncr));
-    
+
+    typedef const_blas_data_mapper<LhsScalar,Index,RowMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
+
     for (Index pi=0; pi<diagSize; pi+=PanelWidth)
     {
       Index actualPanelWidth = (std::min)(PanelWidth, diagSize-pi);
@@ -136,19 +142,19 @@ EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,Con
       if (r>0)
       {
         Index s = IsLower ? 0 : pi + actualPanelWidth;
-        general_matrix_vector_product<Index,LhsScalar,RowMajor,ConjLhs,RhsScalar,ConjRhs,BuiltIn>::run(
+        general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,ConjLhs,RhsScalar,RhsMapper,ConjRhs,BuiltIn>::run(
             actualPanelWidth, r,
-            &lhs.coeffRef(pi,s), lhsStride,
-            &rhs.coeffRef(s), rhsIncr,
+            LhsMapper(&lhs.coeffRef(pi,s), lhsStride),
+            RhsMapper(&rhs.coeffRef(s), rhsIncr),
             &res.coeffRef(pi), resIncr, alpha);
       }
     }
     if(IsLower && rows>diagSize)
     {
-      general_matrix_vector_product<Index,LhsScalar,RowMajor,ConjLhs,RhsScalar,ConjRhs>::run(
+      general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,ConjLhs,RhsScalar,RhsMapper,ConjRhs>::run(
             rows-diagSize, cols,
-            &lhs.coeffRef(diagSize,0), lhsStride,
-            &rhs.coeffRef(0), rhsIncr,
+            LhsMapper(&lhs.coeffRef(diagSize,0), lhsStride),
+            RhsMapper(&rhs.coeffRef(0), rhsIncr),
             &res.coeffRef(diagSize), resIncr, alpha);
     }
   }
@@ -157,83 +163,67 @@ EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,Con
 * Wrapper to product_triangular_vector
 ***************************************************************************/
 
-template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
-struct traits<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,true> >
- : traits<ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,true>, Lhs, Rhs> >
-{};
-
-template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
-struct traits<TriangularProduct<Mode,LhsIsTriangular,Lhs,true,Rhs,false> >
- : traits<ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,true,Rhs,false>, Lhs, Rhs> >
-{};
-
-
-template<int StorageOrder>
+template<int Mode,int StorageOrder>
 struct trmv_selector;
 
 } // end namespace internal
 
+namespace internal {
+
 template<int Mode, typename Lhs, typename Rhs>
-struct TriangularProduct<Mode,true,Lhs,false,Rhs,true>
-  : public ProductBase<TriangularProduct<Mode,true,Lhs,false,Rhs,true>, Lhs, Rhs >
+struct triangular_product_impl<Mode,true,Lhs,false,Rhs,true>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(TriangularProduct)
-
-  TriangularProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
+  template<typename Dest> static void run(Dest& dst, const Lhs &lhs, const Rhs &rhs, const typename Dest::Scalar& alpha)
   {
-    eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
+    eigen_assert(dst.rows()==lhs.rows() && dst.cols()==rhs.cols());
   
-    internal::trmv_selector<(int(internal::traits<Lhs>::Flags)&RowMajorBit) ? RowMajor : ColMajor>::run(*this, dst, alpha);
+    internal::trmv_selector<Mode,(int(internal::traits<Lhs>::Flags)&RowMajorBit) ? RowMajor : ColMajor>::run(lhs, rhs, dst, alpha);
   }
 };
 
 template<int Mode, typename Lhs, typename Rhs>
-struct TriangularProduct<Mode,false,Lhs,true,Rhs,false>
-  : public ProductBase<TriangularProduct<Mode,false,Lhs,true,Rhs,false>, Lhs, Rhs >
+struct triangular_product_impl<Mode,false,Lhs,true,Rhs,false>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(TriangularProduct)
-
-  TriangularProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
+  template<typename Dest> static void run(Dest& dst, const Lhs &lhs, const Rhs &rhs, const typename Dest::Scalar& alpha)
   {
-    eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
+    eigen_assert(dst.rows()==lhs.rows() && dst.cols()==rhs.cols());
 
-    typedef TriangularProduct<(Mode & (UnitDiag|ZeroDiag)) | ((Mode & Lower) ? Upper : Lower),true,Transpose<const Rhs>,false,Transpose<const Lhs>,true> TriangularProductTranspose;
     Transpose<Dest> dstT(dst);
-    internal::trmv_selector<(int(internal::traits<Rhs>::Flags)&RowMajorBit) ? ColMajor : RowMajor>::run(
-      TriangularProductTranspose(m_rhs.transpose(),m_lhs.transpose()), dstT, alpha);
+    internal::trmv_selector<(Mode & (UnitDiag|ZeroDiag)) | ((Mode & Lower) ? Upper : Lower),
+                            (int(internal::traits<Rhs>::Flags)&RowMajorBit) ? ColMajor : RowMajor>
+            ::run(rhs.transpose(),lhs.transpose(), dstT, alpha);
   }
 };
 
+} // end namespace internal
+
 namespace internal {
 
 // TODO: find a way to factorize this piece of code with gemv_selector since the logic is exactly the same.
   
-template<> struct trmv_selector<ColMajor>
+template<int Mode> struct trmv_selector<Mode,ColMajor>
 {
-  template<int Mode, typename Lhs, typename Rhs, typename Dest>
-  static void run(const TriangularProduct<Mode,true,Lhs,false,Rhs,true>& prod, Dest& dest, const typename TriangularProduct<Mode,true,Lhs,false,Rhs,true>::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef TriangularProduct<Mode,true,Lhs,false,Rhs,true> ProductType;
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::LhsScalar   LhsScalar;
-    typedef typename ProductType::RhsScalar   RhsScalar;
-    typedef typename ProductType::Scalar      ResScalar;
-    typedef typename ProductType::RealScalar  RealScalar;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
-    typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
-
-    typename internal::add_const_on_value_type<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
-    typename internal::add_const_on_value_type<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
-
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
+    typedef typename Lhs::Scalar      LhsScalar;
+    typedef typename Rhs::Scalar      RhsScalar;
+    typedef typename Dest::Scalar     ResScalar;
+    typedef typename Dest::RealScalar RealScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    
+    typedef Map<Matrix<ResScalar,Dynamic,1>, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits<ResScalar>::size)> MappedDest;
+
+    typename internal::add_const_on_value_type<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
+
+    LhsScalar lhs_alpha = LhsBlasTraits::extractScalarFactor(lhs);
+    RhsScalar rhs_alpha = RhsBlasTraits::extractScalarFactor(rhs);
+    ResScalar actualAlpha = alpha * lhs_alpha * rhs_alpha;
 
     enum {
       // FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
@@ -247,7 +237,7 @@ template<> struct trmv_selector<ColMajor>
 
     bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
     bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
-    
+
     RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
 
     ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
@@ -267,7 +257,7 @@ template<> struct trmv_selector<ColMajor>
       else
         MappedDest(actualDestPtr, dest.size()) = dest;
     }
-    
+
     internal::triangular_matrix_vector_product
       <Index,Mode,
        LhsScalar, LhsBlasTraits::NeedToConjugate,
@@ -285,36 +275,42 @@ template<> struct trmv_selector<ColMajor>
       else
         dest = MappedDest(actualDestPtr, dest.size());
     }
+
+    if ( ((Mode&UnitDiag)==UnitDiag) && (lhs_alpha!=LhsScalar(1)) )
+    {
+      Index diagSize = (std::min)(lhs.rows(),lhs.cols());
+      dest.head(diagSize) -= (lhs_alpha-LhsScalar(1))*rhs.head(diagSize);
+    }
   }
 };
 
-template<> struct trmv_selector<RowMajor>
+template<int Mode> struct trmv_selector<Mode,RowMajor>
 {
-  template<int Mode, typename Lhs, typename Rhs, typename Dest>
-  static void run(const TriangularProduct<Mode,true,Lhs,false,Rhs,true>& prod, Dest& dest, const typename TriangularProduct<Mode,true,Lhs,false,Rhs,true>::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef TriangularProduct<Mode,true,Lhs,false,Rhs,true> ProductType;
-    typedef typename ProductType::LhsScalar LhsScalar;
-    typedef typename ProductType::RhsScalar RhsScalar;
-    typedef typename ProductType::Scalar    ResScalar;
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::_ActualRhsType _ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
-
-    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
-    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
-
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
+    typedef typename Lhs::Scalar      LhsScalar;
+    typedef typename Rhs::Scalar      RhsScalar;
+    typedef typename Dest::Scalar     ResScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
+
+    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
+    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
+
+    LhsScalar lhs_alpha = LhsBlasTraits::extractScalarFactor(lhs);
+    RhsScalar rhs_alpha = RhsBlasTraits::extractScalarFactor(rhs);
+    ResScalar actualAlpha = alpha * lhs_alpha * rhs_alpha;
 
     enum {
-      DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1
+      DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1
     };
 
-    gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
+    gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
 
     ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
         DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
@@ -322,12 +318,12 @@ template<> struct trmv_selector<RowMajor>
     if(!DirectlyUseRhs)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = actualRhs.size();
+      Index size = actualRhs.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
-      Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
+      Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
     }
-    
+
     internal::triangular_matrix_vector_product
       <Index,Mode,
        LhsScalar, LhsBlasTraits::NeedToConjugate,
@@ -338,6 +334,12 @@ template<> struct trmv_selector<RowMajor>
             actualRhsPtr,1,
             dest.data(),dest.innerStride(),
             actualAlpha);
+
+    if ( ((Mode&UnitDiag)==UnitDiag) && (lhs_alpha!=LhsScalar(1)) )
+    {
+      Index diagSize = (std::min)(lhs.rows(),lhs.cols());
+      dest.head(diagSize) -= (lhs_alpha-LhsScalar(1))*rhs.head(diagSize);
+    }
   }
 };
 
diff --git a/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixVector_BLAS.h b/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixVector_BLAS.h
new file mode 100644
index 00000000000..3d47a2b94cb
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/products/TriangularMatrixVector_BLAS.h
@@ -0,0 +1,255 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   Triangular matrix-vector product functionality based on ?TRMV.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_TRIANGULAR_MATRIX_VECTOR_BLAS_H
+#define EIGEN_TRIANGULAR_MATRIX_VECTOR_BLAS_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/**********************************************************************
+* This file implements triangular matrix-vector multiplication using BLAS
+**********************************************************************/
+
+// trmv/hemv specialization
+
+template<typename Index, int Mode, typename LhsScalar, bool ConjLhs, typename RhsScalar, bool ConjRhs, int StorageOrder>
+struct triangular_matrix_vector_product_trmv :
+  triangular_matrix_vector_product<Index,Mode,LhsScalar,ConjLhs,RhsScalar,ConjRhs,StorageOrder,BuiltIn> {};
+
+#define EIGEN_BLAS_TRMV_SPECIALIZE(Scalar) \
+template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \
+struct triangular_matrix_vector_product<Index,Mode,Scalar,ConjLhs,Scalar,ConjRhs,ColMajor,Specialized> { \
+ static void run(Index _rows, Index _cols, const Scalar* _lhs, Index lhsStride, \
+                                     const Scalar* _rhs, Index rhsIncr, Scalar* _res, Index resIncr, Scalar alpha) { \
+      triangular_matrix_vector_product_trmv<Index,Mode,Scalar,ConjLhs,Scalar,ConjRhs,ColMajor>::run( \
+        _rows, _cols, _lhs, lhsStride, _rhs, rhsIncr, _res, resIncr, alpha); \
+  } \
+}; \
+template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \
+struct triangular_matrix_vector_product<Index,Mode,Scalar,ConjLhs,Scalar,ConjRhs,RowMajor,Specialized> { \
+ static void run(Index _rows, Index _cols, const Scalar* _lhs, Index lhsStride, \
+                                     const Scalar* _rhs, Index rhsIncr, Scalar* _res, Index resIncr, Scalar alpha) { \
+      triangular_matrix_vector_product_trmv<Index,Mode,Scalar,ConjLhs,Scalar,ConjRhs,RowMajor>::run( \
+        _rows, _cols, _lhs, lhsStride, _rhs, rhsIncr, _res, resIncr, alpha); \
+  } \
+};
+
+EIGEN_BLAS_TRMV_SPECIALIZE(double)
+EIGEN_BLAS_TRMV_SPECIALIZE(float)
+EIGEN_BLAS_TRMV_SPECIALIZE(dcomplex)
+EIGEN_BLAS_TRMV_SPECIALIZE(scomplex)
+
+// implements col-major: res += alpha * op(triangular) * vector
+#define EIGEN_BLAS_TRMV_CM(EIGTYPE, BLASTYPE, EIGPREFIX, BLASPREFIX, BLASPOSTFIX) \
+template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \
+struct triangular_matrix_vector_product_trmv<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,ColMajor> { \
+  enum { \
+    IsLower = (Mode&Lower) == Lower, \
+    SetDiag = (Mode&(ZeroDiag|UnitDiag)) ? 0 : 1, \
+    IsUnitDiag  = (Mode&UnitDiag) ? 1 : 0, \
+    IsZeroDiag  = (Mode&ZeroDiag) ? 1 : 0, \
+    LowUp = IsLower ? Lower : Upper \
+  }; \
+ static void run(Index _rows, Index _cols, const EIGTYPE* _lhs, Index lhsStride, \
+                 const EIGTYPE* _rhs, Index rhsIncr, EIGTYPE* _res, Index resIncr, EIGTYPE alpha) \
+ { \
+   if (ConjLhs || IsZeroDiag) { \
+     triangular_matrix_vector_product<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,ColMajor,BuiltIn>::run( \
+       _rows, _cols, _lhs, lhsStride, _rhs, rhsIncr, _res, resIncr, alpha); \
+     return; \
+   }\
+   Index size = (std::min)(_rows,_cols); \
+   Index rows = IsLower ? _rows : size; \
+   Index cols = IsLower ? size : _cols; \
+\
+   typedef VectorX##EIGPREFIX VectorRhs; \
+   EIGTYPE *x, *y;\
+\
+/* Set x*/ \
+   Map<const VectorRhs, 0, InnerStride<> > rhs(_rhs,cols,InnerStride<>(rhsIncr)); \
+   VectorRhs x_tmp; \
+   if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \
+   x = x_tmp.data(); \
+\
+/* Square part handling */\
+\
+   char trans, uplo, diag; \
+   BlasIndex m, n, lda, incx, incy; \
+   EIGTYPE const *a; \
+   EIGTYPE beta(1); \
+\
+/* Set m, n */ \
+   n = convert_index<BlasIndex>(size); \
+   lda = convert_index<BlasIndex>(lhsStride); \
+   incx = 1; \
+   incy = convert_index<BlasIndex>(resIncr); \
+\
+/* Set uplo, trans and diag*/ \
+   trans = 'N'; \
+   uplo = IsLower ? 'L' : 'U'; \
+   diag = IsUnitDiag ? 'U' : 'N'; \
+\
+/* call ?TRMV*/ \
+   BLASPREFIX##trmv##BLASPOSTFIX(&uplo, &trans, &diag, &n, (const BLASTYPE*)_lhs, &lda, (BLASTYPE*)x, &incx); \
+\
+/* Add op(a_tr)rhs into res*/ \
+   BLASPREFIX##axpy##BLASPOSTFIX(&n, (const BLASTYPE*)&numext::real_ref(alpha),(const BLASTYPE*)x, &incx, (BLASTYPE*)_res, &incy); \
+/* Non-square case - doesn't fit to BLAS ?TRMV. Fall to default triangular product*/ \
+   if (size<(std::max)(rows,cols)) { \
+     if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \
+     x = x_tmp.data(); \
+     if (size<rows) { \
+       y = _res + size*resIncr; \
+       a = _lhs + size; \
+       m = convert_index<BlasIndex>(rows-size); \
+       n = convert_index<BlasIndex>(size); \
+     } \
+     else { \
+       x += size; \
+       y = _res; \
+       a = _lhs + size*lda; \
+       m = convert_index<BlasIndex>(size); \
+       n = convert_index<BlasIndex>(cols-size); \
+     } \
+     BLASPREFIX##gemv##BLASPOSTFIX(&trans, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)x, &incx, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)y, &incy); \
+   } \
+  } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRMV_CM(double,   double, d,  d,)
+EIGEN_BLAS_TRMV_CM(dcomplex, MKL_Complex16, cd, z,)
+EIGEN_BLAS_TRMV_CM(float,    float,  f,  s,)
+EIGEN_BLAS_TRMV_CM(scomplex, MKL_Complex8,  cf, c,)
+#else
+EIGEN_BLAS_TRMV_CM(double,   double, d,  d, _)
+EIGEN_BLAS_TRMV_CM(dcomplex, double, cd, z, _)
+EIGEN_BLAS_TRMV_CM(float,    float,  f,  s, _)
+EIGEN_BLAS_TRMV_CM(scomplex, float,  cf, c, _)
+#endif
+
+// implements row-major: res += alpha * op(triangular) * vector
+#define EIGEN_BLAS_TRMV_RM(EIGTYPE, BLASTYPE, EIGPREFIX, BLASPREFIX, BLASPOSTFIX) \
+template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \
+struct triangular_matrix_vector_product_trmv<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,RowMajor> { \
+  enum { \
+    IsLower = (Mode&Lower) == Lower, \
+    SetDiag = (Mode&(ZeroDiag|UnitDiag)) ? 0 : 1, \
+    IsUnitDiag  = (Mode&UnitDiag) ? 1 : 0, \
+    IsZeroDiag  = (Mode&ZeroDiag) ? 1 : 0, \
+    LowUp = IsLower ? Lower : Upper \
+  }; \
+ static void run(Index _rows, Index _cols, const EIGTYPE* _lhs, Index lhsStride, \
+                 const EIGTYPE* _rhs, Index rhsIncr, EIGTYPE* _res, Index resIncr, EIGTYPE alpha) \
+ { \
+   if (IsZeroDiag) { \
+     triangular_matrix_vector_product<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,RowMajor,BuiltIn>::run( \
+       _rows, _cols, _lhs, lhsStride, _rhs, rhsIncr, _res, resIncr, alpha); \
+     return; \
+   }\
+   Index size = (std::min)(_rows,_cols); \
+   Index rows = IsLower ? _rows : size; \
+   Index cols = IsLower ? size : _cols; \
+\
+   typedef VectorX##EIGPREFIX VectorRhs; \
+   EIGTYPE *x, *y;\
+\
+/* Set x*/ \
+   Map<const VectorRhs, 0, InnerStride<> > rhs(_rhs,cols,InnerStride<>(rhsIncr)); \
+   VectorRhs x_tmp; \
+   if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \
+   x = x_tmp.data(); \
+\
+/* Square part handling */\
+\
+   char trans, uplo, diag; \
+   BlasIndex m, n, lda, incx, incy; \
+   EIGTYPE const *a; \
+   EIGTYPE beta(1); \
+\
+/* Set m, n */ \
+   n = convert_index<BlasIndex>(size); \
+   lda = convert_index<BlasIndex>(lhsStride); \
+   incx = 1; \
+   incy = convert_index<BlasIndex>(resIncr); \
+\
+/* Set uplo, trans and diag*/ \
+   trans = ConjLhs ? 'C' : 'T'; \
+   uplo = IsLower ? 'U' : 'L'; \
+   diag = IsUnitDiag ? 'U' : 'N'; \
+\
+/* call ?TRMV*/ \
+   BLASPREFIX##trmv##BLASPOSTFIX(&uplo, &trans, &diag, &n, (const BLASTYPE*)_lhs, &lda, (BLASTYPE*)x, &incx); \
+\
+/* Add op(a_tr)rhs into res*/ \
+   BLASPREFIX##axpy##BLASPOSTFIX(&n, (const BLASTYPE*)&numext::real_ref(alpha),(const BLASTYPE*)x, &incx, (BLASTYPE*)_res, &incy); \
+/* Non-square case - doesn't fit to BLAS ?TRMV. Fall to default triangular product*/ \
+   if (size<(std::max)(rows,cols)) { \
+     if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \
+     x = x_tmp.data(); \
+     if (size<rows) { \
+       y = _res + size*resIncr; \
+       a = _lhs + size*lda; \
+       m = convert_index<BlasIndex>(rows-size); \
+       n = convert_index<BlasIndex>(size); \
+     } \
+     else { \
+       x += size; \
+       y = _res; \
+       a = _lhs + size; \
+       m = convert_index<BlasIndex>(size); \
+       n = convert_index<BlasIndex>(cols-size); \
+     } \
+     BLASPREFIX##gemv##BLASPOSTFIX(&trans, &n, &m, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)x, &incx, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)y, &incy); \
+   } \
+  } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRMV_RM(double,   double, d,  d,)
+EIGEN_BLAS_TRMV_RM(dcomplex, MKL_Complex16, cd, z,)
+EIGEN_BLAS_TRMV_RM(float,    float,  f,  s,)
+EIGEN_BLAS_TRMV_RM(scomplex, MKL_Complex8,  cf, c,)
+#else
+EIGEN_BLAS_TRMV_RM(double,   double, d,  d,_)
+EIGEN_BLAS_TRMV_RM(dcomplex, double, cd, z,_)
+EIGEN_BLAS_TRMV_RM(float,    float,  f,  s,_)
+EIGEN_BLAS_TRMV_RM(scomplex, float,  cf, c,_)
+#endif
+
+} // end namespase internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TRIANGULAR_MATRIX_VECTOR_BLAS_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/TriangularSolverMatrix.h b/extern/Eigen3/Eigen/src/Core/products/TriangularSolverMatrix.h
index 04240ab5032..223c38b8656 100644
--- a/extern/Eigen3/Eigen/src/Core/products/TriangularSolverMatrix.h
+++ b/extern/Eigen3/Eigen/src/Core/products/TriangularSolverMatrix.h
@@ -52,10 +52,14 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
     level3_blocking<Scalar,Scalar>& blocking)
   {
     Index cols = otherSize;
-    const_blas_data_mapper<Scalar, Index, TriStorageOrder> tri(_tri,triStride);
-    blas_data_mapper<Scalar, Index, ColMajor> other(_other,otherStride);
+
+    typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> TriMapper;
+    typedef blas_data_mapper<Scalar, Index, ColMajor> OtherMapper;
+    TriMapper tri(_tri, triStride);
+    OtherMapper other(_other, otherStride);
 
     typedef gebp_traits<Scalar,Scalar> Traits;
+
     enum {
       SmallPanelWidth   = EIGEN_PLAIN_ENUM_MAX(Traits::mr,Traits::nr),
       IsLower = (Mode&Lower) == Lower
@@ -66,22 +70,20 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
 
     std::size_t sizeA = kc*mc;
     std::size_t sizeB = kc*cols;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
 
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockW, sizeW, blocking.blockW());
 
     conj_if<Conjugate> conj;
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, Conjugate, false> gebp_kernel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, TriStorageOrder> pack_lhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr, ColMajor, false, true> pack_rhs;
+    gebp_kernel<Scalar, Scalar, Index, OtherMapper, Traits::mr, Traits::nr, Conjugate, false> gebp_kernel;
+    gemm_pack_lhs<Scalar, Index, TriMapper, Traits::mr, Traits::LhsProgress, TriStorageOrder> pack_lhs;
+    gemm_pack_rhs<Scalar, Index, OtherMapper, Traits::nr, ColMajor, false, true> pack_rhs;
 
     // the goal here is to subdivise the Rhs panels such that we keep some cache
     // coherence when accessing the rhs elements
-    std::ptrdiff_t l1, l2;
-    manage_caching_sizes(GetAction, &l1, &l2);
-    Index subcols = cols>0 ? l2/(4 * sizeof(Scalar) * otherStride) : 0;
+    std::ptrdiff_t l1, l2, l3;
+    manage_caching_sizes(GetAction, &l1, &l2, &l3);
+    Index subcols = cols>0 ? l2/(4 * sizeof(Scalar) * std::max<Index>(otherStride,size)) : 0;
     subcols = std::max<Index>((subcols/Traits::nr)*Traits::nr, Traits::nr);
 
     for(Index k2=IsLower ? 0 : size;
@@ -115,8 +117,9 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
           {
             // TODO write a small kernel handling this (can be shared with trsv)
             Index i  = IsLower ? k2+k1+k : k2-k1-k-1;
-            Index s  = IsLower ? k2+k1 : i+1;
             Index rs = actualPanelWidth - k - 1; // remaining size
+            Index s  = TriStorageOrder==RowMajor ? (IsLower ? k2+k1 : i+1)
+                                                 :  IsLower ? i+1 : i-rs;
 
             Scalar a = (Mode & UnitDiag) ? Scalar(1) : Scalar(1)/conj(tri(i,i));
             for (Index j=j2; j<j2+actual_cols; ++j)
@@ -133,7 +136,6 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
               }
               else
               {
-                Index s = IsLower ? i+1 : i-rs;
                 Scalar b = (other(i,j) *= a);
                 Scalar* r = &other(s,j);
                 const Scalar* l = &tri(s,i);
@@ -148,17 +150,17 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
           Index blockBOffset = IsLower ? k1 : lengthTarget;
 
           // update the respective rows of B from other
-          pack_rhs(blockB+actual_kc*j2, &other(startBlock,j2), otherStride, actualPanelWidth, actual_cols, actual_kc, blockBOffset);
+          pack_rhs(blockB+actual_kc*j2, other.getSubMapper(startBlock,j2), actualPanelWidth, actual_cols, actual_kc, blockBOffset);
 
           // GEBP
           if (lengthTarget>0)
           {
             Index startTarget  = IsLower ? k2+k1+actualPanelWidth : k2-actual_kc;
 
-            pack_lhs(blockA, &tri(startTarget,startBlock), triStride, actualPanelWidth, lengthTarget);
+            pack_lhs(blockA, tri.getSubMapper(startTarget,startBlock), actualPanelWidth, lengthTarget);
 
-            gebp_kernel(&other(startTarget,j2), otherStride, blockA, blockB+actual_kc*j2, lengthTarget, actualPanelWidth, actual_cols, Scalar(-1),
-                        actualPanelWidth, actual_kc, 0, blockBOffset, blockW);
+            gebp_kernel(other.getSubMapper(startTarget,j2), blockA, blockB+actual_kc*j2, lengthTarget, actualPanelWidth, actual_cols, Scalar(-1),
+                        actualPanelWidth, actual_kc, 0, blockBOffset);
           }
         }
       }
@@ -172,16 +174,16 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
           const Index actual_mc = (std::min)(mc,end-i2);
           if (actual_mc>0)
           {
-            pack_lhs(blockA, &tri(i2, IsLower ? k2 : k2-kc), triStride, actual_kc, actual_mc);
+            pack_lhs(blockA, tri.getSubMapper(i2, IsLower ? k2 : k2-kc), actual_kc, actual_mc);
 
-            gebp_kernel(_other+i2, otherStride, blockA, blockB, actual_mc, actual_kc, cols, Scalar(-1), -1, -1, 0, 0, blockW);
+            gebp_kernel(other.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc, cols, Scalar(-1), -1, -1, 0, 0);
           }
         }
       }
     }
   }
 
-/* Optimized triangular solver with multiple left hand sides and the trinagular matrix on the right
+/* Optimized triangular solver with multiple left hand sides and the triangular matrix on the right
  */
 template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder>
 struct triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor>
@@ -200,8 +202,12 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
     level3_blocking<Scalar,Scalar>& blocking)
   {
     Index rows = otherSize;
-    const_blas_data_mapper<Scalar, Index, TriStorageOrder> rhs(_tri,triStride);
-    blas_data_mapper<Scalar, Index, ColMajor> lhs(_other,otherStride);
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+
+    typedef blas_data_mapper<Scalar, Index, ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> RhsMapper;
+    LhsMapper lhs(_other, otherStride);
+    RhsMapper rhs(_tri, triStride);
 
     typedef gebp_traits<Scalar,Scalar> Traits;
     enum {
@@ -215,17 +221,15 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
 
     std::size_t sizeA = kc*mc;
     std::size_t sizeB = kc*size;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
 
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockW, sizeW, blocking.blockW());
 
     conj_if<Conjugate> conj;
-    gebp_kernel<Scalar,Scalar, Index, Traits::mr, Traits::nr, false, Conjugate> gebp_kernel;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder,false,true> pack_rhs_panel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, ColMajor, false, true> pack_lhs_panel;
+    gebp_kernel<Scalar, Scalar, Index, LhsMapper, Traits::mr, Traits::nr, false, Conjugate> gebp_kernel;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr, RhsStorageOrder> pack_rhs;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr, RhsStorageOrder,false,true> pack_rhs_panel;
+    gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, ColMajor, false, true> pack_lhs_panel;
 
     for(Index k2=IsLower ? size : 0;
         IsLower ? k2>0 : k2<size;
@@ -238,7 +242,7 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
       Index rs = IsLower ? actual_k2 : size - actual_k2 - actual_kc;
       Scalar* geb = blockB+actual_kc*actual_kc;
 
-      if (rs>0) pack_rhs(geb, &rhs(actual_k2,startPanel), triStride, actual_kc, rs);
+      if (rs>0) pack_rhs(geb, rhs.getSubMapper(actual_k2,startPanel), actual_kc, rs);
 
       // triangular packing (we only pack the panels off the diagonal,
       // neglecting the blocks overlapping the diagonal
@@ -252,7 +256,7 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
 
           if (panelLength>0)
           pack_rhs_panel(blockB+j2*actual_kc,
-                         &rhs(actual_k2+panelOffset, actual_j2), triStride,
+                         rhs.getSubMapper(actual_k2+panelOffset, actual_j2),
                          panelLength, actualPanelWidth,
                          actual_kc, panelOffset);
         }
@@ -280,13 +284,12 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
             // GEBP
             if(panelLength>0)
             {
-              gebp_kernel(&lhs(i2,absolute_j2), otherStride,
+              gebp_kernel(lhs.getSubMapper(i2,absolute_j2),
                           blockA, blockB+j2*actual_kc,
                           actual_mc, panelLength, actualPanelWidth,
                           Scalar(-1),
                           actual_kc, actual_kc, // strides
-                          panelOffset, panelOffset, // offsets
-                          blockW);  // workspace
+                          panelOffset, panelOffset); // offsets
             }
 
             // unblocked triangular solve
@@ -304,23 +307,23 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
               }
               if((Mode & UnitDiag)==0)
               {
-                Scalar b = conj(rhs(j,j));
+                Scalar inv_rjj = RealScalar(1)/conj(rhs(j,j));
                 for (Index i=0; i<actual_mc; ++i)
-                  r[i] /= b;
+                  r[i] *= inv_rjj;
               }
             }
 
             // pack the just computed part of lhs to A
-            pack_lhs_panel(blockA, _other+absolute_j2*otherStride+i2, otherStride,
+            pack_lhs_panel(blockA, LhsMapper(_other+absolute_j2*otherStride+i2, otherStride),
                            actualPanelWidth, actual_mc,
                            actual_kc, j2);
           }
         }
 
         if (rs>0)
-          gebp_kernel(_other+i2+startPanel*otherStride, otherStride, blockA, geb,
+          gebp_kernel(lhs.getSubMapper(i2, startPanel), blockA, geb,
                       actual_mc, actual_kc, rs, Scalar(-1),
-                      -1, -1, 0, 0, blockW);
+                      -1, -1, 0, 0);
       }
     }
   }
diff --git a/extern/Eigen3/Eigen/src/Core/products/TriangularSolverMatrix_BLAS.h b/extern/Eigen3/Eigen/src/Core/products/TriangularSolverMatrix_BLAS.h
new file mode 100644
index 00000000000..f0775116ac3
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Core/products/TriangularSolverMatrix_BLAS.h
@@ -0,0 +1,163 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   Triangular matrix * matrix product functionality based on ?TRMM.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_TRIANGULAR_SOLVER_MATRIX_BLAS_H
+#define EIGEN_TRIANGULAR_SOLVER_MATRIX_BLAS_H
+
+namespace Eigen {
+
+namespace internal {
+
+// implements LeftSide op(triangular)^-1 * general
+#define EIGEN_BLAS_TRSM_L(EIGTYPE, BLASTYPE, BLASFUNC) \
+template <typename Index, int Mode, bool Conjugate, int TriStorageOrder> \
+struct triangular_solve_matrix<EIGTYPE,Index,OnTheLeft,Mode,Conjugate,TriStorageOrder,ColMajor> \
+{ \
+  enum { \
+    IsLower = (Mode&Lower) == Lower, \
+    IsUnitDiag  = (Mode&UnitDiag) ? 1 : 0, \
+    IsZeroDiag  = (Mode&ZeroDiag) ? 1 : 0, \
+    conjA = ((TriStorageOrder==ColMajor) && Conjugate) ? 1 : 0 \
+  }; \
+  static void run( \
+      Index size, Index otherSize, \
+      const EIGTYPE* _tri, Index triStride, \
+      EIGTYPE* _other, Index otherStride, level3_blocking<EIGTYPE,EIGTYPE>& /*blocking*/) \
+  { \
+   BlasIndex m = convert_index<BlasIndex>(size), n = convert_index<BlasIndex>(otherSize), lda, ldb; \
+   char side = 'L', uplo, diag='N', transa; \
+   /* Set alpha_ */ \
+   EIGTYPE alpha(1); \
+   ldb = convert_index<BlasIndex>(otherStride);\
+\
+   const EIGTYPE *a; \
+/* Set trans */ \
+   transa = (TriStorageOrder==RowMajor) ? ((Conjugate) ? 'C' : 'T') : 'N'; \
+/* Set uplo */ \
+   uplo = IsLower ? 'L' : 'U'; \
+   if (TriStorageOrder==RowMajor) uplo = (uplo == 'L') ? 'U' : 'L'; \
+/* Set a, lda */ \
+   typedef Matrix<EIGTYPE, Dynamic, Dynamic, TriStorageOrder> MatrixTri; \
+   Map<const MatrixTri, 0, OuterStride<> > tri(_tri,size,size,OuterStride<>(triStride)); \
+   MatrixTri a_tmp; \
+\
+   if (conjA) { \
+     a_tmp = tri.conjugate(); \
+     a = a_tmp.data(); \
+     lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
+   } else { \
+     a = _tri; \
+     lda = convert_index<BlasIndex>(triStride); \
+   } \
+   if (IsUnitDiag) diag='U'; \
+/* call ?trsm*/ \
+   BLASFUNC(&side, &uplo, &transa, &diag, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (BLASTYPE*)_other, &ldb); \
+ } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRSM_L(double,   double, dtrsm)
+EIGEN_BLAS_TRSM_L(dcomplex, MKL_Complex16, ztrsm)
+EIGEN_BLAS_TRSM_L(float,    float,  strsm)
+EIGEN_BLAS_TRSM_L(scomplex, MKL_Complex8, ctrsm)
+#else
+EIGEN_BLAS_TRSM_L(double,   double, dtrsm_)
+EIGEN_BLAS_TRSM_L(dcomplex, double, ztrsm_)
+EIGEN_BLAS_TRSM_L(float,    float,  strsm_)
+EIGEN_BLAS_TRSM_L(scomplex, float,  ctrsm_)
+#endif
+
+// implements RightSide general * op(triangular)^-1
+#define EIGEN_BLAS_TRSM_R(EIGTYPE, BLASTYPE, BLASFUNC) \
+template <typename Index, int Mode, bool Conjugate, int TriStorageOrder> \
+struct triangular_solve_matrix<EIGTYPE,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor> \
+{ \
+  enum { \
+    IsLower = (Mode&Lower) == Lower, \
+    IsUnitDiag  = (Mode&UnitDiag) ? 1 : 0, \
+    IsZeroDiag  = (Mode&ZeroDiag) ? 1 : 0, \
+    conjA = ((TriStorageOrder==ColMajor) && Conjugate) ? 1 : 0 \
+  }; \
+  static void run( \
+      Index size, Index otherSize, \
+      const EIGTYPE* _tri, Index triStride, \
+      EIGTYPE* _other, Index otherStride, level3_blocking<EIGTYPE,EIGTYPE>& /*blocking*/) \
+  { \
+   BlasIndex m = convert_index<BlasIndex>(otherSize), n = convert_index<BlasIndex>(size), lda, ldb; \
+   char side = 'R', uplo, diag='N', transa; \
+   /* Set alpha_ */ \
+   EIGTYPE alpha(1); \
+   ldb = convert_index<BlasIndex>(otherStride);\
+\
+   const EIGTYPE *a; \
+/* Set trans */ \
+   transa = (TriStorageOrder==RowMajor) ? ((Conjugate) ? 'C' : 'T') : 'N'; \
+/* Set uplo */ \
+   uplo = IsLower ? 'L' : 'U'; \
+   if (TriStorageOrder==RowMajor) uplo = (uplo == 'L') ? 'U' : 'L'; \
+/* Set a, lda */ \
+   typedef Matrix<EIGTYPE, Dynamic, Dynamic, TriStorageOrder> MatrixTri; \
+   Map<const MatrixTri, 0, OuterStride<> > tri(_tri,size,size,OuterStride<>(triStride)); \
+   MatrixTri a_tmp; \
+\
+   if (conjA) { \
+     a_tmp = tri.conjugate(); \
+     a = a_tmp.data(); \
+     lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
+   } else { \
+     a = _tri; \
+     lda = convert_index<BlasIndex>(triStride); \
+   } \
+   if (IsUnitDiag) diag='U'; \
+/* call ?trsm*/ \
+   BLASFUNC(&side, &uplo, &transa, &diag, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (BLASTYPE*)_other, &ldb); \
+   /*std::cout << "TRMS_L specialization!\n";*/ \
+ } \
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRSM_R(double,   double, dtrsm)
+EIGEN_BLAS_TRSM_R(dcomplex, MKL_Complex16, ztrsm)
+EIGEN_BLAS_TRSM_R(float,    float,  strsm)
+EIGEN_BLAS_TRSM_R(scomplex, MKL_Complex8,  ctrsm)
+#else
+EIGEN_BLAS_TRSM_R(double,   double, dtrsm_)
+EIGEN_BLAS_TRSM_R(dcomplex, double, ztrsm_)
+EIGEN_BLAS_TRSM_R(float,    float,  strsm_)
+EIGEN_BLAS_TRSM_R(scomplex, float,  ctrsm_)
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TRIANGULAR_SOLVER_MATRIX_BLAS_H
diff --git a/extern/Eigen3/Eigen/src/Core/products/TriangularSolverVector.h b/extern/Eigen3/Eigen/src/Core/products/TriangularSolverVector.h
index ce4d1008801..b994759b268 100644
--- a/extern/Eigen3/Eigen/src/Core/products/TriangularSolverVector.h
+++ b/extern/Eigen3/Eigen/src/Core/products/TriangularSolverVector.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_TRIANGULAR_SOLVER_VECTOR_H
 #define EIGEN_TRIANGULAR_SOLVER_VECTOR_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -25,7 +25,7 @@ struct triangular_solve_vector<LhsScalar, RhsScalar, Index, OnTheRight, Mode, Co
       >::run(size, _lhs, lhsStride, rhs);
   }
 };
-    
+
 // forward and backward substitution, row-major, rhs is a vector
 template<typename LhsScalar, typename RhsScalar, typename Index, int Mode, bool Conjugate>
 struct triangular_solve_vector<LhsScalar, RhsScalar, Index, OnTheLeft, Mode, Conjugate, RowMajor>
@@ -37,6 +37,10 @@ struct triangular_solve_vector<LhsScalar, RhsScalar, Index, OnTheLeft, Mode, Con
   {
     typedef Map<const Matrix<LhsScalar,Dynamic,Dynamic,RowMajor>, 0, OuterStride<> > LhsMap;
     const LhsMap lhs(_lhs,size,size,OuterStride<>(lhsStride));
+
+    typedef const_blas_data_mapper<LhsScalar,Index,RowMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,ColMajor> RhsMapper;
+
     typename internal::conditional<
                           Conjugate,
                           const CwiseUnaryOp<typename internal::scalar_conjugate_op<LhsScalar>,LhsMap>,
@@ -58,10 +62,10 @@ struct triangular_solve_vector<LhsScalar, RhsScalar, Index, OnTheLeft, Mode, Con
         Index startRow = IsLower ? pi : pi-actualPanelWidth;
         Index startCol = IsLower ? 0 : pi;
 
-        general_matrix_vector_product<Index,LhsScalar,RowMajor,Conjugate,RhsScalar,false>::run(
+        general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,Conjugate,RhsScalar,RhsMapper,false>::run(
           actualPanelWidth, r,
-          &lhs.coeffRef(startRow,startCol), lhsStride,
-          rhs + startCol, 1,
+          LhsMapper(&lhs.coeffRef(startRow,startCol), lhsStride),
+          RhsMapper(rhs + startCol, 1),
           rhs + startRow, 1,
           RhsScalar(-1));
       }
@@ -72,7 +76,7 @@ struct triangular_solve_vector<LhsScalar, RhsScalar, Index, OnTheLeft, Mode, Con
         Index s = IsLower ? pi   : i+1;
         if (k>0)
           rhs[i] -= (cjLhs.row(i).segment(s,k).transpose().cwiseProduct(Map<const Matrix<RhsScalar,Dynamic,1> >(rhs+s,k))).sum();
-        
+
         if(!(Mode & UnitDiag))
           rhs[i] /= cjLhs(i,i);
       }
@@ -91,6 +95,8 @@ struct triangular_solve_vector<LhsScalar, RhsScalar, Index, OnTheLeft, Mode, Con
   {
     typedef Map<const Matrix<LhsScalar,Dynamic,Dynamic,ColMajor>, 0, OuterStride<> > LhsMap;
     const LhsMap lhs(_lhs,size,size,OuterStride<>(lhsStride));
+    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,ColMajor> RhsMapper;
     typename internal::conditional<Conjugate,
                                    const CwiseUnaryOp<typename internal::scalar_conjugate_op<LhsScalar>,LhsMap>,
                                    const LhsMap&
@@ -122,10 +128,10 @@ struct triangular_solve_vector<LhsScalar, RhsScalar, Index, OnTheLeft, Mode, Con
         // let's directly call the low level product function because:
         // 1 - it is faster to compile
         // 2 - it is slighlty faster at runtime
-        general_matrix_vector_product<Index,LhsScalar,ColMajor,Conjugate,RhsScalar,false>::run(
+        general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,Conjugate,RhsScalar,RhsMapper,false>::run(
             r, actualPanelWidth,
-            &lhs.coeffRef(endBlock,startBlock), lhsStride,
-            rhs+startBlock, 1,
+            LhsMapper(&lhs.coeffRef(endBlock,startBlock), lhsStride),
+            RhsMapper(rhs+startBlock, 1),
             rhs+endBlock, 1, RhsScalar(-1));
       }
     }
diff --git a/extern/Eigen3/Eigen/src/Core/util/BlasUtil.h b/extern/Eigen3/Eigen/src/Core/util/BlasUtil.h
index a28f16fa04b..6e6ee119b64 100644
--- a/extern/Eigen3/Eigen/src/Core/util/BlasUtil.h
+++ b/extern/Eigen3/Eigen/src/Core/util/BlasUtil.h
@@ -18,13 +18,13 @@ namespace Eigen {
 namespace internal {
 
 // forward declarations
-template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjugateLhs=false, bool ConjugateRhs=false>
+template<typename LhsScalar, typename RhsScalar, typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs=false, bool ConjugateRhs=false>
 struct gebp_kernel;
 
-template<typename Scalar, typename Index, int nr, int StorageOrder, bool Conjugate = false, bool PanelMode=false>
+template<typename Scalar, typename Index, typename DataMapper, int nr, int StorageOrder, bool Conjugate = false, bool PanelMode=false>
 struct gemm_pack_rhs;
 
-template<typename Scalar, typename Index, int Pack1, int Pack2, int StorageOrder, bool Conjugate = false, bool PanelMode = false>
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, int StorageOrder, bool Conjugate = false, bool PanelMode = false>
 struct gemm_pack_lhs;
 
 template<
@@ -34,7 +34,9 @@ template<
   int ResStorageOrder>
 struct general_matrix_matrix_product;
 
-template<typename Index, typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version=Specialized>
+template<typename Index,
+         typename LhsScalar, typename LhsMapper, int LhsStorageOrder, bool ConjugateLhs,
+         typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version=Specialized>
 struct general_matrix_vector_product;
 
 
@@ -42,22 +44,35 @@ template<bool Conjugate> struct conj_if;
 
 template<> struct conj_if<true> {
   template<typename T>
-  inline T operator()(const T& x) { return numext::conj(x); }
+  inline T operator()(const T& x) const { return numext::conj(x); }
   template<typename T>
-  inline T pconj(const T& x) { return internal::pconj(x); }
+  inline T pconj(const T& x) const { return internal::pconj(x); }
 };
 
 template<> struct conj_if<false> {
   template<typename T>
-  inline const T& operator()(const T& x) { return x; }
+  inline const T& operator()(const T& x) const { return x; }
   template<typename T>
-  inline const T& pconj(const T& x) { return x; }
+  inline const T& pconj(const T& x) const { return x; }
+};
+
+// Generic implementation for custom complex types.
+template<typename LhsScalar, typename RhsScalar, bool ConjLhs, bool ConjRhs>
+struct conj_helper
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar>::ReturnType Scalar;
+
+  EIGEN_STRONG_INLINE Scalar pmadd(const LhsScalar& x, const RhsScalar& y, const Scalar& c) const
+  { return padd(c, pmul(x,y)); }
+
+  EIGEN_STRONG_INLINE Scalar pmul(const LhsScalar& x, const RhsScalar& y) const
+  { return conj_if<ConjLhs>()(x) *  conj_if<ConjRhs>()(y); }
 };
 
 template<typename Scalar> struct conj_helper<Scalar,Scalar,false,false>
 {
-  EIGEN_STRONG_INLINE Scalar pmadd(const Scalar& x, const Scalar& y, const Scalar& c) const { return internal::pmadd(x,y,c); }
-  EIGEN_STRONG_INLINE Scalar pmul(const Scalar& x, const Scalar& y) const { return internal::pmul(x,y); }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar pmadd(const Scalar& x, const Scalar& y, const Scalar& c) const { return internal::pmadd(x,y,c); }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar pmul(const Scalar& x, const Scalar& y) const { return internal::pmul(x,y); }
 };
 
 template<typename RealScalar> struct conj_helper<std::complex<RealScalar>, std::complex<RealScalar>, false,true>
@@ -109,39 +124,142 @@ template<typename RealScalar,bool Conj> struct conj_helper<RealScalar, std::comp
 };
 
 template<typename From,typename To> struct get_factor {
-  static EIGEN_STRONG_INLINE To run(const From& x) { return x; }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE To run(const From& x) { return To(x); }
 };
 
 template<typename Scalar> struct get_factor<Scalar,typename NumTraits<Scalar>::Real> {
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE typename NumTraits<Scalar>::Real run(const Scalar& x) { return numext::real(x); }
 };
 
+
+template<typename Scalar, typename Index>
+class BlasVectorMapper {
+  public:
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE BlasVectorMapper(Scalar *data) : m_data(data) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Scalar operator()(Index i) const {
+    return m_data[i];
+  }
+  template <typename Packet, int AlignmentType>
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet load(Index i) const {
+    return ploadt<Packet, AlignmentType>(m_data + i);
+  }
+
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC bool aligned(Index i) const {
+    return (UIntPtr(m_data+i)%sizeof(Packet))==0;
+  }
+
+  protected:
+  Scalar* m_data;
+};
+
+template<typename Scalar, typename Index, int AlignmentType>
+class BlasLinearMapper {
+  public:
+  typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename packet_traits<Scalar>::half HalfPacket;
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE BlasLinearMapper(Scalar *data) : m_data(data) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void prefetch(int i) const {
+    internal::prefetch(&operator()(i));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Scalar& operator()(Index i) const {
+    return m_data[i];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacket(Index i) const {
+    return ploadt<Packet, AlignmentType>(m_data + i);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE HalfPacket loadHalfPacket(Index i) const {
+    return ploadt<HalfPacket, AlignmentType>(m_data + i);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void storePacket(Index i, const Packet &p) const {
+    pstoret<Scalar, Packet, AlignmentType>(m_data + i, p);
+  }
+
+  protected:
+  Scalar *m_data;
+};
+
 // Lightweight helper class to access matrix coefficients.
-// Yes, this is somehow redundant with Map<>, but this version is much much lighter,
-// and so I hope better compilation performance (time and code quality).
-template<typename Scalar, typename Index, int StorageOrder>
-class blas_data_mapper
-{
+template<typename Scalar, typename Index, int StorageOrder, int AlignmentType = Unaligned>
+class blas_data_mapper {
   public:
-    blas_data_mapper(Scalar* data, Index stride) : m_data(data), m_stride(stride) {}
-    EIGEN_STRONG_INLINE Scalar& operator()(Index i, Index j)
-    { return m_data[StorageOrder==RowMajor ? j + i*m_stride : i + j*m_stride]; }
+  typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename packet_traits<Scalar>::half HalfPacket;
+
+  typedef BlasLinearMapper<Scalar, Index, AlignmentType> LinearMapper;
+  typedef BlasVectorMapper<Scalar, Index> VectorMapper;
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE blas_data_mapper(Scalar* data, Index stride) : m_data(data), m_stride(stride) {}
+
+  EIGEN_DEVICE_FUNC  EIGEN_ALWAYS_INLINE blas_data_mapper<Scalar, Index, StorageOrder, AlignmentType>
+  getSubMapper(Index i, Index j) const {
+    return blas_data_mapper<Scalar, Index, StorageOrder, AlignmentType>(&operator()(i, j), m_stride);
+  }
+
+  EIGEN_DEVICE_FUNC  EIGEN_ALWAYS_INLINE LinearMapper getLinearMapper(Index i, Index j) const {
+    return LinearMapper(&operator()(i, j));
+  }
+
+  EIGEN_DEVICE_FUNC  EIGEN_ALWAYS_INLINE VectorMapper getVectorMapper(Index i, Index j) const {
+    return VectorMapper(&operator()(i, j));
+  }
+
+
+  EIGEN_DEVICE_FUNC
+  EIGEN_ALWAYS_INLINE Scalar& operator()(Index i, Index j) const {
+    return m_data[StorageOrder==RowMajor ? j + i*m_stride : i + j*m_stride];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacket(Index i, Index j) const {
+    return ploadt<Packet, AlignmentType>(&operator()(i, j));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE HalfPacket loadHalfPacket(Index i, Index j) const {
+    return ploadt<HalfPacket, AlignmentType>(&operator()(i, j));
+  }
+
+  template<typename SubPacket>
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void scatterPacket(Index i, Index j, const SubPacket &p) const {
+    pscatter<Scalar, SubPacket>(&operator()(i, j), p, m_stride);
+  }
+
+  template<typename SubPacket>
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE SubPacket gatherPacket(Index i, Index j) const {
+    return pgather<Scalar, SubPacket>(&operator()(i, j), m_stride);
+  }
+
+  EIGEN_DEVICE_FUNC const Index stride() const { return m_stride; }
+  EIGEN_DEVICE_FUNC const Scalar* data() const { return m_data; }
+
+  EIGEN_DEVICE_FUNC Index firstAligned(Index size) const {
+    if (UIntPtr(m_data)%sizeof(Scalar)) {
+      return -1;
+    }
+    return internal::first_default_aligned(m_data, size);
+  }
+
   protected:
-    Scalar* EIGEN_RESTRICT m_data;
-    Index m_stride;
+  Scalar* EIGEN_RESTRICT m_data;
+  const Index m_stride;
 };
 
 // lightweight helper class to access matrix coefficients (const version)
 template<typename Scalar, typename Index, int StorageOrder>
-class const_blas_data_mapper
-{
+class const_blas_data_mapper : public blas_data_mapper<const Scalar, Index, StorageOrder> {
   public:
-    const_blas_data_mapper(const Scalar* data, Index stride) : m_data(data), m_stride(stride) {}
-    EIGEN_STRONG_INLINE const Scalar& operator()(Index i, Index j) const
-    { return m_data[StorageOrder==RowMajor ? j + i*m_stride : i + j*m_stride]; }
-  protected:
-    const Scalar* EIGEN_RESTRICT m_data;
-    Index m_stride;
+  EIGEN_ALWAYS_INLINE const_blas_data_mapper(const Scalar *data, Index stride) : blas_data_mapper<const Scalar, Index, StorageOrder>(data, stride) {}
+
+  EIGEN_ALWAYS_INLINE const_blas_data_mapper<Scalar, Index, StorageOrder> getSubMapper(Index i, Index j) const {
+    return const_blas_data_mapper<Scalar, Index, StorageOrder>(&(this->operator()(i, j)), this->m_stride);
+  }
 };
 
 
@@ -188,17 +306,33 @@ struct blas_traits<CwiseUnaryOp<scalar_conjugate_op<Scalar>, NestedXpr> >
 };
 
 // pop scalar multiple
-template<typename Scalar, typename NestedXpr>
-struct blas_traits<CwiseUnaryOp<scalar_multiple_op<Scalar>, NestedXpr> >
+template<typename Scalar, typename NestedXpr, typename Plain>
+struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain>, NestedXpr> >
  : blas_traits<NestedXpr>
 {
   typedef blas_traits<NestedXpr> Base;
-  typedef CwiseUnaryOp<scalar_multiple_op<Scalar>, NestedXpr> XprType;
+  typedef CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain>, NestedXpr> XprType;
   typedef typename Base::ExtractType ExtractType;
-  static inline ExtractType extract(const XprType& x) { return Base::extract(x.nestedExpression()); }
+  static inline ExtractType extract(const XprType& x) { return Base::extract(x.rhs()); }
   static inline Scalar extractScalarFactor(const XprType& x)
-  { return x.functor().m_other * Base::extractScalarFactor(x.nestedExpression()); }
+  { return x.lhs().functor().m_other * Base::extractScalarFactor(x.rhs()); }
 };
+template<typename Scalar, typename NestedXpr, typename Plain>
+struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, NestedXpr, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain> > >
+ : blas_traits<NestedXpr>
+{
+  typedef blas_traits<NestedXpr> Base;
+  typedef CwiseBinaryOp<scalar_product_op<Scalar>, NestedXpr, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain> > XprType;
+  typedef typename Base::ExtractType ExtractType;
+  static inline ExtractType extract(const XprType& x) { return Base::extract(x.lhs()); }
+  static inline Scalar extractScalarFactor(const XprType& x)
+  { return Base::extractScalarFactor(x.lhs()) * x.rhs().functor().m_other; }
+};
+template<typename Scalar, typename Plain1, typename Plain2>
+struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain1>,
+                                                            const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain2> > >
+ : blas_traits<CwiseNullaryOp<scalar_constant_op<Scalar>,Plain1> >
+{};
 
 // pop opposite
 template<typename Scalar, typename NestedXpr>
@@ -230,7 +364,7 @@ struct blas_traits<Transpose<NestedXpr> >
   enum {
     IsTransposed = Base::IsTransposed ? 0 : 1
   };
-  static inline ExtractType extract(const XprType& x) { return Base::extract(x.nestedExpression()); }
+  static inline ExtractType extract(const XprType& x) { return ExtractType(Base::extract(x.nestedExpression())); }
   static inline Scalar extractScalarFactor(const XprType& x) { return Base::extractScalarFactor(x.nestedExpression()); }
 };
 
diff --git a/extern/Eigen3/Eigen/src/Core/util/Constants.h b/extern/Eigen3/Eigen/src/Core/util/Constants.h
index 1e6277c4f96..7587d684243 100644
--- a/extern/Eigen3/Eigen/src/Core/util/Constants.h
+++ b/extern/Eigen3/Eigen/src/Core/util/Constants.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2007-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -30,6 +30,14 @@ const int DynamicIndex = 0xffffff;
   */
 const int Infinity = -1;
 
+/** This value means that the cost to evaluate an expression coefficient is either very expensive or
+  * cannot be known at compile time.
+  *
+  * This value has to be positive to (1) simplify cost computation, and (2) allow to distinguish between a very expensive and very very expensive expressions.
+  * It thus must also be large enough to make sure unrolling won't happen and that sub expressions will be evaluated, but not too large to avoid overflow.
+  */
+const int HugeCost = 10000;
+
 /** \defgroup flags Flags
   * \ingroup Core_Module
   *
@@ -48,19 +56,19 @@ const int Infinity = -1;
   * for a matrix, this means that the storage order is row-major.
   * If this bit is not set, the storage order is column-major.
   * For an expression, this determines the storage order of
-  * the matrix created by evaluation of that expression. 
-  * \sa \ref TopicStorageOrders */
+  * the matrix created by evaluation of that expression.
+  * \sa \blank  \ref TopicStorageOrders */
 const unsigned int RowMajorBit = 0x1;
 
 /** \ingroup flags
-  *
   * means the expression should be evaluated by the calling expression */
 const unsigned int EvalBeforeNestingBit = 0x2;
 
 /** \ingroup flags
-  *
+  * \deprecated
   * means the expression should be evaluated before any assignment */
-const unsigned int EvalBeforeAssigningBit = 0x4;
+EIGEN_DEPRECATED
+const unsigned int EvalBeforeAssigningBit = 0x4; // FIXME deprecated
 
 /** \ingroup flags
   *
@@ -141,17 +149,46 @@ const unsigned int LvalueBit = 0x20;
   */
 const unsigned int DirectAccessBit = 0x40;
 
-/** \ingroup flags
+/** \deprecated \ingroup flags
   *
-  * means the first coefficient packet is guaranteed to be aligned */
-const unsigned int AlignedBit = 0x80;
+  * means the first coefficient packet is guaranteed to be aligned.
+  * An expression cannot has the AlignedBit without the PacketAccessBit flag.
+  * In other words, this means we are allow to perform an aligned packet access to the first element regardless
+  * of the expression kind:
+  * \code
+  * expression.packet<Aligned>(0);
+  * \endcode
+  */
+EIGEN_DEPRECATED const unsigned int AlignedBit = 0x80;
 
 const unsigned int NestByRefBit = 0x100;
 
+/** \ingroup flags
+  *
+  * for an expression, this means that the storage order
+  * can be either row-major or column-major.
+  * The precise choice will be decided at evaluation time or when
+  * combined with other expressions.
+  * \sa \blank  \ref RowMajorBit, \ref TopicStorageOrders */
+const unsigned int NoPreferredStorageOrderBit = 0x200;
+
+/** \ingroup flags
+  *
+  * Means that the underlying coefficients can be accessed through pointers to the sparse (un)compressed storage format,
+  * that is, the expression provides:
+  * \code
+    inline const Scalar* valuePtr() const;
+    inline const Index* innerIndexPtr() const;
+    inline const Index* outerIndexPtr() const;
+    inline const Index* innerNonZeroPtr() const;
+    \endcode
+  */
+const unsigned int CompressedAccessBit = 0x400;
+
+
 // list of flags that are inherited by default
 const unsigned int HereditaryBits = RowMajorBit
-                                  | EvalBeforeNestingBit
-                                  | EvalBeforeAssigningBit;
+                                  | EvalBeforeNestingBit;
 
 /** \defgroup enums Enumerations
   * \ingroup Core_Module
@@ -160,9 +197,9 @@ const unsigned int HereditaryBits = RowMajorBit
   */
 
 /** \ingroup enums
-  * Enum containing possible values for the \p Mode parameter of 
-  * MatrixBase::selfadjointView() and MatrixBase::triangularView(). */
-enum {
+  * Enum containing possible values for the \c Mode or \c UpLo parameter of
+  * MatrixBase::selfadjointView() and MatrixBase::triangularView(), and selfadjoint solvers. */
+enum UpLoType {
   /** View matrix as a lower triangular matrix. */
   Lower=0x1,                      
   /** View matrix as an upper triangular matrix. */
@@ -186,12 +223,31 @@ enum {
 };
 
 /** \ingroup enums
-  * Enum for indicating whether an object is aligned or not. */
-enum { 
-  /** Object is not correctly aligned for vectorization. */
-  Unaligned=0, 
-  /** Object is aligned for vectorization. */
-  Aligned=1 
+  * Enum for indicating whether a buffer is aligned or not. */
+enum AlignmentType {
+  Unaligned=0,        /**< Data pointer has no specific alignment. */
+  Aligned8=8,         /**< Data pointer is aligned on a 8 bytes boundary. */
+  Aligned16=16,       /**< Data pointer is aligned on a 16 bytes boundary. */
+  Aligned32=32,       /**< Data pointer is aligned on a 32 bytes boundary. */
+  Aligned64=64,       /**< Data pointer is aligned on a 64 bytes boundary. */
+  Aligned128=128,     /**< Data pointer is aligned on a 128 bytes boundary. */
+  AlignedMask=255,
+  Aligned=16,         /**< \deprecated Synonym for Aligned16. */
+#if EIGEN_MAX_ALIGN_BYTES==128
+  AlignedMax = Aligned128
+#elif EIGEN_MAX_ALIGN_BYTES==64
+  AlignedMax = Aligned64
+#elif EIGEN_MAX_ALIGN_BYTES==32
+  AlignedMax = Aligned32
+#elif EIGEN_MAX_ALIGN_BYTES==16
+  AlignedMax = Aligned16
+#elif EIGEN_MAX_ALIGN_BYTES==8
+  AlignedMax = Aligned8
+#elif EIGEN_MAX_ALIGN_BYTES==0
+  AlignedMax = Unaligned
+#else
+#error Invalid value for EIGEN_MAX_ALIGN_BYTES
+#endif
 };
 
 /** \ingroup enums
@@ -217,7 +273,7 @@ enum DirectionType {
 
 /** \internal \ingroup enums
   * Enum to specify how to traverse the entries of a matrix. */
-enum {
+enum TraversalType {
   /** \internal Default traversal, no vectorization, no index-based access */
   DefaultTraversal,
   /** \internal No vectorization, use index-based access to have only one for loop instead of 2 nested loops */
@@ -239,7 +295,7 @@ enum {
 
 /** \internal \ingroup enums
   * Enum to specify whether to unroll loops when traversing over the entries of a matrix. */
-enum {
+enum UnrollingType {
   /** \internal Do not unroll loops. */
   NoUnrolling,
   /** \internal Unroll only the inner loop, but not the outer loop. */
@@ -251,7 +307,7 @@ enum {
 
 /** \internal \ingroup enums
   * Enum to specify whether to use the default (built-in) implementation or the specialization. */
-enum {
+enum SpecializedType {
   Specialized,
   BuiltIn
 };
@@ -259,7 +315,7 @@ enum {
 /** \ingroup enums
   * Enum containing possible values for the \p _Options template parameter of
   * Matrix, Array and BandMatrix. */
-enum {
+enum StorageOptions {
   /** Storage order is column major (see \ref TopicStorageOrders). */
   ColMajor = 0,
   /** Storage order is row major (see \ref TopicStorageOrders). */
@@ -272,7 +328,7 @@ enum {
 
 /** \ingroup enums
   * Enum for specifying whether to apply or solve on the left or right. */
-enum {
+enum SideType {
   /** Apply transformation on the left. */
   OnTheLeft = 1,  
   /** Apply transformation on the right. */
@@ -297,7 +353,7 @@ enum Default_t    { Default };
 
 /** \internal \ingroup enums
   * Used in AmbiVector. */
-enum {
+enum AmbiVectorMode {
   IsDense         = 0,
   IsSparse
 };
@@ -406,10 +462,16 @@ namespace Architecture
     Generic = 0x0,
     SSE = 0x1,
     AltiVec = 0x2,
+    VSX = 0x3,
+    NEON = 0x4,
 #if defined EIGEN_VECTORIZE_SSE
     Target = SSE
 #elif defined EIGEN_VECTORIZE_ALTIVEC
     Target = AltiVec
+#elif defined EIGEN_VECTORIZE_VSX
+    Target = VSX
+#elif defined EIGEN_VECTORIZE_NEON
+    Target = NEON
 #else
     Target = Generic
 #endif
@@ -417,8 +479,9 @@ namespace Architecture
 }
 
 /** \internal \ingroup enums
-  * Enum used as template parameter in GeneralProduct. */
-enum { CoeffBasedProductMode, LazyCoeffBasedProductMode, OuterProduct, InnerProduct, GemvProduct, GemmProduct };
+  * Enum used as template parameter in Product and product evaluators. */
+enum ProductImplType
+{ DefaultProduct=0, LazyProduct, AliasFreeProduct, CoeffBasedProductMode, LazyCoeffBasedProductMode, OuterProduct, InnerProduct, GemvProduct, GemmProduct };
 
 /** \internal \ingroup enums
   * Enum used in experimental parallel implementation. */
@@ -427,24 +490,57 @@ enum Action {GetAction, SetAction};
 /** The type used to identify a dense storage. */
 struct Dense {};
 
+/** The type used to identify a general sparse storage. */
+struct Sparse {};
+
+/** The type used to identify a general solver (factored) storage. */
+struct SolverStorage {};
+
+/** The type used to identify a permutation storage. */
+struct PermutationStorage {};
+
+/** The type used to identify a permutation storage. */
+struct TranspositionsStorage {};
+
 /** The type used to identify a matrix expression */
 struct MatrixXpr {};
 
 /** The type used to identify an array expression */
 struct ArrayXpr {};
 
+// An evaluator must define its shape. By default, it can be one of the following:
+struct DenseShape             { static std::string debugName() { return "DenseShape"; } };
+struct SolverShape            { static std::string debugName() { return "SolverShape"; } };
+struct HomogeneousShape       { static std::string debugName() { return "HomogeneousShape"; } };
+struct DiagonalShape          { static std::string debugName() { return "DiagonalShape"; } };
+struct BandShape              { static std::string debugName() { return "BandShape"; } };
+struct TriangularShape        { static std::string debugName() { return "TriangularShape"; } };
+struct SelfAdjointShape       { static std::string debugName() { return "SelfAdjointShape"; } };
+struct PermutationShape       { static std::string debugName() { return "PermutationShape"; } };
+struct TranspositionsShape    { static std::string debugName() { return "TranspositionsShape"; } };
+struct SparseShape            { static std::string debugName() { return "SparseShape"; } };
+
 namespace internal {
-  /** \internal
-  * Constants for comparison functors
-  */
-  enum ComparisonName {
-    cmp_EQ = 0,
-    cmp_LT = 1,
-    cmp_LE = 2,
-    cmp_UNORD = 3,
-    cmp_NEQ = 4
-  };
-}
+
+  // random access iterators based on coeff*() accessors.
+struct IndexBased {};
+
+// evaluator based on iterators to access coefficients. 
+struct IteratorBased {};
+
+/** \internal
+ * Constants for comparison functors
+ */
+enum ComparisonName {
+  cmp_EQ = 0,
+  cmp_LT = 1,
+  cmp_LE = 2,
+  cmp_UNORD = 3,
+  cmp_NEQ = 4,
+  cmp_GT = 5,
+  cmp_GE = 6
+};
+} // end namespace internal
 
 } // end namespace Eigen
 
diff --git a/extern/Eigen3/Eigen/src/Core/util/DisableStupidWarnings.h b/extern/Eigen3/Eigen/src/Core/util/DisableStupidWarnings.h
index 6a0bf0629c5..351bd6c6000 100644
--- a/extern/Eigen3/Eigen/src/Core/util/DisableStupidWarnings.h
+++ b/extern/Eigen3/Eigen/src/Core/util/DisableStupidWarnings.h
@@ -10,24 +10,31 @@
   // 4244 - 'argument' : conversion from 'type1' to 'type2', possible loss of data
   // 4273 - QtAlignedMalloc, inconsistent DLL linkage
   // 4324 - structure was padded due to declspec(align())
+  // 4503 - decorated name length exceeded, name was truncated
   // 4512 - assignment operator could not be generated
   // 4522 - 'class' : multiple assignment operators specified
   // 4700 - uninitialized local variable 'xyz' used
+  // 4714 - function marked as __forceinline not inlined
   // 4717 - 'function' : recursive on all control paths, function will cause runtime stack overflow
+  // 4800 - 'type' : forcing value to bool 'true' or 'false' (performance warning)
   #ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS
     #pragma warning( push )
   #endif
-  #pragma warning( disable : 4100 4101 4127 4181 4211 4244 4273 4324 4512 4522 4700 4717 )
+  #pragma warning( disable : 4100 4101 4127 4181 4211 4244 4273 4324 4503 4512 4522 4700 4714 4717 4800)
+
 #elif defined __INTEL_COMPILER
   // 2196 - routine is both "inline" and "noinline" ("noinline" assumed)
   //        ICC 12 generates this warning even without any inline keyword, when defining class methods 'inline' i.e. inside of class body
   //        typedef that may be a reference type.
   // 279  - controlling expression is constant
   //        ICC 12 generates this warning on assert(constant_expression_depending_on_template_params) and frankly this is a legitimate use case.
+  // 1684 - conversion from pointer to same-sized integral type (potential portability problem)
+  // 2259 - non-pointer conversion from "Eigen::Index={ptrdiff_t={long}}" to "int" may lose significant bits
   #ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS
     #pragma warning push
   #endif
-  #pragma warning disable 2196 279
+  #pragma warning disable 2196 279 1684 2259
+
 #elif defined __clang__
   // -Wconstant-logical-operand - warning: use of logical && with constant operand; switch to bitwise & or remove constant
   //     this is really a stupid warning as it warns on compile-time expressions involving enums
@@ -35,6 +42,42 @@
     #pragma clang diagnostic push
   #endif
   #pragma clang diagnostic ignored "-Wconstant-logical-operand"
+
+#elif defined __GNUC__
+
+  #if (!defined(EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS)) &&  (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))
+    #pragma GCC diagnostic push
+  #endif
+  // g++ warns about local variables shadowing member functions, which is too strict
+  #pragma GCC diagnostic ignored "-Wshadow"
+  #if __GNUC__ == 4 && __GNUC_MINOR__ < 8
+    // Until g++-4.7 there are warnings when comparing unsigned int vs 0, even in templated functions:
+    #pragma GCC diagnostic ignored "-Wtype-limits"
+  #endif
+  #if __GNUC__>=6
+    #pragma GCC diagnostic ignored "-Wignored-attributes"
+  #endif
+
+#endif
+
+#if defined __NVCC__
+  // Disable the "statement is unreachable" message
+  #pragma diag_suppress code_is_unreachable
+  // Disable the "dynamic initialization in unreachable code" message
+  #pragma diag_suppress initialization_not_reachable
+  // Disable the "invalid error number" message that we get with older versions of nvcc
+  #pragma diag_suppress 1222
+  // Disable the "calling a __host__ function from a __host__ __device__ function is not allowed" messages (yes, there are many of them and they seem to change with every version of the compiler)
+  #pragma diag_suppress 2527
+  #pragma diag_suppress 2529
+  #pragma diag_suppress 2651
+  #pragma diag_suppress 2653
+  #pragma diag_suppress 2668
+  #pragma diag_suppress 2669
+  #pragma diag_suppress 2670
+  #pragma diag_suppress 2671
+  #pragma diag_suppress 2735
+  #pragma diag_suppress 2737
 #endif
 
 #endif // not EIGEN_WARNINGS_DISABLED
diff --git a/extern/Eigen3/Eigen/src/Core/util/ForwardDeclarations.h b/extern/Eigen3/Eigen/src/Core/util/ForwardDeclarations.h
index f277720077b..ea107393a7d 100644
--- a/extern/Eigen3/Eigen/src/Core/util/ForwardDeclarations.h
+++ b/extern/Eigen3/Eigen/src/Core/util/ForwardDeclarations.h
@@ -36,6 +36,10 @@ template<typename Derived> struct accessors_level
   };
 };
 
+template<typename T> struct evaluator_traits;
+
+template< typename T> struct evaluator;
+
 } // end namespace internal
 
 template<typename T> struct NumTraits;
@@ -51,18 +55,18 @@ class DenseCoeffsBase;
 
 template<typename _Scalar, int _Rows, int _Cols,
          int _Options = AutoAlign |
-#if defined(__GNUC__) && __GNUC__==3 && __GNUC_MINOR__==4
+#if EIGEN_GNUC_AT(3,4)
     // workaround a bug in at least gcc 3.4.6
     // the innermost ?: ternary operator is misparsed. We write it slightly
     // differently and this makes gcc 3.4.6 happy, but it's ugly.
     // The error would only show up with EIGEN_DEFAULT_TO_ROW_MAJOR is defined
     // (when EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION is RowMajor)
-                          ( (_Rows==1 && _Cols!=1) ? RowMajor
+                          ( (_Rows==1 && _Cols!=1) ? Eigen::RowMajor
                           : !(_Cols==1 && _Rows!=1) ?  EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION
-                          : ColMajor ),
+                          : Eigen::ColMajor ),
 #else
-                          ( (_Rows==1 && _Cols!=1) ? RowMajor
-                          : (_Cols==1 && _Rows!=1) ? ColMajor
+                          ( (_Rows==1 && _Cols!=1) ? Eigen::RowMajor
+                          : (_Cols==1 && _Rows!=1) ? Eigen::ColMajor
                           : EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION ),
 #endif
          int _MaxRows = _Rows,
@@ -87,10 +91,11 @@ template<typename NullaryOp, typename MatrixType>         class CwiseNullaryOp;
 template<typename UnaryOp,   typename MatrixType>         class CwiseUnaryOp;
 template<typename ViewOp,    typename MatrixType>         class CwiseUnaryView;
 template<typename BinaryOp,  typename Lhs, typename Rhs>  class CwiseBinaryOp;
-template<typename BinOp,     typename Lhs, typename Rhs>  class SelfCwiseBinaryOp;
-template<typename Derived,   typename Lhs, typename Rhs>  class ProductBase;
-template<typename Lhs, typename Rhs, int Mode>            class GeneralProduct;
-template<typename Lhs, typename Rhs, int NestingFlags>    class CoeffBasedProduct;
+template<typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>  class CwiseTernaryOp;
+template<typename Decomposition, typename Rhstype>        class Solve;
+template<typename XprType>                                class Inverse;
+
+template<typename Lhs, typename Rhs, int Option = DefaultProduct> class Product;
 
 template<typename Derived> class DiagonalBase;
 template<typename _DiagonalVectorType> class DiagonalWrapper;
@@ -108,7 +113,12 @@ template<typename Derived,
          int Level = internal::accessors_level<Derived>::has_write_access ? WriteAccessors : ReadOnlyAccessors
 > class MapBase;
 template<int InnerStrideAtCompileTime, int OuterStrideAtCompileTime> class Stride;
+template<int Value = Dynamic> class InnerStride;
+template<int Value = Dynamic> class OuterStride;
 template<typename MatrixType, int MapOptions=Unaligned, typename StrideType = Stride<0,0> > class Map;
+template<typename Derived> class RefBase;
+template<typename PlainObjectType, int Options = 0,
+         typename StrideType = typename internal::conditional<PlainObjectType::IsVectorAtCompileTime,InnerStride<1>,OuterStride<> >::type > class Ref;
 
 template<typename Derived> class TriangularBase;
 template<typename MatrixType, unsigned int Mode> class TriangularView;
@@ -119,10 +129,10 @@ template<typename MatrixType> struct CommaInitializer;
 template<typename Derived> class ReturnByValue;
 template<typename ExpressionType> class ArrayWrapper;
 template<typename ExpressionType> class MatrixWrapper;
+template<typename Derived> class SolverBase;
+template<typename XprType> class InnerIterator;
 
 namespace internal {
-template<typename DecompositionType, typename Rhs> struct solve_retval_base;
-template<typename DecompositionType, typename Rhs> struct solve_retval;
 template<typename DecompositionType> struct kernel_retval_base;
 template<typename DecompositionType> struct kernel_retval;
 template<typename DecompositionType> struct image_retval_base;
@@ -135,6 +145,21 @@ template<typename _Scalar, int Rows=Dynamic, int Cols=Dynamic, int Supers=Dynami
 
 namespace internal {
 template<typename Lhs, typename Rhs> struct product_type;
+
+template<bool> struct EnableIf;
+
+/** \internal
+  * \class product_evaluator
+  * Products need their own evaluator with more template arguments allowing for
+  * easier partial template specializations.
+  */
+template< typename T,
+          int ProductTag = internal::product_type<typename T::Lhs,typename T::Rhs>::ret,
+          typename LhsShape = typename evaluator_traits<typename T::Lhs>::Shape,
+          typename RhsShape = typename evaluator_traits<typename T::Rhs>::Shape,
+          typename LhsScalar = typename traits<typename T::Lhs>::Scalar,
+          typename RhsScalar = typename traits<typename T::Rhs>::Scalar
+        > struct product_evaluator;
 }
 
 template<typename Lhs, typename Rhs,
@@ -150,9 +175,11 @@ namespace internal {
 // with optional conjugation of the arguments.
 template<typename LhsScalar, typename RhsScalar, bool ConjLhs=false, bool ConjRhs=false> struct conj_helper;
 
-template<typename Scalar> struct scalar_sum_op;
-template<typename Scalar> struct scalar_difference_op;
-template<typename LhsScalar,typename RhsScalar> struct scalar_conj_product_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_sum_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_difference_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_conj_product_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_min_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_max_op;
 template<typename Scalar> struct scalar_opposite_op;
 template<typename Scalar> struct scalar_conjugate_op;
 template<typename Scalar> struct scalar_real_op;
@@ -160,6 +187,7 @@ template<typename Scalar> struct scalar_imag_op;
 template<typename Scalar> struct scalar_abs_op;
 template<typename Scalar> struct scalar_abs2_op;
 template<typename Scalar> struct scalar_sqrt_op;
+template<typename Scalar> struct scalar_rsqrt_op;
 template<typename Scalar> struct scalar_exp_op;
 template<typename Scalar> struct scalar_log_op;
 template<typename Scalar> struct scalar_cos_op;
@@ -167,24 +195,29 @@ template<typename Scalar> struct scalar_sin_op;
 template<typename Scalar> struct scalar_acos_op;
 template<typename Scalar> struct scalar_asin_op;
 template<typename Scalar> struct scalar_tan_op;
-template<typename Scalar> struct scalar_pow_op;
 template<typename Scalar> struct scalar_inverse_op;
 template<typename Scalar> struct scalar_square_op;
 template<typename Scalar> struct scalar_cube_op;
 template<typename Scalar, typename NewType> struct scalar_cast_op;
-template<typename Scalar> struct scalar_multiple_op;
-template<typename Scalar> struct scalar_quotient1_op;
-template<typename Scalar> struct scalar_min_op;
-template<typename Scalar> struct scalar_max_op;
 template<typename Scalar> struct scalar_random_op;
-template<typename Scalar> struct scalar_add_op;
 template<typename Scalar> struct scalar_constant_op;
 template<typename Scalar> struct scalar_identity_op;
-
+template<typename Scalar,bool iscpx> struct scalar_sign_op;
+template<typename Scalar,typename ScalarExponent> struct scalar_pow_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_hypot_op;
 template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_product_op;
-template<typename LhsScalar,typename RhsScalar> struct scalar_multiple2_op;
 template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_quotient_op;
 
+// SpecialFunctions module
+template<typename Scalar> struct scalar_lgamma_op;
+template<typename Scalar> struct scalar_digamma_op;
+template<typename Scalar> struct scalar_erf_op;
+template<typename Scalar> struct scalar_erfc_op;
+template<typename Scalar> struct scalar_igamma_op;
+template<typename Scalar> struct scalar_igammac_op;
+template<typename Scalar> struct scalar_zeta_op;
+template<typename Scalar> struct scalar_betainc_op;
+
 } // end namespace internal
 
 struct IOFormat;
@@ -192,18 +225,18 @@ struct IOFormat;
 // Array module
 template<typename _Scalar, int _Rows, int _Cols,
          int _Options = AutoAlign |
-#if defined(__GNUC__) && __GNUC__==3 && __GNUC_MINOR__==4
+#if EIGEN_GNUC_AT(3,4)
     // workaround a bug in at least gcc 3.4.6
     // the innermost ?: ternary operator is misparsed. We write it slightly
     // differently and this makes gcc 3.4.6 happy, but it's ugly.
     // The error would only show up with EIGEN_DEFAULT_TO_ROW_MAJOR is defined
     // (when EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION is RowMajor)
-                          ( (_Rows==1 && _Cols!=1) ? RowMajor
+                          ( (_Rows==1 && _Cols!=1) ? Eigen::RowMajor
                           : !(_Cols==1 && _Rows!=1) ?  EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION
-                          : ColMajor ),
+                          : Eigen::ColMajor ),
 #else
-                          ( (_Rows==1 && _Cols!=1) ? RowMajor
-                          : (_Cols==1 && _Rows!=1) ? ColMajor
+                          ( (_Rows==1 && _Cols!=1) ? Eigen::RowMajor
+                          : (_Cols==1 && _Rows!=1) ? Eigen::ColMajor
                           : EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION ),
 #endif
          int _MaxRows = _Rows, int _MaxCols = _Cols> class Array;
@@ -221,7 +254,9 @@ template<typename MatrixType> struct inverse_impl;
 template<typename MatrixType> class HouseholderQR;
 template<typename MatrixType> class ColPivHouseholderQR;
 template<typename MatrixType> class FullPivHouseholderQR;
+template<typename MatrixType> class CompleteOrthogonalDecomposition;
 template<typename MatrixType, int QRPreconditioner = ColPivHouseholderQRPreconditioner> class JacobiSVD;
+template<typename MatrixType> class BDCSVD;
 template<typename MatrixType, int UpLo = Lower> class LLT;
 template<typename MatrixType, int UpLo = Lower> class LDLT;
 template<typename VectorsType, typename CoeffsType, int Side=OnTheLeft> class HouseholderSequence;
@@ -234,39 +269,16 @@ template<typename Derived> class QuaternionBase;
 template<typename Scalar> class Rotation2D;
 template<typename Scalar> class AngleAxis;
 template<typename Scalar,int Dim> class Translation;
-
-// Sparse module:
-template<typename Derived> class SparseMatrixBase;
-
-#ifdef EIGEN2_SUPPORT
-template<typename Derived, int _Dim> class eigen2_RotationBase;
-template<typename Lhs, typename Rhs> class eigen2_Cross;
-template<typename Scalar> class eigen2_Quaternion;
-template<typename Scalar> class eigen2_Rotation2D;
-template<typename Scalar> class eigen2_AngleAxis;
-template<typename Scalar,int Dim> class eigen2_Transform;
-template <typename _Scalar, int _AmbientDim> class eigen2_ParametrizedLine;
-template <typename _Scalar, int _AmbientDim> class eigen2_Hyperplane;
-template<typename Scalar,int Dim> class eigen2_Translation;
-template<typename Scalar,int Dim> class eigen2_Scaling;
-#endif
-
-#if EIGEN2_SUPPORT_STAGE < STAGE20_RESOLVE_API_CONFLICTS
-template<typename Scalar> class Quaternion;
-template<typename Scalar,int Dim> class Transform;
-template <typename _Scalar, int _AmbientDim> class ParametrizedLine;
-template <typename _Scalar, int _AmbientDim> class Hyperplane;
-template<typename Scalar,int Dim> class Scaling;
-#endif
-
-#if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
+template<typename Scalar,int Dim> class AlignedBox;
 template<typename Scalar, int Options = AutoAlign> class Quaternion;
 template<typename Scalar,int Dim,int Mode,int _Options=AutoAlign> class Transform;
 template <typename _Scalar, int _AmbientDim, int Options=AutoAlign> class ParametrizedLine;
 template <typename _Scalar, int _AmbientDim, int Options=AutoAlign> class Hyperplane;
 template<typename Scalar> class UniformScaling;
 template<typename MatrixType,int Direction> class Homogeneous;
-#endif
+
+// Sparse module:
+template<typename Derived> class SparseMatrixBase;
 
 // MatrixFunctions module
 template<typename Derived> struct MatrixExponentialReturnValue;
@@ -274,7 +286,7 @@ template<typename Derived> class MatrixFunctionReturnValue;
 template<typename Derived> class MatrixSquareRootReturnValue;
 template<typename Derived> class MatrixLogarithmReturnValue;
 template<typename Derived> class MatrixPowerReturnValue;
-template<typename Derived, typename Lhs, typename Rhs> class MatrixPowerProduct;
+template<typename Derived> class MatrixComplexPowerReturnValue;
 
 namespace internal {
 template <typename Scalar>
@@ -285,18 +297,6 @@ struct stem_function
 };
 }
 
-
-#ifdef EIGEN2_SUPPORT
-template<typename ExpressionType> class Cwise;
-template<typename MatrixType> class Minor;
-template<typename MatrixType> class LU;
-template<typename MatrixType> class QR;
-template<typename MatrixType> class SVD;
-namespace internal {
-template<typename MatrixType, unsigned int Mode> struct eigen2_part_return_type;
-}
-#endif
-
 } // end namespace Eigen
 
 #endif // EIGEN_FORWARDDECLARATIONS_H
diff --git a/extern/Eigen3/Eigen/src/Core/util/MKL_support.h b/extern/Eigen3/Eigen/src/Core/util/MKL_support.h
index 1ef3b61db14..b7d6ecc76e8 100644
--- a/extern/Eigen3/Eigen/src/Core/util/MKL_support.h
+++ b/extern/Eigen3/Eigen/src/Core/util/MKL_support.h
@@ -49,10 +49,11 @@
   #define EIGEN_USE_LAPACKE
 #endif
 
-#if defined(EIGEN_USE_BLAS) || defined(EIGEN_USE_LAPACKE) || defined(EIGEN_USE_MKL_VML)
+#if defined(EIGEN_USE_MKL_VML) && !defined(EIGEN_USE_MKL)
   #define EIGEN_USE_MKL
 #endif
 
+
 #if defined EIGEN_USE_MKL
 #   include <mkl.h> 
 /*Check IMKL version for compatibility: < 10.3 is not usable with Eigen*/
@@ -64,7 +65,6 @@
 #   ifndef EIGEN_USE_MKL
     /*If the MKL version is too old, undef everything*/
 #       undef   EIGEN_USE_MKL_ALL
-#       undef   EIGEN_USE_BLAS
 #       undef   EIGEN_USE_LAPACKE
 #       undef   EIGEN_USE_MKL_VML
 #       undef   EIGEN_USE_LAPACKE_STRICT
@@ -73,7 +73,7 @@
 #endif
 
 #if defined EIGEN_USE_MKL
-#include <mkl_lapacke.h>
+
 #define EIGEN_MKL_VML_THRESHOLD 128
 
 /* MKL_DOMAIN_BLAS, etc are defined only in 10.3 update 7 */
@@ -107,52 +107,24 @@
 #else
 #define EIGEN_MKL_DOMAIN_PARDISO MKL_PARDISO
 #endif
+#endif
+
+#if defined(EIGEN_USE_BLAS) && !defined(EIGEN_USE_MKL)
+#include "../../misc/blas.h"
+#endif
 
 namespace Eigen {
 
 typedef std::complex<double> dcomplex;
 typedef std::complex<float>  scomplex;
 
-namespace internal {
-
-template<typename MKLType, typename EigenType>
-static inline void assign_scalar_eig2mkl(MKLType& mklScalar, const EigenType& eigenScalar) {
-  mklScalar=eigenScalar;
-}
-
-template<typename MKLType, typename EigenType>
-static inline void assign_conj_scalar_eig2mkl(MKLType& mklScalar, const EigenType& eigenScalar) {
-  mklScalar=eigenScalar;
-}
-
-template <>
-inline void assign_scalar_eig2mkl<MKL_Complex16,dcomplex>(MKL_Complex16& mklScalar, const dcomplex& eigenScalar) {
-  mklScalar.real=eigenScalar.real();
-  mklScalar.imag=eigenScalar.imag();
-}
-
-template <>
-inline void assign_scalar_eig2mkl<MKL_Complex8,scomplex>(MKL_Complex8& mklScalar, const scomplex& eigenScalar) {
-  mklScalar.real=eigenScalar.real();
-  mklScalar.imag=eigenScalar.imag();
-}
-
-template <>
-inline void assign_conj_scalar_eig2mkl<MKL_Complex16,dcomplex>(MKL_Complex16& mklScalar, const dcomplex& eigenScalar) {
-  mklScalar.real=eigenScalar.real();
-  mklScalar.imag=-eigenScalar.imag();
-}
-
-template <>
-inline void assign_conj_scalar_eig2mkl<MKL_Complex8,scomplex>(MKL_Complex8& mklScalar, const scomplex& eigenScalar) {
-  mklScalar.real=eigenScalar.real();
-  mklScalar.imag=-eigenScalar.imag();
-}
-
-} // end namespace internal
+#if defined(EIGEN_USE_MKL)
+typedef MKL_INT BlasIndex;
+#else
+typedef int BlasIndex;
+#endif
 
 } // end namespace Eigen
 
-#endif
 
 #endif // EIGEN_MKL_SUPPORT_H
diff --git a/extern/Eigen3/Eigen/src/Core/util/Macros.h b/extern/Eigen3/Eigen/src/Core/util/Macros.h
index 53fb5fae420..aa054a0b7ff 100644
--- a/extern/Eigen3/Eigen/src/Core/util/Macros.h
+++ b/extern/Eigen3/Eigen/src/Core/util/Macros.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -12,115 +12,461 @@
 #define EIGEN_MACROS_H
 
 #define EIGEN_WORLD_VERSION 3
-#define EIGEN_MAJOR_VERSION 2
+#define EIGEN_MAJOR_VERSION 3
 #define EIGEN_MINOR_VERSION 7
 
 #define EIGEN_VERSION_AT_LEAST(x,y,z) (EIGEN_WORLD_VERSION>x || (EIGEN_WORLD_VERSION>=x && \
                                       (EIGEN_MAJOR_VERSION>y || (EIGEN_MAJOR_VERSION>=y && \
                                                                  EIGEN_MINOR_VERSION>=z))))
+
+// Compiler identification, EIGEN_COMP_*
+
+/// \internal EIGEN_COMP_GNUC set to 1 for all compilers compatible with GCC
 #ifdef __GNUC__
-  #define EIGEN_GNUC_AT_LEAST(x,y) ((__GNUC__==x && __GNUC_MINOR__>=y) || __GNUC__>x)
+  #define EIGEN_COMP_GNUC 1
 #else
-  #define EIGEN_GNUC_AT_LEAST(x,y) 0
+  #define EIGEN_COMP_GNUC 0
 #endif
- 
-#ifdef __GNUC__
-  #define EIGEN_GNUC_AT_MOST(x,y) ((__GNUC__==x && __GNUC_MINOR__<=y) || __GNUC__<x)
+
+/// \internal EIGEN_COMP_CLANG set to major+minor version (e.g., 307 for clang 3.7) if the compiler is clang
+#if defined(__clang__)
+  #define EIGEN_COMP_CLANG (__clang_major__*100+__clang_minor__)
 #else
-  #define EIGEN_GNUC_AT_MOST(x,y) 0
+  #define EIGEN_COMP_CLANG 0
 #endif
 
-#if EIGEN_GNUC_AT_MOST(4,3) && !defined(__clang__)
-  // see bug 89
-  #define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 0
+
+/// \internal EIGEN_COMP_LLVM set to 1 if the compiler backend is llvm
+#if defined(__llvm__)
+  #define EIGEN_COMP_LLVM 1
 #else
-  #define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 1
+  #define EIGEN_COMP_LLVM 0
+#endif
+
+/// \internal EIGEN_COMP_ICC set to __INTEL_COMPILER if the compiler is Intel compiler, 0 otherwise
+#if defined(__INTEL_COMPILER)
+  #define EIGEN_COMP_ICC __INTEL_COMPILER
+#else
+  #define EIGEN_COMP_ICC 0
+#endif
+
+/// \internal EIGEN_COMP_MINGW set to 1 if the compiler is mingw
+#if defined(__MINGW32__)
+  #define EIGEN_COMP_MINGW 1
+#else
+  #define EIGEN_COMP_MINGW 0
+#endif
+
+/// \internal EIGEN_COMP_SUNCC set to 1 if the compiler is Solaris Studio
+#if defined(__SUNPRO_CC)
+  #define EIGEN_COMP_SUNCC 1
+#else
+  #define EIGEN_COMP_SUNCC 0
+#endif
+
+/// \internal EIGEN_COMP_MSVC set to _MSC_VER if the compiler is Microsoft Visual C++, 0 otherwise.
+#if defined(_MSC_VER)
+  #define EIGEN_COMP_MSVC _MSC_VER
+#else
+  #define EIGEN_COMP_MSVC 0
+#endif
+
+// For the record, here is a table summarizing the possible values for EIGEN_COMP_MSVC:
+//  name  ver   MSC_VER
+//  2008    9      1500
+//  2010   10      1600
+//  2012   11      1700
+//  2013   12      1800
+//  2015   14      1900
+//  "15"   15      1900
+
+/// \internal EIGEN_COMP_MSVC_STRICT set to 1 if the compiler is really Microsoft Visual C++ and not ,e.g., ICC or clang-cl
+#if EIGEN_COMP_MSVC && !(EIGEN_COMP_ICC || EIGEN_COMP_LLVM || EIGEN_COMP_CLANG)
+  #define EIGEN_COMP_MSVC_STRICT _MSC_VER
+#else
+  #define EIGEN_COMP_MSVC_STRICT 0
 #endif
 
-#if defined(__GNUC__) && (__GNUC__ <= 3)
+/// \internal EIGEN_COMP_IBM set to 1 if the compiler is IBM XL C++
+#if defined(__IBMCPP__) || defined(__xlc__)
+  #define EIGEN_COMP_IBM 1
+#else
+  #define EIGEN_COMP_IBM 0
+#endif
+
+/// \internal EIGEN_COMP_PGI set to 1 if the compiler is Portland Group Compiler
+#if defined(__PGI)
+  #define EIGEN_COMP_PGI 1
+#else
+  #define EIGEN_COMP_PGI 0
+#endif
+
+/// \internal EIGEN_COMP_ARM set to 1 if the compiler is ARM Compiler
+#if defined(__CC_ARM) || defined(__ARMCC_VERSION)
+  #define EIGEN_COMP_ARM 1
+#else
+  #define EIGEN_COMP_ARM 0
+#endif
+
+/// \internal EIGEN_COMP_ARM set to 1 if the compiler is ARM Compiler
+#if defined(__EMSCRIPTEN__)
+  #define EIGEN_COMP_EMSCRIPTEN 1
+#else
+  #define EIGEN_COMP_EMSCRIPTEN 0
+#endif
+
+
+/// \internal EIGEN_GNUC_STRICT set to 1 if the compiler is really GCC and not a compatible compiler (e.g., ICC, clang, mingw, etc.)
+#if EIGEN_COMP_GNUC && !(EIGEN_COMP_CLANG || EIGEN_COMP_ICC || EIGEN_COMP_MINGW || EIGEN_COMP_PGI || EIGEN_COMP_IBM || EIGEN_COMP_ARM || EIGEN_COMP_EMSCRIPTEN)
+  #define EIGEN_COMP_GNUC_STRICT 1
+#else
+  #define EIGEN_COMP_GNUC_STRICT 0
+#endif
+
+
+#if EIGEN_COMP_GNUC
+  #define EIGEN_GNUC_AT_LEAST(x,y) ((__GNUC__==x && __GNUC_MINOR__>=y) || __GNUC__>x)
+  #define EIGEN_GNUC_AT_MOST(x,y)  ((__GNUC__==x && __GNUC_MINOR__<=y) || __GNUC__<x)
+  #define EIGEN_GNUC_AT(x,y)       ( __GNUC__==x && __GNUC_MINOR__==y )
+#else
+  #define EIGEN_GNUC_AT_LEAST(x,y) 0
+  #define EIGEN_GNUC_AT_MOST(x,y)  0
+  #define EIGEN_GNUC_AT(x,y)       0
+#endif
+
+// FIXME: could probably be removed as we do not support gcc 3.x anymore
+#if EIGEN_COMP_GNUC && (__GNUC__ <= 3)
 #define EIGEN_GCC3_OR_OLDER 1
 #else
 #define EIGEN_GCC3_OR_OLDER 0
 #endif
 
-// 16 byte alignment is only useful for vectorization. Since it affects the ABI, we need to enable
-// 16 byte alignment on all platforms where vectorization might be enabled. In theory we could always
-// enable alignment, but it can be a cause of problems on some platforms, so we just disable it in
-// certain common platform (compiler+architecture combinations) to avoid these problems.
-// Only static alignment is really problematic (relies on nonstandard compiler extensions that don't
-// work everywhere, for example don't work on GCC/ARM), try to keep heap alignment even
-// when we have to disable static alignment.
-#if defined(__GNUC__) && !(defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) || defined(__ppc__) || defined(__ia64__))
-#define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1
+
+// Architecture identification, EIGEN_ARCH_*
+
+#if defined(__x86_64__) || defined(_M_X64) || defined(__amd64)
+  #define EIGEN_ARCH_x86_64 1
 #else
-#define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 0
+  #define EIGEN_ARCH_x86_64 0
 #endif
 
-// static alignment is completely disabled with GCC 3, Sun Studio, and QCC/QNX
-#if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT \
- && !EIGEN_GCC3_OR_OLDER \
- && !defined(__SUNPRO_CC) \
- && !defined(__QNXNTO__)
-  #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 1
+#if defined(__i386__) || defined(_M_IX86) || defined(_X86_) || defined(__i386)
+  #define EIGEN_ARCH_i386 1
 #else
-  #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 0
+  #define EIGEN_ARCH_i386 0
 #endif
 
-#ifdef EIGEN_DONT_ALIGN
-  #ifndef EIGEN_DONT_ALIGN_STATICALLY
-    #define EIGEN_DONT_ALIGN_STATICALLY
-  #endif
-  #define EIGEN_ALIGN 0
+#if EIGEN_ARCH_x86_64 || EIGEN_ARCH_i386
+  #define EIGEN_ARCH_i386_OR_x86_64 1
 #else
-  #define EIGEN_ALIGN 1
+  #define EIGEN_ARCH_i386_OR_x86_64 0
 #endif
 
-// EIGEN_ALIGN_STATICALLY is the true test whether we want to align arrays on the stack or not. It takes into account both the user choice to explicitly disable
-// alignment (EIGEN_DONT_ALIGN_STATICALLY) and the architecture config (EIGEN_ARCH_WANTS_STACK_ALIGNMENT). Henceforth, only EIGEN_ALIGN_STATICALLY should be used.
-#if EIGEN_ARCH_WANTS_STACK_ALIGNMENT && !defined(EIGEN_DONT_ALIGN_STATICALLY)
-  #define EIGEN_ALIGN_STATICALLY 1
+/// \internal EIGEN_ARCH_ARM set to 1 if the architecture is ARM
+#if defined(__arm__)
+  #define EIGEN_ARCH_ARM 1
 #else
-  #define EIGEN_ALIGN_STATICALLY 0
-  #ifndef EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT
-    #define EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT
-  #endif
+  #define EIGEN_ARCH_ARM 0
+#endif
+
+/// \internal EIGEN_ARCH_ARM64 set to 1 if the architecture is ARM64
+#if defined(__aarch64__)
+  #define EIGEN_ARCH_ARM64 1
+#else
+  #define EIGEN_ARCH_ARM64 0
+#endif
+
+#if EIGEN_ARCH_ARM || EIGEN_ARCH_ARM64
+  #define EIGEN_ARCH_ARM_OR_ARM64 1
+#else
+  #define EIGEN_ARCH_ARM_OR_ARM64 0
+#endif
+
+/// \internal EIGEN_ARCH_MIPS set to 1 if the architecture is MIPS
+#if defined(__mips__) || defined(__mips)
+  #define EIGEN_ARCH_MIPS 1
+#else
+  #define EIGEN_ARCH_MIPS 0
+#endif
+
+/// \internal EIGEN_ARCH_SPARC set to 1 if the architecture is SPARC
+#if defined(__sparc__) || defined(__sparc)
+  #define EIGEN_ARCH_SPARC 1
+#else
+  #define EIGEN_ARCH_SPARC 0
+#endif
+
+/// \internal EIGEN_ARCH_IA64 set to 1 if the architecture is Intel Itanium
+#if defined(__ia64__)
+  #define EIGEN_ARCH_IA64 1
+#else
+  #define EIGEN_ARCH_IA64 0
+#endif
+
+/// \internal EIGEN_ARCH_PPC set to 1 if the architecture is PowerPC
+#if defined(__powerpc__) || defined(__ppc__) || defined(_M_PPC)
+  #define EIGEN_ARCH_PPC 1
+#else
+  #define EIGEN_ARCH_PPC 0
+#endif
+
+
+
+// Operating system identification, EIGEN_OS_*
+
+/// \internal EIGEN_OS_UNIX set to 1 if the OS is a unix variant
+#if defined(__unix__) || defined(__unix)
+  #define EIGEN_OS_UNIX 1
+#else
+  #define EIGEN_OS_UNIX 0
+#endif
+
+/// \internal EIGEN_OS_LINUX set to 1 if the OS is based on Linux kernel
+#if defined(__linux__)
+  #define EIGEN_OS_LINUX 1
+#else
+  #define EIGEN_OS_LINUX 0
+#endif
+
+/// \internal EIGEN_OS_ANDROID set to 1 if the OS is Android
+// note: ANDROID is defined when using ndk_build, __ANDROID__ is defined when using a standalone toolchain.
+#if defined(__ANDROID__) || defined(ANDROID)
+  #define EIGEN_OS_ANDROID 1
+#else
+  #define EIGEN_OS_ANDROID 0
+#endif
+
+/// \internal EIGEN_OS_GNULINUX set to 1 if the OS is GNU Linux and not Linux-based OS (e.g., not android)
+#if defined(__gnu_linux__) && !(EIGEN_OS_ANDROID)
+  #define EIGEN_OS_GNULINUX 1
+#else
+  #define EIGEN_OS_GNULINUX 0
+#endif
+
+/// \internal EIGEN_OS_BSD set to 1 if the OS is a BSD variant
+#if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__bsdi__) || defined(__DragonFly__)
+  #define EIGEN_OS_BSD 1
+#else
+  #define EIGEN_OS_BSD 0
+#endif
+
+/// \internal EIGEN_OS_MAC set to 1 if the OS is MacOS
+#if defined(__APPLE__)
+  #define EIGEN_OS_MAC 1
+#else
+  #define EIGEN_OS_MAC 0
+#endif
+
+/// \internal EIGEN_OS_QNX set to 1 if the OS is QNX
+#if defined(__QNX__)
+  #define EIGEN_OS_QNX 1
+#else
+  #define EIGEN_OS_QNX 0
+#endif
+
+/// \internal EIGEN_OS_WIN set to 1 if the OS is Windows based
+#if defined(_WIN32)
+  #define EIGEN_OS_WIN 1
+#else
+  #define EIGEN_OS_WIN 0
+#endif
+
+/// \internal EIGEN_OS_WIN64 set to 1 if the OS is Windows 64bits
+#if defined(_WIN64)
+  #define EIGEN_OS_WIN64 1
+#else
+  #define EIGEN_OS_WIN64 0
 #endif
 
+/// \internal EIGEN_OS_WINCE set to 1 if the OS is Windows CE
+#if defined(_WIN32_WCE)
+  #define EIGEN_OS_WINCE 1
+#else
+  #define EIGEN_OS_WINCE 0
+#endif
+
+/// \internal EIGEN_OS_CYGWIN set to 1 if the OS is Windows/Cygwin
+#if defined(__CYGWIN__)
+  #define EIGEN_OS_CYGWIN 1
+#else
+  #define EIGEN_OS_CYGWIN 0
+#endif
+
+/// \internal EIGEN_OS_WIN_STRICT set to 1 if the OS is really Windows and not some variants
+#if EIGEN_OS_WIN && !( EIGEN_OS_WINCE || EIGEN_OS_CYGWIN )
+  #define EIGEN_OS_WIN_STRICT 1
+#else
+  #define EIGEN_OS_WIN_STRICT 0
+#endif
+
+/// \internal EIGEN_OS_SUN set to 1 if the OS is SUN
+#if (defined(sun) || defined(__sun)) && !(defined(__SVR4) || defined(__svr4__))
+  #define EIGEN_OS_SUN 1
+#else
+  #define EIGEN_OS_SUN 0
+#endif
+
+/// \internal EIGEN_OS_SOLARIS set to 1 if the OS is Solaris
+#if (defined(sun) || defined(__sun)) && (defined(__SVR4) || defined(__svr4__))
+  #define EIGEN_OS_SOLARIS 1
+#else
+  #define EIGEN_OS_SOLARIS 0
+#endif
+
+
+
+#if EIGEN_GNUC_AT_MOST(4,3) && !EIGEN_COMP_CLANG
+  // see bug 89
+  #define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 0
+#else
+  #define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 1
+#endif
+
+// This macro can be used to prevent from macro expansion, e.g.:
+//   std::max EIGEN_NOT_A_MACRO(a,b)
+#define EIGEN_NOT_A_MACRO
+
 #ifdef EIGEN_DEFAULT_TO_ROW_MAJOR
-#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION RowMajor
+#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION Eigen::RowMajor
 #else
-#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION ColMajor
+#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION Eigen::ColMajor
 #endif
 
 #ifndef EIGEN_DEFAULT_DENSE_INDEX_TYPE
 #define EIGEN_DEFAULT_DENSE_INDEX_TYPE std::ptrdiff_t
 #endif
 
+// Cross compiler wrapper around LLVM's __has_builtin
+#ifdef __has_builtin
+#  define EIGEN_HAS_BUILTIN(x) __has_builtin(x)
+#else
+#  define EIGEN_HAS_BUILTIN(x) 0
+#endif
+
 // A Clang feature extension to determine compiler features.
 // We use it to determine 'cxx_rvalue_references'
 #ifndef __has_feature
 # define __has_feature(x) 0
 #endif
 
+// Upperbound on the C++ version to use.
+// Expected values are 03, 11, 14, 17, etc.
+// By default, let's use an arbitrarily large C++ version.
+#ifndef EIGEN_MAX_CPP_VER
+#define EIGEN_MAX_CPP_VER 99
+#endif
+
+#if EIGEN_MAX_CPP_VER>=11 && (defined(__cplusplus) && (__cplusplus >= 201103L) || EIGEN_COMP_MSVC >= 1900)
+#define EIGEN_HAS_CXX11 1
+#else
+#define EIGEN_HAS_CXX11 0
+#endif
+
+
 // Do we support r-value references?
-#if (__has_feature(cxx_rvalue_references) || \
-     defined(__GXX_EXPERIMENTAL_CXX0X__) || \
-     (defined(_MSC_VER) && _MSC_VER >= 1600))
-  #define EIGEN_HAVE_RVALUE_REFERENCES
+#ifndef EIGEN_HAS_RVALUE_REFERENCES
+#if EIGEN_MAX_CPP_VER>=11 && \
+    (__has_feature(cxx_rvalue_references) || \
+    (defined(__cplusplus) && __cplusplus >= 201103L) || \
+    (EIGEN_COMP_MSVC >= 1600))
+  #define EIGEN_HAS_RVALUE_REFERENCES 1
+#else
+  #define EIGEN_HAS_RVALUE_REFERENCES 0
+#endif
 #endif
 
+// Does the compiler support C99?
+#ifndef EIGEN_HAS_C99_MATH
+#if EIGEN_MAX_CPP_VER>=11 && \
+    ((defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901))       \
+  || (defined(__GNUC__) && defined(_GLIBCXX_USE_C99)) \
+  || (defined(_LIBCPP_VERSION) && !defined(_MSC_VER)))
+  #define EIGEN_HAS_C99_MATH 1
+#else
+  #define EIGEN_HAS_C99_MATH 0
+#endif
+#endif
 
-// Cross compiler wrapper around LLVM's __has_builtin
-#ifdef __has_builtin
-#  define EIGEN_HAS_BUILTIN(x) __has_builtin(x)
+// Does the compiler support result_of?
+#ifndef EIGEN_HAS_STD_RESULT_OF
+#if EIGEN_MAX_CPP_VER>=11 && ((__has_feature(cxx_lambdas) || (defined(__cplusplus) && __cplusplus >= 201103L)))
+#define EIGEN_HAS_STD_RESULT_OF 1
 #else
-#  define EIGEN_HAS_BUILTIN(x) 0
+#define EIGEN_HAS_STD_RESULT_OF 0
+#endif
+#endif
+
+// Does the compiler support variadic templates?
+#ifndef EIGEN_HAS_VARIADIC_TEMPLATES
+#if EIGEN_MAX_CPP_VER>=11 && (__cplusplus > 199711L || EIGEN_COMP_MSVC >= 1900) \
+  && (!defined(__NVCC__) || !EIGEN_ARCH_ARM_OR_ARM64 || (EIGEN_CUDACC_VER >= 80000) )
+    // ^^ Disable the use of variadic templates when compiling with versions of nvcc older than 8.0 on ARM devices:
+    //    this prevents nvcc from crashing when compiling Eigen on Tegra X1
+#define EIGEN_HAS_VARIADIC_TEMPLATES 1
+#else
+#define EIGEN_HAS_VARIADIC_TEMPLATES 0
+#endif
+#endif
+
+// Does the compiler fully support const expressions? (as in c++14)
+#ifndef EIGEN_HAS_CONSTEXPR
+
+#ifdef __CUDACC__
+// Const expressions are supported provided that c++11 is enabled and we're using either clang or nvcc 7.5 or above
+#if EIGEN_MAX_CPP_VER>=14 && (__cplusplus > 199711L && (EIGEN_COMP_CLANG || EIGEN_CUDACC_VER >= 70500))
+  #define EIGEN_HAS_CONSTEXPR 1
+#endif
+#elif EIGEN_MAX_CPP_VER>=14 && (__has_feature(cxx_relaxed_constexpr) || (defined(__cplusplus) && __cplusplus >= 201402L) || \
+  (EIGEN_GNUC_AT_LEAST(4,8) && (__cplusplus > 199711L)))
+#define EIGEN_HAS_CONSTEXPR 1
+#endif
+
+#ifndef EIGEN_HAS_CONSTEXPR
+#define EIGEN_HAS_CONSTEXPR 0
+#endif
+
+#endif
+
+// Does the compiler support C++11 math?
+// Let's be conservative and enable the default C++11 implementation only if we are sure it exists
+#ifndef EIGEN_HAS_CXX11_MATH
+  #if EIGEN_MAX_CPP_VER>=11 && ((__cplusplus > 201103L) || (__cplusplus >= 201103L) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_CLANG || EIGEN_COMP_MSVC || EIGEN_COMP_ICC)  \
+      && (EIGEN_ARCH_i386_OR_x86_64) && (EIGEN_OS_GNULINUX || EIGEN_OS_WIN_STRICT || EIGEN_OS_MAC))
+    #define EIGEN_HAS_CXX11_MATH 1
+  #else
+    #define EIGEN_HAS_CXX11_MATH 0
+  #endif
+#endif
+
+// Does the compiler support proper C++11 containers?
+#ifndef EIGEN_HAS_CXX11_CONTAINERS
+  #if    EIGEN_MAX_CPP_VER>=11 && \
+         ((__cplusplus > 201103L) \
+      || ((__cplusplus >= 201103L) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_CLANG || EIGEN_COMP_ICC>=1400)) \
+      || EIGEN_COMP_MSVC >= 1900)
+    #define EIGEN_HAS_CXX11_CONTAINERS 1
+  #else
+    #define EIGEN_HAS_CXX11_CONTAINERS 0
+  #endif
+#endif
+
+// Does the compiler support C++11 noexcept?
+#ifndef EIGEN_HAS_CXX11_NOEXCEPT
+  #if    EIGEN_MAX_CPP_VER>=11 && \
+         (__has_feature(cxx_noexcept) \
+      || (__cplusplus > 201103L) \
+      || ((__cplusplus >= 201103L) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_CLANG || EIGEN_COMP_ICC>=1400)) \
+      || EIGEN_COMP_MSVC >= 1900)
+    #define EIGEN_HAS_CXX11_NOEXCEPT 1
+  #else
+    #define EIGEN_HAS_CXX11_NOEXCEPT 0
+  #endif
 #endif
 
 /** Allows to disable some optimizations which might affect the accuracy of the result.
   * Such optimization are enabled by default, and set EIGEN_FAST_MATH to 0 to disable them.
   * They currently include:
-  *   - single precision Cwise::sin() and Cwise::cos() when SSE vectorization is enabled.
+  *   - single precision ArrayBase::sin() and ArrayBase::cos() for SSE and AVX vectorization.
   */
 #ifndef EIGEN_FAST_MATH
 #define EIGEN_FAST_MATH 1
@@ -132,6 +478,8 @@
 #define EIGEN_CAT2(a,b) a ## b
 #define EIGEN_CAT(a,b) EIGEN_CAT2(a,b)
 
+#define EIGEN_COMMA ,
+
 // convert a token to a string
 #define EIGEN_MAKESTRING2(a) #a
 #define EIGEN_MAKESTRING(a) EIGEN_MAKESTRING2(a)
@@ -139,34 +487,37 @@
 // EIGEN_STRONG_INLINE is a stronger version of the inline, using __forceinline on MSVC,
 // but it still doesn't use GCC's always_inline. This is useful in (common) situations where MSVC needs forceinline
 // but GCC is still doing fine with just inline.
-#if (defined _MSC_VER) || (defined __INTEL_COMPILER)
+#ifndef EIGEN_STRONG_INLINE
+#if EIGEN_COMP_MSVC || EIGEN_COMP_ICC
 #define EIGEN_STRONG_INLINE __forceinline
 #else
 #define EIGEN_STRONG_INLINE inline
 #endif
+#endif
 
 // EIGEN_ALWAYS_INLINE is the stronget, it has the effect of making the function inline and adding every possible
 // attribute to maximize inlining. This should only be used when really necessary: in particular,
 // it uses __attribute__((always_inline)) on GCC, which most of the time is useless and can severely harm compile times.
 // FIXME with the always_inline attribute,
-// gcc 3.4.x reports the following compilation error:
+// gcc 3.4.x and 4.1 reports the following compilation error:
 //   Eval.h:91: sorry, unimplemented: inlining failed in call to 'const Eigen::Eval<Derived> Eigen::MatrixBase<Scalar, Derived>::eval() const'
 //    : function body not available
-#if EIGEN_GNUC_AT_LEAST(4,0)
+//   See also bug 1367
+#if EIGEN_GNUC_AT_LEAST(4,2)
 #define EIGEN_ALWAYS_INLINE __attribute__((always_inline)) inline
 #else
 #define EIGEN_ALWAYS_INLINE EIGEN_STRONG_INLINE
 #endif
 
-#if (defined __GNUC__)
+#if EIGEN_COMP_GNUC
 #define EIGEN_DONT_INLINE __attribute__((noinline))
-#elif (defined _MSC_VER)
+#elif EIGEN_COMP_MSVC
 #define EIGEN_DONT_INLINE __declspec(noinline)
 #else
 #define EIGEN_DONT_INLINE
 #endif
 
-#if (defined __GNUC__)
+#if EIGEN_COMP_GNUC
 #define EIGEN_PERMISSIVE_EXPR __extension__
 #else
 #define EIGEN_PERMISSIVE_EXPR
@@ -235,15 +586,15 @@
 #endif
 
 #ifdef EIGEN_NO_DEBUG
-#define EIGEN_ONLY_USED_FOR_DEBUG(x) (void)x
+#define EIGEN_ONLY_USED_FOR_DEBUG(x) EIGEN_UNUSED_VARIABLE(x)
 #else
 #define EIGEN_ONLY_USED_FOR_DEBUG(x)
 #endif
 
 #ifndef EIGEN_NO_DEPRECATED_WARNING
-  #if (defined __GNUC__)
+  #if EIGEN_COMP_GNUC
     #define EIGEN_DEPRECATED __attribute__((deprecated))
-  #elif (defined _MSC_VER)
+  #elif EIGEN_COMP_MSVC
     #define EIGEN_DEPRECATED __declspec(deprecated)
   #else
     #define EIGEN_DEPRECATED
@@ -252,7 +603,7 @@
   #define EIGEN_DEPRECATED
 #endif
 
-#if (defined __GNUC__)
+#if EIGEN_COMP_GNUC
 #define EIGEN_UNUSED __attribute__((unused))
 #else
 #define EIGEN_UNUSED
@@ -261,19 +612,33 @@
 // Suppresses 'unused variable' warnings.
 namespace Eigen {
   namespace internal {
-    template<typename T> void ignore_unused_variable(const T&) {}
+    template<typename T> EIGEN_DEVICE_FUNC void ignore_unused_variable(const T&) {}
   }
 }
 #define EIGEN_UNUSED_VARIABLE(var) Eigen::internal::ignore_unused_variable(var);
 
 #if !defined(EIGEN_ASM_COMMENT)
-  #if (defined __GNUC__) && ( defined(__i386__) || defined(__x86_64__) )
+  #if EIGEN_COMP_GNUC && (EIGEN_ARCH_i386_OR_x86_64 || EIGEN_ARCH_ARM_OR_ARM64)
     #define EIGEN_ASM_COMMENT(X)  __asm__("#" X)
   #else
     #define EIGEN_ASM_COMMENT(X)
   #endif
 #endif
 
+
+//------------------------------------------------------------------------------------------
+// Static and dynamic alignment control
+//
+// The main purpose of this section is to define EIGEN_MAX_ALIGN_BYTES and EIGEN_MAX_STATIC_ALIGN_BYTES
+// as the maximal boundary in bytes on which dynamically and statically allocated data may be alignment respectively.
+// The values of EIGEN_MAX_ALIGN_BYTES and EIGEN_MAX_STATIC_ALIGN_BYTES can be specified by the user. If not,
+// a default value is automatically computed based on architecture, compiler, and OS.
+//
+// This section also defines macros EIGEN_ALIGN_TO_BOUNDARY(N) and the shortcuts EIGEN_ALIGN{8,16,32,_MAX}
+// to be used to declare statically aligned buffers.
+//------------------------------------------------------------------------------------------
+
+
 /* EIGEN_ALIGN_TO_BOUNDARY(n) forces data to be n-byte aligned. This is used to satisfy SIMD requirements.
  * However, we do that EVEN if vectorization (EIGEN_VECTORIZE) is disabled,
  * so that vectorization doesn't affect binary compatibility.
@@ -281,28 +646,149 @@ namespace Eigen {
  * If we made alignment depend on whether or not EIGEN_VECTORIZE is defined, it would be impossible to link
  * vectorized and non-vectorized code.
  */
-#if (defined __GNUC__) || (defined __PGI) || (defined __IBMCPP__) || (defined __ARMCC_VERSION)
+#if (defined __CUDACC__)
+  #define EIGEN_ALIGN_TO_BOUNDARY(n) __align__(n)
+#elif EIGEN_COMP_GNUC || EIGEN_COMP_PGI || EIGEN_COMP_IBM || EIGEN_COMP_ARM
   #define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n)))
-#elif (defined _MSC_VER)
+#elif EIGEN_COMP_MSVC
   #define EIGEN_ALIGN_TO_BOUNDARY(n) __declspec(align(n))
-#elif (defined __SUNPRO_CC)
+#elif EIGEN_COMP_SUNCC
   // FIXME not sure about this one:
   #define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n)))
 #else
   #error Please tell me what is the equivalent of __attribute__((aligned(n))) for your compiler
 #endif
 
+// If the user explicitly disable vectorization, then we also disable alignment
+#if defined(EIGEN_DONT_VECTORIZE)
+  #define EIGEN_IDEAL_MAX_ALIGN_BYTES 0
+#elif defined(EIGEN_VECTORIZE_AVX512)
+  // 64 bytes static alignmeent is preferred only if really required
+  #define EIGEN_IDEAL_MAX_ALIGN_BYTES 64
+#elif defined(__AVX__)
+  // 32 bytes static alignmeent is preferred only if really required
+  #define EIGEN_IDEAL_MAX_ALIGN_BYTES 32
+#else
+  #define EIGEN_IDEAL_MAX_ALIGN_BYTES 16
+#endif
+
+
+// EIGEN_MIN_ALIGN_BYTES defines the minimal value for which the notion of explicit alignment makes sense
+#define EIGEN_MIN_ALIGN_BYTES 16
+
+// Defined the boundary (in bytes) on which the data needs to be aligned. Note
+// that unless EIGEN_ALIGN is defined and not equal to 0, the data may not be
+// aligned at all regardless of the value of this #define.
+
+#if (defined(EIGEN_DONT_ALIGN_STATICALLY) || defined(EIGEN_DONT_ALIGN))  && defined(EIGEN_MAX_STATIC_ALIGN_BYTES) && EIGEN_MAX_STATIC_ALIGN_BYTES>0
+#error EIGEN_MAX_STATIC_ALIGN_BYTES and EIGEN_DONT_ALIGN[_STATICALLY] are both defined with EIGEN_MAX_STATIC_ALIGN_BYTES!=0. Use EIGEN_MAX_STATIC_ALIGN_BYTES=0 as a synonym of EIGEN_DONT_ALIGN_STATICALLY.
+#endif
+
+// EIGEN_DONT_ALIGN_STATICALLY and EIGEN_DONT_ALIGN are deprectated
+// They imply EIGEN_MAX_STATIC_ALIGN_BYTES=0
+#if defined(EIGEN_DONT_ALIGN_STATICALLY) || defined(EIGEN_DONT_ALIGN)
+  #ifdef EIGEN_MAX_STATIC_ALIGN_BYTES
+    #undef EIGEN_MAX_STATIC_ALIGN_BYTES
+  #endif
+  #define EIGEN_MAX_STATIC_ALIGN_BYTES 0
+#endif
+
+#ifndef EIGEN_MAX_STATIC_ALIGN_BYTES
+
+  // Try to automatically guess what is the best default value for EIGEN_MAX_STATIC_ALIGN_BYTES
+
+  // 16 byte alignment is only useful for vectorization. Since it affects the ABI, we need to enable
+  // 16 byte alignment on all platforms where vectorization might be enabled. In theory we could always
+  // enable alignment, but it can be a cause of problems on some platforms, so we just disable it in
+  // certain common platform (compiler+architecture combinations) to avoid these problems.
+  // Only static alignment is really problematic (relies on nonstandard compiler extensions),
+  // try to keep heap alignment even when we have to disable static alignment.
+  #if EIGEN_COMP_GNUC && !(EIGEN_ARCH_i386_OR_x86_64 || EIGEN_ARCH_ARM_OR_ARM64 || EIGEN_ARCH_PPC || EIGEN_ARCH_IA64)
+  #define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1
+  #elif EIGEN_ARCH_ARM_OR_ARM64 && EIGEN_COMP_GNUC_STRICT && EIGEN_GNUC_AT_MOST(4, 6)
+  // Old versions of GCC on ARM, at least 4.4, were once seen to have buggy static alignment support.
+  // Not sure which version fixed it, hopefully it doesn't affect 4.7, which is still somewhat in use.
+  // 4.8 and newer seem definitely unaffected.
+  #define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1
+  #else
+  #define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 0
+  #endif
+
+  // static alignment is completely disabled with GCC 3, Sun Studio, and QCC/QNX
+  #if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT \
+  && !EIGEN_GCC3_OR_OLDER \
+  && !EIGEN_COMP_SUNCC \
+  && !EIGEN_OS_QNX
+    #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 1
+  #else
+    #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 0
+  #endif
+
+  #if EIGEN_ARCH_WANTS_STACK_ALIGNMENT
+    #define EIGEN_MAX_STATIC_ALIGN_BYTES EIGEN_IDEAL_MAX_ALIGN_BYTES
+  #else
+    #define EIGEN_MAX_STATIC_ALIGN_BYTES 0
+  #endif
+
+#endif
+
+// If EIGEN_MAX_ALIGN_BYTES is defined, then it is considered as an upper bound for EIGEN_MAX_ALIGN_BYTES
+#if defined(EIGEN_MAX_ALIGN_BYTES) && EIGEN_MAX_ALIGN_BYTES<EIGEN_MAX_STATIC_ALIGN_BYTES
+#undef EIGEN_MAX_STATIC_ALIGN_BYTES
+#define EIGEN_MAX_STATIC_ALIGN_BYTES EIGEN_MAX_ALIGN_BYTES
+#endif
+
+#if EIGEN_MAX_STATIC_ALIGN_BYTES==0 && !defined(EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT)
+  #define EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT
+#endif
+
+// At this stage, EIGEN_MAX_STATIC_ALIGN_BYTES>0 is the true test whether we want to align arrays on the stack or not.
+// It takes into account both the user choice to explicitly enable/disable alignment (by settting EIGEN_MAX_STATIC_ALIGN_BYTES)
+// and the architecture config (EIGEN_ARCH_WANTS_STACK_ALIGNMENT).
+// Henceforth, only EIGEN_MAX_STATIC_ALIGN_BYTES should be used.
+
+
+// Shortcuts to EIGEN_ALIGN_TO_BOUNDARY
 #define EIGEN_ALIGN8  EIGEN_ALIGN_TO_BOUNDARY(8)
 #define EIGEN_ALIGN16 EIGEN_ALIGN_TO_BOUNDARY(16)
+#define EIGEN_ALIGN32 EIGEN_ALIGN_TO_BOUNDARY(32)
+#define EIGEN_ALIGN64 EIGEN_ALIGN_TO_BOUNDARY(64)
+#if EIGEN_MAX_STATIC_ALIGN_BYTES>0
+#define EIGEN_ALIGN_MAX EIGEN_ALIGN_TO_BOUNDARY(EIGEN_MAX_STATIC_ALIGN_BYTES)
+#else
+#define EIGEN_ALIGN_MAX
+#endif
+
 
-#if EIGEN_ALIGN_STATICALLY
-#define EIGEN_USER_ALIGN_TO_BOUNDARY(n) EIGEN_ALIGN_TO_BOUNDARY(n)
-#define EIGEN_USER_ALIGN16 EIGEN_ALIGN16
+// Dynamic alignment control
+
+#if defined(EIGEN_DONT_ALIGN) && defined(EIGEN_MAX_ALIGN_BYTES) && EIGEN_MAX_ALIGN_BYTES>0
+#error EIGEN_MAX_ALIGN_BYTES and EIGEN_DONT_ALIGN are both defined with EIGEN_MAX_ALIGN_BYTES!=0. Use EIGEN_MAX_ALIGN_BYTES=0 as a synonym of EIGEN_DONT_ALIGN.
+#endif
+
+#ifdef EIGEN_DONT_ALIGN
+  #ifdef EIGEN_MAX_ALIGN_BYTES
+    #undef EIGEN_MAX_ALIGN_BYTES
+  #endif
+  #define EIGEN_MAX_ALIGN_BYTES 0
+#elif !defined(EIGEN_MAX_ALIGN_BYTES)
+  #define EIGEN_MAX_ALIGN_BYTES EIGEN_IDEAL_MAX_ALIGN_BYTES
+#endif
+
+#if EIGEN_IDEAL_MAX_ALIGN_BYTES > EIGEN_MAX_ALIGN_BYTES
+#define EIGEN_DEFAULT_ALIGN_BYTES EIGEN_IDEAL_MAX_ALIGN_BYTES
 #else
-#define EIGEN_USER_ALIGN_TO_BOUNDARY(n)
-#define EIGEN_USER_ALIGN16
+#define EIGEN_DEFAULT_ALIGN_BYTES EIGEN_MAX_ALIGN_BYTES
+#endif
+
+
+#ifndef EIGEN_UNALIGNED_VECTORIZE
+#define EIGEN_UNALIGNED_VECTORIZE 1
 #endif
 
+//----------------------------------------------------------------------
+
+
 #ifdef EIGEN_DONT_USE_RESTRICT_KEYWORD
   #define EIGEN_RESTRICT
 #endif
@@ -328,25 +814,27 @@ namespace Eigen {
 // just an empty macro !
 #define EIGEN_EMPTY
 
-#if defined(_MSC_VER) && (_MSC_VER < 1900) && (!defined(__INTEL_COMPILER))
-#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
-  using Base::operator =;
-#elif defined(__clang__) // workaround clang bug (see http://forum.kde.org/viewtopic.php?f=74&t=102653)
-#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
-  using Base::operator =; \
-  EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) { Base::operator=(other); return *this; } \
-  template <typename OtherDerived> \
-  EIGEN_STRONG_INLINE Derived& operator=(const DenseBase<OtherDerived>& other) { Base::operator=(other.derived()); return *this; }
-#else
-#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
-  using Base::operator =; \
-  EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) \
-  { \
-    Base::operator=(other); \
-    return *this; \
-  }
+#if EIGEN_COMP_MSVC_STRICT && (EIGEN_COMP_MSVC < 1900 || EIGEN_CUDACC_VER>0)
+  // for older MSVC versions, as well as 1900 && CUDA 8, using the base operator is sufficient (cf Bugs 1000, 1324)
+  #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
+    using Base::operator =;
+#elif EIGEN_COMP_CLANG // workaround clang bug (see http://forum.kde.org/viewtopic.php?f=74&t=102653)
+  #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
+    using Base::operator =; \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) { Base::operator=(other); return *this; } \
+    template <typename OtherDerived> \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const DenseBase<OtherDerived>& other) { Base::operator=(other.derived()); return *this; }
+#else
+  #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
+    using Base::operator =; \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) \
+    { \
+      Base::operator=(other); \
+      return *this; \
+    }
 #endif
 
+
 /** \internal
  * \brief Macro to manually inherit assignment operators.
  * This is necessary, because the implicitly defined assignment operator gets deleted when a custom operator= is defined.
@@ -365,32 +853,12 @@ namespace Eigen {
   typedef typename Eigen::internal::traits<Derived>::Scalar Scalar; /*!< \brief Numeric type, e.g. float, double, int or std::complex<float>. */ \
   typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; /*!< \brief The underlying numeric type for composed scalar types. \details In cases where Scalar is e.g. std::complex<T>, T were corresponding to RealScalar. */ \
   typedef typename Base::CoeffReturnType CoeffReturnType; /*!< \brief The return type for coefficient access. \details Depending on whether the object allows direct coefficient access (e.g. for a MatrixXd), this type is either 'const Scalar&' or simply 'Scalar' for objects that do not allow direct coefficient access. */ \
-  typedef typename Eigen::internal::nested<Derived>::type Nested; \
-  typedef typename Eigen::internal::traits<Derived>::StorageKind StorageKind; \
-  typedef typename Eigen::internal::traits<Derived>::Index Index; \
-  enum { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \
-        ColsAtCompileTime = Eigen::internal::traits<Derived>::ColsAtCompileTime, \
-        Flags = Eigen::internal::traits<Derived>::Flags, \
-        CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \
-        SizeAtCompileTime = Base::SizeAtCompileTime, \
-        MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \
-        IsVectorAtCompileTime = Base::IsVectorAtCompileTime };
-
-
-#define EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \
-  typedef typename Eigen::internal::traits<Derived>::Scalar Scalar; /*!< \brief Numeric type, e.g. float, double, int or std::complex<float>. */ \
-  typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; /*!< \brief The underlying numeric type for composed scalar types. \details In cases where Scalar is e.g. std::complex<T>, T were corresponding to RealScalar. */ \
-  typedef typename Base::PacketScalar PacketScalar; \
-  typedef typename Base::CoeffReturnType CoeffReturnType; /*!< \brief The return type for coefficient access. \details Depending on whether the object allows direct coefficient access (e.g. for a MatrixXd), this type is either 'const Scalar&' or simply 'Scalar' for objects that do not allow direct coefficient access. */ \
-  typedef typename Eigen::internal::nested<Derived>::type Nested; \
+  typedef typename Eigen::internal::ref_selector<Derived>::type Nested; \
   typedef typename Eigen::internal::traits<Derived>::StorageKind StorageKind; \
-  typedef typename Eigen::internal::traits<Derived>::Index Index; \
+  typedef typename Eigen::internal::traits<Derived>::StorageIndex StorageIndex; \
   enum { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \
         ColsAtCompileTime = Eigen::internal::traits<Derived>::ColsAtCompileTime, \
-        MaxRowsAtCompileTime = Eigen::internal::traits<Derived>::MaxRowsAtCompileTime, \
-        MaxColsAtCompileTime = Eigen::internal::traits<Derived>::MaxColsAtCompileTime, \
         Flags = Eigen::internal::traits<Derived>::Flags, \
-        CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \
         SizeAtCompileTime = Base::SizeAtCompileTime, \
         MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \
         IsVectorAtCompileTime = Base::IsVectorAtCompileTime }; \
@@ -398,6 +866,12 @@ namespace Eigen {
   using Base::const_cast_derived;
 
 
+// FIXME Maybe the EIGEN_DENSE_PUBLIC_INTERFACE could be removed as importing PacketScalar is rarely needed
+#define EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \
+  EIGEN_GENERIC_PUBLIC_INTERFACE(Derived) \
+  typedef typename Base::PacketScalar PacketScalar;
+
+
 #define EIGEN_PLAIN_ENUM_MIN(a,b) (((int)a <= (int)b) ? (int)a : (int)b)
 #define EIGEN_PLAIN_ENUM_MAX(a,b) (((int)a >= (int)b) ? (int)a : (int)b)
 
@@ -423,24 +897,14 @@ namespace Eigen {
 #define EIGEN_SIZE_MAX(a,b) (((int)a == Dynamic || (int)b == Dynamic) ? Dynamic \
                            : ((int)a >= (int)b) ? (int)a : (int)b)
 
-#define EIGEN_ADD_COST(a,b) int(a)==Dynamic || int(b)==Dynamic ? Dynamic : int(a)+int(b)
-
 #define EIGEN_LOGICAL_XOR(a,b) (((a) || (b)) && !((a) && (b)))
 
 #define EIGEN_IMPLIES(a,b) (!(a) || (b))
 
-#define EIGEN_MAKE_CWISE_BINARY_OP(METHOD,FUNCTOR) \
-  template<typename OtherDerived> \
-  EIGEN_STRONG_INLINE const CwiseBinaryOp<FUNCTOR<Scalar>, const Derived, const OtherDerived> \
-  (METHOD)(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \
-  { \
-    return CwiseBinaryOp<FUNCTOR<Scalar>, const Derived, const OtherDerived>(derived(), other.derived()); \
-  }
-
-// the expression type of a cwise product
-#define EIGEN_CWISE_PRODUCT_RETURN_TYPE(LHS,RHS) \
+// the expression type of a standard coefficient wise binary operation
+#define EIGEN_CWISE_BINARY_RETURN_TYPE(LHS,RHS,OPNAME) \
     CwiseBinaryOp< \
-      internal::scalar_product_op< \
+      EIGEN_CAT(EIGEN_CAT(internal::scalar_,OPNAME),_op)< \
           typename internal::traits<LHS>::Scalar, \
           typename internal::traits<RHS>::Scalar \
       >, \
@@ -448,4 +912,90 @@ namespace Eigen {
       const RHS \
     >
 
+#define EIGEN_MAKE_CWISE_BINARY_OP(METHOD,OPNAME) \
+  template<typename OtherDerived> \
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,OPNAME) \
+  (METHOD)(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \
+  { \
+    return EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,OPNAME)(derived(), other.derived()); \
+  }
+
+#define EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,TYPEA,TYPEB) \
+  (Eigen::internal::has_ReturnType<Eigen::ScalarBinaryOpTraits<TYPEA,TYPEB,EIGEN_CAT(EIGEN_CAT(Eigen::internal::scalar_,OPNAME),_op)<TYPEA,TYPEB>  > >::value)
+
+#define EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(EXPR,SCALAR,OPNAME) \
+  CwiseBinaryOp<EIGEN_CAT(EIGEN_CAT(internal::scalar_,OPNAME),_op)<typename internal::traits<EXPR>::Scalar,SCALAR>, const EXPR, \
+                const typename internal::plain_constant_type<EXPR,SCALAR>::type>
+
+#define EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(SCALAR,EXPR,OPNAME) \
+  CwiseBinaryOp<EIGEN_CAT(EIGEN_CAT(internal::scalar_,OPNAME),_op)<SCALAR,typename internal::traits<EXPR>::Scalar>, \
+                const typename internal::plain_constant_type<EXPR,SCALAR>::type, const EXPR>
+
+// Workaround for MSVC 2010 (see ML thread "patch with compile for for MSVC 2010")
+#if EIGEN_COMP_MSVC_STRICT<=1600
+#define EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(X) typename internal::enable_if<true,X>::type
+#else
+#define EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(X) X
+#endif
+
+#define EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(METHOD,OPNAME) \
+  template <typename T> EIGEN_DEVICE_FUNC inline \
+  EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename internal::promote_scalar_arg<Scalar EIGEN_COMMA T EIGEN_COMMA EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,Scalar,T)>::type,OPNAME))\
+  (METHOD)(const T& scalar) const { \
+    typedef typename internal::promote_scalar_arg<Scalar,T,EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,Scalar,T)>::type PromotedT; \
+    return EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,PromotedT,OPNAME)(derived(), \
+           typename internal::plain_constant_type<Derived,PromotedT>::type(derived().rows(), derived().cols(), internal::scalar_constant_op<PromotedT>(scalar))); \
+  }
+
+#define EIGEN_MAKE_SCALAR_BINARY_OP_ONTHELEFT(METHOD,OPNAME) \
+  template <typename T> EIGEN_DEVICE_FUNC inline friend \
+  EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename internal::promote_scalar_arg<Scalar EIGEN_COMMA T EIGEN_COMMA EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,T,Scalar)>::type,Derived,OPNAME)) \
+  (METHOD)(const T& scalar, const StorageBaseType& matrix) { \
+    typedef typename internal::promote_scalar_arg<Scalar,T,EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,T,Scalar)>::type PromotedT; \
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(PromotedT,Derived,OPNAME)( \
+           typename internal::plain_constant_type<Derived,PromotedT>::type(matrix.derived().rows(), matrix.derived().cols(), internal::scalar_constant_op<PromotedT>(scalar)), matrix.derived()); \
+  }
+
+#define EIGEN_MAKE_SCALAR_BINARY_OP(METHOD,OPNAME) \
+  EIGEN_MAKE_SCALAR_BINARY_OP_ONTHELEFT(METHOD,OPNAME) \
+  EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(METHOD,OPNAME)
+
+
+#ifdef EIGEN_EXCEPTIONS
+#  define EIGEN_THROW_X(X) throw X
+#  define EIGEN_THROW throw
+#  define EIGEN_TRY try
+#  define EIGEN_CATCH(X) catch (X)
+#else
+#  ifdef __CUDA_ARCH__
+#    define EIGEN_THROW_X(X) asm("trap;")
+#    define EIGEN_THROW asm("trap;")
+#  else
+#    define EIGEN_THROW_X(X) std::abort()
+#    define EIGEN_THROW std::abort()
+#  endif
+#  define EIGEN_TRY if (true)
+#  define EIGEN_CATCH(X) else
+#endif
+
+
+#if EIGEN_HAS_CXX11_NOEXCEPT
+#   define EIGEN_INCLUDE_TYPE_TRAITS
+#   define EIGEN_NOEXCEPT noexcept
+#   define EIGEN_NOEXCEPT_IF(x) noexcept(x)
+#   define EIGEN_NO_THROW noexcept(true)
+#   define EIGEN_EXCEPTION_SPEC(X) noexcept(false)
+#else
+#   define EIGEN_NOEXCEPT
+#   define EIGEN_NOEXCEPT_IF(x)
+#   define EIGEN_NO_THROW throw()
+#   if EIGEN_COMP_MSVC
+      // MSVC does not support exception specifications (warning C4290),
+      // and they are deprecated in c++11 anyway.
+#     define EIGEN_EXCEPTION_SPEC(X) throw()
+#   else
+#     define EIGEN_EXCEPTION_SPEC(X) throw(X)
+#   endif
+#endif
+
 #endif // EIGEN_MACROS_H
diff --git a/extern/Eigen3/Eigen/src/Core/util/Memory.h b/extern/Eigen3/Eigen/src/Core/util/Memory.h
index b9af5cf8c7b..291383c581a 100644
--- a/extern/Eigen3/Eigen/src/Core/util/Memory.h
+++ b/extern/Eigen3/Eigen/src/Core/util/Memory.h
@@ -1,11 +1,12 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2008-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
 // Copyright (C) 2009 Kenneth Riddile <kfriddile@yahoo.com>
 // Copyright (C) 2010 Hauke Heibel <hauke.heibel@gmail.com>
 // Copyright (C) 2010 Thomas Capricelli <orzel@freehackers.org>
+// Copyright (C) 2013 Pavel Holoborodko <pavel@holoborodko.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -31,7 +32,7 @@
 // page 114, "[The] LP64 model [...] is used by all 64-bit UNIX ports" so it's indeed
 // quite safe, at least within the context of glibc, to equate 64-bit with LP64.
 #if defined(__GLIBC__) && ((__GLIBC__>=2 && __GLIBC_MINOR__ >= 8) || __GLIBC__>2) \
- && defined(__LP64__) && ! defined( __SANITIZE_ADDRESS__ )
+ && defined(__LP64__) && ! defined( __SANITIZE_ADDRESS__ ) && (EIGEN_DEFAULT_ALIGN_BYTES == 16)
   #define EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED 1
 #else
   #define EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED 0
@@ -41,15 +42,15 @@
 //   See http://svn.freebsd.org/viewvc/base/stable/6/lib/libc/stdlib/malloc.c?view=markup
 // FreeBSD 7 seems to have 16-byte aligned malloc except on ARM and MIPS architectures
 //   See http://svn.freebsd.org/viewvc/base/stable/7/lib/libc/stdlib/malloc.c?view=markup
-#if defined(__FreeBSD__) && !defined(__arm__) && !defined(__mips__)
+#if defined(__FreeBSD__) && !(EIGEN_ARCH_ARM || EIGEN_ARCH_MIPS) && (EIGEN_DEFAULT_ALIGN_BYTES == 16)
   #define EIGEN_FREEBSD_MALLOC_ALREADY_ALIGNED 1
 #else
   #define EIGEN_FREEBSD_MALLOC_ALREADY_ALIGNED 0
 #endif
 
-#if defined(__APPLE__) \
- || defined(_WIN64) \
- || EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED \
+#if (EIGEN_OS_MAC && (EIGEN_DEFAULT_ALIGN_BYTES == 16))     \
+ || (EIGEN_OS_WIN64 && (EIGEN_DEFAULT_ALIGN_BYTES == 16))   \
+ || EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED              \
  || EIGEN_FREEBSD_MALLOC_ALREADY_ALIGNED
   #define EIGEN_MALLOC_ALREADY_ALIGNED 1
 #else
@@ -58,37 +59,18 @@
 
 #endif
 
-// See bug 554 (http://eigen.tuxfamily.org/bz/show_bug.cgi?id=554)
-// It seems to be unsafe to check _POSIX_ADVISORY_INFO without including unistd.h first.
-// Currently, let's include it only on unix systems:
-#if defined(__unix__) || defined(__unix)
-  #include <unistd.h>
-  #if ((defined __QNXNTO__) || (defined _GNU_SOURCE) || (defined __PGI) || ((defined _XOPEN_SOURCE) && (_XOPEN_SOURCE >= 600))) && (defined _POSIX_ADVISORY_INFO) && (_POSIX_ADVISORY_INFO > 0)
-    #define EIGEN_HAS_POSIX_MEMALIGN 1
-  #endif
-#endif
-
-#ifndef EIGEN_HAS_POSIX_MEMALIGN
-  #define EIGEN_HAS_POSIX_MEMALIGN 0
-#endif
-
-#ifdef EIGEN_VECTORIZE_SSE
-  #define EIGEN_HAS_MM_MALLOC 1
-#else
-  #define EIGEN_HAS_MM_MALLOC 0
-#endif
-
 namespace Eigen {
 
 namespace internal {
 
+EIGEN_DEVICE_FUNC 
 inline void throw_std_bad_alloc()
 {
   #ifdef EIGEN_EXCEPTIONS
     throw std::bad_alloc();
   #else
-    std::size_t huge = -1;
-    new int[huge];
+    std::size_t huge = static_cast<std::size_t>(-1);
+    ::operator new(huge);
   #endif
 }
 
@@ -103,9 +85,9 @@ inline void throw_std_bad_alloc()
   */
 inline void* handmade_aligned_malloc(std::size_t size)
 {
-  void *original = std::malloc(size+16);
+  void *original = std::malloc(size+EIGEN_DEFAULT_ALIGN_BYTES);
   if (original == 0) return 0;
-  void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(15))) + 16);
+  void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1))) + EIGEN_DEFAULT_ALIGN_BYTES);
   *(reinterpret_cast<void**>(aligned) - 1) = original;
   return aligned;
 }
@@ -118,7 +100,7 @@ inline void handmade_aligned_free(void *ptr)
 
 /** \internal
   * \brief Reallocates aligned memory.
-  * Since we know that our handmade version is based on std::realloc
+  * Since we know that our handmade version is based on std::malloc
   * we can use std::realloc to implement efficient reallocation.
   */
 inline void* handmade_aligned_realloc(void* ptr, std::size_t size, std::size_t = 0)
@@ -126,9 +108,9 @@ inline void* handmade_aligned_realloc(void* ptr, std::size_t size, std::size_t =
   if (ptr == 0) return handmade_aligned_malloc(size);
   void *original = *(reinterpret_cast<void**>(ptr) - 1);
   std::ptrdiff_t previous_offset = static_cast<char *>(ptr)-static_cast<char *>(original);
-  original = std::realloc(original,size+16);
+  original = std::realloc(original,size+EIGEN_DEFAULT_ALIGN_BYTES);
   if (original == 0) return 0;
-  void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(15))) + 16);
+  void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1))) + EIGEN_DEFAULT_ALIGN_BYTES);
   void *previous_aligned = static_cast<char *>(original)+previous_offset;
   if(aligned!=previous_aligned)
     std::memmove(aligned, previous_aligned, size);
@@ -138,92 +120,46 @@ inline void* handmade_aligned_realloc(void* ptr, std::size_t size, std::size_t =
 }
 
 /*****************************************************************************
-*** Implementation of generic aligned realloc (when no realloc can be used)***
-*****************************************************************************/
-
-void* aligned_malloc(std::size_t size);
-void  aligned_free(void *ptr);
-
-/** \internal
-  * \brief Reallocates aligned memory.
-  * Allows reallocation with aligned ptr types. This implementation will
-  * always create a new memory chunk and copy the old data.
-  */
-inline void* generic_aligned_realloc(void* ptr, size_t size, size_t old_size)
-{
-  if (ptr==0)
-    return aligned_malloc(size);
-
-  if (size==0)
-  {
-    aligned_free(ptr);
-    return 0;
-  }
-
-  void* newptr = aligned_malloc(size);
-  if (newptr == 0)
-  {
-    #ifdef EIGEN_HAS_ERRNO
-    errno = ENOMEM; // according to the standard
-    #endif
-    return 0;
-  }
-
-  if (ptr != 0)
-  {
-    std::memcpy(newptr, ptr, (std::min)(size,old_size));
-    aligned_free(ptr);
-  }
-
-  return newptr;
-}
-
-/*****************************************************************************
 *** Implementation of portable aligned versions of malloc/free/realloc     ***
 *****************************************************************************/
 
 #ifdef EIGEN_NO_MALLOC
-inline void check_that_malloc_is_allowed()
+EIGEN_DEVICE_FUNC inline void check_that_malloc_is_allowed()
 {
   eigen_assert(false && "heap allocation is forbidden (EIGEN_NO_MALLOC is defined)");
 }
 #elif defined EIGEN_RUNTIME_NO_MALLOC
-inline bool is_malloc_allowed_impl(bool update, bool new_value = false)
+EIGEN_DEVICE_FUNC inline bool is_malloc_allowed_impl(bool update, bool new_value = false)
 {
   static bool value = true;
   if (update == 1)
     value = new_value;
   return value;
 }
-inline bool is_malloc_allowed() { return is_malloc_allowed_impl(false); }
-inline bool set_is_malloc_allowed(bool new_value) { return is_malloc_allowed_impl(true, new_value); }
-inline void check_that_malloc_is_allowed()
+EIGEN_DEVICE_FUNC inline bool is_malloc_allowed() { return is_malloc_allowed_impl(false); }
+EIGEN_DEVICE_FUNC inline bool set_is_malloc_allowed(bool new_value) { return is_malloc_allowed_impl(true, new_value); }
+EIGEN_DEVICE_FUNC inline void check_that_malloc_is_allowed()
 {
   eigen_assert(is_malloc_allowed() && "heap allocation is forbidden (EIGEN_RUNTIME_NO_MALLOC is defined and g_is_malloc_allowed is false)");
 }
 #else 
-inline void check_that_malloc_is_allowed()
+EIGEN_DEVICE_FUNC inline void check_that_malloc_is_allowed()
 {}
 #endif
 
-/** \internal Allocates \a size bytes. The returned pointer is guaranteed to have 16 bytes alignment.
+/** \internal Allocates \a size bytes. The returned pointer is guaranteed to have 16 or 32 bytes alignment depending on the requirements.
   * On allocation error, the returned pointer is null, and std::bad_alloc is thrown.
   */
-inline void* aligned_malloc(size_t size)
+EIGEN_DEVICE_FUNC inline void* aligned_malloc(std::size_t size)
 {
   check_that_malloc_is_allowed();
 
   void *result;
-  #if !EIGEN_ALIGN
-    result = std::malloc(size);
-  #elif EIGEN_MALLOC_ALREADY_ALIGNED
+  #if (EIGEN_DEFAULT_ALIGN_BYTES==0) || EIGEN_MALLOC_ALREADY_ALIGNED
     result = std::malloc(size);
-  #elif EIGEN_HAS_POSIX_MEMALIGN
-    if(posix_memalign(&result, 16, size)) result = 0;
-  #elif EIGEN_HAS_MM_MALLOC
-    result = _mm_malloc(size, 16);
-  #elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
-    result = _aligned_malloc(size, 16);
+    #if EIGEN_DEFAULT_ALIGN_BYTES==16
+    eigen_assert((size<16 || (std::size_t(result)%16)==0) && "System's malloc returned an unaligned pointer. Compile with EIGEN_MALLOC_ALREADY_ALIGNED=0 to fallback to handmade alignd memory allocator.");
+    #endif
   #else
     result = handmade_aligned_malloc(size);
   #endif
@@ -235,50 +171,27 @@ inline void* aligned_malloc(size_t size)
 }
 
 /** \internal Frees memory allocated with aligned_malloc. */
-inline void aligned_free(void *ptr)
+EIGEN_DEVICE_FUNC inline void aligned_free(void *ptr)
 {
-  #if !EIGEN_ALIGN
-    std::free(ptr);
-  #elif EIGEN_MALLOC_ALREADY_ALIGNED
-    std::free(ptr);
-  #elif EIGEN_HAS_POSIX_MEMALIGN
+  #if (EIGEN_DEFAULT_ALIGN_BYTES==0) || EIGEN_MALLOC_ALREADY_ALIGNED
     std::free(ptr);
-  #elif EIGEN_HAS_MM_MALLOC
-    _mm_free(ptr);
-  #elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
-    _aligned_free(ptr);
   #else
     handmade_aligned_free(ptr);
   #endif
 }
 
 /**
-* \internal
-* \brief Reallocates an aligned block of memory.
-* \throws std::bad_alloc on allocation failure
-**/
-inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
+  * \internal
+  * \brief Reallocates an aligned block of memory.
+  * \throws std::bad_alloc on allocation failure
+  */
+inline void* aligned_realloc(void *ptr, std::size_t new_size, std::size_t old_size)
 {
   EIGEN_UNUSED_VARIABLE(old_size);
 
   void *result;
-#if !EIGEN_ALIGN
-  result = std::realloc(ptr,new_size);
-#elif EIGEN_MALLOC_ALREADY_ALIGNED
+#if (EIGEN_DEFAULT_ALIGN_BYTES==0) || EIGEN_MALLOC_ALREADY_ALIGNED
   result = std::realloc(ptr,new_size);
-#elif EIGEN_HAS_POSIX_MEMALIGN
-  result = generic_aligned_realloc(ptr,new_size,old_size);
-#elif EIGEN_HAS_MM_MALLOC
-  // The defined(_mm_free) is just here to verify that this MSVC version
-  // implements _mm_malloc/_mm_free based on the corresponding _aligned_
-  // functions. This may not always be the case and we just try to be safe.
-  #if defined(_MSC_VER) && (!defined(_WIN32_WCE)) && defined(_mm_free)
-    result = _aligned_realloc(ptr,new_size,16);
-  #else
-    result = generic_aligned_realloc(ptr,new_size,old_size);
-  #endif
-#elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
-  result = _aligned_realloc(ptr,new_size,16);
 #else
   result = handmade_aligned_realloc(ptr,new_size,old_size);
 #endif
@@ -296,12 +209,12 @@ inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
 /** \internal Allocates \a size bytes. If Align is true, then the returned ptr is 16-byte-aligned.
   * On allocation error, the returned pointer is null, and a std::bad_alloc is thrown.
   */
-template<bool Align> inline void* conditional_aligned_malloc(size_t size)
+template<bool Align> EIGEN_DEVICE_FUNC inline void* conditional_aligned_malloc(std::size_t size)
 {
   return aligned_malloc(size);
 }
 
-template<> inline void* conditional_aligned_malloc<false>(size_t size)
+template<> EIGEN_DEVICE_FUNC inline void* conditional_aligned_malloc<false>(std::size_t size)
 {
   check_that_malloc_is_allowed();
 
@@ -312,22 +225,22 @@ template<> inline void* conditional_aligned_malloc<false>(size_t size)
 }
 
 /** \internal Frees memory allocated with conditional_aligned_malloc */
-template<bool Align> inline void conditional_aligned_free(void *ptr)
+template<bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_free(void *ptr)
 {
   aligned_free(ptr);
 }
 
-template<> inline void conditional_aligned_free<false>(void *ptr)
+template<> EIGEN_DEVICE_FUNC inline void conditional_aligned_free<false>(void *ptr)
 {
   std::free(ptr);
 }
 
-template<bool Align> inline void* conditional_aligned_realloc(void* ptr, size_t new_size, size_t old_size)
+template<bool Align> inline void* conditional_aligned_realloc(void* ptr, std::size_t new_size, std::size_t old_size)
 {
   return aligned_realloc(ptr, new_size, old_size);
 }
 
-template<> inline void* conditional_aligned_realloc<false>(void* ptr, size_t new_size, size_t)
+template<> inline void* conditional_aligned_realloc<false>(void* ptr, std::size_t new_size, std::size_t)
 {
   return std::realloc(ptr, new_size);
 }
@@ -336,33 +249,43 @@ template<> inline void* conditional_aligned_realloc<false>(void* ptr, size_t new
 *** Construction/destruction of array elements                             ***
 *****************************************************************************/
 
-/** \internal Constructs the elements of an array.
-  * The \a size parameter tells on how many objects to call the constructor of T.
-  */
-template<typename T> inline T* construct_elements_of_array(T *ptr, size_t size)
-{
-  for (size_t i=0; i < size; ++i) ::new (ptr + i) T;
-  return ptr;
-}
-
 /** \internal Destructs the elements of an array.
   * The \a size parameters tells on how many objects to call the destructor of T.
   */
-template<typename T> inline void destruct_elements_of_array(T *ptr, size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline void destruct_elements_of_array(T *ptr, std::size_t size)
 {
   // always destruct an array starting from the end.
   if(ptr)
     while(size) ptr[--size].~T();
 }
 
+/** \internal Constructs the elements of an array.
+  * The \a size parameter tells on how many objects to call the constructor of T.
+  */
+template<typename T> EIGEN_DEVICE_FUNC inline T* construct_elements_of_array(T *ptr, std::size_t size)
+{
+  std::size_t i;
+  EIGEN_TRY
+  {
+      for (i = 0; i < size; ++i) ::new (ptr + i) T;
+      return ptr;
+  }
+  EIGEN_CATCH(...)
+  {
+    destruct_elements_of_array(ptr, i);
+    EIGEN_THROW;
+  }
+  return NULL;
+}
+
 /*****************************************************************************
 *** Implementation of aligned new/delete-like functions                    ***
 *****************************************************************************/
 
 template<typename T>
-EIGEN_ALWAYS_INLINE void check_size_for_overflow(size_t size)
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void check_size_for_overflow(std::size_t size)
 {
-  if(size > size_t(-1) / sizeof(T))
+  if(size > std::size_t(-1) / sizeof(T))
     throw_std_bad_alloc();
 }
 
@@ -370,24 +293,42 @@ EIGEN_ALWAYS_INLINE void check_size_for_overflow(size_t size)
   * On allocation error, the returned pointer is undefined, but a std::bad_alloc is thrown.
   * The default constructor of T is called.
   */
-template<typename T> inline T* aligned_new(size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline T* aligned_new(std::size_t size)
 {
   check_size_for_overflow<T>(size);
   T *result = reinterpret_cast<T*>(aligned_malloc(sizeof(T)*size));
-  return construct_elements_of_array(result, size);
+  EIGEN_TRY
+  {
+    return construct_elements_of_array(result, size);
+  }
+  EIGEN_CATCH(...)
+  {
+    aligned_free(result);
+    EIGEN_THROW;
+  }
+  return result;
 }
 
-template<typename T, bool Align> inline T* conditional_aligned_new(size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_new(std::size_t size)
 {
   check_size_for_overflow<T>(size);
   T *result = reinterpret_cast<T*>(conditional_aligned_malloc<Align>(sizeof(T)*size));
-  return construct_elements_of_array(result, size);
+  EIGEN_TRY
+  {
+    return construct_elements_of_array(result, size);
+  }
+  EIGEN_CATCH(...)
+  {
+    conditional_aligned_free<Align>(result);
+    EIGEN_THROW;
+  }
+  return result;
 }
 
 /** \internal Deletes objects constructed with aligned_new
   * The \a size parameters tells on how many objects to call the destructor of T.
   */
-template<typename T> inline void aligned_delete(T *ptr, size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline void aligned_delete(T *ptr, std::size_t size)
 {
   destruct_elements_of_array<T>(ptr, size);
   aligned_free(ptr);
@@ -396,13 +337,13 @@ template<typename T> inline void aligned_delete(T *ptr, size_t size)
 /** \internal Deletes objects constructed with conditional_aligned_new
   * The \a size parameters tells on how many objects to call the destructor of T.
   */
-template<typename T, bool Align> inline void conditional_aligned_delete(T *ptr, size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_delete(T *ptr, std::size_t size)
 {
   destruct_elements_of_array<T>(ptr, size);
   conditional_aligned_free<Align>(ptr);
 }
 
-template<typename T, bool Align> inline T* conditional_aligned_realloc_new(T* pts, size_t new_size, size_t old_size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_realloc_new(T* pts, std::size_t new_size, std::size_t old_size)
 {
   check_size_for_overflow<T>(new_size);
   check_size_for_overflow<T>(old_size);
@@ -410,23 +351,43 @@ template<typename T, bool Align> inline T* conditional_aligned_realloc_new(T* pt
     destruct_elements_of_array(pts+new_size, old_size-new_size);
   T *result = reinterpret_cast<T*>(conditional_aligned_realloc<Align>(reinterpret_cast<void*>(pts), sizeof(T)*new_size, sizeof(T)*old_size));
   if(new_size > old_size)
-    construct_elements_of_array(result+old_size, new_size-old_size);
+  {
+    EIGEN_TRY
+    {
+      construct_elements_of_array(result+old_size, new_size-old_size);
+    }
+    EIGEN_CATCH(...)
+    {
+      conditional_aligned_free<Align>(result);
+      EIGEN_THROW;
+    }
+  }
   return result;
 }
 
 
-template<typename T, bool Align> inline T* conditional_aligned_new_auto(size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_new_auto(std::size_t size)
 {
   if(size==0)
     return 0; // short-cut. Also fixes Bug 884
   check_size_for_overflow<T>(size);
   T *result = reinterpret_cast<T*>(conditional_aligned_malloc<Align>(sizeof(T)*size));
   if(NumTraits<T>::RequireInitialization)
-    construct_elements_of_array(result, size);
+  {
+    EIGEN_TRY
+    {
+      construct_elements_of_array(result, size);
+    }
+    EIGEN_CATCH(...)
+    {
+      conditional_aligned_free<Align>(result);
+      EIGEN_THROW;
+    }
+  }
   return result;
 }
 
-template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(T* pts, size_t new_size, size_t old_size)
+template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(T* pts, std::size_t new_size, std::size_t old_size)
 {
   check_size_for_overflow<T>(new_size);
   check_size_for_overflow<T>(old_size);
@@ -434,11 +395,21 @@ template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(
     destruct_elements_of_array(pts+new_size, old_size-new_size);
   T *result = reinterpret_cast<T*>(conditional_aligned_realloc<Align>(reinterpret_cast<void*>(pts), sizeof(T)*new_size, sizeof(T)*old_size));
   if(NumTraits<T>::RequireInitialization && (new_size > old_size))
-    construct_elements_of_array(result+old_size, new_size-old_size);
+  {
+    EIGEN_TRY
+    {
+      construct_elements_of_array(result+old_size, new_size-old_size);
+    }
+    EIGEN_CATCH(...)
+    {
+      conditional_aligned_free<Align>(result);
+      EIGEN_THROW;
+    }
+  }
   return result;
 }
 
-template<typename T, bool Align> inline void conditional_aligned_delete_auto(T *ptr, size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_delete_auto(T *ptr, std::size_t size)
 {
   if(NumTraits<T>::RequireInitialization)
     destruct_elements_of_array<T>(ptr, size);
@@ -447,51 +418,62 @@ template<typename T, bool Align> inline void conditional_aligned_delete_auto(T *
 
 /****************************************************************************/
 
-/** \internal Returns the index of the first element of the array that is well aligned for vectorization.
+/** \internal Returns the index of the first element of the array that is well aligned with respect to the requested \a Alignment.
   *
+  * \tparam Alignment requested alignment in Bytes.
   * \param array the address of the start of the array
   * \param size the size of the array
   *
-  * \note If no element of the array is well aligned, the size of the array is returned. Typically,
-  * for example with SSE, "well aligned" means 16-byte-aligned. If vectorization is disabled or if the
+  * \note If no element of the array is well aligned or the requested alignment is not a multiple of a scalar,
+  * the size of the array is returned. For example with SSE, the requested alignment is typically 16-bytes. If
   * packet size for the given scalar type is 1, then everything is considered well-aligned.
   *
-  * \note If the scalar type is vectorizable, we rely on the following assumptions: sizeof(Scalar) is a
-  * power of 2, the packet size in bytes is also a power of 2, and is a multiple of sizeof(Scalar). On the
-  * other hand, we do not assume that the array address is a multiple of sizeof(Scalar), as that fails for
+  * \note Otherwise, if the Alignment is larger that the scalar size, we rely on the assumptions that sizeof(Scalar) is a
+  * power of 2. On the other hand, we do not assume that the array address is a multiple of sizeof(Scalar), as that fails for
   * example with Scalar=double on certain 32-bit platforms, see bug #79.
   *
   * There is also the variant first_aligned(const MatrixBase&) defined in DenseCoeffsBase.h.
+  * \sa first_default_aligned()
   */
-template<typename Scalar, typename Index>
-static inline Index first_aligned(const Scalar* array, Index size)
+template<int Alignment, typename Scalar, typename Index>
+EIGEN_DEVICE_FUNC inline Index first_aligned(const Scalar* array, Index size)
 {
-  static const Index PacketSize = packet_traits<Scalar>::size;
-  static const Index PacketAlignedMask = PacketSize-1;
+  const Index ScalarSize = sizeof(Scalar);
+  const Index AlignmentSize = Alignment / ScalarSize;
+  const Index AlignmentMask = AlignmentSize-1;
 
-  if(PacketSize==1)
+  if(AlignmentSize<=1)
   {
-    // Either there is no vectorization, or a packet consists of exactly 1 scalar so that all elements
-    // of the array have the same alignment.
+    // Either the requested alignment if smaller than a scalar, or it exactly match a 1 scalar
+    // so that all elements of the array have the same alignment.
     return 0;
   }
-  else if(size_t(array) & (sizeof(Scalar)-1))
+  else if( (UIntPtr(array) & (sizeof(Scalar)-1)) || (Alignment%ScalarSize)!=0)
   {
-    // There is vectorization for this scalar type, but the array is not aligned to the size of a single scalar.
+    // The array is not aligned to the size of a single scalar, or the requested alignment is not a multiple of the scalar size.
     // Consequently, no element of the array is well aligned.
     return size;
   }
   else
   {
-    return std::min<Index>( (PacketSize - (Index((size_t(array)/sizeof(Scalar))) & PacketAlignedMask))
-                           & PacketAlignedMask, size);
+    Index first = (AlignmentSize - (Index((UIntPtr(array)/sizeof(Scalar))) & AlignmentMask)) & AlignmentMask;
+    return (first < size) ? first : size;
   }
 }
 
+/** \internal Returns the index of the first element of the array that is well aligned with respect the largest packet requirement.
+   * \sa first_aligned(Scalar*,Index) and first_default_aligned(DenseBase<Derived>) */
+template<typename Scalar, typename Index>
+EIGEN_DEVICE_FUNC inline Index first_default_aligned(const Scalar* array, Index size)
+{
+  typedef typename packet_traits<Scalar>::type DefaultPacketType;
+  return first_aligned<unpacket_traits<DefaultPacketType>::alignment>(array, size);
+}
+
 /** \internal Returns the smallest integer multiple of \a base and greater or equal to \a size
   */ 
 template<typename Index> 
-inline static Index first_multiple(Index size, Index base)
+inline Index first_multiple(Index size, Index base)
 {
   return ((size+base-1)/base)*base;
 }
@@ -500,21 +482,59 @@ inline static Index first_multiple(Index size, Index base)
 // use memcpy on trivial types, i.e., on types that does not require an initialization ctor.
 template<typename T, bool UseMemcpy> struct smart_copy_helper;
 
-template<typename T> void smart_copy(const T* start, const T* end, T* target)
+template<typename T> EIGEN_DEVICE_FUNC void smart_copy(const T* start, const T* end, T* target)
 {
   smart_copy_helper<T,!NumTraits<T>::RequireInitialization>::run(start, end, target);
 }
 
 template<typename T> struct smart_copy_helper<T,true> {
-  static inline void run(const T* start, const T* end, T* target)
-  { memcpy(target, start, std::ptrdiff_t(end)-std::ptrdiff_t(start)); }
+  EIGEN_DEVICE_FUNC static inline void run(const T* start, const T* end, T* target)
+  {
+    IntPtr size = IntPtr(end)-IntPtr(start);
+    if(size==0) return;
+    eigen_internal_assert(start!=0 && end!=0 && target!=0);
+    std::memcpy(target, start, size);
+  }
 };
 
 template<typename T> struct smart_copy_helper<T,false> {
-  static inline void run(const T* start, const T* end, T* target)
+  EIGEN_DEVICE_FUNC static inline void run(const T* start, const T* end, T* target)
   { std::copy(start, end, target); }
 };
 
+// intelligent memmove. falls back to std::memmove for POD types, uses std::copy otherwise. 
+template<typename T, bool UseMemmove> struct smart_memmove_helper;
+
+template<typename T> void smart_memmove(const T* start, const T* end, T* target)
+{
+  smart_memmove_helper<T,!NumTraits<T>::RequireInitialization>::run(start, end, target);
+}
+
+template<typename T> struct smart_memmove_helper<T,true> {
+  static inline void run(const T* start, const T* end, T* target)
+  {
+    IntPtr size = IntPtr(end)-IntPtr(start);
+    if(size==0) return;
+    eigen_internal_assert(start!=0 && end!=0 && target!=0);
+    std::memmove(target, start, size);
+  }
+};
+
+template<typename T> struct smart_memmove_helper<T,false> {
+  static inline void run(const T* start, const T* end, T* target)
+  { 
+    if (UIntPtr(target) < UIntPtr(start))
+    {
+      std::copy(start, end, target);
+    }
+    else                                 
+    {
+      std::ptrdiff_t count = (std::ptrdiff_t(end)-std::ptrdiff_t(start)) / sizeof(T);
+      std::copy_backward(start, end, target + count); 
+    }
+  }
+};
+
 
 /*****************************************************************************
 *** Implementation of runtime stack allocation (falling back to malloc)    ***
@@ -523,16 +543,16 @@ template<typename T> struct smart_copy_helper<T,false> {
 // you can overwrite Eigen's default behavior regarding alloca by defining EIGEN_ALLOCA
 // to the appropriate stack allocation function
 #ifndef EIGEN_ALLOCA
-  #if (defined __linux__) || (defined __APPLE__) || (defined alloca)
+  #if EIGEN_OS_LINUX || EIGEN_OS_MAC || (defined alloca)
     #define EIGEN_ALLOCA alloca
-  #elif defined(_MSC_VER)
+  #elif EIGEN_COMP_MSVC
     #define EIGEN_ALLOCA _alloca
   #endif
 #endif
 
 // This helper class construct the allocated memory, and takes care of destructing and freeing the handled data
 // at destruction time. In practice this helper class is mainly useful to avoid memory leak in case of exceptions.
-template<typename T> class aligned_stack_memory_handler
+template<typename T> class aligned_stack_memory_handler : noncopyable
 {
   public:
     /* Creates a stack_memory_handler responsible for the buffer \a ptr of size \a size.
@@ -541,7 +561,7 @@ template<typename T> class aligned_stack_memory_handler
      * In this case, the buffer elements will also be destructed when this handler will be destructed.
      * Finally, if \a dealloc is true, then the pointer \a ptr is freed.
      **/
-    aligned_stack_memory_handler(T* ptr, size_t size, bool dealloc)
+    aligned_stack_memory_handler(T* ptr, std::size_t size, bool dealloc)
       : m_ptr(ptr), m_size(size), m_deallocate(dealloc)
     {
       if(NumTraits<T>::RequireInitialization && m_ptr)
@@ -556,10 +576,34 @@ template<typename T> class aligned_stack_memory_handler
     }
   protected:
     T* m_ptr;
-    size_t m_size;
+    std::size_t m_size;
     bool m_deallocate;
 };
 
+template<typename T> class scoped_array : noncopyable
+{
+  T* m_ptr;
+public:
+  explicit scoped_array(std::ptrdiff_t size)
+  {
+    m_ptr = new T[size];
+  }
+  ~scoped_array()
+  {
+    delete[] m_ptr;
+  }
+  T& operator[](std::ptrdiff_t i) { return m_ptr[i]; }
+  const T& operator[](std::ptrdiff_t i) const { return m_ptr[i]; }
+  T* &ptr() { return m_ptr; }
+  const T* ptr() const { return m_ptr; }
+  operator const T*() const { return m_ptr; }
+};
+
+template<typename T> void swap(scoped_array<T> &a,scoped_array<T> &b)
+{
+  std::swap(a.ptr(),b.ptr());
+}
+    
 } // end namespace internal
 
 /** \internal
@@ -578,11 +622,13 @@ template<typename T> class aligned_stack_memory_handler
   * The underlying stack allocation function can controlled with the EIGEN_ALLOCA preprocessor token.
   */
 #ifdef EIGEN_ALLOCA
-
-  #if defined(__arm__) || defined(_WIN32)
-    #define EIGEN_ALIGNED_ALLOCA(SIZE) reinterpret_cast<void*>((reinterpret_cast<size_t>(EIGEN_ALLOCA(SIZE+16)) & ~(size_t(15))) + 16)
+  
+  #if EIGEN_DEFAULT_ALIGN_BYTES>0
+    // We always manually re-align the result of EIGEN_ALLOCA.
+    // If alloca is already aligned, the compiler should be smart enough to optimize away the re-alignment.
+    #define EIGEN_ALIGNED_ALLOCA(SIZE) reinterpret_cast<void*>((internal::UIntPtr(EIGEN_ALLOCA(SIZE+EIGEN_DEFAULT_ALIGN_BYTES-1)) + EIGEN_DEFAULT_ALIGN_BYTES-1) & ~(std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1)))
   #else
-    #define EIGEN_ALIGNED_ALLOCA EIGEN_ALLOCA
+    #define EIGEN_ALIGNED_ALLOCA(SIZE) EIGEN_ALLOCA(SIZE)
   #endif
 
   #define ei_declare_aligned_stack_constructed_variable(TYPE,NAME,SIZE,BUFFER) \
@@ -607,39 +653,33 @@ template<typename T> class aligned_stack_memory_handler
 *** Implementation of EIGEN_MAKE_ALIGNED_OPERATOR_NEW [_IF]                ***
 *****************************************************************************/
 
-#if EIGEN_ALIGN
-  #ifdef EIGEN_EXCEPTIONS
-    #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
-      void* operator new(size_t size, const std::nothrow_t&) throw() { \
-        try { return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); } \
-        catch (...) { return 0; } \
+#if EIGEN_MAX_ALIGN_BYTES!=0
+  #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
+      void* operator new(std::size_t size, const std::nothrow_t&) EIGEN_NO_THROW { \
+        EIGEN_TRY { return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); } \
+        EIGEN_CATCH (...) { return 0; } \
       }
-  #else
-    #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
-      void* operator new(size_t size, const std::nothrow_t&) throw() { \
-        return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
-      }
-  #endif
-
   #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign) \
-      void *operator new(size_t size) { \
+      void *operator new(std::size_t size) { \
         return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
       } \
-      void *operator new[](size_t size) { \
+      void *operator new[](std::size_t size) { \
         return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
       } \
-      void operator delete(void * ptr) throw() { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
-      void operator delete[](void * ptr) throw() { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
+      void operator delete(void * ptr) EIGEN_NO_THROW { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
+      void operator delete[](void * ptr) EIGEN_NO_THROW { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
+      void operator delete(void * ptr, std::size_t /* sz */) EIGEN_NO_THROW { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
+      void operator delete[](void * ptr, std::size_t /* sz */) EIGEN_NO_THROW { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
       /* in-place new and delete. since (at least afaik) there is no actual   */ \
       /* memory allocated we can safely let the default implementation handle */ \
       /* this particular case. */ \
-      static void *operator new(size_t size, void *ptr) { return ::operator new(size,ptr); } \
-      static void *operator new[](size_t size, void* ptr) { return ::operator new[](size,ptr); } \
-      void operator delete(void * memory, void *ptr) throw() { return ::operator delete(memory,ptr); } \
-      void operator delete[](void * memory, void *ptr) throw() { return ::operator delete[](memory,ptr); } \
+      static void *operator new(std::size_t size, void *ptr) { return ::operator new(size,ptr); } \
+      static void *operator new[](std::size_t size, void* ptr) { return ::operator new[](size,ptr); } \
+      void operator delete(void * memory, void *ptr) EIGEN_NO_THROW { return ::operator delete(memory,ptr); } \
+      void operator delete[](void * memory, void *ptr) EIGEN_NO_THROW { return ::operator delete[](memory,ptr); } \
       /* nothrow-new (returns zero instead of std::bad_alloc) */ \
       EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
-      void operator delete(void *ptr, const std::nothrow_t&) throw() { \
+      void operator delete(void *ptr, const std::nothrow_t&) EIGEN_NO_THROW { \
         Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); \
       } \
       typedef void eigen_aligned_operator_new_marker_type;
@@ -649,14 +689,22 @@ template<typename T> class aligned_stack_memory_handler
 
 #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(true)
 #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(Scalar,Size) \
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(bool(((Size)!=Eigen::Dynamic) && ((sizeof(Scalar)*(Size))%16==0)))
+  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(bool(((Size)!=Eigen::Dynamic) && ((sizeof(Scalar)*(Size))%EIGEN_MAX_ALIGN_BYTES==0)))
 
 /****************************************************************************/
 
 /** \class aligned_allocator
 * \ingroup Core_Module
 *
-* \brief STL compatible allocator to use with with 16 byte aligned types
+* \brief STL compatible allocator to use with types requiring a non standrad alignment.
+*
+* The memory is aligned as for dynamically aligned matrix/array types such as MatrixXd.
+* By default, it will thus provide at least 16 bytes alignment and more in following cases:
+*  - 32 bytes alignment if AVX is enabled.
+*  - 64 bytes alignment if AVX512 is enabled.
+*
+* This can be controled using the \c EIGEN_MAX_ALIGN_BYTES macro as documented
+* \link TopicPreprocessorDirectivesPerformance there \endlink.
 *
 * Example:
 * \code
@@ -667,96 +715,64 @@ template<typename T> class aligned_stack_memory_handler
 * std::map< int, Vector3f > my_map_vec3;
 * \endcode
 *
-* \sa \ref TopicStlContainers.
+* \sa \blank \ref TopicStlContainers.
 */
 template<class T>
-class aligned_allocator
+class aligned_allocator : public std::allocator<T>
 {
 public:
-    typedef size_t    size_type;
-    typedef std::ptrdiff_t difference_type;
-    typedef T*        pointer;
-    typedef const T*  const_pointer;
-    typedef T&        reference;
-    typedef const T&  const_reference;
-    typedef T         value_type;
-
-    template<class U>
-    struct rebind
-    {
-        typedef aligned_allocator<U> other;
-    };
-
-    pointer address( reference value ) const
-    {
-        return &value;
-    }
-
-    const_pointer address( const_reference value ) const
-    {
-        return &value;
-    }
-
-    aligned_allocator()
-    {
-    }
-
-    aligned_allocator( const aligned_allocator& )
-    {
-    }
-
-    template<class U>
-    aligned_allocator( const aligned_allocator<U>& )
-    {
-    }
-
-    ~aligned_allocator()
-    {
-    }
-
-    size_type max_size() const
-    {
-        return (std::numeric_limits<size_type>::max)();
-    }
+  typedef std::size_t     size_type;
+  typedef std::ptrdiff_t  difference_type;
+  typedef T*              pointer;
+  typedef const T*        const_pointer;
+  typedef T&              reference;
+  typedef const T&        const_reference;
+  typedef T               value_type;
+
+  template<class U>
+  struct rebind
+  {
+    typedef aligned_allocator<U> other;
+  };
 
-    pointer allocate( size_type num, const void* hint = 0 )
-    {
-        EIGEN_UNUSED_VARIABLE(hint);
-        internal::check_size_for_overflow<T>(num);
-        return static_cast<pointer>( internal::aligned_malloc( num * sizeof(T) ) );
-    }
+  aligned_allocator() : std::allocator<T>() {}
 
-    void construct( pointer p, const T& value )
-    {
-        ::new( p ) T( value );
-    }
+  aligned_allocator(const aligned_allocator& other) : std::allocator<T>(other) {}
 
-    void destroy( pointer p )
-    {
-        p->~T();
-    }
+  template<class U>
+  aligned_allocator(const aligned_allocator<U>& other) : std::allocator<T>(other) {}
 
-    void deallocate( pointer p, size_type /*num*/ )
-    {
-        internal::aligned_free( p );
-    }
+  ~aligned_allocator() {}
 
-    bool operator!=(const aligned_allocator<T>& ) const
-    { return false; }
+  pointer allocate(size_type num, const void* /*hint*/ = 0)
+  {
+    internal::check_size_for_overflow<T>(num);
+    size_type size = num * sizeof(T);
+#if EIGEN_COMP_GNUC_STRICT && EIGEN_GNUC_AT_LEAST(7,0)
+    // workaround gcc bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87544
+    // It triggered eigen/Eigen/src/Core/util/Memory.h:189:12: warning: argument 1 value '18446744073709551612' exceeds maximum object size 9223372036854775807
+    if(size>=std::size_t((std::numeric_limits<std::ptrdiff_t>::max)()))
+      return 0;
+    else
+#endif
+      return static_cast<pointer>( internal::aligned_malloc(size) );
+  }
 
-    bool operator==(const aligned_allocator<T>& ) const
-    { return true; }
+  void deallocate(pointer p, size_type /*num*/)
+  {
+    internal::aligned_free(p);
+  }
 };
 
 //---------- Cache sizes ----------
 
 #if !defined(EIGEN_NO_CPUID)
-#  if defined(__GNUC__) && ( defined(__i386__) || defined(__x86_64__) )
-#    if defined(__PIC__) && defined(__i386__)
+#  if EIGEN_COMP_GNUC && EIGEN_ARCH_i386_OR_x86_64
+#    if defined(__PIC__) && EIGEN_ARCH_i386
        // Case for x86 with PIC
 #      define EIGEN_CPUID(abcd,func,id) \
          __asm__ __volatile__ ("xchgl %%ebx, %k1;cpuid; xchgl %%ebx,%k1": "=a" (abcd[0]), "=&r" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "a" (func), "c" (id));
-#    elif defined(__PIC__) && defined(__x86_64__)
+#    elif defined(__PIC__) && EIGEN_ARCH_x86_64
        // Case for x64 with PIC. In theory this is only a problem with recent gcc and with medium or large code model, not with the default small code model.
        // However, we cannot detect which code model is used, and the xchg overhead is negligible anyway.
 #      define EIGEN_CPUID(abcd,func,id) \
@@ -766,8 +782,8 @@ public:
 #      define EIGEN_CPUID(abcd,func,id) \
          __asm__ __volatile__ ("cpuid": "=a" (abcd[0]), "=b" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "0" (func), "2" (id) );
 #    endif
-#  elif defined(_MSC_VER)
-#    if (_MSC_VER > 1500) && ( defined(_M_IX86) || defined(_M_X64) )
+#  elif EIGEN_COMP_MSVC
+#    if (EIGEN_COMP_MSVC > 1500) && EIGEN_ARCH_i386_OR_x86_64
 #      define EIGEN_CPUID(abcd,func,id) __cpuidex((int*)abcd,func,id)
 #    endif
 #  endif
diff --git a/extern/Eigen3/Eigen/src/Core/util/Meta.h b/extern/Eigen3/Eigen/src/Core/util/Meta.h
index 71d58710871..d31e9541122 100644
--- a/extern/Eigen3/Eigen/src/Core/util/Meta.h
+++ b/extern/Eigen3/Eigen/src/Core/util/Meta.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -11,8 +11,27 @@
 #ifndef EIGEN_META_H
 #define EIGEN_META_H
 
+#if defined(__CUDA_ARCH__)
+#include <cfloat>
+#include <math_constants.h>
+#endif
+
+#if EIGEN_COMP_ICC>=1600 &&  __cplusplus >= 201103L
+#include <cstdint>
+#endif
+
 namespace Eigen {
 
+typedef EIGEN_DEFAULT_DENSE_INDEX_TYPE DenseIndex;
+
+/**
+ * \brief The Index type as used for the API.
+ * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
+ * \sa \blank \ref TopicPreprocessorDirectives, StorageIndex.
+ */
+
+typedef EIGEN_DEFAULT_DENSE_INDEX_TYPE Index;
+
 namespace internal {
 
 /** \internal
@@ -22,6 +41,16 @@ namespace internal {
   * we however don't want to add a dependency to Boost.
   */
 
+// Only recent versions of ICC complain about using ptrdiff_t to hold pointers,
+// and older versions do not provide *intptr_t types.
+#if EIGEN_COMP_ICC>=1600 &&  __cplusplus >= 201103L
+typedef std::intptr_t  IntPtr;
+typedef std::uintptr_t UIntPtr;
+#else
+typedef std::ptrdiff_t IntPtr;
+typedef std::size_t UIntPtr;
+#endif
+
 struct true_type {  enum { value = 1 }; };
 struct false_type { enum { value = 0 }; };
 
@@ -68,6 +97,40 @@ template<> struct is_arithmetic<unsigned int>  { enum { value = true }; };
 template<> struct is_arithmetic<signed long>   { enum { value = true }; };
 template<> struct is_arithmetic<unsigned long> { enum { value = true }; };
 
+template<typename T> struct is_integral        { enum { value = false }; };
+template<> struct is_integral<bool>            { enum { value = true }; };
+template<> struct is_integral<char>            { enum { value = true }; };
+template<> struct is_integral<signed char>     { enum { value = true }; };
+template<> struct is_integral<unsigned char>   { enum { value = true }; };
+template<> struct is_integral<signed short>    { enum { value = true }; };
+template<> struct is_integral<unsigned short>  { enum { value = true }; };
+template<> struct is_integral<signed int>      { enum { value = true }; };
+template<> struct is_integral<unsigned int>    { enum { value = true }; };
+template<> struct is_integral<signed long>     { enum { value = true }; };
+template<> struct is_integral<unsigned long>   { enum { value = true }; };
+
+#if EIGEN_HAS_CXX11
+using std::make_unsigned;
+#else
+// TODO: Possibly improve this implementation of make_unsigned.
+// It is currently used only by
+// template<typename Scalar> struct random_default_impl<Scalar, false, true>.
+template<typename> struct make_unsigned;
+template<> struct make_unsigned<char>             { typedef unsigned char type; };
+template<> struct make_unsigned<signed char>      { typedef unsigned char type; };
+template<> struct make_unsigned<unsigned char>    { typedef unsigned char type; };
+template<> struct make_unsigned<signed short>     { typedef unsigned short type; };
+template<> struct make_unsigned<unsigned short>   { typedef unsigned short type; };
+template<> struct make_unsigned<signed int>       { typedef unsigned int type; };
+template<> struct make_unsigned<unsigned int>     { typedef unsigned int type; };
+template<> struct make_unsigned<signed long>      { typedef unsigned long type; };
+template<> struct make_unsigned<unsigned long>    { typedef unsigned long type; };
+#if EIGEN_COMP_MSVC
+template<> struct make_unsigned<signed __int64>   { typedef unsigned __int64 type; };
+template<> struct make_unsigned<unsigned __int64> { typedef unsigned __int64 type; };
+#endif
+#endif
+
 template <typename T> struct add_const { typedef const T type; };
 template <typename T> struct add_const<T&> { typedef T& type; };
 
@@ -80,29 +143,163 @@ template<typename T> struct add_const_on_value_type<T*>        { typedef T const
 template<typename T> struct add_const_on_value_type<T* const>  { typedef T const* const type; };
 template<typename T> struct add_const_on_value_type<T const* const>  { typedef T const* const type; };
 
+
+template<typename From, typename To>
+struct is_convertible_impl
+{
+private:
+  struct any_conversion
+  {
+    template <typename T> any_conversion(const volatile T&);
+    template <typename T> any_conversion(T&);
+  };
+  struct yes {int a[1];};
+  struct no  {int a[2];};
+
+  static yes test(const To&, int);
+  static no  test(any_conversion, ...);
+
+public:
+  static From ms_from;
+#ifdef __INTEL_COMPILER
+  #pragma warning push
+  #pragma warning ( disable : 2259 )
+#endif
+  enum { value = sizeof(test(ms_from, 0))==sizeof(yes) };
+#ifdef __INTEL_COMPILER
+  #pragma warning pop
+#endif
+};
+
+template<typename From, typename To>
+struct is_convertible
+{
+  enum { value = is_convertible_impl<typename remove_all<From>::type,
+                                     typename remove_all<To  >::type>::value };
+};
+
 /** \internal Allows to enable/disable an overload
   * according to a compile time condition.
   */
-template<bool Condition, typename T> struct enable_if;
+template<bool Condition, typename T=void> struct enable_if;
 
 template<typename T> struct enable_if<true,T>
 { typedef T type; };
 
+#if defined(__CUDA_ARCH__)
+#if !defined(__FLT_EPSILON__)
+#define __FLT_EPSILON__ FLT_EPSILON
+#define __DBL_EPSILON__ DBL_EPSILON
+#endif
+
+namespace device {
+
+template<typename T> struct numeric_limits
+{
+  EIGEN_DEVICE_FUNC
+  static T epsilon() { return 0; }
+  static T (max)() { assert(false && "Highest not supported for this type"); }
+  static T (min)() { assert(false && "Lowest not supported for this type"); }
+  static T infinity() { assert(false && "Infinity not supported for this type"); }
+  static T quiet_NaN() { assert(false && "quiet_NaN not supported for this type"); }
+};
+template<> struct numeric_limits<float>
+{
+  EIGEN_DEVICE_FUNC
+  static float epsilon() { return __FLT_EPSILON__; }
+  EIGEN_DEVICE_FUNC
+  static float (max)() { return CUDART_MAX_NORMAL_F; }
+  EIGEN_DEVICE_FUNC
+  static float (min)() { return FLT_MIN; }
+  EIGEN_DEVICE_FUNC
+  static float infinity() { return CUDART_INF_F; }
+  EIGEN_DEVICE_FUNC
+  static float quiet_NaN() { return CUDART_NAN_F; }
+};
+template<> struct numeric_limits<double>
+{
+  EIGEN_DEVICE_FUNC
+  static double epsilon() { return __DBL_EPSILON__; }
+  EIGEN_DEVICE_FUNC
+  static double (max)() { return DBL_MAX; }
+  EIGEN_DEVICE_FUNC
+  static double (min)() { return DBL_MIN; }
+  EIGEN_DEVICE_FUNC
+  static double infinity() { return CUDART_INF; }
+  EIGEN_DEVICE_FUNC
+  static double quiet_NaN() { return CUDART_NAN; }
+};
+template<> struct numeric_limits<int>
+{
+  EIGEN_DEVICE_FUNC
+  static int epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static int (max)() { return INT_MAX; }
+  EIGEN_DEVICE_FUNC
+  static int (min)() { return INT_MIN; }
+};
+template<> struct numeric_limits<unsigned int>
+{
+  EIGEN_DEVICE_FUNC
+  static unsigned int epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static unsigned int (max)() { return UINT_MAX; }
+  EIGEN_DEVICE_FUNC
+  static unsigned int (min)() { return 0; }
+};
+template<> struct numeric_limits<long>
+{
+  EIGEN_DEVICE_FUNC
+  static long epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static long (max)() { return LONG_MAX; }
+  EIGEN_DEVICE_FUNC
+  static long (min)() { return LONG_MIN; }
+};
+template<> struct numeric_limits<unsigned long>
+{
+  EIGEN_DEVICE_FUNC
+  static unsigned long epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static unsigned long (max)() { return ULONG_MAX; }
+  EIGEN_DEVICE_FUNC
+  static unsigned long (min)() { return 0; }
+};
+template<> struct numeric_limits<long long>
+{
+  EIGEN_DEVICE_FUNC
+  static long long epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static long long (max)() { return LLONG_MAX; }
+  EIGEN_DEVICE_FUNC
+  static long long (min)() { return LLONG_MIN; }
+};
+template<> struct numeric_limits<unsigned long long>
+{
+  EIGEN_DEVICE_FUNC
+  static unsigned long long epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static unsigned long long (max)() { return ULLONG_MAX; }
+  EIGEN_DEVICE_FUNC
+  static unsigned long long (min)() { return 0; }
+};
+
+}
 
+#endif
 
 /** \internal
   * A base class do disable default copy ctor and copy assignement operator.
   */
 class noncopyable
 {
-  noncopyable(const noncopyable&);
-  const noncopyable& operator=(const noncopyable&);
+  EIGEN_DEVICE_FUNC noncopyable(const noncopyable&);
+  EIGEN_DEVICE_FUNC const noncopyable& operator=(const noncopyable&);
 protected:
-  noncopyable() {}
-  ~noncopyable() {}
+  EIGEN_DEVICE_FUNC noncopyable() {}
+  EIGEN_DEVICE_FUNC ~noncopyable() {}
 };
 
-
 /** \internal
   * Convenient struct to get the result type of a unary or binary functor.
   *
@@ -110,14 +307,20 @@ protected:
   * upcoming next STL generation (using a templated result member).
   * If none of these members is provided, then the type of the first argument is returned. FIXME, that behavior is a pretty bad hack.
   */
-template<typename T> struct result_of {};
+#if EIGEN_HAS_STD_RESULT_OF
+template<typename T> struct result_of {
+  typedef typename std::result_of<T>::type type1;
+  typedef typename remove_all<type1>::type type;
+};
+#else
+template<typename T> struct result_of { };
 
 struct has_none {int a[1];};
 struct has_std_result_type {int a[2];};
 struct has_tr1_result {int a[3];};
 
 template<typename Func, typename ArgType, int SizeOf=sizeof(has_none)>
-struct unary_result_of_select {typedef ArgType type;};
+struct unary_result_of_select {typedef typename internal::remove_all<ArgType>::type type;};
 
 template<typename Func, typename ArgType>
 struct unary_result_of_select<Func, ArgType, sizeof(has_std_result_type)> {typedef typename Func::result_type type;};
@@ -128,10 +331,10 @@ struct unary_result_of_select<Func, ArgType, sizeof(has_tr1_result)> {typedef ty
 template<typename Func, typename ArgType>
 struct result_of<Func(ArgType)> {
     template<typename T>
-    static has_std_result_type testFunctor(T const *, typename T::result_type const * = 0);
+    static has_std_result_type    testFunctor(T const *, typename T::result_type const * = 0);
     template<typename T>
-    static has_tr1_result      testFunctor(T const *, typename T::template result<T(ArgType)>::type const * = 0);
-    static has_none            testFunctor(...);
+    static has_tr1_result         testFunctor(T const *, typename T::template result<T(ArgType)>::type const * = 0);
+    static has_none               testFunctor(...);
 
     // note that the following indirection is needed for gcc-3.3
     enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
@@ -139,7 +342,7 @@ struct result_of<Func(ArgType)> {
 };
 
 template<typename Func, typename ArgType0, typename ArgType1, int SizeOf=sizeof(has_none)>
-struct binary_result_of_select {typedef ArgType0 type;};
+struct binary_result_of_select {typedef typename internal::remove_all<ArgType0>::type type;};
 
 template<typename Func, typename ArgType0, typename ArgType1>
 struct binary_result_of_select<Func, ArgType0, ArgType1, sizeof(has_std_result_type)>
@@ -152,16 +355,83 @@ struct binary_result_of_select<Func, ArgType0, ArgType1, sizeof(has_tr1_result)>
 template<typename Func, typename ArgType0, typename ArgType1>
 struct result_of<Func(ArgType0,ArgType1)> {
     template<typename T>
-    static has_std_result_type testFunctor(T const *, typename T::result_type const * = 0);
+    static has_std_result_type    testFunctor(T const *, typename T::result_type const * = 0);
     template<typename T>
-    static has_tr1_result      testFunctor(T const *, typename T::template result<T(ArgType0,ArgType1)>::type const * = 0);
-    static has_none            testFunctor(...);
+    static has_tr1_result         testFunctor(T const *, typename T::template result<T(ArgType0,ArgType1)>::type const * = 0);
+    static has_none               testFunctor(...);
 
     // note that the following indirection is needed for gcc-3.3
     enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
     typedef typename binary_result_of_select<Func, ArgType0, ArgType1, FunctorType>::type type;
 };
 
+template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2, int SizeOf=sizeof(has_none)>
+struct ternary_result_of_select {typedef typename internal::remove_all<ArgType0>::type type;};
+
+template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
+struct ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, sizeof(has_std_result_type)>
+{typedef typename Func::result_type type;};
+
+template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
+struct ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, sizeof(has_tr1_result)>
+{typedef typename Func::template result<Func(ArgType0,ArgType1,ArgType2)>::type type;};
+
+template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
+struct result_of<Func(ArgType0,ArgType1,ArgType2)> {
+    template<typename T>
+    static has_std_result_type    testFunctor(T const *, typename T::result_type const * = 0);
+    template<typename T>
+    static has_tr1_result         testFunctor(T const *, typename T::template result<T(ArgType0,ArgType1,ArgType2)>::type const * = 0);
+    static has_none               testFunctor(...);
+
+    // note that the following indirection is needed for gcc-3.3
+    enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
+    typedef typename ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, FunctorType>::type type;
+};
+#endif
+
+struct meta_yes { char a[1]; };
+struct meta_no  { char a[2]; };
+
+// Check whether T::ReturnType does exist
+template <typename T>
+struct has_ReturnType
+{
+  template <typename C> static meta_yes testFunctor(typename C::ReturnType const *);
+  template <typename C> static meta_no testFunctor(...);
+
+  enum { value = sizeof(testFunctor<T>(0)) == sizeof(meta_yes) };
+};
+
+template<typename T> const T* return_ptr();
+
+template <typename T, typename IndexType=Index>
+struct has_nullary_operator
+{
+  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()())>0)>::type * = 0);
+  static meta_no testFunctor(...);
+
+  enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
+};
+
+template <typename T, typename IndexType=Index>
+struct has_unary_operator
+{
+  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()(IndexType(0)))>0)>::type * = 0);
+  static meta_no testFunctor(...);
+
+  enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
+};
+
+template <typename T, typename IndexType=Index>
+struct has_binary_operator
+{
+  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()(IndexType(0),IndexType(0)))>0)>::type * = 0);
+  static meta_no testFunctor(...);
+
+  enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
+};
+
 /** \internal In short, it computes int(sqrt(\a Y)) with \a Y an integer.
   * Usage example: \code meta_sqrt<1023>::ret \endcode
   */
@@ -185,37 +455,26 @@ class meta_sqrt
 template<int Y, int InfX, int SupX>
 class meta_sqrt<Y, InfX, SupX, true> { public:  enum { ret = (SupX*SupX <= Y) ? SupX : InfX }; };
 
-/** \internal determines whether the product of two numeric types is allowed and what the return type is */
-template<typename T, typename U> struct scalar_product_traits
-{
-  enum { Defined = 0 };
-};
 
-template<typename T> struct scalar_product_traits<T,T>
+/** \internal Computes the least common multiple of two positive integer A and B
+  * at compile-time. It implements a naive algorithm testing all multiples of A.
+  * It thus works better if A>=B.
+  */
+template<int A, int B, int K=1, bool Done = ((A*K)%B)==0>
+struct meta_least_common_multiple
 {
-  enum {
-    // Cost = NumTraits<T>::MulCost,
-    Defined = 1
-  };
-  typedef T ReturnType;
+  enum { ret = meta_least_common_multiple<A,B,K+1>::ret };
 };
-
-template<typename T> struct scalar_product_traits<T,std::complex<T> >
+template<int A, int B, int K>
+struct meta_least_common_multiple<A,B,K,true>
 {
-  enum {
-    // Cost = 2*NumTraits<T>::MulCost,
-    Defined = 1
-  };
-  typedef std::complex<T> ReturnType;
+  enum { ret = A*K };
 };
 
-template<typename T> struct scalar_product_traits<std::complex<T>, T>
+/** \internal determines whether the product of two numeric types is allowed and what the return type is */
+template<typename T, typename U> struct scalar_product_traits
 {
-  enum {
-    // Cost = 2*NumTraits<T>::MulCost,
-    Defined = 1
-  };
-  typedef std::complex<T> ReturnType;
+  enum { Defined = 0 };
 };
 
 // FIXME quick workaround around current limitation of result_of
@@ -224,19 +483,51 @@ template<typename T> struct scalar_product_traits<std::complex<T>, T>
 // typedef typename scalar_product_traits<typename remove_all<ArgType0>::type, typename remove_all<ArgType1>::type>::ReturnType type;
 // };
 
-template<typename T> struct is_diagonal
-{ enum { ret = false }; };
+} // end namespace internal
+
+namespace numext {
+  
+#if defined(__CUDA_ARCH__)
+template<typename T> EIGEN_DEVICE_FUNC   void swap(T &a, T &b) { T tmp = b; b = a; a = tmp; }
+#else
+template<typename T> EIGEN_STRONG_INLINE void swap(T &a, T &b) { std::swap(a,b); }
+#endif
+
+#if defined(__CUDA_ARCH__)
+using internal::device::numeric_limits;
+#else
+using std::numeric_limits;
+#endif
+
+// Integer division with rounding up.
+// T is assumed to be an integer type with a>=0, and b>0
+template<typename T>
+T div_ceil(const T &a, const T &b)
+{
+  return (a+b-1) / b;
+}
 
-template<typename T> struct is_diagonal<DiagonalBase<T> >
-{ enum { ret = true }; };
+// The aim of the following functions is to bypass -Wfloat-equal warnings
+// when we really want a strict equality comparison on floating points.
+template<typename X, typename Y> EIGEN_STRONG_INLINE
+bool equal_strict(const X& x,const Y& y) { return x == y; }
 
-template<typename T> struct is_diagonal<DiagonalWrapper<T> >
-{ enum { ret = true }; };
+template<> EIGEN_STRONG_INLINE
+bool equal_strict(const float& x,const float& y) { return std::equal_to<float>()(x,y); }
 
-template<typename T, int S> struct is_diagonal<DiagonalMatrix<T,S> >
-{ enum { ret = true }; };
+template<> EIGEN_STRONG_INLINE
+bool equal_strict(const double& x,const double& y) { return std::equal_to<double>()(x,y); }
 
-} // end namespace internal
+template<typename X, typename Y> EIGEN_STRONG_INLINE
+bool not_equal_strict(const X& x,const Y& y) { return x != y; }
+
+template<> EIGEN_STRONG_INLINE
+bool not_equal_strict(const float& x,const float& y) { return std::not_equal_to<float>()(x,y); }
+
+template<> EIGEN_STRONG_INLINE
+bool not_equal_strict(const double& x,const double& y) { return std::not_equal_to<double>()(x,y); }
+
+} // end namespace numext
 
 } // end namespace Eigen
 
diff --git a/extern/Eigen3/Eigen/src/Core/util/ReenableStupidWarnings.h b/extern/Eigen3/Eigen/src/Core/util/ReenableStupidWarnings.h
index 5ddfbd4aa68..ecc82b7c8df 100644
--- a/extern/Eigen3/Eigen/src/Core/util/ReenableStupidWarnings.h
+++ b/extern/Eigen3/Eigen/src/Core/util/ReenableStupidWarnings.h
@@ -8,7 +8,20 @@
     #pragma warning pop
   #elif defined __clang__
     #pragma clang diagnostic pop
+  #elif defined __GNUC__  &&  (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))
+    #pragma GCC diagnostic pop
   #endif
+
+  #if defined __NVCC__
+//    Don't reenable the diagnostic messages, as it turns out these messages need
+//    to be disabled at the point of the template instantiation (i.e the user code)
+//    otherwise they'll be triggered by nvcc.
+//    #pragma diag_default code_is_unreachable
+//    #pragma diag_default initialization_not_reachable
+//    #pragma diag_default 2651
+//    #pragma diag_default 2653
+  #endif
+
 #endif
 
 #endif // EIGEN_WARNINGS_DISABLED
diff --git a/extern/Eigen3/Eigen/src/Core/util/StaticAssert.h b/extern/Eigen3/Eigen/src/Core/util/StaticAssert.h
index bac5d9fe92b..500e47792a4 100644
--- a/extern/Eigen3/Eigen/src/Core/util/StaticAssert.h
+++ b/extern/Eigen3/Eigen/src/Core/util/StaticAssert.h
@@ -24,9 +24,10 @@
  *
  */
 
+#ifndef EIGEN_STATIC_ASSERT
 #ifndef EIGEN_NO_STATIC_ASSERT
 
-  #if defined(__GXX_EXPERIMENTAL_CXX0X__) || (defined(_MSC_VER) && (_MSC_VER >= 1600))
+  #if EIGEN_MAX_CPP_VER>=11 && (__has_feature(cxx_static_assert) || (defined(__cplusplus) && __cplusplus >= 201103L) || (EIGEN_COMP_MSVC >= 1600))
 
     // if native static_assert is enabled, let's use it
     #define EIGEN_STATIC_ASSERT(X,MSG) static_assert(X,#MSG);
@@ -44,55 +45,65 @@
     struct static_assertion<true>
     {
       enum {
-        YOU_TRIED_CALLING_A_VECTOR_METHOD_ON_A_MATRIX,
-        YOU_MIXED_VECTORS_OF_DIFFERENT_SIZES,
-        YOU_MIXED_MATRICES_OF_DIFFERENT_SIZES,
-        THIS_METHOD_IS_ONLY_FOR_VECTORS_OF_A_SPECIFIC_SIZE,
-        THIS_METHOD_IS_ONLY_FOR_MATRICES_OF_A_SPECIFIC_SIZE,
-        THIS_METHOD_IS_ONLY_FOR_OBJECTS_OF_A_SPECIFIC_SIZE,
-        YOU_MADE_A_PROGRAMMING_MISTAKE,
-        EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT,
-        EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE,
-        YOU_CALLED_A_FIXED_SIZE_METHOD_ON_A_DYNAMIC_SIZE_MATRIX_OR_VECTOR,
-        YOU_CALLED_A_DYNAMIC_SIZE_METHOD_ON_A_FIXED_SIZE_MATRIX_OR_VECTOR,
-        UNALIGNED_LOAD_AND_STORE_OPERATIONS_UNIMPLEMENTED_ON_ALTIVEC,
-        THIS_FUNCTION_IS_NOT_FOR_INTEGER_NUMERIC_TYPES,
-        FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED,
-        NUMERIC_TYPE_MUST_BE_REAL,
-        COEFFICIENT_WRITE_ACCESS_TO_SELFADJOINT_NOT_SUPPORTED,
-        WRITING_TO_TRIANGULAR_PART_WITH_UNIT_DIAGONAL_IS_NOT_SUPPORTED,
-        THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE,
-        INVALID_MATRIX_PRODUCT,
-        INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS,
-        INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION,
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY,
-        THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES,
-        THIS_METHOD_IS_ONLY_FOR_ROW_MAJOR_MATRICES,
-        INVALID_MATRIX_TEMPLATE_PARAMETERS,
-        INVALID_MATRIXBASE_TEMPLATE_PARAMETERS,
-        BOTH_MATRICES_MUST_HAVE_THE_SAME_STORAGE_ORDER,
-        THIS_METHOD_IS_ONLY_FOR_DIAGONAL_MATRIX,
-        THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE,
-        THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_WITH_DIRECT_MEMORY_ACCESS_SUCH_AS_MAP_OR_PLAIN_MATRICES,
-        YOU_ALREADY_SPECIFIED_THIS_STRIDE,
-        INVALID_STORAGE_ORDER_FOR_THIS_VECTOR_EXPRESSION,
-        THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD,
-        PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1,
-        THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS,
-        YOU_CANNOT_MIX_ARRAYS_AND_MATRICES,
-        YOU_PERFORMED_AN_INVALID_TRANSFORMATION_CONVERSION,
-        THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY,
-        YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT,
-        THIS_METHOD_IS_ONLY_FOR_1x1_EXPRESSIONS,
-        THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL,
-        THIS_METHOD_IS_ONLY_FOR_ARRAYS_NOT_MATRICES,
-        YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED,
-        YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED,
-        THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE,
-        THE_STORAGE_ORDER_OF_BOTH_SIDES_MUST_MATCH,
-        OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG,
-        IMPLICIT_CONVERSION_TO_SCALAR_IS_FOR_INNER_PRODUCT_ONLY,
-        STORAGE_LAYOUT_DOES_NOT_MATCH
+        YOU_TRIED_CALLING_A_VECTOR_METHOD_ON_A_MATRIX=1,
+        YOU_MIXED_VECTORS_OF_DIFFERENT_SIZES=1,
+        YOU_MIXED_MATRICES_OF_DIFFERENT_SIZES=1,
+        THIS_METHOD_IS_ONLY_FOR_VECTORS_OF_A_SPECIFIC_SIZE=1,
+        THIS_METHOD_IS_ONLY_FOR_MATRICES_OF_A_SPECIFIC_SIZE=1,
+        THIS_METHOD_IS_ONLY_FOR_OBJECTS_OF_A_SPECIFIC_SIZE=1,
+        OUT_OF_RANGE_ACCESS=1,
+        YOU_MADE_A_PROGRAMMING_MISTAKE=1,
+        EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT=1,
+        EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE=1,
+        YOU_CALLED_A_FIXED_SIZE_METHOD_ON_A_DYNAMIC_SIZE_MATRIX_OR_VECTOR=1,
+        YOU_CALLED_A_DYNAMIC_SIZE_METHOD_ON_A_FIXED_SIZE_MATRIX_OR_VECTOR=1,
+        UNALIGNED_LOAD_AND_STORE_OPERATIONS_UNIMPLEMENTED_ON_ALTIVEC=1,
+        THIS_FUNCTION_IS_NOT_FOR_INTEGER_NUMERIC_TYPES=1,
+        FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED=1,
+        NUMERIC_TYPE_MUST_BE_REAL=1,
+        COEFFICIENT_WRITE_ACCESS_TO_SELFADJOINT_NOT_SUPPORTED=1,
+        WRITING_TO_TRIANGULAR_PART_WITH_UNIT_DIAGONAL_IS_NOT_SUPPORTED=1,
+        THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE=1,
+        INVALID_MATRIX_PRODUCT=1,
+        INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS=1,
+        INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION=1,
+        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY=1,
+        THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES=1,
+        THIS_METHOD_IS_ONLY_FOR_ROW_MAJOR_MATRICES=1,
+        INVALID_MATRIX_TEMPLATE_PARAMETERS=1,
+        INVALID_MATRIXBASE_TEMPLATE_PARAMETERS=1,
+        BOTH_MATRICES_MUST_HAVE_THE_SAME_STORAGE_ORDER=1,
+        THIS_METHOD_IS_ONLY_FOR_DIAGONAL_MATRIX=1,
+        THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE=1,
+        THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_WITH_DIRECT_MEMORY_ACCESS_SUCH_AS_MAP_OR_PLAIN_MATRICES=1,
+        YOU_ALREADY_SPECIFIED_THIS_STRIDE=1,
+        INVALID_STORAGE_ORDER_FOR_THIS_VECTOR_EXPRESSION=1,
+        THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD=1,
+        PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1=1,
+        THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS=1,
+        YOU_CANNOT_MIX_ARRAYS_AND_MATRICES=1,
+        YOU_PERFORMED_AN_INVALID_TRANSFORMATION_CONVERSION=1,
+        THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY=1,
+        YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT=1,
+        THIS_METHOD_IS_ONLY_FOR_1x1_EXPRESSIONS=1,
+        THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS=1,
+        THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL=1,
+        THIS_METHOD_IS_ONLY_FOR_ARRAYS_NOT_MATRICES=1,
+        YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED=1,
+        YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED=1,
+        THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE=1,
+        THE_STORAGE_ORDER_OF_BOTH_SIDES_MUST_MATCH=1,
+        OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG=1,
+        IMPLICIT_CONVERSION_TO_SCALAR_IS_FOR_INNER_PRODUCT_ONLY=1,
+        STORAGE_LAYOUT_DOES_NOT_MATCH=1,
+        EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT__INVALID_COST_VALUE=1,
+        THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS=1,
+        MATRIX_FREE_CONJUGATE_GRADIENT_IS_COMPATIBLE_WITH_UPPER_UNION_LOWER_MODE_ONLY=1,
+        THIS_TYPE_IS_NOT_SUPPORTED=1,
+        STORAGE_KIND_MUST_MATCH=1,
+        STORAGE_INDEX_MUST_MATCH=1,
+        CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY=1,
+        SELFADJOINTVIEW_ACCEPTS_UPPER_AND_LOWER_MODE_ONLY=1
       };
     };
 
@@ -103,15 +114,15 @@
     // Specialized implementation for MSVC to avoid "conditional
     // expression is constant" warnings.  This implementation doesn't
     // appear to work under GCC, hence the multiple implementations.
-    #ifdef _MSC_VER
+    #if EIGEN_COMP_MSVC
 
       #define EIGEN_STATIC_ASSERT(CONDITION,MSG) \
         {Eigen::internal::static_assertion<bool(CONDITION)>::MSG;}
 
     #else
-
+      // In some cases clang interprets bool(CONDITION) as function declaration
       #define EIGEN_STATIC_ASSERT(CONDITION,MSG) \
-        if (Eigen::internal::static_assertion<bool(CONDITION)>::MSG) {}
+        if (Eigen::internal::static_assertion<static_cast<bool>(CONDITION)>::MSG) {}
 
     #endif
 
@@ -122,7 +133,7 @@
   #define EIGEN_STATIC_ASSERT(CONDITION,MSG) eigen_assert((CONDITION) && #MSG);
 
 #endif // EIGEN_NO_STATIC_ASSERT
-
+#endif // EIGEN_STATIC_ASSERT
 
 // static assertion failing if the type \a TYPE is not a vector type
 #define EIGEN_STATIC_ASSERT_VECTOR_ONLY(TYPE) \
@@ -159,7 +170,7 @@
 
 #define EIGEN_PREDICATE_SAME_MATRIX_SIZE(TYPE0,TYPE1) \
      ( \
-        (int(TYPE0::SizeAtCompileTime)==0 && int(TYPE1::SizeAtCompileTime)==0) \
+        (int(Eigen::internal::size_of_xpr_at_compile_time<TYPE0>::ret)==0 && int(Eigen::internal::size_of_xpr_at_compile_time<TYPE1>::ret)==0) \
     || (\
           (int(TYPE0::RowsAtCompileTime)==Eigen::Dynamic \
         || int(TYPE1::RowsAtCompileTime)==Eigen::Dynamic \
@@ -170,13 +181,8 @@
        ) \
      )
 
-#ifdef EIGEN2_SUPPORT
-  #define EIGEN_STATIC_ASSERT_NON_INTEGER(TYPE) \
-    eigen_assert(!NumTraits<Scalar>::IsInteger);
-#else
-  #define EIGEN_STATIC_ASSERT_NON_INTEGER(TYPE) \
+#define EIGEN_STATIC_ASSERT_NON_INTEGER(TYPE) \
     EIGEN_STATIC_ASSERT(!NumTraits<TYPE>::IsInteger, THIS_FUNCTION_IS_NOT_FOR_INTEGER_NUMERIC_TYPES)
-#endif
 
 
 // static assertion failing if it is guaranteed at compile-time that the two matrix expression types have different sizes
@@ -191,18 +197,22 @@
                           THIS_METHOD_IS_ONLY_FOR_1x1_EXPRESSIONS)
 
 #define EIGEN_STATIC_ASSERT_LVALUE(Derived) \
-      EIGEN_STATIC_ASSERT(internal::is_lvalue<Derived>::value, \
+      EIGEN_STATIC_ASSERT(Eigen::internal::is_lvalue<Derived>::value, \
                           THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY)
 
 #define EIGEN_STATIC_ASSERT_ARRAYXPR(Derived) \
-      EIGEN_STATIC_ASSERT((internal::is_same<typename internal::traits<Derived>::XprKind, ArrayXpr>::value), \
+      EIGEN_STATIC_ASSERT((Eigen::internal::is_same<typename Eigen::internal::traits<Derived>::XprKind, ArrayXpr>::value), \
                           THIS_METHOD_IS_ONLY_FOR_ARRAYS_NOT_MATRICES)
 
 #define EIGEN_STATIC_ASSERT_SAME_XPR_KIND(Derived1, Derived2) \
-      EIGEN_STATIC_ASSERT((internal::is_same<typename internal::traits<Derived1>::XprKind, \
-                                             typename internal::traits<Derived2>::XprKind \
+      EIGEN_STATIC_ASSERT((Eigen::internal::is_same<typename Eigen::internal::traits<Derived1>::XprKind, \
+                                             typename Eigen::internal::traits<Derived2>::XprKind \
                                             >::value), \
                           YOU_CANNOT_MIX_ARRAYS_AND_MATRICES)
 
+// Check that a cost value is positive, and that is stay within a reasonable range
+// TODO this check could be enabled for internal debugging only
+#define EIGEN_INTERNAL_CHECK_COST_VALUE(C) \
+      EIGEN_STATIC_ASSERT((C)>=0 && (C)<=HugeCost*HugeCost, EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT__INVALID_COST_VALUE);
 
 #endif // EIGEN_STATIC_ASSERT_H
diff --git a/extern/Eigen3/Eigen/src/Core/util/XprHelper.h b/extern/Eigen3/Eigen/src/Core/util/XprHelper.h
index d05f8e5f6f4..ba5bd186d29 100644
--- a/extern/Eigen3/Eigen/src/Core/util/XprHelper.h
+++ b/extern/Eigen3/Eigen/src/Core/util/XprHelper.h
@@ -14,20 +14,77 @@
 // just a workaround because GCC seems to not really like empty structs
 // FIXME: gcc 4.3 generates bad code when strict-aliasing is enabled
 // so currently we simply disable this optimization for gcc 4.3
-#if (defined __GNUG__) && !((__GNUC__==4) && (__GNUC_MINOR__==3))
+#if EIGEN_COMP_GNUC && !EIGEN_GNUC_AT(4,3)
   #define EIGEN_EMPTY_STRUCT_CTOR(X) \
-    EIGEN_STRONG_INLINE X() {} \
-    EIGEN_STRONG_INLINE X(const X& ) {}
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE X() {} \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE X(const X& ) {}
 #else
   #define EIGEN_EMPTY_STRUCT_CTOR(X)
 #endif
 
 namespace Eigen {
 
-typedef EIGEN_DEFAULT_DENSE_INDEX_TYPE DenseIndex;
-
 namespace internal {
 
+template<typename IndexDest, typename IndexSrc>
+EIGEN_DEVICE_FUNC
+inline IndexDest convert_index(const IndexSrc& idx) {
+  // for sizeof(IndexDest)>=sizeof(IndexSrc) compilers should be able to optimize this away:
+  eigen_internal_assert(idx <= NumTraits<IndexDest>::highest() && "Index value to big for target type");
+  return IndexDest(idx);
+}
+
+
+// promote_scalar_arg is an helper used in operation between an expression and a scalar, like:
+//    expression * scalar
+// Its role is to determine how the type T of the scalar operand should be promoted given the scalar type ExprScalar of the given expression.
+// The IsSupported template parameter must be provided by the caller as: internal::has_ReturnType<ScalarBinaryOpTraits<ExprScalar,T,op> >::value using the proper order for ExprScalar and T.
+// Then the logic is as follows:
+//  - if the operation is natively supported as defined by IsSupported, then the scalar type is not promoted, and T is returned.
+//  - otherwise, NumTraits<ExprScalar>::Literal is returned if T is implicitly convertible to NumTraits<ExprScalar>::Literal AND that this does not imply a float to integer conversion.
+//  - otherwise, ExprScalar is returned if T is implicitly convertible to ExprScalar AND that this does not imply a float to integer conversion.
+//  - In all other cases, the promoted type is not defined, and the respective operation is thus invalid and not available (SFINAE).
+template<typename ExprScalar,typename T, bool IsSupported>
+struct promote_scalar_arg;
+
+template<typename S,typename T>
+struct promote_scalar_arg<S,T,true>
+{
+  typedef T type;
+};
+
+// Recursively check safe conversion to PromotedType, and then ExprScalar if they are different.
+template<typename ExprScalar,typename T,typename PromotedType,
+  bool ConvertibleToLiteral = internal::is_convertible<T,PromotedType>::value,
+  bool IsSafe = NumTraits<T>::IsInteger || !NumTraits<PromotedType>::IsInteger>
+struct promote_scalar_arg_unsupported;
+
+// Start recursion with NumTraits<ExprScalar>::Literal
+template<typename S,typename T>
+struct promote_scalar_arg<S,T,false> : promote_scalar_arg_unsupported<S,T,typename NumTraits<S>::Literal> {};
+
+// We found a match!
+template<typename S,typename T, typename PromotedType>
+struct promote_scalar_arg_unsupported<S,T,PromotedType,true,true>
+{
+  typedef PromotedType type;
+};
+
+// No match, but no real-to-integer issues, and ExprScalar and current PromotedType are different,
+// so let's try to promote to ExprScalar
+template<typename ExprScalar,typename T, typename PromotedType>
+struct promote_scalar_arg_unsupported<ExprScalar,T,PromotedType,false,true>
+   : promote_scalar_arg_unsupported<ExprScalar,T,ExprScalar>
+{};
+
+// Unsafe real-to-integer, let's stop.
+template<typename S,typename T, typename PromotedType, bool ConvertibleToLiteral>
+struct promote_scalar_arg_unsupported<S,T,PromotedType,ConvertibleToLiteral,false> {};
+
+// T is not even convertible to ExprScalar, let's stop.
+template<typename S,typename T>
+struct promote_scalar_arg_unsupported<S,T,S,false,true> {};
+
 //classes inheriting no_assignment_operator don't generate a default operator=.
 class no_assignment_operator
 {
@@ -50,19 +107,19 @@ template<typename T, int Value> class variable_if_dynamic
 {
   public:
     EIGEN_EMPTY_STRUCT_CTOR(variable_if_dynamic)
-    explicit variable_if_dynamic(T v) { EIGEN_ONLY_USED_FOR_DEBUG(v); assert(v == T(Value)); }
-    static T value() { return T(Value); }
-    void setValue(T) {}
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit variable_if_dynamic(T v) { EIGEN_ONLY_USED_FOR_DEBUG(v); eigen_assert(v == T(Value)); }
+    EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE T value() { return T(Value); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void setValue(T) {}
 };
 
 template<typename T> class variable_if_dynamic<T, Dynamic>
 {
     T m_value;
-    variable_if_dynamic() { assert(false); }
+    EIGEN_DEVICE_FUNC variable_if_dynamic() { eigen_assert(false); }
   public:
-    explicit variable_if_dynamic(T value) : m_value(value) {}
-    T value() const { return m_value; }
-    void setValue(T value) { m_value = value; }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit variable_if_dynamic(T value) : m_value(value) {}
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T value() const { return m_value; }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void setValue(T value) { m_value = value; }
 };
 
 /** \internal like variable_if_dynamic but for DynamicIndex
@@ -71,19 +128,19 @@ template<typename T, int Value> class variable_if_dynamicindex
 {
   public:
     EIGEN_EMPTY_STRUCT_CTOR(variable_if_dynamicindex)
-    explicit variable_if_dynamicindex(T v) { EIGEN_ONLY_USED_FOR_DEBUG(v); assert(v == T(Value)); }
-    static T value() { return T(Value); }
-    void setValue(T) {}
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit variable_if_dynamicindex(T v) { EIGEN_ONLY_USED_FOR_DEBUG(v); eigen_assert(v == T(Value)); }
+    EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE T value() { return T(Value); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void setValue(T) {}
 };
 
 template<typename T> class variable_if_dynamicindex<T, DynamicIndex>
 {
     T m_value;
-    variable_if_dynamicindex() { assert(false); }
+    EIGEN_DEVICE_FUNC variable_if_dynamicindex() { eigen_assert(false); }
   public:
-    explicit variable_if_dynamicindex(T value) : m_value(value) {}
-    T value() const { return m_value; }
-    void setValue(T value) { m_value = value; }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit variable_if_dynamicindex(T value) : m_value(value) {}
+    EIGEN_DEVICE_FUNC T EIGEN_STRONG_INLINE value() const { return m_value; }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void setValue(T value) { m_value = value; }
 };
 
 template<typename T> struct functor_traits
@@ -101,7 +158,73 @@ template<typename T> struct packet_traits;
 template<typename T> struct unpacket_traits
 {
   typedef T type;
-  enum {size=1};
+  typedef T half;
+  enum
+  {
+    size = 1,
+    alignment = 1
+  };
+};
+
+template<int Size, typename PacketType,
+         bool Stop = Size==Dynamic || (Size%unpacket_traits<PacketType>::size)==0 || is_same<PacketType,typename unpacket_traits<PacketType>::half>::value>
+struct find_best_packet_helper;
+
+template< int Size, typename PacketType>
+struct find_best_packet_helper<Size,PacketType,true>
+{
+  typedef PacketType type;
+};
+
+template<int Size, typename PacketType>
+struct find_best_packet_helper<Size,PacketType,false>
+{
+  typedef typename find_best_packet_helper<Size,typename unpacket_traits<PacketType>::half>::type type;
+};
+
+template<typename T, int Size>
+struct find_best_packet
+{
+  typedef typename find_best_packet_helper<Size,typename packet_traits<T>::type>::type type;
+};
+
+#if EIGEN_MAX_STATIC_ALIGN_BYTES>0
+template<int ArrayBytes, int AlignmentBytes,
+         bool Match     =  bool((ArrayBytes%AlignmentBytes)==0),
+         bool TryHalf   =  bool(EIGEN_MIN_ALIGN_BYTES<AlignmentBytes) >
+struct compute_default_alignment_helper
+{
+  enum { value = 0 };
+};
+
+template<int ArrayBytes, int AlignmentBytes, bool TryHalf>
+struct compute_default_alignment_helper<ArrayBytes, AlignmentBytes, true, TryHalf> // Match
+{
+  enum { value = AlignmentBytes };
+};
+
+template<int ArrayBytes, int AlignmentBytes>
+struct compute_default_alignment_helper<ArrayBytes, AlignmentBytes, false, true> // Try-half
+{
+  // current packet too large, try with an half-packet
+  enum { value = compute_default_alignment_helper<ArrayBytes, AlignmentBytes/2>::value };
+};
+#else
+// If static alignment is disabled, no need to bother.
+// This also avoids a division by zero in "bool Match =  bool((ArrayBytes%AlignmentBytes)==0)"
+template<int ArrayBytes, int AlignmentBytes>
+struct compute_default_alignment_helper
+{
+  enum { value = 0 };
+};
+#endif
+
+template<typename T, int Size> struct compute_default_alignment {
+  enum { value = compute_default_alignment_helper<Size*sizeof(T),EIGEN_MAX_STATIC_ALIGN_BYTES>::value };
+};
+
+template<typename T> struct compute_default_alignment<T,Dynamic> {
+  enum { value = EIGEN_MAX_ALIGN_BYTES };
 };
 
 template<typename _Scalar, int _Rows, int _Cols,
@@ -127,35 +250,12 @@ template<typename _Scalar, int _Rows, int _Cols,
 template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
 class compute_matrix_flags
 {
-    enum {
-      row_major_bit = Options&RowMajor ? RowMajorBit : 0,
-      is_dynamic_size_storage = MaxRows==Dynamic || MaxCols==Dynamic,
-
-      aligned_bit =
-      (
-            ((Options&DontAlign)==0)
-        && (
-#if EIGEN_ALIGN_STATICALLY
-             ((!is_dynamic_size_storage) && (((MaxCols*MaxRows*int(sizeof(Scalar))) % 16) == 0))
-#else
-             0
-#endif
-
-          ||
-
-#if EIGEN_ALIGN
-             is_dynamic_size_storage
-#else
-             0
-#endif
-
-          )
-      ) ? AlignedBit : 0,
-      packet_access_bit = packet_traits<Scalar>::Vectorizable && aligned_bit ? PacketAccessBit : 0
-    };
-
+    enum { row_major_bit = Options&RowMajor ? RowMajorBit : 0 };
   public:
-    enum { ret = LinearAccessBit | LvalueBit | DirectAccessBit | NestByRefBit | packet_access_bit | row_major_bit | aligned_bit };
+    // FIXME currently we still have to handle DirectAccessBit at the expression level to handle DenseCoeffsBase<>
+    // and then propagate this information to the evaluator's flags.
+    // However, I (Gael) think that DirectAccessBit should only matter at the evaluation stage.
+    enum { ret = DirectAccessBit | LvalueBit | NestByRefBit | row_major_bit };
 };
 
 template<int _Rows, int _Cols> struct size_at_compile_time
@@ -163,34 +263,43 @@ template<int _Rows, int _Cols> struct size_at_compile_time
   enum { ret = (_Rows==Dynamic || _Cols==Dynamic) ? Dynamic : _Rows * _Cols };
 };
 
+template<typename XprType> struct size_of_xpr_at_compile_time
+{
+  enum { ret = size_at_compile_time<traits<XprType>::RowsAtCompileTime,traits<XprType>::ColsAtCompileTime>::ret };
+};
+
 /* plain_matrix_type : the difference from eval is that plain_matrix_type is always a plain matrix type,
  * whereas eval is a const reference in the case of a matrix
  */
 
 template<typename T, typename StorageKind = typename traits<T>::StorageKind> struct plain_matrix_type;
-template<typename T, typename BaseClassType> struct plain_matrix_type_dense;
+template<typename T, typename BaseClassType, int Flags> struct plain_matrix_type_dense;
 template<typename T> struct plain_matrix_type<T,Dense>
 {
-  typedef typename plain_matrix_type_dense<T,typename traits<T>::XprKind>::type type;
+  typedef typename plain_matrix_type_dense<T,typename traits<T>::XprKind, traits<T>::Flags>::type type;
+};
+template<typename T> struct plain_matrix_type<T,DiagonalShape>
+{
+  typedef typename T::PlainObject type;
 };
 
-template<typename T> struct plain_matrix_type_dense<T,MatrixXpr>
+template<typename T, int Flags> struct plain_matrix_type_dense<T,MatrixXpr,Flags>
 {
   typedef Matrix<typename traits<T>::Scalar,
                 traits<T>::RowsAtCompileTime,
                 traits<T>::ColsAtCompileTime,
-                AutoAlign | (traits<T>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                AutoAlign | (Flags&RowMajorBit ? RowMajor : ColMajor),
                 traits<T>::MaxRowsAtCompileTime,
                 traits<T>::MaxColsAtCompileTime
           > type;
 };
 
-template<typename T> struct plain_matrix_type_dense<T,ArrayXpr>
+template<typename T, int Flags> struct plain_matrix_type_dense<T,ArrayXpr,Flags>
 {
   typedef Array<typename traits<T>::Scalar,
                 traits<T>::RowsAtCompileTime,
                 traits<T>::ColsAtCompileTime,
-                AutoAlign | (traits<T>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                AutoAlign | (Flags&RowMajorBit ? RowMajor : ColMajor),
                 traits<T>::MaxRowsAtCompileTime,
                 traits<T>::MaxColsAtCompileTime
           > type;
@@ -215,6 +324,11 @@ template<typename T> struct eval<T,Dense>
 //           > type;
 };
 
+template<typename T> struct eval<T,DiagonalShape>
+{
+  typedef typename plain_matrix_type<T>::type type;
+};
+
 // for matrices, no need to evaluate, just use a const reference to avoid a useless copy
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 struct eval<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>, Dense>
@@ -229,6 +343,15 @@ struct eval<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>, Dense>
 };
 
 
+/* similar to plain_matrix_type, but using the evaluator's Flags */
+template<typename T, typename StorageKind = typename traits<T>::StorageKind> struct plain_object_eval;
+
+template<typename T>
+struct plain_object_eval<T,Dense>
+{
+  typedef typename plain_matrix_type_dense<T,typename traits<T>::XprKind, evaluator<T>::Flags>::type type;
+};
+
 
 /* plain_matrix_type_column_major : same as plain_matrix_type but guaranteed to be column-major
  */
@@ -266,9 +389,6 @@ template<typename T> struct plain_matrix_type_row_major
           > type;
 };
 
-// we should be able to get rid of this one too
-template<typename T> struct must_nest_by_value { enum { ret = false }; };
-
 /** \internal The reference selector for template expressions. The idea is that we don't
   * need to use references for expressions since they are light weight proxy
   * objects which should generate no copying overhead. */
@@ -280,6 +400,12 @@ struct ref_selector
     T const&,
     const T
   >::type type;
+  
+  typedef typename conditional<
+    bool(traits<T>::Flags & NestByRefBit),
+    T &,
+    T
+  >::type non_const_type;
 };
 
 /** \internal Adds the const qualifier on the value-type of T2 if and only if T1 is a const type */
@@ -293,54 +419,41 @@ struct transfer_constness
   >::type type;
 };
 
-/** \internal Determines how a given expression should be nested into another one.
+
+// However, we still need a mechanism to detect whether an expression which is evaluated multiple time
+// has to be evaluated into a temporary.
+// That's the purpose of this new nested_eval helper:
+/** \internal Determines how a given expression should be nested when evaluated multiple times.
   * For example, when you do a * (b+c), Eigen will determine how the expression b+c should be
-  * nested into the bigger product expression. The choice is between nesting the expression b+c as-is, or
+  * evaluated into the bigger product expression. The choice is between nesting the expression b+c as-is, or
   * evaluating that expression b+c into a temporary variable d, and nest d so that the resulting expression is
   * a*d. Evaluating can be beneficial for example if every coefficient access in the resulting expression causes
   * many coefficient accesses in the nested expressions -- as is the case with matrix product for example.
   *
-  * \param T the type of the expression being nested
-  * \param n the number of coefficient accesses in the nested expression for each coefficient access in the bigger expression.
-  *
-  * Note that if no evaluation occur, then the constness of T is preserved.
-  *
-  * Example. Suppose that a, b, and c are of type Matrix3d. The user forms the expression a*(b+c).
-  * b+c is an expression "sum of matrices", which we will denote by S. In order to determine how to nest it,
-  * the Product expression uses: nested<S, 3>::ret, which turns out to be Matrix3d because the internal logic of
-  * nested determined that in this case it was better to evaluate the expression b+c into a temporary. On the other hand,
-  * since a is of type Matrix3d, the Product expression nests it as nested<Matrix3d, 3>::ret, which turns out to be
-  * const Matrix3d&, because the internal logic of nested determined that since a was already a matrix, there was no point
-  * in copying it into another matrix.
+  * \tparam T the type of the expression being nested.
+  * \tparam n the number of coefficient accesses in the nested expression for each coefficient access in the bigger expression.
+  * \tparam PlainObject the type of the temporary if needed.
   */
-template<typename T, int n=1, typename PlainObject = typename eval<T>::type> struct nested
+template<typename T, int n, typename PlainObject = typename plain_object_eval<T>::type> struct nested_eval
 {
   enum {
-    // for the purpose of this test, to keep it reasonably simple, we arbitrarily choose a value of Dynamic values.
-    // the choice of 10000 makes it larger than any practical fixed value and even most dynamic values.
-    // in extreme cases where these assumptions would be wrong, we would still at worst suffer performance issues
-    // (poor choice of temporaries).
-    // it's important that this value can still be squared without integer overflowing.
-    DynamicAsInteger = 10000,
     ScalarReadCost = NumTraits<typename traits<T>::Scalar>::ReadCost,
-    ScalarReadCostAsInteger = ScalarReadCost == Dynamic ? int(DynamicAsInteger) : int(ScalarReadCost),
-    CoeffReadCost = traits<T>::CoeffReadCost,
-    CoeffReadCostAsInteger = CoeffReadCost == Dynamic ? int(DynamicAsInteger) : int(CoeffReadCost),
-    NAsInteger = n == Dynamic ? int(DynamicAsInteger) : n,
-    CostEvalAsInteger   = (NAsInteger+1) * ScalarReadCostAsInteger + CoeffReadCostAsInteger,
-    CostNoEvalAsInteger = NAsInteger * CoeffReadCostAsInteger
+    CoeffReadCost = evaluator<T>::CoeffReadCost,  // NOTE What if an evaluator evaluate itself into a tempory?
+                                                  //      Then CoeffReadCost will be small (e.g., 1) but we still have to evaluate, especially if n>1.
+                                                  //      This situation is already taken care by the EvalBeforeNestingBit flag, which is turned ON
+                                                  //      for all evaluator creating a temporary. This flag is then propagated by the parent evaluators.
+                                                  //      Another solution could be to count the number of temps?
+    NAsInteger = n == Dynamic ? HugeCost : n,
+    CostEval   = (NAsInteger+1) * ScalarReadCost + CoeffReadCost,
+    CostNoEval = NAsInteger * CoeffReadCost,
+    Evaluate = (int(evaluator<T>::Flags) & EvalBeforeNestingBit) || (int(CostEval) < int(CostNoEval))
   };
 
-  typedef typename conditional<
-      ( (int(traits<T>::Flags) & EvalBeforeNestingBit) ||
-        int(CostEvalAsInteger) < int(CostNoEvalAsInteger)
-      ),
-      PlainObject,
-      typename ref_selector<T>::type
-  >::type type;
+  typedef typename conditional<Evaluate, PlainObject, typename ref_selector<T>::type>::type type;
 };
 
 template<typename T>
+EIGEN_DEVICE_FUNC
 inline T* const_cast_ptr(const T* ptr)
 {
   return const_cast<T*>(ptr);
@@ -364,30 +477,13 @@ struct dense_xpr_base<Derived, ArrayXpr>
   typedef ArrayBase<Derived> type;
 };
 
-/** \internal Helper base class to add a scalar multiple operator
-  * overloads for complex types */
-template<typename Derived, typename Scalar, typename OtherScalar, typename BaseType,
-         bool EnableIt = !is_same<Scalar,OtherScalar>::value >
-struct special_scalar_op_base : public BaseType
-{
-  // dummy operator* so that the
-  // "using special_scalar_op_base::operator*" compiles
-  void operator*() const;
-};
+template<typename Derived, typename XprKind = typename traits<Derived>::XprKind, typename StorageKind = typename traits<Derived>::StorageKind>
+struct generic_xpr_base;
 
-template<typename Derived,typename Scalar,typename OtherScalar, typename BaseType>
-struct special_scalar_op_base<Derived,Scalar,OtherScalar,BaseType,true>  : public BaseType
+template<typename Derived, typename XprKind>
+struct generic_xpr_base<Derived, XprKind, Dense>
 {
-  const CwiseUnaryOp<scalar_multiple2_op<Scalar,OtherScalar>, Derived>
-  operator*(const OtherScalar& scalar) const
-  {
-    return CwiseUnaryOp<scalar_multiple2_op<Scalar,OtherScalar>, Derived>
-      (*static_cast<const Derived*>(this), scalar_multiple2_op<Scalar,OtherScalar>(scalar));
-  }
-
-  inline friend const CwiseUnaryOp<scalar_multiple2_op<Scalar,OtherScalar>, Derived>
-  operator*(const OtherScalar& scalar, const Derived& matrix)
-  { return static_cast<const special_scalar_op_base&>(matrix).operator*(scalar); }
+  typedef typename dense_xpr_base<Derived,XprKind>::type type;
 };
 
 template<typename XprType, typename CastType> struct cast_return_type
@@ -405,9 +501,79 @@ template <typename A> struct promote_storage_type<A,A>
 {
   typedef A ret;
 };
+template <typename A> struct promote_storage_type<A, const A>
+{
+  typedef A ret;
+};
+template <typename A> struct promote_storage_type<const A, A>
+{
+  typedef A ret;
+};
+
+/** \internal Specify the "storage kind" of applying a coefficient-wise
+  * binary operations between two expressions of kinds A and B respectively.
+  * The template parameter Functor permits to specialize the resulting storage kind wrt to
+  * the functor.
+  * The default rules are as follows:
+  * \code
+  * A      op A      -> A
+  * A      op dense  -> dense
+  * dense  op B      -> dense
+  * sparse op dense  -> sparse
+  * dense  op sparse -> sparse
+  * \endcode
+  */
+template <typename A, typename B, typename Functor> struct cwise_promote_storage_type;
+
+template <typename A, typename Functor>                   struct cwise_promote_storage_type<A,A,Functor>                                      { typedef A      ret; };
+template <typename Functor>                               struct cwise_promote_storage_type<Dense,Dense,Functor>                              { typedef Dense  ret; };
+template <typename A, typename Functor>                   struct cwise_promote_storage_type<A,Dense,Functor>                                  { typedef Dense  ret; };
+template <typename B, typename Functor>                   struct cwise_promote_storage_type<Dense,B,Functor>                                  { typedef Dense  ret; };
+template <typename Functor>                               struct cwise_promote_storage_type<Sparse,Dense,Functor>                             { typedef Sparse ret; };
+template <typename Functor>                               struct cwise_promote_storage_type<Dense,Sparse,Functor>                             { typedef Sparse ret; };
+
+template <typename LhsKind, typename RhsKind, int LhsOrder, int RhsOrder> struct cwise_promote_storage_order {
+  enum { value = LhsOrder };
+};
+
+template <typename LhsKind, int LhsOrder, int RhsOrder>   struct cwise_promote_storage_order<LhsKind,Sparse,LhsOrder,RhsOrder>                { enum { value = RhsOrder }; };
+template <typename RhsKind, int LhsOrder, int RhsOrder>   struct cwise_promote_storage_order<Sparse,RhsKind,LhsOrder,RhsOrder>                { enum { value = LhsOrder }; };
+template <int Order>                                      struct cwise_promote_storage_order<Sparse,Sparse,Order,Order>                       { enum { value = Order }; };
+
+
+/** \internal Specify the "storage kind" of multiplying an expression of kind A with kind B.
+  * The template parameter ProductTag permits to specialize the resulting storage kind wrt to
+  * some compile-time properties of the product: GemmProduct, GemvProduct, OuterProduct, InnerProduct.
+  * The default rules are as follows:
+  * \code
+  *  K * K            -> K
+  *  dense * K        -> dense
+  *  K * dense        -> dense
+  *  diag * K         -> K
+  *  K * diag         -> K
+  *  Perm * K         -> K
+  * K * Perm          -> K
+  * \endcode
+  */
+template <typename A, typename B, int ProductTag> struct product_promote_storage_type;
+
+template <typename A, int ProductTag> struct product_promote_storage_type<A,                  A,                  ProductTag> { typedef A     ret;};
+template <int ProductTag>             struct product_promote_storage_type<Dense,              Dense,              ProductTag> { typedef Dense ret;};
+template <typename A, int ProductTag> struct product_promote_storage_type<A,                  Dense,              ProductTag> { typedef Dense ret; };
+template <typename B, int ProductTag> struct product_promote_storage_type<Dense,              B,                  ProductTag> { typedef Dense ret; };
+
+template <typename A, int ProductTag> struct product_promote_storage_type<A,                  DiagonalShape,      ProductTag> { typedef A ret; };
+template <typename B, int ProductTag> struct product_promote_storage_type<DiagonalShape,      B,                  ProductTag> { typedef B ret; };
+template <int ProductTag>             struct product_promote_storage_type<Dense,              DiagonalShape,      ProductTag> { typedef Dense ret; };
+template <int ProductTag>             struct product_promote_storage_type<DiagonalShape,      Dense,              ProductTag> { typedef Dense ret; };
+
+template <typename A, int ProductTag> struct product_promote_storage_type<A,                  PermutationStorage, ProductTag> { typedef A ret; };
+template <typename B, int ProductTag> struct product_promote_storage_type<PermutationStorage, B,                  ProductTag> { typedef B ret; };
+template <int ProductTag>             struct product_promote_storage_type<Dense,              PermutationStorage, ProductTag> { typedef Dense ret; };
+template <int ProductTag>             struct product_promote_storage_type<PermutationStorage, Dense,              ProductTag> { typedef Dense ret; };
 
 /** \internal gives the plain matrix or array type to store a row/column/diagonal of a matrix type.
-  * \param Scalar optional parameter allowing to pass a different scalar type than the one of the MatrixType.
+  * \tparam Scalar optional parameter allowing to pass a different scalar type than the one of the MatrixType.
   */
 template<typename ExpressionType, typename Scalar = typename ExpressionType::Scalar>
 struct plain_row_type
@@ -455,15 +621,201 @@ struct plain_diag_type
   >::type type;
 };
 
+template<typename Expr,typename Scalar = typename Expr::Scalar>
+struct plain_constant_type
+{
+  enum { Options = (traits<Expr>::Flags&RowMajorBit)?RowMajor:0 };
+
+  typedef Array<Scalar,  traits<Expr>::RowsAtCompileTime,   traits<Expr>::ColsAtCompileTime,
+                Options, traits<Expr>::MaxRowsAtCompileTime,traits<Expr>::MaxColsAtCompileTime> array_type;
+
+  typedef Matrix<Scalar,  traits<Expr>::RowsAtCompileTime,   traits<Expr>::ColsAtCompileTime,
+                 Options, traits<Expr>::MaxRowsAtCompileTime,traits<Expr>::MaxColsAtCompileTime> matrix_type;
+
+  typedef CwiseNullaryOp<scalar_constant_op<Scalar>, const typename conditional<is_same< typename traits<Expr>::XprKind, MatrixXpr >::value, matrix_type, array_type>::type > type;
+};
+
 template<typename ExpressionType>
 struct is_lvalue
 {
-  enum { value = !bool(is_const<ExpressionType>::value) &&
+  enum { value = (!bool(is_const<ExpressionType>::value)) &&
                  bool(traits<ExpressionType>::Flags & LvalueBit) };
 };
 
+template<typename T> struct is_diagonal
+{ enum { ret = false }; };
+
+template<typename T> struct is_diagonal<DiagonalBase<T> >
+{ enum { ret = true }; };
+
+template<typename T> struct is_diagonal<DiagonalWrapper<T> >
+{ enum { ret = true }; };
+
+template<typename T, int S> struct is_diagonal<DiagonalMatrix<T,S> >
+{ enum { ret = true }; };
+
+template<typename S1, typename S2> struct glue_shapes;
+template<> struct glue_shapes<DenseShape,TriangularShape> { typedef TriangularShape type;  };
+
+template<typename T1, typename T2>
+bool is_same_dense(const T1 &mat1, const T2 &mat2, typename enable_if<has_direct_access<T1>::ret&&has_direct_access<T2>::ret, T1>::type * = 0)
+{
+  return (mat1.data()==mat2.data()) && (mat1.innerStride()==mat2.innerStride()) && (mat1.outerStride()==mat2.outerStride());
+}
+
+template<typename T1, typename T2>
+bool is_same_dense(const T1 &, const T2 &, typename enable_if<!(has_direct_access<T1>::ret&&has_direct_access<T2>::ret), T1>::type * = 0)
+{
+  return false;
+}
+
+// Internal helper defining the cost of a scalar division for the type T.
+// The default heuristic can be specialized for each scalar type and architecture.
+template<typename T,bool Vectorized=false,typename EnaleIf = void>
+struct scalar_div_cost {
+  enum { value = 8*NumTraits<T>::MulCost };
+};
+
+template<typename T,bool Vectorized>
+struct scalar_div_cost<std::complex<T>, Vectorized> {
+  enum { value = 2*scalar_div_cost<T>::value
+               + 6*NumTraits<T>::MulCost
+               + 3*NumTraits<T>::AddCost
+  };
+};
+
+
+template<bool Vectorized>
+struct scalar_div_cost<signed long,Vectorized,typename conditional<sizeof(long)==8,void,false_type>::type> { enum { value = 24 }; };
+template<bool Vectorized>
+struct scalar_div_cost<unsigned long,Vectorized,typename conditional<sizeof(long)==8,void,false_type>::type> { enum { value = 21 }; };
+
+
+#ifdef EIGEN_DEBUG_ASSIGN
+std::string demangle_traversal(int t)
+{
+  if(t==DefaultTraversal) return "DefaultTraversal";
+  if(t==LinearTraversal) return "LinearTraversal";
+  if(t==InnerVectorizedTraversal) return "InnerVectorizedTraversal";
+  if(t==LinearVectorizedTraversal) return "LinearVectorizedTraversal";
+  if(t==SliceVectorizedTraversal) return "SliceVectorizedTraversal";
+  return "?";
+}
+std::string demangle_unrolling(int t)
+{
+  if(t==NoUnrolling) return "NoUnrolling";
+  if(t==InnerUnrolling) return "InnerUnrolling";
+  if(t==CompleteUnrolling) return "CompleteUnrolling";
+  return "?";
+}
+std::string demangle_flags(int f)
+{
+  std::string res;
+  if(f&RowMajorBit)                 res += " | RowMajor";
+  if(f&PacketAccessBit)             res += " | Packet";
+  if(f&LinearAccessBit)             res += " | Linear";
+  if(f&LvalueBit)                   res += " | Lvalue";
+  if(f&DirectAccessBit)             res += " | Direct";
+  if(f&NestByRefBit)                res += " | NestByRef";
+  if(f&NoPreferredStorageOrderBit)  res += " | NoPreferredStorageOrderBit";
+  
+  return res;
+}
+#endif
+
 } // end namespace internal
 
+
+/** \class ScalarBinaryOpTraits
+  * \ingroup Core_Module
+  *
+  * \brief Determines whether the given binary operation of two numeric types is allowed and what the scalar return type is.
+  *
+  * This class permits to control the scalar return type of any binary operation performed on two different scalar types through (partial) template specializations.
+  *
+  * For instance, let \c U1, \c U2 and \c U3 be three user defined scalar types for which most operations between instances of \c U1 and \c U2 returns an \c U3.
+  * You can let %Eigen knows that by defining:
+    \code
+    template<typename BinaryOp>
+    struct ScalarBinaryOpTraits<U1,U2,BinaryOp> { typedef U3 ReturnType;  };
+    template<typename BinaryOp>
+    struct ScalarBinaryOpTraits<U2,U1,BinaryOp> { typedef U3 ReturnType;  };
+    \endcode
+  * You can then explicitly disable some particular operations to get more explicit error messages:
+    \code
+    template<>
+    struct ScalarBinaryOpTraits<U1,U2,internal::scalar_max_op<U1,U2> > {};
+    \endcode
+  * Or customize the return type for individual operation:
+    \code
+    template<>
+    struct ScalarBinaryOpTraits<U1,U2,internal::scalar_sum_op<U1,U2> > { typedef U1 ReturnType; };
+    \endcode
+  *
+  * By default, the following generic combinations are supported:
+  <table class="manual">
+  <tr><th>ScalarA</th><th>ScalarB</th><th>BinaryOp</th><th>ReturnType</th><th>Note</th></tr>
+  <tr            ><td>\c T </td><td>\c T </td><td>\c * </td><td>\c T </td><td></td></tr>
+  <tr class="alt"><td>\c NumTraits<T>::Real </td><td>\c T </td><td>\c * </td><td>\c T </td><td>Only if \c NumTraits<T>::IsComplex </td></tr>
+  <tr            ><td>\c T </td><td>\c NumTraits<T>::Real </td><td>\c * </td><td>\c T </td><td>Only if \c NumTraits<T>::IsComplex </td></tr>
+  </table>
+  *
+  * \sa CwiseBinaryOp
+  */
+template<typename ScalarA, typename ScalarB, typename BinaryOp=internal::scalar_product_op<ScalarA,ScalarB> >
+struct ScalarBinaryOpTraits
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+  // for backward compatibility, use the hints given by the (deprecated) internal::scalar_product_traits class.
+  : internal::scalar_product_traits<ScalarA,ScalarB>
+#endif // EIGEN_PARSED_BY_DOXYGEN
+{};
+
+template<typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<T,T,BinaryOp>
+{
+  typedef T ReturnType;
+};
+
+template <typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<T, typename NumTraits<typename internal::enable_if<NumTraits<T>::IsComplex,T>::type>::Real, BinaryOp>
+{
+  typedef T ReturnType;
+};
+template <typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<typename NumTraits<typename internal::enable_if<NumTraits<T>::IsComplex,T>::type>::Real, T, BinaryOp>
+{
+  typedef T ReturnType;
+};
+
+// For Matrix * Permutation
+template<typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<T,void,BinaryOp>
+{
+  typedef T ReturnType;
+};
+
+// For Permutation * Matrix
+template<typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<void,T,BinaryOp>
+{
+  typedef T ReturnType;
+};
+
+// for Permutation*Permutation
+template<typename BinaryOp>
+struct ScalarBinaryOpTraits<void,void,BinaryOp>
+{
+  typedef void ReturnType;
+};
+
+// We require Lhs and Rhs to have "compatible" scalar types.
+// It is tempting to always allow mixing different types but remember that this is often impossible in the vectorized paths.
+// So allowing mixing different types gives very unexpected errors when enabling vectorization, when the user tries to
+// add together a float matrix and a double matrix.
+#define EIGEN_CHECK_BINARY_COMPATIBILIY(BINOP,LHS,RHS) \
+  EIGEN_STATIC_ASSERT((Eigen::internal::has_ReturnType<ScalarBinaryOpTraits<LHS, RHS,BINOP> >::value), \
+    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+    
 } // end namespace Eigen
 
 #endif // EIGEN_XPRHELPER_H
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/ComplexEigenSolver.h b/extern/Eigen3/Eigen/src/Eigenvalues/ComplexEigenSolver.h
index 417c72944e1..dc5fae06a3c 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/ComplexEigenSolver.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/ComplexEigenSolver.h
@@ -60,7 +60,7 @@ template<typename _MatrixType> class ComplexEigenSolver
     /** \brief Scalar type for matrices of type #MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Complex scalar type for #MatrixType.
       *
@@ -104,7 +104,7 @@ template<typename _MatrixType> class ComplexEigenSolver
       * according to the specified problem \a size.
       * \sa ComplexEigenSolver()
       */
-    ComplexEigenSolver(Index size)
+    explicit ComplexEigenSolver(Index size)
             : m_eivec(size, size),
               m_eivalues(size),
               m_schur(size),
@@ -122,7 +122,8 @@ template<typename _MatrixType> class ComplexEigenSolver
       *
       * This constructor calls compute() to compute the eigendecomposition.
       */
-      ComplexEigenSolver(const MatrixType& matrix, bool computeEigenvectors = true)
+    template<typename InputType>
+    explicit ComplexEigenSolver(const EigenBase<InputType>& matrix, bool computeEigenvectors = true)
             : m_eivec(matrix.rows(),matrix.cols()),
               m_eivalues(matrix.cols()),
               m_schur(matrix.rows()),
@@ -130,7 +131,7 @@ template<typename _MatrixType> class ComplexEigenSolver
               m_eigenvectorsOk(false),
               m_matX(matrix.rows(),matrix.cols())
     {
-      compute(matrix, computeEigenvectors);
+      compute(matrix.derived(), computeEigenvectors);
     }
 
     /** \brief Returns the eigenvectors of given matrix.
@@ -208,7 +209,8 @@ template<typename _MatrixType> class ComplexEigenSolver
       * Example: \include ComplexEigenSolver_compute.cpp
       * Output: \verbinclude ComplexEigenSolver_compute.out
       */
-    ComplexEigenSolver& compute(const MatrixType& matrix, bool computeEigenvectors = true);
+    template<typename InputType>
+    ComplexEigenSolver& compute(const EigenBase<InputType>& matrix, bool computeEigenvectors = true);
 
     /** \brief Reports whether previous computation was successful.
       *
@@ -248,14 +250,15 @@ template<typename _MatrixType> class ComplexEigenSolver
     EigenvectorType m_matX;
 
   private:
-    void doComputeEigenvectors(const RealScalar& matrixnorm);
+    void doComputeEigenvectors(RealScalar matrixnorm);
     void sortEigenvalues(bool computeEigenvectors);
 };
 
 
 template<typename MatrixType>
+template<typename InputType>
 ComplexEigenSolver<MatrixType>& 
-ComplexEigenSolver<MatrixType>::compute(const MatrixType& matrix, bool computeEigenvectors)
+ComplexEigenSolver<MatrixType>::compute(const EigenBase<InputType>& matrix, bool computeEigenvectors)
 {
   check_template_parameters();
   
@@ -264,13 +267,13 @@ ComplexEigenSolver<MatrixType>::compute(const MatrixType& matrix, bool computeEi
 
   // Do a complex Schur decomposition, A = U T U^*
   // The eigenvalues are on the diagonal of T.
-  m_schur.compute(matrix, computeEigenvectors);
+  m_schur.compute(matrix.derived(), computeEigenvectors);
 
   if(m_schur.info() == Success)
   {
     m_eivalues = m_schur.matrixT().diagonal();
     if(computeEigenvectors)
-      doComputeEigenvectors(matrix.norm());
+      doComputeEigenvectors(m_schur.matrixT().norm());
     sortEigenvalues(computeEigenvectors);
   }
 
@@ -281,10 +284,12 @@ ComplexEigenSolver<MatrixType>::compute(const MatrixType& matrix, bool computeEi
 
 
 template<typename MatrixType>
-void ComplexEigenSolver<MatrixType>::doComputeEigenvectors(const RealScalar& matrixnorm)
+void ComplexEigenSolver<MatrixType>::doComputeEigenvectors(RealScalar matrixnorm)
 {
   const Index n = m_eivalues.size();
 
+  matrixnorm = numext::maxi(matrixnorm,(std::numeric_limits<RealScalar>::min)());
+
   // Compute X such that T = X D X^(-1), where D is the diagonal of T.
   // The matrix X is unit triangular.
   m_matX = EigenvectorType::Zero(n, n);
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/ComplexSchur.h b/extern/Eigen3/Eigen/src/Eigenvalues/ComplexSchur.h
index 89e6cade334..7f38919f778 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/ComplexSchur.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/ComplexSchur.h
@@ -63,7 +63,7 @@ template<typename _MatrixType> class ComplexSchur
     /** \brief Scalar type for matrices of type \p _MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Complex scalar type for \p _MatrixType. 
       *
@@ -91,7 +91,7 @@ template<typename _MatrixType> class ComplexSchur
       *
       * \sa compute() for an example.
       */
-    ComplexSchur(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime)
+    explicit ComplexSchur(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime)
       : m_matT(size,size),
         m_matU(size,size),
         m_hess(size),
@@ -109,7 +109,8 @@ template<typename _MatrixType> class ComplexSchur
       *
       * \sa matrixT() and matrixU() for examples.
       */
-    ComplexSchur(const MatrixType& matrix, bool computeU = true)
+    template<typename InputType>
+    explicit ComplexSchur(const EigenBase<InputType>& matrix, bool computeU = true)
       : m_matT(matrix.rows(),matrix.cols()),
         m_matU(matrix.rows(),matrix.cols()),
         m_hess(matrix.rows()),
@@ -117,7 +118,7 @@ template<typename _MatrixType> class ComplexSchur
         m_matUisUptodate(false),
         m_maxIters(-1)
     {
-      compute(matrix, computeU);
+      compute(matrix.derived(), computeU);
     }
 
     /** \brief Returns the unitary matrix in the Schur decomposition. 
@@ -186,7 +187,8 @@ template<typename _MatrixType> class ComplexSchur
       *
       * \sa compute(const MatrixType&, bool, Index)
       */
-    ComplexSchur& compute(const MatrixType& matrix, bool computeU = true);
+    template<typename InputType>
+    ComplexSchur& compute(const EigenBase<InputType>& matrix, bool computeU = true);
     
     /** \brief Compute Schur decomposition from a given Hessenberg matrix
      *  \param[in] matrixH Matrix in Hessenberg form H
@@ -313,14 +315,15 @@ typename ComplexSchur<MatrixType>::ComplexScalar ComplexSchur<MatrixType>::compu
 
 
 template<typename MatrixType>
-ComplexSchur<MatrixType>& ComplexSchur<MatrixType>::compute(const MatrixType& matrix, bool computeU)
+template<typename InputType>
+ComplexSchur<MatrixType>& ComplexSchur<MatrixType>::compute(const EigenBase<InputType>& matrix, bool computeU)
 {
   m_matUisUptodate = false;
   eigen_assert(matrix.cols() == matrix.rows());
 
   if(matrix.cols() == 1)
   {
-    m_matT = matrix.template cast<ComplexScalar>();
+    m_matT = matrix.derived().template cast<ComplexScalar>();
     if(computeU)  m_matU = ComplexMatrixType::Identity(1,1);
     m_info = Success;
     m_isInitialized = true;
@@ -328,7 +331,7 @@ ComplexSchur<MatrixType>& ComplexSchur<MatrixType>::compute(const MatrixType& ma
     return *this;
   }
 
-  internal::complex_schur_reduce_to_hessenberg<MatrixType, NumTraits<Scalar>::IsComplex>::run(*this, matrix, computeU);
+  internal::complex_schur_reduce_to_hessenberg<MatrixType, NumTraits<Scalar>::IsComplex>::run(*this, matrix.derived(), computeU);
   computeFromHessenberg(m_matT, m_matU, computeU);
   return *this;
 }
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/ComplexSchur_LAPACKE.h b/extern/Eigen3/Eigen/src/Eigenvalues/ComplexSchur_LAPACKE.h
new file mode 100644
index 00000000000..4980a3ede0a
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/ComplexSchur_LAPACKE.h
@@ -0,0 +1,91 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *    Complex Schur needed to complex unsymmetrical eigenvalues/eigenvectors.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_COMPLEX_SCHUR_LAPACKE_H
+#define EIGEN_COMPLEX_SCHUR_LAPACKE_H
+
+namespace Eigen { 
+
+/** \internal Specialization for the data types supported by LAPACKe */
+
+#define EIGEN_LAPACKE_SCHUR_COMPLEX(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX, LAPACKE_PREFIX_U, EIGCOLROW, LAPACKE_COLROW) \
+template<> template<typename InputType> inline \
+ComplexSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >& \
+ComplexSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >::compute(const EigenBase<InputType>& matrix, bool computeU) \
+{ \
+  typedef Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> MatrixType; \
+  typedef MatrixType::RealScalar RealScalar; \
+  typedef std::complex<RealScalar> ComplexScalar; \
+\
+  eigen_assert(matrix.cols() == matrix.rows()); \
+\
+  m_matUisUptodate = false; \
+  if(matrix.cols() == 1) \
+  { \
+    m_matT = matrix.derived().template cast<ComplexScalar>(); \
+    if(computeU)  m_matU = ComplexMatrixType::Identity(1,1); \
+      m_info = Success; \
+      m_isInitialized = true; \
+      m_matUisUptodate = computeU; \
+      return *this; \
+  } \
+  lapack_int n = internal::convert_index<lapack_int>(matrix.cols()), sdim, info; \
+  lapack_int matrix_order = LAPACKE_COLROW; \
+  char jobvs, sort='N'; \
+  LAPACK_##LAPACKE_PREFIX_U##_SELECT1 select = 0; \
+  jobvs = (computeU) ? 'V' : 'N'; \
+  m_matU.resize(n, n); \
+  lapack_int ldvs  = internal::convert_index<lapack_int>(m_matU.outerStride()); \
+  m_matT = matrix; \
+  lapack_int lda = internal::convert_index<lapack_int>(m_matT.outerStride()); \
+  Matrix<EIGTYPE, Dynamic, Dynamic> w; \
+  w.resize(n, 1);\
+  info = LAPACKE_##LAPACKE_PREFIX##gees( matrix_order, jobvs, sort, select, n, (LAPACKE_TYPE*)m_matT.data(), lda, &sdim, (LAPACKE_TYPE*)w.data(), (LAPACKE_TYPE*)m_matU.data(), ldvs ); \
+  if(info == 0) \
+    m_info = Success; \
+  else \
+    m_info = NoConvergence; \
+\
+  m_isInitialized = true; \
+  m_matUisUptodate = computeU; \
+  return *this; \
+\
+}
+
+EIGEN_LAPACKE_SCHUR_COMPLEX(dcomplex, lapack_complex_double, z, Z, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SCHUR_COMPLEX(scomplex, lapack_complex_float,  c, C, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SCHUR_COMPLEX(dcomplex, lapack_complex_double, z, Z, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SCHUR_COMPLEX(scomplex, lapack_complex_float,  c, C, RowMajor, LAPACK_ROW_MAJOR)
+
+} // end namespace Eigen
+
+#endif // EIGEN_COMPLEX_SCHUR_LAPACKE_H
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/EigenSolver.h b/extern/Eigen3/Eigen/src/Eigenvalues/EigenSolver.h
index 20c59a7a2e6..f205b185dee 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/EigenSolver.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/EigenSolver.h
@@ -79,7 +79,7 @@ template<typename _MatrixType> class EigenSolver
     /** \brief Scalar type for matrices of type #MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Complex scalar type for #MatrixType. 
       *
@@ -110,7 +110,7 @@ template<typename _MatrixType> class EigenSolver
       *
       * \sa compute() for an example.
       */
- EigenSolver() : m_eivec(), m_eivalues(), m_isInitialized(false), m_realSchur(), m_matT(), m_tmp() {}
+    EigenSolver() : m_eivec(), m_eivalues(), m_isInitialized(false), m_realSchur(), m_matT(), m_tmp() {}
 
     /** \brief Default constructor with memory preallocation
       *
@@ -118,7 +118,7 @@ template<typename _MatrixType> class EigenSolver
       * according to the specified problem \a size.
       * \sa EigenSolver()
       */
-    EigenSolver(Index size)
+    explicit EigenSolver(Index size)
       : m_eivec(size, size),
         m_eivalues(size),
         m_isInitialized(false),
@@ -143,7 +143,8 @@ template<typename _MatrixType> class EigenSolver
       *
       * \sa compute()
       */
-    EigenSolver(const MatrixType& matrix, bool computeEigenvectors = true)
+    template<typename InputType>
+    explicit EigenSolver(const EigenBase<InputType>& matrix, bool computeEigenvectors = true)
       : m_eivec(matrix.rows(), matrix.cols()),
         m_eivalues(matrix.cols()),
         m_isInitialized(false),
@@ -152,7 +153,7 @@ template<typename _MatrixType> class EigenSolver
         m_matT(matrix.rows(), matrix.cols()), 
         m_tmp(matrix.cols())
     {
-      compute(matrix, computeEigenvectors);
+      compute(matrix.derived(), computeEigenvectors);
     }
 
     /** \brief Returns the eigenvectors of given matrix. 
@@ -273,12 +274,14 @@ template<typename _MatrixType> class EigenSolver
       * Example: \include EigenSolver_compute.cpp
       * Output: \verbinclude EigenSolver_compute.out
       */
-    EigenSolver& compute(const MatrixType& matrix, bool computeEigenvectors = true);
+    template<typename InputType>
+    EigenSolver& compute(const EigenBase<InputType>& matrix, bool computeEigenvectors = true);
 
+    /** \returns NumericalIssue if the input contains INF or NaN values or overflow occured. Returns Success otherwise. */
     ComputationInfo info() const
     {
       eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
-      return m_realSchur.info();
+      return m_info;
     }
 
     /** \brief Sets the maximum number of iterations allowed. */
@@ -309,6 +312,7 @@ template<typename _MatrixType> class EigenSolver
     EigenvalueType m_eivalues;
     bool m_isInitialized;
     bool m_eigenvectorsOk;
+    ComputationInfo m_info;
     RealSchur<MatrixType> m_realSchur;
     MatrixType m_matT;
 
@@ -320,11 +324,12 @@ template<typename MatrixType>
 MatrixType EigenSolver<MatrixType>::pseudoEigenvalueMatrix() const
 {
   eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
+  const RealScalar precision = RealScalar(2)*NumTraits<RealScalar>::epsilon();
   Index n = m_eivalues.rows();
   MatrixType matD = MatrixType::Zero(n,n);
   for (Index i=0; i<n; ++i)
   {
-    if (internal::isMuchSmallerThan(numext::imag(m_eivalues.coeff(i)), numext::real(m_eivalues.coeff(i))))
+    if (internal::isMuchSmallerThan(numext::imag(m_eivalues.coeff(i)), numext::real(m_eivalues.coeff(i)), precision))
       matD.coeffRef(i,i) = numext::real(m_eivalues.coeff(i));
     else
     {
@@ -341,11 +346,12 @@ typename EigenSolver<MatrixType>::EigenvectorsType EigenSolver<MatrixType>::eige
 {
   eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
   eigen_assert(m_eigenvectorsOk && "The eigenvectors have not been computed together with the eigenvalues.");
+  const RealScalar precision = RealScalar(2)*NumTraits<RealScalar>::epsilon();
   Index n = m_eivec.cols();
   EigenvectorsType matV(n,n);
   for (Index j=0; j<n; ++j)
   {
-    if (internal::isMuchSmallerThan(numext::imag(m_eivalues.coeff(j)), numext::real(m_eivalues.coeff(j))) || j+1==n)
+    if (internal::isMuchSmallerThan(numext::imag(m_eivalues.coeff(j)), numext::real(m_eivalues.coeff(j)), precision) || j+1==n)
     {
       // we have a real eigen value
       matV.col(j) = m_eivec.col(j).template cast<ComplexScalar>();
@@ -368,19 +374,23 @@ typename EigenSolver<MatrixType>::EigenvectorsType EigenSolver<MatrixType>::eige
 }
 
 template<typename MatrixType>
+template<typename InputType>
 EigenSolver<MatrixType>& 
-EigenSolver<MatrixType>::compute(const MatrixType& matrix, bool computeEigenvectors)
+EigenSolver<MatrixType>::compute(const EigenBase<InputType>& matrix, bool computeEigenvectors)
 {
   check_template_parameters();
   
   using std::sqrt;
   using std::abs;
+  using numext::isfinite;
   eigen_assert(matrix.cols() == matrix.rows());
 
   // Reduce to real Schur form.
-  m_realSchur.compute(matrix, computeEigenvectors);
+  m_realSchur.compute(matrix.derived(), computeEigenvectors);
+  
+  m_info = m_realSchur.info();
 
-  if (m_realSchur.info() == Success)
+  if (m_info == Success)
   {
     m_matT = m_realSchur.matrixT();
     if (computeEigenvectors)
@@ -394,14 +404,40 @@ EigenSolver<MatrixType>::compute(const MatrixType& matrix, bool computeEigenvect
       if (i == matrix.cols() - 1 || m_matT.coeff(i+1, i) == Scalar(0)) 
       {
         m_eivalues.coeffRef(i) = m_matT.coeff(i, i);
+        if(!(isfinite)(m_eivalues.coeffRef(i)))
+        {
+          m_isInitialized = true;
+          m_eigenvectorsOk = false;
+          m_info = NumericalIssue;
+          return *this;
+        }
         ++i;
       }
       else
       {
         Scalar p = Scalar(0.5) * (m_matT.coeff(i, i) - m_matT.coeff(i+1, i+1));
-        Scalar z = sqrt(abs(p * p + m_matT.coeff(i+1, i) * m_matT.coeff(i, i+1)));
+        Scalar z;
+        // Compute z = sqrt(abs(p * p + m_matT.coeff(i+1, i) * m_matT.coeff(i, i+1)));
+        // without overflow
+        {
+          Scalar t0 = m_matT.coeff(i+1, i);
+          Scalar t1 = m_matT.coeff(i, i+1);
+          Scalar maxval = numext::maxi<Scalar>(abs(p),numext::maxi<Scalar>(abs(t0),abs(t1)));
+          t0 /= maxval;
+          t1 /= maxval;
+          Scalar p0 = p/maxval;
+          z = maxval * sqrt(abs(p0 * p0 + t0 * t1));
+        }
+        
         m_eivalues.coeffRef(i)   = ComplexScalar(m_matT.coeff(i+1, i+1) + p, z);
         m_eivalues.coeffRef(i+1) = ComplexScalar(m_matT.coeff(i+1, i+1) + p, -z);
+        if(!((isfinite)(m_eivalues.coeffRef(i)) && (isfinite)(m_eivalues.coeffRef(i+1))))
+        {
+          m_isInitialized = true;
+          m_eigenvectorsOk = false;
+          m_info = NumericalIssue;
+          return *this;
+        }
         i += 2;
       }
     }
@@ -417,26 +453,6 @@ EigenSolver<MatrixType>::compute(const MatrixType& matrix, bool computeEigenvect
   return *this;
 }
 
-// Complex scalar division.
-template<typename Scalar>
-std::complex<Scalar> cdiv(const Scalar& xr, const Scalar& xi, const Scalar& yr, const Scalar& yi)
-{
-  using std::abs;
-  Scalar r,d;
-  if (abs(yr) > abs(yi))
-  {
-      r = yi/yr;
-      d = yr + r*yi;
-      return std::complex<Scalar>((xr + r*xi)/d, (xi - r*xr)/d);
-  }
-  else
-  {
-      r = yr/yi;
-      d = yi + r*yr;
-      return std::complex<Scalar>((r*xr + xi)/d, (r*xi - xr)/d);
-  }
-}
-
 
 template<typename MatrixType>
 void EigenSolver<MatrixType>::doComputeEigenvectors()
@@ -453,7 +469,7 @@ void EigenSolver<MatrixType>::doComputeEigenvectors()
   }
   
   // Backsubstitute to find vectors of upper triangular form
-  if (norm == 0.0)
+  if (norm == Scalar(0))
   {
     return;
   }
@@ -469,13 +485,13 @@ void EigenSolver<MatrixType>::doComputeEigenvectors()
       Scalar lastr(0), lastw(0);
       Index l = n;
 
-      m_matT.coeffRef(n,n) = 1.0;
+      m_matT.coeffRef(n,n) = Scalar(1);
       for (Index i = n-1; i >= 0; i--)
       {
         Scalar w = m_matT.coeff(i,i) - p;
         Scalar r = m_matT.row(i).segment(l,n-l+1).dot(m_matT.col(n).segment(l, n-l+1));
 
-        if (m_eivalues.coeff(i).imag() < 0.0)
+        if (m_eivalues.coeff(i).imag() < Scalar(0))
         {
           lastw = w;
           lastr = r;
@@ -483,9 +499,9 @@ void EigenSolver<MatrixType>::doComputeEigenvectors()
         else
         {
           l = i;
-          if (m_eivalues.coeff(i).imag() == 0.0)
+          if (m_eivalues.coeff(i).imag() == Scalar(0))
           {
-            if (w != 0.0)
+            if (w != Scalar(0))
               m_matT.coeffRef(i,n) = -r / w;
             else
               m_matT.coeffRef(i,n) = -r / (eps * norm);
@@ -523,19 +539,19 @@ void EigenSolver<MatrixType>::doComputeEigenvectors()
       }
       else
       {
-        std::complex<Scalar> cc = cdiv<Scalar>(0.0,-m_matT.coeff(n-1,n),m_matT.coeff(n-1,n-1)-p,q);
+        ComplexScalar cc = ComplexScalar(Scalar(0),-m_matT.coeff(n-1,n)) / ComplexScalar(m_matT.coeff(n-1,n-1)-p,q);
         m_matT.coeffRef(n-1,n-1) = numext::real(cc);
         m_matT.coeffRef(n-1,n) = numext::imag(cc);
       }
-      m_matT.coeffRef(n,n-1) = 0.0;
-      m_matT.coeffRef(n,n) = 1.0;
+      m_matT.coeffRef(n,n-1) = Scalar(0);
+      m_matT.coeffRef(n,n) = Scalar(1);
       for (Index i = n-2; i >= 0; i--)
       {
         Scalar ra = m_matT.row(i).segment(l, n-l+1).dot(m_matT.col(n-1).segment(l, n-l+1));
         Scalar sa = m_matT.row(i).segment(l, n-l+1).dot(m_matT.col(n).segment(l, n-l+1));
         Scalar w = m_matT.coeff(i,i) - p;
 
-        if (m_eivalues.coeff(i).imag() < 0.0)
+        if (m_eivalues.coeff(i).imag() < Scalar(0))
         {
           lastw = w;
           lastra = ra;
@@ -546,7 +562,7 @@ void EigenSolver<MatrixType>::doComputeEigenvectors()
           l = i;
           if (m_eivalues.coeff(i).imag() == RealScalar(0))
           {
-            std::complex<Scalar> cc = cdiv(-ra,-sa,w,q);
+            ComplexScalar cc = ComplexScalar(-ra,-sa) / ComplexScalar(w,q);
             m_matT.coeffRef(i,n-1) = numext::real(cc);
             m_matT.coeffRef(i,n) = numext::imag(cc);
           }
@@ -557,10 +573,10 @@ void EigenSolver<MatrixType>::doComputeEigenvectors()
             Scalar y = m_matT.coeff(i+1,i);
             Scalar vr = (m_eivalues.coeff(i).real() - p) * (m_eivalues.coeff(i).real() - p) + m_eivalues.coeff(i).imag() * m_eivalues.coeff(i).imag() - q * q;
             Scalar vi = (m_eivalues.coeff(i).real() - p) * Scalar(2) * q;
-            if ((vr == 0.0) && (vi == 0.0))
+            if ((vr == Scalar(0)) && (vi == Scalar(0)))
               vr = eps * norm * (abs(w) + abs(q) + abs(x) + abs(y) + abs(lastw));
 
-            std::complex<Scalar> cc = cdiv(x*lastra-lastw*ra+q*sa,x*lastsa-lastw*sa-q*ra,vr,vi);
+            ComplexScalar cc = ComplexScalar(x*lastra-lastw*ra+q*sa,x*lastsa-lastw*sa-q*ra) / ComplexScalar(vr,vi);
             m_matT.coeffRef(i,n-1) = numext::real(cc);
             m_matT.coeffRef(i,n) = numext::imag(cc);
             if (abs(x) > (abs(lastw) + abs(q)))
@@ -570,15 +586,14 @@ void EigenSolver<MatrixType>::doComputeEigenvectors()
             }
             else
             {
-              cc = cdiv(-lastra-y*m_matT.coeff(i,n-1),-lastsa-y*m_matT.coeff(i,n),lastw,q);
+              cc = ComplexScalar(-lastra-y*m_matT.coeff(i,n-1),-lastsa-y*m_matT.coeff(i,n)) / ComplexScalar(lastw,q);
               m_matT.coeffRef(i+1,n-1) = numext::real(cc);
               m_matT.coeffRef(i+1,n) = numext::imag(cc);
             }
           }
 
           // Overflow control
-          using std::max;
-          Scalar t = (max)(abs(m_matT.coeff(i,n-1)),abs(m_matT.coeff(i,n)));
+          Scalar t = numext::maxi<Scalar>(abs(m_matT.coeff(i,n-1)),abs(m_matT.coeff(i,n)));
           if ((eps * t) * t > Scalar(1))
             m_matT.block(i, n-1, size-i, 2) /= t;
 
@@ -590,7 +605,7 @@ void EigenSolver<MatrixType>::doComputeEigenvectors()
     }
     else
     {
-      eigen_assert(0 && "Internal bug in EigenSolver"); // this should not happen
+      eigen_assert(0 && "Internal bug in EigenSolver (INF or NaN has not been detected)"); // this should not happen
     }
   }
 
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/GeneralizedEigenSolver.h b/extern/Eigen3/Eigen/src/Eigenvalues/GeneralizedEigenSolver.h
index 956e80d9edc..87d789b3f4a 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/GeneralizedEigenSolver.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/GeneralizedEigenSolver.h
@@ -1,8 +1,9 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2012 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2012-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2010,2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
+// Copyright (C) 2016 Tobias Wood <tobias@spinicist.org.uk>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -72,7 +73,7 @@ template<typename _MatrixType> class GeneralizedEigenSolver
     /** \brief Scalar type for matrices of type #MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Complex scalar type for #MatrixType. 
       *
@@ -89,7 +90,7 @@ template<typename _MatrixType> class GeneralizedEigenSolver
       */
     typedef Matrix<Scalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> VectorType;
 
-    /** \brief Type for vector of complex scalar values eigenvalues as returned by betas().
+    /** \brief Type for vector of complex scalar values eigenvalues as returned by alphas().
       *
       * This is a column vector with entries of type #ComplexScalar.
       * The length of the vector is the size of #MatrixType.
@@ -114,7 +115,14 @@ template<typename _MatrixType> class GeneralizedEigenSolver
       *
       * \sa compute() for an example.
       */
-    GeneralizedEigenSolver() : m_eivec(), m_alphas(), m_betas(), m_isInitialized(false), m_realQZ(), m_matS(), m_tmp() {}
+    GeneralizedEigenSolver()
+      : m_eivec(),
+        m_alphas(),
+        m_betas(),
+        m_valuesOkay(false),
+        m_vectorsOkay(false),
+        m_realQZ()
+    {}
 
     /** \brief Default constructor with memory preallocation
       *
@@ -122,14 +130,13 @@ template<typename _MatrixType> class GeneralizedEigenSolver
       * according to the specified problem \a size.
       * \sa GeneralizedEigenSolver()
       */
-    GeneralizedEigenSolver(Index size)
+    explicit GeneralizedEigenSolver(Index size)
       : m_eivec(size, size),
         m_alphas(size),
         m_betas(size),
-        m_isInitialized(false),
-        m_eigenvectorsOk(false),
+        m_valuesOkay(false),
+        m_vectorsOkay(false),
         m_realQZ(size),
-        m_matS(size, size),
         m_tmp(size)
     {}
 
@@ -149,10 +156,9 @@ template<typename _MatrixType> class GeneralizedEigenSolver
       : m_eivec(A.rows(), A.cols()),
         m_alphas(A.cols()),
         m_betas(A.cols()),
-        m_isInitialized(false),
-        m_eigenvectorsOk(false),
+        m_valuesOkay(false),
+        m_vectorsOkay(false),
         m_realQZ(A.cols()),
-        m_matS(A.rows(), A.cols()),
         m_tmp(A.cols())
     {
       compute(A, B, computeEigenvectors);
@@ -160,22 +166,20 @@ template<typename _MatrixType> class GeneralizedEigenSolver
 
     /* \brief Returns the computed generalized eigenvectors.
       *
-      * \returns  %Matrix whose columns are the (possibly complex) eigenvectors.
+      * \returns  %Matrix whose columns are the (possibly complex) right eigenvectors.
+      * i.e. the eigenvectors that solve (A - l*B)x = 0. The ordering matches the eigenvalues.
       *
       * \pre Either the constructor 
       * GeneralizedEigenSolver(const MatrixType&,const MatrixType&, bool) or the member function
       * compute(const MatrixType&, const MatrixType& bool) has been called before, and
       * \p computeEigenvectors was set to true (the default).
       *
-      * Column \f$ k \f$ of the returned matrix is an eigenvector corresponding
-      * to eigenvalue number \f$ k \f$ as returned by eigenvalues().  The
-      * eigenvectors are normalized to have (Euclidean) norm equal to one. The
-      * matrix returned by this function is the matrix \f$ V \f$ in the
-      * generalized eigendecomposition \f$ A = B V D V^{-1} \f$, if it exists.
-      *
       * \sa eigenvalues()
       */
-//    EigenvectorsType eigenvectors() const;
+    EigenvectorsType eigenvectors() const {
+      eigen_assert(m_vectorsOkay && "Eigenvectors for GeneralizedEigenSolver were not calculated.");
+      return m_eivec;
+    }
 
     /** \brief Returns an expression of the computed generalized eigenvalues.
       *
@@ -197,7 +201,7 @@ template<typename _MatrixType> class GeneralizedEigenSolver
       */
     EigenvalueType eigenvalues() const
     {
-      eigen_assert(m_isInitialized && "GeneralizedEigenSolver is not initialized.");
+      eigen_assert(m_valuesOkay && "GeneralizedEigenSolver is not initialized.");
       return EigenvalueType(m_alphas,m_betas);
     }
 
@@ -208,7 +212,7 @@ template<typename _MatrixType> class GeneralizedEigenSolver
       * \sa betas(), eigenvalues() */
     ComplexVectorType alphas() const
     {
-      eigen_assert(m_isInitialized && "GeneralizedEigenSolver is not initialized.");
+      eigen_assert(m_valuesOkay && "GeneralizedEigenSolver is not initialized.");
       return m_alphas;
     }
 
@@ -219,7 +223,7 @@ template<typename _MatrixType> class GeneralizedEigenSolver
       * \sa alphas(), eigenvalues() */
     VectorType betas() const
     {
-      eigen_assert(m_isInitialized && "GeneralizedEigenSolver is not initialized.");
+      eigen_assert(m_valuesOkay && "GeneralizedEigenSolver is not initialized.");
       return m_betas;
     }
 
@@ -250,7 +254,7 @@ template<typename _MatrixType> class GeneralizedEigenSolver
 
     ComputationInfo info() const
     {
-      eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
+      eigen_assert(m_valuesOkay && "EigenSolver is not initialized.");
       return m_realQZ.info();
     }
 
@@ -270,29 +274,14 @@ template<typename _MatrixType> class GeneralizedEigenSolver
       EIGEN_STATIC_ASSERT(!NumTraits<Scalar>::IsComplex, NUMERIC_TYPE_MUST_BE_REAL);
     }
     
-    MatrixType m_eivec;
+    EigenvectorsType m_eivec;
     ComplexVectorType m_alphas;
     VectorType m_betas;
-    bool m_isInitialized;
-    bool m_eigenvectorsOk;
+    bool m_valuesOkay, m_vectorsOkay;
     RealQZ<MatrixType> m_realQZ;
-    MatrixType m_matS;
-
-    typedef Matrix<Scalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> ColumnVectorType;
-    ColumnVectorType m_tmp;
+    ComplexVectorType m_tmp;
 };
 
-//template<typename MatrixType>
-//typename GeneralizedEigenSolver<MatrixType>::EigenvectorsType GeneralizedEigenSolver<MatrixType>::eigenvectors() const
-//{
-//  eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
-//  eigen_assert(m_eigenvectorsOk && "The eigenvectors have not been computed together with the eigenvalues.");
-//  Index n = m_eivec.cols();
-//  EigenvectorsType matV(n,n);
-//  // TODO
-//  return matV;
-//}
-
 template<typename MatrixType>
 GeneralizedEigenSolver<MatrixType>&
 GeneralizedEigenSolver<MatrixType>::compute(const MatrixType& A, const MatrixType& B, bool computeEigenvectors)
@@ -302,46 +291,125 @@ GeneralizedEigenSolver<MatrixType>::compute(const MatrixType& A, const MatrixTyp
   using std::sqrt;
   using std::abs;
   eigen_assert(A.cols() == A.rows() && B.cols() == A.rows() && B.cols() == B.rows());
-
+  Index size = A.cols();
+  m_valuesOkay = false;
+  m_vectorsOkay = false;
   // Reduce to generalized real Schur form:
   // A = Q S Z and B = Q T Z
   m_realQZ.compute(A, B, computeEigenvectors);
-
   if (m_realQZ.info() == Success)
   {
-    m_matS = m_realQZ.matrixS();
+    // Resize storage
+    m_alphas.resize(size);
+    m_betas.resize(size);
     if (computeEigenvectors)
-      m_eivec = m_realQZ.matrixZ().transpose();
-  
-    // Compute eigenvalues from matS
-    m_alphas.resize(A.cols());
-    m_betas.resize(A.cols());
+    {
+      m_eivec.resize(size,size);
+      m_tmp.resize(size);
+    }
+
+    // Aliases:
+    Map<VectorType> v(reinterpret_cast<Scalar*>(m_tmp.data()), size);
+    ComplexVectorType &cv = m_tmp;
+    const MatrixType &mS = m_realQZ.matrixS();
+    const MatrixType &mT = m_realQZ.matrixT();
+
     Index i = 0;
-    while (i < A.cols())
+    while (i < size)
     {
-      if (i == A.cols() - 1 || m_matS.coeff(i+1, i) == Scalar(0))
+      if (i == size - 1 || mS.coeff(i+1, i) == Scalar(0))
       {
-        m_alphas.coeffRef(i) = m_matS.coeff(i, i);
-        m_betas.coeffRef(i)  = m_realQZ.matrixT().coeff(i,i);
+        // Real eigenvalue
+        m_alphas.coeffRef(i) = mS.diagonal().coeff(i);
+        m_betas.coeffRef(i)  = mT.diagonal().coeff(i);
+        if (computeEigenvectors)
+        {
+          v.setConstant(Scalar(0.0));
+          v.coeffRef(i) = Scalar(1.0);
+          // For singular eigenvalues do nothing more
+          if(abs(m_betas.coeffRef(i)) >= (std::numeric_limits<RealScalar>::min)())
+          {
+            // Non-singular eigenvalue
+            const Scalar alpha = real(m_alphas.coeffRef(i));
+            const Scalar beta = m_betas.coeffRef(i);
+            for (Index j = i-1; j >= 0; j--)
+            {
+              const Index st = j+1;
+              const Index sz = i-j;
+              if (j > 0 && mS.coeff(j, j-1) != Scalar(0))
+              {
+                // 2x2 block
+                Matrix<Scalar, 2, 1> rhs = (alpha*mT.template block<2,Dynamic>(j-1,st,2,sz) - beta*mS.template block<2,Dynamic>(j-1,st,2,sz)) .lazyProduct( v.segment(st,sz) );
+                Matrix<Scalar, 2, 2> lhs = beta * mS.template block<2,2>(j-1,j-1) - alpha * mT.template block<2,2>(j-1,j-1);
+                v.template segment<2>(j-1) = lhs.partialPivLu().solve(rhs);
+                j--;
+              }
+              else
+              {
+                v.coeffRef(j) = -v.segment(st,sz).transpose().cwiseProduct(beta*mS.block(j,st,1,sz) - alpha*mT.block(j,st,1,sz)).sum() / (beta*mS.coeffRef(j,j) - alpha*mT.coeffRef(j,j));
+              }
+            }
+          }
+          m_eivec.col(i).real().noalias() = m_realQZ.matrixZ().transpose() * v;
+          m_eivec.col(i).real().normalize();
+          m_eivec.col(i).imag().setConstant(0);
+        }
         ++i;
       }
       else
       {
-        Scalar p = Scalar(0.5) * (m_matS.coeff(i, i) - m_matS.coeff(i+1, i+1));
-        Scalar z = sqrt(abs(p * p + m_matS.coeff(i+1, i) * m_matS.coeff(i, i+1)));
-        m_alphas.coeffRef(i)   = ComplexScalar(m_matS.coeff(i+1, i+1) + p, z);
-        m_alphas.coeffRef(i+1) = ComplexScalar(m_matS.coeff(i+1, i+1) + p, -z);
-
-        m_betas.coeffRef(i)   = m_realQZ.matrixT().coeff(i,i);
-        m_betas.coeffRef(i+1) = m_realQZ.matrixT().coeff(i,i);
+        // We need to extract the generalized eigenvalues of the pair of a general 2x2 block S and a positive diagonal 2x2 block T
+        // Then taking beta=T_00*T_11, we can avoid any division, and alpha is the eigenvalues of A = (U^-1 * S * U) * diag(T_11,T_00):
+
+        // T =  [a 0]
+        //      [0 b]
+        RealScalar a = mT.diagonal().coeff(i),
+                   b = mT.diagonal().coeff(i+1);
+        const RealScalar beta = m_betas.coeffRef(i) = m_betas.coeffRef(i+1) = a*b;
+
+        // ^^ NOTE: using diagonal()(i) instead of coeff(i,i) workarounds a MSVC bug.
+        Matrix<RealScalar,2,2> S2 = mS.template block<2,2>(i,i) * Matrix<Scalar,2,1>(b,a).asDiagonal();
+
+        Scalar p = Scalar(0.5) * (S2.coeff(0,0) - S2.coeff(1,1));
+        Scalar z = sqrt(abs(p * p + S2.coeff(1,0) * S2.coeff(0,1)));
+        const ComplexScalar alpha = ComplexScalar(S2.coeff(1,1) + p, (beta > 0) ? z : -z);
+        m_alphas.coeffRef(i)   = conj(alpha);
+        m_alphas.coeffRef(i+1) = alpha;
+
+        if (computeEigenvectors) {
+          // Compute eigenvector in position (i+1) and then position (i) is just the conjugate
+          cv.setZero();
+          cv.coeffRef(i+1) = Scalar(1.0);
+          // here, the "static_cast" workaound expression template issues.
+          cv.coeffRef(i) = -(static_cast<Scalar>(beta*mS.coeffRef(i,i+1)) - alpha*mT.coeffRef(i,i+1))
+                          / (static_cast<Scalar>(beta*mS.coeffRef(i,i))   - alpha*mT.coeffRef(i,i));
+          for (Index j = i-1; j >= 0; j--)
+          {
+            const Index st = j+1;
+            const Index sz = i+1-j;
+            if (j > 0 && mS.coeff(j, j-1) != Scalar(0))
+            {
+              // 2x2 block
+              Matrix<ComplexScalar, 2, 1> rhs = (alpha*mT.template block<2,Dynamic>(j-1,st,2,sz) - beta*mS.template block<2,Dynamic>(j-1,st,2,sz)) .lazyProduct( cv.segment(st,sz) );
+              Matrix<ComplexScalar, 2, 2> lhs = beta * mS.template block<2,2>(j-1,j-1) - alpha * mT.template block<2,2>(j-1,j-1);
+              cv.template segment<2>(j-1) = lhs.partialPivLu().solve(rhs);
+              j--;
+            } else {
+              cv.coeffRef(j) =  cv.segment(st,sz).transpose().cwiseProduct(beta*mS.block(j,st,1,sz) - alpha*mT.block(j,st,1,sz)).sum()
+                              / (alpha*mT.coeffRef(j,j) - static_cast<Scalar>(beta*mS.coeffRef(j,j)));
+            }
+          }
+          m_eivec.col(i+1).noalias() = (m_realQZ.matrixZ().transpose() * cv);
+          m_eivec.col(i+1).normalize();
+          m_eivec.col(i) = m_eivec.col(i+1).conjugate();
+        }
         i += 2;
       }
     }
-  }
-
-  m_isInitialized = true;
-  m_eigenvectorsOk = false;//computeEigenvectors;
 
+    m_valuesOkay = true;
+    m_vectorsOkay = computeEigenvectors;
+  }
   return *this;
 }
 
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h b/extern/Eigen3/Eigen/src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h
index 07bf1ea0956..5f6bb82898b 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h
@@ -50,7 +50,6 @@ class GeneralizedSelfAdjointEigenSolver : public SelfAdjointEigenSolver<_MatrixT
     typedef SelfAdjointEigenSolver<_MatrixType> Base;
   public:
 
-    typedef typename Base::Index Index;
     typedef _MatrixType MatrixType;
 
     /** \brief Default constructor for fixed-size matrices.
@@ -74,7 +73,7 @@ class GeneralizedSelfAdjointEigenSolver : public SelfAdjointEigenSolver<_MatrixT
       *
       * \sa compute() for an example
       */
-    GeneralizedSelfAdjointEigenSolver(Index size)
+    explicit GeneralizedSelfAdjointEigenSolver(Index size)
         : Base(size)
     {}
 
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/HessenbergDecomposition.h b/extern/Eigen3/Eigen/src/Eigenvalues/HessenbergDecomposition.h
index 3db0c0106c9..f647f69b060 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/HessenbergDecomposition.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/HessenbergDecomposition.h
@@ -71,7 +71,7 @@ template<typename _MatrixType> class HessenbergDecomposition
 
     /** \brief Scalar type for matrices of type #MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Type for vector of Householder coefficients.
       *
@@ -97,7 +97,7 @@ template<typename _MatrixType> class HessenbergDecomposition
       *
       * \sa compute() for an example.
       */
-    HessenbergDecomposition(Index size = Size==Dynamic ? 2 : Size)
+    explicit HessenbergDecomposition(Index size = Size==Dynamic ? 2 : Size)
       : m_matrix(size,size),
         m_temp(size),
         m_isInitialized(false)
@@ -115,8 +115,9 @@ template<typename _MatrixType> class HessenbergDecomposition
       *
       * \sa matrixH() for an example.
       */
-    HessenbergDecomposition(const MatrixType& matrix)
-      : m_matrix(matrix),
+    template<typename InputType>
+    explicit HessenbergDecomposition(const EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
         m_temp(matrix.rows()),
         m_isInitialized(false)
     {
@@ -147,9 +148,10 @@ template<typename _MatrixType> class HessenbergDecomposition
       * Example: \include HessenbergDecomposition_compute.cpp
       * Output: \verbinclude HessenbergDecomposition_compute.out
       */
-    HessenbergDecomposition& compute(const MatrixType& matrix)
+    template<typename InputType>
+    HessenbergDecomposition& compute(const EigenBase<InputType>& matrix)
     {
-      m_matrix = matrix;
+      m_matrix = matrix.derived();
       if(matrix.rows()<2)
       {
         m_isInitialized = true;
@@ -337,7 +339,6 @@ namespace internal {
 template<typename MatrixType> struct HessenbergDecompositionMatrixHReturnType
 : public ReturnByValue<HessenbergDecompositionMatrixHReturnType<MatrixType> >
 {
-    typedef typename MatrixType::Index Index;
   public:
     /** \brief Constructor.
       *
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/MatrixBaseEigenvalues.h b/extern/Eigen3/Eigen/src/Eigenvalues/MatrixBaseEigenvalues.h
index 4fec8af0a3e..e4e42607113 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/MatrixBaseEigenvalues.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/MatrixBaseEigenvalues.h
@@ -66,7 +66,6 @@ template<typename Derived>
 inline typename MatrixBase<Derived>::EigenvaluesReturnType
 MatrixBase<Derived>::eigenvalues() const
 {
-  typedef typename internal::traits<Derived>::Scalar Scalar;
   return internal::eigenvalues_selector<Derived, NumTraits<Scalar>::IsComplex>::run(derived());
 }
 
@@ -88,7 +87,6 @@ template<typename MatrixType, unsigned int UpLo>
 inline typename SelfAdjointView<MatrixType, UpLo>::EigenvaluesReturnType
 SelfAdjointView<MatrixType, UpLo>::eigenvalues() const
 {
-  typedef typename SelfAdjointView<MatrixType, UpLo>::PlainObject PlainObject;
   PlainObject thisAsMatrix(*this);
   return SelfAdjointEigenSolver<PlainObject>(thisAsMatrix, false).eigenvalues();
 }
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/RealQZ.h b/extern/Eigen3/Eigen/src/Eigenvalues/RealQZ.h
index aa3833ebad1..b3a910dd9f3 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/RealQZ.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/RealQZ.h
@@ -67,7 +67,7 @@ namespace Eigen {
       };
       typedef typename MatrixType::Scalar Scalar;
       typedef std::complex<typename NumTraits<Scalar>::Real> ComplexScalar;
-      typedef typename MatrixType::Index Index;
+      typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
       typedef Matrix<ComplexScalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> EigenvalueType;
       typedef Matrix<Scalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> ColumnVectorType;
@@ -83,7 +83,7 @@ namespace Eigen {
        *
        * \sa compute() for an example.
        */
-      RealQZ(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime) : 
+      explicit RealQZ(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime) :
         m_S(size, size),
         m_T(size, size),
         m_Q(size, size),
@@ -276,7 +276,7 @@ namespace Eigen {
 
   /** \internal Look for single small sub-diagonal element S(res, res-1) and return res (or 0) */
   template<typename MatrixType>
-    inline typename MatrixType::Index RealQZ<MatrixType>::findSmallSubdiagEntry(Index iu)
+    inline Index RealQZ<MatrixType>::findSmallSubdiagEntry(Index iu)
     {
       using std::abs;
       Index res = iu;
@@ -294,7 +294,7 @@ namespace Eigen {
 
   /** \internal Look for single small diagonal element T(res, res) for res between f and l, and return res (or f-1)  */
   template<typename MatrixType>
-    inline typename MatrixType::Index RealQZ<MatrixType>::findSmallDiagEntry(Index f, Index l)
+    inline Index RealQZ<MatrixType>::findSmallDiagEntry(Index f, Index l)
     {
       using std::abs;
       Index res = l;
@@ -315,8 +315,8 @@ namespace Eigen {
       const Index dim=m_S.cols();
       if (abs(m_S.coeff(i+1,i))==Scalar(0))
         return;
-      Index z = findSmallDiagEntry(i,i+1);
-      if (z==i-1)
+      Index j = findSmallDiagEntry(i,i+1);
+      if (j==i-1)
       {
         // block of (S T^{-1})
         Matrix2s STi = m_T.template block<2,2>(i,i).template triangularView<Upper>().
@@ -352,7 +352,7 @@ namespace Eigen {
       }
       else
       {
-        pushDownZero(z,i,i+1);
+        pushDownZero(j,i,i+1);
       }
     }
 
@@ -552,7 +552,6 @@ namespace Eigen {
       m_T.coeffRef(l,l-1) = Scalar(0.0);
     }
 
-
   template<typename MatrixType>
     RealQZ<MatrixType>& RealQZ<MatrixType>::compute(const MatrixType& A_in, const MatrixType& B_in, bool computeQZ)
     {
@@ -616,6 +615,37 @@ namespace Eigen {
       }
       // check if we converged before reaching iterations limit
       m_info = (local_iter<m_maxIters) ? Success : NoConvergence;
+
+      // For each non triangular 2x2 diagonal block of S,
+      //    reduce the respective 2x2 diagonal block of T to positive diagonal form using 2x2 SVD.
+      // This step is not mandatory for QZ, but it does help further extraction of eigenvalues/eigenvectors,
+      // and is in par with Lapack/Matlab QZ.
+      if(m_info==Success)
+      {
+        for(Index i=0; i<dim-1; ++i)
+        {
+          if(m_S.coeff(i+1, i) != Scalar(0))
+          {
+            JacobiRotation<Scalar> j_left, j_right;
+            internal::real_2x2_jacobi_svd(m_T, i, i+1, &j_left, &j_right);
+
+            // Apply resulting Jacobi rotations
+            m_S.applyOnTheLeft(i,i+1,j_left);
+            m_S.applyOnTheRight(i,i+1,j_right);
+            m_T.applyOnTheLeft(i,i+1,j_left);
+            m_T.applyOnTheRight(i,i+1,j_right);
+            m_T(i+1,i) = m_T(i,i+1) = Scalar(0);
+
+            if(m_computeQZ) {
+              m_Q.applyOnTheRight(i,i+1,j_left.transpose());
+              m_Z.applyOnTheLeft(i,i+1,j_right.transpose());
+            }
+
+            i++;
+          }
+        }
+      }
+
       return *this;
     } // end compute
 
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/RealSchur.h b/extern/Eigen3/Eigen/src/Eigenvalues/RealSchur.h
index 16d3875372e..17ea903f5f1 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/RealSchur.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/RealSchur.h
@@ -64,7 +64,7 @@ template<typename _MatrixType> class RealSchur
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef std::complex<typename NumTraits<Scalar>::Real> ComplexScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     typedef Matrix<ComplexScalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> EigenvalueType;
     typedef Matrix<Scalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> ColumnVectorType;
@@ -80,7 +80,7 @@ template<typename _MatrixType> class RealSchur
       *
       * \sa compute() for an example.
       */
-    RealSchur(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime)
+    explicit RealSchur(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime)
             : m_matT(size, size),
               m_matU(size, size),
               m_workspaceVector(size),
@@ -100,7 +100,8 @@ template<typename _MatrixType> class RealSchur
       * Example: \include RealSchur_RealSchur_MatrixType.cpp
       * Output: \verbinclude RealSchur_RealSchur_MatrixType.out
       */
-    RealSchur(const MatrixType& matrix, bool computeU = true)
+    template<typename InputType>
+    explicit RealSchur(const EigenBase<InputType>& matrix, bool computeU = true)
             : m_matT(matrix.rows(),matrix.cols()),
               m_matU(matrix.rows(),matrix.cols()),
               m_workspaceVector(matrix.rows()),
@@ -109,7 +110,7 @@ template<typename _MatrixType> class RealSchur
               m_matUisUptodate(false),
               m_maxIters(-1)
     {
-      compute(matrix, computeU);
+      compute(matrix.derived(), computeU);
     }
 
     /** \brief Returns the orthogonal matrix in the Schur decomposition. 
@@ -165,7 +166,8 @@ template<typename _MatrixType> class RealSchur
       *
       * \sa compute(const MatrixType&, bool, Index)
       */
-    RealSchur& compute(const MatrixType& matrix, bool computeU = true);
+    template<typename InputType>
+    RealSchur& compute(const EigenBase<InputType>& matrix, bool computeU = true);
 
     /** \brief Computes Schur decomposition of a Hessenberg matrix H = Z T Z^T
      *  \param[in] matrixH Matrix in Hessenberg form H
@@ -243,26 +245,45 @@ template<typename _MatrixType> class RealSchur
 
 
 template<typename MatrixType>
-RealSchur<MatrixType>& RealSchur<MatrixType>::compute(const MatrixType& matrix, bool computeU)
+template<typename InputType>
+RealSchur<MatrixType>& RealSchur<MatrixType>::compute(const EigenBase<InputType>& matrix, bool computeU)
 {
+  const Scalar considerAsZero = (std::numeric_limits<Scalar>::min)();
+
   eigen_assert(matrix.cols() == matrix.rows());
   Index maxIters = m_maxIters;
   if (maxIters == -1)
     maxIters = m_maxIterationsPerRow * matrix.rows();
 
+  Scalar scale = matrix.derived().cwiseAbs().maxCoeff();
+  if(scale<considerAsZero)
+  {
+    m_matT.setZero(matrix.rows(),matrix.cols());
+    if(computeU)
+      m_matU.setIdentity(matrix.rows(),matrix.cols());
+    m_info = Success;
+    m_isInitialized = true;
+    m_matUisUptodate = computeU;
+    return *this;
+  }
+
   // Step 1. Reduce to Hessenberg form
-  m_hess.compute(matrix);
+  m_hess.compute(matrix.derived()/scale);
 
   // Step 2. Reduce to real Schur form  
   computeFromHessenberg(m_hess.matrixH(), m_hess.matrixQ(), computeU);
+
+  m_matT *= scale;
   
   return *this;
 }
 template<typename MatrixType>
 template<typename HessMatrixType, typename OrthMatrixType>
 RealSchur<MatrixType>& RealSchur<MatrixType>::computeFromHessenberg(const HessMatrixType& matrixH, const OrthMatrixType& matrixQ,  bool computeU)
-{  
-  m_matT = matrixH; 
+{
+  using std::abs;
+
+  m_matT = matrixH;
   if(computeU)
     m_matU = matrixQ;
   
@@ -282,7 +303,7 @@ RealSchur<MatrixType>& RealSchur<MatrixType>::computeFromHessenberg(const HessMa
   Scalar exshift(0);   // sum of exceptional shifts
   Scalar norm = computeNormOfT();
 
-  if(norm!=0)
+  if(norm!=Scalar(0))
   {
     while (iu >= 0)
     {
@@ -306,7 +327,7 @@ RealSchur<MatrixType>& RealSchur<MatrixType>::computeFromHessenberg(const HessMa
       else // No convergence yet
       {
         // The firstHouseholderVector vector has to be initialized to something to get rid of a silly GCC warning (-O1 -Wall -DNDEBUG )
-        Vector3s firstHouseholderVector(0,0,0), shiftInfo;
+        Vector3s firstHouseholderVector = Vector3s::Zero(), shiftInfo;
         computeShift(iu, iter, exshift, shiftInfo);
         iter = iter + 1;
         totalIter = totalIter + 1;
@@ -343,7 +364,7 @@ inline typename MatrixType::Scalar RealSchur<MatrixType>::computeNormOfT()
 
 /** \internal Look for single small sub-diagonal element and returns its index */
 template<typename MatrixType>
-inline typename MatrixType::Index RealSchur<MatrixType>::findSmallSubdiagEntry(Index iu)
+inline Index RealSchur<MatrixType>::findSmallSubdiagEntry(Index iu)
 {
   using std::abs;
   Index res = iu;
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/RealSchur_LAPACKE.h b/extern/Eigen3/Eigen/src/Eigenvalues/RealSchur_LAPACKE.h
new file mode 100644
index 00000000000..2c22517155e
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/RealSchur_LAPACKE.h
@@ -0,0 +1,77 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *    Real Schur needed to real unsymmetrical eigenvalues/eigenvectors.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_REAL_SCHUR_LAPACKE_H
+#define EIGEN_REAL_SCHUR_LAPACKE_H
+
+namespace Eigen { 
+
+/** \internal Specialization for the data types supported by LAPACKe */
+
+#define EIGEN_LAPACKE_SCHUR_REAL(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX, LAPACKE_PREFIX_U, EIGCOLROW, LAPACKE_COLROW) \
+template<> template<typename InputType> inline \
+RealSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >& \
+RealSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >::compute(const EigenBase<InputType>& matrix, bool computeU) \
+{ \
+  eigen_assert(matrix.cols() == matrix.rows()); \
+\
+  lapack_int n = internal::convert_index<lapack_int>(matrix.cols()), sdim, info; \
+  lapack_int matrix_order = LAPACKE_COLROW; \
+  char jobvs, sort='N'; \
+  LAPACK_##LAPACKE_PREFIX_U##_SELECT2 select = 0; \
+  jobvs = (computeU) ? 'V' : 'N'; \
+  m_matU.resize(n, n); \
+  lapack_int ldvs  = internal::convert_index<lapack_int>(m_matU.outerStride()); \
+  m_matT = matrix; \
+  lapack_int lda = internal::convert_index<lapack_int>(m_matT.outerStride()); \
+  Matrix<EIGTYPE, Dynamic, Dynamic> wr, wi; \
+  wr.resize(n, 1); wi.resize(n, 1); \
+  info = LAPACKE_##LAPACKE_PREFIX##gees( matrix_order, jobvs, sort, select, n, (LAPACKE_TYPE*)m_matT.data(), lda, &sdim, (LAPACKE_TYPE*)wr.data(), (LAPACKE_TYPE*)wi.data(), (LAPACKE_TYPE*)m_matU.data(), ldvs ); \
+  if(info == 0) \
+    m_info = Success; \
+  else \
+    m_info = NoConvergence; \
+\
+  m_isInitialized = true; \
+  m_matUisUptodate = computeU; \
+  return *this; \
+\
+}
+
+EIGEN_LAPACKE_SCHUR_REAL(double,   double, d, D, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SCHUR_REAL(float,    float,  s, S, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SCHUR_REAL(double,   double, d, D, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SCHUR_REAL(float,    float,  s, S, RowMajor, LAPACK_ROW_MAJOR)
+
+} // end namespace Eigen
+
+#endif // EIGEN_REAL_SCHUR_LAPACKE_H
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h b/extern/Eigen3/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h
index 1131c8af8ad..9ddd553f2f1 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h
@@ -20,6 +20,8 @@ class GeneralizedSelfAdjointEigenSolver;
 
 namespace internal {
 template<typename SolverType,int Size,bool IsComplex> struct direct_selfadjoint_eigenvalues;
+template<typename MatrixType, typename DiagType, typename SubDiagType>
+ComputationInfo computeFromTridiagonal_impl(DiagType& diag, SubDiagType& subdiag, const Index maxIterations, bool computeEigenvectors, MatrixType& eivec);
 }
 
 /** \eigenvalues_module \ingroup Eigenvalues_Module
@@ -79,7 +81,7 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
     
     /** \brief Scalar type for matrices of type \p _MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
     
     typedef Matrix<Scalar,Size,Size,ColMajor,MaxColsAtCompileTime,MaxColsAtCompileTime> EigenvectorsType;
 
@@ -100,6 +102,7 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       */
     typedef typename internal::plain_col_type<MatrixType, RealScalar>::type RealVectorType;
     typedef Tridiagonalization<MatrixType> TridiagonalizationType;
+    typedef typename TridiagonalizationType::SubDiagonalType SubDiagonalType;
 
     /** \brief Default constructor for fixed-size matrices.
       *
@@ -111,6 +114,7 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       * Example: \include SelfAdjointEigenSolver_SelfAdjointEigenSolver.cpp
       * Output: \verbinclude SelfAdjointEigenSolver_SelfAdjointEigenSolver.out
       */
+    EIGEN_DEVICE_FUNC
     SelfAdjointEigenSolver()
         : m_eivec(),
           m_eivalues(),
@@ -130,7 +134,8 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       *
       * \sa compute() for an example
       */
-    SelfAdjointEigenSolver(Index size)
+    EIGEN_DEVICE_FUNC
+    explicit SelfAdjointEigenSolver(Index size)
         : m_eivec(size, size),
           m_eivalues(size),
           m_subdiag(size > 1 ? size - 1 : 1),
@@ -152,13 +157,15 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       *
       * \sa compute(const MatrixType&, int)
       */
-    SelfAdjointEigenSolver(const MatrixType& matrix, int options = ComputeEigenvectors)
+    template<typename InputType>
+    EIGEN_DEVICE_FUNC
+    explicit SelfAdjointEigenSolver(const EigenBase<InputType>& matrix, int options = ComputeEigenvectors)
       : m_eivec(matrix.rows(), matrix.cols()),
         m_eivalues(matrix.cols()),
         m_subdiag(matrix.rows() > 1 ? matrix.rows() - 1 : 1),
         m_isInitialized(false)
     {
-      compute(matrix, options);
+      compute(matrix.derived(), options);
     }
 
     /** \brief Computes eigendecomposition of given matrix.
@@ -191,24 +198,45 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       *
       * \sa SelfAdjointEigenSolver(const MatrixType&, int)
       */
-    SelfAdjointEigenSolver& compute(const MatrixType& matrix, int options = ComputeEigenvectors);
+    template<typename InputType>
+    EIGEN_DEVICE_FUNC
+    SelfAdjointEigenSolver& compute(const EigenBase<InputType>& matrix, int options = ComputeEigenvectors);
     
-    /** \brief Computes eigendecomposition of given matrix using a direct algorithm
+    /** \brief Computes eigendecomposition of given matrix using a closed-form algorithm
       *
       * This is a variant of compute(const MatrixType&, int options) which
       * directly solves the underlying polynomial equation.
       * 
-      * Currently only 3x3 matrices for which the sizes are known at compile time are supported (e.g., Matrix3d).
+      * Currently only 2x2 and 3x3 matrices for which the sizes are known at compile time are supported (e.g., Matrix3d).
       * 
-      * This method is usually significantly faster than the QR algorithm
+      * This method is usually significantly faster than the QR iterative algorithm
       * but it might also be less accurate. It is also worth noting that
       * for 3x3 matrices it involves trigonometric operations which are
       * not necessarily available for all scalar types.
+      * 
+      * For the 3x3 case, we observed the following worst case relative error regarding the eigenvalues:
+      *   - double: 1e-8
+      *   - float:  1e-3
       *
       * \sa compute(const MatrixType&, int options)
       */
+    EIGEN_DEVICE_FUNC
     SelfAdjointEigenSolver& computeDirect(const MatrixType& matrix, int options = ComputeEigenvectors);
 
+    /**
+      *\brief Computes the eigen decomposition from a tridiagonal symmetric matrix
+      *
+      * \param[in] diag The vector containing the diagonal of the matrix.
+      * \param[in] subdiag The subdiagonal of the matrix.
+      * \param[in] options Can be #ComputeEigenvectors (default) or #EigenvaluesOnly.
+      * \returns Reference to \c *this
+      *
+      * This function assumes that the matrix has been reduced to tridiagonal form.
+      *
+      * \sa compute(const MatrixType&, int) for more information
+      */
+    SelfAdjointEigenSolver& computeFromTridiagonal(const RealVectorType& diag, const SubDiagonalType& subdiag , int options=ComputeEigenvectors);
+
     /** \brief Returns the eigenvectors of given matrix.
       *
       * \returns  A const reference to the matrix whose columns are the eigenvectors.
@@ -227,6 +255,7 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       *
       * \sa eigenvalues()
       */
+    EIGEN_DEVICE_FUNC
     const EigenvectorsType& eigenvectors() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
@@ -249,6 +278,7 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       *
       * \sa eigenvectors(), MatrixBase::eigenvalues()
       */
+    EIGEN_DEVICE_FUNC
     const RealVectorType& eigenvalues() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
@@ -270,9 +300,9 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       * Example: \include SelfAdjointEigenSolver_operatorSqrt.cpp
       * Output: \verbinclude SelfAdjointEigenSolver_operatorSqrt.out
       *
-      * \sa operatorInverseSqrt(),
-      *     \ref MatrixFunctions_Module "MatrixFunctions Module"
+      * \sa operatorInverseSqrt(), <a href="unsupported/group__MatrixFunctions__Module.html">MatrixFunctions Module</a>
       */
+    EIGEN_DEVICE_FUNC
     MatrixType operatorSqrt() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
@@ -295,9 +325,9 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       * Example: \include SelfAdjointEigenSolver_operatorInverseSqrt.cpp
       * Output: \verbinclude SelfAdjointEigenSolver_operatorInverseSqrt.out
       *
-      * \sa operatorSqrt(), MatrixBase::inverse(),
-      *     \ref MatrixFunctions_Module "MatrixFunctions Module"
+      * \sa operatorSqrt(), MatrixBase::inverse(), <a href="unsupported/group__MatrixFunctions__Module.html">MatrixFunctions Module</a>
       */
+    EIGEN_DEVICE_FUNC
     MatrixType operatorInverseSqrt() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
@@ -309,6 +339,7 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       *
       * \returns \c Success if computation was succesful, \c NoConvergence otherwise.
       */
+    EIGEN_DEVICE_FUNC
     ComputationInfo info() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
@@ -322,36 +353,6 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
       */
     static const int m_maxIterations = 30;
 
-    #ifdef EIGEN2_SUPPORT
-    SelfAdjointEigenSolver(const MatrixType& matrix, bool computeEigenvectors)
-      : m_eivec(matrix.rows(), matrix.cols()),
-        m_eivalues(matrix.cols()),
-        m_subdiag(matrix.rows() > 1 ? matrix.rows() - 1 : 1),
-        m_isInitialized(false)
-    {
-      compute(matrix, computeEigenvectors);
-    }
-    
-    SelfAdjointEigenSolver(const MatrixType& matA, const MatrixType& matB, bool computeEigenvectors = true)
-        : m_eivec(matA.cols(), matA.cols()),
-          m_eivalues(matA.cols()),
-          m_subdiag(matA.cols() > 1 ? matA.cols() - 1 : 1),
-          m_isInitialized(false)
-    {
-      static_cast<GeneralizedSelfAdjointEigenSolver<MatrixType>*>(this)->compute(matA, matB, computeEigenvectors ? ComputeEigenvectors : EigenvaluesOnly);
-    }
-    
-    void compute(const MatrixType& matrix, bool computeEigenvectors)
-    {
-      compute(matrix, computeEigenvectors ? ComputeEigenvectors : EigenvaluesOnly);
-    }
-
-    void compute(const MatrixType& matA, const MatrixType& matB, bool computeEigenvectors = true)
-    {
-      compute(matA, matB, computeEigenvectors ? ComputeEigenvectors : EigenvaluesOnly);
-    }
-    #endif // EIGEN2_SUPPORT
-
   protected:
     static void check_template_parameters()
     {
@@ -366,6 +367,7 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
     bool m_eigenvectorsOk;
 };
 
+namespace internal {
 /** \internal
   *
   * \eigenvalues_module \ingroup Eigenvalues_Module
@@ -373,8 +375,12 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
   * Performs a QR step on a tridiagonal symmetric matrix represented as a
   * pair of two vectors \a diag and \a subdiag.
   *
-  * \param matA the input selfadjoint matrix
-  * \param hCoeffs returned Householder coefficients
+  * \param diag the diagonal part of the input selfadjoint tridiagonal matrix
+  * \param subdiag the sub-diagonal part of the input selfadjoint tridiagonal matrix
+  * \param start starting index of the submatrix to work on
+  * \param end last+1 index of the submatrix to work on
+  * \param matrixQ pointer to the column-major matrix holding the eigenvectors, can be 0
+  * \param n size of the input matrix
   *
   * For compilation efficiency reasons, this procedure does not use eigen expression
   * for its arguments.
@@ -382,17 +388,21 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
   * Implemented from Golub's "Matrix Computations", algorithm 8.3.2:
   * "implicit symmetric QR step with Wilkinson shift"
   */
-namespace internal {
-template<typename RealScalar, typename Scalar, typename Index>
+template<int StorageOrder,typename RealScalar, typename Scalar, typename Index>
+EIGEN_DEVICE_FUNC
 static void tridiagonal_qr_step(RealScalar* diag, RealScalar* subdiag, Index start, Index end, Scalar* matrixQ, Index n);
 }
 
 template<typename MatrixType>
+template<typename InputType>
+EIGEN_DEVICE_FUNC
 SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
-::compute(const MatrixType& matrix, int options)
+::compute(const EigenBase<InputType>& a_matrix, int options)
 {
   check_template_parameters();
   
+  const InputType &matrix(a_matrix.derived());
+  
   using std::abs;
   eigen_assert(matrix.cols() == matrix.rows());
   eigen_assert((options&~(EigVecMask|GenEigMask))==0
@@ -404,7 +414,8 @@ SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
 
   if(n==1)
   {
-    m_eivalues.coeffRef(0,0) = numext::real(matrix.coeff(0,0));
+    m_eivec = matrix;
+    m_eivalues.coeffRef(0,0) = numext::real(m_eivec.coeff(0,0));
     if(computeEigenvectors)
       m_eivec.setOnes(n,n);
     m_info = Success;
@@ -424,19 +435,74 @@ SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
   mat.template triangularView<Lower>() /= scale;
   m_subdiag.resize(n-1);
   internal::tridiagonalization_inplace(mat, diag, m_subdiag, computeEigenvectors);
+
+  m_info = internal::computeFromTridiagonal_impl(diag, m_subdiag, m_maxIterations, computeEigenvectors, m_eivec);
   
+  // scale back the eigen values
+  m_eivalues *= scale;
+
+  m_isInitialized = true;
+  m_eigenvectorsOk = computeEigenvectors;
+  return *this;
+}
+
+template<typename MatrixType>
+SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
+::computeFromTridiagonal(const RealVectorType& diag, const SubDiagonalType& subdiag , int options)
+{
+  //TODO : Add an option to scale the values beforehand
+  bool computeEigenvectors = (options&ComputeEigenvectors)==ComputeEigenvectors;
+
+  m_eivalues = diag;
+  m_subdiag = subdiag;
+  if (computeEigenvectors)
+  {
+    m_eivec.setIdentity(diag.size(), diag.size());
+  }
+  m_info = internal::computeFromTridiagonal_impl(m_eivalues, m_subdiag, m_maxIterations, computeEigenvectors, m_eivec);
+
+  m_isInitialized = true;
+  m_eigenvectorsOk = computeEigenvectors;
+  return *this;
+}
+
+namespace internal {
+/**
+  * \internal
+  * \brief Compute the eigendecomposition from a tridiagonal matrix
+  *
+  * \param[in,out] diag : On input, the diagonal of the matrix, on output the eigenvalues
+  * \param[in,out] subdiag : The subdiagonal part of the matrix (entries are modified during the decomposition)
+  * \param[in] maxIterations : the maximum number of iterations
+  * \param[in] computeEigenvectors : whether the eigenvectors have to be computed or not
+  * \param[out] eivec : The matrix to store the eigenvectors if computeEigenvectors==true. Must be allocated on input.
+  * \returns \c Success or \c NoConvergence
+  */
+template<typename MatrixType, typename DiagType, typename SubDiagType>
+ComputationInfo computeFromTridiagonal_impl(DiagType& diag, SubDiagType& subdiag, const Index maxIterations, bool computeEigenvectors, MatrixType& eivec)
+{
+  using std::abs;
+
+  ComputationInfo info;
+  typedef typename MatrixType::Scalar Scalar;
+
+  Index n = diag.size();
   Index end = n-1;
   Index start = 0;
   Index iter = 0; // total number of iterations
-
+  
+  typedef typename DiagType::RealScalar RealScalar;
+  const RealScalar considerAsZero = (std::numeric_limits<RealScalar>::min)();
+  const RealScalar precision = RealScalar(2)*NumTraits<RealScalar>::epsilon();
+  
   while (end>0)
   {
     for (Index i = start; i<end; ++i)
-      if (internal::isMuchSmallerThan(abs(m_subdiag[i]),(abs(diag[i])+abs(diag[i+1]))))
-        m_subdiag[i] = 0;
+      if (internal::isMuchSmallerThan(abs(subdiag[i]),(abs(diag[i])+abs(diag[i+1])),precision) || abs(subdiag[i]) <= considerAsZero)
+        subdiag[i] = 0;
 
     // find the largest unreduced block
-    while (end>0 && m_subdiag[end-1]==0)
+    while (end>0 && subdiag[end-1]==RealScalar(0))
     {
       end--;
     }
@@ -445,51 +511,42 @@ SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
 
     // if we spent too many iterations, we give up
     iter++;
-    if(iter > m_maxIterations * n) break;
+    if(iter > maxIterations * n) break;
 
     start = end - 1;
-    while (start>0 && m_subdiag[start-1]!=0)
+    while (start>0 && subdiag[start-1]!=0)
       start--;
 
-    internal::tridiagonal_qr_step(diag.data(), m_subdiag.data(), start, end, computeEigenvectors ? m_eivec.data() : (Scalar*)0, n);
+    internal::tridiagonal_qr_step<MatrixType::Flags&RowMajorBit ? RowMajor : ColMajor>(diag.data(), subdiag.data(), start, end, computeEigenvectors ? eivec.data() : (Scalar*)0, n);
   }
-
-  if (iter <= m_maxIterations * n)
-    m_info = Success;
+  if (iter <= maxIterations * n)
+    info = Success;
   else
-    m_info = NoConvergence;
+    info = NoConvergence;
 
   // Sort eigenvalues and corresponding vectors.
   // TODO make the sort optional ?
   // TODO use a better sort algorithm !!
-  if (m_info == Success)
+  if (info == Success)
   {
     for (Index i = 0; i < n-1; ++i)
     {
       Index k;
-      m_eivalues.segment(i,n-i).minCoeff(&k);
+      diag.segment(i,n-i).minCoeff(&k);
       if (k > 0)
       {
-        std::swap(m_eivalues[i], m_eivalues[k+i]);
+        std::swap(diag[i], diag[k+i]);
         if(computeEigenvectors)
-          m_eivec.col(i).swap(m_eivec.col(k+i));
+          eivec.col(i).swap(eivec.col(k+i));
       }
     }
   }
-  
-  // scale back the eigen values
-  m_eivalues *= scale;
-
-  m_isInitialized = true;
-  m_eigenvectorsOk = computeEigenvectors;
-  return *this;
+  return info;
 }
-
-
-namespace internal {
   
 template<typename SolverType,int Size,bool IsComplex> struct direct_selfadjoint_eigenvalues
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(SolverType& eig, const typename SolverType::MatrixType& A, int options)
   { eig.compute(A,options); }
 };
@@ -499,21 +556,22 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,3
   typedef typename SolverType::MatrixType MatrixType;
   typedef typename SolverType::RealVectorType VectorType;
   typedef typename SolverType::Scalar Scalar;
-  typedef typename MatrixType::Index Index;
   typedef typename SolverType::EigenvectorsType EigenvectorsType;
   
+
   /** \internal
    * Computes the roots of the characteristic polynomial of \a m.
    * For numerical stability m.trace() should be near zero and to avoid over- or underflow m should be normalized.
    */
+  EIGEN_DEVICE_FUNC
   static inline void computeRoots(const MatrixType& m, VectorType& roots)
   {
-    using std::sqrt;
-    using std::atan2;
-    using std::cos;
-    using std::sin;
-    const Scalar s_inv3 = Scalar(1.0)/Scalar(3.0);
-    const Scalar s_sqrt3 = sqrt(Scalar(3.0));
+    EIGEN_USING_STD_MATH(sqrt)
+    EIGEN_USING_STD_MATH(atan2)
+    EIGEN_USING_STD_MATH(cos)
+    EIGEN_USING_STD_MATH(sin)
+    const Scalar s_inv3 = Scalar(1)/Scalar(3);
+    const Scalar s_sqrt3 = sqrt(Scalar(3));
 
     // The characteristic equation is x^3 - c2*x^2 + c1*x - c0 = 0.  The
     // eigenvalues are the roots to this equation, all guaranteed to be
@@ -526,14 +584,12 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,3
     // and in solving the equation for the roots in closed form.
     Scalar c2_over_3 = c2*s_inv3;
     Scalar a_over_3 = (c2*c2_over_3 - c1)*s_inv3;
-    if(a_over_3<Scalar(0))
-      a_over_3 = Scalar(0);
+    a_over_3 = numext::maxi(a_over_3, Scalar(0));
 
     Scalar half_b = Scalar(0.5)*(c0 + c2_over_3*(Scalar(2)*c2_over_3*c2_over_3 - c1));
 
     Scalar q = a_over_3*a_over_3*a_over_3 - half_b*half_b;
-    if(q<Scalar(0))
-      q = Scalar(0);
+    q = numext::maxi(q, Scalar(0));
 
     // Compute the eigenvalues by solving for the roots of the polynomial.
     Scalar rho = sqrt(a_over_3);
@@ -546,6 +602,7 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,3
     roots(2) = c2_over_3 + Scalar(2)*rho*cos_theta;
   }
 
+  EIGEN_DEVICE_FUNC
   static inline bool extract_kernel(MatrixType& mat, Ref<VectorType> res, Ref<VectorType> representative)
   {
     using std::abs;
@@ -565,6 +622,7 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,3
     return true;
   }
 
+  EIGEN_DEVICE_FUNC
   static inline void run(SolverType& solver, const MatrixType& mat, int options)
   {
     eigen_assert(mat.cols() == 3 && mat.cols() == mat.rows());
@@ -606,7 +664,7 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,3
         Index k(0), l(2);
         if(d0 > d1)
         {
-          std::swap(k,l);
+          numext::swap(k,l);
           d0 = d1;
         }
 
@@ -650,13 +708,15 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,3
 };
 
 // 2x2 direct eigenvalues decomposition, code from Hauke Heibel
-template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,2,false>
+template<typename SolverType> 
+struct direct_selfadjoint_eigenvalues<SolverType,2,false>
 {
   typedef typename SolverType::MatrixType MatrixType;
   typedef typename SolverType::RealVectorType VectorType;
   typedef typename SolverType::Scalar Scalar;
   typedef typename SolverType::EigenvectorsType EigenvectorsType;
   
+  EIGEN_DEVICE_FUNC
   static inline void computeRoots(const MatrixType& m, VectorType& roots)
   {
     using std::sqrt;
@@ -666,11 +726,12 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,2
     roots(1) = t1 + t0;
   }
   
+  EIGEN_DEVICE_FUNC
   static inline void run(SolverType& solver, const MatrixType& mat, int options)
   {
-    using std::sqrt;
-    using std::abs;
-
+    EIGEN_USING_STD_MATH(sqrt);
+    EIGEN_USING_STD_MATH(abs);
+    
     eigen_assert(mat.cols() == 2 && mat.cols() == mat.rows());
     eigen_assert((options&~(EigVecMask|GenEigMask))==0
             && (options&EigVecMask)!=EigVecMask
@@ -680,14 +741,18 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,2
     EigenvectorsType& eivecs = solver.m_eivec;
     VectorType& eivals = solver.m_eivalues;
   
-    // map the matrix coefficients to [-1:1] to avoid over- and underflow.
-    Scalar scale = mat.cwiseAbs().maxCoeff();
-    scale = (std::max)(scale,Scalar(1));
-    MatrixType scaledMat = mat / scale;
-    
+    // Shift the matrix to the mean eigenvalue and map the matrix coefficients to [-1:1] to avoid over- and underflow.
+    Scalar shift = mat.trace() / Scalar(2);
+    MatrixType scaledMat = mat;
+    scaledMat.coeffRef(0,1) = mat.coeff(1,0);
+    scaledMat.diagonal().array() -= shift;
+    Scalar scale = scaledMat.cwiseAbs().maxCoeff();
+    if(scale > Scalar(0))
+      scaledMat /= scale;
+
     // Compute the eigenvalues
     computeRoots(scaledMat,eivals);
-    
+
     // compute the eigen vectors
     if(computeEigenvectors)
     {
@@ -715,10 +780,11 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,2
         eivecs.col(0) << eivecs.col(1).unitOrthogonal();
       }
     }
-    
+
     // Rescale back to the original size.
     eivals *= scale;
-    
+    eivals.array() += shift;
+
     solver.m_info = Success;
     solver.m_isInitialized = true;
     solver.m_eigenvectorsOk = computeEigenvectors;
@@ -728,6 +794,7 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,2
 }
 
 template<typename MatrixType>
+EIGEN_DEVICE_FUNC
 SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
 ::computeDirect(const MatrixType& matrix, int options)
 {
@@ -736,7 +803,8 @@ SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
 }
 
 namespace internal {
-template<typename RealScalar, typename Scalar, typename Index>
+template<int StorageOrder,typename RealScalar, typename Scalar, typename Index>
+EIGEN_DEVICE_FUNC
 static void tridiagonal_qr_step(RealScalar* diag, RealScalar* subdiag, Index start, Index end, Scalar* matrixQ, Index n)
 {
   using std::abs;
@@ -748,14 +816,14 @@ static void tridiagonal_qr_step(RealScalar* diag, RealScalar* subdiag, Index sta
 //   RealScalar mu = diag[end] - e2 / (td + (td>0 ? 1 : -1) * sqrt(td*td + e2));
   // This explain the following, somewhat more complicated, version:
   RealScalar mu = diag[end];
-  if(td==0)
+  if(td==RealScalar(0))
     mu -= abs(e);
   else
   {
     RealScalar e2 = numext::abs2(subdiag[end-1]);
     RealScalar h = numext::hypot(td,e);
-    if(e2==0)  mu -= (e / (td + (td>0 ? 1 : -1))) * (e / h);
-    else       mu -= e2 / (td + (td>0 ? h : -h));
+    if(e2==RealScalar(0)) mu -= (e / (td + (td>RealScalar(0) ? RealScalar(1) : RealScalar(-1)))) * (e / h);
+    else                  mu -= e2 / (td + (td>RealScalar(0) ? h : -h));
   }
   
   RealScalar x = diag[start] - mu;
@@ -788,7 +856,8 @@ static void tridiagonal_qr_step(RealScalar* diag, RealScalar* subdiag, Index sta
     // apply the givens rotation to the unit matrix Q = Q * G
     if (matrixQ)
     {
-      Map<Matrix<Scalar,Dynamic,Dynamic,ColMajor> > q(matrixQ,n,n);
+      // FIXME if StorageOrder == RowMajor this operation is not very efficient
+      Map<Matrix<Scalar,Dynamic,Dynamic,StorageOrder> > q(matrixQ,n,n);
       q.applyOnTheRight(k,k+1,rot);
     }
   }
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h b/extern/Eigen3/Eigen/src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h
new file mode 100644
index 00000000000..b0c947dc07b
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h
@@ -0,0 +1,87 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *    Self-adjoint eigenvalues/eigenvectors.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_SAEIGENSOLVER_LAPACKE_H
+#define EIGEN_SAEIGENSOLVER_LAPACKE_H
+
+namespace Eigen { 
+
+/** \internal Specialization for the data types supported by LAPACKe */
+
+#define EIGEN_LAPACKE_EIG_SELFADJ_2(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_NAME, EIGCOLROW ) \
+template<> template<typename InputType> inline \
+SelfAdjointEigenSolver<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >& \
+SelfAdjointEigenSolver<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >::compute(const EigenBase<InputType>& matrix, int options) \
+{ \
+  eigen_assert(matrix.cols() == matrix.rows()); \
+  eigen_assert((options&~(EigVecMask|GenEigMask))==0 \
+          && (options&EigVecMask)!=EigVecMask \
+          && "invalid option parameter"); \
+  bool computeEigenvectors = (options&ComputeEigenvectors)==ComputeEigenvectors; \
+  lapack_int n = internal::convert_index<lapack_int>(matrix.cols()), lda, info; \
+  m_eivalues.resize(n,1); \
+  m_subdiag.resize(n-1); \
+  m_eivec = matrix; \
+\
+  if(n==1) \
+  { \
+    m_eivalues.coeffRef(0,0) = numext::real(m_eivec.coeff(0,0)); \
+    if(computeEigenvectors) m_eivec.setOnes(n,n); \
+    m_info = Success; \
+    m_isInitialized = true; \
+    m_eigenvectorsOk = computeEigenvectors; \
+    return *this; \
+  } \
+\
+  lda = internal::convert_index<lapack_int>(m_eivec.outerStride()); \
+  char jobz, uplo='L'/*, range='A'*/; \
+  jobz = computeEigenvectors ? 'V' : 'N'; \
+\
+  info = LAPACKE_##LAPACKE_NAME( LAPACK_COL_MAJOR, jobz, uplo, n, (LAPACKE_TYPE*)m_eivec.data(), lda, (LAPACKE_RTYPE*)m_eivalues.data() ); \
+  m_info = (info==0) ? Success : NoConvergence; \
+  m_isInitialized = true; \
+  m_eigenvectorsOk = computeEigenvectors; \
+  return *this; \
+}
+
+#define EIGEN_LAPACKE_EIG_SELFADJ(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_NAME )              \
+        EIGEN_LAPACKE_EIG_SELFADJ_2(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_NAME, ColMajor )  \
+        EIGEN_LAPACKE_EIG_SELFADJ_2(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_NAME, RowMajor ) 
+
+EIGEN_LAPACKE_EIG_SELFADJ(double,   double,                double, dsyev)
+EIGEN_LAPACKE_EIG_SELFADJ(float,    float,                 float,  ssyev)
+EIGEN_LAPACKE_EIG_SELFADJ(dcomplex, lapack_complex_double, double, zheev)
+EIGEN_LAPACKE_EIG_SELFADJ(scomplex, lapack_complex_float,  float,  cheev)
+
+} // end namespace Eigen
+
+#endif // EIGEN_SAEIGENSOLVER_H
diff --git a/extern/Eigen3/Eigen/src/Eigenvalues/Tridiagonalization.h b/extern/Eigen3/Eigen/src/Eigenvalues/Tridiagonalization.h
index 192278d685d..1d102c17bc1 100644
--- a/extern/Eigen3/Eigen/src/Eigenvalues/Tridiagonalization.h
+++ b/extern/Eigen3/Eigen/src/Eigenvalues/Tridiagonalization.h
@@ -18,8 +18,10 @@ namespace internal {
 template<typename MatrixType> struct TridiagonalizationMatrixTReturnType;
 template<typename MatrixType>
 struct traits<TridiagonalizationMatrixTReturnType<MatrixType> >
+  : public traits<typename MatrixType::PlainObject>
 {
-  typedef typename MatrixType::PlainObject ReturnType;
+  typedef typename MatrixType::PlainObject ReturnType; // FIXME shall it be a BandMatrix?
+  enum { Flags = 0 };
 };
 
 template<typename MatrixType, typename CoeffVectorType>
@@ -67,7 +69,7 @@ template<typename _MatrixType> class Tridiagonalization
 
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     enum {
       Size = MatrixType::RowsAtCompileTime,
@@ -89,10 +91,8 @@ template<typename _MatrixType> class Tridiagonalization
             >::type DiagonalReturnType;
 
     typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
-              typename internal::add_const_on_value_type<typename Diagonal<
-                Block<const MatrixType,SizeMinusOne,SizeMinusOne> >::RealReturnType>::type,
-              const Diagonal<
-                Block<const MatrixType,SizeMinusOne,SizeMinusOne> >
+              typename internal::add_const_on_value_type<typename Diagonal<const MatrixType, -1>::RealReturnType>::type,
+              const Diagonal<const MatrixType, -1>
             >::type SubDiagonalReturnType;
 
     /** \brief Return type of matrixQ() */
@@ -110,7 +110,7 @@ template<typename _MatrixType> class Tridiagonalization
       *
       * \sa compute() for an example.
       */
-    Tridiagonalization(Index size = Size==Dynamic ? 2 : Size)
+    explicit Tridiagonalization(Index size = Size==Dynamic ? 2 : Size)
       : m_matrix(size,size),
         m_hCoeffs(size > 1 ? size-1 : 1),
         m_isInitialized(false)
@@ -126,8 +126,9 @@ template<typename _MatrixType> class Tridiagonalization
       * Example: \include Tridiagonalization_Tridiagonalization_MatrixType.cpp
       * Output: \verbinclude Tridiagonalization_Tridiagonalization_MatrixType.out
       */
-    Tridiagonalization(const MatrixType& matrix)
-      : m_matrix(matrix),
+    template<typename InputType>
+    explicit Tridiagonalization(const EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
         m_hCoeffs(matrix.cols() > 1 ? matrix.cols()-1 : 1),
         m_isInitialized(false)
     {
@@ -152,9 +153,10 @@ template<typename _MatrixType> class Tridiagonalization
       * Example: \include Tridiagonalization_compute.cpp
       * Output: \verbinclude Tridiagonalization_compute.out
       */
-    Tridiagonalization& compute(const MatrixType& matrix)
+    template<typename InputType>
+    Tridiagonalization& compute(const EigenBase<InputType>& matrix)
     {
-      m_matrix = matrix;
+      m_matrix = matrix.derived();
       m_hCoeffs.resize(matrix.rows()-1, 1);
       internal::tridiagonalization_inplace(m_matrix, m_hCoeffs);
       m_isInitialized = true;
@@ -305,7 +307,7 @@ typename Tridiagonalization<MatrixType>::DiagonalReturnType
 Tridiagonalization<MatrixType>::diagonal() const
 {
   eigen_assert(m_isInitialized && "Tridiagonalization is not initialized.");
-  return m_matrix.diagonal();
+  return m_matrix.diagonal().real();
 }
 
 template<typename MatrixType>
@@ -313,8 +315,7 @@ typename Tridiagonalization<MatrixType>::SubDiagonalReturnType
 Tridiagonalization<MatrixType>::subDiagonal() const
 {
   eigen_assert(m_isInitialized && "Tridiagonalization is not initialized.");
-  Index n = m_matrix.rows();
-  return Block<const MatrixType,SizeMinusOne,SizeMinusOne>(m_matrix, 1, 0, n-1,n-1).diagonal();
+  return m_matrix.template diagonal<-1>().real();
 }
 
 namespace internal {
@@ -346,7 +347,6 @@ template<typename MatrixType, typename CoeffVectorType>
 void tridiagonalization_inplace(MatrixType& matA, CoeffVectorType& hCoeffs)
 {
   using numext::conj;
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   typedef typename MatrixType::RealScalar RealScalar;
   Index n = matA.rows();
@@ -367,10 +367,10 @@ void tridiagonalization_inplace(MatrixType& matA, CoeffVectorType& hCoeffs)
     hCoeffs.tail(n-i-1).noalias() = (matA.bottomRightCorner(remainingSize,remainingSize).template selfadjointView<Lower>()
                                   * (conj(h) * matA.col(i).tail(remainingSize)));
 
-    hCoeffs.tail(n-i-1) += (conj(h)*Scalar(-0.5)*(hCoeffs.tail(remainingSize).dot(matA.col(i).tail(remainingSize)))) * matA.col(i).tail(n-i-1);
+    hCoeffs.tail(n-i-1) += (conj(h)*RealScalar(-0.5)*(hCoeffs.tail(remainingSize).dot(matA.col(i).tail(remainingSize)))) * matA.col(i).tail(n-i-1);
 
     matA.bottomRightCorner(remainingSize, remainingSize).template selfadjointView<Lower>()
-      .rankUpdate(matA.col(i).tail(remainingSize), hCoeffs.tail(remainingSize), -1);
+      .rankUpdate(matA.col(i).tail(remainingSize), hCoeffs.tail(remainingSize), Scalar(-1));
 
     matA.col(i).coeffRef(i+1) = beta;
     hCoeffs.coeffRef(i) = h;
@@ -438,7 +438,6 @@ struct tridiagonalization_inplace_selector
 {
   typedef typename Tridiagonalization<MatrixType>::CoeffVectorType CoeffVectorType;
   typedef typename Tridiagonalization<MatrixType>::HouseholderSequenceType HouseholderSequenceType;
-  typedef typename MatrixType::Index Index;
   template<typename DiagonalType, typename SubDiagonalType>
   static void run(MatrixType& mat, DiagonalType& diag, SubDiagonalType& subdiag, bool extractQ)
   {
@@ -467,9 +466,10 @@ struct tridiagonalization_inplace_selector<MatrixType,3,false>
   static void run(MatrixType& mat, DiagonalType& diag, SubDiagonalType& subdiag, bool extractQ)
   {
     using std::sqrt;
+    const RealScalar tol = (std::numeric_limits<RealScalar>::min)();
     diag[0] = mat(0,0);
     RealScalar v1norm2 = numext::abs2(mat(2,0));
-    if(v1norm2 == RealScalar(0))
+    if(v1norm2 <= tol)
     {
       diag[1] = mat(1,1);
       diag[2] = mat(2,2);
@@ -526,7 +526,6 @@ struct tridiagonalization_inplace_selector<MatrixType,1,IsComplex>
 template<typename MatrixType> struct TridiagonalizationMatrixTReturnType
 : public ReturnByValue<TridiagonalizationMatrixTReturnType<MatrixType> >
 {
-    typedef typename MatrixType::Index Index;
   public:
     /** \brief Constructor.
       *
diff --git a/extern/Eigen3/Eigen/src/Geometry/AlignedBox.h b/extern/Eigen3/Eigen/src/Geometry/AlignedBox.h
index 7e1cd9eb79c..066eae4f921 100644
--- a/extern/Eigen3/Eigen/src/Geometry/AlignedBox.h
+++ b/extern/Eigen3/Eigen/src/Geometry/AlignedBox.h
@@ -34,10 +34,11 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
   enum { AmbientDimAtCompileTime = _AmbientDim };
   typedef _Scalar                                   Scalar;
   typedef NumTraits<Scalar>                         ScalarTraits;
-  typedef DenseIndex                                Index;
+  typedef Eigen::Index                              Index; ///< \deprecated since Eigen 3.3
   typedef typename ScalarTraits::Real               RealScalar;
-  typedef typename ScalarTraits::NonInteger      NonInteger;
+  typedef typename ScalarTraits::NonInteger         NonInteger;
   typedef Matrix<Scalar,AmbientDimAtCompileTime,1>  VectorType;
+  typedef CwiseBinaryOp<internal::scalar_sum_op<Scalar>, const VectorType, const VectorType> VectorTypeSum;
 
   /** Define constants to name the corners of a 1D, 2D or 3D axis aligned bounding box */
   enum CornerType
@@ -61,77 +62,76 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
 
 
   /** Default constructor initializing a null box. */
-  inline AlignedBox()
+  EIGEN_DEVICE_FUNC inline AlignedBox()
   { if (AmbientDimAtCompileTime!=Dynamic) setEmpty(); }
 
   /** Constructs a null box with \a _dim the dimension of the ambient space. */
-  inline explicit AlignedBox(Index _dim) : m_min(_dim), m_max(_dim)
+  EIGEN_DEVICE_FUNC inline explicit AlignedBox(Index _dim) : m_min(_dim), m_max(_dim)
   { setEmpty(); }
 
   /** Constructs a box with extremities \a _min and \a _max.
    * \warning If either component of \a _min is larger than the same component of \a _max, the constructed box is empty. */
   template<typename OtherVectorType1, typename OtherVectorType2>
-  inline AlignedBox(const OtherVectorType1& _min, const OtherVectorType2& _max) : m_min(_min), m_max(_max) {}
+  EIGEN_DEVICE_FUNC inline AlignedBox(const OtherVectorType1& _min, const OtherVectorType2& _max) : m_min(_min), m_max(_max) {}
 
   /** Constructs a box containing a single point \a p. */
   template<typename Derived>
-  inline explicit AlignedBox(const MatrixBase<Derived>& p) : m_min(p), m_max(m_min)
+  EIGEN_DEVICE_FUNC inline explicit AlignedBox(const MatrixBase<Derived>& p) : m_min(p), m_max(m_min)
   { }
 
-  ~AlignedBox() {}
+  EIGEN_DEVICE_FUNC ~AlignedBox() {}
 
   /** \returns the dimension in which the box holds */
-  inline Index dim() const { return AmbientDimAtCompileTime==Dynamic ? m_min.size() : Index(AmbientDimAtCompileTime); }
+  EIGEN_DEVICE_FUNC inline Index dim() const { return AmbientDimAtCompileTime==Dynamic ? m_min.size() : Index(AmbientDimAtCompileTime); }
 
   /** \deprecated use isEmpty() */
-  inline bool isNull() const { return isEmpty(); }
+  EIGEN_DEVICE_FUNC inline bool isNull() const { return isEmpty(); }
 
   /** \deprecated use setEmpty() */
-  inline void setNull() { setEmpty(); }
+  EIGEN_DEVICE_FUNC inline void setNull() { setEmpty(); }
 
   /** \returns true if the box is empty.
    * \sa setEmpty */
-  inline bool isEmpty() const { return (m_min.array() > m_max.array()).any(); }
+  EIGEN_DEVICE_FUNC inline bool isEmpty() const { return (m_min.array() > m_max.array()).any(); }
 
   /** Makes \c *this an empty box.
    * \sa isEmpty */
-  inline void setEmpty()
+  EIGEN_DEVICE_FUNC inline void setEmpty()
   {
     m_min.setConstant( ScalarTraits::highest() );
     m_max.setConstant( ScalarTraits::lowest() );
   }
 
   /** \returns the minimal corner */
-  inline const VectorType& (min)() const { return m_min; }
+  EIGEN_DEVICE_FUNC inline const VectorType& (min)() const { return m_min; }
   /** \returns a non const reference to the minimal corner */
-  inline VectorType& (min)() { return m_min; }
+  EIGEN_DEVICE_FUNC inline VectorType& (min)() { return m_min; }
   /** \returns the maximal corner */
-  inline const VectorType& (max)() const { return m_max; }
+  EIGEN_DEVICE_FUNC inline const VectorType& (max)() const { return m_max; }
   /** \returns a non const reference to the maximal corner */
-  inline VectorType& (max)() { return m_max; }
+  EIGEN_DEVICE_FUNC inline VectorType& (max)() { return m_max; }
 
   /** \returns the center of the box */
-  inline const CwiseUnaryOp<internal::scalar_quotient1_op<Scalar>,
-                            const CwiseBinaryOp<internal::scalar_sum_op<Scalar>, const VectorType, const VectorType> >
+  EIGEN_DEVICE_FUNC inline const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(VectorTypeSum, RealScalar, quotient)
   center() const
-  { return (m_min+m_max)/2; }
+  { return (m_min+m_max)/RealScalar(2); }
 
   /** \returns the lengths of the sides of the bounding box.
     * Note that this function does not get the same
     * result for integral or floating scalar types: see
     */
-  inline const CwiseBinaryOp< internal::scalar_difference_op<Scalar>, const VectorType, const VectorType> sizes() const
+  EIGEN_DEVICE_FUNC inline const CwiseBinaryOp< internal::scalar_difference_op<Scalar,Scalar>, const VectorType, const VectorType> sizes() const
   { return m_max - m_min; }
 
   /** \returns the volume of the bounding box */
-  inline Scalar volume() const
+  EIGEN_DEVICE_FUNC inline Scalar volume() const
   { return sizes().prod(); }
 
   /** \returns an expression for the bounding box diagonal vector
     * if the length of the diagonal is needed: diagonal().norm()
     * will provide it.
     */
-  inline CwiseBinaryOp< internal::scalar_difference_op<Scalar>, const VectorType, const VectorType> diagonal() const
+  EIGEN_DEVICE_FUNC inline CwiseBinaryOp< internal::scalar_difference_op<Scalar,Scalar>, const VectorType, const VectorType> diagonal() const
   { return sizes(); }
 
   /** \returns the vertex of the bounding box at the corner defined by
@@ -143,7 +143,7 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
     * For 3D bounding boxes, the following names are added:
     * BottomLeftCeil, BottomRightCeil, TopLeftCeil, TopRightCeil.
     */
-  inline VectorType corner(CornerType corner) const
+  EIGEN_DEVICE_FUNC inline VectorType corner(CornerType corner) const
   {
     EIGEN_STATIC_ASSERT(_AmbientDim <= 3, THIS_METHOD_IS_ONLY_FOR_VECTORS_OF_A_SPECIFIC_SIZE);
 
@@ -161,7 +161,7 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
 
   /** \returns a random point inside the bounding box sampled with
    * a uniform distribution */
-  inline VectorType sample() const
+  EIGEN_DEVICE_FUNC inline VectorType sample() const
   {
     VectorType r(dim());
     for(Index d=0; d<dim(); ++d)
@@ -179,27 +179,27 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
 
   /** \returns true if the point \a p is inside the box \c *this. */
   template<typename Derived>
-  inline bool contains(const MatrixBase<Derived>& p) const
+  EIGEN_DEVICE_FUNC inline bool contains(const MatrixBase<Derived>& p) const
   {
-    typename internal::nested<Derived,2>::type p_n(p.derived());
+    typename internal::nested_eval<Derived,2>::type p_n(p.derived());
     return (m_min.array()<=p_n.array()).all() && (p_n.array()<=m_max.array()).all();
   }
 
   /** \returns true if the box \a b is entirely inside the box \c *this. */
-  inline bool contains(const AlignedBox& b) const
+  EIGEN_DEVICE_FUNC inline bool contains(const AlignedBox& b) const
   { return (m_min.array()<=(b.min)().array()).all() && ((b.max)().array()<=m_max.array()).all(); }
 
   /** \returns true if the box \a b is intersecting the box \c *this.
    * \sa intersection, clamp */
-  inline bool intersects(const AlignedBox& b) const
+  EIGEN_DEVICE_FUNC inline bool intersects(const AlignedBox& b) const
   { return (m_min.array()<=(b.max)().array()).all() && ((b.min)().array()<=m_max.array()).all(); }
 
   /** Extends \c *this such that it contains the point \a p and returns a reference to \c *this.
    * \sa extend(const AlignedBox&) */
   template<typename Derived>
-  inline AlignedBox& extend(const MatrixBase<Derived>& p)
+  EIGEN_DEVICE_FUNC inline AlignedBox& extend(const MatrixBase<Derived>& p)
   {
-    typename internal::nested<Derived,2>::type p_n(p.derived());
+    typename internal::nested_eval<Derived,2>::type p_n(p.derived());
     m_min = m_min.cwiseMin(p_n);
     m_max = m_max.cwiseMax(p_n);
     return *this;
@@ -207,7 +207,7 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
 
   /** Extends \c *this such that it contains the box \a b and returns a reference to \c *this.
    * \sa merged, extend(const MatrixBase&) */
-  inline AlignedBox& extend(const AlignedBox& b)
+  EIGEN_DEVICE_FUNC inline AlignedBox& extend(const AlignedBox& b)
   {
     m_min = m_min.cwiseMin(b.m_min);
     m_max = m_max.cwiseMax(b.m_max);
@@ -217,7 +217,7 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
   /** Clamps \c *this by the box \a b and returns a reference to \c *this.
    * \note If the boxes don't intersect, the resulting box is empty.
    * \sa intersection(), intersects() */
-  inline AlignedBox& clamp(const AlignedBox& b)
+  EIGEN_DEVICE_FUNC inline AlignedBox& clamp(const AlignedBox& b)
   {
     m_min = m_min.cwiseMax(b.m_min);
     m_max = m_max.cwiseMin(b.m_max);
@@ -227,20 +227,20 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
   /** Returns an AlignedBox that is the intersection of \a b and \c *this
    * \note If the boxes don't intersect, the resulting box is empty.
    * \sa intersects(), clamp, contains()  */
-  inline AlignedBox intersection(const AlignedBox& b) const
+  EIGEN_DEVICE_FUNC inline AlignedBox intersection(const AlignedBox& b) const
   {return AlignedBox(m_min.cwiseMax(b.m_min), m_max.cwiseMin(b.m_max)); }
 
   /** Returns an AlignedBox that is the union of \a b and \c *this.
    * \note Merging with an empty box may result in a box bigger than \c *this. 
    * \sa extend(const AlignedBox&) */
-  inline AlignedBox merged(const AlignedBox& b) const
+  EIGEN_DEVICE_FUNC inline AlignedBox merged(const AlignedBox& b) const
   { return AlignedBox(m_min.cwiseMin(b.m_min), m_max.cwiseMax(b.m_max)); }
 
   /** Translate \c *this by the vector \a t and returns a reference to \c *this. */
   template<typename Derived>
-  inline AlignedBox& translate(const MatrixBase<Derived>& a_t)
+  EIGEN_DEVICE_FUNC inline AlignedBox& translate(const MatrixBase<Derived>& a_t)
   {
-    const typename internal::nested<Derived,2>::type t(a_t.derived());
+    const typename internal::nested_eval<Derived,2>::type t(a_t.derived());
     m_min += t;
     m_max += t;
     return *this;
@@ -251,28 +251,28 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
     * \sa exteriorDistance(const MatrixBase&), squaredExteriorDistance(const AlignedBox&)
     */
   template<typename Derived>
-  inline Scalar squaredExteriorDistance(const MatrixBase<Derived>& p) const;
+  EIGEN_DEVICE_FUNC inline Scalar squaredExteriorDistance(const MatrixBase<Derived>& p) const;
 
   /** \returns the squared distance between the boxes \a b and \c *this,
     * and zero if the boxes intersect.
     * \sa exteriorDistance(const AlignedBox&), squaredExteriorDistance(const MatrixBase&)
     */
-  inline Scalar squaredExteriorDistance(const AlignedBox& b) const;
+  EIGEN_DEVICE_FUNC inline Scalar squaredExteriorDistance(const AlignedBox& b) const;
 
   /** \returns the distance between the point \a p and the box \c *this,
     * and zero if \a p is inside the box.
     * \sa squaredExteriorDistance(const MatrixBase&), exteriorDistance(const AlignedBox&)
     */
   template<typename Derived>
-  inline NonInteger exteriorDistance(const MatrixBase<Derived>& p) const
-  { using std::sqrt; return sqrt(NonInteger(squaredExteriorDistance(p))); }
+  EIGEN_DEVICE_FUNC inline NonInteger exteriorDistance(const MatrixBase<Derived>& p) const
+  { EIGEN_USING_STD_MATH(sqrt) return sqrt(NonInteger(squaredExteriorDistance(p))); }
 
   /** \returns the distance between the boxes \a b and \c *this,
     * and zero if the boxes intersect.
     * \sa squaredExteriorDistance(const AlignedBox&), exteriorDistance(const MatrixBase&)
     */
-  inline NonInteger exteriorDistance(const AlignedBox& b) const
-  { using std::sqrt; return sqrt(NonInteger(squaredExteriorDistance(b))); }
+  EIGEN_DEVICE_FUNC inline NonInteger exteriorDistance(const AlignedBox& b) const
+  { EIGEN_USING_STD_MATH(sqrt) return sqrt(NonInteger(squaredExteriorDistance(b))); }
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -280,7 +280,7 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<AlignedBox,
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<AlignedBox,
            AlignedBox<NewScalarType,AmbientDimAtCompileTime> >::type cast() const
   {
     return typename internal::cast_return_type<AlignedBox,
@@ -289,7 +289,7 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit AlignedBox(const AlignedBox<OtherScalarType,AmbientDimAtCompileTime>& other)
+  EIGEN_DEVICE_FUNC inline explicit AlignedBox(const AlignedBox<OtherScalarType,AmbientDimAtCompileTime>& other)
   {
     m_min = (other.min)().template cast<Scalar>();
     m_max = (other.max)().template cast<Scalar>();
@@ -299,7 +299,7 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const AlignedBox& other, const RealScalar& prec = ScalarTraits::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const AlignedBox& other, const RealScalar& prec = ScalarTraits::dummy_precision()) const
   { return m_min.isApprox(other.m_min, prec) && m_max.isApprox(other.m_max, prec); }
 
 protected:
@@ -311,9 +311,9 @@ protected:
 
 template<typename Scalar,int AmbientDim>
 template<typename Derived>
-inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const MatrixBase<Derived>& a_p) const
+EIGEN_DEVICE_FUNC inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const MatrixBase<Derived>& a_p) const
 {
-  typename internal::nested<Derived,2*AmbientDim>::type p(a_p.derived());
+  typename internal::nested_eval<Derived,2*AmbientDim>::type p(a_p.derived());
   Scalar dist2(0);
   Scalar aux;
   for (Index k=0; k<dim(); ++k)
@@ -333,7 +333,7 @@ inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const Matri
 }
 
 template<typename Scalar,int AmbientDim>
-inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const AlignedBox& b) const
+EIGEN_DEVICE_FUNC inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const AlignedBox& b) const
 {
   Scalar dist2(0);
   Scalar aux;
diff --git a/extern/Eigen3/Eigen/src/Geometry/AngleAxis.h b/extern/Eigen3/Eigen/src/Geometry/AngleAxis.h
index bbf6a7ed8ed..83ee1be4616 100644
--- a/extern/Eigen3/Eigen/src/Geometry/AngleAxis.h
+++ b/extern/Eigen3/Eigen/src/Geometry/AngleAxis.h
@@ -69,50 +69,61 @@ protected:
 public:
 
   /** Default constructor without initialization. */
-  AngleAxis() {}
+  EIGEN_DEVICE_FUNC AngleAxis() {}
   /** Constructs and initialize the angle-axis rotation from an \a angle in radian
     * and an \a axis which \b must \b be \b normalized.
     *
     * \warning If the \a axis vector is not normalized, then the angle-axis object
     *          represents an invalid rotation. */
   template<typename Derived>
+  EIGEN_DEVICE_FUNC 
   inline AngleAxis(const Scalar& angle, const MatrixBase<Derived>& axis) : m_axis(axis), m_angle(angle) {}
-  /** Constructs and initialize the angle-axis rotation from a quaternion \a q. */
-  template<typename QuatDerived> inline explicit AngleAxis(const QuaternionBase<QuatDerived>& q) { *this = q; }
+  /** Constructs and initialize the angle-axis rotation from a quaternion \a q.
+    * This function implicitly normalizes the quaternion \a q.
+    */
+  template<typename QuatDerived> 
+  EIGEN_DEVICE_FUNC inline explicit AngleAxis(const QuaternionBase<QuatDerived>& q) { *this = q; }
   /** Constructs and initialize the angle-axis rotation from a 3x3 rotation matrix. */
   template<typename Derived>
-  inline explicit AngleAxis(const MatrixBase<Derived>& m) { *this = m; }
+  EIGEN_DEVICE_FUNC inline explicit AngleAxis(const MatrixBase<Derived>& m) { *this = m; }
 
-  Scalar angle() const { return m_angle; }
-  Scalar& angle() { return m_angle; }
+  /** \returns the value of the rotation angle in radian */
+  EIGEN_DEVICE_FUNC Scalar angle() const { return m_angle; }
+  /** \returns a read-write reference to the stored angle in radian */
+  EIGEN_DEVICE_FUNC Scalar& angle() { return m_angle; }
 
-  const Vector3& axis() const { return m_axis; }
-  Vector3& axis() { return m_axis; }
+  /** \returns the rotation axis */
+  EIGEN_DEVICE_FUNC const Vector3& axis() const { return m_axis; }
+  /** \returns a read-write reference to the stored rotation axis.
+    *
+    * \warning The rotation axis must remain a \b unit vector.
+    */
+  EIGEN_DEVICE_FUNC Vector3& axis() { return m_axis; }
 
   /** Concatenates two rotations */
-  inline QuaternionType operator* (const AngleAxis& other) const
+  EIGEN_DEVICE_FUNC inline QuaternionType operator* (const AngleAxis& other) const
   { return QuaternionType(*this) * QuaternionType(other); }
 
   /** Concatenates two rotations */
-  inline QuaternionType operator* (const QuaternionType& other) const
+  EIGEN_DEVICE_FUNC inline QuaternionType operator* (const QuaternionType& other) const
   { return QuaternionType(*this) * other; }
 
   /** Concatenates two rotations */
-  friend inline QuaternionType operator* (const QuaternionType& a, const AngleAxis& b)
+  friend EIGEN_DEVICE_FUNC inline QuaternionType operator* (const QuaternionType& a, const AngleAxis& b)
   { return a * QuaternionType(b); }
 
   /** \returns the inverse rotation, i.e., an angle-axis with opposite rotation angle */
-  AngleAxis inverse() const
+  EIGEN_DEVICE_FUNC AngleAxis inverse() const
   { return AngleAxis(-m_angle, m_axis); }
 
   template<class QuatDerived>
-  AngleAxis& operator=(const QuaternionBase<QuatDerived>& q);
+  EIGEN_DEVICE_FUNC AngleAxis& operator=(const QuaternionBase<QuatDerived>& q);
   template<typename Derived>
-  AngleAxis& operator=(const MatrixBase<Derived>& m);
+  EIGEN_DEVICE_FUNC AngleAxis& operator=(const MatrixBase<Derived>& m);
 
   template<typename Derived>
-  AngleAxis& fromRotationMatrix(const MatrixBase<Derived>& m);
-  Matrix3 toRotationMatrix(void) const;
+  EIGEN_DEVICE_FUNC AngleAxis& fromRotationMatrix(const MatrixBase<Derived>& m);
+  EIGEN_DEVICE_FUNC Matrix3 toRotationMatrix(void) const;
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -120,24 +131,24 @@ public:
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<AngleAxis,AngleAxis<NewScalarType> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<AngleAxis,AngleAxis<NewScalarType> >::type cast() const
   { return typename internal::cast_return_type<AngleAxis,AngleAxis<NewScalarType> >::type(*this); }
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit AngleAxis(const AngleAxis<OtherScalarType>& other)
+  EIGEN_DEVICE_FUNC inline explicit AngleAxis(const AngleAxis<OtherScalarType>& other)
   {
     m_axis = other.axis().template cast<Scalar>();
     m_angle = Scalar(other.angle());
   }
 
-  static inline const AngleAxis Identity() { return AngleAxis(Scalar(0), Vector3::UnitX()); }
+  EIGEN_DEVICE_FUNC static inline const AngleAxis Identity() { return AngleAxis(Scalar(0), Vector3::UnitX()); }
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const AngleAxis& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const AngleAxis& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_axis.isApprox(other.m_axis, prec) && internal::isApprox(m_angle,other.m_angle, prec); }
 };
 
@@ -149,29 +160,32 @@ typedef AngleAxis<float> AngleAxisf;
 typedef AngleAxis<double> AngleAxisd;
 
 /** Set \c *this from a \b unit quaternion.
-  * The axis is normalized.
+  *
+  * The resulting axis is normalized, and the computed angle is in the [0,pi] range.
   * 
-  * \warning As any other method dealing with quaternion, if the input quaternion
-  *          is not normalized then the result is undefined.
+  * This function implicitly normalizes the quaternion \a q.
   */
 template<typename Scalar>
 template<typename QuatDerived>
-AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const QuaternionBase<QuatDerived>& q)
+EIGEN_DEVICE_FUNC AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const QuaternionBase<QuatDerived>& q)
 {
-  using std::acos;
-  using std::min;
-  using std::max;
-  using std::sqrt;
-  Scalar n2 = q.vec().squaredNorm();
-  if (n2 < NumTraits<Scalar>::dummy_precision()*NumTraits<Scalar>::dummy_precision())
+  EIGEN_USING_STD_MATH(atan2)
+  EIGEN_USING_STD_MATH(abs)
+  Scalar n = q.vec().norm();
+  if(n<NumTraits<Scalar>::epsilon())
+    n = q.vec().stableNorm();
+
+  if (n != Scalar(0))
   {
-    m_angle = Scalar(0);
-    m_axis << Scalar(1), Scalar(0), Scalar(0);
+    m_angle = Scalar(2)*atan2(n, abs(q.w()));
+    if(q.w() < Scalar(0))
+      n = -n;
+    m_axis  = q.vec() / n;
   }
   else
   {
-    m_angle = Scalar(2)*acos((min)((max)(Scalar(-1),q.w()),Scalar(1)));
-    m_axis = q.vec() / sqrt(n2);
+    m_angle = Scalar(0);
+    m_axis << Scalar(1), Scalar(0), Scalar(0);
   }
   return *this;
 }
@@ -180,7 +194,7 @@ AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const QuaternionBase<QuatDerived
   */
 template<typename Scalar>
 template<typename Derived>
-AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const MatrixBase<Derived>& mat)
+EIGEN_DEVICE_FUNC AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const MatrixBase<Derived>& mat)
 {
   // Since a direct conversion would not be really faster,
   // let's use the robust Quaternion implementation:
@@ -192,7 +206,7 @@ AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const MatrixBase<Derived>& mat)
 **/
 template<typename Scalar>
 template<typename Derived>
-AngleAxis<Scalar>& AngleAxis<Scalar>::fromRotationMatrix(const MatrixBase<Derived>& mat)
+EIGEN_DEVICE_FUNC AngleAxis<Scalar>& AngleAxis<Scalar>::fromRotationMatrix(const MatrixBase<Derived>& mat)
 {
   return *this = QuaternionType(mat);
 }
@@ -201,10 +215,10 @@ AngleAxis<Scalar>& AngleAxis<Scalar>::fromRotationMatrix(const MatrixBase<Derive
   */
 template<typename Scalar>
 typename AngleAxis<Scalar>::Matrix3
-AngleAxis<Scalar>::toRotationMatrix(void) const
+EIGEN_DEVICE_FUNC AngleAxis<Scalar>::toRotationMatrix(void) const
 {
-  using std::sin;
-  using std::cos;
+  EIGEN_USING_STD_MATH(sin)
+  EIGEN_USING_STD_MATH(cos)
   Matrix3 res;
   Vector3 sin_axis  = sin(m_angle) * m_axis;
   Scalar c = cos(m_angle);
diff --git a/extern/Eigen3/Eigen/src/Geometry/EulerAngles.h b/extern/Eigen3/Eigen/src/Geometry/EulerAngles.h
index 82802fb43cf..c633268af2c 100644
--- a/extern/Eigen3/Eigen/src/Geometry/EulerAngles.h
+++ b/extern/Eigen3/Eigen/src/Geometry/EulerAngles.h
@@ -33,12 +33,12 @@ namespace Eigen {
   * \sa class AngleAxis
   */
 template<typename Derived>
-inline Matrix<typename MatrixBase<Derived>::Scalar,3,1>
+EIGEN_DEVICE_FUNC inline Matrix<typename MatrixBase<Derived>::Scalar,3,1>
 MatrixBase<Derived>::eulerAngles(Index a0, Index a1, Index a2) const
 {
-  using std::atan2;
-  using std::sin;
-  using std::cos;
+  EIGEN_USING_STD_MATH(atan2)
+  EIGEN_USING_STD_MATH(sin)
+  EIGEN_USING_STD_MATH(cos)
   /* Implemented from Graphics Gems IV */
   EIGEN_STATIC_ASSERT_MATRIX_SPECIFIC_SIZE(Derived,3,3)
 
@@ -55,7 +55,12 @@ MatrixBase<Derived>::eulerAngles(Index a0, Index a1, Index a2) const
     res[0] = atan2(coeff(j,i), coeff(k,i));
     if((odd && res[0]<Scalar(0)) || ((!odd) && res[0]>Scalar(0)))
     {
-      res[0] = (res[0] > Scalar(0)) ? res[0] - Scalar(M_PI) : res[0] + Scalar(M_PI);
+      if(res[0] > Scalar(0)) {
+        res[0] -= Scalar(EIGEN_PI);
+      }
+      else {
+        res[0] += Scalar(EIGEN_PI);
+      }
       Scalar s2 = Vector2(coeff(j,i), coeff(k,i)).norm();
       res[1] = -atan2(s2, coeff(i,i));
     }
@@ -84,7 +89,12 @@ MatrixBase<Derived>::eulerAngles(Index a0, Index a1, Index a2) const
     res[0] = atan2(coeff(j,k), coeff(k,k));
     Scalar c2 = Vector2(coeff(i,i), coeff(i,j)).norm();
     if((odd && res[0]<Scalar(0)) || ((!odd) && res[0]>Scalar(0))) {
-      res[0] = (res[0] > Scalar(0)) ? res[0] - Scalar(M_PI) : res[0] + Scalar(M_PI);
+      if(res[0] > Scalar(0)) {
+        res[0] -= Scalar(EIGEN_PI);
+      }
+      else {
+        res[0] += Scalar(EIGEN_PI);
+      }
       res[1] = atan2(-coeff(i,k), -c2);
     }
     else
diff --git a/extern/Eigen3/Eigen/src/Geometry/Homogeneous.h b/extern/Eigen3/Eigen/src/Geometry/Homogeneous.h
index 372e422b92c..5f0da1a9e86 100644
--- a/extern/Eigen3/Eigen/src/Geometry/Homogeneous.h
+++ b/extern/Eigen3/Eigen/src/Geometry/Homogeneous.h
@@ -34,7 +34,7 @@ struct traits<Homogeneous<MatrixType,Direction> >
  : traits<MatrixType>
 {
   typedef typename traits<MatrixType>::StorageKind StorageKind;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsPlusOne = (MatrixType::RowsAtCompileTime != Dynamic) ?
@@ -48,8 +48,7 @@ struct traits<Homogeneous<MatrixType,Direction> >
     TmpFlags = _MatrixTypeNested::Flags & HereditaryBits,
     Flags = ColsAtCompileTime==1 ? (TmpFlags & ~RowMajorBit)
           : RowsAtCompileTime==1 ? (TmpFlags | RowMajorBit)
-          : TmpFlags,
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost
+          : TmpFlags
   };
 };
 
@@ -59,102 +58,117 @@ template<typename MatrixType,typename Rhs> struct homogeneous_right_product_impl
 } // end namespace internal
 
 template<typename MatrixType,int _Direction> class Homogeneous
-  : internal::no_assignment_operator, public MatrixBase<Homogeneous<MatrixType,_Direction> >
+  : public MatrixBase<Homogeneous<MatrixType,_Direction> >, internal::no_assignment_operator
 {
   public:
 
+    typedef MatrixType NestedExpression;
     enum { Direction = _Direction };
 
     typedef MatrixBase<Homogeneous> Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Homogeneous)
 
-    inline Homogeneous(const MatrixType& matrix)
+    EIGEN_DEVICE_FUNC explicit inline Homogeneous(const MatrixType& matrix)
       : m_matrix(matrix)
     {}
 
-    inline Index rows() const { return m_matrix.rows() + (int(Direction)==Vertical   ? 1 : 0); }
-    inline Index cols() const { return m_matrix.cols() + (int(Direction)==Horizontal ? 1 : 0); }
-
-    inline Scalar coeff(Index row, Index col) const
-    {
-      if(  (int(Direction)==Vertical   && row==m_matrix.rows())
-        || (int(Direction)==Horizontal && col==m_matrix.cols()))
-        return Scalar(1);
-      return m_matrix.coeff(row, col);
-    }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_matrix.rows() + (int(Direction)==Vertical   ? 1 : 0); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_matrix.cols() + (int(Direction)==Horizontal ? 1 : 0); }
+    
+    EIGEN_DEVICE_FUNC const NestedExpression& nestedExpression() const { return m_matrix; }
 
     template<typename Rhs>
-    inline const internal::homogeneous_right_product_impl<Homogeneous,Rhs>
+    EIGEN_DEVICE_FUNC inline const Product<Homogeneous,Rhs>
     operator* (const MatrixBase<Rhs>& rhs) const
     {
       eigen_assert(int(Direction)==Horizontal);
-      return internal::homogeneous_right_product_impl<Homogeneous,Rhs>(m_matrix,rhs.derived());
+      return Product<Homogeneous,Rhs>(*this,rhs.derived());
     }
 
     template<typename Lhs> friend
-    inline const internal::homogeneous_left_product_impl<Homogeneous,Lhs>
+    EIGEN_DEVICE_FUNC inline const Product<Lhs,Homogeneous>
     operator* (const MatrixBase<Lhs>& lhs, const Homogeneous& rhs)
     {
       eigen_assert(int(Direction)==Vertical);
-      return internal::homogeneous_left_product_impl<Homogeneous,Lhs>(lhs.derived(),rhs.m_matrix);
+      return Product<Lhs,Homogeneous>(lhs.derived(),rhs);
     }
 
     template<typename Scalar, int Dim, int Mode, int Options> friend
-    inline const internal::homogeneous_left_product_impl<Homogeneous,Transform<Scalar,Dim,Mode,Options> >
+    EIGEN_DEVICE_FUNC inline const Product<Transform<Scalar,Dim,Mode,Options>, Homogeneous >
     operator* (const Transform<Scalar,Dim,Mode,Options>& lhs, const Homogeneous& rhs)
     {
       eigen_assert(int(Direction)==Vertical);
-      return internal::homogeneous_left_product_impl<Homogeneous,Transform<Scalar,Dim,Mode,Options> >(lhs,rhs.m_matrix);
+      return Product<Transform<Scalar,Dim,Mode,Options>, Homogeneous>(lhs,rhs);
+    }
+
+    template<typename Func>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::result_of<Func(Scalar,Scalar)>::type
+    redux(const Func& func) const
+    {
+      return func(m_matrix.redux(func), Scalar(1));
     }
 
   protected:
     typename MatrixType::Nested m_matrix;
 };
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
+  *
+  * \returns a vector expression that is one longer than the vector argument, with the value 1 symbolically appended as the last coefficient.
   *
-  * \return an expression of the equivalent homogeneous vector
+  * This can be used to convert affine coordinates to homogeneous coordinates.
   *
   * \only_for_vectors
   *
   * Example: \include MatrixBase_homogeneous.cpp
   * Output: \verbinclude MatrixBase_homogeneous.out
   *
-  * \sa class Homogeneous
+  * \sa VectorwiseOp::homogeneous(), class Homogeneous
   */
 template<typename Derived>
-inline typename MatrixBase<Derived>::HomogeneousReturnType
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::HomogeneousReturnType
 MatrixBase<Derived>::homogeneous() const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  return derived();
+  return HomogeneousReturnType(derived());
 }
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
-  * \returns a matrix expression of homogeneous column (or row) vectors
+  * \returns an expression where the value 1 is symbolically appended as the final coefficient to each column (or row) of the matrix.
+  *
+  * This can be used to convert affine coordinates to homogeneous coordinates.
   *
   * Example: \include VectorwiseOp_homogeneous.cpp
   * Output: \verbinclude VectorwiseOp_homogeneous.out
   *
-  * \sa MatrixBase::homogeneous() */
+  * \sa MatrixBase::homogeneous(), class Homogeneous */
 template<typename ExpressionType, int Direction>
-inline Homogeneous<ExpressionType,Direction>
+EIGEN_DEVICE_FUNC inline Homogeneous<ExpressionType,Direction>
 VectorwiseOp<ExpressionType,Direction>::homogeneous() const
 {
-  return _expression();
+  return HomogeneousReturnType(_expression());
 }
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
-  * \returns an expression of the homogeneous normalized vector of \c *this
+  * \brief homogeneous normalization
+  *
+  * \returns a vector expression of the N-1 first coefficients of \c *this divided by that last coefficient.
+  *
+  * This can be used to convert homogeneous coordinates to affine coordinates.
+  *
+  * It is essentially a shortcut for:
+  * \code
+    this->head(this->size()-1)/this->coeff(this->size()-1);
+    \endcode
   *
   * Example: \include MatrixBase_hnormalized.cpp
   * Output: \verbinclude MatrixBase_hnormalized.out
   *
   * \sa VectorwiseOp::hnormalized() */
 template<typename Derived>
-inline const typename MatrixBase<Derived>::HNormalizedReturnType
+EIGEN_DEVICE_FUNC inline const typename MatrixBase<Derived>::HNormalizedReturnType
 MatrixBase<Derived>::hnormalized() const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
@@ -163,16 +177,22 @@ MatrixBase<Derived>::hnormalized() const
     ColsAtCompileTime==1?1:size()-1) / coeff(size()-1);
 }
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
+  *
+  * \brief column or row-wise homogeneous normalization
   *
-  * \returns an expression of the homogeneous normalized vector of \c *this
+  * \returns an expression of the first N-1 coefficients of each column (or row) of \c *this divided by the last coefficient of each column (or row).
+  *
+  * This can be used to convert homogeneous coordinates to affine coordinates.
+  *
+  * It is conceptually equivalent to calling MatrixBase::hnormalized() to each column (or row) of \c *this.
   *
   * Example: \include DirectionWise_hnormalized.cpp
   * Output: \verbinclude DirectionWise_hnormalized.out
   *
   * \sa MatrixBase::hnormalized() */
 template<typename ExpressionType, int Direction>
-inline const typename VectorwiseOp<ExpressionType,Direction>::HNormalizedReturnType
+EIGEN_DEVICE_FUNC inline const typename VectorwiseOp<ExpressionType,Direction>::HNormalizedReturnType
 VectorwiseOp<ExpressionType,Direction>::hnormalized() const
 {
   return HNormalized_Block(_expression(),0,0,
@@ -196,7 +216,7 @@ template<typename MatrixOrTransformType>
 struct take_matrix_for_product
 {
   typedef MatrixOrTransformType type;
-  static const type& run(const type &x) { return x; }
+  EIGEN_DEVICE_FUNC static const type& run(const type &x) { return x; }
 };
 
 template<typename Scalar, int Dim, int Mode,int Options>
@@ -204,7 +224,7 @@ struct take_matrix_for_product<Transform<Scalar, Dim, Mode, Options> >
 {
   typedef Transform<Scalar, Dim, Mode, Options> TransformType;
   typedef typename internal::add_const<typename TransformType::ConstAffinePart>::type type;
-  static type run (const TransformType& x) { return x.affine(); }
+  EIGEN_DEVICE_FUNC static type run (const TransformType& x) { return x.affine(); }
 };
 
 template<typename Scalar, int Dim, int Options>
@@ -212,7 +232,7 @@ struct take_matrix_for_product<Transform<Scalar, Dim, Projective, Options> >
 {
   typedef Transform<Scalar, Dim, Projective, Options> TransformType;
   typedef typename TransformType::MatrixType type;
-  static const type& run (const TransformType& x) { return x.matrix(); }
+  EIGEN_DEVICE_FUNC static const type& run (const TransformType& x) { return x.matrix(); }
 };
 
 template<typename MatrixType,typename Lhs>
@@ -237,16 +257,15 @@ struct homogeneous_left_product_impl<Homogeneous<MatrixType,Vertical>,Lhs>
   typedef typename traits<homogeneous_left_product_impl>::LhsMatrixType LhsMatrixType;
   typedef typename remove_all<LhsMatrixType>::type LhsMatrixTypeCleaned;
   typedef typename remove_all<typename LhsMatrixTypeCleaned::Nested>::type LhsMatrixTypeNested;
-  typedef typename MatrixType::Index Index;
-  homogeneous_left_product_impl(const Lhs& lhs, const MatrixType& rhs)
+  EIGEN_DEVICE_FUNC homogeneous_left_product_impl(const Lhs& lhs, const MatrixType& rhs)
     : m_lhs(take_matrix_for_product<Lhs>::run(lhs)),
       m_rhs(rhs)
   {}
 
-  inline Index rows() const { return m_lhs.rows(); }
-  inline Index cols() const { return m_rhs.cols(); }
+  EIGEN_DEVICE_FUNC inline Index rows() const { return m_lhs.rows(); }
+  EIGEN_DEVICE_FUNC inline Index cols() const { return m_rhs.cols(); }
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  template<typename Dest> EIGEN_DEVICE_FUNC void evalTo(Dest& dst) const
   {
     // FIXME investigate how to allow lazy evaluation of this product when possible
     dst = Block<const LhsMatrixTypeNested,
@@ -277,15 +296,14 @@ struct homogeneous_right_product_impl<Homogeneous<MatrixType,Horizontal>,Rhs>
   : public ReturnByValue<homogeneous_right_product_impl<Homogeneous<MatrixType,Horizontal>,Rhs> >
 {
   typedef typename remove_all<typename Rhs::Nested>::type RhsNested;
-  typedef typename MatrixType::Index Index;
-  homogeneous_right_product_impl(const MatrixType& lhs, const Rhs& rhs)
+  EIGEN_DEVICE_FUNC homogeneous_right_product_impl(const MatrixType& lhs, const Rhs& rhs)
     : m_lhs(lhs), m_rhs(rhs)
   {}
 
-  inline Index rows() const { return m_lhs.rows(); }
-  inline Index cols() const { return m_rhs.cols(); }
+  EIGEN_DEVICE_FUNC inline Index rows() const { return m_lhs.rows(); }
+  EIGEN_DEVICE_FUNC inline Index cols() const { return m_rhs.cols(); }
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  template<typename Dest> EIGEN_DEVICE_FUNC void evalTo(Dest& dst) const
   {
     // FIXME investigate how to allow lazy evaluation of this product when possible
     dst = m_lhs * Block<const RhsNested,
@@ -300,6 +318,178 @@ struct homogeneous_right_product_impl<Homogeneous<MatrixType,Horizontal>,Rhs>
   typename Rhs::Nested m_rhs;
 };
 
+template<typename ArgType,int Direction>
+struct evaluator_traits<Homogeneous<ArgType,Direction> >
+{
+  typedef typename storage_kind_to_evaluator_kind<typename ArgType::StorageKind>::Kind Kind;
+  typedef HomogeneousShape Shape;  
+};
+
+template<> struct AssignmentKind<DenseShape,HomogeneousShape> { typedef Dense2Dense Kind; };
+
+
+template<typename ArgType,int Direction>
+struct unary_evaluator<Homogeneous<ArgType,Direction>, IndexBased>
+  : evaluator<typename Homogeneous<ArgType,Direction>::PlainObject >
+{
+  typedef Homogeneous<ArgType,Direction> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& op)
+    : Base(), m_temp(op)
+  {
+    ::new (static_cast<Base*>(this)) Base(m_temp);
+  }
+
+protected:
+  PlainObject m_temp;
+};
+
+// dense = homogeneous
+template< typename DstXprType, typename ArgType, typename Scalar>
+struct Assignment<DstXprType, Homogeneous<ArgType,Vertical>, internal::assign_op<Scalar,typename ArgType::Scalar>, Dense2Dense>
+{
+  typedef Homogeneous<ArgType,Vertical> SrcXprType;
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename ArgType::Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    dst.template topRows<ArgType::RowsAtCompileTime>(src.nestedExpression().rows()) = src.nestedExpression();
+    dst.row(dst.rows()-1).setOnes();
+  }
+};
+
+// dense = homogeneous
+template< typename DstXprType, typename ArgType, typename Scalar>
+struct Assignment<DstXprType, Homogeneous<ArgType,Horizontal>, internal::assign_op<Scalar,typename ArgType::Scalar>, Dense2Dense>
+{
+  typedef Homogeneous<ArgType,Horizontal> SrcXprType;
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename ArgType::Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    dst.template leftCols<ArgType::ColsAtCompileTime>(src.nestedExpression().cols()) = src.nestedExpression();
+    dst.col(dst.cols()-1).setOnes();
+  }
+};
+
+template<typename LhsArg, typename Rhs, int ProductTag>
+struct generic_product_impl<Homogeneous<LhsArg,Horizontal>, Rhs, HomogeneousShape, DenseShape, ProductTag>
+{
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC static void evalTo(Dest& dst, const Homogeneous<LhsArg,Horizontal>& lhs, const Rhs& rhs)
+  {
+    homogeneous_right_product_impl<Homogeneous<LhsArg,Horizontal>, Rhs>(lhs.nestedExpression(), rhs).evalTo(dst);
+  }
+};
+
+template<typename Lhs,typename Rhs>
+struct homogeneous_right_product_refactoring_helper
+{
+  enum {
+    Dim  = Lhs::ColsAtCompileTime,
+    Rows = Lhs::RowsAtCompileTime
+  };
+  typedef typename Rhs::template ConstNRowsBlockXpr<Dim>::Type          LinearBlockConst;
+  typedef typename remove_const<LinearBlockConst>::type                 LinearBlock;
+  typedef typename Rhs::ConstRowXpr                                     ConstantColumn;
+  typedef Replicate<const ConstantColumn,Rows,1>                        ConstantBlock;
+  typedef Product<Lhs,LinearBlock,LazyProduct>                          LinearProduct;
+  typedef CwiseBinaryOp<internal::scalar_sum_op<typename Lhs::Scalar,typename Rhs::Scalar>, const LinearProduct, const ConstantBlock> Xpr;
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, HomogeneousShape, DenseShape>
+ : public evaluator<typename homogeneous_right_product_refactoring_helper<typename Lhs::NestedExpression,Rhs>::Xpr>
+{
+  typedef Product<Lhs, Rhs, LazyProduct> XprType;
+  typedef homogeneous_right_product_refactoring_helper<typename Lhs::NestedExpression,Rhs> helper;
+  typedef typename helper::ConstantBlock ConstantBlock;
+  typedef typename helper::Xpr RefactoredXpr;
+  typedef evaluator<RefactoredXpr> Base;
+  
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(  xpr.lhs().nestedExpression() .lazyProduct(  xpr.rhs().template topRows<helper::Dim>(xpr.lhs().nestedExpression().cols()) )
+            + ConstantBlock(xpr.rhs().row(xpr.rhs().rows()-1),xpr.lhs().rows(), 1) )
+  {}
+};
+
+template<typename Lhs, typename RhsArg, int ProductTag>
+struct generic_product_impl<Lhs, Homogeneous<RhsArg,Vertical>, DenseShape, HomogeneousShape, ProductTag>
+{
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC static void evalTo(Dest& dst, const Lhs& lhs, const Homogeneous<RhsArg,Vertical>& rhs)
+  {
+    homogeneous_left_product_impl<Homogeneous<RhsArg,Vertical>, Lhs>(lhs, rhs.nestedExpression()).evalTo(dst);
+  }
+};
+
+// TODO: the following specialization is to address a regression from 3.2 to 3.3
+// In the future, this path should be optimized.
+template<typename Lhs, typename RhsArg, int ProductTag>
+struct generic_product_impl<Lhs, Homogeneous<RhsArg,Vertical>, TriangularShape, HomogeneousShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Homogeneous<RhsArg,Vertical>& rhs)
+  {
+    dst.noalias() = lhs * rhs.eval();
+  }
+};
+
+template<typename Lhs,typename Rhs>
+struct homogeneous_left_product_refactoring_helper
+{
+  enum {
+    Dim = Rhs::RowsAtCompileTime,
+    Cols = Rhs::ColsAtCompileTime
+  };
+  typedef typename Lhs::template ConstNColsBlockXpr<Dim>::Type          LinearBlockConst;
+  typedef typename remove_const<LinearBlockConst>::type                 LinearBlock;
+  typedef typename Lhs::ConstColXpr                                     ConstantColumn;
+  typedef Replicate<const ConstantColumn,1,Cols>                        ConstantBlock;
+  typedef Product<LinearBlock,Rhs,LazyProduct>                          LinearProduct;
+  typedef CwiseBinaryOp<internal::scalar_sum_op<typename Lhs::Scalar,typename Rhs::Scalar>, const LinearProduct, const ConstantBlock> Xpr;
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape, HomogeneousShape>
+ : public evaluator<typename homogeneous_left_product_refactoring_helper<Lhs,typename Rhs::NestedExpression>::Xpr>
+{
+  typedef Product<Lhs, Rhs, LazyProduct> XprType;
+  typedef homogeneous_left_product_refactoring_helper<Lhs,typename Rhs::NestedExpression> helper;
+  typedef typename helper::ConstantBlock ConstantBlock;
+  typedef typename helper::Xpr RefactoredXpr;
+  typedef evaluator<RefactoredXpr> Base;
+  
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(   xpr.lhs().template leftCols<helper::Dim>(xpr.rhs().nestedExpression().rows()) .lazyProduct( xpr.rhs().nestedExpression() )
+            + ConstantBlock(xpr.lhs().col(xpr.lhs().cols()-1),1,xpr.rhs().cols()) )
+  {}
+};
+
+template<typename Scalar, int Dim, int Mode,int Options, typename RhsArg, int ProductTag>
+struct generic_product_impl<Transform<Scalar,Dim,Mode,Options>, Homogeneous<RhsArg,Vertical>, DenseShape, HomogeneousShape, ProductTag>
+{
+  typedef Transform<Scalar,Dim,Mode,Options> TransformType;
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC static void evalTo(Dest& dst, const TransformType& lhs, const Homogeneous<RhsArg,Vertical>& rhs)
+  {
+    homogeneous_left_product_impl<Homogeneous<RhsArg,Vertical>, TransformType>(lhs, rhs.nestedExpression()).evalTo(dst);
+  }
+};
+
+template<typename ExpressionType, int Side, bool Transposed>
+struct permutation_matrix_product<ExpressionType, Side, Transposed, HomogeneousShape>
+  : public permutation_matrix_product<ExpressionType, Side, Transposed, DenseShape>
+{};
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Geometry/Hyperplane.h b/extern/Eigen3/Eigen/src/Geometry/Hyperplane.h
index 00b7c4300fd..05929b29941 100644
--- a/extern/Eigen3/Eigen/src/Geometry/Hyperplane.h
+++ b/extern/Eigen3/Eigen/src/Geometry/Hyperplane.h
@@ -22,8 +22,8 @@ namespace Eigen {
   * A hyperplane is an affine subspace of dimension n-1 in a space of dimension n.
   * For example, a hyperplane in a plane is a line; a hyperplane in 3-space is a plane.
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
-  * \param _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients
+  * \tparam _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
   *             Notice that the dimension of the hyperplane is _AmbientDim-1.
   *
   * This class represents an hyperplane as the zero set of the implicit equation
@@ -41,7 +41,7 @@ public:
   };
   typedef _Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef DenseIndex Index;
+  typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
   typedef Matrix<Scalar,AmbientDimAtCompileTime,1> VectorType;
   typedef Matrix<Scalar,Index(AmbientDimAtCompileTime)==Dynamic
                         ? Dynamic
@@ -50,21 +50,21 @@ public:
   typedef const Block<const Coefficients,AmbientDimAtCompileTime,1> ConstNormalReturnType;
 
   /** Default constructor without initialization */
-  inline Hyperplane() {}
+  EIGEN_DEVICE_FUNC inline Hyperplane() {}
   
   template<int OtherOptions>
-  Hyperplane(const Hyperplane<Scalar,AmbientDimAtCompileTime,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC Hyperplane(const Hyperplane<Scalar,AmbientDimAtCompileTime,OtherOptions>& other)
    : m_coeffs(other.coeffs())
   {}
 
   /** Constructs a dynamic-size hyperplane with \a _dim the dimension
     * of the ambient space */
-  inline explicit Hyperplane(Index _dim) : m_coeffs(_dim+1) {}
+  EIGEN_DEVICE_FUNC inline explicit Hyperplane(Index _dim) : m_coeffs(_dim+1) {}
 
   /** Construct a plane from its normal \a n and a point \a e onto the plane.
     * \warning the vector normal is assumed to be normalized.
     */
-  inline Hyperplane(const VectorType& n, const VectorType& e)
+  EIGEN_DEVICE_FUNC inline Hyperplane(const VectorType& n, const VectorType& e)
     : m_coeffs(n.size()+1)
   {
     normal() = n;
@@ -75,7 +75,7 @@ public:
     * such that the algebraic equation of the plane is \f$ n \cdot x + d = 0 \f$.
     * \warning the vector normal is assumed to be normalized.
     */
-  inline Hyperplane(const VectorType& n, const Scalar& d)
+  EIGEN_DEVICE_FUNC inline Hyperplane(const VectorType& n, const Scalar& d)
     : m_coeffs(n.size()+1)
   {
     normal() = n;
@@ -85,7 +85,7 @@ public:
   /** Constructs a hyperplane passing through the two points. If the dimension of the ambient space
     * is greater than 2, then there isn't uniqueness, so an arbitrary choice is made.
     */
-  static inline Hyperplane Through(const VectorType& p0, const VectorType& p1)
+  EIGEN_DEVICE_FUNC static inline Hyperplane Through(const VectorType& p0, const VectorType& p1)
   {
     Hyperplane result(p0.size());
     result.normal() = (p1 - p0).unitOrthogonal();
@@ -96,7 +96,7 @@ public:
   /** Constructs a hyperplane passing through the three points. The dimension of the ambient space
     * is required to be exactly 3.
     */
-  static inline Hyperplane Through(const VectorType& p0, const VectorType& p1, const VectorType& p2)
+  EIGEN_DEVICE_FUNC static inline Hyperplane Through(const VectorType& p0, const VectorType& p1, const VectorType& p2)
   {
     EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(VectorType, 3)
     Hyperplane result(p0.size());
@@ -120,19 +120,19 @@ public:
     * so an arbitrary choice is made.
     */
   // FIXME to be consitent with the rest this could be implemented as a static Through function ??
-  explicit Hyperplane(const ParametrizedLine<Scalar, AmbientDimAtCompileTime>& parametrized)
+  EIGEN_DEVICE_FUNC explicit Hyperplane(const ParametrizedLine<Scalar, AmbientDimAtCompileTime>& parametrized)
   {
     normal() = parametrized.direction().unitOrthogonal();
     offset() = -parametrized.origin().dot(normal());
   }
 
-  ~Hyperplane() {}
+  EIGEN_DEVICE_FUNC ~Hyperplane() {}
 
   /** \returns the dimension in which the plane holds */
-  inline Index dim() const { return AmbientDimAtCompileTime==Dynamic ? m_coeffs.size()-1 : Index(AmbientDimAtCompileTime); }
+  EIGEN_DEVICE_FUNC inline Index dim() const { return AmbientDimAtCompileTime==Dynamic ? m_coeffs.size()-1 : Index(AmbientDimAtCompileTime); }
 
   /** normalizes \c *this */
-  void normalize(void)
+  EIGEN_DEVICE_FUNC void normalize(void)
   {
     m_coeffs /= normal().norm();
   }
@@ -140,45 +140,45 @@ public:
   /** \returns the signed distance between the plane \c *this and a point \a p.
     * \sa absDistance()
     */
-  inline Scalar signedDistance(const VectorType& p) const { return normal().dot(p) + offset(); }
+  EIGEN_DEVICE_FUNC inline Scalar signedDistance(const VectorType& p) const { return normal().dot(p) + offset(); }
 
   /** \returns the absolute distance between the plane \c *this and a point \a p.
     * \sa signedDistance()
     */
-  inline Scalar absDistance(const VectorType& p) const { using std::abs; return abs(signedDistance(p)); }
+  EIGEN_DEVICE_FUNC inline Scalar absDistance(const VectorType& p) const { return numext::abs(signedDistance(p)); }
 
   /** \returns the projection of a point \a p onto the plane \c *this.
     */
-  inline VectorType projection(const VectorType& p) const { return p - signedDistance(p) * normal(); }
+  EIGEN_DEVICE_FUNC inline VectorType projection(const VectorType& p) const { return p - signedDistance(p) * normal(); }
 
   /** \returns a constant reference to the unit normal vector of the plane, which corresponds
     * to the linear part of the implicit equation.
     */
-  inline ConstNormalReturnType normal() const { return ConstNormalReturnType(m_coeffs,0,0,dim(),1); }
+  EIGEN_DEVICE_FUNC inline ConstNormalReturnType normal() const { return ConstNormalReturnType(m_coeffs,0,0,dim(),1); }
 
   /** \returns a non-constant reference to the unit normal vector of the plane, which corresponds
     * to the linear part of the implicit equation.
     */
-  inline NormalReturnType normal() { return NormalReturnType(m_coeffs,0,0,dim(),1); }
+  EIGEN_DEVICE_FUNC inline NormalReturnType normal() { return NormalReturnType(m_coeffs,0,0,dim(),1); }
 
   /** \returns the distance to the origin, which is also the "constant term" of the implicit equation
     * \warning the vector normal is assumed to be normalized.
     */
-  inline const Scalar& offset() const { return m_coeffs.coeff(dim()); }
+  EIGEN_DEVICE_FUNC inline const Scalar& offset() const { return m_coeffs.coeff(dim()); }
 
   /** \returns a non-constant reference to the distance to the origin, which is also the constant part
     * of the implicit equation */
-  inline Scalar& offset() { return m_coeffs(dim()); }
+  EIGEN_DEVICE_FUNC inline Scalar& offset() { return m_coeffs(dim()); }
 
   /** \returns a constant reference to the coefficients c_i of the plane equation:
     * \f$ c_0*x_0 + ... + c_{d-1}*x_{d-1} + c_d = 0 \f$
     */
-  inline const Coefficients& coeffs() const { return m_coeffs; }
+  EIGEN_DEVICE_FUNC inline const Coefficients& coeffs() const { return m_coeffs; }
 
   /** \returns a non-constant reference to the coefficients c_i of the plane equation:
     * \f$ c_0*x_0 + ... + c_{d-1}*x_{d-1} + c_d = 0 \f$
     */
-  inline Coefficients& coeffs() { return m_coeffs; }
+  EIGEN_DEVICE_FUNC inline Coefficients& coeffs() { return m_coeffs; }
 
   /** \returns the intersection of *this with \a other.
     *
@@ -186,16 +186,15 @@ public:
     *
     * \note If \a other is approximately parallel to *this, this method will return any point on *this.
     */
-  VectorType intersection(const Hyperplane& other) const
+  EIGEN_DEVICE_FUNC VectorType intersection(const Hyperplane& other) const
   {
-    using std::abs;
     EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(VectorType, 2)
     Scalar det = coeffs().coeff(0) * other.coeffs().coeff(1) - coeffs().coeff(1) * other.coeffs().coeff(0);
     // since the line equations ax+by=c are normalized with a^2+b^2=1, the following tests
     // whether the two lines are approximately parallel.
     if(internal::isMuchSmallerThan(det, Scalar(1)))
     {   // special case where the two lines are approximately parallel. Pick any point on the first line.
-        if(abs(coeffs().coeff(1))>abs(coeffs().coeff(0)))
+        if(numext::abs(coeffs().coeff(1))>numext::abs(coeffs().coeff(0)))
             return VectorType(coeffs().coeff(1), -coeffs().coeff(2)/coeffs().coeff(1)-coeffs().coeff(0));
         else
             return VectorType(-coeffs().coeff(2)/coeffs().coeff(0)-coeffs().coeff(1), coeffs().coeff(0));
@@ -215,10 +214,13 @@ public:
     *               or a more generic #Affine transformation. The default is #Affine.
     */
   template<typename XprType>
-  inline Hyperplane& transform(const MatrixBase<XprType>& mat, TransformTraits traits = Affine)
+  EIGEN_DEVICE_FUNC inline Hyperplane& transform(const MatrixBase<XprType>& mat, TransformTraits traits = Affine)
   {
     if (traits==Affine)
+    {
       normal() = mat.inverse().transpose() * normal();
+      m_coeffs /= normal().norm();
+    }
     else if (traits==Isometry)
       normal() = mat * normal();
     else
@@ -236,7 +238,7 @@ public:
     *               Other kind of transformations are not supported.
     */
   template<int TrOptions>
-  inline Hyperplane& transform(const Transform<Scalar,AmbientDimAtCompileTime,Affine,TrOptions>& t,
+  EIGEN_DEVICE_FUNC inline Hyperplane& transform(const Transform<Scalar,AmbientDimAtCompileTime,Affine,TrOptions>& t,
                                 TransformTraits traits = Affine)
   {
     transform(t.linear(), traits);
@@ -250,7 +252,7 @@ public:
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Hyperplane,
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Hyperplane,
            Hyperplane<NewScalarType,AmbientDimAtCompileTime,Options> >::type cast() const
   {
     return typename internal::cast_return_type<Hyperplane,
@@ -259,7 +261,7 @@ public:
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType,int OtherOptions>
-  inline explicit Hyperplane(const Hyperplane<OtherScalarType,AmbientDimAtCompileTime,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC inline explicit Hyperplane(const Hyperplane<OtherScalarType,AmbientDimAtCompileTime,OtherOptions>& other)
   { m_coeffs = other.coeffs().template cast<Scalar>(); }
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
@@ -267,7 +269,7 @@ public:
     *
     * \sa MatrixBase::isApprox() */
   template<int OtherOptions>
-  bool isApprox(const Hyperplane<Scalar,AmbientDimAtCompileTime,OtherOptions>& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const Hyperplane<Scalar,AmbientDimAtCompileTime,OtherOptions>& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_coeffs.isApprox(other.m_coeffs, prec); }
 
 protected:
diff --git a/extern/Eigen3/Eigen/src/Geometry/OrthoMethods.h b/extern/Eigen3/Eigen/src/Geometry/OrthoMethods.h
index 556bc81604e..a035e6310a7 100644
--- a/extern/Eigen3/Eigen/src/Geometry/OrthoMethods.h
+++ b/extern/Eigen3/Eigen/src/Geometry/OrthoMethods.h
@@ -13,16 +13,24 @@
 
 namespace Eigen { 
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
   * \returns the cross product of \c *this and \a other
   *
   * Here is a very good explanation of cross-product: http://xkcd.com/199/
+  * 
+  * With complex numbers, the cross product is implemented as
+  * \f$ (\mathbf{a}+i\mathbf{b}) \times (\mathbf{c}+i\mathbf{d}) = (\mathbf{a} \times \mathbf{c} - \mathbf{b} \times \mathbf{d}) - i(\mathbf{a} \times \mathbf{d} - \mathbf{b} \times \mathbf{c})\f$
+  * 
   * \sa MatrixBase::cross3()
   */
 template<typename Derived>
 template<typename OtherDerived>
-inline typename MatrixBase<Derived>::template cross_product_return_type<OtherDerived>::type
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::template cross_product_return_type<OtherDerived>::type
+#else
+inline typename MatrixBase<Derived>::PlainObject
+#endif
 MatrixBase<Derived>::cross(const MatrixBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Derived,3)
@@ -30,8 +38,8 @@ MatrixBase<Derived>::cross(const MatrixBase<OtherDerived>& other) const
 
   // Note that there is no need for an expression here since the compiler
   // optimize such a small temporary very well (even within a complex expression)
-  typename internal::nested<Derived,2>::type lhs(derived());
-  typename internal::nested<OtherDerived,2>::type rhs(other.derived());
+  typename internal::nested_eval<Derived,2>::type lhs(derived());
+  typename internal::nested_eval<OtherDerived,2>::type rhs(other.derived());
   return typename cross_product_return_type<OtherDerived>::type(
     numext::conj(lhs.coeff(1) * rhs.coeff(2) - lhs.coeff(2) * rhs.coeff(1)),
     numext::conj(lhs.coeff(2) * rhs.coeff(0) - lhs.coeff(0) * rhs.coeff(2)),
@@ -45,7 +53,7 @@ template< int Arch,typename VectorLhs,typename VectorRhs,
           typename Scalar = typename VectorLhs::Scalar,
           bool Vectorizable = bool((VectorLhs::Flags&VectorRhs::Flags)&PacketAccessBit)>
 struct cross3_impl {
-  static inline typename internal::plain_matrix_type<VectorLhs>::type
+  EIGEN_DEVICE_FUNC static inline typename internal::plain_matrix_type<VectorLhs>::type
   run(const VectorLhs& lhs, const VectorRhs& rhs)
   {
     return typename internal::plain_matrix_type<VectorLhs>::type(
@@ -59,7 +67,7 @@ struct cross3_impl {
 
 }
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
   * \returns the cross product of \c *this and \a other using only the x, y, and z coefficients
   *
@@ -70,14 +78,14 @@ struct cross3_impl {
   */
 template<typename Derived>
 template<typename OtherDerived>
-inline typename MatrixBase<Derived>::PlainObject
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::PlainObject
 MatrixBase<Derived>::cross3(const MatrixBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Derived,4)
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,4)
 
-  typedef typename internal::nested<Derived,2>::type DerivedNested;
-  typedef typename internal::nested<OtherDerived,2>::type OtherDerivedNested;
+  typedef typename internal::nested_eval<Derived,2>::type DerivedNested;
+  typedef typename internal::nested_eval<OtherDerived,2>::type OtherDerivedNested;
   DerivedNested lhs(derived());
   OtherDerivedNested rhs(other.derived());
 
@@ -86,38 +94,42 @@ MatrixBase<Derived>::cross3(const MatrixBase<OtherDerived>& other) const
                         typename internal::remove_all<OtherDerivedNested>::type>::run(lhs,rhs);
 }
 
-/** \returns a matrix expression of the cross product of each column or row
+/** \geometry_module \ingroup Geometry_Module
+  *
+  * \returns a matrix expression of the cross product of each column or row
   * of the referenced expression with the \a other vector.
   *
   * The referenced matrix must have one dimension equal to 3.
   * The result matrix has the same dimensions than the referenced one.
   *
-  * \geometry_module
-  *
   * \sa MatrixBase::cross() */
 template<typename ExpressionType, int Direction>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC 
 const typename VectorwiseOp<ExpressionType,Direction>::CrossReturnType
 VectorwiseOp<ExpressionType,Direction>::cross(const MatrixBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,3)
   EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
     YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+  
+  typename internal::nested_eval<ExpressionType,2>::type mat(_expression());
+  typename internal::nested_eval<OtherDerived,2>::type vec(other.derived());
 
   CrossReturnType res(_expression().rows(),_expression().cols());
   if(Direction==Vertical)
   {
     eigen_assert(CrossReturnType::RowsAtCompileTime==3 && "the matrix must have exactly 3 rows");
-    res.row(0) = (_expression().row(1) * other.coeff(2) - _expression().row(2) * other.coeff(1)).conjugate();
-    res.row(1) = (_expression().row(2) * other.coeff(0) - _expression().row(0) * other.coeff(2)).conjugate();
-    res.row(2) = (_expression().row(0) * other.coeff(1) - _expression().row(1) * other.coeff(0)).conjugate();
+    res.row(0) = (mat.row(1) * vec.coeff(2) - mat.row(2) * vec.coeff(1)).conjugate();
+    res.row(1) = (mat.row(2) * vec.coeff(0) - mat.row(0) * vec.coeff(2)).conjugate();
+    res.row(2) = (mat.row(0) * vec.coeff(1) - mat.row(1) * vec.coeff(0)).conjugate();
   }
   else
   {
     eigen_assert(CrossReturnType::ColsAtCompileTime==3 && "the matrix must have exactly 3 columns");
-    res.col(0) = (_expression().col(1) * other.coeff(2) - _expression().col(2) * other.coeff(1)).conjugate();
-    res.col(1) = (_expression().col(2) * other.coeff(0) - _expression().col(0) * other.coeff(2)).conjugate();
-    res.col(2) = (_expression().col(0) * other.coeff(1) - _expression().col(1) * other.coeff(0)).conjugate();
+    res.col(0) = (mat.col(1) * vec.coeff(2) - mat.col(2) * vec.coeff(1)).conjugate();
+    res.col(1) = (mat.col(2) * vec.coeff(0) - mat.col(0) * vec.coeff(2)).conjugate();
+    res.col(2) = (mat.col(0) * vec.coeff(1) - mat.col(1) * vec.coeff(0)).conjugate();
   }
   return res;
 }
@@ -130,8 +142,8 @@ struct unitOrthogonal_selector
   typedef typename plain_matrix_type<Derived>::type VectorType;
   typedef typename traits<Derived>::Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef typename Derived::Index Index;
   typedef Matrix<Scalar,2,1> Vector2;
+  EIGEN_DEVICE_FUNC
   static inline VectorType run(const Derived& src)
   {
     VectorType perp = VectorType::Zero(src.size());
@@ -154,6 +166,7 @@ struct unitOrthogonal_selector<Derived,3>
   typedef typename plain_matrix_type<Derived>::type VectorType;
   typedef typename traits<Derived>::Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline VectorType run(const Derived& src)
   {
     VectorType perp;
@@ -192,13 +205,16 @@ template<typename Derived>
 struct unitOrthogonal_selector<Derived,2>
 {
   typedef typename plain_matrix_type<Derived>::type VectorType;
+  EIGEN_DEVICE_FUNC
   static inline VectorType run(const Derived& src)
   { return VectorType(-numext::conj(src.y()), numext::conj(src.x())).normalized(); }
 };
 
 } // end namespace internal
 
-/** \returns a unit vector which is orthogonal to \c *this
+/** \geometry_module \ingroup Geometry_Module
+  *
+  * \returns a unit vector which is orthogonal to \c *this
   *
   * The size of \c *this must be at least 2. If the size is exactly 2,
   * then the returned vector is a counter clock wise rotation of \c *this, i.e., (-y,x).normalized().
@@ -206,7 +222,7 @@ struct unitOrthogonal_selector<Derived,2>
   * \sa cross()
   */
 template<typename Derived>
-typename MatrixBase<Derived>::PlainObject
+EIGEN_DEVICE_FUNC typename MatrixBase<Derived>::PlainObject
 MatrixBase<Derived>::unitOrthogonal() const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
diff --git a/extern/Eigen3/Eigen/src/Geometry/ParametrizedLine.h b/extern/Eigen3/Eigen/src/Geometry/ParametrizedLine.h
index 77fa228e6a5..1e985d8cde2 100644
--- a/extern/Eigen3/Eigen/src/Geometry/ParametrizedLine.h
+++ b/extern/Eigen3/Eigen/src/Geometry/ParametrizedLine.h
@@ -23,8 +23,8 @@ namespace Eigen {
   * direction vector \f$ \mathbf{d} \f$ such that the line corresponds to
   * the set \f$ l(t) = \mathbf{o} + t \mathbf{d} \f$, \f$ t \in \mathbf{R} \f$.
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
-  * \param _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients
+  * \tparam _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 class ParametrizedLine
@@ -37,49 +37,49 @@ public:
   };
   typedef _Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef DenseIndex Index;
+  typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
   typedef Matrix<Scalar,AmbientDimAtCompileTime,1,Options> VectorType;
 
   /** Default constructor without initialization */
-  inline ParametrizedLine() {}
+  EIGEN_DEVICE_FUNC inline ParametrizedLine() {}
   
   template<int OtherOptions>
-  ParametrizedLine(const ParametrizedLine<Scalar,AmbientDimAtCompileTime,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC ParametrizedLine(const ParametrizedLine<Scalar,AmbientDimAtCompileTime,OtherOptions>& other)
    : m_origin(other.origin()), m_direction(other.direction())
   {}
 
   /** Constructs a dynamic-size line with \a _dim the dimension
     * of the ambient space */
-  inline explicit ParametrizedLine(Index _dim) : m_origin(_dim), m_direction(_dim) {}
+  EIGEN_DEVICE_FUNC inline explicit ParametrizedLine(Index _dim) : m_origin(_dim), m_direction(_dim) {}
 
   /** Initializes a parametrized line of direction \a direction and origin \a origin.
     * \warning the vector direction is assumed to be normalized.
     */
-  ParametrizedLine(const VectorType& origin, const VectorType& direction)
+  EIGEN_DEVICE_FUNC ParametrizedLine(const VectorType& origin, const VectorType& direction)
     : m_origin(origin), m_direction(direction) {}
 
   template <int OtherOptions>
-  explicit ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane);
+  EIGEN_DEVICE_FUNC explicit ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane);
 
   /** Constructs a parametrized line going from \a p0 to \a p1. */
-  static inline ParametrizedLine Through(const VectorType& p0, const VectorType& p1)
+  EIGEN_DEVICE_FUNC static inline ParametrizedLine Through(const VectorType& p0, const VectorType& p1)
   { return ParametrizedLine(p0, (p1-p0).normalized()); }
 
-  ~ParametrizedLine() {}
+  EIGEN_DEVICE_FUNC ~ParametrizedLine() {}
 
   /** \returns the dimension in which the line holds */
-  inline Index dim() const { return m_direction.size(); }
+  EIGEN_DEVICE_FUNC inline Index dim() const { return m_direction.size(); }
 
-  const VectorType& origin() const { return m_origin; }
-  VectorType& origin() { return m_origin; }
+  EIGEN_DEVICE_FUNC const VectorType& origin() const { return m_origin; }
+  EIGEN_DEVICE_FUNC VectorType& origin() { return m_origin; }
 
-  const VectorType& direction() const { return m_direction; }
-  VectorType& direction() { return m_direction; }
+  EIGEN_DEVICE_FUNC const VectorType& direction() const { return m_direction; }
+  EIGEN_DEVICE_FUNC VectorType& direction() { return m_direction; }
 
   /** \returns the squared distance of a point \a p to its projection onto the line \c *this.
     * \sa distance()
     */
-  RealScalar squaredDistance(const VectorType& p) const
+  EIGEN_DEVICE_FUNC RealScalar squaredDistance(const VectorType& p) const
   {
     VectorType diff = p - origin();
     return (diff - direction().dot(diff) * direction()).squaredNorm();
@@ -87,22 +87,22 @@ public:
   /** \returns the distance of a point \a p to its projection onto the line \c *this.
     * \sa squaredDistance()
     */
-  RealScalar distance(const VectorType& p) const { using std::sqrt; return sqrt(squaredDistance(p)); }
+  EIGEN_DEVICE_FUNC RealScalar distance(const VectorType& p) const { EIGEN_USING_STD_MATH(sqrt) return sqrt(squaredDistance(p)); }
 
   /** \returns the projection of a point \a p onto the line \c *this. */
-  VectorType projection(const VectorType& p) const
+  EIGEN_DEVICE_FUNC VectorType projection(const VectorType& p) const
   { return origin() + direction().dot(p-origin()) * direction(); }
 
-  VectorType pointAt(const Scalar& t) const;
+  EIGEN_DEVICE_FUNC VectorType pointAt(const Scalar& t) const;
   
   template <int OtherOptions>
-  Scalar intersectionParameter(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
+  EIGEN_DEVICE_FUNC Scalar intersectionParameter(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
  
   template <int OtherOptions>
-  Scalar intersection(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
+  EIGEN_DEVICE_FUNC Scalar intersection(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
   
   template <int OtherOptions>
-  VectorType intersectionPoint(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
+  EIGEN_DEVICE_FUNC VectorType intersectionPoint(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -110,7 +110,7 @@ public:
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<ParametrizedLine,
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<ParametrizedLine,
            ParametrizedLine<NewScalarType,AmbientDimAtCompileTime,Options> >::type cast() const
   {
     return typename internal::cast_return_type<ParametrizedLine,
@@ -119,7 +119,7 @@ public:
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType,int OtherOptions>
-  inline explicit ParametrizedLine(const ParametrizedLine<OtherScalarType,AmbientDimAtCompileTime,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC inline explicit ParametrizedLine(const ParametrizedLine<OtherScalarType,AmbientDimAtCompileTime,OtherOptions>& other)
   {
     m_origin = other.origin().template cast<Scalar>();
     m_direction = other.direction().template cast<Scalar>();
@@ -129,7 +129,7 @@ public:
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const ParametrizedLine& other, typename NumTraits<Scalar>::Real prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const ParametrizedLine& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_origin.isApprox(other.m_origin, prec) && m_direction.isApprox(other.m_direction, prec); }
 
 protected:
@@ -143,7 +143,7 @@ protected:
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 template <int OtherOptions>
-inline ParametrizedLine<_Scalar, _AmbientDim,_Options>::ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim,OtherOptions>& hyperplane)
+EIGEN_DEVICE_FUNC inline ParametrizedLine<_Scalar, _AmbientDim,_Options>::ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim,OtherOptions>& hyperplane)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(VectorType, 2)
   direction() = hyperplane.normal().unitOrthogonal();
@@ -153,7 +153,7 @@ inline ParametrizedLine<_Scalar, _AmbientDim,_Options>::ParametrizedLine(const H
 /** \returns the point at \a t along this line
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
-inline typename ParametrizedLine<_Scalar, _AmbientDim,_Options>::VectorType
+EIGEN_DEVICE_FUNC inline typename ParametrizedLine<_Scalar, _AmbientDim,_Options>::VectorType
 ParametrizedLine<_Scalar, _AmbientDim,_Options>::pointAt(const _Scalar& t) const
 {
   return origin() + (direction()*t); 
@@ -163,7 +163,7 @@ ParametrizedLine<_Scalar, _AmbientDim,_Options>::pointAt(const _Scalar& t) const
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 template <int OtherOptions>
-inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersectionParameter(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
+EIGEN_DEVICE_FUNC inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersectionParameter(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
 {
   return -(hyperplane.offset()+hyperplane.normal().dot(origin()))
           / hyperplane.normal().dot(direction());
@@ -175,7 +175,7 @@ inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersectionPara
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 template <int OtherOptions>
-inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersection(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
+EIGEN_DEVICE_FUNC inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersection(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
 {
   return intersectionParameter(hyperplane);
 }
@@ -184,7 +184,7 @@ inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersection(con
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 template <int OtherOptions>
-inline typename ParametrizedLine<_Scalar, _AmbientDim,_Options>::VectorType
+EIGEN_DEVICE_FUNC inline typename ParametrizedLine<_Scalar, _AmbientDim,_Options>::VectorType
 ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersectionPoint(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
 {
   return pointAt(intersectionParameter(hyperplane));
diff --git a/extern/Eigen3/Eigen/src/Geometry/Quaternion.h b/extern/Eigen3/Eigen/src/Geometry/Quaternion.h
index 25ed17bb690..c3fd8c3e0f8 100644
--- a/extern/Eigen3/Eigen/src/Geometry/Quaternion.h
+++ b/extern/Eigen3/Eigen/src/Geometry/Quaternion.h
@@ -34,14 +34,20 @@ struct quaternionbase_assign_impl;
 template<class Derived>
 class QuaternionBase : public RotationBase<Derived, 3>
 {
+ public:
   typedef RotationBase<Derived, 3> Base;
-public:
+
   using Base::operator*;
   using Base::derived;
 
   typedef typename internal::traits<Derived>::Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
   typedef typename internal::traits<Derived>::Coefficients Coefficients;
+  typedef typename Coefficients::CoeffReturnType CoeffReturnType;
+  typedef typename internal::conditional<bool(internal::traits<Derived>::Flags&LvalueBit),
+                                        Scalar&, CoeffReturnType>::type NonConstCoeffReturnType;
+
+
   enum {
     Flags = Eigen::internal::traits<Derived>::Flags
   };
@@ -57,37 +63,37 @@ public:
 
 
   /** \returns the \c x coefficient */
-  inline Scalar x() const { return this->derived().coeffs().coeff(0); }
+  EIGEN_DEVICE_FUNC inline CoeffReturnType x() const { return this->derived().coeffs().coeff(0); }
   /** \returns the \c y coefficient */
-  inline Scalar y() const { return this->derived().coeffs().coeff(1); }
+  EIGEN_DEVICE_FUNC inline CoeffReturnType y() const { return this->derived().coeffs().coeff(1); }
   /** \returns the \c z coefficient */
-  inline Scalar z() const { return this->derived().coeffs().coeff(2); }
+  EIGEN_DEVICE_FUNC inline CoeffReturnType z() const { return this->derived().coeffs().coeff(2); }
   /** \returns the \c w coefficient */
-  inline Scalar w() const { return this->derived().coeffs().coeff(3); }
+  EIGEN_DEVICE_FUNC inline CoeffReturnType w() const { return this->derived().coeffs().coeff(3); }
 
-  /** \returns a reference to the \c x coefficient */
-  inline Scalar& x() { return this->derived().coeffs().coeffRef(0); }
-  /** \returns a reference to the \c y coefficient */
-  inline Scalar& y() { return this->derived().coeffs().coeffRef(1); }
-  /** \returns a reference to the \c z coefficient */
-  inline Scalar& z() { return this->derived().coeffs().coeffRef(2); }
-  /** \returns a reference to the \c w coefficient */
-  inline Scalar& w() { return this->derived().coeffs().coeffRef(3); }
+  /** \returns a reference to the \c x coefficient (if Derived is a non-const lvalue) */
+  EIGEN_DEVICE_FUNC inline NonConstCoeffReturnType x() { return this->derived().coeffs().x(); }
+  /** \returns a reference to the \c y coefficient (if Derived is a non-const lvalue) */
+  EIGEN_DEVICE_FUNC inline NonConstCoeffReturnType y() { return this->derived().coeffs().y(); }
+  /** \returns a reference to the \c z coefficient (if Derived is a non-const lvalue) */
+  EIGEN_DEVICE_FUNC inline NonConstCoeffReturnType z() { return this->derived().coeffs().z(); }
+  /** \returns a reference to the \c w coefficient (if Derived is a non-const lvalue) */
+  EIGEN_DEVICE_FUNC inline NonConstCoeffReturnType w() { return this->derived().coeffs().w(); }
 
   /** \returns a read-only vector expression of the imaginary part (x,y,z) */
-  inline const VectorBlock<const Coefficients,3> vec() const { return coeffs().template head<3>(); }
+  EIGEN_DEVICE_FUNC inline const VectorBlock<const Coefficients,3> vec() const { return coeffs().template head<3>(); }
 
   /** \returns a vector expression of the imaginary part (x,y,z) */
-  inline VectorBlock<Coefficients,3> vec() { return coeffs().template head<3>(); }
+  EIGEN_DEVICE_FUNC inline VectorBlock<Coefficients,3> vec() { return coeffs().template head<3>(); }
 
   /** \returns a read-only vector expression of the coefficients (x,y,z,w) */
-  inline const typename internal::traits<Derived>::Coefficients& coeffs() const { return derived().coeffs(); }
+  EIGEN_DEVICE_FUNC inline const typename internal::traits<Derived>::Coefficients& coeffs() const { return derived().coeffs(); }
 
   /** \returns a vector expression of the coefficients (x,y,z,w) */
-  inline typename internal::traits<Derived>::Coefficients& coeffs() { return derived().coeffs(); }
+  EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Coefficients& coeffs() { return derived().coeffs(); }
 
-  EIGEN_STRONG_INLINE QuaternionBase<Derived>& operator=(const QuaternionBase<Derived>& other);
-  template<class OtherDerived> EIGEN_STRONG_INLINE Derived& operator=(const QuaternionBase<OtherDerived>& other);
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE QuaternionBase<Derived>& operator=(const QuaternionBase<Derived>& other);
+  template<class OtherDerived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const QuaternionBase<OtherDerived>& other);
 
 // disabled this copy operator as it is giving very strange compilation errors when compiling
 // test_stdvector with GCC 4.4.2. This looks like a GCC bug though, so feel free to re-enable it if it's
@@ -96,72 +102,72 @@ public:
 //  Derived& operator=(const QuaternionBase& other)
 //  { return operator=<Derived>(other); }
 
-  Derived& operator=(const AngleAxisType& aa);
-  template<class OtherDerived> Derived& operator=(const MatrixBase<OtherDerived>& m);
+  EIGEN_DEVICE_FUNC Derived& operator=(const AngleAxisType& aa);
+  template<class OtherDerived> EIGEN_DEVICE_FUNC Derived& operator=(const MatrixBase<OtherDerived>& m);
 
   /** \returns a quaternion representing an identity rotation
     * \sa MatrixBase::Identity()
     */
-  static inline Quaternion<Scalar> Identity() { return Quaternion<Scalar>(Scalar(1), Scalar(0), Scalar(0), Scalar(0)); }
+  EIGEN_DEVICE_FUNC static inline Quaternion<Scalar> Identity() { return Quaternion<Scalar>(Scalar(1), Scalar(0), Scalar(0), Scalar(0)); }
 
   /** \sa QuaternionBase::Identity(), MatrixBase::setIdentity()
     */
-  inline QuaternionBase& setIdentity() { coeffs() << Scalar(0), Scalar(0), Scalar(0), Scalar(1); return *this; }
+  EIGEN_DEVICE_FUNC inline QuaternionBase& setIdentity() { coeffs() << Scalar(0), Scalar(0), Scalar(0), Scalar(1); return *this; }
 
   /** \returns the squared norm of the quaternion's coefficients
     * \sa QuaternionBase::norm(), MatrixBase::squaredNorm()
     */
-  inline Scalar squaredNorm() const { return coeffs().squaredNorm(); }
+  EIGEN_DEVICE_FUNC inline Scalar squaredNorm() const { return coeffs().squaredNorm(); }
 
   /** \returns the norm of the quaternion's coefficients
     * \sa QuaternionBase::squaredNorm(), MatrixBase::norm()
     */
-  inline Scalar norm() const { return coeffs().norm(); }
+  EIGEN_DEVICE_FUNC inline Scalar norm() const { return coeffs().norm(); }
 
   /** Normalizes the quaternion \c *this
     * \sa normalized(), MatrixBase::normalize() */
-  inline void normalize() { coeffs().normalize(); }
+  EIGEN_DEVICE_FUNC inline void normalize() { coeffs().normalize(); }
   /** \returns a normalized copy of \c *this
     * \sa normalize(), MatrixBase::normalized() */
-  inline Quaternion<Scalar> normalized() const { return Quaternion<Scalar>(coeffs().normalized()); }
+  EIGEN_DEVICE_FUNC inline Quaternion<Scalar> normalized() const { return Quaternion<Scalar>(coeffs().normalized()); }
 
     /** \returns the dot product of \c *this and \a other
     * Geometrically speaking, the dot product of two unit quaternions
     * corresponds to the cosine of half the angle between the two rotations.
     * \sa angularDistance()
     */
-  template<class OtherDerived> inline Scalar dot(const QuaternionBase<OtherDerived>& other) const { return coeffs().dot(other.coeffs()); }
+  template<class OtherDerived> EIGEN_DEVICE_FUNC inline Scalar dot(const QuaternionBase<OtherDerived>& other) const { return coeffs().dot(other.coeffs()); }
 
-  template<class OtherDerived> Scalar angularDistance(const QuaternionBase<OtherDerived>& other) const;
+  template<class OtherDerived> EIGEN_DEVICE_FUNC Scalar angularDistance(const QuaternionBase<OtherDerived>& other) const;
 
   /** \returns an equivalent 3x3 rotation matrix */
-  Matrix3 toRotationMatrix() const;
+  EIGEN_DEVICE_FUNC Matrix3 toRotationMatrix() const;
 
   /** \returns the quaternion which transform \a a into \a b through a rotation */
   template<typename Derived1, typename Derived2>
-  Derived& setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b);
+  EIGEN_DEVICE_FUNC Derived& setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b);
 
-  template<class OtherDerived> EIGEN_STRONG_INLINE Quaternion<Scalar> operator* (const QuaternionBase<OtherDerived>& q) const;
-  template<class OtherDerived> EIGEN_STRONG_INLINE Derived& operator*= (const QuaternionBase<OtherDerived>& q);
+  template<class OtherDerived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Quaternion<Scalar> operator* (const QuaternionBase<OtherDerived>& q) const;
+  template<class OtherDerived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator*= (const QuaternionBase<OtherDerived>& q);
 
   /** \returns the quaternion describing the inverse rotation */
-  Quaternion<Scalar> inverse() const;
+  EIGEN_DEVICE_FUNC Quaternion<Scalar> inverse() const;
 
   /** \returns the conjugated quaternion */
-  Quaternion<Scalar> conjugate() const;
+  EIGEN_DEVICE_FUNC Quaternion<Scalar> conjugate() const;
 
-  template<class OtherDerived> Quaternion<Scalar> slerp(const Scalar& t, const QuaternionBase<OtherDerived>& other) const;
+  template<class OtherDerived> EIGEN_DEVICE_FUNC Quaternion<Scalar> slerp(const Scalar& t, const QuaternionBase<OtherDerived>& other) const;
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
   template<class OtherDerived>
-  bool isApprox(const QuaternionBase<OtherDerived>& other, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const QuaternionBase<OtherDerived>& other, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const
   { return coeffs().isApprox(other.coeffs(), prec); }
 
-	/** return the result vector of \a v through the rotation*/
-  EIGEN_STRONG_INLINE Vector3 _transformVector(const Vector3& v) const;
+  /** return the result vector of \a v through the rotation*/
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Vector3 _transformVector(const Vector3& v) const;
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -169,7 +175,7 @@ public:
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Derived,Quaternion<NewScalarType> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Derived,Quaternion<NewScalarType> >::type cast() const
   {
     return typename internal::cast_return_type<Derived,Quaternion<NewScalarType> >::type(derived());
   }
@@ -216,8 +222,8 @@ struct traits<Quaternion<_Scalar,_Options> >
   typedef _Scalar Scalar;
   typedef Matrix<_Scalar,4,1,_Options> Coefficients;
   enum{
-    IsAligned = internal::traits<Coefficients>::Flags & AlignedBit,
-    Flags = IsAligned ? (AlignedBit | LvalueBit) : LvalueBit
+    Alignment = internal::traits<Coefficients>::Alignment,
+    Flags = LvalueBit
   };
 };
 }
@@ -225,10 +231,10 @@ struct traits<Quaternion<_Scalar,_Options> >
 template<typename _Scalar, int _Options>
 class Quaternion : public QuaternionBase<Quaternion<_Scalar,_Options> >
 {
+public:
   typedef QuaternionBase<Quaternion<_Scalar,_Options> > Base;
-  enum { IsAligned = internal::traits<Quaternion>::IsAligned };
+  enum { NeedsAlignment = internal::traits<Quaternion>::Alignment>0 };
 
-public:
   typedef _Scalar Scalar;
 
   EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Quaternion)
@@ -238,7 +244,7 @@ public:
   typedef typename Base::AngleAxisType AngleAxisType;
 
   /** Default constructor leaving the quaternion uninitialized. */
-  inline Quaternion() {}
+  EIGEN_DEVICE_FUNC inline Quaternion() {}
 
   /** Constructs and initializes the quaternion \f$ w+xi+yj+zk \f$ from
     * its four coefficients \a w, \a x, \a y and \a z.
@@ -247,36 +253,42 @@ public:
     * while internally the coefficients are stored in the following order:
     * [\c x, \c y, \c z, \c w]
     */
-  inline Quaternion(const Scalar& w, const Scalar& x, const Scalar& y, const Scalar& z) : m_coeffs(x, y, z, w){}
+  EIGEN_DEVICE_FUNC inline Quaternion(const Scalar& w, const Scalar& x, const Scalar& y, const Scalar& z) : m_coeffs(x, y, z, w){}
 
   /** Constructs and initialize a quaternion from the array data */
-  inline Quaternion(const Scalar* data) : m_coeffs(data) {}
+  EIGEN_DEVICE_FUNC explicit inline Quaternion(const Scalar* data) : m_coeffs(data) {}
 
   /** Copy constructor */
-  template<class Derived> EIGEN_STRONG_INLINE Quaternion(const QuaternionBase<Derived>& other) { this->Base::operator=(other); }
+  template<class Derived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Quaternion(const QuaternionBase<Derived>& other) { this->Base::operator=(other); }
 
   /** Constructs and initializes a quaternion from the angle-axis \a aa */
-  explicit inline Quaternion(const AngleAxisType& aa) { *this = aa; }
+  EIGEN_DEVICE_FUNC explicit inline Quaternion(const AngleAxisType& aa) { *this = aa; }
 
   /** Constructs and initializes a quaternion from either:
     *  - a rotation matrix expression,
     *  - a 4D vector expression representing quaternion coefficients.
     */
   template<typename Derived>
-  explicit inline Quaternion(const MatrixBase<Derived>& other) { *this = other; }
+  EIGEN_DEVICE_FUNC explicit inline Quaternion(const MatrixBase<Derived>& other) { *this = other; }
 
   /** Explicit copy constructor with scalar conversion */
   template<typename OtherScalar, int OtherOptions>
-  explicit inline Quaternion(const Quaternion<OtherScalar, OtherOptions>& other)
+  EIGEN_DEVICE_FUNC explicit inline Quaternion(const Quaternion<OtherScalar, OtherOptions>& other)
   { m_coeffs = other.coeffs().template cast<Scalar>(); }
 
+  EIGEN_DEVICE_FUNC static Quaternion UnitRandom();
+
   template<typename Derived1, typename Derived2>
-  static Quaternion FromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b);
+  EIGEN_DEVICE_FUNC static Quaternion FromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b);
 
-  inline Coefficients& coeffs() { return m_coeffs;}
-  inline const Coefficients& coeffs() const { return m_coeffs;}
+  EIGEN_DEVICE_FUNC inline Coefficients& coeffs() { return m_coeffs;}
+  EIGEN_DEVICE_FUNC inline const Coefficients& coeffs() const { return m_coeffs;}
 
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(IsAligned)
+  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(bool(NeedsAlignment))
+  
+#ifdef EIGEN_QUATERNION_PLUGIN
+# include EIGEN_QUATERNION_PLUGIN
+#endif
 
 protected:
   Coefficients m_coeffs;
@@ -336,9 +348,9 @@ template<typename _Scalar, int _Options>
 class Map<const Quaternion<_Scalar>, _Options >
   : public QuaternionBase<Map<const Quaternion<_Scalar>, _Options> >
 {
+  public:
     typedef QuaternionBase<Map<const Quaternion<_Scalar>, _Options> > Base;
 
-  public:
     typedef _Scalar Scalar;
     typedef typename internal::traits<Map>::Coefficients Coefficients;
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Map)
@@ -350,9 +362,9 @@ class Map<const Quaternion<_Scalar>, _Options >
       * \code *coeffs == {x, y, z, w} \endcode
       *
       * If the template parameter _Options is set to #Aligned, then the pointer coeffs must be aligned. */
-    EIGEN_STRONG_INLINE Map(const Scalar* coeffs) : m_coeffs(coeffs) {}
+    EIGEN_DEVICE_FUNC explicit EIGEN_STRONG_INLINE Map(const Scalar* coeffs) : m_coeffs(coeffs) {}
 
-    inline const Coefficients& coeffs() const { return m_coeffs;}
+    EIGEN_DEVICE_FUNC inline const Coefficients& coeffs() const { return m_coeffs;}
 
   protected:
     const Coefficients m_coeffs;
@@ -373,9 +385,9 @@ template<typename _Scalar, int _Options>
 class Map<Quaternion<_Scalar>, _Options >
   : public QuaternionBase<Map<Quaternion<_Scalar>, _Options> >
 {
+  public:
     typedef QuaternionBase<Map<Quaternion<_Scalar>, _Options> > Base;
 
-  public:
     typedef _Scalar Scalar;
     typedef typename internal::traits<Map>::Coefficients Coefficients;
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Map)
@@ -387,10 +399,10 @@ class Map<Quaternion<_Scalar>, _Options >
       * \code *coeffs == {x, y, z, w} \endcode
       *
       * If the template parameter _Options is set to #Aligned, then the pointer coeffs must be aligned. */
-    EIGEN_STRONG_INLINE Map(Scalar* coeffs) : m_coeffs(coeffs) {}
+    EIGEN_DEVICE_FUNC explicit EIGEN_STRONG_INLINE Map(Scalar* coeffs) : m_coeffs(coeffs) {}
 
-    inline Coefficients& coeffs() { return m_coeffs; }
-    inline const Coefficients& coeffs() const { return m_coeffs; }
+    EIGEN_DEVICE_FUNC inline Coefficients& coeffs() { return m_coeffs; }
+    EIGEN_DEVICE_FUNC inline const Coefficients& coeffs() const { return m_coeffs; }
 
   protected:
     Coefficients m_coeffs;
@@ -416,9 +428,9 @@ typedef Map<Quaternion<double>, Aligned>  QuaternionMapAlignedd;
 // Generic Quaternion * Quaternion product
 // This product can be specialized for a given architecture via the Arch template argument.
 namespace internal {
-template<int Arch, class Derived1, class Derived2, typename Scalar, int _Options> struct quat_product
+template<int Arch, class Derived1, class Derived2, typename Scalar> struct quat_product
 {
-  static EIGEN_STRONG_INLINE Quaternion<Scalar> run(const QuaternionBase<Derived1>& a, const QuaternionBase<Derived2>& b){
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Quaternion<Scalar> run(const QuaternionBase<Derived1>& a, const QuaternionBase<Derived2>& b){
     return Quaternion<Scalar>
     (
       a.w() * b.w() - a.x() * b.x() - a.y() * b.y() - a.z() * b.z(),
@@ -433,20 +445,19 @@ template<int Arch, class Derived1, class Derived2, typename Scalar, int _Options
 /** \returns the concatenation of two rotations as a quaternion-quaternion product */
 template <class Derived>
 template <class OtherDerived>
-EIGEN_STRONG_INLINE Quaternion<typename internal::traits<Derived>::Scalar>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Quaternion<typename internal::traits<Derived>::Scalar>
 QuaternionBase<Derived>::operator* (const QuaternionBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT((internal::is_same<typename Derived::Scalar, typename OtherDerived::Scalar>::value),
    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
   return internal::quat_product<Architecture::Target, Derived, OtherDerived,
-                         typename internal::traits<Derived>::Scalar,
-                         internal::traits<Derived>::IsAligned && internal::traits<OtherDerived>::IsAligned>::run(*this, other);
+                         typename internal::traits<Derived>::Scalar>::run(*this, other);
 }
 
 /** \sa operator*(Quaternion) */
 template <class Derived>
 template <class OtherDerived>
-EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator*= (const QuaternionBase<OtherDerived>& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator*= (const QuaternionBase<OtherDerived>& other)
 {
   derived() = derived() * other.derived();
   return derived();
@@ -460,7 +471,7 @@ EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator*= (const Quaterni
   *   - Via a Matrix3: 24 + 15n
   */
 template <class Derived>
-EIGEN_STRONG_INLINE typename QuaternionBase<Derived>::Vector3
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename QuaternionBase<Derived>::Vector3
 QuaternionBase<Derived>::_transformVector(const Vector3& v) const
 {
     // Note that this algorithm comes from the optimization by hand
@@ -474,7 +485,7 @@ QuaternionBase<Derived>::_transformVector(const Vector3& v) const
 }
 
 template<class Derived>
-EIGEN_STRONG_INLINE QuaternionBase<Derived>& QuaternionBase<Derived>::operator=(const QuaternionBase<Derived>& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE QuaternionBase<Derived>& QuaternionBase<Derived>::operator=(const QuaternionBase<Derived>& other)
 {
   coeffs() = other.coeffs();
   return derived();
@@ -482,7 +493,7 @@ EIGEN_STRONG_INLINE QuaternionBase<Derived>& QuaternionBase<Derived>::operator=(
 
 template<class Derived>
 template<class OtherDerived>
-EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const QuaternionBase<OtherDerived>& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const QuaternionBase<OtherDerived>& other)
 {
   coeffs() = other.coeffs();
   return derived();
@@ -491,10 +502,10 @@ EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const Quaternion
 /** Set \c *this from an angle-axis \a aa and returns a reference to \c *this
   */
 template<class Derived>
-EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const AngleAxisType& aa)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const AngleAxisType& aa)
 {
-  using std::cos;
-  using std::sin;
+  EIGEN_USING_STD_MATH(cos)
+  EIGEN_USING_STD_MATH(sin)
   Scalar ha = Scalar(0.5)*aa.angle(); // Scalar(0.5) to suppress precision loss warnings
   this->w() = cos(ha);
   this->vec() = sin(ha) * aa.axis();
@@ -509,7 +520,7 @@ EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const AngleAxisT
 
 template<class Derived>
 template<class MatrixDerived>
-inline Derived& QuaternionBase<Derived>::operator=(const MatrixBase<MatrixDerived>& xpr)
+EIGEN_DEVICE_FUNC inline Derived& QuaternionBase<Derived>::operator=(const MatrixBase<MatrixDerived>& xpr)
 {
   EIGEN_STATIC_ASSERT((internal::is_same<typename Derived::Scalar, typename MatrixDerived::Scalar>::value),
    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
@@ -521,7 +532,7 @@ inline Derived& QuaternionBase<Derived>::operator=(const MatrixBase<MatrixDerive
   * be normalized, otherwise the result is undefined.
   */
 template<class Derived>
-inline typename QuaternionBase<Derived>::Matrix3
+EIGEN_DEVICE_FUNC inline typename QuaternionBase<Derived>::Matrix3
 QuaternionBase<Derived>::toRotationMatrix(void) const
 {
   // NOTE if inlined, then gcc 4.2 and 4.4 get rid of the temporary (not gcc 4.3 !!)
@@ -568,10 +579,9 @@ QuaternionBase<Derived>::toRotationMatrix(void) const
   */
 template<class Derived>
 template<typename Derived1, typename Derived2>
-inline Derived& QuaternionBase<Derived>::setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b)
+EIGEN_DEVICE_FUNC inline Derived& QuaternionBase<Derived>::setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b)
 {
-  using std::max;
-  using std::sqrt;
+  EIGEN_USING_STD_MATH(sqrt)
   Vector3 v0 = a.normalized();
   Vector3 v1 = b.normalized();
   Scalar c = v1.dot(v0);
@@ -586,7 +596,7 @@ inline Derived& QuaternionBase<Derived>::setFromTwoVectors(const MatrixBase<Deri
   //    which yields a singular value problem
   if (c < Scalar(-1)+NumTraits<Scalar>::dummy_precision())
   {
-    c = (max)(c,Scalar(-1));
+    c = numext::maxi(c,Scalar(-1));
     Matrix<Scalar,2,3> m; m << v0.transpose(), v1.transpose();
     JacobiSVD<Matrix<Scalar,2,3> > svd(m, ComputeFullV);
     Vector3 axis = svd.matrixV().col(2);
@@ -605,6 +615,24 @@ inline Derived& QuaternionBase<Derived>::setFromTwoVectors(const MatrixBase<Deri
   return derived();
 }
 
+/** \returns a random unit quaternion following a uniform distribution law on SO(3)
+  *
+  * \note The implementation is based on http://planning.cs.uiuc.edu/node198.html
+  */
+template<typename Scalar, int Options>
+EIGEN_DEVICE_FUNC Quaternion<Scalar,Options> Quaternion<Scalar,Options>::UnitRandom()
+{
+  EIGEN_USING_STD_MATH(sqrt)
+  EIGEN_USING_STD_MATH(sin)
+  EIGEN_USING_STD_MATH(cos)
+  const Scalar u1 = internal::random<Scalar>(0, 1),
+               u2 = internal::random<Scalar>(0, 2*EIGEN_PI),
+               u3 = internal::random<Scalar>(0, 2*EIGEN_PI);
+  const Scalar a = sqrt(1 - u1),
+               b = sqrt(u1);
+  return Quaternion (a * sin(u2), a * cos(u2), b * sin(u3), b * cos(u3));
+}
+
 
 /** Returns a quaternion representing a rotation between
   * the two arbitrary vectors \a a and \a b. In other words, the built
@@ -618,7 +646,7 @@ inline Derived& QuaternionBase<Derived>::setFromTwoVectors(const MatrixBase<Deri
   */
 template<typename Scalar, int Options>
 template<typename Derived1, typename Derived2>
-Quaternion<Scalar,Options> Quaternion<Scalar,Options>::FromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b)
+EIGEN_DEVICE_FUNC Quaternion<Scalar,Options> Quaternion<Scalar,Options>::FromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b)
 {
     Quaternion quat;
     quat.setFromTwoVectors(a, b);
@@ -633,7 +661,7 @@ Quaternion<Scalar,Options> Quaternion<Scalar,Options>::FromTwoVectors(const Matr
   * \sa QuaternionBase::conjugate()
   */
 template <class Derived>
-inline Quaternion<typename internal::traits<Derived>::Scalar> QuaternionBase<Derived>::inverse() const
+EIGEN_DEVICE_FUNC inline Quaternion<typename internal::traits<Derived>::Scalar> QuaternionBase<Derived>::inverse() const
 {
   // FIXME should this function be called multiplicativeInverse and conjugate() be called inverse() or opposite()  ??
   Scalar n2 = this->squaredNorm();
@@ -646,6 +674,16 @@ inline Quaternion<typename internal::traits<Derived>::Scalar> QuaternionBase<Der
   }
 }
 
+// Generic conjugate of a Quaternion
+namespace internal {
+template<int Arch, class Derived, typename Scalar> struct quat_conj
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Quaternion<Scalar> run(const QuaternionBase<Derived>& q){
+    return Quaternion<Scalar>(q.w(),-q.x(),-q.y(),-q.z());
+  }
+};
+}
+                         
 /** \returns the conjugate of the \c *this which is equal to the multiplicative inverse
   * if the quaternion is normalized.
   * The conjugate of a quaternion represents the opposite rotation.
@@ -653,10 +691,12 @@ inline Quaternion<typename internal::traits<Derived>::Scalar> QuaternionBase<Der
   * \sa Quaternion2::inverse()
   */
 template <class Derived>
-inline Quaternion<typename internal::traits<Derived>::Scalar>
+EIGEN_DEVICE_FUNC inline Quaternion<typename internal::traits<Derived>::Scalar>
 QuaternionBase<Derived>::conjugate() const
 {
-  return Quaternion<Scalar>(this->w(),-this->x(),-this->y(),-this->z());
+  return internal::quat_conj<Architecture::Target, Derived,
+                         typename internal::traits<Derived>::Scalar>::run(*this);
+                         
 }
 
 /** \returns the angle (in radian) between two rotations
@@ -664,13 +704,12 @@ QuaternionBase<Derived>::conjugate() const
   */
 template <class Derived>
 template <class OtherDerived>
-inline typename internal::traits<Derived>::Scalar
+EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Scalar
 QuaternionBase<Derived>::angularDistance(const QuaternionBase<OtherDerived>& other) const
 {
-  using std::atan2;
-  using std::abs;
+  EIGEN_USING_STD_MATH(atan2)
   Quaternion<Scalar> d = (*this) * other.conjugate();
-  return Scalar(2) * atan2( d.vec().norm(), abs(d.w()) );
+  return Scalar(2) * atan2( d.vec().norm(), numext::abs(d.w()) );
 }
 
  
@@ -683,15 +722,14 @@ QuaternionBase<Derived>::angularDistance(const QuaternionBase<OtherDerived>& oth
   */
 template <class Derived>
 template <class OtherDerived>
-Quaternion<typename internal::traits<Derived>::Scalar>
+EIGEN_DEVICE_FUNC Quaternion<typename internal::traits<Derived>::Scalar>
 QuaternionBase<Derived>::slerp(const Scalar& t, const QuaternionBase<OtherDerived>& other) const
 {
-  using std::acos;
-  using std::sin;
-  using std::abs;
-  static const Scalar one = Scalar(1) - NumTraits<Scalar>::epsilon();
+  EIGEN_USING_STD_MATH(acos)
+  EIGEN_USING_STD_MATH(sin)
+  const Scalar one = Scalar(1) - NumTraits<Scalar>::epsilon();
   Scalar d = this->dot(other);
-  Scalar absD = abs(d);
+  Scalar absD = numext::abs(d);
 
   Scalar scale0;
   Scalar scale1;
@@ -722,10 +760,10 @@ template<typename Other>
 struct quaternionbase_assign_impl<Other,3,3>
 {
   typedef typename Other::Scalar Scalar;
-  typedef DenseIndex Index;
-  template<class Derived> static inline void run(QuaternionBase<Derived>& q, const Other& mat)
+  template<class Derived> EIGEN_DEVICE_FUNC static inline void run(QuaternionBase<Derived>& q, const Other& a_mat)
   {
-    using std::sqrt;
+    const typename internal::nested_eval<Other,2>::type mat(a_mat);
+    EIGEN_USING_STD_MATH(sqrt)
     // This algorithm comes from  "Quaternion Calculus and Fast Animation",
     // Ken Shoemake, 1987 SIGGRAPH course notes
     Scalar t = mat.trace();
@@ -740,13 +778,13 @@ struct quaternionbase_assign_impl<Other,3,3>
     }
     else
     {
-      DenseIndex i = 0;
+      Index i = 0;
       if (mat.coeff(1,1) > mat.coeff(0,0))
         i = 1;
       if (mat.coeff(2,2) > mat.coeff(i,i))
         i = 2;
-      DenseIndex j = (i+1)%3;
-      DenseIndex k = (j+1)%3;
+      Index j = (i+1)%3;
+      Index k = (j+1)%3;
 
       t = sqrt(mat.coeff(i,i)-mat.coeff(j,j)-mat.coeff(k,k) + Scalar(1.0));
       q.coeffs().coeffRef(i) = Scalar(0.5) * t;
@@ -763,7 +801,7 @@ template<typename Other>
 struct quaternionbase_assign_impl<Other,4,1>
 {
   typedef typename Other::Scalar Scalar;
-  template<class Derived> static inline void run(QuaternionBase<Derived>& q, const Other& vec)
+  template<class Derived> EIGEN_DEVICE_FUNC static inline void run(QuaternionBase<Derived>& q, const Other& vec)
   {
     q.coeffs() = vec;
   }
diff --git a/extern/Eigen3/Eigen/src/Geometry/Rotation2D.h b/extern/Eigen3/Eigen/src/Geometry/Rotation2D.h
index a2d59fce10f..884b7d0ee95 100644
--- a/extern/Eigen3/Eigen/src/Geometry/Rotation2D.h
+++ b/extern/Eigen3/Eigen/src/Geometry/Rotation2D.h
@@ -18,7 +18,7 @@ namespace Eigen {
   *
   * \brief Represents a rotation/orientation in a 2 dimensional space.
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients
   *
   * This class is equivalent to a single scalar representing a counter clock wise rotation
   * as a single angle in radian. It provides some additional features such as the automatic
@@ -59,41 +59,79 @@ protected:
 public:
 
   /** Construct a 2D counter clock wise rotation from the angle \a a in radian. */
-  inline Rotation2D(const Scalar& a) : m_angle(a) {}
+  EIGEN_DEVICE_FUNC explicit inline Rotation2D(const Scalar& a) : m_angle(a) {}
   
   /** Default constructor wihtout initialization. The represented rotation is undefined. */
-  Rotation2D() {}
+  EIGEN_DEVICE_FUNC Rotation2D() {}
+
+  /** Construct a 2D rotation from a 2x2 rotation matrix \a mat.
+    *
+    * \sa fromRotationMatrix()
+    */
+  template<typename Derived>
+  EIGEN_DEVICE_FUNC explicit Rotation2D(const MatrixBase<Derived>& m)
+  {
+    fromRotationMatrix(m.derived());
+  }
 
   /** \returns the rotation angle */
-  inline Scalar angle() const { return m_angle; }
+  EIGEN_DEVICE_FUNC inline Scalar angle() const { return m_angle; }
 
   /** \returns a read-write reference to the rotation angle */
-  inline Scalar& angle() { return m_angle; }
+  EIGEN_DEVICE_FUNC inline Scalar& angle() { return m_angle; }
+  
+  /** \returns the rotation angle in [0,2pi] */
+  EIGEN_DEVICE_FUNC inline Scalar smallestPositiveAngle() const {
+    Scalar tmp = numext::fmod(m_angle,Scalar(2*EIGEN_PI));
+    return tmp<Scalar(0) ? tmp + Scalar(2*EIGEN_PI) : tmp;
+  }
+  
+  /** \returns the rotation angle in [-pi,pi] */
+  EIGEN_DEVICE_FUNC inline Scalar smallestAngle() const {
+    Scalar tmp = numext::fmod(m_angle,Scalar(2*EIGEN_PI));
+    if(tmp>Scalar(EIGEN_PI))       tmp -= Scalar(2*EIGEN_PI);
+    else if(tmp<-Scalar(EIGEN_PI)) tmp += Scalar(2*EIGEN_PI);
+    return tmp;
+  }
 
   /** \returns the inverse rotation */
-  inline Rotation2D inverse() const { return -m_angle; }
+  EIGEN_DEVICE_FUNC inline Rotation2D inverse() const { return Rotation2D(-m_angle); }
 
   /** Concatenates two rotations */
-  inline Rotation2D operator*(const Rotation2D& other) const
-  { return m_angle + other.m_angle; }
+  EIGEN_DEVICE_FUNC inline Rotation2D operator*(const Rotation2D& other) const
+  { return Rotation2D(m_angle + other.m_angle); }
 
   /** Concatenates two rotations */
-  inline Rotation2D& operator*=(const Rotation2D& other)
+  EIGEN_DEVICE_FUNC inline Rotation2D& operator*=(const Rotation2D& other)
   { m_angle += other.m_angle; return *this; }
 
   /** Applies the rotation to a 2D vector */
-  Vector2 operator* (const Vector2& vec) const
+  EIGEN_DEVICE_FUNC Vector2 operator* (const Vector2& vec) const
   { return toRotationMatrix() * vec; }
   
   template<typename Derived>
-  Rotation2D& fromRotationMatrix(const MatrixBase<Derived>& m);
-  Matrix2 toRotationMatrix() const;
+  EIGEN_DEVICE_FUNC Rotation2D& fromRotationMatrix(const MatrixBase<Derived>& m);
+  EIGEN_DEVICE_FUNC Matrix2 toRotationMatrix() const;
+
+  /** Set \c *this from a 2x2 rotation matrix \a mat.
+    * In other words, this function extract the rotation angle from the rotation matrix.
+    *
+    * This method is an alias for fromRotationMatrix()
+    *
+    * \sa fromRotationMatrix()
+    */
+  template<typename Derived>
+  EIGEN_DEVICE_FUNC Rotation2D& operator=(const  MatrixBase<Derived>& m)
+  { return fromRotationMatrix(m.derived()); }
 
   /** \returns the spherical interpolation between \c *this and \a other using
     * parameter \a t. It is in fact equivalent to a linear interpolation.
     */
-  inline Rotation2D slerp(const Scalar& t, const Rotation2D& other) const
-  { return m_angle * (1-t) + other.angle() * t; }
+  EIGEN_DEVICE_FUNC inline Rotation2D slerp(const Scalar& t, const Rotation2D& other) const
+  {
+    Scalar dist = Rotation2D(other.m_angle-m_angle).smallestAngle();
+    return Rotation2D(m_angle + dist*t);
+  }
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -101,24 +139,25 @@ public:
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Rotation2D,Rotation2D<NewScalarType> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Rotation2D,Rotation2D<NewScalarType> >::type cast() const
   { return typename internal::cast_return_type<Rotation2D,Rotation2D<NewScalarType> >::type(*this); }
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit Rotation2D(const Rotation2D<OtherScalarType>& other)
+  EIGEN_DEVICE_FUNC inline explicit Rotation2D(const Rotation2D<OtherScalarType>& other)
   {
     m_angle = Scalar(other.angle());
   }
 
-  static inline Rotation2D Identity() { return Rotation2D(0); }
+  EIGEN_DEVICE_FUNC static inline Rotation2D Identity() { return Rotation2D(0); }
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const Rotation2D& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const Rotation2D& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return internal::isApprox(m_angle,other.m_angle, prec); }
+  
 };
 
 /** \ingroup Geometry_Module
@@ -134,9 +173,9 @@ typedef Rotation2D<double> Rotation2Dd;
   */
 template<typename Scalar>
 template<typename Derived>
-Rotation2D<Scalar>& Rotation2D<Scalar>::fromRotationMatrix(const MatrixBase<Derived>& mat)
+EIGEN_DEVICE_FUNC Rotation2D<Scalar>& Rotation2D<Scalar>::fromRotationMatrix(const MatrixBase<Derived>& mat)
 {
-  using std::atan2;
+  EIGEN_USING_STD_MATH(atan2)
   EIGEN_STATIC_ASSERT(Derived::RowsAtCompileTime==2 && Derived::ColsAtCompileTime==2,YOU_MADE_A_PROGRAMMING_MISTAKE)
   m_angle = atan2(mat.coeff(1,0), mat.coeff(0,0));
   return *this;
@@ -146,10 +185,10 @@ Rotation2D<Scalar>& Rotation2D<Scalar>::fromRotationMatrix(const MatrixBase<Deri
   */
 template<typename Scalar>
 typename Rotation2D<Scalar>::Matrix2
-Rotation2D<Scalar>::toRotationMatrix(void) const
+EIGEN_DEVICE_FUNC Rotation2D<Scalar>::toRotationMatrix(void) const
 {
-  using std::sin;
-  using std::cos;
+  EIGEN_USING_STD_MATH(sin)
+  EIGEN_USING_STD_MATH(cos)
   Scalar sinA = sin(m_angle);
   Scalar cosA = cos(m_angle);
   return (Matrix2() << cosA, -sinA, sinA, cosA).finished();
diff --git a/extern/Eigen3/Eigen/src/Geometry/RotationBase.h b/extern/Eigen3/Eigen/src/Geometry/RotationBase.h
index b88661de6b1..f0ee0bd03c5 100644
--- a/extern/Eigen3/Eigen/src/Geometry/RotationBase.h
+++ b/extern/Eigen3/Eigen/src/Geometry/RotationBase.h
@@ -22,8 +22,8 @@ struct rotation_base_generic_product_selector;
   *
   * \brief Common base class for compact rotation representations
   *
-  * \param Derived is the derived type, i.e., a rotation type
-  * \param _Dim the dimension of the space
+  * \tparam Derived is the derived type, i.e., a rotation type
+  * \tparam _Dim the dimension of the space
   */
 template<typename Derived, int _Dim>
 class RotationBase
@@ -38,26 +38,26 @@ class RotationBase
     typedef Matrix<Scalar,Dim,1> VectorType;
 
   public:
-    inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
-    inline Derived& derived() { return *static_cast<Derived*>(this); }
+    EIGEN_DEVICE_FUNC inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    EIGEN_DEVICE_FUNC inline Derived& derived() { return *static_cast<Derived*>(this); }
 
     /** \returns an equivalent rotation matrix */
-    inline RotationMatrixType toRotationMatrix() const { return derived().toRotationMatrix(); }
+    EIGEN_DEVICE_FUNC inline RotationMatrixType toRotationMatrix() const { return derived().toRotationMatrix(); }
 
     /** \returns an equivalent rotation matrix 
       * This function is added to be conform with the Transform class' naming scheme.
       */
-    inline RotationMatrixType matrix() const { return derived().toRotationMatrix(); }
+    EIGEN_DEVICE_FUNC inline RotationMatrixType matrix() const { return derived().toRotationMatrix(); }
 
     /** \returns the inverse rotation */
-    inline Derived inverse() const { return derived().inverse(); }
+    EIGEN_DEVICE_FUNC inline Derived inverse() const { return derived().inverse(); }
 
     /** \returns the concatenation of the rotation \c *this with a translation \a t */
-    inline Transform<Scalar,Dim,Isometry> operator*(const Translation<Scalar,Dim>& t) const
+    EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Isometry> operator*(const Translation<Scalar,Dim>& t) const
     { return Transform<Scalar,Dim,Isometry>(*this) * t; }
 
     /** \returns the concatenation of the rotation \c *this with a uniform scaling \a s */
-    inline RotationMatrixType operator*(const UniformScaling<Scalar>& s) const
+    EIGEN_DEVICE_FUNC inline RotationMatrixType operator*(const UniformScaling<Scalar>& s) const
     { return toRotationMatrix() * s.factor(); }
 
     /** \returns the concatenation of the rotation \c *this with a generic expression \a e
@@ -67,17 +67,17 @@ class RotationBase
       *  - a vector of size Dim
       */
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE typename internal::rotation_base_generic_product_selector<Derived,OtherDerived,OtherDerived::IsVectorAtCompileTime>::ReturnType
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::rotation_base_generic_product_selector<Derived,OtherDerived,OtherDerived::IsVectorAtCompileTime>::ReturnType
     operator*(const EigenBase<OtherDerived>& e) const
     { return internal::rotation_base_generic_product_selector<Derived,OtherDerived>::run(derived(), e.derived()); }
 
     /** \returns the concatenation of a linear transformation \a l with the rotation \a r */
     template<typename OtherDerived> friend
-    inline RotationMatrixType operator*(const EigenBase<OtherDerived>& l, const Derived& r)
+    EIGEN_DEVICE_FUNC inline RotationMatrixType operator*(const EigenBase<OtherDerived>& l, const Derived& r)
     { return l.derived() * r.toRotationMatrix(); }
 
     /** \returns the concatenation of a scaling \a l with the rotation \a r */
-    friend inline Transform<Scalar,Dim,Affine> operator*(const DiagonalMatrix<Scalar,Dim>& l, const Derived& r)
+    EIGEN_DEVICE_FUNC friend inline Transform<Scalar,Dim,Affine> operator*(const DiagonalMatrix<Scalar,Dim>& l, const Derived& r)
     { 
       Transform<Scalar,Dim,Affine> res(r);
       res.linear().applyOnTheLeft(l);
@@ -86,11 +86,11 @@ class RotationBase
 
     /** \returns the concatenation of the rotation \c *this with a transformation \a t */
     template<int Mode, int Options>
-    inline Transform<Scalar,Dim,Mode> operator*(const Transform<Scalar,Dim,Mode,Options>& t) const
+    EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode> operator*(const Transform<Scalar,Dim,Mode,Options>& t) const
     { return toRotationMatrix() * t; }
 
     template<typename OtherVectorType>
-    inline VectorType _transformVector(const OtherVectorType& v) const
+    EIGEN_DEVICE_FUNC inline VectorType _transformVector(const OtherVectorType& v) const
     { return toRotationMatrix() * v; }
 };
 
@@ -102,7 +102,7 @@ struct rotation_base_generic_product_selector<RotationDerived,MatrixType,false>
 {
   enum { Dim = RotationDerived::Dim };
   typedef Matrix<typename RotationDerived::Scalar,Dim,Dim> ReturnType;
-  static inline ReturnType run(const RotationDerived& r, const MatrixType& m)
+  EIGEN_DEVICE_FUNC static inline ReturnType run(const RotationDerived& r, const MatrixType& m)
   { return r.toRotationMatrix() * m; }
 };
 
@@ -110,7 +110,7 @@ template<typename RotationDerived, typename Scalar, int Dim, int MaxDim>
 struct rotation_base_generic_product_selector< RotationDerived, DiagonalMatrix<Scalar,Dim,MaxDim>, false >
 {
   typedef Transform<Scalar,Dim,Affine> ReturnType;
-  static inline ReturnType run(const RotationDerived& r, const DiagonalMatrix<Scalar,Dim,MaxDim>& m)
+  EIGEN_DEVICE_FUNC static inline ReturnType run(const RotationDerived& r, const DiagonalMatrix<Scalar,Dim,MaxDim>& m)
   {
     ReturnType res(r);
     res.linear() *= m;
@@ -123,7 +123,7 @@ struct rotation_base_generic_product_selector<RotationDerived,OtherVectorType,tr
 {
   enum { Dim = RotationDerived::Dim };
   typedef Matrix<typename RotationDerived::Scalar,Dim,1> ReturnType;
-  static EIGEN_STRONG_INLINE ReturnType run(const RotationDerived& r, const OtherVectorType& v)
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE ReturnType run(const RotationDerived& r, const OtherVectorType& v)
   {
     return r._transformVector(v);
   }
@@ -137,7 +137,7 @@ struct rotation_base_generic_product_selector<RotationDerived,OtherVectorType,tr
   */
 template<typename _Scalar, int _Rows, int _Cols, int _Storage, int _MaxRows, int _MaxCols>
 template<typename OtherDerived>
-Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>
+EIGEN_DEVICE_FUNC Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>
 ::Matrix(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
 {
   EIGEN_STATIC_ASSERT_MATRIX_SPECIFIC_SIZE(Matrix,int(OtherDerived::Dim),int(OtherDerived::Dim))
@@ -150,7 +150,7 @@ Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>
   */
 template<typename _Scalar, int _Rows, int _Cols, int _Storage, int _MaxRows, int _MaxCols>
 template<typename OtherDerived>
-Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>&
+EIGEN_DEVICE_FUNC Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>&
 Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>
 ::operator=(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
 {
@@ -164,8 +164,8 @@ namespace internal {
   *
   * Helper function to return an arbitrary rotation object to a rotation matrix.
   *
-  * \param Scalar the numeric type of the matrix coefficients
-  * \param Dim the dimension of the current space
+  * \tparam Scalar the numeric type of the matrix coefficients
+  * \tparam Dim the dimension of the current space
   *
   * It returns a Dim x Dim fixed size matrix.
   *
@@ -179,20 +179,20 @@ namespace internal {
   * \sa class Transform, class Rotation2D, class Quaternion, class AngleAxis
   */
 template<typename Scalar, int Dim>
-static inline Matrix<Scalar,2,2> toRotationMatrix(const Scalar& s)
+EIGEN_DEVICE_FUNC static inline Matrix<Scalar,2,2> toRotationMatrix(const Scalar& s)
 {
   EIGEN_STATIC_ASSERT(Dim==2,YOU_MADE_A_PROGRAMMING_MISTAKE)
   return Rotation2D<Scalar>(s).toRotationMatrix();
 }
 
 template<typename Scalar, int Dim, typename OtherDerived>
-static inline Matrix<Scalar,Dim,Dim> toRotationMatrix(const RotationBase<OtherDerived,Dim>& r)
+EIGEN_DEVICE_FUNC static inline Matrix<Scalar,Dim,Dim> toRotationMatrix(const RotationBase<OtherDerived,Dim>& r)
 {
   return r.toRotationMatrix();
 }
 
 template<typename Scalar, int Dim, typename OtherDerived>
-static inline const MatrixBase<OtherDerived>& toRotationMatrix(const MatrixBase<OtherDerived>& mat)
+EIGEN_DEVICE_FUNC static inline const MatrixBase<OtherDerived>& toRotationMatrix(const MatrixBase<OtherDerived>& mat)
 {
   EIGEN_STATIC_ASSERT(OtherDerived::RowsAtCompileTime==Dim && OtherDerived::ColsAtCompileTime==Dim,
     YOU_MADE_A_PROGRAMMING_MISTAKE)
diff --git a/extern/Eigen3/Eigen/src/Geometry/Scaling.h b/extern/Eigen3/Eigen/src/Geometry/Scaling.h
index 1c25f36fe62..f58ca03d94a 100644
--- a/extern/Eigen3/Eigen/src/Geometry/Scaling.h
+++ b/extern/Eigen3/Eigen/src/Geometry/Scaling.h
@@ -18,7 +18,7 @@ namespace Eigen {
   *
   * \brief Represents a generic uniform scaling transformation
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients.
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients.
   *
   * This class represent a uniform scaling transformation. It is the return
   * type of Scaling(Scalar), and most of the time this is the only way it
@@ -62,10 +62,10 @@ public:
   template<int Dim, int Mode, int Options>
   inline Transform<Scalar,Dim,(int(Mode)==int(Isometry)?Affine:Mode)> operator* (const Transform<Scalar,Dim, Mode, Options>& t) const
   {
-   Transform<Scalar,Dim,(int(Mode)==int(Isometry)?Affine:Mode)> res = t;
-   res.prescale(factor());
-   return res;
-}
+    Transform<Scalar,Dim,(int(Mode)==int(Isometry)?Affine:Mode)> res = t;
+    res.prescale(factor());
+    return res;
+  }
 
   /** Concatenates a uniform scaling and a linear transformation matrix */
   // TODO returns an expression
@@ -104,40 +104,44 @@ public:
 
 };
 
-/** Concatenates a linear transformation matrix and a uniform scaling */
+/** \addtogroup Geometry_Module */
+//@{
+
+/** Concatenates a linear transformation matrix and a uniform scaling
+  * \relates UniformScaling
+  */
 // NOTE this operator is defiend in MatrixBase and not as a friend function
 // of UniformScaling to fix an internal crash of Intel's ICC
-template<typename Derived> typename MatrixBase<Derived>::ScalarMultipleReturnType
-MatrixBase<Derived>::operator*(const UniformScaling<Scalar>& s) const
-{ return derived() * s.factor(); }
+template<typename Derived,typename Scalar>
+EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,Scalar,product)
+operator*(const MatrixBase<Derived>& matrix, const UniformScaling<Scalar>& s)
+{ return matrix.derived() * s.factor(); }
 
 /** Constructs a uniform scaling from scale factor \a s */
-static inline UniformScaling<float> Scaling(float s) { return UniformScaling<float>(s); }
+inline UniformScaling<float> Scaling(float s) { return UniformScaling<float>(s); }
 /** Constructs a uniform scaling from scale factor \a s */
-static inline UniformScaling<double> Scaling(double s) { return UniformScaling<double>(s); }
+inline UniformScaling<double> Scaling(double s) { return UniformScaling<double>(s); }
 /** Constructs a uniform scaling from scale factor \a s */
 template<typename RealScalar>
-static inline UniformScaling<std::complex<RealScalar> > Scaling(const std::complex<RealScalar>& s)
+inline UniformScaling<std::complex<RealScalar> > Scaling(const std::complex<RealScalar>& s)
 { return UniformScaling<std::complex<RealScalar> >(s); }
 
 /** Constructs a 2D axis aligned scaling */
 template<typename Scalar>
-static inline DiagonalMatrix<Scalar,2> Scaling(const Scalar& sx, const Scalar& sy)
+inline DiagonalMatrix<Scalar,2> Scaling(const Scalar& sx, const Scalar& sy)
 { return DiagonalMatrix<Scalar,2>(sx, sy); }
 /** Constructs a 3D axis aligned scaling */
 template<typename Scalar>
-static inline DiagonalMatrix<Scalar,3> Scaling(const Scalar& sx, const Scalar& sy, const Scalar& sz)
+inline DiagonalMatrix<Scalar,3> Scaling(const Scalar& sx, const Scalar& sy, const Scalar& sz)
 { return DiagonalMatrix<Scalar,3>(sx, sy, sz); }
 
 /** Constructs an axis aligned scaling expression from vector expression \a coeffs
   * This is an alias for coeffs.asDiagonal()
   */
 template<typename Derived>
-static inline const DiagonalWrapper<const Derived> Scaling(const MatrixBase<Derived>& coeffs)
+inline const DiagonalWrapper<const Derived> Scaling(const MatrixBase<Derived>& coeffs)
 { return coeffs.asDiagonal(); }
 
-/** \addtogroup Geometry_Module */
-//@{
 /** \deprecated */
 typedef DiagonalMatrix<float, 2> AlignedScaling2f;
 /** \deprecated */
diff --git a/extern/Eigen3/Eigen/src/Geometry/Transform.h b/extern/Eigen3/Eigen/src/Geometry/Transform.h
index e786e535695..3f31ee45df9 100644
--- a/extern/Eigen3/Eigen/src/Geometry/Transform.h
+++ b/extern/Eigen3/Eigen/src/Geometry/Transform.h
@@ -32,7 +32,8 @@ template< typename TransformType,
           typename MatrixType,
           int Case = transform_traits<TransformType>::IsProjective ? 0
                    : int(MatrixType::RowsAtCompileTime) == int(transform_traits<TransformType>::HDim) ? 1
-                   : 2>
+                   : 2,
+          int RhsCols = MatrixType::ColsAtCompileTime>
 struct transform_right_product_impl;
 
 template< typename Other,
@@ -62,6 +63,22 @@ struct transform_construct_from_matrix;
 
 template<typename TransformType> struct transform_take_affine_part;
 
+template<typename _Scalar, int _Dim, int _Mode, int _Options>
+struct traits<Transform<_Scalar,_Dim,_Mode,_Options> >
+{
+  typedef _Scalar Scalar;
+  typedef Eigen::Index StorageIndex;
+  typedef Dense StorageKind;
+  enum {
+    Dim1 = _Dim==Dynamic ? _Dim : _Dim + 1,
+    RowsAtCompileTime = _Mode==Projective ? Dim1 : _Dim,
+    ColsAtCompileTime = Dim1,
+    MaxRowsAtCompileTime = RowsAtCompileTime,
+    MaxColsAtCompileTime = ColsAtCompileTime,
+    Flags = 0
+  };
+};
+
 template<int Mode> struct transform_make_affine;
 
 } // end namespace internal
@@ -102,15 +119,15 @@ template<int Mode> struct transform_make_affine;
   *
   * However, unlike a plain matrix, the Transform class provides many features
   * simplifying both its assembly and usage. In particular, it can be composed
-  * with any other transformations (Transform,Translation,RotationBase,Matrix)
+  * with any other transformations (Transform,Translation,RotationBase,DiagonalMatrix)
   * and can be directly used to transform implicit homogeneous vectors. All these
   * operations are handled via the operator*. For the composition of transformations,
   * its principle consists to first convert the right/left hand sides of the product
   * to a compatible (Dim+1)^2 matrix and then perform a pure matrix product.
   * Of course, internally, operator* tries to perform the minimal number of operations
   * according to the nature of each terms. Likewise, when applying the transform
-  * to non homogeneous vectors, the latters are automatically promoted to homogeneous
-  * one before doing the matrix product. The convertions to homogeneous representations
+  * to points, the latters are automatically promoted to homogeneous vectors
+  * before doing the matrix product. The conventions to homogeneous representations
   * are performed as follow:
   *
   * \b Translation t (Dim)x(1):
@@ -124,7 +141,7 @@ template<int Mode> struct transform_make_affine;
   * R & 0\\
   * 0\,...\,0 & 1
   * \end{array} \right) \f$
-  *
+  *<!--
   * \b Linear \b Matrix L (Dim)x(Dim):
   * \f$ \left( \begin{array}{cc}
   * L & 0\\
@@ -136,14 +153,20 @@ template<int Mode> struct transform_make_affine;
   * A\\
   * 0\,...\,0\,1
   * \end{array} \right) \f$
+  *-->
+  * \b Scaling \b DiagonalMatrix S (Dim)x(Dim):
+  * \f$ \left( \begin{array}{cc}
+  * S & 0\\
+  * 0\,...\,0 & 1
+  * \end{array} \right) \f$
   *
-  * \b Column \b vector v (Dim)x(1):
+  * \b Column \b point v (Dim)x(1):
   * \f$ \left( \begin{array}{c}
   * v\\
   * 1
   * \end{array} \right) \f$
   *
-  * \b Set \b of \b column \b vectors V1...Vn (Dim)x(n):
+  * \b Set \b of \b column \b points V1...Vn (Dim)x(n):
   * \f$ \left( \begin{array}{ccc}
   * v_1 & ... & v_n\\
   * 1 & ... & 1
@@ -170,7 +193,7 @@ template<int Mode> struct transform_make_affine;
   * preprocessor token EIGEN_QT_SUPPORT is defined.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_TRANSFORM_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_TRANSFORM_PLUGIN.
   *
   * \sa class Matrix, class Quaternion
   */
@@ -188,7 +211,8 @@ public:
   };
   /** the scalar type of the coefficients */
   typedef _Scalar Scalar;
-  typedef DenseIndex Index;
+  typedef Eigen::Index StorageIndex;
+  typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
   /** type of the matrix used to represent the transformation */
   typedef typename internal::make_proper_matrix_type<Scalar,Rows,HDim,Options>::type MatrixType;
   /** constified MatrixType */
@@ -210,9 +234,9 @@ public:
   /** type of a vector */
   typedef Matrix<Scalar,Dim,1> VectorType;
   /** type of a read/write reference to the translation part of the rotation */
-  typedef Block<MatrixType,Dim,1,int(Mode)==(AffineCompact)> TranslationPart;
+  typedef Block<MatrixType,Dim,1,!(internal::traits<MatrixType>::Flags & RowMajorBit)> TranslationPart;
   /** type of a read reference to the translation part of the rotation */
-  typedef const Block<ConstMatrixType,Dim,1,int(Mode)==(AffineCompact)> ConstTranslationPart;
+  typedef const Block<ConstMatrixType,Dim,1,!(internal::traits<MatrixType>::Flags & RowMajorBit)> ConstTranslationPart;
   /** corresponding translation type */
   typedef Translation<Scalar,Dim> TranslationType;
   
@@ -229,43 +253,43 @@ public:
 
   /** Default constructor without initialization of the meaningful coefficients.
     * If Mode==Affine, then the last row is set to [0 ... 0 1] */
-  inline Transform()
+  EIGEN_DEVICE_FUNC inline Transform()
   {
     check_template_params();
     internal::transform_make_affine<(int(Mode)==Affine) ? Affine : AffineCompact>::run(m_matrix);
   }
 
-  inline Transform(const Transform& other)
+  EIGEN_DEVICE_FUNC inline Transform(const Transform& other)
   {
     check_template_params();
     m_matrix = other.m_matrix;
   }
 
-  inline explicit Transform(const TranslationType& t)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const TranslationType& t)
   {
     check_template_params();
     *this = t;
   }
-  inline explicit Transform(const UniformScaling<Scalar>& s)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const UniformScaling<Scalar>& s)
   {
     check_template_params();
     *this = s;
   }
   template<typename Derived>
-  inline explicit Transform(const RotationBase<Derived, Dim>& r)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const RotationBase<Derived, Dim>& r)
   {
     check_template_params();
     *this = r;
   }
 
-  inline Transform& operator=(const Transform& other)
+  EIGEN_DEVICE_FUNC inline Transform& operator=(const Transform& other)
   { m_matrix = other.m_matrix; return *this; }
 
   typedef internal::transform_take_affine_part<Transform> take_affine_part;
 
   /** Constructs and initializes a transformation from a Dim^2 or a (Dim+1)^2 matrix. */
   template<typename OtherDerived>
-  inline explicit Transform(const EigenBase<OtherDerived>& other)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const EigenBase<OtherDerived>& other)
   {
     EIGEN_STATIC_ASSERT((internal::is_same<Scalar,typename OtherDerived::Scalar>::value),
       YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY);
@@ -276,7 +300,7 @@ public:
 
   /** Set \c *this from a Dim^2 or (Dim+1)^2 matrix. */
   template<typename OtherDerived>
-  inline Transform& operator=(const EigenBase<OtherDerived>& other)
+  EIGEN_DEVICE_FUNC inline Transform& operator=(const EigenBase<OtherDerived>& other)
   {
     EIGEN_STATIC_ASSERT((internal::is_same<Scalar,typename OtherDerived::Scalar>::value),
       YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY);
@@ -286,7 +310,7 @@ public:
   }
   
   template<int OtherOptions>
-  inline Transform(const Transform<Scalar,Dim,Mode,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC inline Transform(const Transform<Scalar,Dim,Mode,OtherOptions>& other)
   {
     check_template_params();
     // only the options change, we can directly copy the matrices
@@ -294,7 +318,7 @@ public:
   }
 
   template<int OtherMode,int OtherOptions>
-  inline Transform(const Transform<Scalar,Dim,OtherMode,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC inline Transform(const Transform<Scalar,Dim,OtherMode,OtherOptions>& other)
   {
     check_template_params();
     // prevent conversions as:
@@ -335,14 +359,14 @@ public:
   }
 
   template<typename OtherDerived>
-  Transform(const ReturnByValue<OtherDerived>& other)
+  EIGEN_DEVICE_FUNC Transform(const ReturnByValue<OtherDerived>& other)
   {
     check_template_params();
     other.evalTo(*this);
   }
 
   template<typename OtherDerived>
-  Transform& operator=(const ReturnByValue<OtherDerived>& other)
+  EIGEN_DEVICE_FUNC Transform& operator=(const ReturnByValue<OtherDerived>& other)
   {
     other.evalTo(*this);
     return *this;
@@ -356,60 +380,76 @@ public:
   inline Transform& operator=(const QTransform& other);
   inline QTransform toQTransform(void) const;
   #endif
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return int(Mode)==int(Projective) ? m_matrix.cols() : (m_matrix.cols()-1); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_matrix.cols(); }
 
   /** shortcut for m_matrix(row,col);
     * \sa MatrixBase::operator(Index,Index) const */
-  inline Scalar operator() (Index row, Index col) const { return m_matrix(row,col); }
+  EIGEN_DEVICE_FUNC inline Scalar operator() (Index row, Index col) const { return m_matrix(row,col); }
   /** shortcut for m_matrix(row,col);
     * \sa MatrixBase::operator(Index,Index) */
-  inline Scalar& operator() (Index row, Index col) { return m_matrix(row,col); }
+  EIGEN_DEVICE_FUNC inline Scalar& operator() (Index row, Index col) { return m_matrix(row,col); }
 
   /** \returns a read-only expression of the transformation matrix */
-  inline const MatrixType& matrix() const { return m_matrix; }
+  EIGEN_DEVICE_FUNC inline const MatrixType& matrix() const { return m_matrix; }
   /** \returns a writable expression of the transformation matrix */
-  inline MatrixType& matrix() { return m_matrix; }
+  EIGEN_DEVICE_FUNC inline MatrixType& matrix() { return m_matrix; }
 
   /** \returns a read-only expression of the linear part of the transformation */
-  inline ConstLinearPart linear() const { return ConstLinearPart(m_matrix,0,0); }
+  EIGEN_DEVICE_FUNC inline ConstLinearPart linear() const { return ConstLinearPart(m_matrix,0,0); }
   /** \returns a writable expression of the linear part of the transformation */
-  inline LinearPart linear() { return LinearPart(m_matrix,0,0); }
+  EIGEN_DEVICE_FUNC inline LinearPart linear() { return LinearPart(m_matrix,0,0); }
 
   /** \returns a read-only expression of the Dim x HDim affine part of the transformation */
-  inline ConstAffinePart affine() const { return take_affine_part::run(m_matrix); }
+  EIGEN_DEVICE_FUNC inline ConstAffinePart affine() const { return take_affine_part::run(m_matrix); }
   /** \returns a writable expression of the Dim x HDim affine part of the transformation */
-  inline AffinePart affine() { return take_affine_part::run(m_matrix); }
+  EIGEN_DEVICE_FUNC inline AffinePart affine() { return take_affine_part::run(m_matrix); }
 
   /** \returns a read-only expression of the translation vector of the transformation */
-  inline ConstTranslationPart translation() const { return ConstTranslationPart(m_matrix,0,Dim); }
+  EIGEN_DEVICE_FUNC inline ConstTranslationPart translation() const { return ConstTranslationPart(m_matrix,0,Dim); }
   /** \returns a writable expression of the translation vector of the transformation */
-  inline TranslationPart translation() { return TranslationPart(m_matrix,0,Dim); }
+  EIGEN_DEVICE_FUNC inline TranslationPart translation() { return TranslationPart(m_matrix,0,Dim); }
 
-  /** \returns an expression of the product between the transform \c *this and a matrix expression \a other
+  /** \returns an expression of the product between the transform \c *this and a matrix expression \a other.
     *
-    * The right hand side \a other might be either:
-    * \li a vector of size Dim,
+    * The right-hand-side \a other can be either:
     * \li an homogeneous vector of size Dim+1,
-    * \li a set of vectors of size Dim x Dynamic,
-    * \li a set of homogeneous vectors of size Dim+1 x Dynamic,
-    * \li a linear transformation matrix of size Dim x Dim,
-    * \li an affine transformation matrix of size Dim x Dim+1,
+    * \li a set of homogeneous vectors of size Dim+1 x N,
     * \li a transformation matrix of size Dim+1 x Dim+1.
+    *
+    * Moreover, if \c *this represents an affine transformation (i.e., Mode!=Projective), then \a other can also be:
+    * \li a point of size Dim (computes: \code this->linear() * other + this->translation()\endcode),
+    * \li a set of N points as a Dim x N matrix (computes: \code (this->linear() * other).colwise() + this->translation()\endcode),
+    *
+    * In all cases, the return type is a matrix or vector of same sizes as the right-hand-side \a other.
+    *
+    * If you want to interpret \a other as a linear or affine transformation, then first convert it to a Transform<> type,
+    * or do your own cooking.
+    *
+    * Finally, if you want to apply Affine transformations to vectors, then explicitly apply the linear part only:
+    * \code
+    * Affine3f A;
+    * Vector3f v1, v2;
+    * v2 = A.linear() * v1;
+    * \endcode
+    *
     */
   // note: this function is defined here because some compilers cannot find the respective declaration
   template<typename OtherDerived>
-  EIGEN_STRONG_INLINE const typename internal::transform_right_product_impl<Transform, OtherDerived>::ResultType
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename internal::transform_right_product_impl<Transform, OtherDerived>::ResultType
   operator * (const EigenBase<OtherDerived> &other) const
   { return internal::transform_right_product_impl<Transform, OtherDerived>::run(*this,other.derived()); }
 
   /** \returns the product expression of a transformation matrix \a a times a transform \a b
     *
-    * The left hand side \a other might be either:
+    * The left hand side \a other can be either:
     * \li a linear transformation matrix of size Dim x Dim,
     * \li an affine transformation matrix of size Dim x Dim+1,
     * \li a general transformation matrix of size Dim+1 x Dim+1.
     */
   template<typename OtherDerived> friend
-  inline const typename internal::transform_left_product_impl<OtherDerived,Mode,Options,_Dim,_Dim+1>::ResultType
+  EIGEN_DEVICE_FUNC inline const typename internal::transform_left_product_impl<OtherDerived,Mode,Options,_Dim,_Dim+1>::ResultType
     operator * (const EigenBase<OtherDerived> &a, const Transform &b)
   { return internal::transform_left_product_impl<OtherDerived,Mode,Options,Dim,HDim>::run(a.derived(),b); }
 
@@ -420,11 +460,11 @@ public:
     * mode is no isometry. In that case, the returned transform is an affinity.
     */
   template<typename DiagonalDerived>
-  inline const TransformTimeDiagonalReturnType
+  EIGEN_DEVICE_FUNC inline const TransformTimeDiagonalReturnType
     operator * (const DiagonalBase<DiagonalDerived> &b) const
   {
     TransformTimeDiagonalReturnType res(*this);
-    res.linear() *= b;
+    res.linearExt() *= b;
     return res;
   }
 
@@ -435,7 +475,7 @@ public:
     * mode is no isometry. In that case, the returned transform is an affinity.
     */
   template<typename DiagonalDerived>
-  friend inline TransformTimeDiagonalReturnType
+  EIGEN_DEVICE_FUNC friend inline TransformTimeDiagonalReturnType
     operator * (const DiagonalBase<DiagonalDerived> &a, const Transform &b)
   {
     TransformTimeDiagonalReturnType res;
@@ -447,15 +487,15 @@ public:
   }
 
   template<typename OtherDerived>
-  inline Transform& operator*=(const EigenBase<OtherDerived>& other) { return *this = *this * other; }
+  EIGEN_DEVICE_FUNC inline Transform& operator*=(const EigenBase<OtherDerived>& other) { return *this = *this * other; }
 
   /** Concatenates two transformations */
-  inline const Transform operator * (const Transform& other) const
+  EIGEN_DEVICE_FUNC inline const Transform operator * (const Transform& other) const
   {
     return internal::transform_transform_product_impl<Transform,Transform>::run(*this,other);
   }
   
-  #ifdef __INTEL_COMPILER
+  #if EIGEN_COMP_ICC
 private:
   // this intermediate structure permits to workaround a bug in ICC 11:
   //   error: template instantiation resulted in unexpected function type of "Eigen::Transform<double, 3, 32, 0>
@@ -482,7 +522,7 @@ public:
   #else
   /** Concatenates two different transformations */
   template<int OtherMode,int OtherOptions>
-  inline typename internal::transform_transform_product_impl<Transform,Transform<Scalar,Dim,OtherMode,OtherOptions> >::ResultType
+  EIGEN_DEVICE_FUNC inline typename internal::transform_transform_product_impl<Transform,Transform<Scalar,Dim,OtherMode,OtherOptions> >::ResultType
     operator * (const Transform<Scalar,Dim,OtherMode,OtherOptions>& other) const
   {
     return internal::transform_transform_product_impl<Transform,Transform<Scalar,Dim,OtherMode,OtherOptions> >::run(*this,other);
@@ -490,79 +530,98 @@ public:
   #endif
 
   /** \sa MatrixBase::setIdentity() */
-  void setIdentity() { m_matrix.setIdentity(); }
+  EIGEN_DEVICE_FUNC void setIdentity() { m_matrix.setIdentity(); }
 
   /**
    * \brief Returns an identity transformation.
    * \todo In the future this function should be returning a Transform expression.
    */
-  static const Transform Identity()
+  EIGEN_DEVICE_FUNC static const Transform Identity()
   {
     return Transform(MatrixType::Identity());
   }
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC 
   inline Transform& scale(const MatrixBase<OtherDerived> &other);
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   inline Transform& prescale(const MatrixBase<OtherDerived> &other);
 
-  inline Transform& scale(const Scalar& s);
-  inline Transform& prescale(const Scalar& s);
+  EIGEN_DEVICE_FUNC inline Transform& scale(const Scalar& s);
+  EIGEN_DEVICE_FUNC inline Transform& prescale(const Scalar& s);
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   inline Transform& translate(const MatrixBase<OtherDerived> &other);
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   inline Transform& pretranslate(const MatrixBase<OtherDerived> &other);
 
   template<typename RotationType>
+  EIGEN_DEVICE_FUNC
   inline Transform& rotate(const RotationType& rotation);
 
   template<typename RotationType>
+  EIGEN_DEVICE_FUNC
   inline Transform& prerotate(const RotationType& rotation);
 
-  Transform& shear(const Scalar& sx, const Scalar& sy);
-  Transform& preshear(const Scalar& sx, const Scalar& sy);
+  EIGEN_DEVICE_FUNC Transform& shear(const Scalar& sx, const Scalar& sy);
+  EIGEN_DEVICE_FUNC Transform& preshear(const Scalar& sx, const Scalar& sy);
 
-  inline Transform& operator=(const TranslationType& t);
+  EIGEN_DEVICE_FUNC inline Transform& operator=(const TranslationType& t);
+  
+  EIGEN_DEVICE_FUNC
   inline Transform& operator*=(const TranslationType& t) { return translate(t.vector()); }
-  inline Transform operator*(const TranslationType& t) const;
+  
+  EIGEN_DEVICE_FUNC inline Transform operator*(const TranslationType& t) const;
 
+  EIGEN_DEVICE_FUNC 
   inline Transform& operator=(const UniformScaling<Scalar>& t);
+  
+  EIGEN_DEVICE_FUNC
   inline Transform& operator*=(const UniformScaling<Scalar>& s) { return scale(s.factor()); }
-  inline Transform<Scalar,Dim,(int(Mode)==int(Isometry)?int(Affine):int(Mode))> operator*(const UniformScaling<Scalar>& s) const
+  
+  EIGEN_DEVICE_FUNC
+  inline TransformTimeDiagonalReturnType operator*(const UniformScaling<Scalar>& s) const
   {
-    Transform<Scalar,Dim,(int(Mode)==int(Isometry)?int(Affine):int(Mode)),Options> res = *this;
+    TransformTimeDiagonalReturnType res = *this;
     res.scale(s.factor());
     return res;
   }
 
-  inline Transform& operator*=(const DiagonalMatrix<Scalar,Dim>& s) { linear() *= s; return *this; }
+  EIGEN_DEVICE_FUNC
+  inline Transform& operator*=(const DiagonalMatrix<Scalar,Dim>& s) { linearExt() *= s; return *this; }
 
   template<typename Derived>
-  inline Transform& operator=(const RotationBase<Derived,Dim>& r);
+  EIGEN_DEVICE_FUNC inline Transform& operator=(const RotationBase<Derived,Dim>& r);
   template<typename Derived>
-  inline Transform& operator*=(const RotationBase<Derived,Dim>& r) { return rotate(r.toRotationMatrix()); }
+  EIGEN_DEVICE_FUNC inline Transform& operator*=(const RotationBase<Derived,Dim>& r) { return rotate(r.toRotationMatrix()); }
   template<typename Derived>
-  inline Transform operator*(const RotationBase<Derived,Dim>& r) const;
+  EIGEN_DEVICE_FUNC inline Transform operator*(const RotationBase<Derived,Dim>& r) const;
 
-  const LinearMatrixType rotation() const;
+  EIGEN_DEVICE_FUNC const LinearMatrixType rotation() const;
   template<typename RotationMatrixType, typename ScalingMatrixType>
+  EIGEN_DEVICE_FUNC
   void computeRotationScaling(RotationMatrixType *rotation, ScalingMatrixType *scaling) const;
   template<typename ScalingMatrixType, typename RotationMatrixType>
+  EIGEN_DEVICE_FUNC
   void computeScalingRotation(ScalingMatrixType *scaling, RotationMatrixType *rotation) const;
 
   template<typename PositionDerived, typename OrientationType, typename ScaleDerived>
+  EIGEN_DEVICE_FUNC
   Transform& fromPositionOrientationScale(const MatrixBase<PositionDerived> &position,
     const OrientationType& orientation, const MatrixBase<ScaleDerived> &scale);
 
+  EIGEN_DEVICE_FUNC
   inline Transform inverse(TransformTraits traits = (TransformTraits)Mode) const;
 
   /** \returns a const pointer to the column major internal matrix */
-  const Scalar* data() const { return m_matrix.data(); }
+  EIGEN_DEVICE_FUNC const Scalar* data() const { return m_matrix.data(); }
   /** \returns a non-const pointer to the column major internal matrix */
-  Scalar* data() { return m_matrix.data(); }
+  EIGEN_DEVICE_FUNC Scalar* data() { return m_matrix.data(); }
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -570,12 +629,12 @@ public:
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Transform,Transform<NewScalarType,Dim,Mode,Options> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Transform,Transform<NewScalarType,Dim,Mode,Options> >::type cast() const
   { return typename internal::cast_return_type<Transform,Transform<NewScalarType,Dim,Mode,Options> >::type(*this); }
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit Transform(const Transform<OtherScalarType,Dim,Mode,Options>& other)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const Transform<OtherScalarType,Dim,Mode,Options>& other)
   {
     check_template_params();
     m_matrix = other.matrix().template cast<Scalar>();
@@ -585,12 +644,12 @@ public:
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const Transform& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const Transform& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_matrix.isApprox(other.m_matrix, prec); }
 
   /** Sets the last row to [0 ... 0 1]
     */
-  void makeAffine()
+  EIGEN_DEVICE_FUNC void makeAffine()
   {
     internal::transform_make_affine<int(Mode)>::run(m_matrix);
   }
@@ -599,26 +658,26 @@ public:
     * \returns the Dim x Dim linear part if the transformation is affine,
     *          and the HDim x Dim part for projective transformations.
     */
-  inline Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,Dim> linearExt()
+  EIGEN_DEVICE_FUNC inline Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,Dim> linearExt()
   { return m_matrix.template block<int(Mode)==int(Projective)?HDim:Dim,Dim>(0,0); }
   /** \internal
     * \returns the Dim x Dim linear part if the transformation is affine,
     *          and the HDim x Dim part for projective transformations.
     */
-  inline const Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,Dim> linearExt() const
+  EIGEN_DEVICE_FUNC inline const Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,Dim> linearExt() const
   { return m_matrix.template block<int(Mode)==int(Projective)?HDim:Dim,Dim>(0,0); }
 
   /** \internal
     * \returns the translation part if the transformation is affine,
     *          and the last column for projective transformations.
     */
-  inline Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,1> translationExt()
+  EIGEN_DEVICE_FUNC inline Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,1> translationExt()
   { return m_matrix.template block<int(Mode)==int(Projective)?HDim:Dim,1>(0,Dim); }
   /** \internal
     * \returns the translation part if the transformation is affine,
     *          and the last column for projective transformations.
     */
-  inline const Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,1> translationExt() const
+  EIGEN_DEVICE_FUNC inline const Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,1> translationExt() const
   { return m_matrix.template block<int(Mode)==int(Projective)?HDim:Dim,1>(0,Dim); }
 
 
@@ -628,7 +687,7 @@ public:
   
 protected:
   #ifndef EIGEN_PARSED_BY_DOXYGEN
-    static EIGEN_STRONG_INLINE void check_template_params()
+    EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void check_template_params()
     {
       EIGEN_STATIC_ASSERT((Options & (DontAlign|RowMajor)) == Options, INVALID_MATRIX_TEMPLATE_PARAMETERS)
     }
@@ -696,9 +755,13 @@ template<typename Scalar, int Dim, int Mode,int Options>
 Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const QMatrix& other)
 {
   EIGEN_STATIC_ASSERT(Dim==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
-  m_matrix << other.m11(), other.m21(), other.dx(),
-              other.m12(), other.m22(), other.dy(),
-              0, 0, 1;
+  if (Mode == int(AffineCompact))
+    m_matrix << other.m11(), other.m21(), other.dx(),
+                other.m12(), other.m22(), other.dy();
+  else
+    m_matrix << other.m11(), other.m21(), other.dx(),
+                other.m12(), other.m22(), other.dy(),
+                0, 0, 1;
   return *this;
 }
 
@@ -777,7 +840,7 @@ QTransform Transform<Scalar,Dim,Mode,Options>::toQTransform(void) const
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename OtherDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::scale(const MatrixBase<OtherDerived> &other)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
@@ -791,7 +854,7 @@ Transform<Scalar,Dim,Mode,Options>::scale(const MatrixBase<OtherDerived> &other)
   * \sa prescale(Scalar)
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::scale(const Scalar& s)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::scale(const Scalar& s)
 {
   EIGEN_STATIC_ASSERT(Mode!=int(Isometry), THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS)
   linearExt() *= s;
@@ -804,12 +867,12 @@ inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::s
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename OtherDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::prescale(const MatrixBase<OtherDerived> &other)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
   EIGEN_STATIC_ASSERT(Mode!=int(Isometry), THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS)
-  m_matrix.template block<Dim,HDim>(0,0).noalias() = (other.asDiagonal() * m_matrix.template block<Dim,HDim>(0,0));
+  affine().noalias() = (other.asDiagonal() * affine());
   return *this;
 }
 
@@ -818,7 +881,7 @@ Transform<Scalar,Dim,Mode,Options>::prescale(const MatrixBase<OtherDerived> &oth
   * \sa scale(Scalar)
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::prescale(const Scalar& s)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::prescale(const Scalar& s)
 {
   EIGEN_STATIC_ASSERT(Mode!=int(Isometry), THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS)
   m_matrix.template topRows<Dim>() *= s;
@@ -831,7 +894,7 @@ inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::p
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename OtherDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::translate(const MatrixBase<OtherDerived> &other)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
@@ -845,7 +908,7 @@ Transform<Scalar,Dim,Mode,Options>::translate(const MatrixBase<OtherDerived> &ot
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename OtherDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::pretranslate(const MatrixBase<OtherDerived> &other)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
@@ -875,7 +938,7 @@ Transform<Scalar,Dim,Mode,Options>::pretranslate(const MatrixBase<OtherDerived>
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename RotationType>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::rotate(const RotationType& rotation)
 {
   linearExt() *= internal::toRotationMatrix<Scalar,Dim>(rotation);
@@ -891,7 +954,7 @@ Transform<Scalar,Dim,Mode,Options>::rotate(const RotationType& rotation)
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename RotationType>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::prerotate(const RotationType& rotation)
 {
   m_matrix.template block<Dim,HDim>(0,0) = internal::toRotationMatrix<Scalar,Dim>(rotation)
@@ -905,7 +968,7 @@ Transform<Scalar,Dim,Mode,Options>::prerotate(const RotationType& rotation)
   * \sa preshear()
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::shear(const Scalar& sx, const Scalar& sy)
 {
   EIGEN_STATIC_ASSERT(int(Dim)==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
@@ -921,7 +984,7 @@ Transform<Scalar,Dim,Mode,Options>::shear(const Scalar& sx, const Scalar& sy)
   * \sa shear()
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::preshear(const Scalar& sx, const Scalar& sy)
 {
   EIGEN_STATIC_ASSERT(int(Dim)==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
@@ -935,7 +998,7 @@ Transform<Scalar,Dim,Mode,Options>::preshear(const Scalar& sx, const Scalar& sy)
 ******************************************************/
 
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const TranslationType& t)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const TranslationType& t)
 {
   linear().setIdentity();
   translation() = t.vector();
@@ -944,7 +1007,7 @@ inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::o
 }
 
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::operator*(const TranslationType& t) const
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::operator*(const TranslationType& t) const
 {
   Transform res = *this;
   res.translate(t.vector());
@@ -952,7 +1015,7 @@ inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::op
 }
 
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const UniformScaling<Scalar>& s)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const UniformScaling<Scalar>& s)
 {
   m_matrix.setZero();
   linear().diagonal().fill(s.factor());
@@ -962,7 +1025,7 @@ inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::o
 
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename Derived>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const RotationBase<Derived,Dim>& r)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const RotationBase<Derived,Dim>& r)
 {
   linear() = internal::toRotationMatrix<Scalar,Dim>(r);
   translation().setZero();
@@ -972,7 +1035,7 @@ inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::o
 
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename Derived>
-inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::operator*(const RotationBase<Derived,Dim>& r) const
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::operator*(const RotationBase<Derived,Dim>& r) const
 {
   Transform res = *this;
   res.rotate(r.derived());
@@ -991,7 +1054,7 @@ inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::op
   * \sa computeRotationScaling(), computeScalingRotation(), class SVD
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-const typename Transform<Scalar,Dim,Mode,Options>::LinearMatrixType
+EIGEN_DEVICE_FUNC const typename Transform<Scalar,Dim,Mode,Options>::LinearMatrixType
 Transform<Scalar,Dim,Mode,Options>::rotation() const
 {
   LinearMatrixType result;
@@ -1013,7 +1076,7 @@ Transform<Scalar,Dim,Mode,Options>::rotation() const
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename RotationMatrixType, typename ScalingMatrixType>
-void Transform<Scalar,Dim,Mode,Options>::computeRotationScaling(RotationMatrixType *rotation, ScalingMatrixType *scaling) const
+EIGEN_DEVICE_FUNC void Transform<Scalar,Dim,Mode,Options>::computeRotationScaling(RotationMatrixType *rotation, ScalingMatrixType *scaling) const
 {
   JacobiSVD<LinearMatrixType> svd(linear(), ComputeFullU | ComputeFullV);
 
@@ -1029,7 +1092,7 @@ void Transform<Scalar,Dim,Mode,Options>::computeRotationScaling(RotationMatrixTy
   }
 }
 
-/** decomposes the linear part of the transformation as a product rotation x scaling, the scaling being
+/** decomposes the linear part of the transformation as a product scaling x rotation, the scaling being
   * not necessarily positive.
   *
   * If either pointer is zero, the corresponding computation is skipped.
@@ -1042,7 +1105,7 @@ void Transform<Scalar,Dim,Mode,Options>::computeRotationScaling(RotationMatrixTy
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename ScalingMatrixType, typename RotationMatrixType>
-void Transform<Scalar,Dim,Mode,Options>::computeScalingRotation(ScalingMatrixType *scaling, RotationMatrixType *rotation) const
+EIGEN_DEVICE_FUNC void Transform<Scalar,Dim,Mode,Options>::computeScalingRotation(ScalingMatrixType *scaling, RotationMatrixType *rotation) const
 {
   JacobiSVD<LinearMatrixType> svd(linear(), ComputeFullU | ComputeFullV);
 
@@ -1063,7 +1126,7 @@ void Transform<Scalar,Dim,Mode,Options>::computeScalingRotation(ScalingMatrixTyp
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename PositionDerived, typename OrientationType, typename ScaleDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::fromPositionOrientationScale(const MatrixBase<PositionDerived> &position,
   const OrientationType& orientation, const MatrixBase<ScaleDerived> &scale)
 {
@@ -1080,7 +1143,7 @@ template<int Mode>
 struct transform_make_affine
 {
   template<typename MatrixType>
-  static void run(MatrixType &mat)
+  EIGEN_DEVICE_FUNC static void run(MatrixType &mat)
   {
     static const int Dim = MatrixType::ColsAtCompileTime-1;
     mat.template block<1,Dim>(Dim,0).setZero();
@@ -1091,21 +1154,21 @@ struct transform_make_affine
 template<>
 struct transform_make_affine<AffineCompact>
 {
-  template<typename MatrixType> static void run(MatrixType &) { }
+  template<typename MatrixType> EIGEN_DEVICE_FUNC static void run(MatrixType &) { }
 };
     
 // selector needed to avoid taking the inverse of a 3x4 matrix
 template<typename TransformType, int Mode=TransformType::Mode>
 struct projective_transform_inverse
 {
-  static inline void run(const TransformType&, TransformType&)
+  EIGEN_DEVICE_FUNC static inline void run(const TransformType&, TransformType&)
   {}
 };
 
 template<typename TransformType>
 struct projective_transform_inverse<TransformType, Projective>
 {
-  static inline void run(const TransformType& m, TransformType& res)
+  EIGEN_DEVICE_FUNC static inline void run(const TransformType& m, TransformType& res)
   {
     res.matrix() = m.matrix().inverse();
   }
@@ -1135,7 +1198,7 @@ struct projective_transform_inverse<TransformType, Projective>
   * \sa MatrixBase::inverse()
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-Transform<Scalar,Dim,Mode,Options>
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>
 Transform<Scalar,Dim,Mode,Options>::inverse(TransformTraits hint) const
 {
   Transform res;
@@ -1244,8 +1307,8 @@ struct transform_product_result
   };
 };
 
-template< typename TransformType, typename MatrixType >
-struct transform_right_product_impl< TransformType, MatrixType, 0 >
+template< typename TransformType, typename MatrixType, int RhsCols>
+struct transform_right_product_impl< TransformType, MatrixType, 0, RhsCols>
 {
   typedef typename MatrixType::PlainObject ResultType;
 
@@ -1255,8 +1318,8 @@ struct transform_right_product_impl< TransformType, MatrixType, 0 >
   }
 };
 
-template< typename TransformType, typename MatrixType >
-struct transform_right_product_impl< TransformType, MatrixType, 1 >
+template< typename TransformType, typename MatrixType, int RhsCols>
+struct transform_right_product_impl< TransformType, MatrixType, 1, RhsCols>
 {
   enum { 
     Dim = TransformType::Dim, 
@@ -1281,8 +1344,8 @@ struct transform_right_product_impl< TransformType, MatrixType, 1 >
   }
 };
 
-template< typename TransformType, typename MatrixType >
-struct transform_right_product_impl< TransformType, MatrixType, 2 >
+template< typename TransformType, typename MatrixType, int RhsCols>
+struct transform_right_product_impl< TransformType, MatrixType, 2, RhsCols>
 {
   enum { 
     Dim = TransformType::Dim, 
@@ -1305,6 +1368,30 @@ struct transform_right_product_impl< TransformType, MatrixType, 2 >
   }
 };
 
+template< typename TransformType, typename MatrixType >
+struct transform_right_product_impl< TransformType, MatrixType, 2, 1> // rhs is a vector of size Dim
+{
+  typedef typename TransformType::MatrixType TransformMatrix;
+  enum {
+    Dim = TransformType::Dim,
+    HDim = TransformType::HDim,
+    OtherRows = MatrixType::RowsAtCompileTime,
+    WorkingRows = EIGEN_PLAIN_ENUM_MIN(TransformMatrix::RowsAtCompileTime,HDim)
+  };
+
+  typedef typename MatrixType::PlainObject ResultType;
+
+  static EIGEN_STRONG_INLINE ResultType run(const TransformType& T, const MatrixType& other)
+  {
+    EIGEN_STATIC_ASSERT(OtherRows==Dim, YOU_MIXED_MATRICES_OF_DIFFERENT_SIZES);
+
+    Matrix<typename ResultType::Scalar, Dim+1, 1> rhs;
+    rhs.template head<Dim>() = other; rhs[Dim] = typename ResultType::Scalar(1);
+    Matrix<typename ResultType::Scalar, WorkingRows, 1> res(T.matrix() * rhs);
+    return res.template head<Dim>();
+  }
+};
+
 /**********************************************************
 ***   Specializations of operator* with lhs EigenBase   ***
 **********************************************************/
diff --git a/extern/Eigen3/Eigen/src/Geometry/Translation.h b/extern/Eigen3/Eigen/src/Geometry/Translation.h
index 7fda179cc35..51d9a82ebbd 100644
--- a/extern/Eigen3/Eigen/src/Geometry/Translation.h
+++ b/extern/Eigen3/Eigen/src/Geometry/Translation.h
@@ -18,8 +18,8 @@ namespace Eigen {
   *
   * \brief Represents a translation transformation
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients.
-  * \param _Dim the  dimension of the space, can be a compile time value or Dynamic
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients.
+  * \tparam _Dim the  dimension of the space, can be a compile time value or Dynamic
   *
   * \note This class is not aimed to be used to store a translation transformation,
   * but rather to make easier the constructions and updates of Transform objects.
@@ -51,16 +51,16 @@ protected:
 public:
 
   /** Default constructor without initialization. */
-  Translation() {}
+  EIGEN_DEVICE_FUNC Translation() {}
   /**  */
-  inline Translation(const Scalar& sx, const Scalar& sy)
+  EIGEN_DEVICE_FUNC inline Translation(const Scalar& sx, const Scalar& sy)
   {
     eigen_assert(Dim==2);
     m_coeffs.x() = sx;
     m_coeffs.y() = sy;
   }
   /**  */
-  inline Translation(const Scalar& sx, const Scalar& sy, const Scalar& sz)
+  EIGEN_DEVICE_FUNC inline Translation(const Scalar& sx, const Scalar& sy, const Scalar& sz)
   {
     eigen_assert(Dim==3);
     m_coeffs.x() = sx;
@@ -68,48 +68,48 @@ public:
     m_coeffs.z() = sz;
   }
   /** Constructs and initialize the translation transformation from a vector of translation coefficients */
-  explicit inline Translation(const VectorType& vector) : m_coeffs(vector) {}
+  EIGEN_DEVICE_FUNC explicit inline Translation(const VectorType& vector) : m_coeffs(vector) {}
 
   /** \brief Retruns the x-translation by value. **/
-  inline Scalar x() const { return m_coeffs.x(); }
+  EIGEN_DEVICE_FUNC inline Scalar x() const { return m_coeffs.x(); }
   /** \brief Retruns the y-translation by value. **/
-  inline Scalar y() const { return m_coeffs.y(); }
+  EIGEN_DEVICE_FUNC inline Scalar y() const { return m_coeffs.y(); }
   /** \brief Retruns the z-translation by value. **/
-  inline Scalar z() const { return m_coeffs.z(); }
+  EIGEN_DEVICE_FUNC inline Scalar z() const { return m_coeffs.z(); }
 
   /** \brief Retruns the x-translation as a reference. **/
-  inline Scalar& x() { return m_coeffs.x(); }
+  EIGEN_DEVICE_FUNC inline Scalar& x() { return m_coeffs.x(); }
   /** \brief Retruns the y-translation as a reference. **/
-  inline Scalar& y() { return m_coeffs.y(); }
+  EIGEN_DEVICE_FUNC inline Scalar& y() { return m_coeffs.y(); }
   /** \brief Retruns the z-translation as a reference. **/
-  inline Scalar& z() { return m_coeffs.z(); }
+  EIGEN_DEVICE_FUNC inline Scalar& z() { return m_coeffs.z(); }
 
-  const VectorType& vector() const { return m_coeffs; }
-  VectorType& vector() { return m_coeffs; }
+  EIGEN_DEVICE_FUNC const VectorType& vector() const { return m_coeffs; }
+  EIGEN_DEVICE_FUNC VectorType& vector() { return m_coeffs; }
 
-  const VectorType& translation() const { return m_coeffs; }
-  VectorType& translation() { return m_coeffs; }
+  EIGEN_DEVICE_FUNC const VectorType& translation() const { return m_coeffs; }
+  EIGEN_DEVICE_FUNC VectorType& translation() { return m_coeffs; }
 
   /** Concatenates two translation */
-  inline Translation operator* (const Translation& other) const
+  EIGEN_DEVICE_FUNC inline Translation operator* (const Translation& other) const
   { return Translation(m_coeffs + other.m_coeffs); }
 
   /** Concatenates a translation and a uniform scaling */
-  inline AffineTransformType operator* (const UniformScaling<Scalar>& other) const;
+  EIGEN_DEVICE_FUNC inline AffineTransformType operator* (const UniformScaling<Scalar>& other) const;
 
   /** Concatenates a translation and a linear transformation */
   template<typename OtherDerived>
-  inline AffineTransformType operator* (const EigenBase<OtherDerived>& linear) const;
+  EIGEN_DEVICE_FUNC inline AffineTransformType operator* (const EigenBase<OtherDerived>& linear) const;
 
   /** Concatenates a translation and a rotation */
   template<typename Derived>
-  inline IsometryTransformType operator*(const RotationBase<Derived,Dim>& r) const
+  EIGEN_DEVICE_FUNC inline IsometryTransformType operator*(const RotationBase<Derived,Dim>& r) const
   { return *this * IsometryTransformType(r); }
 
   /** \returns the concatenation of a linear transformation \a l with the translation \a t */
   // its a nightmare to define a templated friend function outside its declaration
   template<typename OtherDerived> friend
-  inline AffineTransformType operator*(const EigenBase<OtherDerived>& linear, const Translation& t)
+  EIGEN_DEVICE_FUNC inline AffineTransformType operator*(const EigenBase<OtherDerived>& linear, const Translation& t)
   {
     AffineTransformType res;
     res.matrix().setZero();
@@ -122,7 +122,7 @@ public:
 
   /** Concatenates a translation and a transformation */
   template<int Mode, int Options>
-  inline Transform<Scalar,Dim,Mode> operator* (const Transform<Scalar,Dim,Mode,Options>& t) const
+  EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode> operator* (const Transform<Scalar,Dim,Mode,Options>& t) const
   {
     Transform<Scalar,Dim,Mode> res = t;
     res.pretranslate(m_coeffs);
@@ -130,8 +130,10 @@ public:
   }
 
   /** Applies translation to vector */
-  inline VectorType operator* (const VectorType& other) const
-  { return m_coeffs + other; }
+  template<typename Derived>
+  inline typename internal::enable_if<Derived::IsVectorAtCompileTime,VectorType>::type
+  operator* (const MatrixBase<Derived>& vec) const
+  { return m_coeffs + vec.derived(); }
 
   /** \returns the inverse translation (opposite) */
   Translation inverse() const { return Translation(-m_coeffs); }
@@ -150,19 +152,19 @@ public:
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Translation,Translation<NewScalarType,Dim> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Translation,Translation<NewScalarType,Dim> >::type cast() const
   { return typename internal::cast_return_type<Translation,Translation<NewScalarType,Dim> >::type(*this); }
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit Translation(const Translation<OtherScalarType,Dim>& other)
+  EIGEN_DEVICE_FUNC inline explicit Translation(const Translation<OtherScalarType,Dim>& other)
   { m_coeffs = other.vector().template cast<Scalar>(); }
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const Translation& other, typename NumTraits<Scalar>::Real prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const Translation& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_coeffs.isApprox(other.m_coeffs, prec); }
 
 };
@@ -176,7 +178,7 @@ typedef Translation<double,3> Translation3d;
 //@}
 
 template<typename Scalar, int Dim>
-inline typename Translation<Scalar,Dim>::AffineTransformType
+EIGEN_DEVICE_FUNC inline typename Translation<Scalar,Dim>::AffineTransformType
 Translation<Scalar,Dim>::operator* (const UniformScaling<Scalar>& other) const
 {
   AffineTransformType res;
@@ -189,7 +191,7 @@ Translation<Scalar,Dim>::operator* (const UniformScaling<Scalar>& other) const
 
 template<typename Scalar, int Dim>
 template<typename OtherDerived>
-inline typename Translation<Scalar,Dim>::AffineTransformType
+EIGEN_DEVICE_FUNC inline typename Translation<Scalar,Dim>::AffineTransformType
 Translation<Scalar,Dim>::operator* (const EigenBase<OtherDerived>& linear) const
 {
   AffineTransformType res;
diff --git a/extern/Eigen3/Eigen/src/Geometry/Umeyama.h b/extern/Eigen3/Eigen/src/Geometry/Umeyama.h
index 5e20662f803..7e933fca13e 100644
--- a/extern/Eigen3/Eigen/src/Geometry/Umeyama.h
+++ b/extern/Eigen3/Eigen/src/Geometry/Umeyama.h
@@ -97,7 +97,6 @@ umeyama(const MatrixBase<Derived>& src, const MatrixBase<OtherDerived>& dst, boo
   typedef typename internal::umeyama_transform_matrix_type<Derived, OtherDerived>::type TransformationMatrixType;
   typedef typename internal::traits<TransformationMatrixType>::Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef typename Derived::Index Index;
 
   EIGEN_STATIC_ASSERT(!NumTraits<Scalar>::IsComplex, NUMERIC_TYPE_MUST_BE_REAL)
   EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename internal::traits<OtherDerived>::Scalar>::value),
@@ -136,22 +135,12 @@ umeyama(const MatrixBase<Derived>& src, const MatrixBase<OtherDerived>& dst, boo
 
   // Eq. (39)
   VectorType S = VectorType::Ones(m);
-  if (sigma.determinant()<Scalar(0)) S(m-1) = Scalar(-1);
+
+  if  ( svd.matrixU().determinant() * svd.matrixV().determinant() < 0 )
+    S(m-1) = -1;
 
   // Eq. (40) and (43)
-  const VectorType& d = svd.singularValues();
-  Index rank = 0; for (Index i=0; i<m; ++i) if (!internal::isMuchSmallerThan(d.coeff(i),d.coeff(0))) ++rank;
-  if (rank == m-1) {
-    if ( svd.matrixU().determinant() * svd.matrixV().determinant() > Scalar(0) ) {
-      Rt.block(0,0,m,m).noalias() = svd.matrixU()*svd.matrixV().transpose();
-    } else {
-      const Scalar s = S(m-1); S(m-1) = Scalar(-1);
-      Rt.block(0,0,m,m).noalias() = svd.matrixU() * S.asDiagonal() * svd.matrixV().transpose();
-      S(m-1) = s;
-    }
-  } else {
-    Rt.block(0,0,m,m).noalias() = svd.matrixU() * S.asDiagonal() * svd.matrixV().transpose();
-  }
+  Rt.block(0,0,m,m).noalias() = svd.matrixU() * S.asDiagonal() * svd.matrixV().transpose();
 
   if (with_scaling)
   {
diff --git a/extern/Eigen3/Eigen/src/Geometry/arch/Geometry_SSE.h b/extern/Eigen3/Eigen/src/Geometry/arch/Geometry_SSE.h
index 3d8284f2d0c..f68cab5834f 100644
--- a/extern/Eigen3/Eigen/src/Geometry/arch/Geometry_SSE.h
+++ b/extern/Eigen3/Eigen/src/Geometry/arch/Geometry_SSE.h
@@ -16,39 +16,63 @@ namespace Eigen {
 namespace internal {
 
 template<class Derived, class OtherDerived>
-struct quat_product<Architecture::SSE, Derived, OtherDerived, float, Aligned>
+struct quat_product<Architecture::SSE, Derived, OtherDerived, float>
 {
+  enum {
+    AAlignment = traits<Derived>::Alignment,
+    BAlignment = traits<OtherDerived>::Alignment,
+    ResAlignment = traits<Quaternion<float> >::Alignment
+  };
   static inline Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
   {
-    const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0,0,0,0x80000000));
     Quaternion<float> res;
-    __m128 a = _a.coeffs().template packet<Aligned>(0);
-    __m128 b = _b.coeffs().template packet<Aligned>(0);
-    __m128 flip1 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),
-                                         vec4f_swizzle1(b,2,0,1,2)),mask);
-    __m128 flip2 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),
-                                         vec4f_swizzle1(b,0,1,2,1)),mask);
-    pstore(&res.x(),
+    const __m128 mask = _mm_setr_ps(0.f,0.f,0.f,-0.f);
+    __m128 a = _a.coeffs().template packet<AAlignment>(0);
+    __m128 b = _b.coeffs().template packet<BAlignment>(0);
+    __m128 s1 = _mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),vec4f_swizzle1(b,2,0,1,2));
+    __m128 s2 = _mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),vec4f_swizzle1(b,0,1,2,1));
+    pstoret<float,Packet4f,ResAlignment>(
+              &res.x(),
               _mm_add_ps(_mm_sub_ps(_mm_mul_ps(a,vec4f_swizzle1(b,3,3,3,3)),
                                     _mm_mul_ps(vec4f_swizzle1(a,2,0,1,0),
                                                vec4f_swizzle1(b,1,2,0,0))),
-                         _mm_add_ps(flip1,flip2)));
+                         _mm_xor_ps(mask,_mm_add_ps(s1,s2))));
+    
     return res;
   }
 };
 
+template<class Derived>
+struct quat_conj<Architecture::SSE, Derived, float>
+{
+  enum {
+    ResAlignment = traits<Quaternion<float> >::Alignment
+  };
+  static inline Quaternion<float> run(const QuaternionBase<Derived>& q)
+  {
+    Quaternion<float> res;
+    const __m128 mask = _mm_setr_ps(-0.f,-0.f,-0.f,0.f);
+    pstoret<float,Packet4f,ResAlignment>(&res.x(), _mm_xor_ps(mask, q.coeffs().template packet<traits<Derived>::Alignment>(0)));
+    return res;
+  }
+};
+
+
 template<typename VectorLhs,typename VectorRhs>
 struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
 {
+  enum {
+    ResAlignment = traits<typename plain_matrix_type<VectorLhs>::type>::Alignment
+  };
   static inline typename plain_matrix_type<VectorLhs>::type
   run(const VectorLhs& lhs, const VectorRhs& rhs)
   {
-    __m128 a = lhs.template packet<VectorLhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
-    __m128 b = rhs.template packet<VectorRhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
+    __m128 a = lhs.template packet<traits<VectorLhs>::Alignment>(0);
+    __m128 b = rhs.template packet<traits<VectorRhs>::Alignment>(0);
     __m128 mul1=_mm_mul_ps(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3));
     __m128 mul2=_mm_mul_ps(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3));
     typename plain_matrix_type<VectorLhs>::type res;
-    pstore(&res.x(),_mm_sub_ps(mul1,mul2));
+    pstoret<float,Packet4f,ResAlignment>(&res.x(),_mm_sub_ps(mul1,mul2));
     return res;
   }
 };
@@ -57,8 +81,13 @@ struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
 
 
 template<class Derived, class OtherDerived>
-struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Aligned>
+struct quat_product<Architecture::SSE, Derived, OtherDerived, double>
 {
+  enum {
+    BAlignment = traits<OtherDerived>::Alignment,
+    ResAlignment = traits<Quaternion<double> >::Alignment
+  };
+
   static inline Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
   {
   const Packet2d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
@@ -66,8 +95,8 @@ struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Aligned>
   Quaternion<double> res;
 
   const double* a = _a.coeffs().data();
-  Packet2d b_xy = _b.coeffs().template packet<Aligned>(0);
-  Packet2d b_zw = _b.coeffs().template packet<Aligned>(2);
+  Packet2d b_xy = _b.coeffs().template packet<BAlignment>(0);
+  Packet2d b_zw = _b.coeffs().template packet<BAlignment>(2);
   Packet2d a_xx = pset1<Packet2d>(a[0]);
   Packet2d a_yy = pset1<Packet2d>(a[1]);
   Packet2d a_zz = pset1<Packet2d>(a[2]);
@@ -85,9 +114,9 @@ struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Aligned>
   t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw));
 #ifdef EIGEN_VECTORIZE_SSE3
   EIGEN_UNUSED_VARIABLE(mask)
-  pstore(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
+  pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
 #else
-  pstore(&res.x(), padd(t1, pxor(mask,preverse(t2))));
+  pstoret<double,Packet2d,ResAlignment>(&res.x(), padd(t1, pxor(mask,preverse(t2))));
 #endif
   
   /*
@@ -99,15 +128,32 @@ struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Aligned>
   t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy));
 #ifdef EIGEN_VECTORIZE_SSE3
   EIGEN_UNUSED_VARIABLE(mask)
-  pstore(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
+  pstoret<double,Packet2d,ResAlignment>(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
 #else
-  pstore(&res.z(), psub(t1, pxor(mask,preverse(t2))));
+  pstoret<double,Packet2d,ResAlignment>(&res.z(), psub(t1, pxor(mask,preverse(t2))));
 #endif
 
   return res;
 }
 };
 
+template<class Derived>
+struct quat_conj<Architecture::SSE, Derived, double>
+{
+  enum {
+    ResAlignment = traits<Quaternion<double> >::Alignment
+  };
+  static inline Quaternion<double> run(const QuaternionBase<Derived>& q)
+  {
+    Quaternion<double> res;
+    const __m128d mask0 = _mm_setr_pd(-0.,-0.);
+    const __m128d mask2 = _mm_setr_pd(-0.,0.);
+    pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_xor_pd(mask0, q.coeffs().template packet<traits<Derived>::Alignment>(0)));
+    pstoret<double,Packet2d,ResAlignment>(&res.z(), _mm_xor_pd(mask2, q.coeffs().template packet<traits<Derived>::Alignment>(2)));
+    return res;
+  }
+};
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/Householder/BlockHouseholder.h b/extern/Eigen3/Eigen/src/Householder/BlockHouseholder.h
index 60dbea5f56a..01a7ed1884d 100644
--- a/extern/Eigen3/Eigen/src/Householder/BlockHouseholder.h
+++ b/extern/Eigen3/Eigen/src/Householder/BlockHouseholder.h
@@ -16,48 +16,83 @@
 namespace Eigen { 
 
 namespace internal {
+  
+/** \internal */
+// template<typename TriangularFactorType,typename VectorsType,typename CoeffsType>
+// void make_block_householder_triangular_factor(TriangularFactorType& triFactor, const VectorsType& vectors, const CoeffsType& hCoeffs)
+// {
+//   typedef typename VectorsType::Scalar Scalar;
+//   const Index nbVecs = vectors.cols();
+//   eigen_assert(triFactor.rows() == nbVecs && triFactor.cols() == nbVecs && vectors.rows()>=nbVecs);
+// 
+//   for(Index i = 0; i < nbVecs; i++)
+//   {
+//     Index rs = vectors.rows() - i;
+//     // Warning, note that hCoeffs may alias with vectors.
+//     // It is then necessary to copy it before modifying vectors(i,i). 
+//     typename CoeffsType::Scalar h = hCoeffs(i);
+//     // This hack permits to pass trough nested Block<> and Transpose<> expressions.
+//     Scalar *Vii_ptr = const_cast<Scalar*>(vectors.data() + vectors.outerStride()*i + vectors.innerStride()*i);
+//     Scalar Vii = *Vii_ptr;
+//     *Vii_ptr = Scalar(1);
+//     triFactor.col(i).head(i).noalias() = -h * vectors.block(i, 0, rs, i).adjoint()
+//                                        * vectors.col(i).tail(rs);
+//     *Vii_ptr = Vii;
+//     // FIXME add .noalias() once the triangular product can work inplace
+//     triFactor.col(i).head(i) = triFactor.block(0,0,i,i).template triangularView<Upper>()
+//                              * triFactor.col(i).head(i);
+//     triFactor(i,i) = hCoeffs(i);
+//   }
+// }
 
 /** \internal */
+// This variant avoid modifications in vectors
 template<typename TriangularFactorType,typename VectorsType,typename CoeffsType>
 void make_block_householder_triangular_factor(TriangularFactorType& triFactor, const VectorsType& vectors, const CoeffsType& hCoeffs)
 {
-  typedef typename TriangularFactorType::Index Index;
-  typedef typename VectorsType::Scalar Scalar;
   const Index nbVecs = vectors.cols();
   eigen_assert(triFactor.rows() == nbVecs && triFactor.cols() == nbVecs && vectors.rows()>=nbVecs);
 
-  for(Index i = 0; i < nbVecs; i++)
+  for(Index i = nbVecs-1; i >=0 ; --i)
   {
-    Index rs = vectors.rows() - i;
-    Scalar Vii = vectors(i,i);
-    vectors.const_cast_derived().coeffRef(i,i) = Scalar(1);
-    triFactor.col(i).head(i).noalias() = -hCoeffs(i) * vectors.block(i, 0, rs, i).adjoint()
-                                       * vectors.col(i).tail(rs);
-    vectors.const_cast_derived().coeffRef(i, i) = Vii;
-    // FIXME add .noalias() once the triangular product can work inplace
-    triFactor.col(i).head(i) = triFactor.block(0,0,i,i).template triangularView<Upper>()
-                             * triFactor.col(i).head(i);
+    Index rs = vectors.rows() - i - 1;
+    Index rt = nbVecs-i-1;
+
+    if(rt>0)
+    {
+      triFactor.row(i).tail(rt).noalias() = -hCoeffs(i) * vectors.col(i).tail(rs).adjoint()
+                                                        * vectors.bottomRightCorner(rs, rt).template triangularView<UnitLower>();
+            
+      // FIXME add .noalias() once the triangular product can work inplace
+      triFactor.row(i).tail(rt) = triFactor.row(i).tail(rt) * triFactor.bottomRightCorner(rt,rt).template triangularView<Upper>();
+      
+    }
     triFactor(i,i) = hCoeffs(i);
   }
 }
 
-/** \internal */
+/** \internal
+  * if forward then perform   mat = H0 * H1 * H2 * mat
+  * otherwise perform         mat = H2 * H1 * H0 * mat
+  */
 template<typename MatrixType,typename VectorsType,typename CoeffsType>
-void apply_block_householder_on_the_left(MatrixType& mat, const VectorsType& vectors, const CoeffsType& hCoeffs)
+void apply_block_householder_on_the_left(MatrixType& mat, const VectorsType& vectors, const CoeffsType& hCoeffs, bool forward)
 {
-  typedef typename MatrixType::Index Index;
   enum { TFactorSize = MatrixType::ColsAtCompileTime };
   Index nbVecs = vectors.cols();
-  Matrix<typename MatrixType::Scalar, TFactorSize, TFactorSize, ColMajor> T(nbVecs,nbVecs);
-  make_block_householder_triangular_factor(T, vectors, hCoeffs);
-
-  const TriangularView<const VectorsType, UnitLower>& V(vectors);
+  Matrix<typename MatrixType::Scalar, TFactorSize, TFactorSize, RowMajor> T(nbVecs,nbVecs);
+  
+  if(forward) make_block_householder_triangular_factor(T, vectors, hCoeffs);
+  else        make_block_householder_triangular_factor(T, vectors, hCoeffs.conjugate());  
+  const TriangularView<const VectorsType, UnitLower> V(vectors);
 
   // A -= V T V^* A
-  Matrix<typename MatrixType::Scalar,VectorsType::ColsAtCompileTime,MatrixType::ColsAtCompileTime,0,
+  Matrix<typename MatrixType::Scalar,VectorsType::ColsAtCompileTime,MatrixType::ColsAtCompileTime,
+         (VectorsType::MaxColsAtCompileTime==1 && MatrixType::MaxColsAtCompileTime!=1)?RowMajor:ColMajor,
          VectorsType::MaxColsAtCompileTime,MatrixType::MaxColsAtCompileTime> tmp = V.adjoint() * mat;
   // FIXME add .noalias() once the triangular product can work inplace
-  tmp = T.template triangularView<Upper>().adjoint() * tmp;
+  if(forward) tmp = T.template triangularView<Upper>()           * tmp;
+  else        tmp = T.template triangularView<Upper>().adjoint() * tmp;
   mat.noalias() -= V * tmp;
 }
 
diff --git a/extern/Eigen3/Eigen/src/Householder/Householder.h b/extern/Eigen3/Eigen/src/Householder/Householder.h
index 32112af9bfd..80de2c3052c 100644
--- a/extern/Eigen3/Eigen/src/Householder/Householder.h
+++ b/extern/Eigen3/Eigen/src/Householder/Householder.h
@@ -75,8 +75,9 @@ void MatrixBase<Derived>::makeHouseholder(
   
   RealScalar tailSqNorm = size()==1 ? RealScalar(0) : tail.squaredNorm();
   Scalar c0 = coeff(0);
+  const RealScalar tol = (std::numeric_limits<RealScalar>::min)();
 
-  if(tailSqNorm == RealScalar(0) && numext::imag(c0)==RealScalar(0))
+  if(tailSqNorm <= tol && numext::abs2(numext::imag(c0))<=tol)
   {
     tau = RealScalar(0);
     beta = numext::real(c0);
@@ -118,7 +119,7 @@ void MatrixBase<Derived>::applyHouseholderOnTheLeft(
   {
     *this *= Scalar(1)-tau;
   }
-  else
+  else if(tau!=Scalar(0))
   {
     Map<typename internal::plain_row_type<PlainObject>::type> tmp(workspace,cols());
     Block<Derived, EssentialPart::SizeAtCompileTime, Derived::ColsAtCompileTime> bottom(derived(), 1, 0, rows()-1, cols());
@@ -155,7 +156,7 @@ void MatrixBase<Derived>::applyHouseholderOnTheRight(
   {
     *this *= Scalar(1)-tau;
   }
-  else
+  else if(tau!=Scalar(0))
   {
     Map<typename internal::plain_col_type<PlainObject>::type> tmp(workspace,rows());
     Block<Derived, Derived::RowsAtCompileTime, EssentialPart::SizeAtCompileTime> right(derived(), 0, 1, rows(), cols()-1);
diff --git a/extern/Eigen3/Eigen/src/Householder/HouseholderSequence.h b/extern/Eigen3/Eigen/src/Householder/HouseholderSequence.h
index d800ca1fa42..3ce0a693d9b 100644
--- a/extern/Eigen3/Eigen/src/Householder/HouseholderSequence.h
+++ b/extern/Eigen3/Eigen/src/Householder/HouseholderSequence.h
@@ -60,7 +60,7 @@ template<typename VectorsType, typename CoeffsType, int Side>
 struct traits<HouseholderSequence<VectorsType,CoeffsType,Side> >
 {
   typedef typename VectorsType::Scalar Scalar;
-  typedef typename VectorsType::Index Index;
+  typedef typename VectorsType::StorageIndex StorageIndex;
   typedef typename VectorsType::StorageKind StorageKind;
   enum {
     RowsAtCompileTime = Side==OnTheLeft ? traits<VectorsType>::RowsAtCompileTime
@@ -73,12 +73,20 @@ struct traits<HouseholderSequence<VectorsType,CoeffsType,Side> >
   };
 };
 
+struct HouseholderSequenceShape {};
+
+template<typename VectorsType, typename CoeffsType, int Side>
+struct evaluator_traits<HouseholderSequence<VectorsType,CoeffsType,Side> >
+  : public evaluator_traits_base<HouseholderSequence<VectorsType,CoeffsType,Side> >
+{
+  typedef HouseholderSequenceShape Shape;
+};
+
 template<typename VectorsType, typename CoeffsType, int Side>
 struct hseq_side_dependent_impl
 {
   typedef Block<const VectorsType, Dynamic, 1> EssentialVectorType;
   typedef HouseholderSequence<VectorsType, CoeffsType, OnTheLeft> HouseholderSequenceType;
-  typedef typename VectorsType::Index Index;
   static inline const EssentialVectorType essentialVector(const HouseholderSequenceType& h, Index k)
   {
     Index start = k+1+h.m_shift;
@@ -91,7 +99,6 @@ struct hseq_side_dependent_impl<VectorsType, CoeffsType, OnTheRight>
 {
   typedef Transpose<Block<const VectorsType, 1, Dynamic> > EssentialVectorType;
   typedef HouseholderSequence<VectorsType, CoeffsType, OnTheRight> HouseholderSequenceType;
-  typedef typename VectorsType::Index Index;
   static inline const EssentialVectorType essentialVector(const HouseholderSequenceType& h, Index k)
   {
     Index start = k+1+h.m_shift;
@@ -101,7 +108,7 @@ struct hseq_side_dependent_impl<VectorsType, CoeffsType, OnTheRight>
 
 template<typename OtherScalarType, typename MatrixType> struct matrix_type_times_scalar_type
 {
-  typedef typename scalar_product_traits<OtherScalarType, typename MatrixType::Scalar>::ReturnType
+  typedef typename ScalarBinaryOpTraits<OtherScalarType, typename MatrixType::Scalar>::ReturnType
     ResultScalar;
   typedef Matrix<ResultScalar, MatrixType::RowsAtCompileTime, MatrixType::ColsAtCompileTime,
                  0, MatrixType::MaxRowsAtCompileTime, MatrixType::MaxColsAtCompileTime> Type;
@@ -122,7 +129,6 @@ template<typename VectorsType, typename CoeffsType, int Side> class HouseholderS
       MaxColsAtCompileTime = internal::traits<HouseholderSequence>::MaxColsAtCompileTime
     };
     typedef typename internal::traits<HouseholderSequence>::Scalar Scalar;
-    typedef typename VectorsType::Index Index;
 
     typedef HouseholderSequence<
       typename internal::conditional<NumTraits<Scalar>::IsComplex,
@@ -237,8 +243,7 @@ template<typename VectorsType, typename CoeffsType, int Side> class HouseholderS
     {
       workspace.resize(rows());
       Index vecs = m_length;
-      if(    internal::is_same<typename internal::remove_all<VectorsType>::type,Dest>::value
-          && internal::extract_data(dst) == internal::extract_data(m_vectors))
+      if(internal::is_same_dense(dst,m_vectors))
       {
         // in-place
         dst.diagonal().setOnes();
@@ -299,7 +304,7 @@ template<typename VectorsType, typename CoeffsType, int Side> class HouseholderS
     /** \internal */
     template<typename Dest> inline void applyThisOnTheLeft(Dest& dst) const
     {
-      Matrix<Scalar,1,Dest::ColsAtCompileTime,RowMajor,1,Dest::MaxColsAtCompileTime> workspace(dst.cols());
+      Matrix<Scalar,1,Dest::ColsAtCompileTime,RowMajor,1,Dest::MaxColsAtCompileTime> workspace;
       applyThisOnTheLeft(dst, workspace);
     }
 
@@ -307,12 +312,36 @@ template<typename VectorsType, typename CoeffsType, int Side> class HouseholderS
     template<typename Dest, typename Workspace>
     inline void applyThisOnTheLeft(Dest& dst, Workspace& workspace) const
     {
-      workspace.resize(dst.cols());
-      for(Index k = 0; k < m_length; ++k)
+      const Index BlockSize = 48;
+      // if the entries are large enough, then apply the reflectors by block
+      if(m_length>=BlockSize && dst.cols()>1)
       {
-        Index actual_k = m_trans ? k : m_length-k-1;
-        dst.bottomRows(rows()-m_shift-actual_k)
-           .applyHouseholderOnTheLeft(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
+        for(Index i = 0; i < m_length; i+=BlockSize)
+        {
+          Index end = m_trans ? (std::min)(m_length,i+BlockSize) : m_length-i;
+          Index k = m_trans ? i : (std::max)(Index(0),end-BlockSize);
+          Index bs = end-k;
+          Index start = k + m_shift;
+          
+          typedef Block<typename internal::remove_all<VectorsType>::type,Dynamic,Dynamic> SubVectorsType;
+          SubVectorsType sub_vecs1(m_vectors.const_cast_derived(), Side==OnTheRight ? k : start,
+                                                                   Side==OnTheRight ? start : k,
+                                                                   Side==OnTheRight ? bs : m_vectors.rows()-start,
+                                                                   Side==OnTheRight ? m_vectors.cols()-start : bs);
+          typename internal::conditional<Side==OnTheRight, Transpose<SubVectorsType>, SubVectorsType&>::type sub_vecs(sub_vecs1);
+          Block<Dest,Dynamic,Dynamic> sub_dst(dst,dst.rows()-rows()+m_shift+k,0, rows()-m_shift-k,dst.cols());
+          apply_block_householder_on_the_left(sub_dst, sub_vecs, m_coeffs.segment(k, bs), !m_trans);
+        }
+      }
+      else
+      {
+        workspace.resize(dst.cols());
+        for(Index k = 0; k < m_length; ++k)
+        {
+          Index actual_k = m_trans ? k : m_length-k-1;
+          dst.bottomRows(rows()-m_shift-actual_k)
+            .applyHouseholderOnTheLeft(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
+        }
       }
     }
 
diff --git a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h
index 1f3c060d028..f66c846ef79 100644
--- a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h
+++ b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -17,33 +17,37 @@ namespace Eigen {
   *
   * This class allows to approximately solve for A.x = b problems assuming A is a diagonal matrix.
   * In other words, this preconditioner neglects all off diagonal entries and, in Eigen's language, solves for:
-  * \code
-  * A.diagonal().asDiagonal() . x = b
-  * \endcode
+    \code
+    A.diagonal().asDiagonal() . x = b
+    \endcode
   *
   * \tparam _Scalar the type of the scalar.
   *
+  * \implsparsesolverconcept
+  *
   * This preconditioner is suitable for both selfadjoint and general problems.
   * The diagonal entries are pre-inverted and stored into a dense vector.
   *
   * \note A variant that has yet to be implemented would attempt to preserve the norm of each column.
   *
+  * \sa class LeastSquareDiagonalPreconditioner, class ConjugateGradient
   */
 template <typename _Scalar>
 class DiagonalPreconditioner
 {
     typedef _Scalar Scalar;
     typedef Matrix<Scalar,Dynamic,1> Vector;
-    typedef typename Vector::Index Index;
-
   public:
-    // this typedef is only to export the scalar type and compile-time dimensions to solve_retval
-    typedef Matrix<Scalar,Dynamic,Dynamic> MatrixType;
+    typedef typename Vector::StorageIndex StorageIndex;
+    enum {
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
+    };
 
     DiagonalPreconditioner() : m_isInitialized(false) {}
 
     template<typename MatType>
-    DiagonalPreconditioner(const MatType& mat) : m_invdiag(mat.cols())
+    explicit DiagonalPreconditioner(const MatType& mat) : m_invdiag(mat.cols())
     {
       compute(mat);
     }
@@ -80,46 +84,117 @@ class DiagonalPreconditioner
       return factorize(mat);
     }
 
+    /** \internal */
     template<typename Rhs, typename Dest>
-    void _solve(const Rhs& b, Dest& x) const
+    void _solve_impl(const Rhs& b, Dest& x) const
     {
       x = m_invdiag.array() * b.array() ;
     }
 
-    template<typename Rhs> inline const internal::solve_retval<DiagonalPreconditioner, Rhs>
+    template<typename Rhs> inline const Solve<DiagonalPreconditioner, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "DiagonalPreconditioner is not initialized.");
       eigen_assert(m_invdiag.size()==b.rows()
                 && "DiagonalPreconditioner::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<DiagonalPreconditioner, Rhs>(*this, b.derived());
+      return Solve<DiagonalPreconditioner, Rhs>(*this, b.derived());
     }
+    
+    ComputationInfo info() { return Success; }
 
   protected:
     Vector m_invdiag;
     bool m_isInitialized;
 };
 
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<DiagonalPreconditioner<_MatrixType>, Rhs>
-  : solve_retval_base<DiagonalPreconditioner<_MatrixType>, Rhs>
+/** \ingroup IterativeLinearSolvers_Module
+  * \brief Jacobi preconditioner for LeastSquaresConjugateGradient
+  *
+  * This class allows to approximately solve for A' A x  = A' b problems assuming A' A is a diagonal matrix.
+  * In other words, this preconditioner neglects all off diagonal entries and, in Eigen's language, solves for:
+    \code
+    (A.adjoint() * A).diagonal().asDiagonal() * x = b
+    \endcode
+  *
+  * \tparam _Scalar the type of the scalar.
+  *
+  * \implsparsesolverconcept
+  *
+  * The diagonal entries are pre-inverted and stored into a dense vector.
+  * 
+  * \sa class LeastSquaresConjugateGradient, class DiagonalPreconditioner
+  */
+template <typename _Scalar>
+class LeastSquareDiagonalPreconditioner : public DiagonalPreconditioner<_Scalar>
 {
-  typedef DiagonalPreconditioner<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
+    typedef _Scalar Scalar;
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    typedef DiagonalPreconditioner<_Scalar> Base;
+    using Base::m_invdiag;
+  public:
 
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
+    LeastSquareDiagonalPreconditioner() : Base() {}
 
-}
+    template<typename MatType>
+    explicit LeastSquareDiagonalPreconditioner(const MatType& mat) : Base()
+    {
+      compute(mat);
+    }
+
+    template<typename MatType>
+    LeastSquareDiagonalPreconditioner& analyzePattern(const MatType& )
+    {
+      return *this;
+    }
+    
+    template<typename MatType>
+    LeastSquareDiagonalPreconditioner& factorize(const MatType& mat)
+    {
+      // Compute the inverse squared-norm of each column of mat
+      m_invdiag.resize(mat.cols());
+      if(MatType::IsRowMajor)
+      {
+        m_invdiag.setZero();
+        for(Index j=0; j<mat.outerSize(); ++j)
+        {
+          for(typename MatType::InnerIterator it(mat,j); it; ++it)
+            m_invdiag(it.index()) += numext::abs2(it.value());
+        }
+        for(Index j=0; j<mat.cols(); ++j)
+          if(numext::real(m_invdiag(j))>RealScalar(0))
+            m_invdiag(j) = RealScalar(1)/numext::real(m_invdiag(j));
+      }
+      else
+      {
+        for(Index j=0; j<mat.outerSize(); ++j)
+        {
+          RealScalar sum = mat.col(j).squaredNorm();
+          if(sum>RealScalar(0))
+            m_invdiag(j) = RealScalar(1)/sum;
+          else
+            m_invdiag(j) = RealScalar(1);
+        }
+      }
+      Base::m_isInitialized = true;
+      return *this;
+    }
+    
+    template<typename MatType>
+    LeastSquareDiagonalPreconditioner& compute(const MatType& mat)
+    {
+      return factorize(mat);
+    }
+    
+    ComputationInfo info() { return Success; }
+
+  protected:
+};
 
 /** \ingroup IterativeLinearSolvers_Module
   * \brief A naive preconditioner which approximates any matrix as the identity matrix
   *
+  * \implsparsesolverconcept
+  *
   * \sa class DiagonalPreconditioner
   */
 class IdentityPreconditioner
@@ -129,7 +204,7 @@ class IdentityPreconditioner
     IdentityPreconditioner() {}
 
     template<typename MatrixType>
-    IdentityPreconditioner(const MatrixType& ) {}
+    explicit IdentityPreconditioner(const MatrixType& ) {}
     
     template<typename MatrixType>
     IdentityPreconditioner& analyzePattern(const MatrixType& ) { return *this; }
@@ -142,6 +217,8 @@ class IdentityPreconditioner
     
     template<typename Rhs>
     inline const Rhs& solve(const Rhs& b) const { return b; }
+    
+    ComputationInfo info() { return Success; }
 };
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/BiCGSTAB.h b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/BiCGSTAB.h
index 5512219076b..454f4681497 100644
--- a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/BiCGSTAB.h
+++ b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/BiCGSTAB.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -27,7 +27,7 @@ namespace internal {
   */
 template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
 bool bicgstab(const MatrixType& mat, const Rhs& rhs, Dest& x,
-              const Preconditioner& precond, int& iters,
+              const Preconditioner& precond, Index& iters,
               typename Dest::RealScalar& tol_error)
 {
   using std::sqrt;
@@ -36,9 +36,9 @@ bool bicgstab(const MatrixType& mat, const Rhs& rhs, Dest& x,
   typedef typename Dest::Scalar Scalar;
   typedef Matrix<Scalar,Dynamic,1> VectorType;
   RealScalar tol = tol_error;
-  int maxIters = iters;
+  Index maxIters = iters;
 
-  int n = mat.cols();
+  Index n = mat.cols();
   VectorType r  = rhs - mat * x;
   VectorType r0 = r;
   
@@ -59,20 +59,21 @@ bool bicgstab(const MatrixType& mat, const Rhs& rhs, Dest& x,
 
   VectorType s(n), t(n);
 
-  RealScalar tol2 = tol*tol;
+  RealScalar tol2 = tol*tol*rhs_sqnorm;
   RealScalar eps2 = NumTraits<Scalar>::epsilon()*NumTraits<Scalar>::epsilon();
-  int i = 0;
-  int restarts = 0;
+  Index i = 0;
+  Index restarts = 0;
 
-  while ( r.squaredNorm()/rhs_sqnorm > tol2 && i<maxIters )
+  while ( r.squaredNorm() > tol2 && i<maxIters )
   {
     Scalar rho_old = rho;
 
     rho = r0.dot(r);
     if (abs(rho) < eps2*r0_sqnorm)
     {
-      // The new residual vector became too orthogonal to the arbitrarily choosen direction r0
+      // The new residual vector became too orthogonal to the arbitrarily chosen direction r0
       // Let's restart with a new r0:
+      r  = rhs - mat * x;
       r0 = r;
       rho = r0_sqnorm = r.squaredNorm();
       if(restarts++ == 0)
@@ -131,35 +132,33 @@ struct traits<BiCGSTAB<_MatrixType,_Preconditioner> >
   * \tparam _MatrixType the type of the sparse matrix A, can be a dense or a sparse matrix.
   * \tparam _Preconditioner the type of the preconditioner. Default is DiagonalPreconditioner
   *
+  * \implsparsesolverconcept
+  *
   * The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
   * and setTolerance() methods. The defaults are the size of the problem for the maximal number of iterations
   * and NumTraits<Scalar>::epsilon() for the tolerance.
   * 
+  * The tolerance corresponds to the relative residual error: |Ax-b|/|b|
+  * 
+  * \b Performance: when using sparse matrices, best performance is achied for a row-major sparse matrix format.
+  * Moreover, in this case multi-threading can be exploited if the user code is compiled with OpenMP enabled.
+  * See \ref TopicMultiThreading for details.
+  * 
   * This class can be used as the direct solver classes. Here is a typical usage example:
-  * \code
-  * int n = 10000;
-  * VectorXd x(n), b(n);
-  * SparseMatrix<double> A(n,n);
-  * // fill A and b
-  * BiCGSTAB<SparseMatrix<double> > solver;
-  * solver.compute(A);
-  * x = solver.solve(b);
-  * std::cout << "#iterations:     " << solver.iterations() << std::endl;
-  * std::cout << "estimated error: " << solver.error()      << std::endl;
-  * // update b, and solve again
-  * x = solver.solve(b);
-  * \endcode
+  * \include BiCGSTAB_simple.cpp
   * 
   * By default the iterations start with x=0 as an initial guess of the solution.
   * One can control the start using the solveWithGuess() method.
   * 
+  * BiCGSTAB can also be used in a matrix-free context, see the following \link MatrixfreeSolverExample example \endlink.
+  *
   * \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
   */
 template< typename _MatrixType, typename _Preconditioner>
 class BiCGSTAB : public IterativeSolverBase<BiCGSTAB<_MatrixType,_Preconditioner> >
 {
   typedef IterativeSolverBase<BiCGSTAB> Base;
-  using Base::mp_matrix;
+  using Base::matrix;
   using Base::m_error;
   using Base::m_iterations;
   using Base::m_info;
@@ -167,7 +166,6 @@ class BiCGSTAB : public IterativeSolverBase<BiCGSTAB<_MatrixType,_Preconditioner
 public:
   typedef _MatrixType MatrixType;
   typedef typename MatrixType::Scalar Scalar;
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::RealScalar RealScalar;
   typedef _Preconditioner Preconditioner;
 
@@ -190,35 +188,19 @@ public:
   explicit BiCGSTAB(const EigenBase<MatrixDerived>& A) : Base(A.derived()) {}
 
   ~BiCGSTAB() {}
-  
-  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
-    * \a x0 as an initial solution.
-    *
-    * \sa compute()
-    */
-  template<typename Rhs,typename Guess>
-  inline const internal::solve_retval_with_guess<BiCGSTAB, Rhs, Guess>
-  solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
-  {
-    eigen_assert(m_isInitialized && "BiCGSTAB is not initialized.");
-    eigen_assert(Base::rows()==b.rows()
-              && "BiCGSTAB::solve(): invalid number of rows of the right hand side matrix b");
-    return internal::solve_retval_with_guess
-            <BiCGSTAB, Rhs, Guess>(*this, b.derived(), x0);
-  }
-  
+
   /** \internal */
   template<typename Rhs,typename Dest>
-  void _solveWithGuess(const Rhs& b, Dest& x) const
+  void _solve_with_guess_impl(const Rhs& b, Dest& x) const
   {    
     bool failed = false;
-    for(int j=0; j<b.cols(); ++j)
+    for(Index j=0; j<b.cols(); ++j)
     {
       m_iterations = Base::maxIterations();
       m_error = Base::m_tolerance;
       
       typename Dest::ColXpr xj(x,j);
-      if(!internal::bicgstab(*mp_matrix, b.col(j), xj, Base::m_preconditioner, m_iterations, m_error))
+      if(!internal::bicgstab(matrix(), b.col(j), xj, Base::m_preconditioner, m_iterations, m_error))
         failed = true;
     }
     m_info = failed ? NumericalIssue
@@ -228,36 +210,19 @@ public:
   }
 
   /** \internal */
+  using Base::_solve_impl;
   template<typename Rhs,typename Dest>
-  void _solve(const Rhs& b, Dest& x) const
+  void _solve_impl(const MatrixBase<Rhs>& b, Dest& x) const
   {
-//     x.setZero();
-  x = b;
-    _solveWithGuess(b,x);
+    x.resize(this->rows(),b.cols());
+    x.setZero();
+    _solve_with_guess_impl(b,x);
   }
 
 protected:
 
 };
 
-
-namespace internal {
-
-  template<typename _MatrixType, typename _Preconditioner, typename Rhs>
-struct solve_retval<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
-  : solve_retval_base<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
-{
-  typedef BiCGSTAB<_MatrixType, _Preconditioner> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_BICGSTAB_H
diff --git a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/ConjugateGradient.h b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/ConjugateGradient.h
index 1a7e569c806..f7ce471349a 100644
--- a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/ConjugateGradient.h
+++ b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/ConjugateGradient.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -26,7 +26,7 @@ namespace internal {
 template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
 EIGEN_DONT_INLINE
 void conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
-                        const Preconditioner& precond, int& iters,
+                        const Preconditioner& precond, Index& iters,
                         typename Dest::RealScalar& tol_error)
 {
   using std::sqrt;
@@ -36,9 +36,9 @@ void conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
   typedef Matrix<Scalar,Dynamic,1> VectorType;
   
   RealScalar tol = tol_error;
-  int maxIters = iters;
+  Index maxIters = iters;
   
-  int n = mat.cols();
+  Index n = mat.cols();
 
   VectorType residual = rhs - mat * x; //initial residual
 
@@ -50,7 +50,8 @@ void conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
     tol_error = 0;
     return;
   }
-  RealScalar threshold = tol*tol*rhsNorm2;
+  const RealScalar considerAsZero = (std::numeric_limits<RealScalar>::min)();
+  RealScalar threshold = numext::maxi(tol*tol*rhsNorm2,considerAsZero);
   RealScalar residualNorm2 = residual.squaredNorm();
   if (residualNorm2 < threshold)
   {
@@ -58,31 +59,31 @@ void conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
     tol_error = sqrt(residualNorm2 / rhsNorm2);
     return;
   }
-  
+
   VectorType p(n);
-  p = precond.solve(residual);      //initial search direction
+  p = precond.solve(residual);      // initial search direction
 
   VectorType z(n), tmp(n);
   RealScalar absNew = numext::real(residual.dot(p));  // the square of the absolute value of r scaled by invM
-  int i = 0;
+  Index i = 0;
   while(i < maxIters)
   {
-    tmp.noalias() = mat * p;              // the bottleneck of the algorithm
+    tmp.noalias() = mat * p;                    // the bottleneck of the algorithm
 
-    Scalar alpha = absNew / p.dot(tmp);   // the amount we travel on dir
-    x += alpha * p;                       // update solution
-    residual -= alpha * tmp;              // update residue
+    Scalar alpha = absNew / p.dot(tmp);         // the amount we travel on dir
+    x += alpha * p;                             // update solution
+    residual -= alpha * tmp;                    // update residual
     
     residualNorm2 = residual.squaredNorm();
     if(residualNorm2 < threshold)
       break;
     
-    z = precond.solve(residual);          // approximately solve for "A z = residual"
+    z = precond.solve(residual);                // approximately solve for "A z = residual"
 
     RealScalar absOld = absNew;
     absNew = numext::real(residual.dot(z));     // update the absolute value of r
-    RealScalar beta = absNew / absOld;            // calculate the Gram-Schmidt value used to create the new search direction
-    p = z + beta * p;                             // update search direction
+    RealScalar beta = absNew / absOld;          // calculate the Gram-Schmidt value used to create the new search direction
+    p = z + beta * p;                           // update search direction
     i++;
   }
   tol_error = sqrt(residualNorm2 / rhsNorm2);
@@ -107,45 +108,57 @@ struct traits<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> >
 }
 
 /** \ingroup IterativeLinearSolvers_Module
-  * \brief A conjugate gradient solver for sparse self-adjoint problems
+  * \brief A conjugate gradient solver for sparse (or dense) self-adjoint problems
   *
-  * This class allows to solve for A.x = b sparse linear problems using a conjugate gradient algorithm.
-  * The sparse matrix A must be selfadjoint. The vectors x and b can be either dense or sparse.
+  * This class allows to solve for A.x = b linear problems using an iterative conjugate gradient algorithm.
+  * The matrix A must be selfadjoint. The matrix A and the vectors x and b can be either dense or sparse.
   *
   * \tparam _MatrixType the type of the matrix A, can be a dense or a sparse matrix.
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower,
-  *               Upper, or Lower|Upper in which the full matrix entries will be considered. Default is Lower.
+  *               \c Upper, or \c Lower|Upper in which the full matrix entries will be considered.
+  *               Default is \c Lower, best performance is \c Lower|Upper.
   * \tparam _Preconditioner the type of the preconditioner. Default is DiagonalPreconditioner
   *
+  * \implsparsesolverconcept
+  *
   * The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
   * and setTolerance() methods. The defaults are the size of the problem for the maximal number of iterations
   * and NumTraits<Scalar>::epsilon() for the tolerance.
   * 
+  * The tolerance corresponds to the relative residual error: |Ax-b|/|b|
+  * 
+  * \b Performance: Even though the default value of \c _UpLo is \c Lower, significantly higher performance is
+  * achieved when using a complete matrix and \b Lower|Upper as the \a _UpLo template parameter. Moreover, in this
+  * case multi-threading can be exploited if the user code is compiled with OpenMP enabled.
+  * See \ref TopicMultiThreading for details.
+  * 
   * This class can be used as the direct solver classes. Here is a typical usage example:
-  * \code
-  * int n = 10000;
-  * VectorXd x(n), b(n);
-  * SparseMatrix<double> A(n,n);
-  * // fill A and b
-  * ConjugateGradient<SparseMatrix<double> > cg;
-  * cg.compute(A);
-  * x = cg.solve(b);
-  * std::cout << "#iterations:     " << cg.iterations() << std::endl;
-  * std::cout << "estimated error: " << cg.error()      << std::endl;
-  * // update b, and solve again
-  * x = cg.solve(b);
-  * \endcode
+    \code
+    int n = 10000;
+    VectorXd x(n), b(n);
+    SparseMatrix<double> A(n,n);
+    // fill A and b
+    ConjugateGradient<SparseMatrix<double>, Lower|Upper> cg;
+    cg.compute(A);
+    x = cg.solve(b);
+    std::cout << "#iterations:     " << cg.iterations() << std::endl;
+    std::cout << "estimated error: " << cg.error()      << std::endl;
+    // update b, and solve again
+    x = cg.solve(b);
+    \endcode
   * 
   * By default the iterations start with x=0 as an initial guess of the solution.
   * One can control the start using the solveWithGuess() method.
   * 
-  * \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
+  * ConjugateGradient can also be used in a matrix-free context, see the following \link MatrixfreeSolverExample example \endlink.
+  *
+  * \sa class LeastSquaresConjugateGradient, class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
   */
 template< typename _MatrixType, int _UpLo, typename _Preconditioner>
 class ConjugateGradient : public IterativeSolverBase<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> >
 {
   typedef IterativeSolverBase<ConjugateGradient> Base;
-  using Base::mp_matrix;
+  using Base::matrix;
   using Base::m_error;
   using Base::m_iterations;
   using Base::m_info;
@@ -153,7 +166,6 @@ class ConjugateGradient : public IterativeSolverBase<ConjugateGradient<_MatrixTy
 public:
   typedef _MatrixType MatrixType;
   typedef typename MatrixType::Scalar Scalar;
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::RealScalar RealScalar;
   typedef _Preconditioner Preconditioner;
 
@@ -180,41 +192,36 @@ public:
   explicit ConjugateGradient(const EigenBase<MatrixDerived>& A) : Base(A.derived()) {}
 
   ~ConjugateGradient() {}
-  
-  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
-    * \a x0 as an initial solution.
-    *
-    * \sa compute()
-    */
-  template<typename Rhs,typename Guess>
-  inline const internal::solve_retval_with_guess<ConjugateGradient, Rhs, Guess>
-  solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
-  {
-    eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
-    eigen_assert(Base::rows()==b.rows()
-              && "ConjugateGradient::solve(): invalid number of rows of the right hand side matrix b");
-    return internal::solve_retval_with_guess
-            <ConjugateGradient, Rhs, Guess>(*this, b.derived(), x0);
-  }
 
   /** \internal */
   template<typename Rhs,typename Dest>
-  void _solveWithGuess(const Rhs& b, Dest& x) const
+  void _solve_with_guess_impl(const Rhs& b, Dest& x) const
   {
+    typedef typename Base::MatrixWrapper MatrixWrapper;
+    typedef typename Base::ActualMatrixType ActualMatrixType;
+    enum {
+      TransposeInput  =   (!MatrixWrapper::MatrixFree)
+                      &&  (UpLo==(Lower|Upper))
+                      &&  (!MatrixType::IsRowMajor)
+                      &&  (!NumTraits<Scalar>::IsComplex)
+    };
+    typedef typename internal::conditional<TransposeInput,Transpose<const ActualMatrixType>, ActualMatrixType const&>::type RowMajorWrapper;
+    EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(MatrixWrapper::MatrixFree,UpLo==(Lower|Upper)),MATRIX_FREE_CONJUGATE_GRADIENT_IS_COMPATIBLE_WITH_UPPER_UNION_LOWER_MODE_ONLY);
     typedef typename internal::conditional<UpLo==(Lower|Upper),
-                                           const MatrixType&,
-                                           SparseSelfAdjointView<const MatrixType, UpLo>
-                                          >::type MatrixWrapperType;
+                                           RowMajorWrapper,
+                                           typename MatrixWrapper::template ConstSelfAdjointViewReturnType<UpLo>::Type
+                                          >::type SelfAdjointWrapper;
     m_iterations = Base::maxIterations();
     m_error = Base::m_tolerance;
 
-    for(int j=0; j<b.cols(); ++j)
+    for(Index j=0; j<b.cols(); ++j)
     {
       m_iterations = Base::maxIterations();
       m_error = Base::m_tolerance;
 
       typename Dest::ColXpr xj(x,j);
-      internal::conjugate_gradient(MatrixWrapperType(*mp_matrix), b.col(j), xj, Base::m_preconditioner, m_iterations, m_error);
+      RowMajorWrapper row_mat(matrix());
+      internal::conjugate_gradient(SelfAdjointWrapper(row_mat), b.col(j), xj, Base::m_preconditioner, m_iterations, m_error);
     }
 
     m_isInitialized = true;
@@ -222,35 +229,18 @@ public:
   }
   
   /** \internal */
+  using Base::_solve_impl;
   template<typename Rhs,typename Dest>
-  void _solve(const Rhs& b, Dest& x) const
+  void _solve_impl(const MatrixBase<Rhs>& b, Dest& x) const
   {
     x.setZero();
-    _solveWithGuess(b,x);
+    _solve_with_guess_impl(b.derived(),x);
   }
 
 protected:
 
 };
 
-
-namespace internal {
-
-template<typename _MatrixType, int _UpLo, typename _Preconditioner, typename Rhs>
-struct solve_retval<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner>, Rhs>
-  : solve_retval_base<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner>, Rhs>
-{
-  typedef ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_CONJUGATE_GRADIENT_H
diff --git a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IncompleteCholesky.h b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IncompleteCholesky.h
new file mode 100644
index 00000000000..e45c272b4c4
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IncompleteCholesky.h
@@ -0,0 +1,400 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_INCOMPLETE_CHOlESKY_H
+#define EIGEN_INCOMPLETE_CHOlESKY_H
+
+#include <vector>
+#include <list>
+
+namespace Eigen {  
+/** 
+  * \brief Modified Incomplete Cholesky with dual threshold
+  *
+  * References : C-J. Lin and J. J. Moré, Incomplete Cholesky Factorizations with
+  *              Limited memory, SIAM J. Sci. Comput.  21(1), pp. 24-45, 1999
+  *
+  * \tparam Scalar the scalar type of the input matrices
+  * \tparam _UpLo The triangular part that will be used for the computations. It can be Lower
+    *               or Upper. Default is Lower.
+  * \tparam _OrderingType The ordering method to use, either AMDOrdering<> or NaturalOrdering<>. Default is AMDOrdering<int>,
+  *                       unless EIGEN_MPL2_ONLY is defined, in which case the default is NaturalOrdering<int>.
+  *
+  * \implsparsesolverconcept
+  *
+  * It performs the following incomplete factorization: \f$ S P A P' S \approx L L' \f$
+  * where L is a lower triangular factor, S is a diagonal scaling matrix, and P is a
+  * fill-in reducing permutation as computed by the ordering method.
+  *
+  * \b Shifting \b strategy: Let \f$ B = S P A P' S \f$  be the scaled matrix on which the factorization is carried out,
+  * and \f$ \beta \f$ be the minimum value of the diagonal. If \f$ \beta > 0 \f$ then, the factorization is directly performed
+  * on the matrix B. Otherwise, the factorization is performed on the shifted matrix \f$ B + (\sigma+|\beta| I \f$ where
+  * \f$ \sigma \f$ is the initial shift value as returned and set by setInitialShift() method. The default value is \f$ \sigma = 10^{-3} \f$.
+  * If the factorization fails, then the shift in doubled until it succeed or a maximum of ten attempts. If it still fails, as returned by
+  * the info() method, then you can either increase the initial shift, or better use another preconditioning technique.
+  *
+  */
+template <typename Scalar, int _UpLo = Lower, typename _OrderingType =
+#ifndef EIGEN_MPL2_ONLY
+AMDOrdering<int>
+#else
+NaturalOrdering<int>
+#endif
+>
+class IncompleteCholesky : public SparseSolverBase<IncompleteCholesky<Scalar,_UpLo,_OrderingType> >
+{
+  protected:
+    typedef SparseSolverBase<IncompleteCholesky<Scalar,_UpLo,_OrderingType> > Base;
+    using Base::m_isInitialized;
+  public:
+    typedef typename NumTraits<Scalar>::Real RealScalar; 
+    typedef _OrderingType OrderingType;
+    typedef typename OrderingType::PermutationType PermutationType;
+    typedef typename PermutationType::StorageIndex StorageIndex; 
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> FactorType;
+    typedef Matrix<Scalar,Dynamic,1> VectorSx;
+    typedef Matrix<RealScalar,Dynamic,1> VectorRx;
+    typedef Matrix<StorageIndex,Dynamic, 1> VectorIx;
+    typedef std::vector<std::list<StorageIndex> > VectorList; 
+    enum { UpLo = _UpLo };
+    enum {
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
+    };
+  public:
+
+    /** Default constructor leaving the object in a partly non-initialized stage.
+      *
+      * You must call compute() or the pair analyzePattern()/factorize() to make it valid.
+      *
+      * \sa IncompleteCholesky(const MatrixType&)
+      */
+    IncompleteCholesky() : m_initialShift(1e-3),m_factorizationIsOk(false) {}
+    
+    /** Constructor computing the incomplete factorization for the given matrix \a matrix.
+      */
+    template<typename MatrixType>
+    IncompleteCholesky(const MatrixType& matrix) : m_initialShift(1e-3),m_factorizationIsOk(false)
+    {
+      compute(matrix);
+    }
+    
+    /** \returns number of rows of the factored matrix */
+    Index rows() const { return m_L.rows(); }
+    
+    /** \returns number of columns of the factored matrix */
+    Index cols() const { return m_L.cols(); }
+    
+
+    /** \brief Reports whether previous computation was successful.
+      *
+      * It triggers an assertion if \c *this has not been initialized through the respective constructor,
+      * or a call to compute() or analyzePattern().
+      *
+      * \returns \c Success if computation was successful,
+      *          \c NumericalIssue if the matrix appears to be negative.
+      */
+    ComputationInfo info() const
+    {
+      eigen_assert(m_isInitialized && "IncompleteCholesky is not initialized.");
+      return m_info;
+    }
+    
+    /** \brief Set the initial shift parameter \f$ \sigma \f$.
+      */
+    void setInitialShift(RealScalar shift) { m_initialShift = shift; }
+    
+    /** \brief Computes the fill reducing permutation vector using the sparsity pattern of \a mat
+      */
+    template<typename MatrixType>
+    void analyzePattern(const MatrixType& mat)
+    {
+      OrderingType ord; 
+      PermutationType pinv;
+      ord(mat.template selfadjointView<UpLo>(), pinv); 
+      if(pinv.size()>0) m_perm = pinv.inverse();
+      else              m_perm.resize(0);
+      m_L.resize(mat.rows(), mat.cols());
+      m_analysisIsOk = true;
+      m_isInitialized = true;
+      m_info = Success;
+    }
+    
+    /** \brief Performs the numerical factorization of the input matrix \a mat
+      *
+      * The method analyzePattern() or compute() must have been called beforehand
+      * with a matrix having the same pattern.
+      *
+      * \sa compute(), analyzePattern()
+      */
+    template<typename MatrixType>
+    void factorize(const MatrixType& mat);
+    
+    /** Computes or re-computes the incomplete Cholesky factorization of the input matrix \a mat
+      *
+      * It is a shortcut for a sequential call to the analyzePattern() and factorize() methods.
+      *
+      * \sa analyzePattern(), factorize()
+      */
+    template<typename MatrixType>
+    void compute(const MatrixType& mat)
+    {
+      analyzePattern(mat);
+      factorize(mat);
+    }
+    
+    // internal
+    template<typename Rhs, typename Dest>
+    void _solve_impl(const Rhs& b, Dest& x) const
+    {
+      eigen_assert(m_factorizationIsOk && "factorize() should be called first");
+      if (m_perm.rows() == b.rows())  x = m_perm * b;
+      else                            x = b;
+      x = m_scale.asDiagonal() * x;
+      x = m_L.template triangularView<Lower>().solve(x);
+      x = m_L.adjoint().template triangularView<Upper>().solve(x);
+      x = m_scale.asDiagonal() * x;
+      if (m_perm.rows() == b.rows())
+        x = m_perm.inverse() * x;
+    }
+
+    /** \returns the sparse lower triangular factor L */
+    const FactorType& matrixL() const { eigen_assert("m_factorizationIsOk"); return m_L; }
+
+    /** \returns a vector representing the scaling factor S */
+    const VectorRx& scalingS() const { eigen_assert("m_factorizationIsOk"); return m_scale; }
+
+    /** \returns the fill-in reducing permutation P (can be empty for a natural ordering) */
+    const PermutationType& permutationP() const { eigen_assert("m_analysisIsOk"); return m_perm; }
+
+  protected:
+    FactorType m_L;              // The lower part stored in CSC
+    VectorRx m_scale;            // The vector for scaling the matrix 
+    RealScalar m_initialShift;   // The initial shift parameter
+    bool m_analysisIsOk; 
+    bool m_factorizationIsOk; 
+    ComputationInfo m_info;
+    PermutationType m_perm; 
+
+  private:
+    inline void updateList(Ref<const VectorIx> colPtr, Ref<VectorIx> rowIdx, Ref<VectorSx> vals, const Index& col, const Index& jk, VectorIx& firstElt, VectorList& listCol); 
+}; 
+
+// Based on the following paper:
+//   C-J. Lin and J. J. Moré, Incomplete Cholesky Factorizations with
+//   Limited memory, SIAM J. Sci. Comput.  21(1), pp. 24-45, 1999
+//   http://ftp.mcs.anl.gov/pub/tech_reports/reports/P682.pdf
+template<typename Scalar, int _UpLo, typename OrderingType>
+template<typename _MatrixType>
+void IncompleteCholesky<Scalar,_UpLo, OrderingType>::factorize(const _MatrixType& mat)
+{
+  using std::sqrt;
+  eigen_assert(m_analysisIsOk && "analyzePattern() should be called first"); 
+    
+  // Dropping strategy : Keep only the p largest elements per column, where p is the number of elements in the column of the original matrix. Other strategies will be added
+  
+  // Apply the fill-reducing permutation computed in analyzePattern()
+  if (m_perm.rows() == mat.rows() ) // To detect the null permutation
+  {
+    // The temporary is needed to make sure that the diagonal entry is properly sorted
+    FactorType tmp(mat.rows(), mat.cols());
+    tmp = mat.template selfadjointView<_UpLo>().twistedBy(m_perm);
+    m_L.template selfadjointView<Lower>() = tmp.template selfadjointView<Lower>();
+  }
+  else
+  {
+    m_L.template selfadjointView<Lower>() = mat.template selfadjointView<_UpLo>();
+  }
+  
+  Index n = m_L.cols(); 
+  Index nnz = m_L.nonZeros();
+  Map<VectorSx> vals(m_L.valuePtr(), nnz);         //values
+  Map<VectorIx> rowIdx(m_L.innerIndexPtr(), nnz);  //Row indices
+  Map<VectorIx> colPtr( m_L.outerIndexPtr(), n+1); // Pointer to the beginning of each row
+  VectorIx firstElt(n-1); // for each j, points to the next entry in vals that will be used in the factorization
+  VectorList listCol(n);  // listCol(j) is a linked list of columns to update column j
+  VectorSx col_vals(n);   // Store a  nonzero values in each column
+  VectorIx col_irow(n);   // Row indices of nonzero elements in each column
+  VectorIx col_pattern(n);
+  col_pattern.fill(-1);
+  StorageIndex col_nnz;
+  
+  
+  // Computes the scaling factors 
+  m_scale.resize(n);
+  m_scale.setZero();
+  for (Index j = 0; j < n; j++)
+    for (Index k = colPtr[j]; k < colPtr[j+1]; k++)
+    {
+      m_scale(j) += numext::abs2(vals(k));
+      if(rowIdx[k]!=j)
+        m_scale(rowIdx[k]) += numext::abs2(vals(k));
+    }
+  
+  m_scale = m_scale.cwiseSqrt().cwiseSqrt();
+
+  for (Index j = 0; j < n; ++j)
+    if(m_scale(j)>(std::numeric_limits<RealScalar>::min)())
+      m_scale(j) = RealScalar(1)/m_scale(j);
+    else
+      m_scale(j) = 1;
+
+  // TODO disable scaling if not needed, i.e., if it is roughly uniform? (this will make solve() faster)
+  
+  // Scale and compute the shift for the matrix 
+  RealScalar mindiag = NumTraits<RealScalar>::highest();
+  for (Index j = 0; j < n; j++)
+  {
+    for (Index k = colPtr[j]; k < colPtr[j+1]; k++)
+      vals[k] *= (m_scale(j)*m_scale(rowIdx[k]));
+    eigen_internal_assert(rowIdx[colPtr[j]]==j && "IncompleteCholesky: only the lower triangular part must be stored");
+    mindiag = numext::mini(numext::real(vals[colPtr[j]]), mindiag);
+  }
+
+  FactorType L_save = m_L;
+  
+  RealScalar shift = 0;
+  if(mindiag <= RealScalar(0.))
+    shift = m_initialShift - mindiag;
+
+  m_info = NumericalIssue;
+
+  // Try to perform the incomplete factorization using the current shift
+  int iter = 0;
+  do
+  {
+    // Apply the shift to the diagonal elements of the matrix
+    for (Index j = 0; j < n; j++)
+      vals[colPtr[j]] += shift;
+
+    // jki version of the Cholesky factorization
+    Index j=0;
+    for (; j < n; ++j)
+    {
+      // Left-looking factorization of the j-th column
+      // First, load the j-th column into col_vals
+      Scalar diag = vals[colPtr[j]];  // It is assumed that only the lower part is stored
+      col_nnz = 0;
+      for (Index i = colPtr[j] + 1; i < colPtr[j+1]; i++)
+      {
+        StorageIndex l = rowIdx[i];
+        col_vals(col_nnz) = vals[i];
+        col_irow(col_nnz) = l;
+        col_pattern(l) = col_nnz;
+        col_nnz++;
+      }
+      {
+        typename std::list<StorageIndex>::iterator k;
+        // Browse all previous columns that will update column j
+        for(k = listCol[j].begin(); k != listCol[j].end(); k++)
+        {
+          Index jk = firstElt(*k); // First element to use in the column
+          eigen_internal_assert(rowIdx[jk]==j);
+          Scalar v_j_jk = numext::conj(vals[jk]);
+
+          jk += 1;
+          for (Index i = jk; i < colPtr[*k+1]; i++)
+          {
+            StorageIndex l = rowIdx[i];
+            if(col_pattern[l]<0)
+            {
+              col_vals(col_nnz) = vals[i] * v_j_jk;
+              col_irow[col_nnz] = l;
+              col_pattern(l) = col_nnz;
+              col_nnz++;
+            }
+            else
+              col_vals(col_pattern[l]) -= vals[i] * v_j_jk;
+          }
+          updateList(colPtr,rowIdx,vals, *k, jk, firstElt, listCol);
+        }
+      }
+
+      // Scale the current column
+      if(numext::real(diag) <= 0)
+      {
+        if(++iter>=10)
+          return;
+
+        // increase shift
+        shift = numext::maxi(m_initialShift,RealScalar(2)*shift);
+        // restore m_L, col_pattern, and listCol
+        vals = Map<const VectorSx>(L_save.valuePtr(), nnz);
+        rowIdx = Map<const VectorIx>(L_save.innerIndexPtr(), nnz);
+        colPtr = Map<const VectorIx>(L_save.outerIndexPtr(), n+1);
+        col_pattern.fill(-1);
+        for(Index i=0; i<n; ++i)
+          listCol[i].clear();
+
+        break;
+      }
+
+      RealScalar rdiag = sqrt(numext::real(diag));
+      vals[colPtr[j]] = rdiag;
+      for (Index k = 0; k<col_nnz; ++k)
+      {
+        Index i = col_irow[k];
+        //Scale
+        col_vals(k) /= rdiag;
+        //Update the remaining diagonals with col_vals
+        vals[colPtr[i]] -= numext::abs2(col_vals(k));
+      }
+      // Select the largest p elements
+      // p is the original number of elements in the column (without the diagonal)
+      Index p = colPtr[j+1] - colPtr[j] - 1 ;
+      Ref<VectorSx> cvals = col_vals.head(col_nnz);
+      Ref<VectorIx> cirow = col_irow.head(col_nnz);
+      internal::QuickSplit(cvals,cirow, p);
+      // Insert the largest p elements in the matrix
+      Index cpt = 0;
+      for (Index i = colPtr[j]+1; i < colPtr[j+1]; i++)
+      {
+        vals[i] = col_vals(cpt);
+        rowIdx[i] = col_irow(cpt);
+        // restore col_pattern:
+        col_pattern(col_irow(cpt)) = -1;
+        cpt++;
+      }
+      // Get the first smallest row index and put it after the diagonal element
+      Index jk = colPtr(j)+1;
+      updateList(colPtr,rowIdx,vals,j,jk,firstElt,listCol);
+    }
+
+    if(j==n)
+    {
+      m_factorizationIsOk = true;
+      m_info = Success;
+    }
+  } while(m_info!=Success);
+}
+
+template<typename Scalar, int _UpLo, typename OrderingType>
+inline void IncompleteCholesky<Scalar,_UpLo, OrderingType>::updateList(Ref<const VectorIx> colPtr, Ref<VectorIx> rowIdx, Ref<VectorSx> vals, const Index& col, const Index& jk, VectorIx& firstElt, VectorList& listCol)
+{
+  if (jk < colPtr(col+1) )
+  {
+    Index p = colPtr(col+1) - jk;
+    Index minpos; 
+    rowIdx.segment(jk,p).minCoeff(&minpos);
+    minpos += jk;
+    if (rowIdx(minpos) != rowIdx(jk))
+    {
+      //Swap
+      std::swap(rowIdx(jk),rowIdx(minpos));
+      std::swap(vals(jk),vals(minpos));
+    }
+    firstElt(col) = internal::convert_index<StorageIndex,Index>(jk);
+    listCol[rowIdx(jk)].push_back(internal::convert_index<StorageIndex,Index>(col));
+  }
+}
+
+} // end namespace Eigen 
+
+#endif
diff --git a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IncompleteLUT.h b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IncompleteLUT.h
index d3f37fea2a1..338e6f10a87 100644
--- a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IncompleteLUT.h
+++ b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IncompleteLUT.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -24,7 +25,7 @@ namespace internal {
   * \param ind The array of index for the elements in @p row
   * \param ncut  The number of largest elements to keep
   **/ 
-template <typename VectorV, typename VectorI, typename Index>
+template <typename VectorV, typename VectorI>
 Index QuickSplit(VectorV &row, VectorI &ind, Index ncut)
 {
   typedef typename VectorV::RealScalar RealScalar;
@@ -66,6 +67,8 @@ Index QuickSplit(VectorV &row, VectorI &ind, Index ncut)
   * \class IncompleteLUT
   * \brief Incomplete LU factorization with dual-threshold strategy
   *
+  * \implsparsesolverconcept
+  *
   * During the numerical factorization, two dropping rules are used :
   *  1) any element whose magnitude is less than some tolerance is dropped.
   *    This tolerance is obtained by multiplying the input tolerance @p droptol 
@@ -92,28 +95,36 @@ Index QuickSplit(VectorV &row, VectorI &ind, Index ncut)
   * alternatively, on GMANE:
   *   http://comments.gmane.org/gmane.comp.lib.eigen/3302
   */
-template <typename _Scalar>
-class IncompleteLUT : internal::noncopyable
+template <typename _Scalar, typename _StorageIndex = int>
+class IncompleteLUT : public SparseSolverBase<IncompleteLUT<_Scalar, _StorageIndex> >
 {
+  protected:
+    typedef SparseSolverBase<IncompleteLUT> Base;
+    using Base::m_isInitialized;
+  public:
     typedef _Scalar Scalar;
+    typedef _StorageIndex StorageIndex;
     typedef typename NumTraits<Scalar>::Real RealScalar;
     typedef Matrix<Scalar,Dynamic,1> Vector;
-    typedef SparseMatrix<Scalar,RowMajor> FactorType;
-    typedef SparseMatrix<Scalar,ColMajor> PermutType;
-    typedef typename FactorType::Index Index;
+    typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+    typedef SparseMatrix<Scalar,RowMajor,StorageIndex> FactorType;
+
+    enum {
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
+    };
 
   public:
-    typedef Matrix<Scalar,Dynamic,Dynamic> MatrixType;
     
     IncompleteLUT()
       : m_droptol(NumTraits<Scalar>::dummy_precision()), m_fillfactor(10),
-        m_analysisIsOk(false), m_factorizationIsOk(false), m_isInitialized(false)
+        m_analysisIsOk(false), m_factorizationIsOk(false)
     {}
     
     template<typename MatrixType>
-    IncompleteLUT(const MatrixType& mat, const RealScalar& droptol=NumTraits<Scalar>::dummy_precision(), int fillfactor = 10)
+    explicit IncompleteLUT(const MatrixType& mat, const RealScalar& droptol=NumTraits<Scalar>::dummy_precision(), int fillfactor = 10)
       : m_droptol(droptol),m_fillfactor(fillfactor),
-        m_analysisIsOk(false),m_factorizationIsOk(false),m_isInitialized(false)
+        m_analysisIsOk(false),m_factorizationIsOk(false)
     {
       eigen_assert(fillfactor != 0);
       compute(mat); 
@@ -146,7 +157,7 @@ class IncompleteLUT : internal::noncopyable
       * 
       **/
     template<typename MatrixType>
-    IncompleteLUT<Scalar>& compute(const MatrixType& amat)
+    IncompleteLUT& compute(const MatrixType& amat)
     {
       analyzePattern(amat); 
       factorize(amat);
@@ -157,7 +168,7 @@ class IncompleteLUT : internal::noncopyable
     void setFillfactor(int fillfactor); 
     
     template<typename Rhs, typename Dest>
-    void _solve(const Rhs& b, Dest& x) const
+    void _solve_impl(const Rhs& b, Dest& x) const
     {
       x = m_Pinv * b;
       x = m_lu.template triangularView<UnitLower>().solve(x);
@@ -165,15 +176,6 @@ class IncompleteLUT : internal::noncopyable
       x = m_P * x; 
     }
 
-    template<typename Rhs> inline const internal::solve_retval<IncompleteLUT, Rhs>
-     solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "IncompleteLUT is not initialized.");
-      eigen_assert(cols()==b.rows()
-                && "IncompleteLUT::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<IncompleteLUT, Rhs>(*this, b.derived());
-    }
-
 protected:
 
     /** keeps off-diagonal entries; drops diagonal entries */
@@ -191,18 +193,17 @@ protected:
     int m_fillfactor;
     bool m_analysisIsOk;
     bool m_factorizationIsOk;
-    bool m_isInitialized;
     ComputationInfo m_info;
-    PermutationMatrix<Dynamic,Dynamic,Index> m_P;     // Fill-reducing permutation
-    PermutationMatrix<Dynamic,Dynamic,Index> m_Pinv;  // Inverse permutation
+    PermutationMatrix<Dynamic,Dynamic,StorageIndex> m_P;     // Fill-reducing permutation
+    PermutationMatrix<Dynamic,Dynamic,StorageIndex> m_Pinv;  // Inverse permutation
 };
 
 /**
  * Set control parameter droptol
  *  \param droptol   Drop any element whose magnitude is less than this tolerance 
  **/ 
-template<typename Scalar>
-void IncompleteLUT<Scalar>::setDroptol(const RealScalar& droptol)
+template<typename Scalar, typename StorageIndex>
+void IncompleteLUT<Scalar,StorageIndex>::setDroptol(const RealScalar& droptol)
 {
   this->m_droptol = droptol;   
 }
@@ -211,61 +212,62 @@ void IncompleteLUT<Scalar>::setDroptol(const RealScalar& droptol)
  * Set control parameter fillfactor
  * \param fillfactor  This is used to compute the  number @p fill_in of largest elements to keep on each row. 
  **/ 
-template<typename Scalar>
-void IncompleteLUT<Scalar>::setFillfactor(int fillfactor)
+template<typename Scalar, typename StorageIndex>
+void IncompleteLUT<Scalar,StorageIndex>::setFillfactor(int fillfactor)
 {
   this->m_fillfactor = fillfactor;   
 }
 
-template <typename Scalar>
+template <typename Scalar, typename StorageIndex>
 template<typename _MatrixType>
-void IncompleteLUT<Scalar>::analyzePattern(const _MatrixType& amat)
+void IncompleteLUT<Scalar,StorageIndex>::analyzePattern(const _MatrixType& amat)
 {
   // Compute the Fill-reducing permutation
   // Since ILUT does not perform any numerical pivoting,
   // it is highly preferable to keep the diagonal through symmetric permutations.
 #ifndef EIGEN_MPL2_ONLY
   // To this end, let's symmetrize the pattern and perform AMD on it.
-  SparseMatrix<Scalar,ColMajor, Index> mat1 = amat;
-  SparseMatrix<Scalar,ColMajor, Index> mat2 = amat.transpose();
+  SparseMatrix<Scalar,ColMajor, StorageIndex> mat1 = amat;
+  SparseMatrix<Scalar,ColMajor, StorageIndex> mat2 = amat.transpose();
   // FIXME for a matrix with nearly symmetric pattern, mat2+mat1 is the appropriate choice.
   //       on the other hand for a really non-symmetric pattern, mat2*mat1 should be prefered...
-  SparseMatrix<Scalar,ColMajor, Index> AtA = mat2 + mat1;
-  AMDOrdering<Index> ordering;
+  SparseMatrix<Scalar,ColMajor, StorageIndex> AtA = mat2 + mat1;
+  AMDOrdering<StorageIndex> ordering;
   ordering(AtA,m_P);
   m_Pinv  = m_P.inverse(); // cache the inverse permutation
 #else
   // If AMD is not available, (MPL2-only), then let's use the slower COLAMD routine.
-  SparseMatrix<Scalar,ColMajor, Index> mat1 = amat;
-  COLAMDOrdering<Index> ordering;
+  SparseMatrix<Scalar,ColMajor, StorageIndex> mat1 = amat;
+  COLAMDOrdering<StorageIndex> ordering;
   ordering(mat1,m_Pinv);
   m_P = m_Pinv.inverse();
 #endif
 
   m_analysisIsOk = true;
   m_factorizationIsOk = false;
-  m_isInitialized = false;
+  m_isInitialized = true;
 }
 
-template <typename Scalar>
+template <typename Scalar, typename StorageIndex>
 template<typename _MatrixType>
-void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
+void IncompleteLUT<Scalar,StorageIndex>::factorize(const _MatrixType& amat)
 {
   using std::sqrt;
   using std::swap;
   using std::abs;
+  using internal::convert_index;
 
   eigen_assert((amat.rows() == amat.cols()) && "The factorization should be done on a square matrix");
   Index n = amat.cols();  // Size of the matrix
   m_lu.resize(n,n);
   // Declare Working vectors and variables
   Vector u(n) ;     // real values of the row -- maximum size is n --
-  VectorXi ju(n);   // column position of the values in u -- maximum size  is n
-  VectorXi jr(n);   // Indicate the position of the nonzero elements in the vector u -- A zero location is indicated by -1
+  VectorI ju(n);   // column position of the values in u -- maximum size  is n
+  VectorI jr(n);   // Indicate the position of the nonzero elements in the vector u -- A zero location is indicated by -1
 
   // Apply the fill-reducing permutation
   eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
-  SparseMatrix<Scalar,RowMajor, Index> mat;
+  SparseMatrix<Scalar,RowMajor, StorageIndex> mat;
   mat = amat.twistedBy(m_Pinv);
 
   // Initialization
@@ -274,7 +276,7 @@ void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
   u.fill(0);
 
   // number of largest elements to keep in each row:
-  Index fill_in =   static_cast<Index> (amat.nonZeros()*m_fillfactor)/n+1;
+  Index fill_in = (amat.nonZeros()*m_fillfactor)/n + 1;
   if (fill_in > n) fill_in = n;
 
   // number of largest nonzero elements to keep in the L and the U part of the current row:
@@ -289,9 +291,9 @@ void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
 
     Index sizeu = 1; // number of nonzero elements in the upper part of the current row
     Index sizel = 0; // number of nonzero elements in the lower part of the current row
-    ju(ii)    = ii;
+    ju(ii)    = convert_index<StorageIndex>(ii);
     u(ii)     = 0;
-    jr(ii)    = ii;
+    jr(ii)    = convert_index<StorageIndex>(ii);
     RealScalar rownorm = 0;
 
     typename FactorType::InnerIterator j_it(mat, ii); // Iterate through the current row ii
@@ -301,9 +303,9 @@ void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
       if (k < ii)
       {
         // copy the lower part
-        ju(sizel) = k;
+        ju(sizel) = convert_index<StorageIndex>(k);
         u(sizel) = j_it.value();
-        jr(k) = sizel;
+        jr(k) = convert_index<StorageIndex>(sizel);
         ++sizel;
       }
       else if (k == ii)
@@ -314,9 +316,9 @@ void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
       {
         // copy the upper part
         Index jpos = ii + sizeu;
-        ju(jpos) = k;
+        ju(jpos) = convert_index<StorageIndex>(k);
         u(jpos) = j_it.value();
-        jr(k) = jpos;
+        jr(k) = convert_index<StorageIndex>(jpos);
         ++sizeu;
       }
       rownorm += numext::abs2(j_it.value());
@@ -346,7 +348,8 @@ void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
         // swap the two locations
         Index j = ju(jj);
         swap(ju(jj), ju(k));
-        jr(minrow) = jj;   jr(j) = k;
+        jr(minrow) = convert_index<StorageIndex>(jj);
+        jr(j) = convert_index<StorageIndex>(k);
         swap(u(jj), u(k));
       }
       // Reset this location
@@ -370,8 +373,8 @@ void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
       for (; ki_it; ++ki_it)
       {
         Scalar prod = fact * ki_it.value();
-        Index j       = ki_it.index();
-        Index jpos    = jr(j);
+        Index j     = ki_it.index();
+        Index jpos  = jr(j);
         if (jpos == -1) // fill-in element
         {
           Index newpos;
@@ -387,16 +390,16 @@ void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
             sizel++;
             eigen_internal_assert(sizel<=ii);
           }
-          ju(newpos) = j;
+          ju(newpos) = convert_index<StorageIndex>(j);
           u(newpos) = -prod;
-          jr(j) = newpos;
+          jr(j) = convert_index<StorageIndex>(newpos);
         }
         else
           u(jpos) -= prod;
       }
       // store the pivot element
-      u(len) = fact;
-      ju(len) = minrow;
+      u(len)  = fact;
+      ju(len) = convert_index<StorageIndex>(minrow);
       ++len;
 
       jj++;
@@ -411,7 +414,7 @@ void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
     sizel = len;
     len = (std::min)(sizel, nnzL);
     typename Vector::SegmentReturnType ul(u.segment(0, sizel));
-    typename VectorXi::SegmentReturnType jul(ju.segment(0, sizel));
+    typename VectorI::SegmentReturnType jul(ju.segment(0, sizel));
     internal::QuickSplit(ul, jul, len);
 
     // store the largest m_fill elements of the L part
@@ -440,39 +443,20 @@ void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
     sizeu = len + 1; // +1 to take into account the diagonal element
     len = (std::min)(sizeu, nnzU);
     typename Vector::SegmentReturnType uu(u.segment(ii+1, sizeu-1));
-    typename VectorXi::SegmentReturnType juu(ju.segment(ii+1, sizeu-1));
+    typename VectorI::SegmentReturnType juu(ju.segment(ii+1, sizeu-1));
     internal::QuickSplit(uu, juu, len);
 
     // store the largest elements of the U part
     for(Index k = ii + 1; k < ii + len; k++)
       m_lu.insertBackByOuterInnerUnordered(ii,ju(k)) = u(k);
   }
-
   m_lu.finalize();
   m_lu.makeCompressed();
 
   m_factorizationIsOk = true;
-  m_isInitialized = m_factorizationIsOk;
   m_info = Success;
 }
 
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<IncompleteLUT<_MatrixType>, Rhs>
-  : solve_retval_base<IncompleteLUT<_MatrixType>, Rhs>
-{
-  typedef IncompleteLUT<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_INCOMPLETE_LUT_H
diff --git a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IterativeSolverBase.h b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IterativeSolverBase.h
index 501ef2f8d87..7c2326eb7fd 100644
--- a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IterativeSolverBase.h
+++ b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/IterativeSolverBase.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,29 +12,158 @@
 
 namespace Eigen { 
 
+namespace internal {
+
+template<typename MatrixType>
+struct is_ref_compatible_impl
+{
+private:
+  template <typename T0>
+  struct any_conversion
+  {
+    template <typename T> any_conversion(const volatile T&);
+    template <typename T> any_conversion(T&);
+  };
+  struct yes {int a[1];};
+  struct no  {int a[2];};
+
+  template<typename T>
+  static yes test(const Ref<const T>&, int);
+  template<typename T>
+  static no  test(any_conversion<T>, ...);
+
+public:
+  static MatrixType ms_from;
+  enum { value = sizeof(test<MatrixType>(ms_from, 0))==sizeof(yes) };
+};
+
+template<typename MatrixType>
+struct is_ref_compatible
+{
+  enum { value = is_ref_compatible_impl<typename remove_all<MatrixType>::type>::value };
+};
+
+template<typename MatrixType, bool MatrixFree = !internal::is_ref_compatible<MatrixType>::value>
+class generic_matrix_wrapper;
+
+// We have an explicit matrix at hand, compatible with Ref<>
+template<typename MatrixType>
+class generic_matrix_wrapper<MatrixType,false>
+{
+public:
+  typedef Ref<const MatrixType> ActualMatrixType;
+  template<int UpLo> struct ConstSelfAdjointViewReturnType {
+    typedef typename ActualMatrixType::template ConstSelfAdjointViewReturnType<UpLo>::Type Type;
+  };
+
+  enum {
+    MatrixFree = false
+  };
+
+  generic_matrix_wrapper()
+    : m_dummy(0,0), m_matrix(m_dummy)
+  {}
+
+  template<typename InputType>
+  generic_matrix_wrapper(const InputType &mat)
+    : m_matrix(mat)
+  {}
+
+  const ActualMatrixType& matrix() const
+  {
+    return m_matrix;
+  }
+
+  template<typename MatrixDerived>
+  void grab(const EigenBase<MatrixDerived> &mat)
+  {
+    m_matrix.~Ref<const MatrixType>();
+    ::new (&m_matrix) Ref<const MatrixType>(mat.derived());
+  }
+
+  void grab(const Ref<const MatrixType> &mat)
+  {
+    if(&(mat.derived()) != &m_matrix)
+    {
+      m_matrix.~Ref<const MatrixType>();
+      ::new (&m_matrix) Ref<const MatrixType>(mat);
+    }
+  }
+
+protected:
+  MatrixType m_dummy; // used to default initialize the Ref<> object
+  ActualMatrixType m_matrix;
+};
+
+// MatrixType is not compatible with Ref<> -> matrix-free wrapper
+template<typename MatrixType>
+class generic_matrix_wrapper<MatrixType,true>
+{
+public:
+  typedef MatrixType ActualMatrixType;
+  template<int UpLo> struct ConstSelfAdjointViewReturnType
+  {
+    typedef ActualMatrixType Type;
+  };
+
+  enum {
+    MatrixFree = true
+  };
+
+  generic_matrix_wrapper()
+    : mp_matrix(0)
+  {}
+
+  generic_matrix_wrapper(const MatrixType &mat)
+    : mp_matrix(&mat)
+  {}
+
+  const ActualMatrixType& matrix() const
+  {
+    return *mp_matrix;
+  }
+
+  void grab(const MatrixType &mat)
+  {
+    mp_matrix = &mat;
+  }
+
+protected:
+  const ActualMatrixType *mp_matrix;
+};
+
+}
+
 /** \ingroup IterativeLinearSolvers_Module
   * \brief Base class for linear iterative solvers
   *
   * \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
   */
 template< typename Derived>
-class IterativeSolverBase : internal::noncopyable
+class IterativeSolverBase : public SparseSolverBase<Derived>
 {
+protected:
+  typedef SparseSolverBase<Derived> Base;
+  using Base::m_isInitialized;
+  
 public:
   typedef typename internal::traits<Derived>::MatrixType MatrixType;
   typedef typename internal::traits<Derived>::Preconditioner Preconditioner;
   typedef typename MatrixType::Scalar Scalar;
-  typedef typename MatrixType::Index Index;
+  typedef typename MatrixType::StorageIndex StorageIndex;
   typedef typename MatrixType::RealScalar RealScalar;
 
+  enum {
+    ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+  };
+
 public:
 
-  Derived& derived() { return *static_cast<Derived*>(this); }
-  const Derived& derived() const { return *static_cast<const Derived*>(this); }
+  using Base::derived;
 
   /** Default constructor. */
   IterativeSolverBase()
-    : mp_matrix(0)
   {
     init();
   }
@@ -49,82 +178,90 @@ public:
     * this class becomes invalid. Call compute() to update it with the new
     * matrix A, or modify a copy of A.
     */
-  template<typename InputDerived>
-  IterativeSolverBase(const EigenBase<InputDerived>& A)
+  template<typename MatrixDerived>
+  explicit IterativeSolverBase(const EigenBase<MatrixDerived>& A)
+    : m_matrixWrapper(A.derived())
   {
     init();
-    compute(A.derived());
+    compute(matrix());
   }
 
   ~IterativeSolverBase() {}
   
-  /** Initializes the iterative solver for the sparcity pattern of the matrix \a A for further solving \c Ax=b problems.
+  /** Initializes the iterative solver for the sparsity pattern of the matrix \a A for further solving \c Ax=b problems.
     *
-    * Currently, this function mostly call analyzePattern on the preconditioner. In the future
-    * we might, for instance, implement column reodering for faster matrix vector products.
+    * Currently, this function mostly calls analyzePattern on the preconditioner. In the future
+    * we might, for instance, implement column reordering for faster matrix vector products.
     */
-  template<typename InputDerived>
-  Derived& analyzePattern(const EigenBase<InputDerived>& A)
+  template<typename MatrixDerived>
+  Derived& analyzePattern(const EigenBase<MatrixDerived>& A)
   {
-    grabInput(A.derived());
-    m_preconditioner.analyzePattern(*mp_matrix);
+    grab(A.derived());
+    m_preconditioner.analyzePattern(matrix());
     m_isInitialized = true;
     m_analysisIsOk = true;
-    m_info = Success;
+    m_info = m_preconditioner.info();
     return derived();
   }
   
   /** Initializes the iterative solver with the numerical values of the matrix \a A for further solving \c Ax=b problems.
     *
-    * Currently, this function mostly call factorize on the preconditioner.
+    * Currently, this function mostly calls factorize on the preconditioner.
     *
     * \warning this class stores a reference to the matrix A as well as some
     * precomputed values that depend on it. Therefore, if \a A is changed
     * this class becomes invalid. Call compute() to update it with the new
     * matrix A, or modify a copy of A.
     */
-  template<typename InputDerived>
-  Derived& factorize(const EigenBase<InputDerived>& A)
+  template<typename MatrixDerived>
+  Derived& factorize(const EigenBase<MatrixDerived>& A)
   {
-    grabInput(A.derived());
     eigen_assert(m_analysisIsOk && "You must first call analyzePattern()"); 
-    m_preconditioner.factorize(*mp_matrix);
+    grab(A.derived());
+    m_preconditioner.factorize(matrix());
     m_factorizationIsOk = true;
-    m_info = Success;
+    m_info = m_preconditioner.info();
     return derived();
   }
 
   /** Initializes the iterative solver with the matrix \a A for further solving \c Ax=b problems.
     *
-    * Currently, this function mostly initialized/compute the preconditioner. In the future
-    * we might, for instance, implement column reodering for faster matrix vector products.
+    * Currently, this function mostly initializes/computes the preconditioner. In the future
+    * we might, for instance, implement column reordering for faster matrix vector products.
     *
     * \warning this class stores a reference to the matrix A as well as some
     * precomputed values that depend on it. Therefore, if \a A is changed
     * this class becomes invalid. Call compute() to update it with the new
     * matrix A, or modify a copy of A.
     */
-  template<typename InputDerived>
-  Derived& compute(const EigenBase<InputDerived>& A)
+  template<typename MatrixDerived>
+  Derived& compute(const EigenBase<MatrixDerived>& A)
   {
-    grabInput(A.derived());
-    m_preconditioner.compute(*mp_matrix);
+    grab(A.derived());
+    m_preconditioner.compute(matrix());
     m_isInitialized = true;
     m_analysisIsOk = true;
     m_factorizationIsOk = true;
-    m_info = Success;
+    m_info = m_preconditioner.info();
     return derived();
   }
 
   /** \internal */
-  Index rows() const { return mp_matrix ? mp_matrix->rows() : 0; }
+  Index rows() const { return matrix().rows(); }
+
   /** \internal */
-  Index cols() const { return mp_matrix ? mp_matrix->cols() : 0; }
+  Index cols() const { return matrix().cols(); }
 
-  /** \returns the tolerance threshold used by the stopping criteria */
+  /** \returns the tolerance threshold used by the stopping criteria.
+    * \sa setTolerance()
+    */
   RealScalar tolerance() const { return m_tolerance; }
   
-  /** Sets the tolerance threshold used by the stopping criteria */
+  /** Sets the tolerance threshold used by the stopping criteria.
+    *
+    * This value is used as an upper bound to the relative residual error: |Ax-b|/|b|.
+    * The default value is the machine precision given by NumTraits<Scalar>::epsilon()
+    */
   Derived& setTolerance(const RealScalar& tolerance)
   {
     m_tolerance = tolerance;
@@ -137,58 +274,52 @@ public:
   /** \returns a read-only reference to the preconditioner. */
   const Preconditioner& preconditioner() const { return m_preconditioner; }
 
-  /** \returns the max number of iterations */
-  int maxIterations() const
+  /** \returns the max number of iterations.
+    * It is either the value setted by setMaxIterations or, by default,
+    * twice the number of columns of the matrix.
+    */
+  Index maxIterations() const
   {
-    return (mp_matrix && m_maxIterations<0) ? mp_matrix->cols() : m_maxIterations;
+    return (m_maxIterations<0) ? 2*matrix().cols() : m_maxIterations;
   }
   
-  /** Sets the max number of iterations */
-  Derived& setMaxIterations(int maxIters)
+  /** Sets the max number of iterations.
+    * Default is twice the number of columns of the matrix.
+    */
+  Derived& setMaxIterations(Index maxIters)
   {
     m_maxIterations = maxIters;
     return derived();
   }
 
   /** \returns the number of iterations performed during the last solve */
-  int iterations() const
+  Index iterations() const
   {
     eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
     return m_iterations;
   }
 
-  /** \returns the tolerance error reached during the last solve */
+  /** \returns the tolerance error reached during the last solve.
+    * It is a close approximation of the true relative residual error |Ax-b|/|b|.
+    */
   RealScalar error() const
   {
     eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
     return m_error;
   }
 
-  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-    *
-    * \sa compute()
-    */
-  template<typename Rhs> inline const internal::solve_retval<Derived, Rhs>
-  solve(const MatrixBase<Rhs>& b) const
-  {
-    eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
-    eigen_assert(rows()==b.rows()
-              && "IterativeSolverBase::solve(): invalid number of rows of the right hand side matrix b");
-    return internal::solve_retval<Derived, Rhs>(derived(), b.derived());
-  }
-  
-  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
+  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
+    * and \a x0 as an initial solution.
     *
-    * \sa compute()
+    * \sa solve(), compute()
     */
-  template<typename Rhs>
-  inline const internal::sparse_solve_retval<IterativeSolverBase, Rhs>
-  solve(const SparseMatrixBase<Rhs>& b) const
+  template<typename Rhs,typename Guess>
+  inline const SolveWithGuess<Derived, Rhs, Guess>
+  solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
   {
-    eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
-    eigen_assert(rows()==b.rows()
-              && "IterativeSolverBase::solve(): invalid number of rows of the right hand side matrix b");
-    return internal::sparse_solve_retval<IterativeSolverBase, Rhs>(*this, b.derived());
+    eigen_assert(m_isInitialized && "Solver is not initialized.");
+    eigen_assert(derived().rows()==b.rows() && "solve(): invalid number of rows of the right hand side matrix b");
+    return SolveWithGuess<Derived, Rhs, Guess>(derived(), b.derived(), x0);
   }
 
   /** \returns Success if the iterations converged, and NoConvergence otherwise. */
@@ -199,46 +330,30 @@ public:
   }
   
   /** \internal */
-  template<typename Rhs, typename DestScalar, int DestOptions, typename DestIndex>
-  void _solve_sparse(const Rhs& b, SparseMatrix<DestScalar,DestOptions,DestIndex> &dest) const
+  template<typename Rhs, typename DestDerived>
+  void _solve_impl(const Rhs& b, SparseMatrixBase<DestDerived> &aDest) const
   {
     eigen_assert(rows()==b.rows());
     
-    int rhsCols = b.cols();
-    int size = b.rows();
+    Index rhsCols = b.cols();
+    Index size = b.rows();
+    DestDerived& dest(aDest.derived());
+    typedef typename DestDerived::Scalar DestScalar;
     Eigen::Matrix<DestScalar,Dynamic,1> tb(size);
-    Eigen::Matrix<DestScalar,Dynamic,1> tx(size);
-    for(int k=0; k<rhsCols; ++k)
+    Eigen::Matrix<DestScalar,Dynamic,1> tx(cols());
+    // We do not directly fill dest because sparse expressions have to be free of aliasing issue.
+    // For non square least-square problems, b and dest might not have the same size whereas they might alias each-other.
+    typename DestDerived::PlainObject tmp(cols(),rhsCols);
+    for(Index k=0; k<rhsCols; ++k)
     {
       tb = b.col(k);
       tx = derived().solve(tb);
-      dest.col(k) = tx.sparseView(0);
+      tmp.col(k) = tx.sparseView(0);
     }
+    dest.swap(tmp);
   }
 
 protected:
-
-  template<typename InputDerived>
-  void grabInput(const EigenBase<InputDerived>& A)
-  {
-    // we const cast to prevent the creation of a MatrixType temporary by the compiler.
-    grabInput_impl(A.const_cast_derived());
-  }
-
-  template<typename InputDerived>
-  void grabInput_impl(const EigenBase<InputDerived>& A)
-  {
-    m_copyMatrix = A;
-    mp_matrix = &m_copyMatrix;
-  }
-
-  void grabInput_impl(MatrixType& A)
-  {
-    if(MatrixType::RowsAtCompileTime==Dynamic && MatrixType::ColsAtCompileTime==Dynamic)
-      m_copyMatrix.resize(0,0);
-    mp_matrix = &A;
-  }
-
   void init()
   {
     m_isInitialized = false;
@@ -247,36 +362,33 @@ protected:
     m_maxIterations = -1;
     m_tolerance = NumTraits<Scalar>::epsilon();
   }
-  MatrixType m_copyMatrix;
-  const MatrixType* mp_matrix;
+
+  typedef internal::generic_matrix_wrapper<MatrixType> MatrixWrapper;
+  typedef typename MatrixWrapper::ActualMatrixType ActualMatrixType;
+
+  const ActualMatrixType& matrix() const
+  {
+    return m_matrixWrapper.matrix();
+  }
+  
+  template<typename InputType>
+  void grab(const InputType &A)
+  {
+    m_matrixWrapper.grab(A);
+  }
+  
+  MatrixWrapper m_matrixWrapper;
   Preconditioner m_preconditioner;
 
-  int m_maxIterations;
+  Index m_maxIterations;
   RealScalar m_tolerance;
   
   mutable RealScalar m_error;
-  mutable int m_iterations;
+  mutable Index m_iterations;
   mutable ComputationInfo m_info;
-  mutable bool m_isInitialized, m_analysisIsOk, m_factorizationIsOk;
+  mutable bool m_analysisIsOk, m_factorizationIsOk;
 };
 
-namespace internal {
- 
-template<typename Derived, typename Rhs>
-struct sparse_solve_retval<IterativeSolverBase<Derived>, Rhs>
-  : sparse_solve_retval_base<IterativeSolverBase<Derived>, Rhs>
-{
-  typedef IterativeSolverBase<Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec().derived()._solve_sparse(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_ITERATIVE_SOLVER_BASE_H
diff --git a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/LeastSquareConjugateGradient.h b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/LeastSquareConjugateGradient.h
new file mode 100644
index 00000000000..0aea0e099d2
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/LeastSquareConjugateGradient.h
@@ -0,0 +1,216 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_LEAST_SQUARE_CONJUGATE_GRADIENT_H
+#define EIGEN_LEAST_SQUARE_CONJUGATE_GRADIENT_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/** \internal Low-level conjugate gradient algorithm for least-square problems
+  * \param mat The matrix A
+  * \param rhs The right hand side vector b
+  * \param x On input and initial solution, on output the computed solution.
+  * \param precond A preconditioner being able to efficiently solve for an
+  *                approximation of A'Ax=b (regardless of b)
+  * \param iters On input the max number of iteration, on output the number of performed iterations.
+  * \param tol_error On input the tolerance error, on output an estimation of the relative error.
+  */
+template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
+EIGEN_DONT_INLINE
+void least_square_conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
+                                     const Preconditioner& precond, Index& iters,
+                                     typename Dest::RealScalar& tol_error)
+{
+  using std::sqrt;
+  using std::abs;
+  typedef typename Dest::RealScalar RealScalar;
+  typedef typename Dest::Scalar Scalar;
+  typedef Matrix<Scalar,Dynamic,1> VectorType;
+  
+  RealScalar tol = tol_error;
+  Index maxIters = iters;
+  
+  Index m = mat.rows(), n = mat.cols();
+
+  VectorType residual        = rhs - mat * x;
+  VectorType normal_residual = mat.adjoint() * residual;
+
+  RealScalar rhsNorm2 = (mat.adjoint()*rhs).squaredNorm();
+  if(rhsNorm2 == 0) 
+  {
+    x.setZero();
+    iters = 0;
+    tol_error = 0;
+    return;
+  }
+  RealScalar threshold = tol*tol*rhsNorm2;
+  RealScalar residualNorm2 = normal_residual.squaredNorm();
+  if (residualNorm2 < threshold)
+  {
+    iters = 0;
+    tol_error = sqrt(residualNorm2 / rhsNorm2);
+    return;
+  }
+  
+  VectorType p(n);
+  p = precond.solve(normal_residual);                         // initial search direction
+
+  VectorType z(n), tmp(m);
+  RealScalar absNew = numext::real(normal_residual.dot(p));  // the square of the absolute value of r scaled by invM
+  Index i = 0;
+  while(i < maxIters)
+  {
+    tmp.noalias() = mat * p;
+
+    Scalar alpha = absNew / tmp.squaredNorm();      // the amount we travel on dir
+    x += alpha * p;                                 // update solution
+    residual -= alpha * tmp;                        // update residual
+    normal_residual = mat.adjoint() * residual;     // update residual of the normal equation
+    
+    residualNorm2 = normal_residual.squaredNorm();
+    if(residualNorm2 < threshold)
+      break;
+    
+    z = precond.solve(normal_residual);             // approximately solve for "A'A z = normal_residual"
+
+    RealScalar absOld = absNew;
+    absNew = numext::real(normal_residual.dot(z));  // update the absolute value of r
+    RealScalar beta = absNew / absOld;              // calculate the Gram-Schmidt value used to create the new search direction
+    p = z + beta * p;                               // update search direction
+    i++;
+  }
+  tol_error = sqrt(residualNorm2 / rhsNorm2);
+  iters = i;
+}
+
+}
+
+template< typename _MatrixType,
+          typename _Preconditioner = LeastSquareDiagonalPreconditioner<typename _MatrixType::Scalar> >
+class LeastSquaresConjugateGradient;
+
+namespace internal {
+
+template< typename _MatrixType, typename _Preconditioner>
+struct traits<LeastSquaresConjugateGradient<_MatrixType,_Preconditioner> >
+{
+  typedef _MatrixType MatrixType;
+  typedef _Preconditioner Preconditioner;
+};
+
+}
+
+/** \ingroup IterativeLinearSolvers_Module
+  * \brief A conjugate gradient solver for sparse (or dense) least-square problems
+  *
+  * This class allows to solve for A x = b linear problems using an iterative conjugate gradient algorithm.
+  * The matrix A can be non symmetric and rectangular, but the matrix A' A should be positive-definite to guaranty stability.
+  * Otherwise, the SparseLU or SparseQR classes might be preferable.
+  * The matrix A and the vectors x and b can be either dense or sparse.
+  *
+  * \tparam _MatrixType the type of the matrix A, can be a dense or a sparse matrix.
+  * \tparam _Preconditioner the type of the preconditioner. Default is LeastSquareDiagonalPreconditioner
+  *
+  * \implsparsesolverconcept
+  * 
+  * The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
+  * and setTolerance() methods. The defaults are the size of the problem for the maximal number of iterations
+  * and NumTraits<Scalar>::epsilon() for the tolerance.
+  * 
+  * This class can be used as the direct solver classes. Here is a typical usage example:
+    \code
+    int m=1000000, n = 10000;
+    VectorXd x(n), b(m);
+    SparseMatrix<double> A(m,n);
+    // fill A and b
+    LeastSquaresConjugateGradient<SparseMatrix<double> > lscg;
+    lscg.compute(A);
+    x = lscg.solve(b);
+    std::cout << "#iterations:     " << lscg.iterations() << std::endl;
+    std::cout << "estimated error: " << lscg.error()      << std::endl;
+    // update b, and solve again
+    x = lscg.solve(b);
+    \endcode
+  * 
+  * By default the iterations start with x=0 as an initial guess of the solution.
+  * One can control the start using the solveWithGuess() method.
+  * 
+  * \sa class ConjugateGradient, SparseLU, SparseQR
+  */
+template< typename _MatrixType, typename _Preconditioner>
+class LeastSquaresConjugateGradient : public IterativeSolverBase<LeastSquaresConjugateGradient<_MatrixType,_Preconditioner> >
+{
+  typedef IterativeSolverBase<LeastSquaresConjugateGradient> Base;
+  using Base::matrix;
+  using Base::m_error;
+  using Base::m_iterations;
+  using Base::m_info;
+  using Base::m_isInitialized;
+public:
+  typedef _MatrixType MatrixType;
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename MatrixType::RealScalar RealScalar;
+  typedef _Preconditioner Preconditioner;
+
+public:
+
+  /** Default constructor. */
+  LeastSquaresConjugateGradient() : Base() {}
+
+  /** Initialize the solver with matrix \a A for further \c Ax=b solving.
+    * 
+    * This constructor is a shortcut for the default constructor followed
+    * by a call to compute().
+    * 
+    * \warning this class stores a reference to the matrix A as well as some
+    * precomputed values that depend on it. Therefore, if \a A is changed
+    * this class becomes invalid. Call compute() to update it with the new
+    * matrix A, or modify a copy of A.
+    */
+  template<typename MatrixDerived>
+  explicit LeastSquaresConjugateGradient(const EigenBase<MatrixDerived>& A) : Base(A.derived()) {}
+
+  ~LeastSquaresConjugateGradient() {}
+
+  /** \internal */
+  template<typename Rhs,typename Dest>
+  void _solve_with_guess_impl(const Rhs& b, Dest& x) const
+  {
+    m_iterations = Base::maxIterations();
+    m_error = Base::m_tolerance;
+
+    for(Index j=0; j<b.cols(); ++j)
+    {
+      m_iterations = Base::maxIterations();
+      m_error = Base::m_tolerance;
+
+      typename Dest::ColXpr xj(x,j);
+      internal::least_square_conjugate_gradient(matrix(), b.col(j), xj, Base::m_preconditioner, m_iterations, m_error);
+    }
+
+    m_isInitialized = true;
+    m_info = m_error <= Base::m_tolerance ? Success : NoConvergence;
+  }
+  
+  /** \internal */
+  using Base::_solve_impl;
+  template<typename Rhs,typename Dest>
+  void _solve_impl(const MatrixBase<Rhs>& b, Dest& x) const
+  {
+    x.setZero();
+    _solve_with_guess_impl(b.derived(),x);
+  }
+
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_LEAST_SQUARE_CONJUGATE_GRADIENT_H
diff --git a/extern/Eigen3/Eigen/src/IterativeLinearSolvers/SolveWithGuess.h b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/SolveWithGuess.h
new file mode 100644
index 00000000000..0ace4517717
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/IterativeLinearSolvers/SolveWithGuess.h
@@ -0,0 +1,115 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SOLVEWITHGUESS_H
+#define EIGEN_SOLVEWITHGUESS_H
+
+namespace Eigen {
+
+template<typename Decomposition, typename RhsType, typename GuessType> class SolveWithGuess;
+  
+/** \class SolveWithGuess
+  * \ingroup IterativeLinearSolvers_Module
+  *
+  * \brief Pseudo expression representing a solving operation
+  *
+  * \tparam Decomposition the type of the matrix or decomposion object
+  * \tparam Rhstype the type of the right-hand side
+  *
+  * This class represents an expression of A.solve(B)
+  * and most of the time this is the only way it is used.
+  *
+  */
+namespace internal {
+
+
+template<typename Decomposition, typename RhsType, typename GuessType>
+struct traits<SolveWithGuess<Decomposition, RhsType, GuessType> >
+  : traits<Solve<Decomposition,RhsType> >
+{};
+
+}
+
+
+template<typename Decomposition, typename RhsType, typename GuessType>
+class SolveWithGuess : public internal::generic_xpr_base<SolveWithGuess<Decomposition,RhsType,GuessType>, MatrixXpr, typename internal::traits<RhsType>::StorageKind>::type
+{
+public:
+  typedef typename internal::traits<SolveWithGuess>::Scalar Scalar;
+  typedef typename internal::traits<SolveWithGuess>::PlainObject PlainObject;
+  typedef typename internal::generic_xpr_base<SolveWithGuess<Decomposition,RhsType,GuessType>, MatrixXpr, typename internal::traits<RhsType>::StorageKind>::type Base;
+  typedef typename internal::ref_selector<SolveWithGuess>::type Nested;
+  
+  SolveWithGuess(const Decomposition &dec, const RhsType &rhs, const GuessType &guess)
+    : m_dec(dec), m_rhs(rhs), m_guess(guess)
+  {}
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return m_dec.cols(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_rhs.cols(); }
+
+  EIGEN_DEVICE_FUNC const Decomposition& dec()   const { return m_dec; }
+  EIGEN_DEVICE_FUNC const RhsType&       rhs()   const { return m_rhs; }
+  EIGEN_DEVICE_FUNC const GuessType&     guess() const { return m_guess; }
+
+protected:
+  const Decomposition &m_dec;
+  const RhsType       &m_rhs;
+  const GuessType     &m_guess;
+  
+private:
+  Scalar coeff(Index row, Index col) const;
+  Scalar coeff(Index i) const;
+};
+
+namespace internal {
+
+// Evaluator of SolveWithGuess -> eval into a temporary
+template<typename Decomposition, typename RhsType, typename GuessType>
+struct evaluator<SolveWithGuess<Decomposition,RhsType, GuessType> >
+  : public evaluator<typename SolveWithGuess<Decomposition,RhsType,GuessType>::PlainObject>
+{
+  typedef SolveWithGuess<Decomposition,RhsType,GuessType> SolveType;
+  typedef typename SolveType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  evaluator(const SolveType& solve)
+    : m_result(solve.rows(), solve.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    m_result = solve.guess();
+    solve.dec()._solve_with_guess_impl(solve.rhs(), m_result);
+  }
+  
+protected:  
+  PlainObject m_result;
+};
+
+// Specialization for "dst = dec.solveWithGuess(rhs)"
+// NOTE we need to specialize it for Dense2Dense to avoid ambiguous specialization error and a Sparse2Sparse specialization must exist somewhere
+template<typename DstXprType, typename DecType, typename RhsType, typename GuessType, typename Scalar>
+struct Assignment<DstXprType, SolveWithGuess<DecType,RhsType,GuessType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
+{
+  typedef SolveWithGuess<DecType,RhsType,GuessType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    dst = src.guess();
+    src.dec()._solve_with_guess_impl(src.rhs(), dst/*, src.guess()*/);
+  }
+};
+
+} // end namepsace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SOLVEWITHGUESS_H
diff --git a/extern/Eigen3/Eigen/src/Jacobi/Jacobi.h b/extern/Eigen3/Eigen/src/Jacobi/Jacobi.h
index 956f72d5701..1998c632274 100644
--- a/extern/Eigen3/Eigen/src/Jacobi/Jacobi.h
+++ b/extern/Eigen3/Eigen/src/Jacobi/Jacobi.h
@@ -62,14 +62,14 @@ template<typename Scalar> class JacobiRotation
     JacobiRotation adjoint() const { using numext::conj; return JacobiRotation(conj(m_c), -m_s); }
 
     template<typename Derived>
-    bool makeJacobi(const MatrixBase<Derived>&, typename Derived::Index p, typename Derived::Index q);
+    bool makeJacobi(const MatrixBase<Derived>&, Index p, Index q);
     bool makeJacobi(const RealScalar& x, const Scalar& y, const RealScalar& z);
 
-    void makeGivens(const Scalar& p, const Scalar& q, Scalar* z=0);
+    void makeGivens(const Scalar& p, const Scalar& q, Scalar* r=0);
 
   protected:
-    void makeGivens(const Scalar& p, const Scalar& q, Scalar* z, internal::true_type);
-    void makeGivens(const Scalar& p, const Scalar& q, Scalar* z, internal::false_type);
+    void makeGivens(const Scalar& p, const Scalar& q, Scalar* r, internal::true_type);
+    void makeGivens(const Scalar& p, const Scalar& q, Scalar* r, internal::false_type);
 
     Scalar m_c, m_s;
 };
@@ -84,8 +84,8 @@ bool JacobiRotation<Scalar>::makeJacobi(const RealScalar& x, const Scalar& y, co
 {
   using std::sqrt;
   using std::abs;
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  if(y == Scalar(0))
+  RealScalar deno = RealScalar(2)*abs(y);
+  if(deno < (std::numeric_limits<RealScalar>::min)())
   {
     m_c = Scalar(1);
     m_s = Scalar(0);
@@ -93,7 +93,7 @@ bool JacobiRotation<Scalar>::makeJacobi(const RealScalar& x, const Scalar& y, co
   }
   else
   {
-    RealScalar tau = (x-z)/(RealScalar(2)*abs(y));
+    RealScalar tau = (x-z)/deno;
     RealScalar w = sqrt(numext::abs2(tau) + RealScalar(1));
     RealScalar t;
     if(tau>RealScalar(0))
@@ -123,7 +123,7 @@ bool JacobiRotation<Scalar>::makeJacobi(const RealScalar& x, const Scalar& y, co
   */
 template<typename Scalar>
 template<typename Derived>
-inline bool JacobiRotation<Scalar>::makeJacobi(const MatrixBase<Derived>& m, typename Derived::Index p, typename Derived::Index q)
+inline bool JacobiRotation<Scalar>::makeJacobi(const MatrixBase<Derived>& m, Index p, Index q)
 {
   return makeJacobi(numext::real(m.coeff(p,p)), m.coeff(p,q), numext::real(m.coeff(q,q)));
 }
@@ -132,7 +132,7 @@ inline bool JacobiRotation<Scalar>::makeJacobi(const MatrixBase<Derived>& m, typ
   * \f$ V = \left ( \begin{array}{c} p \\ q \end{array} \right )\f$ yields:
   * \f$ G^* V = \left ( \begin{array}{c} r \\ 0 \end{array} \right )\f$.
   *
-  * The value of \a z is returned if \a z is not null (the default is null).
+  * The value of \a r is returned if \a r is not null (the default is null).
   * Also note that G is built such that the cosine is always real.
   *
   * Example: \include Jacobi_makeGivens.cpp
@@ -145,9 +145,9 @@ inline bool JacobiRotation<Scalar>::makeJacobi(const MatrixBase<Derived>& m, typ
   * \sa MatrixBase::applyOnTheLeft(), MatrixBase::applyOnTheRight()
   */
 template<typename Scalar>
-void JacobiRotation<Scalar>::makeGivens(const Scalar& p, const Scalar& q, Scalar* z)
+void JacobiRotation<Scalar>::makeGivens(const Scalar& p, const Scalar& q, Scalar* r)
 {
-  makeGivens(p, q, z, typename internal::conditional<NumTraits<Scalar>::IsComplex, internal::true_type, internal::false_type>::type());
+  makeGivens(p, q, r, typename internal::conditional<NumTraits<Scalar>::IsComplex, internal::true_type, internal::false_type>::type());
 }
 
 
@@ -255,15 +255,15 @@ void JacobiRotation<Scalar>::makeGivens(const Scalar& p, const Scalar& q, Scalar
 *   Implementation of MatrixBase methods
 ****************************************************************************************/
 
+namespace internal {
 /** \jacobi_module
   * Applies the clock wise 2D rotation \a j to the set of 2D vectors of cordinates \a x and \a y:
   * \f$ \left ( \begin{array}{cc} x \\ y \end{array} \right )  =  J \left ( \begin{array}{cc} x \\ y \end{array} \right ) \f$
   *
   * \sa MatrixBase::applyOnTheLeft(), MatrixBase::applyOnTheRight()
   */
-namespace internal {
 template<typename VectorX, typename VectorY, typename OtherScalar>
-void apply_rotation_in_the_plane(VectorX& _x, VectorY& _y, const JacobiRotation<OtherScalar>& j);
+void apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x, DenseBase<VectorY>& xpr_y, const JacobiRotation<OtherScalar>& j);
 }
 
 /** \jacobi_module
@@ -297,133 +297,162 @@ inline void MatrixBase<Derived>::applyOnTheRight(Index p, Index q, const JacobiR
 }
 
 namespace internal {
-template<typename VectorX, typename VectorY, typename OtherScalar>
-void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(VectorX& _x, VectorY& _y, const JacobiRotation<OtherScalar>& j)
-{
-  typedef typename VectorX::Index Index;
-  typedef typename VectorX::Scalar Scalar;
-  enum { PacketSize = packet_traits<Scalar>::size };
-  typedef typename packet_traits<Scalar>::type Packet;
-  eigen_assert(_x.size() == _y.size());
-  Index size = _x.size();
-  Index incrx = _x.innerStride();
-  Index incry = _y.innerStride();
-
-  Scalar* EIGEN_RESTRICT x = &_x.coeffRef(0);
-  Scalar* EIGEN_RESTRICT y = &_y.coeffRef(0);
-  
-  OtherScalar c = j.c();
-  OtherScalar s = j.s();
-  if (c==OtherScalar(1) && s==OtherScalar(0))
-    return;
 
-  /*** dynamic-size vectorized paths ***/
+template<typename Scalar, typename OtherScalar,
+         int SizeAtCompileTime, int MinAlignment, bool Vectorizable>
+struct apply_rotation_in_the_plane_selector
+{
+  static inline void run(Scalar *x, Index incrx, Scalar *y, Index incry, Index size, OtherScalar c, OtherScalar s)
+  {
+    for(Index i=0; i<size; ++i)
+    {
+      Scalar xi = *x;
+      Scalar yi = *y;
+      *x =  c * xi + numext::conj(s) * yi;
+      *y = -s * xi + numext::conj(c) * yi;
+      x += incrx;
+      y += incry;
+    }
+  }
+};
 
-  if(VectorX::SizeAtCompileTime == Dynamic &&
-    (VectorX::Flags & VectorY::Flags & PacketAccessBit) &&
-    ((incrx==1 && incry==1) || PacketSize == 1))
+template<typename Scalar, typename OtherScalar,
+         int SizeAtCompileTime, int MinAlignment>
+struct apply_rotation_in_the_plane_selector<Scalar,OtherScalar,SizeAtCompileTime,MinAlignment,true /* vectorizable */>
+{
+  static inline void run(Scalar *x, Index incrx, Scalar *y, Index incry, Index size, OtherScalar c, OtherScalar s)
   {
-    // both vectors are sequentially stored in memory => vectorization
-    enum { Peeling = 2 };
+    enum {
+      PacketSize = packet_traits<Scalar>::size,
+      OtherPacketSize = packet_traits<OtherScalar>::size
+    };
+    typedef typename packet_traits<Scalar>::type Packet;
+    typedef typename packet_traits<OtherScalar>::type OtherPacket;
+
+    /*** dynamic-size vectorized paths ***/
+    if(SizeAtCompileTime == Dynamic && ((incrx==1 && incry==1) || PacketSize == 1))
+    {
+      // both vectors are sequentially stored in memory => vectorization
+      enum { Peeling = 2 };
 
-    Index alignedStart = internal::first_aligned(y, size);
-    Index alignedEnd = alignedStart + ((size-alignedStart)/PacketSize)*PacketSize;
+      Index alignedStart = internal::first_default_aligned(y, size);
+      Index alignedEnd = alignedStart + ((size-alignedStart)/PacketSize)*PacketSize;
 
-    const Packet pc = pset1<Packet>(c);
-    const Packet ps = pset1<Packet>(s);
-    conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex,false> pcj;
+      const OtherPacket pc = pset1<OtherPacket>(c);
+      const OtherPacket ps = pset1<OtherPacket>(s);
+      conj_helper<OtherPacket,Packet,NumTraits<OtherScalar>::IsComplex,false> pcj;
+      conj_helper<OtherPacket,Packet,false,false> pm;
 
-    for(Index i=0; i<alignedStart; ++i)
-    {
-      Scalar xi = x[i];
-      Scalar yi = y[i];
-      x[i] =  c * xi + numext::conj(s) * yi;
-      y[i] = -s * xi + numext::conj(c) * yi;
-    }
+      for(Index i=0; i<alignedStart; ++i)
+      {
+        Scalar xi = x[i];
+        Scalar yi = y[i];
+        x[i] =  c * xi + numext::conj(s) * yi;
+        y[i] = -s * xi + numext::conj(c) * yi;
+      }
 
-    Scalar* EIGEN_RESTRICT px = x + alignedStart;
-    Scalar* EIGEN_RESTRICT py = y + alignedStart;
+      Scalar* EIGEN_RESTRICT px = x + alignedStart;
+      Scalar* EIGEN_RESTRICT py = y + alignedStart;
 
-    if(internal::first_aligned(x, size)==alignedStart)
-    {
-      for(Index i=alignedStart; i<alignedEnd; i+=PacketSize)
+      if(internal::first_default_aligned(x, size)==alignedStart)
       {
-        Packet xi = pload<Packet>(px);
-        Packet yi = pload<Packet>(py);
-        pstore(px, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-        pstore(py, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
-        px += PacketSize;
-        py += PacketSize;
+        for(Index i=alignedStart; i<alignedEnd; i+=PacketSize)
+        {
+          Packet xi = pload<Packet>(px);
+          Packet yi = pload<Packet>(py);
+          pstore(px, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+          pstore(py, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
+          px += PacketSize;
+          py += PacketSize;
+        }
       }
-    }
-    else
-    {
-      Index peelingEnd = alignedStart + ((size-alignedStart)/(Peeling*PacketSize))*(Peeling*PacketSize);
-      for(Index i=alignedStart; i<peelingEnd; i+=Peeling*PacketSize)
+      else
       {
-        Packet xi   = ploadu<Packet>(px);
-        Packet xi1  = ploadu<Packet>(px+PacketSize);
-        Packet yi   = pload <Packet>(py);
-        Packet yi1  = pload <Packet>(py+PacketSize);
-        pstoreu(px, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-        pstoreu(px+PacketSize, padd(pmul(pc,xi1),pcj.pmul(ps,yi1)));
-        pstore (py, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
-        pstore (py+PacketSize, psub(pcj.pmul(pc,yi1),pmul(ps,xi1)));
-        px += Peeling*PacketSize;
-        py += Peeling*PacketSize;
+        Index peelingEnd = alignedStart + ((size-alignedStart)/(Peeling*PacketSize))*(Peeling*PacketSize);
+        for(Index i=alignedStart; i<peelingEnd; i+=Peeling*PacketSize)
+        {
+          Packet xi   = ploadu<Packet>(px);
+          Packet xi1  = ploadu<Packet>(px+PacketSize);
+          Packet yi   = pload <Packet>(py);
+          Packet yi1  = pload <Packet>(py+PacketSize);
+          pstoreu(px, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+          pstoreu(px+PacketSize, padd(pm.pmul(pc,xi1),pcj.pmul(ps,yi1)));
+          pstore (py, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
+          pstore (py+PacketSize, psub(pcj.pmul(pc,yi1),pm.pmul(ps,xi1)));
+          px += Peeling*PacketSize;
+          py += Peeling*PacketSize;
+        }
+        if(alignedEnd!=peelingEnd)
+        {
+          Packet xi = ploadu<Packet>(x+peelingEnd);
+          Packet yi = pload <Packet>(y+peelingEnd);
+          pstoreu(x+peelingEnd, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+          pstore (y+peelingEnd, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
+        }
       }
-      if(alignedEnd!=peelingEnd)
+
+      for(Index i=alignedEnd; i<size; ++i)
       {
-        Packet xi = ploadu<Packet>(x+peelingEnd);
-        Packet yi = pload <Packet>(y+peelingEnd);
-        pstoreu(x+peelingEnd, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-        pstore (y+peelingEnd, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
+        Scalar xi = x[i];
+        Scalar yi = y[i];
+        x[i] =  c * xi + numext::conj(s) * yi;
+        y[i] = -s * xi + numext::conj(c) * yi;
       }
     }
 
-    for(Index i=alignedEnd; i<size; ++i)
+    /*** fixed-size vectorized path ***/
+    else if(SizeAtCompileTime != Dynamic && MinAlignment>0) // FIXME should be compared to the required alignment
     {
-      Scalar xi = x[i];
-      Scalar yi = y[i];
-      x[i] =  c * xi + numext::conj(s) * yi;
-      y[i] = -s * xi + numext::conj(c) * yi;
+      const OtherPacket pc = pset1<OtherPacket>(c);
+      const OtherPacket ps = pset1<OtherPacket>(s);
+      conj_helper<OtherPacket,Packet,NumTraits<OtherPacket>::IsComplex,false> pcj;
+      conj_helper<OtherPacket,Packet,false,false> pm;
+      Scalar* EIGEN_RESTRICT px = x;
+      Scalar* EIGEN_RESTRICT py = y;
+      for(Index i=0; i<size; i+=PacketSize)
+      {
+        Packet xi = pload<Packet>(px);
+        Packet yi = pload<Packet>(py);
+        pstore(px, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+        pstore(py, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
+        px += PacketSize;
+        py += PacketSize;
+      }
     }
-  }
 
-  /*** fixed-size vectorized path ***/
-  else if(VectorX::SizeAtCompileTime != Dynamic &&
-          (VectorX::Flags & VectorY::Flags & PacketAccessBit) &&
-          (VectorX::Flags & VectorY::Flags & AlignedBit))
-  {
-    const Packet pc = pset1<Packet>(c);
-    const Packet ps = pset1<Packet>(s);
-    conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex,false> pcj;
-    Scalar* EIGEN_RESTRICT px = x;
-    Scalar* EIGEN_RESTRICT py = y;
-    for(Index i=0; i<size; i+=PacketSize)
+    /*** non-vectorized path ***/
+    else
     {
-      Packet xi = pload<Packet>(px);
-      Packet yi = pload<Packet>(py);
-      pstore(px, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-      pstore(py, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
-      px += PacketSize;
-      py += PacketSize;
+      apply_rotation_in_the_plane_selector<Scalar,OtherScalar,SizeAtCompileTime,MinAlignment,false>::run(x,incrx,y,incry,size,c,s);
     }
   }
+};
 
-  /*** non-vectorized path ***/
-  else
-  {
-    for(Index i=0; i<size; ++i)
-    {
-      Scalar xi = *x;
-      Scalar yi = *y;
-      *x =  c * xi + numext::conj(s) * yi;
-      *y = -s * xi + numext::conj(c) * yi;
-      x += incrx;
-      y += incry;
-    }
-  }
+template<typename VectorX, typename VectorY, typename OtherScalar>
+void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x, DenseBase<VectorY>& xpr_y, const JacobiRotation<OtherScalar>& j)
+{
+  typedef typename VectorX::Scalar Scalar;
+  const bool Vectorizable =    (VectorX::Flags & VectorY::Flags & PacketAccessBit)
+                            && (int(packet_traits<Scalar>::size) == int(packet_traits<OtherScalar>::size));
+
+  eigen_assert(xpr_x.size() == xpr_y.size());
+  Index size = xpr_x.size();
+  Index incrx = xpr_x.derived().innerStride();
+  Index incry = xpr_y.derived().innerStride();
+
+  Scalar* EIGEN_RESTRICT x = &xpr_x.derived().coeffRef(0);
+  Scalar* EIGEN_RESTRICT y = &xpr_y.derived().coeffRef(0);
+  
+  OtherScalar c = j.c();
+  OtherScalar s = j.s();
+  if (c==OtherScalar(1) && s==OtherScalar(0))
+    return;
+
+  apply_rotation_in_the_plane_selector<
+    Scalar,OtherScalar,
+    VectorX::SizeAtCompileTime,
+    EIGEN_PLAIN_ENUM_MIN(evaluator<VectorX>::Alignment, evaluator<VectorY>::Alignment),
+    Vectorizable>::run(x,incrx,y,incry,size,c,s);
 }
 
 } // end namespace internal
diff --git a/extern/Eigen3/Eigen/src/LU/Determinant.h b/extern/Eigen3/Eigen/src/LU/Determinant.h
index bb8e78a8a8a..d6a3c1e5a54 100644
--- a/extern/Eigen3/Eigen/src/LU/Determinant.h
+++ b/extern/Eigen3/Eigen/src/LU/Determinant.h
@@ -92,7 +92,7 @@ template<typename Derived>
 inline typename internal::traits<Derived>::Scalar MatrixBase<Derived>::determinant() const
 {
   eigen_assert(rows() == cols());
-  typedef typename internal::nested<Derived,Base::RowsAtCompileTime>::type Nested;
+  typedef typename internal::nested_eval<Derived,Base::RowsAtCompileTime>::type Nested;
   return internal::determinant_impl<typename internal::remove_all<Nested>::type>::run(derived());
 }
 
diff --git a/extern/Eigen3/Eigen/src/LU/FullPivLU.h b/extern/Eigen3/Eigen/src/LU/FullPivLU.h
index 26bc714475c..03b6af70613 100644
--- a/extern/Eigen3/Eigen/src/LU/FullPivLU.h
+++ b/extern/Eigen3/Eigen/src/LU/FullPivLU.h
@@ -10,7 +10,18 @@
 #ifndef EIGEN_LU_H
 #define EIGEN_LU_H
 
-namespace Eigen { 
+namespace Eigen {
+
+namespace internal {
+template<typename _MatrixType> struct traits<FullPivLU<_MatrixType> >
+ : traits<_MatrixType>
+{
+  typedef MatrixXpr XprKind;
+  typedef SolverStorage StorageKind;
+  enum { Flags = 0 };
+};
+
+} // end namespace internal
 
 /** \ingroup LU_Module
   *
@@ -18,7 +29,7 @@ namespace Eigen {
   *
   * \brief LU decomposition of a matrix with complete pivoting, and related features
   *
-  * \param MatrixType the type of the matrix of which we are computing the LU decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the LU decomposition
   *
   * This class represents a LU decomposition of any matrix, with complete pivoting: the matrix A is
   * decomposed as \f$ A = P^{-1} L U Q^{-1} \f$ where L is unit-lower-triangular, U is
@@ -41,27 +52,28 @@ namespace Eigen {
   * \include class_FullPivLU.cpp
   * Output: \verbinclude class_FullPivLU.out
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
   * \sa MatrixBase::fullPivLu(), MatrixBase::determinant(), MatrixBase::inverse()
   */
 template<typename _MatrixType> class FullPivLU
+  : public SolverBase<FullPivLU<_MatrixType> >
 {
   public:
     typedef _MatrixType MatrixType;
+    typedef SolverBase<FullPivLU> Base;
+
+    EIGEN_GENERIC_PUBLIC_INTERFACE(FullPivLU)
+    // FIXME StorageIndex defined in EIGEN_GENERIC_PUBLIC_INTERFACE should be int
     enum {
-      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
-      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
-    typedef typename MatrixType::Scalar Scalar;
-    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename internal::traits<MatrixType>::StorageKind StorageKind;
-    typedef typename MatrixType::Index Index;
-    typedef typename internal::plain_row_type<MatrixType, Index>::type IntRowVectorType;
-    typedef typename internal::plain_col_type<MatrixType, Index>::type IntColVectorType;
+    typedef typename internal::plain_row_type<MatrixType, StorageIndex>::type IntRowVectorType;
+    typedef typename internal::plain_col_type<MatrixType, StorageIndex>::type IntColVectorType;
     typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime> PermutationQType;
     typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationPType;
+    typedef typename MatrixType::PlainObject PlainObject;
 
     /**
       * \brief Default Constructor.
@@ -84,7 +96,17 @@ template<typename _MatrixType> class FullPivLU
       * \param matrix the matrix of which to compute the LU decomposition.
       *               It is required to be nonzero.
       */
-    FullPivLU(const MatrixType& matrix);
+    template<typename InputType>
+    explicit FullPivLU(const EigenBase<InputType>& matrix);
+
+    /** \brief Constructs a LU factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+      *
+      * \sa FullPivLU(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit FullPivLU(EigenBase<InputType>& matrix);
 
     /** Computes the LU decomposition of the given matrix.
       *
@@ -93,7 +115,12 @@ template<typename _MatrixType> class FullPivLU
       *
       * \returns a reference to *this
       */
-    FullPivLU& compute(const MatrixType& matrix);
+    template<typename InputType>
+    FullPivLU& compute(const EigenBase<InputType>& matrix) {
+      m_lu = matrix.derived();
+      computeInPlace();
+      return *this;
+    }
 
     /** \returns the LU decomposition matrix: the upper-triangular part is U, the
       * unit-lower-triangular part is L (at least for square matrices; in the non-square
@@ -129,7 +156,7 @@ template<typename _MatrixType> class FullPivLU
       *
       * \sa permutationQ()
       */
-    inline const PermutationPType& permutationP() const
+    EIGEN_DEVICE_FUNC inline const PermutationPType& permutationP() const
     {
       eigen_assert(m_isInitialized && "LU is not initialized.");
       return m_p;
@@ -166,7 +193,7 @@ template<typename _MatrixType> class FullPivLU
     }
 
     /** \returns the image of the matrix, also called its column-space. The columns of the returned matrix
-      * will form a basis of the kernel.
+      * will form a basis of the image (column-space).
       *
       * \param originalMatrix the original matrix, of which *this is the LU decomposition.
       *                       The reason why it is needed to pass it here, is that this allows
@@ -210,12 +237,22 @@ template<typename _MatrixType> class FullPivLU
       *
       * \sa TriangularView::solve(), kernel(), inverse()
       */
+    // FIXME this is a copy-paste of the base-class member to add the isInitialized assertion.
     template<typename Rhs>
-    inline const internal::solve_retval<FullPivLU, Rhs>
+    inline const Solve<FullPivLU, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "LU is not initialized.");
-      return internal::solve_retval<FullPivLU, Rhs>(*this, b.derived());
+      return Solve<FullPivLU, Rhs>(*this, b.derived());
+    }
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of which \c *this is
+        the LU decomposition.
+      */
+    inline RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
     }
 
     /** \returns the determinant of the matrix of which
@@ -360,33 +397,46 @@ template<typename _MatrixType> class FullPivLU
       *
       * \sa MatrixBase::inverse()
       */
-    inline const internal::solve_retval<FullPivLU,typename MatrixType::IdentityReturnType> inverse() const
+    inline const Inverse<FullPivLU> inverse() const
     {
       eigen_assert(m_isInitialized && "LU is not initialized.");
       eigen_assert(m_lu.rows() == m_lu.cols() && "You can't take the inverse of a non-square matrix!");
-      return internal::solve_retval<FullPivLU,typename MatrixType::IdentityReturnType>
-               (*this, MatrixType::Identity(m_lu.rows(), m_lu.cols()));
+      return Inverse<FullPivLU>(*this);
     }
 
     MatrixType reconstructedMatrix() const;
 
-    inline Index rows() const { return m_lu.rows(); }
-    inline Index cols() const { return m_lu.cols(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_lu.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_lu.cols(); }
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+
+    template<bool Conjugate, typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const;
+    #endif
 
   protected:
-    
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
-    
+
+    void computeInPlace();
+
     MatrixType m_lu;
     PermutationPType m_p;
     PermutationQType m_q;
     IntColVectorType m_rowsTranspositions;
     IntRowVectorType m_colsTranspositions;
-    Index m_det_pq, m_nonzero_pivots;
+    Index m_nonzero_pivots;
+    RealScalar m_l1_norm;
     RealScalar m_maxpivot, m_prescribedThreshold;
+    signed char m_det_pq;
     bool m_isInitialized, m_usePrescribedThreshold;
 };
 
@@ -409,7 +459,8 @@ FullPivLU<MatrixType>::FullPivLU(Index rows, Index cols)
 }
 
 template<typename MatrixType>
-FullPivLU<MatrixType>::FullPivLU(const MatrixType& matrix)
+template<typename InputType>
+FullPivLU<MatrixType>::FullPivLU(const EigenBase<InputType>& matrix)
   : m_lu(matrix.rows(), matrix.cols()),
     m_p(matrix.rows()),
     m_q(matrix.cols()),
@@ -418,28 +469,41 @@ FullPivLU<MatrixType>::FullPivLU(const MatrixType& matrix)
     m_isInitialized(false),
     m_usePrescribedThreshold(false)
 {
-  compute(matrix);
+  compute(matrix.derived());
+}
+
+template<typename MatrixType>
+template<typename InputType>
+FullPivLU<MatrixType>::FullPivLU(EigenBase<InputType>& matrix)
+  : m_lu(matrix.derived()),
+    m_p(matrix.rows()),
+    m_q(matrix.cols()),
+    m_rowsTranspositions(matrix.rows()),
+    m_colsTranspositions(matrix.cols()),
+    m_isInitialized(false),
+    m_usePrescribedThreshold(false)
+{
+  computeInPlace();
 }
 
 template<typename MatrixType>
-FullPivLU<MatrixType>& FullPivLU<MatrixType>::compute(const MatrixType& matrix)
+void FullPivLU<MatrixType>::computeInPlace()
 {
   check_template_parameters();
-  
+
   // the permutations are stored as int indices, so just to be sure:
-  eigen_assert(matrix.rows()<=NumTraits<int>::highest() && matrix.cols()<=NumTraits<int>::highest());
-  
-  m_isInitialized = true;
-  m_lu = matrix;
+  eigen_assert(m_lu.rows()<=NumTraits<int>::highest() && m_lu.cols()<=NumTraits<int>::highest());
+
+  m_l1_norm = m_lu.cwiseAbs().colwise().sum().maxCoeff();
 
-  const Index size = matrix.diagonalSize();
-  const Index rows = matrix.rows();
-  const Index cols = matrix.cols();
+  const Index size = m_lu.diagonalSize();
+  const Index rows = m_lu.rows();
+  const Index cols = m_lu.cols();
 
   // will store the transpositions, before we accumulate them at the end.
   // can't accumulate on-the-fly because that will be done in reverse order for the rows.
-  m_rowsTranspositions.resize(matrix.rows());
-  m_colsTranspositions.resize(matrix.cols());
+  m_rowsTranspositions.resize(m_lu.rows());
+  m_colsTranspositions.resize(m_lu.cols());
   Index number_of_transpositions = 0; // number of NONTRIVIAL transpositions, i.e. m_rowsTranspositions[i]!=i
 
   m_nonzero_pivots = size; // the generic case is that in which all pivots are nonzero (invertible case)
@@ -451,14 +515,16 @@ FullPivLU<MatrixType>& FullPivLU<MatrixType>::compute(const MatrixType& matrix)
 
     // biggest coefficient in the remaining bottom-right corner (starting at row k, col k)
     Index row_of_biggest_in_corner, col_of_biggest_in_corner;
-    RealScalar biggest_in_corner;
+    typedef internal::scalar_score_coeff_op<Scalar> Scoring;
+    typedef typename Scoring::result_type Score;
+    Score biggest_in_corner;
     biggest_in_corner = m_lu.bottomRightCorner(rows-k, cols-k)
-                        .cwiseAbs()
+                        .unaryExpr(Scoring())
                         .maxCoeff(&row_of_biggest_in_corner, &col_of_biggest_in_corner);
     row_of_biggest_in_corner += k; // correct the values! since they were computed in the corner,
     col_of_biggest_in_corner += k; // need to add k to them.
 
-    if(biggest_in_corner==RealScalar(0))
+    if(biggest_in_corner==Score(0))
     {
       // before exiting, make sure to initialize the still uninitialized transpositions
       // in a sane state without destroying what we already have.
@@ -471,7 +537,8 @@ FullPivLU<MatrixType>& FullPivLU<MatrixType>::compute(const MatrixType& matrix)
       break;
     }
 
-    if(biggest_in_corner > m_maxpivot) m_maxpivot = biggest_in_corner;
+    RealScalar abs_pivot = internal::abs_knowing_score<Scalar>()(m_lu(row_of_biggest_in_corner, col_of_biggest_in_corner), biggest_in_corner);
+    if(abs_pivot > m_maxpivot) m_maxpivot = abs_pivot;
 
     // Now that we've found the pivot, we need to apply the row/col swaps to
     // bring it to the location (k,k).
@@ -508,7 +575,8 @@ FullPivLU<MatrixType>& FullPivLU<MatrixType>::compute(const MatrixType& matrix)
     m_q.applyTranspositionOnTheRight(k, m_colsTranspositions.coeff(k));
 
   m_det_pq = (number_of_transpositions%2) ? -1 : 1;
-  return *this;
+
+  m_isInitialized = true;
 }
 
 template<typename MatrixType>
@@ -671,64 +739,136 @@ struct image_retval<FullPivLU<_MatrixType> >
 
 /***** Implementation of solve() *****************************************************/
 
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<FullPivLU<_MatrixType>, Rhs>
-  : solve_retval_base<FullPivLU<_MatrixType>, Rhs>
+} // end namespace internal
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType>
+template<typename RhsType, typename DstType>
+void FullPivLU<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
 {
-  EIGEN_MAKE_SOLVE_HELPERS(FullPivLU<_MatrixType>,Rhs)
+  /* The decomposition PAQ = LU can be rewritten as A = P^{-1} L U Q^{-1}.
+  * So we proceed as follows:
+  * Step 1: compute c = P * rhs.
+  * Step 2: replace c by the solution x to Lx = c. Exists because L is invertible.
+  * Step 3: replace c by the solution x to Ux = c. May or may not exist.
+  * Step 4: result = Q * c;
+  */
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  const Index rows = this->rows(),
+              cols = this->cols(),
+              nonzero_pivots = this->rank();
+  eigen_assert(rhs.rows() == rows);
+  const Index smalldim = (std::min)(rows, cols);
+
+  if(nonzero_pivots == 0)
   {
-    /* The decomposition PAQ = LU can be rewritten as A = P^{-1} L U Q^{-1}.
-     * So we proceed as follows:
-     * Step 1: compute c = P * rhs.
-     * Step 2: replace c by the solution x to Lx = c. Exists because L is invertible.
-     * Step 3: replace c by the solution x to Ux = c. May or may not exist.
-     * Step 4: result = Q * c;
-     */
-
-    const Index rows = dec().rows(), cols = dec().cols(),
-              nonzero_pivots = dec().nonzeroPivots();
-    eigen_assert(rhs().rows() == rows);
-    const Index smalldim = (std::min)(rows, cols);
-
-    if(nonzero_pivots == 0)
-    {
-      dst.setZero();
-      return;
-    }
+    dst.setZero();
+    return;
+  }
+
+  typename RhsType::PlainObject c(rhs.rows(), rhs.cols());
+
+  // Step 1
+  c = permutationP() * rhs;
 
-    typename Rhs::PlainObject c(rhs().rows(), rhs().cols());
+  // Step 2
+  m_lu.topLeftCorner(smalldim,smalldim)
+      .template triangularView<UnitLower>()
+      .solveInPlace(c.topRows(smalldim));
+  if(rows>cols)
+    c.bottomRows(rows-cols) -= m_lu.bottomRows(rows-cols) * c.topRows(cols);
 
-    // Step 1
-    c = dec().permutationP() * rhs();
+  // Step 3
+  m_lu.topLeftCorner(nonzero_pivots, nonzero_pivots)
+      .template triangularView<Upper>()
+      .solveInPlace(c.topRows(nonzero_pivots));
 
+  // Step 4
+  for(Index i = 0; i < nonzero_pivots; ++i)
+    dst.row(permutationQ().indices().coeff(i)) = c.row(i);
+  for(Index i = nonzero_pivots; i < m_lu.cols(); ++i)
+    dst.row(permutationQ().indices().coeff(i)).setZero();
+}
+
+template<typename _MatrixType>
+template<bool Conjugate, typename RhsType, typename DstType>
+void FullPivLU<_MatrixType>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const
+{
+  /* The decomposition PAQ = LU can be rewritten as A = P^{-1} L U Q^{-1},
+   * and since permutations are real and unitary, we can write this
+   * as   A^T = Q U^T L^T P,
+   * So we proceed as follows:
+   * Step 1: compute c = Q^T rhs.
+   * Step 2: replace c by the solution x to U^T x = c. May or may not exist.
+   * Step 3: replace c by the solution x to L^T x = c.
+   * Step 4: result = P^T c.
+   * If Conjugate is true, replace "^T" by "^*" above.
+   */
+
+  const Index rows = this->rows(), cols = this->cols(),
+    nonzero_pivots = this->rank();
+   eigen_assert(rhs.rows() == cols);
+  const Index smalldim = (std::min)(rows, cols);
+
+  if(nonzero_pivots == 0)
+  {
+    dst.setZero();
+    return;
+  }
+
+  typename RhsType::PlainObject c(rhs.rows(), rhs.cols());
+
+  // Step 1
+  c = permutationQ().inverse() * rhs;
+
+  if (Conjugate) {
     // Step 2
-    dec().matrixLU()
-        .topLeftCorner(smalldim,smalldim)
+    m_lu.topLeftCorner(nonzero_pivots, nonzero_pivots)
+        .template triangularView<Upper>()
+        .adjoint()
+        .solveInPlace(c.topRows(nonzero_pivots));
+    // Step 3
+    m_lu.topLeftCorner(smalldim, smalldim)
         .template triangularView<UnitLower>()
+        .adjoint()
         .solveInPlace(c.topRows(smalldim));
-    if(rows>cols)
-    {
-      c.bottomRows(rows-cols)
-        -= dec().matrixLU().bottomRows(rows-cols)
-         * c.topRows(cols);
-    }
-
-    // Step 3
-    dec().matrixLU()
-        .topLeftCorner(nonzero_pivots, nonzero_pivots)
+  } else {
+    // Step 2
+    m_lu.topLeftCorner(nonzero_pivots, nonzero_pivots)
         .template triangularView<Upper>()
+        .transpose()
         .solveInPlace(c.topRows(nonzero_pivots));
+    // Step 3
+    m_lu.topLeftCorner(smalldim, smalldim)
+        .template triangularView<UnitLower>()
+        .transpose()
+        .solveInPlace(c.topRows(smalldim));
+  }
+
+  // Step 4
+  PermutationPType invp = permutationP().inverse().eval();
+  for(Index i = 0; i < smalldim; ++i)
+    dst.row(invp.indices().coeff(i)) = c.row(i);
+  for(Index i = smalldim; i < rows; ++i)
+    dst.row(invp.indices().coeff(i)).setZero();
+}
+
+#endif
+
+namespace internal {
+
 
-    // Step 4
-    for(Index i = 0; i < nonzero_pivots; ++i)
-      dst.row(dec().permutationQ().indices().coeff(i)) = c.row(i);
-    for(Index i = nonzero_pivots; i < dec().matrixLU().cols(); ++i)
-      dst.row(dec().permutationQ().indices().coeff(i)).setZero();
+/***** Implementation of inverse() *****************************************************/
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<FullPivLU<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename FullPivLU<MatrixType>::Scalar>, Dense2Dense>
+{
+  typedef FullPivLU<MatrixType> LuType;
+  typedef Inverse<LuType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename MatrixType::Scalar> &)
+  {
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.cols()));
   }
 };
-
 } // end namespace internal
 
 /******* MatrixBase methods *****************************************************************/
diff --git a/extern/Eigen3/Eigen/src/LU/InverseImpl.h b/extern/Eigen3/Eigen/src/LU/InverseImpl.h
new file mode 100644
index 00000000000..f49f2336004
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/LU/InverseImpl.h
@@ -0,0 +1,415 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_INVERSE_IMPL_H
+#define EIGEN_INVERSE_IMPL_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/**********************************
+*** General case implementation ***
+**********************************/
+
+template<typename MatrixType, typename ResultType, int Size = MatrixType::RowsAtCompileTime>
+struct compute_inverse
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(const MatrixType& matrix, ResultType& result)
+  {
+    result = matrix.partialPivLu().inverse();
+  }
+};
+
+template<typename MatrixType, typename ResultType, int Size = MatrixType::RowsAtCompileTime>
+struct compute_inverse_and_det_with_check { /* nothing! general case not supported. */ };
+
+/****************************
+*** Size 1 implementation ***
+****************************/
+
+template<typename MatrixType, typename ResultType>
+struct compute_inverse<MatrixType, ResultType, 1>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(const MatrixType& matrix, ResultType& result)
+  {
+    typedef typename MatrixType::Scalar Scalar;
+    internal::evaluator<MatrixType> matrixEval(matrix);
+    result.coeffRef(0,0) = Scalar(1) / matrixEval.coeff(0,0);
+  }
+};
+
+template<typename MatrixType, typename ResultType>
+struct compute_inverse_and_det_with_check<MatrixType, ResultType, 1>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(
+    const MatrixType& matrix,
+    const typename MatrixType::RealScalar& absDeterminantThreshold,
+    ResultType& result,
+    typename ResultType::Scalar& determinant,
+    bool& invertible
+  )
+  {
+    using std::abs;
+    determinant = matrix.coeff(0,0);
+    invertible = abs(determinant) > absDeterminantThreshold;
+    if(invertible) result.coeffRef(0,0) = typename ResultType::Scalar(1) / determinant;
+  }
+};
+
+/****************************
+*** Size 2 implementation ***
+****************************/
+
+template<typename MatrixType, typename ResultType>
+EIGEN_DEVICE_FUNC 
+inline void compute_inverse_size2_helper(
+    const MatrixType& matrix, const typename ResultType::Scalar& invdet,
+    ResultType& result)
+{
+  result.coeffRef(0,0) =  matrix.coeff(1,1) * invdet;
+  result.coeffRef(1,0) = -matrix.coeff(1,0) * invdet;
+  result.coeffRef(0,1) = -matrix.coeff(0,1) * invdet;
+  result.coeffRef(1,1) =  matrix.coeff(0,0) * invdet;
+}
+
+template<typename MatrixType, typename ResultType>
+struct compute_inverse<MatrixType, ResultType, 2>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(const MatrixType& matrix, ResultType& result)
+  {
+    typedef typename ResultType::Scalar Scalar;
+    const Scalar invdet = typename MatrixType::Scalar(1) / matrix.determinant();
+    compute_inverse_size2_helper(matrix, invdet, result);
+  }
+};
+
+template<typename MatrixType, typename ResultType>
+struct compute_inverse_and_det_with_check<MatrixType, ResultType, 2>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(
+    const MatrixType& matrix,
+    const typename MatrixType::RealScalar& absDeterminantThreshold,
+    ResultType& inverse,
+    typename ResultType::Scalar& determinant,
+    bool& invertible
+  )
+  {
+    using std::abs;
+    typedef typename ResultType::Scalar Scalar;
+    determinant = matrix.determinant();
+    invertible = abs(determinant) > absDeterminantThreshold;
+    if(!invertible) return;
+    const Scalar invdet = Scalar(1) / determinant;
+    compute_inverse_size2_helper(matrix, invdet, inverse);
+  }
+};
+
+/****************************
+*** Size 3 implementation ***
+****************************/
+
+template<typename MatrixType, int i, int j>
+EIGEN_DEVICE_FUNC 
+inline typename MatrixType::Scalar cofactor_3x3(const MatrixType& m)
+{
+  enum {
+    i1 = (i+1) % 3,
+    i2 = (i+2) % 3,
+    j1 = (j+1) % 3,
+    j2 = (j+2) % 3
+  };
+  return m.coeff(i1, j1) * m.coeff(i2, j2)
+       - m.coeff(i1, j2) * m.coeff(i2, j1);
+}
+
+template<typename MatrixType, typename ResultType>
+EIGEN_DEVICE_FUNC
+inline void compute_inverse_size3_helper(
+    const MatrixType& matrix,
+    const typename ResultType::Scalar& invdet,
+    const Matrix<typename ResultType::Scalar,3,1>& cofactors_col0,
+    ResultType& result)
+{
+  result.row(0) = cofactors_col0 * invdet;
+  result.coeffRef(1,0) =  cofactor_3x3<MatrixType,0,1>(matrix) * invdet;
+  result.coeffRef(1,1) =  cofactor_3x3<MatrixType,1,1>(matrix) * invdet;
+  result.coeffRef(1,2) =  cofactor_3x3<MatrixType,2,1>(matrix) * invdet;
+  result.coeffRef(2,0) =  cofactor_3x3<MatrixType,0,2>(matrix) * invdet;
+  result.coeffRef(2,1) =  cofactor_3x3<MatrixType,1,2>(matrix) * invdet;
+  result.coeffRef(2,2) =  cofactor_3x3<MatrixType,2,2>(matrix) * invdet;
+}
+
+template<typename MatrixType, typename ResultType>
+struct compute_inverse<MatrixType, ResultType, 3>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(const MatrixType& matrix, ResultType& result)
+  {
+    typedef typename ResultType::Scalar Scalar;
+    Matrix<typename MatrixType::Scalar,3,1> cofactors_col0;
+    cofactors_col0.coeffRef(0) =  cofactor_3x3<MatrixType,0,0>(matrix);
+    cofactors_col0.coeffRef(1) =  cofactor_3x3<MatrixType,1,0>(matrix);
+    cofactors_col0.coeffRef(2) =  cofactor_3x3<MatrixType,2,0>(matrix);
+    const Scalar det = (cofactors_col0.cwiseProduct(matrix.col(0))).sum();
+    const Scalar invdet = Scalar(1) / det;
+    compute_inverse_size3_helper(matrix, invdet, cofactors_col0, result);
+  }
+};
+
+template<typename MatrixType, typename ResultType>
+struct compute_inverse_and_det_with_check<MatrixType, ResultType, 3>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(
+    const MatrixType& matrix,
+    const typename MatrixType::RealScalar& absDeterminantThreshold,
+    ResultType& inverse,
+    typename ResultType::Scalar& determinant,
+    bool& invertible
+  )
+  {
+    using std::abs;
+    typedef typename ResultType::Scalar Scalar;
+    Matrix<Scalar,3,1> cofactors_col0;
+    cofactors_col0.coeffRef(0) =  cofactor_3x3<MatrixType,0,0>(matrix);
+    cofactors_col0.coeffRef(1) =  cofactor_3x3<MatrixType,1,0>(matrix);
+    cofactors_col0.coeffRef(2) =  cofactor_3x3<MatrixType,2,0>(matrix);
+    determinant = (cofactors_col0.cwiseProduct(matrix.col(0))).sum();
+    invertible = abs(determinant) > absDeterminantThreshold;
+    if(!invertible) return;
+    const Scalar invdet = Scalar(1) / determinant;
+    compute_inverse_size3_helper(matrix, invdet, cofactors_col0, inverse);
+  }
+};
+
+/****************************
+*** Size 4 implementation ***
+****************************/
+
+template<typename Derived>
+EIGEN_DEVICE_FUNC 
+inline const typename Derived::Scalar general_det3_helper
+(const MatrixBase<Derived>& matrix, int i1, int i2, int i3, int j1, int j2, int j3)
+{
+  return matrix.coeff(i1,j1)
+         * (matrix.coeff(i2,j2) * matrix.coeff(i3,j3) - matrix.coeff(i2,j3) * matrix.coeff(i3,j2));
+}
+
+template<typename MatrixType, int i, int j>
+EIGEN_DEVICE_FUNC 
+inline typename MatrixType::Scalar cofactor_4x4(const MatrixType& matrix)
+{
+  enum {
+    i1 = (i+1) % 4,
+    i2 = (i+2) % 4,
+    i3 = (i+3) % 4,
+    j1 = (j+1) % 4,
+    j2 = (j+2) % 4,
+    j3 = (j+3) % 4
+  };
+  return general_det3_helper(matrix, i1, i2, i3, j1, j2, j3)
+       + general_det3_helper(matrix, i2, i3, i1, j1, j2, j3)
+       + general_det3_helper(matrix, i3, i1, i2, j1, j2, j3);
+}
+
+template<int Arch, typename Scalar, typename MatrixType, typename ResultType>
+struct compute_inverse_size4
+{
+  EIGEN_DEVICE_FUNC
+  static void run(const MatrixType& matrix, ResultType& result)
+  {
+    result.coeffRef(0,0) =  cofactor_4x4<MatrixType,0,0>(matrix);
+    result.coeffRef(1,0) = -cofactor_4x4<MatrixType,0,1>(matrix);
+    result.coeffRef(2,0) =  cofactor_4x4<MatrixType,0,2>(matrix);
+    result.coeffRef(3,0) = -cofactor_4x4<MatrixType,0,3>(matrix);
+    result.coeffRef(0,2) =  cofactor_4x4<MatrixType,2,0>(matrix);
+    result.coeffRef(1,2) = -cofactor_4x4<MatrixType,2,1>(matrix);
+    result.coeffRef(2,2) =  cofactor_4x4<MatrixType,2,2>(matrix);
+    result.coeffRef(3,2) = -cofactor_4x4<MatrixType,2,3>(matrix);
+    result.coeffRef(0,1) = -cofactor_4x4<MatrixType,1,0>(matrix);
+    result.coeffRef(1,1) =  cofactor_4x4<MatrixType,1,1>(matrix);
+    result.coeffRef(2,1) = -cofactor_4x4<MatrixType,1,2>(matrix);
+    result.coeffRef(3,1) =  cofactor_4x4<MatrixType,1,3>(matrix);
+    result.coeffRef(0,3) = -cofactor_4x4<MatrixType,3,0>(matrix);
+    result.coeffRef(1,3) =  cofactor_4x4<MatrixType,3,1>(matrix);
+    result.coeffRef(2,3) = -cofactor_4x4<MatrixType,3,2>(matrix);
+    result.coeffRef(3,3) =  cofactor_4x4<MatrixType,3,3>(matrix);
+    result /= (matrix.col(0).cwiseProduct(result.row(0).transpose())).sum();
+  }
+};
+
+template<typename MatrixType, typename ResultType>
+struct compute_inverse<MatrixType, ResultType, 4>
+ : compute_inverse_size4<Architecture::Target, typename MatrixType::Scalar,
+                            MatrixType, ResultType>
+{
+};
+
+template<typename MatrixType, typename ResultType>
+struct compute_inverse_and_det_with_check<MatrixType, ResultType, 4>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(
+    const MatrixType& matrix,
+    const typename MatrixType::RealScalar& absDeterminantThreshold,
+    ResultType& inverse,
+    typename ResultType::Scalar& determinant,
+    bool& invertible
+  )
+  {
+    using std::abs;
+    determinant = matrix.determinant();
+    invertible = abs(determinant) > absDeterminantThreshold;
+    if(invertible) compute_inverse<MatrixType, ResultType>::run(matrix, inverse);
+  }
+};
+
+/*************************
+*** MatrixBase methods ***
+*************************/
+
+} // end namespace internal
+
+namespace internal {
+
+// Specialization for "dense = dense_xpr.inverse()"
+template<typename DstXprType, typename XprType>
+struct Assignment<DstXprType, Inverse<XprType>, internal::assign_op<typename DstXprType::Scalar,typename XprType::Scalar>, Dense2Dense>
+{
+  typedef Inverse<XprType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename XprType::Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    
+    const int Size = EIGEN_PLAIN_ENUM_MIN(XprType::ColsAtCompileTime,DstXprType::ColsAtCompileTime);
+    EIGEN_ONLY_USED_FOR_DEBUG(Size);
+    eigen_assert(( (Size<=1) || (Size>4) || (extract_data(src.nestedExpression())!=extract_data(dst)))
+              && "Aliasing problem detected in inverse(), you need to do inverse().eval() here.");
+
+    typedef typename internal::nested_eval<XprType,XprType::ColsAtCompileTime>::type  ActualXprType;
+    typedef typename internal::remove_all<ActualXprType>::type                        ActualXprTypeCleanded;
+    
+    ActualXprType actual_xpr(src.nestedExpression());
+    
+    compute_inverse<ActualXprTypeCleanded, DstXprType>::run(actual_xpr, dst);
+  }
+};
+
+  
+} // end namespace internal
+
+/** \lu_module
+  *
+  * \returns the matrix inverse of this matrix.
+  *
+  * For small fixed sizes up to 4x4, this method uses cofactors.
+  * In the general case, this method uses class PartialPivLU.
+  *
+  * \note This matrix must be invertible, otherwise the result is undefined. If you need an
+  * invertibility check, do the following:
+  * \li for fixed sizes up to 4x4, use computeInverseAndDetWithCheck().
+  * \li for the general case, use class FullPivLU.
+  *
+  * Example: \include MatrixBase_inverse.cpp
+  * Output: \verbinclude MatrixBase_inverse.out
+  *
+  * \sa computeInverseAndDetWithCheck()
+  */
+template<typename Derived>
+inline const Inverse<Derived> MatrixBase<Derived>::inverse() const
+{
+  EIGEN_STATIC_ASSERT(!NumTraits<Scalar>::IsInteger,THIS_FUNCTION_IS_NOT_FOR_INTEGER_NUMERIC_TYPES)
+  eigen_assert(rows() == cols());
+  return Inverse<Derived>(derived());
+}
+
+/** \lu_module
+  *
+  * Computation of matrix inverse and determinant, with invertibility check.
+  *
+  * This is only for fixed-size square matrices of size up to 4x4.
+  *
+  * \param inverse Reference to the matrix in which to store the inverse.
+  * \param determinant Reference to the variable in which to store the determinant.
+  * \param invertible Reference to the bool variable in which to store whether the matrix is invertible.
+  * \param absDeterminantThreshold Optional parameter controlling the invertibility check.
+  *                                The matrix will be declared invertible if the absolute value of its
+  *                                determinant is greater than this threshold.
+  *
+  * Example: \include MatrixBase_computeInverseAndDetWithCheck.cpp
+  * Output: \verbinclude MatrixBase_computeInverseAndDetWithCheck.out
+  *
+  * \sa inverse(), computeInverseWithCheck()
+  */
+template<typename Derived>
+template<typename ResultType>
+inline void MatrixBase<Derived>::computeInverseAndDetWithCheck(
+    ResultType& inverse,
+    typename ResultType::Scalar& determinant,
+    bool& invertible,
+    const RealScalar& absDeterminantThreshold
+  ) const
+{
+  // i'd love to put some static assertions there, but SFINAE means that they have no effect...
+  eigen_assert(rows() == cols());
+  // for 2x2, it's worth giving a chance to avoid evaluating.
+  // for larger sizes, evaluating has negligible cost and limits code size.
+  typedef typename internal::conditional<
+    RowsAtCompileTime == 2,
+    typename internal::remove_all<typename internal::nested_eval<Derived, 2>::type>::type,
+    PlainObject
+  >::type MatrixType;
+  internal::compute_inverse_and_det_with_check<MatrixType, ResultType>::run
+    (derived(), absDeterminantThreshold, inverse, determinant, invertible);
+}
+
+/** \lu_module
+  *
+  * Computation of matrix inverse, with invertibility check.
+  *
+  * This is only for fixed-size square matrices of size up to 4x4.
+  *
+  * \param inverse Reference to the matrix in which to store the inverse.
+  * \param invertible Reference to the bool variable in which to store whether the matrix is invertible.
+  * \param absDeterminantThreshold Optional parameter controlling the invertibility check.
+  *                                The matrix will be declared invertible if the absolute value of its
+  *                                determinant is greater than this threshold.
+  *
+  * Example: \include MatrixBase_computeInverseWithCheck.cpp
+  * Output: \verbinclude MatrixBase_computeInverseWithCheck.out
+  *
+  * \sa inverse(), computeInverseAndDetWithCheck()
+  */
+template<typename Derived>
+template<typename ResultType>
+inline void MatrixBase<Derived>::computeInverseWithCheck(
+    ResultType& inverse,
+    bool& invertible,
+    const RealScalar& absDeterminantThreshold
+  ) const
+{
+  Scalar determinant;
+  // i'd love to put some static assertions there, but SFINAE means that they have no effect...
+  eigen_assert(rows() == cols());
+  computeInverseAndDetWithCheck(inverse,determinant,invertible,absDeterminantThreshold);
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_INVERSE_IMPL_H
diff --git a/extern/Eigen3/Eigen/src/LU/PartialPivLU.h b/extern/Eigen3/Eigen/src/LU/PartialPivLU.h
index 7d1db948c0a..d439618879a 100644
--- a/extern/Eigen3/Eigen/src/LU/PartialPivLU.h
+++ b/extern/Eigen3/Eigen/src/LU/PartialPivLU.h
@@ -11,7 +11,33 @@
 #ifndef EIGEN_PARTIALLU_H
 #define EIGEN_PARTIALLU_H
 
-namespace Eigen { 
+namespace Eigen {
+
+namespace internal {
+template<typename _MatrixType> struct traits<PartialPivLU<_MatrixType> >
+ : traits<_MatrixType>
+{
+  typedef MatrixXpr XprKind;
+  typedef SolverStorage StorageKind;
+  typedef traits<_MatrixType> BaseTraits;
+  enum {
+    Flags = BaseTraits::Flags & RowMajorBit,
+    CoeffReadCost = Dynamic
+  };
+};
+
+template<typename T,typename Derived>
+struct enable_if_ref;
+// {
+//   typedef Derived type;
+// };
+
+template<typename T,typename Derived>
+struct enable_if_ref<Ref<T>,Derived> {
+  typedef Derived type;
+};
+
+} // end namespace internal
 
 /** \ingroup LU_Module
   *
@@ -19,7 +45,7 @@ namespace Eigen {
   *
   * \brief LU decomposition of a matrix with partial pivoting, and related features
   *
-  * \param MatrixType the type of the matrix of which we are computing the LU decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the LU decomposition
   *
   * This class represents a LU decomposition of a \b square \b invertible matrix, with partial pivoting: the matrix A
   * is decomposed as A = PLU where L is unit-lower-triangular, U is upper-triangular, and P
@@ -42,34 +68,33 @@ namespace Eigen {
   *
   * The data of the LU decomposition can be directly accessed through the methods matrixLU(), permutationP().
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
   * \sa MatrixBase::partialPivLu(), MatrixBase::determinant(), MatrixBase::inverse(), MatrixBase::computeInverse(), class FullPivLU
   */
 template<typename _MatrixType> class PartialPivLU
+  : public SolverBase<PartialPivLU<_MatrixType> >
 {
   public:
 
     typedef _MatrixType MatrixType;
+    typedef SolverBase<PartialPivLU> Base;
+    EIGEN_GENERIC_PUBLIC_INTERFACE(PartialPivLU)
+    // FIXME StorageIndex defined in EIGEN_GENERIC_PUBLIC_INTERFACE should be int
     enum {
-      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
-      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
-    typedef typename MatrixType::Scalar Scalar;
-    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename internal::traits<MatrixType>::StorageKind StorageKind;
-    typedef typename MatrixType::Index Index;
     typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationType;
     typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
-
+    typedef typename MatrixType::PlainObject PlainObject;
 
     /**
-    * \brief Default Constructor.
-    *
-    * The default constructor is useful in cases in which the user intends to
-    * perform decompositions via PartialPivLU::compute(const MatrixType&).
-    */
+      * \brief Default Constructor.
+      *
+      * The default constructor is useful in cases in which the user intends to
+      * perform decompositions via PartialPivLU::compute(const MatrixType&).
+      */
     PartialPivLU();
 
     /** \brief Default Constructor with memory preallocation
@@ -78,7 +103,7 @@ template<typename _MatrixType> class PartialPivLU
       * according to the specified problem \a size.
       * \sa PartialPivLU()
       */
-    PartialPivLU(Index size);
+    explicit PartialPivLU(Index size);
 
     /** Constructor.
       *
@@ -87,9 +112,25 @@ template<typename _MatrixType> class PartialPivLU
       * \warning The matrix should have full rank (e.g. if it's square, it should be invertible).
       * If you need to deal with non-full rank, use class FullPivLU instead.
       */
-    PartialPivLU(const MatrixType& matrix);
+    template<typename InputType>
+    explicit PartialPivLU(const EigenBase<InputType>& matrix);
 
-    PartialPivLU& compute(const MatrixType& matrix);
+    /** Constructor for \link InplaceDecomposition inplace decomposition \endlink
+      *
+      * \param matrix the matrix of which to compute the LU decomposition.
+      *
+      * \warning The matrix should have full rank (e.g. if it's square, it should be invertible).
+      * If you need to deal with non-full rank, use class FullPivLU instead.
+      */
+    template<typename InputType>
+    explicit PartialPivLU(EigenBase<InputType>& matrix);
+
+    template<typename InputType>
+    PartialPivLU& compute(const EigenBase<InputType>& matrix) {
+      m_lu = matrix.derived();
+      compute();
+      return *this;
+    }
 
     /** \returns the LU decomposition matrix: the upper-triangular part is U, the
       * unit-lower-triangular part is L (at least for square matrices; in the non-square
@@ -128,12 +169,22 @@ template<typename _MatrixType> class PartialPivLU
       *
       * \sa TriangularView::solve(), inverse(), computeInverse()
       */
+    // FIXME this is a copy-paste of the base-class member to add the isInitialized assertion.
     template<typename Rhs>
-    inline const internal::solve_retval<PartialPivLU, Rhs>
+    inline const Solve<PartialPivLU, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
-      return internal::solve_retval<PartialPivLU, Rhs>(*this, b.derived());
+      return Solve<PartialPivLU, Rhs>(*this, b.derived());
+    }
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of which \c *this is
+        the LU decomposition.
+      */
+    inline RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
     }
 
     /** \returns the inverse of the matrix of which *this is the LU decomposition.
@@ -143,11 +194,10 @@ template<typename _MatrixType> class PartialPivLU
       *
       * \sa MatrixBase::inverse(), LU::inverse()
       */
-    inline const internal::solve_retval<PartialPivLU,typename MatrixType::IdentityReturnType> inverse() const
+    inline const Inverse<PartialPivLU> inverse() const
     {
       eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
-      return internal::solve_retval<PartialPivLU,typename MatrixType::IdentityReturnType>
-               (*this, MatrixType::Identity(m_lu.rows(), m_lu.cols()));
+      return Inverse<PartialPivLU>(*this);
     }
 
     /** \returns the determinant of the matrix of which
@@ -163,24 +213,78 @@ template<typename _MatrixType> class PartialPivLU
       *
       * \sa MatrixBase::determinant()
       */
-    typename internal::traits<MatrixType>::Scalar determinant() const;
+    Scalar determinant() const;
 
     MatrixType reconstructedMatrix() const;
 
     inline Index rows() const { return m_lu.rows(); }
     inline Index cols() const { return m_lu.cols(); }
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const {
+     /* The decomposition PA = LU can be rewritten as A = P^{-1} L U.
+      * So we proceed as follows:
+      * Step 1: compute c = Pb.
+      * Step 2: replace c by the solution x to Lx = c.
+      * Step 3: replace c by the solution x to Ux = c.
+      */
+
+      eigen_assert(rhs.rows() == m_lu.rows());
+
+      // Step 1
+      dst = permutationP() * rhs;
+
+      // Step 2
+      m_lu.template triangularView<UnitLower>().solveInPlace(dst);
+
+      // Step 3
+      m_lu.template triangularView<Upper>().solveInPlace(dst);
+    }
+
+    template<bool Conjugate, typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const {
+     /* The decomposition PA = LU can be rewritten as A = P^{-1} L U.
+      * So we proceed as follows:
+      * Step 1: compute c = Pb.
+      * Step 2: replace c by the solution x to Lx = c.
+      * Step 3: replace c by the solution x to Ux = c.
+      */
+
+      eigen_assert(rhs.rows() == m_lu.cols());
+
+      if (Conjugate) {
+        // Step 1
+        dst = m_lu.template triangularView<Upper>().adjoint().solve(rhs);
+        // Step 2
+        m_lu.template triangularView<UnitLower>().adjoint().solveInPlace(dst);
+      } else {
+        // Step 1
+        dst = m_lu.template triangularView<Upper>().transpose().solve(rhs);
+        // Step 2
+        m_lu.template triangularView<UnitLower>().transpose().solveInPlace(dst);
+      }
+      // Step 3
+      dst = permutationP().transpose() * dst;
+    }
+    #endif
+
   protected:
-    
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
-    
+
+    void compute();
+
     MatrixType m_lu;
     PermutationType m_p;
     TranspositionType m_rowsTranspositions;
-    Index m_det_p;
+    RealScalar m_l1_norm;
+    signed char m_det_p;
     bool m_isInitialized;
 };
 
@@ -189,6 +293,7 @@ PartialPivLU<MatrixType>::PartialPivLU()
   : m_lu(),
     m_p(),
     m_rowsTranspositions(),
+    m_l1_norm(0),
     m_det_p(0),
     m_isInitialized(false)
 {
@@ -199,20 +304,36 @@ PartialPivLU<MatrixType>::PartialPivLU(Index size)
   : m_lu(size, size),
     m_p(size),
     m_rowsTranspositions(size),
+    m_l1_norm(0),
     m_det_p(0),
     m_isInitialized(false)
 {
 }
 
 template<typename MatrixType>
-PartialPivLU<MatrixType>::PartialPivLU(const MatrixType& matrix)
-  : m_lu(matrix.rows(), matrix.rows()),
+template<typename InputType>
+PartialPivLU<MatrixType>::PartialPivLU(const EigenBase<InputType>& matrix)
+  : m_lu(matrix.rows(),matrix.cols()),
     m_p(matrix.rows()),
     m_rowsTranspositions(matrix.rows()),
+    m_l1_norm(0),
     m_det_p(0),
     m_isInitialized(false)
 {
-  compute(matrix);
+  compute(matrix.derived());
+}
+
+template<typename MatrixType>
+template<typename InputType>
+PartialPivLU<MatrixType>::PartialPivLU(EigenBase<InputType>& matrix)
+  : m_lu(matrix.derived()),
+    m_p(matrix.rows()),
+    m_rowsTranspositions(matrix.rows()),
+    m_l1_norm(0),
+    m_det_p(0),
+    m_isInitialized(false)
+{
+  compute();
 }
 
 namespace internal {
@@ -230,7 +351,6 @@ struct partial_lu_impl
   typedef Block<MapLU, Dynamic, Dynamic> MatrixType;
   typedef Block<MatrixType,Dynamic,Dynamic> BlockType;
   typedef typename MatrixType::RealScalar RealScalar;
-  typedef typename MatrixType::Index Index;
 
   /** \internal performs the LU decomposition in-place of the matrix \a lu
     * using an unblocked algorithm.
@@ -244,6 +364,8 @@ struct partial_lu_impl
     */
   static Index unblocked_lu(MatrixType& lu, PivIndex* row_transpositions, PivIndex& nb_transpositions)
   {
+    typedef scalar_score_coeff_op<Scalar> Scoring;
+    typedef typename Scoring::result_type Score;
     const Index rows = lu.rows();
     const Index cols = lu.cols();
     const Index size = (std::min)(rows,cols);
@@ -253,15 +375,15 @@ struct partial_lu_impl
     {
       Index rrows = rows-k-1;
       Index rcols = cols-k-1;
-        
+
       Index row_of_biggest_in_col;
-      RealScalar biggest_in_corner
-        = lu.col(k).tail(rows-k).cwiseAbs().maxCoeff(&row_of_biggest_in_col);
+      Score biggest_in_corner
+        = lu.col(k).tail(rows-k).unaryExpr(Scoring()).maxCoeff(&row_of_biggest_in_col);
       row_of_biggest_in_col += k;
 
       row_transpositions[k] = PivIndex(row_of_biggest_in_col);
 
-      if(biggest_in_corner != RealScalar(0))
+      if(biggest_in_corner != Score(0))
       {
         if(k != row_of_biggest_in_col)
         {
@@ -354,7 +476,7 @@ struct partial_lu_impl
       // update permutations and apply them to A_0
       for(Index i=k; i<k+bs; ++i)
       {
-        Index piv = (row_transpositions[i] += k);
+        Index piv = (row_transpositions[i] += internal::convert_index<PivIndex>(k));
         A_0.row(i).swap(A_0.row(piv));
       }
 
@@ -377,45 +499,44 @@ struct partial_lu_impl
 /** \internal performs the LU decomposition with partial pivoting in-place.
   */
 template<typename MatrixType, typename TranspositionType>
-void partial_lu_inplace(MatrixType& lu, TranspositionType& row_transpositions, typename TranspositionType::Index& nb_transpositions)
+void partial_lu_inplace(MatrixType& lu, TranspositionType& row_transpositions, typename TranspositionType::StorageIndex& nb_transpositions)
 {
   eigen_assert(lu.cols() == row_transpositions.size());
   eigen_assert((&row_transpositions.coeffRef(1)-&row_transpositions.coeffRef(0)) == 1);
 
   partial_lu_impl
-    <typename MatrixType::Scalar, MatrixType::Flags&RowMajorBit?RowMajor:ColMajor, typename TranspositionType::Index>
+    <typename MatrixType::Scalar, MatrixType::Flags&RowMajorBit?RowMajor:ColMajor, typename TranspositionType::StorageIndex>
     ::blocked_lu(lu.rows(), lu.cols(), &lu.coeffRef(0,0), lu.outerStride(), &row_transpositions.coeffRef(0), nb_transpositions);
 }
 
 } // end namespace internal
 
 template<typename MatrixType>
-PartialPivLU<MatrixType>& PartialPivLU<MatrixType>::compute(const MatrixType& matrix)
+void PartialPivLU<MatrixType>::compute()
 {
   check_template_parameters();
-  
+
   // the row permutation is stored as int indices, so just to be sure:
-  eigen_assert(matrix.rows()<NumTraits<int>::highest());
-  
-  m_lu = matrix;
+  eigen_assert(m_lu.rows()<NumTraits<int>::highest());
+
+  m_l1_norm = m_lu.cwiseAbs().colwise().sum().maxCoeff();
 
-  eigen_assert(matrix.rows() == matrix.cols() && "PartialPivLU is only for square (and moreover invertible) matrices");
-  const Index size = matrix.rows();
+  eigen_assert(m_lu.rows() == m_lu.cols() && "PartialPivLU is only for square (and moreover invertible) matrices");
+  const Index size = m_lu.rows();
 
   m_rowsTranspositions.resize(size);
 
-  typename TranspositionType::Index nb_transpositions;
+  typename TranspositionType::StorageIndex nb_transpositions;
   internal::partial_lu_inplace(m_lu, m_rowsTranspositions, nb_transpositions);
   m_det_p = (nb_transpositions%2) ? -1 : 1;
 
   m_p = m_rowsTranspositions;
 
   m_isInitialized = true;
-  return *this;
 }
 
 template<typename MatrixType>
-typename internal::traits<MatrixType>::Scalar PartialPivLU<MatrixType>::determinant() const
+typename PartialPivLU<MatrixType>::Scalar PartialPivLU<MatrixType>::determinant() const
 {
   eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
   return Scalar(m_det_p) * m_lu.diagonal().prod();
@@ -438,38 +559,21 @@ MatrixType PartialPivLU<MatrixType>::reconstructedMatrix() const
   return res;
 }
 
-/***** Implementation of solve() *****************************************************/
+/***** Implementation details *****************************************************/
 
 namespace internal {
 
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<PartialPivLU<_MatrixType>, Rhs>
-  : solve_retval_base<PartialPivLU<_MatrixType>, Rhs>
+/***** Implementation of inverse() *****************************************************/
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<PartialPivLU<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename PartialPivLU<MatrixType>::Scalar>, Dense2Dense>
 {
-  EIGEN_MAKE_SOLVE_HELPERS(PartialPivLU<_MatrixType>,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
+  typedef PartialPivLU<MatrixType> LuType;
+  typedef Inverse<LuType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename LuType::Scalar> &)
   {
-    /* The decomposition PA = LU can be rewritten as A = P^{-1} L U.
-    * So we proceed as follows:
-    * Step 1: compute c = Pb.
-    * Step 2: replace c by the solution x to Lx = c.
-    * Step 3: replace c by the solution x to Ux = c.
-    */
-
-    eigen_assert(rhs().rows() == dec().matrixLU().rows());
-
-    // Step 1
-    dst = dec().permutationP() * rhs();
-
-    // Step 2
-    dec().matrixLU().template triangularView<UnitLower>().solveInPlace(dst);
-
-    // Step 3
-    dec().matrixLU().template triangularView<Upper>().solveInPlace(dst);
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.cols()));
   }
 };
-
 } // end namespace internal
 
 /******** MatrixBase methods *******/
@@ -487,7 +591,6 @@ MatrixBase<Derived>::partialPivLu() const
   return PartialPivLU<PlainObject>(eval());
 }
 
-#if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
 /** \lu_module
   *
   * Synonym of partialPivLu().
@@ -502,7 +605,6 @@ MatrixBase<Derived>::lu() const
 {
   return PartialPivLU<PlainObject>(eval());
 }
-#endif
 
 } // end namespace Eigen
 
diff --git a/extern/Eigen3/Eigen/src/LU/PartialPivLU_LAPACKE.h b/extern/Eigen3/Eigen/src/LU/PartialPivLU_LAPACKE.h
new file mode 100644
index 00000000000..755168a9460
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/LU/PartialPivLU_LAPACKE.h
@@ -0,0 +1,83 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *     LU decomposition with partial pivoting based on LAPACKE_?getrf function.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_PARTIALLU_LAPACK_H
+#define EIGEN_PARTIALLU_LAPACK_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/** \internal Specialization for the data types supported by LAPACKe */
+
+#define EIGEN_LAPACKE_LU_PARTPIV(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX) \
+template<int StorageOrder> \
+struct partial_lu_impl<EIGTYPE, StorageOrder, lapack_int> \
+{ \
+  /* \internal performs the LU decomposition in-place of the matrix represented */ \
+  static lapack_int blocked_lu(Index rows, Index cols, EIGTYPE* lu_data, Index luStride, lapack_int* row_transpositions, lapack_int& nb_transpositions, lapack_int maxBlockSize=256) \
+  { \
+    EIGEN_UNUSED_VARIABLE(maxBlockSize);\
+    lapack_int matrix_order, first_zero_pivot; \
+    lapack_int m, n, lda, *ipiv, info; \
+    EIGTYPE* a; \
+/* Set up parameters for ?getrf */ \
+    matrix_order = StorageOrder==RowMajor ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \
+    lda = convert_index<lapack_int>(luStride); \
+    a = lu_data; \
+    ipiv = row_transpositions; \
+    m = convert_index<lapack_int>(rows); \
+    n = convert_index<lapack_int>(cols); \
+    nb_transpositions = 0; \
+\
+    info = LAPACKE_##LAPACKE_PREFIX##getrf( matrix_order, m, n, (LAPACKE_TYPE*)a, lda, ipiv ); \
+\
+    for(int i=0;i<m;i++) { ipiv[i]--; if (ipiv[i]!=i) nb_transpositions++; } \
+\
+    eigen_assert(info >= 0); \
+/* something should be done with nb_transpositions */ \
+\
+    first_zero_pivot = info; \
+    return first_zero_pivot; \
+  } \
+};
+
+EIGEN_LAPACKE_LU_PARTPIV(double, double, d)
+EIGEN_LAPACKE_LU_PARTPIV(float, float, s)
+EIGEN_LAPACKE_LU_PARTPIV(dcomplex, lapack_complex_double, z)
+EIGEN_LAPACKE_LU_PARTPIV(scomplex, lapack_complex_float,  c)
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PARTIALLU_LAPACK_H
diff --git a/extern/Eigen3/Eigen/src/LU/arch/Inverse_SSE.h b/extern/Eigen3/Eigen/src/LU/arch/Inverse_SSE.h
index 60b7a23763e..ebb64a62b02 100644
--- a/extern/Eigen3/Eigen/src/LU/arch/Inverse_SSE.h
+++ b/extern/Eigen3/Eigen/src/LU/arch/Inverse_SSE.h
@@ -35,13 +35,15 @@ template<typename MatrixType, typename ResultType>
 struct compute_inverse_size4<Architecture::SSE, float, MatrixType, ResultType>
 {
   enum {
-    MatrixAlignment     = bool(MatrixType::Flags&AlignedBit),
-    ResultAlignment     = bool(ResultType::Flags&AlignedBit),
+    MatrixAlignment     = traits<MatrixType>::Alignment,
+    ResultAlignment     = traits<ResultType>::Alignment,
     StorageOrdersMatch  = (MatrixType::Flags&RowMajorBit) == (ResultType::Flags&RowMajorBit)
   };
+  typedef typename conditional<(MatrixType::Flags&LinearAccessBit),MatrixType const &,typename MatrixType::PlainObject>::type ActualMatrixType;
   
-  static void run(const MatrixType& matrix, ResultType& result)
+  static void run(const MatrixType& mat, ResultType& result)
   {
+    ActualMatrixType matrix(mat);
     EIGEN_ALIGN16 const unsigned int _Sign_PNNP[4] = { 0x00000000, 0x80000000, 0x80000000, 0x00000000 };
 
     // Load the full matrix into registers
@@ -151,10 +153,12 @@ struct compute_inverse_size4<Architecture::SSE, float, MatrixType, ResultType>
     iC = _mm_mul_ps(rd,iC);
     iD = _mm_mul_ps(rd,iD);
 
-    result.template writePacket<ResultAlignment>( 0, _mm_shuffle_ps(iA,iB,0x77));
-    result.template writePacket<ResultAlignment>( 4, _mm_shuffle_ps(iA,iB,0x22));
-    result.template writePacket<ResultAlignment>( 8, _mm_shuffle_ps(iC,iD,0x77));
-    result.template writePacket<ResultAlignment>(12, _mm_shuffle_ps(iC,iD,0x22));
+    Index res_stride = result.outerStride();
+    float* res = result.data();
+    pstoret<float, Packet4f, ResultAlignment>(res+0,            _mm_shuffle_ps(iA,iB,0x77));
+    pstoret<float, Packet4f, ResultAlignment>(res+res_stride,   _mm_shuffle_ps(iA,iB,0x22));
+    pstoret<float, Packet4f, ResultAlignment>(res+2*res_stride, _mm_shuffle_ps(iC,iD,0x77));
+    pstoret<float, Packet4f, ResultAlignment>(res+3*res_stride, _mm_shuffle_ps(iC,iD,0x22));
   }
 
 };
@@ -163,18 +167,21 @@ template<typename MatrixType, typename ResultType>
 struct compute_inverse_size4<Architecture::SSE, double, MatrixType, ResultType>
 {
   enum {
-    MatrixAlignment = bool(MatrixType::Flags&AlignedBit),
-    ResultAlignment = bool(ResultType::Flags&AlignedBit),
+    MatrixAlignment     = traits<MatrixType>::Alignment,
+    ResultAlignment     = traits<ResultType>::Alignment,
     StorageOrdersMatch  = (MatrixType::Flags&RowMajorBit) == (ResultType::Flags&RowMajorBit)
   };
-  static void run(const MatrixType& matrix, ResultType& result)
+  typedef typename conditional<(MatrixType::Flags&LinearAccessBit),MatrixType const &,typename MatrixType::PlainObject>::type ActualMatrixType;
+  
+  static void run(const MatrixType& mat, ResultType& result)
   {
+    ActualMatrixType matrix(mat);
     const __m128d _Sign_NP = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
     const __m128d _Sign_PN = _mm_castsi128_pd(_mm_set_epi32(0x80000000,0x0,0x0,0x0));
 
     // The inverse is calculated using "Divide and Conquer" technique. The
     // original matrix is divide into four 2x2 sub-matrices. Since each
-    // register of the matrix holds two element, the smaller matrices are
+    // register of the matrix holds two elements, the smaller matrices are
     // consisted of two registers. Hence we get a better locality of the
     // calculations.
 
@@ -311,14 +318,16 @@ struct compute_inverse_size4<Architecture::SSE, double, MatrixType, ResultType>
     iC1 = _mm_sub_pd(_mm_mul_pd(B1, dC), iC1);
     iC2 = _mm_sub_pd(_mm_mul_pd(B2, dC), iC2);
 
-    result.template writePacket<ResultAlignment>( 0, _mm_mul_pd(_mm_shuffle_pd(iA2, iA1, 3), d1));     // iA# / det
-    result.template writePacket<ResultAlignment>( 4, _mm_mul_pd(_mm_shuffle_pd(iA2, iA1, 0), d2));
-    result.template writePacket<ResultAlignment>( 2, _mm_mul_pd(_mm_shuffle_pd(iB2, iB1, 3), d1));     // iB# / det
-    result.template writePacket<ResultAlignment>( 6, _mm_mul_pd(_mm_shuffle_pd(iB2, iB1, 0), d2));
-    result.template writePacket<ResultAlignment>( 8, _mm_mul_pd(_mm_shuffle_pd(iC2, iC1, 3), d1));     // iC# / det
-    result.template writePacket<ResultAlignment>(12, _mm_mul_pd(_mm_shuffle_pd(iC2, iC1, 0), d2));
-    result.template writePacket<ResultAlignment>(10, _mm_mul_pd(_mm_shuffle_pd(iD2, iD1, 3), d1));     // iD# / det
-    result.template writePacket<ResultAlignment>(14, _mm_mul_pd(_mm_shuffle_pd(iD2, iD1, 0), d2));
+    Index res_stride = result.outerStride();
+    double* res = result.data();
+    pstoret<double, Packet2d, ResultAlignment>(res+0,             _mm_mul_pd(_mm_shuffle_pd(iA2, iA1, 3), d1));
+    pstoret<double, Packet2d, ResultAlignment>(res+res_stride,    _mm_mul_pd(_mm_shuffle_pd(iA2, iA1, 0), d2));
+    pstoret<double, Packet2d, ResultAlignment>(res+2,             _mm_mul_pd(_mm_shuffle_pd(iB2, iB1, 3), d1));
+    pstoret<double, Packet2d, ResultAlignment>(res+res_stride+2,  _mm_mul_pd(_mm_shuffle_pd(iB2, iB1, 0), d2));
+    pstoret<double, Packet2d, ResultAlignment>(res+2*res_stride,  _mm_mul_pd(_mm_shuffle_pd(iC2, iC1, 3), d1));
+    pstoret<double, Packet2d, ResultAlignment>(res+3*res_stride,  _mm_mul_pd(_mm_shuffle_pd(iC2, iC1, 0), d2));
+    pstoret<double, Packet2d, ResultAlignment>(res+2*res_stride+2,_mm_mul_pd(_mm_shuffle_pd(iD2, iD1, 3), d1));
+    pstoret<double, Packet2d, ResultAlignment>(res+3*res_stride+2,_mm_mul_pd(_mm_shuffle_pd(iD2, iD1, 0), d2));
   }
 };
 
diff --git a/extern/Eigen3/Eigen/src/MetisSupport/MetisSupport.h b/extern/Eigen3/Eigen/src/MetisSupport/MetisSupport.h
index f2bbef20c8f..4c15304ad63 100644
--- a/extern/Eigen3/Eigen/src/MetisSupport/MetisSupport.h
+++ b/extern/Eigen3/Eigen/src/MetisSupport/MetisSupport.h
@@ -18,12 +18,12 @@ namespace Eigen {
  * Row (column) i of A is the matperm(i) row (column) of Ap. 
  * WARNING: As computed by METIS, this corresponds to the vector iperm (instead of perm)
  */
-template <typename Index>
+template <typename StorageIndex>
 class MetisOrdering
 {
 public:
-  typedef PermutationMatrix<Dynamic,Dynamic,Index> PermutationType;
-  typedef Matrix<Index,Dynamic,1> IndexVector; 
+  typedef PermutationMatrix<Dynamic,Dynamic,StorageIndex> PermutationType;
+  typedef Matrix<StorageIndex,Dynamic,1> IndexVector; 
   
   template <typename MatrixType>
   void get_symmetrized_graph(const MatrixType& A)
@@ -36,7 +36,7 @@ public:
     Index TotNz = 0; 
     IndexVector visited(m); 
     visited.setConstant(-1); 
-    for (int j = 0; j < m; j++)
+    for (StorageIndex j = 0; j < m; j++)
     {
       // Compute the union structure of of A(j,:) and At(j,:)
       visited(j) = j; // Do not include the diagonal element
@@ -67,8 +67,8 @@ public:
 
     // Now compute the real adjacency list of each column/row 
     visited.setConstant(-1); 
-    Index CurNz = 0; 
-    for (int j = 0; j < m; j++)
+    StorageIndex CurNz = 0; 
+    for (StorageIndex j = 0; j < m; j++)
     {
       m_indexPtr(j) = CurNz; 
       
@@ -76,7 +76,7 @@ public:
       // Add the pattern of row/column j of A to A+At
       for (typename MatrixType::InnerIterator it(A,j); it; ++it)
       {
-        Index idx = it.index(); // Get the row index (for column major) or column index (for row major)
+        StorageIndex idx = it.index(); // Get the row index (for column major) or column index (for row major)
         if (visited(idx) != j ) 
         {
           visited(idx) = j; 
@@ -87,7 +87,7 @@ public:
       //Add the pattern of row/column j of At to A+At
       for (typename MatrixType::InnerIterator it(At, j); it; ++it)
       {
-        Index idx = it.index(); 
+        StorageIndex idx = it.index(); 
         if(visited(idx) != j)
         {
           visited(idx) = j; 
@@ -102,7 +102,7 @@ public:
   template <typename MatrixType>
   void operator() (const MatrixType& A, PermutationType& matperm)
   {
-     Index m = A.cols();
+     StorageIndex m = internal::convert_index<StorageIndex>(A.cols()); // must be StorageIndex, because it is passed by address to METIS
      IndexVector perm(m),iperm(m); 
     // First, symmetrize the matrix graph. 
      get_symmetrized_graph(A); 
diff --git a/extern/Eigen3/Eigen/src/OrderingMethods/Amd.h b/extern/Eigen3/Eigen/src/OrderingMethods/Amd.h
index 70550b8a90a..f91ecb24efc 100644
--- a/extern/Eigen3/Eigen/src/OrderingMethods/Amd.h
+++ b/extern/Eigen3/Eigen/src/OrderingMethods/Amd.h
@@ -8,7 +8,7 @@
 NOTE: this routine has been adapted from the CSparse library:
 
 Copyright (c) 2006, Timothy A. Davis.
-http://www.cise.ufl.edu/research/sparse/CSparse
+http://www.suitesparse.com
 
 CSparse is free software; you can redistribute it and/or
 modify it under the terms of the GNU Lesser General Public
@@ -41,10 +41,10 @@ template<typename T0, typename T1> inline bool amd_marked(const T0* w, const T1&
 template<typename T0, typename T1> inline void amd_mark(const T0* w, const T1& j) { return w[j] = amd_flip(w[j]); }
 
 /* clear w */
-template<typename Index>
-static int cs_wclear (Index mark, Index lemax, Index *w, Index n)
+template<typename StorageIndex>
+static StorageIndex cs_wclear (StorageIndex mark, StorageIndex lemax, StorageIndex *w, StorageIndex n)
 {
-  Index k;
+  StorageIndex k;
   if(mark < 2 || (mark + lemax < 0))
   {
     for(k = 0; k < n; k++)
@@ -56,10 +56,10 @@ static int cs_wclear (Index mark, Index lemax, Index *w, Index n)
 }
 
 /* depth-first search and postorder of a tree rooted at node j */
-template<typename Index>
-Index cs_tdfs(Index j, Index k, Index *head, const Index *next, Index *post, Index *stack)
+template<typename StorageIndex>
+StorageIndex cs_tdfs(StorageIndex j, StorageIndex k, StorageIndex *head, const StorageIndex *next, StorageIndex *post, StorageIndex *stack)
 {
-  int i, p, top = 0;
+  StorageIndex i, p, top = 0;
   if(!head || !next || !post || !stack) return (-1);    /* check inputs */
   stack[0] = j;                 /* place j on the stack */
   while (top >= 0)                /* while (stack is not empty) */
@@ -84,42 +84,45 @@ Index cs_tdfs(Index j, Index k, Index *head, const Index *next, Index *post, Ind
 /** \internal
   * \ingroup OrderingMethods_Module 
   * Approximate minimum degree ordering algorithm.
-  * \returns the permutation P reducing the fill-in of the input matrix \a C
-  * The input matrix \a C must be a selfadjoint compressed column major SparseMatrix object. Both the upper and lower parts have to be stored, but the diagonal entries are optional.
+  *
+  * \param[in] C the input selfadjoint matrix stored in compressed column major format.
+  * \param[out] perm the permutation P reducing the fill-in of the input matrix \a C
+  *
+  * Note that the input matrix \a C must be complete, that is both the upper and lower parts have to be stored, as well as the diagonal entries.
   * On exit the values of C are destroyed */
-template<typename Scalar, typename Index>
-void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,Index>& C, PermutationMatrix<Dynamic,Dynamic,Index>& perm)
+template<typename Scalar, typename StorageIndex>
+void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,StorageIndex>& C, PermutationMatrix<Dynamic,Dynamic,StorageIndex>& perm)
 {
   using std::sqrt;
   
-  int d, dk, dext, lemax = 0, e, elenk, eln, i, j, k, k1,
-      k2, k3, jlast, ln, dense, nzmax, mindeg = 0, nvi, nvj, nvk, mark, wnvi,
-      ok, nel = 0, p, p1, p2, p3, p4, pj, pk, pk1, pk2, pn, q, t;
-  unsigned int h;
+  StorageIndex d, dk, dext, lemax = 0, e, elenk, eln, i, j, k, k1,
+                k2, k3, jlast, ln, dense, nzmax, mindeg = 0, nvi, nvj, nvk, mark, wnvi,
+                ok, nel = 0, p, p1, p2, p3, p4, pj, pk, pk1, pk2, pn, q, t, h;
   
-  Index n = C.cols();
-  dense = std::max<Index> (16, Index(10 * sqrt(double(n))));   /* find dense threshold */
-  dense = std::min<Index> (n-2, dense);
+  StorageIndex n = StorageIndex(C.cols());
+  dense = std::max<StorageIndex> (16, StorageIndex(10 * sqrt(double(n))));   /* find dense threshold */
+  dense = (std::min)(n-2, dense);
   
-  Index cnz = C.nonZeros();
+  StorageIndex cnz = StorageIndex(C.nonZeros());
   perm.resize(n+1);
   t = cnz + cnz/5 + 2*n;                 /* add elbow room to C */
   C.resizeNonZeros(t);
   
-  Index* W       = new Index[8*(n+1)]; /* get workspace */
-  Index* len     = W;
-  Index* nv      = W +   (n+1);
-  Index* next    = W + 2*(n+1);
-  Index* head    = W + 3*(n+1);
-  Index* elen    = W + 4*(n+1);
-  Index* degree  = W + 5*(n+1);
-  Index* w       = W + 6*(n+1);
-  Index* hhead   = W + 7*(n+1);
-  Index* last    = perm.indices().data();                              /* use P as workspace for last */
+  // get workspace
+  ei_declare_aligned_stack_constructed_variable(StorageIndex,W,8*(n+1),0);
+  StorageIndex* len     = W;
+  StorageIndex* nv      = W +   (n+1);
+  StorageIndex* next    = W + 2*(n+1);
+  StorageIndex* head    = W + 3*(n+1);
+  StorageIndex* elen    = W + 4*(n+1);
+  StorageIndex* degree  = W + 5*(n+1);
+  StorageIndex* w       = W + 6*(n+1);
+  StorageIndex* hhead   = W + 7*(n+1);
+  StorageIndex* last    = perm.indices().data();                              /* use P as workspace for last */
   
   /* --- Initialize quotient graph ---------------------------------------- */
-  Index* Cp = C.outerIndexPtr();
-  Index* Ci = C.innerIndexPtr();
+  StorageIndex* Cp = C.outerIndexPtr();
+  StorageIndex* Ci = C.innerIndexPtr();
   for(k = 0; k < n; k++)
     len[k] = Cp[k+1] - Cp[k];
   len[n] = 0;
@@ -136,7 +139,7 @@ void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,Index>& C, Permutation
     elen[i]   = 0;                      // Ek of node i is empty
     degree[i] = len[i];                 // degree of node i
   }
-  mark = internal::cs_wclear<Index>(0, 0, w, n);         /* clear w */
+  mark = internal::cs_wclear<StorageIndex>(0, 0, w, n);         /* clear w */
   
   /* --- Initialize degree lists ------------------------------------------ */
   for(i = 0; i < n; i++)
@@ -260,7 +263,7 @@ void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,Index>& C, Permutation
     elen[k] = -2;                     /* k is now an element */
     
     /* --- Find set differences ----------------------------------------- */
-    mark = internal::cs_wclear<Index>(mark, lemax, w, n);  /* clear w if necessary */
+    mark = internal::cs_wclear<StorageIndex>(mark, lemax, w, n);  /* clear w if necessary */
     for(pk = pk1; pk < pk2; pk++)    /* scan 1: find |Le\Lk| */
     {
       i = Ci[pk];
@@ -330,7 +333,7 @@ void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,Index>& C, Permutation
       }
       else
       {
-        degree[i] = std::min<Index> (degree[i], d);   /* update degree(i) */
+        degree[i] = std::min<StorageIndex> (degree[i], d);   /* update degree(i) */
         Ci[pn] = Ci[p3];         /* move first node to end */
         Ci[p3] = Ci[p1];         /* move 1st el. to end of Ei */
         Ci[p1] = k;               /* add k as 1st element in of Ei */
@@ -338,12 +341,12 @@ void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,Index>& C, Permutation
         h %= n;                    /* finalize hash of i */
         next[i] = hhead[h];      /* place i in hash bucket */
         hhead[h] = i;
-        last[i] = h;              /* save hash of i in last[i] */
+        last[i] = h;      /* save hash of i in last[i] */
       }
     }                                   /* scan2 is done */
     degree[k] = dk;                   /* finalize |Lk| */
-    lemax = std::max<Index>(lemax, dk);
-    mark = internal::cs_wclear<Index>(mark+lemax, lemax, w, n);    /* clear w */
+    lemax = std::max<StorageIndex>(lemax, dk);
+    mark = internal::cs_wclear<StorageIndex>(mark+lemax, lemax, w, n);    /* clear w */
     
     /* --- Supernode detection ------------------------------------------ */
     for(pk = pk1; pk < pk2; pk++)
@@ -391,12 +394,12 @@ void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,Index>& C, Permutation
       if((nvi = -nv[i]) <= 0) continue;/* skip if i is dead */
       nv[i] = nvi;                      /* restore nv[i] */
       d = degree[i] + dk - nvi;         /* compute external degree(i) */
-      d = std::min<Index> (d, n - nel - nvi);
+      d = std::min<StorageIndex> (d, n - nel - nvi);
       if(head[d] != -1) last[head[d]] = i;
       next[i] = head[d];               /* put i back in degree list */
       last[i] = -1;
       head[d] = i;
-      mindeg = std::min<Index> (mindeg, d);       /* find new minimum degree */
+      mindeg = std::min<StorageIndex> (mindeg, d);       /* find new minimum degree */
       degree[i] = d;
       Ci[p++] = i;                      /* place i in Lk */
     }
@@ -429,12 +432,10 @@ void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,Index>& C, Permutation
   }
   for(k = 0, i = 0; i <= n; i++)       /* postorder the assembly tree */
   {
-    if(Cp[i] == -1) k = internal::cs_tdfs<Index>(i, k, head, next, perm.indices().data(), w);
+    if(Cp[i] == -1) k = internal::cs_tdfs<StorageIndex>(i, k, head, next, perm.indices().data(), w);
   }
   
   perm.indices().conservativeResize(n);
-
-  delete[] W;
 }
 
 } // namespace internal
diff --git a/extern/Eigen3/Eigen/src/OrderingMethods/Eigen_Colamd.h b/extern/Eigen3/Eigen/src/OrderingMethods/Eigen_Colamd.h
index 44548f6607c..da85b4d6ea2 100644
--- a/extern/Eigen3/Eigen/src/OrderingMethods/Eigen_Colamd.h
+++ b/extern/Eigen3/Eigen/src/OrderingMethods/Eigen_Colamd.h
@@ -41,12 +41,8 @@
 // 
 //   The colamd/symamd library is available at
 // 
-//       http://www.cise.ufl.edu/research/sparse/colamd/
+//       http://www.suitesparse.com
 
-//   This is the http://www.cise.ufl.edu/research/sparse/colamd/colamd.h
-//   file.  It is required by the colamd.c, colamdmex.c, and symamdmex.c
-//   files, and by any C code that calls the routines whose prototypes are
-//   listed below, or that uses the colamd/symamd definitions listed below.
   
 #ifndef EIGEN_COLAMD_H
 #define EIGEN_COLAMD_H
@@ -102,9 +98,6 @@ namespace internal {
 /* === Definitions ========================================================== */
 /* ========================================================================== */
 
-#define COLAMD_MAX(a,b) (((a) > (b)) ? (a) : (b))
-#define COLAMD_MIN(a,b) (((a) < (b)) ? (a) : (b))
-
 #define ONES_COMPLEMENT(r) (-(r)-1)
 
 /* -------------------------------------------------------------------------- */
@@ -135,54 +128,54 @@ namespace internal {
 /* ========================================================================== */
 
 // == Row and Column structures ==
-template <typename Index>
+template <typename IndexType>
 struct colamd_col
 {
-  Index start ;   /* index for A of first row in this column, or DEAD */
+  IndexType start ;   /* index for A of first row in this column, or DEAD */
   /* if column is dead */
-  Index length ;  /* number of rows in this column */
+  IndexType length ;  /* number of rows in this column */
   union
   {
-    Index thickness ; /* number of original columns represented by this */
+    IndexType thickness ; /* number of original columns represented by this */
     /* col, if the column is alive */
-    Index parent ;  /* parent in parent tree super-column structure, if */
+    IndexType parent ;  /* parent in parent tree super-column structure, if */
     /* the column is dead */
   } shared1 ;
   union
   {
-    Index score ; /* the score used to maintain heap, if col is alive */
-    Index order ; /* pivot ordering of this column, if col is dead */
+    IndexType score ; /* the score used to maintain heap, if col is alive */
+    IndexType order ; /* pivot ordering of this column, if col is dead */
   } shared2 ;
   union
   {
-    Index headhash ;  /* head of a hash bucket, if col is at the head of */
+    IndexType headhash ;  /* head of a hash bucket, if col is at the head of */
     /* a degree list */
-    Index hash ;  /* hash value, if col is not in a degree list */
-    Index prev ;  /* previous column in degree list, if col is in a */
+    IndexType hash ;  /* hash value, if col is not in a degree list */
+    IndexType prev ;  /* previous column in degree list, if col is in a */
     /* degree list (but not at the head of a degree list) */
   } shared3 ;
   union
   {
-    Index degree_next ; /* next column, if col is in a degree list */
-    Index hash_next ;   /* next column, if col is in a hash list */
+    IndexType degree_next ; /* next column, if col is in a degree list */
+    IndexType hash_next ;   /* next column, if col is in a hash list */
   } shared4 ;
   
 };
  
-template <typename Index>
+template <typename IndexType>
 struct Colamd_Row
 {
-  Index start ;   /* index for A of first col in this row */
-  Index length ;  /* number of principal columns in this row */
+  IndexType start ;   /* index for A of first col in this row */
+  IndexType length ;  /* number of principal columns in this row */
   union
   {
-    Index degree ;  /* number of principal & non-principal columns in row */
-    Index p ;   /* used as a row pointer in init_rows_cols () */
+    IndexType degree ;  /* number of principal & non-principal columns in row */
+    IndexType p ;   /* used as a row pointer in init_rows_cols () */
   } shared1 ;
   union
   {
-    Index mark ;  /* for computing set differences and marking dead rows*/
-    Index first_column ;/* first column in row (used in garbage collection) */
+    IndexType mark ;  /* for computing set differences and marking dead rows*/
+    IndexType first_column ;/* first column in row (used in garbage collection) */
   } shared2 ;
   
 };
@@ -202,38 +195,38 @@ struct Colamd_Row
   
   This macro is not needed when using symamd.
   
-  Explicit typecast to Index added Sept. 23, 2002, COLAMD version 2.2, to avoid
+  Explicit typecast to IndexType added Sept. 23, 2002, COLAMD version 2.2, to avoid
   gcc -pedantic warning messages.
 */
-template <typename Index>
-inline Index colamd_c(Index n_col) 
-{ return Index( ((n_col) + 1) * sizeof (colamd_col<Index>) / sizeof (Index) ) ; }
+template <typename IndexType>
+inline IndexType colamd_c(IndexType n_col) 
+{ return IndexType( ((n_col) + 1) * sizeof (colamd_col<IndexType>) / sizeof (IndexType) ) ; }
 
-template <typename Index>
-inline Index  colamd_r(Index n_row)
-{ return Index(((n_row) + 1) * sizeof (Colamd_Row<Index>) / sizeof (Index)); }
+template <typename IndexType>
+inline IndexType  colamd_r(IndexType n_row)
+{ return IndexType(((n_row) + 1) * sizeof (Colamd_Row<IndexType>) / sizeof (IndexType)); }
 
 // Prototypes of non-user callable routines
-template <typename Index>
-static Index init_rows_cols (Index n_row, Index n_col, Colamd_Row<Index> Row [], colamd_col<Index> col [], Index A [], Index p [], Index stats[COLAMD_STATS] ); 
+template <typename IndexType>
+static IndexType init_rows_cols (IndexType n_row, IndexType n_col, Colamd_Row<IndexType> Row [], colamd_col<IndexType> col [], IndexType A [], IndexType p [], IndexType stats[COLAMD_STATS] ); 
 
-template <typename Index>
-static void init_scoring (Index n_row, Index n_col, Colamd_Row<Index> Row [], colamd_col<Index> Col [], Index A [], Index head [], double knobs[COLAMD_KNOBS], Index *p_n_row2, Index *p_n_col2, Index *p_max_deg);
+template <typename IndexType>
+static void init_scoring (IndexType n_row, IndexType n_col, Colamd_Row<IndexType> Row [], colamd_col<IndexType> Col [], IndexType A [], IndexType head [], double knobs[COLAMD_KNOBS], IndexType *p_n_row2, IndexType *p_n_col2, IndexType *p_max_deg);
 
-template <typename Index>
-static Index find_ordering (Index n_row, Index n_col, Index Alen, Colamd_Row<Index> Row [], colamd_col<Index> Col [], Index A [], Index head [], Index n_col2, Index max_deg, Index pfree);
+template <typename IndexType>
+static IndexType find_ordering (IndexType n_row, IndexType n_col, IndexType Alen, Colamd_Row<IndexType> Row [], colamd_col<IndexType> Col [], IndexType A [], IndexType head [], IndexType n_col2, IndexType max_deg, IndexType pfree);
 
-template <typename Index>
-static void order_children (Index n_col, colamd_col<Index> Col [], Index p []);
+template <typename IndexType>
+static void order_children (IndexType n_col, colamd_col<IndexType> Col [], IndexType p []);
 
-template <typename Index>
-static void detect_super_cols (colamd_col<Index> Col [], Index A [], Index head [], Index row_start, Index row_length ) ;
+template <typename IndexType>
+static void detect_super_cols (colamd_col<IndexType> Col [], IndexType A [], IndexType head [], IndexType row_start, IndexType row_length ) ;
 
-template <typename Index>
-static Index garbage_collection (Index n_row, Index n_col, Colamd_Row<Index> Row [], colamd_col<Index> Col [], Index A [], Index *pfree) ;
+template <typename IndexType>
+static IndexType garbage_collection (IndexType n_row, IndexType n_col, Colamd_Row<IndexType> Row [], colamd_col<IndexType> Col [], IndexType A [], IndexType *pfree) ;
 
-template <typename Index>
-static inline  Index clear_mark (Index n_row, Colamd_Row<Index> Row [] ) ;
+template <typename IndexType>
+static inline  IndexType clear_mark (IndexType n_row, Colamd_Row<IndexType> Row [] ) ;
 
 /* === No debugging ========================================================= */
 
@@ -260,8 +253,8 @@ static inline  Index clear_mark (Index n_row, Colamd_Row<Index> Row [] ) ;
  * \param n_col number of columns in A
  * \return recommended value of Alen for use by colamd
  */
-template <typename Index>
-inline Index colamd_recommended ( Index nnz, Index n_row, Index n_col)
+template <typename IndexType>
+inline IndexType colamd_recommended ( IndexType nnz, IndexType n_row, IndexType n_col)
 {
   if ((nnz) < 0 || (n_row) < 0 || (n_col) < 0)
     return (-1);
@@ -325,22 +318,22 @@ static inline void colamd_set_defaults(double knobs[COLAMD_KNOBS])
  * \param knobs parameter settings for colamd
  * \param stats colamd output statistics and error codes
  */
-template <typename Index>
-static bool colamd(Index n_row, Index n_col, Index Alen, Index *A, Index *p, double knobs[COLAMD_KNOBS], Index stats[COLAMD_STATS])
+template <typename IndexType>
+static bool colamd(IndexType n_row, IndexType n_col, IndexType Alen, IndexType *A, IndexType *p, double knobs[COLAMD_KNOBS], IndexType stats[COLAMD_STATS])
 {
   /* === Local variables ================================================== */
   
-  Index i ;     /* loop index */
-  Index nnz ;     /* nonzeros in A */
-  Index Row_size ;    /* size of Row [], in integers */
-  Index Col_size ;    /* size of Col [], in integers */
-  Index need ;      /* minimum required length of A */
-  Colamd_Row<Index> *Row ;   /* pointer into A of Row [0..n_row] array */
-  colamd_col<Index> *Col ;   /* pointer into A of Col [0..n_col] array */
-  Index n_col2 ;    /* number of non-dense, non-empty columns */
-  Index n_row2 ;    /* number of non-dense, non-empty rows */
-  Index ngarbage ;    /* number of garbage collections performed */
-  Index max_deg ;   /* maximum row degree */
+  IndexType i ;     /* loop index */
+  IndexType nnz ;     /* nonzeros in A */
+  IndexType Row_size ;    /* size of Row [], in integers */
+  IndexType Col_size ;    /* size of Col [], in integers */
+  IndexType need ;      /* minimum required length of A */
+  Colamd_Row<IndexType> *Row ;   /* pointer into A of Row [0..n_row] array */
+  colamd_col<IndexType> *Col ;   /* pointer into A of Col [0..n_col] array */
+  IndexType n_col2 ;    /* number of non-dense, non-empty columns */
+  IndexType n_row2 ;    /* number of non-dense, non-empty rows */
+  IndexType ngarbage ;    /* number of garbage collections performed */
+  IndexType max_deg ;   /* maximum row degree */
   double default_knobs [COLAMD_KNOBS] ; /* default knobs array */
   
   
@@ -431,8 +424,8 @@ static bool colamd(Index n_row, Index n_col, Index Alen, Index *A, Index *p, dou
   }
   
   Alen -= Col_size + Row_size ;
-  Col = (colamd_col<Index> *) &A [Alen] ;
-  Row = (Colamd_Row<Index> *) &A [Alen + Col_size] ;
+  Col = (colamd_col<IndexType> *) &A [Alen] ;
+  Row = (Colamd_Row<IndexType> *) &A [Alen + Col_size] ;
 
   /* === Construct the row and column data structures ===================== */
   
@@ -485,29 +478,29 @@ static bool colamd(Index n_row, Index n_col, Index Alen, Index *A, Index *p, dou
   column form of the matrix.  Returns false if the matrix is invalid,
   true otherwise.  Not user-callable.
 */
-template <typename Index>
-static Index init_rows_cols  /* returns true if OK, or false otherwise */
+template <typename IndexType>
+static IndexType init_rows_cols  /* returns true if OK, or false otherwise */
   (
     /* === Parameters ======================================================= */
 
-    Index n_row,      /* number of rows of A */
-    Index n_col,      /* number of columns of A */
-    Colamd_Row<Index> Row [],    /* of size n_row+1 */
-    colamd_col<Index> Col [],    /* of size n_col+1 */
-    Index A [],     /* row indices of A, of size Alen */
-    Index p [],     /* pointers to columns in A, of size n_col+1 */
-    Index stats [COLAMD_STATS]  /* colamd statistics */ 
+    IndexType n_row,      /* number of rows of A */
+    IndexType n_col,      /* number of columns of A */
+    Colamd_Row<IndexType> Row [],    /* of size n_row+1 */
+    colamd_col<IndexType> Col [],    /* of size n_col+1 */
+    IndexType A [],     /* row indices of A, of size Alen */
+    IndexType p [],     /* pointers to columns in A, of size n_col+1 */
+    IndexType stats [COLAMD_STATS]  /* colamd statistics */ 
     )
 {
   /* === Local variables ================================================== */
 
-  Index col ;     /* a column index */
-  Index row ;     /* a row index */
-  Index *cp ;     /* a column pointer */
-  Index *cp_end ;   /* a pointer to the end of a column */
-  Index *rp ;     /* a row pointer */
-  Index *rp_end ;   /* a pointer to the end of a row */
-  Index last_row ;    /* previous row */
+  IndexType col ;     /* a column index */
+  IndexType row ;     /* a row index */
+  IndexType *cp ;     /* a column pointer */
+  IndexType *cp_end ;   /* a pointer to the end of a column */
+  IndexType *rp ;     /* a row pointer */
+  IndexType *rp_end ;   /* a pointer to the end of a row */
+  IndexType last_row ;    /* previous row */
 
   /* === Initialize columns, and check column pointers ==================== */
 
@@ -516,7 +509,7 @@ static Index init_rows_cols  /* returns true if OK, or false otherwise */
     Col [col].start = p [col] ;
     Col [col].length = p [col+1] - p [col] ;
 
-    if (Col [col].length < 0)
+    if ((Col [col].length) < 0) // extra parentheses to work-around gcc bug 10200
     {
       /* column pointers must be non-decreasing */
       stats [COLAMD_STATUS] = COLAMD_ERROR_col_length_negative ;
@@ -701,46 +694,46 @@ static Index init_rows_cols  /* returns true if OK, or false otherwise */
   Kills dense or empty columns and rows, calculates an initial score for
   each column, and places all columns in the degree lists.  Not user-callable.
 */
-template <typename Index>
+template <typename IndexType>
 static void init_scoring
   (
     /* === Parameters ======================================================= */
 
-    Index n_row,      /* number of rows of A */
-    Index n_col,      /* number of columns of A */
-    Colamd_Row<Index> Row [],    /* of size n_row+1 */
-    colamd_col<Index> Col [],    /* of size n_col+1 */
-    Index A [],     /* column form and row form of A */
-    Index head [],    /* of size n_col+1 */
+    IndexType n_row,      /* number of rows of A */
+    IndexType n_col,      /* number of columns of A */
+    Colamd_Row<IndexType> Row [],    /* of size n_row+1 */
+    colamd_col<IndexType> Col [],    /* of size n_col+1 */
+    IndexType A [],     /* column form and row form of A */
+    IndexType head [],    /* of size n_col+1 */
     double knobs [COLAMD_KNOBS],/* parameters */
-    Index *p_n_row2,    /* number of non-dense, non-empty rows */
-    Index *p_n_col2,    /* number of non-dense, non-empty columns */
-    Index *p_max_deg    /* maximum row degree */
+    IndexType *p_n_row2,    /* number of non-dense, non-empty rows */
+    IndexType *p_n_col2,    /* number of non-dense, non-empty columns */
+    IndexType *p_max_deg    /* maximum row degree */
     )
 {
   /* === Local variables ================================================== */
 
-  Index c ;     /* a column index */
-  Index r, row ;    /* a row index */
-  Index *cp ;     /* a column pointer */
-  Index deg ;     /* degree of a row or column */
-  Index *cp_end ;   /* a pointer to the end of a column */
-  Index *new_cp ;   /* new column pointer */
-  Index col_length ;    /* length of pruned column */
-  Index score ;     /* current column score */
-  Index n_col2 ;    /* number of non-dense, non-empty columns */
-  Index n_row2 ;    /* number of non-dense, non-empty rows */
-  Index dense_row_count ; /* remove rows with more entries than this */
-  Index dense_col_count ; /* remove cols with more entries than this */
-  Index min_score ;   /* smallest column score */
-  Index max_deg ;   /* maximum row degree */
-  Index next_col ;    /* Used to add to degree list.*/
+  IndexType c ;     /* a column index */
+  IndexType r, row ;    /* a row index */
+  IndexType *cp ;     /* a column pointer */
+  IndexType deg ;     /* degree of a row or column */
+  IndexType *cp_end ;   /* a pointer to the end of a column */
+  IndexType *new_cp ;   /* new column pointer */
+  IndexType col_length ;    /* length of pruned column */
+  IndexType score ;     /* current column score */
+  IndexType n_col2 ;    /* number of non-dense, non-empty columns */
+  IndexType n_row2 ;    /* number of non-dense, non-empty rows */
+  IndexType dense_row_count ; /* remove rows with more entries than this */
+  IndexType dense_col_count ; /* remove cols with more entries than this */
+  IndexType min_score ;   /* smallest column score */
+  IndexType max_deg ;   /* maximum row degree */
+  IndexType next_col ;    /* Used to add to degree list.*/
 
 
   /* === Extract knobs ==================================================== */
 
-  dense_row_count = COLAMD_MAX (0, COLAMD_MIN (knobs [COLAMD_DENSE_ROW] * n_col, n_col)) ;
-  dense_col_count = COLAMD_MAX (0, COLAMD_MIN (knobs [COLAMD_DENSE_COL] * n_row, n_row)) ;
+  dense_row_count = numext::maxi(IndexType(0), numext::mini(IndexType(knobs [COLAMD_DENSE_ROW] * n_col), n_col)) ;
+  dense_col_count = numext::maxi(IndexType(0), numext::mini(IndexType(knobs [COLAMD_DENSE_COL] * n_row), n_row)) ;
   COLAMD_DEBUG1 (("colamd: densecount: %d %d\n", dense_row_count, dense_col_count)) ;
   max_deg = 0 ;
   n_col2 = n_col ;
@@ -804,7 +797,7 @@ static void init_scoring
     else
     {
       /* keep track of max degree of remaining rows */
-      max_deg = COLAMD_MAX (max_deg, deg) ;
+      max_deg = numext::maxi(max_deg, deg) ;
     }
   }
   COLAMD_DEBUG1 (("colamd: Dense and null rows killed: %d\n", n_row - n_row2)) ;
@@ -842,10 +835,10 @@ static void init_scoring
       /* add row's external degree */
       score += Row [row].shared1.degree - 1 ;
       /* guard against integer overflow */
-      score = COLAMD_MIN (score, n_col) ;
+      score = numext::mini(score, n_col) ;
     }
     /* determine pruned column length */
-    col_length = (Index) (new_cp - &A [Col [c].start]) ;
+    col_length = (IndexType) (new_cp - &A [Col [c].start]) ;
     if (col_length == 0)
     {
       /* a newly-made null column (all rows in this col are "dense" */
@@ -914,7 +907,7 @@ static void init_scoring
       head [score] = c ;
 
       /* see if this score is less than current min */
-      min_score = COLAMD_MIN (min_score, score) ;
+      min_score = numext::mini(min_score, score) ;
 
 
     }
@@ -938,56 +931,56 @@ static void init_scoring
   (no supercolumns on input).  Uses a minimum approximate column minimum
   degree ordering method.  Not user-callable.
 */
-template <typename Index>
-static Index find_ordering /* return the number of garbage collections */
+template <typename IndexType>
+static IndexType find_ordering /* return the number of garbage collections */
   (
     /* === Parameters ======================================================= */
 
-    Index n_row,      /* number of rows of A */
-    Index n_col,      /* number of columns of A */
-    Index Alen,     /* size of A, 2*nnz + n_col or larger */
-    Colamd_Row<Index> Row [],    /* of size n_row+1 */
-    colamd_col<Index> Col [],    /* of size n_col+1 */
-    Index A [],     /* column form and row form of A */
-    Index head [],    /* of size n_col+1 */
-    Index n_col2,     /* Remaining columns to order */
-    Index max_deg,    /* Maximum row degree */
-    Index pfree     /* index of first free slot (2*nnz on entry) */
+    IndexType n_row,      /* number of rows of A */
+    IndexType n_col,      /* number of columns of A */
+    IndexType Alen,     /* size of A, 2*nnz + n_col or larger */
+    Colamd_Row<IndexType> Row [],    /* of size n_row+1 */
+    colamd_col<IndexType> Col [],    /* of size n_col+1 */
+    IndexType A [],     /* column form and row form of A */
+    IndexType head [],    /* of size n_col+1 */
+    IndexType n_col2,     /* Remaining columns to order */
+    IndexType max_deg,    /* Maximum row degree */
+    IndexType pfree     /* index of first free slot (2*nnz on entry) */
     )
 {
   /* === Local variables ================================================== */
 
-  Index k ;     /* current pivot ordering step */
-  Index pivot_col ;   /* current pivot column */
-  Index *cp ;     /* a column pointer */
-  Index *rp ;     /* a row pointer */
-  Index pivot_row ;   /* current pivot row */
-  Index *new_cp ;   /* modified column pointer */
-  Index *new_rp ;   /* modified row pointer */
-  Index pivot_row_start ; /* pointer to start of pivot row */
-  Index pivot_row_degree ;  /* number of columns in pivot row */
-  Index pivot_row_length ;  /* number of supercolumns in pivot row */
-  Index pivot_col_score ; /* score of pivot column */
-  Index needed_memory ;   /* free space needed for pivot row */
-  Index *cp_end ;   /* pointer to the end of a column */
-  Index *rp_end ;   /* pointer to the end of a row */
-  Index row ;     /* a row index */
-  Index col ;     /* a column index */
-  Index max_score ;   /* maximum possible score */
-  Index cur_score ;   /* score of current column */
+  IndexType k ;     /* current pivot ordering step */
+  IndexType pivot_col ;   /* current pivot column */
+  IndexType *cp ;     /* a column pointer */
+  IndexType *rp ;     /* a row pointer */
+  IndexType pivot_row ;   /* current pivot row */
+  IndexType *new_cp ;   /* modified column pointer */
+  IndexType *new_rp ;   /* modified row pointer */
+  IndexType pivot_row_start ; /* pointer to start of pivot row */
+  IndexType pivot_row_degree ;  /* number of columns in pivot row */
+  IndexType pivot_row_length ;  /* number of supercolumns in pivot row */
+  IndexType pivot_col_score ; /* score of pivot column */
+  IndexType needed_memory ;   /* free space needed for pivot row */
+  IndexType *cp_end ;   /* pointer to the end of a column */
+  IndexType *rp_end ;   /* pointer to the end of a row */
+  IndexType row ;     /* a row index */
+  IndexType col ;     /* a column index */
+  IndexType max_score ;   /* maximum possible score */
+  IndexType cur_score ;   /* score of current column */
   unsigned int hash ;   /* hash value for supernode detection */
-  Index head_column ;   /* head of hash bucket */
-  Index first_col ;   /* first column in hash bucket */
-  Index tag_mark ;    /* marker value for mark array */
-  Index row_mark ;    /* Row [row].shared2.mark */
-  Index set_difference ;  /* set difference size of row with pivot row */
-  Index min_score ;   /* smallest column score */
-  Index col_thickness ;   /* "thickness" (no. of columns in a supercol) */
-  Index max_mark ;    /* maximum value of tag_mark */
-  Index pivot_col_thickness ; /* number of columns represented by pivot col */
-  Index prev_col ;    /* Used by Dlist operations. */
-  Index next_col ;    /* Used by Dlist operations. */
-  Index ngarbage ;    /* number of garbage collections performed */
+  IndexType head_column ;   /* head of hash bucket */
+  IndexType first_col ;   /* first column in hash bucket */
+  IndexType tag_mark ;    /* marker value for mark array */
+  IndexType row_mark ;    /* Row [row].shared2.mark */
+  IndexType set_difference ;  /* set difference size of row with pivot row */
+  IndexType min_score ;   /* smallest column score */
+  IndexType col_thickness ;   /* "thickness" (no. of columns in a supercol) */
+  IndexType max_mark ;    /* maximum value of tag_mark */
+  IndexType pivot_col_thickness ; /* number of columns represented by pivot col */
+  IndexType prev_col ;    /* Used by Dlist operations. */
+  IndexType next_col ;    /* Used by Dlist operations. */
+  IndexType ngarbage ;    /* number of garbage collections performed */
 
 
   /* === Initialization and clear mark ==================================== */
@@ -1011,7 +1004,7 @@ static Index find_ordering /* return the number of garbage collections */
     COLAMD_ASSERT (head [min_score] >= COLAMD_EMPTY) ;
 
     /* get pivot column from head of minimum degree list */
-    while (head [min_score] == COLAMD_EMPTY && min_score < n_col)
+    while (min_score < n_col && head [min_score] == COLAMD_EMPTY)
     {
       min_score++ ;
     }
@@ -1040,7 +1033,7 @@ static Index find_ordering /* return the number of garbage collections */
 
     /* === Garbage_collection, if necessary ============================= */
 
-    needed_memory = COLAMD_MIN (pivot_col_score, n_col - k) ;
+    needed_memory = numext::mini(pivot_col_score, n_col - k) ;
     if (pfree + needed_memory >= Alen)
     {
       pfree = Eigen::internal::garbage_collection (n_row, n_col, Row, Col, A, &A [pfree]) ;
@@ -1099,7 +1092,7 @@ static Index find_ordering /* return the number of garbage collections */
 
     /* clear tag on pivot column */
     Col [pivot_col].shared1.thickness = pivot_col_thickness ;
-    max_deg = COLAMD_MAX (max_deg, pivot_row_degree) ;
+    max_deg = numext::maxi(max_deg, pivot_row_degree) ;
 
 
     /* === Kill all rows used to construct pivot row ==================== */
@@ -1273,11 +1266,11 @@ static Index find_ordering /* return the number of garbage collections */
 	/* add set difference */
 	cur_score += row_mark - tag_mark ;
 	/* integer overflow... */
-	cur_score = COLAMD_MIN (cur_score, n_col) ;
+	cur_score = numext::mini(cur_score, n_col) ;
       }
 
       /* recompute the column's length */
-      Col [col].length = (Index) (new_cp - &A [Col [col].start]) ;
+      Col [col].length = (IndexType) (new_cp - &A [Col [col].start]) ;
 
       /* === Further mass elimination ================================= */
 
@@ -1325,7 +1318,7 @@ static Index find_ordering /* return the number of garbage collections */
 	Col [col].shared4.hash_next = first_col ;
 
 	/* save hash function in Col [col].shared3.hash */
-	Col [col].shared3.hash = (Index) hash ;
+	Col [col].shared3.hash = (IndexType) hash ;
 	COLAMD_ASSERT (COL_IS_ALIVE (col)) ;
       }
     }
@@ -1386,7 +1379,7 @@ static Index find_ordering /* return the number of garbage collections */
       cur_score -= Col [col].shared1.thickness ;
 
       /* make sure score is less or equal than the max score */
-      cur_score = COLAMD_MIN (cur_score, max_score) ;
+      cur_score = numext::mini(cur_score, max_score) ;
       COLAMD_ASSERT (cur_score >= 0) ;
 
       /* store updated score */
@@ -1409,7 +1402,7 @@ static Index find_ordering /* return the number of garbage collections */
       head [cur_score] = col ;
 
       /* see if this score is less than current min */
-      min_score = COLAMD_MIN (min_score, cur_score) ;
+      min_score = numext::mini(min_score, cur_score) ;
 
     }
 
@@ -1420,7 +1413,7 @@ static Index find_ordering /* return the number of garbage collections */
       /* update pivot row length to reflect any cols that were killed */
       /* during super-col detection and mass elimination */
       Row [pivot_row].start  = pivot_row_start ;
-      Row [pivot_row].length = (Index) (new_rp - &A[pivot_row_start]) ;
+      Row [pivot_row].length = (IndexType) (new_rp - &A[pivot_row_start]) ;
       Row [pivot_row].shared1.degree = pivot_row_degree ;
       Row [pivot_row].shared2.mark = 0 ;
       /* pivot row is no longer dead */
@@ -1449,22 +1442,22 @@ static Index find_ordering /* return the number of garbage collections */
   taken by this routine is O (n_col), that is, linear in the number of
   columns.  Not user-callable.
 */
-template <typename Index>
+template <typename IndexType>
 static inline  void order_children
 (
   /* === Parameters ======================================================= */
 
-  Index n_col,      /* number of columns of A */
-  colamd_col<Index> Col [],    /* of size n_col+1 */
-  Index p []      /* p [0 ... n_col-1] is the column permutation*/
+  IndexType n_col,      /* number of columns of A */
+  colamd_col<IndexType> Col [],    /* of size n_col+1 */
+  IndexType p []      /* p [0 ... n_col-1] is the column permutation*/
   )
 {
   /* === Local variables ================================================== */
 
-  Index i ;     /* loop counter for all columns */
-  Index c ;     /* column index */
-  Index parent ;    /* index of column's parent */
-  Index order ;     /* column's order */
+  IndexType i ;     /* loop counter for all columns */
+  IndexType c ;     /* column index */
+  IndexType parent ;    /* index of column's parent */
+  IndexType order ;     /* column's order */
 
   /* === Order each non-principal column ================================== */
 
@@ -1550,33 +1543,33 @@ static inline  void order_children
   just been computed in the approximate degree computation.
   Not user-callable.
 */
-template <typename Index>
+template <typename IndexType>
 static void detect_super_cols
 (
   /* === Parameters ======================================================= */
   
-  colamd_col<Index> Col [],    /* of size n_col+1 */
-  Index A [],     /* row indices of A */
-  Index head [],    /* head of degree lists and hash buckets */
-  Index row_start,    /* pointer to set of columns to check */
-  Index row_length    /* number of columns to check */
+  colamd_col<IndexType> Col [],    /* of size n_col+1 */
+  IndexType A [],     /* row indices of A */
+  IndexType head [],    /* head of degree lists and hash buckets */
+  IndexType row_start,    /* pointer to set of columns to check */
+  IndexType row_length    /* number of columns to check */
 )
 {
   /* === Local variables ================================================== */
 
-  Index hash ;      /* hash value for a column */
-  Index *rp ;     /* pointer to a row */
-  Index c ;     /* a column index */
-  Index super_c ;   /* column index of the column to absorb into */
-  Index *cp1 ;      /* column pointer for column super_c */
-  Index *cp2 ;      /* column pointer for column c */
-  Index length ;    /* length of column super_c */
-  Index prev_c ;    /* column preceding c in hash bucket */
-  Index i ;     /* loop counter */
-  Index *rp_end ;   /* pointer to the end of the row */
-  Index col ;     /* a column index in the row to check */
-  Index head_column ;   /* first column in hash bucket or degree list */
-  Index first_col ;   /* first column in hash bucket */
+  IndexType hash ;      /* hash value for a column */
+  IndexType *rp ;     /* pointer to a row */
+  IndexType c ;     /* a column index */
+  IndexType super_c ;   /* column index of the column to absorb into */
+  IndexType *cp1 ;      /* column pointer for column super_c */
+  IndexType *cp2 ;      /* column pointer for column c */
+  IndexType length ;    /* length of column super_c */
+  IndexType prev_c ;    /* column preceding c in hash bucket */
+  IndexType i ;     /* loop counter */
+  IndexType *rp_end ;   /* pointer to the end of the row */
+  IndexType col ;     /* a column index in the row to check */
+  IndexType head_column ;   /* first column in hash bucket or degree list */
+  IndexType first_col ;   /* first column in hash bucket */
 
   /* === Consider each column in the row ================================== */
 
@@ -1701,27 +1694,27 @@ static void detect_super_cols
   itself linear in the number of nonzeros in the input matrix.
   Not user-callable.
 */
-template <typename Index>
-static Index garbage_collection  /* returns the new value of pfree */
+template <typename IndexType>
+static IndexType garbage_collection  /* returns the new value of pfree */
   (
     /* === Parameters ======================================================= */
     
-    Index n_row,      /* number of rows */
-    Index n_col,      /* number of columns */
-    Colamd_Row<Index> Row [],    /* row info */
-    colamd_col<Index> Col [],    /* column info */
-    Index A [],     /* A [0 ... Alen-1] holds the matrix */
-    Index *pfree      /* &A [0] ... pfree is in use */
+    IndexType n_row,      /* number of rows */
+    IndexType n_col,      /* number of columns */
+    Colamd_Row<IndexType> Row [],    /* row info */
+    colamd_col<IndexType> Col [],    /* column info */
+    IndexType A [],     /* A [0 ... Alen-1] holds the matrix */
+    IndexType *pfree      /* &A [0] ... pfree is in use */
     )
 {
   /* === Local variables ================================================== */
 
-  Index *psrc ;     /* source pointer */
-  Index *pdest ;    /* destination pointer */
-  Index j ;     /* counter */
-  Index r ;     /* a row index */
-  Index c ;     /* a column index */
-  Index length ;    /* length of a row or column */
+  IndexType *psrc ;     /* source pointer */
+  IndexType *pdest ;    /* destination pointer */
+  IndexType j ;     /* counter */
+  IndexType r ;     /* a row index */
+  IndexType c ;     /* a column index */
+  IndexType length ;    /* length of a row or column */
 
   /* === Defragment the columns =========================================== */
 
@@ -1734,7 +1727,7 @@ static Index garbage_collection  /* returns the new value of pfree */
 
       /* move and compact the column */
       COLAMD_ASSERT (pdest <= psrc) ;
-      Col [c].start = (Index) (pdest - &A [0]) ;
+      Col [c].start = (IndexType) (pdest - &A [0]) ;
       length = Col [c].length ;
       for (j = 0 ; j < length ; j++)
       {
@@ -1744,7 +1737,7 @@ static Index garbage_collection  /* returns the new value of pfree */
 	  *pdest++ = r ;
 	}
       }
-      Col [c].length = (Index) (pdest - &A [Col [c].start]) ;
+      Col [c].length = (IndexType) (pdest - &A [Col [c].start]) ;
     }
   }
 
@@ -1791,7 +1784,7 @@ static Index garbage_collection  /* returns the new value of pfree */
 
       /* move and compact the row */
       COLAMD_ASSERT (pdest <= psrc) ;
-      Row [r].start = (Index) (pdest - &A [0]) ;
+      Row [r].start = (IndexType) (pdest - &A [0]) ;
       length = Row [r].length ;
       for (j = 0 ; j < length ; j++)
       {
@@ -1801,7 +1794,7 @@ static Index garbage_collection  /* returns the new value of pfree */
 	  *pdest++ = c ;
 	}
       }
-      Row [r].length = (Index) (pdest - &A [Row [r].start]) ;
+      Row [r].length = (IndexType) (pdest - &A [Row [r].start]) ;
 
     }
   }
@@ -1810,7 +1803,7 @@ static Index garbage_collection  /* returns the new value of pfree */
 
   /* === Return the new value of pfree ==================================== */
 
-  return ((Index) (pdest - &A [0])) ;
+  return ((IndexType) (pdest - &A [0])) ;
 }
 
 
@@ -1822,18 +1815,18 @@ static Index garbage_collection  /* returns the new value of pfree */
   Clears the Row [].shared2.mark array, and returns the new tag_mark.
   Return value is the new tag_mark.  Not user-callable.
 */
-template <typename Index>
-static inline  Index clear_mark  /* return the new value for tag_mark */
+template <typename IndexType>
+static inline  IndexType clear_mark  /* return the new value for tag_mark */
   (
       /* === Parameters ======================================================= */
 
-    Index n_row,    /* number of rows in A */
-    Colamd_Row<Index> Row [] /* Row [0 ... n_row-1].shared2.mark is set to zero */
+    IndexType n_row,    /* number of rows in A */
+    Colamd_Row<IndexType> Row [] /* Row [0 ... n_row-1].shared2.mark is set to zero */
     )
 {
   /* === Local variables ================================================== */
 
-  Index r ;
+  IndexType r ;
 
   for (r = 0 ; r < n_row ; r++)
   {
diff --git a/extern/Eigen3/Eigen/src/OrderingMethods/Ordering.h b/extern/Eigen3/Eigen/src/OrderingMethods/Ordering.h
index f3c31f9cbfc..7ea9b14d7eb 100644
--- a/extern/Eigen3/Eigen/src/OrderingMethods/Ordering.h
+++ b/extern/Eigen3/Eigen/src/OrderingMethods/Ordering.h
@@ -19,20 +19,21 @@ namespace internal {
     
 /** \internal
   * \ingroup OrderingMethods_Module
-  * \returns the symmetric pattern A^T+A from the input matrix A. 
+  * \param[in] A the input non-symmetric matrix
+  * \param[out] symmat the symmetric pattern A^T+A from the input matrix \a A.
   * FIXME: The values should not be considered here
   */
 template<typename MatrixType> 
-void ordering_helper_at_plus_a(const MatrixType& mat, MatrixType& symmat)
+void ordering_helper_at_plus_a(const MatrixType& A, MatrixType& symmat)
 {
   MatrixType C;
-  C = mat.transpose(); // NOTE: Could be  costly
+  C = A.transpose(); // NOTE: Could be  costly
   for (int i = 0; i < C.rows(); i++) 
   {
       for (typename MatrixType::InnerIterator it(C, i); it; ++it)
         it.valueRef() = 0.0;
   }
-  symmat = C + mat; 
+  symmat = C + A;
 }
     
 }
@@ -44,14 +45,14 @@ void ordering_helper_at_plus_a(const MatrixType& mat, MatrixType& symmat)
   *
   * Functor computing the \em approximate \em minimum \em degree ordering
   * If the matrix is not structurally symmetric, an ordering of A^T+A is computed
-  * \tparam  Index The type of indices of the matrix 
+  * \tparam  StorageIndex The type of indices of the matrix 
   * \sa COLAMDOrdering
   */
-template <typename Index>
+template <typename StorageIndex>
 class AMDOrdering
 {
   public:
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
     
     /** Compute the permutation vector from a sparse matrix
      * This routine is much faster if the input matrix is column-major     
@@ -60,7 +61,7 @@ class AMDOrdering
     void operator()(const MatrixType& mat, PermutationType& perm)
     {
       // Compute the symmetric pattern
-      SparseMatrix<typename MatrixType::Scalar, ColMajor, Index> symm;
+      SparseMatrix<typename MatrixType::Scalar, ColMajor, StorageIndex> symm;
       internal::ordering_helper_at_plus_a(mat,symm); 
     
       // Call the AMD routine 
@@ -72,7 +73,7 @@ class AMDOrdering
     template <typename SrcType, unsigned int SrcUpLo> 
     void operator()(const SparseSelfAdjointView<SrcType, SrcUpLo>& mat, PermutationType& perm)
     { 
-      SparseMatrix<typename SrcType::Scalar, ColMajor, Index> C; C = mat;
+      SparseMatrix<typename SrcType::Scalar, ColMajor, StorageIndex> C; C = mat;
       
       // Call the AMD routine 
       // m_mat.prune(keep_diag()); //Remove the diagonal elements 
@@ -88,13 +89,13 @@ class AMDOrdering
   * Functor computing the natural ordering (identity)
   * 
   * \note Returns an empty permutation matrix
-  * \tparam  Index The type of indices of the matrix 
+  * \tparam  StorageIndex The type of indices of the matrix 
   */
-template <typename Index>
+template <typename StorageIndex>
 class NaturalOrdering
 {
   public:
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
     
     /** Compute the permutation vector from a column-major sparse matrix */
     template <typename MatrixType>
@@ -108,15 +109,17 @@ class NaturalOrdering
 /** \ingroup OrderingMethods_Module
   * \class COLAMDOrdering
   *
+  * \tparam  StorageIndex The type of indices of the matrix 
+  * 
   * Functor computing the \em column \em approximate \em minimum \em degree ordering 
   * The matrix should be in column-major and \b compressed format (see SparseMatrix::makeCompressed()).
   */
-template<typename Index>
+template<typename StorageIndex>
 class COLAMDOrdering
 {
   public:
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType; 
-    typedef Matrix<Index, Dynamic, 1> IndexVector;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType; 
+    typedef Matrix<StorageIndex, Dynamic, 1> IndexVector;
     
     /** Compute the permutation vector \a perm form the sparse matrix \a mat
       * \warning The input sparse matrix \a mat must be in compressed mode (see SparseMatrix::makeCompressed()).
@@ -126,26 +129,26 @@ class COLAMDOrdering
     {
       eigen_assert(mat.isCompressed() && "COLAMDOrdering requires a sparse matrix in compressed mode. Call .makeCompressed() before passing it to COLAMDOrdering");
       
-      Index m = mat.rows();
-      Index n = mat.cols();
-      Index nnz = mat.nonZeros();
+      StorageIndex m = StorageIndex(mat.rows());
+      StorageIndex n = StorageIndex(mat.cols());
+      StorageIndex nnz = StorageIndex(mat.nonZeros());
       // Get the recommended value of Alen to be used by colamd
-      Index Alen = internal::colamd_recommended(nnz, m, n); 
+      StorageIndex Alen = internal::colamd_recommended(nnz, m, n); 
       // Set the default parameters
       double knobs [COLAMD_KNOBS]; 
-      Index stats [COLAMD_STATS];
+      StorageIndex stats [COLAMD_STATS];
       internal::colamd_set_defaults(knobs);
       
       IndexVector p(n+1), A(Alen); 
-      for(Index i=0; i <= n; i++)   p(i) = mat.outerIndexPtr()[i];
-      for(Index i=0; i < nnz; i++)  A(i) = mat.innerIndexPtr()[i];
+      for(StorageIndex i=0; i <= n; i++)   p(i) = mat.outerIndexPtr()[i];
+      for(StorageIndex i=0; i < nnz; i++)  A(i) = mat.innerIndexPtr()[i];
       // Call Colamd routine to compute the ordering 
-      Index info = internal::colamd(m, n, Alen, A.data(), p.data(), knobs, stats); 
+      StorageIndex info = internal::colamd(m, n, Alen, A.data(), p.data(), knobs, stats); 
       EIGEN_UNUSED_VARIABLE(info);
       eigen_assert( info && "COLAMD failed " );
       
       perm.resize(n);
-      for (Index i = 0; i < n; i++) perm.indices()(p(i)) = i;
+      for (StorageIndex i = 0; i < n; i++) perm.indices()(p(i)) = i;
     }
 };
 
diff --git a/extern/Eigen3/Eigen/src/PaStiXSupport/PaStiXSupport.h b/extern/Eigen3/Eigen/src/PaStiXSupport/PaStiXSupport.h
index a955287d1c9..160d8a5234a 100644
--- a/extern/Eigen3/Eigen/src/PaStiXSupport/PaStiXSupport.h
+++ b/extern/Eigen3/Eigen/src/PaStiXSupport/PaStiXSupport.h
@@ -12,6 +12,14 @@
 
 namespace Eigen { 
 
+#if defined(DCOMPLEX)
+  #define PASTIX_COMPLEX  COMPLEX
+  #define PASTIX_DCOMPLEX DCOMPLEX
+#else
+  #define PASTIX_COMPLEX  std::complex<float>
+  #define PASTIX_DCOMPLEX std::complex<double>
+#endif
+
 /** \ingroup PaStiXSupport_Module
   * \brief Interface to the PaStix solver
   * 
@@ -35,7 +43,7 @@ namespace internal
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
 
   template<typename _MatrixType, int Options>
@@ -44,7 +52,7 @@ namespace internal
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
 
   template<typename _MatrixType, int Options>
@@ -53,35 +61,35 @@ namespace internal
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
   
-  void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, float *vals, int *perm, int * invp, float *x, int nbrhs, int *iparm, double *dparm)
+  inline void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, float *vals, int *perm, int * invp, float *x, int nbrhs, int *iparm, double *dparm)
   {
     if (n == 0) { ptr = NULL; idx = NULL; vals = NULL; }
     if (nbrhs == 0) {x = NULL; nbrhs=1;}
     s_pastix(pastix_data, pastix_comm, n, ptr, idx, vals, perm, invp, x, nbrhs, iparm, dparm); 
   }
   
-  void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, double *vals, int *perm, int * invp, double *x, int nbrhs, int *iparm, double *dparm)
+  inline void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, double *vals, int *perm, int * invp, double *x, int nbrhs, int *iparm, double *dparm)
   {
     if (n == 0) { ptr = NULL; idx = NULL; vals = NULL; }
     if (nbrhs == 0) {x = NULL; nbrhs=1;}
     d_pastix(pastix_data, pastix_comm, n, ptr, idx, vals, perm, invp, x, nbrhs, iparm, dparm); 
   }
   
-  void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, std::complex<float> *vals, int *perm, int * invp, std::complex<float> *x, int nbrhs, int *iparm, double *dparm)
+  inline void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, std::complex<float> *vals, int *perm, int * invp, std::complex<float> *x, int nbrhs, int *iparm, double *dparm)
   {
     if (n == 0) { ptr = NULL; idx = NULL; vals = NULL; }
     if (nbrhs == 0) {x = NULL; nbrhs=1;}
-    c_pastix(pastix_data, pastix_comm, n, ptr, idx, reinterpret_cast<COMPLEX*>(vals), perm, invp, reinterpret_cast<COMPLEX*>(x), nbrhs, iparm, dparm); 
+    c_pastix(pastix_data, pastix_comm, n, ptr, idx, reinterpret_cast<PASTIX_COMPLEX*>(vals), perm, invp, reinterpret_cast<PASTIX_COMPLEX*>(x), nbrhs, iparm, dparm); 
   }
   
-  void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, std::complex<double> *vals, int *perm, int * invp, std::complex<double> *x, int nbrhs, int *iparm, double *dparm)
+  inline void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, std::complex<double> *vals, int *perm, int * invp, std::complex<double> *x, int nbrhs, int *iparm, double *dparm)
   {
     if (n == 0) { ptr = NULL; idx = NULL; vals = NULL; }
     if (nbrhs == 0) {x = NULL; nbrhs=1;}
-    z_pastix(pastix_data, pastix_comm, n, ptr, idx, reinterpret_cast<DCOMPLEX*>(vals), perm, invp, reinterpret_cast<DCOMPLEX*>(x), nbrhs, iparm, dparm); 
+    z_pastix(pastix_data, pastix_comm, n, ptr, idx, reinterpret_cast<PASTIX_DCOMPLEX*>(vals), perm, invp, reinterpret_cast<PASTIX_DCOMPLEX*>(x), nbrhs, iparm, dparm); 
   }
 
   // Convert the matrix  to Fortran-style Numbering
@@ -117,20 +125,30 @@ namespace internal
 // This is the base class to interface with PaStiX functions. 
 // Users should not used this class directly. 
 template <class Derived>
-class PastixBase : internal::noncopyable
+class PastixBase : public SparseSolverBase<Derived>
 {
+  protected:
+    typedef SparseSolverBase<Derived> Base;
+    using Base::derived;
+    using Base::m_isInitialized;
   public:
+    using Base::_solve_impl;
+    
     typedef typename internal::pastix_traits<Derived>::MatrixType _MatrixType;
     typedef _MatrixType MatrixType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar,Dynamic,1> Vector;
     typedef SparseMatrix<Scalar, ColMajor> ColSpMatrix;
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
     
   public:
     
-    PastixBase() : m_initisOk(false), m_analysisIsOk(false), m_factorizationIsOk(false), m_isInitialized(false), m_pastixdata(0), m_size(0)
+    PastixBase() : m_initisOk(false), m_analysisIsOk(false), m_factorizationIsOk(false), m_pastixdata(0), m_size(0)
     {
       init();
     }
@@ -139,39 +157,16 @@ class PastixBase : internal::noncopyable
     {
       clean();
     }
-
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<PastixBase, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "Pastix solver is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "PastixBase::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<PastixBase, Rhs>(*this, b.derived());
-    }
     
     template<typename Rhs,typename Dest>
-    bool _solve (const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) const;
+    bool _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) const;
     
-    Derived& derived()
-    {
-      return *static_cast<Derived*>(this);
-    }
-    const Derived& derived() const
-    {
-      return *static_cast<const Derived*>(this);
-    }
-
     /** Returns a reference to the integer vector IPARM of PaStiX parameters
       * to modify the default parameters. 
       * The statistics related to the different phases of factorization and solve are saved here as well
       * \sa analyzePattern() factorize()
       */
-    Array<Index,IPARM_SIZE,1>& iparm()
+    Array<StorageIndex,IPARM_SIZE,1>& iparm()
     {
       return m_iparm; 
     }
@@ -189,7 +184,7 @@ class PastixBase : internal::noncopyable
       * The statistics related to the different phases of factorization and solve are saved here as well
       * \sa analyzePattern() factorize()
       */
-    Array<RealScalar,IPARM_SIZE,1>& dparm()
+    Array<double,DPARM_SIZE,1>& dparm()
     {
       return m_dparm; 
     }
@@ -220,20 +215,6 @@ class PastixBase : internal::noncopyable
       return m_info;
     }
     
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<PastixBase, Rhs>
-    solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "Pastix LU, LLT or LDLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "PastixBase::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<PastixBase, Rhs>(*this, b.derived());
-    }
-    
   protected:
 
     // Initialize the Pastix data structure, check the matrix
@@ -260,14 +241,13 @@ class PastixBase : internal::noncopyable
     int m_initisOk; 
     int m_analysisIsOk;
     int m_factorizationIsOk;
-    bool m_isInitialized;
     mutable ComputationInfo m_info; 
     mutable pastix_data_t *m_pastixdata; // Data structure for pastix
     mutable int m_comm; // The MPI communicator identifier
-    mutable Matrix<int,IPARM_SIZE,1> m_iparm; // integer vector for the input parameters
-    mutable Matrix<double,DPARM_SIZE,1> m_dparm; // Scalar vector for the input parameters
-    mutable Matrix<Index,Dynamic,1> m_perm;  // Permutation vector
-    mutable Matrix<Index,Dynamic,1> m_invp;  // Inverse permutation vector
+    mutable Array<int,IPARM_SIZE,1> m_iparm; // integer vector for the input parameters
+    mutable Array<double,DPARM_SIZE,1> m_dparm; // Scalar vector for the input parameters
+    mutable Matrix<StorageIndex,Dynamic,1> m_perm;  // Permutation vector
+    mutable Matrix<StorageIndex,Dynamic,1> m_invp;  // Inverse permutation vector
     mutable int m_size; // Size of the matrix 
 }; 
 
@@ -288,7 +268,7 @@ void PastixBase<Derived>::init()
          0, 0, 0, 1, m_iparm.data(), m_dparm.data());
   
   m_iparm[IPARM_MATRIX_VERIFICATION] = API_NO;
-  m_iparm[IPARM_VERBOSE]             = 2;
+  m_iparm[IPARM_VERBOSE]             = API_VERBOSE_NOT;
   m_iparm[IPARM_ORDERING]            = API_ORDER_SCOTCH;
   m_iparm[IPARM_INCOMPLETE]          = API_NO;
   m_iparm[IPARM_OOC_LIMIT]           = 2000;
@@ -320,7 +300,6 @@ void PastixBase<Derived>::compute(ColSpMatrix& mat)
   factorize(mat);
   
   m_iparm(IPARM_MATRIX_VERIFICATION) = API_NO;
-  m_isInitialized = m_factorizationIsOk;
 }
 
 
@@ -333,7 +312,7 @@ void PastixBase<Derived>::analyzePattern(ColSpMatrix& mat)
   if(m_size>0)
     clean();
   
-  m_size = mat.rows();
+  m_size = internal::convert_index<int>(mat.rows());
   m_perm.resize(m_size);
   m_invp.resize(m_size);
   
@@ -362,7 +341,7 @@ void PastixBase<Derived>::factorize(ColSpMatrix& mat)
   eigen_assert(m_analysisIsOk && "The analysis phase should be called before the factorization phase");
   m_iparm(IPARM_START_TASK) = API_TASK_NUMFACT;
   m_iparm(IPARM_END_TASK) = API_TASK_NUMFACT;
-  m_size = mat.rows();
+  m_size = internal::convert_index<int>(mat.rows());
   
   internal::eigen_pastix(&m_pastixdata, MPI_COMM_WORLD, m_size, mat.outerIndexPtr(), mat.innerIndexPtr(),
                mat.valuePtr(), m_perm.data(), m_invp.data(), 0, 0, m_iparm.data(), m_dparm.data());
@@ -385,7 +364,7 @@ void PastixBase<Derived>::factorize(ColSpMatrix& mat)
 /* Solve the system */
 template<typename Base>
 template<typename Rhs,typename Dest>
-bool PastixBase<Base>::_solve (const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) const
+bool PastixBase<Base>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) const
 {
   eigen_assert(m_isInitialized && "The matrix should be factorized first");
   EIGEN_STATIC_ASSERT((Dest::Flags&RowMajorBit)==0,
@@ -398,7 +377,7 @@ bool PastixBase<Base>::_solve (const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) co
     m_iparm[IPARM_START_TASK]          = API_TASK_SOLVE;
     m_iparm[IPARM_END_TASK]            = API_TASK_REFINE;
   
-    internal::eigen_pastix(&m_pastixdata, MPI_COMM_WORLD, x.rows(), 0, 0, 0,
+    internal::eigen_pastix(&m_pastixdata, MPI_COMM_WORLD, internal::convert_index<int>(x.rows()), 0, 0, 0,
                            m_perm.data(), m_invp.data(), &x(0, i), rhs, m_iparm.data(), m_dparm.data());
   }
   
@@ -423,8 +402,10 @@ bool PastixBase<Base>::_solve (const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) co
   * NOTE : Note that if the analysis and factorization phase are called separately, 
   * the input matrix will be symmetrized at each call, hence it is advised to 
   * symmetrize the matrix in a end-user program and set \p IsStrSym to true
-  * 
-  * \sa \ref TutorialSparseDirectSolvers
+  *
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SparseLU
   * 
   */
 template<typename _MatrixType, bool IsStrSym>
@@ -434,7 +415,7 @@ class PastixLU : public PastixBase< PastixLU<_MatrixType> >
     typedef _MatrixType MatrixType;
     typedef PastixBase<PastixLU<MatrixType> > Base;
     typedef typename Base::ColSpMatrix ColSpMatrix;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     
   public:
     PastixLU() : Base()
@@ -442,7 +423,7 @@ class PastixLU : public PastixBase< PastixLU<_MatrixType> >
       init();
     }
     
-    PastixLU(const MatrixType& matrix):Base()
+    explicit PastixLU(const MatrixType& matrix):Base()
     {
       init();
       compute(matrix);
@@ -534,8 +515,10 @@ class PastixLU : public PastixBase< PastixLU<_MatrixType> >
   * 
   * \tparam MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam UpLo The part of the matrix to use : Lower or Upper. The default is Lower as required by PaStiX
-  * 
-  * \sa \ref TutorialSparseDirectSolvers
+  *
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SimplicialLLT
   */
 template<typename _MatrixType, int _UpLo>
 class PastixLLT : public PastixBase< PastixLLT<_MatrixType, _UpLo> >
@@ -552,7 +535,7 @@ class PastixLLT : public PastixBase< PastixLLT<_MatrixType, _UpLo> >
       init();
     }
     
-    PastixLLT(const MatrixType& matrix):Base()
+    explicit PastixLLT(const MatrixType& matrix):Base()
     {
       init();
       compute(matrix);
@@ -598,6 +581,7 @@ class PastixLLT : public PastixBase< PastixLLT<_MatrixType, _UpLo> >
     
     void grabMatrix(const MatrixType& matrix, ColSpMatrix& out)
     {
+      out.resize(matrix.rows(), matrix.cols());
       // Pastix supports only lower, column-major matrices 
       out.template selfadjointView<Lower>() = matrix.template selfadjointView<UpLo>();
       internal::c_to_fortran_numbering(out);
@@ -615,8 +599,10 @@ class PastixLLT : public PastixBase< PastixLLT<_MatrixType, _UpLo> >
   * 
   * \tparam MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam UpLo The part of the matrix to use : Lower or Upper. The default is Lower as required by PaStiX
-  * 
-  * \sa \ref TutorialSparseDirectSolvers
+  *
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SimplicialLDLT
   */
 template<typename _MatrixType, int _UpLo>
 class PastixLDLT : public PastixBase< PastixLDLT<_MatrixType, _UpLo> >
@@ -633,7 +619,7 @@ class PastixLDLT : public PastixBase< PastixLDLT<_MatrixType, _UpLo> >
       init();
     }
     
-    PastixLDLT(const MatrixType& matrix):Base()
+    explicit PastixLDLT(const MatrixType& matrix):Base()
     {
       init();
       compute(matrix);
@@ -681,41 +667,12 @@ class PastixLDLT : public PastixBase< PastixLDLT<_MatrixType, _UpLo> >
     void grabMatrix(const MatrixType& matrix, ColSpMatrix& out)
     {
       // Pastix supports only lower, column-major matrices 
+      out.resize(matrix.rows(), matrix.cols());
       out.template selfadjointView<Lower>() = matrix.template selfadjointView<UpLo>();
       internal::c_to_fortran_numbering(out);
     }
 };
 
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<PastixBase<_MatrixType>, Rhs>
-  : solve_retval_base<PastixBase<_MatrixType>, Rhs>
-{
-  typedef PastixBase<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, typename Rhs>
-struct sparse_solve_retval<PastixBase<_MatrixType>, Rhs>
-  : sparse_solve_retval_base<PastixBase<_MatrixType>, Rhs>
-{
-  typedef PastixBase<_MatrixType> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif
diff --git a/extern/Eigen3/Eigen/src/PardisoSupport/PardisoSupport.h b/extern/Eigen3/Eigen/src/PardisoSupport/PardisoSupport.h
index 18cd7d88aea..091c3970e81 100644
--- a/extern/Eigen3/Eigen/src/PardisoSupport/PardisoSupport.h
+++ b/extern/Eigen3/Eigen/src/PardisoSupport/PardisoSupport.h
@@ -40,13 +40,13 @@ template<typename _MatrixType, int Options=Upper> class PardisoLDLT;
 
 namespace internal
 {
-  template<typename Index>
+  template<typename IndexType>
   struct pardiso_run_selector
   {
-    static Index run( _MKL_DSS_HANDLE_t pt, Index maxfct, Index mnum, Index type, Index phase, Index n, void *a,
-                      Index *ia, Index *ja, Index *perm, Index nrhs, Index *iparm, Index msglvl, void *b, void *x)
+    static IndexType run( _MKL_DSS_HANDLE_t pt, IndexType maxfct, IndexType mnum, IndexType type, IndexType phase, IndexType n, void *a,
+                      IndexType *ia, IndexType *ja, IndexType *perm, IndexType nrhs, IndexType *iparm, IndexType msglvl, void *b, void *x)
     {
-      Index error = 0;
+      IndexType error = 0;
       ::pardiso(pt, &maxfct, &mnum, &type, &phase, &n, a, ia, ja, perm, &nrhs, iparm, &msglvl, b, x, &error);
       return error;
     }
@@ -54,11 +54,11 @@ namespace internal
   template<>
   struct pardiso_run_selector<long long int>
   {
-    typedef long long int Index;
-    static Index run( _MKL_DSS_HANDLE_t pt, Index maxfct, Index mnum, Index type, Index phase, Index n, void *a,
-                      Index *ia, Index *ja, Index *perm, Index nrhs, Index *iparm, Index msglvl, void *b, void *x)
+    typedef long long int IndexType;
+    static IndexType run( _MKL_DSS_HANDLE_t pt, IndexType maxfct, IndexType mnum, IndexType type, IndexType phase, IndexType n, void *a,
+                      IndexType *ia, IndexType *ja, IndexType *perm, IndexType nrhs, IndexType *iparm, IndexType msglvl, void *b, void *x)
     {
-      Index error = 0;
+      IndexType error = 0;
       ::pardiso_64(pt, &maxfct, &mnum, &type, &phase, &n, a, ia, ja, perm, &nrhs, iparm, &msglvl, b, x, &error);
       return error;
     }
@@ -72,7 +72,7 @@ namespace internal
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
 
   template<typename _MatrixType, int Options>
@@ -81,7 +81,7 @@ namespace internal
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
 
   template<typename _MatrixType, int Options>
@@ -90,35 +90,44 @@ namespace internal
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;    
+    typedef typename _MatrixType::StorageIndex StorageIndex;    
   };
 
-}
+} // end namespace internal
 
 template<class Derived>
-class PardisoImpl
+class PardisoImpl : public SparseSolverBase<Derived>
 {
+  protected:
+    typedef SparseSolverBase<Derived> Base;
+    using Base::derived;
+    using Base::m_isInitialized;
+    
     typedef internal::pardiso_traits<Derived> Traits;
   public:
+    using Base::_solve_impl;
+    
     typedef typename Traits::MatrixType MatrixType;
     typedef typename Traits::Scalar Scalar;
     typedef typename Traits::RealScalar RealScalar;
-    typedef typename Traits::Index Index;
-    typedef SparseMatrix<Scalar,RowMajor,Index> SparseMatrixType;
+    typedef typename Traits::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,RowMajor,StorageIndex> SparseMatrixType;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
-    typedef Matrix<Index, 1, MatrixType::ColsAtCompileTime> IntRowVectorType;
-    typedef Matrix<Index, MatrixType::RowsAtCompileTime, 1> IntColVectorType;
-    typedef Array<Index,64,1,DontAlign> ParameterType;
+    typedef Matrix<StorageIndex, 1, MatrixType::ColsAtCompileTime> IntRowVectorType;
+    typedef Matrix<StorageIndex, MatrixType::RowsAtCompileTime, 1> IntColVectorType;
+    typedef Array<StorageIndex,64,1,DontAlign> ParameterType;
     enum {
-      ScalarIsComplex = NumTraits<Scalar>::IsComplex
+      ScalarIsComplex = NumTraits<Scalar>::IsComplex,
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
     };
 
     PardisoImpl()
     {
-      eigen_assert((sizeof(Index) >= sizeof(_INTEGER_t) && sizeof(Index) <= 8) && "Non-supported index type");
+      eigen_assert((sizeof(StorageIndex) >= sizeof(_INTEGER_t) && sizeof(StorageIndex) <= 8) && "Non-supported index type");
       m_iparm.setZero();
       m_msglvl = 0; // No output
-      m_initialized = false;
+      m_isInitialized = false;
     }
 
     ~PardisoImpl()
@@ -136,7 +145,7 @@ class PardisoImpl
       */
     ComputationInfo info() const
     {
-      eigen_assert(m_initialized && "Decomposition is not initialized.");
+      eigen_assert(m_isInitialized && "Decomposition is not initialized.");
       return m_info;
     }
 
@@ -165,54 +174,18 @@ class PardisoImpl
     Derived& factorize(const MatrixType& matrix);
 
     Derived& compute(const MatrixType& matrix);
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<PardisoImpl, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_initialized && "Pardiso solver is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "PardisoImpl::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<PardisoImpl, Rhs>(*this, b.derived());
-    }
 
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<PardisoImpl, Rhs>
-    solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_initialized && "Pardiso solver is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "PardisoImpl::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<PardisoImpl, Rhs>(*this, b.derived());
-    }
-
-    Derived& derived()
-    {
-      return *static_cast<Derived*>(this);
-    }
-    const Derived& derived() const
-    {
-      return *static_cast<const Derived*>(this);
-    }
-
-    template<typename BDerived, typename XDerived>
-    bool _solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived>& x) const;
+    template<typename Rhs,typename Dest>
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
 
   protected:
     void pardisoRelease()
     {
-      if(m_initialized) // Factorization ran at least once
+      if(m_isInitialized) // Factorization ran at least once
       {
-        internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, -1, m_size, 0, 0, 0, m_perm.data(), 0,
-                                                   m_iparm.data(), m_msglvl, 0, 0);
+        internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, -1, internal::convert_index<StorageIndex>(m_size),0, 0, 0, m_perm.data(), 0,
+                                                          m_iparm.data(), m_msglvl, NULL, NULL);
+        m_isInitialized = false;
       }
     }
 
@@ -221,11 +194,11 @@ class PardisoImpl
       m_type = type;
       bool symmetric = std::abs(m_type) < 10;
       m_iparm[0] = 1;   // No solver default
-      m_iparm[1] = 3;   // use Metis for the ordering
-      m_iparm[2] = 1;   // Numbers of processors, value of OMP_NUM_THREADS
+      m_iparm[1] = 2;   // use Metis for the ordering
+      m_iparm[2] = 0;   // Reserved. Set to zero. (??Numbers of processors, value of OMP_NUM_THREADS??)
       m_iparm[3] = 0;   // No iterative-direct algorithm
       m_iparm[4] = 0;   // No user fill-in reducing permutation
-      m_iparm[5] = 0;   // Write solution into x
+      m_iparm[5] = 0;   // Write solution into x, b is left unchanged
       m_iparm[6] = 0;   // Not in use
       m_iparm[7] = 2;   // Max numbers of iterative refinement steps
       m_iparm[8] = 0;   // Not in use
@@ -246,13 +219,16 @@ class PardisoImpl
       m_iparm[26] = 0;  // No matrix checker
       m_iparm[27] = (sizeof(RealScalar) == 4) ? 1 : 0;
       m_iparm[34] = 1;  // C indexing
-      m_iparm[59] = 1;  // Automatic switch between In-Core and Out-of-Core modes
+      m_iparm[36] = 0;  // CSR
+      m_iparm[59] = 0;  // 0 - In-Core ; 1 - Automatic switch between In-Core and Out-of-Core modes ; 2 - Out-of-Core
+      
+      memset(m_pt, 0, sizeof(m_pt));
     }
 
   protected:
     // cached data to reduce reallocation, etc.
     
-    void manageErrorCode(Index error)
+    void manageErrorCode(Index error) const
     {
       switch(error)
       {
@@ -269,16 +245,14 @@ class PardisoImpl
     }
 
     mutable SparseMatrixType m_matrix;
-    ComputationInfo m_info;
-    bool m_initialized, m_analysisIsOk, m_factorizationIsOk;
-    Index m_type, m_msglvl;
+    mutable ComputationInfo m_info;
+    bool m_analysisIsOk, m_factorizationIsOk;
+    StorageIndex m_type, m_msglvl;
     mutable void *m_pt[64];
     mutable ParameterType m_iparm;
     mutable IntColVectorType m_perm;
     Index m_size;
     
-  private:
-    PardisoImpl(PardisoImpl &) {}
 };
 
 template<class Derived>
@@ -288,19 +262,17 @@ Derived& PardisoImpl<Derived>::compute(const MatrixType& a)
   eigen_assert(a.rows() == a.cols());
 
   pardisoRelease();
-  memset(m_pt, 0, sizeof(m_pt));
   m_perm.setZero(m_size);
   derived().getMatrix(a);
   
   Index error;
-  error = internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, 12, m_size,
-                                                     m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
-                                                     m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
-
+  error = internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, 12, internal::convert_index<StorageIndex>(m_size),
+                                                            m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
+                                                            m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
   manageErrorCode(error);
   m_analysisIsOk = true;
   m_factorizationIsOk = true;
-  m_initialized = true;
+  m_isInitialized = true;
   return derived();
 }
 
@@ -311,19 +283,18 @@ Derived& PardisoImpl<Derived>::analyzePattern(const MatrixType& a)
   eigen_assert(m_size == a.cols());
 
   pardisoRelease();
-  memset(m_pt, 0, sizeof(m_pt));
   m_perm.setZero(m_size);
   derived().getMatrix(a);
   
   Index error;
-  error = internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, 11, m_size,
-                                                     m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
-                                                     m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
+  error = internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, 11, internal::convert_index<StorageIndex>(m_size),
+                                                            m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
+                                                            m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
   
   manageErrorCode(error);
   m_analysisIsOk = true;
   m_factorizationIsOk = false;
-  m_initialized = true;
+  m_isInitialized = true;
   return derived();
 }
 
@@ -335,22 +306,25 @@ Derived& PardisoImpl<Derived>::factorize(const MatrixType& a)
   
   derived().getMatrix(a);
 
-  Index error;  
-  error = internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, 22, m_size,
-                                                     m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
-                                                     m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
+  Index error;
+  error = internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, 22, internal::convert_index<StorageIndex>(m_size),
+                                                            m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
+                                                            m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
   
   manageErrorCode(error);
   m_factorizationIsOk = true;
   return derived();
 }
 
-template<class Base>
+template<class Derived>
 template<typename BDerived,typename XDerived>
-bool PardisoImpl<Base>::_solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived>& x) const
+void PardisoImpl<Derived>::_solve_impl(const MatrixBase<BDerived> &b, MatrixBase<XDerived>& x) const
 {
   if(m_iparm[0] == 0) // Factorization was not computed
-    return false;
+  {
+    m_info = InvalidInput;
+    return;
+  }
 
   //Index n = m_matrix.rows();
   Index nrhs = Index(b.cols());
@@ -380,12 +354,12 @@ bool PardisoImpl<Base>::_solve(const MatrixBase<BDerived> &b, MatrixBase<XDerive
   }
   
   Index error;
-  error = internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, 33, m_size,
-                                                     m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
-                                                     m_perm.data(), nrhs, m_iparm.data(), m_msglvl,
-                                                     rhs_ptr, x.derived().data());
+  error = internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, 33, internal::convert_index<StorageIndex>(m_size),
+                                                            m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
+                                                            m_perm.data(), internal::convert_index<StorageIndex>(nrhs), m_iparm.data(), m_msglvl,
+                                                            rhs_ptr, x.derived().data());
 
-  return error==0;
+  manageErrorCode(error);
 }
 
 
@@ -397,15 +371,20 @@ bool PardisoImpl<Base>::_solve(const MatrixBase<BDerived> &b, MatrixBase<XDerive
   * using the Intel MKL PARDISO library. The sparse matrix A must be squared and invertible.
   * The vectors or matrices X and B can be either dense or sparse.
   *
+  * By default, it runs in in-core mode. To enable PARDISO's out-of-core feature, set:
+  * \code solver.pardisoParameterArray()[59] = 1; \endcode
+  *
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SparseLU
   */
 template<typename MatrixType>
 class PardisoLU : public PardisoImpl< PardisoLU<MatrixType> >
 {
   protected:
-    typedef PardisoImpl< PardisoLU<MatrixType> > Base;
+    typedef PardisoImpl<PardisoLU> Base;
     typedef typename Base::Scalar Scalar;
     typedef typename Base::RealScalar RealScalar;
     using Base::pardisoInit;
@@ -423,7 +402,7 @@ class PardisoLU : public PardisoImpl< PardisoLU<MatrixType> >
       pardisoInit(Base::ScalarIsComplex ? 13 : 11);
     }
 
-    PardisoLU(const MatrixType& matrix)
+    explicit PardisoLU(const MatrixType& matrix)
       : Base()
     {
       pardisoInit(Base::ScalarIsComplex ? 13 : 11);
@@ -433,10 +412,8 @@ class PardisoLU : public PardisoImpl< PardisoLU<MatrixType> >
     void getMatrix(const MatrixType& matrix)
     {
       m_matrix = matrix;
+      m_matrix.makeCompressed();
     }
-    
-  private:
-    PardisoLU(PardisoLU& ) {}
 };
 
 /** \ingroup PardisoSupport_Module
@@ -447,11 +424,16 @@ class PardisoLU : public PardisoImpl< PardisoLU<MatrixType> >
   * using the Intel MKL PARDISO library. The sparse matrix A must be selfajoint and positive definite.
   * The vectors or matrices X and B can be either dense or sparse.
   *
+  * By default, it runs in in-core mode. To enable PARDISO's out-of-core feature, set:
+  * \code solver.pardisoParameterArray()[59] = 1; \endcode
+  *
   * \tparam MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam UpLo can be any bitwise combination of Upper, Lower. The default is Upper, meaning only the upper triangular part has to be used.
   *         Upper|Lower can be used to tell both triangular parts can be used as input.
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SimplicialLLT
   */
 template<typename MatrixType, int _UpLo>
 class PardisoLLT : public PardisoImpl< PardisoLLT<MatrixType,_UpLo> >
@@ -459,7 +441,6 @@ class PardisoLLT : public PardisoImpl< PardisoLLT<MatrixType,_UpLo> >
   protected:
     typedef PardisoImpl< PardisoLLT<MatrixType,_UpLo> > Base;
     typedef typename Base::Scalar Scalar;
-    typedef typename Base::Index Index;
     typedef typename Base::RealScalar RealScalar;
     using Base::pardisoInit;
     using Base::m_matrix;
@@ -467,9 +448,9 @@ class PardisoLLT : public PardisoImpl< PardisoLLT<MatrixType,_UpLo> >
 
   public:
 
+    typedef typename Base::StorageIndex StorageIndex;
     enum { UpLo = _UpLo };
     using Base::compute;
-    using Base::solve;
 
     PardisoLLT()
       : Base()
@@ -477,7 +458,7 @@ class PardisoLLT : public PardisoImpl< PardisoLLT<MatrixType,_UpLo> >
       pardisoInit(Base::ScalarIsComplex ? 4 : 2);
     }
 
-    PardisoLLT(const MatrixType& matrix)
+    explicit PardisoLLT(const MatrixType& matrix)
       : Base()
     {
       pardisoInit(Base::ScalarIsComplex ? 4 : 2);
@@ -489,13 +470,11 @@ class PardisoLLT : public PardisoImpl< PardisoLLT<MatrixType,_UpLo> >
     void getMatrix(const MatrixType& matrix)
     {
       // PARDISO supports only upper, row-major matrices
-      PermutationMatrix<Dynamic,Dynamic,Index> p_null;
+      PermutationMatrix<Dynamic,Dynamic,StorageIndex> p_null;
       m_matrix.resize(matrix.rows(), matrix.cols());
       m_matrix.template selfadjointView<Upper>() = matrix.template selfadjointView<UpLo>().twistedBy(p_null);
+      m_matrix.makeCompressed();
     }
-    
-  private:
-    PardisoLLT(PardisoLLT& ) {}
 };
 
 /** \ingroup PardisoSupport_Module
@@ -507,12 +486,17 @@ class PardisoLLT : public PardisoImpl< PardisoLLT<MatrixType,_UpLo> >
   * For complex matrices, A can also be symmetric only, see the \a Options template parameter.
   * The vectors or matrices X and B can be either dense or sparse.
   *
+  * By default, it runs in in-core mode. To enable PARDISO's out-of-core feature, set:
+  * \code solver.pardisoParameterArray()[59] = 1; \endcode
+  *
   * \tparam MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam Options can be any bitwise combination of Upper, Lower, and Symmetric. The default is Upper, meaning only the upper triangular part has to be used.
   *         Symmetric can be used for symmetric, non-selfadjoint complex matrices, the default being to assume a selfadjoint matrix.
   *         Upper|Lower can be used to tell both triangular parts can be used as input.
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SimplicialLDLT
   */
 template<typename MatrixType, int Options>
 class PardisoLDLT : public PardisoImpl< PardisoLDLT<MatrixType,Options> >
@@ -520,7 +504,6 @@ class PardisoLDLT : public PardisoImpl< PardisoLDLT<MatrixType,Options> >
   protected:
     typedef PardisoImpl< PardisoLDLT<MatrixType,Options> > Base;
     typedef typename Base::Scalar Scalar;
-    typedef typename Base::Index Index;
     typedef typename Base::RealScalar RealScalar;
     using Base::pardisoInit;
     using Base::m_matrix;
@@ -528,8 +511,8 @@ class PardisoLDLT : public PardisoImpl< PardisoLDLT<MatrixType,Options> >
 
   public:
 
+    typedef typename Base::StorageIndex StorageIndex;
     using Base::compute;
-    using Base::solve;
     enum { UpLo = Options&(Upper|Lower) };
 
     PardisoLDLT()
@@ -538,7 +521,7 @@ class PardisoLDLT : public PardisoImpl< PardisoLDLT<MatrixType,Options> >
       pardisoInit(Base::ScalarIsComplex ? ( bool(Options&Symmetric) ? 6 : -4 ) : -2);
     }
 
-    PardisoLDLT(const MatrixType& matrix)
+    explicit PardisoLDLT(const MatrixType& matrix)
       : Base()
     {
       pardisoInit(Base::ScalarIsComplex ? ( bool(Options&Symmetric) ? 6 : -4 ) : -2);
@@ -548,45 +531,13 @@ class PardisoLDLT : public PardisoImpl< PardisoLDLT<MatrixType,Options> >
     void getMatrix(const MatrixType& matrix)
     {
       // PARDISO supports only upper, row-major matrices
-      PermutationMatrix<Dynamic,Dynamic,Index> p_null;
+      PermutationMatrix<Dynamic,Dynamic,StorageIndex> p_null;
       m_matrix.resize(matrix.rows(), matrix.cols());
       m_matrix.template selfadjointView<Upper>() = matrix.template selfadjointView<UpLo>().twistedBy(p_null);
+      m_matrix.makeCompressed();
     }
-    
-  private:
-    PardisoLDLT(PardisoLDLT& ) {}
-};
-
-namespace internal {
-  
-template<typename _Derived, typename Rhs>
-struct solve_retval<PardisoImpl<_Derived>, Rhs>
-  : solve_retval_base<PardisoImpl<_Derived>, Rhs>
-{
-  typedef PardisoImpl<_Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
 };
 
-template<typename Derived, typename Rhs>
-struct sparse_solve_retval<PardisoImpl<Derived>, Rhs>
-  : sparse_solve_retval_base<PardisoImpl<Derived>, Rhs>
-{
-  typedef PardisoImpl<Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_PARDISOSUPPORT_H
diff --git a/extern/Eigen3/Eigen/src/QR/ColPivHouseholderQR.h b/extern/Eigen3/Eigen/src/QR/ColPivHouseholderQR.h
index 567eab7cda5..a7b47d55dc5 100644
--- a/extern/Eigen3/Eigen/src/QR/ColPivHouseholderQR.h
+++ b/extern/Eigen3/Eigen/src/QR/ColPivHouseholderQR.h
@@ -11,7 +11,16 @@
 #ifndef EIGEN_COLPIVOTINGHOUSEHOLDERQR_H
 #define EIGEN_COLPIVOTINGHOUSEHOLDERQR_H
 
-namespace Eigen { 
+namespace Eigen {
+
+namespace internal {
+template<typename _MatrixType> struct traits<ColPivHouseholderQR<_MatrixType> >
+ : traits<_MatrixType>
+{
+  enum { Flags = 0 };
+};
+
+} // end namespace internal
 
 /** \ingroup QR_Module
   *
@@ -19,19 +28,21 @@ namespace Eigen {
   *
   * \brief Householder rank-revealing QR decomposition of a matrix with column-pivoting
   *
-  * \param MatrixType the type of the matrix of which we are computing the QR decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the QR decomposition
   *
   * This class performs a rank-revealing QR decomposition of a matrix \b A into matrices \b P, \b Q and \b R
-  * such that 
+  * such that
   * \f[
   *  \mathbf{A} \, \mathbf{P} = \mathbf{Q} \, \mathbf{R}
   * \f]
-  * by using Householder transformations. Here, \b P is a permutation matrix, \b Q a unitary matrix and \b R an 
+  * by using Householder transformations. Here, \b P is a permutation matrix, \b Q a unitary matrix and \b R an
   * upper triangular matrix.
   *
   * This decomposition performs column pivoting in order to be rank-revealing and improve
   * numerical stability. It is slower than HouseholderQR, and faster than FullPivHouseholderQR.
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
   * \sa MatrixBase::colPivHouseholderQr()
   */
 template<typename _MatrixType> class ColPivHouseholderQR
@@ -42,25 +53,25 @@ template<typename _MatrixType> class ColPivHouseholderQR
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime, Options, MaxRowsAtCompileTime, MaxRowsAtCompileTime> MatrixQType;
+    // FIXME should be int
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
     typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime> PermutationType;
     typedef typename internal::plain_row_type<MatrixType, Index>::type IntRowVectorType;
     typedef typename internal::plain_row_type<MatrixType>::type RowVectorType;
     typedef typename internal::plain_row_type<MatrixType, RealScalar>::type RealRowVectorType;
     typedef HouseholderSequence<MatrixType,typename internal::remove_all<typename HCoeffsType::ConjugateReturnType>::type> HouseholderSequenceType;
-    
+    typedef typename MatrixType::PlainObject PlainObject;
+
   private:
-    
-    typedef typename PermutationType::Index PermIndexType;
-    
+
+    typedef typename PermutationType::StorageIndex PermIndexType;
+
   public:
 
     /**
@@ -75,7 +86,8 @@ template<typename _MatrixType> class ColPivHouseholderQR
         m_colsPermutation(),
         m_colsTranspositions(),
         m_temp(),
-        m_colSqNorms(),
+        m_colNormsUpdated(),
+        m_colNormsDirect(),
         m_isInitialized(false),
         m_usePrescribedThreshold(false) {}
 
@@ -91,7 +103,8 @@ template<typename _MatrixType> class ColPivHouseholderQR
         m_colsPermutation(PermIndexType(cols)),
         m_colsTranspositions(cols),
         m_temp(cols),
-        m_colSqNorms(cols),
+        m_colNormsUpdated(cols),
+        m_colNormsDirect(cols),
         m_isInitialized(false),
         m_usePrescribedThreshold(false) {}
 
@@ -99,25 +112,48 @@ template<typename _MatrixType> class ColPivHouseholderQR
       *
       * This constructor computes the QR factorization of the matrix \a matrix by calling
       * the method compute(). It is a short cut for:
-      * 
+      *
       * \code
       * ColPivHouseholderQR<MatrixType> qr(matrix.rows(), matrix.cols());
       * qr.compute(matrix);
       * \endcode
-      * 
+      *
       * \sa compute()
       */
-    ColPivHouseholderQR(const MatrixType& matrix)
+    template<typename InputType>
+    explicit ColPivHouseholderQR(const EigenBase<InputType>& matrix)
       : m_qr(matrix.rows(), matrix.cols()),
         m_hCoeffs((std::min)(matrix.rows(),matrix.cols())),
         m_colsPermutation(PermIndexType(matrix.cols())),
         m_colsTranspositions(matrix.cols()),
         m_temp(matrix.cols()),
-        m_colSqNorms(matrix.cols()),
+        m_colNormsUpdated(matrix.cols()),
+        m_colNormsDirect(matrix.cols()),
         m_isInitialized(false),
         m_usePrescribedThreshold(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+    /** \brief Constructs a QR factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+      *
+      * \sa ColPivHouseholderQR(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit ColPivHouseholderQR(EigenBase<InputType>& matrix)
+      : m_qr(matrix.derived()),
+        m_hCoeffs((std::min)(matrix.rows(),matrix.cols())),
+        m_colsPermutation(PermIndexType(matrix.cols())),
+        m_colsTranspositions(matrix.cols()),
+        m_temp(matrix.cols()),
+        m_colNormsUpdated(matrix.cols()),
+        m_colNormsDirect(matrix.cols()),
+        m_isInitialized(false),
+        m_usePrescribedThreshold(false)
+    {
+      computeInPlace();
     }
 
     /** This method finds a solution x to the equation Ax=b, where A is the matrix of which
@@ -127,9 +163,6 @@ template<typename _MatrixType> class ColPivHouseholderQR
       *
       * \returns a solution.
       *
-      * \note The case where b is a matrix is not yet implemented. Also, this
-      *       code is space inefficient.
-      *
       * \note_about_checking_solutions
       *
       * \note_about_arbitrary_choice_of_solution
@@ -138,17 +171,17 @@ template<typename _MatrixType> class ColPivHouseholderQR
       * Output: \verbinclude ColPivHouseholderQR_solve.out
       */
     template<typename Rhs>
-    inline const internal::solve_retval<ColPivHouseholderQR, Rhs>
+    inline const Solve<ColPivHouseholderQR, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "ColPivHouseholderQR is not initialized.");
-      return internal::solve_retval<ColPivHouseholderQR, Rhs>(*this, b.derived());
+      return Solve<ColPivHouseholderQR, Rhs>(*this, b.derived());
     }
 
-    HouseholderSequenceType householderQ(void) const;
-    HouseholderSequenceType matrixQ(void) const
+    HouseholderSequenceType householderQ() const;
+    HouseholderSequenceType matrixQ() const
     {
-      return householderQ(); 
+      return householderQ();
     }
 
     /** \returns a reference to the matrix where the Householder QR decomposition is stored
@@ -158,14 +191,14 @@ template<typename _MatrixType> class ColPivHouseholderQR
       eigen_assert(m_isInitialized && "ColPivHouseholderQR is not initialized.");
       return m_qr;
     }
-    
-    /** \returns a reference to the matrix where the result Householder QR is stored 
-     * \warning The strict lower part of this matrix contains internal values. 
+
+    /** \returns a reference to the matrix where the result Householder QR is stored
+     * \warning The strict lower part of this matrix contains internal values.
      * Only the upper triangular part should be referenced. To get it, use
      * \code matrixR().template triangularView<Upper>() \endcode
-     * For rank-deficient matrices, use 
-     * \code 
-     * matrixR().topLeftCorner(rank(), rank()).template triangularView<Upper>() 
+     * For rank-deficient matrices, use
+     * \code
+     * matrixR().topLeftCorner(rank(), rank()).template triangularView<Upper>()
      * \endcode
      */
     const MatrixType& matrixR() const
@@ -173,8 +206,9 @@ template<typename _MatrixType> class ColPivHouseholderQR
       eigen_assert(m_isInitialized && "ColPivHouseholderQR is not initialized.");
       return m_qr;
     }
-    
-    ColPivHouseholderQR& compute(const MatrixType& matrix);
+
+    template<typename InputType>
+    ColPivHouseholderQR& compute(const EigenBase<InputType>& matrix);
 
     /** \returns a const reference to the column permutation matrix */
     const PermutationType& colsPermutation() const
@@ -284,20 +318,17 @@ template<typename _MatrixType> class ColPivHouseholderQR
       * \note If this matrix is not invertible, the returned matrix has undefined coefficients.
       *       Use isInvertible() to first determine whether this matrix is invertible.
       */
-    inline const
-    internal::solve_retval<ColPivHouseholderQR, typename MatrixType::IdentityReturnType>
-    inverse() const
+    inline const Inverse<ColPivHouseholderQR> inverse() const
     {
       eigen_assert(m_isInitialized && "ColPivHouseholderQR is not initialized.");
-      return internal::solve_retval<ColPivHouseholderQR,typename MatrixType::IdentityReturnType>
-               (*this, MatrixType::Identity(m_qr.rows(), m_qr.cols()));
+      return Inverse<ColPivHouseholderQR>(*this);
     }
 
     inline Index rows() const { return m_qr.rows(); }
     inline Index cols() const { return m_qr.cols(); }
-    
+
     /** \returns a const reference to the vector of Householder coefficients used to represent the factor \c Q.
-      * 
+      *
       * For advanced uses only.
       */
     const HCoeffsType& hCoeffs() const { return m_hCoeffs; }
@@ -370,12 +401,12 @@ template<typename _MatrixType> class ColPivHouseholderQR
       *          diagonal coefficient of R.
       */
     RealScalar maxPivot() const { return m_maxpivot; }
-    
+
     /** \brief Reports whether the QR factorization was succesful.
       *
-      * \note This function always returns \c Success. It is provided for compatibility 
+      * \note This function always returns \c Success. It is provided for compatibility
       * with other factorization routines.
-      * \returns \c Success 
+      * \returns \c Success
       */
     ComputationInfo info() const
     {
@@ -383,19 +414,30 @@ template<typename _MatrixType> class ColPivHouseholderQR
       return Success;
     }
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
+
   protected:
-    
+
+    friend class CompleteOrthogonalDecomposition<MatrixType>;
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
-    
+
+    void computeInPlace();
+
     MatrixType m_qr;
     HCoeffsType m_hCoeffs;
     PermutationType m_colsPermutation;
     IntRowVectorType m_colsTranspositions;
     RowVectorType m_temp;
-    RealRowVectorType m_colSqNorms;
+    RealRowVectorType m_colNormsUpdated;
+    RealRowVectorType m_colNormsDirect;
     bool m_isInitialized, m_usePrescribedThreshold;
     RealScalar m_prescribedThreshold, m_maxpivot;
     Index m_nonzero_pivots;
@@ -426,51 +468,57 @@ typename MatrixType::RealScalar ColPivHouseholderQR<MatrixType>::logAbsDetermina
   * \sa class ColPivHouseholderQR, ColPivHouseholderQR(const MatrixType&)
   */
 template<typename MatrixType>
-ColPivHouseholderQR<MatrixType>& ColPivHouseholderQR<MatrixType>::compute(const MatrixType& matrix)
+template<typename InputType>
+ColPivHouseholderQR<MatrixType>& ColPivHouseholderQR<MatrixType>::compute(const EigenBase<InputType>& matrix)
+{
+  m_qr = matrix.derived();
+  computeInPlace();
+  return *this;
+}
+
+template<typename MatrixType>
+void ColPivHouseholderQR<MatrixType>::computeInPlace()
 {
   check_template_parameters();
-  
-  using std::abs;
-  Index rows = matrix.rows();
-  Index cols = matrix.cols();
-  Index size = matrix.diagonalSize();
-  
+
   // the column permutation is stored as int indices, so just to be sure:
-  eigen_assert(cols<=NumTraits<int>::highest());
+  eigen_assert(m_qr.cols()<=NumTraits<int>::highest());
+
+  using std::abs;
+
+  Index rows = m_qr.rows();
+  Index cols = m_qr.cols();
+  Index size = m_qr.diagonalSize();
 
-  m_qr = matrix;
   m_hCoeffs.resize(size);
 
   m_temp.resize(cols);
 
-  m_colsTranspositions.resize(matrix.cols());
+  m_colsTranspositions.resize(m_qr.cols());
   Index number_of_transpositions = 0;
 
-  m_colSqNorms.resize(cols);
-  for(Index k = 0; k < cols; ++k)
-    m_colSqNorms.coeffRef(k) = m_qr.col(k).squaredNorm();
+  m_colNormsUpdated.resize(cols);
+  m_colNormsDirect.resize(cols);
+  for (Index k = 0; k < cols; ++k) {
+    // colNormsDirect(k) caches the most recent directly computed norm of
+    // column k.
+    m_colNormsDirect.coeffRef(k) = m_qr.col(k).norm();
+    m_colNormsUpdated.coeffRef(k) = m_colNormsDirect.coeffRef(k);
+  }
 
-  RealScalar threshold_helper = m_colSqNorms.maxCoeff() * numext::abs2(NumTraits<Scalar>::epsilon()) / RealScalar(rows);
+  RealScalar threshold_helper =  numext::abs2<RealScalar>(m_colNormsUpdated.maxCoeff() * NumTraits<RealScalar>::epsilon()) / RealScalar(rows);
+  RealScalar norm_downdate_threshold = numext::sqrt(NumTraits<RealScalar>::epsilon());
 
   m_nonzero_pivots = size; // the generic case is that in which all pivots are nonzero (invertible case)
   m_maxpivot = RealScalar(0);
 
   for(Index k = 0; k < size; ++k)
   {
-    // first, we look up in our table m_colSqNorms which column has the biggest squared norm
+    // first, we look up in our table m_colNormsUpdated which column has the biggest norm
     Index biggest_col_index;
-    RealScalar biggest_col_sq_norm = m_colSqNorms.tail(cols-k).maxCoeff(&biggest_col_index);
+    RealScalar biggest_col_sq_norm = numext::abs2(m_colNormsUpdated.tail(cols-k).maxCoeff(&biggest_col_index));
     biggest_col_index += k;
 
-    // since our table m_colSqNorms accumulates imprecision at every step, we must now recompute
-    // the actual squared norm of the selected column.
-    // Note that not doing so does result in solve() sometimes returning inf/nan values
-    // when running the unit test with 1000 repetitions.
-    biggest_col_sq_norm = m_qr.col(biggest_col_index).tail(rows-k).squaredNorm();
-
-    // we store that back into our table: it can't hurt to correct our table.
-    m_colSqNorms.coeffRef(biggest_col_index) = biggest_col_sq_norm;
-
     // Track the number of meaningful pivots but do not stop the decomposition to make
     // sure that the initial matrix is properly reproduced. See bug 941.
     if(m_nonzero_pivots==size && biggest_col_sq_norm < threshold_helper * RealScalar(rows-k))
@@ -480,7 +528,8 @@ ColPivHouseholderQR<MatrixType>& ColPivHouseholderQR<MatrixType>::compute(const
     m_colsTranspositions.coeffRef(k) = biggest_col_index;
     if(k != biggest_col_index) {
       m_qr.col(k).swap(m_qr.col(biggest_col_index));
-      std::swap(m_colSqNorms.coeffRef(k), m_colSqNorms.coeffRef(biggest_col_index));
+      std::swap(m_colNormsUpdated.coeffRef(k), m_colNormsUpdated.coeffRef(biggest_col_index));
+      std::swap(m_colNormsDirect.coeffRef(k), m_colNormsDirect.coeffRef(biggest_col_index));
       ++number_of_transpositions;
     }
 
@@ -498,8 +547,28 @@ ColPivHouseholderQR<MatrixType>& ColPivHouseholderQR<MatrixType>::compute(const
     m_qr.bottomRightCorner(rows-k, cols-k-1)
         .applyHouseholderOnTheLeft(m_qr.col(k).tail(rows-k-1), m_hCoeffs.coeffRef(k), &m_temp.coeffRef(k+1));
 
-    // update our table of squared norms of the columns
-    m_colSqNorms.tail(cols-k-1) -= m_qr.row(k).tail(cols-k-1).cwiseAbs2();
+    // update our table of norms of the columns
+    for (Index j = k + 1; j < cols; ++j) {
+      // The following implements the stable norm downgrade step discussed in
+      // http://www.netlib.org/lapack/lawnspdf/lawn176.pdf
+      // and used in LAPACK routines xGEQPF and xGEQP3.
+      // See lines 278-297 in http://www.netlib.org/lapack/explore-html/dc/df4/sgeqpf_8f_source.html
+      if (m_colNormsUpdated.coeffRef(j) != RealScalar(0)) {
+        RealScalar temp = abs(m_qr.coeffRef(k, j)) / m_colNormsUpdated.coeffRef(j);
+        temp = (RealScalar(1) + temp) * (RealScalar(1) - temp);
+        temp = temp <  RealScalar(0) ? RealScalar(0) : temp;
+        RealScalar temp2 = temp * numext::abs2<RealScalar>(m_colNormsUpdated.coeffRef(j) /
+                                                           m_colNormsDirect.coeffRef(j));
+        if (temp2 <= norm_downdate_threshold) {
+          // The updated norm has become too inaccurate so re-compute the column
+          // norm directly.
+          m_colNormsDirect.coeffRef(j) = m_qr.col(j).tail(rows - k - 1).norm();
+          m_colNormsUpdated.coeffRef(j) = m_colNormsDirect.coeffRef(j);
+        } else {
+          m_colNormsUpdated.coeffRef(j) *= numext::sqrt(temp);
+        }
+      }
+    }
   }
 
   m_colsPermutation.setIdentity(PermIndexType(cols));
@@ -508,46 +577,50 @@ ColPivHouseholderQR<MatrixType>& ColPivHouseholderQR<MatrixType>::compute(const
 
   m_det_pq = (number_of_transpositions%2) ? -1 : 1;
   m_isInitialized = true;
-
-  return *this;
 }
 
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<ColPivHouseholderQR<_MatrixType>, Rhs>
-  : solve_retval_base<ColPivHouseholderQR<_MatrixType>, Rhs>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType>
+template<typename RhsType, typename DstType>
+void ColPivHouseholderQR<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
 {
-  EIGEN_MAKE_SOLVE_HELPERS(ColPivHouseholderQR<_MatrixType>,Rhs)
+  eigen_assert(rhs.rows() == rows());
+
+  const Index nonzero_pivots = nonzeroPivots();
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  if(nonzero_pivots == 0)
   {
-    eigen_assert(rhs().rows() == dec().rows());
+    dst.setZero();
+    return;
+  }
 
-    const Index cols = dec().cols(),
-				nonzero_pivots = dec().nonzeroPivots();
+  typename RhsType::PlainObject c(rhs);
 
-    if(nonzero_pivots == 0)
-    {
-      dst.setZero();
-      return;
-    }
+  // Note that the matrix Q = H_0^* H_1^*... so its inverse is Q^* = (H_0 H_1 ...)^T
+  c.applyOnTheLeft(householderSequence(m_qr, m_hCoeffs)
+                    .setLength(nonzero_pivots)
+                    .transpose()
+    );
 
-    typename Rhs::PlainObject c(rhs());
+  m_qr.topLeftCorner(nonzero_pivots, nonzero_pivots)
+      .template triangularView<Upper>()
+      .solveInPlace(c.topRows(nonzero_pivots));
 
-    // Note that the matrix Q = H_0^* H_1^*... so its inverse is Q^* = (H_0 H_1 ...)^T
-    c.applyOnTheLeft(householderSequence(dec().matrixQR(), dec().hCoeffs())
-                     .setLength(dec().nonzeroPivots())
-		     .transpose()
-      );
+  for(Index i = 0; i < nonzero_pivots; ++i) dst.row(m_colsPermutation.indices().coeff(i)) = c.row(i);
+  for(Index i = nonzero_pivots; i < cols(); ++i) dst.row(m_colsPermutation.indices().coeff(i)).setZero();
+}
+#endif
 
-    dec().matrixR()
-       .topLeftCorner(nonzero_pivots, nonzero_pivots)
-       .template triangularView<Upper>()
-       .solveInPlace(c.topRows(nonzero_pivots));
+namespace internal {
 
-    for(Index i = 0; i < nonzero_pivots; ++i) dst.row(dec().colsPermutation().indices().coeff(i)) = c.row(i);
-    for(Index i = nonzero_pivots; i < cols; ++i) dst.row(dec().colsPermutation().indices().coeff(i)).setZero();
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<ColPivHouseholderQR<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename ColPivHouseholderQR<MatrixType>::Scalar>, Dense2Dense>
+{
+  typedef ColPivHouseholderQR<MatrixType> QrType;
+  typedef Inverse<QrType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename QrType::Scalar> &)
+  {
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.cols()));
   }
 };
 
diff --git a/extern/Eigen3/Eigen/src/QR/ColPivHouseholderQR_LAPACKE.h b/extern/Eigen3/Eigen/src/QR/ColPivHouseholderQR_LAPACKE.h
new file mode 100644
index 00000000000..4e9651f83d5
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/QR/ColPivHouseholderQR_LAPACKE.h
@@ -0,0 +1,97 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *    Householder QR decomposition of a matrix with column pivoting based on
+ *    LAPACKE_?geqp3 function.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_COLPIVOTINGHOUSEHOLDERQR_LAPACKE_H
+#define EIGEN_COLPIVOTINGHOUSEHOLDERQR_LAPACKE_H
+
+namespace Eigen { 
+
+/** \internal Specialization for the data types supported by LAPACKe */
+
+#define EIGEN_LAPACKE_QR_COLPIV(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX, EIGCOLROW, LAPACKE_COLROW) \
+template<> template<typename InputType> inline \
+ColPivHouseholderQR<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic> >& \
+ColPivHouseholderQR<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic> >::compute( \
+              const EigenBase<InputType>& matrix) \
+\
+{ \
+  using std::abs; \
+  typedef Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic> MatrixType; \
+  typedef MatrixType::RealScalar RealScalar; \
+  Index rows = matrix.rows();\
+  Index cols = matrix.cols();\
+\
+  m_qr = matrix;\
+  Index size = m_qr.diagonalSize();\
+  m_hCoeffs.resize(size);\
+\
+  m_colsTranspositions.resize(cols);\
+  /*Index number_of_transpositions = 0;*/ \
+\
+  m_nonzero_pivots = 0; \
+  m_maxpivot = RealScalar(0);\
+  m_colsPermutation.resize(cols); \
+  m_colsPermutation.indices().setZero(); \
+\
+  lapack_int lda = internal::convert_index<lapack_int,Index>(m_qr.outerStride()); \
+  lapack_int matrix_order = LAPACKE_COLROW; \
+  LAPACKE_##LAPACKE_PREFIX##geqp3( matrix_order, internal::convert_index<lapack_int,Index>(rows), internal::convert_index<lapack_int,Index>(cols), \
+                              (LAPACKE_TYPE*)m_qr.data(), lda, (lapack_int*)m_colsPermutation.indices().data(), (LAPACKE_TYPE*)m_hCoeffs.data()); \
+  m_isInitialized = true; \
+  m_maxpivot=m_qr.diagonal().cwiseAbs().maxCoeff(); \
+  m_hCoeffs.adjointInPlace(); \
+  RealScalar premultiplied_threshold = abs(m_maxpivot) * threshold(); \
+  lapack_int *perm = m_colsPermutation.indices().data(); \
+  for(Index i=0;i<size;i++) { \
+    m_nonzero_pivots += (abs(m_qr.coeff(i,i)) > premultiplied_threshold);\
+  } \
+  for(Index i=0;i<cols;i++) perm[i]--;\
+\
+  /*m_det_pq = (number_of_transpositions%2) ? -1 : 1;  // TODO: It's not needed now; fix upon availability in Eigen */ \
+\
+  return *this; \
+}
+
+EIGEN_LAPACKE_QR_COLPIV(double,   double,        d, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(float,    float,         s, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(dcomplex, lapack_complex_double, z, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(scomplex, lapack_complex_float,  c, ColMajor, LAPACK_COL_MAJOR)
+
+EIGEN_LAPACKE_QR_COLPIV(double,   double,        d, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(float,    float,         s, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(dcomplex, lapack_complex_double, z, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(scomplex, lapack_complex_float,  c, RowMajor, LAPACK_ROW_MAJOR)
+
+} // end namespace Eigen
+
+#endif // EIGEN_COLPIVOTINGHOUSEHOLDERQR_LAPACKE_H
diff --git a/extern/Eigen3/Eigen/src/QR/CompleteOrthogonalDecomposition.h b/extern/Eigen3/Eigen/src/QR/CompleteOrthogonalDecomposition.h
new file mode 100644
index 00000000000..34c637b70a9
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/QR/CompleteOrthogonalDecomposition.h
@@ -0,0 +1,562 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Rasmus Munk Larsen <rmlarsen@google.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_COMPLETEORTHOGONALDECOMPOSITION_H
+#define EIGEN_COMPLETEORTHOGONALDECOMPOSITION_H
+
+namespace Eigen {
+
+namespace internal {
+template <typename _MatrixType>
+struct traits<CompleteOrthogonalDecomposition<_MatrixType> >
+    : traits<_MatrixType> {
+  enum { Flags = 0 };
+};
+
+}  // end namespace internal
+
+/** \ingroup QR_Module
+  *
+  * \class CompleteOrthogonalDecomposition
+  *
+  * \brief Complete orthogonal decomposition (COD) of a matrix.
+  *
+  * \param MatrixType the type of the matrix of which we are computing the COD.
+  *
+  * This class performs a rank-revealing complete orthogonal decomposition of a
+  * matrix  \b A into matrices \b P, \b Q, \b T, and \b Z such that
+  * \f[
+  *  \mathbf{A} \, \mathbf{P} = \mathbf{Q} \,
+  *                     \begin{bmatrix} \mathbf{T} &  \mathbf{0} \\
+  *                                     \mathbf{0} & \mathbf{0} \end{bmatrix} \, \mathbf{Z}
+  * \f]
+  * by using Householder transformations. Here, \b P is a permutation matrix,
+  * \b Q and \b Z are unitary matrices and \b T an upper triangular matrix of
+  * size rank-by-rank. \b A may be rank deficient.
+  *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
+  * \sa MatrixBase::completeOrthogonalDecomposition()
+  */
+template <typename _MatrixType>
+class CompleteOrthogonalDecomposition {
+ public:
+  typedef _MatrixType MatrixType;
+  enum {
+    RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+    ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+  };
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename MatrixType::RealScalar RealScalar;
+  typedef typename MatrixType::StorageIndex StorageIndex;
+  typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
+  typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime>
+      PermutationType;
+  typedef typename internal::plain_row_type<MatrixType, Index>::type
+      IntRowVectorType;
+  typedef typename internal::plain_row_type<MatrixType>::type RowVectorType;
+  typedef typename internal::plain_row_type<MatrixType, RealScalar>::type
+      RealRowVectorType;
+  typedef HouseholderSequence<
+      MatrixType, typename internal::remove_all<
+                      typename HCoeffsType::ConjugateReturnType>::type>
+      HouseholderSequenceType;
+  typedef typename MatrixType::PlainObject PlainObject;
+
+ private:
+  typedef typename PermutationType::Index PermIndexType;
+
+ public:
+  /**
+   * \brief Default Constructor.
+   *
+   * The default constructor is useful in cases in which the user intends to
+   * perform decompositions via
+   * \c CompleteOrthogonalDecomposition::compute(const* MatrixType&).
+   */
+  CompleteOrthogonalDecomposition() : m_cpqr(), m_zCoeffs(), m_temp() {}
+
+  /** \brief Default Constructor with memory preallocation
+   *
+   * Like the default constructor but with preallocation of the internal data
+   * according to the specified problem \a size.
+   * \sa CompleteOrthogonalDecomposition()
+   */
+  CompleteOrthogonalDecomposition(Index rows, Index cols)
+      : m_cpqr(rows, cols), m_zCoeffs((std::min)(rows, cols)), m_temp(cols) {}
+
+  /** \brief Constructs a complete orthogonal decomposition from a given
+   * matrix.
+   *
+   * This constructor computes the complete orthogonal decomposition of the
+   * matrix \a matrix by calling the method compute(). The default
+   * threshold for rank determination will be used. It is a short cut for:
+   *
+   * \code
+   * CompleteOrthogonalDecomposition<MatrixType> cod(matrix.rows(),
+   *                                                 matrix.cols());
+   * cod.setThreshold(Default);
+   * cod.compute(matrix);
+   * \endcode
+   *
+   * \sa compute()
+   */
+  template <typename InputType>
+  explicit CompleteOrthogonalDecomposition(const EigenBase<InputType>& matrix)
+      : m_cpqr(matrix.rows(), matrix.cols()),
+        m_zCoeffs((std::min)(matrix.rows(), matrix.cols())),
+        m_temp(matrix.cols())
+  {
+    compute(matrix.derived());
+  }
+
+  /** \brief Constructs a complete orthogonal decomposition from a given matrix
+    *
+    * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+    *
+    * \sa CompleteOrthogonalDecomposition(const EigenBase&)
+    */
+  template<typename InputType>
+  explicit CompleteOrthogonalDecomposition(EigenBase<InputType>& matrix)
+    : m_cpqr(matrix.derived()),
+      m_zCoeffs((std::min)(matrix.rows(), matrix.cols())),
+      m_temp(matrix.cols())
+  {
+    computeInPlace();
+  }
+
+
+  /** This method computes the minimum-norm solution X to a least squares
+   * problem \f[\mathrm{minimize} \|A X - B\|, \f] where \b A is the matrix of
+   * which \c *this is the complete orthogonal decomposition.
+   *
+   * \param b the right-hand sides of the problem to solve.
+   *
+   * \returns a solution.
+   *
+   */
+  template <typename Rhs>
+  inline const Solve<CompleteOrthogonalDecomposition, Rhs> solve(
+      const MatrixBase<Rhs>& b) const {
+    eigen_assert(m_cpqr.m_isInitialized &&
+                 "CompleteOrthogonalDecomposition is not initialized.");
+    return Solve<CompleteOrthogonalDecomposition, Rhs>(*this, b.derived());
+  }
+
+  HouseholderSequenceType householderQ(void) const;
+  HouseholderSequenceType matrixQ(void) const { return m_cpqr.householderQ(); }
+
+  /** \returns the matrix \b Z.
+   */
+  MatrixType matrixZ() const {
+    MatrixType Z = MatrixType::Identity(m_cpqr.cols(), m_cpqr.cols());
+    applyZAdjointOnTheLeftInPlace(Z);
+    return Z.adjoint();
+  }
+
+  /** \returns a reference to the matrix where the complete orthogonal
+   * decomposition is stored
+   */
+  const MatrixType& matrixQTZ() const { return m_cpqr.matrixQR(); }
+
+  /** \returns a reference to the matrix where the complete orthogonal
+   * decomposition is stored.
+   * \warning The strict lower part and \code cols() - rank() \endcode right
+   * columns of this matrix contains internal values.
+   * Only the upper triangular part should be referenced. To get it, use
+   * \code matrixT().template triangularView<Upper>() \endcode
+   * For rank-deficient matrices, use
+   * \code
+   * matrixR().topLeftCorner(rank(), rank()).template triangularView<Upper>()
+   * \endcode
+   */
+  const MatrixType& matrixT() const { return m_cpqr.matrixQR(); }
+
+  template <typename InputType>
+  CompleteOrthogonalDecomposition& compute(const EigenBase<InputType>& matrix) {
+    // Compute the column pivoted QR factorization A P = Q R.
+    m_cpqr.compute(matrix);
+    computeInPlace();
+    return *this;
+  }
+
+  /** \returns a const reference to the column permutation matrix */
+  const PermutationType& colsPermutation() const {
+    return m_cpqr.colsPermutation();
+  }
+
+  /** \returns the absolute value of the determinant of the matrix of which
+   * *this is the complete orthogonal decomposition. It has only linear
+   * complexity (that is, O(n) where n is the dimension of the square matrix)
+   * as the complete orthogonal decomposition has already been computed.
+   *
+   * \note This is only for square matrices.
+   *
+   * \warning a determinant can be very big or small, so for matrices
+   * of large enough dimension, there is a risk of overflow/underflow.
+   * One way to work around that is to use logAbsDeterminant() instead.
+   *
+   * \sa logAbsDeterminant(), MatrixBase::determinant()
+   */
+  typename MatrixType::RealScalar absDeterminant() const;
+
+  /** \returns the natural log of the absolute value of the determinant of the
+   * matrix of which *this is the complete orthogonal decomposition. It has
+   * only linear complexity (that is, O(n) where n is the dimension of the
+   * square matrix) as the complete orthogonal decomposition has already been
+   * computed.
+   *
+   * \note This is only for square matrices.
+   *
+   * \note This method is useful to work around the risk of overflow/underflow
+   * that's inherent to determinant computation.
+   *
+   * \sa absDeterminant(), MatrixBase::determinant()
+   */
+  typename MatrixType::RealScalar logAbsDeterminant() const;
+
+  /** \returns the rank of the matrix of which *this is the complete orthogonal
+   * decomposition.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline Index rank() const { return m_cpqr.rank(); }
+
+  /** \returns the dimension of the kernel of the matrix of which *this is the
+   * complete orthogonal decomposition.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline Index dimensionOfKernel() const { return m_cpqr.dimensionOfKernel(); }
+
+  /** \returns true if the matrix of which *this is the decomposition represents
+   * an injective linear map, i.e. has trivial kernel; false otherwise.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline bool isInjective() const { return m_cpqr.isInjective(); }
+
+  /** \returns true if the matrix of which *this is the decomposition represents
+   * a surjective linear map; false otherwise.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline bool isSurjective() const { return m_cpqr.isSurjective(); }
+
+  /** \returns true if the matrix of which *this is the complete orthogonal
+   * decomposition is invertible.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline bool isInvertible() const { return m_cpqr.isInvertible(); }
+
+  /** \returns the pseudo-inverse of the matrix of which *this is the complete
+   * orthogonal decomposition.
+   * \warning: Do not compute \c this->pseudoInverse()*rhs to solve a linear systems.
+   * It is more efficient and numerically stable to call \c this->solve(rhs).
+   */
+  inline const Inverse<CompleteOrthogonalDecomposition> pseudoInverse() const
+  {
+    return Inverse<CompleteOrthogonalDecomposition>(*this);
+  }
+
+  inline Index rows() const { return m_cpqr.rows(); }
+  inline Index cols() const { return m_cpqr.cols(); }
+
+  /** \returns a const reference to the vector of Householder coefficients used
+   * to represent the factor \c Q.
+   *
+   * For advanced uses only.
+   */
+  inline const HCoeffsType& hCoeffs() const { return m_cpqr.hCoeffs(); }
+
+  /** \returns a const reference to the vector of Householder coefficients
+   * used to represent the factor \c Z.
+   *
+   * For advanced uses only.
+   */
+  const HCoeffsType& zCoeffs() const { return m_zCoeffs; }
+
+  /** Allows to prescribe a threshold to be used by certain methods, such as
+   * rank(), who need to determine when pivots are to be considered nonzero.
+   * Most be called before calling compute().
+   *
+   * When it needs to get the threshold value, Eigen calls threshold(). By
+   * default, this uses a formula to automatically determine a reasonable
+   * threshold. Once you have called the present method
+   * setThreshold(const RealScalar&), your value is used instead.
+   *
+   * \param threshold The new value to use as the threshold.
+   *
+   * A pivot will be considered nonzero if its absolute value is strictly
+   * greater than
+   *  \f$ \vert pivot \vert \leqslant threshold \times \vert maxpivot \vert \f$
+   * where maxpivot is the biggest pivot.
+   *
+   * If you want to come back to the default behavior, call
+   * setThreshold(Default_t)
+   */
+  CompleteOrthogonalDecomposition& setThreshold(const RealScalar& threshold) {
+    m_cpqr.setThreshold(threshold);
+    return *this;
+  }
+
+  /** Allows to come back to the default behavior, letting Eigen use its default
+   * formula for determining the threshold.
+   *
+   * You should pass the special object Eigen::Default as parameter here.
+   * \code qr.setThreshold(Eigen::Default); \endcode
+   *
+   * See the documentation of setThreshold(const RealScalar&).
+   */
+  CompleteOrthogonalDecomposition& setThreshold(Default_t) {
+    m_cpqr.setThreshold(Default);
+    return *this;
+  }
+
+  /** Returns the threshold that will be used by certain methods such as rank().
+   *
+   * See the documentation of setThreshold(const RealScalar&).
+   */
+  RealScalar threshold() const { return m_cpqr.threshold(); }
+
+  /** \returns the number of nonzero pivots in the complete orthogonal
+   * decomposition. Here nonzero is meant in the exact sense, not in a
+   * fuzzy sense. So that notion isn't really intrinsically interesting,
+   * but it is still useful when implementing algorithms.
+   *
+   * \sa rank()
+   */
+  inline Index nonzeroPivots() const { return m_cpqr.nonzeroPivots(); }
+
+  /** \returns the absolute value of the biggest pivot, i.e. the biggest
+   *          diagonal coefficient of R.
+   */
+  inline RealScalar maxPivot() const { return m_cpqr.maxPivot(); }
+
+  /** \brief Reports whether the complete orthogonal decomposition was
+   * succesful.
+   *
+   * \note This function always returns \c Success. It is provided for
+   * compatibility
+   * with other factorization routines.
+   * \returns \c Success
+   */
+  ComputationInfo info() const {
+    eigen_assert(m_cpqr.m_isInitialized && "Decomposition is not initialized.");
+    return Success;
+  }
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+  template <typename RhsType, typename DstType>
+  EIGEN_DEVICE_FUNC void _solve_impl(const RhsType& rhs, DstType& dst) const;
+#endif
+
+ protected:
+  static void check_template_parameters() {
+    EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
+  }
+
+  void computeInPlace();
+
+  /** Overwrites \b rhs with \f$ \mathbf{Z}^* * \mathbf{rhs} \f$.
+   */
+  template <typename Rhs>
+  void applyZAdjointOnTheLeftInPlace(Rhs& rhs) const;
+
+  ColPivHouseholderQR<MatrixType> m_cpqr;
+  HCoeffsType m_zCoeffs;
+  RowVectorType m_temp;
+};
+
+template <typename MatrixType>
+typename MatrixType::RealScalar
+CompleteOrthogonalDecomposition<MatrixType>::absDeterminant() const {
+  return m_cpqr.absDeterminant();
+}
+
+template <typename MatrixType>
+typename MatrixType::RealScalar
+CompleteOrthogonalDecomposition<MatrixType>::logAbsDeterminant() const {
+  return m_cpqr.logAbsDeterminant();
+}
+
+/** Performs the complete orthogonal decomposition of the given matrix \a
+ * matrix. The result of the factorization is stored into \c *this, and a
+ * reference to \c *this is returned.
+ *
+ * \sa class CompleteOrthogonalDecomposition,
+ * CompleteOrthogonalDecomposition(const MatrixType&)
+ */
+template <typename MatrixType>
+void CompleteOrthogonalDecomposition<MatrixType>::computeInPlace()
+{
+  check_template_parameters();
+
+  // the column permutation is stored as int indices, so just to be sure:
+  eigen_assert(m_cpqr.cols() <= NumTraits<int>::highest());
+
+  const Index rank = m_cpqr.rank();
+  const Index cols = m_cpqr.cols();
+  const Index rows = m_cpqr.rows();
+  m_zCoeffs.resize((std::min)(rows, cols));
+  m_temp.resize(cols);
+
+  if (rank < cols) {
+    // We have reduced the (permuted) matrix to the form
+    //   [R11 R12]
+    //   [ 0  R22]
+    // where R11 is r-by-r (r = rank) upper triangular, R12 is
+    // r-by-(n-r), and R22 is empty or the norm of R22 is negligible.
+    // We now compute the complete orthogonal decomposition by applying
+    // Householder transformations from the right to the upper trapezoidal
+    // matrix X = [R11 R12] to zero out R12 and obtain the factorization
+    // [R11 R12] = [T11 0] * Z, where T11 is r-by-r upper triangular and
+    // Z = Z(0) * Z(1) ... Z(r-1) is an n-by-n orthogonal matrix.
+    // We store the data representing Z in R12 and m_zCoeffs.
+    for (Index k = rank - 1; k >= 0; --k) {
+      if (k != rank - 1) {
+        // Given the API for Householder reflectors, it is more convenient if
+        // we swap the leading parts of columns k and r-1 (zero-based) to form
+        // the matrix X_k = [X(0:k, k), X(0:k, r:n)]
+        m_cpqr.m_qr.col(k).head(k + 1).swap(
+            m_cpqr.m_qr.col(rank - 1).head(k + 1));
+      }
+      // Construct Householder reflector Z(k) to zero out the last row of X_k,
+      // i.e. choose Z(k) such that
+      // [X(k, k), X(k, r:n)] * Z(k) = [beta, 0, .., 0].
+      RealScalar beta;
+      m_cpqr.m_qr.row(k)
+          .tail(cols - rank + 1)
+          .makeHouseholderInPlace(m_zCoeffs(k), beta);
+      m_cpqr.m_qr(k, rank - 1) = beta;
+      if (k > 0) {
+        // Apply Z(k) to the first k rows of X_k
+        m_cpqr.m_qr.topRightCorner(k, cols - rank + 1)
+            .applyHouseholderOnTheRight(
+                m_cpqr.m_qr.row(k).tail(cols - rank).transpose(), m_zCoeffs(k),
+                &m_temp(0));
+      }
+      if (k != rank - 1) {
+        // Swap X(0:k,k) back to its proper location.
+        m_cpqr.m_qr.col(k).head(k + 1).swap(
+            m_cpqr.m_qr.col(rank - 1).head(k + 1));
+      }
+    }
+  }
+}
+
+template <typename MatrixType>
+template <typename Rhs>
+void CompleteOrthogonalDecomposition<MatrixType>::applyZAdjointOnTheLeftInPlace(
+    Rhs& rhs) const {
+  const Index cols = this->cols();
+  const Index nrhs = rhs.cols();
+  const Index rank = this->rank();
+  Matrix<typename MatrixType::Scalar, Dynamic, 1> temp((std::max)(cols, nrhs));
+  for (Index k = 0; k < rank; ++k) {
+    if (k != rank - 1) {
+      rhs.row(k).swap(rhs.row(rank - 1));
+    }
+    rhs.middleRows(rank - 1, cols - rank + 1)
+        .applyHouseholderOnTheLeft(
+            matrixQTZ().row(k).tail(cols - rank).adjoint(), zCoeffs()(k),
+            &temp(0));
+    if (k != rank - 1) {
+      rhs.row(k).swap(rhs.row(rank - 1));
+    }
+  }
+}
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template <typename _MatrixType>
+template <typename RhsType, typename DstType>
+void CompleteOrthogonalDecomposition<_MatrixType>::_solve_impl(
+    const RhsType& rhs, DstType& dst) const {
+  eigen_assert(rhs.rows() == this->rows());
+
+  const Index rank = this->rank();
+  if (rank == 0) {
+    dst.setZero();
+    return;
+  }
+
+  // Compute c = Q^* * rhs
+  // Note that the matrix Q = H_0^* H_1^*... so its inverse is
+  // Q^* = (H_0 H_1 ...)^T
+  typename RhsType::PlainObject c(rhs);
+  c.applyOnTheLeft(
+      householderSequence(matrixQTZ(), hCoeffs()).setLength(rank).transpose());
+
+  // Solve T z = c(1:rank, :)
+  dst.topRows(rank) = matrixT()
+                          .topLeftCorner(rank, rank)
+                          .template triangularView<Upper>()
+                          .solve(c.topRows(rank));
+
+  const Index cols = this->cols();
+  if (rank < cols) {
+    // Compute y = Z^* * [ z ]
+    //                   [ 0 ]
+    dst.bottomRows(cols - rank).setZero();
+    applyZAdjointOnTheLeftInPlace(dst);
+  }
+
+  // Undo permutation to get x = P^{-1} * y.
+  dst = colsPermutation() * dst;
+}
+#endif
+
+namespace internal {
+
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<CompleteOrthogonalDecomposition<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename CompleteOrthogonalDecomposition<MatrixType>::Scalar>, Dense2Dense>
+{
+  typedef CompleteOrthogonalDecomposition<MatrixType> CodType;
+  typedef Inverse<CodType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename CodType::Scalar> &)
+  {
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.rows()));
+  }
+};
+
+} // end namespace internal
+
+/** \returns the matrix Q as a sequence of householder transformations */
+template <typename MatrixType>
+typename CompleteOrthogonalDecomposition<MatrixType>::HouseholderSequenceType
+CompleteOrthogonalDecomposition<MatrixType>::householderQ() const {
+  return m_cpqr.householderQ();
+}
+
+/** \return the complete orthogonal decomposition of \c *this.
+  *
+  * \sa class CompleteOrthogonalDecomposition
+  */
+template <typename Derived>
+const CompleteOrthogonalDecomposition<typename MatrixBase<Derived>::PlainObject>
+MatrixBase<Derived>::completeOrthogonalDecomposition() const {
+  return CompleteOrthogonalDecomposition<PlainObject>(eval());
+}
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_COMPLETEORTHOGONALDECOMPOSITION_H
diff --git a/extern/Eigen3/Eigen/src/QR/FullPivHouseholderQR.h b/extern/Eigen3/Eigen/src/QR/FullPivHouseholderQR.h
index 0b39966e145..e489bddc2db 100644
--- a/extern/Eigen3/Eigen/src/QR/FullPivHouseholderQR.h
+++ b/extern/Eigen3/Eigen/src/QR/FullPivHouseholderQR.h
@@ -15,6 +15,12 @@ namespace Eigen {
 
 namespace internal {
 
+template<typename _MatrixType> struct traits<FullPivHouseholderQR<_MatrixType> >
+ : traits<_MatrixType>
+{
+  enum { Flags = 0 };
+};
+
 template<typename MatrixType> struct FullPivHouseholderQRMatrixQReturnType;
 
 template<typename MatrixType>
@@ -23,7 +29,7 @@ struct traits<FullPivHouseholderQRMatrixQReturnType<MatrixType> >
   typedef typename MatrixType::PlainObject ReturnType;
 };
 
-}
+} // end namespace internal
 
 /** \ingroup QR_Module
   *
@@ -31,19 +37,21 @@ struct traits<FullPivHouseholderQRMatrixQReturnType<MatrixType> >
   *
   * \brief Householder rank-revealing QR decomposition of a matrix with full pivoting
   *
-  * \param MatrixType the type of the matrix of which we are computing the QR decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the QR decomposition
   *
-  * This class performs a rank-revealing QR decomposition of a matrix \b A into matrices \b P, \b Q and \b R
+  * This class performs a rank-revealing QR decomposition of a matrix \b A into matrices \b P, \b P', \b Q and \b R
   * such that 
   * \f[
-  *  \mathbf{A} \, \mathbf{P} = \mathbf{Q} \, \mathbf{R}
+  *  \mathbf{P} \, \mathbf{A} \, \mathbf{P}' = \mathbf{Q} \, \mathbf{R}
   * \f]
-  * by using Householder transformations. Here, \b P is a permutation matrix, \b Q a unitary matrix and \b R an 
-  * upper triangular matrix.
+  * by using Householder transformations. Here, \b P and \b P' are permutation matrices, \b Q a unitary matrix 
+  * and \b R an upper triangular matrix.
   *
   * This decomposition performs a very prudent full pivoting in order to be rank-revealing and achieve optimal
   * numerical stability. The trade-off is that it is slower than HouseholderQR and ColPivHouseholderQR.
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
   * \sa MatrixBase::fullPivHouseholderQr()
   */
 template<typename _MatrixType> class FullPivHouseholderQR
@@ -54,21 +62,22 @@ template<typename _MatrixType> class FullPivHouseholderQR
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    // FIXME should be int
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef internal::FullPivHouseholderQRMatrixQReturnType<MatrixType> MatrixQReturnType;
     typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
-    typedef Matrix<Index, 1,
+    typedef Matrix<StorageIndex, 1,
                    EIGEN_SIZE_MIN_PREFER_DYNAMIC(ColsAtCompileTime,RowsAtCompileTime), RowMajor, 1,
                    EIGEN_SIZE_MIN_PREFER_FIXED(MaxColsAtCompileTime,MaxRowsAtCompileTime)> IntDiagSizeVectorType;
     typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime> PermutationType;
     typedef typename internal::plain_row_type<MatrixType>::type RowVectorType;
     typedef typename internal::plain_col_type<MatrixType>::type ColVectorType;
+    typedef typename MatrixType::PlainObject PlainObject;
 
     /** \brief Default Constructor.
       *
@@ -113,7 +122,8 @@ template<typename _MatrixType> class FullPivHouseholderQR
       * 
       * \sa compute()
       */
-    FullPivHouseholderQR(const MatrixType& matrix)
+    template<typename InputType>
+    explicit FullPivHouseholderQR(const EigenBase<InputType>& matrix)
       : m_qr(matrix.rows(), matrix.cols()),
         m_hCoeffs((std::min)(matrix.rows(), matrix.cols())),
         m_rows_transpositions((std::min)(matrix.rows(), matrix.cols())),
@@ -123,7 +133,27 @@ template<typename _MatrixType> class FullPivHouseholderQR
         m_isInitialized(false),
         m_usePrescribedThreshold(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+    /** \brief Constructs a QR factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+      *
+      * \sa FullPivHouseholderQR(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit FullPivHouseholderQR(EigenBase<InputType>& matrix)
+      : m_qr(matrix.derived()),
+        m_hCoeffs((std::min)(matrix.rows(), matrix.cols())),
+        m_rows_transpositions((std::min)(matrix.rows(), matrix.cols())),
+        m_cols_transpositions((std::min)(matrix.rows(), matrix.cols())),
+        m_cols_permutation(matrix.cols()),
+        m_temp(matrix.cols()),
+        m_isInitialized(false),
+        m_usePrescribedThreshold(false)
+    {
+      computeInPlace();
     }
 
     /** This method finds a solution x to the equation Ax=b, where A is the matrix of which
@@ -134,9 +164,6 @@ template<typename _MatrixType> class FullPivHouseholderQR
       * \returns the exact or least-square solution if the rank is greater or equal to the number of columns of A,
       * and an arbitrary solution otherwise.
       *
-      * \note The case where b is a matrix is not yet implemented. Also, this
-      *       code is space inefficient.
-      *
       * \note_about_checking_solutions
       *
       * \note_about_arbitrary_choice_of_solution
@@ -145,11 +172,11 @@ template<typename _MatrixType> class FullPivHouseholderQR
       * Output: \verbinclude FullPivHouseholderQR_solve.out
       */
     template<typename Rhs>
-    inline const internal::solve_retval<FullPivHouseholderQR, Rhs>
+    inline const Solve<FullPivHouseholderQR, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "FullPivHouseholderQR is not initialized.");
-      return internal::solve_retval<FullPivHouseholderQR, Rhs>(*this, b.derived());
+      return Solve<FullPivHouseholderQR, Rhs>(*this, b.derived());
     }
 
     /** \returns Expression object representing the matrix Q
@@ -164,7 +191,8 @@ template<typename _MatrixType> class FullPivHouseholderQR
       return m_qr;
     }
 
-    FullPivHouseholderQR& compute(const MatrixType& matrix);
+    template<typename InputType>
+    FullPivHouseholderQR& compute(const EigenBase<InputType>& matrix);
 
     /** \returns a const reference to the column permutation matrix */
     const PermutationType& colsPermutation() const
@@ -280,13 +308,11 @@ template<typename _MatrixType> class FullPivHouseholderQR
       *
       * \note If this matrix is not invertible, the returned matrix has undefined coefficients.
       *       Use isInvertible() to first determine whether this matrix is invertible.
-      */    inline const
-    internal::solve_retval<FullPivHouseholderQR, typename MatrixType::IdentityReturnType>
-    inverse() const
+      */
+    inline const Inverse<FullPivHouseholderQR> inverse() const
     {
       eigen_assert(m_isInitialized && "FullPivHouseholderQR is not initialized.");
-      return internal::solve_retval<FullPivHouseholderQR,typename MatrixType::IdentityReturnType>
-               (*this, MatrixType::Identity(m_qr.rows(), m_qr.cols()));
+      return Inverse<FullPivHouseholderQR>(*this);
     }
 
     inline Index rows() const { return m_qr.rows(); }
@@ -366,6 +392,12 @@ template<typename _MatrixType> class FullPivHouseholderQR
       *          diagonal coefficient of U.
       */
     RealScalar maxPivot() const { return m_maxpivot; }
+    
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
 
   protected:
     
@@ -374,6 +406,8 @@ template<typename _MatrixType> class FullPivHouseholderQR
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
     
+    void computeInPlace();
+    
     MatrixType m_qr;
     HCoeffsType m_hCoeffs;
     IntDiagSizeVectorType m_rows_transpositions;
@@ -411,16 +445,25 @@ typename MatrixType::RealScalar FullPivHouseholderQR<MatrixType>::logAbsDetermin
   * \sa class FullPivHouseholderQR, FullPivHouseholderQR(const MatrixType&)
   */
 template<typename MatrixType>
-FullPivHouseholderQR<MatrixType>& FullPivHouseholderQR<MatrixType>::compute(const MatrixType& matrix)
+template<typename InputType>
+FullPivHouseholderQR<MatrixType>& FullPivHouseholderQR<MatrixType>::compute(const EigenBase<InputType>& matrix)
+{
+  m_qr = matrix.derived();
+  computeInPlace();
+  return *this;
+}
+
+template<typename MatrixType>
+void FullPivHouseholderQR<MatrixType>::computeInPlace()
 {
   check_template_parameters();
-  
+
   using std::abs;
-  Index rows = matrix.rows();
-  Index cols = matrix.cols();
+  Index rows = m_qr.rows();
+  Index cols = m_qr.cols();
   Index size = (std::min)(rows,cols);
 
-  m_qr = matrix;
+  
   m_hCoeffs.resize(size);
 
   m_temp.resize(cols);
@@ -439,13 +482,15 @@ FullPivHouseholderQR<MatrixType>& FullPivHouseholderQR<MatrixType>::compute(cons
   for (Index k = 0; k < size; ++k)
   {
     Index row_of_biggest_in_corner, col_of_biggest_in_corner;
-    RealScalar biggest_in_corner;
+    typedef internal::scalar_score_coeff_op<Scalar> Scoring;
+    typedef typename Scoring::result_type Score;
 
-    biggest_in_corner = m_qr.bottomRightCorner(rows-k, cols-k)
-                            .cwiseAbs()
-                            .maxCoeff(&row_of_biggest_in_corner, &col_of_biggest_in_corner);
+    Score score = m_qr.bottomRightCorner(rows-k, cols-k)
+                      .unaryExpr(Scoring())
+                      .maxCoeff(&row_of_biggest_in_corner, &col_of_biggest_in_corner);
     row_of_biggest_in_corner += k;
     col_of_biggest_in_corner += k;
+    RealScalar biggest_in_corner = internal::abs_knowing_score<Scalar>()(m_qr(row_of_biggest_in_corner, col_of_biggest_in_corner), score);
     if(k==0) biggest = biggest_in_corner;
 
     // if the corner is negligible, then we have less than full rank, and we can finish early
@@ -489,50 +534,55 @@ FullPivHouseholderQR<MatrixType>& FullPivHouseholderQR<MatrixType>::compute(cons
 
   m_det_pq = (number_of_transpositions%2) ? -1 : 1;
   m_isInitialized = true;
-
-  return *this;
 }
 
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<FullPivHouseholderQR<_MatrixType>, Rhs>
-  : solve_retval_base<FullPivHouseholderQR<_MatrixType>, Rhs>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType>
+template<typename RhsType, typename DstType>
+void FullPivHouseholderQR<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
 {
-  EIGEN_MAKE_SOLVE_HELPERS(FullPivHouseholderQR<_MatrixType>,Rhs)
+  eigen_assert(rhs.rows() == rows());
+  const Index l_rank = rank();
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  // FIXME introduce nonzeroPivots() and use it here. and more generally,
+  // make the same improvements in this dec as in FullPivLU.
+  if(l_rank==0)
   {
-    const Index rows = dec().rows(), cols = dec().cols();
-    eigen_assert(rhs().rows() == rows);
+    dst.setZero();
+    return;
+  }
 
-    // FIXME introduce nonzeroPivots() and use it here. and more generally,
-    // make the same improvements in this dec as in FullPivLU.
-    if(dec().rank()==0)
-    {
-      dst.setZero();
-      return;
-    }
+  typename RhsType::PlainObject c(rhs);
 
-    typename Rhs::PlainObject c(rhs());
+  Matrix<Scalar,1,RhsType::ColsAtCompileTime> temp(rhs.cols());
+  for (Index k = 0; k < l_rank; ++k)
+  {
+    Index remainingSize = rows()-k;
+    c.row(k).swap(c.row(m_rows_transpositions.coeff(k)));
+    c.bottomRightCorner(remainingSize, rhs.cols())
+      .applyHouseholderOnTheLeft(m_qr.col(k).tail(remainingSize-1),
+                               m_hCoeffs.coeff(k), &temp.coeffRef(0));
+  }
 
-    Matrix<Scalar,1,Rhs::ColsAtCompileTime> temp(rhs().cols());
-    for (Index k = 0; k < dec().rank(); ++k)
-    {
-      Index remainingSize = rows-k;
-      c.row(k).swap(c.row(dec().rowsTranspositions().coeff(k)));
-      c.bottomRightCorner(remainingSize, rhs().cols())
-       .applyHouseholderOnTheLeft(dec().matrixQR().col(k).tail(remainingSize-1),
-                                  dec().hCoeffs().coeff(k), &temp.coeffRef(0));
-    }
+  m_qr.topLeftCorner(l_rank, l_rank)
+      .template triangularView<Upper>()
+      .solveInPlace(c.topRows(l_rank));
 
-    dec().matrixQR()
-       .topLeftCorner(dec().rank(), dec().rank())
-       .template triangularView<Upper>()
-       .solveInPlace(c.topRows(dec().rank()));
+  for(Index i = 0; i < l_rank; ++i) dst.row(m_cols_permutation.indices().coeff(i)) = c.row(i);
+  for(Index i = l_rank; i < cols(); ++i) dst.row(m_cols_permutation.indices().coeff(i)).setZero();
+}
+#endif
 
-    for(Index i = 0; i < dec().rank(); ++i) dst.row(dec().colsPermutation().indices().coeff(i)) = c.row(i);
-    for(Index i = dec().rank(); i < cols; ++i) dst.row(dec().colsPermutation().indices().coeff(i)).setZero();
+namespace internal {
+  
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<FullPivHouseholderQR<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename FullPivHouseholderQR<MatrixType>::Scalar>, Dense2Dense>
+{
+  typedef FullPivHouseholderQR<MatrixType> QrType;
+  typedef Inverse<QrType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename QrType::Scalar> &)
+  {    
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.cols()));
   }
 };
 
@@ -546,7 +596,6 @@ template<typename MatrixType> struct FullPivHouseholderQRMatrixQReturnType
   : public ReturnByValue<FullPivHouseholderQRMatrixQReturnType<MatrixType> >
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename FullPivHouseholderQR<MatrixType>::IntDiagSizeVectorType IntDiagSizeVectorType;
   typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
   typedef Matrix<typename MatrixType::Scalar, 1, MatrixType::RowsAtCompileTime, RowMajor, 1,
@@ -558,7 +607,7 @@ public:
     : m_qr(qr),
       m_hCoeffs(hCoeffs),
       m_rowsTranspositions(rowsTranspositions)
-      {}
+  {}
 
   template <typename ResultType>
   void evalTo(ResultType& result) const
@@ -588,8 +637,8 @@ public:
     }
   }
 
-    Index rows() const { return m_qr.rows(); }
-    Index cols() const { return m_qr.rows(); }
+  Index rows() const { return m_qr.rows(); }
+  Index cols() const { return m_qr.rows(); }
 
 protected:
   typename MatrixType::Nested m_qr;
@@ -597,6 +646,11 @@ protected:
   typename IntDiagSizeVectorType::Nested m_rowsTranspositions;
 };
 
+// template<typename MatrixType>
+// struct evaluator<FullPivHouseholderQRMatrixQReturnType<MatrixType> >
+//  : public evaluator<ReturnByValue<FullPivHouseholderQRMatrixQReturnType<MatrixType> > >
+// {};
+
 } // end namespace internal
 
 template<typename MatrixType>
diff --git a/extern/Eigen3/Eigen/src/QR/HouseholderQR.h b/extern/Eigen3/Eigen/src/QR/HouseholderQR.h
index 343a6649934..3513d995cb6 100644
--- a/extern/Eigen3/Eigen/src/QR/HouseholderQR.h
+++ b/extern/Eigen3/Eigen/src/QR/HouseholderQR.h
@@ -21,7 +21,7 @@ namespace Eigen {
   *
   * \brief Householder QR decomposition of a matrix
   *
-  * \param MatrixType the type of the matrix of which we are computing the QR decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the QR decomposition
   *
   * This class performs a QR decomposition of a matrix \b A into matrices \b Q and \b R
   * such that 
@@ -37,6 +37,8 @@ namespace Eigen {
   * This Householder QR decomposition is faster, but less numerically stable and less feature-full than
   * FullPivHouseholderQR or ColPivHouseholderQR.
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  *
   * \sa MatrixBase::householderQr()
   */
 template<typename _MatrixType> class HouseholderQR
@@ -47,13 +49,13 @@ template<typename _MatrixType> class HouseholderQR
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    // FIXME should be int
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime, (MatrixType::Flags&RowMajorBit) ? RowMajor : ColMajor, MaxRowsAtCompileTime, MaxRowsAtCompileTime> MatrixQType;
     typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
     typedef typename internal::plain_row_type<MatrixType>::type RowVectorType;
@@ -91,13 +93,32 @@ template<typename _MatrixType> class HouseholderQR
       * 
       * \sa compute()
       */
-    HouseholderQR(const MatrixType& matrix)
+    template<typename InputType>
+    explicit HouseholderQR(const EigenBase<InputType>& matrix)
       : m_qr(matrix.rows(), matrix.cols()),
         m_hCoeffs((std::min)(matrix.rows(),matrix.cols())),
         m_temp(matrix.cols()),
         m_isInitialized(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+
+    /** \brief Constructs a QR factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when
+      * \c MatrixType is a Eigen::Ref.
+      *
+      * \sa HouseholderQR(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit HouseholderQR(EigenBase<InputType>& matrix)
+      : m_qr(matrix.derived()),
+        m_hCoeffs((std::min)(matrix.rows(),matrix.cols())),
+        m_temp(matrix.cols()),
+        m_isInitialized(false)
+    {
+      computeInPlace();
     }
 
     /** This method finds a solution x to the equation Ax=b, where A is the matrix of which
@@ -107,9 +128,6 @@ template<typename _MatrixType> class HouseholderQR
       *
       * \returns a solution.
       *
-      * \note The case where b is a matrix is not yet implemented. Also, this
-      *       code is space inefficient.
-      *
       * \note_about_checking_solutions
       *
       * \note_about_arbitrary_choice_of_solution
@@ -118,11 +136,11 @@ template<typename _MatrixType> class HouseholderQR
       * Output: \verbinclude HouseholderQR_solve.out
       */
     template<typename Rhs>
-    inline const internal::solve_retval<HouseholderQR, Rhs>
+    inline const Solve<HouseholderQR, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "HouseholderQR is not initialized.");
-      return internal::solve_retval<HouseholderQR, Rhs>(*this, b.derived());
+      return Solve<HouseholderQR, Rhs>(*this, b.derived());
     }
 
     /** This method returns an expression of the unitary matrix Q as a sequence of Householder transformations.
@@ -148,7 +166,12 @@ template<typename _MatrixType> class HouseholderQR
         return m_qr;
     }
 
-    HouseholderQR& compute(const MatrixType& matrix);
+    template<typename InputType>
+    HouseholderQR& compute(const EigenBase<InputType>& matrix) {
+      m_qr = matrix.derived();
+      computeInPlace();
+      return *this;
+    }
 
     /** \returns the absolute value of the determinant of the matrix of which
       * *this is the QR decomposition. It has only linear complexity
@@ -187,6 +210,12 @@ template<typename _MatrixType> class HouseholderQR
       * For advanced uses only.
       */
     const HCoeffsType& hCoeffs() const { return m_hCoeffs; }
+    
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
 
   protected:
     
@@ -194,6 +223,8 @@ template<typename _MatrixType> class HouseholderQR
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
+
+    void computeInPlace();
     
     MatrixType m_qr;
     HCoeffsType m_hCoeffs;
@@ -224,7 +255,6 @@ namespace internal {
 template<typename MatrixQR, typename HCoeffs>
 void householder_qr_inplace_unblocked(MatrixQR& mat, HCoeffs& hCoeffs, typename MatrixQR::Scalar* tempData = 0)
 {
-  typedef typename MatrixQR::Index Index;
   typedef typename MatrixQR::Scalar Scalar;
   typedef typename MatrixQR::RealScalar RealScalar;
   Index rows = mat.rows();
@@ -263,11 +293,9 @@ template<typename MatrixQR, typename HCoeffs,
 struct householder_qr_inplace_blocked
 {
   // This is specialized for MKL-supported Scalar types in HouseholderQR_MKL.h
-  static void run(MatrixQR& mat, HCoeffs& hCoeffs,
-      typename MatrixQR::Index maxBlockSize=32,
+  static void run(MatrixQR& mat, HCoeffs& hCoeffs, Index maxBlockSize=32,
       typename MatrixQR::Scalar* tempData = 0)
   {
-    typedef typename MatrixQR::Index Index;
     typedef typename MatrixQR::Scalar Scalar;
     typedef Block<MatrixQR,Dynamic,Dynamic> BlockType;
 
@@ -289,8 +317,8 @@ struct householder_qr_inplace_blocked
     for (k = 0; k < size; k += blockSize)
     {
       Index bs = (std::min)(size-k,blockSize);  // actual size of the block
-      Index tcols = cols - k - bs;            // trailing columns
-      Index brows = rows-k;                   // rows of the block
+      Index tcols = cols - k - bs;              // trailing columns
+      Index brows = rows-k;                     // rows of the block
 
       // partition the matrix:
       //        A00 | A01 | A02
@@ -308,43 +336,38 @@ struct householder_qr_inplace_blocked
       if(tcols)
       {
         BlockType A21_22 = mat.block(k,k+bs,brows,tcols);
-        apply_block_householder_on_the_left(A21_22,A11_21,hCoeffsSegment.adjoint());
+        apply_block_householder_on_the_left(A21_22,A11_21,hCoeffsSegment, false); // false == backward
       }
     }
   }
 };
 
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<HouseholderQR<_MatrixType>, Rhs>
-  : solve_retval_base<HouseholderQR<_MatrixType>, Rhs>
-{
-  EIGEN_MAKE_SOLVE_HELPERS(HouseholderQR<_MatrixType>,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    const Index rows = dec().rows(), cols = dec().cols();
-    const Index rank = (std::min)(rows, cols);
-    eigen_assert(rhs().rows() == rows);
+} // end namespace internal
 
-    typename Rhs::PlainObject c(rhs());
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType>
+template<typename RhsType, typename DstType>
+void HouseholderQR<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
+{
+  const Index rank = (std::min)(rows(), cols());
+  eigen_assert(rhs.rows() == rows());
 
-    // Note that the matrix Q = H_0^* H_1^*... so its inverse is Q^* = (H_0 H_1 ...)^T
-    c.applyOnTheLeft(householderSequence(
-      dec().matrixQR().leftCols(rank),
-      dec().hCoeffs().head(rank)).transpose()
-    );
+  typename RhsType::PlainObject c(rhs);
 
-    dec().matrixQR()
-       .topLeftCorner(rank, rank)
-       .template triangularView<Upper>()
-       .solveInPlace(c.topRows(rank));
+  // Note that the matrix Q = H_0^* H_1^*... so its inverse is Q^* = (H_0 H_1 ...)^T
+  c.applyOnTheLeft(householderSequence(
+    m_qr.leftCols(rank),
+    m_hCoeffs.head(rank)).transpose()
+  );
 
-    dst.topRows(rank) = c.topRows(rank);
-    dst.bottomRows(cols-rank).setZero();
-  }
-};
+  m_qr.topLeftCorner(rank, rank)
+      .template triangularView<Upper>()
+      .solveInPlace(c.topRows(rank));
 
-} // end namespace internal
+  dst.topRows(rank) = c.topRows(rank);
+  dst.bottomRows(cols()-rank).setZero();
+}
+#endif
 
 /** Performs the QR factorization of the given matrix \a matrix. The result of
   * the factorization is stored into \c *this, and a reference to \c *this
@@ -353,15 +376,14 @@ struct solve_retval<HouseholderQR<_MatrixType>, Rhs>
   * \sa class HouseholderQR, HouseholderQR(const MatrixType&)
   */
 template<typename MatrixType>
-HouseholderQR<MatrixType>& HouseholderQR<MatrixType>::compute(const MatrixType& matrix)
+void HouseholderQR<MatrixType>::computeInPlace()
 {
   check_template_parameters();
   
-  Index rows = matrix.rows();
-  Index cols = matrix.cols();
+  Index rows = m_qr.rows();
+  Index cols = m_qr.cols();
   Index size = (std::min)(rows,cols);
 
-  m_qr = matrix;
   m_hCoeffs.resize(size);
 
   m_temp.resize(cols);
@@ -369,7 +391,6 @@ HouseholderQR<MatrixType>& HouseholderQR<MatrixType>::compute(const MatrixType&
   internal::householder_qr_inplace_blocked<MatrixType, HCoeffsType>::run(m_qr, m_hCoeffs, 48, m_temp.data());
 
   m_isInitialized = true;
-  return *this;
 }
 
 /** \return the Householder QR decomposition of \c *this.
diff --git a/extern/Eigen3/Eigen/src/QR/HouseholderQR_LAPACKE.h b/extern/Eigen3/Eigen/src/QR/HouseholderQR_LAPACKE.h
new file mode 100644
index 00000000000..1dc7d5363f1
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/QR/HouseholderQR_LAPACKE.h
@@ -0,0 +1,68 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *    Householder QR decomposition of a matrix w/o pivoting based on
+ *    LAPACKE_?geqrf function.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_QR_LAPACKE_H
+#define EIGEN_QR_LAPACKE_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/** \internal Specialization for the data types supported by LAPACKe */
+
+#define EIGEN_LAPACKE_QR_NOPIV(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX) \
+template<typename MatrixQR, typename HCoeffs> \
+struct householder_qr_inplace_blocked<MatrixQR, HCoeffs, EIGTYPE, true> \
+{ \
+  static void run(MatrixQR& mat, HCoeffs& hCoeffs, Index = 32, \
+      typename MatrixQR::Scalar* = 0) \
+  { \
+    lapack_int m = (lapack_int) mat.rows(); \
+    lapack_int n = (lapack_int) mat.cols(); \
+    lapack_int lda = (lapack_int) mat.outerStride(); \
+    lapack_int matrix_order = (MatrixQR::IsRowMajor) ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \
+    LAPACKE_##LAPACKE_PREFIX##geqrf( matrix_order, m, n, (LAPACKE_TYPE*)mat.data(), lda, (LAPACKE_TYPE*)hCoeffs.data()); \
+    hCoeffs.adjointInPlace(); \
+  } \
+};
+
+EIGEN_LAPACKE_QR_NOPIV(double, double, d)
+EIGEN_LAPACKE_QR_NOPIV(float, float, s)
+EIGEN_LAPACKE_QR_NOPIV(dcomplex, lapack_complex_double, z)
+EIGEN_LAPACKE_QR_NOPIV(scomplex, lapack_complex_float, c)
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_QR_LAPACKE_H
diff --git a/extern/Eigen3/Eigen/src/SPQRSupport/SuiteSparseQRSupport.h b/extern/Eigen3/Eigen/src/SPQRSupport/SuiteSparseQRSupport.h
index 36138101d74..953d57c9d76 100644
--- a/extern/Eigen3/Eigen/src/SPQRSupport/SuiteSparseQRSupport.h
+++ b/extern/Eigen3/Eigen/src/SPQRSupport/SuiteSparseQRSupport.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2012 Desire Nuentsa <desire.nuentsa_wakam@inria.fr>
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -32,45 +33,54 @@ namespace Eigen {
   } // End namespace internal
   
 /**
- * \ingroup SPQRSupport_Module
- * \class SPQR
- * \brief Sparse QR factorization based on SuiteSparseQR library
- * 
- * This class is used to perform a multithreaded and multifrontal rank-revealing QR decomposition 
- * of sparse matrices. The result is then used to solve linear leasts_square systems.
- * Clearly, a QR factorization is returned such that A*P = Q*R where :
- * 
- * P is the column permutation. Use colsPermutation() to get it.
- * 
- * Q is the orthogonal matrix represented as Householder reflectors. 
- * Use matrixQ() to get an expression and matrixQ().transpose() to get the transpose.
- * You can then apply it to a vector.
- * 
- * R is the sparse triangular factor. Use matrixQR() to get it as SparseMatrix.
- * NOTE : The Index type of R is always SuiteSparse_long. You can get it with SPQR::Index
- * 
- * \tparam _MatrixType The type of the sparse matrix A, must be a column-major SparseMatrix<>
- * NOTE 
- * 
- */
+  * \ingroup SPQRSupport_Module
+  * \class SPQR
+  * \brief Sparse QR factorization based on SuiteSparseQR library
+  *
+  * This class is used to perform a multithreaded and multifrontal rank-revealing QR decomposition
+  * of sparse matrices. The result is then used to solve linear leasts_square systems.
+  * Clearly, a QR factorization is returned such that A*P = Q*R where :
+  *
+  * P is the column permutation. Use colsPermutation() to get it.
+  *
+  * Q is the orthogonal matrix represented as Householder reflectors.
+  * Use matrixQ() to get an expression and matrixQ().transpose() to get the transpose.
+  * You can then apply it to a vector.
+  *
+  * R is the sparse triangular factor. Use matrixQR() to get it as SparseMatrix.
+  * NOTE : The Index type of R is always SuiteSparse_long. You can get it with SPQR::Index
+  *
+  * \tparam _MatrixType The type of the sparse matrix A, must be a column-major SparseMatrix<>
+  *
+  * \implsparsesolverconcept
+  *
+  *
+  */
 template<typename _MatrixType>
-class SPQR
+class SPQR : public SparseSolverBase<SPQR<_MatrixType> >
 {
+  protected:
+    typedef SparseSolverBase<SPQR<_MatrixType> > Base;
+    using Base::m_isInitialized;
   public:
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef SuiteSparse_long Index ;
-    typedef SparseMatrix<Scalar, ColMajor, Index> MatrixType;
-    typedef PermutationMatrix<Dynamic, Dynamic> PermutationType;
+    typedef SuiteSparse_long StorageIndex ;
+    typedef SparseMatrix<Scalar, ColMajor, StorageIndex> MatrixType;
+    typedef Map<PermutationMatrix<Dynamic, Dynamic, StorageIndex> > PermutationType;
+    enum {
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
+    };
   public:
     SPQR() 
-      : m_isInitialized(false), m_ordering(SPQR_ORDERING_DEFAULT), m_allow_tol(SPQR_DEFAULT_TOL), m_tolerance (NumTraits<Scalar>::epsilon()), m_useDefaultThreshold(true)
+      : m_ordering(SPQR_ORDERING_DEFAULT), m_allow_tol(SPQR_DEFAULT_TOL), m_tolerance (NumTraits<Scalar>::epsilon()), m_useDefaultThreshold(true)
     { 
       cholmod_l_start(&m_cc);
     }
     
-    SPQR(const _MatrixType& matrix)
-    : m_isInitialized(false), m_ordering(SPQR_ORDERING_DEFAULT), m_allow_tol(SPQR_DEFAULT_TOL), m_tolerance (NumTraits<Scalar>::epsilon()), m_useDefaultThreshold(true)
+    explicit SPQR(const _MatrixType& matrix)
+    : m_ordering(SPQR_ORDERING_DEFAULT), m_allow_tol(SPQR_DEFAULT_TOL), m_tolerance (NumTraits<Scalar>::epsilon()), m_useDefaultThreshold(true)
     {
       cholmod_l_start(&m_cc);
       compute(matrix);
@@ -103,23 +113,22 @@ class SPQR
       RealScalar pivotThreshold = m_tolerance;
       if(m_useDefaultThreshold) 
       {
-        using std::max;
         RealScalar max2Norm = 0.0;
-        for (int j = 0; j < mat.cols(); j++) max2Norm = (max)(max2Norm, mat.col(j).norm());
+        for (int j = 0; j < mat.cols(); j++) max2Norm = numext::maxi(max2Norm, mat.col(j).norm());
         if(max2Norm==RealScalar(0))
           max2Norm = RealScalar(1);
         pivotThreshold = 20 * (mat.rows() + mat.cols()) * max2Norm * NumTraits<RealScalar>::epsilon();
       }
-      
       cholmod_sparse A; 
       A = viewAsCholmod(mat);
+      m_rows = matrix.rows();
       Index col = matrix.cols();
       m_rank = SuiteSparseQR<Scalar>(m_ordering, pivotThreshold, col, &A, 
                              &m_cR, &m_E, &m_H, &m_HPinv, &m_HTau, &m_cc);
 
       if (!m_cR)
       {
-        m_info = NumericalIssue; 
+        m_info = NumericalIssue;
         m_isInitialized = false;
         return;
       }
@@ -130,28 +139,15 @@ class SPQR
     /** 
      * Get the number of rows of the input matrix and the Q matrix
      */
-    inline Index rows() const {return m_cR->nrow; }
+    inline Index rows() const {return m_rows; }
     
     /** 
      * Get the number of columns of the input matrix. 
      */
     inline Index cols() const { return m_cR->ncol; }
-   
-      /** \returns the solution X of \f$ A X = B \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<SPQR, Rhs> solve(const MatrixBase<Rhs>& B) const 
-    {
-      eigen_assert(m_isInitialized && " The QR factorization should be computed first, call compute()");
-      eigen_assert(this->rows()==B.rows()
-                    && "SPQR::solve(): invalid number of rows of the right hand side matrix B");
-          return internal::solve_retval<SPQR, Rhs>(*this, B.derived());
-    }
     
     template<typename Rhs, typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
     {
       eigen_assert(m_isInitialized && " The QR factorization should be computed first, call compute()");
       eigen_assert(b.cols()==1 && "This method is for vectors only");
@@ -184,7 +180,7 @@ class SPQR
     {
       eigen_assert(m_isInitialized && " The QR factorization should be computed first, call compute()");
       if(!m_isRUpToDate) {
-        m_R = viewAsEigen<Scalar,ColMajor, typename MatrixType::Index>(*m_cR);
+        m_R = viewAsEigen<Scalar,ColMajor, typename MatrixType::StorageIndex>(*m_cR);
         m_isRUpToDate = true;
       }
       return m_R;
@@ -198,11 +194,7 @@ class SPQR
     PermutationType colsPermutation() const
     { 
       eigen_assert(m_isInitialized && "Decomposition is not initialized.");
-      Index n = m_cR->ncol;
-      PermutationType colsPerm(n);
-      for(Index j = 0; j <n; j++) colsPerm.indices()(j) = m_E[j];
-      return colsPerm; 
-      
+      return PermutationType(m_E, m_cR->ncol);
     }
     /**
      * Gets the rank of the matrix. 
@@ -237,7 +229,6 @@ class SPQR
       return m_info;
     }
   protected:
-    bool m_isInitialized;
     bool m_analysisIsOk;
     bool m_factorizationIsOk;
     mutable bool m_isRUpToDate;
@@ -247,13 +238,14 @@ class SPQR
     RealScalar m_tolerance; // treat columns with 2-norm below this tolerance as zero
     mutable cholmod_sparse *m_cR; // The sparse R factor in cholmod format
     mutable MatrixType m_R; // The sparse matrix R in Eigen format
-    mutable Index *m_E; // The permutation applied to columns
+    mutable StorageIndex *m_E; // The permutation applied to columns
     mutable cholmod_sparse *m_H;  //The householder vectors
-    mutable Index *m_HPinv; // The row permutation of H
+    mutable StorageIndex *m_HPinv; // The row permutation of H
     mutable cholmod_dense *m_HTau; // The Householder coefficients
     mutable Index m_rank; // The rank of the matrix
     mutable cholmod_common m_cc; // Workspace and parameters
     bool m_useDefaultThreshold;     // Use default threshold
+    Index m_rows;
     template<typename ,typename > friend struct SPQR_QProduct;
 };
 
@@ -261,7 +253,7 @@ template <typename SPQRType, typename Derived>
 struct SPQR_QProduct : ReturnByValue<SPQR_QProduct<SPQRType,Derived> >
 {
   typedef typename SPQRType::Scalar Scalar;
-  typedef typename SPQRType::Index Index;
+  typedef typename SPQRType::StorageIndex StorageIndex;
   //Define the constructor to get reference to argument types
   SPQR_QProduct(const SPQRType& spqr, const Derived& other, bool transpose) : m_spqr(spqr),m_other(other),m_transpose(transpose) {}
   
@@ -317,22 +309,5 @@ struct SPQRMatrixQTransposeReturnType{
   const SPQRType& m_spqr;
 };
 
-namespace internal {
-  
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<SPQR<_MatrixType>, Rhs>
-  : solve_retval_base<SPQR<_MatrixType>, Rhs>
-{
-  typedef SPQR<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 }// End namespace Eigen
 #endif
diff --git a/extern/Eigen3/Eigen/src/SVD/BDCSVD.h b/extern/Eigen3/Eigen/src/SVD/BDCSVD.h
new file mode 100644
index 00000000000..1134d66e7e9
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/SVD/BDCSVD.h
@@ -0,0 +1,1246 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+// 
+// We used the "A Divide-And-Conquer Algorithm for the Bidiagonal SVD"
+// research report written by Ming Gu and Stanley C.Eisenstat
+// The code variable names correspond to the names they used in their 
+// report
+//
+// Copyright (C) 2013 Gauthier Brun <brun.gauthier@gmail.com>
+// Copyright (C) 2013 Nicolas Carre <nicolas.carre@ensimag.fr>
+// Copyright (C) 2013 Jean Ceccato <jean.ceccato@ensimag.fr>
+// Copyright (C) 2013 Pierre Zoppitelli <pierre.zoppitelli@ensimag.fr>
+// Copyright (C) 2013 Jitse Niesen <jitse@maths.leeds.ac.uk>
+// Copyright (C) 2014-2017 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_BDCSVD_H
+#define EIGEN_BDCSVD_H
+// #define EIGEN_BDCSVD_DEBUG_VERBOSE
+// #define EIGEN_BDCSVD_SANITY_CHECKS
+
+namespace Eigen {
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+IOFormat bdcsvdfmt(8, 0, ", ", "\n", "  [", "]");
+#endif
+  
+template<typename _MatrixType> class BDCSVD;
+
+namespace internal {
+
+template<typename _MatrixType> 
+struct traits<BDCSVD<_MatrixType> >
+{
+  typedef _MatrixType MatrixType;
+};  
+
+} // end namespace internal
+  
+  
+/** \ingroup SVD_Module
+ *
+ *
+ * \class BDCSVD
+ *
+ * \brief class Bidiagonal Divide and Conquer SVD
+ *
+ * \tparam _MatrixType the type of the matrix of which we are computing the SVD decomposition
+ *
+ * This class first reduces the input matrix to bi-diagonal form using class UpperBidiagonalization,
+ * and then performs a divide-and-conquer diagonalization. Small blocks are diagonalized using class JacobiSVD.
+ * You can control the switching size with the setSwitchSize() method, default is 16.
+ * For small matrice (<16), it is thus preferable to directly use JacobiSVD. For larger ones, BDCSVD is highly
+ * recommended and can several order of magnitude faster.
+ *
+ * \warning this algorithm is unlikely to provide accurate result when compiled with unsafe math optimizations.
+ * For instance, this concerns Intel's compiler (ICC), which perfroms such optimization by default unless
+ * you compile with the \c -fp-model \c precise option. Likewise, the \c -ffast-math option of GCC or clang will
+ * significantly degrade the accuracy.
+ *
+ * \sa class JacobiSVD
+ */
+template<typename _MatrixType> 
+class BDCSVD : public SVDBase<BDCSVD<_MatrixType> >
+{
+  typedef SVDBase<BDCSVD> Base;
+    
+public:
+  using Base::rows;
+  using Base::cols;
+  using Base::computeU;
+  using Base::computeV;
+  
+  typedef _MatrixType MatrixType;
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
+  typedef typename NumTraits<RealScalar>::Literal Literal;
+  enum {
+    RowsAtCompileTime = MatrixType::RowsAtCompileTime, 
+    ColsAtCompileTime = MatrixType::ColsAtCompileTime, 
+    DiagSizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime, ColsAtCompileTime), 
+    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime, 
+    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime, 
+    MaxDiagSizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(MaxRowsAtCompileTime, MaxColsAtCompileTime), 
+    MatrixOptions = MatrixType::Options
+  };
+
+  typedef typename Base::MatrixUType MatrixUType;
+  typedef typename Base::MatrixVType MatrixVType;
+  typedef typename Base::SingularValuesType SingularValuesType;
+  
+  typedef Matrix<Scalar, Dynamic, Dynamic, ColMajor> MatrixX;
+  typedef Matrix<RealScalar, Dynamic, Dynamic, ColMajor> MatrixXr;
+  typedef Matrix<RealScalar, Dynamic, 1> VectorType;
+  typedef Array<RealScalar, Dynamic, 1> ArrayXr;
+  typedef Array<Index,1,Dynamic> ArrayXi;
+  typedef Ref<ArrayXr> ArrayRef;
+  typedef Ref<ArrayXi> IndicesRef;
+
+  /** \brief Default Constructor.
+   *
+   * The default constructor is useful in cases in which the user intends to
+   * perform decompositions via BDCSVD::compute(const MatrixType&).
+   */
+  BDCSVD() : m_algoswap(16), m_numIters(0)
+  {}
+
+
+  /** \brief Default Constructor with memory preallocation
+   *
+   * Like the default constructor but with preallocation of the internal data
+   * according to the specified problem size.
+   * \sa BDCSVD()
+   */
+  BDCSVD(Index rows, Index cols, unsigned int computationOptions = 0)
+    : m_algoswap(16), m_numIters(0)
+  {
+    allocate(rows, cols, computationOptions);
+  }
+
+  /** \brief Constructor performing the decomposition of given matrix.
+   *
+   * \param matrix the matrix to decompose
+   * \param computationOptions optional parameter allowing to specify if you want full or thin U or V unitaries to be computed.
+   *                           By default, none is computed. This is a bit - field, the possible bits are #ComputeFullU, #ComputeThinU, 
+   *                           #ComputeFullV, #ComputeThinV.
+   *
+   * Thin unitaries are only available if your matrix type has a Dynamic number of columns (for example MatrixXf). They also are not
+   * available with the (non - default) FullPivHouseholderQR preconditioner.
+   */
+  BDCSVD(const MatrixType& matrix, unsigned int computationOptions = 0)
+    : m_algoswap(16), m_numIters(0)
+  {
+    compute(matrix, computationOptions);
+  }
+
+  ~BDCSVD() 
+  {
+  }
+  
+  /** \brief Method performing the decomposition of given matrix using custom options.
+   *
+   * \param matrix the matrix to decompose
+   * \param computationOptions optional parameter allowing to specify if you want full or thin U or V unitaries to be computed.
+   *                           By default, none is computed. This is a bit - field, the possible bits are #ComputeFullU, #ComputeThinU, 
+   *                           #ComputeFullV, #ComputeThinV.
+   *
+   * Thin unitaries are only available if your matrix type has a Dynamic number of columns (for example MatrixXf). They also are not
+   * available with the (non - default) FullPivHouseholderQR preconditioner.
+   */
+  BDCSVD& compute(const MatrixType& matrix, unsigned int computationOptions);
+
+  /** \brief Method performing the decomposition of given matrix using current options.
+   *
+   * \param matrix the matrix to decompose
+   *
+   * This method uses the current \a computationOptions, as already passed to the constructor or to compute(const MatrixType&, unsigned int).
+   */
+  BDCSVD& compute(const MatrixType& matrix)
+  {
+    return compute(matrix, this->m_computationOptions);
+  }
+
+  void setSwitchSize(int s) 
+  {
+    eigen_assert(s>3 && "BDCSVD the size of the algo switch has to be greater than 3");
+    m_algoswap = s;
+  }
+ 
+private:
+  void allocate(Index rows, Index cols, unsigned int computationOptions);
+  void divide(Index firstCol, Index lastCol, Index firstRowW, Index firstColW, Index shift);
+  void computeSVDofM(Index firstCol, Index n, MatrixXr& U, VectorType& singVals, MatrixXr& V);
+  void computeSingVals(const ArrayRef& col0, const ArrayRef& diag, const IndicesRef& perm, VectorType& singVals, ArrayRef shifts, ArrayRef mus);
+  void perturbCol0(const ArrayRef& col0, const ArrayRef& diag, const IndicesRef& perm, const VectorType& singVals, const ArrayRef& shifts, const ArrayRef& mus, ArrayRef zhat);
+  void computeSingVecs(const ArrayRef& zhat, const ArrayRef& diag, const IndicesRef& perm, const VectorType& singVals, const ArrayRef& shifts, const ArrayRef& mus, MatrixXr& U, MatrixXr& V);
+  void deflation43(Index firstCol, Index shift, Index i, Index size);
+  void deflation44(Index firstColu , Index firstColm, Index firstRowW, Index firstColW, Index i, Index j, Index size);
+  void deflation(Index firstCol, Index lastCol, Index k, Index firstRowW, Index firstColW, Index shift);
+  template<typename HouseholderU, typename HouseholderV, typename NaiveU, typename NaiveV>
+  void copyUV(const HouseholderU &householderU, const HouseholderV &householderV, const NaiveU &naiveU, const NaiveV &naivev);
+  void structured_update(Block<MatrixXr,Dynamic,Dynamic> A, const MatrixXr &B, Index n1);
+  static RealScalar secularEq(RealScalar x, const ArrayRef& col0, const ArrayRef& diag, const IndicesRef &perm, const ArrayRef& diagShifted, RealScalar shift);
+
+protected:
+  MatrixXr m_naiveU, m_naiveV;
+  MatrixXr m_computed;
+  Index m_nRec;
+  ArrayXr m_workspace;
+  ArrayXi m_workspaceI;
+  int m_algoswap;
+  bool m_isTranspose, m_compU, m_compV;
+  
+  using Base::m_singularValues;
+  using Base::m_diagSize;
+  using Base::m_computeFullU;
+  using Base::m_computeFullV;
+  using Base::m_computeThinU;
+  using Base::m_computeThinV;
+  using Base::m_matrixU;
+  using Base::m_matrixV;
+  using Base::m_isInitialized;
+  using Base::m_nonzeroSingularValues;
+
+public:  
+  int m_numIters;
+}; //end class BDCSVD
+
+
+// Method to allocate and initialize matrix and attributes
+template<typename MatrixType>
+void BDCSVD<MatrixType>::allocate(Index rows, Index cols, unsigned int computationOptions)
+{
+  m_isTranspose = (cols > rows);
+
+  if (Base::allocate(rows, cols, computationOptions))
+    return;
+  
+  m_computed = MatrixXr::Zero(m_diagSize + 1, m_diagSize );
+  m_compU = computeV();
+  m_compV = computeU();
+  if (m_isTranspose)
+    std::swap(m_compU, m_compV);
+  
+  if (m_compU) m_naiveU = MatrixXr::Zero(m_diagSize + 1, m_diagSize + 1 );
+  else         m_naiveU = MatrixXr::Zero(2, m_diagSize + 1 );
+  
+  if (m_compV) m_naiveV = MatrixXr::Zero(m_diagSize, m_diagSize);
+  
+  m_workspace.resize((m_diagSize+1)*(m_diagSize+1)*3);
+  m_workspaceI.resize(3*m_diagSize);
+}// end allocate
+
+template<typename MatrixType>
+BDCSVD<MatrixType>& BDCSVD<MatrixType>::compute(const MatrixType& matrix, unsigned int computationOptions) 
+{
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "\n\n\n======================================================================================================================\n\n\n";
+#endif
+  allocate(matrix.rows(), matrix.cols(), computationOptions);
+  using std::abs;
+
+  const RealScalar considerZero = (std::numeric_limits<RealScalar>::min)();
+  
+  //**** step -1 - If the problem is too small, directly falls back to JacobiSVD and return
+  if(matrix.cols() < m_algoswap)
+  {
+    // FIXME this line involves temporaries
+    JacobiSVD<MatrixType> jsvd(matrix,computationOptions);
+    if(computeU()) m_matrixU = jsvd.matrixU();
+    if(computeV()) m_matrixV = jsvd.matrixV();
+    m_singularValues = jsvd.singularValues();
+    m_nonzeroSingularValues = jsvd.nonzeroSingularValues();
+    m_isInitialized = true;
+    return *this;
+  }
+  
+  //**** step 0 - Copy the input matrix and apply scaling to reduce over/under-flows
+  RealScalar scale = matrix.cwiseAbs().maxCoeff();
+  if(scale==Literal(0)) scale = Literal(1);
+  MatrixX copy;
+  if (m_isTranspose) copy = matrix.adjoint()/scale;
+  else               copy = matrix/scale;
+  
+  //**** step 1 - Bidiagonalization
+  // FIXME this line involves temporaries
+  internal::UpperBidiagonalization<MatrixX> bid(copy);
+
+  //**** step 2 - Divide & Conquer
+  m_naiveU.setZero();
+  m_naiveV.setZero();
+  // FIXME this line involves a temporary matrix
+  m_computed.topRows(m_diagSize) = bid.bidiagonal().toDenseMatrix().transpose();
+  m_computed.template bottomRows<1>().setZero();
+  divide(0, m_diagSize - 1, 0, 0, 0);
+
+  //**** step 3 - Copy singular values and vectors
+  for (int i=0; i<m_diagSize; i++)
+  {
+    RealScalar a = abs(m_computed.coeff(i, i));
+    m_singularValues.coeffRef(i) = a * scale;
+    if (a<considerZero)
+    {
+      m_nonzeroSingularValues = i;
+      m_singularValues.tail(m_diagSize - i - 1).setZero();
+      break;
+    }
+    else if (i == m_diagSize - 1)
+    {
+      m_nonzeroSingularValues = i + 1;
+      break;
+    }
+  }
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+//   std::cout << "m_naiveU\n" << m_naiveU << "\n\n";
+//   std::cout << "m_naiveV\n" << m_naiveV << "\n\n";
+#endif
+  if(m_isTranspose) copyUV(bid.householderV(), bid.householderU(), m_naiveV, m_naiveU);
+  else              copyUV(bid.householderU(), bid.householderV(), m_naiveU, m_naiveV);
+
+  m_isInitialized = true;
+  return *this;
+}// end compute
+
+
+template<typename MatrixType>
+template<typename HouseholderU, typename HouseholderV, typename NaiveU, typename NaiveV>
+void BDCSVD<MatrixType>::copyUV(const HouseholderU &householderU, const HouseholderV &householderV, const NaiveU &naiveU, const NaiveV &naiveV)
+{
+  // Note exchange of U and V: m_matrixU is set from m_naiveV and vice versa
+  if (computeU())
+  {
+    Index Ucols = m_computeThinU ? m_diagSize : householderU.cols();
+    m_matrixU = MatrixX::Identity(householderU.cols(), Ucols);
+    m_matrixU.topLeftCorner(m_diagSize, m_diagSize) = naiveV.template cast<Scalar>().topLeftCorner(m_diagSize, m_diagSize);
+    householderU.applyThisOnTheLeft(m_matrixU); // FIXME this line involves a temporary buffer
+  }
+  if (computeV())
+  {
+    Index Vcols = m_computeThinV ? m_diagSize : householderV.cols();
+    m_matrixV = MatrixX::Identity(householderV.cols(), Vcols);
+    m_matrixV.topLeftCorner(m_diagSize, m_diagSize) = naiveU.template cast<Scalar>().topLeftCorner(m_diagSize, m_diagSize);
+    householderV.applyThisOnTheLeft(m_matrixV); // FIXME this line involves a temporary buffer
+  }
+}
+
+/** \internal
+  * Performs A = A * B exploiting the special structure of the matrix A. Splitting A as:
+  *  A = [A1]
+  *      [A2]
+  * such that A1.rows()==n1, then we assume that at least half of the columns of A1 and A2 are zeros.
+  * We can thus pack them prior to the the matrix product. However, this is only worth the effort if the matrix is large
+  * enough.
+  */
+template<typename MatrixType>
+void BDCSVD<MatrixType>::structured_update(Block<MatrixXr,Dynamic,Dynamic> A, const MatrixXr &B, Index n1)
+{
+  Index n = A.rows();
+  if(n>100)
+  {
+    // If the matrices are large enough, let's exploit the sparse structure of A by
+    // splitting it in half (wrt n1), and packing the non-zero columns.
+    Index n2 = n - n1;
+    Map<MatrixXr> A1(m_workspace.data()      , n1, n);
+    Map<MatrixXr> A2(m_workspace.data()+ n1*n, n2, n);
+    Map<MatrixXr> B1(m_workspace.data()+  n*n, n,  n);
+    Map<MatrixXr> B2(m_workspace.data()+2*n*n, n,  n);
+    Index k1=0, k2=0;
+    for(Index j=0; j<n; ++j)
+    {
+      if( (A.col(j).head(n1).array()!=Literal(0)).any() )
+      {
+        A1.col(k1) = A.col(j).head(n1);
+        B1.row(k1) = B.row(j);
+        ++k1;
+      }
+      if( (A.col(j).tail(n2).array()!=Literal(0)).any() )
+      {
+        A2.col(k2) = A.col(j).tail(n2);
+        B2.row(k2) = B.row(j);
+        ++k2;
+      }
+    }
+  
+    A.topRows(n1).noalias()    = A1.leftCols(k1) * B1.topRows(k1);
+    A.bottomRows(n2).noalias() = A2.leftCols(k2) * B2.topRows(k2);
+  }
+  else
+  {
+    Map<MatrixXr,Aligned> tmp(m_workspace.data(),n,n);
+    tmp.noalias() = A*B;
+    A = tmp;
+  }
+}
+
+// The divide algorithm is done "in place", we are always working on subsets of the same matrix. The divide methods takes as argument the 
+// place of the submatrix we are currently working on.
+
+//@param firstCol : The Index of the first column of the submatrix of m_computed and for m_naiveU;
+//@param lastCol : The Index of the last column of the submatrix of m_computed and for m_naiveU; 
+// lastCol + 1 - firstCol is the size of the submatrix.
+//@param firstRowW : The Index of the first row of the matrix W that we are to change. (see the reference paper section 1 for more information on W)
+//@param firstRowW : Same as firstRowW with the column.
+//@param shift : Each time one takes the left submatrix, one must add 1 to the shift. Why? Because! We actually want the last column of the U submatrix 
+// to become the first column (*coeff) and to shift all the other columns to the right. There are more details on the reference paper.
+template<typename MatrixType>
+void BDCSVD<MatrixType>::divide (Index firstCol, Index lastCol, Index firstRowW, Index firstColW, Index shift)
+{
+  // requires rows = cols + 1;
+  using std::pow;
+  using std::sqrt;
+  using std::abs;
+  const Index n = lastCol - firstCol + 1;
+  const Index k = n/2;
+  const RealScalar considerZero = (std::numeric_limits<RealScalar>::min)();
+  RealScalar alphaK;
+  RealScalar betaK; 
+  RealScalar r0; 
+  RealScalar lambda, phi, c0, s0;
+  VectorType l, f;
+  // We use the other algorithm which is more efficient for small 
+  // matrices.
+  if (n < m_algoswap)
+  {
+    // FIXME this line involves temporaries
+    JacobiSVD<MatrixXr> b(m_computed.block(firstCol, firstCol, n + 1, n), ComputeFullU | (m_compV ? ComputeFullV : 0));
+    if (m_compU)
+      m_naiveU.block(firstCol, firstCol, n + 1, n + 1).real() = b.matrixU();
+    else 
+    {
+      m_naiveU.row(0).segment(firstCol, n + 1).real() = b.matrixU().row(0);
+      m_naiveU.row(1).segment(firstCol, n + 1).real() = b.matrixU().row(n);
+    }
+    if (m_compV) m_naiveV.block(firstRowW, firstColW, n, n).real() = b.matrixV();
+    m_computed.block(firstCol + shift, firstCol + shift, n + 1, n).setZero();
+    m_computed.diagonal().segment(firstCol + shift, n) = b.singularValues().head(n);
+    return;
+  }
+  // We use the divide and conquer algorithm
+  alphaK =  m_computed(firstCol + k, firstCol + k);
+  betaK = m_computed(firstCol + k + 1, firstCol + k);
+  // The divide must be done in that order in order to have good results. Divide change the data inside the submatrices
+  // and the divide of the right submatrice reads one column of the left submatrice. That's why we need to treat the 
+  // right submatrix before the left one. 
+  divide(k + 1 + firstCol, lastCol, k + 1 + firstRowW, k + 1 + firstColW, shift);
+  divide(firstCol, k - 1 + firstCol, firstRowW, firstColW + 1, shift + 1);
+
+  if (m_compU)
+  {
+    lambda = m_naiveU(firstCol + k, firstCol + k);
+    phi = m_naiveU(firstCol + k + 1, lastCol + 1);
+  } 
+  else 
+  {
+    lambda = m_naiveU(1, firstCol + k);
+    phi = m_naiveU(0, lastCol + 1);
+  }
+  r0 = sqrt((abs(alphaK * lambda) * abs(alphaK * lambda)) + abs(betaK * phi) * abs(betaK * phi));
+  if (m_compU)
+  {
+    l = m_naiveU.row(firstCol + k).segment(firstCol, k);
+    f = m_naiveU.row(firstCol + k + 1).segment(firstCol + k + 1, n - k - 1);
+  } 
+  else 
+  {
+    l = m_naiveU.row(1).segment(firstCol, k);
+    f = m_naiveU.row(0).segment(firstCol + k + 1, n - k - 1);
+  }
+  if (m_compV) m_naiveV(firstRowW+k, firstColW) = Literal(1);
+  if (r0<considerZero)
+  {
+    c0 = Literal(1);
+    s0 = Literal(0);
+  }
+  else
+  {
+    c0 = alphaK * lambda / r0;
+    s0 = betaK * phi / r0;
+  }
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+  
+  if (m_compU)
+  {
+    MatrixXr q1 (m_naiveU.col(firstCol + k).segment(firstCol, k + 1));     
+    // we shiftW Q1 to the right
+    for (Index i = firstCol + k - 1; i >= firstCol; i--) 
+      m_naiveU.col(i + 1).segment(firstCol, k + 1) = m_naiveU.col(i).segment(firstCol, k + 1);
+    // we shift q1 at the left with a factor c0
+    m_naiveU.col(firstCol).segment( firstCol, k + 1) = (q1 * c0);
+    // last column = q1 * - s0
+    m_naiveU.col(lastCol + 1).segment(firstCol, k + 1) = (q1 * ( - s0));
+    // first column = q2 * s0
+    m_naiveU.col(firstCol).segment(firstCol + k + 1, n - k) = m_naiveU.col(lastCol + 1).segment(firstCol + k + 1, n - k) * s0; 
+    // q2 *= c0
+    m_naiveU.col(lastCol + 1).segment(firstCol + k + 1, n - k) *= c0;
+  } 
+  else 
+  {
+    RealScalar q1 = m_naiveU(0, firstCol + k);
+    // we shift Q1 to the right
+    for (Index i = firstCol + k - 1; i >= firstCol; i--) 
+      m_naiveU(0, i + 1) = m_naiveU(0, i);
+    // we shift q1 at the left with a factor c0
+    m_naiveU(0, firstCol) = (q1 * c0);
+    // last column = q1 * - s0
+    m_naiveU(0, lastCol + 1) = (q1 * ( - s0));
+    // first column = q2 * s0
+    m_naiveU(1, firstCol) = m_naiveU(1, lastCol + 1) *s0; 
+    // q2 *= c0
+    m_naiveU(1, lastCol + 1) *= c0;
+    m_naiveU.row(1).segment(firstCol + 1, k).setZero();
+    m_naiveU.row(0).segment(firstCol + k + 1, n - k - 1).setZero();
+  }
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+  
+  m_computed(firstCol + shift, firstCol + shift) = r0;
+  m_computed.col(firstCol + shift).segment(firstCol + shift + 1, k) = alphaK * l.transpose().real();
+  m_computed.col(firstCol + shift).segment(firstCol + shift + k + 1, n - k - 1) = betaK * f.transpose().real();
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  ArrayXr tmp1 = (m_computed.block(firstCol+shift, firstCol+shift, n, n)).jacobiSvd().singularValues();
+#endif
+  // Second part: try to deflate singular values in combined matrix
+  deflation(firstCol, lastCol, k, firstRowW, firstColW, shift);
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  ArrayXr tmp2 = (m_computed.block(firstCol+shift, firstCol+shift, n, n)).jacobiSvd().singularValues();
+  std::cout << "\n\nj1 = " << tmp1.transpose().format(bdcsvdfmt) << "\n";
+  std::cout << "j2 = " << tmp2.transpose().format(bdcsvdfmt) << "\n\n";
+  std::cout << "err:      " << ((tmp1-tmp2).abs()>1e-12*tmp2.abs()).transpose() << "\n";
+  static int count = 0;
+  std::cout << "# " << ++count << "\n\n";
+  assert((tmp1-tmp2).matrix().norm() < 1e-14*tmp2.matrix().norm());
+//   assert(count<681);
+//   assert(((tmp1-tmp2).abs()<1e-13*tmp2.abs()).all());
+#endif
+  
+  // Third part: compute SVD of combined matrix
+  MatrixXr UofSVD, VofSVD;
+  VectorType singVals;
+  computeSVDofM(firstCol + shift, n, UofSVD, singVals, VofSVD);
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(UofSVD.allFinite());
+  assert(VofSVD.allFinite());
+#endif
+  
+  if (m_compU)
+    structured_update(m_naiveU.block(firstCol, firstCol, n + 1, n + 1), UofSVD, (n+2)/2);
+  else
+  {
+    Map<Matrix<RealScalar,2,Dynamic>,Aligned> tmp(m_workspace.data(),2,n+1);
+    tmp.noalias() = m_naiveU.middleCols(firstCol, n+1) * UofSVD;
+    m_naiveU.middleCols(firstCol, n + 1) = tmp;
+  }
+  
+  if (m_compV)  structured_update(m_naiveV.block(firstRowW, firstColW, n, n), VofSVD, (n+1)/2);
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+  
+  m_computed.block(firstCol + shift, firstCol + shift, n, n).setZero();
+  m_computed.block(firstCol + shift, firstCol + shift, n, n).diagonal() = singVals;
+}// end divide
+
+// Compute SVD of m_computed.block(firstCol, firstCol, n + 1, n); this block only has non-zeros in
+// the first column and on the diagonal and has undergone deflation, so diagonal is in increasing
+// order except for possibly the (0,0) entry. The computed SVD is stored U, singVals and V, except
+// that if m_compV is false, then V is not computed. Singular values are sorted in decreasing order.
+//
+// TODO Opportunities for optimization: better root finding algo, better stopping criterion, better
+// handling of round-off errors, be consistent in ordering
+// For instance, to solve the secular equation using FMM, see http://www.stat.uchicago.edu/~lekheng/courses/302/classics/greengard-rokhlin.pdf
+template <typename MatrixType>
+void BDCSVD<MatrixType>::computeSVDofM(Index firstCol, Index n, MatrixXr& U, VectorType& singVals, MatrixXr& V)
+{
+  const RealScalar considerZero = (std::numeric_limits<RealScalar>::min)();
+  using std::abs;
+  ArrayRef col0 = m_computed.col(firstCol).segment(firstCol, n);
+  m_workspace.head(n) =  m_computed.block(firstCol, firstCol, n, n).diagonal();
+  ArrayRef diag = m_workspace.head(n);
+  diag(0) = Literal(0);
+
+  // Allocate space for singular values and vectors
+  singVals.resize(n);
+  U.resize(n+1, n+1);
+  if (m_compV) V.resize(n, n);
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  if (col0.hasNaN() || diag.hasNaN())
+    std::cout << "\n\nHAS NAN\n\n";
+#endif
+  
+  // Many singular values might have been deflated, the zero ones have been moved to the end,
+  // but others are interleaved and we must ignore them at this stage.
+  // To this end, let's compute a permutation skipping them:
+  Index actual_n = n;
+  while(actual_n>1 && diag(actual_n-1)==Literal(0)) --actual_n;
+  Index m = 0; // size of the deflated problem
+  for(Index k=0;k<actual_n;++k)
+    if(abs(col0(k))>considerZero)
+      m_workspaceI(m++) = k;
+  Map<ArrayXi> perm(m_workspaceI.data(),m);
+  
+  Map<ArrayXr> shifts(m_workspace.data()+1*n, n);
+  Map<ArrayXr> mus(m_workspace.data()+2*n, n);
+  Map<ArrayXr> zhat(m_workspace.data()+3*n, n);
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "computeSVDofM using:\n";
+  std::cout << "  z: " << col0.transpose() << "\n";
+  std::cout << "  d: " << diag.transpose() << "\n";
+#endif
+  
+  // Compute singVals, shifts, and mus
+  computeSingVals(col0, diag, perm, singVals, shifts, mus);
+  
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "  j:        " << (m_computed.block(firstCol, firstCol, n, n)).jacobiSvd().singularValues().transpose().reverse() << "\n\n";
+  std::cout << "  sing-val: " << singVals.transpose() << "\n";
+  std::cout << "  mu:       " << mus.transpose() << "\n";
+  std::cout << "  shift:    " << shifts.transpose() << "\n";
+  
+  {
+    Index actual_n = n;
+    while(actual_n>1 && abs(col0(actual_n-1))<considerZero) --actual_n;
+    std::cout << "\n\n    mus:    " << mus.head(actual_n).transpose() << "\n\n";
+    std::cout << "    check1 (expect0) : " << ((singVals.array()-(shifts+mus)) / singVals.array()).head(actual_n).transpose() << "\n\n";
+    std::cout << "    check2 (>0)      : " << ((singVals.array()-diag) / singVals.array()).head(actual_n).transpose() << "\n\n";
+    std::cout << "    check3 (>0)      : " << ((diag.segment(1,actual_n-1)-singVals.head(actual_n-1).array()) / singVals.head(actual_n-1).array()).transpose() << "\n\n\n";
+    std::cout << "    check4 (>0)      : " << ((singVals.segment(1,actual_n-1)-singVals.head(actual_n-1))).transpose() << "\n\n\n";
+  }
+#endif
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(singVals.allFinite());
+  assert(mus.allFinite());
+  assert(shifts.allFinite());
+#endif
+  
+  // Compute zhat
+  perturbCol0(col0, diag, perm, singVals, shifts, mus, zhat);
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "  zhat: " << zhat.transpose() << "\n";
+#endif
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(zhat.allFinite());
+#endif
+  
+  computeSingVecs(zhat, diag, perm, singVals, shifts, mus, U, V);
+  
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "U^T U: " << (U.transpose() * U - MatrixXr(MatrixXr::Identity(U.cols(),U.cols()))).norm() << "\n";
+  std::cout << "V^T V: " << (V.transpose() * V - MatrixXr(MatrixXr::Identity(V.cols(),V.cols()))).norm() << "\n";
+#endif
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(U.allFinite());
+  assert(V.allFinite());
+  assert((U.transpose() * U - MatrixXr(MatrixXr::Identity(U.cols(),U.cols()))).norm() < 1e-14 * n);
+  assert((V.transpose() * V - MatrixXr(MatrixXr::Identity(V.cols(),V.cols()))).norm() < 1e-14 * n);
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+  
+  // Because of deflation, the singular values might not be completely sorted.
+  // Fortunately, reordering them is a O(n) problem
+  for(Index i=0; i<actual_n-1; ++i)
+  {
+    if(singVals(i)>singVals(i+1))
+    {
+      using std::swap;
+      swap(singVals(i),singVals(i+1));
+      U.col(i).swap(U.col(i+1));
+      if(m_compV) V.col(i).swap(V.col(i+1));
+    }
+  }
+  
+  // Reverse order so that singular values in increased order
+  // Because of deflation, the zeros singular-values are already at the end
+  singVals.head(actual_n).reverseInPlace();
+  U.leftCols(actual_n).rowwise().reverseInPlace();
+  if (m_compV) V.leftCols(actual_n).rowwise().reverseInPlace();
+  
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  JacobiSVD<MatrixXr> jsvd(m_computed.block(firstCol, firstCol, n, n) );
+  std::cout << "  * j:        " << jsvd.singularValues().transpose() << "\n\n";
+  std::cout << "  * sing-val: " << singVals.transpose() << "\n";
+//   std::cout << "  * err:      " << ((jsvd.singularValues()-singVals)>1e-13*singVals.norm()).transpose() << "\n";
+#endif
+}
+
+template <typename MatrixType>
+typename BDCSVD<MatrixType>::RealScalar BDCSVD<MatrixType>::secularEq(RealScalar mu, const ArrayRef& col0, const ArrayRef& diag, const IndicesRef &perm, const ArrayRef& diagShifted, RealScalar shift)
+{
+  Index m = perm.size();
+  RealScalar res = Literal(1);
+  for(Index i=0; i<m; ++i)
+  {
+    Index j = perm(i);
+    // The following expression could be rewritten to involve only a single division,
+    // but this would make the expression more sensitive to overflow.
+    res += (col0(j) / (diagShifted(j) - mu)) * (col0(j) / (diag(j) + shift + mu));
+  }
+  return res;
+
+}
+
+template <typename MatrixType>
+void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& diag, const IndicesRef &perm,
+                                         VectorType& singVals, ArrayRef shifts, ArrayRef mus)
+{
+  using std::abs;
+  using std::swap;
+  using std::sqrt;
+
+  Index n = col0.size();
+  Index actual_n = n;
+  // Note that here actual_n is computed based on col0(i)==0 instead of diag(i)==0 as above
+  // because 1) we have diag(i)==0 => col0(i)==0 and 2) if col0(i)==0, then diag(i) is already a singular value.
+  while(actual_n>1 && col0(actual_n-1)==Literal(0)) --actual_n;
+
+  for (Index k = 0; k < n; ++k)
+  {
+    if (col0(k) == Literal(0) || actual_n==1)
+    {
+      // if col0(k) == 0, then entry is deflated, so singular value is on diagonal
+      // if actual_n==1, then the deflated problem is already diagonalized
+      singVals(k) = k==0 ? col0(0) : diag(k);
+      mus(k) = Literal(0);
+      shifts(k) = k==0 ? col0(0) : diag(k);
+      continue;
+    } 
+
+    // otherwise, use secular equation to find singular value
+    RealScalar left = diag(k);
+    RealScalar right; // was: = (k != actual_n-1) ? diag(k+1) : (diag(actual_n-1) + col0.matrix().norm());
+    if(k==actual_n-1)
+      right = (diag(actual_n-1) + col0.matrix().norm());
+    else
+    {
+      // Skip deflated singular values,
+      // recall that at this stage we assume that z[j]!=0 and all entries for which z[j]==0 have been put aside.
+      // This should be equivalent to using perm[]
+      Index l = k+1;
+      while(col0(l)==Literal(0)) { ++l; eigen_internal_assert(l<actual_n); }
+      right = diag(l);
+    }
+
+    // first decide whether it's closer to the left end or the right end
+    RealScalar mid = left + (right-left) / Literal(2);
+    RealScalar fMid = secularEq(mid, col0, diag, perm, diag, Literal(0));
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+    std::cout << right-left << "\n";
+    std::cout << "fMid = " << fMid << " " << secularEq(mid-left, col0, diag, perm, diag-left, left) << " " << secularEq(mid-right, col0, diag, perm, diag-right, right)   << "\n";
+    std::cout << "     = " << secularEq(0.1*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.2*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.3*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.4*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.49*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.5*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.51*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.6*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.7*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.8*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.9*(left+right), col0, diag, perm, diag, 0) << "\n";
+#endif
+    RealScalar shift = (k == actual_n-1 || fMid > Literal(0)) ? left : right;
+    
+    // measure everything relative to shift
+    Map<ArrayXr> diagShifted(m_workspace.data()+4*n, n);
+    diagShifted = diag - shift;
+    
+    // initial guess
+    RealScalar muPrev, muCur;
+    if (shift == left)
+    {
+      muPrev = (right - left) * RealScalar(0.1);
+      if (k == actual_n-1) muCur = right - left;
+      else                 muCur = (right - left) * RealScalar(0.5);
+    }
+    else
+    {
+      muPrev = -(right - left) * RealScalar(0.1);
+      muCur = -(right - left) * RealScalar(0.5);
+    }
+
+    RealScalar fPrev = secularEq(muPrev, col0, diag, perm, diagShifted, shift);
+    RealScalar fCur = secularEq(muCur, col0, diag, perm, diagShifted, shift);
+    if (abs(fPrev) < abs(fCur))
+    {
+      swap(fPrev, fCur);
+      swap(muPrev, muCur);
+    }
+
+    // rational interpolation: fit a function of the form a / mu + b through the two previous
+    // iterates and use its zero to compute the next iterate
+    bool useBisection = fPrev*fCur>Literal(0);
+    while (fCur!=Literal(0) && abs(muCur - muPrev) > Literal(8) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(muCur), abs(muPrev)) && abs(fCur - fPrev)>NumTraits<RealScalar>::epsilon() && !useBisection)
+    {
+      ++m_numIters;
+
+      // Find a and b such that the function f(mu) = a / mu + b matches the current and previous samples.
+      RealScalar a = (fCur - fPrev) / (Literal(1)/muCur - Literal(1)/muPrev);
+      RealScalar b = fCur - a / muCur;
+      // And find mu such that f(mu)==0:
+      RealScalar muZero = -a/b;
+      RealScalar fZero = secularEq(muZero, col0, diag, perm, diagShifted, shift);
+      
+      muPrev = muCur;
+      fPrev = fCur;
+      muCur = muZero;
+      fCur = fZero;
+      
+      
+      if (shift == left  && (muCur < Literal(0) || muCur > right - left)) useBisection = true;
+      if (shift == right && (muCur < -(right - left) || muCur > Literal(0))) useBisection = true;
+      if (abs(fCur)>abs(fPrev)) useBisection = true;
+    }
+
+    // fall back on bisection method if rational interpolation did not work
+    if (useBisection)
+    {
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+      std::cout << "useBisection for k = " << k << ", actual_n = " << actual_n << "\n";
+#endif
+      RealScalar leftShifted, rightShifted;
+      if (shift == left)
+      {
+        // to avoid overflow, we must have mu > max(real_min, |z(k)|/sqrt(real_max)),
+        // the factor 2 is to be more conservative
+        leftShifted = numext::maxi<RealScalar>( (std::numeric_limits<RealScalar>::min)(), Literal(2) * abs(col0(k)) / sqrt((std::numeric_limits<RealScalar>::max)()) );
+
+        // check that we did it right:
+        eigen_internal_assert( (numext::isfinite)( (col0(k)/leftShifted)*(col0(k)/(diag(k)+shift+leftShifted)) ) );
+        // I don't understand why the case k==0 would be special there:
+        // if (k == 0) rightShifted = right - left; else
+        rightShifted = (k==actual_n-1) ? right : ((right - left) * RealScalar(0.51)); // theoretically we can take 0.5, but let's be safe
+      }
+      else
+      {
+        leftShifted = -(right - left) * RealScalar(0.51);
+        if(k+1<n)
+          rightShifted = -numext::maxi<RealScalar>( (std::numeric_limits<RealScalar>::min)(), abs(col0(k+1)) / sqrt((std::numeric_limits<RealScalar>::max)()) );
+        else
+          rightShifted = -(std::numeric_limits<RealScalar>::min)();
+      }
+      
+      RealScalar fLeft = secularEq(leftShifted, col0, diag, perm, diagShifted, shift);
+
+#if defined EIGEN_INTERNAL_DEBUGGING || defined EIGEN_BDCSVD_DEBUG_VERBOSE
+      RealScalar fRight = secularEq(rightShifted, col0, diag, perm, diagShifted, shift);
+#endif
+
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+      if(!(fLeft * fRight<0))
+      {
+        std::cout << "fLeft: " << leftShifted << " - " << diagShifted.head(10).transpose()  << "\n ; " << bool(left==shift) << " " << (left-shift) << "\n";
+        std::cout << k << " : " <<  fLeft << " * " << fRight << " == " << fLeft * fRight << "  ;  " << left << " - " << right << " -> " <<  leftShifted << " " << rightShifted << "   shift=" << shift << "\n";
+      }
+#endif
+      eigen_internal_assert(fLeft * fRight < Literal(0));
+      
+      while (rightShifted - leftShifted > Literal(2) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(leftShifted), abs(rightShifted)))
+      {
+        RealScalar midShifted = (leftShifted + rightShifted) / Literal(2);
+        fMid = secularEq(midShifted, col0, diag, perm, diagShifted, shift);
+        if (fLeft * fMid < Literal(0))
+        {
+          rightShifted = midShifted;
+        }
+        else
+        {
+          leftShifted = midShifted;
+          fLeft = fMid;
+        }
+      }
+
+      muCur = (leftShifted + rightShifted) / Literal(2);
+    }
+      
+    singVals[k] = shift + muCur;
+    shifts[k] = shift;
+    mus[k] = muCur;
+
+    // perturb singular value slightly if it equals diagonal entry to avoid division by zero later
+    // (deflation is supposed to avoid this from happening)
+    // - this does no seem to be necessary anymore -
+//     if (singVals[k] == left) singVals[k] *= 1 + NumTraits<RealScalar>::epsilon();
+//     if (singVals[k] == right) singVals[k] *= 1 - NumTraits<RealScalar>::epsilon();
+  }
+}
+
+
+// zhat is perturbation of col0 for which singular vectors can be computed stably (see Section 3.1)
+template <typename MatrixType>
+void BDCSVD<MatrixType>::perturbCol0
+   (const ArrayRef& col0, const ArrayRef& diag, const IndicesRef &perm, const VectorType& singVals,
+    const ArrayRef& shifts, const ArrayRef& mus, ArrayRef zhat)
+{
+  using std::sqrt;
+  Index n = col0.size();
+  Index m = perm.size();
+  if(m==0)
+  {
+    zhat.setZero();
+    return;
+  }
+  Index last = perm(m-1);
+  // The offset permits to skip deflated entries while computing zhat
+  for (Index k = 0; k < n; ++k)
+  {
+    if (col0(k) == Literal(0)) // deflated
+      zhat(k) = Literal(0);
+    else
+    {
+      // see equation (3.6)
+      RealScalar dk = diag(k);
+      RealScalar prod = (singVals(last) + dk) * (mus(last) + (shifts(last) - dk));
+
+      for(Index l = 0; l<m; ++l)
+      {
+        Index i = perm(l);
+        if(i!=k)
+        {
+          Index j = i<k ? i : perm(l-1);
+          prod *= ((singVals(j)+dk) / ((diag(i)+dk))) * ((mus(j)+(shifts(j)-dk)) / ((diag(i)-dk)));
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+          if(i!=k && std::abs(((singVals(j)+dk)*(mus(j)+(shifts(j)-dk)))/((diag(i)+dk)*(diag(i)-dk)) - 1) > 0.9 )
+            std::cout << "     " << ((singVals(j)+dk)*(mus(j)+(shifts(j)-dk)))/((diag(i)+dk)*(diag(i)-dk)) << " == (" << (singVals(j)+dk) << " * " << (mus(j)+(shifts(j)-dk))
+                       << ") / (" << (diag(i)+dk) << " * " << (diag(i)-dk) << ")\n";
+#endif
+        }
+      }
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+      std::cout << "zhat(" << k << ") =  sqrt( " << prod << ")  ;  " << (singVals(last) + dk) << " * " << mus(last) + shifts(last) << " - " << dk << "\n";
+#endif
+      RealScalar tmp = sqrt(prod);
+      zhat(k) = col0(k) > Literal(0) ? tmp : -tmp;
+    }
+  }
+}
+
+// compute singular vectors
+template <typename MatrixType>
+void BDCSVD<MatrixType>::computeSingVecs
+   (const ArrayRef& zhat, const ArrayRef& diag, const IndicesRef &perm, const VectorType& singVals,
+    const ArrayRef& shifts, const ArrayRef& mus, MatrixXr& U, MatrixXr& V)
+{
+  Index n = zhat.size();
+  Index m = perm.size();
+  
+  for (Index k = 0; k < n; ++k)
+  {
+    if (zhat(k) == Literal(0))
+    {
+      U.col(k) = VectorType::Unit(n+1, k);
+      if (m_compV) V.col(k) = VectorType::Unit(n, k);
+    }
+    else
+    {
+      U.col(k).setZero();
+      for(Index l=0;l<m;++l)
+      {
+        Index i = perm(l);
+        U(i,k) = zhat(i)/(((diag(i) - shifts(k)) - mus(k)) )/( (diag(i) + singVals[k]));
+      }
+      U(n,k) = Literal(0);
+      U.col(k).normalize();
+    
+      if (m_compV)
+      {
+        V.col(k).setZero();
+        for(Index l=1;l<m;++l)
+        {
+          Index i = perm(l);
+          V(i,k) = diag(i) * zhat(i) / (((diag(i) - shifts(k)) - mus(k)) )/( (diag(i) + singVals[k]));
+        }
+        V(0,k) = Literal(-1);
+        V.col(k).normalize();
+      }
+    }
+  }
+  U.col(n) = VectorType::Unit(n+1, n);
+}
+
+
+// page 12_13
+// i >= 1, di almost null and zi non null.
+// We use a rotation to zero out zi applied to the left of M
+template <typename MatrixType>
+void BDCSVD<MatrixType>::deflation43(Index firstCol, Index shift, Index i, Index size)
+{
+  using std::abs;
+  using std::sqrt;
+  using std::pow;
+  Index start = firstCol + shift;
+  RealScalar c = m_computed(start, start);
+  RealScalar s = m_computed(start+i, start);
+  RealScalar r = numext::hypot(c,s);
+  if (r == Literal(0))
+  {
+    m_computed(start+i, start+i) = Literal(0);
+    return;
+  }
+  m_computed(start,start) = r;  
+  m_computed(start+i, start) = Literal(0);
+  m_computed(start+i, start+i) = Literal(0);
+  
+  JacobiRotation<RealScalar> J(c/r,-s/r);
+  if (m_compU)  m_naiveU.middleRows(firstCol, size+1).applyOnTheRight(firstCol, firstCol+i, J);
+  else          m_naiveU.applyOnTheRight(firstCol, firstCol+i, J);
+}// end deflation 43
+
+
+// page 13
+// i,j >= 1, i!=j and |di - dj| < epsilon * norm2(M)
+// We apply two rotations to have zj = 0;
+// TODO deflation44 is still broken and not properly tested
+template <typename MatrixType>
+void BDCSVD<MatrixType>::deflation44(Index firstColu , Index firstColm, Index firstRowW, Index firstColW, Index i, Index j, Index size)
+{
+  using std::abs;
+  using std::sqrt;
+  using std::conj;
+  using std::pow;
+  RealScalar c = m_computed(firstColm+i, firstColm);
+  RealScalar s = m_computed(firstColm+j, firstColm);
+  RealScalar r = sqrt(numext::abs2(c) + numext::abs2(s));
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "deflation 4.4: " << i << "," << j << " -> " << c << " " << s << " " << r << " ; "
+    << m_computed(firstColm + i-1, firstColm)  << " "
+    << m_computed(firstColm + i, firstColm)  << " "
+    << m_computed(firstColm + i+1, firstColm) << " "
+    << m_computed(firstColm + i+2, firstColm) << "\n";
+  std::cout << m_computed(firstColm + i-1, firstColm + i-1)  << " "
+    << m_computed(firstColm + i, firstColm+i)  << " "
+    << m_computed(firstColm + i+1, firstColm+i+1) << " "
+    << m_computed(firstColm + i+2, firstColm+i+2) << "\n";
+#endif
+  if (r==Literal(0))
+  {
+    m_computed(firstColm + i, firstColm + i) = m_computed(firstColm + j, firstColm + j);
+    return;
+  }
+  c/=r;
+  s/=r;
+  m_computed(firstColm + i, firstColm) = r;  
+  m_computed(firstColm + j, firstColm + j) = m_computed(firstColm + i, firstColm + i);
+  m_computed(firstColm + j, firstColm) = Literal(0);
+
+  JacobiRotation<RealScalar> J(c,-s);
+  if (m_compU)  m_naiveU.middleRows(firstColu, size+1).applyOnTheRight(firstColu + i, firstColu + j, J);
+  else          m_naiveU.applyOnTheRight(firstColu+i, firstColu+j, J);
+  if (m_compV)  m_naiveV.middleRows(firstRowW, size).applyOnTheRight(firstColW + i, firstColW + j, J);
+}// end deflation 44
+
+
+// acts on block from (firstCol+shift, firstCol+shift) to (lastCol+shift, lastCol+shift) [inclusive]
+template <typename MatrixType>
+void BDCSVD<MatrixType>::deflation(Index firstCol, Index lastCol, Index k, Index firstRowW, Index firstColW, Index shift)
+{
+  using std::sqrt;
+  using std::abs;
+  const Index length = lastCol + 1 - firstCol;
+  
+  Block<MatrixXr,Dynamic,1> col0(m_computed, firstCol+shift, firstCol+shift, length, 1);
+  Diagonal<MatrixXr> fulldiag(m_computed);
+  VectorBlock<Diagonal<MatrixXr>,Dynamic> diag(fulldiag, firstCol+shift, length);
+  
+  const RealScalar considerZero = (std::numeric_limits<RealScalar>::min)();
+  RealScalar maxDiag = diag.tail((std::max)(Index(1),length-1)).cwiseAbs().maxCoeff();
+  RealScalar epsilon_strict = numext::maxi<RealScalar>(considerZero,NumTraits<RealScalar>::epsilon() * maxDiag);
+  RealScalar epsilon_coarse = Literal(8) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(col0.cwiseAbs().maxCoeff(), maxDiag);
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE  
+  std::cout << "\ndeflate:" << diag.head(k+1).transpose() << "  |  " << diag.segment(k+1,length-k-1).transpose() << "\n";
+#endif
+  
+  //condition 4.1
+  if (diag(0) < epsilon_coarse)
+  { 
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+    std::cout << "deflation 4.1, because " << diag(0) << " < " << epsilon_coarse << "\n";
+#endif
+    diag(0) = epsilon_coarse;
+  }
+
+  //condition 4.2
+  for (Index i=1;i<length;++i)
+    if (abs(col0(i)) < epsilon_strict)
+    {
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+      std::cout << "deflation 4.2, set z(" << i << ") to zero because " << abs(col0(i)) << " < " << epsilon_strict << "  (diag(" << i << ")=" << diag(i) << ")\n";
+#endif
+      col0(i) = Literal(0);
+    }
+
+  //condition 4.3
+  for (Index i=1;i<length; i++)
+    if (diag(i) < epsilon_coarse)
+    {
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+      std::cout << "deflation 4.3, cancel z(" << i << ")=" << col0(i) << " because diag(" << i << ")=" << diag(i) << " < " << epsilon_coarse << "\n";
+#endif
+      deflation43(firstCol, shift, i, length);
+    }
+
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "to be sorted: " << diag.transpose() << "\n\n";
+#endif
+  {
+    // Check for total deflation
+    // If we have a total deflation, then we have to consider col0(0)==diag(0) as a singular value during sorting
+    bool total_deflation = (col0.tail(length-1).array()<considerZero).all();
+    
+    // Sort the diagonal entries, since diag(1:k-1) and diag(k:length) are already sorted, let's do a sorted merge.
+    // First, compute the respective permutation.
+    Index *permutation = m_workspaceI.data();
+    {
+      permutation[0] = 0;
+      Index p = 1;
+      
+      // Move deflated diagonal entries at the end.
+      for(Index i=1; i<length; ++i)
+        if(abs(diag(i))<considerZero)
+          permutation[p++] = i;
+        
+      Index i=1, j=k+1;
+      for( ; p < length; ++p)
+      {
+             if (i > k)             permutation[p] = j++;
+        else if (j >= length)       permutation[p] = i++;
+        else if (diag(i) < diag(j)) permutation[p] = j++;
+        else                        permutation[p] = i++;
+      }
+    }
+    
+    // If we have a total deflation, then we have to insert diag(0) at the right place
+    if(total_deflation)
+    {
+      for(Index i=1; i<length; ++i)
+      {
+        Index pi = permutation[i];
+        if(abs(diag(pi))<considerZero || diag(0)<diag(pi))
+          permutation[i-1] = permutation[i];
+        else
+        {
+          permutation[i-1] = 0;
+          break;
+        }
+      }
+    }
+    
+    // Current index of each col, and current column of each index
+    Index *realInd = m_workspaceI.data()+length;
+    Index *realCol = m_workspaceI.data()+2*length;
+    
+    for(int pos = 0; pos< length; pos++)
+    {
+      realCol[pos] = pos;
+      realInd[pos] = pos;
+    }
+    
+    for(Index i = total_deflation?0:1; i < length; i++)
+    {
+      const Index pi = permutation[length - (total_deflation ? i+1 : i)];
+      const Index J = realCol[pi];
+      
+      using std::swap;
+      // swap diagonal and first column entries:
+      swap(diag(i), diag(J));
+      if(i!=0 && J!=0) swap(col0(i), col0(J));
+
+      // change columns
+      if (m_compU) m_naiveU.col(firstCol+i).segment(firstCol, length + 1).swap(m_naiveU.col(firstCol+J).segment(firstCol, length + 1));
+      else         m_naiveU.col(firstCol+i).segment(0, 2)                .swap(m_naiveU.col(firstCol+J).segment(0, 2));
+      if (m_compV) m_naiveV.col(firstColW + i).segment(firstRowW, length).swap(m_naiveV.col(firstColW + J).segment(firstRowW, length));
+
+      //update real pos
+      const Index realI = realInd[i];
+      realCol[realI] = J;
+      realCol[pi] = i;
+      realInd[J] = realI;
+      realInd[i] = pi;
+    }
+  }
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "sorted: " << diag.transpose().format(bdcsvdfmt) << "\n";
+  std::cout << "      : " << col0.transpose() << "\n\n";
+#endif
+    
+  //condition 4.4
+  {
+    Index i = length-1;
+    while(i>0 && (abs(diag(i))<considerZero || abs(col0(i))<considerZero)) --i;
+    for(; i>1;--i)
+       if( (diag(i) - diag(i-1)) < NumTraits<RealScalar>::epsilon()*maxDiag )
+      {
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+        std::cout << "deflation 4.4 with i = " << i << " because " << (diag(i) - diag(i-1)) << " < " << NumTraits<RealScalar>::epsilon()*diag(i) << "\n";
+#endif
+        eigen_internal_assert(abs(diag(i) - diag(i-1))<epsilon_coarse && " diagonal entries are not properly sorted");
+        deflation44(firstCol, firstCol + shift, firstRowW, firstColW, i-1, i, length);
+      }
+  }
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  for(Index j=2;j<length;++j)
+    assert(diag(j-1)<=diag(j) || abs(diag(j))<considerZero);
+#endif
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+}//end deflation
+
+#ifndef __CUDACC__
+/** \svd_module
+  *
+  * \return the singular value decomposition of \c *this computed by Divide & Conquer algorithm
+  *
+  * \sa class BDCSVD
+  */
+template<typename Derived>
+BDCSVD<typename MatrixBase<Derived>::PlainObject>
+MatrixBase<Derived>::bdcSvd(unsigned int computationOptions) const
+{
+  return BDCSVD<PlainObject>(*this, computationOptions);
+}
+#endif
+
+} // end namespace Eigen
+
+#endif
diff --git a/extern/Eigen3/Eigen/src/SVD/JacobiSVD.h b/extern/Eigen3/Eigen/src/SVD/JacobiSVD.h
index 1b29774190a..43488b1e0c2 100644
--- a/extern/Eigen3/Eigen/src/SVD/JacobiSVD.h
+++ b/extern/Eigen3/Eigen/src/SVD/JacobiSVD.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2009-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2013-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -51,7 +52,6 @@ template<typename MatrixType, int QRPreconditioner, int Case>
 class qr_preconditioner_impl<MatrixType, QRPreconditioner, Case, false>
 {
 public:
-  typedef typename MatrixType::Index Index;
   void allocate(const JacobiSVD<MatrixType, QRPreconditioner>&) {}
   bool run(JacobiSVD<MatrixType, QRPreconditioner>&, const MatrixType&)
   {
@@ -65,7 +65,6 @@ template<typename MatrixType>
 class qr_preconditioner_impl<MatrixType, FullPivHouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   enum
   {
@@ -106,7 +105,6 @@ template<typename MatrixType>
 class qr_preconditioner_impl<MatrixType, FullPivHouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   enum
   {
@@ -114,9 +112,11 @@ public:
     ColsAtCompileTime = MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-    Options = MatrixType::Options
+    TrOptions = RowsAtCompileTime==1 ? (MatrixType::Options & ~(RowMajor))
+              : ColsAtCompileTime==1 ? (MatrixType::Options |   RowMajor)
+              : MatrixType::Options
   };
-  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, Options, MaxColsAtCompileTime, MaxRowsAtCompileTime>
+  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, TrOptions, MaxColsAtCompileTime, MaxRowsAtCompileTime>
           TransposeTypeWithSameStorageOrder;
 
   void allocate(const JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner>& svd)
@@ -156,8 +156,6 @@ template<typename MatrixType>
 class qr_preconditioner_impl<MatrixType, ColPivHouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
-
   void allocate(const JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd)
   {
     if (svd.rows() != m_qr.rows() || svd.cols() != m_qr.cols())
@@ -197,7 +195,6 @@ template<typename MatrixType>
 class qr_preconditioner_impl<MatrixType, ColPivHouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   enum
   {
@@ -205,10 +202,12 @@ public:
     ColsAtCompileTime = MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-    Options = MatrixType::Options
+    TrOptions = RowsAtCompileTime==1 ? (MatrixType::Options & ~(RowMajor))
+              : ColsAtCompileTime==1 ? (MatrixType::Options |   RowMajor)
+              : MatrixType::Options
   };
 
-  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, Options, MaxColsAtCompileTime, MaxRowsAtCompileTime>
+  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, TrOptions, MaxColsAtCompileTime, MaxRowsAtCompileTime>
           TransposeTypeWithSameStorageOrder;
 
   void allocate(const JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd)
@@ -256,8 +255,6 @@ template<typename MatrixType>
 class qr_preconditioner_impl<MatrixType, HouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
-
   void allocate(const JacobiSVD<MatrixType, HouseholderQRPreconditioner>& svd)
   {
     if (svd.rows() != m_qr.rows() || svd.cols() != m_qr.cols())
@@ -296,7 +293,6 @@ template<typename MatrixType>
 class qr_preconditioner_impl<MatrixType, HouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   enum
   {
@@ -358,8 +354,8 @@ template<typename MatrixType, int QRPreconditioner>
 struct svd_precondition_2x2_block_to_be_real<MatrixType, QRPreconditioner, false>
 {
   typedef JacobiSVD<MatrixType, QRPreconditioner> SVD;
-  typedef typename SVD::Index Index;
-  static void run(typename SVD::WorkMatrixType&, SVD&, Index, Index) {}
+  typedef typename MatrixType::RealScalar RealScalar;
+  static bool run(typename SVD::WorkMatrixType&, SVD&, Index, Index, RealScalar&) { return true; }
 };
 
 template<typename MatrixType, int QRPreconditioner>
@@ -368,20 +364,30 @@ struct svd_precondition_2x2_block_to_be_real<MatrixType, QRPreconditioner, true>
   typedef JacobiSVD<MatrixType, QRPreconditioner> SVD;
   typedef typename MatrixType::Scalar Scalar;
   typedef typename MatrixType::RealScalar RealScalar;
-  typedef typename SVD::Index Index;
-  static void run(typename SVD::WorkMatrixType& work_matrix, SVD& svd, Index p, Index q)
+  static bool run(typename SVD::WorkMatrixType& work_matrix, SVD& svd, Index p, Index q, RealScalar& maxDiagEntry)
   {
     using std::sqrt;
+    using std::abs;
     Scalar z;
     JacobiRotation<Scalar> rot;
     RealScalar n = sqrt(numext::abs2(work_matrix.coeff(p,p)) + numext::abs2(work_matrix.coeff(q,p)));
-    
+
+    const RealScalar considerAsZero = (std::numeric_limits<RealScalar>::min)();
+    const RealScalar precision = NumTraits<Scalar>::epsilon();
+
     if(n==0)
     {
-      z = abs(work_matrix.coeff(p,q)) / work_matrix.coeff(p,q);
-      work_matrix.row(p) *= z;
-      if(svd.computeU()) svd.m_matrixU.col(p) *= conj(z);
-      if(work_matrix.coeff(q,q)!=Scalar(0))
+      // make sure first column is zero
+      work_matrix.coeffRef(p,p) = work_matrix.coeffRef(q,p) = Scalar(0);
+
+      if(abs(numext::imag(work_matrix.coeff(p,q)))>considerAsZero)
+      {
+        // work_matrix.coeff(p,q) can be zero if work_matrix.coeff(q,p) is not zero but small enough to underflow when computing n
+        z = abs(work_matrix.coeff(p,q)) / work_matrix.coeff(p,q);
+        work_matrix.row(p) *= z;
+        if(svd.computeU()) svd.m_matrixU.col(p) *= conj(z);
+      }
+      if(abs(numext::imag(work_matrix.coeff(q,q)))>considerAsZero)
       {
         z = abs(work_matrix.coeff(q,q)) / work_matrix.coeff(q,q);
         work_matrix.row(q) *= z;
@@ -395,52 +401,33 @@ struct svd_precondition_2x2_block_to_be_real<MatrixType, QRPreconditioner, true>
       rot.s() = work_matrix.coeff(q,p) / n;
       work_matrix.applyOnTheLeft(p,q,rot);
       if(svd.computeU()) svd.m_matrixU.applyOnTheRight(p,q,rot.adjoint());
-      if(work_matrix.coeff(p,q) != Scalar(0))
+      if(abs(numext::imag(work_matrix.coeff(p,q)))>considerAsZero)
       {
-        Scalar z = abs(work_matrix.coeff(p,q)) / work_matrix.coeff(p,q);
+        z = abs(work_matrix.coeff(p,q)) / work_matrix.coeff(p,q);
         work_matrix.col(q) *= z;
         if(svd.computeV()) svd.m_matrixV.col(q) *= z;
       }
-      if(work_matrix.coeff(q,q) != Scalar(0))
+      if(abs(numext::imag(work_matrix.coeff(q,q)))>considerAsZero)
       {
         z = abs(work_matrix.coeff(q,q)) / work_matrix.coeff(q,q);
         work_matrix.row(q) *= z;
         if(svd.computeU()) svd.m_matrixU.col(q) *= conj(z);
       }
     }
+
+    // update largest diagonal entry
+    maxDiagEntry = numext::maxi<RealScalar>(maxDiagEntry,numext::maxi<RealScalar>(abs(work_matrix.coeff(p,p)), abs(work_matrix.coeff(q,q))));
+    // and check whether the 2x2 block is already diagonal
+    RealScalar threshold = numext::maxi<RealScalar>(considerAsZero, precision * maxDiagEntry);
+    return abs(work_matrix.coeff(p,q))>threshold || abs(work_matrix.coeff(q,p)) > threshold;
   }
 };
 
-template<typename MatrixType, typename RealScalar, typename Index>
-void real_2x2_jacobi_svd(const MatrixType& matrix, Index p, Index q,
-                            JacobiRotation<RealScalar> *j_left,
-                            JacobiRotation<RealScalar> *j_right)
+template<typename _MatrixType, int QRPreconditioner> 
+struct traits<JacobiSVD<_MatrixType,QRPreconditioner> >
 {
-  using std::sqrt;
-  using std::abs;
-  Matrix<RealScalar,2,2> m;
-  m << numext::real(matrix.coeff(p,p)), numext::real(matrix.coeff(p,q)),
-       numext::real(matrix.coeff(q,p)), numext::real(matrix.coeff(q,q));
-  JacobiRotation<RealScalar> rot1;
-  RealScalar t = m.coeff(0,0) + m.coeff(1,1);
-  RealScalar d = m.coeff(1,0) - m.coeff(0,1);
-  if(t == RealScalar(0))
-  {
-    rot1.c() = RealScalar(0);
-    rot1.s() = d > RealScalar(0) ? RealScalar(1) : RealScalar(-1);
-  }
-  else
-  {
-    RealScalar t2d2 = numext::hypot(t,d);
-    rot1.c() = abs(t)/t2d2;
-    rot1.s() = d/t2d2;
-    if(t<RealScalar(0))
-      rot1.s() = -rot1.s();
-  }
-  m.applyOnTheLeft(0,1,rot1);
-  j_right->makeJacobi(m,0,1);
-  *j_left  = rot1 * j_right->transpose();
-}
+  typedef _MatrixType MatrixType;
+};
 
 } // end namespace internal
 
@@ -451,8 +438,8 @@ void real_2x2_jacobi_svd(const MatrixType& matrix, Index p, Index q,
   *
   * \brief Two-sided Jacobi SVD decomposition of a rectangular matrix
   *
-  * \param MatrixType the type of the matrix of which we are computing the SVD decomposition
-  * \param QRPreconditioner this optional parameter allows to specify the type of QR decomposition that will be used internally
+  * \tparam _MatrixType the type of the matrix of which we are computing the SVD decomposition
+  * \tparam QRPreconditioner this optional parameter allows to specify the type of QR decomposition that will be used internally
   *                        for the R-SVD step for non-square matrices. See discussion of possible values below.
   *
   * SVD decomposition consists in decomposing any n-by-p matrix \a A as a product
@@ -498,13 +485,14 @@ void real_2x2_jacobi_svd(const MatrixType& matrix, Index p, Index q,
   * \sa MatrixBase::jacobiSvd()
   */
 template<typename _MatrixType, int QRPreconditioner> class JacobiSVD
+ : public SVDBase<JacobiSVD<_MatrixType,QRPreconditioner> >
 {
+    typedef SVDBase<JacobiSVD> Base;
   public:
 
     typedef _MatrixType MatrixType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
@@ -515,13 +503,10 @@ template<typename _MatrixType, int QRPreconditioner> class JacobiSVD
       MatrixOptions = MatrixType::Options
     };
 
-    typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime,
-                   MatrixOptions, MaxRowsAtCompileTime, MaxRowsAtCompileTime>
-            MatrixUType;
-    typedef Matrix<Scalar, ColsAtCompileTime, ColsAtCompileTime,
-                   MatrixOptions, MaxColsAtCompileTime, MaxColsAtCompileTime>
-            MatrixVType;
-    typedef typename internal::plain_diag_type<MatrixType, RealScalar>::type SingularValuesType;
+    typedef typename Base::MatrixUType MatrixUType;
+    typedef typename Base::MatrixVType MatrixVType;
+    typedef typename Base::SingularValuesType SingularValuesType;
+    
     typedef typename internal::plain_row_type<MatrixType>::type RowType;
     typedef typename internal::plain_col_type<MatrixType>::type ColType;
     typedef Matrix<Scalar, DiagSizeAtCompileTime, DiagSizeAtCompileTime,
@@ -534,11 +519,6 @@ template<typename _MatrixType, int QRPreconditioner> class JacobiSVD
       * perform decompositions via JacobiSVD::compute(const MatrixType&).
       */
     JacobiSVD()
-      : m_isInitialized(false),
-        m_isAllocated(false),
-        m_usePrescribedThreshold(false),
-        m_computationOptions(0),
-        m_rows(-1), m_cols(-1), m_diagSize(0)
     {}
 
 
@@ -549,11 +529,6 @@ template<typename _MatrixType, int QRPreconditioner> class JacobiSVD
       * \sa JacobiSVD()
       */
     JacobiSVD(Index rows, Index cols, unsigned int computationOptions = 0)
-      : m_isInitialized(false),
-        m_isAllocated(false),
-        m_usePrescribedThreshold(false),
-        m_computationOptions(0),
-        m_rows(-1), m_cols(-1)
     {
       allocate(rows, cols, computationOptions);
     }
@@ -568,12 +543,7 @@ template<typename _MatrixType, int QRPreconditioner> class JacobiSVD
      * Thin unitaries are only available if your matrix type has a Dynamic number of columns (for example MatrixXf). They also are not
      * available with the (non-default) FullPivHouseholderQR preconditioner.
      */
-    JacobiSVD(const MatrixType& matrix, unsigned int computationOptions = 0)
-      : m_isInitialized(false),
-        m_isAllocated(false),
-        m_usePrescribedThreshold(false),
-        m_computationOptions(0),
-        m_rows(-1), m_cols(-1)
+    explicit JacobiSVD(const MatrixType& matrix, unsigned int computationOptions = 0)
     {
       compute(matrix, computationOptions);
     }
@@ -601,164 +571,33 @@ template<typename _MatrixType, int QRPreconditioner> class JacobiSVD
       return compute(matrix, m_computationOptions);
     }
 
-    /** \returns the \a U matrix.
-     *
-     * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p,
-     * the U matrix is n-by-n if you asked for #ComputeFullU, and is n-by-m if you asked for #ComputeThinU.
-     *
-     * The \a m first columns of \a U are the left singular vectors of the matrix being decomposed.
-     *
-     * This method asserts that you asked for \a U to be computed.
-     */
-    const MatrixUType& matrixU() const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      eigen_assert(computeU() && "This JacobiSVD decomposition didn't compute U. Did you ask for it?");
-      return m_matrixU;
-    }
-
-    /** \returns the \a V matrix.
-     *
-     * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p,
-     * the V matrix is p-by-p if you asked for #ComputeFullV, and is p-by-m if you asked for ComputeThinV.
-     *
-     * The \a m first columns of \a V are the right singular vectors of the matrix being decomposed.
-     *
-     * This method asserts that you asked for \a V to be computed.
-     */
-    const MatrixVType& matrixV() const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      eigen_assert(computeV() && "This JacobiSVD decomposition didn't compute V. Did you ask for it?");
-      return m_matrixV;
-    }
-
-    /** \returns the vector of singular values.
-     *
-     * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p, the
-     * returned vector has size \a m.  Singular values are always sorted in decreasing order.
-     */
-    const SingularValuesType& singularValues() const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      return m_singularValues;
-    }
-
-    /** \returns true if \a U (full or thin) is asked for in this SVD decomposition */
-    inline bool computeU() const { return m_computeFullU || m_computeThinU; }
-    /** \returns true if \a V (full or thin) is asked for in this SVD decomposition */
-    inline bool computeV() const { return m_computeFullV || m_computeThinV; }
-
-    /** \returns a (least squares) solution of \f$ A x = b \f$ using the current SVD decomposition of A.
-      *
-      * \param b the right-hand-side of the equation to solve.
-      *
-      * \note Solving requires both U and V to be computed. Thin U and V are enough, there is no need for full U or V.
-      *
-      * \note SVD solving is implicitly least-squares. Thus, this method serves both purposes of exact solving and least-squares solving.
-      * In other words, the returned solution is guaranteed to minimize the Euclidean norm \f$ \Vert A x - b \Vert \f$.
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<JacobiSVD, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      eigen_assert(computeU() && computeV() && "JacobiSVD::solve() requires both unitaries U and V to be computed (thin unitaries suffice).");
-      return internal::solve_retval<JacobiSVD, Rhs>(*this, b.derived());
-    }
-
-    /** \returns the number of singular values that are not exactly 0 */
-    Index nonzeroSingularValues() const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      return m_nonzeroSingularValues;
-    }
-    
-    /** \returns the rank of the matrix of which \c *this is the SVD.
-      *
-      * \note This method has to determine which singular values should be considered nonzero.
-      *       For that, it uses the threshold value that you can control by calling
-      *       setThreshold(const RealScalar&).
-      */
-    inline Index rank() const
-    {
-      using std::abs;
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      if(m_singularValues.size()==0) return 0;
-      RealScalar premultiplied_threshold = m_singularValues.coeff(0) * threshold();
-      Index i = m_nonzeroSingularValues-1;
-      while(i>=0 && m_singularValues.coeff(i) < premultiplied_threshold) --i;
-      return i+1;
-    }
-    
-    /** Allows to prescribe a threshold to be used by certain methods, such as rank() and solve(),
-      * which need to determine when singular values are to be considered nonzero.
-      * This is not used for the SVD decomposition itself.
-      *
-      * When it needs to get the threshold value, Eigen calls threshold().
-      * The default is \c NumTraits<Scalar>::epsilon()
-      *
-      * \param threshold The new value to use as the threshold.
-      *
-      * A singular value will be considered nonzero if its value is strictly greater than
-      *  \f$ \vert singular value \vert \leqslant threshold \times \vert max singular value \vert \f$.
-      *
-      * If you want to come back to the default behavior, call setThreshold(Default_t)
-      */
-    JacobiSVD& setThreshold(const RealScalar& threshold)
-    {
-      m_usePrescribedThreshold = true;
-      m_prescribedThreshold = threshold;
-      return *this;
-    }
-
-    /** Allows to come back to the default behavior, letting Eigen use its default formula for
-      * determining the threshold.
-      *
-      * You should pass the special object Eigen::Default as parameter here.
-      * \code svd.setThreshold(Eigen::Default); \endcode
-      *
-      * See the documentation of setThreshold(const RealScalar&).
-      */
-    JacobiSVD& setThreshold(Default_t)
-    {
-      m_usePrescribedThreshold = false;
-      return *this;
-    }
-
-    /** Returns the threshold that will be used by certain methods such as rank().
-      *
-      * See the documentation of setThreshold(const RealScalar&).
-      */
-    RealScalar threshold() const
-    {
-      eigen_assert(m_isInitialized || m_usePrescribedThreshold);
-      return m_usePrescribedThreshold ? m_prescribedThreshold
-                                      : (std::max<Index>)(1,m_diagSize)*NumTraits<Scalar>::epsilon();
-    }
-
-    inline Index rows() const { return m_rows; }
-    inline Index cols() const { return m_cols; }
+    using Base::computeU;
+    using Base::computeV;
+    using Base::rows;
+    using Base::cols;
+    using Base::rank;
 
   private:
     void allocate(Index rows, Index cols, unsigned int computationOptions);
-    
-    static void check_template_parameters()
-    {
-      EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
-    }
 
   protected:
-    MatrixUType m_matrixU;
-    MatrixVType m_matrixV;
-    SingularValuesType m_singularValues;
+    using Base::m_matrixU;
+    using Base::m_matrixV;
+    using Base::m_singularValues;
+    using Base::m_isInitialized;
+    using Base::m_isAllocated;
+    using Base::m_usePrescribedThreshold;
+    using Base::m_computeFullU;
+    using Base::m_computeThinU;
+    using Base::m_computeFullV;
+    using Base::m_computeThinV;
+    using Base::m_computationOptions;
+    using Base::m_nonzeroSingularValues;
+    using Base::m_rows;
+    using Base::m_cols;
+    using Base::m_diagSize;
+    using Base::m_prescribedThreshold;
     WorkMatrixType m_workMatrix;
-    bool m_isInitialized, m_isAllocated, m_usePrescribedThreshold;
-    bool m_computeFullU, m_computeThinU;
-    bool m_computeFullV, m_computeThinV;
-    unsigned int m_computationOptions;
-    Index m_nonzeroSingularValues, m_rows, m_cols, m_diagSize;
-    RealScalar m_prescribedThreshold;
 
     template<typename __MatrixType, int _QRPreconditioner, bool _IsComplex>
     friend struct internal::svd_precondition_2x2_block_to_be_real;
@@ -816,15 +655,13 @@ void JacobiSVD<MatrixType, QRPreconditioner>::allocate(Index rows, Index cols, u
   
   if(m_cols>m_rows)   m_qr_precond_morecols.allocate(*this);
   if(m_rows>m_cols)   m_qr_precond_morerows.allocate(*this);
-  if(m_cols!=m_cols)  m_scaledMatrix.resize(rows,cols);
+  if(m_rows!=m_cols)  m_scaledMatrix.resize(rows,cols);
 }
 
 template<typename MatrixType, int QRPreconditioner>
 JacobiSVD<MatrixType, QRPreconditioner>&
 JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsigned int computationOptions)
 {
-  check_template_parameters();
-  
   using std::abs;
   allocate(matrix.rows(), matrix.cols(), computationOptions);
 
@@ -832,8 +669,8 @@ JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsig
   // only worsening the precision of U and V as we accumulate more rotations
   const RealScalar precision = RealScalar(2) * NumTraits<Scalar>::epsilon();
 
-  // limit for very small denormal numbers to be considered zero in order to avoid infinite loops (see bug 286)
-  const RealScalar considerAsZero = RealScalar(2) * std::numeric_limits<RealScalar>::denorm_min();
+  // limit for denormal numbers to be considered zero in order to avoid infinite loops (see bug 286)
+  const RealScalar considerAsZero = (std::numeric_limits<RealScalar>::min)();
 
   // Scaling factor to reduce over/under-flows
   RealScalar scale = matrix.cwiseAbs().maxCoeff();
@@ -857,6 +694,7 @@ JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsig
   }
 
   /*** step 2. The main Jacobi SVD iteration. ***/
+  RealScalar maxDiagEntry = m_workMatrix.cwiseAbs().diagonal().maxCoeff();
 
   bool finished = false;
   while(!finished)
@@ -872,25 +710,27 @@ JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsig
         // if this 2x2 sub-matrix is not diagonal already...
         // notice that this comparison will evaluate to false if any NaN is involved, ensuring that NaN's don't
         // keep us iterating forever. Similarly, small denormal numbers are considered zero.
-        using std::max;
-        RealScalar threshold = (max)(considerAsZero, precision * (max)(abs(m_workMatrix.coeff(p,p)),
-                                                                       abs(m_workMatrix.coeff(q,q))));
-        // We compare both values to threshold instead of calling max to be robust to NaN (See bug 791)
+        RealScalar threshold = numext::maxi<RealScalar>(considerAsZero, precision * maxDiagEntry);
         if(abs(m_workMatrix.coeff(p,q))>threshold || abs(m_workMatrix.coeff(q,p)) > threshold)
         {
           finished = false;
-
           // perform SVD decomposition of 2x2 sub-matrix corresponding to indices p,q to make it diagonal
-          internal::svd_precondition_2x2_block_to_be_real<MatrixType, QRPreconditioner>::run(m_workMatrix, *this, p, q);
-          JacobiRotation<RealScalar> j_left, j_right;
-          internal::real_2x2_jacobi_svd(m_workMatrix, p, q, &j_left, &j_right);
-
-          // accumulate resulting Jacobi rotations
-          m_workMatrix.applyOnTheLeft(p,q,j_left);
-          if(computeU()) m_matrixU.applyOnTheRight(p,q,j_left.transpose());
-
-          m_workMatrix.applyOnTheRight(p,q,j_right);
-          if(computeV()) m_matrixV.applyOnTheRight(p,q,j_right);
+          // the complex to real operation returns true if the updated 2x2 block is not already diagonal
+          if(internal::svd_precondition_2x2_block_to_be_real<MatrixType, QRPreconditioner>::run(m_workMatrix, *this, p, q, maxDiagEntry))
+          {
+            JacobiRotation<RealScalar> j_left, j_right;
+            internal::real_2x2_jacobi_svd(m_workMatrix, p, q, &j_left, &j_right);
+
+            // accumulate resulting Jacobi rotations
+            m_workMatrix.applyOnTheLeft(p,q,j_left);
+            if(computeU()) m_matrixU.applyOnTheRight(p,q,j_left.transpose());
+
+            m_workMatrix.applyOnTheRight(p,q,j_right);
+            if(computeV()) m_matrixV.applyOnTheRight(p,q,j_right);
+
+            // keep track of the largest diagonal coefficient
+            maxDiagEntry = numext::maxi<RealScalar>(maxDiagEntry,numext::maxi<RealScalar>(abs(m_workMatrix.coeff(p,p)), abs(m_workMatrix.coeff(q,q))));
+          }
         }
       }
     }
@@ -900,10 +740,25 @@ JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsig
 
   for(Index i = 0; i < m_diagSize; ++i)
   {
-    RealScalar a = abs(m_workMatrix.coeff(i,i));
-    m_singularValues.coeffRef(i) = a;
-    if(computeU() && (a!=RealScalar(0))) m_matrixU.col(i) *= m_workMatrix.coeff(i,i)/a;
+    // For a complex matrix, some diagonal coefficients might note have been
+    // treated by svd_precondition_2x2_block_to_be_real, and the imaginary part
+    // of some diagonal entry might not be null.
+    if(NumTraits<Scalar>::IsComplex && abs(numext::imag(m_workMatrix.coeff(i,i)))>considerAsZero)
+    {
+      RealScalar a = abs(m_workMatrix.coeff(i,i));
+      m_singularValues.coeffRef(i) = abs(a);
+      if(computeU()) m_matrixU.col(i) *= m_workMatrix.coeff(i,i)/a;
+    }
+    else
+    {
+      // m_workMatrix.coeff(i,i) is already real, no difficulty:
+      RealScalar a = numext::real(m_workMatrix.coeff(i,i));
+      m_singularValues.coeffRef(i) = abs(a);
+      if(computeU() && (a<RealScalar(0))) m_matrixU.col(i) = -m_matrixU.col(i);
+    }
   }
+  
+  m_singularValues *= scale;
 
   /*** step 4. Sort singular values in descending order and compute the number of nonzero singular values ***/
 
@@ -925,38 +780,11 @@ JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsig
       if(computeV()) m_matrixV.col(pos).swap(m_matrixV.col(i));
     }
   }
-  
-  m_singularValues *= scale;
 
   m_isInitialized = true;
   return *this;
 }
 
-namespace internal {
-template<typename _MatrixType, int QRPreconditioner, typename Rhs>
-struct solve_retval<JacobiSVD<_MatrixType, QRPreconditioner>, Rhs>
-  : solve_retval_base<JacobiSVD<_MatrixType, QRPreconditioner>, Rhs>
-{
-  typedef JacobiSVD<_MatrixType, QRPreconditioner> JacobiSVDType;
-  EIGEN_MAKE_SOLVE_HELPERS(JacobiSVDType,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    eigen_assert(rhs().rows() == dec().rows());
-
-    // A = U S V^*
-    // So A^{-1} = V S^{-1} U^*
-
-    Matrix<Scalar, Dynamic, Rhs::ColsAtCompileTime, 0, _MatrixType::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime> tmp;
-    Index rank = dec().rank();
-    
-    tmp.noalias() = dec().matrixU().leftCols(rank).adjoint() * rhs();
-    tmp = dec().singularValues().head(rank).asDiagonal().inverse() * tmp;
-    dst = dec().matrixV().leftCols(rank) * tmp;
-  }
-};
-} // end namespace internal
-
 /** \svd_module
   *
   * \return the singular value decomposition of \c *this computed by two-sided
diff --git a/extern/Eigen3/Eigen/src/SVD/JacobiSVD_LAPACKE.h b/extern/Eigen3/Eigen/src/SVD/JacobiSVD_LAPACKE.h
new file mode 100644
index 00000000000..ff0516f6116
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/SVD/JacobiSVD_LAPACKE.h
@@ -0,0 +1,91 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *    Singular Value Decomposition - SVD.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_JACOBISVD_LAPACKE_H
+#define EIGEN_JACOBISVD_LAPACKE_H
+
+namespace Eigen { 
+
+/** \internal Specialization for the data types supported by LAPACKe */
+
+#define EIGEN_LAPACKE_SVD(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_PREFIX, EIGCOLROW, LAPACKE_COLROW) \
+template<> inline \
+JacobiSVD<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>, ColPivHouseholderQRPreconditioner>& \
+JacobiSVD<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>, ColPivHouseholderQRPreconditioner>::compute(const Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>& matrix, unsigned int computationOptions) \
+{ \
+  typedef Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic> MatrixType; \
+  /*typedef MatrixType::Scalar Scalar;*/ \
+  /*typedef MatrixType::RealScalar RealScalar;*/ \
+  allocate(matrix.rows(), matrix.cols(), computationOptions); \
+\
+  /*const RealScalar precision = RealScalar(2) * NumTraits<Scalar>::epsilon();*/ \
+  m_nonzeroSingularValues = m_diagSize; \
+\
+  lapack_int lda = internal::convert_index<lapack_int>(matrix.outerStride()), ldu, ldvt; \
+  lapack_int matrix_order = LAPACKE_COLROW; \
+  char jobu, jobvt; \
+  LAPACKE_TYPE *u, *vt, dummy; \
+  jobu  = (m_computeFullU) ? 'A' : (m_computeThinU) ? 'S' : 'N'; \
+  jobvt = (m_computeFullV) ? 'A' : (m_computeThinV) ? 'S' : 'N'; \
+  if (computeU()) { \
+    ldu  = internal::convert_index<lapack_int>(m_matrixU.outerStride()); \
+    u    = (LAPACKE_TYPE*)m_matrixU.data(); \
+  } else { ldu=1; u=&dummy; }\
+  MatrixType localV; \
+  lapack_int vt_rows = (m_computeFullV) ? internal::convert_index<lapack_int>(m_cols) : (m_computeThinV) ? internal::convert_index<lapack_int>(m_diagSize) : 1; \
+  if (computeV()) { \
+    localV.resize(vt_rows, m_cols); \
+    ldvt  = internal::convert_index<lapack_int>(localV.outerStride()); \
+    vt   = (LAPACKE_TYPE*)localV.data(); \
+  } else { ldvt=1; vt=&dummy; }\
+  Matrix<LAPACKE_RTYPE, Dynamic, Dynamic> superb; superb.resize(m_diagSize, 1); \
+  MatrixType m_temp; m_temp = matrix; \
+  LAPACKE_##LAPACKE_PREFIX##gesvd( matrix_order, jobu, jobvt, internal::convert_index<lapack_int>(m_rows), internal::convert_index<lapack_int>(m_cols), (LAPACKE_TYPE*)m_temp.data(), lda, (LAPACKE_RTYPE*)m_singularValues.data(), u, ldu, vt, ldvt, superb.data()); \
+  if (computeV()) m_matrixV = localV.adjoint(); \
+ /* for(int i=0;i<m_diagSize;i++) if (m_singularValues.coeffRef(i) < precision) { m_nonzeroSingularValues--; m_singularValues.coeffRef(i)=RealScalar(0);}*/ \
+  m_isInitialized = true; \
+  return *this; \
+}
+
+EIGEN_LAPACKE_SVD(double,   double,                double, d, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SVD(float,    float,                 float , s, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SVD(dcomplex, lapack_complex_double, double, z, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SVD(scomplex, lapack_complex_float,  float , c, ColMajor, LAPACK_COL_MAJOR)
+
+EIGEN_LAPACKE_SVD(double,   double,                double, d, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SVD(float,    float,                 float , s, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SVD(dcomplex, lapack_complex_double, double, z, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SVD(scomplex, lapack_complex_float,  float , c, RowMajor, LAPACK_ROW_MAJOR)
+
+} // end namespace Eigen
+
+#endif // EIGEN_JACOBISVD_LAPACKE_H
diff --git a/extern/Eigen3/Eigen/src/SVD/SVDBase.h b/extern/Eigen3/Eigen/src/SVD/SVDBase.h
new file mode 100644
index 00000000000..3d1ef373ea7
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/SVD/SVDBase.h
@@ -0,0 +1,315 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// Copyright (C) 2013 Gauthier Brun <brun.gauthier@gmail.com>
+// Copyright (C) 2013 Nicolas Carre <nicolas.carre@ensimag.fr>
+// Copyright (C) 2013 Jean Ceccato <jean.ceccato@ensimag.fr>
+// Copyright (C) 2013 Pierre Zoppitelli <pierre.zoppitelli@ensimag.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SVDBASE_H
+#define EIGEN_SVDBASE_H
+
+namespace Eigen {
+/** \ingroup SVD_Module
+ *
+ *
+ * \class SVDBase
+ *
+ * \brief Base class of SVD algorithms
+ *
+ * \tparam Derived the type of the actual SVD decomposition
+ *
+ * SVD decomposition consists in decomposing any n-by-p matrix \a A as a product
+ *   \f[ A = U S V^* \f]
+ * where \a U is a n-by-n unitary, \a V is a p-by-p unitary, and \a S is a n-by-p real positive matrix which is zero outside of its main diagonal;
+ * the diagonal entries of S are known as the \em singular \em values of \a A and the columns of \a U and \a V are known as the left
+ * and right \em singular \em vectors of \a A respectively.
+ *
+ * Singular values are always sorted in decreasing order.
+ *
+ * 
+ * You can ask for only \em thin \a U or \a V to be computed, meaning the following. In case of a rectangular n-by-p matrix, letting \a m be the
+ * smaller value among \a n and \a p, there are only \a m singular vectors; the remaining columns of \a U and \a V do not correspond to actual
+ * singular vectors. Asking for \em thin \a U or \a V means asking for only their \a m first columns to be formed. So \a U is then a n-by-m matrix,
+ * and \a V is then a p-by-m matrix. Notice that thin \a U and \a V are all you need for (least squares) solving.
+ *  
+ * If the input matrix has inf or nan coefficients, the result of the computation is undefined, but the computation is guaranteed to
+ * terminate in finite (and reasonable) time.
+ * \sa class BDCSVD, class JacobiSVD
+ */
+template<typename Derived>
+class SVDBase
+{
+
+public:
+  typedef typename internal::traits<Derived>::MatrixType MatrixType;
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
+  typedef typename MatrixType::StorageIndex StorageIndex;
+  typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+  enum {
+    RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+    ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+    DiagSizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime,ColsAtCompileTime),
+    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
+    MaxDiagSizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(MaxRowsAtCompileTime,MaxColsAtCompileTime),
+    MatrixOptions = MatrixType::Options
+  };
+
+  typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime, MatrixOptions, MaxRowsAtCompileTime, MaxRowsAtCompileTime> MatrixUType;
+  typedef Matrix<Scalar, ColsAtCompileTime, ColsAtCompileTime, MatrixOptions, MaxColsAtCompileTime, MaxColsAtCompileTime> MatrixVType;
+  typedef typename internal::plain_diag_type<MatrixType, RealScalar>::type SingularValuesType;
+  
+  Derived& derived() { return *static_cast<Derived*>(this); }
+  const Derived& derived() const { return *static_cast<const Derived*>(this); }
+
+  /** \returns the \a U matrix.
+   *
+   * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p,
+   * the U matrix is n-by-n if you asked for \link Eigen::ComputeFullU ComputeFullU \endlink, and is n-by-m if you asked for \link Eigen::ComputeThinU ComputeThinU \endlink.
+   *
+   * The \a m first columns of \a U are the left singular vectors of the matrix being decomposed.
+   *
+   * This method asserts that you asked for \a U to be computed.
+   */
+  const MatrixUType& matrixU() const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    eigen_assert(computeU() && "This SVD decomposition didn't compute U. Did you ask for it?");
+    return m_matrixU;
+  }
+
+  /** \returns the \a V matrix.
+   *
+   * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p,
+   * the V matrix is p-by-p if you asked for \link Eigen::ComputeFullV ComputeFullV \endlink, and is p-by-m if you asked for \link Eigen::ComputeThinV ComputeThinV \endlink.
+   *
+   * The \a m first columns of \a V are the right singular vectors of the matrix being decomposed.
+   *
+   * This method asserts that you asked for \a V to be computed.
+   */
+  const MatrixVType& matrixV() const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    eigen_assert(computeV() && "This SVD decomposition didn't compute V. Did you ask for it?");
+    return m_matrixV;
+  }
+
+  /** \returns the vector of singular values.
+   *
+   * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p, the
+   * returned vector has size \a m.  Singular values are always sorted in decreasing order.
+   */
+  const SingularValuesType& singularValues() const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    return m_singularValues;
+  }
+
+  /** \returns the number of singular values that are not exactly 0 */
+  Index nonzeroSingularValues() const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    return m_nonzeroSingularValues;
+  }
+  
+  /** \returns the rank of the matrix of which \c *this is the SVD.
+    *
+    * \note This method has to determine which singular values should be considered nonzero.
+    *       For that, it uses the threshold value that you can control by calling
+    *       setThreshold(const RealScalar&).
+    */
+  inline Index rank() const
+  {
+    using std::abs;
+    eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
+    if(m_singularValues.size()==0) return 0;
+    RealScalar premultiplied_threshold = numext::maxi<RealScalar>(m_singularValues.coeff(0) * threshold(), (std::numeric_limits<RealScalar>::min)());
+    Index i = m_nonzeroSingularValues-1;
+    while(i>=0 && m_singularValues.coeff(i) < premultiplied_threshold) --i;
+    return i+1;
+  }
+  
+  /** Allows to prescribe a threshold to be used by certain methods, such as rank() and solve(),
+    * which need to determine when singular values are to be considered nonzero.
+    * This is not used for the SVD decomposition itself.
+    *
+    * When it needs to get the threshold value, Eigen calls threshold().
+    * The default is \c NumTraits<Scalar>::epsilon()
+    *
+    * \param threshold The new value to use as the threshold.
+    *
+    * A singular value will be considered nonzero if its value is strictly greater than
+    *  \f$ \vert singular value \vert \leqslant threshold \times \vert max singular value \vert \f$.
+    *
+    * If you want to come back to the default behavior, call setThreshold(Default_t)
+    */
+  Derived& setThreshold(const RealScalar& threshold)
+  {
+    m_usePrescribedThreshold = true;
+    m_prescribedThreshold = threshold;
+    return derived();
+  }
+
+  /** Allows to come back to the default behavior, letting Eigen use its default formula for
+    * determining the threshold.
+    *
+    * You should pass the special object Eigen::Default as parameter here.
+    * \code svd.setThreshold(Eigen::Default); \endcode
+    *
+    * See the documentation of setThreshold(const RealScalar&).
+    */
+  Derived& setThreshold(Default_t)
+  {
+    m_usePrescribedThreshold = false;
+    return derived();
+  }
+
+  /** Returns the threshold that will be used by certain methods such as rank().
+    *
+    * See the documentation of setThreshold(const RealScalar&).
+    */
+  RealScalar threshold() const
+  {
+    eigen_assert(m_isInitialized || m_usePrescribedThreshold);
+    // this temporary is needed to workaround a MSVC issue
+    Index diagSize = (std::max<Index>)(1,m_diagSize);
+    return m_usePrescribedThreshold ? m_prescribedThreshold
+                                    : diagSize*NumTraits<Scalar>::epsilon();
+  }
+
+  /** \returns true if \a U (full or thin) is asked for in this SVD decomposition */
+  inline bool computeU() const { return m_computeFullU || m_computeThinU; }
+  /** \returns true if \a V (full or thin) is asked for in this SVD decomposition */
+  inline bool computeV() const { return m_computeFullV || m_computeThinV; }
+
+  inline Index rows() const { return m_rows; }
+  inline Index cols() const { return m_cols; }
+  
+  /** \returns a (least squares) solution of \f$ A x = b \f$ using the current SVD decomposition of A.
+    *
+    * \param b the right-hand-side of the equation to solve.
+    *
+    * \note Solving requires both U and V to be computed. Thin U and V are enough, there is no need for full U or V.
+    *
+    * \note SVD solving is implicitly least-squares. Thus, this method serves both purposes of exact solving and least-squares solving.
+    * In other words, the returned solution is guaranteed to minimize the Euclidean norm \f$ \Vert A x - b \Vert \f$.
+    */
+  template<typename Rhs>
+  inline const Solve<Derived, Rhs>
+  solve(const MatrixBase<Rhs>& b) const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    eigen_assert(computeU() && computeV() && "SVD::solve() requires both unitaries U and V to be computed (thin unitaries suffice).");
+    return Solve<Derived, Rhs>(derived(), b.derived());
+  }
+  
+  #ifndef EIGEN_PARSED_BY_DOXYGEN
+  template<typename RhsType, typename DstType>
+  EIGEN_DEVICE_FUNC
+  void _solve_impl(const RhsType &rhs, DstType &dst) const;
+  #endif
+
+protected:
+  
+  static void check_template_parameters()
+  {
+    EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
+  }
+  
+  // return true if already allocated
+  bool allocate(Index rows, Index cols, unsigned int computationOptions) ;
+
+  MatrixUType m_matrixU;
+  MatrixVType m_matrixV;
+  SingularValuesType m_singularValues;
+  bool m_isInitialized, m_isAllocated, m_usePrescribedThreshold;
+  bool m_computeFullU, m_computeThinU;
+  bool m_computeFullV, m_computeThinV;
+  unsigned int m_computationOptions;
+  Index m_nonzeroSingularValues, m_rows, m_cols, m_diagSize;
+  RealScalar m_prescribedThreshold;
+
+  /** \brief Default Constructor.
+   *
+   * Default constructor of SVDBase
+   */
+  SVDBase()
+    : m_isInitialized(false),
+      m_isAllocated(false),
+      m_usePrescribedThreshold(false),
+      m_computationOptions(0),
+      m_rows(-1), m_cols(-1), m_diagSize(0)
+  {
+    check_template_parameters();
+  }
+
+
+};
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename Derived>
+template<typename RhsType, typename DstType>
+void SVDBase<Derived>::_solve_impl(const RhsType &rhs, DstType &dst) const
+{
+  eigen_assert(rhs.rows() == rows());
+
+  // A = U S V^*
+  // So A^{-1} = V S^{-1} U^*
+
+  Matrix<Scalar, Dynamic, RhsType::ColsAtCompileTime, 0, MatrixType::MaxRowsAtCompileTime, RhsType::MaxColsAtCompileTime> tmp;
+  Index l_rank = rank();
+  tmp.noalias() =  m_matrixU.leftCols(l_rank).adjoint() * rhs;
+  tmp = m_singularValues.head(l_rank).asDiagonal().inverse() * tmp;
+  dst = m_matrixV.leftCols(l_rank) * tmp;
+}
+#endif
+
+template<typename MatrixType>
+bool SVDBase<MatrixType>::allocate(Index rows, Index cols, unsigned int computationOptions)
+{
+  eigen_assert(rows >= 0 && cols >= 0);
+
+  if (m_isAllocated &&
+      rows == m_rows &&
+      cols == m_cols &&
+      computationOptions == m_computationOptions)
+  {
+    return true;
+  }
+
+  m_rows = rows;
+  m_cols = cols;
+  m_isInitialized = false;
+  m_isAllocated = true;
+  m_computationOptions = computationOptions;
+  m_computeFullU = (computationOptions & ComputeFullU) != 0;
+  m_computeThinU = (computationOptions & ComputeThinU) != 0;
+  m_computeFullV = (computationOptions & ComputeFullV) != 0;
+  m_computeThinV = (computationOptions & ComputeThinV) != 0;
+  eigen_assert(!(m_computeFullU && m_computeThinU) && "SVDBase: you can't ask for both full and thin U");
+  eigen_assert(!(m_computeFullV && m_computeThinV) && "SVDBase: you can't ask for both full and thin V");
+  eigen_assert(EIGEN_IMPLIES(m_computeThinU || m_computeThinV, MatrixType::ColsAtCompileTime==Dynamic) &&
+	       "SVDBase: thin U and V are only available when your matrix has a dynamic number of columns.");
+
+  m_diagSize = (std::min)(m_rows, m_cols);
+  m_singularValues.resize(m_diagSize);
+  if(RowsAtCompileTime==Dynamic)
+    m_matrixU.resize(m_rows, m_computeFullU ? m_rows : m_computeThinU ? m_diagSize : 0);
+  if(ColsAtCompileTime==Dynamic)
+    m_matrixV.resize(m_cols, m_computeFullV ? m_cols : m_computeThinV ? m_diagSize : 0);
+
+  return false;
+}
+
+}// end namespace
+
+#endif // EIGEN_SVDBASE_H
diff --git a/extern/Eigen3/Eigen/src/SVD/UpperBidiagonalization.h b/extern/Eigen3/Eigen/src/SVD/UpperBidiagonalization.h
index 587de37a5ad..11ac847e1db 100644
--- a/extern/Eigen3/Eigen/src/SVD/UpperBidiagonalization.h
+++ b/extern/Eigen3/Eigen/src/SVD/UpperBidiagonalization.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2013-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -28,15 +29,15 @@ template<typename _MatrixType> class UpperBidiagonalization
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
     typedef Matrix<Scalar, 1, ColsAtCompileTime> RowVectorType;
     typedef Matrix<Scalar, RowsAtCompileTime, 1> ColVectorType;
-    typedef BandMatrix<RealScalar, ColsAtCompileTime, ColsAtCompileTime, 1, 0> BidiagonalType;
+    typedef BandMatrix<RealScalar, ColsAtCompileTime, ColsAtCompileTime, 1, 0, RowMajor> BidiagonalType;
     typedef Matrix<Scalar, ColsAtCompileTime, 1> DiagVectorType;
     typedef Matrix<Scalar, ColsAtCompileTimeMinusOne, 1> SuperDiagVectorType;
     typedef HouseholderSequence<
               const MatrixType,
-              CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, const Diagonal<const MatrixType,0> >
+              const typename internal::remove_all<typename Diagonal<const MatrixType,0>::ConjugateReturnType>::type
             > HouseholderUSequenceType;
     typedef HouseholderSequence<
               const typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type,
@@ -52,7 +53,7 @@ template<typename _MatrixType> class UpperBidiagonalization
     */
     UpperBidiagonalization() : m_householder(), m_bidiagonal(), m_isInitialized(false) {}
 
-    UpperBidiagonalization(const MatrixType& matrix)
+    explicit UpperBidiagonalization(const MatrixType& matrix)
       : m_householder(matrix.rows(), matrix.cols()),
         m_bidiagonal(matrix.cols(), matrix.cols()),
         m_isInitialized(false)
@@ -61,6 +62,7 @@ template<typename _MatrixType> class UpperBidiagonalization
     }
     
     UpperBidiagonalization& compute(const MatrixType& matrix);
+    UpperBidiagonalization& computeUnblocked(const MatrixType& matrix);
     
     const MatrixType& householder() const { return m_householder; }
     const BidiagonalType& bidiagonal() const { return m_bidiagonal; }
@@ -85,45 +87,309 @@ template<typename _MatrixType> class UpperBidiagonalization
     bool m_isInitialized;
 };
 
-template<typename _MatrixType>
-UpperBidiagonalization<_MatrixType>& UpperBidiagonalization<_MatrixType>::compute(const _MatrixType& matrix)
+// Standard upper bidiagonalization without fancy optimizations
+// This version should be faster for small matrix size
+template<typename MatrixType>
+void upperbidiagonalization_inplace_unblocked(MatrixType& mat,
+                                              typename MatrixType::RealScalar *diagonal,
+                                              typename MatrixType::RealScalar *upper_diagonal,
+                                              typename MatrixType::Scalar* tempData = 0)
 {
-  Index rows = matrix.rows();
-  Index cols = matrix.cols();
-  
-  eigen_assert(rows >= cols && "UpperBidiagonalization is only for matrices satisfying rows>=cols.");
-  
-  m_householder = matrix;
+  typedef typename MatrixType::Scalar Scalar;
 
-  ColVectorType temp(rows);
+  Index rows = mat.rows();
+  Index cols = mat.cols();
+
+  typedef Matrix<Scalar,Dynamic,1,ColMajor,MatrixType::MaxRowsAtCompileTime,1> TempType;
+  TempType tempVector;
+  if(tempData==0)
+  {
+    tempVector.resize(rows);
+    tempData = tempVector.data();
+  }
 
   for (Index k = 0; /* breaks at k==cols-1 below */ ; ++k)
   {
     Index remainingRows = rows - k;
     Index remainingCols = cols - k - 1;
 
-    // construct left householder transform in-place in m_householder
-    m_householder.col(k).tail(remainingRows)
-                 .makeHouseholderInPlace(m_householder.coeffRef(k,k),
-                                         m_bidiagonal.template diagonal<0>().coeffRef(k));
-    // apply householder transform to remaining part of m_householder on the left
-    m_householder.bottomRightCorner(remainingRows, remainingCols)
-                 .applyHouseholderOnTheLeft(m_householder.col(k).tail(remainingRows-1),
-                                            m_householder.coeff(k,k),
-                                            temp.data());
+    // construct left householder transform in-place in A
+    mat.col(k).tail(remainingRows)
+       .makeHouseholderInPlace(mat.coeffRef(k,k), diagonal[k]);
+    // apply householder transform to remaining part of A on the left
+    mat.bottomRightCorner(remainingRows, remainingCols)
+       .applyHouseholderOnTheLeft(mat.col(k).tail(remainingRows-1), mat.coeff(k,k), tempData);
 
     if(k == cols-1) break;
+
+    // construct right householder transform in-place in mat
+    mat.row(k).tail(remainingCols)
+       .makeHouseholderInPlace(mat.coeffRef(k,k+1), upper_diagonal[k]);
+    // apply householder transform to remaining part of mat on the left
+    mat.bottomRightCorner(remainingRows-1, remainingCols)
+       .applyHouseholderOnTheRight(mat.row(k).tail(remainingCols-1).transpose(), mat.coeff(k,k+1), tempData);
+  }
+}
+
+/** \internal
+  * Helper routine for the block reduction to upper bidiagonal form.
+  *
+  * Let's partition the matrix A:
+  * 
+  *      | A00 A01 |
+  *  A = |         |
+  *      | A10 A11 |
+  *
+  * This function reduces to bidiagonal form the left \c rows x \a blockSize vertical panel [A00/A10]
+  * and the \a blockSize x \c cols horizontal panel [A00 A01] of the matrix \a A. The bottom-right block A11
+  * is updated using matrix-matrix products:
+  *   A22 -= V * Y^T - X * U^T
+  * where V and U contains the left and right Householder vectors. U and V are stored in A10, and A01
+  * respectively, and the update matrices X and Y are computed during the reduction.
+  * 
+  */
+template<typename MatrixType>
+void upperbidiagonalization_blocked_helper(MatrixType& A,
+                                           typename MatrixType::RealScalar *diagonal,
+                                           typename MatrixType::RealScalar *upper_diagonal,
+                                           Index bs,
+                                           Ref<Matrix<typename MatrixType::Scalar, Dynamic, Dynamic,
+                                                      traits<MatrixType>::Flags & RowMajorBit> > X,
+                                           Ref<Matrix<typename MatrixType::Scalar, Dynamic, Dynamic,
+                                                      traits<MatrixType>::Flags & RowMajorBit> > Y)
+{
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename MatrixType::RealScalar RealScalar;
+  typedef typename NumTraits<RealScalar>::Literal Literal;
+  enum { StorageOrder = traits<MatrixType>::Flags & RowMajorBit };
+  typedef InnerStride<int(StorageOrder) == int(ColMajor) ? 1 : Dynamic> ColInnerStride;
+  typedef InnerStride<int(StorageOrder) == int(ColMajor) ? Dynamic : 1> RowInnerStride;
+  typedef Ref<Matrix<Scalar, Dynamic, 1>, 0, ColInnerStride>    SubColumnType;
+  typedef Ref<Matrix<Scalar, 1, Dynamic>, 0, RowInnerStride>    SubRowType;
+  typedef Ref<Matrix<Scalar, Dynamic, Dynamic, StorageOrder > > SubMatType;
+  
+  Index brows = A.rows();
+  Index bcols = A.cols();
+
+  Scalar tau_u, tau_u_prev(0), tau_v;
+
+  for(Index k = 0; k < bs; ++k)
+  {
+    Index remainingRows = brows - k;
+    Index remainingCols = bcols - k - 1;
+
+    SubMatType X_k1( X.block(k,0, remainingRows,k) );
+    SubMatType V_k1( A.block(k,0, remainingRows,k) );
+
+    // 1 - update the k-th column of A
+    SubColumnType v_k = A.col(k).tail(remainingRows);
+          v_k -= V_k1 * Y.row(k).head(k).adjoint();
+    if(k) v_k -= X_k1 * A.col(k).head(k);
+    
+    // 2 - construct left Householder transform in-place
+    v_k.makeHouseholderInPlace(tau_v, diagonal[k]);
+       
+    if(k+1<bcols)
+    {
+      SubMatType Y_k  ( Y.block(k+1,0, remainingCols, k+1) );
+      SubMatType U_k1 ( A.block(0,k+1, k,remainingCols) );
+      
+      // this eases the application of Householder transforAions
+      // A(k,k) will store tau_v later
+      A(k,k) = Scalar(1);
+
+      // 3 - Compute y_k^T = tau_v * ( A^T*v_k - Y_k-1*V_k-1^T*v_k - U_k-1*X_k-1^T*v_k )
+      {
+        SubColumnType y_k( Y.col(k).tail(remainingCols) );
+        
+        // let's use the begining of column k of Y as a temporary vector
+        SubColumnType tmp( Y.col(k).head(k) );
+        y_k.noalias()  = A.block(k,k+1, remainingRows,remainingCols).adjoint() * v_k; // bottleneck
+        tmp.noalias()  = V_k1.adjoint()  * v_k;
+        y_k.noalias() -= Y_k.leftCols(k) * tmp;
+        tmp.noalias()  = X_k1.adjoint()  * v_k;
+        y_k.noalias() -= U_k1.adjoint()  * tmp;
+        y_k *= numext::conj(tau_v);
+      }
+
+      // 4 - update k-th row of A (it will become u_k)
+      SubRowType u_k( A.row(k).tail(remainingCols) );
+      u_k = u_k.conjugate();
+      {
+        u_k -= Y_k * A.row(k).head(k+1).adjoint();
+        if(k) u_k -= U_k1.adjoint() * X.row(k).head(k).adjoint();
+      }
+
+      // 5 - construct right Householder transform in-place
+      u_k.makeHouseholderInPlace(tau_u, upper_diagonal[k]);
+
+      // this eases the application of Householder transformations
+      // A(k,k+1) will store tau_u later
+      A(k,k+1) = Scalar(1);
+
+      // 6 - Compute x_k = tau_u * ( A*u_k - X_k-1*U_k-1^T*u_k - V_k*Y_k^T*u_k )
+      {
+        SubColumnType x_k ( X.col(k).tail(remainingRows-1) );
+        
+        // let's use the begining of column k of X as a temporary vectors
+        // note that tmp0 and tmp1 overlaps
+        SubColumnType tmp0 ( X.col(k).head(k) ),
+                      tmp1 ( X.col(k).head(k+1) );
+                    
+        x_k.noalias()   = A.block(k+1,k+1, remainingRows-1,remainingCols) * u_k.transpose(); // bottleneck
+        tmp0.noalias()  = U_k1 * u_k.transpose();
+        x_k.noalias()  -= X_k1.bottomRows(remainingRows-1) * tmp0;
+        tmp1.noalias()  = Y_k.adjoint() * u_k.transpose();
+        x_k.noalias()  -= A.block(k+1,0, remainingRows-1,k+1) * tmp1;
+        x_k *= numext::conj(tau_u);
+        tau_u = numext::conj(tau_u);
+        u_k = u_k.conjugate();
+      }
+
+      if(k>0) A.coeffRef(k-1,k) = tau_u_prev;
+      tau_u_prev = tau_u;
+    }
+    else
+      A.coeffRef(k-1,k) = tau_u_prev;
+
+    A.coeffRef(k,k) = tau_v;
+  }
+  
+  if(bs<bcols)
+    A.coeffRef(bs-1,bs) = tau_u_prev;
+
+  // update A22
+  if(bcols>bs && brows>bs)
+  {
+    SubMatType A11( A.bottomRightCorner(brows-bs,bcols-bs) );
+    SubMatType A10( A.block(bs,0, brows-bs,bs) );
+    SubMatType A01( A.block(0,bs, bs,bcols-bs) );
+    Scalar tmp = A01(bs-1,0);
+    A01(bs-1,0) = Literal(1);
+    A11.noalias() -= A10 * Y.topLeftCorner(bcols,bs).bottomRows(bcols-bs).adjoint();
+    A11.noalias() -= X.topLeftCorner(brows,bs).bottomRows(brows-bs) * A01;
+    A01(bs-1,0) = tmp;
+  }
+}
+
+/** \internal
+  *
+  * Implementation of a block-bidiagonal reduction.
+  * It is based on the following paper:
+  *   The Design of a Parallel Dense Linear Algebra Software Library: Reduction to Hessenberg, Tridiagonal, and Bidiagonal Form.
+  *   by Jaeyoung Choi, Jack J. Dongarra, David W. Walker. (1995)
+  *   section 3.3
+  */
+template<typename MatrixType, typename BidiagType>
+void upperbidiagonalization_inplace_blocked(MatrixType& A, BidiagType& bidiagonal,
+                                            Index maxBlockSize=32,
+                                            typename MatrixType::Scalar* /*tempData*/ = 0)
+{
+  typedef typename MatrixType::Scalar Scalar;
+  typedef Block<MatrixType,Dynamic,Dynamic> BlockType;
+
+  Index rows = A.rows();
+  Index cols = A.cols();
+  Index size = (std::min)(rows, cols);
+
+  // X and Y are work space
+  enum { StorageOrder = traits<MatrixType>::Flags & RowMajorBit };
+  Matrix<Scalar,
+         MatrixType::RowsAtCompileTime,
+         Dynamic,
+         StorageOrder,
+         MatrixType::MaxRowsAtCompileTime> X(rows,maxBlockSize);
+  Matrix<Scalar,
+         MatrixType::ColsAtCompileTime,
+         Dynamic,
+         StorageOrder,
+         MatrixType::MaxColsAtCompileTime> Y(cols,maxBlockSize);
+  Index blockSize = (std::min)(maxBlockSize,size);
+
+  Index k = 0;
+  for(k = 0; k < size; k += blockSize)
+  {
+    Index bs = (std::min)(size-k,blockSize);  // actual size of the block
+    Index brows = rows - k;                   // rows of the block
+    Index bcols = cols - k;                   // columns of the block
+
+    // partition the matrix A:
+    // 
+    //      | A00 A01 A02 |
+    //      |             |
+    // A  = | A10 A11 A12 |
+    //      |             |
+    //      | A20 A21 A22 |
+    //
+    // where A11 is a bs x bs diagonal block,
+    // and let:
+    //      | A11 A12 |
+    //  B = |         |
+    //      | A21 A22 |
+
+    BlockType B = A.block(k,k,brows,bcols);
     
-    // construct right householder transform in-place in m_householder
-    m_householder.row(k).tail(remainingCols)
-                 .makeHouseholderInPlace(m_householder.coeffRef(k,k+1),
-                                         m_bidiagonal.template diagonal<1>().coeffRef(k));
-    // apply householder transform to remaining part of m_householder on the left
-    m_householder.bottomRightCorner(remainingRows-1, remainingCols)
-                 .applyHouseholderOnTheRight(m_householder.row(k).tail(remainingCols-1).transpose(),
-                                             m_householder.coeff(k,k+1),
-                                             temp.data());
+    // This stage performs the bidiagonalization of A11, A21, A12, and updating of A22.
+    // Finally, the algorithm continue on the updated A22.
+    //
+    // However, if B is too small, or A22 empty, then let's use an unblocked strategy
+    if(k+bs==cols || bcols<48) // somewhat arbitrary threshold
+    {
+      upperbidiagonalization_inplace_unblocked(B,
+                                               &(bidiagonal.template diagonal<0>().coeffRef(k)),
+                                               &(bidiagonal.template diagonal<1>().coeffRef(k)),
+                                               X.data()
+                                              );
+      break; // We're done
+    }
+    else
+    {
+      upperbidiagonalization_blocked_helper<BlockType>( B,
+                                                        &(bidiagonal.template diagonal<0>().coeffRef(k)),
+                                                        &(bidiagonal.template diagonal<1>().coeffRef(k)),
+                                                        bs,
+                                                        X.topLeftCorner(brows,bs),
+                                                        Y.topLeftCorner(bcols,bs)
+                                                      );
+    }
   }
+}
+
+template<typename _MatrixType>
+UpperBidiagonalization<_MatrixType>& UpperBidiagonalization<_MatrixType>::computeUnblocked(const _MatrixType& matrix)
+{
+  Index rows = matrix.rows();
+  Index cols = matrix.cols();
+  EIGEN_ONLY_USED_FOR_DEBUG(cols);
+
+  eigen_assert(rows >= cols && "UpperBidiagonalization is only for Arices satisfying rows>=cols.");
+
+  m_householder = matrix;
+
+  ColVectorType temp(rows);
+
+  upperbidiagonalization_inplace_unblocked(m_householder,
+                                           &(m_bidiagonal.template diagonal<0>().coeffRef(0)),
+                                           &(m_bidiagonal.template diagonal<1>().coeffRef(0)),
+                                           temp.data());
+
+  m_isInitialized = true;
+  return *this;
+}
+
+template<typename _MatrixType>
+UpperBidiagonalization<_MatrixType>& UpperBidiagonalization<_MatrixType>::compute(const _MatrixType& matrix)
+{
+  Index rows = matrix.rows();
+  Index cols = matrix.cols();
+  EIGEN_ONLY_USED_FOR_DEBUG(rows);
+  EIGEN_ONLY_USED_FOR_DEBUG(cols);
+
+  eigen_assert(rows >= cols && "UpperBidiagonalization is only for Arices satisfying rows>=cols.");
+
+  m_householder = matrix;
+  upperbidiagonalization_inplace_blocked(m_householder, m_bidiagonal);
+            
   m_isInitialized = true;
   return *this;
 }
diff --git a/extern/Eigen3/Eigen/src/SparseCholesky/SimplicialCholesky.h b/extern/Eigen3/Eigen/src/SparseCholesky/SimplicialCholesky.h
index e1f96ba5a14..2907f65296b 100644
--- a/extern/Eigen3/Eigen/src/SparseCholesky/SimplicialCholesky.h
+++ b/extern/Eigen3/Eigen/src/SparseCholesky/SimplicialCholesky.h
@@ -17,43 +17,74 @@ enum SimplicialCholeskyMode {
   SimplicialCholeskyLDLT
 };
 
+namespace internal {
+  template<typename CholMatrixType, typename InputMatrixType>
+  struct simplicial_cholesky_grab_input {
+    typedef CholMatrixType const * ConstCholMatrixPtr;
+    static void run(const InputMatrixType& input, ConstCholMatrixPtr &pmat, CholMatrixType &tmp)
+    {
+      tmp = input;
+      pmat = &tmp;
+    }
+  };
+  
+  template<typename MatrixType>
+  struct simplicial_cholesky_grab_input<MatrixType,MatrixType> {
+    typedef MatrixType const * ConstMatrixPtr;
+    static void run(const MatrixType& input, ConstMatrixPtr &pmat, MatrixType &/*tmp*/)
+    {
+      pmat = &input;
+    }
+  };
+} // end namespace internal
+
 /** \ingroup SparseCholesky_Module
-  * \brief A direct sparse Cholesky factorizations
+  * \brief A base class for direct sparse Cholesky factorizations
   *
-  * These classes provide LL^T and LDL^T Cholesky factorizations of sparse matrices that are
-  * selfadjoint and positive definite. The factorization allows for solving A.X = B where
+  * This is a base class for LL^T and LDL^T Cholesky factorizations of sparse matrices that are
+  * selfadjoint and positive definite. These factorizations allow for solving A.X = B where
   * X and B can be either dense or sparse.
   * 
   * In order to reduce the fill-in, a symmetric permutation P is applied prior to the factorization
   * such that the factorized matrix is P A P^-1.
   *
-  * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
-  * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
-  *               or Upper. Default is Lower.
+  * \tparam Derived the type of the derived class, that is the actual factorization type.
   *
   */
 template<typename Derived>
-class SimplicialCholeskyBase : internal::noncopyable
+class SimplicialCholeskyBase : public SparseSolverBase<Derived>
 {
+    typedef SparseSolverBase<Derived> Base;
+    using Base::m_isInitialized;
+    
   public:
     typedef typename internal::traits<Derived>::MatrixType MatrixType;
     typedef typename internal::traits<Derived>::OrderingType OrderingType;
     enum { UpLo = internal::traits<Derived>::UpLo };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef SparseMatrix<Scalar,ColMajor,Index> CholMatrixType;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> CholMatrixType;
+    typedef CholMatrixType const * ConstCholMatrixPtr;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
+    typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
 
   public:
+    
+    using Base::derived;
 
     /** Default constructor */
     SimplicialCholeskyBase()
-      : m_info(Success), m_isInitialized(false), m_shiftOffset(0), m_shiftScale(1)
+      : m_info(Success), m_shiftOffset(0), m_shiftScale(1)
     {}
 
-    SimplicialCholeskyBase(const MatrixType& matrix)
-      : m_info(Success), m_isInitialized(false), m_shiftOffset(0), m_shiftScale(1)
+    explicit SimplicialCholeskyBase(const MatrixType& matrix)
+      : m_info(Success), m_shiftOffset(0), m_shiftScale(1)
     {
       derived().compute(matrix);
     }
@@ -79,42 +110,14 @@ class SimplicialCholeskyBase : internal::noncopyable
       return m_info;
     }
     
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<SimplicialCholeskyBase, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "Simplicial LLT or LDLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "SimplicialCholeskyBase::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<SimplicialCholeskyBase, Rhs>(*this, b.derived());
-    }
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<SimplicialCholeskyBase, Rhs>
-    solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "Simplicial LLT or LDLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "SimplicialCholesky::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<SimplicialCholeskyBase, Rhs>(*this, b.derived());
-    }
-    
     /** \returns the permutation P
       * \sa permutationPinv() */
-    const PermutationMatrix<Dynamic,Dynamic,Index>& permutationP() const
+    const PermutationMatrix<Dynamic,Dynamic,StorageIndex>& permutationP() const
     { return m_P; }
     
     /** \returns the inverse P^-1 of the permutation P
       * \sa permutationP() */
-    const PermutationMatrix<Dynamic,Dynamic,Index>& permutationPinv() const
+    const PermutationMatrix<Dynamic,Dynamic,StorageIndex>& permutationPinv() const
     { return m_Pinv; }
 
     /** Sets the shift parameters that will be used to adjust the diagonal coefficients during the numerical factorization.
@@ -150,7 +153,7 @@ class SimplicialCholeskyBase : internal::noncopyable
 
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
     {
       eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
       eigen_assert(m_matrix.rows()==b.rows());
@@ -175,6 +178,12 @@ class SimplicialCholeskyBase : internal::noncopyable
       if(m_P.size()>0)
         dest = m_Pinv * dest;
     }
+    
+    template<typename Rhs,typename Dest>
+    void _solve_impl(const SparseMatrixBase<Rhs> &b, SparseMatrixBase<Dest> &dest) const
+    {
+      internal::solve_sparse_through_dense_panels(derived(), b, dest);
+    }
 
 #endif // EIGEN_PARSED_BY_DOXYGEN
 
@@ -186,20 +195,33 @@ class SimplicialCholeskyBase : internal::noncopyable
     {
       eigen_assert(matrix.rows()==matrix.cols());
       Index size = matrix.cols();
-      CholMatrixType ap(size,size);
-      ordering(matrix, ap);
-      analyzePattern_preordered(ap, DoLDLT);
-      factorize_preordered<DoLDLT>(ap);
+      CholMatrixType tmp(size,size);
+      ConstCholMatrixPtr pmat;
+      ordering(matrix, pmat, tmp);
+      analyzePattern_preordered(*pmat, DoLDLT);
+      factorize_preordered<DoLDLT>(*pmat);
     }
     
     template<bool DoLDLT>
     void factorize(const MatrixType& a)
     {
       eigen_assert(a.rows()==a.cols());
-      int size = a.cols();
-      CholMatrixType ap(size,size);
-      ap.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>().twistedBy(m_P);
-      factorize_preordered<DoLDLT>(ap);
+      Index size = a.cols();
+      CholMatrixType tmp(size,size);
+      ConstCholMatrixPtr pmat;
+      
+      if(m_P.size()==0 && (UpLo&Upper)==Upper)
+      {
+        // If there is no ordering, try to directly use the input matrix without any copy
+        internal::simplicial_cholesky_grab_input<CholMatrixType,MatrixType>::run(a, pmat, tmp);
+      }
+      else
+      {
+        tmp.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>().twistedBy(m_P);
+        pmat = &tmp;
+      }
+      
+      factorize_preordered<DoLDLT>(*pmat);
     }
 
     template<bool DoLDLT>
@@ -208,14 +230,15 @@ class SimplicialCholeskyBase : internal::noncopyable
     void analyzePattern(const MatrixType& a, bool doLDLT)
     {
       eigen_assert(a.rows()==a.cols());
-      int size = a.cols();
-      CholMatrixType ap(size,size);
-      ordering(a, ap);
-      analyzePattern_preordered(ap,doLDLT);
+      Index size = a.cols();
+      CholMatrixType tmp(size,size);
+      ConstCholMatrixPtr pmat;
+      ordering(a, pmat, tmp);
+      analyzePattern_preordered(*pmat,doLDLT);
     }
     void analyzePattern_preordered(const CholMatrixType& a, bool doLDLT);
     
-    void ordering(const MatrixType& a, CholMatrixType& ap);
+    void ordering(const MatrixType& a, ConstCholMatrixPtr &pmat, CholMatrixType& ap);
 
     /** keeps off-diagonal entries; drops diagonal entries */
     struct keep_diag {
@@ -226,24 +249,23 @@ class SimplicialCholeskyBase : internal::noncopyable
     };
 
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     bool m_factorizationIsOk;
     bool m_analysisIsOk;
     
     CholMatrixType m_matrix;
     VectorType m_diag;                                // the diagonal coefficients (LDLT mode)
-    VectorXi m_parent;                                // elimination tree
-    VectorXi m_nonZerosPerCol;
-    PermutationMatrix<Dynamic,Dynamic,Index> m_P;     // the permutation
-    PermutationMatrix<Dynamic,Dynamic,Index> m_Pinv;  // the inverse permutation
+    VectorI m_parent;                                 // elimination tree
+    VectorI m_nonZerosPerCol;
+    PermutationMatrix<Dynamic,Dynamic,StorageIndex> m_P;     // the permutation
+    PermutationMatrix<Dynamic,Dynamic,StorageIndex> m_Pinv;  // the inverse permutation
 
     RealScalar m_shiftOffset;
     RealScalar m_shiftScale;
 };
 
-template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::Index> > class SimplicialLLT;
-template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::Index> > class SimplicialLDLT;
-template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::Index> > class SimplicialCholesky;
+template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::StorageIndex> > class SimplicialLLT;
+template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::StorageIndex> > class SimplicialLDLT;
+template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::StorageIndex> > class SimplicialCholesky;
 
 namespace internal {
 
@@ -253,12 +275,12 @@ template<typename _MatrixType, int _UpLo, typename _Ordering> struct traits<Simp
   typedef _Ordering OrderingType;
   enum { UpLo = _UpLo };
   typedef typename MatrixType::Scalar                         Scalar;
-  typedef typename MatrixType::Index                          Index;
-  typedef SparseMatrix<Scalar, ColMajor, Index>               CholMatrixType;
-  typedef SparseTriangularView<CholMatrixType, Eigen::Lower>  MatrixL;
-  typedef SparseTriangularView<typename CholMatrixType::AdjointReturnType, Eigen::Upper>   MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m; }
-  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  typedef typename MatrixType::StorageIndex                   StorageIndex;
+  typedef SparseMatrix<Scalar, ColMajor, StorageIndex>        CholMatrixType;
+  typedef TriangularView<const CholMatrixType, Eigen::Lower>  MatrixL;
+  typedef TriangularView<const typename CholMatrixType::AdjointReturnType, Eigen::Upper>   MatrixU;
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
 };
 
 template<typename _MatrixType,int _UpLo, typename _Ordering> struct traits<SimplicialLDLT<_MatrixType,_UpLo,_Ordering> >
@@ -267,12 +289,12 @@ template<typename _MatrixType,int _UpLo, typename _Ordering> struct traits<Simpl
   typedef _Ordering OrderingType;
   enum { UpLo = _UpLo };
   typedef typename MatrixType::Scalar                             Scalar;
-  typedef typename MatrixType::Index                              Index;
-  typedef SparseMatrix<Scalar, ColMajor, Index>                   CholMatrixType;
-  typedef SparseTriangularView<CholMatrixType, Eigen::UnitLower>  MatrixL;
-  typedef SparseTriangularView<typename CholMatrixType::AdjointReturnType, Eigen::UnitUpper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m; }
-  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  typedef typename MatrixType::StorageIndex                       StorageIndex;
+  typedef SparseMatrix<Scalar, ColMajor, StorageIndex>            CholMatrixType;
+  typedef TriangularView<const CholMatrixType, Eigen::UnitLower>  MatrixL;
+  typedef TriangularView<const typename CholMatrixType::AdjointReturnType, Eigen::UnitUpper> MatrixU;
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
 };
 
 template<typename _MatrixType, int _UpLo, typename _Ordering> struct traits<SimplicialCholesky<_MatrixType,_UpLo,_Ordering> >
@@ -300,6 +322,8 @@ template<typename _MatrixType, int _UpLo, typename _Ordering> struct traits<Simp
   *               or Upper. Default is Lower.
   * \tparam _Ordering The ordering method to use, either AMDOrdering<> or NaturalOrdering<>. Default is AMDOrdering<>
   *
+  * \implsparsesolverconcept
+  *
   * \sa class SimplicialLDLT, class AMDOrdering, class NaturalOrdering
   */
 template<typename _MatrixType, int _UpLo, typename _Ordering>
@@ -311,7 +335,7 @@ public:
     typedef SimplicialCholeskyBase<SimplicialLLT> Base;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef SparseMatrix<Scalar,ColMajor,Index> CholMatrixType;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
     typedef internal::traits<SimplicialLLT> Traits;
@@ -321,7 +345,7 @@ public:
     /** Default constructor */
     SimplicialLLT() : Base() {}
     /** Constructs and performs the LLT factorization of \a matrix */
-    SimplicialLLT(const MatrixType& matrix)
+    explicit SimplicialLLT(const MatrixType& matrix)
         : Base(matrix) {}
 
     /** \returns an expression of the factor L */
@@ -389,6 +413,8 @@ public:
   *               or Upper. Default is Lower.
   * \tparam _Ordering The ordering method to use, either AMDOrdering<> or NaturalOrdering<>. Default is AMDOrdering<>
   *
+  * \implsparsesolverconcept
+  *
   * \sa class SimplicialLLT, class AMDOrdering, class NaturalOrdering
   */
 template<typename _MatrixType, int _UpLo, typename _Ordering>
@@ -400,8 +426,8 @@ public:
     typedef SimplicialCholeskyBase<SimplicialLDLT> Base;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef SparseMatrix<Scalar,ColMajor,Index> CholMatrixType;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> CholMatrixType;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
     typedef internal::traits<SimplicialLDLT> Traits;
     typedef typename Traits::MatrixL  MatrixL;
@@ -411,7 +437,7 @@ public:
     SimplicialLDLT() : Base() {}
 
     /** Constructs and performs the LLT factorization of \a matrix */
-    SimplicialLDLT(const MatrixType& matrix)
+    explicit SimplicialLDLT(const MatrixType& matrix)
         : Base(matrix) {}
 
     /** \returns a vector expression of the diagonal D */
@@ -482,8 +508,8 @@ public:
     typedef SimplicialCholeskyBase<SimplicialCholesky> Base;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef SparseMatrix<Scalar,ColMajor,Index> CholMatrixType;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> CholMatrixType;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
     typedef internal::traits<SimplicialCholesky> Traits;
     typedef internal::traits<SimplicialLDLT<MatrixType,UpLo> > LDLTTraits;
@@ -491,7 +517,7 @@ public:
   public:
     SimplicialCholesky() : Base(), m_LDLT(true) {}
 
-    SimplicialCholesky(const MatrixType& matrix)
+    explicit SimplicialCholesky(const MatrixType& matrix)
       : Base(), m_LDLT(true)
     {
       compute(matrix);
@@ -560,7 +586,7 @@ public:
 
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
     {
       eigen_assert(Base::m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
       eigen_assert(Base::m_matrix.rows()==b.rows());
@@ -596,6 +622,13 @@ public:
         dest = Base::m_Pinv * dest;
     }
     
+    /** \internal */
+    template<typename Rhs,typename Dest>
+    void _solve_impl(const SparseMatrixBase<Rhs> &b, SparseMatrixBase<Dest> &dest) const
+    {
+      internal::solve_sparse_through_dense_panels(*this, b, dest);
+    }
+    
     Scalar determinant() const
     {
       if(m_LDLT)
@@ -614,58 +647,43 @@ public:
 };
 
 template<typename Derived>
-void SimplicialCholeskyBase<Derived>::ordering(const MatrixType& a, CholMatrixType& ap)
+void SimplicialCholeskyBase<Derived>::ordering(const MatrixType& a, ConstCholMatrixPtr &pmat, CholMatrixType& ap)
 {
   eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
-  // Note that amd compute the inverse permutation
+  pmat = &ap;
+  // Note that ordering methods compute the inverse permutation
+  if(!internal::is_same<OrderingType,NaturalOrdering<Index> >::value)
   {
-    CholMatrixType C;
-    C = a.template selfadjointView<UpLo>();
+    {
+      CholMatrixType C;
+      C = a.template selfadjointView<UpLo>();
+      
+      OrderingType ordering;
+      ordering(C,m_Pinv);
+    }
+
+    if(m_Pinv.size()>0) m_P = m_Pinv.inverse();
+    else                m_P.resize(0);
     
-    OrderingType ordering;
-    ordering(C,m_Pinv);
+    ap.resize(size,size);
+    ap.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>().twistedBy(m_P);
   }
-
-  if(m_Pinv.size()>0)
-    m_P = m_Pinv.inverse();
   else
+  {
+    m_Pinv.resize(0);
     m_P.resize(0);
-
-  ap.resize(size,size);
-  ap.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>().twistedBy(m_P);
+    if(int(UpLo)==int(Lower) || MatrixType::IsRowMajor)
+    {
+      // we have to transpose the lower part to to the upper one
+      ap.resize(size,size);
+      ap.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>();
+    }
+    else
+      internal::simplicial_cholesky_grab_input<CholMatrixType,MatrixType>::run(a, pmat, ap);
+  }  
 }
 
-namespace internal {
-  
-template<typename Derived, typename Rhs>
-struct solve_retval<SimplicialCholeskyBase<Derived>, Rhs>
-  : solve_retval_base<SimplicialCholeskyBase<Derived>, Rhs>
-{
-  typedef SimplicialCholeskyBase<Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec().derived()._solve(rhs(),dst);
-  }
-};
-
-template<typename Derived, typename Rhs>
-struct sparse_solve_retval<SimplicialCholeskyBase<Derived>, Rhs>
-  : sparse_solve_retval_base<SimplicialCholeskyBase<Derived>, Rhs>
-{
-  typedef SimplicialCholeskyBase<Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_SIMPLICIAL_CHOLESKY_H
diff --git a/extern/Eigen3/Eigen/src/SparseCholesky/SimplicialCholesky_impl.h b/extern/Eigen3/Eigen/src/SparseCholesky/SimplicialCholesky_impl.h
index 7aaf702beed..31e06995b8a 100644
--- a/extern/Eigen3/Eigen/src/SparseCholesky/SimplicialCholesky_impl.h
+++ b/extern/Eigen3/Eigen/src/SparseCholesky/SimplicialCholesky_impl.h
@@ -50,14 +50,14 @@ namespace Eigen {
 template<typename Derived>
 void SimplicialCholeskyBase<Derived>::analyzePattern_preordered(const CholMatrixType& ap, bool doLDLT)
 {
-  const Index size = ap.rows();
+  const StorageIndex size = StorageIndex(ap.rows());
   m_matrix.resize(size, size);
   m_parent.resize(size);
   m_nonZerosPerCol.resize(size);
 
-  ei_declare_aligned_stack_constructed_variable(Index, tags, size, 0);
+  ei_declare_aligned_stack_constructed_variable(StorageIndex, tags, size, 0);
 
-  for(Index k = 0; k < size; ++k)
+  for(StorageIndex k = 0; k < size; ++k)
   {
     /* L(k,:) pattern: all nodes reachable in etree from nz in A(0:k-1,k) */
     m_parent[k] = -1;             /* parent of k is not yet known */
@@ -65,7 +65,7 @@ void SimplicialCholeskyBase<Derived>::analyzePattern_preordered(const CholMatrix
     m_nonZerosPerCol[k] = 0;      /* count of nonzeros in column k of L */
     for(typename CholMatrixType::InnerIterator it(ap,k); it; ++it)
     {
-      Index i = it.index();
+      StorageIndex i = it.index();
       if(i < k)
       {
         /* follow path from i to root of etree, stop at flagged node */
@@ -82,9 +82,9 @@ void SimplicialCholeskyBase<Derived>::analyzePattern_preordered(const CholMatrix
   }
 
   /* construct Lp index array from m_nonZerosPerCol column counts */
-  Index* Lp = m_matrix.outerIndexPtr();
+  StorageIndex* Lp = m_matrix.outerIndexPtr();
   Lp[0] = 0;
-  for(Index k = 0; k < size; ++k)
+  for(StorageIndex k = 0; k < size; ++k)
     Lp[k+1] = Lp[k] + m_nonZerosPerCol[k] + (doLDLT ? 0 : 1);
 
   m_matrix.resizeNonZeros(Lp[size]);
@@ -104,31 +104,31 @@ void SimplicialCholeskyBase<Derived>::factorize_preordered(const CholMatrixType&
 
   eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
   eigen_assert(ap.rows()==ap.cols());
-  const Index size = ap.rows();
-  eigen_assert(m_parent.size()==size);
-  eigen_assert(m_nonZerosPerCol.size()==size);
+  eigen_assert(m_parent.size()==ap.rows());
+  eigen_assert(m_nonZerosPerCol.size()==ap.rows());
 
-  const Index* Lp = m_matrix.outerIndexPtr();
-  Index* Li = m_matrix.innerIndexPtr();
+  const StorageIndex size = StorageIndex(ap.rows());
+  const StorageIndex* Lp = m_matrix.outerIndexPtr();
+  StorageIndex* Li = m_matrix.innerIndexPtr();
   Scalar* Lx = m_matrix.valuePtr();
 
   ei_declare_aligned_stack_constructed_variable(Scalar, y, size, 0);
-  ei_declare_aligned_stack_constructed_variable(Index,  pattern, size, 0);
-  ei_declare_aligned_stack_constructed_variable(Index,  tags, size, 0);
+  ei_declare_aligned_stack_constructed_variable(StorageIndex,  pattern, size, 0);
+  ei_declare_aligned_stack_constructed_variable(StorageIndex,  tags, size, 0);
 
   bool ok = true;
   m_diag.resize(DoLDLT ? size : 0);
 
-  for(Index k = 0; k < size; ++k)
+  for(StorageIndex k = 0; k < size; ++k)
   {
     // compute nonzero pattern of kth row of L, in topological order
     y[k] = 0.0;                     // Y(0:k) is now all zero
-    Index top = size;               // stack for pattern is empty
+    StorageIndex top = size;               // stack for pattern is empty
     tags[k] = k;                    // mark node k as visited
     m_nonZerosPerCol[k] = 0;        // count of nonzeros in column k of L
-    for(typename MatrixType::InnerIterator it(ap,k); it; ++it)
+    for(typename CholMatrixType::InnerIterator it(ap,k); it; ++it)
     {
-      Index i = it.index();
+      StorageIndex i = it.index();
       if(i <= k)
       {
         y[i] += numext::conj(it.value());            /* scatter A(i,k) into Y (sum duplicates) */
diff --git a/extern/Eigen3/Eigen/src/SparseCore/AmbiVector.h b/extern/Eigen3/Eigen/src/SparseCore/AmbiVector.h
index 220c6451cb9..e0295f2af12 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/AmbiVector.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/AmbiVector.h
@@ -19,15 +19,15 @@ namespace internal {
   *
   * See BasicSparseLLT and SparseProduct for usage examples.
   */
-template<typename _Scalar, typename _Index>
+template<typename _Scalar, typename _StorageIndex>
 class AmbiVector
 {
   public:
     typedef _Scalar Scalar;
-    typedef _Index Index;
+    typedef _StorageIndex StorageIndex;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
-    AmbiVector(Index size)
+    explicit AmbiVector(Index size)
       : m_buffer(0), m_zero(0), m_size(0), m_allocatedSize(0), m_allocatedElements(0), m_mode(-1)
     {
       resize(size);
@@ -39,7 +39,7 @@ class AmbiVector
     Index nonZeros() const;
 
     /** Specifies a sub-vector to work on */
-    void setBounds(Index start, Index end) { m_start = start; m_end = end; }
+    void setBounds(Index start, Index end) { m_start = convert_index(start); m_end = convert_index(end); }
 
     void setZero();
 
@@ -55,12 +55,16 @@ class AmbiVector
     {
       if (m_allocatedSize < size)
         reallocate(size);
-      m_size = size;
+      m_size = convert_index(size);
     }
 
-    Index size() const { return m_size; }
+    StorageIndex size() const { return m_size; }
 
   protected:
+    StorageIndex convert_index(Index idx)
+    {
+      return internal::convert_index<StorageIndex>(idx);
+    }
 
     void reallocate(Index size)
     {
@@ -70,15 +74,15 @@ class AmbiVector
       if (size<1000)
       {
         Index allocSize = (size * sizeof(ListEl) + sizeof(Scalar) - 1)/sizeof(Scalar);
-        m_allocatedElements = (allocSize*sizeof(Scalar))/sizeof(ListEl);
+        m_allocatedElements = convert_index((allocSize*sizeof(Scalar))/sizeof(ListEl));
         m_buffer = new Scalar[allocSize];
       }
       else
       {
-        m_allocatedElements = (size*sizeof(Scalar))/sizeof(ListEl);
+        m_allocatedElements = convert_index((size*sizeof(Scalar))/sizeof(ListEl));
         m_buffer = new Scalar[size];
       }
-      m_size = size;
+      m_size = convert_index(size);
       m_start = 0;
       m_end = m_size;
     }
@@ -86,11 +90,11 @@ class AmbiVector
     void reallocateSparse()
     {
       Index copyElements = m_allocatedElements;
-      m_allocatedElements = (std::min)(Index(m_allocatedElements*1.5),m_size);
+      m_allocatedElements = (std::min)(StorageIndex(m_allocatedElements*1.5),m_size);
       Index allocSize = m_allocatedElements * sizeof(ListEl);
       allocSize = (allocSize + sizeof(Scalar) - 1)/sizeof(Scalar);
       Scalar* newBuffer = new Scalar[allocSize];
-      memcpy(newBuffer,  m_buffer,  copyElements * sizeof(ListEl));
+      std::memcpy(newBuffer,  m_buffer,  copyElements * sizeof(ListEl));
       delete[] m_buffer;
       m_buffer = newBuffer;
     }
@@ -99,30 +103,30 @@ class AmbiVector
     // element type of the linked list
     struct ListEl
     {
-      Index next;
-      Index index;
+      StorageIndex next;
+      StorageIndex index;
       Scalar value;
     };
 
     // used to store data in both mode
     Scalar* m_buffer;
     Scalar m_zero;
-    Index m_size;
-    Index m_start;
-    Index m_end;
-    Index m_allocatedSize;
-    Index m_allocatedElements;
-    Index m_mode;
+    StorageIndex m_size;
+    StorageIndex m_start;
+    StorageIndex m_end;
+    StorageIndex m_allocatedSize;
+    StorageIndex m_allocatedElements;
+    StorageIndex m_mode;
 
     // linked list mode
-    Index m_llStart;
-    Index m_llCurrent;
-    Index m_llSize;
+    StorageIndex m_llStart;
+    StorageIndex m_llCurrent;
+    StorageIndex m_llSize;
 };
 
 /** \returns the number of non zeros in the current sub vector */
-template<typename _Scalar,typename _Index>
-_Index AmbiVector<_Scalar,_Index>::nonZeros() const
+template<typename _Scalar,typename _StorageIndex>
+Index AmbiVector<_Scalar,_StorageIndex>::nonZeros() const
 {
   if (m_mode==IsSparse)
     return m_llSize;
@@ -130,8 +134,8 @@ _Index AmbiVector<_Scalar,_Index>::nonZeros() const
     return m_end - m_start;
 }
 
-template<typename _Scalar,typename _Index>
-void AmbiVector<_Scalar,_Index>::init(double estimatedDensity)
+template<typename _Scalar,typename _StorageIndex>
+void AmbiVector<_Scalar,_StorageIndex>::init(double estimatedDensity)
 {
   if (estimatedDensity>0.1)
     init(IsDense);
@@ -139,8 +143,8 @@ void AmbiVector<_Scalar,_Index>::init(double estimatedDensity)
     init(IsSparse);
 }
 
-template<typename _Scalar,typename _Index>
-void AmbiVector<_Scalar,_Index>::init(int mode)
+template<typename _Scalar,typename _StorageIndex>
+void AmbiVector<_Scalar,_StorageIndex>::init(int mode)
 {
   m_mode = mode;
   if (m_mode==IsSparse)
@@ -155,15 +159,15 @@ void AmbiVector<_Scalar,_Index>::init(int mode)
   *
   * Don't worry, this function is extremely cheap.
   */
-template<typename _Scalar,typename _Index>
-void AmbiVector<_Scalar,_Index>::restart()
+template<typename _Scalar,typename _StorageIndex>
+void AmbiVector<_Scalar,_StorageIndex>::restart()
 {
   m_llCurrent = m_llStart;
 }
 
 /** Set all coefficients of current subvector to zero */
-template<typename _Scalar,typename _Index>
-void AmbiVector<_Scalar,_Index>::setZero()
+template<typename _Scalar,typename _StorageIndex>
+void AmbiVector<_Scalar,_StorageIndex>::setZero()
 {
   if (m_mode==IsDense)
   {
@@ -178,8 +182,8 @@ void AmbiVector<_Scalar,_Index>::setZero()
   }
 }
 
-template<typename _Scalar,typename _Index>
-_Scalar& AmbiVector<_Scalar,_Index>::coeffRef(_Index i)
+template<typename _Scalar,typename _StorageIndex>
+_Scalar& AmbiVector<_Scalar,_StorageIndex>::coeffRef(Index i)
 {
   if (m_mode==IsDense)
     return m_buffer[i];
@@ -195,7 +199,7 @@ _Scalar& AmbiVector<_Scalar,_Index>::coeffRef(_Index i)
       m_llCurrent = 0;
       ++m_llSize;
       llElements[0].value = Scalar(0);
-      llElements[0].index = i;
+      llElements[0].index = convert_index(i);
       llElements[0].next = -1;
       return llElements[0].value;
     }
@@ -204,7 +208,7 @@ _Scalar& AmbiVector<_Scalar,_Index>::coeffRef(_Index i)
       // this is going to be the new first element of the list
       ListEl& el = llElements[m_llSize];
       el.value = Scalar(0);
-      el.index = i;
+      el.index = convert_index(i);
       el.next = m_llStart;
       m_llStart = m_llSize;
       ++m_llSize;
@@ -213,7 +217,7 @@ _Scalar& AmbiVector<_Scalar,_Index>::coeffRef(_Index i)
     }
     else
     {
-      Index nextel = llElements[m_llCurrent].next;
+      StorageIndex nextel = llElements[m_llCurrent].next;
       eigen_assert(i>=llElements[m_llCurrent].index && "you must call restart() before inserting an element with lower or equal index");
       while (nextel >= 0 && llElements[nextel].index<=i)
       {
@@ -237,7 +241,7 @@ _Scalar& AmbiVector<_Scalar,_Index>::coeffRef(_Index i)
         // let's insert a new coefficient
         ListEl& el = llElements[m_llSize];
         el.value = Scalar(0);
-        el.index = i;
+        el.index = convert_index(i);
         el.next = llElements[m_llCurrent].next;
         llElements[m_llCurrent].next = m_llSize;
         ++m_llSize;
@@ -247,8 +251,8 @@ _Scalar& AmbiVector<_Scalar,_Index>::coeffRef(_Index i)
   }
 }
 
-template<typename _Scalar,typename _Index>
-_Scalar& AmbiVector<_Scalar,_Index>::coeff(_Index i)
+template<typename _Scalar,typename _StorageIndex>
+_Scalar& AmbiVector<_Scalar,_StorageIndex>::coeff(Index i)
 {
   if (m_mode==IsDense)
     return m_buffer[i];
@@ -275,8 +279,8 @@ _Scalar& AmbiVector<_Scalar,_Index>::coeff(_Index i)
 }
 
 /** Iterator over the nonzero coefficients */
-template<typename _Scalar,typename _Index>
-class AmbiVector<_Scalar,_Index>::Iterator
+template<typename _Scalar,typename _StorageIndex>
+class AmbiVector<_Scalar,_StorageIndex>::Iterator
 {
   public:
     typedef _Scalar Scalar;
@@ -288,7 +292,7 @@ class AmbiVector<_Scalar,_Index>::Iterator
       * In practice, all coefficients having a magnitude smaller than \a epsilon
       * are skipped.
       */
-    Iterator(const AmbiVector& vec, const RealScalar& epsilon = 0)
+    explicit Iterator(const AmbiVector& vec, const RealScalar& epsilon = 0)
       : m_vector(vec)
     {
       using std::abs;
@@ -320,7 +324,7 @@ class AmbiVector<_Scalar,_Index>::Iterator
       }
     }
 
-    Index index() const { return m_cachedIndex; }
+    StorageIndex index() const { return m_cachedIndex; }
     Scalar value() const { return m_cachedValue; }
 
     operator bool() const { return m_cachedIndex>=0; }
@@ -332,7 +336,7 @@ class AmbiVector<_Scalar,_Index>::Iterator
       {
         do {
           ++m_cachedIndex;
-        } while (m_cachedIndex<m_vector.m_end && abs(m_vector.m_buffer[m_cachedIndex])<m_epsilon);
+        } while (m_cachedIndex<m_vector.m_end && abs(m_vector.m_buffer[m_cachedIndex])<=m_epsilon);
         if (m_cachedIndex<m_vector.m_end)
           m_cachedValue = m_vector.m_buffer[m_cachedIndex];
         else
@@ -343,7 +347,7 @@ class AmbiVector<_Scalar,_Index>::Iterator
         ListEl* EIGEN_RESTRICT llElements = reinterpret_cast<ListEl*>(m_vector.m_buffer);
         do {
           m_currentEl = llElements[m_currentEl].next;
-        } while (m_currentEl>=0 && abs(llElements[m_currentEl].value)<m_epsilon);
+        } while (m_currentEl>=0 && abs(llElements[m_currentEl].value)<=m_epsilon);
         if (m_currentEl<0)
         {
           m_cachedIndex = -1;
@@ -359,9 +363,9 @@ class AmbiVector<_Scalar,_Index>::Iterator
 
   protected:
     const AmbiVector& m_vector; // the target vector
-    Index m_currentEl;            // the current element in sparse/linked-list mode
+    StorageIndex m_currentEl;   // the current element in sparse/linked-list mode
     RealScalar m_epsilon;       // epsilon used to prune zero coefficients
-    Index m_cachedIndex;          // current coordinate
+    StorageIndex m_cachedIndex; // current coordinate
     Scalar m_cachedValue;       // current value
     bool m_isDense;             // mode of the vector
 };
diff --git a/extern/Eigen3/Eigen/src/SparseCore/CompressedStorage.h b/extern/Eigen3/Eigen/src/SparseCore/CompressedStorage.h
index a667cb56ebe..d89fa0dae47 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/CompressedStorage.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/CompressedStorage.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -18,13 +18,13 @@ namespace internal {
   * Stores a sparse set of values as a list of values and a list of indices.
   *
   */
-template<typename _Scalar,typename _Index>
+template<typename _Scalar,typename _StorageIndex>
 class CompressedStorage
 {
   public:
 
     typedef _Scalar Scalar;
-    typedef _Index Index;
+    typedef _StorageIndex StorageIndex;
 
   protected:
 
@@ -36,7 +36,7 @@ class CompressedStorage
       : m_values(0), m_indices(0), m_size(0), m_allocatedSize(0)
     {}
 
-    CompressedStorage(size_t size)
+    explicit CompressedStorage(Index size)
       : m_values(0), m_indices(0), m_size(0), m_allocatedSize(0)
     {
       resize(size);
@@ -51,8 +51,11 @@ class CompressedStorage
     CompressedStorage& operator=(const CompressedStorage& other)
     {
       resize(other.size());
-      internal::smart_copy(other.m_values,  other.m_values  + m_size, m_values);
-      internal::smart_copy(other.m_indices, other.m_indices + m_size, m_indices);
+      if(other.size()>0)
+      {
+        internal::smart_copy(other.m_values,  other.m_values  + m_size, m_values);
+        internal::smart_copy(other.m_indices, other.m_indices + m_size, m_indices);
+      }
       return *this;
     }
 
@@ -70,9 +73,9 @@ class CompressedStorage
       delete[] m_indices;
     }
 
-    void reserve(size_t size)
+    void reserve(Index size)
     {
-      size_t newAllocatedSize = m_size + size;
+      Index newAllocatedSize = m_size + size;
       if (newAllocatedSize > m_allocatedSize)
         reallocate(newAllocatedSize);
     }
@@ -83,39 +86,40 @@ class CompressedStorage
         reallocate(m_size);
     }
 
-    void resize(size_t size, double reserveSizeFactor = 0)
+    void resize(Index size, double reserveSizeFactor = 0)
     {
       if (m_allocatedSize<size)
-        reallocate(size + size_t(reserveSizeFactor*double(size)));
+      {
+        Index realloc_size = (std::min<Index>)(NumTraits<StorageIndex>::highest(),  size + Index(reserveSizeFactor*double(size)));
+        if(realloc_size<size)
+          internal::throw_std_bad_alloc();
+        reallocate(realloc_size);
+      }
       m_size = size;
     }
 
     void append(const Scalar& v, Index i)
     {
-      Index id = static_cast<Index>(m_size);
+      Index id = m_size;
       resize(m_size+1, 1);
       m_values[id] = v;
-      m_indices[id] = i;
+      m_indices[id] = internal::convert_index<StorageIndex>(i);
     }
 
-    inline size_t size() const { return m_size; }
-    inline size_t allocatedSize() const { return m_allocatedSize; }
+    inline Index size() const { return m_size; }
+    inline Index allocatedSize() const { return m_allocatedSize; }
     inline void clear() { m_size = 0; }
 
-    inline Scalar& value(size_t i) { return m_values[i]; }
-    inline const Scalar& value(size_t i) const { return m_values[i]; }
+    const Scalar* valuePtr() const { return m_values; }
+    Scalar* valuePtr() { return m_values; }
+    const StorageIndex* indexPtr() const { return m_indices; }
+    StorageIndex* indexPtr() { return m_indices; }
 
-    inline Index& index(size_t i) { return m_indices[i]; }
-    inline const Index& index(size_t i) const { return m_indices[i]; }
+    inline Scalar& value(Index i) { eigen_internal_assert(m_values!=0); return m_values[i]; }
+    inline const Scalar& value(Index i) const { eigen_internal_assert(m_values!=0); return m_values[i]; }
 
-    static CompressedStorage Map(Index* indices, Scalar* values, size_t size)
-    {
-      CompressedStorage res;
-      res.m_indices = indices;
-      res.m_values = values;
-      res.m_allocatedSize = res.m_size = size;
-      return res;
-    }
+    inline StorageIndex& index(Index i) { eigen_internal_assert(m_indices!=0); return m_indices[i]; }
+    inline const StorageIndex& index(Index i) const { eigen_internal_assert(m_indices!=0); return m_indices[i]; }
 
     /** \returns the largest \c k such that for all \c j in [0,k) index[\c j]\<\a key */
     inline Index searchLowerIndex(Index key) const
@@ -124,17 +128,17 @@ class CompressedStorage
     }
 
     /** \returns the largest \c k in [start,end) such that for all \c j in [start,k) index[\c j]\<\a key */
-    inline Index searchLowerIndex(size_t start, size_t end, Index key) const
+    inline Index searchLowerIndex(Index start, Index end, Index key) const
     {
       while(end>start)
       {
-        size_t mid = (end+start)>>1;
+        Index mid = (end+start)>>1;
         if (m_indices[mid]<key)
           start = mid+1;
         else
           end = mid;
       }
-      return static_cast<Index>(start);
+      return start;
     }
 
     /** \returns the stored value at index \a key
@@ -147,20 +151,20 @@ class CompressedStorage
         return m_values[m_size-1];
       // ^^  optimization: let's first check if it is the last coefficient
       // (very common in high level algorithms)
-      const size_t id = searchLowerIndex(0,m_size-1,key);
+      const Index id = searchLowerIndex(0,m_size-1,key);
       return ((id<m_size) && (m_indices[id]==key)) ? m_values[id] : defaultValue;
     }
 
     /** Like at(), but the search is performed in the range [start,end) */
-    inline Scalar atInRange(size_t start, size_t end, Index key, const Scalar& defaultValue = Scalar(0)) const
+    inline Scalar atInRange(Index start, Index end, Index key, const Scalar &defaultValue = Scalar(0)) const
     {
       if (start>=end)
-        return Scalar(0);
+        return defaultValue;
       else if (end>start && key==m_indices[end-1])
         return m_values[end-1];
       // ^^  optimization: let's first check if it is the last coefficient
       // (very common in high level algorithms)
-      const size_t id = searchLowerIndex(start,end-1,key);
+      const Index id = searchLowerIndex(start,end-1,key);
       return ((id<end) && (m_indices[id]==key)) ? m_values[id] : defaultValue;
     }
 
@@ -169,16 +173,35 @@ class CompressedStorage
       * such that the keys are sorted. */
     inline Scalar& atWithInsertion(Index key, const Scalar& defaultValue = Scalar(0))
     {
-      size_t id = searchLowerIndex(0,m_size,key);
+      Index id = searchLowerIndex(0,m_size,key);
       if (id>=m_size || m_indices[id]!=key)
       {
-        resize(m_size+1,1);
-        for (size_t j=m_size-1; j>id; --j)
+        if (m_allocatedSize<m_size+1)
+        {
+          m_allocatedSize = 2*(m_size+1);
+          internal::scoped_array<Scalar> newValues(m_allocatedSize);
+          internal::scoped_array<StorageIndex> newIndices(m_allocatedSize);
+
+          // copy first chunk
+          internal::smart_copy(m_values,  m_values +id, newValues.ptr());
+          internal::smart_copy(m_indices, m_indices+id, newIndices.ptr());
+
+          // copy the rest
+          if(m_size>id)
+          {
+            internal::smart_copy(m_values +id,  m_values +m_size, newValues.ptr() +id+1);
+            internal::smart_copy(m_indices+id,  m_indices+m_size, newIndices.ptr()+id+1);
+          }
+          std::swap(m_values,newValues.ptr());
+          std::swap(m_indices,newIndices.ptr());
+        }
+        else if(m_size>id)
         {
-          m_indices[j] = m_indices[j-1];
-          m_values[j] = m_values[j-1];
+          internal::smart_memmove(m_values +id, m_values +m_size, m_values +id+1);
+          internal::smart_memmove(m_indices+id, m_indices+m_size, m_indices+id+1);
         }
-        m_indices[id] = key;
+        m_size++;
+        m_indices[id] = internal::convert_index<StorageIndex>(key);
         m_values[id] = defaultValue;
       }
       return m_values[id];
@@ -186,9 +209,9 @@ class CompressedStorage
 
     void prune(const Scalar& reference, const RealScalar& epsilon = NumTraits<RealScalar>::dummy_precision())
     {
-      size_t k = 0;
-      size_t n = size();
-      for (size_t i=0; i<n; ++i)
+      Index k = 0;
+      Index n = size();
+      for (Index i=0; i<n; ++i)
       {
         if (!internal::isMuchSmallerThan(value(i), reference, epsilon))
         {
@@ -202,27 +225,29 @@ class CompressedStorage
 
   protected:
 
-    inline void reallocate(size_t size)
+    inline void reallocate(Index size)
     {
-      Scalar* newValues  = new Scalar[size];
-      Index* newIndices = new Index[size];
-      size_t copySize = (std::min)(size, m_size);
-      // copy
-      internal::smart_copy(m_values, m_values+copySize, newValues);
-      internal::smart_copy(m_indices, m_indices+copySize, newIndices);
-      // delete old stuff
-      delete[] m_values;
-      delete[] m_indices;
-      m_values = newValues;
-      m_indices = newIndices;
+      #ifdef EIGEN_SPARSE_COMPRESSED_STORAGE_REALLOCATE_PLUGIN
+        EIGEN_SPARSE_COMPRESSED_STORAGE_REALLOCATE_PLUGIN
+      #endif
+      eigen_internal_assert(size!=m_allocatedSize);
+      internal::scoped_array<Scalar> newValues(size);
+      internal::scoped_array<StorageIndex> newIndices(size);
+      Index copySize = (std::min)(size, m_size);
+      if (copySize>0) {
+        internal::smart_copy(m_values, m_values+copySize, newValues.ptr());
+        internal::smart_copy(m_indices, m_indices+copySize, newIndices.ptr());
+      }
+      std::swap(m_values,newValues.ptr());
+      std::swap(m_indices,newIndices.ptr());
       m_allocatedSize = size;
     }
 
   protected:
     Scalar* m_values;
-    Index* m_indices;
-    size_t m_size;
-    size_t m_allocatedSize;
+    StorageIndex* m_indices;
+    Index m_size;
+    Index m_allocatedSize;
 
 };
 
diff --git a/extern/Eigen3/Eigen/src/SparseCore/ConservativeSparseSparseProduct.h b/extern/Eigen3/Eigen/src/SparseCore/ConservativeSparseSparseProduct.h
index 5c320e2d2db..9db119b67f9 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/ConservativeSparseSparseProduct.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/ConservativeSparseSparseProduct.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -15,27 +15,33 @@ namespace Eigen {
 namespace internal {
 
 template<typename Lhs, typename Rhs, typename ResultType>
-static void conservative_sparse_sparse_product_impl(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+static void conservative_sparse_sparse_product_impl(const Lhs& lhs, const Rhs& rhs, ResultType& res, bool sortedInsertion = false)
 {
-  typedef typename remove_all<Lhs>::type::Scalar Scalar;
-  typedef typename remove_all<Lhs>::type::Index Index;
+  typedef typename remove_all<Lhs>::type::Scalar LhsScalar;
+  typedef typename remove_all<Rhs>::type::Scalar RhsScalar;
+  typedef typename remove_all<ResultType>::type::Scalar ResScalar;
 
   // make sure to call innerSize/outerSize since we fake the storage order.
   Index rows = lhs.innerSize();
   Index cols = rhs.outerSize();
   eigen_assert(lhs.outerSize() == rhs.innerSize());
-
-  std::vector<bool> mask(rows,false);
-  Matrix<Scalar,Dynamic,1> values(rows);
-  Matrix<Index,Dynamic,1>  indices(rows);
-
+  
+  ei_declare_aligned_stack_constructed_variable(bool,   mask,     rows, 0);
+  ei_declare_aligned_stack_constructed_variable(ResScalar, values,   rows, 0);
+  ei_declare_aligned_stack_constructed_variable(Index,  indices,  rows, 0);
+  
+  std::memset(mask,0,sizeof(bool)*rows);
+
+  evaluator<Lhs> lhsEval(lhs);
+  evaluator<Rhs> rhsEval(rhs);
+  
   // estimate the number of non zero entries
   // given a rhs column containing Y non zeros, we assume that the respective Y columns
   // of the lhs differs in average of one non zeros, thus the number of non zeros for
   // the product of a rhs column with the lhs is X+Y where X is the average number of non zero
   // per column of the lhs.
   // Therefore, we have nnz(lhs*rhs) = nnz(lhs) + nnz(rhs)
-  Index estimated_nnz_prod = lhs.nonZeros() + rhs.nonZeros();
+  Index estimated_nnz_prod = lhsEval.nonZerosEstimate() + rhsEval.nonZerosEstimate();
 
   res.setZero();
   res.reserve(Index(estimated_nnz_prod));
@@ -45,14 +51,14 @@ static void conservative_sparse_sparse_product_impl(const Lhs& lhs, const Rhs& r
 
     res.startVec(j);
     Index nnz = 0;
-    for (typename Rhs::InnerIterator rhsIt(rhs, j); rhsIt; ++rhsIt)
+    for (typename evaluator<Rhs>::InnerIterator rhsIt(rhsEval, j); rhsIt; ++rhsIt)
     {
-      Scalar y = rhsIt.value();
+      RhsScalar y = rhsIt.value();
       Index k = rhsIt.index();
-      for (typename Lhs::InnerIterator lhsIt(lhs, k); lhsIt; ++lhsIt)
+      for (typename evaluator<Lhs>::InnerIterator lhsIt(lhsEval, k); lhsIt; ++lhsIt)
       {
         Index i = lhsIt.index();
-        Scalar x = lhsIt.value();
+        LhsScalar x = lhsIt.value();
         if(!mask[i])
         {
           mask[i] = true;
@@ -64,53 +70,51 @@ static void conservative_sparse_sparse_product_impl(const Lhs& lhs, const Rhs& r
           values[i] += x * y;
       }
     }
-
-    // unordered insertion
-    for(Index k=0; k<nnz; ++k)
-    {
-      Index i = indices[k];
-      res.insertBackByOuterInnerUnordered(j,i) = values[i];
-      mask[i] = false;
-    }
-
-#if 0
-    // alternative ordered insertion code:
-
-    Index t200 = rows/(log2(200)*1.39);
-    Index t = (rows*100)/139;
-
-    // FIXME reserve nnz non zeros
-    // FIXME implement fast sort algorithms for very small nnz
-    // if the result is sparse enough => use a quick sort
-    // otherwise => loop through the entire vector
-    // In order to avoid to perform an expensive log2 when the
-    // result is clearly very sparse we use a linear bound up to 200.
-    //if((nnz<200 && nnz<t200) || nnz * log2(nnz) < t)
-    //res.startVec(j);
-    if(true)
+    if(!sortedInsertion)
     {
-      if(nnz>1) std::sort(indices.data(),indices.data()+nnz);
+      // unordered insertion
       for(Index k=0; k<nnz; ++k)
       {
         Index i = indices[k];
-        res.insertBackByOuterInner(j,i) = values[i];
+        res.insertBackByOuterInnerUnordered(j,i) = values[i];
         mask[i] = false;
       }
     }
     else
     {
-      // dense path
-      for(Index i=0; i<rows; ++i)
+      // alternative ordered insertion code:
+      const Index t200 = rows/11; // 11 == (log2(200)*1.39)
+      const Index t = (rows*100)/139;
+
+      // FIXME reserve nnz non zeros
+      // FIXME implement faster sorting algorithms for very small nnz
+      // if the result is sparse enough => use a quick sort
+      // otherwise => loop through the entire vector
+      // In order to avoid to perform an expensive log2 when the
+      // result is clearly very sparse we use a linear bound up to 200.
+      if((nnz<200 && nnz<t200) || nnz * numext::log2(int(nnz)) < t)
       {
-        if(mask[i])
+        if(nnz>1) std::sort(indices,indices+nnz);
+        for(Index k=0; k<nnz; ++k)
         {
-          mask[i] = false;
+          Index i = indices[k];
           res.insertBackByOuterInner(j,i) = values[i];
+          mask[i] = false;
+        }
+      }
+      else
+      {
+        // dense path
+        for(Index i=0; i<rows; ++i)
+        {
+          if(mask[i])
+          {
+            mask[i] = false;
+            res.insertBackByOuterInner(j,i) = values[i];
+          }
         }
       }
     }
-#endif
-
   }
   res.finalize();
 }
@@ -134,13 +138,28 @@ struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,C
 
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
-    ColMajorMatrix resCol(lhs.rows(),rhs.cols());
-    internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrix>(lhs, rhs, resCol);
-    // sort the non zeros:
-    RowMajorMatrix resRow(resCol);
-    res = resRow;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorMatrixAux;
+    typedef typename sparse_eval<ColMajorMatrixAux,ResultType::RowsAtCompileTime,ResultType::ColsAtCompileTime,ColMajorMatrixAux::Flags>::type ColMajorMatrix;
+    
+    // If the result is tall and thin (in the extreme case a column vector)
+    // then it is faster to sort the coefficients inplace instead of transposing twice.
+    // FIXME, the following heuristic is probably not very good.
+    if(lhs.rows()>rhs.cols())
+    {
+      ColMajorMatrix resCol(lhs.rows(),rhs.cols());
+      // perform sorted insertion
+      internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrix>(lhs, rhs, resCol, true);
+      res = resCol.markAsRValue();
+    }
+    else
+    {
+      ColMajorMatrixAux resCol(lhs.rows(),rhs.cols());
+      // ressort to transpose to sort the entries
+      internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrixAux>(lhs, rhs, resCol, false);
+      RowMajorMatrix resRow(resCol);
+      res = resRow.markAsRValue();
+    }
   }
 };
 
@@ -149,11 +168,12 @@ struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,C
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-     typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
-     RowMajorMatrix rhsRow = rhs;
-     RowMajorMatrix resRow(lhs.rows(), rhs.cols());
-     internal::conservative_sparse_sparse_product_impl<RowMajorMatrix,Lhs,RowMajorMatrix>(rhsRow, lhs, resRow);
-     res = resRow;
+    typedef SparseMatrix<typename Rhs::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorRhs;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorRes;
+    RowMajorRhs rhsRow = rhs;
+    RowMajorRes resRow(lhs.rows(), rhs.cols());
+    internal::conservative_sparse_sparse_product_impl<RowMajorRhs,Lhs,RowMajorRes>(rhsRow, lhs, resRow);
+    res = resRow;
   }
 };
 
@@ -162,10 +182,11 @@ struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,R
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
-    RowMajorMatrix lhsRow = lhs;
-    RowMajorMatrix resRow(lhs.rows(), rhs.cols());
-    internal::conservative_sparse_sparse_product_impl<Rhs,RowMajorMatrix,RowMajorMatrix>(rhs, lhsRow, resRow);
+    typedef SparseMatrix<typename Lhs::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorLhs;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorRes;
+    RowMajorLhs lhsRow = lhs;
+    RowMajorRes resRow(lhs.rows(), rhs.cols());
+    internal::conservative_sparse_sparse_product_impl<Rhs,RowMajorLhs,RowMajorRes>(rhs, lhsRow, resRow);
     res = resRow;
   }
 };
@@ -175,7 +196,7 @@ struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,R
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorMatrix;
     RowMajorMatrix resRow(lhs.rows(), rhs.cols());
     internal::conservative_sparse_sparse_product_impl<Rhs,Lhs,RowMajorMatrix>(rhs, lhs, resRow);
     res = resRow;
@@ -190,7 +211,7 @@ struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,C
 
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorMatrix;
     ColMajorMatrix resCol(lhs.rows(), rhs.cols());
     internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrix>(lhs, rhs, resCol);
     res = resCol;
@@ -202,10 +223,11 @@ struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,C
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
-    ColMajorMatrix lhsCol = lhs;
-    ColMajorMatrix resCol(lhs.rows(), rhs.cols());
-    internal::conservative_sparse_sparse_product_impl<ColMajorMatrix,Rhs,ColMajorMatrix>(lhsCol, rhs, resCol);
+    typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorLhs;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorRes;
+    ColMajorLhs lhsCol = lhs;
+    ColMajorRes resCol(lhs.rows(), rhs.cols());
+    internal::conservative_sparse_sparse_product_impl<ColMajorLhs,Rhs,ColMajorRes>(lhsCol, rhs, resCol);
     res = resCol;
   }
 };
@@ -215,10 +237,11 @@ struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,R
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
-    ColMajorMatrix rhsCol = rhs;
-    ColMajorMatrix resCol(lhs.rows(), rhs.cols());
-    internal::conservative_sparse_sparse_product_impl<Lhs,ColMajorMatrix,ColMajorMatrix>(lhs, rhsCol, resCol);
+    typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorRhs;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorRes;
+    ColMajorRhs rhsCol = rhs;
+    ColMajorRes resCol(lhs.rows(), rhs.cols());
+    internal::conservative_sparse_sparse_product_impl<Lhs,ColMajorRhs,ColMajorRes>(lhs, rhsCol, resCol);
     res = resCol;
   }
 };
@@ -228,8 +251,8 @@ struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,R
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorMatrix;
     RowMajorMatrix resRow(lhs.rows(),rhs.cols());
     internal::conservative_sparse_sparse_product_impl<Rhs,Lhs,RowMajorMatrix>(rhs, lhs, resRow);
     // sort the non zeros:
@@ -240,6 +263,90 @@ struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,R
 
 } // end namespace internal
 
+
+namespace internal {
+
+template<typename Lhs, typename Rhs, typename ResultType>
+static void sparse_sparse_to_dense_product_impl(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+{
+  typedef typename remove_all<Lhs>::type::Scalar LhsScalar;
+  typedef typename remove_all<Rhs>::type::Scalar RhsScalar;
+  Index cols = rhs.outerSize();
+  eigen_assert(lhs.outerSize() == rhs.innerSize());
+
+  evaluator<Lhs> lhsEval(lhs);
+  evaluator<Rhs> rhsEval(rhs);
+
+  for (Index j=0; j<cols; ++j)
+  {
+    for (typename evaluator<Rhs>::InnerIterator rhsIt(rhsEval, j); rhsIt; ++rhsIt)
+    {
+      RhsScalar y = rhsIt.value();
+      Index k = rhsIt.index();
+      for (typename evaluator<Lhs>::InnerIterator lhsIt(lhsEval, k); lhsIt; ++lhsIt)
+      {
+        Index i = lhsIt.index();
+        LhsScalar x = lhsIt.value();
+        res.coeffRef(i,j) += x * y;
+      }
+    }
+  }
+}
+
+
+} // end namespace internal
+
+namespace internal {
+
+template<typename Lhs, typename Rhs, typename ResultType,
+  int LhsStorageOrder = (traits<Lhs>::Flags&RowMajorBit) ? RowMajor : ColMajor,
+  int RhsStorageOrder = (traits<Rhs>::Flags&RowMajorBit) ? RowMajor : ColMajor>
+struct sparse_sparse_to_dense_product_selector;
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_to_dense_product_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor>
+{
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+  {
+    internal::sparse_sparse_to_dense_product_impl<Lhs,Rhs,ResultType>(lhs, rhs, res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_to_dense_product_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor>
+{
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+  {
+    typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorLhs;
+    ColMajorLhs lhsCol(lhs);
+    internal::sparse_sparse_to_dense_product_impl<ColMajorLhs,Rhs,ResultType>(lhsCol, rhs, res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_to_dense_product_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor>
+{
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+  {
+    typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorRhs;
+    ColMajorRhs rhsCol(rhs);
+    internal::sparse_sparse_to_dense_product_impl<Lhs,ColMajorRhs,ResultType>(lhs, rhsCol, res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_to_dense_product_selector<Lhs,Rhs,ResultType,RowMajor,RowMajor>
+{
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+  {
+    Transpose<ResultType> trRes(res);
+    internal::sparse_sparse_to_dense_product_impl<Rhs,Lhs,Transpose<ResultType> >(rhs, lhs, trRes);
+  }
+};
+
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_CONSERVATIVESPARSESPARSEPRODUCT_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/MappedSparseMatrix.h b/extern/Eigen3/Eigen/src/SparseCore/MappedSparseMatrix.h
index ab1a266a905..67718c85be9 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/MappedSparseMatrix.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/MappedSparseMatrix.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -10,9 +10,10 @@
 #ifndef EIGEN_MAPPED_SPARSEMATRIX_H
 #define EIGEN_MAPPED_SPARSEMATRIX_H
 
-namespace Eigen { 
+namespace Eigen {
 
-/** \class MappedSparseMatrix
+/** \deprecated Use Map<SparseMatrix<> >
+  * \class MappedSparseMatrix
   *
   * \brief Sparse matrix
   *
@@ -22,160 +23,45 @@ namespace Eigen {
   *
   */
 namespace internal {
-template<typename _Scalar, int _Flags, typename _Index>
-struct traits<MappedSparseMatrix<_Scalar, _Flags, _Index> > : traits<SparseMatrix<_Scalar, _Flags, _Index> >
+template<typename _Scalar, int _Flags, typename _StorageIndex>
+struct traits<MappedSparseMatrix<_Scalar, _Flags, _StorageIndex> > : traits<SparseMatrix<_Scalar, _Flags, _StorageIndex> >
 {};
-}
+} // end namespace internal
 
-template<typename _Scalar, int _Flags, typename _Index>
+template<typename _Scalar, int _Flags, typename _StorageIndex>
 class MappedSparseMatrix
-  : public SparseMatrixBase<MappedSparseMatrix<_Scalar, _Flags, _Index> >
+  : public Map<SparseMatrix<_Scalar, _Flags, _StorageIndex> >
 {
-  public:
-    EIGEN_SPARSE_PUBLIC_INTERFACE(MappedSparseMatrix)
-    enum { IsRowMajor = Base::IsRowMajor };
-
-  protected:
-
-    Index   m_outerSize;
-    Index   m_innerSize;
-    Index   m_nnz;
-    Index*  m_outerIndex;
-    Index*  m_innerIndices;
-    Scalar* m_values;
+    typedef Map<SparseMatrix<_Scalar, _Flags, _StorageIndex> > Base;
 
   public:
-
-    inline Index rows() const { return IsRowMajor ? m_outerSize : m_innerSize; }
-    inline Index cols() const { return IsRowMajor ? m_innerSize : m_outerSize; }
-    inline Index innerSize() const { return m_innerSize; }
-    inline Index outerSize() const { return m_outerSize; }
     
-    bool isCompressed() const { return true; }
-
-    //----------------------------------------
-    // direct access interface
-    inline const Scalar* valuePtr() const { return m_values; }
-    inline Scalar* valuePtr() { return m_values; }
-
-    inline const Index* innerIndexPtr() const { return m_innerIndices; }
-    inline Index* innerIndexPtr() { return m_innerIndices; }
-
-    inline const Index* outerIndexPtr() const { return m_outerIndex; }
-    inline Index* outerIndexPtr() { return m_outerIndex; }
-    //----------------------------------------
-
-    inline Scalar coeff(Index row, Index col) const
-    {
-      const Index outer = IsRowMajor ? row : col;
-      const Index inner = IsRowMajor ? col : row;
+    typedef typename Base::StorageIndex StorageIndex;
+    typedef typename Base::Scalar Scalar;
 
-      Index start = m_outerIndex[outer];
-      Index end = m_outerIndex[outer+1];
-      if (start==end)
-        return Scalar(0);
-      else if (end>0 && inner==m_innerIndices[end-1])
-        return m_values[end-1];
-      // ^^  optimization: let's first check if it is the last coefficient
-      // (very common in high level algorithms)
-
-      const Index* r = std::lower_bound(&m_innerIndices[start],&m_innerIndices[end-1],inner);
-      const Index id = r-&m_innerIndices[0];
-      return ((*r==inner) && (id<end)) ? m_values[id] : Scalar(0);
-    }
-
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      const Index outer = IsRowMajor ? row : col;
-      const Index inner = IsRowMajor ? col : row;
-
-      Index start = m_outerIndex[outer];
-      Index end = m_outerIndex[outer+1];
-      eigen_assert(end>=start && "you probably called coeffRef on a non finalized matrix");
-      eigen_assert(end>start && "coeffRef cannot be called on a zero coefficient");
-      Index* r = std::lower_bound(&m_innerIndices[start],&m_innerIndices[end],inner);
-      const Index id = r-&m_innerIndices[0];
-      eigen_assert((*r==inner) && (id<end) && "coeffRef cannot be called on a zero coefficient");
-      return m_values[id];
-    }
-
-    class InnerIterator;
-    class ReverseInnerIterator;
-
-    /** \returns the number of non zero coefficients */
-    inline Index nonZeros() const  { return m_nnz; }
-
-    inline MappedSparseMatrix(Index rows, Index cols, Index nnz, Index* outerIndexPtr, Index* innerIndexPtr, Scalar* valuePtr)
-      : m_outerSize(IsRowMajor?rows:cols), m_innerSize(IsRowMajor?cols:rows), m_nnz(nnz), m_outerIndex(outerIndexPtr),
-        m_innerIndices(innerIndexPtr), m_values(valuePtr)
+    inline MappedSparseMatrix(Index rows, Index cols, Index nnz, StorageIndex* outerIndexPtr, StorageIndex* innerIndexPtr, Scalar* valuePtr, StorageIndex* innerNonZeroPtr = 0)
+      : Base(rows, cols, nnz, outerIndexPtr, innerIndexPtr, valuePtr, innerNonZeroPtr)
     {}
 
     /** Empty destructor */
     inline ~MappedSparseMatrix() {}
 };
 
-template<typename Scalar, int _Flags, typename _Index>
-class MappedSparseMatrix<Scalar,_Flags,_Index>::InnerIterator
-{
-  public:
-    InnerIterator(const MappedSparseMatrix& mat, Index outer)
-      : m_matrix(mat),
-        m_outer(outer),
-        m_id(mat.outerIndexPtr()[outer]),
-        m_start(m_id),
-        m_end(mat.outerIndexPtr()[outer+1])
-    {}
-
-    inline InnerIterator& operator++() { m_id++; return *this; }
-
-    inline Scalar value() const { return m_matrix.valuePtr()[m_id]; }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_matrix.valuePtr()[m_id]); }
-
-    inline Index index() const { return m_matrix.innerIndexPtr()[m_id]; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    inline operator bool() const { return (m_id < m_end) && (m_id>=m_start); }
-
-  protected:
-    const MappedSparseMatrix& m_matrix;
-    const Index m_outer;
-    Index m_id;
-    const Index m_start;
-    const Index m_end;
-};
+namespace internal {
 
-template<typename Scalar, int _Flags, typename _Index>
-class MappedSparseMatrix<Scalar,_Flags,_Index>::ReverseInnerIterator
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct evaluator<MappedSparseMatrix<_Scalar,_Options,_StorageIndex> >
+  : evaluator<SparseCompressedBase<MappedSparseMatrix<_Scalar,_Options,_StorageIndex> > >
 {
-  public:
-    ReverseInnerIterator(const MappedSparseMatrix& mat, Index outer)
-      : m_matrix(mat),
-        m_outer(outer),
-        m_id(mat.outerIndexPtr()[outer+1]),
-        m_start(mat.outerIndexPtr()[outer]),
-        m_end(m_id)
-    {}
-
-    inline ReverseInnerIterator& operator--() { m_id--; return *this; }
-
-    inline Scalar value() const { return m_matrix.valuePtr()[m_id-1]; }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_matrix.valuePtr()[m_id-1]); }
-
-    inline Index index() const { return m_matrix.innerIndexPtr()[m_id-1]; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    inline operator bool() const { return (m_id <= m_end) && (m_id>m_start); }
-
-  protected:
-    const MappedSparseMatrix& m_matrix;
-    const Index m_outer;
-    Index m_id;
-    const Index m_start;
-    const Index m_end;
+  typedef MappedSparseMatrix<_Scalar,_Options,_StorageIndex> XprType;
+  typedef evaluator<SparseCompressedBase<XprType> > Base;
+  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
 };
 
+}
+
 } // end namespace Eigen
 
 #endif // EIGEN_MAPPED_SPARSEMATRIX_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseAssign.h b/extern/Eigen3/Eigen/src/SparseCore/SparseAssign.h
index e69de29bb2d..18352a847b2 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseAssign.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseAssign.h
@@ -0,0 +1,216 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSEASSIGN_H
+#define EIGEN_SPARSEASSIGN_H
+
+namespace Eigen { 
+
+template<typename Derived>    
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
+{
+  internal::call_assignment_no_alias(derived(), other.derived());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
+{
+  // TODO use the evaluator mechanism
+  other.evalTo(derived());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+inline Derived& SparseMatrixBase<Derived>::operator=(const SparseMatrixBase<OtherDerived>& other)
+{
+  // by default sparse evaluation do not alias, so we can safely bypass the generic call_assignment routine
+  internal::Assignment<Derived,OtherDerived,internal::assign_op<Scalar,typename OtherDerived::Scalar> >
+          ::run(derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+inline Derived& SparseMatrixBase<Derived>::operator=(const Derived& other)
+{
+  internal::call_assignment_no_alias(derived(), other.derived());
+  return derived();
+}
+
+namespace internal {
+
+template<>
+struct storage_kind_to_evaluator_kind<Sparse> {
+  typedef IteratorBased Kind;
+};
+
+template<>
+struct storage_kind_to_shape<Sparse> {
+  typedef SparseShape Shape;
+};
+
+struct Sparse2Sparse {};
+struct Sparse2Dense  {};
+
+template<> struct AssignmentKind<SparseShape, SparseShape>           { typedef Sparse2Sparse Kind; };
+template<> struct AssignmentKind<SparseShape, SparseTriangularShape> { typedef Sparse2Sparse Kind; };
+template<> struct AssignmentKind<DenseShape,  SparseShape>           { typedef Sparse2Dense  Kind; };
+template<> struct AssignmentKind<DenseShape,  SparseTriangularShape> { typedef Sparse2Dense  Kind; };
+
+
+template<typename DstXprType, typename SrcXprType>
+void assign_sparse_to_sparse(DstXprType &dst, const SrcXprType &src)
+{
+  typedef typename DstXprType::Scalar Scalar;
+  typedef internal::evaluator<DstXprType> DstEvaluatorType;
+  typedef internal::evaluator<SrcXprType> SrcEvaluatorType;
+
+  SrcEvaluatorType srcEvaluator(src);
+
+  const bool transpose = (DstEvaluatorType::Flags & RowMajorBit) != (SrcEvaluatorType::Flags & RowMajorBit);
+  const Index outerEvaluationSize = (SrcEvaluatorType::Flags&RowMajorBit) ? src.rows() : src.cols();
+  if ((!transpose) && src.isRValue())
+  {
+    // eval without temporary
+    dst.resize(src.rows(), src.cols());
+    dst.setZero();
+    dst.reserve((std::max)(src.rows(),src.cols())*2);
+    for (Index j=0; j<outerEvaluationSize; ++j)
+    {
+      dst.startVec(j);
+      for (typename SrcEvaluatorType::InnerIterator it(srcEvaluator, j); it; ++it)
+      {
+        Scalar v = it.value();
+        dst.insertBackByOuterInner(j,it.index()) = v;
+      }
+    }
+    dst.finalize();
+  }
+  else
+  {
+    // eval through a temporary
+    eigen_assert(( ((internal::traits<DstXprType>::SupportedAccessPatterns & OuterRandomAccessPattern)==OuterRandomAccessPattern) ||
+              (!((DstEvaluatorType::Flags & RowMajorBit) != (SrcEvaluatorType::Flags & RowMajorBit)))) &&
+              "the transpose operation is supposed to be handled in SparseMatrix::operator=");
+
+    enum { Flip = (DstEvaluatorType::Flags & RowMajorBit) != (SrcEvaluatorType::Flags & RowMajorBit) };
+
+    
+    DstXprType temp(src.rows(), src.cols());
+
+    temp.reserve((std::max)(src.rows(),src.cols())*2);
+    for (Index j=0; j<outerEvaluationSize; ++j)
+    {
+      temp.startVec(j);
+      for (typename SrcEvaluatorType::InnerIterator it(srcEvaluator, j); it; ++it)
+      {
+        Scalar v = it.value();
+        temp.insertBackByOuterInner(Flip?it.index():j,Flip?j:it.index()) = v;
+      }
+    }
+    temp.finalize();
+
+    dst = temp.markAsRValue();
+  }
+}
+
+// Generic Sparse to Sparse assignment
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Sparse2Sparse>
+{
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    assign_sparse_to_sparse(dst.derived(), src.derived());
+  }
+};
+
+// Generic Sparse to Dense assignment
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Sparse2Dense>
+{
+  static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+    if(internal::is_same<Functor,internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> >::value)
+      dst.setZero();
+    
+    internal::evaluator<SrcXprType> srcEval(src);
+    resize_if_allowed(dst, src, func);
+    internal::evaluator<DstXprType> dstEval(dst);
+    
+    const Index outerEvaluationSize = (internal::evaluator<SrcXprType>::Flags&RowMajorBit) ? src.rows() : src.cols();
+    for (Index j=0; j<outerEvaluationSize; ++j)
+      for (typename internal::evaluator<SrcXprType>::InnerIterator i(srcEval,j); i; ++i)
+        func.assignCoeff(dstEval.coeffRef(i.row(),i.col()), i.value());
+  }
+};
+
+// Specialization for "dst = dec.solve(rhs)"
+// NOTE we need to specialize it for Sparse2Sparse to avoid ambiguous specialization error
+template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
+struct Assignment<DstXprType, Solve<DecType,RhsType>, internal::assign_op<Scalar,Scalar>, Sparse2Sparse>
+{
+  typedef Solve<DecType,RhsType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    src.dec()._solve_impl(src.rhs(), dst);
+  }
+};
+
+struct Diagonal2Sparse {};
+
+template<> struct AssignmentKind<SparseShape,DiagonalShape> { typedef Diagonal2Sparse Kind; };
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Sparse>
+{
+  typedef typename DstXprType::StorageIndex StorageIndex;
+  typedef typename DstXprType::Scalar Scalar;
+  typedef Array<StorageIndex,Dynamic,1> ArrayXI;
+  typedef Array<Scalar,Dynamic,1> ArrayXS;
+  template<int Options>
+  static void run(SparseMatrix<Scalar,Options,StorageIndex> &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    Index size = src.diagonal().size();
+    dst.makeCompressed();
+    dst.resizeNonZeros(size);
+    Map<ArrayXI>(dst.innerIndexPtr(), size).setLinSpaced(0,StorageIndex(size)-1);
+    Map<ArrayXI>(dst.outerIndexPtr(), size+1).setLinSpaced(0,StorageIndex(size));
+    Map<ArrayXS>(dst.valuePtr(), size) = src.diagonal();
+  }
+  
+  template<typename DstDerived>
+  static void run(SparseMatrixBase<DstDerived> &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    dst.diagonal() = src.diagonal();
+  }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() += src.diagonal(); }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() -= src.diagonal(); }
+};
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSEASSIGN_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseBlock.h b/extern/Eigen3/Eigen/src/SparseCore/SparseBlock.h
index 0c90bafbeab..511e92b2f9b 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseBlock.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseBlock.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -10,8 +10,9 @@
 #ifndef EIGEN_SPARSE_BLOCK_H
 #define EIGEN_SPARSE_BLOCK_H
 
-namespace Eigen { 
+namespace Eigen {
 
+// Subset of columns or rows
 template<typename XprType, int BlockRows, int BlockCols>
 class BlockImpl<XprType,BlockRows,BlockCols,true,Sparse>
   : public SparseMatrixBase<Block<XprType,BlockRows,BlockCols,true> >
@@ -22,185 +23,189 @@ public:
     enum { IsRowMajor = internal::traits<BlockType>::IsRowMajor };
 protected:
     enum { OuterSize = IsRowMajor ? BlockRows : BlockCols };
+    typedef SparseMatrixBase<BlockType> Base;
+    using Base::convert_index;
 public:
     EIGEN_SPARSE_PUBLIC_INTERFACE(BlockType)
-    
-    class InnerIterator: public XprType::InnerIterator
-    {
-        typedef typename BlockImpl::Index Index;
-      public:
-        inline InnerIterator(const BlockType& xpr, Index outer)
-          : XprType::InnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
-    class ReverseInnerIterator: public XprType::ReverseInnerIterator
-    {
-        typedef typename BlockImpl::Index Index;
-      public:
-        inline ReverseInnerIterator(const BlockType& xpr, Index outer)
-          : XprType::ReverseInnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
 
-    inline BlockImpl(const XprType& xpr, int i)
-      : m_matrix(xpr), m_outerStart(i), m_outerSize(OuterSize)
+    inline BlockImpl(XprType& xpr, Index i)
+      : m_matrix(xpr), m_outerStart(convert_index(i)), m_outerSize(OuterSize)
     {}
 
-    inline BlockImpl(const XprType& xpr, int startRow, int startCol, int blockRows, int blockCols)
-      : m_matrix(xpr), m_outerStart(IsRowMajor ? startRow : startCol), m_outerSize(IsRowMajor ? blockRows : blockCols)
+    inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+      : m_matrix(xpr), m_outerStart(convert_index(IsRowMajor ? startRow : startCol)), m_outerSize(convert_index(IsRowMajor ? blockRows : blockCols))
     {}
-    
-    inline const Scalar coeff(int row, int col) const
+
+    EIGEN_STRONG_INLINE Index rows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
+    EIGEN_STRONG_INLINE Index cols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
+
+    Index nonZeros() const
     {
-      return m_matrix.coeff(row + IsRowMajor ? m_outerStart : 0, col +IsRowMajor ? 0 :  m_outerStart);
+      typedef internal::evaluator<XprType> EvaluatorType;
+      EvaluatorType matEval(m_matrix);
+      Index nnz = 0;
+      Index end = m_outerStart + m_outerSize.value();
+      for(Index j=m_outerStart; j<end; ++j)
+        for(typename EvaluatorType::InnerIterator it(matEval, j); it; ++it)
+          ++nnz;
+      return nnz;
     }
-    
-    inline const Scalar coeff(int index) const
+
+    inline const Scalar coeff(Index row, Index col) const
+    {
+      return m_matrix.coeff(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
+    }
+
+    inline const Scalar coeff(Index index) const
     {
       return m_matrix.coeff(IsRowMajor ? m_outerStart : index, IsRowMajor ? index :  m_outerStart);
     }
 
-    EIGEN_STRONG_INLINE Index rows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
+    inline const XprType& nestedExpression() const { return m_matrix; }
+    inline XprType& nestedExpression() { return m_matrix; }
+    Index startRow() const { return IsRowMajor ? m_outerStart : 0; }
+    Index startCol() const { return IsRowMajor ? 0 : m_outerStart; }
+    Index blockRows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
+    Index blockCols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
 
   protected:
 
-    typename XprType::Nested m_matrix;
+    typename internal::ref_selector<XprType>::non_const_type m_matrix;
     Index m_outerStart;
     const internal::variable_if_dynamic<Index, OuterSize> m_outerSize;
 
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
-  private:
-    Index nonZeros() const;
+  protected:
+    // Disable assignment with clear error message.
+    // Note that simply removing operator= yields compilation errors with ICC+MSVC
+    template<typename T>
+    BlockImpl& operator=(const T&)
+    {
+      EIGEN_STATIC_ASSERT(sizeof(T)==0, THIS_SPARSE_BLOCK_SUBEXPRESSION_IS_READ_ONLY);
+      return *this;
+    }
 };
 
 
 /***************************************************************************
-* specialisation for SparseMatrix
+* specialization for SparseMatrix
 ***************************************************************************/
 
-template<typename _Scalar, int _Options, typename _Index, int BlockRows, int BlockCols>
-class BlockImpl<SparseMatrix<_Scalar, _Options, _Index>,BlockRows,BlockCols,true,Sparse>
-  : public SparseMatrixBase<Block<SparseMatrix<_Scalar, _Options, _Index>,BlockRows,BlockCols,true> >
+namespace internal {
+
+template<typename SparseMatrixType, int BlockRows, int BlockCols>
+class sparse_matrix_block_impl
+  : public SparseCompressedBase<Block<SparseMatrixType,BlockRows,BlockCols,true> >
 {
-    typedef SparseMatrix<_Scalar, _Options, _Index> SparseMatrixType;
     typedef typename internal::remove_all<typename SparseMatrixType::Nested>::type _MatrixTypeNested;
     typedef Block<SparseMatrixType, BlockRows, BlockCols, true> BlockType;
-    typedef Block<const SparseMatrixType, BlockRows, BlockCols, true> ConstBlockType;
+    typedef SparseCompressedBase<Block<SparseMatrixType,BlockRows,BlockCols,true> > Base;
+    using Base::convert_index;
 public:
     enum { IsRowMajor = internal::traits<BlockType>::IsRowMajor };
     EIGEN_SPARSE_PUBLIC_INTERFACE(BlockType)
 protected:
+    typedef typename Base::IndexVector IndexVector;
     enum { OuterSize = IsRowMajor ? BlockRows : BlockCols };
 public:
-    
-    class InnerIterator: public SparseMatrixType::InnerIterator
-    {
-      public:
-        inline InnerIterator(const BlockType& xpr, Index outer)
-          : SparseMatrixType::InnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
-    class ReverseInnerIterator: public SparseMatrixType::ReverseInnerIterator
-    {
-      public:
-        inline ReverseInnerIterator(const BlockType& xpr, Index outer)
-          : SparseMatrixType::ReverseInnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
 
-    inline BlockImpl(const SparseMatrixType& xpr, int i)
-      : m_matrix(xpr), m_outerStart(i), m_outerSize(OuterSize)
+    inline sparse_matrix_block_impl(SparseMatrixType& xpr, Index i)
+      : m_matrix(xpr), m_outerStart(convert_index(i)), m_outerSize(OuterSize)
     {}
 
-    inline BlockImpl(const SparseMatrixType& xpr, int startRow, int startCol, int blockRows, int blockCols)
-      : m_matrix(xpr), m_outerStart(IsRowMajor ? startRow : startCol), m_outerSize(IsRowMajor ? blockRows : blockCols)
+    inline sparse_matrix_block_impl(SparseMatrixType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+      : m_matrix(xpr), m_outerStart(convert_index(IsRowMajor ? startRow : startCol)), m_outerSize(convert_index(IsRowMajor ? blockRows : blockCols))
     {}
 
     template<typename OtherDerived>
     inline BlockType& operator=(const SparseMatrixBase<OtherDerived>& other)
     {
       typedef typename internal::remove_all<typename SparseMatrixType::Nested>::type _NestedMatrixType;
-      _NestedMatrixType& matrix = const_cast<_NestedMatrixType&>(m_matrix);;
-      // This assignement is slow if this vector set is not empty
+      _NestedMatrixType& matrix = m_matrix;
+      // This assignment is slow if this vector set is not empty
       // and/or it is not at the end of the nonzeros of the underlying matrix.
 
       // 1 - eval to a temporary to avoid transposition and/or aliasing issues
-      SparseMatrix<Scalar, IsRowMajor ? RowMajor : ColMajor, Index> tmp(other);
+      Ref<const SparseMatrix<Scalar, IsRowMajor ? RowMajor : ColMajor, StorageIndex> > tmp(other.derived());
+      eigen_internal_assert(tmp.outerSize()==m_outerSize.value());
 
       // 2 - let's check whether there is enough allocated memory
       Index nnz           = tmp.nonZeros();
-      Index start         = m_outerStart==0 ? 0 : matrix.outerIndexPtr()[m_outerStart]; // starting position of the current block
-      Index end           = m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()]; // ending posiiton of the current block
+      Index start         = m_outerStart==0 ? 0 : m_matrix.outerIndexPtr()[m_outerStart]; // starting position of the current block
+      Index end           = m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()]; // ending position of the current block
       Index block_size    = end - start;                                                // available room in the current block
       Index tail_size     = m_matrix.outerIndexPtr()[m_matrix.outerSize()] - end;
-      
+
       Index free_size     = m_matrix.isCompressed()
                           ? Index(matrix.data().allocatedSize()) + block_size
                           : block_size;
 
-      if(nnz>free_size) 
+      Index tmp_start = tmp.outerIndexPtr()[0];
+
+      bool update_trailing_pointers = false;
+      if(nnz>free_size)
       {
         // realloc manually to reduce copies
         typename SparseMatrixType::Storage newdata(m_matrix.data().allocatedSize() - block_size + nnz);
 
-        std::memcpy(&newdata.value(0), &m_matrix.data().value(0), start*sizeof(Scalar));
-        std::memcpy(&newdata.index(0), &m_matrix.data().index(0), start*sizeof(Index));
+        internal::smart_copy(m_matrix.valuePtr(),       m_matrix.valuePtr() + start,      newdata.valuePtr());
+        internal::smart_copy(m_matrix.innerIndexPtr(),  m_matrix.innerIndexPtr() + start, newdata.indexPtr());
 
-        std::memcpy(&newdata.value(start), &tmp.data().value(0), nnz*sizeof(Scalar));
-        std::memcpy(&newdata.index(start), &tmp.data().index(0), nnz*sizeof(Index));
+        internal::smart_copy(tmp.valuePtr() + tmp_start,      tmp.valuePtr() + tmp_start + nnz,       newdata.valuePtr() + start);
+        internal::smart_copy(tmp.innerIndexPtr() + tmp_start, tmp.innerIndexPtr() + tmp_start + nnz,  newdata.indexPtr() + start);
+
+        internal::smart_copy(matrix.valuePtr()+end,       matrix.valuePtr()+end + tail_size,      newdata.valuePtr()+start+nnz);
+        internal::smart_copy(matrix.innerIndexPtr()+end,  matrix.innerIndexPtr()+end + tail_size, newdata.indexPtr()+start+nnz);
 
-        std::memcpy(&newdata.value(start+nnz), &matrix.data().value(end), tail_size*sizeof(Scalar));
-        std::memcpy(&newdata.index(start+nnz), &matrix.data().index(end), tail_size*sizeof(Index));
-        
         newdata.resize(m_matrix.outerIndexPtr()[m_matrix.outerSize()] - block_size + nnz);
 
         matrix.data().swap(newdata);
+
+        update_trailing_pointers = true;
       }
       else
       {
-        // no need to realloc, simply copy the tail at its respective position and insert tmp
-        matrix.data().resize(start + nnz + tail_size);
+        if(m_matrix.isCompressed())
+        {
+          // no need to realloc, simply copy the tail at its respective position and insert tmp
+          matrix.data().resize(start + nnz + tail_size);
 
-        std::memmove(&matrix.data().value(start+nnz), &matrix.data().value(end), tail_size*sizeof(Scalar));
-        std::memmove(&matrix.data().index(start+nnz), &matrix.data().index(end), tail_size*sizeof(Index));
+          internal::smart_memmove(matrix.valuePtr()+end,      matrix.valuePtr() + end+tail_size,      matrix.valuePtr() + start+nnz);
+          internal::smart_memmove(matrix.innerIndexPtr()+end, matrix.innerIndexPtr() + end+tail_size, matrix.innerIndexPtr() + start+nnz);
 
-        std::memcpy(&matrix.data().value(start), &tmp.data().value(0), nnz*sizeof(Scalar));
-        std::memcpy(&matrix.data().index(start), &tmp.data().index(0), nnz*sizeof(Index));
+          update_trailing_pointers = true;
+        }
+
+        internal::smart_copy(tmp.valuePtr() + tmp_start,      tmp.valuePtr() + tmp_start + nnz,       matrix.valuePtr() + start);
+        internal::smart_copy(tmp.innerIndexPtr() + tmp_start, tmp.innerIndexPtr() + tmp_start + nnz,  matrix.innerIndexPtr() + start);
       }
-      
-      // update innerNonZeros
-      if(!m_matrix.isCompressed())
-        for(Index j=0; j<m_outerSize.value(); ++j)
-          matrix.innerNonZeroPtr()[m_outerStart+j] = tmp.innerVector(j).nonZeros();
-
-      // update outer index pointers
-      Index p = start;
-      for(Index k=0; k<m_outerSize.value(); ++k)
+
+      // update outer index pointers and innerNonZeros
+      if(IsVectorAtCompileTime)
       {
-        matrix.outerIndexPtr()[m_outerStart+k] = p;
-        p += tmp.innerVector(k).nonZeros();
+        if(!m_matrix.isCompressed())
+          matrix.innerNonZeroPtr()[m_outerStart] = StorageIndex(nnz);
+        matrix.outerIndexPtr()[m_outerStart] = StorageIndex(start);
       }
-      std::ptrdiff_t offset = nnz - block_size;
-      for(Index k = m_outerStart + m_outerSize.value(); k<=matrix.outerSize(); ++k)
+      else
       {
-        matrix.outerIndexPtr()[k] += offset;
+        StorageIndex p = StorageIndex(start);
+        for(Index k=0; k<m_outerSize.value(); ++k)
+        {
+          StorageIndex nnz_k = internal::convert_index<StorageIndex>(tmp.innerVector(k).nonZeros());
+          if(!m_matrix.isCompressed())
+            matrix.innerNonZeroPtr()[m_outerStart+k] = nnz_k;
+          matrix.outerIndexPtr()[m_outerStart+k] = p;
+          p += nnz_k;
+        }
+      }
+
+      if(update_trailing_pointers)
+      {
+        StorageIndex offset = internal::convert_index<StorageIndex>(nnz - block_size);
+        for(Index k = m_outerStart + m_outerSize.value(); k<=matrix.outerSize(); ++k)
+        {
+          matrix.outerIndexPtr()[k] += offset;
+        }
       }
 
       return derived();
@@ -212,50 +217,46 @@ public:
     }
 
     inline const Scalar* valuePtr() const
-    { return m_matrix.valuePtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
+    { return m_matrix.valuePtr(); }
     inline Scalar* valuePtr()
-    { return m_matrix.const_cast_derived().valuePtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
+    { return m_matrix.valuePtr(); }
 
-    inline const Index* innerIndexPtr() const
-    { return m_matrix.innerIndexPtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
-    inline Index* innerIndexPtr()
-    { return m_matrix.const_cast_derived().innerIndexPtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
+    inline const StorageIndex* innerIndexPtr() const
+    { return m_matrix.innerIndexPtr(); }
+    inline StorageIndex* innerIndexPtr()
+    { return m_matrix.innerIndexPtr(); }
 
-    inline const Index* outerIndexPtr() const
+    inline const StorageIndex* outerIndexPtr() const
+    { return m_matrix.outerIndexPtr() + m_outerStart; }
+    inline StorageIndex* outerIndexPtr()
     { return m_matrix.outerIndexPtr() + m_outerStart; }
-    inline Index* outerIndexPtr()
-    { return m_matrix.const_cast_derived().outerIndexPtr() + m_outerStart; }
 
-    Index nonZeros() const
-    {
-      if(m_matrix.isCompressed())
-        return  std::size_t(m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()])
-              - std::size_t(m_matrix.outerIndexPtr()[m_outerStart]);
-      else if(m_outerSize.value()==0)
-        return 0;
-      else
-        return Map<const Matrix<Index,OuterSize,1> >(m_matrix.innerNonZeroPtr()+m_outerStart, m_outerSize.value()).sum();
-    }
-    
-    inline Scalar& coeffRef(int row, int col)
+    inline const StorageIndex* innerNonZeroPtr() const
+    { return isCompressed() ? 0 : (m_matrix.innerNonZeroPtr()+m_outerStart); }
+    inline StorageIndex* innerNonZeroPtr()
+    { return isCompressed() ? 0 : (m_matrix.innerNonZeroPtr()+m_outerStart); }
+
+    bool isCompressed() const { return m_matrix.innerNonZeroPtr()==0; }
+
+    inline Scalar& coeffRef(Index row, Index col)
     {
-      return m_matrix.const_cast_derived().coeffRef(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
+      return m_matrix.coeffRef(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
     }
-    
-    inline const Scalar coeff(int row, int col) const
+
+    inline const Scalar coeff(Index row, Index col) const
     {
       return m_matrix.coeff(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
     }
-    
-    inline const Scalar coeff(int index) const
+
+    inline const Scalar coeff(Index index) const
     {
       return m_matrix.coeff(IsRowMajor ? m_outerStart : index, IsRowMajor ? index :  m_outerStart);
     }
 
     const Scalar& lastCoeff() const
     {
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(BlockImpl);
-      eigen_assert(nonZeros()>0);
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(sparse_matrix_block_impl);
+      eigen_assert(Base::nonZeros()>0);
       if(m_matrix.isCompressed())
         return m_matrix.valuePtr()[m_matrix.outerIndexPtr()[m_outerStart+1]-1];
       else
@@ -265,109 +266,62 @@ public:
     EIGEN_STRONG_INLINE Index rows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
     EIGEN_STRONG_INLINE Index cols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
 
+    inline const SparseMatrixType& nestedExpression() const { return m_matrix; }
+    inline SparseMatrixType& nestedExpression() { return m_matrix; }
+    Index startRow() const { return IsRowMajor ? m_outerStart : 0; }
+    Index startCol() const { return IsRowMajor ? 0 : m_outerStart; }
+    Index blockRows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
+    Index blockCols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
+
   protected:
 
-    typename SparseMatrixType::Nested m_matrix;
+    typename internal::ref_selector<SparseMatrixType>::non_const_type m_matrix;
     Index m_outerStart;
     const internal::variable_if_dynamic<Index, OuterSize> m_outerSize;
 
 };
 
+} // namespace internal
 
-template<typename _Scalar, int _Options, typename _Index, int BlockRows, int BlockCols>
-class BlockImpl<const SparseMatrix<_Scalar, _Options, _Index>,BlockRows,BlockCols,true,Sparse>
-  : public SparseMatrixBase<Block<const SparseMatrix<_Scalar, _Options, _Index>,BlockRows,BlockCols,true> >
+template<typename _Scalar, int _Options, typename _StorageIndex, int BlockRows, int BlockCols>
+class BlockImpl<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true,Sparse>
+  : public internal::sparse_matrix_block_impl<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols>
 {
-    typedef SparseMatrix<_Scalar, _Options, _Index> SparseMatrixType;
-    typedef typename internal::remove_all<typename SparseMatrixType::Nested>::type _MatrixTypeNested;
-    typedef Block<const SparseMatrixType, BlockRows, BlockCols, true> BlockType;
 public:
-    enum { IsRowMajor = internal::traits<BlockType>::IsRowMajor };
-    EIGEN_SPARSE_PUBLIC_INTERFACE(BlockType)
-protected:
-    enum { OuterSize = IsRowMajor ? BlockRows : BlockCols };
-public:
-    
-    class InnerIterator: public SparseMatrixType::InnerIterator
-    {
-      public:
-        inline InnerIterator(const BlockType& xpr, Index outer)
-          : SparseMatrixType::InnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
-    class ReverseInnerIterator: public SparseMatrixType::ReverseInnerIterator
-    {
-      public:
-        inline ReverseInnerIterator(const BlockType& xpr, Index outer)
-          : SparseMatrixType::ReverseInnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
-
-    inline BlockImpl(const SparseMatrixType& xpr, int i)
-      : m_matrix(xpr), m_outerStart(i), m_outerSize(OuterSize)
-    {}
-
-    inline BlockImpl(const SparseMatrixType& xpr, int startRow, int startCol, int blockRows, int blockCols)
-      : m_matrix(xpr), m_outerStart(IsRowMajor ? startRow : startCol), m_outerSize(IsRowMajor ? blockRows : blockCols)
-    {}
-
-    inline const Scalar* valuePtr() const
-    { return m_matrix.valuePtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
-
-    inline const Index* innerIndexPtr() const
-    { return m_matrix.innerIndexPtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
-
-    inline const Index* outerIndexPtr() const
-    { return m_matrix.outerIndexPtr() + m_outerStart; }
-
-    Index nonZeros() const
-    {
-      if(m_matrix.isCompressed())
-        return  std::size_t(m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()])
-              - std::size_t(m_matrix.outerIndexPtr()[m_outerStart]);
-      else if(m_outerSize.value()==0)
-        return 0;
-      else
-        return Map<const Matrix<Index,OuterSize,1> >(m_matrix.innerNonZeroPtr()+m_outerStart, m_outerSize.value()).sum();
-    }
-    
-    inline const Scalar coeff(int row, int col) const
-    {
-      return m_matrix.coeff(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
-    }
-    
-    inline const Scalar coeff(int index) const
-    {
-      return m_matrix.coeff(IsRowMajor ? m_outerStart : index, IsRowMajor ? index :  m_outerStart);
-    }
-
-    const Scalar& lastCoeff() const
-    {
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(BlockImpl);
-      eigen_assert(nonZeros()>0);
-      if(m_matrix.isCompressed())
-        return m_matrix.valuePtr()[m_matrix.outerIndexPtr()[m_outerStart+1]-1];
-      else
-        return m_matrix.valuePtr()[m_matrix.outerIndexPtr()[m_outerStart]+m_matrix.innerNonZeroPtr()[m_outerStart]-1];
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
-
-  protected:
-
-    typename SparseMatrixType::Nested m_matrix;
-    Index m_outerStart;
-    const internal::variable_if_dynamic<Index, OuterSize> m_outerSize;
+  typedef _StorageIndex StorageIndex;
+  typedef SparseMatrix<_Scalar, _Options, _StorageIndex> SparseMatrixType;
+  typedef internal::sparse_matrix_block_impl<SparseMatrixType,BlockRows,BlockCols> Base;
+  inline BlockImpl(SparseMatrixType& xpr, Index i)
+    : Base(xpr, i)
+  {}
+
+  inline BlockImpl(SparseMatrixType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+    : Base(xpr, startRow, startCol, blockRows, blockCols)
+  {}
+
+  using Base::operator=;
+};
 
+template<typename _Scalar, int _Options, typename _StorageIndex, int BlockRows, int BlockCols>
+class BlockImpl<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true,Sparse>
+  : public internal::sparse_matrix_block_impl<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols>
+{
+public:
+  typedef _StorageIndex StorageIndex;
+  typedef const SparseMatrix<_Scalar, _Options, _StorageIndex> SparseMatrixType;
+  typedef internal::sparse_matrix_block_impl<SparseMatrixType,BlockRows,BlockCols> Base;
+  inline BlockImpl(SparseMatrixType& xpr, Index i)
+    : Base(xpr, i)
+  {}
+
+  inline BlockImpl(SparseMatrixType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+    : Base(xpr, startRow, startCol, blockRows, blockCols)
+  {}
+
+  using Base::operator=;
+private:
+  template<typename Derived> BlockImpl(const SparseMatrixBase<Derived>& xpr, Index i);
+  template<typename Derived> BlockImpl(const SparseMatrixBase<Derived>& xpr);
 };
 
 //----------
@@ -396,7 +350,7 @@ SparseMatrixBase<Derived>::innerVectors(Index outerStart, Index outerSize)
   return Block<Derived,Dynamic,Dynamic,true>(derived(),
                                              IsRowMajor ? outerStart : 0, IsRowMajor ? 0 : outerStart,
                                              IsRowMajor ? outerSize : rows(), IsRowMajor ? cols() : outerSize);
-  
+
 }
 
 /** \returns the \a outer -th column (resp. row) of the matrix \c *this if \c *this
@@ -409,129 +363,241 @@ SparseMatrixBase<Derived>::innerVectors(Index outerStart, Index outerSize) const
   return Block<const Derived,Dynamic,Dynamic,true>(derived(),
                                                   IsRowMajor ? outerStart : 0, IsRowMajor ? 0 : outerStart,
                                                   IsRowMajor ? outerSize : rows(), IsRowMajor ? cols() : outerSize);
-  
+
 }
 
 /** Generic implementation of sparse Block expression.
-  * Real-only. 
+  * Real-only.
   */
 template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
 class BlockImpl<XprType,BlockRows,BlockCols,InnerPanel,Sparse>
   : public SparseMatrixBase<Block<XprType,BlockRows,BlockCols,InnerPanel> >, internal::no_assignment_operator
 {
-  typedef typename internal::remove_all<typename XprType::Nested>::type _MatrixTypeNested;
-  typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
+    typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
+    typedef SparseMatrixBase<BlockType> Base;
+    using Base::convert_index;
 public:
     enum { IsRowMajor = internal::traits<BlockType>::IsRowMajor };
     EIGEN_SPARSE_PUBLIC_INTERFACE(BlockType)
 
+    typedef typename internal::remove_all<typename XprType::Nested>::type _MatrixTypeNested;
+
     /** Column or Row constructor
       */
-    inline BlockImpl(const XprType& xpr, int i)
+    inline BlockImpl(XprType& xpr, Index i)
       : m_matrix(xpr),
-        m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0),
-        m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0),
+        m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? convert_index(i) : 0),
+        m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? convert_index(i) : 0),
         m_blockRows(BlockRows==1 ? 1 : xpr.rows()),
         m_blockCols(BlockCols==1 ? 1 : xpr.cols())
     {}
 
     /** Dynamic-size constructor
       */
-    inline BlockImpl(const XprType& xpr, int startRow, int startCol, int blockRows, int blockCols)
-      : m_matrix(xpr), m_startRow(startRow), m_startCol(startCol), m_blockRows(blockRows), m_blockCols(blockCols)
+    inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+      : m_matrix(xpr), m_startRow(convert_index(startRow)), m_startCol(convert_index(startCol)), m_blockRows(convert_index(blockRows)), m_blockCols(convert_index(blockCols))
     {}
 
-    inline int rows() const { return m_blockRows.value(); }
-    inline int cols() const { return m_blockCols.value(); }
+    inline Index rows() const { return m_blockRows.value(); }
+    inline Index cols() const { return m_blockCols.value(); }
 
-    inline Scalar& coeffRef(int row, int col)
+    inline Scalar& coeffRef(Index row, Index col)
     {
-      return m_matrix.const_cast_derived()
-               .coeffRef(row + m_startRow.value(), col + m_startCol.value());
+      return m_matrix.coeffRef(row + m_startRow.value(), col + m_startCol.value());
     }
 
-    inline const Scalar coeff(int row, int col) const
+    inline const Scalar coeff(Index row, Index col) const
     {
       return m_matrix.coeff(row + m_startRow.value(), col + m_startCol.value());
     }
 
-    inline Scalar& coeffRef(int index)
+    inline Scalar& coeffRef(Index index)
     {
-      return m_matrix.const_cast_derived()
-             .coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                       m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_matrix.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                               m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
-    inline const Scalar coeff(int index) const
+    inline const Scalar coeff(Index index) const
     {
-      return m_matrix
-             .coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                    m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_matrix.coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                            m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
-    
-    inline const _MatrixTypeNested& nestedExpression() const { return m_matrix; }
-    
-    class InnerIterator : public _MatrixTypeNested::InnerIterator
-    {
-      typedef typename _MatrixTypeNested::InnerIterator Base;
-      const BlockType& m_block;
-      Index m_end;
-    public:
-
-      EIGEN_STRONG_INLINE InnerIterator(const BlockType& block, Index outer)
-        : Base(block.derived().nestedExpression(), outer + (IsRowMajor ? block.m_startRow.value() : block.m_startCol.value())),
-          m_block(block),
-          m_end(IsRowMajor ? block.m_startCol.value()+block.m_blockCols.value() : block.m_startRow.value()+block.m_blockRows.value())
-      {
-        while( (Base::operator bool()) && (Base::index() < (IsRowMajor ? m_block.m_startCol.value() : m_block.m_startRow.value())) )
-          Base::operator++();
-      }
 
-      inline Index index()  const { return Base::index() - (IsRowMajor ? m_block.m_startCol.value() : m_block.m_startRow.value()); }
-      inline Index outer()  const { return Base::outer() - (IsRowMajor ? m_block.m_startRow.value() : m_block.m_startCol.value()); }
-      inline Index row()    const { return Base::row()   - m_block.m_startRow.value(); }
-      inline Index col()    const { return Base::col()   - m_block.m_startCol.value(); }
-      
-      inline operator bool() const { return Base::operator bool() && Base::index() < m_end; }
-    };
-    class ReverseInnerIterator : public _MatrixTypeNested::ReverseInnerIterator
-    {
-      typedef typename _MatrixTypeNested::ReverseInnerIterator Base;
-      const BlockType& m_block;
-      Index m_begin;
-    public:
-
-      EIGEN_STRONG_INLINE ReverseInnerIterator(const BlockType& block, Index outer)
-        : Base(block.derived().nestedExpression(), outer + (IsRowMajor ? block.m_startRow.value() : block.m_startCol.value())),
-          m_block(block),
-          m_begin(IsRowMajor ? block.m_startCol.value() : block.m_startRow.value())
-      {
-        while( (Base::operator bool()) && (Base::index() >= (IsRowMajor ? m_block.m_startCol.value()+block.m_blockCols.value() : m_block.m_startRow.value()+block.m_blockRows.value())) )
-          Base::operator--();
-      }
+    inline const XprType& nestedExpression() const { return m_matrix; }
+    inline XprType& nestedExpression() { return m_matrix; }
+    Index startRow() const { return m_startRow.value(); }
+    Index startCol() const { return m_startCol.value(); }
+    Index blockRows() const { return m_blockRows.value(); }
+    Index blockCols() const { return m_blockCols.value(); }
 
-      inline Index index()  const { return Base::index() - (IsRowMajor ? m_block.m_startCol.value() : m_block.m_startRow.value()); }
-      inline Index outer()  const { return Base::outer() - (IsRowMajor ? m_block.m_startRow.value() : m_block.m_startCol.value()); }
-      inline Index row()    const { return Base::row()   - m_block.m_startRow.value(); }
-      inline Index col()    const { return Base::col()   - m_block.m_startCol.value(); }
-      
-      inline operator bool() const { return Base::operator bool() && Base::index() >= m_begin; }
-    };
   protected:
-    friend class InnerIterator;
-    friend class ReverseInnerIterator;
-    
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
+//     friend class internal::GenericSparseBlockInnerIteratorImpl<XprType,BlockRows,BlockCols,InnerPanel>;
+    friend struct internal::unary_evaluator<Block<XprType,BlockRows,BlockCols,InnerPanel>, internal::IteratorBased, Scalar >;
 
-    typename XprType::Nested m_matrix;
+    Index nonZeros() const { return Dynamic; }
+
+    typename internal::ref_selector<XprType>::non_const_type m_matrix;
     const internal::variable_if_dynamic<Index, XprType::RowsAtCompileTime == 1 ? 0 : Dynamic> m_startRow;
     const internal::variable_if_dynamic<Index, XprType::ColsAtCompileTime == 1 ? 0 : Dynamic> m_startCol;
     const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_blockRows;
     const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_blockCols;
 
+  protected:
+    // Disable assignment with clear error message.
+    // Note that simply removing operator= yields compilation errors with ICC+MSVC
+    template<typename T>
+    BlockImpl& operator=(const T&)
+    {
+      EIGEN_STATIC_ASSERT(sizeof(T)==0, THIS_SPARSE_BLOCK_SUBEXPRESSION_IS_READ_ONLY);
+      return *this;
+    }
+
+};
+
+namespace internal {
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+struct unary_evaluator<Block<ArgType,BlockRows,BlockCols,InnerPanel>, IteratorBased >
+ : public evaluator_base<Block<ArgType,BlockRows,BlockCols,InnerPanel> >
+{
+    class InnerVectorInnerIterator;
+    class OuterVectorInnerIterator;
+  public:
+    typedef Block<ArgType,BlockRows,BlockCols,InnerPanel> XprType;
+    typedef typename XprType::StorageIndex StorageIndex;
+    typedef typename XprType::Scalar Scalar;
+
+    enum {
+      IsRowMajor = XprType::IsRowMajor,
+
+      OuterVector =  (BlockCols==1 && ArgType::IsRowMajor)
+                    | // FIXME | instead of || to please GCC 4.4.0 stupid warning "suggest parentheses around &&".
+                      // revert to || as soon as not needed anymore.
+                     (BlockRows==1 && !ArgType::IsRowMajor),
+
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+      Flags = XprType::Flags
+    };
+
+    typedef typename internal::conditional<OuterVector,OuterVectorInnerIterator,InnerVectorInnerIterator>::type InnerIterator;
+
+    explicit unary_evaluator(const XprType& op)
+      : m_argImpl(op.nestedExpression()), m_block(op)
+    {}
+
+    inline Index nonZerosEstimate() const {
+      Index nnz = m_block.nonZeros();
+      if(nnz<0)
+        return m_argImpl.nonZerosEstimate() * m_block.size() / m_block.nestedExpression().size();
+      return nnz;
+    }
+
+  protected:
+    typedef typename evaluator<ArgType>::InnerIterator EvalIterator;
+
+    evaluator<ArgType> m_argImpl;
+    const XprType &m_block;
+};
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+class unary_evaluator<Block<ArgType,BlockRows,BlockCols,InnerPanel>, IteratorBased>::InnerVectorInnerIterator
+ : public EvalIterator
+{
+  enum { IsRowMajor = unary_evaluator::IsRowMajor };
+  const XprType& m_block;
+  Index m_end;
+public:
+
+  EIGEN_STRONG_INLINE InnerVectorInnerIterator(const unary_evaluator& aEval, Index outer)
+    : EvalIterator(aEval.m_argImpl, outer + (IsRowMajor ? aEval.m_block.startRow() : aEval.m_block.startCol())),
+      m_block(aEval.m_block),
+      m_end(IsRowMajor ? aEval.m_block.startCol()+aEval.m_block.blockCols() : aEval.m_block.startRow()+aEval.m_block.blockRows())
+  {
+    while( (EvalIterator::operator bool()) && (EvalIterator::index() < (IsRowMajor ? m_block.startCol() : m_block.startRow())) )
+      EvalIterator::operator++();
+  }
+
+  inline StorageIndex index() const { return EvalIterator::index() - convert_index<StorageIndex>(IsRowMajor ? m_block.startCol() : m_block.startRow()); }
+  inline Index outer()  const { return EvalIterator::outer() - (IsRowMajor ? m_block.startRow() : m_block.startCol()); }
+  inline Index row()    const { return EvalIterator::row()   - m_block.startRow(); }
+  inline Index col()    const { return EvalIterator::col()   - m_block.startCol(); }
+
+  inline operator bool() const { return EvalIterator::operator bool() && EvalIterator::index() < m_end; }
+};
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+class unary_evaluator<Block<ArgType,BlockRows,BlockCols,InnerPanel>, IteratorBased>::OuterVectorInnerIterator
+{
+  enum { IsRowMajor = unary_evaluator::IsRowMajor };
+  const unary_evaluator& m_eval;
+  Index m_outerPos;
+  const Index m_innerIndex;
+  Index m_end;
+  EvalIterator m_it;
+public:
+
+  EIGEN_STRONG_INLINE OuterVectorInnerIterator(const unary_evaluator& aEval, Index outer)
+    : m_eval(aEval),
+      m_outerPos( (IsRowMajor ? aEval.m_block.startCol() : aEval.m_block.startRow()) ),
+      m_innerIndex(IsRowMajor ? aEval.m_block.startRow() : aEval.m_block.startCol()),
+      m_end(IsRowMajor ? aEval.m_block.startCol()+aEval.m_block.blockCols() : aEval.m_block.startRow()+aEval.m_block.blockRows()),
+      m_it(m_eval.m_argImpl, m_outerPos)
+  {
+    EIGEN_UNUSED_VARIABLE(outer);
+    eigen_assert(outer==0);
+
+    while(m_it && m_it.index() < m_innerIndex) ++m_it;
+    if((!m_it) || (m_it.index()!=m_innerIndex))
+      ++(*this);
+  }
+
+  inline StorageIndex index() const { return convert_index<StorageIndex>(m_outerPos - (IsRowMajor ? m_eval.m_block.startCol() : m_eval.m_block.startRow())); }
+  inline Index outer()  const { return 0; }
+  inline Index row()    const { return IsRowMajor ? 0 : index(); }
+  inline Index col()    const { return IsRowMajor ? index() : 0; }
+
+  inline Scalar value() const { return m_it.value(); }
+  inline Scalar& valueRef() { return m_it.valueRef(); }
+
+  inline OuterVectorInnerIterator& operator++()
+  {
+    // search next non-zero entry
+    while(++m_outerPos<m_end)
+    {
+      // Restart iterator at the next inner-vector:
+      m_it.~EvalIterator();
+      ::new (&m_it) EvalIterator(m_eval.m_argImpl, m_outerPos);
+      // search for the key m_innerIndex in the current outer-vector
+      while(m_it && m_it.index() < m_innerIndex) ++m_it;
+      if(m_it && m_it.index()==m_innerIndex) break;
+    }
+    return *this;
+  }
+
+  inline operator bool() const { return m_outerPos < m_end; }
 };
 
+template<typename _Scalar, int _Options, typename _StorageIndex, int BlockRows, int BlockCols>
+struct unary_evaluator<Block<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true>, IteratorBased>
+  : evaluator<SparseCompressedBase<Block<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true> > >
+{
+  typedef Block<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true> XprType;
+  typedef evaluator<SparseCompressedBase<XprType> > Base;
+  explicit unary_evaluator(const XprType &xpr) : Base(xpr) {}
+};
+
+template<typename _Scalar, int _Options, typename _StorageIndex, int BlockRows, int BlockCols>
+struct unary_evaluator<Block<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true>, IteratorBased>
+  : evaluator<SparseCompressedBase<Block<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true> > >
+{
+  typedef Block<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true> XprType;
+  typedef evaluator<SparseCompressedBase<XprType> > Base;
+  explicit unary_evaluator(const XprType &xpr) : Base(xpr) {}
+};
+
+} // end namespace internal
+
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSE_BLOCK_H
-
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseColEtree.h b/extern/Eigen3/Eigen/src/SparseCore/SparseColEtree.h
index f8745f46100..ebe02d1ab02 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseColEtree.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseColEtree.h
@@ -58,30 +58,29 @@ Index etree_find (Index i, IndexVector& pp)
   * \param perm The permutation to apply to the column of \b mat
   */
 template <typename MatrixType, typename IndexVector>
-int coletree(const MatrixType& mat, IndexVector& parent, IndexVector& firstRowElt, typename MatrixType::Index *perm=0)
+int coletree(const MatrixType& mat, IndexVector& parent, IndexVector& firstRowElt, typename MatrixType::StorageIndex *perm=0)
 {
-  typedef typename MatrixType::Index Index;
-  Index nc = mat.cols(); // Number of columns 
-  Index m = mat.rows();
-  Index diagSize = (std::min)(nc,m);
+  typedef typename MatrixType::StorageIndex StorageIndex;
+  StorageIndex nc = convert_index<StorageIndex>(mat.cols()); // Number of columns
+  StorageIndex m = convert_index<StorageIndex>(mat.rows());
+  StorageIndex diagSize = (std::min)(nc,m);
   IndexVector root(nc); // root of subtree of etree 
   root.setZero();
   IndexVector pp(nc); // disjoint sets 
   pp.setZero(); // Initialize disjoint sets 
   parent.resize(mat.cols());
   //Compute first nonzero column in each row 
-  Index row,col; 
   firstRowElt.resize(m);
   firstRowElt.setConstant(nc);
   firstRowElt.segment(0, diagSize).setLinSpaced(diagSize, 0, diagSize-1);
   bool found_diag;
-  for (col = 0; col < nc; col++)
+  for (StorageIndex col = 0; col < nc; col++)
   {
-    Index pcol = col;
+    StorageIndex pcol = col;
     if(perm) pcol  = perm[col];
     for (typename MatrixType::InnerIterator it(mat, pcol); it; ++it)
     { 
-      row = it.row();
+      Index row = it.row();
       firstRowElt(row) = (std::min)(firstRowElt(row), col);
     }
   }
@@ -89,8 +88,8 @@ int coletree(const MatrixType& mat, IndexVector& parent, IndexVector& firstRowEl
           except use (firstRowElt[r],c) in place of an edge (r,c) of A.
     Thus each row clique in A'*A is replaced by a star
     centered at its first vertex, which has the same fill. */
-  Index rset, cset, rroot; 
-  for (col = 0; col < nc; col++) 
+  StorageIndex rset, cset, rroot;
+  for (StorageIndex col = 0; col < nc; col++) 
   {
     found_diag = col>=m;
     pp(col) = col; 
@@ -99,7 +98,7 @@ int coletree(const MatrixType& mat, IndexVector& parent, IndexVector& firstRowEl
     parent(col) = nc; 
     /* The diagonal element is treated here even if it does not exist in the matrix
      * hence the loop is executed once more */ 
-    Index pcol = col;
+    StorageIndex pcol = col;
     if(perm) pcol  = perm[col];
     for (typename MatrixType::InnerIterator it(mat, pcol); it||!found_diag; ++it)
     { //  A sequence of interleaved find and union is performed 
@@ -107,7 +106,7 @@ int coletree(const MatrixType& mat, IndexVector& parent, IndexVector& firstRowEl
       if(it) i = it.index();
       if (i == col) found_diag = true;
       
-      row = firstRowElt(i);
+      StorageIndex row = firstRowElt(i);
       if (row >= col) continue; 
       rset = internal::etree_find(row, pp); // Find the name of the set containing row
       rroot = root(rset);
@@ -127,10 +126,11 @@ int coletree(const MatrixType& mat, IndexVector& parent, IndexVector& firstRowEl
   * Depth-first search from vertex n.  No recursion.
   * This routine was contributed by Cédric Doucet, CEDRAT Group, Meylan, France.
 */
-template <typename Index, typename IndexVector>
-void nr_etdfs (Index n, IndexVector& parent, IndexVector& first_kid, IndexVector& next_kid, IndexVector& post, Index postnum)
+template <typename IndexVector>
+void nr_etdfs (typename IndexVector::Scalar n, IndexVector& parent, IndexVector& first_kid, IndexVector& next_kid, IndexVector& post, typename IndexVector::Scalar postnum)
 {
-  Index current = n, first, next;
+  typedef typename IndexVector::Scalar StorageIndex;
+  StorageIndex current = n, first, next;
   while (postnum != n) 
   {
     // No kid for the current node
@@ -174,22 +174,22 @@ void nr_etdfs (Index n, IndexVector& parent, IndexVector& first_kid, IndexVector
   * \param parent Input tree
   * \param post postordered tree
   */
-template <typename Index, typename IndexVector>
-void treePostorder(Index n, IndexVector& parent, IndexVector& post)
+template <typename IndexVector>
+void treePostorder(typename IndexVector::Scalar n, IndexVector& parent, IndexVector& post)
 {
+  typedef typename IndexVector::Scalar StorageIndex;
   IndexVector first_kid, next_kid; // Linked list of children 
-  Index postnum; 
+  StorageIndex postnum; 
   // Allocate storage for working arrays and results 
   first_kid.resize(n+1); 
   next_kid.setZero(n+1);
   post.setZero(n+1);
   
   // Set up structure describing children
-  Index v, dad; 
   first_kid.setConstant(-1); 
-  for (v = n-1; v >= 0; v--) 
+  for (StorageIndex v = n-1; v >= 0; v--) 
   {
-    dad = parent(v);
+    StorageIndex dad = parent(v);
     next_kid(v) = first_kid(dad); 
     first_kid(dad) = v; 
   }
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseCompressedBase.h b/extern/Eigen3/Eigen/src/SparseCore/SparseCompressedBase.h
new file mode 100644
index 00000000000..5ccb4665614
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseCompressedBase.h
@@ -0,0 +1,341 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSE_COMPRESSED_BASE_H
+#define EIGEN_SPARSE_COMPRESSED_BASE_H
+
+namespace Eigen { 
+
+template<typename Derived> class SparseCompressedBase;
+  
+namespace internal {
+
+template<typename Derived>
+struct traits<SparseCompressedBase<Derived> > : traits<Derived>
+{};
+
+} // end namespace internal
+
+/** \ingroup SparseCore_Module
+  * \class SparseCompressedBase
+  * \brief Common base class for sparse [compressed]-{row|column}-storage format.
+  *
+  * This class defines the common interface for all derived classes implementing the compressed sparse storage format, such as:
+  *  - SparseMatrix
+  *  - Ref<SparseMatrixType,Options>
+  *  - Map<SparseMatrixType>
+  *
+  */
+template<typename Derived>
+class SparseCompressedBase
+  : public SparseMatrixBase<Derived>
+{
+  public:
+    typedef SparseMatrixBase<Derived> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(SparseCompressedBase)
+    using Base::operator=;
+    using Base::IsRowMajor;
+    
+    class InnerIterator;
+    class ReverseInnerIterator;
+    
+  protected:
+    typedef typename Base::IndexVector IndexVector;
+    Eigen::Map<IndexVector> innerNonZeros() { return Eigen::Map<IndexVector>(innerNonZeroPtr(), isCompressed()?0:derived().outerSize()); }
+    const  Eigen::Map<const IndexVector> innerNonZeros() const { return Eigen::Map<const IndexVector>(innerNonZeroPtr(), isCompressed()?0:derived().outerSize()); }
+        
+  public:
+    
+    /** \returns the number of non zero coefficients */
+    inline Index nonZeros() const
+    {
+      if(Derived::IsVectorAtCompileTime && outerIndexPtr()==0)
+        return derived().nonZeros();
+      else if(isCompressed())
+        return outerIndexPtr()[derived().outerSize()]-outerIndexPtr()[0];
+      else if(derived().outerSize()==0)
+        return 0;
+      else
+        return innerNonZeros().sum();
+    }
+    
+    /** \returns a const pointer to the array of values.
+      * This function is aimed at interoperability with other libraries.
+      * \sa innerIndexPtr(), outerIndexPtr() */
+    inline const Scalar* valuePtr() const { return derived().valuePtr(); }
+    /** \returns a non-const pointer to the array of values.
+      * This function is aimed at interoperability with other libraries.
+      * \sa innerIndexPtr(), outerIndexPtr() */
+    inline Scalar* valuePtr() { return derived().valuePtr(); }
+
+    /** \returns a const pointer to the array of inner indices.
+      * This function is aimed at interoperability with other libraries.
+      * \sa valuePtr(), outerIndexPtr() */
+    inline const StorageIndex* innerIndexPtr() const { return derived().innerIndexPtr(); }
+    /** \returns a non-const pointer to the array of inner indices.
+      * This function is aimed at interoperability with other libraries.
+      * \sa valuePtr(), outerIndexPtr() */
+    inline StorageIndex* innerIndexPtr() { return derived().innerIndexPtr(); }
+
+    /** \returns a const pointer to the array of the starting positions of the inner vectors.
+      * This function is aimed at interoperability with other libraries.
+      * \warning it returns the null pointer 0 for SparseVector
+      * \sa valuePtr(), innerIndexPtr() */
+    inline const StorageIndex* outerIndexPtr() const { return derived().outerIndexPtr(); }
+    /** \returns a non-const pointer to the array of the starting positions of the inner vectors.
+      * This function is aimed at interoperability with other libraries.
+      * \warning it returns the null pointer 0 for SparseVector
+      * \sa valuePtr(), innerIndexPtr() */
+    inline StorageIndex* outerIndexPtr() { return derived().outerIndexPtr(); }
+
+    /** \returns a const pointer to the array of the number of non zeros of the inner vectors.
+      * This function is aimed at interoperability with other libraries.
+      * \warning it returns the null pointer 0 in compressed mode */
+    inline const StorageIndex* innerNonZeroPtr() const { return derived().innerNonZeroPtr(); }
+    /** \returns a non-const pointer to the array of the number of non zeros of the inner vectors.
+      * This function is aimed at interoperability with other libraries.
+      * \warning it returns the null pointer 0 in compressed mode */
+    inline StorageIndex* innerNonZeroPtr() { return derived().innerNonZeroPtr(); }
+    
+    /** \returns whether \c *this is in compressed form. */
+    inline bool isCompressed() const { return innerNonZeroPtr()==0; }
+
+    /** \returns a read-only view of the stored coefficients as a 1D array expression.
+      *
+      * \warning this method is for \b compressed \b storage \b only, and it will trigger an assertion otherwise.
+      *
+      * \sa valuePtr(), isCompressed() */
+    const Map<const Array<Scalar,Dynamic,1> > coeffs() const { eigen_assert(isCompressed()); return Array<Scalar,Dynamic,1>::Map(valuePtr(),nonZeros()); }
+
+    /** \returns a read-write view of the stored coefficients as a 1D array expression
+      *
+      * \warning this method is for \b compressed \b storage \b only, and it will trigger an assertion otherwise.
+      *
+      * Here is an example:
+      * \include SparseMatrix_coeffs.cpp
+      * and the output is:
+      * \include SparseMatrix_coeffs.out
+      *
+      * \sa valuePtr(), isCompressed() */
+    Map<Array<Scalar,Dynamic,1> > coeffs() { eigen_assert(isCompressed()); return Array<Scalar,Dynamic,1>::Map(valuePtr(),nonZeros()); }
+
+  protected:
+    /** Default constructor. Do nothing. */
+    SparseCompressedBase() {}
+  private:
+    template<typename OtherDerived> explicit SparseCompressedBase(const SparseCompressedBase<OtherDerived>&);
+};
+
+template<typename Derived>
+class SparseCompressedBase<Derived>::InnerIterator
+{
+  public:
+    InnerIterator()
+      : m_values(0), m_indices(0), m_outer(0), m_id(0), m_end(0)
+    {}
+
+    InnerIterator(const InnerIterator& other)
+      : m_values(other.m_values), m_indices(other.m_indices), m_outer(other.m_outer), m_id(other.m_id), m_end(other.m_end)
+    {}
+
+    InnerIterator& operator=(const InnerIterator& other)
+    {
+      m_values = other.m_values;
+      m_indices = other.m_indices;
+      const_cast<OuterType&>(m_outer).setValue(other.m_outer.value());
+      m_id = other.m_id;
+      m_end = other.m_end;
+      return *this;
+    }
+
+    InnerIterator(const SparseCompressedBase& mat, Index outer)
+      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(outer)
+    {
+      if(Derived::IsVectorAtCompileTime && mat.outerIndexPtr()==0)
+      {
+        m_id = 0;
+        m_end = mat.nonZeros();
+      }
+      else
+      {
+        m_id = mat.outerIndexPtr()[outer];
+        if(mat.isCompressed())
+          m_end = mat.outerIndexPtr()[outer+1];
+        else
+          m_end = m_id + mat.innerNonZeroPtr()[outer];
+      }
+    }
+
+    explicit InnerIterator(const SparseCompressedBase& mat)
+      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(0), m_id(0), m_end(mat.nonZeros())
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
+    }
+
+    explicit InnerIterator(const internal::CompressedStorage<Scalar,StorageIndex>& data)
+      : m_values(data.valuePtr()), m_indices(data.indexPtr()), m_outer(0), m_id(0), m_end(data.size())
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
+    }
+
+    inline InnerIterator& operator++() { m_id++; return *this; }
+
+    inline const Scalar& value() const { return m_values[m_id]; }
+    inline Scalar& valueRef() { return const_cast<Scalar&>(m_values[m_id]); }
+
+    inline StorageIndex index() const { return m_indices[m_id]; }
+    inline Index outer() const { return m_outer.value(); }
+    inline Index row() const { return IsRowMajor ? m_outer.value() : index(); }
+    inline Index col() const { return IsRowMajor ? index() : m_outer.value(); }
+
+    inline operator bool() const { return (m_id < m_end); }
+
+  protected:
+    const Scalar* m_values;
+    const StorageIndex* m_indices;
+    typedef internal::variable_if_dynamic<Index,Derived::IsVectorAtCompileTime?0:Dynamic> OuterType;
+    const OuterType m_outer;
+    Index m_id;
+    Index m_end;
+  private:
+    // If you get here, then you're not using the right InnerIterator type, e.g.:
+    //   SparseMatrix<double,RowMajor> A;
+    //   SparseMatrix<double>::InnerIterator it(A,0);
+    template<typename T> InnerIterator(const SparseMatrixBase<T>&, Index outer);
+};
+
+template<typename Derived>
+class SparseCompressedBase<Derived>::ReverseInnerIterator
+{
+  public:
+    ReverseInnerIterator(const SparseCompressedBase& mat, Index outer)
+      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(outer)
+    {
+      if(Derived::IsVectorAtCompileTime && mat.outerIndexPtr()==0)
+      {
+        m_start = 0;
+        m_id = mat.nonZeros();
+      }
+      else
+      {
+        m_start = mat.outerIndexPtr()[outer];
+        if(mat.isCompressed())
+          m_id = mat.outerIndexPtr()[outer+1];
+        else
+          m_id = m_start + mat.innerNonZeroPtr()[outer];
+      }
+    }
+
+    explicit ReverseInnerIterator(const SparseCompressedBase& mat)
+      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(0), m_start(0), m_id(mat.nonZeros())
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
+    }
+
+    explicit ReverseInnerIterator(const internal::CompressedStorage<Scalar,StorageIndex>& data)
+      : m_values(data.valuePtr()), m_indices(data.indexPtr()), m_outer(0), m_start(0), m_id(data.size())
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
+    }
+
+    inline ReverseInnerIterator& operator--() { --m_id; return *this; }
+
+    inline const Scalar& value() const { return m_values[m_id-1]; }
+    inline Scalar& valueRef() { return const_cast<Scalar&>(m_values[m_id-1]); }
+
+    inline StorageIndex index() const { return m_indices[m_id-1]; }
+    inline Index outer() const { return m_outer.value(); }
+    inline Index row() const { return IsRowMajor ? m_outer.value() : index(); }
+    inline Index col() const { return IsRowMajor ? index() : m_outer.value(); }
+
+    inline operator bool() const { return (m_id > m_start); }
+
+  protected:
+    const Scalar* m_values;
+    const StorageIndex* m_indices;
+    typedef internal::variable_if_dynamic<Index,Derived::IsVectorAtCompileTime?0:Dynamic> OuterType;
+    const OuterType m_outer;
+    Index m_start;
+    Index m_id;
+};
+
+namespace internal {
+
+template<typename Derived>
+struct evaluator<SparseCompressedBase<Derived> >
+  : evaluator_base<Derived>
+{
+  typedef typename Derived::Scalar Scalar;
+  typedef typename Derived::InnerIterator InnerIterator;
+  
+  enum {
+    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Flags = Derived::Flags
+  };
+  
+  evaluator() : m_matrix(0), m_zero(0)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  explicit evaluator(const Derived &mat) : m_matrix(&mat), m_zero(0)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_matrix->nonZeros();
+  }
+  
+  operator Derived&() { return m_matrix->const_cast_derived(); }
+  operator const Derived&() const { return *m_matrix; }
+  
+  typedef typename DenseCoeffsBase<Derived,ReadOnlyAccessors>::CoeffReturnType CoeffReturnType;
+  const Scalar& coeff(Index row, Index col) const
+  {
+    Index p = find(row,col);
+
+    if(p==Dynamic)
+      return m_zero;
+    else
+      return m_matrix->const_cast_derived().valuePtr()[p];
+  }
+
+  Scalar& coeffRef(Index row, Index col)
+  {
+    Index p = find(row,col);
+    eigen_assert(p!=Dynamic && "written coefficient does not exist");
+    return m_matrix->const_cast_derived().valuePtr()[p];
+  }
+
+protected:
+
+  Index find(Index row, Index col) const
+  {
+    eigen_internal_assert(row>=0 && row<m_matrix->rows() && col>=0 && col<m_matrix->cols());
+
+    const Index outer = Derived::IsRowMajor ? row : col;
+    const Index inner = Derived::IsRowMajor ? col : row;
+
+    Index start = m_matrix->outerIndexPtr()[outer];
+    Index end = m_matrix->isCompressed() ? m_matrix->outerIndexPtr()[outer+1] : m_matrix->outerIndexPtr()[outer] + m_matrix->innerNonZeroPtr()[outer];
+    eigen_assert(end>=start && "you are using a non finalized sparse matrix or written coefficient does not exist");
+    const Index p = std::lower_bound(m_matrix->innerIndexPtr()+start, m_matrix->innerIndexPtr()+end,inner) - m_matrix->innerIndexPtr();
+
+    return ((p<end) && (m_matrix->innerIndexPtr()[p]==inner)) ? p : Dynamic;
+  }
+
+  const Derived *m_matrix;
+  const Scalar m_zero;
+};
+
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSE_COMPRESSED_BASE_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseCwiseBinaryOp.h b/extern/Eigen3/Eigen/src/SparseCore/SparseCwiseBinaryOp.h
index 4ca9128337f..e315e355060 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseCwiseBinaryOp.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseCwiseBinaryOp.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -28,89 +28,57 @@ namespace Eigen {
 //                         generic      sparse
 //  4 - dense op dense     product      dense
 //                         generic      dense
-
-namespace internal {
-
-template<> struct promote_storage_type<Dense,Sparse>
-{ typedef Sparse ret; };
-
-template<> struct promote_storage_type<Sparse,Dense>
-{ typedef Sparse ret; };
-
-template<typename BinaryOp, typename Lhs, typename Rhs, typename Derived,
-  typename _LhsStorageMode = typename traits<Lhs>::StorageKind,
-  typename _RhsStorageMode = typename traits<Rhs>::StorageKind>
-class sparse_cwise_binary_op_inner_iterator_selector;
-
-} // end namespace internal
+//
+// TODO to ease compiler job, we could specialize product/quotient with a scalar
+//      and fallback to cwise-unary evaluator using bind1st_op and bind2nd_op.
 
 template<typename BinaryOp, typename Lhs, typename Rhs>
 class CwiseBinaryOpImpl<BinaryOp, Lhs, Rhs, Sparse>
   : public SparseMatrixBase<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
 {
   public:
-    class InnerIterator;
-    class ReverseInnerIterator;
     typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> Derived;
+    typedef SparseMatrixBase<Derived> Base;
     EIGEN_SPARSE_PUBLIC_INTERFACE(Derived)
     CwiseBinaryOpImpl()
     {
-      typedef typename internal::traits<Lhs>::StorageKind LhsStorageKind;
-      typedef typename internal::traits<Rhs>::StorageKind RhsStorageKind;
       EIGEN_STATIC_ASSERT((
-                (!internal::is_same<LhsStorageKind,RhsStorageKind>::value)
-            ||  ((Lhs::Flags&RowMajorBit) == (Rhs::Flags&RowMajorBit))),
+                (!internal::is_same<typename internal::traits<Lhs>::StorageKind,
+                                    typename internal::traits<Rhs>::StorageKind>::value)
+            ||  ((internal::evaluator<Lhs>::Flags&RowMajorBit) == (internal::evaluator<Rhs>::Flags&RowMajorBit))),
             THE_STORAGE_ORDER_OF_BOTH_SIDES_MUST_MATCH);
     }
 };
 
-template<typename BinaryOp, typename Lhs, typename Rhs>
-class CwiseBinaryOpImpl<BinaryOp,Lhs,Rhs,Sparse>::InnerIterator
-  : public internal::sparse_cwise_binary_op_inner_iterator_selector<BinaryOp,Lhs,Rhs,typename CwiseBinaryOpImpl<BinaryOp,Lhs,Rhs,Sparse>::InnerIterator>
-{
-  public:
-    typedef typename Lhs::Index Index;
-    typedef internal::sparse_cwise_binary_op_inner_iterator_selector<
-      BinaryOp,Lhs,Rhs, InnerIterator> Base;
-
-    // NOTE: we have to prefix Index by "typename Lhs::" to avoid an ICE with VC11
-    EIGEN_STRONG_INLINE InnerIterator(const CwiseBinaryOpImpl& binOp, typename Lhs::Index outer)
-      : Base(binOp.derived(),outer)
-    {}
-};
-
-/***************************************************************************
-* Implementation of inner-iterators
-***************************************************************************/
-
-// template<typename T> struct internal::func_is_conjunction { enum { ret = false }; };
-// template<typename T> struct internal::func_is_conjunction<internal::scalar_product_op<T> > { enum { ret = true }; };
-
-// TODO generalize the internal::scalar_product_op specialization to all conjunctions if any !
-
 namespace internal {
 
-// sparse - sparse  (generic)
-template<typename BinaryOp, typename Lhs, typename Rhs, typename Derived>
-class sparse_cwise_binary_op_inner_iterator_selector<BinaryOp, Lhs, Rhs, Derived, Sparse, Sparse>
-{
-    typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> CwiseBinaryXpr;
-    typedef typename traits<CwiseBinaryXpr>::Scalar Scalar;
-    typedef typename traits<CwiseBinaryXpr>::_LhsNested _LhsNested;
-    typedef typename traits<CwiseBinaryXpr>::_RhsNested _RhsNested;
-    typedef typename _LhsNested::InnerIterator LhsIterator;
-    typedef typename _RhsNested::InnerIterator RhsIterator;
-    typedef typename Lhs::Index Index;
+  
+// Generic "sparse OP sparse"
+template<typename XprType> struct binary_sparse_evaluator;
 
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IteratorBased, IteratorBased>
+  : evaluator_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+{
+protected:
+  typedef typename evaluator<Lhs>::InnerIterator  LhsIterator;
+  typedef typename evaluator<Rhs>::InnerIterator  RhsIterator;
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef typename traits<XprType>::Scalar Scalar;
+  typedef typename XprType::StorageIndex StorageIndex;
+public:
+
+  class InnerIterator
+  {
   public:
-
-    EIGEN_STRONG_INLINE sparse_cwise_binary_op_inner_iterator_selector(const CwiseBinaryXpr& xpr, Index outer)
-      : m_lhsIter(xpr.lhs(),outer), m_rhsIter(xpr.rhs(),outer), m_functor(xpr.functor())
+    
+    EIGEN_STRONG_INLINE InnerIterator(const binary_evaluator& aEval, Index outer)
+      : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor)
     {
       this->operator++();
     }
 
-    EIGEN_STRONG_INLINE Derived& operator++()
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
     {
       if (m_lhsIter && m_rhsIter && (m_lhsIter.index() == m_rhsIter.index()))
       {
@@ -136,12 +104,13 @@ class sparse_cwise_binary_op_inner_iterator_selector<BinaryOp, Lhs, Rhs, Derived
         m_value = 0; // this is to avoid a compilation warning
         m_id = -1;
       }
-      return *static_cast<Derived*>(this);
+      return *this;
     }
 
     EIGEN_STRONG_INLINE Scalar value() const { return m_value; }
 
-    EIGEN_STRONG_INLINE Index index() const { return m_id; }
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_id; }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return Lhs::IsRowMajor ? m_lhsIter.row() : index(); }
     EIGEN_STRONG_INLINE Index col() const { return Lhs::IsRowMajor ? index() : m_lhsIter.col(); }
 
@@ -152,25 +121,303 @@ class sparse_cwise_binary_op_inner_iterator_selector<BinaryOp, Lhs, Rhs, Derived
     RhsIterator m_rhsIter;
     const BinaryOp& m_functor;
     Scalar m_value;
-    Index m_id;
+    StorageIndex m_id;
+  };
+  
+  
+  enum {
+    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    Flags = XprType::Flags
+  };
+  
+  explicit binary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_lhsImpl.nonZerosEstimate() + m_rhsImpl.nonZerosEstimate();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<Lhs> m_lhsImpl;
+  evaluator<Rhs> m_rhsImpl;
 };
 
-// sparse - sparse  (product)
-template<typename T, typename Lhs, typename Rhs, typename Derived>
-class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs, Rhs, Derived, Sparse, Sparse>
+// dense op sparse
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IteratorBased>
+  : evaluator_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
 {
-    typedef scalar_product_op<T> BinaryFunc;
-    typedef CwiseBinaryOp<BinaryFunc, Lhs, Rhs> CwiseBinaryXpr;
-    typedef typename CwiseBinaryXpr::Scalar Scalar;
-    typedef typename traits<CwiseBinaryXpr>::_LhsNested _LhsNested;
-    typedef typename _LhsNested::InnerIterator LhsIterator;
-    typedef typename traits<CwiseBinaryXpr>::_RhsNested _RhsNested;
-    typedef typename _RhsNested::InnerIterator RhsIterator;
-    typedef typename Lhs::Index Index;
+protected:
+  typedef typename evaluator<Rhs>::InnerIterator  RhsIterator;
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef typename traits<XprType>::Scalar Scalar;
+  typedef typename XprType::StorageIndex StorageIndex;
+public:
+
+  class InnerIterator
+  {
+    enum { IsRowMajor = (int(Rhs::Flags)&RowMajorBit)==RowMajorBit };
   public:
 
-    EIGEN_STRONG_INLINE sparse_cwise_binary_op_inner_iterator_selector(const CwiseBinaryXpr& xpr, Index outer)
-      : m_lhsIter(xpr.lhs(),outer), m_rhsIter(xpr.rhs(),outer), m_functor(xpr.functor())
+    EIGEN_STRONG_INLINE InnerIterator(const binary_evaluator& aEval, Index outer)
+      : m_lhsEval(aEval.m_lhsImpl), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor), m_value(0), m_id(-1), m_innerSize(aEval.m_expr.rhs().innerSize())
+    {
+      this->operator++();
+    }
+
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
+    {
+      ++m_id;
+      if(m_id<m_innerSize)
+      {
+        Scalar lhsVal = m_lhsEval.coeff(IsRowMajor?m_rhsIter.outer():m_id,
+                                        IsRowMajor?m_id:m_rhsIter.outer());
+        if(m_rhsIter && m_rhsIter.index()==m_id)
+        {
+          m_value = m_functor(lhsVal, m_rhsIter.value());
+          ++m_rhsIter;
+        }
+        else
+          m_value = m_functor(lhsVal, Scalar(0));
+      }
+
+      return *this;
+    }
+
+    EIGEN_STRONG_INLINE Scalar value() const { eigen_internal_assert(m_id<m_innerSize); return m_value; }
+
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_id; }
+    EIGEN_STRONG_INLINE Index outer() const { return m_rhsIter.outer(); }
+    EIGEN_STRONG_INLINE Index row() const { return IsRowMajor ? m_rhsIter.outer() : m_id; }
+    EIGEN_STRONG_INLINE Index col() const { return IsRowMajor ? m_id : m_rhsIter.outer(); }
+
+    EIGEN_STRONG_INLINE operator bool() const { return m_id<m_innerSize; }
+
+  protected:
+    const evaluator<Lhs> &m_lhsEval;
+    RhsIterator m_rhsIter;
+    const BinaryOp& m_functor;
+    Scalar m_value;
+    StorageIndex m_id;
+    StorageIndex m_innerSize;
+  };
+
+
+  enum {
+    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    // Expose storage order of the sparse expression
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(Rhs::Flags)&RowMajorBit)
+  };
+
+  explicit binary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()),
+      m_rhsImpl(xpr.rhs()),
+      m_expr(xpr)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  inline Index nonZerosEstimate() const {
+    return m_expr.size();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<Lhs> m_lhsImpl;
+  evaluator<Rhs> m_rhsImpl;
+  const XprType &m_expr;
+};
+
+// sparse op dense
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IteratorBased, IndexBased>
+  : evaluator_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+{
+protected:
+  typedef typename evaluator<Lhs>::InnerIterator  LhsIterator;
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef typename traits<XprType>::Scalar Scalar;
+  typedef typename XprType::StorageIndex StorageIndex;
+public:
+
+  class InnerIterator
+  {
+    enum { IsRowMajor = (int(Lhs::Flags)&RowMajorBit)==RowMajorBit };
+  public:
+
+    EIGEN_STRONG_INLINE InnerIterator(const binary_evaluator& aEval, Index outer)
+      : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsEval(aEval.m_rhsImpl), m_functor(aEval.m_functor), m_value(0), m_id(-1), m_innerSize(aEval.m_expr.lhs().innerSize())
+    {
+      this->operator++();
+    }
+
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
+    {
+      ++m_id;
+      if(m_id<m_innerSize)
+      {
+        Scalar rhsVal = m_rhsEval.coeff(IsRowMajor?m_lhsIter.outer():m_id,
+                                        IsRowMajor?m_id:m_lhsIter.outer());
+        if(m_lhsIter && m_lhsIter.index()==m_id)
+        {
+          m_value = m_functor(m_lhsIter.value(), rhsVal);
+          ++m_lhsIter;
+        }
+        else
+          m_value = m_functor(Scalar(0),rhsVal);
+      }
+
+      return *this;
+    }
+
+    EIGEN_STRONG_INLINE Scalar value() const { eigen_internal_assert(m_id<m_innerSize); return m_value; }
+
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_id; }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
+    EIGEN_STRONG_INLINE Index row() const { return IsRowMajor ? m_lhsIter.outer() : m_id; }
+    EIGEN_STRONG_INLINE Index col() const { return IsRowMajor ? m_id : m_lhsIter.outer(); }
+
+    EIGEN_STRONG_INLINE operator bool() const { return m_id<m_innerSize; }
+
+  protected:
+    LhsIterator m_lhsIter;
+    const evaluator<Rhs> &m_rhsEval;
+    const BinaryOp& m_functor;
+    Scalar m_value;
+    StorageIndex m_id;
+    StorageIndex m_innerSize;
+  };
+
+
+  enum {
+    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    // Expose storage order of the sparse expression
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(Lhs::Flags)&RowMajorBit)
+  };
+
+  explicit binary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()),
+      m_rhsImpl(xpr.rhs()),
+      m_expr(xpr)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  inline Index nonZerosEstimate() const {
+    return m_expr.size();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<Lhs> m_lhsImpl;
+  evaluator<Rhs> m_rhsImpl;
+  const XprType &m_expr;
+};
+
+template<typename T,
+         typename LhsKind   = typename evaluator_traits<typename T::Lhs>::Kind,
+         typename RhsKind   = typename evaluator_traits<typename T::Rhs>::Kind,
+         typename LhsScalar = typename traits<typename T::Lhs>::Scalar,
+         typename RhsScalar = typename traits<typename T::Rhs>::Scalar> struct sparse_conjunction_evaluator;
+
+// "sparse .* sparse"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IteratorBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "dense .* sparse"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IndexBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "sparse .* dense"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IteratorBased, IndexBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+// "sparse ./ dense"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_quotient_op<T1,T2>, Lhs, Rhs>, IteratorBased, IndexBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_quotient_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_quotient_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+// "sparse && sparse"
+template<typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs>, IteratorBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "dense && sparse"
+template<typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs>, IndexBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "sparse && dense"
+template<typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs>, IteratorBased, IndexBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+// "sparse ^ sparse"
+template<typename XprType>
+struct sparse_conjunction_evaluator<XprType, IteratorBased, IteratorBased>
+  : evaluator_base<XprType>
+{
+protected:
+  typedef typename XprType::Functor BinaryOp;
+  typedef typename XprType::Lhs LhsArg;
+  typedef typename XprType::Rhs RhsArg;
+  typedef typename evaluator<LhsArg>::InnerIterator  LhsIterator;
+  typedef typename evaluator<RhsArg>::InnerIterator  RhsIterator;
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename traits<XprType>::Scalar Scalar;
+public:
+
+  class InnerIterator
+  {
+  public:
+    
+    EIGEN_STRONG_INLINE InnerIterator(const sparse_conjunction_evaluator& aEval, Index outer)
+      : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor)
     {
       while (m_lhsIter && m_rhsIter && (m_lhsIter.index() != m_rhsIter.index()))
       {
@@ -181,7 +428,7 @@ class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs,
       }
     }
 
-    EIGEN_STRONG_INLINE Derived& operator++()
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
     {
       ++m_lhsIter;
       ++m_rhsIter;
@@ -192,12 +439,13 @@ class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs,
         else
           ++m_rhsIter;
       }
-      return *static_cast<Derived*>(this);
+      return *this;
     }
-
+    
     EIGEN_STRONG_INLINE Scalar value() const { return m_functor(m_lhsIter.value(), m_rhsIter.value()); }
 
-    EIGEN_STRONG_INLINE Index index() const { return m_lhsIter.index(); }
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_lhsIter.index(); }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return m_lhsIter.row(); }
     EIGEN_STRONG_INLINE Index col() const { return m_lhsIter.col(); }
 
@@ -206,91 +454,184 @@ class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs,
   protected:
     LhsIterator m_lhsIter;
     RhsIterator m_rhsIter;
-    const BinaryFunc& m_functor;
+    const BinaryOp& m_functor;
+  };
+  
+  
+  enum {
+    CoeffReadCost = evaluator<LhsArg>::CoeffReadCost + evaluator<RhsArg>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    Flags = XprType::Flags
+  };
+  
+  explicit sparse_conjunction_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return (std::min)(m_lhsImpl.nonZerosEstimate(), m_rhsImpl.nonZerosEstimate());
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<LhsArg> m_lhsImpl;
+  evaluator<RhsArg> m_rhsImpl;
 };
 
-// sparse - dense  (product)
-template<typename T, typename Lhs, typename Rhs, typename Derived>
-class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs, Rhs, Derived, Sparse, Dense>
+// "dense ^ sparse"
+template<typename XprType>
+struct sparse_conjunction_evaluator<XprType, IndexBased, IteratorBased>
+  : evaluator_base<XprType>
 {
-    typedef scalar_product_op<T> BinaryFunc;
-    typedef CwiseBinaryOp<BinaryFunc, Lhs, Rhs> CwiseBinaryXpr;
-    typedef typename CwiseBinaryXpr::Scalar Scalar;
-    typedef typename traits<CwiseBinaryXpr>::_LhsNested _LhsNested;
-    typedef typename traits<CwiseBinaryXpr>::RhsNested RhsNested;
-    typedef typename _LhsNested::InnerIterator LhsIterator;
-    typedef typename Lhs::Index Index;
-    enum { IsRowMajor = (int(Lhs::Flags)&RowMajorBit)==RowMajorBit };
-  public:
+protected:
+  typedef typename XprType::Functor BinaryOp;
+  typedef typename XprType::Lhs LhsArg;
+  typedef typename XprType::Rhs RhsArg;
+  typedef evaluator<LhsArg> LhsEvaluator;
+  typedef typename evaluator<RhsArg>::InnerIterator  RhsIterator;
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename traits<XprType>::Scalar Scalar;
+public:
+
+  class InnerIterator
+  {
+    enum { IsRowMajor = (int(RhsArg::Flags)&RowMajorBit)==RowMajorBit };
 
-    EIGEN_STRONG_INLINE sparse_cwise_binary_op_inner_iterator_selector(const CwiseBinaryXpr& xpr, Index outer)
-      : m_rhs(xpr.rhs()), m_lhsIter(xpr.lhs(),outer), m_functor(xpr.functor()), m_outer(outer)
+  public:
+    
+    EIGEN_STRONG_INLINE InnerIterator(const sparse_conjunction_evaluator& aEval, Index outer)
+      : m_lhsEval(aEval.m_lhsImpl), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor), m_outer(outer)
     {}
 
-    EIGEN_STRONG_INLINE Derived& operator++()
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
     {
-      ++m_lhsIter;
-      return *static_cast<Derived*>(this);
+      ++m_rhsIter;
+      return *this;
     }
 
     EIGEN_STRONG_INLINE Scalar value() const
-    { return m_functor(m_lhsIter.value(),
-                       m_rhs.coeff(IsRowMajor?m_outer:m_lhsIter.index(),IsRowMajor?m_lhsIter.index():m_outer)); }
+    { return m_functor(m_lhsEval.coeff(IsRowMajor?m_outer:m_rhsIter.index(),IsRowMajor?m_rhsIter.index():m_outer), m_rhsIter.value()); }
 
-    EIGEN_STRONG_INLINE Index index() const { return m_lhsIter.index(); }
-    EIGEN_STRONG_INLINE Index row() const { return m_lhsIter.row(); }
-    EIGEN_STRONG_INLINE Index col() const { return m_lhsIter.col(); }
-
-    EIGEN_STRONG_INLINE operator bool() const { return m_lhsIter; }
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_rhsIter.index(); }
+    EIGEN_STRONG_INLINE Index outer() const { return m_rhsIter.outer(); }
+    EIGEN_STRONG_INLINE Index row() const { return m_rhsIter.row(); }
+    EIGEN_STRONG_INLINE Index col() const { return m_rhsIter.col(); }
 
+    EIGEN_STRONG_INLINE operator bool() const { return m_rhsIter; }
+    
   protected:
-    RhsNested m_rhs;
-    LhsIterator m_lhsIter;
-    const BinaryFunc m_functor;
+    const LhsEvaluator &m_lhsEval;
+    RhsIterator m_rhsIter;
+    const BinaryOp& m_functor;
     const Index m_outer;
+  };
+  
+  
+  enum {
+    CoeffReadCost = evaluator<LhsArg>::CoeffReadCost + evaluator<RhsArg>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    // Expose storage order of the sparse expression
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(RhsArg::Flags)&RowMajorBit)
+  };
+  
+  explicit sparse_conjunction_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_rhsImpl.nonZerosEstimate();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<LhsArg> m_lhsImpl;
+  evaluator<RhsArg> m_rhsImpl;
 };
 
-// sparse - dense  (product)
-template<typename T, typename Lhs, typename Rhs, typename Derived>
-class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs, Rhs, Derived, Dense, Sparse>
+// "sparse ^ dense"
+template<typename XprType>
+struct sparse_conjunction_evaluator<XprType, IteratorBased, IndexBased>
+  : evaluator_base<XprType>
 {
-    typedef scalar_product_op<T> BinaryFunc;
-    typedef CwiseBinaryOp<BinaryFunc, Lhs, Rhs> CwiseBinaryXpr;
-    typedef typename CwiseBinaryXpr::Scalar Scalar;
-    typedef typename traits<CwiseBinaryXpr>::_RhsNested _RhsNested;
-    typedef typename _RhsNested::InnerIterator RhsIterator;
-    typedef typename Lhs::Index Index;
+protected:
+  typedef typename XprType::Functor BinaryOp;
+  typedef typename XprType::Lhs LhsArg;
+  typedef typename XprType::Rhs RhsArg;
+  typedef typename evaluator<LhsArg>::InnerIterator LhsIterator;
+  typedef evaluator<RhsArg> RhsEvaluator;
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename traits<XprType>::Scalar Scalar;
+public:
+
+  class InnerIterator
+  {
+    enum { IsRowMajor = (int(LhsArg::Flags)&RowMajorBit)==RowMajorBit };
 
-    enum { IsRowMajor = (int(Rhs::Flags)&RowMajorBit)==RowMajorBit };
   public:
-
-    EIGEN_STRONG_INLINE sparse_cwise_binary_op_inner_iterator_selector(const CwiseBinaryXpr& xpr, Index outer)
-      : m_xpr(xpr), m_rhsIter(xpr.rhs(),outer), m_functor(xpr.functor()), m_outer(outer)
+    
+    EIGEN_STRONG_INLINE InnerIterator(const sparse_conjunction_evaluator& aEval, Index outer)
+      : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsEval(aEval.m_rhsImpl), m_functor(aEval.m_functor), m_outer(outer)
     {}
 
-    EIGEN_STRONG_INLINE Derived& operator++()
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
     {
-      ++m_rhsIter;
-      return *static_cast<Derived*>(this);
+      ++m_lhsIter;
+      return *this;
     }
 
     EIGEN_STRONG_INLINE Scalar value() const
-    { return m_functor(m_xpr.lhs().coeff(IsRowMajor?m_outer:m_rhsIter.index(),IsRowMajor?m_rhsIter.index():m_outer), m_rhsIter.value()); }
-
-    EIGEN_STRONG_INLINE Index index() const { return m_rhsIter.index(); }
-    EIGEN_STRONG_INLINE Index row() const { return m_rhsIter.row(); }
-    EIGEN_STRONG_INLINE Index col() const { return m_rhsIter.col(); }
+    { return m_functor(m_lhsIter.value(),
+                       m_rhsEval.coeff(IsRowMajor?m_outer:m_lhsIter.index(),IsRowMajor?m_lhsIter.index():m_outer)); }
 
-    EIGEN_STRONG_INLINE operator bool() const { return m_rhsIter; }
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_lhsIter.index(); }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
+    EIGEN_STRONG_INLINE Index row() const { return m_lhsIter.row(); }
+    EIGEN_STRONG_INLINE Index col() const { return m_lhsIter.col(); }
 
+    EIGEN_STRONG_INLINE operator bool() const { return m_lhsIter; }
+    
   protected:
-    const CwiseBinaryXpr& m_xpr;
-    RhsIterator m_rhsIter;
-    const BinaryFunc& m_functor;
+    LhsIterator m_lhsIter;
+    const evaluator<RhsArg> &m_rhsEval;
+    const BinaryOp& m_functor;
     const Index m_outer;
+  };
+  
+  
+  enum {
+    CoeffReadCost = evaluator<LhsArg>::CoeffReadCost + evaluator<RhsArg>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    // Expose storage order of the sparse expression
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(LhsArg::Flags)&RowMajorBit)
+  };
+  
+  explicit sparse_conjunction_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_lhsImpl.nonZerosEstimate();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<LhsArg> m_lhsImpl;
+  evaluator<RhsArg> m_rhsImpl;
 };
 
-} // end namespace internal
+}
 
 /***************************************************************************
 * Implementation of SparseMatrixBase and SparseCwise functions/operators
@@ -298,6 +639,22 @@ class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs,
 
 template<typename Derived>
 template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 SparseMatrixBase<Derived>::operator-=(const SparseMatrixBase<OtherDerived> &other)
 {
@@ -314,12 +671,56 @@ SparseMatrixBase<Derived>::operator+=(const SparseMatrixBase<OtherDerived>& othe
 
 template<typename Derived>
 template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator+=(const DiagonalBase<OtherDerived>& other)
+{
+  call_assignment_no_alias(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator-=(const DiagonalBase<OtherDerived>& other)
+{
+  call_assignment_no_alias(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+    
+template<typename Derived>
+template<typename OtherDerived>
 EIGEN_STRONG_INLINE const typename SparseMatrixBase<Derived>::template CwiseProductDenseReturnType<OtherDerived>::Type
 SparseMatrixBase<Derived>::cwiseProduct(const MatrixBase<OtherDerived> &other) const
 {
   return typename CwiseProductDenseReturnType<OtherDerived>::Type(derived(), other.derived());
 }
 
+template<typename DenseDerived, typename SparseDerived>
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_sum_op<typename DenseDerived::Scalar,typename SparseDerived::Scalar>, const DenseDerived, const SparseDerived>
+operator+(const MatrixBase<DenseDerived> &a, const SparseMatrixBase<SparseDerived> &b)
+{
+  return CwiseBinaryOp<internal::scalar_sum_op<typename DenseDerived::Scalar,typename SparseDerived::Scalar>, const DenseDerived, const SparseDerived>(a.derived(), b.derived());
+}
+
+template<typename SparseDerived, typename DenseDerived>
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_sum_op<typename SparseDerived::Scalar,typename DenseDerived::Scalar>, const SparseDerived, const DenseDerived>
+operator+(const SparseMatrixBase<SparseDerived> &a, const MatrixBase<DenseDerived> &b)
+{
+  return CwiseBinaryOp<internal::scalar_sum_op<typename SparseDerived::Scalar,typename DenseDerived::Scalar>, const SparseDerived, const DenseDerived>(a.derived(), b.derived());
+}
+
+template<typename DenseDerived, typename SparseDerived>
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_difference_op<typename DenseDerived::Scalar,typename SparseDerived::Scalar>, const DenseDerived, const SparseDerived>
+operator-(const MatrixBase<DenseDerived> &a, const SparseMatrixBase<SparseDerived> &b)
+{
+  return CwiseBinaryOp<internal::scalar_difference_op<typename DenseDerived::Scalar,typename SparseDerived::Scalar>, const DenseDerived, const SparseDerived>(a.derived(), b.derived());
+}
+
+template<typename SparseDerived, typename DenseDerived>
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_difference_op<typename SparseDerived::Scalar,typename DenseDerived::Scalar>, const SparseDerived, const DenseDerived>
+operator-(const SparseMatrixBase<SparseDerived> &a, const MatrixBase<DenseDerived> &b)
+{
+  return CwiseBinaryOp<internal::scalar_difference_op<typename SparseDerived::Scalar,typename DenseDerived::Scalar>, const SparseDerived, const DenseDerived>(a.derived(), b.derived());
+}
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSE_CWISE_BINARY_OP_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseCwiseUnaryOp.h b/extern/Eigen3/Eigen/src/SparseCore/SparseCwiseUnaryOp.h
index 5a50c780303..ea797379012 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseCwiseUnaryOp.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseCwiseUnaryOp.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,138 +12,121 @@
 
 namespace Eigen { 
 
-template<typename UnaryOp, typename MatrixType>
-class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>
-  : public SparseMatrixBase<CwiseUnaryOp<UnaryOp, MatrixType> >
+namespace internal {
+  
+template<typename UnaryOp, typename ArgType>
+struct unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>
+  : public evaluator_base<CwiseUnaryOp<UnaryOp,ArgType> >
 {
   public:
+    typedef CwiseUnaryOp<UnaryOp, ArgType> XprType;
 
     class InnerIterator;
-    class ReverseInnerIterator;
-
-    typedef CwiseUnaryOp<UnaryOp, MatrixType> Derived;
-    EIGEN_SPARSE_PUBLIC_INTERFACE(Derived)
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& op) : m_functor(op.functor()), m_argImpl(op.nestedExpression())
+    {
+      EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
+      EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+    }
+    
+    inline Index nonZerosEstimate() const {
+      return m_argImpl.nonZerosEstimate();
+    }
 
   protected:
-    typedef typename internal::traits<Derived>::_XprTypeNested _MatrixTypeNested;
-    typedef typename _MatrixTypeNested::InnerIterator MatrixTypeIterator;
-    typedef typename _MatrixTypeNested::ReverseInnerIterator MatrixTypeReverseIterator;
+    typedef typename evaluator<ArgType>::InnerIterator        EvalIterator;
+    
+    const UnaryOp m_functor;
+    evaluator<ArgType> m_argImpl;
 };
 
-template<typename UnaryOp, typename MatrixType>
-class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::InnerIterator
-    : public CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeIterator
+template<typename UnaryOp, typename ArgType>
+class unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::InnerIterator
+    : public unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::EvalIterator
 {
-    typedef typename CwiseUnaryOpImpl::Scalar Scalar;
-    typedef typename CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeIterator Base;
+    typedef typename XprType::Scalar Scalar;
+    typedef typename unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::EvalIterator Base;
   public:
 
-    EIGEN_STRONG_INLINE InnerIterator(const CwiseUnaryOpImpl& unaryOp, typename CwiseUnaryOpImpl::Index outer)
-      : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
+    EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& unaryOp, Index outer)
+      : Base(unaryOp.m_argImpl,outer), m_functor(unaryOp.m_functor)
     {}
 
     EIGEN_STRONG_INLINE InnerIterator& operator++()
     { Base::operator++(); return *this; }
 
-    EIGEN_STRONG_INLINE typename CwiseUnaryOpImpl::Scalar value() const { return m_functor(Base::value()); }
-
-  protected:
-    const UnaryOp m_functor;
-  private:
-    typename CwiseUnaryOpImpl::Scalar& valueRef();
-};
-
-template<typename UnaryOp, typename MatrixType>
-class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::ReverseInnerIterator
-    : public CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeReverseIterator
-{
-    typedef typename CwiseUnaryOpImpl::Scalar Scalar;
-    typedef typename CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeReverseIterator Base;
-  public:
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator(const CwiseUnaryOpImpl& unaryOp, typename CwiseUnaryOpImpl::Index outer)
-      : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
-    {}
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
-    { Base::operator--(); return *this; }
-
-    EIGEN_STRONG_INLINE typename CwiseUnaryOpImpl::Scalar value() const { return m_functor(Base::value()); }
+    EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
 
   protected:
     const UnaryOp m_functor;
   private:
-    typename CwiseUnaryOpImpl::Scalar& valueRef();
+    Scalar& valueRef();
 };
 
-template<typename ViewOp, typename MatrixType>
-class CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>
-  : public SparseMatrixBase<CwiseUnaryView<ViewOp, MatrixType> >
+template<typename ViewOp, typename ArgType>
+struct unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>
+  : public evaluator_base<CwiseUnaryView<ViewOp,ArgType> >
 {
   public:
+    typedef CwiseUnaryView<ViewOp, ArgType> XprType;
 
     class InnerIterator;
-    class ReverseInnerIterator;
-
-    typedef CwiseUnaryView<ViewOp, MatrixType> Derived;
-    EIGEN_SPARSE_PUBLIC_INTERFACE(Derived)
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<ViewOp>::Cost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& op) : m_functor(op.functor()), m_argImpl(op.nestedExpression())
+    {
+      EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<ViewOp>::Cost);
+      EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+    }
 
   protected:
-    typedef typename internal::traits<Derived>::_MatrixTypeNested _MatrixTypeNested;
-    typedef typename _MatrixTypeNested::InnerIterator MatrixTypeIterator;
-    typedef typename _MatrixTypeNested::ReverseInnerIterator MatrixTypeReverseIterator;
+    typedef typename evaluator<ArgType>::InnerIterator        EvalIterator;
+    
+    const ViewOp m_functor;
+    evaluator<ArgType> m_argImpl;
 };
 
-template<typename ViewOp, typename MatrixType>
-class CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::InnerIterator
-    : public CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeIterator
+template<typename ViewOp, typename ArgType>
+class unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::InnerIterator
+    : public unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::EvalIterator
 {
-    typedef typename CwiseUnaryViewImpl::Scalar Scalar;
-    typedef typename CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeIterator Base;
+    typedef typename XprType::Scalar Scalar;
+    typedef typename unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::EvalIterator Base;
   public:
 
-    EIGEN_STRONG_INLINE InnerIterator(const CwiseUnaryViewImpl& unaryOp, typename CwiseUnaryViewImpl::Index outer)
-      : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
+    EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& unaryOp, Index outer)
+      : Base(unaryOp.m_argImpl,outer), m_functor(unaryOp.m_functor)
     {}
 
     EIGEN_STRONG_INLINE InnerIterator& operator++()
     { Base::operator++(); return *this; }
 
-    EIGEN_STRONG_INLINE typename CwiseUnaryViewImpl::Scalar value() const { return m_functor(Base::value()); }
-    EIGEN_STRONG_INLINE typename CwiseUnaryViewImpl::Scalar& valueRef() { return m_functor(Base::valueRef()); }
+    EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
+    EIGEN_STRONG_INLINE Scalar& valueRef() { return m_functor(Base::valueRef()); }
 
   protected:
     const ViewOp m_functor;
 };
 
-template<typename ViewOp, typename MatrixType>
-class CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::ReverseInnerIterator
-    : public CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeReverseIterator
-{
-    typedef typename CwiseUnaryViewImpl::Scalar Scalar;
-    typedef typename CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeReverseIterator Base;
-  public:
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator(const CwiseUnaryViewImpl& unaryOp, typename CwiseUnaryViewImpl::Index outer)
-      : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
-    {}
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
-    { Base::operator--(); return *this; }
-
-    EIGEN_STRONG_INLINE typename CwiseUnaryViewImpl::Scalar value() const { return m_functor(Base::value()); }
-    EIGEN_STRONG_INLINE typename CwiseUnaryViewImpl::Scalar& valueRef() { return m_functor(Base::valueRef()); }
-
-  protected:
-    const ViewOp m_functor;
-};
+} // end namespace internal
 
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
 SparseMatrixBase<Derived>::operator*=(const Scalar& other)
 {
+  typedef typename internal::evaluator<Derived>::InnerIterator EvalIterator;
+  internal::evaluator<Derived> thisEval(derived());
   for (Index j=0; j<outerSize(); ++j)
-    for (typename Derived::InnerIterator i(derived(),j); i; ++i)
+    for (EvalIterator i(thisEval,j); i; ++i)
       i.valueRef() *= other;
   return derived();
 }
@@ -152,8 +135,10 @@ template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
 SparseMatrixBase<Derived>::operator/=(const Scalar& other)
 {
+  typedef typename internal::evaluator<Derived>::InnerIterator EvalIterator;
+  internal::evaluator<Derived> thisEval(derived());
   for (Index j=0; j<outerSize(); ++j)
-    for (typename Derived::InnerIterator i(derived(),j); i; ++i)
+    for (EvalIterator i(thisEval,j); i; ++i)
       i.valueRef() /= other;
   return derived();
 }
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseDenseProduct.h b/extern/Eigen3/Eigen/src/SparseCore/SparseDenseProduct.h
index ccb6ae7b788..0547db59689 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseDenseProduct.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseDenseProduct.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,196 +12,93 @@
 
 namespace Eigen { 
 
-template<typename Lhs, typename Rhs, int InnerSize> struct SparseDenseProductReturnType
-{
-  typedef SparseTimeDenseProduct<Lhs,Rhs> Type;
-};
-
-template<typename Lhs, typename Rhs> struct SparseDenseProductReturnType<Lhs,Rhs,1>
-{
-  typedef typename internal::conditional<
-    Lhs::IsRowMajor,
-    SparseDenseOuterProduct<Rhs,Lhs,true>,
-    SparseDenseOuterProduct<Lhs,Rhs,false> >::type Type;
-};
-
-template<typename Lhs, typename Rhs, int InnerSize> struct DenseSparseProductReturnType
-{
-  typedef DenseTimeSparseProduct<Lhs,Rhs> Type;
-};
-
-template<typename Lhs, typename Rhs> struct DenseSparseProductReturnType<Lhs,Rhs,1>
-{
-  typedef typename internal::conditional<
-    Rhs::IsRowMajor,
-    SparseDenseOuterProduct<Rhs,Lhs,true>,
-    SparseDenseOuterProduct<Lhs,Rhs,false> >::type Type;
-};
-
 namespace internal {
 
-template<typename Lhs, typename Rhs, bool Tr>
-struct traits<SparseDenseOuterProduct<Lhs,Rhs,Tr> >
-{
-  typedef Sparse StorageKind;
-  typedef typename scalar_product_traits<typename traits<Lhs>::Scalar,
-                                         typename traits<Rhs>::Scalar>::ReturnType Scalar;
-  typedef typename Lhs::Index Index;
-  typedef typename Lhs::Nested LhsNested;
-  typedef typename Rhs::Nested RhsNested;
-  typedef typename remove_all<LhsNested>::type _LhsNested;
-  typedef typename remove_all<RhsNested>::type _RhsNested;
-
-  enum {
-    LhsCoeffReadCost = traits<_LhsNested>::CoeffReadCost,
-    RhsCoeffReadCost = traits<_RhsNested>::CoeffReadCost,
-
-    RowsAtCompileTime    = Tr ? int(traits<Rhs>::RowsAtCompileTime)     : int(traits<Lhs>::RowsAtCompileTime),
-    ColsAtCompileTime    = Tr ? int(traits<Lhs>::ColsAtCompileTime)     : int(traits<Rhs>::ColsAtCompileTime),
-    MaxRowsAtCompileTime = Tr ? int(traits<Rhs>::MaxRowsAtCompileTime)  : int(traits<Lhs>::MaxRowsAtCompileTime),
-    MaxColsAtCompileTime = Tr ? int(traits<Lhs>::MaxColsAtCompileTime)  : int(traits<Rhs>::MaxColsAtCompileTime),
-
-    Flags = Tr ? RowMajorBit : 0,
-
-    CoeffReadCost = LhsCoeffReadCost + RhsCoeffReadCost + NumTraits<Scalar>::MulCost
-  };
-};
-
-} // end namespace internal
-
-template<typename Lhs, typename Rhs, bool Tr>
-class SparseDenseOuterProduct
- : public SparseMatrixBase<SparseDenseOuterProduct<Lhs,Rhs,Tr> >
-{
-  public:
-
-    typedef SparseMatrixBase<SparseDenseOuterProduct> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(SparseDenseOuterProduct)
-    typedef internal::traits<SparseDenseOuterProduct> Traits;
-
-  private:
-
-    typedef typename Traits::LhsNested LhsNested;
-    typedef typename Traits::RhsNested RhsNested;
-    typedef typename Traits::_LhsNested _LhsNested;
-    typedef typename Traits::_RhsNested _RhsNested;
-
-  public:
-
-    class InnerIterator;
-
-    EIGEN_STRONG_INLINE SparseDenseOuterProduct(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      EIGEN_STATIC_ASSERT(!Tr,YOU_MADE_A_PROGRAMMING_MISTAKE);
-    }
-
-    EIGEN_STRONG_INLINE SparseDenseOuterProduct(const Rhs& rhs, const Lhs& lhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      EIGEN_STATIC_ASSERT(Tr,YOU_MADE_A_PROGRAMMING_MISTAKE);
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return Tr ? m_rhs.rows() : m_lhs.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return Tr ? m_lhs.cols() : m_rhs.cols(); }
-
-    EIGEN_STRONG_INLINE const _LhsNested& lhs() const { return m_lhs; }
-    EIGEN_STRONG_INLINE const _RhsNested& rhs() const { return m_rhs; }
-
-  protected:
-    LhsNested m_lhs;
-    RhsNested m_rhs;
-};
-
-template<typename Lhs, typename Rhs, bool Transpose>
-class SparseDenseOuterProduct<Lhs,Rhs,Transpose>::InnerIterator : public _LhsNested::InnerIterator
-{
-    typedef typename _LhsNested::InnerIterator Base;
-    typedef typename SparseDenseOuterProduct::Index Index;
-  public:
-    EIGEN_STRONG_INLINE InnerIterator(const SparseDenseOuterProduct& prod, Index outer)
-      : Base(prod.lhs(), 0), m_outer(outer), m_factor(get(prod.rhs(), outer, typename internal::traits<Rhs>::StorageKind() ))
-    { }
-
-    inline Index outer() const { return m_outer; }
-    inline Index row() const { return Transpose ? m_outer : Base::index(); }
-    inline Index col() const { return Transpose ? Base::index() : m_outer; }
-
-    inline Scalar value() const { return Base::value() * m_factor; }
-
-  protected:
-    static Scalar get(const _RhsNested &rhs, Index outer, Dense = Dense())
-    {
-      return rhs.coeff(outer);
-    }
-    
-    static Scalar get(const _RhsNested &rhs, Index outer, Sparse = Sparse())
-    {
-      typename Traits::_RhsNested::InnerIterator it(rhs, outer);
-      if (it && it.index()==0)
-        return it.value();
-      
-      return Scalar(0);
-    }
-    
-    Index m_outer;
-    Scalar m_factor;
-};
-
-namespace internal {
-template<typename Lhs, typename Rhs>
-struct traits<SparseTimeDenseProduct<Lhs,Rhs> >
- : traits<ProductBase<SparseTimeDenseProduct<Lhs,Rhs>, Lhs, Rhs> >
-{
-  typedef Dense StorageKind;
-  typedef MatrixXpr XprKind;
-};
+template <> struct product_promote_storage_type<Sparse,Dense, OuterProduct> { typedef Sparse ret; };
+template <> struct product_promote_storage_type<Dense,Sparse, OuterProduct> { typedef Sparse ret; };
 
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType,
+         typename AlphaType,
          int LhsStorageOrder = ((SparseLhsType::Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor,
          bool ColPerCol = ((DenseRhsType::Flags&RowMajorBit)==0) || DenseRhsType::ColsAtCompileTime==1>
 struct sparse_time_dense_product_impl;
 
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType>
-struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, RowMajor, true>
+struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, typename DenseResType::Scalar, RowMajor, true>
 {
   typedef typename internal::remove_all<SparseLhsType>::type Lhs;
   typedef typename internal::remove_all<DenseRhsType>::type Rhs;
   typedef typename internal::remove_all<DenseResType>::type Res;
-  typedef typename Lhs::Index Index;
-  typedef typename Lhs::InnerIterator LhsInnerIterator;
+  typedef typename evaluator<Lhs>::InnerIterator LhsInnerIterator;
+  typedef evaluator<Lhs> LhsEval;
   static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha)
   {
+    LhsEval lhsEval(lhs);
+    
+    Index n = lhs.outerSize();
+#ifdef EIGEN_HAS_OPENMP
+    Eigen::initParallel();
+    Index threads = Eigen::nbThreads();
+#endif
+    
     for(Index c=0; c<rhs.cols(); ++c)
     {
-      Index n = lhs.outerSize();
-      for(Index j=0; j<n; ++j)
+#ifdef EIGEN_HAS_OPENMP
+      // This 20000 threshold has been found experimentally on 2D and 3D Poisson problems.
+      // It basically represents the minimal amount of work to be done to be worth it.
+      if(threads>1 && lhsEval.nonZerosEstimate() > 20000)
+      {
+        #pragma omp parallel for schedule(dynamic,(n+threads*4-1)/(threads*4)) num_threads(threads)
+        for(Index i=0; i<n; ++i)
+          processRow(lhsEval,rhs,res,alpha,i,c);
+      }
+      else
+#endif
       {
-        typename Res::Scalar tmp(0);
-        for(LhsInnerIterator it(lhs,j); it ;++it)
-          tmp += it.value() * rhs.coeff(it.index(),c);
-        res.coeffRef(j,c) += alpha * tmp;
+        for(Index i=0; i<n; ++i)
+          processRow(lhsEval,rhs,res,alpha,i,c);
       }
     }
   }
+  
+  static void processRow(const LhsEval& lhsEval, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha, Index i, Index col)
+  {
+    typename Res::Scalar tmp(0);
+    for(LhsInnerIterator it(lhsEval,i); it ;++it)
+      tmp += it.value() * rhs.coeff(it.index(),col);
+    res.coeffRef(i,col) += alpha * tmp;
+  }
+  
 };
 
-template<typename SparseLhsType, typename DenseRhsType, typename DenseResType>
-struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, ColMajor, true>
+// FIXME: what is the purpose of the following specialization? Is it for the BlockedSparse format?
+// -> let's disable it for now as it is conflicting with generic scalar*matrix and matrix*scalar operators
+// template<typename T1, typename T2/*, int _Options, typename _StrideType*/>
+// struct ScalarBinaryOpTraits<T1, Ref<T2/*, _Options, _StrideType*/> >
+// {
+//   enum {
+//     Defined = 1
+//   };
+//   typedef typename CwiseUnaryOp<scalar_multiple2_op<T1, typename T2::Scalar>, T2>::PlainObject ReturnType;
+// };
+
+template<typename SparseLhsType, typename DenseRhsType, typename DenseResType, typename AlphaType>
+struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, AlphaType, ColMajor, true>
 {
   typedef typename internal::remove_all<SparseLhsType>::type Lhs;
   typedef typename internal::remove_all<DenseRhsType>::type Rhs;
   typedef typename internal::remove_all<DenseResType>::type Res;
-  typedef typename Lhs::InnerIterator LhsInnerIterator;
-  typedef typename Lhs::Index Index;
-  static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha)
+  typedef typename evaluator<Lhs>::InnerIterator LhsInnerIterator;
+  static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const AlphaType& alpha)
   {
+    evaluator<Lhs> lhsEval(lhs);
     for(Index c=0; c<rhs.cols(); ++c)
     {
       for(Index j=0; j<lhs.outerSize(); ++j)
       {
-        typename Res::Scalar rhs_j = alpha * rhs.coeff(j,c);
-        for(LhsInnerIterator it(lhs,j); it ;++it)
+//        typename Res::Scalar rhs_j = alpha * rhs.coeff(j,c);
+        typename ScalarBinaryOpTraits<AlphaType, typename Rhs::Scalar>::ReturnType rhs_j(alpha * rhs.coeff(j,c));
+        for(LhsInnerIterator it(lhsEval,j); it ;++it)
           res.coeffRef(it.index(),c) += it.value() * rhs_j;
       }
     }
@@ -209,38 +106,38 @@ struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, C
 };
 
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType>
-struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, RowMajor, false>
+struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, typename DenseResType::Scalar, RowMajor, false>
 {
   typedef typename internal::remove_all<SparseLhsType>::type Lhs;
   typedef typename internal::remove_all<DenseRhsType>::type Rhs;
   typedef typename internal::remove_all<DenseResType>::type Res;
-  typedef typename Lhs::InnerIterator LhsInnerIterator;
-  typedef typename Lhs::Index Index;
+  typedef typename evaluator<Lhs>::InnerIterator LhsInnerIterator;
   static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha)
   {
+    evaluator<Lhs> lhsEval(lhs);
     for(Index j=0; j<lhs.outerSize(); ++j)
     {
       typename Res::RowXpr res_j(res.row(j));
-      for(LhsInnerIterator it(lhs,j); it ;++it)
+      for(LhsInnerIterator it(lhsEval,j); it ;++it)
         res_j += (alpha*it.value()) * rhs.row(it.index());
     }
   }
 };
 
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType>
-struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, ColMajor, false>
+struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, typename DenseResType::Scalar, ColMajor, false>
 {
   typedef typename internal::remove_all<SparseLhsType>::type Lhs;
   typedef typename internal::remove_all<DenseRhsType>::type Rhs;
   typedef typename internal::remove_all<DenseResType>::type Res;
-  typedef typename Lhs::InnerIterator LhsInnerIterator;
-  typedef typename Lhs::Index Index;
+  typedef typename evaluator<Lhs>::InnerIterator LhsInnerIterator;
   static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha)
   {
+    evaluator<Lhs> lhsEval(lhs);
     for(Index j=0; j<lhs.outerSize(); ++j)
     {
       typename Rhs::ConstRowXpr rhs_j(rhs.row(j));
-      for(LhsInnerIterator it(lhs,j); it ;++it)
+      for(LhsInnerIterator it(lhsEval,j); it ;++it)
         res.row(it.index()) += (alpha*it.value()) * rhs_j;
     }
   }
@@ -249,63 +146,175 @@ struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, C
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType,typename AlphaType>
 inline void sparse_time_dense_product(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const AlphaType& alpha)
 {
-  sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType>::run(lhs, rhs, res, alpha);
+  sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, AlphaType>::run(lhs, rhs, res, alpha);
 }
 
 } // end namespace internal
 
-template<typename Lhs, typename Rhs>
-class SparseTimeDenseProduct
-  : public ProductBase<SparseTimeDenseProduct<Lhs,Rhs>, Lhs, Rhs>
-{
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(SparseTimeDenseProduct)
-
-    SparseTimeDenseProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {}
-
-    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
-    {
-      internal::sparse_time_dense_product(m_lhs, m_rhs, dest, alpha);
-    }
+namespace internal {
 
-  private:
-    SparseTimeDenseProduct& operator=(const SparseTimeDenseProduct&);
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseShape, DenseShape, ProductType>
+ : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,SparseShape,DenseShape,ProductType> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    typedef typename nested_eval<Lhs,((Rhs::Flags&RowMajorBit)==0) ? 1 : Rhs::ColsAtCompileTime>::type LhsNested;
+    typedef typename nested_eval<Rhs,((Lhs::Flags&RowMajorBit)==0) ? 1 : Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhs);
+    internal::sparse_time_dense_product(lhsNested, rhsNested, dst, alpha);
+  }
 };
 
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseTriangularShape, DenseShape, ProductType>
+  : generic_product_impl<Lhs, Rhs, SparseShape, DenseShape, ProductType>
+{};
 
-// dense = dense * sparse
-namespace internal {
-template<typename Lhs, typename Rhs>
-struct traits<DenseTimeSparseProduct<Lhs,Rhs> >
- : traits<ProductBase<DenseTimeSparseProduct<Lhs,Rhs>, Lhs, Rhs> >
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, DenseShape, SparseShape, ProductType>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,SparseShape,ProductType> >
 {
-  typedef Dense StorageKind;
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dst>
+  static void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    typedef typename nested_eval<Lhs,((Rhs::Flags&RowMajorBit)==0) ? Dynamic : 1>::type LhsNested;
+    typedef typename nested_eval<Rhs,((Lhs::Flags&RowMajorBit)==RowMajorBit) ? 1 : Lhs::RowsAtCompileTime>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhs);
+    
+    // transpose everything
+    Transpose<Dst> dstT(dst);
+    internal::sparse_time_dense_product(rhsNested.transpose(), lhsNested.transpose(), dstT, alpha);
+  }
 };
-} // end namespace internal
 
-template<typename Lhs, typename Rhs>
-class DenseTimeSparseProduct
-  : public ProductBase<DenseTimeSparseProduct<Lhs,Rhs>, Lhs, Rhs>
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, DenseShape, SparseTriangularShape, ProductType>
+  : generic_product_impl<Lhs, Rhs, DenseShape, SparseShape, ProductType>
+{};
+
+template<typename LhsT, typename RhsT, bool NeedToTranspose>
+struct sparse_dense_outer_product_evaluator
 {
+protected:
+  typedef typename conditional<NeedToTranspose,RhsT,LhsT>::type Lhs1;
+  typedef typename conditional<NeedToTranspose,LhsT,RhsT>::type ActualRhs;
+  typedef Product<LhsT,RhsT,DefaultProduct> ProdXprType;
+  
+  // if the actual left-hand side is a dense vector,
+  // then build a sparse-view so that we can seamlessly iterate over it.
+  typedef typename conditional<is_same<typename internal::traits<Lhs1>::StorageKind,Sparse>::value,
+            Lhs1, SparseView<Lhs1> >::type ActualLhs;
+  typedef typename conditional<is_same<typename internal::traits<Lhs1>::StorageKind,Sparse>::value,
+            Lhs1 const&, SparseView<Lhs1> >::type LhsArg;
+            
+  typedef evaluator<ActualLhs> LhsEval;
+  typedef evaluator<ActualRhs> RhsEval;
+  typedef typename evaluator<ActualLhs>::InnerIterator LhsIterator;
+  typedef typename ProdXprType::Scalar Scalar;
+  
+public:
+  enum {
+    Flags = NeedToTranspose ? RowMajorBit : 0,
+    CoeffReadCost = HugeCost
+  };
+  
+  class InnerIterator : public LhsIterator
+  {
   public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(DenseTimeSparseProduct)
-
-    DenseTimeSparseProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
+    InnerIterator(const sparse_dense_outer_product_evaluator &xprEval, Index outer)
+      : LhsIterator(xprEval.m_lhsXprImpl, 0),
+        m_outer(outer),
+        m_empty(false),
+        m_factor(get(xprEval.m_rhsXprImpl, outer, typename internal::traits<ActualRhs>::StorageKind() ))
     {}
+    
+    EIGEN_STRONG_INLINE Index outer() const { return m_outer; }
+    EIGEN_STRONG_INLINE Index row()   const { return NeedToTranspose ? m_outer : LhsIterator::index(); }
+    EIGEN_STRONG_INLINE Index col()   const { return NeedToTranspose ? LhsIterator::index() : m_outer; }
 
-    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+    EIGEN_STRONG_INLINE Scalar value() const { return LhsIterator::value() * m_factor; }
+    EIGEN_STRONG_INLINE operator bool() const { return LhsIterator::operator bool() && (!m_empty); }
+    
+  protected:
+    Scalar get(const RhsEval &rhs, Index outer, Dense = Dense()) const
+    {
+      return rhs.coeff(outer);
+    }
+    
+    Scalar get(const RhsEval &rhs, Index outer, Sparse = Sparse())
     {
-      Transpose<const _LhsNested> lhs_t(m_lhs);
-      Transpose<const _RhsNested> rhs_t(m_rhs);
-      Transpose<Dest> dest_t(dest);
-      internal::sparse_time_dense_product(rhs_t, lhs_t, dest_t, alpha);
+      typename RhsEval::InnerIterator it(rhs, outer);
+      if (it && it.index()==0 && it.value()!=Scalar(0))
+        return it.value();
+      m_empty = true;
+      return Scalar(0);
     }
+    
+    Index m_outer;
+    bool m_empty;
+    Scalar m_factor;
+  };
+  
+  sparse_dense_outer_product_evaluator(const Lhs1 &lhs, const ActualRhs &rhs)
+     : m_lhs(lhs), m_lhsXprImpl(m_lhs), m_rhsXprImpl(rhs)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  // transpose case
+  sparse_dense_outer_product_evaluator(const ActualRhs &rhs, const Lhs1 &lhs)
+     : m_lhs(lhs), m_lhsXprImpl(m_lhs), m_rhsXprImpl(rhs)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+    
+protected:
+  const LhsArg m_lhs;
+  evaluator<ActualLhs> m_lhsXprImpl;
+  evaluator<ActualRhs> m_rhsXprImpl;
+};
 
-  private:
-    DenseTimeSparseProduct& operator=(const DenseTimeSparseProduct&);
+// sparse * dense outer product
+template<typename Lhs, typename Rhs>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, OuterProduct, SparseShape, DenseShape>
+  : sparse_dense_outer_product_evaluator<Lhs,Rhs, Lhs::IsRowMajor>
+{
+  typedef sparse_dense_outer_product_evaluator<Lhs,Rhs, Lhs::IsRowMajor> Base;
+  
+  typedef Product<Lhs, Rhs> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+
+  explicit product_evaluator(const XprType& xpr)
+    : Base(xpr.lhs(), xpr.rhs())
+  {}
+  
 };
 
+template<typename Lhs, typename Rhs>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, OuterProduct, DenseShape, SparseShape>
+  : sparse_dense_outer_product_evaluator<Lhs,Rhs, Rhs::IsRowMajor>
+{
+  typedef sparse_dense_outer_product_evaluator<Lhs,Rhs, Rhs::IsRowMajor> Base;
+  
+  typedef Product<Lhs, Rhs> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+
+  explicit product_evaluator(const XprType& xpr)
+    : Base(xpr.lhs(), xpr.rhs())
+  {}
+  
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSEDENSEPRODUCT_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseDiagonalProduct.h b/extern/Eigen3/Eigen/src/SparseCore/SparseDiagonalProduct.h
index 1bb590e64d4..941c03be3de 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseDiagonalProduct.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseDiagonalProduct.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -26,171 +26,113 @@ namespace Eigen {
 
 namespace internal {
 
-template<typename Lhs, typename Rhs>
-struct traits<SparseDiagonalProduct<Lhs, Rhs> >
-{
-  typedef typename remove_all<Lhs>::type _Lhs;
-  typedef typename remove_all<Rhs>::type _Rhs;
-  typedef typename _Lhs::Scalar Scalar;
-  typedef typename promote_index_type<typename traits<Lhs>::Index,
-                                         typename traits<Rhs>::Index>::type Index;
-  typedef Sparse StorageKind;
-  typedef MatrixXpr XprKind;
-  enum {
-    RowsAtCompileTime = _Lhs::RowsAtCompileTime,
-    ColsAtCompileTime = _Rhs::ColsAtCompileTime,
-
-    MaxRowsAtCompileTime = _Lhs::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = _Rhs::MaxColsAtCompileTime,
-
-    SparseFlags = is_diagonal<_Lhs>::ret ? int(_Rhs::Flags) : int(_Lhs::Flags),
-    Flags = (SparseFlags&RowMajorBit),
-    CoeffReadCost = Dynamic
-  };
+enum {
+  SDP_AsScalarProduct,
+  SDP_AsCwiseProduct
 };
+  
+template<typename SparseXprType, typename DiagonalCoeffType, int SDP_Tag>
+struct sparse_diagonal_product_evaluator;
 
-enum {SDP_IsDiagonal, SDP_IsSparseRowMajor, SDP_IsSparseColMajor};
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType, int RhsMode, int LhsMode>
-class sparse_diagonal_product_inner_iterator_selector;
-
-} // end namespace internal
-
-template<typename Lhs, typename Rhs>
-class SparseDiagonalProduct
-  : public SparseMatrixBase<SparseDiagonalProduct<Lhs,Rhs> >,
-    internal::no_assignment_operator
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, ProductTag, DiagonalShape, SparseShape>
+  : public sparse_diagonal_product_evaluator<Rhs, typename Lhs::DiagonalVectorType, Rhs::Flags&RowMajorBit?SDP_AsScalarProduct:SDP_AsCwiseProduct>
 {
-    typedef typename Lhs::Nested LhsNested;
-    typedef typename Rhs::Nested RhsNested;
-
-    typedef typename internal::remove_all<LhsNested>::type _LhsNested;
-    typedef typename internal::remove_all<RhsNested>::type _RhsNested;
-
-    enum {
-      LhsMode = internal::is_diagonal<_LhsNested>::ret ? internal::SDP_IsDiagonal
-              : (_LhsNested::Flags&RowMajorBit) ? internal::SDP_IsSparseRowMajor : internal::SDP_IsSparseColMajor,
-      RhsMode = internal::is_diagonal<_RhsNested>::ret ? internal::SDP_IsDiagonal
-              : (_RhsNested::Flags&RowMajorBit) ? internal::SDP_IsSparseRowMajor : internal::SDP_IsSparseColMajor
-    };
-
-  public:
-
-    EIGEN_SPARSE_PUBLIC_INTERFACE(SparseDiagonalProduct)
-
-    typedef internal::sparse_diagonal_product_inner_iterator_selector
-                      <_LhsNested,_RhsNested,SparseDiagonalProduct,LhsMode,RhsMode> InnerIterator;
-    
-    // We do not want ReverseInnerIterator for diagonal-sparse products,
-    // but this dummy declaration is needed to make diag * sparse * diag compile.
-    class ReverseInnerIterator;
-
-    EIGEN_STRONG_INLINE SparseDiagonalProduct(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      eigen_assert(lhs.cols() == rhs.rows() && "invalid sparse matrix * diagonal matrix product");
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
-
-    EIGEN_STRONG_INLINE const _LhsNested& lhs() const { return m_lhs; }
-    EIGEN_STRONG_INLINE const _RhsNested& rhs() const { return m_rhs; }
-
-  protected:
-    LhsNested m_lhs;
-    RhsNested m_rhs;
+  typedef Product<Lhs, Rhs, DefaultProduct> XprType;
+  enum { CoeffReadCost = HugeCost, Flags = Rhs::Flags&RowMajorBit, Alignment = 0 }; // FIXME CoeffReadCost & Flags
+  
+  typedef sparse_diagonal_product_evaluator<Rhs, typename Lhs::DiagonalVectorType, Rhs::Flags&RowMajorBit?SDP_AsScalarProduct:SDP_AsCwiseProduct> Base;
+  explicit product_evaluator(const XprType& xpr) : Base(xpr.rhs(), xpr.lhs().diagonal()) {}
 };
 
-namespace internal {
-
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType>
-class sparse_diagonal_product_inner_iterator_selector
-<Lhs,Rhs,SparseDiagonalProductType,SDP_IsDiagonal,SDP_IsSparseRowMajor>
-  : public CwiseUnaryOp<scalar_multiple_op<typename Lhs::Scalar>,const Rhs>::InnerIterator
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, ProductTag, SparseShape, DiagonalShape>
+  : public sparse_diagonal_product_evaluator<Lhs, Transpose<const typename Rhs::DiagonalVectorType>, Lhs::Flags&RowMajorBit?SDP_AsCwiseProduct:SDP_AsScalarProduct>
 {
-    typedef typename CwiseUnaryOp<scalar_multiple_op<typename Lhs::Scalar>,const Rhs>::InnerIterator Base;
-    typedef typename Lhs::Index Index;
-  public:
-    inline sparse_diagonal_product_inner_iterator_selector(
-              const SparseDiagonalProductType& expr, Index outer)
-      : Base(expr.rhs()*(expr.lhs().diagonal().coeff(outer)), outer)
-    {}
+  typedef Product<Lhs, Rhs, DefaultProduct> XprType;
+  enum { CoeffReadCost = HugeCost, Flags = Lhs::Flags&RowMajorBit, Alignment = 0 }; // FIXME CoeffReadCost & Flags
+  
+  typedef sparse_diagonal_product_evaluator<Lhs, Transpose<const typename Rhs::DiagonalVectorType>, Lhs::Flags&RowMajorBit?SDP_AsCwiseProduct:SDP_AsScalarProduct> Base;
+  explicit product_evaluator(const XprType& xpr) : Base(xpr.lhs(), xpr.rhs().diagonal().transpose()) {}
 };
 
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType>
-class sparse_diagonal_product_inner_iterator_selector
-<Lhs,Rhs,SparseDiagonalProductType,SDP_IsDiagonal,SDP_IsSparseColMajor>
-  : public CwiseBinaryOp<
-      scalar_product_op<typename Lhs::Scalar>,
-      const typename Rhs::ConstInnerVectorReturnType,
-      const typename Lhs::DiagonalVectorType>::InnerIterator
+template<typename SparseXprType, typename DiagonalCoeffType>
+struct sparse_diagonal_product_evaluator<SparseXprType, DiagonalCoeffType, SDP_AsScalarProduct>
 {
-    typedef typename CwiseBinaryOp<
-      scalar_product_op<typename Lhs::Scalar>,
-      const typename Rhs::ConstInnerVectorReturnType,
-      const typename Lhs::DiagonalVectorType>::InnerIterator Base;
-    typedef typename Lhs::Index Index;
-    Index m_outer;
+protected:
+  typedef typename evaluator<SparseXprType>::InnerIterator SparseXprInnerIterator;
+  typedef typename SparseXprType::Scalar Scalar;
+  
+public:
+  class InnerIterator : public SparseXprInnerIterator
+  {
   public:
-    inline sparse_diagonal_product_inner_iterator_selector(
-              const SparseDiagonalProductType& expr, Index outer)
-      : Base(expr.rhs().innerVector(outer) .cwiseProduct(expr.lhs().diagonal()), 0), m_outer(outer)
+    InnerIterator(const sparse_diagonal_product_evaluator &xprEval, Index outer)
+      : SparseXprInnerIterator(xprEval.m_sparseXprImpl, outer),
+        m_coeff(xprEval.m_diagCoeffImpl.coeff(outer))
     {}
     
-    inline Index outer() const { return m_outer; }
-    inline Index col() const { return m_outer; }
-};
+    EIGEN_STRONG_INLINE Scalar value() const { return m_coeff * SparseXprInnerIterator::value(); }
+  protected:
+    typename DiagonalCoeffType::Scalar m_coeff;
+  };
+  
+  sparse_diagonal_product_evaluator(const SparseXprType &sparseXpr, const DiagonalCoeffType &diagCoeff)
+    : m_sparseXprImpl(sparseXpr), m_diagCoeffImpl(diagCoeff)
+  {}
 
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType>
-class sparse_diagonal_product_inner_iterator_selector
-<Lhs,Rhs,SparseDiagonalProductType,SDP_IsSparseColMajor,SDP_IsDiagonal>
-  : public CwiseUnaryOp<scalar_multiple_op<typename Rhs::Scalar>,const Lhs>::InnerIterator
-{
-    typedef typename CwiseUnaryOp<scalar_multiple_op<typename Rhs::Scalar>,const Lhs>::InnerIterator Base;
-    typedef typename Lhs::Index Index;
-  public:
-    inline sparse_diagonal_product_inner_iterator_selector(
-              const SparseDiagonalProductType& expr, Index outer)
-      : Base(expr.lhs()*expr.rhs().diagonal().coeff(outer), outer)
-    {}
+  Index nonZerosEstimate() const { return m_sparseXprImpl.nonZerosEstimate(); }
+    
+protected:
+  evaluator<SparseXprType> m_sparseXprImpl;
+  evaluator<DiagonalCoeffType> m_diagCoeffImpl;
 };
 
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType>
-class sparse_diagonal_product_inner_iterator_selector
-<Lhs,Rhs,SparseDiagonalProductType,SDP_IsSparseRowMajor,SDP_IsDiagonal>
-  : public CwiseBinaryOp<
-      scalar_product_op<typename Rhs::Scalar>,
-      const typename Lhs::ConstInnerVectorReturnType,
-      const Transpose<const typename Rhs::DiagonalVectorType> >::InnerIterator
+
+template<typename SparseXprType, typename DiagCoeffType>
+struct sparse_diagonal_product_evaluator<SparseXprType, DiagCoeffType, SDP_AsCwiseProduct>
 {
-    typedef typename CwiseBinaryOp<
-      scalar_product_op<typename Rhs::Scalar>,
-      const typename Lhs::ConstInnerVectorReturnType,
-      const Transpose<const typename Rhs::DiagonalVectorType> >::InnerIterator Base;
-    typedef typename Lhs::Index Index;
-    Index m_outer;
+  typedef typename SparseXprType::Scalar Scalar;
+  typedef typename SparseXprType::StorageIndex StorageIndex;
+  
+  typedef typename nested_eval<DiagCoeffType,SparseXprType::IsRowMajor ? SparseXprType::RowsAtCompileTime
+                                                                       : SparseXprType::ColsAtCompileTime>::type DiagCoeffNested;
+  
+  class InnerIterator
+  {
+    typedef typename evaluator<SparseXprType>::InnerIterator SparseXprIter;
   public:
-    inline sparse_diagonal_product_inner_iterator_selector(
-              const SparseDiagonalProductType& expr, Index outer)
-      : Base(expr.lhs().innerVector(outer) .cwiseProduct(expr.rhs().diagonal().transpose()), 0), m_outer(outer)
+    InnerIterator(const sparse_diagonal_product_evaluator &xprEval, Index outer)
+      : m_sparseIter(xprEval.m_sparseXprEval, outer), m_diagCoeffNested(xprEval.m_diagCoeffNested)
     {}
     
-    inline Index outer() const { return m_outer; }
-    inline Index row() const { return m_outer; }
+    inline Scalar value() const { return m_sparseIter.value() * m_diagCoeffNested.coeff(index()); }
+    inline StorageIndex index() const  { return m_sparseIter.index(); }
+    inline Index outer() const  { return m_sparseIter.outer(); }
+    inline Index col() const    { return SparseXprType::IsRowMajor ? m_sparseIter.index() : m_sparseIter.outer(); }
+    inline Index row() const    { return SparseXprType::IsRowMajor ? m_sparseIter.outer() : m_sparseIter.index(); }
+    
+    EIGEN_STRONG_INLINE InnerIterator& operator++() { ++m_sparseIter; return *this; }
+    inline operator bool() const  { return m_sparseIter; }
+    
+  protected:
+    SparseXprIter m_sparseIter;
+    DiagCoeffNested m_diagCoeffNested;
+  };
+  
+  sparse_diagonal_product_evaluator(const SparseXprType &sparseXpr, const DiagCoeffType &diagCoeff)
+    : m_sparseXprEval(sparseXpr), m_diagCoeffNested(diagCoeff)
+  {}
+
+  Index nonZerosEstimate() const { return m_sparseXprEval.nonZerosEstimate(); }
+    
+protected:
+  evaluator<SparseXprType> m_sparseXprEval;
+  DiagCoeffNested m_diagCoeffNested;
 };
 
 } // end namespace internal
 
-// SparseMatrixBase functions
-
-template<typename Derived>
-template<typename OtherDerived>
-const SparseDiagonalProduct<Derived,OtherDerived>
-SparseMatrixBase<Derived>::operator*(const DiagonalBase<OtherDerived> &other) const
-{
-  return SparseDiagonalProduct<Derived,OtherDerived>(this->derived(), other.derived());
-}
-
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSE_DIAGONAL_PRODUCT_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseDot.h b/extern/Eigen3/Eigen/src/SparseCore/SparseDot.h
index db39c9aecc7..38bc4aa9ea0 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseDot.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseDot.h
@@ -26,7 +26,8 @@ SparseMatrixBase<Derived>::dot(const MatrixBase<OtherDerived>& other) const
   eigen_assert(size() == other.size());
   eigen_assert(other.size()>0 && "you are using a non initialized vector");
 
-  typename Derived::InnerIterator i(derived(),0);
+  internal::evaluator<Derived> thisEval(derived());
+  typename internal::evaluator<Derived>::InnerIterator i(thisEval, 0);
   Scalar res(0);
   while (i)
   {
@@ -49,16 +50,12 @@ SparseMatrixBase<Derived>::dot(const SparseMatrixBase<OtherDerived>& other) cons
 
   eigen_assert(size() == other.size());
 
-  typedef typename Derived::Nested  Nested;
-  typedef typename OtherDerived::Nested  OtherNested;
-  typedef typename internal::remove_all<Nested>::type  NestedCleaned;
-  typedef typename internal::remove_all<OtherNested>::type  OtherNestedCleaned;
+  internal::evaluator<Derived> thisEval(derived());
+  typename internal::evaluator<Derived>::InnerIterator i(thisEval, 0);
+  
+  internal::evaluator<OtherDerived>  otherEval(other.derived());
+  typename internal::evaluator<OtherDerived>::InnerIterator j(otherEval, 0);
 
-  Nested nthis(derived());
-  OtherNested nother(other.derived());
-
-  typename NestedCleaned::InnerIterator i(nthis,0);
-  typename OtherNestedCleaned::InnerIterator j(nother,0);
   Scalar res(0);
   while (i && j)
   {
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseFuzzy.h b/extern/Eigen3/Eigen/src/SparseCore/SparseFuzzy.h
index 45f36e9eb90..7d47eb94d2f 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseFuzzy.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseFuzzy.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -10,17 +10,20 @@
 #ifndef EIGEN_SPARSE_FUZZY_H
 #define EIGEN_SPARSE_FUZZY_H
 
-// template<typename Derived>
-// template<typename OtherDerived>
-// bool SparseMatrixBase<Derived>::isApprox(
-//   const OtherDerived& other,
-//   typename NumTraits<Scalar>::Real prec
-// ) const
-// {
-//   const typename internal::nested<Derived,2>::type nested(derived());
-//   const typename internal::nested<OtherDerived,2>::type otherNested(other.derived());
-//   return    (nested - otherNested).cwise().abs2().sum()
-//          <= prec * prec * (std::min)(nested.cwise().abs2().sum(), otherNested.cwise().abs2().sum());
-// }
+namespace Eigen {
+  
+template<typename Derived>
+template<typename OtherDerived>
+bool SparseMatrixBase<Derived>::isApprox(const SparseMatrixBase<OtherDerived>& other, const RealScalar &prec) const
+{
+  const typename internal::nested_eval<Derived,2,PlainObject>::type actualA(derived());
+  typename internal::conditional<bool(IsRowMajor)==bool(OtherDerived::IsRowMajor),
+    const typename internal::nested_eval<OtherDerived,2,PlainObject>::type,
+    const PlainObject>::type actualB(other.derived());
+
+  return (actualA - actualB).squaredNorm() <= prec * prec * numext::mini(actualA.squaredNorm(), actualB.squaredNorm());
+}
+
+} // end namespace Eigen
 
 #endif // EIGEN_SPARSE_FUZZY_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseMap.h b/extern/Eigen3/Eigen/src/SparseCore/SparseMap.h
new file mode 100644
index 00000000000..f99be3379db
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseMap.h
@@ -0,0 +1,305 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSE_MAP_H
+#define EIGEN_SPARSE_MAP_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct traits<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : public traits<SparseMatrix<MatScalar,MatOptions,MatIndex> >
+{
+  typedef SparseMatrix<MatScalar,MatOptions,MatIndex> PlainObjectType;
+  typedef traits<PlainObjectType> TraitsBase;
+  enum {
+    Flags = TraitsBase::Flags & (~NestByRefBit)
+  };
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct traits<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : public traits<SparseMatrix<MatScalar,MatOptions,MatIndex> >
+{
+  typedef SparseMatrix<MatScalar,MatOptions,MatIndex> PlainObjectType;
+  typedef traits<PlainObjectType> TraitsBase;
+  enum {
+    Flags = TraitsBase::Flags & (~ (NestByRefBit | LvalueBit))
+  };
+};
+
+} // end namespace internal
+
+template<typename Derived,
+         int Level = internal::accessors_level<Derived>::has_write_access ? WriteAccessors : ReadOnlyAccessors
+> class SparseMapBase;
+
+/** \ingroup SparseCore_Module
+  * class SparseMapBase
+  * \brief Common base class for Map and Ref instance of sparse matrix and vector.
+  */
+template<typename Derived>
+class SparseMapBase<Derived,ReadOnlyAccessors>
+  : public SparseCompressedBase<Derived>
+{
+  public:
+    typedef SparseCompressedBase<Derived> Base;
+    typedef typename Base::Scalar Scalar;
+    typedef typename Base::StorageIndex StorageIndex;
+    enum { IsRowMajor = Base::IsRowMajor };
+    using Base::operator=;
+  protected:
+    
+    typedef typename internal::conditional<
+                         bool(internal::is_lvalue<Derived>::value),
+                         Scalar *, const Scalar *>::type ScalarPointer;
+    typedef typename internal::conditional<
+                         bool(internal::is_lvalue<Derived>::value),
+                         StorageIndex *, const StorageIndex *>::type IndexPointer;
+
+    Index   m_outerSize;
+    Index   m_innerSize;
+    Array<StorageIndex,2,1>  m_zero_nnz;
+    IndexPointer  m_outerIndex;
+    IndexPointer  m_innerIndices;
+    ScalarPointer m_values;
+    IndexPointer  m_innerNonZeros;
+
+  public:
+
+    /** \copydoc SparseMatrixBase::rows() */
+    inline Index rows() const { return IsRowMajor ? m_outerSize : m_innerSize; }
+    /** \copydoc SparseMatrixBase::cols() */
+    inline Index cols() const { return IsRowMajor ? m_innerSize : m_outerSize; }
+    /** \copydoc SparseMatrixBase::innerSize() */
+    inline Index innerSize() const { return m_innerSize; }
+    /** \copydoc SparseMatrixBase::outerSize() */
+    inline Index outerSize() const { return m_outerSize; }
+    /** \copydoc SparseCompressedBase::nonZeros */
+    inline Index nonZeros() const { return m_zero_nnz[1]; }
+    
+    /** \copydoc SparseCompressedBase::isCompressed */
+    bool isCompressed() const { return m_innerNonZeros==0; }
+
+    //----------------------------------------
+    // direct access interface
+    /** \copydoc SparseMatrix::valuePtr */
+    inline const Scalar* valuePtr() const { return m_values; }
+    /** \copydoc SparseMatrix::innerIndexPtr */
+    inline const StorageIndex* innerIndexPtr() const { return m_innerIndices; }
+    /** \copydoc SparseMatrix::outerIndexPtr */
+    inline const StorageIndex* outerIndexPtr() const { return m_outerIndex; }
+    /** \copydoc SparseMatrix::innerNonZeroPtr */
+    inline const StorageIndex* innerNonZeroPtr() const { return m_innerNonZeros; }
+    //----------------------------------------
+
+    /** \copydoc SparseMatrix::coeff */
+    inline Scalar coeff(Index row, Index col) const
+    {
+      const Index outer = IsRowMajor ? row : col;
+      const Index inner = IsRowMajor ? col : row;
+
+      Index start = m_outerIndex[outer];
+      Index end = isCompressed() ? m_outerIndex[outer+1] : start + m_innerNonZeros[outer];
+      if (start==end)
+        return Scalar(0);
+      else if (end>0 && inner==m_innerIndices[end-1])
+        return m_values[end-1];
+      // ^^  optimization: let's first check if it is the last coefficient
+      // (very common in high level algorithms)
+
+      const StorageIndex* r = std::lower_bound(&m_innerIndices[start],&m_innerIndices[end-1],inner);
+      const Index id = r-&m_innerIndices[0];
+      return ((*r==inner) && (id<end)) ? m_values[id] : Scalar(0);
+    }
+
+    inline SparseMapBase(Index rows, Index cols, Index nnz, IndexPointer outerIndexPtr, IndexPointer innerIndexPtr,
+                              ScalarPointer valuePtr, IndexPointer innerNonZerosPtr = 0)
+      : m_outerSize(IsRowMajor?rows:cols), m_innerSize(IsRowMajor?cols:rows), m_zero_nnz(0,internal::convert_index<StorageIndex>(nnz)), m_outerIndex(outerIndexPtr),
+        m_innerIndices(innerIndexPtr), m_values(valuePtr), m_innerNonZeros(innerNonZerosPtr)
+    {}
+
+    // for vectors
+    inline SparseMapBase(Index size, Index nnz, IndexPointer innerIndexPtr, ScalarPointer valuePtr)
+      : m_outerSize(1), m_innerSize(size), m_zero_nnz(0,internal::convert_index<StorageIndex>(nnz)), m_outerIndex(m_zero_nnz.data()),
+        m_innerIndices(innerIndexPtr), m_values(valuePtr), m_innerNonZeros(0)
+    {}
+
+    /** Empty destructor */
+    inline ~SparseMapBase() {}
+
+  protected:
+    inline SparseMapBase() {}
+};
+
+/** \ingroup SparseCore_Module
+  * class SparseMapBase
+  * \brief Common base class for writable Map and Ref instance of sparse matrix and vector.
+  */
+template<typename Derived>
+class SparseMapBase<Derived,WriteAccessors>
+  : public SparseMapBase<Derived,ReadOnlyAccessors>
+{
+    typedef MapBase<Derived, ReadOnlyAccessors> ReadOnlyMapBase;
+    
+  public:
+    typedef SparseMapBase<Derived, ReadOnlyAccessors> Base;
+    typedef typename Base::Scalar Scalar;
+    typedef typename Base::StorageIndex StorageIndex;
+    enum { IsRowMajor = Base::IsRowMajor };
+    
+    using Base::operator=;
+
+  public:
+    
+    //----------------------------------------
+    // direct access interface
+    using Base::valuePtr;
+    using Base::innerIndexPtr;
+    using Base::outerIndexPtr;
+    using Base::innerNonZeroPtr;
+    /** \copydoc SparseMatrix::valuePtr */
+    inline Scalar* valuePtr()              { return Base::m_values; }
+    /** \copydoc SparseMatrix::innerIndexPtr */
+    inline StorageIndex* innerIndexPtr()   { return Base::m_innerIndices; }
+    /** \copydoc SparseMatrix::outerIndexPtr */
+    inline StorageIndex* outerIndexPtr()   { return Base::m_outerIndex; }
+    /** \copydoc SparseMatrix::innerNonZeroPtr */
+    inline StorageIndex* innerNonZeroPtr() { return Base::m_innerNonZeros; }
+    //----------------------------------------
+
+    /** \copydoc SparseMatrix::coeffRef */
+    inline Scalar& coeffRef(Index row, Index col)
+    {
+      const Index outer = IsRowMajor ? row : col;
+      const Index inner = IsRowMajor ? col : row;
+
+      Index start = Base::m_outerIndex[outer];
+      Index end = Base::isCompressed() ? Base::m_outerIndex[outer+1] : start + Base::m_innerNonZeros[outer];
+      eigen_assert(end>=start && "you probably called coeffRef on a non finalized matrix");
+      eigen_assert(end>start && "coeffRef cannot be called on a zero coefficient");
+      StorageIndex* r = std::lower_bound(&Base::m_innerIndices[start],&Base::m_innerIndices[end],inner);
+      const Index id = r - &Base::m_innerIndices[0];
+      eigen_assert((*r==inner) && (id<end) && "coeffRef cannot be called on a zero coefficient");
+      return const_cast<Scalar*>(Base::m_values)[id];
+    }
+    
+    inline SparseMapBase(Index rows, Index cols, Index nnz, StorageIndex* outerIndexPtr, StorageIndex* innerIndexPtr,
+                         Scalar* valuePtr, StorageIndex* innerNonZerosPtr = 0)
+      : Base(rows, cols, nnz, outerIndexPtr, innerIndexPtr, valuePtr, innerNonZerosPtr)
+    {}
+
+    // for vectors
+    inline SparseMapBase(Index size, Index nnz, StorageIndex* innerIndexPtr, Scalar* valuePtr)
+      : Base(size, nnz, innerIndexPtr, valuePtr)
+    {}
+
+    /** Empty destructor */
+    inline ~SparseMapBase() {}
+
+  protected:
+    inline SparseMapBase() {}
+};
+
+/** \ingroup SparseCore_Module
+  *
+  * \brief Specialization of class Map for SparseMatrix-like storage.
+  *
+  * \tparam SparseMatrixType the equivalent sparse matrix type of the referenced data, it must be a template instance of class SparseMatrix.
+  *
+  * \sa class Map, class SparseMatrix, class Ref<SparseMatrixType,Options>
+  */
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType>
+  : public SparseMapBase<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+#else
+template<typename SparseMatrixType>
+class Map<SparseMatrixType>
+  : public SparseMapBase<Derived,WriteAccessors>
+#endif
+{
+  public:
+    typedef SparseMapBase<Map> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Map)
+    enum { IsRowMajor = Base::IsRowMajor };
+
+  public:
+
+    /** Constructs a read-write Map to a sparse matrix of size \a rows x \a cols, containing \a nnz non-zero coefficients,
+      * stored as a sparse format as defined by the pointers \a outerIndexPtr, \a innerIndexPtr, and \a valuePtr.
+      * If the optional parameter \a innerNonZerosPtr is the null pointer, then a standard compressed format is assumed.
+      *
+      * This constructor is available only if \c SparseMatrixType is non-const.
+      *
+      * More details on the expected storage schemes are given in the \ref TutorialSparse "manual pages".
+      */
+    inline Map(Index rows, Index cols, Index nnz, StorageIndex* outerIndexPtr,
+               StorageIndex* innerIndexPtr, Scalar* valuePtr, StorageIndex* innerNonZerosPtr = 0)
+      : Base(rows, cols, nnz, outerIndexPtr, innerIndexPtr, valuePtr, innerNonZerosPtr)
+    {}
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** Empty destructor */
+    inline ~Map() {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType>
+  : public SparseMapBase<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+{
+  public:
+    typedef SparseMapBase<Map> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Map)
+    enum { IsRowMajor = Base::IsRowMajor };
+
+  public:
+#endif
+    /** This is the const version of the above constructor.
+      *
+      * This constructor is available only if \c SparseMatrixType is const, e.g.:
+      * \code Map<const SparseMatrix<double> >  \endcode
+      */
+    inline Map(Index rows, Index cols, Index nnz, const StorageIndex* outerIndexPtr,
+               const StorageIndex* innerIndexPtr, const Scalar* valuePtr, const StorageIndex* innerNonZerosPtr = 0)
+      : Base(rows, cols, nnz, outerIndexPtr, innerIndexPtr, valuePtr, innerNonZerosPtr)
+    {}
+
+    /** Empty destructor */
+    inline ~Map() {}
+};
+
+namespace internal {
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSE_MAP_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseMatrix.h b/extern/Eigen3/Eigen/src/SparseCore/SparseMatrix.h
index 2ff2015512f..0a2490bcc3d 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseMatrix.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseMatrix.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -32,18 +32,22 @@ namespace Eigen {
   * \tparam _Scalar the scalar type, i.e. the type of the coefficients
   * \tparam _Options Union of bit flags controlling the storage scheme. Currently the only possibility
   *                 is ColMajor or RowMajor. The default is 0 which means column-major.
-  * \tparam _Index the type of the indices. It has to be a \b signed type (e.g., short, int, std::ptrdiff_t). Default is \c int.
+  * \tparam _StorageIndex the type of the indices. It has to be a \b signed type (e.g., short, int, std::ptrdiff_t). Default is \c int.
+  *
+  * \warning In %Eigen 3.2, the undocumented type \c SparseMatrix::Index was improperly defined as the storage index type (e.g., int),
+  *          whereas it is now (starting from %Eigen 3.3) deprecated and always defined as Eigen::Index.
+  *          Codes making use of \c SparseMatrix::Index, might thus likely have to be changed to use \c SparseMatrix::StorageIndex instead.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_SPARSEMATRIX_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_SPARSEMATRIX_PLUGIN.
   */
 
 namespace internal {
-template<typename _Scalar, int _Options, typename _Index>
-struct traits<SparseMatrix<_Scalar, _Options, _Index> >
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct traits<SparseMatrix<_Scalar, _Options, _StorageIndex> >
 {
   typedef _Scalar Scalar;
-  typedef _Index Index;
+  typedef _StorageIndex StorageIndex;
   typedef Sparse StorageKind;
   typedef MatrixXpr XprKind;
   enum {
@@ -51,22 +55,21 @@ struct traits<SparseMatrix<_Scalar, _Options, _Index> >
     ColsAtCompileTime = Dynamic,
     MaxRowsAtCompileTime = Dynamic,
     MaxColsAtCompileTime = Dynamic,
-    Flags = _Options | NestByRefBit | LvalueBit,
-    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Flags = _Options | NestByRefBit | LvalueBit | CompressedAccessBit,
     SupportedAccessPatterns = InnerRandomAccessPattern
   };
 };
 
-template<typename _Scalar, int _Options, typename _Index, int DiagIndex>
-struct traits<Diagonal<const SparseMatrix<_Scalar, _Options, _Index>, DiagIndex> >
+template<typename _Scalar, int _Options, typename _StorageIndex, int DiagIndex>
+struct traits<Diagonal<SparseMatrix<_Scalar, _Options, _StorageIndex>, DiagIndex> >
 {
-  typedef SparseMatrix<_Scalar, _Options, _Index> MatrixType;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef SparseMatrix<_Scalar, _Options, _StorageIndex> MatrixType;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
 
   typedef _Scalar Scalar;
   typedef Dense StorageKind;
-  typedef _Index Index;
+  typedef _StorageIndex StorageIndex;
   typedef MatrixXpr XprKind;
 
   enum {
@@ -74,47 +77,61 @@ struct traits<Diagonal<const SparseMatrix<_Scalar, _Options, _Index>, DiagIndex>
     ColsAtCompileTime = 1,
     MaxRowsAtCompileTime = Dynamic,
     MaxColsAtCompileTime = 1,
-    Flags = 0,
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost*10
+    Flags = LvalueBit
+  };
+};
+
+template<typename _Scalar, int _Options, typename _StorageIndex, int DiagIndex>
+struct traits<Diagonal<const SparseMatrix<_Scalar, _Options, _StorageIndex>, DiagIndex> >
+ : public traits<Diagonal<SparseMatrix<_Scalar, _Options, _StorageIndex>, DiagIndex> >
+{
+  enum {
+    Flags = 0
   };
 };
 
 } // end namespace internal
 
-template<typename _Scalar, int _Options, typename _Index>
+template<typename _Scalar, int _Options, typename _StorageIndex>
 class SparseMatrix
-  : public SparseMatrixBase<SparseMatrix<_Scalar, _Options, _Index> >
+  : public SparseCompressedBase<SparseMatrix<_Scalar, _Options, _StorageIndex> >
 {
+    typedef SparseCompressedBase<SparseMatrix> Base;
+    using Base::convert_index;
+    friend class SparseVector<_Scalar,0,_StorageIndex>;
   public:
+    using Base::isCompressed;
+    using Base::nonZeros;
     EIGEN_SPARSE_PUBLIC_INTERFACE(SparseMatrix)
-    EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseMatrix, +=)
-    EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseMatrix, -=)
+    using Base::operator+=;
+    using Base::operator-=;
 
     typedef MappedSparseMatrix<Scalar,Flags> Map;
+    typedef Diagonal<SparseMatrix> DiagonalReturnType;
+    typedef Diagonal<const SparseMatrix> ConstDiagonalReturnType;
+    typedef typename Base::InnerIterator InnerIterator;
+    typedef typename Base::ReverseInnerIterator ReverseInnerIterator;
+    
+
     using Base::IsRowMajor;
-    typedef internal::CompressedStorage<Scalar,Index> Storage;
+    typedef internal::CompressedStorage<Scalar,StorageIndex> Storage;
     enum {
       Options = _Options
     };
 
+    typedef typename Base::IndexVector IndexVector;
+    typedef typename Base::ScalarVector ScalarVector;
   protected:
-
     typedef SparseMatrix<Scalar,(Flags&~RowMajorBit)|(IsRowMajor?RowMajorBit:0)> TransposedSparseMatrix;
 
     Index m_outerSize;
     Index m_innerSize;
-    Index* m_outerIndex;
-    Index* m_innerNonZeros;     // optional, if null then the data is compressed
+    StorageIndex* m_outerIndex;
+    StorageIndex* m_innerNonZeros;     // optional, if null then the data is compressed
     Storage m_data;
-    
-    Eigen::Map<Matrix<Index,Dynamic,1> > innerNonZeros() { return Eigen::Map<Matrix<Index,Dynamic,1> >(m_innerNonZeros, m_innerNonZeros?m_outerSize:0); }
-    const  Eigen::Map<const Matrix<Index,Dynamic,1> > innerNonZeros() const { return Eigen::Map<const Matrix<Index,Dynamic,1> >(m_innerNonZeros, m_innerNonZeros?m_outerSize:0); }
 
   public:
     
-    /** \returns whether \c *this is in compressed form. */
-    inline bool isCompressed() const { return m_innerNonZeros==0; }
-
     /** \returns the number of rows of the matrix */
     inline Index rows() const { return IsRowMajor ? m_outerSize : m_innerSize; }
     /** \returns the number of columns of the matrix */
@@ -128,38 +145,38 @@ class SparseMatrix
     /** \returns a const pointer to the array of values.
       * This function is aimed at interoperability with other libraries.
       * \sa innerIndexPtr(), outerIndexPtr() */
-    inline const Scalar* valuePtr() const { return &m_data.value(0); }
+    inline const Scalar* valuePtr() const { return m_data.valuePtr(); }
     /** \returns a non-const pointer to the array of values.
       * This function is aimed at interoperability with other libraries.
       * \sa innerIndexPtr(), outerIndexPtr() */
-    inline Scalar* valuePtr() { return &m_data.value(0); }
+    inline Scalar* valuePtr() { return m_data.valuePtr(); }
 
     /** \returns a const pointer to the array of inner indices.
       * This function is aimed at interoperability with other libraries.
       * \sa valuePtr(), outerIndexPtr() */
-    inline const Index* innerIndexPtr() const { return &m_data.index(0); }
+    inline const StorageIndex* innerIndexPtr() const { return m_data.indexPtr(); }
     /** \returns a non-const pointer to the array of inner indices.
       * This function is aimed at interoperability with other libraries.
       * \sa valuePtr(), outerIndexPtr() */
-    inline Index* innerIndexPtr() { return &m_data.index(0); }
+    inline StorageIndex* innerIndexPtr() { return m_data.indexPtr(); }
 
     /** \returns a const pointer to the array of the starting positions of the inner vectors.
       * This function is aimed at interoperability with other libraries.
       * \sa valuePtr(), innerIndexPtr() */
-    inline const Index* outerIndexPtr() const { return m_outerIndex; }
+    inline const StorageIndex* outerIndexPtr() const { return m_outerIndex; }
     /** \returns a non-const pointer to the array of the starting positions of the inner vectors.
       * This function is aimed at interoperability with other libraries.
       * \sa valuePtr(), innerIndexPtr() */
-    inline Index* outerIndexPtr() { return m_outerIndex; }
+    inline StorageIndex* outerIndexPtr() { return m_outerIndex; }
 
     /** \returns a const pointer to the array of the number of non zeros of the inner vectors.
       * This function is aimed at interoperability with other libraries.
       * \warning it returns the null pointer 0 in compressed mode */
-    inline const Index* innerNonZeroPtr() const { return m_innerNonZeros; }
+    inline const StorageIndex* innerNonZeroPtr() const { return m_innerNonZeros; }
     /** \returns a non-const pointer to the array of the number of non zeros of the inner vectors.
       * This function is aimed at interoperability with other libraries.
       * \warning it returns the null pointer 0 in compressed mode */
-    inline Index* innerNonZeroPtr() { return m_innerNonZeros; }
+    inline StorageIndex* innerNonZeroPtr() { return m_innerNonZeros; }
 
     /** \internal */
     inline Storage& data() { return m_data; }
@@ -175,7 +192,7 @@ class SparseMatrix
       const Index outer = IsRowMajor ? row : col;
       const Index inner = IsRowMajor ? col : row;
       Index end = m_innerNonZeros ? m_outerIndex[outer] + m_innerNonZeros[outer] : m_outerIndex[outer+1];
-      return m_data.atInRange(m_outerIndex[outer], end, inner);
+      return m_data.atInRange(m_outerIndex[outer], end, StorageIndex(inner));
     }
 
     /** \returns a non-const reference to the value of the matrix at position \a i, \a j
@@ -198,7 +215,7 @@ class SparseMatrix
       eigen_assert(end>=start && "you probably called coeffRef on a non finalized matrix");
       if(end<=start)
         return insert(row,col);
-      const Index p = m_data.searchLowerIndex(start,end-1,inner);
+      const Index p = m_data.searchLowerIndex(start,end-1,StorageIndex(inner));
       if((p<end) && (m_data.index(p)==inner))
         return m_data.value(p);
       else
@@ -209,45 +226,34 @@ class SparseMatrix
       * The non zero coefficient must \b not already exist.
       *
       * If the matrix \c *this is in compressed mode, then \c *this is turned into uncompressed
-      * mode while reserving room for 2 non zeros per inner vector. It is strongly recommended to first
-      * call reserve(const SizesType &) to reserve a more appropriate number of elements per
-      * inner vector that better match your scenario.
+      * mode while reserving room for 2 x this->innerSize() non zeros if reserve(Index) has not been called earlier.
+      * In this case, the insertion procedure is optimized for a \e sequential insertion mode where elements are assumed to be
+      * inserted by increasing outer-indices.
+      * 
+      * If that's not the case, then it is strongly recommended to either use a triplet-list to assemble the matrix, or to first
+      * call reserve(const SizesType &) to reserve the appropriate number of non-zero elements per inner vector.
       *
-      * This function performs a sorted insertion in O(1) if the elements of each inner vector are
-      * inserted in increasing inner index order, and in O(nnz_j) for a random insertion.
+      * Assuming memory has been appropriately reserved, this function performs a sorted insertion in O(1)
+      * if the elements of each inner vector are inserted in increasing inner index order, and in O(nnz_j) for a random insertion.
       *
       */
-    Scalar& insert(Index row, Index col)
-    {
-      eigen_assert(row>=0 && row<rows() && col>=0 && col<cols());
-      
-      if(isCompressed())
-      {
-        reserve(Matrix<Index,Dynamic,1>::Constant(outerSize(), 2));
-      }
-      return insertUncompressed(row,col);
-    }
+    Scalar& insert(Index row, Index col);
 
   public:
 
-    class InnerIterator;
-    class ReverseInnerIterator;
-
-    /** Removes all non zeros but keep allocated memory */
+    /** Removes all non zeros but keep allocated memory
+      *
+      * This function does not free the currently allocated memory. To release as much as memory as possible,
+      * call \code mat.data().squeeze(); \endcode after resizing it.
+      * 
+      * \sa resize(Index,Index), data()
+      */
     inline void setZero()
     {
       m_data.clear();
-      memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(Index));
-      if(m_innerNonZeros)
-        memset(m_innerNonZeros, 0, (m_outerSize)*sizeof(Index));
-    }
-
-    /** \returns the number of non zero coefficients */
-    inline Index nonZeros() const
-    {
+      memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(StorageIndex));
       if(m_innerNonZeros)
-        return innerNonZeros().sum();
-      return static_cast<Index>(m_data.size());
+        memset(m_innerNonZeros, 0, (m_outerSize)*sizeof(StorageIndex));
     }
 
     /** Preallocates \a reserveSize non zeros.
@@ -262,22 +268,25 @@ class SparseMatrix
     #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** Preallocates \a reserveSize[\c j] non zeros for each column (resp. row) \c j.
       *
-      * This function turns the matrix in non-compressed mode */
+      * This function turns the matrix in non-compressed mode.
+      * 
+      * The type \c SizesType must expose the following interface:
+        \code
+        typedef value_type;
+        const value_type& operator[](i) const;
+        \endcode
+      * for \c i in the [0,this->outerSize()[ range.
+      * Typical choices include std::vector<int>, Eigen::VectorXi, Eigen::VectorXi::Constant, etc.
+      */
     template<class SizesType>
     inline void reserve(const SizesType& reserveSizes);
     #else
     template<class SizesType>
-    inline void reserve(const SizesType& reserveSizes, const typename SizesType::value_type& enableif = typename SizesType::value_type())
-    {
-      EIGEN_UNUSED_VARIABLE(enableif);
-      reserveInnerVectors(reserveSizes);
-    }
-    template<class SizesType>
-    inline void reserve(const SizesType& reserveSizes, const typename SizesType::Scalar& enableif =
-    #if (!defined(_MSC_VER)) || (_MSC_VER>=1500) // MSVC 2005 fails to compile with this typename
+    inline void reserve(const SizesType& reserveSizes, const typename SizesType::value_type& enableif =
+    #if (!EIGEN_COMP_MSVC) || (EIGEN_COMP_MSVC>=1500) // MSVC 2005 fails to compile with this typename
         typename
     #endif
-        SizesType::Scalar())
+        SizesType::value_type())
     {
       EIGEN_UNUSED_VARIABLE(enableif);
       reserveInnerVectors(reserveSizes);
@@ -289,15 +298,15 @@ class SparseMatrix
     {
       if(isCompressed())
       {
-        std::size_t totalReserveSize = 0;
+        Index totalReserveSize = 0;
         // turn the matrix into non-compressed mode
-        m_innerNonZeros = static_cast<Index*>(std::malloc(m_outerSize * sizeof(Index)));
+        m_innerNonZeros = static_cast<StorageIndex*>(std::malloc(m_outerSize * sizeof(StorageIndex)));
         if (!m_innerNonZeros) internal::throw_std_bad_alloc();
         
         // temporarily use m_innerSizes to hold the new starting points.
-        Index* newOuterIndex = m_innerNonZeros;
+        StorageIndex* newOuterIndex = m_innerNonZeros;
         
-        Index count = 0;
+        StorageIndex count = 0;
         for(Index j=0; j<m_outerSize; ++j)
         {
           newOuterIndex[j] = count;
@@ -305,10 +314,10 @@ class SparseMatrix
           totalReserveSize += reserveSizes[j];
         }
         m_data.reserve(totalReserveSize);
-        Index previousOuterIndex = m_outerIndex[m_outerSize];
+        StorageIndex previousOuterIndex = m_outerIndex[m_outerSize];
         for(Index j=m_outerSize-1; j>=0; --j)
         {
-          Index innerNNZ = previousOuterIndex - m_outerIndex[j];
+          StorageIndex innerNNZ = previousOuterIndex - m_outerIndex[j];
           for(Index i=innerNNZ-1; i>=0; --i)
           {
             m_data.index(newOuterIndex[j]+i) = m_data.index(m_outerIndex[j]+i);
@@ -324,15 +333,15 @@ class SparseMatrix
       }
       else
       {
-        Index* newOuterIndex = static_cast<Index*>(std::malloc((m_outerSize+1)*sizeof(Index)));
+        StorageIndex* newOuterIndex = static_cast<StorageIndex*>(std::malloc((m_outerSize+1)*sizeof(StorageIndex)));
         if (!newOuterIndex) internal::throw_std_bad_alloc();
         
-        Index count = 0;
+        StorageIndex count = 0;
         for(Index j=0; j<m_outerSize; ++j)
         {
           newOuterIndex[j] = count;
-          Index alreadyReserved = (m_outerIndex[j+1]-m_outerIndex[j]) - m_innerNonZeros[j];
-          Index toReserve = std::max<Index>(reserveSizes[j], alreadyReserved);
+          StorageIndex alreadyReserved = (m_outerIndex[j+1]-m_outerIndex[j]) - m_innerNonZeros[j];
+          StorageIndex toReserve = std::max<StorageIndex>(reserveSizes[j], alreadyReserved);
           count += toReserve + m_innerNonZeros[j];
         }
         newOuterIndex[m_outerSize] = count;
@@ -343,7 +352,7 @@ class SparseMatrix
           Index offset = newOuterIndex[j] - m_outerIndex[j];
           if(offset>0)
           {
-            Index innerNNZ = m_innerNonZeros[j];
+            StorageIndex innerNNZ = m_innerNonZeros[j];
             for(Index i=innerNNZ-1; i>=0; --i)
             {
               m_data.index(newOuterIndex[j]+i) = m_data.index(m_outerIndex[j]+i);
@@ -380,11 +389,11 @@ class SparseMatrix
       * \sa insertBack, startVec */
     inline Scalar& insertBackByOuterInner(Index outer, Index inner)
     {
-      eigen_assert(size_t(m_outerIndex[outer+1]) == m_data.size() && "Invalid ordered insertion (invalid outer index)");
+      eigen_assert(Index(m_outerIndex[outer+1]) == m_data.size() && "Invalid ordered insertion (invalid outer index)");
       eigen_assert( (m_outerIndex[outer+1]-m_outerIndex[outer]==0 || m_data.index(m_data.size()-1)<inner) && "Invalid ordered insertion (invalid inner index)");
       Index p = m_outerIndex[outer+1];
       ++m_outerIndex[outer+1];
-      m_data.append(0, inner);
+      m_data.append(Scalar(0), inner);
       return m_data.value(p);
     }
 
@@ -394,7 +403,7 @@ class SparseMatrix
     {
       Index p = m_outerIndex[outer+1];
       ++m_outerIndex[outer+1];
-      m_data.append(0, inner);
+      m_data.append(Scalar(0), inner);
       return m_data.value(p);
     }
 
@@ -414,7 +423,7 @@ class SparseMatrix
     {
       if(isCompressed())
       {
-        Index size = static_cast<Index>(m_data.size());
+        StorageIndex size = internal::convert_index<StorageIndex>(m_data.size());
         Index i = m_outerSize;
         // find the last filled column
         while (i>=0 && m_outerIndex[i]==0)
@@ -433,7 +442,13 @@ class SparseMatrix
     template<typename InputIterators>
     void setFromTriplets(const InputIterators& begin, const InputIterators& end);
 
-    void sumupDuplicates();
+    template<typename InputIterators,typename DupFunctor>
+    void setFromTriplets(const InputIterators& begin, const InputIterators& end, DupFunctor dup_func);
+
+    void sumupDuplicates() { collapseDuplicates(internal::scalar_sum_op<Scalar,Scalar>()); }
+
+    template<typename DupFunctor>
+    void collapseDuplicates(DupFunctor dup_func = DupFunctor());
 
     //---
     
@@ -451,6 +466,8 @@ class SparseMatrix
       if(isCompressed())
         return;
       
+      eigen_internal_assert(m_outerIndex!=0 && m_outerSize>0);
+      
       Index oldStart = m_outerIndex[1];
       m_outerIndex[1] = m_innerNonZeros[0];
       for(Index j=1; j<m_outerSize; ++j)
@@ -479,7 +496,7 @@ class SparseMatrix
     {
       if(m_innerNonZeros != 0)
         return; 
-      m_innerNonZeros = static_cast<Index*>(std::malloc(m_outerSize * sizeof(Index)));
+      m_innerNonZeros = static_cast<StorageIndex*>(std::malloc(m_outerSize * sizeof(StorageIndex)));
       for (Index i = 0; i < m_outerSize; i++)
       {
         m_innerNonZeros[i] = m_outerIndex[i+1] - m_outerIndex[i]; 
@@ -503,10 +520,9 @@ class SparseMatrix
     void prune(const KeepFunc& keep = KeepFunc())
     {
       // TODO optimize the uncompressed mode to avoid moving and allocating the data twice
-      // TODO also implement a unit test
       makeCompressed();
 
-      Index k = 0;
+      StorageIndex k = 0;
       for(Index j=0; j<m_outerSize; ++j)
       {
         Index previousStart = m_outerIndex[j];
@@ -527,7 +543,12 @@ class SparseMatrix
     }
 
     /** Resizes the matrix to a \a rows x \a cols matrix leaving old values untouched.
-      * \sa resizeNonZeros(Index), reserve(), setZero()
+      *
+      * If the sizes of the matrix are decreased, then the matrix is turned to \b uncompressed-mode
+      * and the storage of the out of bounds coefficients is kept and reserved.
+      * Call makeCompressed() to pack the entries and squeeze extra memory.
+      *
+      * \sa reserve(), setZero(), makeCompressed()
       */
     void conservativeResize(Index rows, Index cols) 
     {
@@ -539,13 +560,13 @@ class SparseMatrix
 
       Index innerChange = IsRowMajor ? cols - this->cols() : rows - this->rows();
       Index outerChange = IsRowMajor ? rows - this->rows() : cols - this->cols();
-      Index newInnerSize = IsRowMajor ? cols : rows;
+      StorageIndex newInnerSize = convert_index(IsRowMajor ? cols : rows);
 
       // Deals with inner non zeros
       if (m_innerNonZeros)
       {
         // Resize m_innerNonZeros
-        Index *newInnerNonZeros = static_cast<Index*>(std::realloc(m_innerNonZeros, (m_outerSize + outerChange) * sizeof(Index)));
+        StorageIndex *newInnerNonZeros = static_cast<StorageIndex*>(std::realloc(m_innerNonZeros, (m_outerSize + outerChange) * sizeof(StorageIndex)));
         if (!newInnerNonZeros) internal::throw_std_bad_alloc();
         m_innerNonZeros = newInnerNonZeros;
         
@@ -555,7 +576,7 @@ class SparseMatrix
       else if (innerChange < 0) 
       {
         // Inner size decreased: allocate a new m_innerNonZeros
-        m_innerNonZeros = static_cast<Index*>(std::malloc((m_outerSize+outerChange+1) * sizeof(Index)));
+        m_innerNonZeros = static_cast<StorageIndex*>(std::malloc((m_outerSize+outerChange+1) * sizeof(StorageIndex)));
         if (!m_innerNonZeros) internal::throw_std_bad_alloc();
         for(Index i = 0; i < m_outerSize; i++)
           m_innerNonZeros[i] = m_outerIndex[i+1] - m_outerIndex[i];
@@ -566,8 +587,8 @@ class SparseMatrix
       {
         for(Index i = 0; i < m_outerSize + (std::min)(outerChange, Index(0)); i++)
         {
-          Index &n = m_innerNonZeros[i];
-          Index start = m_outerIndex[i];
+          StorageIndex &n = m_innerNonZeros[i];
+          StorageIndex start = m_outerIndex[i];
           while (n > 0 && m_data.index(start+n-1) >= newInnerSize) --n; 
         }
       }
@@ -578,12 +599,12 @@ class SparseMatrix
       if (outerChange == 0)
         return;
           
-      Index *newOuterIndex = static_cast<Index*>(std::realloc(m_outerIndex, (m_outerSize + outerChange + 1) * sizeof(Index)));
+      StorageIndex *newOuterIndex = static_cast<StorageIndex*>(std::realloc(m_outerIndex, (m_outerSize + outerChange + 1) * sizeof(StorageIndex)));
       if (!newOuterIndex) internal::throw_std_bad_alloc();
       m_outerIndex = newOuterIndex;
       if (outerChange > 0)
       {
-        Index last = m_outerSize == 0 ? 0 : m_outerIndex[m_outerSize];
+        StorageIndex last = m_outerSize == 0 ? 0 : m_outerIndex[m_outerSize];
         for(Index i=m_outerSize; i<m_outerSize+outerChange+1; i++)          
           m_outerIndex[i] = last; 
       }
@@ -591,7 +612,11 @@ class SparseMatrix
     }
     
     /** Resizes the matrix to a \a rows x \a cols matrix and initializes it to zero.
-      * \sa resizeNonZeros(Index), reserve(), setZero()
+      * 
+      * This function does not free the currently allocated memory. To release as much as memory as possible,
+      * call \code mat.data().squeeze(); \endcode after resizing it.
+      * 
+      * \sa reserve(), setZero()
       */
     void resize(Index rows, Index cols)
     {
@@ -601,7 +626,7 @@ class SparseMatrix
       if (m_outerSize != outerSize || m_outerSize==0)
       {
         std::free(m_outerIndex);
-        m_outerIndex = static_cast<Index*>(std::malloc((outerSize + 1) * sizeof(Index)));
+        m_outerIndex = static_cast<StorageIndex*>(std::malloc((outerSize + 1) * sizeof(StorageIndex)));
         if (!m_outerIndex) internal::throw_std_bad_alloc();
         
         m_outerSize = outerSize;
@@ -611,19 +636,24 @@ class SparseMatrix
         std::free(m_innerNonZeros);
         m_innerNonZeros = 0;
       }
-      memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(Index));
+      memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(StorageIndex));
     }
 
     /** \internal
       * Resize the nonzero vector to \a size */
     void resizeNonZeros(Index size)
     {
-      // TODO remove this function
       m_data.resize(size);
     }
 
-    /** \returns a const expression of the diagonal coefficients */
-    const Diagonal<const SparseMatrix> diagonal() const { return *this; }
+    /** \returns a const expression of the diagonal coefficients. */
+    const ConstDiagonalReturnType diagonal() const { return ConstDiagonalReturnType(*this); }
+    
+    /** \returns a read-write expression of the diagonal coefficients.
+      * \warning If the diagonal entries are written, then all diagonal
+      * entries \b must already exist, otherwise an assertion will be raised.
+      */
+    DiagonalReturnType diagonal() { return DiagonalReturnType(*this); }
 
     /** Default constructor yielding an empty \c 0 \c x \c 0 matrix */
     inline SparseMatrix()
@@ -649,7 +679,16 @@ class SparseMatrix
       EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
         YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
       check_template_parameters();
-      *this = other.derived();
+      const bool needToTranspose = (Flags & RowMajorBit) != (internal::evaluator<OtherDerived>::Flags & RowMajorBit);
+      if (needToTranspose)
+        *this = other.derived();
+      else
+      {
+        #ifdef EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+          EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+        #endif
+        internal::call_assignment_no_alias(*this, other.derived());
+      }
     }
     
     /** Constructs a sparse matrix from the sparse selfadjoint view \a other */
@@ -658,7 +697,7 @@ class SparseMatrix
       : m_outerSize(0), m_innerSize(0), m_outerIndex(0), m_innerNonZeros(0)
     {
       check_template_parameters();
-      *this = other;
+      Base::operator=(other);
     }
 
     /** Copy constructor (it performs a deep copy) */
@@ -678,6 +717,15 @@ class SparseMatrix
       initAssignment(other);
       other.evalTo(*this);
     }
+    
+    /** \brief Copy constructor with in-place evaluation */
+    template<typename OtherDerived>
+    explicit SparseMatrix(const DiagonalBase<OtherDerived>& other)
+      : Base(), m_outerSize(0), m_innerSize(0), m_outerIndex(0), m_innerNonZeros(0)
+    {
+      check_template_parameters();
+      *this = other.derived();
+    }
 
     /** Swaps the content of two sparse matrices of the same type.
       * This is a fast operation that simply swaps the underlying pointers and parameters. */
@@ -697,9 +745,9 @@ class SparseMatrix
     {
       eigen_assert(rows() == cols() && "ONLY FOR SQUARED MATRICES");
       this->m_data.resize(rows());
-      Eigen::Map<Matrix<Index, Dynamic, 1> >(&this->m_data.index(0), rows()).setLinSpaced(0, rows()-1);
-      Eigen::Map<Matrix<Scalar, Dynamic, 1> >(&this->m_data.value(0), rows()).setOnes();
-      Eigen::Map<Matrix<Index, Dynamic, 1> >(this->m_outerIndex, rows()+1).setLinSpaced(0, rows());
+      Eigen::Map<IndexVector>(this->m_data.indexPtr(), rows()).setLinSpaced(0, StorageIndex(rows()-1));
+      Eigen::Map<ScalarVector>(this->m_data.valuePtr(), rows()).setOnes();
+      Eigen::Map<IndexVector>(this->m_outerIndex, rows()+1).setLinSpaced(0, StorageIndex(rows()));
       std::free(m_innerNonZeros);
       m_innerNonZeros = 0;
     }
@@ -711,10 +759,13 @@ class SparseMatrix
       }
       else if(this!=&other)
       {
+        #ifdef EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+          EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+        #endif
         initAssignment(other);
         if(other.isCompressed())
         {
-          memcpy(m_outerIndex, other.m_outerIndex, (m_outerSize+1)*sizeof(Index));
+          internal::smart_copy(other.m_outerIndex, other.m_outerIndex + m_outerSize + 1, m_outerIndex);
           m_data = other.m_data;
         }
         else
@@ -725,22 +776,11 @@ class SparseMatrix
       return *this;
     }
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename Lhs, typename Rhs>
-    inline SparseMatrix& operator=(const SparseSparseProduct<Lhs,Rhs>& product)
-    { return Base::operator=(product); }
-    
-    template<typename OtherDerived>
-    inline SparseMatrix& operator=(const ReturnByValue<OtherDerived>& other)
-    {
-      initAssignment(other);
-      return Base::operator=(other.derived());
-    }
-    
+#ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename OtherDerived>
     inline SparseMatrix& operator=(const EigenBase<OtherDerived>& other)
     { return Base::operator=(other.derived()); }
-    #endif
+#endif // EIGEN_PARSED_BY_DOXYGEN
 
     template<typename OtherDerived>
     EIGEN_DONT_INLINE SparseMatrix& operator=(const SparseMatrixBase<OtherDerived>& other);
@@ -750,30 +790,38 @@ class SparseMatrix
       EIGEN_DBG_SPARSE(
         s << "Nonzero entries:\n";
         if(m.isCompressed())
+        {
           for (Index i=0; i<m.nonZeros(); ++i)
             s << "(" << m.m_data.value(i) << "," << m.m_data.index(i) << ") ";
+        }
         else
+        {
           for (Index i=0; i<m.outerSize(); ++i)
           {
             Index p = m.m_outerIndex[i];
             Index pe = m.m_outerIndex[i]+m.m_innerNonZeros[i];
             Index k=p;
-            for (; k<pe; ++k)
+            for (; k<pe; ++k) {
               s << "(" << m.m_data.value(k) << "," << m.m_data.index(k) << ") ";
-            for (; k<m.m_outerIndex[i+1]; ++k)
+            }
+            for (; k<m.m_outerIndex[i+1]; ++k) {
               s << "(_,_) ";
+            }
           }
+        }
         s << std::endl;
         s << std::endl;
         s << "Outer pointers:\n";
-        for (Index i=0; i<m.outerSize(); ++i)
+        for (Index i=0; i<m.outerSize(); ++i) {
           s << m.m_outerIndex[i] << " ";
+        }
         s << " $" << std::endl;
         if(!m.isCompressed())
         {
           s << "Inner non zeros:\n";
-          for (Index i=0; i<m.outerSize(); ++i)
+          for (Index i=0; i<m.outerSize(); ++i) {
             s << m.m_innerNonZeros[i] << " ";
+          }
           s << " $" << std::endl;
         }
         s << std::endl;
@@ -789,10 +837,8 @@ class SparseMatrix
       std::free(m_innerNonZeros);
     }
 
-#ifndef EIGEN_PARSED_BY_DOXYGEN
     /** Overloaded for performance */
     Scalar sum() const;
-#endif
     
 #   ifdef EIGEN_SPARSEMATRIX_PLUGIN
 #     include EIGEN_SPARSEMATRIX_PLUGIN
@@ -819,15 +865,15 @@ protected:
       * A vector object that is equal to 0 everywhere but v at the position i */
     class SingletonVector
     {
-        Index m_index;
-        Index m_value;
+        StorageIndex m_index;
+        StorageIndex m_value;
       public:
-        typedef Index value_type;
+        typedef StorageIndex value_type;
         SingletonVector(Index i, Index v)
-          : m_index(i), m_value(v)
+          : m_index(convert_index(i)), m_value(convert_index(v))
         {}
 
-        Index operator[](Index i) const { return i==m_index ? m_value : 0; }
+        StorageIndex operator[](Index i) const { return i==m_index ? m_value : 0; }
     };
 
     /** \internal
@@ -846,14 +892,14 @@ public:
       eigen_assert(m_innerNonZeros[outer]<=(m_outerIndex[outer+1] - m_outerIndex[outer]));
 
       Index p = m_outerIndex[outer] + m_innerNonZeros[outer]++;
-      m_data.index(p) = inner;
-      return (m_data.value(p) = 0);
+      m_data.index(p) = convert_index(inner);
+      return (m_data.value(p) = Scalar(0));
     }
 
 private:
   static void check_template_parameters()
   {
-    EIGEN_STATIC_ASSERT(NumTraits<Index>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE);
+    EIGEN_STATIC_ASSERT(NumTraits<StorageIndex>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE);
     EIGEN_STATIC_ASSERT((Options&(ColMajor|RowMajor))==Options,INVALID_MATRIX_TEMPLATE_PARAMETERS);
   }
 
@@ -868,87 +914,20 @@ private:
   };
 };
 
-template<typename Scalar, int _Options, typename _Index>
-class SparseMatrix<Scalar,_Options,_Index>::InnerIterator
-{
-  public:
-    InnerIterator(const SparseMatrix& mat, Index outer)
-      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(outer), m_id(mat.m_outerIndex[outer])
-    {
-      if(mat.isCompressed())
-        m_end = mat.m_outerIndex[outer+1];
-      else
-        m_end = m_id + mat.m_innerNonZeros[outer];
-    }
-
-    inline InnerIterator& operator++() { m_id++; return *this; }
-
-    inline const Scalar& value() const { return m_values[m_id]; }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_values[m_id]); }
-
-    inline Index index() const { return m_indices[m_id]; }
-    inline Index outer() const { return m_outer; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    inline operator bool() const { return (m_id < m_end); }
-
-  protected:
-    const Scalar* m_values;
-    const Index* m_indices;
-    const Index m_outer;
-    Index m_id;
-    Index m_end;
-};
-
-template<typename Scalar, int _Options, typename _Index>
-class SparseMatrix<Scalar,_Options,_Index>::ReverseInnerIterator
-{
-  public:
-    ReverseInnerIterator(const SparseMatrix& mat, Index outer)
-      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(outer), m_start(mat.m_outerIndex[outer])
-    {
-      if(mat.isCompressed())
-        m_id = mat.m_outerIndex[outer+1];
-      else
-        m_id = m_start + mat.m_innerNonZeros[outer];
-    }
-
-    inline ReverseInnerIterator& operator--() { --m_id; return *this; }
-
-    inline const Scalar& value() const { return m_values[m_id-1]; }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_values[m_id-1]); }
-
-    inline Index index() const { return m_indices[m_id-1]; }
-    inline Index outer() const { return m_outer; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    inline operator bool() const { return (m_id > m_start); }
-
-  protected:
-    const Scalar* m_values;
-    const Index* m_indices;
-    const Index m_outer;
-    Index m_id;
-    const Index m_start;
-};
-
 namespace internal {
 
-template<typename InputIterator, typename SparseMatrixType>
-void set_from_triplets(const InputIterator& begin, const InputIterator& end, SparseMatrixType& mat, int Options = 0)
+template<typename InputIterator, typename SparseMatrixType, typename DupFunctor>
+void set_from_triplets(const InputIterator& begin, const InputIterator& end, SparseMatrixType& mat, DupFunctor dup_func)
 {
-  EIGEN_UNUSED_VARIABLE(Options);
   enum { IsRowMajor = SparseMatrixType::IsRowMajor };
   typedef typename SparseMatrixType::Scalar Scalar;
-  typedef typename SparseMatrixType::Index Index;
-  SparseMatrix<Scalar,IsRowMajor?ColMajor:RowMajor,Index> trMat(mat.rows(),mat.cols());
+  typedef typename SparseMatrixType::StorageIndex StorageIndex;
+  SparseMatrix<Scalar,IsRowMajor?ColMajor:RowMajor,StorageIndex> trMat(mat.rows(),mat.cols());
 
   if(begin!=end)
   {
     // pass 1: count the nnz per inner-vector
-    Matrix<Index,Dynamic,1> wi(trMat.outerSize());
+    typename SparseMatrixType::IndexVector wi(trMat.outerSize());
     wi.setZero();
     for(InputIterator it(begin); it!=end; ++it)
     {
@@ -962,7 +941,7 @@ void set_from_triplets(const InputIterator& begin, const InputIterator& end, Spa
       trMat.insertBackUncompressed(it->row(),it->col()) = it->value();
 
     // pass 3:
-    trMat.sumupDuplicates();
+    trMat.collapseDuplicates(dup_func);
   }
 
   // pass 4: transposed copy -> implicit sorting
@@ -1009,26 +988,43 @@ void set_from_triplets(const InputIterator& begin, const InputIterator& end, Spa
   * an abstract iterator over a complex data-structure that would be expensive to evaluate. The triplets should rather
   * be explicitely stored into a std::vector for instance.
   */
-template<typename Scalar, int _Options, typename _Index>
+template<typename Scalar, int _Options, typename _StorageIndex>
 template<typename InputIterators>
-void SparseMatrix<Scalar,_Options,_Index>::setFromTriplets(const InputIterators& begin, const InputIterators& end)
+void SparseMatrix<Scalar,_Options,_StorageIndex>::setFromTriplets(const InputIterators& begin, const InputIterators& end)
+{
+  internal::set_from_triplets<InputIterators, SparseMatrix<Scalar,_Options,_StorageIndex> >(begin, end, *this, internal::scalar_sum_op<Scalar,Scalar>());
+}
+
+/** The same as setFromTriplets but when duplicates are met the functor \a dup_func is applied:
+  * \code
+  * value = dup_func(OldValue, NewValue)
+  * \endcode 
+  * Here is a C++11 example keeping the latest entry only:
+  * \code
+  * mat.setFromTriplets(triplets.begin(), triplets.end(), [] (const Scalar&,const Scalar &b) { return b; });
+  * \endcode
+  */
+template<typename Scalar, int _Options, typename _StorageIndex>
+template<typename InputIterators,typename DupFunctor>
+void SparseMatrix<Scalar,_Options,_StorageIndex>::setFromTriplets(const InputIterators& begin, const InputIterators& end, DupFunctor dup_func)
 {
-  internal::set_from_triplets(begin, end, *this);
+  internal::set_from_triplets<InputIterators, SparseMatrix<Scalar,_Options,_StorageIndex>, DupFunctor>(begin, end, *this, dup_func);
 }
 
 /** \internal */
-template<typename Scalar, int _Options, typename _Index>
-void SparseMatrix<Scalar,_Options,_Index>::sumupDuplicates()
+template<typename Scalar, int _Options, typename _StorageIndex>
+template<typename DupFunctor>
+void SparseMatrix<Scalar,_Options,_StorageIndex>::collapseDuplicates(DupFunctor dup_func)
 {
   eigen_assert(!isCompressed());
   // TODO, in practice we should be able to use m_innerNonZeros for that task
-  Matrix<Index,Dynamic,1> wi(innerSize());
+  IndexVector wi(innerSize());
   wi.fill(-1);
-  Index count = 0;
+  StorageIndex count = 0;
   // for each inner-vector, wi[inner_index] will hold the position of first element into the index/value buffers
   for(Index j=0; j<outerSize(); ++j)
   {
-    Index start   = count;
+    StorageIndex start   = count;
     Index oldEnd  = m_outerIndex[j]+m_innerNonZeros[j];
     for(Index k=m_outerIndex[j]; k<oldEnd; ++k)
     {
@@ -1036,7 +1032,7 @@ void SparseMatrix<Scalar,_Options,_Index>::sumupDuplicates()
       if(wi(i)>=start)
       {
         // we already meet this entry => accumulate it
-        m_data.value(wi(i)) += m_data.value(k);
+        m_data.value(wi(i)) = dup_func(m_data.value(wi(i)), m_data.value(k));
       }
       else
       {
@@ -1056,39 +1052,48 @@ void SparseMatrix<Scalar,_Options,_Index>::sumupDuplicates()
   m_data.resize(m_outerIndex[m_outerSize]);
 }
 
-template<typename Scalar, int _Options, typename _Index>
+template<typename Scalar, int _Options, typename _StorageIndex>
 template<typename OtherDerived>
-EIGEN_DONT_INLINE SparseMatrix<Scalar,_Options,_Index>& SparseMatrix<Scalar,_Options,_Index>::operator=(const SparseMatrixBase<OtherDerived>& other)
+EIGEN_DONT_INLINE SparseMatrix<Scalar,_Options,_StorageIndex>& SparseMatrix<Scalar,_Options,_StorageIndex>::operator=(const SparseMatrixBase<OtherDerived>& other)
 {
   EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
         YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-  
-  const bool needToTranspose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
+
+  #ifdef EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+    EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+  #endif
+      
+  const bool needToTranspose = (Flags & RowMajorBit) != (internal::evaluator<OtherDerived>::Flags & RowMajorBit);
   if (needToTranspose)
   {
+    #ifdef EIGEN_SPARSE_TRANSPOSED_COPY_PLUGIN
+      EIGEN_SPARSE_TRANSPOSED_COPY_PLUGIN
+    #endif
     // two passes algorithm:
     //  1 - compute the number of coeffs per dest inner vector
     //  2 - do the actual copy/eval
     // Since each coeff of the rhs has to be evaluated twice, let's evaluate it if needed
-    typedef typename internal::nested<OtherDerived,2>::type OtherCopy;
+    typedef typename internal::nested_eval<OtherDerived,2,typename internal::plain_matrix_type<OtherDerived>::type >::type OtherCopy;
     typedef typename internal::remove_all<OtherCopy>::type _OtherCopy;
+    typedef internal::evaluator<_OtherCopy> OtherCopyEval;
     OtherCopy otherCopy(other.derived());
+    OtherCopyEval otherCopyEval(otherCopy);
 
     SparseMatrix dest(other.rows(),other.cols());
-    Eigen::Map<Matrix<Index, Dynamic, 1> > (dest.m_outerIndex,dest.outerSize()).setZero();
+    Eigen::Map<IndexVector> (dest.m_outerIndex,dest.outerSize()).setZero();
 
     // pass 1
     // FIXME the above copy could be merged with that pass
     for (Index j=0; j<otherCopy.outerSize(); ++j)
-      for (typename _OtherCopy::InnerIterator it(otherCopy, j); it; ++it)
+      for (typename OtherCopyEval::InnerIterator it(otherCopyEval, j); it; ++it)
         ++dest.m_outerIndex[it.index()];
 
     // prefix sum
-    Index count = 0;
-    Matrix<Index,Dynamic,1> positions(dest.outerSize());
+    StorageIndex count = 0;
+    IndexVector positions(dest.outerSize());
     for (Index j=0; j<dest.outerSize(); ++j)
     {
-      Index tmp = dest.m_outerIndex[j];
+      StorageIndex tmp = dest.m_outerIndex[j];
       dest.m_outerIndex[j] = count;
       positions[j] = count;
       count += tmp;
@@ -1097,9 +1102,9 @@ EIGEN_DONT_INLINE SparseMatrix<Scalar,_Options,_Index>& SparseMatrix<Scalar,_Opt
     // alloc
     dest.m_data.resize(count);
     // pass 2
-    for (Index j=0; j<otherCopy.outerSize(); ++j)
+    for (StorageIndex j=0; j<otherCopy.outerSize(); ++j)
     {
-      for (typename _OtherCopy::InnerIterator it(otherCopy, j); it; ++it)
+      for (typename OtherCopyEval::InnerIterator it(otherCopyEval, j); it; ++it)
       {
         Index pos = positions[it.index()]++;
         dest.m_data.index(pos) = j;
@@ -1112,26 +1117,148 @@ EIGEN_DONT_INLINE SparseMatrix<Scalar,_Options,_Index>& SparseMatrix<Scalar,_Opt
   else
   {
     if(other.isRValue())
+    {
       initAssignment(other.derived());
+    }
     // there is no special optimization
     return Base::operator=(other.derived());
   }
 }
 
-template<typename _Scalar, int _Options, typename _Index>
-EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& SparseMatrix<_Scalar,_Options,_Index>::insertUncompressed(Index row, Index col)
+template<typename _Scalar, int _Options, typename _StorageIndex>
+typename SparseMatrix<_Scalar,_Options,_StorageIndex>::Scalar& SparseMatrix<_Scalar,_Options,_StorageIndex>::insert(Index row, Index col)
+{
+  eigen_assert(row>=0 && row<rows() && col>=0 && col<cols());
+  
+  const Index outer = IsRowMajor ? row : col;
+  const Index inner = IsRowMajor ? col : row;
+  
+  if(isCompressed())
+  {
+    if(nonZeros()==0)
+    {
+      // reserve space if not already done
+      if(m_data.allocatedSize()==0)
+        m_data.reserve(2*m_innerSize);
+      
+      // turn the matrix into non-compressed mode
+      m_innerNonZeros = static_cast<StorageIndex*>(std::malloc(m_outerSize * sizeof(StorageIndex)));
+      if(!m_innerNonZeros) internal::throw_std_bad_alloc();
+      
+      memset(m_innerNonZeros, 0, (m_outerSize)*sizeof(StorageIndex));
+      
+      // pack all inner-vectors to the end of the pre-allocated space
+      // and allocate the entire free-space to the first inner-vector
+      StorageIndex end = convert_index(m_data.allocatedSize());
+      for(Index j=1; j<=m_outerSize; ++j)
+        m_outerIndex[j] = end;
+    }
+    else
+    {
+      // turn the matrix into non-compressed mode
+      m_innerNonZeros = static_cast<StorageIndex*>(std::malloc(m_outerSize * sizeof(StorageIndex)));
+      if(!m_innerNonZeros) internal::throw_std_bad_alloc();
+      for(Index j=0; j<m_outerSize; ++j)
+        m_innerNonZeros[j] = m_outerIndex[j+1]-m_outerIndex[j];
+    }
+  }
+  
+  // check whether we can do a fast "push back" insertion
+  Index data_end = m_data.allocatedSize();
+  
+  // First case: we are filling a new inner vector which is packed at the end.
+  // We assume that all remaining inner-vectors are also empty and packed to the end.
+  if(m_outerIndex[outer]==data_end)
+  {
+    eigen_internal_assert(m_innerNonZeros[outer]==0);
+    
+    // pack previous empty inner-vectors to end of the used-space
+    // and allocate the entire free-space to the current inner-vector.
+    StorageIndex p = convert_index(m_data.size());
+    Index j = outer;
+    while(j>=0 && m_innerNonZeros[j]==0)
+      m_outerIndex[j--] = p;
+    
+    // push back the new element
+    ++m_innerNonZeros[outer];
+    m_data.append(Scalar(0), inner);
+    
+    // check for reallocation
+    if(data_end != m_data.allocatedSize())
+    {
+      // m_data has been reallocated
+      //  -> move remaining inner-vectors back to the end of the free-space
+      //     so that the entire free-space is allocated to the current inner-vector.
+      eigen_internal_assert(data_end < m_data.allocatedSize());
+      StorageIndex new_end = convert_index(m_data.allocatedSize());
+      for(Index k=outer+1; k<=m_outerSize; ++k)
+        if(m_outerIndex[k]==data_end)
+          m_outerIndex[k] = new_end;
+    }
+    return m_data.value(p);
+  }
+  
+  // Second case: the next inner-vector is packed to the end
+  // and the current inner-vector end match the used-space.
+  if(m_outerIndex[outer+1]==data_end && m_outerIndex[outer]+m_innerNonZeros[outer]==m_data.size())
+  {
+    eigen_internal_assert(outer+1==m_outerSize || m_innerNonZeros[outer+1]==0);
+    
+    // add space for the new element
+    ++m_innerNonZeros[outer];
+    m_data.resize(m_data.size()+1);
+    
+    // check for reallocation
+    if(data_end != m_data.allocatedSize())
+    {
+      // m_data has been reallocated
+      //  -> move remaining inner-vectors back to the end of the free-space
+      //     so that the entire free-space is allocated to the current inner-vector.
+      eigen_internal_assert(data_end < m_data.allocatedSize());
+      StorageIndex new_end = convert_index(m_data.allocatedSize());
+      for(Index k=outer+1; k<=m_outerSize; ++k)
+        if(m_outerIndex[k]==data_end)
+          m_outerIndex[k] = new_end;
+    }
+    
+    // and insert it at the right position (sorted insertion)
+    Index startId = m_outerIndex[outer];
+    Index p = m_outerIndex[outer]+m_innerNonZeros[outer]-1;
+    while ( (p > startId) && (m_data.index(p-1) > inner) )
+    {
+      m_data.index(p) = m_data.index(p-1);
+      m_data.value(p) = m_data.value(p-1);
+      --p;
+    }
+    
+    m_data.index(p) = convert_index(inner);
+    return (m_data.value(p) = 0);
+  }
+  
+  if(m_data.size() != m_data.allocatedSize())
+  {
+    // make sure the matrix is compatible to random un-compressed insertion:
+    m_data.resize(m_data.allocatedSize());
+    this->reserveInnerVectors(Array<StorageIndex,Dynamic,1>::Constant(m_outerSize, 2));
+  }
+  
+  return insertUncompressed(row,col);
+}
+    
+template<typename _Scalar, int _Options, typename _StorageIndex>
+EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_StorageIndex>::Scalar& SparseMatrix<_Scalar,_Options,_StorageIndex>::insertUncompressed(Index row, Index col)
 {
   eigen_assert(!isCompressed());
 
   const Index outer = IsRowMajor ? row : col;
-  const Index inner = IsRowMajor ? col : row;
+  const StorageIndex inner = convert_index(IsRowMajor ? col : row);
 
   Index room = m_outerIndex[outer+1] - m_outerIndex[outer];
-  Index innerNNZ = m_innerNonZeros[outer];
+  StorageIndex innerNNZ = m_innerNonZeros[outer];
   if(innerNNZ>=room)
   {
     // this inner vector is full, we need to reallocate the whole buffer :(
-    reserve(SingletonVector(outer,std::max<Index>(2,innerNNZ)));
+    reserve(SingletonVector(outer,std::max<StorageIndex>(2,innerNNZ)));
   }
 
   Index startId = m_outerIndex[outer];
@@ -1142,16 +1269,16 @@ EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& Sparse
     m_data.value(p) = m_data.value(p-1);
     --p;
   }
-  eigen_assert((p<=startId || m_data.index(p-1)!=inner) && "you cannot insert an element that already exist, you must call coeffRef to this end");
+  eigen_assert((p<=startId || m_data.index(p-1)!=inner) && "you cannot insert an element that already exists, you must call coeffRef to this end");
 
   m_innerNonZeros[outer]++;
 
   m_data.index(p) = inner;
-  return (m_data.value(p) = 0);
+  return (m_data.value(p) = Scalar(0));
 }
 
-template<typename _Scalar, int _Options, typename _Index>
-EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& SparseMatrix<_Scalar,_Options,_Index>::insertCompressed(Index row, Index col)
+template<typename _Scalar, int _Options, typename _StorageIndex>
+EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_StorageIndex>::Scalar& SparseMatrix<_Scalar,_Options,_StorageIndex>::insertCompressed(Index row, Index col)
 {
   eigen_assert(isCompressed());
 
@@ -1164,7 +1291,7 @@ EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& Sparse
     // we start a new inner vector
     while (previousOuter>=0 && m_outerIndex[previousOuter]==0)
     {
-      m_outerIndex[previousOuter] = static_cast<Index>(m_data.size());
+      m_outerIndex[previousOuter] = convert_index(m_data.size());
       --previousOuter;
     }
     m_outerIndex[outer+1] = m_outerIndex[outer];
@@ -1174,11 +1301,11 @@ EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& Sparse
   // starts with: [ 0 0 0 0 0 1 ...] and we are inserted in, e.g.,
   // the 2nd inner vector...
   bool isLastVec = (!(previousOuter==-1 && m_data.size()!=0))
-                && (size_t(m_outerIndex[outer+1]) == m_data.size());
+                && (std::size_t(m_outerIndex[outer+1]) == m_data.size());
 
-  size_t startId = m_outerIndex[outer];
-  // FIXME let's make sure sizeof(long int) == sizeof(size_t)
-  size_t p = m_outerIndex[outer+1];
+  std::size_t startId = m_outerIndex[outer];
+  // FIXME let's make sure sizeof(long int) == sizeof(std::size_t)
+  std::size_t p = m_outerIndex[outer+1];
   ++m_outerIndex[outer+1];
 
   double reallocRatio = 1;
@@ -1254,7 +1381,21 @@ EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& Sparse
   }
 
   m_data.index(p) = inner;
-  return (m_data.value(p) = 0);
+  return (m_data.value(p) = Scalar(0));
+}
+
+namespace internal {
+
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct evaluator<SparseMatrix<_Scalar,_Options,_StorageIndex> >
+  : evaluator<SparseCompressedBase<SparseMatrix<_Scalar,_Options,_StorageIndex> > >
+{
+  typedef evaluator<SparseCompressedBase<SparseMatrix<_Scalar,_Options,_StorageIndex> > > Base;
+  typedef SparseMatrix<_Scalar,_Options,_StorageIndex> SparseMatrixType;
+  evaluator() : Base() {}
+  explicit evaluator(const SparseMatrixType &mat) : Base(mat) {}
+};
+
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseMatrixBase.h b/extern/Eigen3/Eigen/src/SparseCore/SparseMatrixBase.h
index 9341d9ad2c0..c6b548f11ad 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseMatrixBase.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseMatrixBase.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -18,38 +18,41 @@ namespace Eigen {
   *
   * \brief Base class of any sparse matrices or sparse expressions
   *
-  * \tparam Derived
+  * \tparam Derived is the derived type, e.g. a sparse matrix type, or an expression, etc.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_SPARSEMATRIXBASE_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_SPARSEMATRIXBASE_PLUGIN.
   */
 template<typename Derived> class SparseMatrixBase
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-  : public internal::special_scalar_op_base<Derived,typename internal::traits<Derived>::Scalar,
-                                            typename NumTraits<typename internal::traits<Derived>::Scalar>::Real,
-                                            EigenBase<Derived> >
-#else
   : public EigenBase<Derived>
-#endif // not EIGEN_PARSED_BY_DOXYGEN
 {
   public:
 
     typedef typename internal::traits<Derived>::Scalar Scalar;
+    
+    /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc.
+      *
+      * It is an alias for the Scalar type */
+    typedef Scalar value_type;
+    
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
+
+    /** The integer type used to \b store indices within a SparseMatrix.
+      * For a \c SparseMatrix<Scalar,Options,IndexType> it an alias of the third template parameter \c IndexType. */
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+
     typedef typename internal::add_const_on_value_type_if_arithmetic<
                          typename internal::packet_traits<Scalar>::type
                      >::type PacketReturnType;
 
     typedef SparseMatrixBase StorageBaseType;
+
+    typedef Matrix<StorageIndex,Dynamic,1> IndexVector;
+    typedef Matrix<Scalar,Dynamic,1> ScalarVector;
     
     template<typename OtherDerived>
-    Derived& operator=(const EigenBase<OtherDerived> &other)
-    {
-      other.derived().evalTo(derived());
-      return derived();
-    }
+    Derived& operator=(const EigenBase<OtherDerived> &other);
 
     enum {
 
@@ -89,11 +92,6 @@ template<typename Derived> class SparseMatrixBase
           * constructed from this one. See the \ref flags "list of flags".
           */
 
-      CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
-        /**< This is a rough measure of how expensive it is to read one coefficient from
-          * this expression.
-          */
-
       IsRowMajor = Flags&RowMajorBit ? 1 : 0,
       
       InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime)
@@ -109,10 +107,11 @@ template<typename Derived> class SparseMatrixBase
                         CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, Eigen::Transpose<const Derived> >,
                         Transpose<const Derived>
                      >::type AdjointReturnType;
+    typedef Transpose<Derived> TransposeReturnType;
+    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
 
-
-    typedef SparseMatrix<Scalar, Flags&RowMajorBit ? RowMajor : ColMajor, Index> PlainObject;
-
+    // FIXME storage order do not match evaluator storage order
+    typedef SparseMatrix<Scalar, Flags&RowMajorBit ? RowMajor : ColMajor, StorageIndex> PlainObject;
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** This is the "real scalar" type; if the \a Scalar type is already real numbers
@@ -130,6 +129,8 @@ template<typename Derived> class SparseMatrixBase
     /** \internal Represents a matrix with all coefficients equal to one another*/
     typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Matrix<Scalar,Dynamic,Dynamic> > ConstantReturnType;
 
+    /** type of the equivalent dense matrix */
+    typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> DenseMatrixType;
     /** type of the equivalent square matrix */
     typedef Matrix<Scalar,EIGEN_SIZE_MAX(RowsAtCompileTime,ColsAtCompileTime),
                           EIGEN_SIZE_MAX(RowsAtCompileTime,ColsAtCompileTime)> SquareMatrixType;
@@ -139,11 +140,20 @@ template<typename Derived> class SparseMatrixBase
     inline Derived& const_cast_derived() const
     { return *static_cast<Derived*>(const_cast<SparseMatrixBase*>(this)); }
 
-    typedef internal::special_scalar_op_base<Derived, Scalar, RealScalar, EigenBase<Derived> > Base;
-    using Base::operator*;
+    typedef EigenBase<Derived> Base;
+
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::SparseMatrixBase
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+#define EIGEN_DOC_UNARY_ADDONS(METHOD,OP)           /** <p>This method does not change the sparsity of \c *this: the OP is applied to explicitly stored coefficients only. \sa SparseCompressedBase::coeffs() </p> */
+#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL      /** <p> \warning This method returns a read-only expression for any sparse matrices. \sa \ref TutorialSparse_SubMatrices "Sparse block operations" </p> */
+#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND) /** <p> \warning This method returns a read-write expression for COND sparse matrices only. Otherwise, the returned expression is read-only. \sa \ref TutorialSparse_SubMatrices "Sparse block operations" </p> */
+#else
+#define EIGEN_DOC_UNARY_ADDONS(X,Y)
+#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND)
+#endif
 #   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/CommonCwiseBinaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
@@ -152,8 +162,10 @@ template<typename Derived> class SparseMatrixBase
 #   ifdef EIGEN_SPARSEMATRIXBASE_PLUGIN
 #     include EIGEN_SPARSEMATRIXBASE_PLUGIN
 #   endif
-#   undef EIGEN_CURRENT_STORAGE_BASE_CLASS
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
+#undef EIGEN_DOC_UNARY_ADDONS
+#undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF
 
     /** \returns the number of rows. \sa cols() */
     inline Index rows() const { return derived().rows(); }
@@ -162,9 +174,6 @@ template<typename Derived> class SparseMatrixBase
     /** \returns the number of coefficients, which is \a rows()*cols().
       * \sa rows(), cols(). */
     inline Index size() const { return rows() * cols(); }
-    /** \returns the number of nonzero coefficients which is in practice the number
-      * of stored coefficients. */
-    inline Index nonZeros() const { return derived().nonZeros(); }
     /** \returns true if either the number of rows or the number of columns is equal to 1.
       * In other words, this function returns
       * \code rows()==1 || cols()==1 \endcode
@@ -184,93 +193,23 @@ template<typename Derived> class SparseMatrixBase
 
     
     template<typename OtherDerived>
-    Derived& operator=(const ReturnByValue<OtherDerived>& other)
-    {
-      other.evalTo(derived());
-      return derived();
-    }
-
+    Derived& operator=(const ReturnByValue<OtherDerived>& other);
 
     template<typename OtherDerived>
-    inline Derived& operator=(const SparseMatrixBase<OtherDerived>& other)
-    {
-      return assign(other.derived());
-    }
+    inline Derived& operator=(const SparseMatrixBase<OtherDerived>& other);
 
-    inline Derived& operator=(const Derived& other)
-    {
-//       if (other.isRValue())
-//         derived().swap(other.const_cast_derived());
-//       else
-      return assign(other.derived());
-    }
+    inline Derived& operator=(const Derived& other);
 
   protected:
 
     template<typename OtherDerived>
-    inline Derived& assign(const OtherDerived& other)
-    {
-      const bool transpose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
-      const Index outerSize = (int(OtherDerived::Flags) & RowMajorBit) ? other.rows() : other.cols();
-      if ((!transpose) && other.isRValue())
-      {
-        // eval without temporary
-        derived().resize(other.rows(), other.cols());
-        derived().setZero();
-        derived().reserve((std::max)(this->rows(),this->cols())*2);
-        for (Index j=0; j<outerSize; ++j)
-        {
-          derived().startVec(j);
-          for (typename OtherDerived::InnerIterator it(other, j); it; ++it)
-          {
-            Scalar v = it.value();
-            derived().insertBackByOuterInner(j,it.index()) = v;
-          }
-        }
-        derived().finalize();
-      }
-      else
-      {
-        assignGeneric(other);
-      }
-      return derived();
-    }
+    inline Derived& assign(const OtherDerived& other);
 
     template<typename OtherDerived>
-    inline void assignGeneric(const OtherDerived& other)
-    {
-      //const bool transpose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
-      eigen_assert(( ((internal::traits<Derived>::SupportedAccessPatterns&OuterRandomAccessPattern)==OuterRandomAccessPattern) ||
-                  (!((Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit)))) &&
-                  "the transpose operation is supposed to be handled in SparseMatrix::operator=");
-
-      enum { Flip = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit) };
-
-      const Index outerSize = other.outerSize();
-      //typedef typename internal::conditional<transpose, LinkedVectorMatrix<Scalar,Flags&RowMajorBit>, Derived>::type TempType;
-      // thanks to shallow copies, we always eval to a tempary
-      Derived temp(other.rows(), other.cols());
-
-      temp.reserve((std::max)(this->rows(),this->cols())*2);
-      for (Index j=0; j<outerSize; ++j)
-      {
-        temp.startVec(j);
-        for (typename OtherDerived::InnerIterator it(other.derived(), j); it; ++it)
-        {
-          Scalar v = it.value();
-          temp.insertBackByOuterInner(Flip?it.index():j,Flip?j:it.index()) = v;
-        }
-      }
-      temp.finalize();
-
-      derived() = temp.markAsRValue();
-    }
+    inline void assignGeneric(const OtherDerived& other);
 
   public:
 
-    template<typename Lhs, typename Rhs>
-    inline Derived& operator=(const SparseSparseProduct<Lhs,Rhs>& product);
-
     friend std::ostream & operator << (std::ostream & s, const SparseMatrixBase& m)
     {
       typedef typename Derived::Nested Nested;
@@ -278,11 +217,12 @@ template<typename Derived> class SparseMatrixBase
 
       if (Flags&RowMajorBit)
       {
-        const Nested nm(m.derived());
+        Nested nm(m.derived());
+        internal::evaluator<NestedCleaned> thisEval(nm);
         for (Index row=0; row<nm.outerSize(); ++row)
         {
           Index col = 0;
-          for (typename NestedCleaned::InnerIterator it(nm.derived(), row); it; ++it)
+          for (typename internal::evaluator<NestedCleaned>::InnerIterator it(thisEval, row); it; ++it)
           {
             for ( ; col<it.index(); ++col)
               s << "0 ";
@@ -296,10 +236,11 @@ template<typename Derived> class SparseMatrixBase
       }
       else
       {
-        const Nested nm(m.derived());
+        Nested nm(m.derived());
+        internal::evaluator<NestedCleaned> thisEval(nm);
         if (m.cols() == 1) {
           Index row = 0;
-          for (typename NestedCleaned::InnerIterator it(nm.derived(), 0); it; ++it)
+          for (typename internal::evaluator<NestedCleaned>::InnerIterator it(thisEval, 0); it; ++it)
           {
             for ( ; row<it.index(); ++row)
               s << "0" << std::endl;
@@ -311,8 +252,8 @@ template<typename Derived> class SparseMatrixBase
         }
         else
         {
-          SparseMatrix<Scalar, RowMajorBit, Index> trans = m;
-          s << static_cast<const SparseMatrixBase<SparseMatrix<Scalar, RowMajorBit, Index> >&>(trans);
+          SparseMatrix<Scalar, RowMajorBit, StorageIndex> trans = m;
+          s << static_cast<const SparseMatrixBase<SparseMatrix<Scalar, RowMajorBit, StorageIndex> >&>(trans);
         }
       }
       return s;
@@ -322,12 +263,22 @@ template<typename Derived> class SparseMatrixBase
     Derived& operator+=(const SparseMatrixBase<OtherDerived>& other);
     template<typename OtherDerived>
     Derived& operator-=(const SparseMatrixBase<OtherDerived>& other);
+    
+    template<typename OtherDerived>
+    Derived& operator+=(const DiagonalBase<OtherDerived>& other);
+    template<typename OtherDerived>
+    Derived& operator-=(const DiagonalBase<OtherDerived>& other);
+
+    template<typename OtherDerived>
+    Derived& operator+=(const EigenBase<OtherDerived> &other);
+    template<typename OtherDerived>
+    Derived& operator-=(const EigenBase<OtherDerived> &other);
 
     Derived& operator*=(const Scalar& other);
     Derived& operator/=(const Scalar& other);
 
     template<typename OtherDerived> struct CwiseProductDenseReturnType {
-      typedef CwiseBinaryOp<internal::scalar_product_op<typename internal::scalar_product_traits<
+      typedef CwiseBinaryOp<internal::scalar_product_op<typename ScalarBinaryOpTraits<
                                                           typename internal::traits<Derived>::Scalar,
                                                           typename internal::traits<OtherDerived>::Scalar
                                                         >::ReturnType>,
@@ -340,36 +291,37 @@ template<typename Derived> class SparseMatrixBase
     EIGEN_STRONG_INLINE const typename CwiseProductDenseReturnType<OtherDerived>::Type
     cwiseProduct(const MatrixBase<OtherDerived> &other) const;
 
-    // sparse * sparse
-    template<typename OtherDerived>
-    const typename SparseSparseProductReturnType<Derived,OtherDerived>::Type
-    operator*(const SparseMatrixBase<OtherDerived> &other) const;
-
     // sparse * diagonal
     template<typename OtherDerived>
-    const SparseDiagonalProduct<Derived,OtherDerived>
-    operator*(const DiagonalBase<OtherDerived> &other) const;
+    const Product<Derived,OtherDerived>
+    operator*(const DiagonalBase<OtherDerived> &other) const
+    { return Product<Derived,OtherDerived>(derived(), other.derived()); }
 
     // diagonal * sparse
     template<typename OtherDerived> friend
-    const SparseDiagonalProduct<OtherDerived,Derived>
+    const Product<OtherDerived,Derived>
     operator*(const DiagonalBase<OtherDerived> &lhs, const SparseMatrixBase& rhs)
-    { return SparseDiagonalProduct<OtherDerived,Derived>(lhs.derived(), rhs.derived()); }
-
-    /** dense * sparse (return a dense object unless it is an outer product) */
-    template<typename OtherDerived> friend
-    const typename DenseSparseProductReturnType<OtherDerived,Derived>::Type
-    operator*(const MatrixBase<OtherDerived>& lhs, const Derived& rhs)
-    { return typename DenseSparseProductReturnType<OtherDerived,Derived>::Type(lhs.derived(),rhs); }
-
-    /** sparse * dense (returns a dense object unless it is an outer product) */
+    { return Product<OtherDerived,Derived>(lhs.derived(), rhs.derived()); }
+    
+    // sparse * sparse
+    template<typename OtherDerived>
+    const Product<Derived,OtherDerived,AliasFreeProduct>
+    operator*(const SparseMatrixBase<OtherDerived> &other) const;
+    
+    // sparse * dense
     template<typename OtherDerived>
-    const typename SparseDenseProductReturnType<Derived,OtherDerived>::Type
+    const Product<Derived,OtherDerived>
     operator*(const MatrixBase<OtherDerived> &other) const
-    { return typename SparseDenseProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived()); }
+    { return Product<Derived,OtherDerived>(derived(), other.derived()); }
+    
+    // dense * sparse
+    template<typename OtherDerived> friend
+    const Product<OtherDerived,Derived>
+    operator*(const MatrixBase<OtherDerived> &lhs, const SparseMatrixBase& rhs)
+    { return Product<OtherDerived,Derived>(lhs.derived(), rhs.derived()); }
     
      /** \returns an expression of P H P^-1 where H is the matrix represented by \c *this */
-    SparseSymmetricPermutationProduct<Derived,Upper|Lower> twistedBy(const PermutationMatrix<Dynamic,Dynamic,Index>& perm) const
+    SparseSymmetricPermutationProduct<Derived,Upper|Lower> twistedBy(const PermutationMatrix<Dynamic,Dynamic,StorageIndex>& perm) const
     {
       return SparseSymmetricPermutationProduct<Derived,Upper|Lower>(derived(), perm);
     }
@@ -377,22 +329,16 @@ template<typename Derived> class SparseMatrixBase
     template<typename OtherDerived>
     Derived& operator*=(const SparseMatrixBase<OtherDerived>& other);
 
-    #ifdef EIGEN2_SUPPORT
-    // deprecated
-    template<typename OtherDerived>
-    typename internal::plain_matrix_type_column_major<OtherDerived>::type
-    solveTriangular(const MatrixBase<OtherDerived>& other) const;
-
-    // deprecated
-    template<typename OtherDerived>
-    void solveTriangularInPlace(MatrixBase<OtherDerived>& other) const;
-    #endif // EIGEN2_SUPPORT
-
     template<int Mode>
-    inline const SparseTriangularView<Derived, Mode> triangularView() const;
+    inline const TriangularView<const Derived, Mode> triangularView() const;
+    
+    template<unsigned int UpLo> struct SelfAdjointViewReturnType { typedef SparseSelfAdjointView<Derived, UpLo> Type; };
+    template<unsigned int UpLo> struct ConstSelfAdjointViewReturnType { typedef const SparseSelfAdjointView<const Derived, UpLo> Type; };
 
-    template<unsigned int UpLo> inline const SparseSelfAdjointView<Derived, UpLo> selfadjointView() const;
-    template<unsigned int UpLo> inline SparseSelfAdjointView<Derived, UpLo> selfadjointView();
+    template<unsigned int UpLo> inline 
+    typename ConstSelfAdjointViewReturnType<UpLo>::Type selfadjointView() const;
+    template<unsigned int UpLo> inline
+    typename SelfAdjointViewReturnType<UpLo>::Type selfadjointView();
 
     template<typename OtherDerived> Scalar dot(const MatrixBase<OtherDerived>& other) const;
     template<typename OtherDerived> Scalar dot(const SparseMatrixBase<OtherDerived>& other) const;
@@ -400,9 +346,9 @@ template<typename Derived> class SparseMatrixBase
     RealScalar norm()  const;
     RealScalar blueNorm() const;
 
-    Transpose<Derived> transpose() { return derived(); }
-    const Transpose<const Derived> transpose() const { return derived(); }
-    const AdjointReturnType adjoint() const { return transpose(); }
+    TransposeReturnType transpose() { return TransposeReturnType(derived()); }
+    const ConstTransposeReturnType transpose() const { return ConstTransposeReturnType(derived()); }
+    const AdjointReturnType adjoint() const { return AdjointReturnType(transpose()); }
 
     // inner-vector
     typedef Block<Derived,IsRowMajor?1:Dynamic,IsRowMajor?Dynamic:1,true>       InnerVectorReturnType;
@@ -416,25 +362,14 @@ template<typename Derived> class SparseMatrixBase
     InnerVectorsReturnType innerVectors(Index outerStart, Index outerSize);
     const ConstInnerVectorsReturnType innerVectors(Index outerStart, Index outerSize) const;
 
-    /** \internal use operator= */
-    template<typename DenseDerived>
-    void evalTo(MatrixBase<DenseDerived>& dst) const
-    {
-      dst.setZero();
-      for (Index j=0; j<outerSize(); ++j)
-        for (typename Derived::InnerIterator i(derived(),j); i; ++i)
-          dst.coeffRef(i.row(),i.col()) = i.value();
-    }
-
-    Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> toDense() const
+    DenseMatrixType toDense() const
     {
-      return derived();
+      return DenseMatrixType(derived());
     }
 
     template<typename OtherDerived>
     bool isApprox(const SparseMatrixBase<OtherDerived>& other,
-                  const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const
-    { return toDense().isApprox(other.toDense(),prec); }
+                  const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
 
     template<typename OtherDerived>
     bool isApprox(const MatrixBase<OtherDerived>& other,
@@ -450,10 +385,19 @@ template<typename Derived> class SparseMatrixBase
     { return typename internal::eval<Derived>::type(derived()); }
 
     Scalar sum() const;
+    
+    inline const SparseView<Derived>
+    pruned(const Scalar& reference = Scalar(0), const RealScalar& epsilon = NumTraits<Scalar>::dummy_precision()) const;
 
   protected:
 
     bool m_isRValue;
+
+    static inline StorageIndex convert_index(const Index idx) {
+      return internal::convert_index<StorageIndex>(idx);
+    }
+  private:
+    template<typename Dest> void evalTo(Dest &) const;
 };
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparsePermutation.h b/extern/Eigen3/Eigen/src/SparseCore/SparsePermutation.h
index 75e21000959..ef38357aefd 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparsePermutation.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparsePermutation.h
@@ -16,131 +16,161 @@ namespace Eigen {
 
 namespace internal {
 
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct traits<permut_sparsematrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
+template<typename ExpressionType, int Side, bool Transposed>
+struct permutation_matrix_product<ExpressionType, Side, Transposed, SparseShape>
 {
-  typedef typename remove_all<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
-  typedef typename MatrixTypeNestedCleaned::Scalar Scalar;
-  typedef typename MatrixTypeNestedCleaned::Index Index;
-  enum {
-    SrcStorageOrder = MatrixTypeNestedCleaned::Flags&RowMajorBit ? RowMajor : ColMajor,
-    MoveOuter = SrcStorageOrder==RowMajor ? Side==OnTheLeft : Side==OnTheRight
-  };
+    typedef typename nested_eval<ExpressionType, 1>::type MatrixType;
+    typedef typename remove_all<MatrixType>::type MatrixTypeCleaned;
 
-  typedef typename internal::conditional<MoveOuter,
-        SparseMatrix<Scalar,SrcStorageOrder,Index>,
-        SparseMatrix<Scalar,int(SrcStorageOrder)==RowMajor?ColMajor:RowMajor,Index> >::type ReturnType;
-};
-
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct permut_sparsematrix_product_retval
- : public ReturnByValue<permut_sparsematrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
-{
-    typedef typename remove_all<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
-    typedef typename MatrixTypeNestedCleaned::Scalar Scalar;
-    typedef typename MatrixTypeNestedCleaned::Index Index;
+    typedef typename MatrixTypeCleaned::Scalar Scalar;
+    typedef typename MatrixTypeCleaned::StorageIndex StorageIndex;
 
     enum {
-      SrcStorageOrder = MatrixTypeNestedCleaned::Flags&RowMajorBit ? RowMajor : ColMajor,
+      SrcStorageOrder = MatrixTypeCleaned::Flags&RowMajorBit ? RowMajor : ColMajor,
       MoveOuter = SrcStorageOrder==RowMajor ? Side==OnTheLeft : Side==OnTheRight
     };
+    
+    typedef typename internal::conditional<MoveOuter,
+        SparseMatrix<Scalar,SrcStorageOrder,StorageIndex>,
+        SparseMatrix<Scalar,int(SrcStorageOrder)==RowMajor?ColMajor:RowMajor,StorageIndex> >::type ReturnType;
 
-    permut_sparsematrix_product_retval(const PermutationType& perm, const MatrixType& matrix)
-      : m_permutation(perm), m_matrix(matrix)
-    {}
-
-    inline int rows() const { return m_matrix.rows(); }
-    inline int cols() const { return m_matrix.cols(); }
-
-    template<typename Dest> inline void evalTo(Dest& dst) const
+    template<typename Dest,typename PermutationType>
+    static inline void run(Dest& dst, const PermutationType& perm, const ExpressionType& xpr)
     {
+      MatrixType mat(xpr);
       if(MoveOuter)
       {
-        SparseMatrix<Scalar,SrcStorageOrder,Index> tmp(m_matrix.rows(), m_matrix.cols());
-        Matrix<Index,Dynamic,1> sizes(m_matrix.outerSize());
-        for(Index j=0; j<m_matrix.outerSize(); ++j)
+        SparseMatrix<Scalar,SrcStorageOrder,StorageIndex> tmp(mat.rows(), mat.cols());
+        Matrix<StorageIndex,Dynamic,1> sizes(mat.outerSize());
+        for(Index j=0; j<mat.outerSize(); ++j)
         {
-          Index jp = m_permutation.indices().coeff(j);
-          sizes[((Side==OnTheLeft) ^ Transposed) ? jp : j] = m_matrix.innerVector(((Side==OnTheRight) ^ Transposed) ? jp : j).nonZeros();
+          Index jp = perm.indices().coeff(j);
+          sizes[((Side==OnTheLeft) ^ Transposed) ? jp : j] = StorageIndex(mat.innerVector(((Side==OnTheRight) ^ Transposed) ? jp : j).nonZeros());
         }
         tmp.reserve(sizes);
-        for(Index j=0; j<m_matrix.outerSize(); ++j)
+        for(Index j=0; j<mat.outerSize(); ++j)
         {
-          Index jp = m_permutation.indices().coeff(j);
+          Index jp = perm.indices().coeff(j);
           Index jsrc = ((Side==OnTheRight) ^ Transposed) ? jp : j;
           Index jdst = ((Side==OnTheLeft) ^ Transposed) ? jp : j;
-          for(typename MatrixTypeNestedCleaned::InnerIterator it(m_matrix,jsrc); it; ++it)
+          for(typename MatrixTypeCleaned::InnerIterator it(mat,jsrc); it; ++it)
             tmp.insertByOuterInner(jdst,it.index()) = it.value();
         }
         dst = tmp;
       }
       else
       {
-        SparseMatrix<Scalar,int(SrcStorageOrder)==RowMajor?ColMajor:RowMajor,Index> tmp(m_matrix.rows(), m_matrix.cols());
-        Matrix<Index,Dynamic,1> sizes(tmp.outerSize());
+        SparseMatrix<Scalar,int(SrcStorageOrder)==RowMajor?ColMajor:RowMajor,StorageIndex> tmp(mat.rows(), mat.cols());
+        Matrix<StorageIndex,Dynamic,1> sizes(tmp.outerSize());
         sizes.setZero();
-        PermutationMatrix<Dynamic,Dynamic,Index> perm;
+        PermutationMatrix<Dynamic,Dynamic,StorageIndex> perm_cpy;
         if((Side==OnTheLeft) ^ Transposed)
-          perm = m_permutation;
+          perm_cpy = perm;
         else
-          perm = m_permutation.transpose();
+          perm_cpy = perm.transpose();
 
-        for(Index j=0; j<m_matrix.outerSize(); ++j)
-          for(typename MatrixTypeNestedCleaned::InnerIterator it(m_matrix,j); it; ++it)
-            sizes[perm.indices().coeff(it.index())]++;
+        for(Index j=0; j<mat.outerSize(); ++j)
+          for(typename MatrixTypeCleaned::InnerIterator it(mat,j); it; ++it)
+            sizes[perm_cpy.indices().coeff(it.index())]++;
         tmp.reserve(sizes);
-        for(Index j=0; j<m_matrix.outerSize(); ++j)
-          for(typename MatrixTypeNestedCleaned::InnerIterator it(m_matrix,j); it; ++it)
-            tmp.insertByOuterInner(perm.indices().coeff(it.index()),j) = it.value();
+        for(Index j=0; j<mat.outerSize(); ++j)
+          for(typename MatrixTypeCleaned::InnerIterator it(mat,j); it; ++it)
+            tmp.insertByOuterInner(perm_cpy.indices().coeff(it.index()),j) = it.value();
         dst = tmp;
       }
     }
-
-  protected:
-    const PermutationType& m_permutation;
-    typename MatrixType::Nested m_matrix;
 };
 
 }
 
+namespace internal {
+
+template <int ProductTag> struct product_promote_storage_type<Sparse,             PermutationStorage, ProductTag> { typedef Sparse ret; };
+template <int ProductTag> struct product_promote_storage_type<PermutationStorage, Sparse,             ProductTag> { typedef Sparse ret; };
+
+// TODO, the following two overloads are only needed to define the right temporary type through 
+// typename traits<permutation_sparse_matrix_product<Rhs,Lhs,OnTheRight,false> >::ReturnType
+// whereas it should be correctly handled by traits<Product<> >::PlainObject
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, AliasFreeProduct>, ProductTag, PermutationShape, SparseShape>
+  : public evaluator<typename permutation_matrix_product<Rhs,OnTheLeft,false,SparseShape>::ReturnType>
+{
+  typedef Product<Lhs, Rhs, AliasFreeProduct> XprType;
+  typedef typename permutation_matrix_product<Rhs,OnTheLeft,false,SparseShape>::ReturnType PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  enum {
+    Flags = Base::Flags | EvalBeforeNestingBit
+  };
+
+  explicit product_evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    generic_product_impl<Lhs, Rhs, PermutationShape, SparseShape, ProductTag>::evalTo(m_result, xpr.lhs(), xpr.rhs());
+  }
 
+protected:
+  PlainObject m_result;
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, AliasFreeProduct>, ProductTag, SparseShape, PermutationShape >
+  : public evaluator<typename permutation_matrix_product<Lhs,OnTheRight,false,SparseShape>::ReturnType>
+{
+  typedef Product<Lhs, Rhs, AliasFreeProduct> XprType;
+  typedef typename permutation_matrix_product<Lhs,OnTheRight,false,SparseShape>::ReturnType PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  enum {
+    Flags = Base::Flags | EvalBeforeNestingBit
+  };
+
+  explicit product_evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    generic_product_impl<Lhs, Rhs, SparseShape, PermutationShape, ProductTag>::evalTo(m_result, xpr.lhs(), xpr.rhs());
+  }
+
+protected:
+  PlainObject m_result;
+};
+
+} // end namespace internal
 
 /** \returns the matrix with the permutation applied to the columns
   */
 template<typename SparseDerived, typename PermDerived>
-inline const internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheRight, false>
+inline const Product<SparseDerived, PermDerived, AliasFreeProduct>
 operator*(const SparseMatrixBase<SparseDerived>& matrix, const PermutationBase<PermDerived>& perm)
-{
-  return internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheRight, false>(perm, matrix.derived());
-}
+{ return Product<SparseDerived, PermDerived, AliasFreeProduct>(matrix.derived(), perm.derived()); }
 
 /** \returns the matrix with the permutation applied to the rows
   */
 template<typename SparseDerived, typename PermDerived>
-inline const internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheLeft, false>
+inline const Product<PermDerived, SparseDerived, AliasFreeProduct>
 operator*( const PermutationBase<PermDerived>& perm, const SparseMatrixBase<SparseDerived>& matrix)
-{
-  return internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheLeft, false>(perm, matrix.derived());
-}
-
+{ return  Product<PermDerived, SparseDerived, AliasFreeProduct>(perm.derived(), matrix.derived()); }
 
 
 /** \returns the matrix with the inverse permutation applied to the columns.
   */
-template<typename SparseDerived, typename PermDerived>
-inline const internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheRight, true>
-operator*(const SparseMatrixBase<SparseDerived>& matrix, const Transpose<PermutationBase<PermDerived> >& tperm)
+template<typename SparseDerived, typename PermutationType>
+inline const Product<SparseDerived, Inverse<PermutationType>, AliasFreeProduct>
+operator*(const SparseMatrixBase<SparseDerived>& matrix, const InverseImpl<PermutationType, PermutationStorage>& tperm)
 {
-  return internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheRight, true>(tperm.nestedPermutation(), matrix.derived());
+  return Product<SparseDerived, Inverse<PermutationType>, AliasFreeProduct>(matrix.derived(), tperm.derived());
 }
 
 /** \returns the matrix with the inverse permutation applied to the rows.
   */
-template<typename SparseDerived, typename PermDerived>
-inline const internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheLeft, true>
-operator*(const Transpose<PermutationBase<PermDerived> >& tperm, const SparseMatrixBase<SparseDerived>& matrix)
+template<typename SparseDerived, typename PermutationType>
+inline const Product<Inverse<PermutationType>, SparseDerived, AliasFreeProduct>
+operator*(const InverseImpl<PermutationType,PermutationStorage>& tperm, const SparseMatrixBase<SparseDerived>& matrix)
 {
-  return internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheLeft, true>(tperm.nestedPermutation(), matrix.derived());
+  return Product<Inverse<PermutationType>, SparseDerived, AliasFreeProduct>(tperm.derived(), matrix.derived());
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseProduct.h b/extern/Eigen3/Eigen/src/SparseCore/SparseProduct.h
index cf766307008..4cbf68781a8 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseProduct.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseProduct.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,158 +12,6 @@
 
 namespace Eigen { 
 
-template<typename Lhs, typename Rhs>
-struct SparseSparseProductReturnType
-{
-  typedef typename internal::traits<Lhs>::Scalar Scalar;
-  typedef typename internal::traits<Lhs>::Index Index;
-  enum {
-    LhsRowMajor = internal::traits<Lhs>::Flags & RowMajorBit,
-    RhsRowMajor = internal::traits<Rhs>::Flags & RowMajorBit,
-    TransposeRhs = (!LhsRowMajor) && RhsRowMajor,
-    TransposeLhs = LhsRowMajor && (!RhsRowMajor)
-  };
-
-  typedef typename internal::conditional<TransposeLhs,
-    SparseMatrix<Scalar,0,Index>,
-    typename internal::nested<Lhs,Rhs::RowsAtCompileTime>::type>::type LhsNested;
-
-  typedef typename internal::conditional<TransposeRhs,
-    SparseMatrix<Scalar,0,Index>,
-    typename internal::nested<Rhs,Lhs::RowsAtCompileTime>::type>::type RhsNested;
-
-  typedef SparseSparseProduct<LhsNested, RhsNested> Type;
-};
-
-namespace internal {
-template<typename LhsNested, typename RhsNested>
-struct traits<SparseSparseProduct<LhsNested, RhsNested> >
-{
-  typedef MatrixXpr XprKind;
-  // clean the nested types:
-  typedef typename remove_all<LhsNested>::type _LhsNested;
-  typedef typename remove_all<RhsNested>::type _RhsNested;
-  typedef typename _LhsNested::Scalar Scalar;
-  typedef typename promote_index_type<typename traits<_LhsNested>::Index,
-                                         typename traits<_RhsNested>::Index>::type Index;
-
-  enum {
-    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
-    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
-    LhsFlags = _LhsNested::Flags,
-    RhsFlags = _RhsNested::Flags,
-
-    RowsAtCompileTime    = _LhsNested::RowsAtCompileTime,
-    ColsAtCompileTime    = _RhsNested::ColsAtCompileTime,
-    MaxRowsAtCompileTime = _LhsNested::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = _RhsNested::MaxColsAtCompileTime,
-
-    InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(_LhsNested::ColsAtCompileTime, _RhsNested::RowsAtCompileTime),
-
-    EvalToRowMajor = (RhsFlags & LhsFlags & RowMajorBit),
-
-    RemovedBits = ~(EvalToRowMajor ? 0 : RowMajorBit),
-
-    Flags = (int(LhsFlags | RhsFlags) & HereditaryBits & RemovedBits)
-          | EvalBeforeAssigningBit
-          | EvalBeforeNestingBit,
-
-    CoeffReadCost = Dynamic
-  };
-
-  typedef Sparse StorageKind;
-};
-
-} // end namespace internal
-
-template<typename LhsNested, typename RhsNested>
-class SparseSparseProduct : internal::no_assignment_operator,
-  public SparseMatrixBase<SparseSparseProduct<LhsNested, RhsNested> >
-{
-  public:
-
-    typedef SparseMatrixBase<SparseSparseProduct> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(SparseSparseProduct)
-
-  private:
-
-    typedef typename internal::traits<SparseSparseProduct>::_LhsNested _LhsNested;
-    typedef typename internal::traits<SparseSparseProduct>::_RhsNested _RhsNested;
-
-  public:
-
-    template<typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE SparseSparseProduct(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs), m_tolerance(0), m_conservative(true)
-    {
-      init();
-    }
-
-    template<typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE SparseSparseProduct(const Lhs& lhs, const Rhs& rhs, const RealScalar& tolerance)
-      : m_lhs(lhs), m_rhs(rhs), m_tolerance(tolerance), m_conservative(false)
-    {
-      init();
-    }
-
-    SparseSparseProduct pruned(const Scalar& reference = 0, const RealScalar& epsilon = NumTraits<RealScalar>::dummy_precision()) const
-    {
-      using std::abs;
-      return SparseSparseProduct(m_lhs,m_rhs,abs(reference)*epsilon);
-    }
-
-    template<typename Dest>
-    void evalTo(Dest& result) const
-    {
-      if(m_conservative)
-        internal::conservative_sparse_sparse_product_selector<_LhsNested, _RhsNested, Dest>::run(lhs(),rhs(),result);
-      else
-        internal::sparse_sparse_product_with_pruning_selector<_LhsNested, _RhsNested, Dest>::run(lhs(),rhs(),result,m_tolerance);
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
-
-    EIGEN_STRONG_INLINE const _LhsNested& lhs() const { return m_lhs; }
-    EIGEN_STRONG_INLINE const _RhsNested& rhs() const { return m_rhs; }
-
-  protected:
-    void init()
-    {
-      eigen_assert(m_lhs.cols() == m_rhs.rows());
-
-      enum {
-        ProductIsValid = _LhsNested::ColsAtCompileTime==Dynamic
-                      || _RhsNested::RowsAtCompileTime==Dynamic
-                      || int(_LhsNested::ColsAtCompileTime)==int(_RhsNested::RowsAtCompileTime),
-        AreVectors = _LhsNested::IsVectorAtCompileTime && _RhsNested::IsVectorAtCompileTime,
-        SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(_LhsNested,_RhsNested)
-      };
-      // note to the lost user:
-      //    * for a dot product use: v1.dot(v2)
-      //    * for a coeff-wise product use: v1.cwise()*v2
-      EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
-        INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
-      EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
-        INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
-      EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
-    }
-
-    LhsNested m_lhs;
-    RhsNested m_rhs;
-    RealScalar m_tolerance;
-    bool m_conservative;
-};
-
-// sparse = sparse * sparse
-template<typename Derived>
-template<typename Lhs, typename Rhs>
-inline Derived& SparseMatrixBase<Derived>::operator=(const SparseSparseProduct<Lhs,Rhs>& product)
-{
-  product.evalTo(derived());
-  return derived();
-}
-
 /** \returns an expression of the product of two sparse matrices.
   * By default a conservative product preserving the symbolic non zeros is performed.
   * The automatic pruning of the small values can be achieved by calling the pruned() function
@@ -177,12 +25,145 @@ inline Derived& SparseMatrixBase<Derived>::operator=(const SparseSparseProduct<L
   * */
 template<typename Derived>
 template<typename OtherDerived>
-inline const typename SparseSparseProductReturnType<Derived,OtherDerived>::Type
+inline const Product<Derived,OtherDerived,AliasFreeProduct>
 SparseMatrixBase<Derived>::operator*(const SparseMatrixBase<OtherDerived> &other) const
 {
-  return typename SparseSparseProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
+  return Product<Derived,OtherDerived,AliasFreeProduct>(derived(), other.derived());
 }
 
+namespace internal {
+
+// sparse * sparse
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseShape, SparseShape, ProductType>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    evalTo(dst, lhs, rhs, typename evaluator_traits<Dest>::Shape());
+  }
+
+  // dense += sparse * sparse
+  template<typename Dest,typename ActualLhs>
+  static void addTo(Dest& dst, const ActualLhs& lhs, const Rhs& rhs, typename enable_if<is_same<typename evaluator_traits<Dest>::Shape,DenseShape>::value,int*>::type* = 0)
+  {
+    typedef typename nested_eval<ActualLhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhs);
+    internal::sparse_sparse_to_dense_product_selector<typename remove_all<LhsNested>::type,
+                                                      typename remove_all<RhsNested>::type, Dest>::run(lhsNested,rhsNested,dst);
+  }
+
+  // dense -= sparse * sparse
+  template<typename Dest>
+  static void subTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, typename enable_if<is_same<typename evaluator_traits<Dest>::Shape,DenseShape>::value,int*>::type* = 0)
+  {
+    addTo(dst, -lhs, rhs);
+  }
+
+protected:
+
+  // sparse = sparse * sparse
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, SparseShape)
+  {
+    typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhs);
+    internal::conservative_sparse_sparse_product_selector<typename remove_all<LhsNested>::type,
+                                                          typename remove_all<RhsNested>::type, Dest>::run(lhsNested,rhsNested,dst);
+  }
+
+  // dense = sparse * sparse
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, DenseShape)
+  {
+    dst.setZero();
+    addTo(dst, lhs, rhs);
+  }
+};
+
+// sparse * sparse-triangular
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseShape, SparseTriangularShape, ProductType>
+ : public generic_product_impl<Lhs, Rhs, SparseShape, SparseShape, ProductType>
+{};
+
+// sparse-triangular * sparse
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseTriangularShape, SparseShape, ProductType>
+ : public generic_product_impl<Lhs, Rhs, SparseShape, SparseShape, ProductType>
+{};
+
+// dense = sparse-product (can be sparse*sparse, sparse*perm, etc.)
+template< typename DstXprType, typename Lhs, typename Rhs>
+struct Assignment<DstXprType, Product<Lhs,Rhs,AliasFreeProduct>, internal::assign_op<typename DstXprType::Scalar,typename Product<Lhs,Rhs,AliasFreeProduct>::Scalar>, Sparse2Dense>
+{
+  typedef Product<Lhs,Rhs,AliasFreeProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    
+    generic_product_impl<Lhs, Rhs>::evalTo(dst,src.lhs(),src.rhs());
+  }
+};
+
+// dense += sparse-product (can be sparse*sparse, sparse*perm, etc.)
+template< typename DstXprType, typename Lhs, typename Rhs>
+struct Assignment<DstXprType, Product<Lhs,Rhs,AliasFreeProduct>, internal::add_assign_op<typename DstXprType::Scalar,typename Product<Lhs,Rhs,AliasFreeProduct>::Scalar>, Sparse2Dense>
+{
+  typedef Product<Lhs,Rhs,AliasFreeProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &)
+  {
+    generic_product_impl<Lhs, Rhs>::addTo(dst,src.lhs(),src.rhs());
+  }
+};
+
+// dense -= sparse-product (can be sparse*sparse, sparse*perm, etc.)
+template< typename DstXprType, typename Lhs, typename Rhs>
+struct Assignment<DstXprType, Product<Lhs,Rhs,AliasFreeProduct>, internal::sub_assign_op<typename DstXprType::Scalar,typename Product<Lhs,Rhs,AliasFreeProduct>::Scalar>, Sparse2Dense>
+{
+  typedef Product<Lhs,Rhs,AliasFreeProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &)
+  {
+    generic_product_impl<Lhs, Rhs>::subTo(dst,src.lhs(),src.rhs());
+  }
+};
+
+template<typename Lhs, typename Rhs, int Options>
+struct unary_evaluator<SparseView<Product<Lhs, Rhs, Options> >, IteratorBased>
+ : public evaluator<typename Product<Lhs, Rhs, DefaultProduct>::PlainObject>
+{
+  typedef SparseView<Product<Lhs, Rhs, Options> > XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  explicit unary_evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    using std::abs;
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(xpr.nestedExpression().lhs());
+    RhsNested rhsNested(xpr.nestedExpression().rhs());
+
+    internal::sparse_sparse_product_with_pruning_selector<typename remove_all<LhsNested>::type,
+                                                          typename remove_all<RhsNested>::type, PlainObject>::run(lhsNested,rhsNested,m_result,
+                                                                                                                  abs(xpr.reference())*xpr.epsilon());
+  }
+
+protected:
+  PlainObject m_result;
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSEPRODUCT_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseRedux.h b/extern/Eigen3/Eigen/src/SparseCore/SparseRedux.h
index f3da93a71d4..45877496270 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseRedux.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseRedux.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -18,8 +18,9 @@ SparseMatrixBase<Derived>::sum() const
 {
   eigen_assert(rows()>0 && cols()>0 && "you are using a non initialized matrix");
   Scalar res(0);
+  internal::evaluator<Derived> thisEval(derived());
   for (Index j=0; j<outerSize(); ++j)
-    for (typename Derived::InnerIterator iter(derived(),j); iter; ++iter)
+    for (typename internal::evaluator<Derived>::InnerIterator iter(thisEval,j); iter; ++iter)
       res += iter.value();
   return res;
 }
@@ -29,7 +30,10 @@ typename internal::traits<SparseMatrix<_Scalar,_Options,_Index> >::Scalar
 SparseMatrix<_Scalar,_Options,_Index>::sum() const
 {
   eigen_assert(rows()>0 && cols()>0 && "you are using a non initialized matrix");
-  return Matrix<Scalar,1,Dynamic>::Map(&m_data.value(0), m_data.size()).sum();
+  if(this->isCompressed())
+    return Matrix<Scalar,1,Dynamic>::Map(m_data.valuePtr(), m_data.size()).sum();
+  else
+    return Base::sum();
 }
 
 template<typename _Scalar, int _Options, typename _Index>
@@ -37,7 +41,7 @@ typename internal::traits<SparseVector<_Scalar,_Options, _Index> >::Scalar
 SparseVector<_Scalar,_Options,_Index>::sum() const
 {
   eigen_assert(rows()>0 && cols()>0 && "you are using a non initialized matrix");
-  return Matrix<Scalar,1,Dynamic>::Map(&m_data.value(0), m_data.size()).sum();
+  return Matrix<Scalar,1,Dynamic>::Map(m_data.valuePtr(), m_data.size()).sum();
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseRef.h b/extern/Eigen3/Eigen/src/SparseCore/SparseRef.h
new file mode 100644
index 00000000000..d91f38f97ce
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseRef.h
@@ -0,0 +1,397 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSE_REF_H
+#define EIGEN_SPARSE_REF_H
+
+namespace Eigen {
+
+enum {
+  StandardCompressedFormat = 2 /**< used by Ref<SparseMatrix> to specify whether the input storage must be in standard compressed form */
+};
+  
+namespace internal {
+
+template<typename Derived> class SparseRefBase;
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int _Options, typename _StrideType>
+struct traits<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+  : public traits<SparseMatrix<MatScalar,MatOptions,MatIndex> >
+{
+  typedef SparseMatrix<MatScalar,MatOptions,MatIndex> PlainObjectType;
+  enum {
+    Options = _Options,
+    Flags = traits<PlainObjectType>::Flags | CompressedAccessBit | NestByRefBit
+  };
+
+  template<typename Derived> struct match {
+    enum {
+      StorageOrderMatch = PlainObjectType::IsVectorAtCompileTime || Derived::IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)),
+      MatchAtCompileTime = (Derived::Flags&CompressedAccessBit) && StorageOrderMatch
+    };
+    typedef typename internal::conditional<MatchAtCompileTime,internal::true_type,internal::false_type>::type type;
+  };
+  
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int _Options, typename _StrideType>
+struct traits<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+  : public traits<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+{
+  enum {
+    Flags = (traits<SparseMatrix<MatScalar,MatOptions,MatIndex> >::Flags | CompressedAccessBit | NestByRefBit) & ~LvalueBit
+  };
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int _Options, typename _StrideType>
+struct traits<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+  : public traits<SparseVector<MatScalar,MatOptions,MatIndex> >
+{
+  typedef SparseVector<MatScalar,MatOptions,MatIndex> PlainObjectType;
+  enum {
+    Options = _Options,
+    Flags = traits<PlainObjectType>::Flags | CompressedAccessBit | NestByRefBit
+  };
+
+  template<typename Derived> struct match {
+    enum {
+      MatchAtCompileTime = (Derived::Flags&CompressedAccessBit) && Derived::IsVectorAtCompileTime
+    };
+    typedef typename internal::conditional<MatchAtCompileTime,internal::true_type,internal::false_type>::type type;
+  };
+
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int _Options, typename _StrideType>
+struct traits<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+  : public traits<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+{
+  enum {
+    Flags = (traits<SparseVector<MatScalar,MatOptions,MatIndex> >::Flags | CompressedAccessBit | NestByRefBit) & ~LvalueBit
+  };
+};
+
+template<typename Derived>
+struct traits<SparseRefBase<Derived> > : public traits<Derived> {};
+
+template<typename Derived> class SparseRefBase
+  : public SparseMapBase<Derived>
+{
+public:
+
+  typedef SparseMapBase<Derived> Base;
+  EIGEN_SPARSE_PUBLIC_INTERFACE(SparseRefBase)
+
+  SparseRefBase()
+    : Base(RowsAtCompileTime==Dynamic?0:RowsAtCompileTime,ColsAtCompileTime==Dynamic?0:ColsAtCompileTime, 0, 0, 0, 0, 0)
+  {}
+  
+protected:
+
+  template<typename Expression>
+  void construct(Expression& expr)
+  {
+    if(expr.outerIndexPtr()==0)
+      ::new (static_cast<Base*>(this)) Base(expr.size(), expr.nonZeros(), expr.innerIndexPtr(), expr.valuePtr());
+    else
+      ::new (static_cast<Base*>(this)) Base(expr.rows(), expr.cols(), expr.nonZeros(), expr.outerIndexPtr(), expr.innerIndexPtr(), expr.valuePtr(), expr.innerNonZeroPtr());
+  }
+};
+
+} // namespace internal
+
+
+/** 
+  * \ingroup SparseCore_Module
+  *
+  * \brief A sparse matrix expression referencing an existing sparse expression
+  *
+  * \tparam SparseMatrixType the equivalent sparse matrix type of the referenced data, it must be a template instance of class SparseMatrix.
+  * \tparam Options specifies whether the a standard compressed format is required \c Options is  \c #StandardCompressedFormat, or \c 0.
+  *                The default is \c 0.
+  *
+  * \sa class Ref
+  */
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType >
+  : public internal::SparseRefBase<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType > >
+#else
+template<typename SparseMatrixType, int Options>
+class Ref<SparseMatrixType, Options>
+  : public SparseMapBase<Derived,WriteAccessors> // yes, that's weird to use Derived here, but that works!
+#endif
+{
+    typedef SparseMatrix<MatScalar,MatOptions,MatIndex> PlainObjectType;
+    typedef internal::traits<Ref> Traits;
+    template<int OtherOptions>
+    inline Ref(const SparseMatrix<MatScalar,OtherOptions,MatIndex>& expr);
+    template<int OtherOptions>
+    inline Ref(const MappedSparseMatrix<MatScalar,OtherOptions,MatIndex>& expr);
+  public:
+
+    typedef internal::SparseRefBase<Ref> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
+
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<int OtherOptions>
+    inline Ref(SparseMatrix<MatScalar,OtherOptions,MatIndex>& expr)
+    {
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<SparseMatrix<MatScalar,OtherOptions,MatIndex> >::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      eigen_assert( ((Options & int(StandardCompressedFormat))==0) || (expr.isCompressed()) );
+      Base::construct(expr.derived());
+    }
+    
+    template<int OtherOptions>
+    inline Ref(MappedSparseMatrix<MatScalar,OtherOptions,MatIndex>& expr)
+    {
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<SparseMatrix<MatScalar,OtherOptions,MatIndex> >::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      eigen_assert( ((Options & int(StandardCompressedFormat))==0) || (expr.isCompressed()) );
+      Base::construct(expr.derived());
+    }
+    
+    template<typename Derived>
+    inline Ref(const SparseCompressedBase<Derived>& expr)
+    #else
+    /** Implicit constructor from any sparse expression (2D matrix or 1D vector) */
+    template<typename Derived>
+    inline Ref(SparseCompressedBase<Derived>& expr)
+    #endif
+    {
+      EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      eigen_assert( ((Options & int(StandardCompressedFormat))==0) || (expr.isCompressed()) );
+      Base::construct(expr.const_cast_derived());
+    }
+};
+
+// this is the const ref version
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType>
+  : public internal::SparseRefBase<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+{
+    typedef SparseMatrix<MatScalar,MatOptions,MatIndex> TPlainObjectType;
+    typedef internal::traits<Ref> Traits;
+  public:
+
+    typedef internal::SparseRefBase<Ref> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
+
+    template<typename Derived>
+    inline Ref(const SparseMatrixBase<Derived>& expr) : m_hasCopy(false)
+    {
+      construct(expr.derived(), typename Traits::template match<Derived>::type());
+    }
+
+    inline Ref(const Ref& other) : Base(other), m_hasCopy(false) {
+      // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
+    }
+
+    template<typename OtherRef>
+    inline Ref(const RefBase<OtherRef>& other) : m_hasCopy(false) {
+      construct(other.derived(), typename Traits::template match<OtherRef>::type());
+    }
+
+    ~Ref() {
+      if(m_hasCopy) {
+        TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+        obj->~TPlainObjectType();
+      }
+    }
+
+  protected:
+
+    template<typename Expression>
+    void construct(const Expression& expr,internal::true_type)
+    {
+      if((Options & int(StandardCompressedFormat)) && (!expr.isCompressed()))
+      {
+        TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+        ::new (obj) TPlainObjectType(expr);
+        m_hasCopy = true;
+        Base::construct(*obj);
+      }
+      else
+      {
+        Base::construct(expr);
+      }
+    }
+
+    template<typename Expression>
+    void construct(const Expression& expr, internal::false_type)
+    {
+      TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+      ::new (obj) TPlainObjectType(expr);
+      m_hasCopy = true;
+      Base::construct(*obj);
+    }
+
+  protected:
+    char m_object_bytes[sizeof(TPlainObjectType)];
+    bool m_hasCopy;
+};
+
+
+
+/**
+  * \ingroup SparseCore_Module
+  *
+  * \brief A sparse vector expression referencing an existing sparse vector expression
+  *
+  * \tparam SparseVectorType the equivalent sparse vector type of the referenced data, it must be a template instance of class SparseVector.
+  *
+  * \sa class Ref
+  */
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType >
+  : public internal::SparseRefBase<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType > >
+#else
+template<typename SparseVectorType>
+class Ref<SparseVectorType>
+  : public SparseMapBase<Derived,WriteAccessors>
+#endif
+{
+    typedef SparseVector<MatScalar,MatOptions,MatIndex> PlainObjectType;
+    typedef internal::traits<Ref> Traits;
+    template<int OtherOptions>
+    inline Ref(const SparseVector<MatScalar,OtherOptions,MatIndex>& expr);
+  public:
+
+    typedef internal::SparseRefBase<Ref> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<int OtherOptions>
+    inline Ref(SparseVector<MatScalar,OtherOptions,MatIndex>& expr)
+    {
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<SparseVector<MatScalar,OtherOptions,MatIndex> >::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      Base::construct(expr.derived());
+    }
+
+    template<typename Derived>
+    inline Ref(const SparseCompressedBase<Derived>& expr)
+    #else
+    /** Implicit constructor from any 1D sparse vector expression */
+    template<typename Derived>
+    inline Ref(SparseCompressedBase<Derived>& expr)
+    #endif
+    {
+      EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      Base::construct(expr.const_cast_derived());
+    }
+};
+
+// this is the const ref version
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType>
+  : public internal::SparseRefBase<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+{
+    typedef SparseVector<MatScalar,MatOptions,MatIndex> TPlainObjectType;
+    typedef internal::traits<Ref> Traits;
+  public:
+
+    typedef internal::SparseRefBase<Ref> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
+
+    template<typename Derived>
+    inline Ref(const SparseMatrixBase<Derived>& expr) : m_hasCopy(false)
+    {
+      construct(expr.derived(), typename Traits::template match<Derived>::type());
+    }
+
+    inline Ref(const Ref& other) : Base(other), m_hasCopy(false) {
+      // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
+    }
+
+    template<typename OtherRef>
+    inline Ref(const RefBase<OtherRef>& other) : m_hasCopy(false) {
+      construct(other.derived(), typename Traits::template match<OtherRef>::type());
+    }
+
+    ~Ref() {
+      if(m_hasCopy) {
+        TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+        obj->~TPlainObjectType();
+      }
+    }
+
+  protected:
+
+    template<typename Expression>
+    void construct(const Expression& expr,internal::true_type)
+    {
+      Base::construct(expr);
+    }
+
+    template<typename Expression>
+    void construct(const Expression& expr, internal::false_type)
+    {
+      TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+      ::new (obj) TPlainObjectType(expr);
+      m_hasCopy = true;
+      Base::construct(*obj);
+    }
+
+  protected:
+    char m_object_bytes[sizeof(TPlainObjectType)];
+    bool m_hasCopy;
+};
+
+namespace internal {
+
+// FIXME shall we introduce a general evaluatior_ref that we can specialize for any sparse object once, and thus remove this copy-pasta thing...
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSE_REF_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseSelfAdjointView.h b/extern/Eigen3/Eigen/src/SparseCore/SparseSelfAdjointView.h
index 0eda96bc471..65611b3d4ca 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseSelfAdjointView.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseSelfAdjointView.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -11,14 +11,14 @@
 #define EIGEN_SPARSE_SELFADJOINTVIEW_H
 
 namespace Eigen { 
-
+  
 /** \ingroup SparseCore_Module
   * \class SparseSelfAdjointView
   *
   * \brief Pseudo expression to manipulate a triangular sparse matrix as a selfadjoint matrix.
   *
   * \param MatrixType the type of the dense matrix storing the coefficients
-  * \param UpLo can be either \c #Lower or \c #Upper
+  * \param Mode can be either \c #Lower or \c #Upper
   *
   * This class is an expression of a sefladjoint matrix from a triangular part of a matrix
   * with given dense storage of the coefficients. It is the return type of MatrixBase::selfadjointView()
@@ -26,38 +26,40 @@ namespace Eigen {
   *
   * \sa SparseMatrixBase::selfadjointView()
   */
-template<typename Lhs, typename Rhs, int UpLo>
-class SparseSelfAdjointTimeDenseProduct;
-
-template<typename Lhs, typename Rhs, int UpLo>
-class DenseTimeSparseSelfAdjointProduct;
-
 namespace internal {
   
-template<typename MatrixType, unsigned int UpLo>
-struct traits<SparseSelfAdjointView<MatrixType,UpLo> > : traits<MatrixType> {
+template<typename MatrixType, unsigned int Mode>
+struct traits<SparseSelfAdjointView<MatrixType,Mode> > : traits<MatrixType> {
 };
 
-template<int SrcUpLo,int DstUpLo,typename MatrixType,int DestOrder>
-void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::Index>& _dest, const typename MatrixType::Index* perm = 0);
+template<int SrcMode,int DstMode,typename MatrixType,int DestOrder>
+void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm = 0);
 
-template<int UpLo,typename MatrixType,int DestOrder>
-void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::Index>& _dest, const typename MatrixType::Index* perm = 0);
+template<int Mode,typename MatrixType,int DestOrder>
+void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm = 0);
 
 }
 
-template<typename MatrixType, unsigned int UpLo> class SparseSelfAdjointView
-  : public EigenBase<SparseSelfAdjointView<MatrixType,UpLo> >
+template<typename MatrixType, unsigned int _Mode> class SparseSelfAdjointView
+  : public EigenBase<SparseSelfAdjointView<MatrixType,_Mode> >
 {
   public:
-
+    
+    enum {
+      Mode = _Mode,
+      TransposeMode = ((Mode & Upper) ? Lower : 0) | ((Mode & Lower) ? Upper : 0),
+      RowsAtCompileTime = internal::traits<SparseSelfAdjointView>::RowsAtCompileTime,
+      ColsAtCompileTime = internal::traits<SparseSelfAdjointView>::ColsAtCompileTime
+    };
+
+    typedef EigenBase<SparseSelfAdjointView> Base;
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
-    typedef Matrix<Index,Dynamic,1> VectorI;
-    typedef typename MatrixType::Nested MatrixTypeNested;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+    typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
     typedef typename internal::remove_all<MatrixTypeNested>::type _MatrixTypeNested;
-
-    inline SparseSelfAdjointView(const MatrixType& matrix) : m_matrix(matrix)
+    
+    explicit inline SparseSelfAdjointView(MatrixType& matrix) : m_matrix(matrix)
     {
       eigen_assert(rows()==cols() && "SelfAdjointView is only for squared matrices");
     }
@@ -67,7 +69,7 @@ template<typename MatrixType, unsigned int UpLo> class SparseSelfAdjointView
 
     /** \internal \returns a reference to the nested matrix */
     const _MatrixTypeNested& matrix() const { return m_matrix; }
-    _MatrixTypeNested& matrix() { return m_matrix.const_cast_derived(); }
+    typename internal::remove_reference<MatrixTypeNested>::type& matrix() { return m_matrix; }
 
     /** \returns an expression of the matrix product between a sparse self-adjoint matrix \c *this and a sparse matrix \a rhs.
       *
@@ -75,10 +77,10 @@ template<typename MatrixType, unsigned int UpLo> class SparseSelfAdjointView
       * Indeed, the SparseSelfadjointView operand is first copied into a temporary SparseMatrix before computing the product.
       */
     template<typename OtherDerived>
-    SparseSparseProduct<typename OtherDerived::PlainObject, OtherDerived>
+    Product<SparseSelfAdjointView, OtherDerived>
     operator*(const SparseMatrixBase<OtherDerived>& rhs) const
     {
-      return SparseSparseProduct<typename OtherDerived::PlainObject, OtherDerived>(*this, rhs.derived());
+      return Product<SparseSelfAdjointView, OtherDerived>(*this, rhs.derived());
     }
 
     /** \returns an expression of the matrix product between a sparse matrix \a lhs and a sparse self-adjoint matrix \a rhs.
@@ -87,26 +89,26 @@ template<typename MatrixType, unsigned int UpLo> class SparseSelfAdjointView
       * Indeed, the SparseSelfadjointView operand is first copied into a temporary SparseMatrix before computing the product.
       */
     template<typename OtherDerived> friend
-    SparseSparseProduct<OtherDerived, typename OtherDerived::PlainObject >
+    Product<OtherDerived, SparseSelfAdjointView>
     operator*(const SparseMatrixBase<OtherDerived>& lhs, const SparseSelfAdjointView& rhs)
     {
-      return SparseSparseProduct<OtherDerived, typename OtherDerived::PlainObject>(lhs.derived(), rhs);
+      return Product<OtherDerived, SparseSelfAdjointView>(lhs.derived(), rhs);
     }
     
     /** Efficient sparse self-adjoint matrix times dense vector/matrix product */
     template<typename OtherDerived>
-    SparseSelfAdjointTimeDenseProduct<MatrixType,OtherDerived,UpLo>
+    Product<SparseSelfAdjointView,OtherDerived>
     operator*(const MatrixBase<OtherDerived>& rhs) const
     {
-      return SparseSelfAdjointTimeDenseProduct<MatrixType,OtherDerived,UpLo>(m_matrix, rhs.derived());
+      return Product<SparseSelfAdjointView,OtherDerived>(*this, rhs.derived());
     }
 
     /** Efficient dense vector/matrix times sparse self-adjoint matrix product */
     template<typename OtherDerived> friend
-    DenseTimeSparseSelfAdjointProduct<OtherDerived,MatrixType,UpLo>
+    Product<OtherDerived,SparseSelfAdjointView>
     operator*(const MatrixBase<OtherDerived>& lhs, const SparseSelfAdjointView& rhs)
     {
-      return DenseTimeSparseSelfAdjointProduct<OtherDerived,_MatrixTypeNested,UpLo>(lhs.derived(), rhs.m_matrix);
+      return Product<OtherDerived,SparseSelfAdjointView>(lhs.derived(), rhs);
     }
 
     /** Perform a symmetric rank K update of the selfadjoint matrix \c *this:
@@ -120,56 +122,48 @@ template<typename MatrixType, unsigned int UpLo> class SparseSelfAdjointView
     template<typename DerivedU>
     SparseSelfAdjointView& rankUpdate(const SparseMatrixBase<DerivedU>& u, const Scalar& alpha = Scalar(1));
     
-    /** \internal triggered by sparse_matrix = SparseSelfadjointView; */
-    template<typename DestScalar,int StorageOrder> void evalTo(SparseMatrix<DestScalar,StorageOrder,Index>& _dest) const
-    {
-      internal::permute_symm_to_fullsymm<UpLo>(m_matrix, _dest);
-    }
-    
-    template<typename DestScalar> void evalTo(DynamicSparseMatrix<DestScalar,ColMajor,Index>& _dest) const
-    {
-      // TODO directly evaluate into _dest;
-      SparseMatrix<DestScalar,ColMajor,Index> tmp(_dest.rows(),_dest.cols());
-      internal::permute_symm_to_fullsymm<UpLo>(m_matrix, tmp);
-      _dest = tmp;
-    }
-    
     /** \returns an expression of P H P^-1 */
-    SparseSymmetricPermutationProduct<_MatrixTypeNested,UpLo> twistedBy(const PermutationMatrix<Dynamic,Dynamic,Index>& perm) const
+    // TODO implement twists in a more evaluator friendly fashion
+    SparseSymmetricPermutationProduct<_MatrixTypeNested,Mode> twistedBy(const PermutationMatrix<Dynamic,Dynamic,StorageIndex>& perm) const
     {
-      return SparseSymmetricPermutationProduct<_MatrixTypeNested,UpLo>(m_matrix, perm);
+      return SparseSymmetricPermutationProduct<_MatrixTypeNested,Mode>(m_matrix, perm);
     }
-    
-    template<typename SrcMatrixType,int SrcUpLo>
-    SparseSelfAdjointView& operator=(const SparseSymmetricPermutationProduct<SrcMatrixType,SrcUpLo>& permutedMatrix)
+
+    template<typename SrcMatrixType,int SrcMode>
+    SparseSelfAdjointView& operator=(const SparseSymmetricPermutationProduct<SrcMatrixType,SrcMode>& permutedMatrix)
     {
-      permutedMatrix.evalTo(*this);
+      internal::call_assignment_no_alias_no_transpose(*this, permutedMatrix);
       return *this;
     }
 
-
     SparseSelfAdjointView& operator=(const SparseSelfAdjointView& src)
     {
-      PermutationMatrix<Dynamic> pnull;
+      PermutationMatrix<Dynamic,Dynamic,StorageIndex> pnull;
       return *this = src.twistedBy(pnull);
     }
 
-    template<typename SrcMatrixType,unsigned int SrcUpLo>
-    SparseSelfAdjointView& operator=(const SparseSelfAdjointView<SrcMatrixType,SrcUpLo>& src)
+    template<typename SrcMatrixType,unsigned int SrcMode>
+    SparseSelfAdjointView& operator=(const SparseSelfAdjointView<SrcMatrixType,SrcMode>& src)
     {
-      PermutationMatrix<Dynamic> pnull;
+      PermutationMatrix<Dynamic,Dynamic,StorageIndex> pnull;
       return *this = src.twistedBy(pnull);
     }
     
-
-    // const SparseLLT<PlainObject, UpLo> llt() const;
-    // const SparseLDLT<PlainObject, UpLo> ldlt() const;
-
+    void resize(Index rows, Index cols)
+    {
+      EIGEN_ONLY_USED_FOR_DEBUG(rows);
+      EIGEN_ONLY_USED_FOR_DEBUG(cols);
+      eigen_assert(rows == this->rows() && cols == this->cols()
+                && "SparseSelfadjointView::resize() does not actually allow to resize.");
+    }
+    
   protected:
 
-    typename MatrixType::Nested m_matrix;
-    mutable VectorI m_countPerRow;
-    mutable VectorI m_countPerCol;
+    MatrixTypeNested m_matrix;
+    //mutable VectorI m_countPerRow;
+    //mutable VectorI m_countPerCol;
+  private:
+    template<typename Dest> void evalTo(Dest &) const;
 };
 
 /***************************************************************************
@@ -178,146 +172,268 @@ template<typename MatrixType, unsigned int UpLo> class SparseSelfAdjointView
 
 template<typename Derived>
 template<unsigned int UpLo>
-const SparseSelfAdjointView<Derived, UpLo> SparseMatrixBase<Derived>::selfadjointView() const
+typename SparseMatrixBase<Derived>::template ConstSelfAdjointViewReturnType<UpLo>::Type SparseMatrixBase<Derived>::selfadjointView() const
 {
-  return derived();
+  return SparseSelfAdjointView<const Derived, UpLo>(derived());
 }
 
 template<typename Derived>
 template<unsigned int UpLo>
-SparseSelfAdjointView<Derived, UpLo> SparseMatrixBase<Derived>::selfadjointView()
+typename SparseMatrixBase<Derived>::template SelfAdjointViewReturnType<UpLo>::Type SparseMatrixBase<Derived>::selfadjointView()
 {
-  return derived();
+  return SparseSelfAdjointView<Derived, UpLo>(derived());
 }
 
 /***************************************************************************
 * Implementation of SparseSelfAdjointView methods
 ***************************************************************************/
 
-template<typename MatrixType, unsigned int UpLo>
+template<typename MatrixType, unsigned int Mode>
 template<typename DerivedU>
-SparseSelfAdjointView<MatrixType,UpLo>&
-SparseSelfAdjointView<MatrixType,UpLo>::rankUpdate(const SparseMatrixBase<DerivedU>& u, const Scalar& alpha)
+SparseSelfAdjointView<MatrixType,Mode>&
+SparseSelfAdjointView<MatrixType,Mode>::rankUpdate(const SparseMatrixBase<DerivedU>& u, const Scalar& alpha)
 {
-  SparseMatrix<Scalar,MatrixType::Flags&RowMajorBit?RowMajor:ColMajor> tmp = u * u.adjoint();
+  SparseMatrix<Scalar,(MatrixType::Flags&RowMajorBit)?RowMajor:ColMajor> tmp = u * u.adjoint();
   if(alpha==Scalar(0))
-    m_matrix.const_cast_derived() = tmp.template triangularView<UpLo>();
+    m_matrix = tmp.template triangularView<Mode>();
   else
-    m_matrix.const_cast_derived() += alpha * tmp.template triangularView<UpLo>();
+    m_matrix += alpha * tmp.template triangularView<Mode>();
 
   return *this;
 }
 
+namespace internal {
+  
+// TODO currently a selfadjoint expression has the form SelfAdjointView<.,.>
+//      in the future selfadjoint-ness should be defined by the expression traits
+//      such that Transpose<SelfAdjointView<.,.> > is valid. (currently TriangularBase::transpose() is overloaded to make it work)
+template<typename MatrixType, unsigned int Mode>
+struct evaluator_traits<SparseSelfAdjointView<MatrixType,Mode> >
+{
+  typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
+  typedef SparseSelfAdjointShape Shape;
+};
+
+struct SparseSelfAdjoint2Sparse {};
+
+template<> struct AssignmentKind<SparseShape,SparseSelfAdjointShape> { typedef SparseSelfAdjoint2Sparse Kind; };
+template<> struct AssignmentKind<SparseSelfAdjointShape,SparseShape> { typedef Sparse2Sparse Kind; };
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, SparseSelfAdjoint2Sparse>
+{
+  typedef typename DstXprType::StorageIndex StorageIndex;
+  typedef internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> AssignOpType;
+
+  template<typename DestScalar,int StorageOrder>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src, const AssignOpType&/*func*/)
+  {
+    internal::permute_symm_to_fullsymm<SrcXprType::Mode>(src.matrix(), dst);
+  }
+
+  // FIXME: the handling of += and -= in sparse matrices should be cleanup so that next two overloads could be reduced to:
+  template<typename DestScalar,int StorageOrder,typename AssignFunc>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src, const AssignFunc& func)
+  {
+    SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
+    run(tmp, src, AssignOpType());
+    call_assignment_no_alias_no_transpose(dst, tmp, func);
+  }
+
+  template<typename DestScalar,int StorageOrder>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src,
+                  const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>& /* func */)
+  {
+    SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
+    run(tmp, src, AssignOpType());
+    dst += tmp;
+  }
+
+  template<typename DestScalar,int StorageOrder>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src,
+                  const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>& /* func */)
+  {
+    SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
+    run(tmp, src, AssignOpType());
+    dst -= tmp;
+  }
+  
+  template<typename DestScalar>
+  static void run(DynamicSparseMatrix<DestScalar,ColMajor,StorageIndex>& dst, const SrcXprType &src, const AssignOpType&/*func*/)
+  {
+    // TODO directly evaluate into dst;
+    SparseMatrix<DestScalar,ColMajor,StorageIndex> tmp(dst.rows(),dst.cols());
+    internal::permute_symm_to_fullsymm<SrcXprType::Mode>(src.matrix(), tmp);
+    dst = tmp;
+  }
+};
+
+} // end namespace internal
+
 /***************************************************************************
 * Implementation of sparse self-adjoint time dense matrix
 ***************************************************************************/
 
 namespace internal {
-template<typename Lhs, typename Rhs, int UpLo>
-struct traits<SparseSelfAdjointTimeDenseProduct<Lhs,Rhs,UpLo> >
- : traits<ProductBase<SparseSelfAdjointTimeDenseProduct<Lhs,Rhs,UpLo>, Lhs, Rhs> >
-{
-  typedef Dense StorageKind;
-};
-}
 
-template<typename Lhs, typename Rhs, int UpLo>
-class SparseSelfAdjointTimeDenseProduct
-  : public ProductBase<SparseSelfAdjointTimeDenseProduct<Lhs,Rhs,UpLo>, Lhs, Rhs>
+template<int Mode, typename SparseLhsType, typename DenseRhsType, typename DenseResType, typename AlphaType>
+inline void sparse_selfadjoint_time_dense_product(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const AlphaType& alpha)
 {
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(SparseSelfAdjointTimeDenseProduct)
-
-    SparseSelfAdjointTimeDenseProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {}
+  EIGEN_ONLY_USED_FOR_DEBUG(alpha);
+  
+  typedef typename internal::nested_eval<SparseLhsType,DenseRhsType::MaxColsAtCompileTime>::type SparseLhsTypeNested;
+  typedef typename internal::remove_all<SparseLhsTypeNested>::type SparseLhsTypeNestedCleaned;
+  typedef evaluator<SparseLhsTypeNestedCleaned> LhsEval;
+  typedef typename LhsEval::InnerIterator LhsIterator;
+  typedef typename SparseLhsType::Scalar LhsScalar;
+  
+  enum {
+    LhsIsRowMajor = (LhsEval::Flags&RowMajorBit)==RowMajorBit,
+    ProcessFirstHalf =
+              ((Mode&(Upper|Lower))==(Upper|Lower))
+          || ( (Mode&Upper) && !LhsIsRowMajor)
+          || ( (Mode&Lower) && LhsIsRowMajor),
+    ProcessSecondHalf = !ProcessFirstHalf
+  };
+  
+  SparseLhsTypeNested lhs_nested(lhs);
+  LhsEval lhsEval(lhs_nested);
 
-    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+  // work on one column at once
+  for (Index k=0; k<rhs.cols(); ++k)
+  {
+    for (Index j=0; j<lhs.outerSize(); ++j)
     {
-      EIGEN_ONLY_USED_FOR_DEBUG(alpha);
-      // TODO use alpha
-      eigen_assert(alpha==Scalar(1) && "alpha != 1 is not implemented yet, sorry");
-      typedef typename internal::remove_all<Lhs>::type _Lhs;
-      typedef typename _Lhs::InnerIterator LhsInnerIterator;
-      enum {
-        LhsIsRowMajor = (_Lhs::Flags&RowMajorBit)==RowMajorBit,
-        ProcessFirstHalf =
-                 ((UpLo&(Upper|Lower))==(Upper|Lower))
-              || ( (UpLo&Upper) && !LhsIsRowMajor)
-              || ( (UpLo&Lower) && LhsIsRowMajor),
-        ProcessSecondHalf = !ProcessFirstHalf
-      };
-      for (Index j=0; j<m_lhs.outerSize(); ++j)
+      LhsIterator i(lhsEval,j);
+      // handle diagonal coeff
+      if (ProcessSecondHalf)
       {
-        LhsInnerIterator i(m_lhs,j);
-        if (ProcessSecondHalf)
-        {
-          while (i && i.index()<j) ++i;
-          if(i && i.index()==j)
-          {
-            dest.row(j) += i.value() * m_rhs.row(j);
-            ++i;
-          }
-        }
-        for(; (ProcessFirstHalf ? i && i.index() < j : i) ; ++i)
+        while (i && i.index()<j) ++i;
+        if(i && i.index()==j)
         {
-          Index a = LhsIsRowMajor ? j : i.index();
-          Index b = LhsIsRowMajor ? i.index() : j;
-          typename Lhs::Scalar v = i.value();
-          dest.row(a) += (v) * m_rhs.row(b);
-          dest.row(b) += numext::conj(v) * m_rhs.row(a);
+          res.coeffRef(j,k) += alpha * i.value() * rhs.coeff(j,k);
+          ++i;
         }
-        if (ProcessFirstHalf && i && (i.index()==j))
-          dest.row(j) += i.value() * m_rhs.row(j);
       }
+
+      // premultiplied rhs for scatters
+      typename ScalarBinaryOpTraits<AlphaType, typename DenseRhsType::Scalar>::ReturnType rhs_j(alpha*rhs(j,k));
+      // accumulator for partial scalar product
+      typename DenseResType::Scalar res_j(0);
+      for(; (ProcessFirstHalf ? i && i.index() < j : i) ; ++i)
+      {
+        LhsScalar lhs_ij = i.value();
+        if(!LhsIsRowMajor) lhs_ij = numext::conj(lhs_ij);
+        res_j += lhs_ij * rhs.coeff(i.index(),k);
+        res(i.index(),k) += numext::conj(lhs_ij) * rhs_j;
+      }
+      res.coeffRef(j,k) += alpha * res_j;
+
+      // handle diagonal coeff
+      if (ProcessFirstHalf && i && (i.index()==j))
+        res.coeffRef(j,k) += alpha * i.value() * rhs.coeff(j,k);
     }
+  }
+}
 
-  private:
-    SparseSelfAdjointTimeDenseProduct& operator=(const SparseSelfAdjointTimeDenseProduct&);
+
+template<typename LhsView, typename Rhs, int ProductType>
+struct generic_product_impl<LhsView, Rhs, SparseSelfAdjointShape, DenseShape, ProductType>
+: generic_product_impl_base<LhsView, Rhs, generic_product_impl<LhsView, Rhs, SparseSelfAdjointShape, DenseShape, ProductType> >
+{
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const LhsView& lhsView, const Rhs& rhs, const typename Dest::Scalar& alpha)
+  {
+    typedef typename LhsView::_MatrixTypeNested Lhs;
+    typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhsView.matrix());
+    RhsNested rhsNested(rhs);
+    
+    internal::sparse_selfadjoint_time_dense_product<LhsView::Mode>(lhsNested, rhsNested, dst, alpha);
+  }
 };
 
-namespace internal {
-template<typename Lhs, typename Rhs, int UpLo>
-struct traits<DenseTimeSparseSelfAdjointProduct<Lhs,Rhs,UpLo> >
- : traits<ProductBase<DenseTimeSparseSelfAdjointProduct<Lhs,Rhs,UpLo>, Lhs, Rhs> >
-{};
-}
+template<typename Lhs, typename RhsView, int ProductType>
+struct generic_product_impl<Lhs, RhsView, DenseShape, SparseSelfAdjointShape, ProductType>
+: generic_product_impl_base<Lhs, RhsView, generic_product_impl<Lhs, RhsView, DenseShape, SparseSelfAdjointShape, ProductType> >
+{
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const RhsView& rhsView, const typename Dest::Scalar& alpha)
+  {
+    typedef typename RhsView::_MatrixTypeNested Rhs;
+    typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhsView.matrix());
+    
+    // transpose everything
+    Transpose<Dest> dstT(dst);
+    internal::sparse_selfadjoint_time_dense_product<RhsView::TransposeMode>(rhsNested.transpose(), lhsNested.transpose(), dstT, alpha);
+  }
+};
 
-template<typename Lhs, typename Rhs, int UpLo>
-class DenseTimeSparseSelfAdjointProduct
-  : public ProductBase<DenseTimeSparseSelfAdjointProduct<Lhs,Rhs,UpLo>, Lhs, Rhs>
+// NOTE: these two overloads are needed to evaluate the sparse selfadjoint view into a full sparse matrix
+// TODO: maybe the copy could be handled by generic_product_impl so that these overloads would not be needed anymore
+
+template<typename LhsView, typename Rhs, int ProductTag>
+struct product_evaluator<Product<LhsView, Rhs, DefaultProduct>, ProductTag, SparseSelfAdjointShape, SparseShape>
+  : public evaluator<typename Product<typename Rhs::PlainObject, Rhs, DefaultProduct>::PlainObject>
 {
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(DenseTimeSparseSelfAdjointProduct)
+  typedef Product<LhsView, Rhs, DefaultProduct> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
 
-    DenseTimeSparseSelfAdjointProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {}
+  product_evaluator(const XprType& xpr)
+    : m_lhs(xpr.lhs()), m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    generic_product_impl<typename Rhs::PlainObject, Rhs, SparseShape, SparseShape, ProductTag>::evalTo(m_result, m_lhs, xpr.rhs());
+  }
+  
+protected:
+  typename Rhs::PlainObject m_lhs;
+  PlainObject m_result;
+};
 
-    template<typename Dest> void scaleAndAddTo(Dest& /*dest*/, const Scalar& /*alpha*/) const
-    {
-      // TODO
-    }
+template<typename Lhs, typename RhsView, int ProductTag>
+struct product_evaluator<Product<Lhs, RhsView, DefaultProduct>, ProductTag, SparseShape, SparseSelfAdjointShape>
+  : public evaluator<typename Product<Lhs, typename Lhs::PlainObject, DefaultProduct>::PlainObject>
+{
+  typedef Product<Lhs, RhsView, DefaultProduct> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
 
-  private:
-    DenseTimeSparseSelfAdjointProduct& operator=(const DenseTimeSparseSelfAdjointProduct&);
+  product_evaluator(const XprType& xpr)
+    : m_rhs(xpr.rhs()), m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    generic_product_impl<Lhs, typename Lhs::PlainObject, SparseShape, SparseShape, ProductTag>::evalTo(m_result, xpr.lhs(), m_rhs);
+  }
+  
+protected:
+  typename Lhs::PlainObject m_rhs;
+  PlainObject m_result;
 };
 
+} // namespace internal
+
 /***************************************************************************
 * Implementation of symmetric copies and permutations
 ***************************************************************************/
 namespace internal {
-  
-template<typename MatrixType, int UpLo>
-struct traits<SparseSymmetricPermutationProduct<MatrixType,UpLo> > : traits<MatrixType> {
-};
 
-template<int UpLo,typename MatrixType,int DestOrder>
-void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::Index>& _dest, const typename MatrixType::Index* perm)
+template<int Mode,typename MatrixType,int DestOrder>
+void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm)
 {
-  typedef typename MatrixType::Index Index;
+  typedef typename MatrixType::StorageIndex StorageIndex;
   typedef typename MatrixType::Scalar Scalar;
-  typedef SparseMatrix<Scalar,DestOrder,Index> Dest;
-  typedef Matrix<Index,Dynamic,1> VectorI;
+  typedef SparseMatrix<Scalar,DestOrder,StorageIndex> Dest;
+  typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+  typedef evaluator<MatrixType> MatEval;
+  typedef typename evaluator<MatrixType>::InnerIterator MatIterator;
   
+  MatEval matEval(mat);
   Dest& dest(_dest.derived());
   enum {
     StorageOrderMatch = int(Dest::IsRowMajor) == int(MatrixType::IsRowMajor)
@@ -331,17 +447,17 @@ void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename Matri
   for(Index j = 0; j<size; ++j)
   {
     Index jp = perm ? perm[j] : j;
-    for(typename MatrixType::InnerIterator it(mat,j); it; ++it)
+    for(MatIterator it(matEval,j); it; ++it)
     {
       Index i = it.index();
       Index r = it.row();
       Index c = it.col();
       Index ip = perm ? perm[i] : i;
-      if(UpLo==(Upper|Lower))
+      if(Mode==(Upper|Lower))
         count[StorageOrderMatch ? jp : ip]++;
       else if(r==c)
         count[ip]++;
-      else if(( UpLo==Lower && r>c) || ( UpLo==Upper && r<c))
+      else if(( Mode==Lower && r>c) || ( Mode==Upper && r<c))
       {
         count[ip]++;
         count[jp]++;
@@ -359,18 +475,18 @@ void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename Matri
     count[j] = dest.outerIndexPtr()[j];
   
   // copy data
-  for(Index j = 0; j<size; ++j)
+  for(StorageIndex j = 0; j<size; ++j)
   {
-    for(typename MatrixType::InnerIterator it(mat,j); it; ++it)
+    for(MatIterator it(matEval,j); it; ++it)
     {
-      Index i = it.index();
+      StorageIndex i = internal::convert_index<StorageIndex>(it.index());
       Index r = it.row();
       Index c = it.col();
       
-      Index jp = perm ? perm[j] : j;
-      Index ip = perm ? perm[i] : i;
+      StorageIndex jp = perm ? perm[j] : j;
+      StorageIndex ip = perm ? perm[i] : i;
       
-      if(UpLo==(Upper|Lower))
+      if(Mode==(Upper|Lower))
       {
         Index k = count[StorageOrderMatch ? jp : ip]++;
         dest.innerIndexPtr()[k] = StorageOrderMatch ? ip : jp;
@@ -382,7 +498,7 @@ void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename Matri
         dest.innerIndexPtr()[k] = ip;
         dest.valuePtr()[k] = it.value();
       }
-      else if(( (UpLo&Lower)==Lower && r>c) || ( (UpLo&Upper)==Upper && r<c))
+      else if(( (Mode&Lower)==Lower && r>c) || ( (Mode&Upper)==Upper && r<c))
       {
         if(!StorageOrderMatch)
           std::swap(ip,jp);
@@ -397,35 +513,40 @@ void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename Matri
   }
 }
 
-template<int _SrcUpLo,int _DstUpLo,typename MatrixType,int DstOrder>
-void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DstOrder,typename MatrixType::Index>& _dest, const typename MatrixType::Index* perm)
+template<int _SrcMode,int _DstMode,typename MatrixType,int DstOrder>
+void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DstOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm)
 {
-  typedef typename MatrixType::Index Index;
+  typedef typename MatrixType::StorageIndex StorageIndex;
   typedef typename MatrixType::Scalar Scalar;
-  SparseMatrix<Scalar,DstOrder,Index>& dest(_dest.derived());
-  typedef Matrix<Index,Dynamic,1> VectorI;
+  SparseMatrix<Scalar,DstOrder,StorageIndex>& dest(_dest.derived());
+  typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+  typedef evaluator<MatrixType> MatEval;
+  typedef typename evaluator<MatrixType>::InnerIterator MatIterator;
+
   enum {
     SrcOrder = MatrixType::IsRowMajor ? RowMajor : ColMajor,
     StorageOrderMatch = int(SrcOrder) == int(DstOrder),
-    DstUpLo = DstOrder==RowMajor ? (_DstUpLo==Upper ? Lower : Upper) : _DstUpLo,
-    SrcUpLo = SrcOrder==RowMajor ? (_SrcUpLo==Upper ? Lower : Upper) : _SrcUpLo
+    DstMode = DstOrder==RowMajor ? (_DstMode==Upper ? Lower : Upper) : _DstMode,
+    SrcMode = SrcOrder==RowMajor ? (_SrcMode==Upper ? Lower : Upper) : _SrcMode
   };
+
+  MatEval matEval(mat);
   
   Index size = mat.rows();
   VectorI count(size);
   count.setZero();
   dest.resize(size,size);
-  for(Index j = 0; j<size; ++j)
+  for(StorageIndex j = 0; j<size; ++j)
   {
-    Index jp = perm ? perm[j] : j;
-    for(typename MatrixType::InnerIterator it(mat,j); it; ++it)
+    StorageIndex jp = perm ? perm[j] : j;
+    for(MatIterator it(matEval,j); it; ++it)
     {
-      Index i = it.index();
-      if((int(SrcUpLo)==int(Lower) && i<j) || (int(SrcUpLo)==int(Upper) && i>j))
+      StorageIndex i = it.index();
+      if((int(SrcMode)==int(Lower) && i<j) || (int(SrcMode)==int(Upper) && i>j))
         continue;
                   
-      Index ip = perm ? perm[i] : i;
-      count[int(DstUpLo)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
+      StorageIndex ip = perm ? perm[i] : i;
+      count[int(DstMode)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
     }
   }
   dest.outerIndexPtr()[0] = 0;
@@ -435,23 +556,23 @@ void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixTyp
   for(Index j=0; j<size; ++j)
     count[j] = dest.outerIndexPtr()[j];
   
-  for(Index j = 0; j<size; ++j)
+  for(StorageIndex j = 0; j<size; ++j)
   {
     
-    for(typename MatrixType::InnerIterator it(mat,j); it; ++it)
+    for(MatIterator it(matEval,j); it; ++it)
     {
-      Index i = it.index();
-      if((int(SrcUpLo)==int(Lower) && i<j) || (int(SrcUpLo)==int(Upper) && i>j))
+      StorageIndex i = it.index();
+      if((int(SrcMode)==int(Lower) && i<j) || (int(SrcMode)==int(Upper) && i>j))
         continue;
                   
-      Index jp = perm ? perm[j] : j;
-      Index ip = perm? perm[i] : i;
+      StorageIndex jp = perm ? perm[j] : j;
+      StorageIndex ip = perm? perm[i] : i;
       
-      Index k = count[int(DstUpLo)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
-      dest.innerIndexPtr()[k] = int(DstUpLo)==int(Lower) ? (std::max)(ip,jp) : (std::min)(ip,jp);
+      Index k = count[int(DstMode)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
+      dest.innerIndexPtr()[k] = int(DstMode)==int(Lower) ? (std::max)(ip,jp) : (std::min)(ip,jp);
       
       if(!StorageOrderMatch) std::swap(ip,jp);
-      if( ((int(DstUpLo)==int(Lower) && ip<jp) || (int(DstUpLo)==int(Upper) && ip>jp)))
+      if( ((int(DstMode)==int(Lower) && ip<jp) || (int(DstMode)==int(Upper) && ip>jp)))
         dest.valuePtr()[k] = numext::conj(it.value());
       else
         dest.valuePtr()[k] = it.value();
@@ -461,19 +582,33 @@ void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixTyp
 
 }
 
-template<typename MatrixType,int UpLo>
+// TODO implement twists in a more evaluator friendly fashion
+
+namespace internal {
+
+template<typename MatrixType, int Mode>
+struct traits<SparseSymmetricPermutationProduct<MatrixType,Mode> > : traits<MatrixType> {
+};
+
+}
+
+template<typename MatrixType,int Mode>
 class SparseSymmetricPermutationProduct
-  : public EigenBase<SparseSymmetricPermutationProduct<MatrixType,UpLo> >
+  : public EigenBase<SparseSymmetricPermutationProduct<MatrixType,Mode> >
 {
   public:
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    enum {
+      RowsAtCompileTime = internal::traits<SparseSymmetricPermutationProduct>::RowsAtCompileTime,
+      ColsAtCompileTime = internal::traits<SparseSymmetricPermutationProduct>::ColsAtCompileTime
+    };
   protected:
-    typedef PermutationMatrix<Dynamic,Dynamic,Index> Perm;
+    typedef PermutationMatrix<Dynamic,Dynamic,StorageIndex> Perm;
   public:
-    typedef Matrix<Index,Dynamic,1> VectorI;
+    typedef Matrix<StorageIndex,Dynamic,1> VectorI;
     typedef typename MatrixType::Nested MatrixTypeNested;
-    typedef typename internal::remove_all<MatrixTypeNested>::type _MatrixTypeNested;
+    typedef typename internal::remove_all<MatrixTypeNested>::type NestedExpression;
     
     SparseSymmetricPermutationProduct(const MatrixType& mat, const Perm& perm)
       : m_matrix(mat), m_perm(perm)
@@ -481,20 +616,9 @@ class SparseSymmetricPermutationProduct
     
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
-    
-    template<typename DestScalar, int Options, typename DstIndex>
-    void evalTo(SparseMatrix<DestScalar,Options,DstIndex>& _dest) const
-    {
-//       internal::permute_symm_to_fullsymm<UpLo>(m_matrix,_dest,m_perm.indices().data());
-      SparseMatrix<DestScalar,(Options&RowMajor)==RowMajor ? ColMajor : RowMajor, DstIndex> tmp;
-      internal::permute_symm_to_fullsymm<UpLo>(m_matrix,tmp,m_perm.indices().data());
-      _dest = tmp;
-    }
-    
-    template<typename DestType,unsigned int DestUpLo> void evalTo(SparseSelfAdjointView<DestType,DestUpLo>& dest) const
-    {
-      internal::permute_symm_to_symm<UpLo,DestUpLo>(m_matrix,dest.matrix(),m_perm.indices().data());
-    }
+        
+    const NestedExpression& matrix() const { return m_matrix; }
+    const Perm& perm() const { return m_perm; }
     
   protected:
     MatrixTypeNested m_matrix;
@@ -502,6 +626,31 @@ class SparseSymmetricPermutationProduct
 
 };
 
+namespace internal {
+  
+template<typename DstXprType, typename MatrixType, int Mode, typename Scalar>
+struct Assignment<DstXprType, SparseSymmetricPermutationProduct<MatrixType,Mode>, internal::assign_op<Scalar,typename MatrixType::Scalar>, Sparse2Sparse>
+{
+  typedef SparseSymmetricPermutationProduct<MatrixType,Mode> SrcXprType;
+  typedef typename DstXprType::StorageIndex DstIndex;
+  template<int Options>
+  static void run(SparseMatrix<Scalar,Options,DstIndex> &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename MatrixType::Scalar> &)
+  {
+    // internal::permute_symm_to_fullsymm<Mode>(m_matrix,_dest,m_perm.indices().data());
+    SparseMatrix<Scalar,(Options&RowMajor)==RowMajor ? ColMajor : RowMajor, DstIndex> tmp;
+    internal::permute_symm_to_fullsymm<Mode>(src.matrix(),tmp,src.perm().indices().data());
+    dst = tmp;
+  }
+  
+  template<typename DestType,unsigned int DestMode>
+  static void run(SparseSelfAdjointView<DestType,DestMode>& dst, const SrcXprType &src, const internal::assign_op<Scalar,typename MatrixType::Scalar> &)
+  {
+    internal::permute_symm_to_symm<Mode,DestMode>(src.matrix(),dst.matrix(),src.perm().indices().data());
+  }
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSE_SELFADJOINTVIEW_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseSolverBase.h b/extern/Eigen3/Eigen/src/SparseCore/SparseSolverBase.h
new file mode 100644
index 00000000000..b4c9a422f05
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseSolverBase.h
@@ -0,0 +1,124 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSESOLVERBASE_H
+#define EIGEN_SPARSESOLVERBASE_H
+
+namespace Eigen { 
+
+namespace internal {
+
+  /** \internal
+  * Helper functions to solve with a sparse right-hand-side and result.
+  * The rhs is decomposed into small vertical panels which are solved through dense temporaries.
+  */
+template<typename Decomposition, typename Rhs, typename Dest>
+typename enable_if<Rhs::ColsAtCompileTime!=1 && Dest::ColsAtCompileTime!=1>::type
+solve_sparse_through_dense_panels(const Decomposition &dec, const Rhs& rhs, Dest &dest)
+{
+  EIGEN_STATIC_ASSERT((Dest::Flags&RowMajorBit)==0,THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
+  typedef typename Dest::Scalar DestScalar;
+  // we process the sparse rhs per block of NbColsAtOnce columns temporarily stored into a dense matrix.
+  static const Index NbColsAtOnce = 4;
+  Index rhsCols = rhs.cols();
+  Index size = rhs.rows();
+  // the temporary matrices do not need more columns than NbColsAtOnce:
+  Index tmpCols = (std::min)(rhsCols, NbColsAtOnce); 
+  Eigen::Matrix<DestScalar,Dynamic,Dynamic> tmp(size,tmpCols);
+  Eigen::Matrix<DestScalar,Dynamic,Dynamic> tmpX(size,tmpCols);
+  for(Index k=0; k<rhsCols; k+=NbColsAtOnce)
+  {
+    Index actualCols = std::min<Index>(rhsCols-k, NbColsAtOnce);
+    tmp.leftCols(actualCols) = rhs.middleCols(k,actualCols);
+    tmpX.leftCols(actualCols) = dec.solve(tmp.leftCols(actualCols));
+    dest.middleCols(k,actualCols) = tmpX.leftCols(actualCols).sparseView();
+  }
+}
+
+// Overload for vector as rhs
+template<typename Decomposition, typename Rhs, typename Dest>
+typename enable_if<Rhs::ColsAtCompileTime==1 || Dest::ColsAtCompileTime==1>::type
+solve_sparse_through_dense_panels(const Decomposition &dec, const Rhs& rhs, Dest &dest)
+{
+  typedef typename Dest::Scalar DestScalar;
+  Index size = rhs.rows();
+  Eigen::Matrix<DestScalar,Dynamic,1> rhs_dense(rhs);
+  Eigen::Matrix<DestScalar,Dynamic,1> dest_dense(size);
+  dest_dense = dec.solve(rhs_dense);
+  dest = dest_dense.sparseView();
+}
+
+} // end namespace internal
+
+/** \class SparseSolverBase
+  * \ingroup SparseCore_Module
+  * \brief A base class for sparse solvers
+  *
+  * \tparam Derived the actual type of the solver.
+  *
+  */
+template<typename Derived>
+class SparseSolverBase : internal::noncopyable
+{
+  public:
+
+    /** Default constructor */
+    SparseSolverBase()
+      : m_isInitialized(false)
+    {}
+
+    ~SparseSolverBase()
+    {}
+
+    Derived& derived() { return *static_cast<Derived*>(this); }
+    const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    
+    /** \returns an expression of the solution x of \f$ A x = b \f$ using the current decomposition of A.
+      *
+      * \sa compute()
+      */
+    template<typename Rhs>
+    inline const Solve<Derived, Rhs>
+    solve(const MatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "Solver is not initialized.");
+      eigen_assert(derived().rows()==b.rows() && "solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<Derived, Rhs>(derived(), b.derived());
+    }
+    
+    /** \returns an expression of the solution x of \f$ A x = b \f$ using the current decomposition of A.
+      *
+      * \sa compute()
+      */
+    template<typename Rhs>
+    inline const Solve<Derived, Rhs>
+    solve(const SparseMatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "Solver is not initialized.");
+      eigen_assert(derived().rows()==b.rows() && "solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<Derived, Rhs>(derived(), b.derived());
+    }
+    
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** \internal default implementation of solving with a sparse rhs */
+    template<typename Rhs,typename Dest>
+    void _solve_impl(const SparseMatrixBase<Rhs> &b, SparseMatrixBase<Dest> &dest) const
+    {
+      internal::solve_sparse_through_dense_panels(derived(), b.derived(), dest.derived());
+    }
+    #endif // EIGEN_PARSED_BY_DOXYGEN
+
+  protected:
+    
+    mutable bool m_isInitialized;
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSESOLVERBASE_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseSparseProductWithPruning.h b/extern/Eigen3/Eigen/src/SparseCore/SparseSparseProductWithPruning.h
index fcc18f5c9c8..88820a48f36 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseSparseProductWithPruning.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseSparseProductWithPruning.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -21,8 +21,9 @@ static void sparse_sparse_product_with_pruning_impl(const Lhs& lhs, const Rhs& r
 {
   // return sparse_sparse_product_with_pruning_impl2(lhs,rhs,res);
 
-  typedef typename remove_all<Lhs>::type::Scalar Scalar;
-  typedef typename remove_all<Lhs>::type::Index Index;
+  typedef typename remove_all<Rhs>::type::Scalar RhsScalar;
+  typedef typename remove_all<ResultType>::type::Scalar ResScalar;
+  typedef typename remove_all<Lhs>::type::StorageIndex StorageIndex;
 
   // make sure to call innerSize/outerSize since we fake the storage order.
   Index rows = lhs.innerSize();
@@ -31,24 +32,27 @@ static void sparse_sparse_product_with_pruning_impl(const Lhs& lhs, const Rhs& r
   eigen_assert(lhs.outerSize() == rhs.innerSize());
 
   // allocate a temporary buffer
-  AmbiVector<Scalar,Index> tempVector(rows);
+  AmbiVector<ResScalar,StorageIndex> tempVector(rows);
 
+  // mimics a resizeByInnerOuter:
+  if(ResultType::IsRowMajor)
+    res.resize(cols, rows);
+  else
+    res.resize(rows, cols);
+  
+  evaluator<Lhs> lhsEval(lhs);
+  evaluator<Rhs> rhsEval(rhs);
+  
   // estimate the number of non zero entries
   // given a rhs column containing Y non zeros, we assume that the respective Y columns
   // of the lhs differs in average of one non zeros, thus the number of non zeros for
   // the product of a rhs column with the lhs is X+Y where X is the average number of non zero
   // per column of the lhs.
   // Therefore, we have nnz(lhs*rhs) = nnz(lhs) + nnz(rhs)
-  Index estimated_nnz_prod = lhs.nonZeros() + rhs.nonZeros();
-
-  // mimics a resizeByInnerOuter:
-  if(ResultType::IsRowMajor)
-    res.resize(cols, rows);
-  else
-    res.resize(rows, cols);
+  Index estimated_nnz_prod = lhsEval.nonZerosEstimate() + rhsEval.nonZerosEstimate();
 
   res.reserve(estimated_nnz_prod);
-  double ratioColRes = double(estimated_nnz_prod)/double(lhs.rows()*rhs.cols());
+  double ratioColRes = double(estimated_nnz_prod)/(double(lhs.rows())*double(rhs.cols()));
   for (Index j=0; j<cols; ++j)
   {
     // FIXME:
@@ -56,18 +60,18 @@ static void sparse_sparse_product_with_pruning_impl(const Lhs& lhs, const Rhs& r
     // let's do a more accurate determination of the nnz ratio for the current column j of res
     tempVector.init(ratioColRes);
     tempVector.setZero();
-    for (typename Rhs::InnerIterator rhsIt(rhs, j); rhsIt; ++rhsIt)
+    for (typename evaluator<Rhs>::InnerIterator rhsIt(rhsEval, j); rhsIt; ++rhsIt)
     {
       // FIXME should be written like this: tmp += rhsIt.value() * lhs.col(rhsIt.index())
       tempVector.restart();
-      Scalar x = rhsIt.value();
-      for (typename Lhs::InnerIterator lhsIt(lhs, rhsIt.index()); lhsIt; ++lhsIt)
+      RhsScalar x = rhsIt.value();
+      for (typename evaluator<Lhs>::InnerIterator lhsIt(lhsEval, rhsIt.index()); lhsIt; ++lhsIt)
       {
         tempVector.coeffRef(lhsIt.index()) += lhsIt.value() * x;
       }
     }
     res.startVec(j);
-    for (typename AmbiVector<Scalar,Index>::Iterator it(tempVector,tolerance); it; ++it)
+    for (typename AmbiVector<ResScalar,StorageIndex>::Iterator it(tempVector,tolerance); it; ++it)
       res.insertBackByOuterInner(j,it.index()) = it.value();
   }
   res.finalize();
@@ -82,7 +86,6 @@ struct sparse_sparse_product_with_pruning_selector;
 template<typename Lhs, typename Rhs, typename ResultType>
 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor,ColMajor>
 {
-  typedef typename traits<typename remove_all<Lhs>::type>::Scalar Scalar;
   typedef typename ResultType::RealScalar RealScalar;
 
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
@@ -100,7 +103,7 @@ struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,C
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
   {
     // we need a col-major matrix to hold the result
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> SparseTemporaryType;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> SparseTemporaryType;
     SparseTemporaryType _res(res.rows(), res.cols());
     internal::sparse_sparse_product_with_pruning_impl<Lhs,Rhs,SparseTemporaryType>(lhs, rhs, _res, tolerance);
     res = _res;
@@ -126,8 +129,8 @@ struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,R
   typedef typename ResultType::RealScalar RealScalar;
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename Lhs::Index> ColMajorMatrixLhs;
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename Lhs::Index> ColMajorMatrixRhs;
+    typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixLhs;
+    typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixRhs;
     ColMajorMatrixLhs colLhs(lhs);
     ColMajorMatrixRhs colRhs(rhs);
     internal::sparse_sparse_product_with_pruning_impl<ColMajorMatrixLhs,ColMajorMatrixRhs,ResultType>(colLhs, colRhs, res, tolerance);
@@ -140,8 +143,53 @@ struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,R
   }
 };
 
-// NOTE the 2 others cases (col row *) must never occur since they are caught
-// by ProductReturnType which transforms it to (col col *) by evaluating rhs.
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor,RowMajor>
+{
+  typedef typename ResultType::RealScalar RealScalar;
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
+  {
+    typedef SparseMatrix<typename Lhs::Scalar,RowMajor,typename Lhs::StorageIndex> RowMajorMatrixLhs;
+    RowMajorMatrixLhs rowLhs(lhs);
+    sparse_sparse_product_with_pruning_selector<RowMajorMatrixLhs,Rhs,ResultType,RowMajor,RowMajor>(rowLhs,rhs,res,tolerance);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor,RowMajor>
+{
+  typedef typename ResultType::RealScalar RealScalar;
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
+  {
+    typedef SparseMatrix<typename Rhs::Scalar,RowMajor,typename Lhs::StorageIndex> RowMajorMatrixRhs;
+    RowMajorMatrixRhs rowRhs(rhs);
+    sparse_sparse_product_with_pruning_selector<Lhs,RowMajorMatrixRhs,ResultType,RowMajor,RowMajor,RowMajor>(lhs,rowRhs,res,tolerance);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor,ColMajor>
+{
+  typedef typename ResultType::RealScalar RealScalar;
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
+  {
+    typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixRhs;
+    ColMajorMatrixRhs colRhs(rhs);
+    internal::sparse_sparse_product_with_pruning_impl<Lhs,ColMajorMatrixRhs,ResultType>(lhs, colRhs, res, tolerance);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor,ColMajor>
+{
+  typedef typename ResultType::RealScalar RealScalar;
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
+  {
+    typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixLhs;
+    ColMajorMatrixLhs colLhs(lhs);
+    internal::sparse_sparse_product_with_pruning_impl<ColMajorMatrixLhs,Rhs,ResultType>(colLhs, rhs, res, tolerance);
+  }
+};
 
 } // end namespace internal
 
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseTranspose.h b/extern/Eigen3/Eigen/src/SparseCore/SparseTranspose.h
index 76d031d52c8..3757d4c6b07 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseTranspose.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseTranspose.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,52 +12,81 @@
 
 namespace Eigen { 
 
-template<typename MatrixType> class TransposeImpl<MatrixType,Sparse>
-  : public SparseMatrixBase<Transpose<MatrixType> >
-{
-    typedef typename internal::remove_all<typename MatrixType::Nested>::type _MatrixTypeNested;
+namespace internal {
+  template<typename MatrixType,int CompressedAccess=int(MatrixType::Flags&CompressedAccessBit)>
+  class SparseTransposeImpl
+    : public SparseMatrixBase<Transpose<MatrixType> >
+  {};
+  
+  template<typename MatrixType>
+  class SparseTransposeImpl<MatrixType,CompressedAccessBit>
+    : public SparseCompressedBase<Transpose<MatrixType> >
+  {
+    typedef SparseCompressedBase<Transpose<MatrixType> > Base;
   public:
-
-    EIGEN_SPARSE_PUBLIC_INTERFACE(Transpose<MatrixType> )
-
-    class InnerIterator;
-    class ReverseInnerIterator;
+    using Base::derived;
+    typedef typename Base::Scalar Scalar;
+    typedef typename Base::StorageIndex StorageIndex;
 
     inline Index nonZeros() const { return derived().nestedExpression().nonZeros(); }
-};
+    
+    inline const Scalar* valuePtr() const { return derived().nestedExpression().valuePtr(); }
+    inline const StorageIndex* innerIndexPtr() const { return derived().nestedExpression().innerIndexPtr(); }
+    inline const StorageIndex* outerIndexPtr() const { return derived().nestedExpression().outerIndexPtr(); }
+    inline const StorageIndex* innerNonZeroPtr() const { return derived().nestedExpression().innerNonZeroPtr(); }
 
-// NOTE: VC10 and VC11 trigger an ICE if don't put typename TransposeImpl<MatrixType,Sparse>:: in front of Index,
-// a typedef typename TransposeImpl<MatrixType,Sparse>::Index Index;
-// does not fix the issue.
-// An alternative is to define the nested class in the parent class itself.
-template<typename MatrixType> class TransposeImpl<MatrixType,Sparse>::InnerIterator
-  : public _MatrixTypeNested::InnerIterator
+    inline Scalar* valuePtr() { return derived().nestedExpression().valuePtr(); }
+    inline StorageIndex* innerIndexPtr() { return derived().nestedExpression().innerIndexPtr(); }
+    inline StorageIndex* outerIndexPtr() { return derived().nestedExpression().outerIndexPtr(); }
+    inline StorageIndex* innerNonZeroPtr() { return derived().nestedExpression().innerNonZeroPtr(); }
+  };
+}
+  
+template<typename MatrixType> class TransposeImpl<MatrixType,Sparse>
+  : public internal::SparseTransposeImpl<MatrixType>
 {
-    typedef typename _MatrixTypeNested::InnerIterator Base;
-    typedef typename TransposeImpl::Index Index;
-  public:
-
-    EIGEN_STRONG_INLINE InnerIterator(const TransposeImpl& trans, typename TransposeImpl<MatrixType,Sparse>::Index outer)
-      : Base(trans.derived().nestedExpression(), outer)
-    {}
-    typename TransposeImpl<MatrixType,Sparse>::Index row() const { return Base::col(); }
-    typename TransposeImpl<MatrixType,Sparse>::Index col() const { return Base::row(); }
+  protected:
+    typedef internal::SparseTransposeImpl<MatrixType> Base;
 };
 
-template<typename MatrixType> class TransposeImpl<MatrixType,Sparse>::ReverseInnerIterator
-  : public _MatrixTypeNested::ReverseInnerIterator
+namespace internal {
+  
+template<typename ArgType>
+struct unary_evaluator<Transpose<ArgType>, IteratorBased>
+  : public evaluator_base<Transpose<ArgType> >
 {
-    typedef typename _MatrixTypeNested::ReverseInnerIterator Base;
-    typedef typename TransposeImpl::Index Index;
+    typedef typename evaluator<ArgType>::InnerIterator        EvalIterator;
   public:
+    typedef Transpose<ArgType> XprType;
+    
+    inline Index nonZerosEstimate() const {
+      return m_argImpl.nonZerosEstimate();
+    }
 
-    EIGEN_STRONG_INLINE ReverseInnerIterator(const TransposeImpl& xpr, typename TransposeImpl<MatrixType,Sparse>::Index outer)
-      : Base(xpr.derived().nestedExpression(), outer)
-    {}
-    typename TransposeImpl<MatrixType,Sparse>::Index row() const { return Base::col(); }
-    typename TransposeImpl<MatrixType,Sparse>::Index col() const { return Base::row(); }
+    class InnerIterator : public EvalIterator
+    {
+    public:
+      EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& unaryOp, Index outer)
+        : EvalIterator(unaryOp.m_argImpl,outer)
+      {}
+      
+      Index row() const { return EvalIterator::col(); }
+      Index col() const { return EvalIterator::row(); }
+    };
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& op) :m_argImpl(op.nestedExpression()) {}
+
+  protected:
+    evaluator<ArgType> m_argImpl;
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSETRANSPOSE_H
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseTriangularView.h b/extern/Eigen3/Eigen/src/SparseCore/SparseTriangularView.h
index 333127b78e8..9ac120266a8 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseTriangularView.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseTriangularView.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -11,19 +11,19 @@
 #ifndef EIGEN_SPARSE_TRIANGULARVIEW_H
 #define EIGEN_SPARSE_TRIANGULARVIEW_H
 
-namespace Eigen { 
-
-namespace internal {
-  
-template<typename MatrixType, int Mode>
-struct traits<SparseTriangularView<MatrixType,Mode> >
-: public traits<MatrixType>
-{};
-
-} // namespace internal
-
-template<typename MatrixType, int Mode> class SparseTriangularView
-  : public SparseMatrixBase<SparseTriangularView<MatrixType,Mode> >
+namespace Eigen {
+
+/** \ingroup SparseCore_Module
+  *
+  * \brief Base class for a triangular part in a \b sparse matrix
+  *
+  * This class is an abstract base class of class TriangularView, and objects of type TriangularViewImpl cannot be instantiated.
+  * It extends class TriangularView with additional methods which are available for sparse expressions only.
+  *
+  * \sa class TriangularView, SparseMatrixBase::triangularView()
+  */
+template<typename MatrixType, unsigned int Mode> class TriangularViewImpl<MatrixType,Mode,Sparse>
+  : public SparseMatrixBase<TriangularView<MatrixType,Mode> >
 {
     enum { SkipFirst = ((Mode&Lower) && !(MatrixType::Flags&RowMajorBit))
                     || ((Mode&Upper) &&  (MatrixType::Flags&RowMajorBit)),
@@ -31,147 +31,157 @@ template<typename MatrixType, int Mode> class SparseTriangularView
            SkipDiag = (Mode&ZeroDiag) ? 1 : 0,
            HasUnitDiag = (Mode&UnitDiag) ? 1 : 0
     };
+    
+    typedef TriangularView<MatrixType,Mode> TriangularViewType;
+    
+  protected:
+    // dummy solve function to make TriangularView happy.
+    void solve() const;
 
+    typedef SparseMatrixBase<TriangularViewType> Base;
   public:
     
-    EIGEN_SPARSE_PUBLIC_INTERFACE(SparseTriangularView)
-
-    class InnerIterator;
-    class ReverseInnerIterator;
-
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
-
+    EIGEN_SPARSE_PUBLIC_INTERFACE(TriangularViewType)
+    
     typedef typename MatrixType::Nested MatrixTypeNested;
     typedef typename internal::remove_reference<MatrixTypeNested>::type MatrixTypeNestedNonRef;
     typedef typename internal::remove_all<MatrixTypeNested>::type MatrixTypeNestedCleaned;
 
-    inline SparseTriangularView(const MatrixType& matrix) : m_matrix(matrix) {}
-
-    /** \internal */
-    inline const MatrixTypeNestedCleaned& nestedExpression() const { return m_matrix; }
-
-    template<typename OtherDerived>
-    typename internal::plain_matrix_type_column_major<OtherDerived>::type
-    solve(const MatrixBase<OtherDerived>& other) const;
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _solve_impl(const RhsType &rhs, DstType &dst) const {
+      if(!(internal::is_same<RhsType,DstType>::value && internal::extract_data(dst) == internal::extract_data(rhs)))
+        dst = rhs;
+      this->solveInPlace(dst);
+    }
 
+    /** Applies the inverse of \c *this to the dense vector or matrix \a other, "in-place" */
     template<typename OtherDerived> void solveInPlace(MatrixBase<OtherDerived>& other) const;
-    template<typename OtherDerived> void solveInPlace(SparseMatrixBase<OtherDerived>& other) const;
 
-  protected:
-    MatrixTypeNested m_matrix;
+    /** Applies the inverse of \c *this to the sparse vector or matrix \a other, "in-place" */
+    template<typename OtherDerived> void solveInPlace(SparseMatrixBase<OtherDerived>& other) const;
+  
 };
 
-template<typename MatrixType, int Mode>
-class SparseTriangularView<MatrixType,Mode>::InnerIterator : public MatrixTypeNestedCleaned::InnerIterator
+namespace internal {
+
+template<typename ArgType, unsigned int Mode>
+struct unary_evaluator<TriangularView<ArgType,Mode>, IteratorBased>
+ : evaluator_base<TriangularView<ArgType,Mode> >
 {
-    typedef typename MatrixTypeNestedCleaned::InnerIterator Base;
-    typedef typename SparseTriangularView::Index Index;
-  public:
+  typedef TriangularView<ArgType,Mode> XprType;
+  
+protected:
+  
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename evaluator<ArgType>::InnerIterator EvalIterator;
+  
+  enum { SkipFirst = ((Mode&Lower) && !(ArgType::Flags&RowMajorBit))
+                    || ((Mode&Upper) &&  (ArgType::Flags&RowMajorBit)),
+         SkipLast = !SkipFirst,
+         SkipDiag = (Mode&ZeroDiag) ? 1 : 0,
+         HasUnitDiag = (Mode&UnitDiag) ? 1 : 0
+  };
+  
+public:
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    Flags = XprType::Flags
+  };
+    
+  explicit unary_evaluator(const XprType &xpr) : m_argImpl(xpr.nestedExpression()), m_arg(xpr.nestedExpression()) {}
+  
+  inline Index nonZerosEstimate() const {
+    return m_argImpl.nonZerosEstimate();
+  }
+  
+  class InnerIterator : public EvalIterator
+  {
+      typedef EvalIterator Base;
+    public:
 
-    EIGEN_STRONG_INLINE InnerIterator(const SparseTriangularView& view, Index outer)
-      : Base(view.nestedExpression(), outer), m_returnOne(false)
-    {
-      if(SkipFirst)
+      EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& xprEval, Index outer)
+        : Base(xprEval.m_argImpl,outer), m_returnOne(false), m_containsDiag(Base::outer()<xprEval.m_arg.innerSize())
       {
-        while((*this) && ((HasUnitDiag||SkipDiag)  ? this->index()<=outer : this->index()<outer))
-          Base::operator++();
-        if(HasUnitDiag)
-          m_returnOne = true;
+        if(SkipFirst)
+        {
+          while((*this) && ((HasUnitDiag||SkipDiag)  ? this->index()<=outer : this->index()<outer))
+            Base::operator++();
+          if(HasUnitDiag)
+            m_returnOne = m_containsDiag;
+        }
+        else if(HasUnitDiag && ((!Base::operator bool()) || Base::index()>=Base::outer()))
+        {
+          if((!SkipFirst) && Base::operator bool())
+            Base::operator++();
+          m_returnOne = m_containsDiag;
+        }
       }
-      else if(HasUnitDiag && ((!Base::operator bool()) || Base::index()>=Base::outer()))
+
+      EIGEN_STRONG_INLINE InnerIterator& operator++()
       {
-        if((!SkipFirst) && Base::operator bool())
+        if(HasUnitDiag && m_returnOne)
+          m_returnOne = false;
+        else
+        {
           Base::operator++();
-        m_returnOne = true;
+          if(HasUnitDiag && (!SkipFirst) && ((!Base::operator bool()) || Base::index()>=Base::outer()))
+          {
+            if((!SkipFirst) && Base::operator bool())
+              Base::operator++();
+            m_returnOne = m_containsDiag;
+          }
+        }
+        return *this;
       }
-    }
-
-    EIGEN_STRONG_INLINE InnerIterator& operator++()
-    {
-      if(HasUnitDiag && m_returnOne)
-        m_returnOne = false;
-      else
+      
+      EIGEN_STRONG_INLINE operator bool() const
       {
-        Base::operator++();
-        if(HasUnitDiag && (!SkipFirst) && ((!Base::operator bool()) || Base::index()>=Base::outer()))
+        if(HasUnitDiag && m_returnOne)
+          return true;
+        if(SkipFirst) return  Base::operator bool();
+        else
         {
-          if((!SkipFirst) && Base::operator bool())
-            Base::operator++();
-          m_returnOne = true;
+          if (SkipDiag) return (Base::operator bool() && this->index() < this->outer());
+          else return (Base::operator bool() && this->index() <= this->outer());
         }
       }
-      return *this;
-    }
-
-    inline Index row() const { return (MatrixType::Flags&RowMajorBit ? Base::outer() : this->index()); }
-    inline Index col() const { return (MatrixType::Flags&RowMajorBit ? this->index() : Base::outer()); }
-    inline Index index() const
-    {
-      if(HasUnitDiag && m_returnOne)  return Base::outer();
-      else                            return Base::index();
-    }
-    inline Scalar value() const
-    {
-      if(HasUnitDiag && m_returnOne)  return Scalar(1);
-      else                            return Base::value();
-    }
 
-    EIGEN_STRONG_INLINE operator bool() const
-    {
-      if(HasUnitDiag && m_returnOne)
-        return true;
-      if(SkipFirst) return  Base::operator bool();
-      else
+//       inline Index row() const { return (ArgType::Flags&RowMajorBit ? Base::outer() : this->index()); }
+//       inline Index col() const { return (ArgType::Flags&RowMajorBit ? this->index() : Base::outer()); }
+      inline StorageIndex index() const
       {
-        if (SkipDiag) return (Base::operator bool() && this->index() < this->outer());
-        else return (Base::operator bool() && this->index() <= this->outer());
+        if(HasUnitDiag && m_returnOne)  return internal::convert_index<StorageIndex>(Base::outer());
+        else                            return Base::index();
       }
-    }
-  protected:
-    bool m_returnOne;
-};
-
-template<typename MatrixType, int Mode>
-class SparseTriangularView<MatrixType,Mode>::ReverseInnerIterator : public MatrixTypeNestedCleaned::ReverseInnerIterator
-{
-    typedef typename MatrixTypeNestedCleaned::ReverseInnerIterator Base;
-    typedef typename SparseTriangularView::Index Index;
-  public:
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator(const SparseTriangularView& view, Index outer)
-      : Base(view.nestedExpression(), outer)
-    {
-      eigen_assert((!HasUnitDiag) && "ReverseInnerIterator does not support yet triangular views with a unit diagonal");
-      if(SkipLast) {
-        while((*this) && (SkipDiag ? this->index()>=outer : this->index()>outer))
-          --(*this);
-      }
-    }
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
-    { Base::operator--(); return *this; }
-
-    inline Index row() const { return Base::row(); }
-    inline Index col() const { return Base::col(); }
-
-    EIGEN_STRONG_INLINE operator bool() const
-    {
-      if (SkipLast) return Base::operator bool() ;
-      else
+      inline Scalar value() const
       {
-        if(SkipDiag) return (Base::operator bool() && this->index() > this->outer());
-        else return (Base::operator bool() && this->index() >= this->outer());
+        if(HasUnitDiag && m_returnOne)  return Scalar(1);
+        else                            return Base::value();
       }
-    }
+
+    protected:
+      bool m_returnOne;
+      bool m_containsDiag;
+    private:
+      Scalar& valueRef();
+  };
+  
+protected:
+  evaluator<ArgType> m_argImpl;
+  const ArgType& m_arg;
 };
 
+} // end namespace internal
+
 template<typename Derived>
 template<int Mode>
-inline const SparseTriangularView<Derived, Mode>
+inline const TriangularView<const Derived, Mode>
 SparseMatrixBase<Derived>::triangularView() const
 {
-  return derived();
+  return TriangularView<const Derived, Mode>(derived());
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseUtil.h b/extern/Eigen3/Eigen/src/SparseCore/SparseUtil.h
index d627546def0..74df0d49642 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseUtil.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseUtil.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -37,42 +37,23 @@ EIGEN_STRONG_INLINE Derived& operator Op(const Other& scalar) \
 }
 
 #define EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATORS(Derived) \
-EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(Derived, =) \
-EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(Derived, +=) \
-EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(Derived, -=) \
-EIGEN_SPARSE_INHERIT_SCALAR_ASSIGNMENT_OPERATOR(Derived, *=) \
-EIGEN_SPARSE_INHERIT_SCALAR_ASSIGNMENT_OPERATOR(Derived, /=)
-
-#define _EIGEN_SPARSE_PUBLIC_INTERFACE(Derived, BaseClass) \
-  typedef BaseClass Base; \
-  typedef typename Eigen::internal::traits<Derived >::Scalar Scalar; \
-  typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; \
-  typedef typename Eigen::internal::nested<Derived >::type Nested; \
-  typedef typename Eigen::internal::traits<Derived >::StorageKind StorageKind; \
-  typedef typename Eigen::internal::traits<Derived >::Index Index; \
-  enum { RowsAtCompileTime = Eigen::internal::traits<Derived >::RowsAtCompileTime, \
-        ColsAtCompileTime = Eigen::internal::traits<Derived >::ColsAtCompileTime, \
-        Flags = Eigen::internal::traits<Derived >::Flags, \
-        CoeffReadCost = Eigen::internal::traits<Derived >::CoeffReadCost, \
-        SizeAtCompileTime = Base::SizeAtCompileTime, \
-        IsVectorAtCompileTime = Base::IsVectorAtCompileTime }; \
-  using Base::derived; \
-  using Base::const_cast_derived;
+EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(Derived, =)
+
 
 #define EIGEN_SPARSE_PUBLIC_INTERFACE(Derived) \
-  _EIGEN_SPARSE_PUBLIC_INTERFACE(Derived, Eigen::SparseMatrixBase<Derived >)
+  EIGEN_GENERIC_PUBLIC_INTERFACE(Derived)
 
+  
 const int CoherentAccessPattern     = 0x1;
 const int InnerRandomAccessPattern  = 0x2 | CoherentAccessPattern;
 const int OuterRandomAccessPattern  = 0x4 | CoherentAccessPattern;
 const int RandomAccessPattern       = 0x8 | OuterRandomAccessPattern | InnerRandomAccessPattern;
 
-template<typename _Scalar, int _Flags = 0, typename _Index = int>  class SparseMatrix;
-template<typename _Scalar, int _Flags = 0, typename _Index = int>  class DynamicSparseMatrix;
-template<typename _Scalar, int _Flags = 0, typename _Index = int>  class SparseVector;
-template<typename _Scalar, int _Flags = 0, typename _Index = int>  class MappedSparseMatrix;
+template<typename _Scalar, int _Flags = 0, typename _StorageIndex = int>  class SparseMatrix;
+template<typename _Scalar, int _Flags = 0, typename _StorageIndex = int>  class DynamicSparseMatrix;
+template<typename _Scalar, int _Flags = 0, typename _StorageIndex = int>  class SparseVector;
+template<typename _Scalar, int _Flags = 0, typename _StorageIndex = int>  class MappedSparseMatrix;
 
-template<typename MatrixType, int Mode>           class SparseTriangularView;
 template<typename MatrixType, unsigned int UpLo>  class SparseSelfAdjointView;
 template<typename Lhs, typename Rhs>              class SparseDiagonalProduct;
 template<typename MatrixType> class SparseView;
@@ -84,42 +65,44 @@ template<typename Lhs, typename Rhs, bool Transpose> class SparseDenseOuterProdu
 
 template<typename Lhs, typename Rhs> struct SparseSparseProductReturnType;
 template<typename Lhs, typename Rhs,
-         int InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(internal::traits<Lhs>::ColsAtCompileTime,internal::traits<Rhs>::RowsAtCompileTime)> struct DenseSparseProductReturnType;         
+         int InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(internal::traits<Lhs>::ColsAtCompileTime,internal::traits<Rhs>::RowsAtCompileTime)> struct DenseSparseProductReturnType;
+         
 template<typename Lhs, typename Rhs,
          int InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(internal::traits<Lhs>::ColsAtCompileTime,internal::traits<Rhs>::RowsAtCompileTime)> struct SparseDenseProductReturnType;
 template<typename MatrixType,int UpLo> class SparseSymmetricPermutationProduct;
 
 namespace internal {
 
-template<typename T,int Rows,int Cols> struct sparse_eval;
+template<typename T,int Rows,int Cols,int Flags> struct sparse_eval;
 
 template<typename T> struct eval<T,Sparse>
-  : public sparse_eval<T, traits<T>::RowsAtCompileTime,traits<T>::ColsAtCompileTime>
+  : sparse_eval<T, traits<T>::RowsAtCompileTime,traits<T>::ColsAtCompileTime,traits<T>::Flags>
 {};
 
-template<typename T,int Cols> struct sparse_eval<T,1,Cols> {
+template<typename T,int Cols,int Flags> struct sparse_eval<T,1,Cols,Flags> {
     typedef typename traits<T>::Scalar _Scalar;
-    typedef typename traits<T>::Index _Index;
+    typedef typename traits<T>::StorageIndex _StorageIndex;
   public:
-    typedef SparseVector<_Scalar, RowMajor, _Index> type;
+    typedef SparseVector<_Scalar, RowMajor, _StorageIndex> type;
 };
 
-template<typename T,int Rows> struct sparse_eval<T,Rows,1> {
+template<typename T,int Rows,int Flags> struct sparse_eval<T,Rows,1,Flags> {
     typedef typename traits<T>::Scalar _Scalar;
-    typedef typename traits<T>::Index _Index;
+    typedef typename traits<T>::StorageIndex _StorageIndex;
   public:
-    typedef SparseVector<_Scalar, ColMajor, _Index> type;
+    typedef SparseVector<_Scalar, ColMajor, _StorageIndex> type;
 };
 
-template<typename T,int Rows,int Cols> struct sparse_eval {
+// TODO this seems almost identical to plain_matrix_type<T, Sparse>
+template<typename T,int Rows,int Cols,int Flags> struct sparse_eval {
     typedef typename traits<T>::Scalar _Scalar;
-    typedef typename traits<T>::Index _Index;
-    enum { _Options = ((traits<T>::Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor };
+    typedef typename traits<T>::StorageIndex _StorageIndex;
+    enum { _Options = ((Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor };
   public:
-    typedef SparseMatrix<_Scalar, _Options, _Index> type;
+    typedef SparseMatrix<_Scalar, _Options, _StorageIndex> type;
 };
 
-template<typename T> struct sparse_eval<T,1,1> {
+template<typename T,int Flags> struct sparse_eval<T,1,1,Flags> {
     typedef typename traits<T>::Scalar _Scalar;
   public:
     typedef Matrix<_Scalar, 1, 1> type;
@@ -128,12 +111,35 @@ template<typename T> struct sparse_eval<T,1,1> {
 template<typename T> struct plain_matrix_type<T,Sparse>
 {
   typedef typename traits<T>::Scalar _Scalar;
-  typedef typename traits<T>::Index _Index;
-  enum { _Options = ((traits<T>::Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor };
+  typedef typename traits<T>::StorageIndex _StorageIndex;
+  enum { _Options = ((evaluator<T>::Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor };
   public:
-    typedef SparseMatrix<_Scalar, _Options, _Index> type;
+    typedef SparseMatrix<_Scalar, _Options, _StorageIndex> type;
+};
+
+template<typename T>
+struct plain_object_eval<T,Sparse>
+  : sparse_eval<T, traits<T>::RowsAtCompileTime,traits<T>::ColsAtCompileTime, evaluator<T>::Flags>
+{};
+
+template<typename Decomposition, typename RhsType>
+struct solve_traits<Decomposition,RhsType,Sparse>
+{
+  typedef typename sparse_eval<RhsType, RhsType::RowsAtCompileTime, RhsType::ColsAtCompileTime,traits<RhsType>::Flags>::type PlainObject;
 };
 
+template<typename Derived>
+struct generic_xpr_base<Derived, MatrixXpr, Sparse>
+{
+  typedef SparseMatrixBase<Derived> type;
+};
+
+struct SparseTriangularShape  { static std::string debugName() { return "SparseTriangularShape"; } };
+struct SparseSelfAdjointShape { static std::string debugName() { return "SparseSelfAdjointShape"; } };
+
+template<> struct glue_shapes<SparseShape,SelfAdjointShape> { typedef SparseSelfAdjointShape type;  };
+template<> struct glue_shapes<SparseShape,TriangularShape > { typedef SparseTriangularShape  type;  };
+
 } // end namespace internal
 
 /** \ingroup SparseCore_Module
@@ -144,26 +150,26 @@ template<typename T> struct plain_matrix_type<T,Sparse>
   *
   * \sa SparseMatrix::setFromTriplets()
   */
-template<typename Scalar, typename Index=typename SparseMatrix<Scalar>::Index >
+template<typename Scalar, typename StorageIndex=typename SparseMatrix<Scalar>::StorageIndex >
 class Triplet
 {
 public:
   Triplet() : m_row(0), m_col(0), m_value(0) {}
 
-  Triplet(const Index& i, const Index& j, const Scalar& v = Scalar(0))
+  Triplet(const StorageIndex& i, const StorageIndex& j, const Scalar& v = Scalar(0))
     : m_row(i), m_col(j), m_value(v)
   {}
 
   /** \returns the row index of the element */
-  const Index& row() const { return m_row; }
+  const StorageIndex& row() const { return m_row; }
 
   /** \returns the column index of the element */
-  const Index& col() const { return m_col; }
+  const StorageIndex& col() const { return m_col; }
 
   /** \returns the value of the element */
   const Scalar& value() const { return m_value; }
 protected:
-  Index m_row, m_col;
+  StorageIndex m_row, m_col;
   Scalar m_value;
 };
 
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseVector.h b/extern/Eigen3/Eigen/src/SparseCore/SparseVector.h
index 49865d0e72f..19b0fbc9d70 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseVector.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseVector.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -22,15 +22,15 @@ namespace Eigen {
   * See http://www.netlib.org/linalg/html_templates/node91.html for details on the storage scheme.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_SPARSEVECTOR_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_SPARSEVECTOR_PLUGIN.
   */
 
 namespace internal {
-template<typename _Scalar, int _Options, typename _Index>
-struct traits<SparseVector<_Scalar, _Options, _Index> >
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct traits<SparseVector<_Scalar, _Options, _StorageIndex> >
 {
   typedef _Scalar Scalar;
-  typedef _Index Index;
+  typedef _StorageIndex StorageIndex;
   typedef Sparse StorageKind;
   typedef MatrixXpr XprKind;
   enum {
@@ -40,8 +40,7 @@ struct traits<SparseVector<_Scalar, _Options, _Index> >
     ColsAtCompileTime = IsColVector ? 1 : Dynamic,
     MaxRowsAtCompileTime = RowsAtCompileTime,
     MaxColsAtCompileTime = ColsAtCompileTime,
-    Flags = _Options | NestByRefBit | LvalueBit | (IsColVector ? 0 : RowMajorBit),
-    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Flags = _Options | NestByRefBit | LvalueBit | (IsColVector ? 0 : RowMajorBit) | CompressedAccessBit,
     SupportedAccessPatterns = InnerRandomAccessPattern
   };
 };
@@ -61,18 +60,18 @@ struct sparse_vector_assign_selector;
 
 }
 
-template<typename _Scalar, int _Options, typename _Index>
+template<typename _Scalar, int _Options, typename _StorageIndex>
 class SparseVector
-  : public SparseMatrixBase<SparseVector<_Scalar, _Options, _Index> >
+  : public SparseCompressedBase<SparseVector<_Scalar, _Options, _StorageIndex> >
 {
-    typedef SparseMatrixBase<SparseVector> SparseBase;
-    
+    typedef SparseCompressedBase<SparseVector> Base;
+    using Base::convert_index;
   public:
     EIGEN_SPARSE_PUBLIC_INTERFACE(SparseVector)
     EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseVector, +=)
     EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseVector, -=)
     
-    typedef internal::CompressedStorage<Scalar,Index> Storage;
+    typedef internal::CompressedStorage<Scalar,StorageIndex> Storage;
     enum { IsColVector = internal::traits<SparseVector>::IsColVector };
     
     enum {
@@ -84,11 +83,16 @@ class SparseVector
     EIGEN_STRONG_INLINE Index innerSize() const { return m_size; }
     EIGEN_STRONG_INLINE Index outerSize() const { return 1; }
 
-    EIGEN_STRONG_INLINE const Scalar* valuePtr() const { return &m_data.value(0); }
-    EIGEN_STRONG_INLINE Scalar* valuePtr() { return &m_data.value(0); }
+    EIGEN_STRONG_INLINE const Scalar* valuePtr() const { return m_data.valuePtr(); }
+    EIGEN_STRONG_INLINE Scalar* valuePtr() { return m_data.valuePtr(); }
 
-    EIGEN_STRONG_INLINE const Index* innerIndexPtr() const { return &m_data.index(0); }
-    EIGEN_STRONG_INLINE Index* innerIndexPtr() { return &m_data.index(0); }
+    EIGEN_STRONG_INLINE const StorageIndex* innerIndexPtr() const { return m_data.indexPtr(); }
+    EIGEN_STRONG_INLINE StorageIndex* innerIndexPtr() { return m_data.indexPtr(); }
+
+    inline const StorageIndex* outerIndexPtr() const { return 0; }
+    inline StorageIndex* outerIndexPtr() { return 0; }
+    inline const StorageIndex* innerNonZeroPtr() const { return 0; }
+    inline StorageIndex* innerNonZeroPtr() { return 0; }
     
     /** \internal */
     inline Storage& data() { return m_data; }
@@ -103,13 +107,13 @@ class SparseVector
     inline Scalar coeff(Index i) const
     {
       eigen_assert(i>=0 && i<m_size);
-      return m_data.at(i);
+      return m_data.at(StorageIndex(i));
     }
 
     inline Scalar& coeffRef(Index row, Index col)
     {
       eigen_assert(IsColVector ? (col==0 && row>=0 && row<m_size) : (row==0 && col>=0 && col<m_size));
-      return coeff(IsColVector ? row : col);
+      return coeffRef(IsColVector ? row : col);
     }
 
     /** \returns a reference to the coefficient value at given index \a i
@@ -121,18 +125,19 @@ class SparseVector
     inline Scalar& coeffRef(Index i)
     {
       eigen_assert(i>=0 && i<m_size);
-      return m_data.atWithInsertion(i);
+
+      return m_data.atWithInsertion(StorageIndex(i));
     }
 
   public:
 
-    class InnerIterator;
-    class ReverseInnerIterator;
+    typedef typename Base::InnerIterator InnerIterator;
+    typedef typename Base::ReverseInnerIterator ReverseInnerIterator;
 
     inline void setZero() { m_data.clear(); }
 
     /** \returns the number of non zero coefficients */
-    inline Index nonZeros() const  { return static_cast<Index>(m_data.size()); }
+    inline Index nonZeros() const  { return m_data.size(); }
 
     inline void startVec(Index outer)
     {
@@ -151,6 +156,18 @@ class SparseVector
       m_data.append(0, i);
       return m_data.value(m_data.size()-1);
     }
+    
+    Scalar& insertBackByOuterInnerUnordered(Index outer, Index inner)
+    {
+      EIGEN_UNUSED_VARIABLE(outer);
+      eigen_assert(outer==0);
+      return insertBackUnordered(inner);
+    }
+    inline Scalar& insertBackUnordered(Index i)
+    {
+      m_data.append(0, i);
+      return m_data.value(m_data.size()-1);
+    }
 
     inline Scalar& insert(Index row, Index col)
     {
@@ -177,7 +194,7 @@ class SparseVector
         m_data.value(p+1) = m_data.value(p);
         --p;
       }
-      m_data.index(p+1) = i;
+      m_data.index(p+1) = convert_index(i);
       m_data.value(p+1) = 0;
       return m_data.value(p+1);
     }
@@ -189,28 +206,59 @@ class SparseVector
 
     inline void finalize() {}
 
+    /** \copydoc SparseMatrix::prune(const Scalar&,const RealScalar&) */
     void prune(const Scalar& reference, const RealScalar& epsilon = NumTraits<RealScalar>::dummy_precision())
     {
       m_data.prune(reference,epsilon);
     }
 
+    /** Resizes the sparse vector to \a rows x \a cols
+      *
+      * This method is provided for compatibility with matrices.
+      * For a column vector, \a cols must be equal to 1.
+      * For a row vector, \a rows must be equal to 1.
+      *
+      * \sa resize(Index)
+      */
     void resize(Index rows, Index cols)
     {
-      eigen_assert(rows==1 || cols==1);
+      eigen_assert((IsColVector ? cols : rows)==1 && "Outer dimension must equal 1");
       resize(IsColVector ? rows : cols);
     }
 
+    /** Resizes the sparse vector to \a newSize
+      * This method deletes all entries, thus leaving an empty sparse vector
+      *
+      * \sa  conservativeResize(), setZero() */
     void resize(Index newSize)
     {
       m_size = newSize;
       m_data.clear();
     }
 
+    /** Resizes the sparse vector to \a newSize, while leaving old values untouched.
+      *
+      * If the size of the vector is decreased, then the storage of the out-of bounds coefficients is kept and reserved.
+      * Call .data().squeeze() to free extra memory.
+      *
+      * \sa reserve(), setZero()
+      */
+    void conservativeResize(Index newSize)
+    {
+      if (newSize < m_size)
+      {
+        Index i = 0;
+        while (i<m_data.size() && m_data.index(i)<newSize) ++i;
+        m_data.resize(i);
+      }
+      m_size = newSize;
+    }
+
     void resizeNonZeros(Index size) { m_data.resize(size); }
 
     inline SparseVector() : m_size(0) { check_template_parameters(); resize(0); }
 
-    inline SparseVector(Index size) : m_size(0) { check_template_parameters(); resize(size); }
+    explicit inline SparseVector(Index size) : m_size(0) { check_template_parameters(); resize(size); }
 
     inline SparseVector(Index rows, Index cols) : m_size(0) { check_template_parameters(); resize(rows,cols); }
 
@@ -218,12 +266,15 @@ class SparseVector
     inline SparseVector(const SparseMatrixBase<OtherDerived>& other)
       : m_size(0)
     {
+      #ifdef EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+        EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+      #endif
       check_template_parameters();
       *this = other.derived();
     }
 
     inline SparseVector(const SparseVector& other)
-      : SparseBase(other), m_size(0)
+      : Base(other), m_size(0)
     {
       check_template_parameters();
       *this = other.derived();
@@ -239,6 +290,14 @@ class SparseVector
       m_data.swap(other.m_data);
     }
 
+    template<int OtherOptions>
+    inline void swap(SparseMatrix<Scalar,OtherOptions,StorageIndex>& other)
+    {
+      eigen_assert(other.outerSize()==1);
+      std::swap(m_size, other.m_innerSize);
+      m_data.swap(other.m_data);
+    }
+
     inline SparseVector& operator=(const SparseVector& other)
     {
       if (other.isRValue())
@@ -337,7 +396,7 @@ protected:
   
     static void check_template_parameters()
     {
-      EIGEN_STATIC_ASSERT(NumTraits<Index>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE);
+      EIGEN_STATIC_ASSERT(NumTraits<StorageIndex>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE);
       EIGEN_STATIC_ASSERT((_Options&(ColMajor|RowMajor))==Options,INVALID_MATRIX_TEMPLATE_PARAMETERS);
     }
     
@@ -345,77 +404,46 @@ protected:
     Index m_size;
 };
 
-template<typename Scalar, int _Options, typename _Index>
-class SparseVector<Scalar,_Options,_Index>::InnerIterator
-{
-  public:
-    InnerIterator(const SparseVector& vec, Index outer=0)
-      : m_data(vec.m_data), m_id(0), m_end(static_cast<Index>(m_data.size()))
-    {
-      EIGEN_UNUSED_VARIABLE(outer);
-      eigen_assert(outer==0);
-    }
-
-    InnerIterator(const internal::CompressedStorage<Scalar,Index>& data)
-      : m_data(data), m_id(0), m_end(static_cast<Index>(m_data.size()))
-    {}
-
-    inline InnerIterator& operator++() { m_id++; return *this; }
-
-    inline Scalar value() const { return m_data.value(m_id); }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_data.value(m_id)); }
-
-    inline Index index() const { return m_data.index(m_id); }
-    inline Index row() const { return IsColVector ? index() : 0; }
-    inline Index col() const { return IsColVector ? 0 : index(); }
-
-    inline operator bool() const { return (m_id < m_end); }
-
-  protected:
-    const internal::CompressedStorage<Scalar,Index>& m_data;
-    Index m_id;
-    const Index m_end;
-};
+namespace internal {
 
-template<typename Scalar, int _Options, typename _Index>
-class SparseVector<Scalar,_Options,_Index>::ReverseInnerIterator
+template<typename _Scalar, int _Options, typename _Index>
+struct evaluator<SparseVector<_Scalar,_Options,_Index> >
+  : evaluator_base<SparseVector<_Scalar,_Options,_Index> >
 {
-  public:
-    ReverseInnerIterator(const SparseVector& vec, Index outer=0)
-      : m_data(vec.m_data), m_id(static_cast<Index>(m_data.size())), m_start(0)
-    {
-      EIGEN_UNUSED_VARIABLE(outer);
-      eigen_assert(outer==0);
-    }
-
-    ReverseInnerIterator(const internal::CompressedStorage<Scalar,Index>& data)
-      : m_data(data), m_id(static_cast<Index>(m_data.size())), m_start(0)
-    {}
-
-    inline ReverseInnerIterator& operator--() { m_id--; return *this; }
-
-    inline Scalar value() const { return m_data.value(m_id-1); }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_data.value(m_id-1)); }
-
-    inline Index index() const { return m_data.index(m_id-1); }
-    inline Index row() const { return IsColVector ? index() : 0; }
-    inline Index col() const { return IsColVector ? 0 : index(); }
-
-    inline operator bool() const { return (m_id > m_start); }
+  typedef SparseVector<_Scalar,_Options,_Index> SparseVectorType;
+  typedef evaluator_base<SparseVectorType> Base;
+  typedef typename SparseVectorType::InnerIterator InnerIterator;
+  typedef typename SparseVectorType::ReverseInnerIterator ReverseInnerIterator;
+  
+  enum {
+    CoeffReadCost = NumTraits<_Scalar>::ReadCost,
+    Flags = SparseVectorType::Flags
+  };
 
-  protected:
-    const internal::CompressedStorage<Scalar,Index>& m_data;
-    Index m_id;
-    const Index m_start;
+  evaluator() : Base() {}
+  
+  explicit evaluator(const SparseVectorType &mat) : m_matrix(&mat)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_matrix->nonZeros();
+  }
+  
+  operator SparseVectorType&() { return m_matrix->const_cast_derived(); }
+  operator const SparseVectorType&() const { return *m_matrix; }
+  
+  const SparseVectorType *m_matrix;
 };
 
-namespace internal {
-
 template< typename Dest, typename Src>
 struct sparse_vector_assign_selector<Dest,Src,SVA_Inner> {
   static void run(Dest& dst, const Src& src) {
     eigen_internal_assert(src.innerSize()==src.size());
-    for(typename Src::InnerIterator it(src, 0); it; ++it)
+    typedef internal::evaluator<Src> SrcEvaluatorType;
+    SrcEvaluatorType srcEval(src);
+    for(typename SrcEvaluatorType::InnerIterator it(srcEval, 0); it; ++it)
       dst.insert(it.index()) = it.value();
   }
 };
@@ -424,9 +452,11 @@ template< typename Dest, typename Src>
 struct sparse_vector_assign_selector<Dest,Src,SVA_Outer> {
   static void run(Dest& dst, const Src& src) {
     eigen_internal_assert(src.outerSize()==src.size());
-    for(typename Dest::Index i=0; i<src.size(); ++i)
+    typedef internal::evaluator<Src> SrcEvaluatorType;
+    SrcEvaluatorType srcEval(src);
+    for(Index i=0; i<src.size(); ++i)
     {
-      typename Src::InnerIterator it(src, i);
+      typename SrcEvaluatorType::InnerIterator it(srcEval, i);
       if(it)
         dst.insert(i) = it.value();
     }
diff --git a/extern/Eigen3/Eigen/src/SparseCore/SparseView.h b/extern/Eigen3/Eigen/src/SparseCore/SparseView.h
index fd8450463fe..7c4aea743e0 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/SparseView.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/SparseView.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2010 Daniel Lowengrub <lowdanie@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -18,7 +18,7 @@ namespace internal {
 template<typename MatrixType>
 struct traits<SparseView<MatrixType> > : traits<MatrixType>
 {
-  typedef typename MatrixType::Index Index;
+  typedef typename MatrixType::StorageIndex StorageIndex;
   typedef Sparse StorageKind;
   enum {
     Flags = int(traits<MatrixType>::Flags) & (RowMajorBit)
@@ -27,71 +27,225 @@ struct traits<SparseView<MatrixType> > : traits<MatrixType>
 
 } // end namespace internal
 
+/** \ingroup SparseCore_Module
+  * \class SparseView
+  *
+  * \brief Expression of a dense or sparse matrix with zero or too small values removed
+  *
+  * \tparam MatrixType the type of the object of which we are removing the small entries
+  *
+  * This class represents an expression of a given dense or sparse matrix with
+  * entries smaller than \c reference * \c epsilon are removed.
+  * It is the return type of MatrixBase::sparseView() and SparseMatrixBase::pruned()
+  * and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::sparseView(), SparseMatrixBase::pruned()
+  */
 template<typename MatrixType>
 class SparseView : public SparseMatrixBase<SparseView<MatrixType> >
 {
   typedef typename MatrixType::Nested MatrixTypeNested;
   typedef typename internal::remove_all<MatrixTypeNested>::type _MatrixTypeNested;
+  typedef SparseMatrixBase<SparseView > Base;
 public:
   EIGEN_SPARSE_PUBLIC_INTERFACE(SparseView)
+  typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
-  SparseView(const MatrixType& mat, const Scalar& m_reference = Scalar(0),
-             typename NumTraits<Scalar>::Real m_epsilon = NumTraits<Scalar>::dummy_precision()) : 
-    m_matrix(mat), m_reference(m_reference), m_epsilon(m_epsilon) {}
-
-  class InnerIterator;
+  explicit SparseView(const MatrixType& mat, const Scalar& reference = Scalar(0),
+                      const RealScalar &epsilon = NumTraits<Scalar>::dummy_precision())
+    : m_matrix(mat), m_reference(reference), m_epsilon(epsilon) {}
 
   inline Index rows() const { return m_matrix.rows(); }
   inline Index cols() const { return m_matrix.cols(); }
 
   inline Index innerSize() const { return m_matrix.innerSize(); }
   inline Index outerSize() const { return m_matrix.outerSize(); }
-
+  
+  /** \returns the nested expression */
+  const typename internal::remove_all<MatrixTypeNested>::type&
+  nestedExpression() const { return m_matrix; }
+  
+  Scalar reference() const { return m_reference; }
+  RealScalar epsilon() const { return m_epsilon; }
+  
 protected:
   MatrixTypeNested m_matrix;
   Scalar m_reference;
-  typename NumTraits<Scalar>::Real m_epsilon;
+  RealScalar m_epsilon;
 };
 
-template<typename MatrixType>
-class SparseView<MatrixType>::InnerIterator : public _MatrixTypeNested::InnerIterator
-{
-  typedef typename SparseView::Index Index;
-public:
-  typedef typename _MatrixTypeNested::InnerIterator IterBase;
-  InnerIterator(const SparseView& view, Index outer) :
-  IterBase(view.m_matrix, outer), m_view(view)
-  {
-    incrementToNonZero();
-  }
-
-  EIGEN_STRONG_INLINE InnerIterator& operator++()
-  {
-    IterBase::operator++();
-    incrementToNonZero();
-    return *this;
-  }
-
-  using IterBase::value;
+namespace internal {
 
-protected:
-  const SparseView& m_view;
+// TODO find a way to unify the two following variants
+// This is tricky because implementing an inner iterator on top of an IndexBased evaluator is
+// not easy because the evaluators do not expose the sizes of the underlying expression.
+  
+template<typename ArgType>
+struct unary_evaluator<SparseView<ArgType>, IteratorBased>
+  : public evaluator_base<SparseView<ArgType> >
+{
+    typedef typename evaluator<ArgType>::InnerIterator EvalIterator;
+  public:
+    typedef SparseView<ArgType> XprType;
+    
+    class InnerIterator : public EvalIterator
+    {
+        typedef typename XprType::Scalar Scalar;
+      public:
+
+        EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& sve, Index outer)
+          : EvalIterator(sve.m_argImpl,outer), m_view(sve.m_view)
+        {
+          incrementToNonZero();
+        }
+
+        EIGEN_STRONG_INLINE InnerIterator& operator++()
+        {
+          EvalIterator::operator++();
+          incrementToNonZero();
+          return *this;
+        }
+
+        using EvalIterator::value;
+
+      protected:
+        const XprType &m_view;
+
+      private:
+        void incrementToNonZero()
+        {
+          while((bool(*this)) && internal::isMuchSmallerThan(value(), m_view.reference(), m_view.epsilon()))
+          {
+            EvalIterator::operator++();
+          }
+        }
+    };
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_view(xpr) {}
+
+  protected:
+    evaluator<ArgType> m_argImpl;
+    const XprType &m_view;
+};
 
-private:
-  void incrementToNonZero()
-  {
-    while((bool(*this)) && internal::isMuchSmallerThan(value(), m_view.m_reference, m_view.m_epsilon))
+template<typename ArgType>
+struct unary_evaluator<SparseView<ArgType>, IndexBased>
+  : public evaluator_base<SparseView<ArgType> >
+{
+  public:
+    typedef SparseView<ArgType> XprType;
+  protected:
+    enum { IsRowMajor = (XprType::Flags&RowMajorBit)==RowMajorBit };
+    typedef typename XprType::Scalar Scalar;
+    typedef typename XprType::StorageIndex StorageIndex;
+  public:
+    
+    class InnerIterator
     {
-      IterBase::operator++();
-    }
-  }
+      public:
+
+        EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& sve, Index outer)
+          : m_sve(sve), m_inner(0), m_outer(outer), m_end(sve.m_view.innerSize())
+        {
+          incrementToNonZero();
+        }
+
+        EIGEN_STRONG_INLINE InnerIterator& operator++()
+        {
+          m_inner++;
+          incrementToNonZero();
+          return *this;
+        }
+
+        EIGEN_STRONG_INLINE Scalar value() const
+        {
+          return (IsRowMajor) ? m_sve.m_argImpl.coeff(m_outer, m_inner)
+                              : m_sve.m_argImpl.coeff(m_inner, m_outer);
+        }
+
+        EIGEN_STRONG_INLINE StorageIndex index() const { return m_inner; }
+        inline Index row() const { return IsRowMajor ? m_outer : index(); }
+        inline Index col() const { return IsRowMajor ? index() : m_outer; }
+
+        EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; }
+
+      protected:
+        const unary_evaluator &m_sve;
+        Index m_inner;
+        const Index m_outer;
+        const Index m_end;
+
+      private:
+        void incrementToNonZero()
+        {
+          while((bool(*this)) && internal::isMuchSmallerThan(value(), m_sve.m_view.reference(), m_sve.m_view.epsilon()))
+          {
+            m_inner++;
+          }
+        }
+    };
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_view(xpr) {}
+
+  protected:
+    evaluator<ArgType> m_argImpl;
+    const XprType &m_view;
 };
 
+} // end namespace internal
+
+/** \ingroup SparseCore_Module
+  *
+  * \returns a sparse expression of the dense expression \c *this with values smaller than
+  * \a reference * \a epsilon removed.
+  *
+  * This method is typically used when prototyping to convert a quickly assembled dense Matrix \c D to a SparseMatrix \c S:
+  * \code
+  * MatrixXd D(n,m);
+  * SparseMatrix<double> S;
+  * S = D.sparseView();             // suppress numerical zeros (exact)
+  * S = D.sparseView(reference);
+  * S = D.sparseView(reference,epsilon);
+  * \endcode
+  * where \a reference is a meaningful non zero reference value,
+  * and \a epsilon is a tolerance factor defaulting to NumTraits<Scalar>::dummy_precision().
+  *
+  * \sa SparseMatrixBase::pruned(), class SparseView */
+template<typename Derived>
+const SparseView<Derived> MatrixBase<Derived>::sparseView(const Scalar& reference,
+                                                          const typename NumTraits<Scalar>::Real& epsilon) const
+{
+  return SparseView<Derived>(derived(), reference, epsilon);
+}
+
+/** \returns an expression of \c *this with values smaller than
+  * \a reference * \a epsilon removed.
+  *
+  * This method is typically used in conjunction with the product of two sparse matrices
+  * to automatically prune the smallest values as follows:
+  * \code
+  * C = (A*B).pruned();             // suppress numerical zeros (exact)
+  * C = (A*B).pruned(ref);
+  * C = (A*B).pruned(ref,epsilon);
+  * \endcode
+  * where \c ref is a meaningful non zero reference value.
+  * */
 template<typename Derived>
-const SparseView<Derived> MatrixBase<Derived>::sparseView(const Scalar& m_reference,
-                                                          const typename NumTraits<Scalar>::Real& m_epsilon) const
+const SparseView<Derived>
+SparseMatrixBase<Derived>::pruned(const Scalar& reference,
+                                  const RealScalar& epsilon) const
 {
-  return SparseView<Derived>(derived(), m_reference, m_epsilon);
+  return SparseView<Derived>(derived(), reference, epsilon);
 }
 
 } // end namespace Eigen
diff --git a/extern/Eigen3/Eigen/src/SparseCore/TriangularSolver.h b/extern/Eigen3/Eigen/src/SparseCore/TriangularSolver.h
index ccc12af7962..f9c56ba7980 100644
--- a/extern/Eigen3/Eigen/src/SparseCore/TriangularSolver.h
+++ b/extern/Eigen3/Eigen/src/SparseCore/TriangularSolver.h
@@ -28,16 +28,19 @@ template<typename Lhs, typename Rhs, int Mode>
 struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Lower,RowMajor>
 {
   typedef typename Rhs::Scalar Scalar;
+  typedef evaluator<Lhs> LhsEval;
+  typedef typename evaluator<Lhs>::InnerIterator LhsIterator;
   static void run(const Lhs& lhs, Rhs& other)
   {
-    for(int col=0 ; col<other.cols() ; ++col)
+    LhsEval lhsEval(lhs);
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
-      for(int i=0; i<lhs.rows(); ++i)
+      for(Index i=0; i<lhs.rows(); ++i)
       {
         Scalar tmp = other.coeff(i,col);
         Scalar lastVal(0);
-        int lastIndex = 0;
-        for(typename Lhs::InnerIterator it(lhs, i); it; ++it)
+        Index lastIndex = 0;
+        for(LhsIterator it(lhsEval, i); it; ++it)
         {
           lastVal = it.value();
           lastIndex = it.index();
@@ -62,15 +65,18 @@ template<typename Lhs, typename Rhs, int Mode>
 struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,RowMajor>
 {
   typedef typename Rhs::Scalar Scalar;
+  typedef evaluator<Lhs> LhsEval;
+  typedef typename evaluator<Lhs>::InnerIterator LhsIterator;
   static void run(const Lhs& lhs, Rhs& other)
   {
-    for(int col=0 ; col<other.cols() ; ++col)
+    LhsEval lhsEval(lhs);
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
-      for(int i=lhs.rows()-1 ; i>=0 ; --i)
+      for(Index i=lhs.rows()-1 ; i>=0 ; --i)
       {
         Scalar tmp = other.coeff(i,col);
         Scalar l_ii(0);
-        typename Lhs::InnerIterator it(lhs, i);
+        LhsIterator it(lhsEval, i);
         while(it && it.index()<i)
           ++it;
         if(!(Mode & UnitDiag))
@@ -86,10 +92,8 @@ struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,RowMajor>
           tmp -= it.value() * other.coeff(it.index(),col);
         }
 
-        if (Mode & UnitDiag)
-          other.coeffRef(i,col) = tmp;
-        else
-          other.coeffRef(i,col) = tmp/l_ii;
+        if (Mode & UnitDiag)  other.coeffRef(i,col) = tmp;
+        else                  other.coeffRef(i,col) = tmp/l_ii;
       }
     }
   }
@@ -100,16 +104,19 @@ template<typename Lhs, typename Rhs, int Mode>
 struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Lower,ColMajor>
 {
   typedef typename Rhs::Scalar Scalar;
+  typedef evaluator<Lhs> LhsEval;
+  typedef typename evaluator<Lhs>::InnerIterator LhsIterator;
   static void run(const Lhs& lhs, Rhs& other)
   {
-    for(int col=0 ; col<other.cols() ; ++col)
+    LhsEval lhsEval(lhs);
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
-      for(int i=0; i<lhs.cols(); ++i)
+      for(Index i=0; i<lhs.cols(); ++i)
       {
         Scalar& tmp = other.coeffRef(i,col);
         if (tmp!=Scalar(0)) // optimization when other is actually sparse
         {
-          typename Lhs::InnerIterator it(lhs, i);
+          LhsIterator it(lhsEval, i);
           while(it && it.index()<i)
             ++it;
           if(!(Mode & UnitDiag))
@@ -132,11 +139,14 @@ template<typename Lhs, typename Rhs, int Mode>
 struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,ColMajor>
 {
   typedef typename Rhs::Scalar Scalar;
+  typedef evaluator<Lhs> LhsEval;
+  typedef typename evaluator<Lhs>::InnerIterator LhsIterator;
   static void run(const Lhs& lhs, Rhs& other)
   {
-    for(int col=0 ; col<other.cols() ; ++col)
+    LhsEval lhsEval(lhs);
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
-      for(int i=lhs.cols()-1; i>=0; --i)
+      for(Index i=lhs.cols()-1; i>=0; --i)
       {
         Scalar& tmp = other.coeffRef(i,col);
         if (tmp!=Scalar(0)) // optimization when other is actually sparse
@@ -144,13 +154,13 @@ struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,ColMajor>
           if(!(Mode & UnitDiag))
           {
             // TODO replace this by a binary search. make sure the binary search is safe for partially sorted elements
-            typename Lhs::ReverseInnerIterator it(lhs, i);
+            LhsIterator it(lhsEval, i);
             while(it && it.index()!=i)
-              --it;
+              ++it;
             eigen_assert(it && it.index()==i);
             other.coeffRef(i,col) /= it.value();
           }
-          typename Lhs::InnerIterator it(lhs, i);
+          LhsIterator it(lhsEval, i);
           for(; it && it.index()<i; ++it)
             other.coeffRef(it.index(), col) -= tmp * it.value();
         }
@@ -161,11 +171,13 @@ struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,ColMajor>
 
 } // end namespace internal
 
-template<typename ExpressionType,int Mode>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+
+template<typename ExpressionType,unsigned int Mode>
 template<typename OtherDerived>
-void SparseTriangularView<ExpressionType,Mode>::solveInPlace(MatrixBase<OtherDerived>& other) const
+void TriangularViewImpl<ExpressionType,Mode,Sparse>::solveInPlace(MatrixBase<OtherDerived>& other) const
 {
-  eigen_assert(m_matrix.cols() == m_matrix.rows() && m_matrix.cols() == other.rows());
+  eigen_assert(derived().cols() == derived().rows() && derived().cols() == other.rows());
   eigen_assert((!(Mode & ZeroDiag)) && bool(Mode & (Upper|Lower)));
 
   enum { copy = internal::traits<OtherDerived>::Flags & RowMajorBit };
@@ -174,21 +186,12 @@ void SparseTriangularView<ExpressionType,Mode>::solveInPlace(MatrixBase<OtherDer
     typename internal::plain_matrix_type_column_major<OtherDerived>::type, OtherDerived&>::type OtherCopy;
   OtherCopy otherCopy(other.derived());
 
-  internal::sparse_solve_triangular_selector<ExpressionType, typename internal::remove_reference<OtherCopy>::type, Mode>::run(m_matrix, otherCopy);
+  internal::sparse_solve_triangular_selector<ExpressionType, typename internal::remove_reference<OtherCopy>::type, Mode>::run(derived().nestedExpression(), otherCopy);
 
   if (copy)
     other = otherCopy;
 }
-
-template<typename ExpressionType,int Mode>
-template<typename OtherDerived>
-typename internal::plain_matrix_type_column_major<OtherDerived>::type
-SparseTriangularView<ExpressionType,Mode>::solve(const MatrixBase<OtherDerived>& other) const
-{
-  typename internal::plain_matrix_type_column_major<OtherDerived>::type res(other);
-  solveInPlace(res);
-  return res;
-}
+#endif
 
 // pure sparse path
 
@@ -208,18 +211,18 @@ template<typename Lhs, typename Rhs, int Mode, int UpLo>
 struct sparse_solve_triangular_sparse_selector<Lhs,Rhs,Mode,UpLo,ColMajor>
 {
   typedef typename Rhs::Scalar Scalar;
-  typedef typename promote_index_type<typename traits<Lhs>::Index,
-                                         typename traits<Rhs>::Index>::type Index;
+  typedef typename promote_index_type<typename traits<Lhs>::StorageIndex,
+                                      typename traits<Rhs>::StorageIndex>::type StorageIndex;
   static void run(const Lhs& lhs, Rhs& other)
   {
     const bool IsLower = (UpLo==Lower);
-    AmbiVector<Scalar,Index> tempVector(other.rows()*2);
+    AmbiVector<Scalar,StorageIndex> tempVector(other.rows()*2);
     tempVector.setBounds(0,other.rows());
 
     Rhs res(other.rows(), other.cols());
     res.reserve(other.nonZeros());
 
-    for(int col=0 ; col<other.cols() ; ++col)
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
       // FIXME estimate number of non zeros
       tempVector.init(.99/*float(other.col(col).nonZeros())/float(other.rows())*/);
@@ -230,7 +233,7 @@ struct sparse_solve_triangular_sparse_selector<Lhs,Rhs,Mode,UpLo,ColMajor>
         tempVector.coeffRef(rhsIt.index()) = rhsIt.value();
       }
 
-      for(int i=IsLower?0:lhs.cols()-1;
+      for(Index i=IsLower?0:lhs.cols()-1;
           IsLower?i<lhs.cols():i>=0;
           i+=IsLower?1:-1)
       {
@@ -267,9 +270,9 @@ struct sparse_solve_triangular_sparse_selector<Lhs,Rhs,Mode,UpLo,ColMajor>
       }
 
 
-      int count = 0;
+      Index count = 0;
       // FIXME compute a reference value to filter zeros
-      for (typename AmbiVector<Scalar,Index>::Iterator it(tempVector/*,1e-12*/); it; ++it)
+      for (typename AmbiVector<Scalar,StorageIndex>::Iterator it(tempVector/*,1e-12*/); it; ++it)
       {
         ++ count;
 //         std::cerr << "fill " << it.index() << ", " << col << "\n";
@@ -286,11 +289,12 @@ struct sparse_solve_triangular_sparse_selector<Lhs,Rhs,Mode,UpLo,ColMajor>
 
 } // end namespace internal
 
-template<typename ExpressionType,int Mode>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename ExpressionType,unsigned int Mode>
 template<typename OtherDerived>
-void SparseTriangularView<ExpressionType,Mode>::solveInPlace(SparseMatrixBase<OtherDerived>& other) const
+void TriangularViewImpl<ExpressionType,Mode,Sparse>::solveInPlace(SparseMatrixBase<OtherDerived>& other) const
 {
-  eigen_assert(m_matrix.cols() == m_matrix.rows() && m_matrix.cols() == other.rows());
+  eigen_assert(derived().cols() == derived().rows() && derived().cols() == other.rows());
   eigen_assert( (!(Mode & ZeroDiag)) && bool(Mode & (Upper|Lower)));
 
 //   enum { copy = internal::traits<OtherDerived>::Flags & RowMajorBit };
@@ -299,35 +303,12 @@ void SparseTriangularView<ExpressionType,Mode>::solveInPlace(SparseMatrixBase<Ot
 //     typename internal::plain_matrix_type_column_major<OtherDerived>::type, OtherDerived&>::type OtherCopy;
 //   OtherCopy otherCopy(other.derived());
 
-  internal::sparse_solve_triangular_sparse_selector<ExpressionType, OtherDerived, Mode>::run(m_matrix, other.derived());
+  internal::sparse_solve_triangular_sparse_selector<ExpressionType, OtherDerived, Mode>::run(derived().nestedExpression(), other.derived());
 
 //   if (copy)
 //     other = otherCopy;
 }
-
-#ifdef EIGEN2_SUPPORT
-
-// deprecated stuff:
-
-/** \deprecated */
-template<typename Derived>
-template<typename OtherDerived>
-void SparseMatrixBase<Derived>::solveTriangularInPlace(MatrixBase<OtherDerived>& other) const
-{
-  this->template triangular<Flags&(Upper|Lower)>().solveInPlace(other);
-}
-
-/** \deprecated */
-template<typename Derived>
-template<typename OtherDerived>
-typename internal::plain_matrix_type_column_major<OtherDerived>::type
-SparseMatrixBase<Derived>::solveTriangular(const MatrixBase<OtherDerived>& other) const
-{
-  typename internal::plain_matrix_type_column_major<OtherDerived>::type res(other);
-  derived().solveTriangularInPlace(res);
-  return res;
-}
-#endif // EIGEN2_SUPPORT
+#endif
 
 } // end namespace Eigen
 
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU.h
index bdc4f193ddb..7104831c03b 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU.h
@@ -2,7 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
-// Copyright (C) 2012 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2012-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -14,7 +14,7 @@
 
 namespace Eigen {
 
-template <typename _MatrixType, typename _OrderingType = COLAMDOrdering<typename _MatrixType::Index> > class SparseLU;
+template <typename _MatrixType, typename _OrderingType = COLAMDOrdering<typename _MatrixType::StorageIndex> > class SparseLU;
 template <typename MappedSparseMatrixType> struct SparseLUMatrixLReturnType;
 template <typename MatrixLType, typename MatrixUType> struct SparseLUMatrixUReturnType;
 
@@ -64,33 +64,45 @@ template <typename MatrixLType, typename MatrixUType> struct SparseLUMatrixURetu
   * 
   * \tparam _MatrixType The type of the sparse matrix. It must be a column-major SparseMatrix<>
   * \tparam _OrderingType The ordering method to use, either AMD, COLAMD or METIS. Default is COLMAD
+  *
+  * \implsparsesolverconcept
   * 
-  * 
-  * \sa \ref TutorialSparseDirectSolvers
+  * \sa \ref TutorialSparseSolverConcept
   * \sa \ref OrderingMethods_Module
   */
 template <typename _MatrixType, typename _OrderingType>
-class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typename _MatrixType::Index>
+class SparseLU : public SparseSolverBase<SparseLU<_MatrixType,_OrderingType> >, public internal::SparseLUImpl<typename _MatrixType::Scalar, typename _MatrixType::StorageIndex>
 {
+  protected:
+    typedef SparseSolverBase<SparseLU<_MatrixType,_OrderingType> > APIBase;
+    using APIBase::m_isInitialized;
   public:
+    using APIBase::_solve_impl;
+    
     typedef _MatrixType MatrixType; 
     typedef _OrderingType OrderingType;
     typedef typename MatrixType::Scalar Scalar; 
     typedef typename MatrixType::RealScalar RealScalar; 
-    typedef typename MatrixType::Index Index; 
-    typedef SparseMatrix<Scalar,ColMajor,Index> NCMatrix;
-    typedef internal::MappedSuperNodalMatrix<Scalar, Index> SCMatrix; 
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> NCMatrix;
+    typedef internal::MappedSuperNodalMatrix<Scalar, StorageIndex> SCMatrix;
     typedef Matrix<Scalar,Dynamic,1> ScalarVector;
-    typedef Matrix<Index,Dynamic,1> IndexVector;
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType;
-    typedef internal::SparseLUImpl<Scalar, Index> Base;
+    typedef Matrix<StorageIndex,Dynamic,1> IndexVector;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
+    typedef internal::SparseLUImpl<Scalar, StorageIndex> Base;
+
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
     
   public:
-    SparseLU():m_isInitialized(true),m_lastError(""),m_Ustore(0,0,0,0,0,0),m_symmetricmode(false),m_diagpivotthresh(1.0),m_detPermR(1)
+    SparseLU():m_lastError(""),m_Ustore(0,0,0,0,0,0),m_symmetricmode(false),m_diagpivotthresh(1.0),m_detPermR(1)
     {
       initperfvalues(); 
     }
-    SparseLU(const MatrixType& matrix):m_isInitialized(true),m_lastError(""),m_Ustore(0,0,0,0,0,0),m_symmetricmode(false),m_diagpivotthresh(1.0),m_detPermR(1)
+    explicit SparseLU(const MatrixType& matrix)
+      : m_lastError(""),m_Ustore(0,0,0,0,0,0),m_symmetricmode(false),m_diagpivotthresh(1.0),m_detPermR(1)
     {
       initperfvalues(); 
       compute(matrix);
@@ -141,9 +153,9 @@ class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typ
       * y = b; matrixU().solveInPlace(y);
       * \endcode
       */
-    SparseLUMatrixUReturnType<SCMatrix,MappedSparseMatrix<Scalar,ColMajor,Index> > matrixU() const
+    SparseLUMatrixUReturnType<SCMatrix,MappedSparseMatrix<Scalar,ColMajor,StorageIndex> > matrixU() const
     {
-      return SparseLUMatrixUReturnType<SCMatrix, MappedSparseMatrix<Scalar,ColMajor,Index> >(m_Lstore, m_Ustore);
+      return SparseLUMatrixUReturnType<SCMatrix, MappedSparseMatrix<Scalar,ColMajor,StorageIndex> >(m_Lstore, m_Ustore);
     }
 
     /**
@@ -168,6 +180,7 @@ class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typ
       m_diagpivotthresh = thresh; 
     }
 
+#ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \returns the solution X of \f$ A X = B \f$ using the current decomposition of A.
       *
       * \warning the destination matrix X in X = this->solve(B) must be colmun-major.
@@ -175,26 +188,8 @@ class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typ
       * \sa compute()
       */
     template<typename Rhs>
-    inline const internal::solve_retval<SparseLU, Rhs> solve(const MatrixBase<Rhs>& B) const 
-    {
-      eigen_assert(m_factorizationIsOk && "SparseLU is not initialized."); 
-      eigen_assert(rows()==B.rows()
-                    && "SparseLU::solve(): invalid number of rows of the right hand side matrix B");
-          return internal::solve_retval<SparseLU, Rhs>(*this, B.derived());
-    }
-
-    /** \returns the solution X of \f$ A X = B \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<SparseLU, Rhs> solve(const SparseMatrixBase<Rhs>& B) const 
-    {
-      eigen_assert(m_factorizationIsOk && "SparseLU is not initialized."); 
-      eigen_assert(rows()==B.rows()
-                    && "SparseLU::solve(): invalid number of rows of the right hand side matrix B");
-          return internal::sparse_solve_retval<SparseLU, Rhs>(*this, B.derived());
-    }
+    inline const Solve<SparseLU, Rhs> solve(const MatrixBase<Rhs>& B) const;
+#endif // EIGEN_PARSED_BY_DOXYGEN
     
     /** \brief Reports whether previous computation was successful.
       *
@@ -219,7 +214,7 @@ class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typ
     }
 
     template<typename Rhs, typename Dest>
-    bool _solve(const MatrixBase<Rhs> &B, MatrixBase<Dest> &X_base) const
+    bool _solve_impl(const MatrixBase<Rhs> &B, MatrixBase<Dest> &X_base) const
     {
       Dest& X(X_base.derived());
       eigen_assert(m_factorizationIsOk && "The matrix should be factorized first");
@@ -255,8 +250,9 @@ class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typ
       *
       * \sa logAbsDeterminant(), signDeterminant()
       */
-     Scalar absDeterminant()
+    Scalar absDeterminant()
     {
+      using std::abs;
       eigen_assert(m_factorizationIsOk && "The matrix should be factorized first.");
       // Initialize with the determinant of the row matrix
       Scalar det = Scalar(1.);
@@ -268,42 +264,43 @@ class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typ
         {
           if(it.index() == j)
           {
-            using std::abs;
             det *= abs(it.value());
             break;
           }
         }
-       }
-       return det;
-     }
+      }
+      return det;
+    }
+
+    /** \returns the natural log of the absolute value of the determinant of the matrix
+      * of which **this is the QR decomposition
+      *
+      * \note This method is useful to work around the risk of overflow/underflow that's
+      * inherent to the determinant computation.
+      *
+      * \sa absDeterminant(), signDeterminant()
+      */
+    Scalar logAbsDeterminant() const
+    {
+      using std::log;
+      using std::abs;
 
-     /** \returns the natural log of the absolute value of the determinant of the matrix
-       * of which **this is the QR decomposition
-       *
-       * \note This method is useful to work around the risk of overflow/underflow that's
-       * inherent to the determinant computation.
-       *
-       * \sa absDeterminant(), signDeterminant()
-       */
-     Scalar logAbsDeterminant() const
-     {
-       eigen_assert(m_factorizationIsOk && "The matrix should be factorized first.");
-       Scalar det = Scalar(0.);
-       for (Index j = 0; j < this->cols(); ++j)
-       {
-         for (typename SCMatrix::InnerIterator it(m_Lstore, j); it; ++it)
-         {
-           if(it.row() < j) continue;
-           if(it.row() == j)
-           {
-             using std::log; using std::abs;
-             det += log(abs(it.value()));
-             break;
-           }
-         }
-       }
-       return det;
-     }
+      eigen_assert(m_factorizationIsOk && "The matrix should be factorized first.");
+      Scalar det = Scalar(0.);
+      for (Index j = 0; j < this->cols(); ++j)
+      {
+        for (typename SCMatrix::InnerIterator it(m_Lstore, j); it; ++it)
+        {
+          if(it.row() < j) continue;
+          if(it.row() == j)
+          {
+            det += log(abs(it.value()));
+            break;
+          }
+        }
+      }
+      return det;
+    }
 
     /** \returns A number representing the sign of the determinant
       *
@@ -355,7 +352,7 @@ class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typ
           }
         }
       }
-      return det * Scalar(m_detPermR * m_detPermC);
+      return (m_detPermR * m_detPermC) > 0 ? det : -det;
     }
 
   protected:
@@ -372,13 +369,12 @@ class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typ
       
     // Variables 
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     bool m_factorizationIsOk;
     bool m_analysisIsOk;
     std::string m_lastError;
     NCMatrix m_mat; // The input (permuted ) matrix 
     SCMatrix m_Lstore; // The lower triangular matrix (supernodal)
-    MappedSparseMatrix<Scalar,ColMajor,Index> m_Ustore; // The upper triangular matrix
+    MappedSparseMatrix<Scalar,ColMajor,StorageIndex> m_Ustore; // The upper triangular matrix
     PermutationType m_perm_c; // Column permutation 
     PermutationType m_perm_r ; // Row permutation
     IndexVector m_etree; // Column elimination tree 
@@ -388,7 +384,7 @@ class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typ
     // SparseLU options 
     bool m_symmetricmode;
     // values for performance 
-    internal::perfvalues<Index> m_perfv; 
+    internal::perfvalues m_perfv;
     RealScalar m_diagpivotthresh; // Specifies the threshold used for a diagonal entry to be an acceptable pivot
     Index m_nnzL, m_nnzU; // Nonzeros in L and U factors
     Index m_detPermR, m_detPermC; // Determinants of the permutation matrices
@@ -417,30 +413,32 @@ void SparseLU<MatrixType, OrderingType>::analyzePattern(const MatrixType& mat)
   
   //TODO  It is possible as in SuperLU to compute row and columns scaling vectors to equilibrate the matrix mat.
   
+  // Firstly, copy the whole input matrix. 
+  m_mat = mat;
+  
+  // Compute fill-in ordering
   OrderingType ord; 
-  ord(mat,m_perm_c);
+  ord(m_mat,m_perm_c);
   
   // Apply the permutation to the column of the input  matrix
-  //First copy the whole input matrix. 
-  m_mat = mat;
-  if (m_perm_c.size()) {
+  if (m_perm_c.size())
+  {
     m_mat.uncompress(); //NOTE: The effect of this command is only to create the InnerNonzeros pointers. FIXME : This vector is filled but not subsequently used.  
-    //Then, permute only the column pointers
-    const Index * outerIndexPtr;
-    if (mat.isCompressed()) outerIndexPtr = mat.outerIndexPtr();
-    else
-    {
-      Index *outerIndexPtr_t = new Index[mat.cols()+1];
-      for(Index i = 0; i <= mat.cols(); i++) outerIndexPtr_t[i] = m_mat.outerIndexPtr()[i];
-      outerIndexPtr = outerIndexPtr_t;
-    }
+    // Then, permute only the column pointers
+    ei_declare_aligned_stack_constructed_variable(StorageIndex,outerIndexPtr,mat.cols()+1,mat.isCompressed()?const_cast<StorageIndex*>(mat.outerIndexPtr()):0);
+    
+    // If the input matrix 'mat' is uncompressed, then the outer-indices do not match the ones of m_mat, and a copy is thus needed.
+    if(!mat.isCompressed()) 
+      IndexVector::Map(outerIndexPtr, mat.cols()+1) = IndexVector::Map(m_mat.outerIndexPtr(),mat.cols()+1);
+    
+    // Apply the permutation and compute the nnz per column.
     for (Index i = 0; i < mat.cols(); i++)
     {
       m_mat.outerIndexPtr()[m_perm_c.indices()(i)] = outerIndexPtr[i];
       m_mat.innerNonZeroPtr()[m_perm_c.indices()(i)] = outerIndexPtr[i+1] - outerIndexPtr[i];
     }
-    if(!mat.isCompressed()) delete[] outerIndexPtr;
   }
+  
   // Compute the column elimination tree of the permuted matrix 
   IndexVector firstRowElt;
   internal::coletree(m_mat, m_etree,firstRowElt); 
@@ -449,7 +447,7 @@ void SparseLU<MatrixType, OrderingType>::analyzePattern(const MatrixType& mat)
   if (!m_symmetricmode) {
     IndexVector post, iwork; 
     // Post order etree
-    internal::treePostorder(m_mat.cols(), m_etree, post); 
+    internal::treePostorder(StorageIndex(m_mat.cols()), m_etree, post); 
       
    
     // Renumber etree in postorder 
@@ -501,7 +499,7 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
   eigen_assert(m_analysisIsOk && "analyzePattern() should be called first"); 
   eigen_assert((matrix.rows() == matrix.cols()) && "Only for squared matrices");
   
-  typedef typename IndexVector::Scalar Index; 
+  m_isInitialized = true;
   
   
   // Apply the column permutation computed in analyzepattern()
@@ -511,11 +509,11 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
   {
     m_mat.uncompress(); //NOTE: The effect of this command is only to create the InnerNonzeros pointers.
     //Then, permute only the column pointers
-    const Index * outerIndexPtr;
+    const StorageIndex * outerIndexPtr;
     if (matrix.isCompressed()) outerIndexPtr = matrix.outerIndexPtr();
     else
     {
-      Index* outerIndexPtr_t = new Index[matrix.cols()+1];
+      StorageIndex* outerIndexPtr_t = new StorageIndex[matrix.cols()+1];
       for(Index i = 0; i <= matrix.cols(); i++) outerIndexPtr_t[i] = m_mat.outerIndexPtr()[i];
       outerIndexPtr = outerIndexPtr_t;
     }
@@ -529,7 +527,7 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
   else 
   { //FIXME This should not be needed if the empty permutation is handled transparently
     m_perm_c.resize(matrix.cols());
-    for(Index i = 0; i < matrix.cols(); ++i) m_perm_c.indices()(i) = i;
+    for(StorageIndex i = 0; i < matrix.cols(); ++i) m_perm_c.indices()(i) = i;
   }
   
   Index m = m_mat.rows();
@@ -694,7 +692,7 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
   // Create supernode matrix L 
   m_Lstore.setInfos(m, n, m_glu.lusup, m_glu.xlusup, m_glu.lsub, m_glu.xlsub, m_glu.supno, m_glu.xsup); 
   // Create the column major upper sparse matrix  U; 
-  new (&m_Ustore) MappedSparseMatrix<Scalar, ColMajor, Index> ( m, n, m_nnzU, m_glu.xusub.data(), m_glu.usub.data(), m_glu.ucol.data() ); 
+  new (&m_Ustore) MappedSparseMatrix<Scalar, ColMajor, StorageIndex> ( m, n, m_nnzU, m_glu.xusub.data(), m_glu.usub.data(), m_glu.ucol.data() );
   
   m_info = Success;
   m_factorizationIsOk = true;
@@ -703,9 +701,8 @@ void SparseLU<MatrixType, OrderingType>::factorize(const MatrixType& matrix)
 template<typename MappedSupernodalType>
 struct SparseLUMatrixLReturnType : internal::no_assignment_operator
 {
-  typedef typename MappedSupernodalType::Index Index;
   typedef typename MappedSupernodalType::Scalar Scalar;
-  SparseLUMatrixLReturnType(const MappedSupernodalType& mapL) : m_mapL(mapL)
+  explicit SparseLUMatrixLReturnType(const MappedSupernodalType& mapL) : m_mapL(mapL)
   { }
   Index rows() { return m_mapL.rows(); }
   Index cols() { return m_mapL.cols(); }
@@ -720,7 +717,6 @@ struct SparseLUMatrixLReturnType : internal::no_assignment_operator
 template<typename MatrixLType, typename MatrixUType>
 struct SparseLUMatrixUReturnType : internal::no_assignment_operator
 {
-  typedef typename MatrixLType::Index Index;
   typedef typename MatrixLType::Scalar Scalar;
   SparseLUMatrixUReturnType(const MatrixLType& mapL, const MatrixUType& mapU)
   : m_mapL(mapL),m_mapU(mapU)
@@ -731,7 +727,7 @@ struct SparseLUMatrixUReturnType : internal::no_assignment_operator
   template<typename Dest>   void solveInPlace(MatrixBase<Dest> &X) const
   {
     Index nrhs = X.cols();
-    Index n = X.rows();
+    Index n    = X.rows();
     // Backward solve with U
     for (Index k = m_mapL.nsuper(); k >= 0; k--)
     {
@@ -750,7 +746,7 @@ struct SparseLUMatrixUReturnType : internal::no_assignment_operator
       else
       {
         Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > A( &(m_mapL.valuePtr()[luptr]), nsupc, nsupc, OuterStride<>(lda) );
-        Map< Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
+        Map< Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
         U = A.template triangularView<Upper>().solve(U);
       }
 
@@ -772,35 +768,6 @@ struct SparseLUMatrixUReturnType : internal::no_assignment_operator
   const MatrixUType& m_mapU;
 };
 
-namespace internal {
-  
-template<typename _MatrixType, typename Derived, typename Rhs>
-struct solve_retval<SparseLU<_MatrixType,Derived>, Rhs>
-  : solve_retval_base<SparseLU<_MatrixType,Derived>, Rhs>
-{
-  typedef SparseLU<_MatrixType,Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, typename Derived, typename Rhs>
-struct sparse_solve_retval<SparseLU<_MatrixType,Derived>, Rhs>
-  : sparse_solve_retval_base<SparseLU<_MatrixType,Derived>, Rhs>
-{
-  typedef SparseLU<_MatrixType,Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-} // end namespace internal
-
 } // End namespace Eigen 
 
 #endif
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLUImpl.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLUImpl.h
index 99d651e40d3..fc0cfc4de1a 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLUImpl.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLUImpl.h
@@ -16,19 +16,19 @@ namespace internal {
   * \class SparseLUImpl
   * Base class for sparseLU
   */
-template <typename Scalar, typename Index>
+template <typename Scalar, typename StorageIndex>
 class SparseLUImpl
 {
   public:
     typedef Matrix<Scalar,Dynamic,1> ScalarVector;
+    typedef Matrix<StorageIndex,Dynamic,1> IndexVector; 
     typedef Matrix<Scalar,Dynamic,Dynamic,ColMajor> ScalarMatrix;
     typedef Map<ScalarMatrix, 0,  OuterStride<> > MappedMatrixBlock;
-    typedef Matrix<Index,Dynamic,1> IndexVector; 
     typedef typename ScalarVector::RealScalar RealScalar; 
     typedef Ref<Matrix<Scalar,Dynamic,1> > BlockScalarVector;
-    typedef Ref<Matrix<Index,Dynamic,1> > BlockIndexVector;
+    typedef Ref<Matrix<StorageIndex,Dynamic,1> > BlockIndexVector;
     typedef LU_GlobalLU_t<IndexVector, ScalarVector> GlobalLU_t; 
-    typedef SparseMatrix<Scalar,ColMajor,Index> MatrixType; 
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> MatrixType; 
     
   protected:
      template <typename VectorType>
@@ -42,7 +42,7 @@ class SparseLUImpl
      Index snode_bmod (const Index jcol, const Index fsupc, ScalarVector& dense, GlobalLU_t& glu);
      Index pivotL(const Index jcol, const RealScalar& diagpivotthresh, IndexVector& perm_r, IndexVector& iperm_c, Index& pivrow, GlobalLU_t& glu);
      template <typename Traits>
-     void dfs_kernel(const Index jj, IndexVector& perm_r,
+     void dfs_kernel(const StorageIndex jj, IndexVector& perm_r,
                     Index& nseg, IndexVector& panel_lsub, IndexVector& segrep,
                     Ref<IndexVector> repfnz_col, IndexVector& xprune, Ref<IndexVector> marker, IndexVector& parent,
                     IndexVector& xplore, GlobalLU_t& glu, Index& nextl_col, Index krow, Traits& traits);
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Memory.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Memory.h
index 45f96d16a8e..4dc42e87ba8 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Memory.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Memory.h
@@ -36,13 +36,12 @@ namespace internal {
   
 enum { LUNoMarker = 3 };
 enum {emptyIdxLU = -1};
-template<typename Index>
 inline Index LUnumTempV(Index& m, Index& w, Index& t, Index& b)
 {
   return (std::max)(m, (t+b)*w);
 }
 
-template< typename Scalar, typename Index>
+template< typename Scalar>
 inline Index LUTempSpace(Index&m, Index& w)
 {
   return (2*w + 4 + LUNoMarker) * m * sizeof(Index) + (w + 1) * m * sizeof(Scalar);
@@ -59,9 +58,9 @@ inline Index LUTempSpace(Index&m, Index& w)
   * \param keep_prev  1: use length  and do not expand the vector; 0: compute new_len and expand
   * \param[in,out] num_expansions Number of times the memory has been expanded
   */
-template <typename Scalar, typename Index>
+template <typename Scalar, typename StorageIndex>
 template <typename VectorType>
-Index  SparseLUImpl<Scalar,Index>::expand(VectorType& vec, Index& length, Index nbElts, Index keep_prev, Index& num_expansions) 
+Index  SparseLUImpl<Scalar,StorageIndex>::expand(VectorType& vec, Index& length, Index nbElts, Index keep_prev, Index& num_expansions) 
 {
   
   float alpha = 1.5; // Ratio of the memory increase 
@@ -148,8 +147,8 @@ Index  SparseLUImpl<Scalar,Index>::expand(VectorType& vec, Index& length, Index
  * \return an estimated size of the required memory if lwork = -1; otherwise, return the size of actually allocated memory when allocation failed, and 0 on success
  * \note Unlike SuperLU, this routine does not support successive factorization with the same pattern and the same row permutation
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::memInit(Index m, Index n, Index annz, Index lwork, Index fillratio, Index panel_size,  GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::memInit(Index m, Index n, Index annz, Index lwork, Index fillratio, Index panel_size,  GlobalLU_t& glu)
 {
   Index& num_expansions = glu.num_expansions; //No memory expansions so far
   num_expansions = 0;
@@ -205,9 +204,9 @@ Index SparseLUImpl<Scalar,Index>::memInit(Index m, Index n, Index annz, Index lw
  * \param num_expansions Number of expansions 
  * \return 0 on success, > 0 size of the memory allocated so far
  */
-template <typename Scalar, typename Index>
+template <typename Scalar, typename StorageIndex>
 template <typename VectorType>
-Index SparseLUImpl<Scalar,Index>::memXpand(VectorType& vec, Index& maxlen, Index nbElts, MemType memtype, Index& num_expansions)
+Index SparseLUImpl<Scalar,StorageIndex>::memXpand(VectorType& vec, Index& maxlen, Index nbElts, MemType memtype, Index& num_expansions)
 {
   Index failed_size; 
   if (memtype == USUB)
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Structs.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Structs.h
index 24d6bf17946..cf5ec449bec 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Structs.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Structs.h
@@ -75,7 +75,7 @@ typedef enum {LUSUP, UCOL, LSUB, USUB, LLVL, ULVL} MemType;
 
 template <typename IndexVector, typename ScalarVector>
 struct LU_GlobalLU_t {
-  typedef typename IndexVector::Scalar Index; 
+  typedef typename IndexVector::Scalar StorageIndex; 
   IndexVector xsup; //First supernode column ... xsup(s) points to the beginning of the s-th supernode
   IndexVector supno; // Supernode number corresponding to this column (column to supernode mapping)
   ScalarVector  lusup; // nonzero values of L ordered by columns 
@@ -93,7 +93,6 @@ struct LU_GlobalLU_t {
 };
 
 // Values to set for performance
-template <typename Index>
 struct perfvalues {
   Index panel_size; // a panel consists of at most <panel_size> consecutive columns
   Index relax; // To control degree of relaxing supernodes. If the number of nodes (columns) 
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h
index 54a56940861..721e1883ba8 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h
@@ -29,20 +29,20 @@ namespace internal {
  * SuperInnerIterator to iterate through all supernodes 
  * Function for triangular solve
  */
-template <typename _Scalar, typename _Index>
+template <typename _Scalar, typename _StorageIndex>
 class MappedSuperNodalMatrix
 {
   public:
     typedef _Scalar Scalar; 
-    typedef _Index Index;
-    typedef Matrix<Index,Dynamic,1> IndexVector; 
+    typedef _StorageIndex StorageIndex;
+    typedef Matrix<StorageIndex,Dynamic,1> IndexVector;
     typedef Matrix<Scalar,Dynamic,1> ScalarVector;
   public:
     MappedSuperNodalMatrix()
     {
       
     }
-    MappedSuperNodalMatrix(Index m, Index n,  ScalarVector& nzval, IndexVector& nzval_colptr, IndexVector& rowind, 
+    MappedSuperNodalMatrix(Index m, Index n,  ScalarVector& nzval, IndexVector& nzval_colptr, IndexVector& rowind,
              IndexVector& rowind_colptr, IndexVector& col_to_sup, IndexVector& sup_to_col )
     {
       setInfos(m, n, nzval, nzval_colptr, rowind, rowind_colptr, col_to_sup, sup_to_col);
@@ -58,7 +58,7 @@ class MappedSuperNodalMatrix
      * FIXME This class will be modified such that it can be use in the course 
      * of the factorization.
      */
-    void setInfos(Index m, Index n, ScalarVector& nzval, IndexVector& nzval_colptr, IndexVector& rowind, 
+    void setInfos(Index m, Index n, ScalarVector& nzval, IndexVector& nzval_colptr, IndexVector& rowind,
              IndexVector& rowind_colptr, IndexVector& col_to_sup, IndexVector& sup_to_col )
     {
       m_row = m;
@@ -96,12 +96,12 @@ class MappedSuperNodalMatrix
     /**
      * Return the pointers to the beginning of each column in \ref valuePtr()
      */
-    Index* colIndexPtr()
+    StorageIndex* colIndexPtr()
     {
       return m_nzval_colptr; 
     }
     
-    const Index* colIndexPtr() const
+    const StorageIndex* colIndexPtr() const
     {
       return m_nzval_colptr; 
     }
@@ -109,9 +109,9 @@ class MappedSuperNodalMatrix
     /**
      * Return the array of compressed row indices of all supernodes
      */
-    Index* rowIndex()  { return m_rowind; }
+    StorageIndex* rowIndex()  { return m_rowind; }
     
-    const Index* rowIndex() const
+    const StorageIndex* rowIndex() const
     {
       return m_rowind; 
     }
@@ -119,9 +119,9 @@ class MappedSuperNodalMatrix
     /**
      * Return the location in \em rowvaluePtr() which starts each column
      */
-    Index* rowIndexPtr() { return m_rowind_colptr; }
+    StorageIndex* rowIndexPtr() { return m_rowind_colptr; }
     
-    const Index* rowIndexPtr() const 
+    const StorageIndex* rowIndexPtr() const
     {
       return m_rowind_colptr; 
     }
@@ -129,18 +129,18 @@ class MappedSuperNodalMatrix
     /** 
      * Return the array of column-to-supernode mapping 
      */
-    Index* colToSup()  { return m_col_to_sup; }
+    StorageIndex* colToSup()  { return m_col_to_sup; }
     
-    const Index* colToSup() const
+    const StorageIndex* colToSup() const
     {
       return m_col_to_sup;       
     }
     /**
      * Return the array of supernode-to-column mapping
      */
-    Index* supToCol() { return m_sup_to_col; }
+    StorageIndex* supToCol() { return m_sup_to_col; }
     
-    const Index* supToCol() const 
+    const StorageIndex* supToCol() const
     {
       return m_sup_to_col;
     }
@@ -148,7 +148,7 @@ class MappedSuperNodalMatrix
     /**
      * Return the number of supernodes
      */
-    Index nsuper() const 
+    Index nsuper() const
     {
       return m_nsuper; 
     }
@@ -162,14 +162,14 @@ class MappedSuperNodalMatrix
     
   protected:
     Index m_row; // Number of rows
-    Index m_col; // Number of columns 
-    Index m_nsuper; // Number of supernodes 
+    Index m_col; // Number of columns
+    Index m_nsuper; // Number of supernodes
     Scalar* m_nzval; //array of nonzero values packed by column
-    Index* m_nzval_colptr; //nzval_colptr[j] Stores the location in nzval[] which starts column j 
-    Index* m_rowind; // Array of compressed row indices of rectangular supernodes
-    Index* m_rowind_colptr; //rowind_colptr[j] stores the location in rowind[] which starts column j
-    Index* m_col_to_sup; // col_to_sup[j] is the supernode number to which column j belongs
-    Index* m_sup_to_col; //sup_to_col[s] points to the starting column of the s-th supernode
+    StorageIndex* m_nzval_colptr; //nzval_colptr[j] Stores the location in nzval[] which starts column j
+    StorageIndex* m_rowind; // Array of compressed row indices of rectangular supernodes
+    StorageIndex* m_rowind_colptr; //rowind_colptr[j] stores the location in rowind[] which starts column j
+    StorageIndex* m_col_to_sup; // col_to_sup[j] is the supernode number to which column j belongs
+    StorageIndex* m_sup_to_col; //sup_to_col[s] points to the starting column of the s-th supernode
     
   private :
 };
@@ -178,13 +178,13 @@ class MappedSuperNodalMatrix
   * \brief InnerIterator class to iterate over nonzero values of the current column in the supernodal matrix L
   * 
   */
-template<typename Scalar, typename Index>
-class MappedSuperNodalMatrix<Scalar,Index>::InnerIterator
+template<typename Scalar, typename StorageIndex>
+class MappedSuperNodalMatrix<Scalar,StorageIndex>::InnerIterator
 {
   public:
      InnerIterator(const MappedSuperNodalMatrix& mat, Index outer)
       : m_matrix(mat),
-        m_outer(outer), 
+        m_outer(outer),
         m_supno(mat.colToSup()[outer]),
         m_idval(mat.colIndexPtr()[outer]),
         m_startidval(m_idval),
@@ -229,14 +229,17 @@ class MappedSuperNodalMatrix<Scalar,Index>::InnerIterator
  * \brief Solve with the supernode triangular matrix
  * 
  */
-template<typename Scalar, typename Index>
+template<typename Scalar, typename Index_>
 template<typename Dest>
-void MappedSuperNodalMatrix<Scalar,Index>::solveInPlace( MatrixBase<Dest>&X) const
+void MappedSuperNodalMatrix<Scalar,Index_>::solveInPlace( MatrixBase<Dest>&X) const
 {
-    Index n = X.rows(); 
-    Index nrhs = X.cols(); 
+    /* Explicit type conversion as the Index type of MatrixBase<Dest> may be wider than Index */
+//    eigen_assert(X.rows() <= NumTraits<Index>::highest());
+//    eigen_assert(X.cols() <= NumTraits<Index>::highest());
+    Index n    = int(X.rows());
+    Index nrhs = Index(X.cols());
     const Scalar * Lval = valuePtr();                 // Nonzero values 
-    Matrix<Scalar,Dynamic,Dynamic, ColMajor> work(n, nrhs);     // working vector
+    Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor> work(n, nrhs);     // working vector
     work.setZero();
     for (Index k = 0; k <= nsuper(); k ++)
     {
@@ -268,12 +271,12 @@ void MappedSuperNodalMatrix<Scalar,Index>::solveInPlace( MatrixBase<Dest>&X) con
         
         // Triangular solve 
         Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > A( &(Lval[luptr]), nsupc, nsupc, OuterStride<>(lda) );
-        Map< Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) ); 
+        Map< Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
         U = A.template triangularView<UnitLower>().solve(U); 
         
         // Matrix-vector product 
         new (&A) Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > ( &(Lval[luptr+nsupc]), nrow, nsupc, OuterStride<>(lda) );
-        work.block(0, 0, nrow, nrhs) = A * U; 
+        work.topRows(nrow).noalias() = A * U;
         
         //Begin Scatter 
         for (Index j = 0; j < nrhs; j++)
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Utils.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Utils.h
index 15352ac33ae..9e3dab44d99 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Utils.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_Utils.h
@@ -17,8 +17,8 @@ namespace internal {
 /**
  * \brief Count Nonzero elements in the factors
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::countnz(const Index n, Index& nnzL, Index& nnzU, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::countnz(const Index n, Index& nnzL, Index& nnzU, GlobalLU_t& glu)
 {
  nnzL = 0; 
  nnzU = (glu.xusub)(n); 
@@ -48,12 +48,12 @@ void SparseLUImpl<Scalar,Index>::countnz(const Index n, Index& nnzL, Index& nnzU
  * and applies permutation to the remaining subscripts
  * 
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::fixupL(const Index n, const IndexVector& perm_r, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::fixupL(const Index n, const IndexVector& perm_r, GlobalLU_t& glu)
 {
   Index fsupc, i, j, k, jstart; 
   
-  Index nextl = 0; 
+  StorageIndex nextl = 0; 
   Index nsuper = (glu.supno)(n); 
   
   // For each supernode 
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_column_bmod.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_column_bmod.h
index cacc7e98712..b57f06802e2 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_column_bmod.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_column_bmod.h
@@ -49,8 +49,9 @@ namespace internal {
  *         > 0 - number of bytes allocated when run out of space
  * 
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::column_bmod(const Index jcol, const Index nseg, BlockScalarVector dense, ScalarVector& tempv, BlockIndexVector segrep, BlockIndexVector repfnz, Index fpanelc, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::column_bmod(const Index jcol, const Index nseg, BlockScalarVector dense, ScalarVector& tempv,
+                                                     BlockIndexVector segrep, BlockIndexVector repfnz, Index fpanelc, GlobalLU_t& glu)
 {
   Index  jsupno, k, ksub, krep, ksupno; 
   Index lptr, nrow, isub, irow, nextlu, new_next, ufirst; 
@@ -137,7 +138,7 @@ Index SparseLUImpl<Scalar,Index>::column_bmod(const Index jcol, const Index nseg
     glu.lusup.segment(nextlu,offset).setZero();
     nextlu += offset;
   }
-  glu.xlusup(jcol + 1) = nextlu;  // close L\U(*,jcol); 
+  glu.xlusup(jcol + 1) = StorageIndex(nextlu);  // close L\U(*,jcol); 
   
   /* For more updates within the panel (also within the current supernode),
    * should start from the first column of the panel, or the first column
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_column_dfs.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_column_dfs.h
index 4c04b0e44e9..c98b30e323f 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_column_dfs.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_column_dfs.h
@@ -30,7 +30,7 @@
 #ifndef SPARSELU_COLUMN_DFS_H
 #define SPARSELU_COLUMN_DFS_H
 
-template <typename Scalar, typename Index> class SparseLUImpl;
+template <typename Scalar, typename StorageIndex> class SparseLUImpl;
 namespace Eigen {
 
 namespace internal {
@@ -39,8 +39,8 @@ template<typename IndexVector, typename ScalarVector>
 struct column_dfs_traits : no_assignment_operator
 {
   typedef typename ScalarVector::Scalar Scalar;
-  typedef typename IndexVector::Scalar Index;
-  column_dfs_traits(Index jcol, Index& jsuper, typename SparseLUImpl<Scalar, Index>::GlobalLU_t& glu, SparseLUImpl<Scalar, Index>& luImpl)
+  typedef typename IndexVector::Scalar StorageIndex;
+  column_dfs_traits(Index jcol, Index& jsuper, typename SparseLUImpl<Scalar, StorageIndex>::GlobalLU_t& glu, SparseLUImpl<Scalar, StorageIndex>& luImpl)
    : m_jcol(jcol), m_jsuper_ref(jsuper), m_glu(glu), m_luImpl(luImpl)
  {}
   bool update_segrep(Index /*krep*/, Index /*jj*/)
@@ -57,8 +57,8 @@ struct column_dfs_traits : no_assignment_operator
   
   Index m_jcol;
   Index& m_jsuper_ref;
-  typename SparseLUImpl<Scalar, Index>::GlobalLU_t& m_glu;
-  SparseLUImpl<Scalar, Index>& m_luImpl;
+  typename SparseLUImpl<Scalar, StorageIndex>::GlobalLU_t& m_glu;
+  SparseLUImpl<Scalar, StorageIndex>& m_luImpl;
 };
 
 
@@ -89,8 +89,10 @@ struct column_dfs_traits : no_assignment_operator
  *         > 0 number of bytes allocated when run out of space
  * 
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::column_dfs(const Index m, const Index jcol, IndexVector& perm_r, Index maxsuper, Index& nseg,  BlockIndexVector lsub_col, IndexVector& segrep, BlockIndexVector repfnz, IndexVector& xprune, IndexVector& marker, IndexVector& parent, IndexVector& xplore, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::column_dfs(const Index m, const Index jcol, IndexVector& perm_r, Index maxsuper, Index& nseg,
+                                                    BlockIndexVector lsub_col, IndexVector& segrep, BlockIndexVector repfnz, IndexVector& xprune,
+                                                    IndexVector& marker, IndexVector& parent, IndexVector& xplore, GlobalLU_t& glu)
 {
   
   Index jsuper = glu.supno(jcol); 
@@ -110,13 +112,13 @@ Index SparseLUImpl<Scalar,Index>::column_dfs(const Index m, const Index jcol, In
     // krow was visited before, go to the next nonz; 
     if (kmark == jcol) continue;
     
-    dfs_kernel(jcol, perm_r, nseg, glu.lsub, segrep, repfnz, xprune, marker2, parent,
+    dfs_kernel(StorageIndex(jcol), perm_r, nseg, glu.lsub, segrep, repfnz, xprune, marker2, parent,
                    xplore, glu, nextl, krow, traits);
   } // for each nonzero ... 
   
-  Index fsupc, jptr, jm1ptr, ito, ifrom, istop;
-  Index nsuper = glu.supno(jcol);
-  Index jcolp1 = jcol + 1;
+  Index fsupc;
+  StorageIndex nsuper = glu.supno(jcol);
+  StorageIndex jcolp1 = StorageIndex(jcol) + 1;
   Index jcolm1 = jcol - 1;
   
   // check to see if j belongs in the same supernode as j-1
@@ -127,8 +129,8 @@ Index SparseLUImpl<Scalar,Index>::column_dfs(const Index m, const Index jcol, In
   else 
   {
     fsupc = glu.xsup(nsuper); 
-    jptr = glu.xlsub(jcol); // Not yet compressed
-    jm1ptr = glu.xlsub(jcolm1); 
+    StorageIndex jptr = glu.xlsub(jcol); // Not yet compressed
+    StorageIndex jm1ptr = glu.xlsub(jcolm1); 
     
     // Use supernodes of type T2 : see SuperLU paper
     if ( (nextl-jptr != jptr-jm1ptr-1) ) jsuper = emptyIdxLU;
@@ -146,13 +148,13 @@ Index SparseLUImpl<Scalar,Index>::column_dfs(const Index m, const Index jcol, In
     { // starts a new supernode 
       if ( (fsupc < jcolm1-1) ) 
       { // >= 3 columns in nsuper
-        ito = glu.xlsub(fsupc+1);
+        StorageIndex ito = glu.xlsub(fsupc+1);
         glu.xlsub(jcolm1) = ito; 
-        istop = ito + jptr - jm1ptr; 
+        StorageIndex istop = ito + jptr - jm1ptr; 
         xprune(jcolm1) = istop; // intialize xprune(jcol-1)
         glu.xlsub(jcol) = istop; 
         
-        for (ifrom = jm1ptr; ifrom < nextl; ++ifrom, ++ito)
+        for (StorageIndex ifrom = jm1ptr; ifrom < nextl; ++ifrom, ++ito)
           glu.lsub(ito) = glu.lsub(ifrom); 
         nextl = ito;  // = istop + length(jcol)
       }
@@ -164,8 +166,8 @@ Index SparseLUImpl<Scalar,Index>::column_dfs(const Index m, const Index jcol, In
   // Tidy up the pointers before exit
   glu.xsup(nsuper+1) = jcolp1; 
   glu.supno(jcolp1) = nsuper; 
-  xprune(jcol) = nextl;  // Intialize upper bound for pruning
-  glu.xlsub(jcolp1) = nextl; 
+  xprune(jcol) = StorageIndex(nextl);  // Intialize upper bound for pruning
+  glu.xlsub(jcolp1) = StorageIndex(nextl); 
   
   return 0; 
 }
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_copy_to_ucol.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_copy_to_ucol.h
index 170610d9f29..c32d8d8b14b 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_copy_to_ucol.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_copy_to_ucol.h
@@ -46,8 +46,9 @@ namespace internal {
  *         > 0 - number of bytes allocated when run out of space
  * 
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::copy_to_ucol(const Index jcol, const Index nseg, IndexVector& segrep, BlockIndexVector repfnz ,IndexVector& perm_r, BlockScalarVector dense, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::copy_to_ucol(const Index jcol, const Index nseg, IndexVector& segrep,
+                                                      BlockIndexVector repfnz ,IndexVector& perm_r, BlockScalarVector dense, GlobalLU_t& glu)
 {  
   Index ksub, krep, ksupno; 
     
@@ -55,7 +56,7 @@ Index SparseLUImpl<Scalar,Index>::copy_to_ucol(const Index jcol, const Index nse
   
   // For each nonzero supernode segment of U[*,j] in topological order 
   Index k = nseg - 1, i; 
-  Index nextu = glu.xusub(jcol); 
+  StorageIndex nextu = glu.xusub(jcol); 
   Index kfnz, isub, segsize; 
   Index new_next,irow; 
   Index fsupc, mem; 
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_gemm_kernel.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_gemm_kernel.h
index 9e4e3e72b70..95ba7413f29 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_gemm_kernel.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_gemm_kernel.h
@@ -21,7 +21,7 @@ namespace internal {
   *  - lda and ldc must be multiples of the respective packet size
   *  - C must have the same alignment as A
   */
-template<typename Scalar,typename Index>
+template<typename Scalar>
 EIGEN_DONT_INLINE
 void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const Scalar* B, Index ldb, Scalar* C, Index ldc)
 {
@@ -39,9 +39,9 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
   };
   Index d_end = (d/RK)*RK;    // number of columns of A (rows of B) suitable for full register blocking
   Index n_end = (n/RN)*RN;    // number of columns of B-C suitable for processing RN columns at once
-  Index i0 = internal::first_aligned(A,m);
+  Index i0 = internal::first_default_aligned(A,m);
   
-  eigen_internal_assert(((lda%PacketSize)==0) && ((ldc%PacketSize)==0) && (i0==internal::first_aligned(C,m)));
+  eigen_internal_assert(((lda%PacketSize)==0) && ((ldc%PacketSize)==0) && (i0==internal::first_default_aligned(C,m)));
   
   // handle the non aligned rows of A and C without any optimization:
   for(Index i=0; i<i0; ++i)
@@ -72,14 +72,14 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
         
         // load and expand a RN x RK block of B
         Packet b00, b10, b20, b30, b01, b11, b21, b31;
-                  b00 = pset1<Packet>(Bc0[0]);
-                  b10 = pset1<Packet>(Bc0[1]);
-        if(RK==4) b20 = pset1<Packet>(Bc0[2]);
-        if(RK==4) b30 = pset1<Packet>(Bc0[3]);
-                  b01 = pset1<Packet>(Bc1[0]);
-                  b11 = pset1<Packet>(Bc1[1]);
-        if(RK==4) b21 = pset1<Packet>(Bc1[2]);
-        if(RK==4) b31 = pset1<Packet>(Bc1[3]);
+                  { b00 = pset1<Packet>(Bc0[0]); }
+                  { b10 = pset1<Packet>(Bc0[1]); }
+        if(RK==4) { b20 = pset1<Packet>(Bc0[2]); }
+        if(RK==4) { b30 = pset1<Packet>(Bc0[3]); }
+                  { b01 = pset1<Packet>(Bc1[0]); }
+                  { b11 = pset1<Packet>(Bc1[1]); }
+        if(RK==4) { b21 = pset1<Packet>(Bc1[2]); }
+        if(RK==4) { b31 = pset1<Packet>(Bc1[3]); }
         
         Packet a0, a1, a2, a3, c0, c1, t0, t1;
         
@@ -106,22 +106,22 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
         
 #define KMADD(c, a, b, tmp) {tmp = b; tmp = pmul(a,tmp); c = padd(c,tmp);}
 #define WORK(I)  \
-                    c0 = pload<Packet>(C0+i+(I)*PacketSize);   \
-                    c1 = pload<Packet>(C1+i+(I)*PacketSize);   \
-                    KMADD(c0, a0, b00, t0)      \
-                    KMADD(c1, a0, b01, t1)      \
-                    a0 = pload<Packet>(A0+i+(I+1)*PacketSize); \
-                    KMADD(c0, a1, b10, t0)      \
-                    KMADD(c1, a1, b11, t1)       \
-                    a1 = pload<Packet>(A1+i+(I+1)*PacketSize); \
-          if(RK==4) KMADD(c0, a2, b20, t0)       \
-          if(RK==4) KMADD(c1, a2, b21, t1)       \
-          if(RK==4) a2 = pload<Packet>(A2+i+(I+1)*PacketSize); \
-          if(RK==4) KMADD(c0, a3, b30, t0)       \
-          if(RK==4) KMADD(c1, a3, b31, t1)       \
-          if(RK==4) a3 = pload<Packet>(A3+i+(I+1)*PacketSize); \
-                    pstore(C0+i+(I)*PacketSize, c0);           \
-                    pstore(C1+i+(I)*PacketSize, c1)
+                     c0 = pload<Packet>(C0+i+(I)*PacketSize);    \
+                     c1 = pload<Packet>(C1+i+(I)*PacketSize);    \
+                     KMADD(c0, a0, b00, t0)                      \
+                     KMADD(c1, a0, b01, t1)                      \
+                     a0 = pload<Packet>(A0+i+(I+1)*PacketSize);  \
+                     KMADD(c0, a1, b10, t0)                      \
+                     KMADD(c1, a1, b11, t1)                      \
+                     a1 = pload<Packet>(A1+i+(I+1)*PacketSize);  \
+          if(RK==4){ KMADD(c0, a2, b20, t0)                     }\
+          if(RK==4){ KMADD(c1, a2, b21, t1)                     }\
+          if(RK==4){ a2 = pload<Packet>(A2+i+(I+1)*PacketSize); }\
+          if(RK==4){ KMADD(c0, a3, b30, t0)                     }\
+          if(RK==4){ KMADD(c1, a3, b31, t1)                     }\
+          if(RK==4){ a3 = pload<Packet>(A3+i+(I+1)*PacketSize); }\
+                     pstore(C0+i+(I)*PacketSize, c0);            \
+                     pstore(C1+i+(I)*PacketSize, c1)
         
         // process rows of A' - C' with aggressive vectorization and peeling 
         for(Index i=0; i<actual_b_end1; i+=PacketSize*8)
@@ -131,14 +131,15 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
                     prefetch((A1+i+(5)*PacketSize));
           if(RK==4) prefetch((A2+i+(5)*PacketSize));
           if(RK==4) prefetch((A3+i+(5)*PacketSize));
-                    WORK(0);
-                    WORK(1);
-                    WORK(2);
-                    WORK(3);
-                    WORK(4);
-                    WORK(5);
-                    WORK(6);
-                    WORK(7);
+
+          WORK(0);
+          WORK(1);
+          WORK(2);
+          WORK(3);
+          WORK(4);
+          WORK(5);
+          WORK(6);
+          WORK(7);
         }
         // process the remaining rows with vectorization only
         for(Index i=actual_b_end1; i<actual_b_end2; i+=PacketSize)
@@ -165,7 +166,7 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
         Bc1 += RK;
       } // peeled loop on k
     } // peeled loop on the columns j
-    // process the last column (we now perform a matrux-vector product)
+    // process the last column (we now perform a matrix-vector product)
     if((n-n_end)>0)
     {
       const Scalar* Bc0 = B+(n-1)*ldb;
@@ -203,16 +204,16 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
         }
         
 #define WORK(I) \
-                  c0 = pload<Packet>(C0+i+(I)*PacketSize);   \
-                  KMADD(c0, a0, b00, t0)       \
-                  a0 = pload<Packet>(A0+i+(I+1)*PacketSize); \
-                  KMADD(c0, a1, b10, t0)       \
-                  a1 = pload<Packet>(A1+i+(I+1)*PacketSize); \
-        if(RK==4) KMADD(c0, a2, b20, t0)       \
-        if(RK==4) a2 = pload<Packet>(A2+i+(I+1)*PacketSize); \
-        if(RK==4) KMADD(c0, a3, b30, t0)       \
-        if(RK==4) a3 = pload<Packet>(A3+i+(I+1)*PacketSize); \
-                  pstore(C0+i+(I)*PacketSize, c0);
+                   c0 = pload<Packet>(C0+i+(I)*PacketSize);     \
+                   KMADD(c0, a0, b00, t0)                       \
+                   a0 = pload<Packet>(A0+i+(I+1)*PacketSize);   \
+                   KMADD(c0, a1, b10, t0)                       \
+                   a1 = pload<Packet>(A1+i+(I+1)*PacketSize);   \
+        if(RK==4){ KMADD(c0, a2, b20, t0)                      }\
+        if(RK==4){ a2 = pload<Packet>(A2+i+(I+1)*PacketSize);  }\
+        if(RK==4){ KMADD(c0, a3, b30, t0)                      }\
+        if(RK==4){ a3 = pload<Packet>(A3+i+(I+1)*PacketSize);  }\
+                   pstore(C0+i+(I)*PacketSize, c0);
         
         // agressive vectorization and peeling
         for(Index i=0; i<actual_b_end1; i+=PacketSize*8)
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_heap_relax_snode.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_heap_relax_snode.h
index 7a4e4305aa9..6f75d500e5f 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_heap_relax_snode.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_heap_relax_snode.h
@@ -42,21 +42,20 @@ namespace internal {
  * \param descendants Number of descendants of each node in the etree
  * \param relax_end last column in a supernode
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::heap_relax_snode (const Index n, IndexVector& et, const Index relax_columns, IndexVector& descendants, IndexVector& relax_end)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::heap_relax_snode (const Index n, IndexVector& et, const Index relax_columns, IndexVector& descendants, IndexVector& relax_end)
 {
   
   // The etree may not be postordered, but its heap ordered  
   IndexVector post;
-  internal::treePostorder(n, et, post); // Post order etree
+  internal::treePostorder(StorageIndex(n), et, post); // Post order etree
   IndexVector inv_post(n+1); 
-  Index i;
-  for (i = 0; i < n+1; ++i) inv_post(post(i)) = i; // inv_post = post.inverse()???
+  for (StorageIndex i = 0; i < n+1; ++i) inv_post(post(i)) = i; // inv_post = post.inverse()???
   
   // Renumber etree in postorder 
   IndexVector iwork(n);
   IndexVector et_save(n+1);
-  for (i = 0; i < n; ++i)
+  for (Index i = 0; i < n; ++i)
   {
     iwork(post(i)) = post(et(i));
   }
@@ -75,10 +74,10 @@ void SparseLUImpl<Scalar,Index>::heap_relax_snode (const Index n, IndexVector& e
   }
   // Identify the relaxed supernodes by postorder traversal of the etree
   Index snode_start; // beginning of a snode 
-  Index k;
+  StorageIndex k;
   Index nsuper_et_post = 0; // Number of relaxed snodes in postordered etree 
   Index nsuper_et = 0; // Number of relaxed snodes in the original etree 
-  Index l; 
+  StorageIndex l; 
   for (j = 0; j < n; )
   {
     parent = et(j);
@@ -90,8 +89,8 @@ void SparseLUImpl<Scalar,Index>::heap_relax_snode (const Index n, IndexVector& e
     }
     // Found a supernode in postordered etree, j is the last column 
     ++nsuper_et_post;
-    k = n;
-    for (i = snode_start; i <= j; ++i)
+    k = StorageIndex(n);
+    for (Index i = snode_start; i <= j; ++i)
       k = (std::min)(k, inv_post(i));
     l = inv_post(j);
     if ( (l - k) == (j - snode_start) )  // Same number of columns in the snode
@@ -102,7 +101,7 @@ void SparseLUImpl<Scalar,Index>::heap_relax_snode (const Index n, IndexVector& e
     }
     else 
     {
-      for (i = snode_start; i <= j; ++i) 
+      for (Index i = snode_start; i <= j; ++i) 
       {
         l = inv_post(i);
         if (descendants(i) == 0) 
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_kernel_bmod.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_kernel_bmod.h
index 6af02675429..8c1b3e8bc67 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_kernel_bmod.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_kernel_bmod.h
@@ -14,30 +14,29 @@
 namespace Eigen {
 namespace internal {
   
-/**
- * \brief Performs numeric block updates from a given supernode to a single column
- * 
- * \param segsize Size of the segment (and blocks ) to use for updates
- * \param[in,out] dense Packed values of the original matrix
- * \param tempv temporary vector to use for updates
- * \param lusup array containing the supernodes
- * \param lda Leading dimension in the supernode
- * \param nrow Number of rows in the rectangular part of the supernode
- * \param lsub compressed row subscripts of supernodes
- * \param lptr pointer to the first column of the current supernode in lsub
- * \param no_zeros Number of nonzeros elements before the diagonal part of the supernode
- * \return 0 on success
- */
 template <int SegSizeAtCompileTime> struct LU_kernel_bmod
 {
-  template <typename BlockScalarVector, typename ScalarVector, typename IndexVector, typename Index>
-  static EIGEN_DONT_INLINE void run(const int segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, Index& luptr, const Index lda,
+  /** \internal
+    * \brief Performs numeric block updates from a given supernode to a single column
+    *
+    * \param segsize Size of the segment (and blocks ) to use for updates
+    * \param[in,out] dense Packed values of the original matrix
+    * \param tempv temporary vector to use for updates
+    * \param lusup array containing the supernodes
+    * \param lda Leading dimension in the supernode
+    * \param nrow Number of rows in the rectangular part of the supernode
+    * \param lsub compressed row subscripts of supernodes
+    * \param lptr pointer to the first column of the current supernode in lsub
+    * \param no_zeros Number of nonzeros elements before the diagonal part of the supernode
+    */
+  template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
+  static EIGEN_DONT_INLINE void run(const Index segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, Index& luptr, const Index lda,
                                     const Index nrow, IndexVector& lsub, const Index lptr, const Index no_zeros);
 };
 
 template <int SegSizeAtCompileTime>
-template <typename BlockScalarVector, typename ScalarVector, typename IndexVector, typename Index>
-EIGEN_DONT_INLINE void LU_kernel_bmod<SegSizeAtCompileTime>::run(const int segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, Index& luptr, const Index lda,
+template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
+EIGEN_DONT_INLINE void LU_kernel_bmod<SegSizeAtCompileTime>::run(const Index segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, Index& luptr, const Index lda,
                                                                   const Index nrow, IndexVector& lsub, const Index lptr, const Index no_zeros)
 {
   typedef typename ScalarVector::Scalar Scalar;
@@ -45,7 +44,7 @@ EIGEN_DONT_INLINE void LU_kernel_bmod<SegSizeAtCompileTime>::run(const int segsi
   // The result of triangular solve is in tempv[*]; 
     // The result of matric-vector update is in dense[*]
   Index isub = lptr + no_zeros; 
-  int i;
+  Index i;
   Index irow;
   for (i = 0; i < ((SegSizeAtCompileTime==Dynamic)?segsize:SegSizeAtCompileTime); i++)
   {
@@ -66,8 +65,8 @@ EIGEN_DONT_INLINE void LU_kernel_bmod<SegSizeAtCompileTime>::run(const int segsi
   const Index PacketSize = internal::packet_traits<Scalar>::size;
   Index ldl = internal::first_multiple(nrow, PacketSize);
   Map<Matrix<Scalar,Dynamic,SegSizeAtCompileTime, ColMajor>, 0, OuterStride<> > B( &(lusup.data()[luptr]), nrow, segsize, OuterStride<>(lda) );
-  Index aligned_offset = internal::first_aligned(tempv.data()+segsize, PacketSize);
-  Index aligned_with_B_offset = (PacketSize-internal::first_aligned(B.data(), PacketSize))%PacketSize;
+  Index aligned_offset = internal::first_default_aligned(tempv.data()+segsize, PacketSize);
+  Index aligned_with_B_offset = (PacketSize-internal::first_default_aligned(B.data(), PacketSize))%PacketSize;
   Map<Matrix<Scalar,Dynamic,1>, 0, OuterStride<> > l(tempv.data()+segsize+aligned_offset+aligned_with_B_offset, nrow, OuterStride<>(ldl) );
   
   l.setZero();
@@ -91,21 +90,22 @@ EIGEN_DONT_INLINE void LU_kernel_bmod<SegSizeAtCompileTime>::run(const int segsi
 
 template <> struct LU_kernel_bmod<1>
 {
-  template <typename BlockScalarVector, typename ScalarVector, typename IndexVector, typename Index>
-  static EIGEN_DONT_INLINE void run(const int /*segsize*/, BlockScalarVector& dense, ScalarVector& /*tempv*/, ScalarVector& lusup, Index& luptr,
+  template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
+  static EIGEN_DONT_INLINE void run(const Index /*segsize*/, BlockScalarVector& dense, ScalarVector& /*tempv*/, ScalarVector& lusup, Index& luptr,
                                     const Index lda, const Index nrow, IndexVector& lsub, const Index lptr, const Index no_zeros);
 };
 
 
-template <typename BlockScalarVector, typename ScalarVector, typename IndexVector, typename Index>
-EIGEN_DONT_INLINE void LU_kernel_bmod<1>::run(const int /*segsize*/, BlockScalarVector& dense, ScalarVector& /*tempv*/, ScalarVector& lusup, Index& luptr,
+template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
+EIGEN_DONT_INLINE void LU_kernel_bmod<1>::run(const Index /*segsize*/, BlockScalarVector& dense, ScalarVector& /*tempv*/, ScalarVector& lusup, Index& luptr,
                                               const Index lda, const Index nrow, IndexVector& lsub, const Index lptr, const Index no_zeros)
 {
   typedef typename ScalarVector::Scalar Scalar;
+  typedef typename IndexVector::Scalar StorageIndex;
   Scalar f = dense(lsub(lptr + no_zeros));
   luptr += lda * no_zeros + no_zeros + 1;
   const Scalar* a(lusup.data() + luptr);
-  const /*typename IndexVector::Scalar*/Index*  irow(lsub.data()+lptr + no_zeros + 1);
+  const StorageIndex*  irow(lsub.data()+lptr + no_zeros + 1);
   Index i = 0;
   for (; i+1 < nrow; i+=2)
   {
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_panel_bmod.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_panel_bmod.h
index 9d2ff290635..822cf32c347 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_panel_bmod.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_panel_bmod.h
@@ -52,8 +52,8 @@ namespace internal {
  * 
  * 
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::panel_bmod(const Index m, const Index w, const Index jcol, 
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::panel_bmod(const Index m, const Index w, const Index jcol, 
                                             const Index nseg, ScalarVector& dense, ScalarVector& tempv,
                                             IndexVector& segrep, IndexVector& repfnz, GlobalLU_t& glu)
 {
@@ -145,7 +145,7 @@ void SparseLUImpl<Scalar,Index>::panel_bmod(const Index m, const Index w, const
       eigen_assert(tempv.size()>w*ldu + nrow*w + 1);
       
       Index ldl = internal::first_multiple<Index>(nrow, PacketSize);
-      Index offset = (PacketSize-internal::first_aligned(B.data(), PacketSize)) % PacketSize;
+      Index offset = (PacketSize-internal::first_default_aligned(B.data(), PacketSize)) % PacketSize;
       MappedMatrixBlock L(tempv.data()+w*ldu+offset, nrow, u_cols, OuterStride<>(ldl));
       
       L.setZero();
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_panel_dfs.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_panel_dfs.h
index dc0054efd25..155df733687 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_panel_dfs.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_panel_dfs.h
@@ -37,11 +37,11 @@ namespace internal {
 template<typename IndexVector>
 struct panel_dfs_traits
 {
-  typedef typename IndexVector::Scalar Index;
-  panel_dfs_traits(Index jcol, Index* marker)
+  typedef typename IndexVector::Scalar StorageIndex;
+  panel_dfs_traits(Index jcol, StorageIndex* marker)
     : m_jcol(jcol), m_marker(marker)
   {}
-  bool update_segrep(Index krep, Index jj)
+  bool update_segrep(Index krep, StorageIndex jj)
   {
     if(m_marker[krep]<m_jcol)
     {
@@ -53,13 +53,13 @@ struct panel_dfs_traits
   void mem_expand(IndexVector& /*glu.lsub*/, Index /*nextl*/, Index /*chmark*/) {}
   enum { ExpandMem = false };
   Index m_jcol;
-  Index* m_marker;
+  StorageIndex* m_marker;
 };
 
 
-template <typename Scalar, typename Index>
+template <typename Scalar, typename StorageIndex>
 template <typename Traits>
-void SparseLUImpl<Scalar,Index>::dfs_kernel(const Index jj, IndexVector& perm_r,
+void SparseLUImpl<Scalar,StorageIndex>::dfs_kernel(const StorageIndex jj, IndexVector& perm_r,
                    Index& nseg, IndexVector& panel_lsub, IndexVector& segrep,
                    Ref<IndexVector> repfnz_col, IndexVector& xprune, Ref<IndexVector> marker, IndexVector& parent,
                    IndexVector& xplore, GlobalLU_t& glu,
@@ -67,14 +67,14 @@ void SparseLUImpl<Scalar,Index>::dfs_kernel(const Index jj, IndexVector& perm_r,
                   )
 {
   
-  Index kmark = marker(krow);
+  StorageIndex kmark = marker(krow);
       
   // For each unmarked krow of jj
   marker(krow) = jj; 
-  Index kperm = perm_r(krow); 
+  StorageIndex kperm = perm_r(krow); 
   if (kperm == emptyIdxLU ) {
     // krow is in L : place it in structure of L(*, jj)
-    panel_lsub(nextl_col++) = krow;  // krow is indexed into A
+    panel_lsub(nextl_col++) = StorageIndex(krow);  // krow is indexed into A
     
     traits.mem_expand(panel_lsub, nextl_col, kmark);
   }
@@ -83,9 +83,9 @@ void SparseLUImpl<Scalar,Index>::dfs_kernel(const Index jj, IndexVector& perm_r,
     // krow is in U : if its supernode-representative krep
     // has been explored, update repfnz(*)
     // krep = supernode representative of the current row
-    Index krep = glu.xsup(glu.supno(kperm)+1) - 1; 
+    StorageIndex krep = glu.xsup(glu.supno(kperm)+1) - 1; 
     // First nonzero element in the current column:
-    Index myfnz = repfnz_col(krep); 
+    StorageIndex myfnz = repfnz_col(krep); 
     
     if (myfnz != emptyIdxLU )
     {
@@ -96,26 +96,26 @@ void SparseLUImpl<Scalar,Index>::dfs_kernel(const Index jj, IndexVector& perm_r,
     else 
     {
       // Otherwise, perform dfs starting at krep
-      Index oldrep = emptyIdxLU; 
+      StorageIndex oldrep = emptyIdxLU; 
       parent(krep) = oldrep; 
       repfnz_col(krep) = kperm; 
-      Index xdfs =  glu.xlsub(krep); 
+      StorageIndex xdfs =  glu.xlsub(krep); 
       Index maxdfs = xprune(krep); 
       
-      Index kpar;
+      StorageIndex kpar;
       do 
       {
         // For each unmarked kchild of krep
         while (xdfs < maxdfs) 
         {
-          Index kchild = glu.lsub(xdfs); 
+          StorageIndex kchild = glu.lsub(xdfs); 
           xdfs++; 
-          Index chmark = marker(kchild); 
+          StorageIndex chmark = marker(kchild); 
           
           if (chmark != jj ) 
           {
             marker(kchild) = jj; 
-            Index chperm = perm_r(kchild); 
+            StorageIndex chperm = perm_r(kchild); 
             
             if (chperm == emptyIdxLU) 
             {
@@ -128,7 +128,7 @@ void SparseLUImpl<Scalar,Index>::dfs_kernel(const Index jj, IndexVector& perm_r,
               // case kchild is in U :
               // chrep = its supernode-rep. If its rep has been explored, 
               // update its repfnz(*)
-              Index chrep = glu.xsup(glu.supno(chperm)+1) - 1; 
+              StorageIndex chrep = glu.xsup(glu.supno(chperm)+1) - 1; 
               myfnz = repfnz_col(chrep); 
               
               if (myfnz != emptyIdxLU) 
@@ -215,8 +215,8 @@ void SparseLUImpl<Scalar,Index>::dfs_kernel(const Index jj, IndexVector& perm_r,
  * 
  */
 
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::panel_dfs(const Index m, const Index w, const Index jcol, MatrixType& A, IndexVector& perm_r, Index& nseg, ScalarVector& dense, IndexVector& panel_lsub, IndexVector& segrep, IndexVector& repfnz, IndexVector& xprune, IndexVector& marker, IndexVector& parent, IndexVector& xplore, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::panel_dfs(const Index m, const Index w, const Index jcol, MatrixType& A, IndexVector& perm_r, Index& nseg, ScalarVector& dense, IndexVector& panel_lsub, IndexVector& segrep, IndexVector& repfnz, IndexVector& xprune, IndexVector& marker, IndexVector& parent, IndexVector& xplore, GlobalLU_t& glu)
 {
   Index nextl_col; // Next available position in panel_lsub[*,jj] 
   
@@ -227,7 +227,7 @@ void SparseLUImpl<Scalar,Index>::panel_dfs(const Index m, const Index w, const I
   panel_dfs_traits<IndexVector> traits(jcol, marker1.data());
   
   // For each column in the panel 
-  for (Index jj = jcol; jj < jcol + w; jj++) 
+  for (StorageIndex jj = StorageIndex(jcol); jj < jcol + w; jj++) 
   {
     nextl_col = (jj - jcol) * m; 
     
@@ -241,7 +241,7 @@ void SparseLUImpl<Scalar,Index>::panel_dfs(const Index m, const Index w, const I
       Index krow = it.row(); 
       dense_col(krow) = it.value();
       
-      Index kmark = marker(krow); 
+      StorageIndex kmark = marker(krow); 
       if (kmark == jj) 
         continue; // krow visited before, go to the next nonzero
       
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_pivotL.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_pivotL.h
index 2e49ef667f4..a86dac93fa9 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_pivotL.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_pivotL.h
@@ -56,8 +56,8 @@ namespace internal {
  * \return 0 if success, i > 0 if U(i,i) is exactly zero 
  * 
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::pivotL(const Index jcol, const RealScalar& diagpivotthresh, IndexVector& perm_r, IndexVector& iperm_c, Index& pivrow, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::pivotL(const Index jcol, const RealScalar& diagpivotthresh, IndexVector& perm_r, IndexVector& iperm_c, Index& pivrow, GlobalLU_t& glu)
 {
   
   Index fsupc = (glu.xsup)((glu.supno)(jcol)); // First column in the supernode containing the column jcol
@@ -67,7 +67,7 @@ Index SparseLUImpl<Scalar,Index>::pivotL(const Index jcol, const RealScalar& dia
   Index lda = glu.xlusup(fsupc+1) - glu.xlusup(fsupc); // leading dimension
   Scalar* lu_sup_ptr = &(glu.lusup.data()[glu.xlusup(fsupc)]); // Start of the current supernode
   Scalar* lu_col_ptr = &(glu.lusup.data()[glu.xlusup(jcol)]); // Start of jcol in the supernode
-  Index* lsub_ptr = &(glu.lsub.data()[lptr]); // Start of row indices of the supernode
+  StorageIndex* lsub_ptr = &(glu.lsub.data()[lptr]); // Start of row indices of the supernode
   
   // Determine the largest abs numerical value for partial pivoting 
   Index diagind = iperm_c(jcol); // diagonal index 
@@ -90,7 +90,7 @@ Index SparseLUImpl<Scalar,Index>::pivotL(const Index jcol, const RealScalar& dia
   if ( pivmax <= RealScalar(0.0) ) {
     // if pivmax == -1, the column is structurally empty, otherwise it is only numerically zero
     pivrow = pivmax < RealScalar(0.0) ? diagind : lsub_ptr[pivptr];
-    perm_r(pivrow) = jcol;
+    perm_r(pivrow) = StorageIndex(jcol);
     return (jcol+1);
   }
   
@@ -105,13 +105,13 @@ Index SparseLUImpl<Scalar,Index>::pivotL(const Index jcol, const RealScalar& dia
       // Diagonal element exists
       using std::abs;
       rtemp = abs(lu_col_ptr[diag]);
-      if (rtemp != 0.0 && rtemp >= thresh) pivptr = diag;
+      if (rtemp != RealScalar(0.0) && rtemp >= thresh) pivptr = diag;
     }
     pivrow = lsub_ptr[pivptr];
   }
   
   // Record pivot row
-  perm_r(pivrow) = jcol; 
+  perm_r(pivrow) = StorageIndex(jcol);
   // Interchange row subscripts
   if (pivptr != nsupc )
   {
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_pruneL.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_pruneL.h
index 66460d16884..ad32fed5e6b 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_pruneL.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_pruneL.h
@@ -49,8 +49,9 @@ namespace internal {
  * \param glu Global LU data
  * 
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::pruneL(const Index jcol, const IndexVector& perm_r, const Index pivrow, const Index nseg, const IndexVector& segrep, BlockIndexVector repfnz, IndexVector& xprune, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::pruneL(const Index jcol, const IndexVector& perm_r, const Index pivrow, const Index nseg,
+                                               const IndexVector& segrep, BlockIndexVector repfnz, IndexVector& xprune, GlobalLU_t& glu)
 {
   // For each supernode-rep irep in U(*,j]
   Index jsupno = glu.supno(jcol); 
@@ -123,7 +124,7 @@ void SparseLUImpl<Scalar,Index>::pruneL(const Index jcol, const IndexVector& per
           }
         } // end while 
         
-        xprune(irep) = kmin;  //Pruning 
+        xprune(irep) = StorageIndex(kmin);  //Pruning 
       } // end if do_prune 
     } // end pruning 
   } // End for each U-segment
diff --git a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_relax_snode.h b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_relax_snode.h
index 58ec32e27e9..c408d01b406 100644
--- a/extern/Eigen3/Eigen/src/SparseLU/SparseLU_relax_snode.h
+++ b/extern/Eigen3/Eigen/src/SparseLU/SparseLU_relax_snode.h
@@ -43,15 +43,15 @@ namespace internal {
  * \param descendants Number of descendants of each node in the etree
  * \param relax_end last column in a supernode
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::relax_snode (const Index n, IndexVector& et, const Index relax_columns, IndexVector& descendants, IndexVector& relax_end)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::relax_snode (const Index n, IndexVector& et, const Index relax_columns, IndexVector& descendants, IndexVector& relax_end)
 {
   
   // compute the number of descendants of each node in the etree
-  Index j, parent; 
+  Index parent; 
   relax_end.setConstant(emptyIdxLU);
   descendants.setZero();
-  for (j = 0; j < n; j++) 
+  for (Index j = 0; j < n; j++) 
   {
     parent = et(j);
     if (parent != n) // not the dummy root
@@ -59,7 +59,7 @@ void SparseLUImpl<Scalar,Index>::relax_snode (const Index n, IndexVector& et, co
   }
   // Identify the relaxed supernodes by postorder traversal of the etree
   Index snode_start; // beginning of a snode 
-  for (j = 0; j < n; )
+  for (Index j = 0; j < n; )
   {
     parent = et(j);
     snode_start = j; 
@@ -69,7 +69,7 @@ void SparseLUImpl<Scalar,Index>::relax_snode (const Index n, IndexVector& et, co
       parent = et(j);
     }
     // Found a supernode in postordered etree, j is the last column 
-    relax_end(snode_start) = j; // Record last column
+    relax_end(snode_start) = StorageIndex(j); // Record last column
     j++;
     // Search for a new leaf
     while (descendants(j) != 0 && j < n) j++;
diff --git a/extern/Eigen3/Eigen/src/SparseQR/SparseQR.h b/extern/Eigen3/Eigen/src/SparseQR/SparseQR.h
index a00bd5db124..7409fcae940 100644
--- a/extern/Eigen3/Eigen/src/SparseQR/SparseQR.h
+++ b/extern/Eigen3/Eigen/src/SparseQR/SparseQR.h
@@ -21,8 +21,12 @@ namespace internal {
   template <typename SparseQRType> struct traits<SparseQRMatrixQReturnType<SparseQRType> >
   {
     typedef typename SparseQRType::MatrixType ReturnType;
-    typedef typename ReturnType::Index Index;
+    typedef typename ReturnType::StorageIndex StorageIndex;
     typedef typename ReturnType::StorageKind StorageKind;
+    enum {
+      RowsAtCompileTime = Dynamic,
+      ColsAtCompileTime = Dynamic
+    };
   };
   template <typename SparseQRType> struct traits<SparseQRMatrixQTransposeReturnType<SparseQRType> >
   {
@@ -48,7 +52,7 @@ namespace internal {
   * rank-revealing permutations. Use colsPermutation() to get it.
   * 
   * Q is the orthogonal matrix represented as products of Householder reflectors. 
-  * Use matrixQ() to get an expression and matrixQ().transpose() to get the transpose.
+  * Use matrixQ() to get an expression and matrixQ().adjoint() to get the adjoint.
   * You can then apply it to a vector.
   * 
   * R is the sparse triangular or trapezoidal matrix. The later occurs when A is rank-deficient.
@@ -58,24 +62,37 @@ namespace internal {
   * \tparam _OrderingType The fill-reducing ordering method. See the \link OrderingMethods_Module 
   *  OrderingMethods \endlink module for the list of built-in and external ordering methods.
   * 
+  * \implsparsesolverconcept
+  *
   * \warning The input sparse matrix A must be in compressed mode (see SparseMatrix::makeCompressed()).
+  * \warning For complex matrices matrixQ().transpose() will actually return the adjoint matrix.
   * 
   */
 template<typename _MatrixType, typename _OrderingType>
-class SparseQR
+class SparseQR : public SparseSolverBase<SparseQR<_MatrixType,_OrderingType> >
 {
+  protected:
+    typedef SparseSolverBase<SparseQR<_MatrixType,_OrderingType> > Base;
+    using Base::m_isInitialized;
   public:
+    using Base::_solve_impl;
     typedef _MatrixType MatrixType;
     typedef _OrderingType OrderingType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef SparseMatrix<Scalar,ColMajor,Index> QRMatrixType;
-    typedef Matrix<Index, Dynamic, 1> IndexVector;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> QRMatrixType;
+    typedef Matrix<StorageIndex, Dynamic, 1> IndexVector;
     typedef Matrix<Scalar, Dynamic, 1> ScalarVector;
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
+
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
+    
   public:
-    SparseQR () : m_isInitialized(false), m_analysisIsok(false), m_lastError(""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
+    SparseQR () :  m_analysisIsok(false), m_lastError(""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
     { }
     
     /** Construct a QR factorization of the matrix \a mat.
@@ -84,7 +101,7 @@ class SparseQR
       * 
       * \sa compute()
       */
-    SparseQR(const MatrixType& mat) : m_isInitialized(false), m_analysisIsok(false), m_lastError(""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
+    explicit SparseQR(const MatrixType& mat) : m_analysisIsok(false), m_lastError(""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
     {
       compute(mat);
     }
@@ -112,6 +129,17 @@ class SparseQR
     inline Index cols() const { return m_pmat.cols();}
     
     /** \returns a const reference to the \b sparse upper triangular matrix R of the QR factorization.
+      * \warning The entries of the returned matrix are not sorted. This means that using it in algorithms
+      *          expecting sorted entries will fail. This include random coefficient accesses (SpaseMatrix::coeff()),
+      *          and coefficient-wise operations. Matrix products and triangular solves are fine though.
+      *
+      * To sort the entries, you can assign it to a row-major matrix, and if a column-major matrix
+      * is required, you can copy it again:
+      * \code
+      * SparseMatrix<double>          R  = qr.matrixR();  // column-major, not sorted!
+      * SparseMatrix<double,RowMajor> Rr = qr.matrixR();  // row-major, sorted
+      * SparseMatrix<double>          Rc = Rr;            // column-major, sorted
+      * \endcode
       */
     const QRMatrixType& matrixR() const { return m_R; }
     
@@ -119,7 +147,7 @@ class SparseQR
       *
       * \sa setPivotThreshold()
       */
-    Index rank() const 
+    Index rank() const
     {
       eigen_assert(m_isInitialized && "The factorization should be called first, use compute()");
       return m_nonzeropivots; 
@@ -162,23 +190,23 @@ class SparseQR
     
     /** \internal */
     template<typename Rhs, typename Dest>
-    bool _solve(const MatrixBase<Rhs> &B, MatrixBase<Dest> &dest) const
+    bool _solve_impl(const MatrixBase<Rhs> &B, MatrixBase<Dest> &dest) const
     {
       eigen_assert(m_isInitialized && "The factorization should be called first, use compute()");
       eigen_assert(this->rows() == B.rows() && "SparseQR::solve() : invalid number of rows in the right hand side matrix");
 
       Index rank = this->rank();
       
-      // Compute Q^T * b;
+      // Compute Q^* * b;
       typename Dest::PlainObject y, b;
-      y = this->matrixQ().transpose() * B; 
+      y = this->matrixQ().adjoint() * B;
       b = y;
       
       // Solve with the triangular matrix R
-      y.resize((std::max)(cols(),Index(y.rows())),y.cols());
+      y.resize((std::max<Index>)(cols(),y.rows()),y.cols());
       y.topRows(rank) = this->matrixR().topLeftCorner(rank, rank).template triangularView<Upper>().solve(b.topRows(rank));
       y.bottomRows(y.rows()-rank).setZero();
-
+      
       // Apply the column permutation
       if (m_perm_c.size())  dest = colsPermutation() * y.topRows(cols());
       else                  dest = y.topRows(cols());
@@ -186,7 +214,6 @@ class SparseQR
       m_info = Success;
       return true;
     }
-    
 
     /** Sets the threshold that is used to determine linearly dependent columns during the factorization.
       *
@@ -204,18 +231,18 @@ class SparseQR
       * \sa compute()
       */
     template<typename Rhs>
-    inline const internal::solve_retval<SparseQR, Rhs> solve(const MatrixBase<Rhs>& B) const 
+    inline const Solve<SparseQR, Rhs> solve(const MatrixBase<Rhs>& B) const 
     {
       eigen_assert(m_isInitialized && "The factorization should be called first, use compute()");
       eigen_assert(this->rows() == B.rows() && "SparseQR::solve() : invalid number of rows in the right hand side matrix");
-      return internal::solve_retval<SparseQR, Rhs>(*this, B.derived());
+      return Solve<SparseQR, Rhs>(*this, B.derived());
     }
     template<typename Rhs>
-    inline const internal::sparse_solve_retval<SparseQR, Rhs> solve(const SparseMatrixBase<Rhs>& B) const
+    inline const Solve<SparseQR, Rhs> solve(const SparseMatrixBase<Rhs>& B) const
     {
           eigen_assert(m_isInitialized && "The factorization should be called first, use compute()");
           eigen_assert(this->rows() == B.rows() && "SparseQR::solve() : invalid number of rows in the right hand side matrix");
-          return internal::sparse_solve_retval<SparseQR, Rhs>(*this, B.derived());
+          return Solve<SparseQR, Rhs>(*this, B.derived());
     }
     
     /** \brief Reports whether previous computation was successful.
@@ -232,8 +259,9 @@ class SparseQR
       return m_info;
     }
 
-  protected:
-    inline void sort_matrix_Q()
+
+    /** \internal */
+    inline void _sort_matrix_Q()
     {
       if(this->m_isQSorted) return;
       // The matrix Q is sorted during the transposition
@@ -244,7 +272,6 @@ class SparseQR
 
     
   protected:
-    bool m_isInitialized;
     bool m_analysisIsok;
     bool m_factorizationIsok;
     mutable ComputationInfo m_info;
@@ -258,14 +285,13 @@ class SparseQR
     PermutationType m_outputPerm_c; // The final column permutation
     RealScalar m_threshold;         // Threshold to determine null Householder reflections
     bool m_useDefaultThreshold;     // Use default threshold
-    Index m_nonzeropivots;          // Number of non zero pivots found 
+    Index m_nonzeropivots;          // Number of non zero pivots found
     IndexVector m_etree;            // Column elimination tree
     IndexVector m_firstRowElt;      // First element in each row
     bool m_isQSorted;               // whether Q is sorted or not
     bool m_isEtreeOk;               // whether the elimination tree match the initial input matrix
     
     template <typename, typename > friend struct SparseQR_QProduct;
-    template <typename > friend struct SparseQRMatrixQReturnType;
     
 };
 
@@ -294,7 +320,7 @@ void SparseQR<MatrixType,OrderingType>::analyzePattern(const MatrixType& mat)
   if (!m_perm_c.size())
   {
     m_perm_c.resize(n);
-    m_perm_c.indices().setLinSpaced(n, 0,n-1);
+    m_perm_c.indices().setLinSpaced(n, 0,StorageIndex(n-1));
   }
   
   // Compute the column elimination tree of the permuted matrix
@@ -323,12 +349,11 @@ template <typename MatrixType, typename OrderingType>
 void SparseQR<MatrixType,OrderingType>::factorize(const MatrixType& mat)
 {
   using std::abs;
-  using std::max;
   
   eigen_assert(m_analysisIsok && "analyzePattern() should be called before this step");
-  Index m = mat.rows();
-  Index n = mat.cols();
-  Index diagSize = (std::min)(m,n);
+  StorageIndex m = StorageIndex(mat.rows());
+  StorageIndex n = StorageIndex(mat.cols());
+  StorageIndex diagSize = (std::min)(m,n);
   IndexVector mark((std::max)(m,n)); mark.setConstant(-1);  // Record the visited nodes
   IndexVector Ridx(n), Qidx(m);                             // Store temporarily the row indexes for the current column of R and Q
   Index nzcolR, nzcolQ;                                     // Number of nonzero for the current column of R and Q
@@ -353,7 +378,7 @@ void SparseQR<MatrixType,OrderingType>::factorize(const MatrixType& mat)
     // otherwise directly use the input matrix
     // 
     IndexVector originalOuterIndicesCpy;
-    const Index *originalOuterIndices = mat.outerIndexPtr();
+    const StorageIndex *originalOuterIndices = mat.outerIndexPtr();
     if(MatrixType::IsRowMajor)
     {
       originalOuterIndicesCpy = IndexVector::Map(m_pmat.outerIndexPtr(),n+1);
@@ -375,7 +400,7 @@ void SparseQR<MatrixType,OrderingType>::factorize(const MatrixType& mat)
   if(m_useDefaultThreshold) 
   {
     RealScalar max2Norm = 0.0;
-    for (int j = 0; j < n; j++) max2Norm = (max)(max2Norm, m_pmat.col(j).norm());
+    for (int j = 0; j < n; j++) max2Norm = numext::maxi(max2Norm, m_pmat.col(j).norm());
     if(max2Norm==RealScalar(0))
       max2Norm = RealScalar(1);
     pivotThreshold = 20 * (m + n) * max2Norm * NumTraits<RealScalar>::epsilon();
@@ -384,11 +409,11 @@ void SparseQR<MatrixType,OrderingType>::factorize(const MatrixType& mat)
   // Initialize the numerical permutation
   m_pivotperm.setIdentity(n);
   
-  Index nonzeroCol = 0; // Record the number of valid pivots
+  StorageIndex nonzeroCol = 0; // Record the number of valid pivots
   m_Q.startVec(0);
 
   // Left looking rank-revealing QR factorization: compute a column of R and Q at a time
-  for (Index col = 0; col < n; ++col)
+  for (StorageIndex col = 0; col < n; ++col)
   {
     mark.setConstant(-1);
     m_R.startVec(col);
@@ -404,12 +429,12 @@ void SparseQR<MatrixType,OrderingType>::factorize(const MatrixType& mat)
     // thus the trick with found_diag that permits to do one more iteration on the diagonal element if this one has not been found.
     for (typename QRMatrixType::InnerIterator itp(m_pmat, col); itp || !found_diag; ++itp)
     {
-      Index curIdx = nonzeroCol;
-      if(itp) curIdx = itp.row();
+      StorageIndex curIdx = nonzeroCol;
+      if(itp) curIdx = StorageIndex(itp.row());
       if(curIdx == nonzeroCol) found_diag = true;
       
       // Get the nonzeros indexes of the current column of R
-      Index st = m_firstRowElt(curIdx); // The traversal of the etree starts here 
+      StorageIndex st = m_firstRowElt(curIdx); // The traversal of the etree starts here
       if (st < 0 )
       {
         m_lastError = "Empty row found during numerical factorization";
@@ -466,7 +491,7 @@ void SparseQR<MatrixType,OrderingType>::factorize(const MatrixType& mat)
       {
         for (typename QRMatrixType::InnerIterator itq(m_Q, curIdx); itq; ++itq)
         {
-          Index iQ = itq.row();
+          StorageIndex iQ = StorageIndex(itq.row());
           if (mark(iQ) != col)
           {
             Qidx(nzcolQ++) = iQ;  // Add this row to the pattern of Q,
@@ -476,7 +501,7 @@ void SparseQR<MatrixType,OrderingType>::factorize(const MatrixType& mat)
       }
     } // End update current column
     
-    Scalar tau = 0;
+    Scalar tau = RealScalar(0);
     RealScalar beta = 0;
     
     if(nonzeroCol < diagSize)
@@ -572,44 +597,15 @@ void SparseQR<MatrixType,OrderingType>::factorize(const MatrixType& mat)
   m_info = Success;
 }
 
-namespace internal {
-  
-template<typename _MatrixType, typename OrderingType, typename Rhs>
-struct solve_retval<SparseQR<_MatrixType,OrderingType>, Rhs>
-  : solve_retval_base<SparseQR<_MatrixType,OrderingType>, Rhs>
-{
-  typedef SparseQR<_MatrixType,OrderingType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-template<typename _MatrixType, typename OrderingType, typename Rhs>
-struct sparse_solve_retval<SparseQR<_MatrixType, OrderingType>, Rhs>
- : sparse_solve_retval_base<SparseQR<_MatrixType, OrderingType>, Rhs>
-{
-  typedef SparseQR<_MatrixType, OrderingType> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec, Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-} // end namespace internal
-
 template <typename SparseQRType, typename Derived>
 struct SparseQR_QProduct : ReturnByValue<SparseQR_QProduct<SparseQRType, Derived> >
 {
   typedef typename SparseQRType::QRMatrixType MatrixType;
   typedef typename SparseQRType::Scalar Scalar;
-  typedef typename SparseQRType::Index Index;
   // Get the references 
   SparseQR_QProduct(const SparseQRType& qr, const Derived& other, bool transpose) : 
   m_qr(qr),m_other(other),m_transpose(transpose) {}
-  inline Index rows() const { return m_transpose ? m_qr.rows() : m_qr.cols(); }
+  inline Index rows() const { return m_qr.matrixQ().rows(); }
   inline Index cols() const { return m_other.cols(); }
   
   // Assign to a vector
@@ -637,7 +633,10 @@ struct SparseQR_QProduct : ReturnByValue<SparseQR_QProduct<SparseQRType, Derived
     }
     else
     {
-      eigen_assert(m_qr.m_Q.rows() == m_other.rows() && "Non conforming object sizes");
+      eigen_assert(m_qr.matrixQ().cols() == m_other.rows() && "Non conforming object sizes");
+
+      res.conservativeResize(rows(), cols());
+
       // Compute res = Q * other column by column
       for(Index j = 0; j < res.cols(); j++)
       {
@@ -646,7 +645,7 @@ struct SparseQR_QProduct : ReturnByValue<SparseQR_QProduct<SparseQRType, Derived
           Scalar tau = Scalar(0);
           tau = m_qr.m_Q.col(k).dot(res.col(j));
           if(tau==Scalar(0)) continue;
-          tau = tau * m_qr.m_hcoeffs(k);
+          tau = tau * numext::conj(m_qr.m_hcoeffs(k));
           res.col(j) -= tau * m_qr.m_Q.col(k);
         }
       }
@@ -655,52 +654,44 @@ struct SparseQR_QProduct : ReturnByValue<SparseQR_QProduct<SparseQRType, Derived
   
   const SparseQRType& m_qr;
   const Derived& m_other;
-  bool m_transpose;
+  bool m_transpose; // TODO this actually means adjoint
 };
 
 template<typename SparseQRType>
 struct SparseQRMatrixQReturnType : public EigenBase<SparseQRMatrixQReturnType<SparseQRType> >
 {  
-  typedef typename SparseQRType::Index Index;
   typedef typename SparseQRType::Scalar Scalar;
   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
-  SparseQRMatrixQReturnType(const SparseQRType& qr) : m_qr(qr) {}
+  enum {
+    RowsAtCompileTime = Dynamic,
+    ColsAtCompileTime = Dynamic
+  };
+  explicit SparseQRMatrixQReturnType(const SparseQRType& qr) : m_qr(qr) {}
   template<typename Derived>
   SparseQR_QProduct<SparseQRType, Derived> operator*(const MatrixBase<Derived>& other)
   {
     return SparseQR_QProduct<SparseQRType,Derived>(m_qr,other.derived(),false);
   }
+  // To use for operations with the adjoint of Q
   SparseQRMatrixQTransposeReturnType<SparseQRType> adjoint() const
   {
     return SparseQRMatrixQTransposeReturnType<SparseQRType>(m_qr);
   }
   inline Index rows() const { return m_qr.rows(); }
-  inline Index cols() const { return (std::min)(m_qr.rows(),m_qr.cols()); }
-  // To use for operations with the transpose of Q
+  inline Index cols() const { return m_qr.rows(); }
+  // To use for operations with the transpose of Q FIXME this is the same as adjoint at the moment
   SparseQRMatrixQTransposeReturnType<SparseQRType> transpose() const
   {
     return SparseQRMatrixQTransposeReturnType<SparseQRType>(m_qr);
   }
-  template<typename Dest> void evalTo(MatrixBase<Dest>& dest) const
-  {
-    dest.derived() = m_qr.matrixQ() * Dest::Identity(m_qr.rows(), m_qr.rows());
-  }
-  template<typename Dest> void evalTo(SparseMatrixBase<Dest>& dest) const
-  {
-    Dest idMat(m_qr.rows(), m_qr.rows());
-    idMat.setIdentity();
-    // Sort the sparse householder reflectors if needed
-    const_cast<SparseQRType *>(&m_qr)->sort_matrix_Q();
-    dest.derived() = SparseQR_QProduct<SparseQRType, Dest>(m_qr, idMat, false);
-  }
-
   const SparseQRType& m_qr;
 };
 
+// TODO this actually represents the adjoint of Q
 template<typename SparseQRType>
 struct SparseQRMatrixQTransposeReturnType
 {
-  SparseQRMatrixQTransposeReturnType(const SparseQRType& qr) : m_qr(qr) {}
+  explicit SparseQRMatrixQTransposeReturnType(const SparseQRType& qr) : m_qr(qr) {}
   template<typename Derived>
   SparseQR_QProduct<SparseQRType,Derived> operator*(const MatrixBase<Derived>& other)
   {
@@ -709,6 +700,46 @@ struct SparseQRMatrixQTransposeReturnType
   const SparseQRType& m_qr;
 };
 
+namespace internal {
+  
+template<typename SparseQRType>
+struct evaluator_traits<SparseQRMatrixQReturnType<SparseQRType> >
+{
+  typedef typename SparseQRType::MatrixType MatrixType;
+  typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
+  typedef SparseShape Shape;
+};
+
+template< typename DstXprType, typename SparseQRType>
+struct Assignment<DstXprType, SparseQRMatrixQReturnType<SparseQRType>, internal::assign_op<typename DstXprType::Scalar,typename DstXprType::Scalar>, Sparse2Sparse>
+{
+  typedef SparseQRMatrixQReturnType<SparseQRType> SrcXprType;
+  typedef typename DstXprType::Scalar Scalar;
+  typedef typename DstXprType::StorageIndex StorageIndex;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &/*func*/)
+  {
+    typename DstXprType::PlainObject idMat(src.rows(), src.cols());
+    idMat.setIdentity();
+    // Sort the sparse householder reflectors if needed
+    const_cast<SparseQRType *>(&src.m_qr)->_sort_matrix_Q();
+    dst = SparseQR_QProduct<SparseQRType, DstXprType>(src.m_qr, idMat, false);
+  }
+};
+
+template< typename DstXprType, typename SparseQRType>
+struct Assignment<DstXprType, SparseQRMatrixQReturnType<SparseQRType>, internal::assign_op<typename DstXprType::Scalar,typename DstXprType::Scalar>, Sparse2Dense>
+{
+  typedef SparseQRMatrixQReturnType<SparseQRType> SrcXprType;
+  typedef typename DstXprType::Scalar Scalar;
+  typedef typename DstXprType::StorageIndex StorageIndex;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &/*func*/)
+  {
+    dst = src.m_qr.matrixQ() * DstXprType::Identity(src.m_qr.rows(), src.m_qr.rows());
+  }
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif
diff --git a/extern/Eigen3/Eigen/src/StlSupport/StdDeque.h b/extern/Eigen3/Eigen/src/StlSupport/StdDeque.h
index aaf66330b17..cf1fedf9273 100644
--- a/extern/Eigen3/Eigen/src/StlSupport/StdDeque.h
+++ b/extern/Eigen3/Eigen/src/StlSupport/StdDeque.h
@@ -13,32 +13,24 @@
 
 #include "details.h"
 
-// Define the explicit instantiation (e.g. necessary for the Intel compiler)
-#if defined(__INTEL_COMPILER) || defined(__GNUC__)
-  #define EIGEN_EXPLICIT_STL_DEQUE_INSTANTIATION(...) template class std::deque<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> >;
-#else
-  #define EIGEN_EXPLICIT_STL_DEQUE_INSTANTIATION(...)
-#endif
-
 /**
  * This section contains a convenience MACRO which allows an easy specialization of
  * std::deque such that for data types with alignment issues the correct allocator
  * is used automatically.
  */
 #define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...) \
-EIGEN_EXPLICIT_STL_DEQUE_INSTANTIATION(__VA_ARGS__) \
 namespace std \
 { \
-  template<typename _Ay> \
-  class deque<__VA_ARGS__, _Ay>  \
+  template<> \
+  class deque<__VA_ARGS__, std::allocator<__VA_ARGS__> >           \
     : public deque<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> > \
   { \
     typedef deque<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> > deque_base; \
   public: \
     typedef __VA_ARGS__ value_type; \
-    typedef typename deque_base::allocator_type allocator_type; \
-    typedef typename deque_base::size_type size_type;  \
-    typedef typename deque_base::iterator iterator;  \
+    typedef deque_base::allocator_type allocator_type; \
+    typedef deque_base::size_type size_type;  \
+    typedef deque_base::iterator iterator;  \
     explicit deque(const allocator_type& a = allocator_type()) : deque_base(a) {}  \
     template<typename InputIterator> \
     deque(InputIterator first, InputIterator last, const allocator_type& a = allocator_type()) : deque_base(first, last, a) {} \
@@ -53,7 +45,7 @@ namespace std \
 }
 
 // check whether we really need the std::deque specialization
-#if !(defined(_GLIBCXX_DEQUE) && (!EIGEN_GNUC_AT_LEAST(4,1))) /* Note that before gcc-4.1 we already have: std::deque::resize(size_type,const T&). */
+#if !EIGEN_HAS_CXX11_CONTAINERS && !(defined(_GLIBCXX_DEQUE) && (!EIGEN_GNUC_AT_LEAST(4,1))) /* Note that before gcc-4.1 we already have: std::deque::resize(size_type,const T&). */
 
 namespace std {
 
diff --git a/extern/Eigen3/Eigen/src/StlSupport/StdList.h b/extern/Eigen3/Eigen/src/StlSupport/StdList.h
index 3c742430c12..e1eba498596 100644
--- a/extern/Eigen3/Eigen/src/StlSupport/StdList.h
+++ b/extern/Eigen3/Eigen/src/StlSupport/StdList.h
@@ -12,32 +12,24 @@
 
 #include "details.h"
 
-// Define the explicit instantiation (e.g. necessary for the Intel compiler)
-#if defined(__INTEL_COMPILER) || defined(__GNUC__)
-  #define EIGEN_EXPLICIT_STL_LIST_INSTANTIATION(...) template class std::list<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> >;
-#else
-  #define EIGEN_EXPLICIT_STL_LIST_INSTANTIATION(...)
-#endif
-
 /**
  * This section contains a convenience MACRO which allows an easy specialization of
  * std::list such that for data types with alignment issues the correct allocator
  * is used automatically.
  */
 #define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...) \
-EIGEN_EXPLICIT_STL_LIST_INSTANTIATION(__VA_ARGS__) \
 namespace std \
 { \
-  template<typename _Ay> \
-  class list<__VA_ARGS__, _Ay>  \
+  template<> \
+  class list<__VA_ARGS__, std::allocator<__VA_ARGS__> >           \
     : public list<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> > \
   { \
     typedef list<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> > list_base; \
   public: \
     typedef __VA_ARGS__ value_type; \
-    typedef typename list_base::allocator_type allocator_type; \
-    typedef typename list_base::size_type size_type;  \
-    typedef typename list_base::iterator iterator;  \
+    typedef list_base::allocator_type allocator_type; \
+    typedef list_base::size_type size_type;  \
+    typedef list_base::iterator iterator;  \
     explicit list(const allocator_type& a = allocator_type()) : list_base(a) {}  \
     template<typename InputIterator> \
     list(InputIterator first, InputIterator last, const allocator_type& a = allocator_type()) : list_base(first, last, a) {} \
@@ -51,8 +43,8 @@ namespace std \
   }; \
 }
 
-// check whether we really need the std::vector specialization
-#if !(defined(_GLIBCXX_VECTOR) && (!EIGEN_GNUC_AT_LEAST(4,1))) /* Note that before gcc-4.1 we already have: std::list::resize(size_type,const T&). */
+// check whether we really need the std::list specialization
+#if !EIGEN_HAS_CXX11_CONTAINERS && !(defined(_GLIBCXX_LIST) && (!EIGEN_GNUC_AT_LEAST(4,1))) /* Note that before gcc-4.1 we already have: std::list::resize(size_type,const T&). */
 
 namespace std
 {
diff --git a/extern/Eigen3/Eigen/src/StlSupport/StdVector.h b/extern/Eigen3/Eigen/src/StlSupport/StdVector.h
index 1894af6b779..ec22821d261 100644
--- a/extern/Eigen3/Eigen/src/StlSupport/StdVector.h
+++ b/extern/Eigen3/Eigen/src/StlSupport/StdVector.h
@@ -44,6 +44,9 @@ namespace std \
   }; \
 }
 
+// Don't specialize if containers are implemented according to C++11
+#if !EIGEN_HAS_CXX11_CONTAINERS
+
 namespace std {
 
 #define EIGEN_STD_VECTOR_SPECIALIZATION_BODY \
@@ -77,7 +80,7 @@ namespace std {
   void resize(size_type new_size)
   { resize(new_size, T()); }
 
-#if defined(_VECTOR_) && (_MSC_VER<1910)
+#if defined(_VECTOR_)
   // workaround MSVC std::vector implementation
   void resize(size_type new_size, const value_type& x)
   {
@@ -110,7 +113,7 @@ namespace std {
       vector_base::insert(vector_base::end(), new_size - vector_base::size(), x);
   }
 #else
-  // either GCC 4.1, MSVC2017 or non-GCC
+  // either GCC 4.1 or non-GCC
   // default implementation which should always work.
   void resize(size_type new_size, const value_type& x)
   {
@@ -122,5 +125,7 @@ namespace std {
 #endif
   };
 }
+#endif // !EIGEN_HAS_CXX11_CONTAINERS
+
 
 #endif // EIGEN_STDVECTOR_H
diff --git a/extern/Eigen3/Eigen/src/StlSupport/details.h b/extern/Eigen3/Eigen/src/StlSupport/details.h
index d8debc7c4f8..2cfd13e03aa 100644
--- a/extern/Eigen3/Eigen/src/StlSupport/details.h
+++ b/extern/Eigen3/Eigen/src/StlSupport/details.h
@@ -22,13 +22,13 @@ namespace Eigen {
   class aligned_allocator_indirection : public EIGEN_ALIGNED_ALLOCATOR<T>
   {
   public:
-    typedef size_t    size_type;
-    typedef ptrdiff_t difference_type;
-    typedef T*        pointer;
-    typedef const T*  const_pointer;
-    typedef T&        reference;
-    typedef const T&  const_reference;
-    typedef T         value_type;
+    typedef std::size_t     size_type;
+    typedef std::ptrdiff_t  difference_type;
+    typedef T*              pointer;
+    typedef const T*        const_pointer;
+    typedef T&              reference;
+    typedef const T&        const_reference;
+    typedef T               value_type;
 
     template<class U>
     struct rebind
@@ -46,7 +46,7 @@ namespace Eigen {
     ~aligned_allocator_indirection() {}
   };
 
-#ifdef _MSC_VER
+#if EIGEN_COMP_MSVC
 
   // sometimes, MSVC detects, at compile time, that the argument x
   // in std::vector::resize(size_t s,T x) won't be aligned and generate an error
diff --git a/extern/Eigen3/Eigen/src/SuperLUSupport/SuperLUSupport.h b/extern/Eigen3/Eigen/src/SuperLUSupport/SuperLUSupport.h
index bcb355760c5..7261c7d0fc8 100644
--- a/extern/Eigen3/Eigen/src/SuperLUSupport/SuperLUSupport.h
+++ b/extern/Eigen3/Eigen/src/SuperLUSupport/SuperLUSupport.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -10,16 +10,16 @@
 #ifndef EIGEN_SUPERLUSUPPORT_H
 #define EIGEN_SUPERLUSUPPORT_H
 
-namespace Eigen { 
+namespace Eigen {
 
+#if defined(SUPERLU_MAJOR_VERSION) && (SUPERLU_MAJOR_VERSION >= 5)
 #define DECL_GSSVX(PREFIX,FLOATTYPE,KEYTYPE)		\
     extern "C" {                                                                                          \
-      typedef struct { FLOATTYPE for_lu; FLOATTYPE total_needed; int expansions; } PREFIX##mem_usage_t;   \
       extern void PREFIX##gssvx(superlu_options_t *, SuperMatrix *, int *, int *, int *,                  \
                                 char *, FLOATTYPE *, FLOATTYPE *, SuperMatrix *, SuperMatrix *,           \
                                 void *, int, SuperMatrix *, SuperMatrix *,                                \
                                 FLOATTYPE *, FLOATTYPE *, FLOATTYPE *, FLOATTYPE *,                       \
-                                PREFIX##mem_usage_t *, SuperLUStat_t *, int *);                           \
+                                GlobalLU_t *, mem_usage_t *, SuperLUStat_t *, int *);                     \
     }                                                                                                     \
     inline float SuperLU_gssvx(superlu_options_t *options, SuperMatrix *A,                                \
          int *perm_c, int *perm_r, int *etree, char *equed,                                               \
@@ -29,12 +29,37 @@ namespace Eigen {
          FLOATTYPE *recip_pivot_growth,                                                                   \
          FLOATTYPE *rcond, FLOATTYPE *ferr, FLOATTYPE *berr,                                              \
          SuperLUStat_t *stats, int *info, KEYTYPE) {                                                      \
-    PREFIX##mem_usage_t mem_usage;                                                                        \
+    mem_usage_t mem_usage;                                                                                \
+    GlobalLU_t gLU;                                                                                       \
+    PREFIX##gssvx(options, A, perm_c, perm_r, etree, equed, R, C, L,                                      \
+         U, work, lwork, B, X, recip_pivot_growth, rcond,                                                 \
+         ferr, berr, &gLU, &mem_usage, stats, info);                                                      \
+    return mem_usage.for_lu; /* bytes used by the factor storage */                                       \
+  }
+#else // version < 5.0
+#define DECL_GSSVX(PREFIX,FLOATTYPE,KEYTYPE)		\
+    extern "C" {                                                                                          \
+      extern void PREFIX##gssvx(superlu_options_t *, SuperMatrix *, int *, int *, int *,                  \
+                                char *, FLOATTYPE *, FLOATTYPE *, SuperMatrix *, SuperMatrix *,           \
+                                void *, int, SuperMatrix *, SuperMatrix *,                                \
+                                FLOATTYPE *, FLOATTYPE *, FLOATTYPE *, FLOATTYPE *,                       \
+                                mem_usage_t *, SuperLUStat_t *, int *);                                   \
+    }                                                                                                     \
+    inline float SuperLU_gssvx(superlu_options_t *options, SuperMatrix *A,                                \
+         int *perm_c, int *perm_r, int *etree, char *equed,                                               \
+         FLOATTYPE *R, FLOATTYPE *C, SuperMatrix *L,                                                      \
+         SuperMatrix *U, void *work, int lwork,                                                           \
+         SuperMatrix *B, SuperMatrix *X,                                                                  \
+         FLOATTYPE *recip_pivot_growth,                                                                   \
+         FLOATTYPE *rcond, FLOATTYPE *ferr, FLOATTYPE *berr,                                              \
+         SuperLUStat_t *stats, int *info, KEYTYPE) {                                                      \
+    mem_usage_t mem_usage;                                                                                \
     PREFIX##gssvx(options, A, perm_c, perm_r, etree, equed, R, C, L,                                      \
          U, work, lwork, B, X, recip_pivot_growth, rcond,                                                 \
          ferr, berr, &mem_usage, stats, info);                                                            \
     return mem_usage.for_lu; /* bytes used by the factor storage */                                       \
   }
+#endif
 
 DECL_GSSVX(s,float,float)
 DECL_GSSVX(c,float,std::complex<float>)
@@ -53,7 +78,7 @@ DECL_GSSVX(z,double,std::complex<double>)
       extern void PREFIX##gsisx(superlu_options_t *, SuperMatrix *, int *, int *, int *,        \
                          char *, FLOATTYPE *, FLOATTYPE *, SuperMatrix *, SuperMatrix *,        \
                          void *, int, SuperMatrix *, SuperMatrix *, FLOATTYPE *, FLOATTYPE *,   \
-                         PREFIX##mem_usage_t *, SuperLUStat_t *, int *);                        \
+                         mem_usage_t *, SuperLUStat_t *, int *);                        \
     }                                                                                           \
     inline float SuperLU_gsisx(superlu_options_t *options, SuperMatrix *A,                      \
          int *perm_c, int *perm_r, int *etree, char *equed,                                     \
@@ -63,7 +88,7 @@ DECL_GSSVX(z,double,std::complex<double>)
          FLOATTYPE *recip_pivot_growth,                                                         \
          FLOATTYPE *rcond,                                                                      \
          SuperLUStat_t *stats, int *info, KEYTYPE) {                                            \
-    PREFIX##mem_usage_t mem_usage;                                                              \
+    mem_usage_t mem_usage;                                                              \
     PREFIX##gsisx(options, A, perm_c, perm_r, etree, equed, R, C, L,                            \
          U, work, lwork, B, X, recip_pivot_growth, rcond,                                       \
          &mem_usage, stats, info);                                                              \
@@ -156,37 +181,38 @@ struct SluMatrix : SuperMatrix
     res.setScalarType<typename MatrixType::Scalar>();
     res.Mtype     = SLU_GE;
 
-    res.nrow      = mat.rows();
-    res.ncol      = mat.cols();
+    res.nrow      = internal::convert_index<int>(mat.rows());
+    res.ncol      = internal::convert_index<int>(mat.cols());
 
-    res.storage.lda       = MatrixType::IsVectorAtCompileTime ? mat.size() : mat.outerStride();
+    res.storage.lda       = internal::convert_index<int>(MatrixType::IsVectorAtCompileTime ? mat.size() : mat.outerStride());
     res.storage.values    = (void*)(mat.data());
     return res;
   }
 
   template<typename MatrixType>
-  static SluMatrix Map(SparseMatrixBase<MatrixType>& mat)
+  static SluMatrix Map(SparseMatrixBase<MatrixType>& a_mat)
   {
+    MatrixType &mat(a_mat.derived());
     SluMatrix res;
     if ((MatrixType::Flags&RowMajorBit)==RowMajorBit)
     {
       res.setStorageType(SLU_NR);
-      res.nrow      = mat.cols();
-      res.ncol      = mat.rows();
+      res.nrow      = internal::convert_index<int>(mat.cols());
+      res.ncol      = internal::convert_index<int>(mat.rows());
     }
     else
     {
       res.setStorageType(SLU_NC);
-      res.nrow      = mat.rows();
-      res.ncol      = mat.cols();
+      res.nrow      = internal::convert_index<int>(mat.rows());
+      res.ncol      = internal::convert_index<int>(mat.cols());
     }
 
     res.Mtype       = SLU_GE;
 
-    res.storage.nnz       = mat.nonZeros();
-    res.storage.values    = mat.derived().valuePtr();
-    res.storage.innerInd  = mat.derived().innerIndexPtr();
-    res.storage.outerInd  = mat.derived().outerIndexPtr();
+    res.storage.nnz       = internal::convert_index<int>(mat.nonZeros());
+    res.storage.values    = mat.valuePtr();
+    res.storage.innerInd  = mat.innerIndexPtr();
+    res.storage.outerInd  = mat.outerIndexPtr();
 
     res.setScalarType<typename MatrixType::Scalar>();
 
@@ -271,8 +297,8 @@ SluMatrix asSluMatrix(MatrixType& mat)
 template<typename Scalar, int Flags, typename Index>
 MappedSparseMatrix<Scalar,Flags,Index> map_superlu(SluMatrix& sluMat)
 {
-  eigen_assert((Flags&RowMajor)==RowMajor && sluMat.Stype == SLU_NR
-         || (Flags&ColMajor)==ColMajor && sluMat.Stype == SLU_NC);
+  eigen_assert(((Flags&RowMajor)==RowMajor && sluMat.Stype == SLU_NR)
+         || ((Flags&ColMajor)==ColMajor && sluMat.Stype == SLU_NC));
 
   Index outerSize = (Flags&RowMajor)==RowMajor ? sluMat.ncol : sluMat.nrow;
 
@@ -288,17 +314,26 @@ MappedSparseMatrix<Scalar,Flags,Index> map_superlu(SluMatrix& sluMat)
   * \brief The base class for the direct and incomplete LU factorization of SuperLU
   */
 template<typename _MatrixType, typename Derived>
-class SuperLUBase : internal::noncopyable
+class SuperLUBase : public SparseSolverBase<Derived>
 {
+  protected:
+    typedef SparseSolverBase<Derived> Base;
+    using Base::derived;
+    using Base::m_isInitialized;
   public:
     typedef _MatrixType MatrixType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar,Dynamic,1> Vector;
     typedef Matrix<int, 1, MatrixType::ColsAtCompileTime> IntRowVectorType;
     typedef Matrix<int, MatrixType::RowsAtCompileTime, 1> IntColVectorType;    
+    typedef Map<PermutationMatrix<Dynamic,Dynamic,int> > PermutationMap;
     typedef SparseMatrix<Scalar> LUMatrixType;
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
 
   public:
 
@@ -309,9 +344,6 @@ class SuperLUBase : internal::noncopyable
       clearFactors();
     }
     
-    Derived& derived() { return *static_cast<Derived*>(this); }
-    const Derived& derived() const { return *static_cast<const Derived*>(this); }
-    
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
     
@@ -335,33 +367,7 @@ class SuperLUBase : internal::noncopyable
       derived().analyzePattern(matrix);
       derived().factorize(matrix);
     }
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<SuperLUBase, Rhs> solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "SuperLU is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "SuperLU::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<SuperLUBase, Rhs>(*this, b.derived());
-    }
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<SuperLUBase, Rhs> solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "SuperLU is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "SuperLU::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<SuperLUBase, Rhs>(*this, b.derived());
-    }
-    
+
     /** Performs a symbolic decomposition on the sparcity of \a matrix.
       *
       * This function is particularly useful when solving for several problems having the same structure.
@@ -386,7 +392,7 @@ class SuperLUBase : internal::noncopyable
     {
       set_default_options(&this->m_sluOptions);
       
-      const int size = a.rows();
+      const Index size = a.rows();
       m_matrix = a;
 
       m_sluA = internal::asSluMatrix(m_matrix);
@@ -405,7 +411,7 @@ class SuperLUBase : internal::noncopyable
       m_sluB.storage.values = 0;
       m_sluB.nrow           = 0;
       m_sluB.ncol           = 0;
-      m_sluB.storage.lda    = size;
+      m_sluB.storage.lda    = internal::convert_index<int>(size);
       m_sluX                = m_sluB;
       
       m_extractedDataAreDirty = true;
@@ -453,7 +459,6 @@ class SuperLUBase : internal::noncopyable
     mutable char m_sluEqued;
 
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     int m_factorizationIsOk;
     int m_analysisIsOk;
     mutable bool m_extractedDataAreDirty;
@@ -473,7 +478,11 @@ class SuperLUBase : internal::noncopyable
   *
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \warning This class is only for the 4.x versions of SuperLU. The 3.x and 5.x versions are not supported.
+  *
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SparseLU
   */
 template<typename _MatrixType>
 class SuperLU : public SuperLUBase<_MatrixType,SuperLU<_MatrixType> >
@@ -483,18 +492,20 @@ class SuperLU : public SuperLUBase<_MatrixType,SuperLU<_MatrixType> >
     typedef _MatrixType MatrixType;
     typedef typename Base::Scalar Scalar;
     typedef typename Base::RealScalar RealScalar;
-    typedef typename Base::Index Index;
+    typedef typename Base::StorageIndex StorageIndex;
     typedef typename Base::IntRowVectorType IntRowVectorType;
-    typedef typename Base::IntColVectorType IntColVectorType;    
+    typedef typename Base::IntColVectorType IntColVectorType;   
+    typedef typename Base::PermutationMap PermutationMap;
     typedef typename Base::LUMatrixType LUMatrixType;
     typedef TriangularView<LUMatrixType, Lower|UnitDiag>  LMatrixType;
-    typedef TriangularView<LUMatrixType,  Upper>           UMatrixType;
+    typedef TriangularView<LUMatrixType,  Upper>          UMatrixType;
 
   public:
+    using Base::_solve_impl;
 
     SuperLU() : Base() { init(); }
 
-    SuperLU(const MatrixType& matrix) : Base()
+    explicit SuperLU(const MatrixType& matrix) : Base()
     {
       init();
       Base::compute(matrix);
@@ -525,11 +536,9 @@ class SuperLU : public SuperLUBase<_MatrixType,SuperLU<_MatrixType> >
       */
     void factorize(const MatrixType& matrix);
     
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
-    #endif // EIGEN_PARSED_BY_DOXYGEN
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
     
     inline const LMatrixType& matrixL() const
     {
@@ -637,12 +646,12 @@ void SuperLU<MatrixType>::factorize(const MatrixType& a)
 
 template<typename MatrixType>
 template<typename Rhs,typename Dest>
-void SuperLU<MatrixType>::_solve(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x) const
+void SuperLU<MatrixType>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x) const
 {
   eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or analyzePattern()/factorize()");
 
-  const int size = m_matrix.rows();
-  const int rhsCols = b.cols();
+  const Index size = m_matrix.rows();
+  const Index rhsCols = b.cols();
   eigen_assert(size==b.rows());
 
   m_sluOptions.Trans = NOTRANS;
@@ -652,8 +661,12 @@ void SuperLU<MatrixType>::_solve(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x)
 
   m_sluFerr.resize(rhsCols);
   m_sluBerr.resize(rhsCols);
-  m_sluB = SluMatrix::Map(b.const_cast_derived());
-  m_sluX = SluMatrix::Map(x.derived());
+  
+  Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b);
+  Ref<const Matrix<typename Dest::Scalar,Dynamic,Dynamic,ColMajor> > x_ref(x);
+  
+  m_sluB = SluMatrix::Map(b_ref.const_cast_derived());
+  m_sluX = SluMatrix::Map(x_ref.const_cast_derived());
   
   typename Rhs::PlainObject b_cpy;
   if(m_sluEqued!='N')
@@ -676,6 +689,10 @@ void SuperLU<MatrixType>::_solve(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x)
                 &m_sluFerr[0], &m_sluBerr[0],
                 &m_sluStat, &info, Scalar());
   StatFree(&m_sluStat);
+  
+  if(x.derived().data() != x_ref.data())
+    x = x_ref;
+  
   m_info = info==0 ? Success : NumericalIssue;
 }
 
@@ -699,7 +716,7 @@ void SuperLUBase<MatrixType,Derived>::extractData() const
     NCformat    *Ustore = static_cast<NCformat*>(m_sluU.Store);
     Scalar      *SNptr;
 
-    const int size = m_matrix.rows();
+    const Index size = m_matrix.rows();
     m_l.resize(size,size);
     m_l.resizeNonZeros(Lstore->nnz);
     m_u.resize(size,size);
@@ -791,6 +808,8 @@ typename SuperLU<MatrixType>::Scalar SuperLU<MatrixType>::determinant() const
         det *= m_u.valuePtr()[lastId];
     }
   }
+  if(PermutationMap(m_p.data(),m_p.size()).determinant()*PermutationMap(m_q.data(),m_q.size()).determinant()<0)
+    det = -det;
   if(m_sluEqued!='N')
     return det/m_sluRscale.prod()/m_sluCscale.prod();
   else
@@ -810,11 +829,13 @@ typename SuperLU<MatrixType>::Scalar SuperLU<MatrixType>::determinant() const
   * This class allows to solve for an approximate solution of A.X = B sparse linear problems via an incomplete LU factorization
   * using the SuperLU library. This class is aimed to be used as a preconditioner of the iterative linear solvers.
   *
-  * \warning This class requires SuperLU 4 or later.
+  * \warning This class is only for the 4.x versions of SuperLU. The 3.x and 5.x versions are not supported.
   *
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   *
-  * \sa \ref TutorialSparseDirectSolvers, class ConjugateGradient, class BiCGSTAB
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class IncompleteLUT, class ConjugateGradient, class BiCGSTAB
   */
 
 template<typename _MatrixType>
@@ -825,9 +846,9 @@ class SuperILU : public SuperLUBase<_MatrixType,SuperILU<_MatrixType> >
     typedef _MatrixType MatrixType;
     typedef typename Base::Scalar Scalar;
     typedef typename Base::RealScalar RealScalar;
-    typedef typename Base::Index Index;
 
   public:
+    using Base::_solve_impl;
 
     SuperILU() : Base() { init(); }
 
@@ -863,7 +884,7 @@ class SuperILU : public SuperLUBase<_MatrixType,SuperILU<_MatrixType> >
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
     #endif // EIGEN_PARSED_BY_DOXYGEN
     
   protected:
@@ -946,9 +967,10 @@ void SuperILU<MatrixType>::factorize(const MatrixType& a)
   m_factorizationIsOk = true;
 }
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 template<typename MatrixType>
 template<typename Rhs,typename Dest>
-void SuperILU<MatrixType>::_solve(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x) const
+void SuperILU<MatrixType>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x) const
 {
   eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or analyzePattern()/factorize()");
 
@@ -962,8 +984,12 @@ void SuperILU<MatrixType>::_solve(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x)
 
   m_sluFerr.resize(rhsCols);
   m_sluBerr.resize(rhsCols);
-  m_sluB = SluMatrix::Map(b.const_cast_derived());
-  m_sluX = SluMatrix::Map(x.derived());
+  
+  Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b);
+  Ref<const Matrix<typename Dest::Scalar,Dynamic,Dynamic,ColMajor> > x_ref(x);
+  
+  m_sluB = SluMatrix::Map(b_ref.const_cast_derived());
+  m_sluX = SluMatrix::Map(x_ref.const_cast_derived());
 
   typename Rhs::PlainObject b_cpy;
   if(m_sluEqued!='N')
@@ -986,40 +1012,15 @@ void SuperILU<MatrixType>::_solve(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x)
                 &recip_pivot_growth, &rcond,
                 &m_sluStat, &info, Scalar());
   StatFree(&m_sluStat);
+  
+  if(x.derived().data() != x_ref.data())
+    x = x_ref;
 
   m_info = info==0 ? Success : NumericalIssue;
 }
 #endif
 
-namespace internal {
-  
-template<typename _MatrixType, typename Derived, typename Rhs>
-struct solve_retval<SuperLUBase<_MatrixType,Derived>, Rhs>
-  : solve_retval_base<SuperLUBase<_MatrixType,Derived>, Rhs>
-{
-  typedef SuperLUBase<_MatrixType,Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec().derived()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, typename Derived, typename Rhs>
-struct sparse_solve_retval<SuperLUBase<_MatrixType,Derived>, Rhs>
-  : sparse_solve_retval_base<SuperLUBase<_MatrixType,Derived>, Rhs>
-{
-  typedef SuperLUBase<_MatrixType,Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
+#endif
 
 } // end namespace Eigen
 
diff --git a/extern/Eigen3/Eigen/src/UmfPackSupport/UmfPackSupport.h b/extern/Eigen3/Eigen/src/UmfPackSupport/UmfPackSupport.h
index 29c60c37875..91c09ab1338 100644
--- a/extern/Eigen3/Eigen/src/UmfPackSupport/UmfPackSupport.h
+++ b/extern/Eigen3/Eigen/src/UmfPackSupport/UmfPackSupport.h
@@ -10,12 +10,37 @@
 #ifndef EIGEN_UMFPACKSUPPORT_H
 #define EIGEN_UMFPACKSUPPORT_H
 
-namespace Eigen { 
+namespace Eigen {
 
 /* TODO extract L, extract U, compute det, etc... */
 
 // generic double/complex<double> wrapper functions:
 
+
+inline void umfpack_defaults(double control[UMFPACK_CONTROL], double)
+{ umfpack_di_defaults(control); }
+
+inline void umfpack_defaults(double control[UMFPACK_CONTROL], std::complex<double>)
+{ umfpack_zi_defaults(control); }
+
+inline void umfpack_report_info(double control[UMFPACK_CONTROL], double info[UMFPACK_INFO], double)
+{ umfpack_di_report_info(control, info);}
+
+inline void umfpack_report_info(double control[UMFPACK_CONTROL], double info[UMFPACK_INFO], std::complex<double>)
+{ umfpack_zi_report_info(control, info);}
+
+inline void umfpack_report_status(double control[UMFPACK_CONTROL], int status, double)
+{ umfpack_di_report_status(control, status);}
+
+inline void umfpack_report_status(double control[UMFPACK_CONTROL], int status, std::complex<double>)
+{ umfpack_zi_report_status(control, status);}
+
+inline void umfpack_report_control(double control[UMFPACK_CONTROL], double)
+{ umfpack_di_report_control(control);}
+
+inline void umfpack_report_control(double control[UMFPACK_CONTROL], std::complex<double>)
+{ umfpack_zi_report_control(control);}
+
 inline void umfpack_free_numeric(void **Numeric, double)
 { umfpack_di_free_numeric(Numeric); *Numeric = 0; }
 
@@ -107,15 +132,6 @@ inline int umfpack_get_determinant(std::complex<double> *Mx, double *Ex, void *N
   return umfpack_zi_get_determinant(&mx_real,0,Ex,NumericHandle,User_Info);
 }
 
-namespace internal {
-  template<typename T> struct umfpack_helper_is_sparse_plain : false_type {};
-  template<typename Scalar, int Options, typename StorageIndex>
-  struct umfpack_helper_is_sparse_plain<SparseMatrix<Scalar,Options,StorageIndex> >
-    : true_type {};
-  template<typename Scalar, int Options, typename StorageIndex>
-  struct umfpack_helper_is_sparse_plain<MappedSparseMatrix<Scalar,Options,StorageIndex> >
-    : true_type {};
-}
 
 /** \ingroup UmfPackSupport_Module
   * \brief A sparse LU factorization and solver based on UmfPack
@@ -128,27 +144,47 @@ namespace internal {
   * Otherwise an expensive copy will be made. You can call the inexpensive makeCompressed() to get a compressed matrix.
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SparseLU
   */
 template<typename _MatrixType>
-class UmfPackLU : internal::noncopyable
+class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
 {
+  protected:
+    typedef SparseSolverBase<UmfPackLU<_MatrixType> > Base;
+    using Base::m_isInitialized;
   public:
+    using Base::_solve_impl;
     typedef _MatrixType MatrixType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar,Dynamic,1> Vector;
     typedef Matrix<int, 1, MatrixType::ColsAtCompileTime> IntRowVectorType;
     typedef Matrix<int, MatrixType::RowsAtCompileTime, 1> IntColVectorType;
     typedef SparseMatrix<Scalar> LUMatrixType;
     typedef SparseMatrix<Scalar,ColMajor,int> UmfpackMatrixType;
+    typedef Ref<const UmfpackMatrixType, StandardCompressedFormat> UmfpackMatrixRef;
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
 
   public:
 
-    UmfPackLU() { init(); }
+    typedef Array<double, UMFPACK_CONTROL, 1> UmfpackControl;
+    typedef Array<double, UMFPACK_INFO, 1> UmfpackInfo;
 
-    UmfPackLU(const MatrixType& matrix)
+    UmfPackLU()
+      : m_dummy(0,0), mp_matrix(m_dummy)
+    {
+      init();
+    }
+
+    template<typename InputMatrixType>
+    explicit UmfPackLU(const InputMatrixType& matrix)
+      : mp_matrix(matrix)
     {
       init();
       compute(matrix);
@@ -160,8 +196,8 @@ class UmfPackLU : internal::noncopyable
       if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
     }
 
-    inline Index rows() const { return m_copyMatrix.rows(); }
-    inline Index cols() const { return m_copyMatrix.cols(); }
+    inline Index rows() const { return mp_matrix.rows(); }
+    inline Index cols() const { return mp_matrix.cols(); }
 
     /** \brief Reports whether previous computation was successful.
       *
@@ -198,7 +234,7 @@ class UmfPackLU : internal::noncopyable
       return m_q;
     }
 
-    /** Computes the sparse Cholesky decomposition of \a matrix 
+    /** Computes the sparse Cholesky decomposition of \a matrix
      *  Note that the matrix should be column-major, and in compressed format for best performance.
      *  \sa SparseMatrix::makeCompressed().
      */
@@ -207,52 +243,59 @@ class UmfPackLU : internal::noncopyable
     {
       if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
       if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
-      grapInput(matrix.derived());
+      grab(matrix.derived());
       analyzePattern_impl();
       factorize_impl();
     }
 
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
+    /** Performs a symbolic decomposition on the sparcity of \a matrix.
       *
-      * \sa compute()
+      * This function is particularly useful when solving for several problems having the same structure.
+      *
+      * \sa factorize(), compute()
       */
-    template<typename Rhs>
-    inline const internal::solve_retval<UmfPackLU, Rhs> solve(const MatrixBase<Rhs>& b) const
+    template<typename InputMatrixType>
+    void analyzePattern(const InputMatrixType& matrix)
     {
-      eigen_assert(m_isInitialized && "UmfPackLU is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "UmfPackLU::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<UmfPackLU, Rhs>(*this, b.derived());
+      if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
+      if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
+
+      grab(matrix.derived());
+
+      analyzePattern_impl();
     }
 
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
+    /** Provides the return status code returned by UmfPack during the numeric
+      * factorization.
       *
-      * \sa compute()
+      * \sa factorize(), compute()
       */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<UmfPackLU, Rhs> solve(const SparseMatrixBase<Rhs>& b) const
+    inline int umfpackFactorizeReturncode() const
     {
-      eigen_assert(m_isInitialized && "UmfPackLU is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "UmfPackLU::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<UmfPackLU, Rhs>(*this, b.derived());
+      eigen_assert(m_numeric && "UmfPackLU: you must first call factorize()");
+      return m_fact_errorCode;
     }
 
-    /** Performs a symbolic decomposition on the sparcity of \a matrix.
+    /** Provides access to the control settings array used by UmfPack.
       *
-      * This function is particularly useful when solving for several problems having the same structure.
+      * If this array contains NaN's, the default values are used.
       *
-      * \sa factorize(), compute()
+      * See UMFPACK documentation for details.
       */
-    template<typename InputMatrixType>
-    void analyzePattern(const InputMatrixType& matrix)
+    inline const UmfpackControl& umfpackControl() const
     {
-      if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
-      if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
-      
-      grapInput(matrix.derived());
+      return m_control;
+    }
 
-      analyzePattern_impl();
+    /** Provides access to the control settings array used by UmfPack.
+      *
+      * If this array contains NaN's, the default values are used.
+      *
+      * See UMFPACK documentation for details.
+      */
+    inline UmfpackControl& umfpackControl()
+    {
+      return m_control;
     }
 
     /** Performs a numeric decomposition of \a matrix
@@ -268,16 +311,42 @@ class UmfPackLU : internal::noncopyable
       if(m_numeric)
         umfpack_free_numeric(&m_numeric,Scalar());
 
-      grapInput(matrix.derived());
-      
+      grab(matrix.derived());
+
       factorize_impl();
     }
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** Prints the current UmfPack control settings.
+      *
+      * \sa umfpackControl()
+      */
+    void umfpackReportControl()
+    {
+      umfpack_report_control(m_control.data(), Scalar());
+    }
+
+    /** Prints statistics collected by UmfPack.
+      *
+      * \sa analyzePattern(), compute()
+      */
+    void umfpackReportInfo()
+    {
+      eigen_assert(m_analysisIsOk && "UmfPackLU: you must first call analyzePattern()");
+      umfpack_report_info(m_control.data(), m_umfpackInfo.data(), Scalar());
+    }
+
+    /** Prints the status of the previous factorization operation performed by UmfPack (symbolic or numerical factorization).
+      *
+      * \sa analyzePattern(), compute()
+      */
+    void umfpackReportStatus() {
+      eigen_assert(m_analysisIsOk && "UmfPackLU: you must first call analyzePattern()");
+      umfpack_report_status(m_control.data(), m_fact_errorCode, Scalar());
+    }
+
     /** \internal */
     template<typename BDerived,typename XDerived>
-    bool _solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const;
-    #endif
+    bool _solve_impl(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const;
 
     Scalar determinant() const;
 
@@ -291,92 +360,75 @@ class UmfPackLU : internal::noncopyable
       m_isInitialized         = false;
       m_numeric               = 0;
       m_symbolic              = 0;
-      m_outerIndexPtr         = 0;
-      m_innerIndexPtr         = 0;
-      m_valuePtr              = 0;
       m_extractedDataAreDirty = true;
+
+      umfpack_defaults(m_control.data(), Scalar());
     }
-    
-    template<typename InputMatrixType>
-    void grapInput_impl(const InputMatrixType& mat, internal::true_type)
-    {
-      m_copyMatrix.resize(mat.rows(), mat.cols());
-      if( ((MatrixType::Flags&RowMajorBit)==RowMajorBit) || sizeof(typename MatrixType::Index)!=sizeof(int) || !mat.isCompressed() )
-      {
-        // non supported input -> copy
-        m_copyMatrix = mat;
-        m_outerIndexPtr = m_copyMatrix.outerIndexPtr();
-        m_innerIndexPtr = m_copyMatrix.innerIndexPtr();
-        m_valuePtr      = m_copyMatrix.valuePtr();
-      }
-      else
-      {
-        m_outerIndexPtr = mat.outerIndexPtr();
-        m_innerIndexPtr = mat.innerIndexPtr();
-        m_valuePtr      = mat.valuePtr();
-      }
-    }
-    
-    template<typename InputMatrixType>
-    void grapInput_impl(const InputMatrixType& mat, internal::false_type)
-    {
-      m_copyMatrix = mat;
-      m_outerIndexPtr = m_copyMatrix.outerIndexPtr();
-      m_innerIndexPtr = m_copyMatrix.innerIndexPtr();
-      m_valuePtr      = m_copyMatrix.valuePtr();
-    }
-    
-    template<typename InputMatrixType>
-    void grapInput(const InputMatrixType& mat)
-    {
-      grapInput_impl(mat, internal::umfpack_helper_is_sparse_plain<InputMatrixType>());
-    }
-    
+
     void analyzePattern_impl()
     {
-      int errorCode = 0;
-      errorCode = umfpack_symbolic(m_copyMatrix.rows(), m_copyMatrix.cols(), m_outerIndexPtr, m_innerIndexPtr, m_valuePtr,
-                                   &m_symbolic, 0, 0);
+      m_fact_errorCode = umfpack_symbolic(internal::convert_index<int>(mp_matrix.rows()),
+                                          internal::convert_index<int>(mp_matrix.cols()),
+                                          mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
+                                          &m_symbolic, m_control.data(), m_umfpackInfo.data());
 
       m_isInitialized = true;
-      m_info = errorCode ? InvalidInput : Success;
+      m_info = m_fact_errorCode ? InvalidInput : Success;
       m_analysisIsOk = true;
       m_factorizationIsOk = false;
       m_extractedDataAreDirty = true;
     }
-    
+
     void factorize_impl()
     {
-      int errorCode;
-      errorCode = umfpack_numeric(m_outerIndexPtr, m_innerIndexPtr, m_valuePtr,
-                                  m_symbolic, &m_numeric, 0, 0);
 
-      m_info = errorCode ? NumericalIssue : Success;
+      m_fact_errorCode = umfpack_numeric(mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
+                                         m_symbolic, &m_numeric, m_control.data(), m_umfpackInfo.data());
+
+      m_info = m_fact_errorCode == UMFPACK_OK ? Success : NumericalIssue;
       m_factorizationIsOk = true;
       m_extractedDataAreDirty = true;
     }
 
+    template<typename MatrixDerived>
+    void grab(const EigenBase<MatrixDerived> &A)
+    {
+      mp_matrix.~UmfpackMatrixRef();
+      ::new (&mp_matrix) UmfpackMatrixRef(A.derived());
+    }
+
+    void grab(const UmfpackMatrixRef &A)
+    {
+      if(&(A.derived()) != &mp_matrix)
+      {
+        mp_matrix.~UmfpackMatrixRef();
+        ::new (&mp_matrix) UmfpackMatrixRef(A);
+      }
+    }
+
     // cached data to reduce reallocation, etc.
     mutable LUMatrixType m_l;
+    int m_fact_errorCode;
+    UmfpackControl m_control;
+    mutable UmfpackInfo m_umfpackInfo;
+
     mutable LUMatrixType m_u;
     mutable IntColVectorType m_p;
     mutable IntRowVectorType m_q;
 
-    UmfpackMatrixType m_copyMatrix;
-    const Scalar* m_valuePtr;
-    const int* m_outerIndexPtr;
-    const int* m_innerIndexPtr;
+    UmfpackMatrixType m_dummy;
+    UmfpackMatrixRef mp_matrix;
+
     void* m_numeric;
     void* m_symbolic;
 
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     int m_factorizationIsOk;
     int m_analysisIsOk;
     mutable bool m_extractedDataAreDirty;
-    
+
   private:
-    UmfPackLU(UmfPackLU& ) { }
+    UmfPackLU(const UmfPackLU& ) { }
 };
 
 
@@ -418,19 +470,30 @@ typename UmfPackLU<MatrixType>::Scalar UmfPackLU<MatrixType>::determinant() cons
 
 template<typename MatrixType>
 template<typename BDerived,typename XDerived>
-bool UmfPackLU<MatrixType>::_solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const
+bool UmfPackLU<MatrixType>::_solve_impl(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const
 {
-  const int rhsCols = b.cols();
+  Index rhsCols = b.cols();
   eigen_assert((BDerived::Flags&RowMajorBit)==0 && "UmfPackLU backend does not support non col-major rhs yet");
   eigen_assert((XDerived::Flags&RowMajorBit)==0 && "UmfPackLU backend does not support non col-major result yet");
   eigen_assert(b.derived().data() != x.derived().data() && " Umfpack does not support inplace solve");
-  
+
   int errorCode;
+  Scalar* x_ptr = 0;
+  Matrix<Scalar,Dynamic,1> x_tmp;
+  if(x.innerStride()!=1)
+  {
+    x_tmp.resize(x.rows());
+    x_ptr = x_tmp.data();
+  }
   for (int j=0; j<rhsCols; ++j)
   {
+    if(x.innerStride()==1)
+      x_ptr = &x.col(j).coeffRef(0);
     errorCode = umfpack_solve(UMFPACK_A,
-        m_outerIndexPtr, m_innerIndexPtr, m_valuePtr,
-        &x.col(j).coeffRef(0), &b.const_cast_derived().col(j).coeffRef(0), m_numeric, 0, 0);
+        mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
+        x_ptr, &b.const_cast_derived().col(j).coeffRef(0), m_numeric, m_control.data(), m_umfpackInfo.data());
+    if(x.innerStride()!=1)
+      x.col(j) = x_tmp;
     if (errorCode!=0)
       return false;
   }
@@ -438,37 +501,6 @@ bool UmfPackLU<MatrixType>::_solve(const MatrixBase<BDerived> &b, MatrixBase<XDe
   return true;
 }
 
-
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<UmfPackLU<_MatrixType>, Rhs>
-  : solve_retval_base<UmfPackLU<_MatrixType>, Rhs>
-{
-  typedef UmfPackLU<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, typename Rhs>
-struct sparse_solve_retval<UmfPackLU<_MatrixType>, Rhs>
-  : sparse_solve_retval_base<UmfPackLU<_MatrixType>, Rhs>
-{
-  typedef UmfPackLU<_MatrixType> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_UMFPACKSUPPORT_H
diff --git a/extern/Eigen3/Eigen/src/misc/Image.h b/extern/Eigen3/Eigen/src/misc/Image.h
index 75c5f433a8a..b8b8a045529 100644
--- a/extern/Eigen3/Eigen/src/misc/Image.h
+++ b/extern/Eigen3/Eigen/src/misc/Image.h
@@ -38,7 +38,6 @@ template<typename _DecompositionType> struct image_retval_base
   typedef _DecompositionType DecompositionType;
   typedef typename DecompositionType::MatrixType MatrixType;
   typedef ReturnByValue<image_retval_base> Base;
-  typedef typename Base::Index Index;
 
   image_retval_base(const DecompositionType& dec, const MatrixType& originalMatrix)
     : m_dec(dec), m_rank(dec.rank()),
@@ -69,7 +68,6 @@ template<typename _DecompositionType> struct image_retval_base
   typedef typename DecompositionType::MatrixType MatrixType; \
   typedef typename MatrixType::Scalar Scalar; \
   typedef typename MatrixType::RealScalar RealScalar; \
-  typedef typename MatrixType::Index Index; \
   typedef Eigen::internal::image_retval_base<DecompositionType> Base; \
   using Base::dec; \
   using Base::originalMatrix; \
diff --git a/extern/Eigen3/Eigen/src/misc/Kernel.h b/extern/Eigen3/Eigen/src/misc/Kernel.h
index b9e1518fd49..bef5d6ff583 100644
--- a/extern/Eigen3/Eigen/src/misc/Kernel.h
+++ b/extern/Eigen3/Eigen/src/misc/Kernel.h
@@ -39,9 +39,8 @@ template<typename _DecompositionType> struct kernel_retval_base
 {
   typedef _DecompositionType DecompositionType;
   typedef ReturnByValue<kernel_retval_base> Base;
-  typedef typename Base::Index Index;
 
-  kernel_retval_base(const DecompositionType& dec)
+  explicit kernel_retval_base(const DecompositionType& dec)
     : m_dec(dec),
       m_rank(dec.rank()),
       m_cols(m_rank==dec.cols() ? 1 : dec.cols() - m_rank)
@@ -68,7 +67,6 @@ template<typename _DecompositionType> struct kernel_retval_base
   typedef typename DecompositionType::MatrixType MatrixType; \
   typedef typename MatrixType::Scalar Scalar; \
   typedef typename MatrixType::RealScalar RealScalar; \
-  typedef typename MatrixType::Index Index; \
   typedef Eigen::internal::kernel_retval_base<DecompositionType> Base; \
   using Base::dec; \
   using Base::rank; \
diff --git a/extern/Eigen3/Eigen/src/misc/RealSvd2x2.h b/extern/Eigen3/Eigen/src/misc/RealSvd2x2.h
new file mode 100644
index 00000000000..abb4d3c2fc7
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/misc/RealSvd2x2.h
@@ -0,0 +1,55 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2013-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_REALSVD2X2_H
+#define EIGEN_REALSVD2X2_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<typename MatrixType, typename RealScalar, typename Index>
+void real_2x2_jacobi_svd(const MatrixType& matrix, Index p, Index q,
+                         JacobiRotation<RealScalar> *j_left,
+                         JacobiRotation<RealScalar> *j_right)
+{
+  using std::sqrt;
+  using std::abs;
+  Matrix<RealScalar,2,2> m;
+  m << numext::real(matrix.coeff(p,p)), numext::real(matrix.coeff(p,q)),
+       numext::real(matrix.coeff(q,p)), numext::real(matrix.coeff(q,q));
+  JacobiRotation<RealScalar> rot1;
+  RealScalar t = m.coeff(0,0) + m.coeff(1,1);
+  RealScalar d = m.coeff(1,0) - m.coeff(0,1);
+
+  if(abs(d) < (std::numeric_limits<RealScalar>::min)())
+  {
+    rot1.s() = RealScalar(0);
+    rot1.c() = RealScalar(1);
+  }
+  else
+  {
+    // If d!=0, then t/d cannot overflow because the magnitude of the
+    // entries forming d are not too small compared to the ones forming t.
+    RealScalar u = t / d;
+    RealScalar tmp = sqrt(RealScalar(1) + numext::abs2(u));
+    rot1.s() = RealScalar(1) / tmp;
+    rot1.c() = u / tmp;
+  }
+  m.applyOnTheLeft(0,1,rot1);
+  j_right->makeJacobi(m,0,1);
+  *j_left = rot1 * j_right->transpose();
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_REALSVD2X2_H
diff --git a/extern/Eigen3/Eigen/src/misc/blas.h b/extern/Eigen3/Eigen/src/misc/blas.h
index 6fce99ed5c4..25215b15e8a 100644
--- a/extern/Eigen3/Eigen/src/misc/blas.h
+++ b/extern/Eigen3/Eigen/src/misc/blas.h
@@ -30,15 +30,15 @@ int  BLASFUNC(cdotcw)  (int *, float  *, int *, float  *, int *, float*);
 int  BLASFUNC(zdotuw)  (int *, double  *, int *, double  *, int *, double*);
 int  BLASFUNC(zdotcw)  (int *, double  *, int *, double  *, int *, double*);
 
-int    BLASFUNC(saxpy) (int *, float  *, float  *, int *, float  *, int *);
-int    BLASFUNC(daxpy) (int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(qaxpy) (int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(caxpy) (int *, float  *, float  *, int *, float  *, int *);
-int    BLASFUNC(zaxpy) (int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(xaxpy) (int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(caxpyc)(int *, float  *, float  *, int *, float  *, int *);
-int    BLASFUNC(zaxpyc)(int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(xaxpyc)(int *, double *, double *, int *, double *, int *);
+int    BLASFUNC(saxpy) (const int *, const float  *, const float  *, const int *, float  *, const int *);
+int    BLASFUNC(daxpy) (const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(qaxpy) (const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(caxpy) (const int *, const float  *, const float  *, const int *, float  *, const int *);
+int    BLASFUNC(zaxpy) (const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(xaxpy) (const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(caxpyc)(const int *, const float  *, const float  *, const int *, float  *, const int *);
+int    BLASFUNC(zaxpyc)(const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(xaxpyc)(const int *, const double *, const double *, const int *, double *, const int *);
 
 int    BLASFUNC(scopy) (int *, float  *, int *, float  *, int *);
 int    BLASFUNC(dcopy) (int *, double *, int *, double *, int *);
@@ -177,31 +177,19 @@ int BLASFUNC(xgeru)(int *,    int *, double *, double *, int *,
 int BLASFUNC(xgerc)(int *,    int *, double *, double *, int *,
 		    double *, int *, double *, int *);
 
-int BLASFUNC(sgemv)(char *, int *, int *, float  *, float  *, int *,
-		    float  *, int *, float  *, float  *, int *);
-int BLASFUNC(dgemv)(char *, int *, int *, double *, double *, int *,
-		    double *, int *, double *, double *, int *);
-int BLASFUNC(qgemv)(char *, int *, int *, double *, double *, int *,
-		    double *, int *, double *, double *, int *);
-int BLASFUNC(cgemv)(char *, int *, int *, float  *, float  *, int *,
-		    float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zgemv)(char *, int *, int *, double *, double *, int *,
-		    double *, int *, double *, double *, int *);
-int BLASFUNC(xgemv)(char *, int *, int *, double *, double *, int *,
-		    double *, int *, double *, double *, int *);
+int BLASFUNC(sgemv)(const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dgemv)(const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qgemv)(const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(cgemv)(const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zgemv)(const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xgemv)(const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
-int BLASFUNC(strsv) (char *, char *, char *, int *, float  *, int *,
-		     float  *, int *);
-int BLASFUNC(dtrsv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(qtrsv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(ctrsv) (char *, char *, char *, int *, float  *, int *,
-		     float  *, int *);
-int BLASFUNC(ztrsv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(xtrsv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
+int BLASFUNC(strsv) (const char *, const char *, const char *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(dtrsv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(qtrsv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(ctrsv) (const char *, const char *, const char *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(ztrsv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(xtrsv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
 
 int BLASFUNC(stpsv) (char *, char *, char *, int *, float  *, float  *, int *);
 int BLASFUNC(dtpsv) (char *, char *, char *, int *, double *, double *, int *);
@@ -210,18 +198,12 @@ int BLASFUNC(ctpsv) (char *, char *, char *, int *, float  *, float  *, int *);
 int BLASFUNC(ztpsv) (char *, char *, char *, int *, double *, double *, int *);
 int BLASFUNC(xtpsv) (char *, char *, char *, int *, double *, double *, int *);
 
-int BLASFUNC(strmv) (char *, char *, char *, int *, float  *, int *,
-		     float  *, int *);
-int BLASFUNC(dtrmv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(qtrmv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(ctrmv) (char *, char *, char *, int *, float  *, int *,
-		     float  *, int *);
-int BLASFUNC(ztrmv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(xtrmv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
+int BLASFUNC(strmv) (const char *, const char *, const char *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(dtrmv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(qtrmv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(ctrmv) (const char *, const char *, const char *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(ztrmv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(xtrmv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
 
 int BLASFUNC(stpmv) (char *, char *, char *, int *, float  *, float  *, int *);
 int BLASFUNC(dtpmv) (char *, char *, char *, int *, double *, double *, int *);
@@ -244,18 +226,9 @@ int BLASFUNC(ctbsv) (char *, char *, char *, int *, int *, float  *, int *, floa
 int BLASFUNC(ztbsv) (char *, char *, char *, int *, int *, double *, int *, double *, int *);
 int BLASFUNC(xtbsv) (char *, char *, char *, int *, int *, double *, int *, double *, int *);
 
-int BLASFUNC(ssymv) (char *, int *, float  *, float *, int *,
-		     float  *, int *, float *, float *, int *);
-int BLASFUNC(dsymv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(qsymv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(csymv) (char *, int *, float  *, float *, int *,
-		     float  *, int *, float *, float *, int *);
-int BLASFUNC(zsymv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(xsymv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
+int BLASFUNC(ssymv) (const char *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dsymv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qsymv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
 int BLASFUNC(sspmv) (char *, int *, float  *, float *,
 		     float  *, int *, float *, float *, int *);
@@ -263,38 +236,17 @@ int BLASFUNC(dspmv) (char *, int *, double  *, double *,
 		     double  *, int *, double *, double *, int *);
 int BLASFUNC(qspmv) (char *, int *, double  *, double *,
 		     double  *, int *, double *, double *, int *);
-int BLASFUNC(cspmv) (char *, int *, float  *, float *,
-		     float  *, int *, float *, float *, int *);
-int BLASFUNC(zspmv) (char *, int *, double  *, double *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(xspmv) (char *, int *, double  *, double *,
-		     double  *, int *, double *, double *, int *);
 
-int BLASFUNC(ssyr) (char *, int *, float   *, float  *, int *,
-		    float  *, int *);
-int BLASFUNC(dsyr) (char *, int *, double  *, double *, int *,
-		    double *, int *);
-int BLASFUNC(qsyr) (char *, int *, double  *, double *, int *,
-		    double *, int *);
-int BLASFUNC(csyr) (char *, int *, float   *, float  *, int *,
-		    float  *, int *);
-int BLASFUNC(zsyr) (char *, int *, double  *, double *, int *,
-		    double *, int *);
-int BLASFUNC(xsyr) (char *, int *, double  *, double *, int *,
-		    double *, int *);
+int BLASFUNC(ssyr) (const char *, const int *, const float   *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(dsyr) (const char *, const int *, const double  *, const double *, const int *, double *, const int *);
+int BLASFUNC(qsyr) (const char *, const int *, const double  *, const double *, const int *, double *, const int *);
 
-int BLASFUNC(ssyr2) (char *, int *, float   *,
-		     float  *, int *, float  *, int *, float  *, int *);
-int BLASFUNC(dsyr2) (char *, int *, double  *,
-		     double *, int *, double *, int *, double *, int *);
-int BLASFUNC(qsyr2) (char *, int *, double  *,
-		     double *, int *, double *, int *, double *, int *);
-int BLASFUNC(csyr2) (char *, int *, float   *,
-		     float  *, int *, float  *, int *, float  *, int *);
-int BLASFUNC(zsyr2) (char *, int *, double  *,
-		     double *, int *, double *, int *, double *, int *);
-int BLASFUNC(xsyr2) (char *, int *, double  *,
-		     double *, int *, double *, int *, double *, int *);
+int BLASFUNC(ssyr2) (const char *, const int *, const float   *, const float  *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(dsyr2) (const char *, const int *, const double  *, const double *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(qsyr2) (const char *, const int *, const double  *, const double *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(csyr2) (const char *, const int *, const float   *, const float  *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(zsyr2) (const char *, const int *, const double  *, const double *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(xsyr2) (const char *, const int *, const double  *, const double *, const int *, const double *, const int *, double *, const int *);
 
 int BLASFUNC(sspr) (char *, int *, float   *, float  *, int *,
 		    float  *);
@@ -302,12 +254,6 @@ int BLASFUNC(dspr) (char *, int *, double  *, double *, int *,
 		    double *);
 int BLASFUNC(qspr) (char *, int *, double  *, double *, int *,
 		    double *);
-int BLASFUNC(cspr) (char *, int *, float   *, float  *, int *,
-		    float  *);
-int BLASFUNC(zspr) (char *, int *, double  *, double *, int *,
-		    double *);
-int BLASFUNC(xspr) (char *, int *, double  *, double *, int *,
-		    double *);
 
 int BLASFUNC(sspr2) (char *, int *, float   *,
 		     float  *, int *, float  *, int *, float  *);
@@ -347,12 +293,9 @@ int BLASFUNC(zhpr2) (char *, int *, double  *,
 int BLASFUNC(xhpr2) (char *, int *, double  *,
 		     double *, int *, double *, int *, double *);
 
-int BLASFUNC(chemv) (char *, int *, float  *, float *, int *,
-		     float  *, int *, float *, float *, int *);
-int BLASFUNC(zhemv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(xhemv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
+int BLASFUNC(chemv) (const char *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zhemv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xhemv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
 int BLASFUNC(chpmv) (char *, int *, float  *, float *,
 		     float  *, int *, float *, float *, int *);
@@ -401,18 +344,12 @@ int BLASFUNC(xhbmv)(char *, int *, int *, double *, double *, int *,
 
 /* Level 3 routines */
 
-int BLASFUNC(sgemm)(char *, char *, int *, int *, int *, float *,
-	   float  *, int *, float  *, int *, float  *, float  *, int *);
-int BLASFUNC(dgemm)(char *, char *, int *, int *, int *, double *,
-	   double *, int *, double *, int *, double *, double *, int *);
-int BLASFUNC(qgemm)(char *, char *, int *, int *, int *, double *,
-	   double *, int *, double *, int *, double *, double *, int *);
-int BLASFUNC(cgemm)(char *, char *, int *, int *, int *, float *,
-	   float  *, int *, float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zgemm)(char *, char *, int *, int *, int *, double *,
-	   double *, int *, double *, int *, double *, double *, int *);
-int BLASFUNC(xgemm)(char *, char *, int *, int *, int *, double *,
-	   double *, int *, double *, int *, double *, double *, int *);
+int BLASFUNC(sgemm)(const char *, const char *, const int *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dgemm)(const char *, const char *, const int *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qgemm)(const char *, const char *, const int *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(cgemm)(const char *, const char *, const int *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zgemm)(const char *, const char *, const int *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xgemm)(const char *, const char *, const int *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
 int BLASFUNC(cgemm3m)(char *, char *, int *, int *, int *, float *,
 	   float  *, int *, float  *, int *, float  *, float  *, int *);
@@ -434,84 +371,48 @@ int BLASFUNC(zge2mm)(char *, char *, char *, int *, int *,
 		     double *, double  *, int *, double  *, int *,
 		     double *, double  *, int *);
 
-int BLASFUNC(strsm)(char *, char *, char *, char *, int *, int *,
-	   float *,  float *, int *, float *, int *);
-int BLASFUNC(dtrsm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(qtrsm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(ctrsm)(char *, char *, char *, char *, int *, int *,
-	   float *,  float *, int *, float *, int *);
-int BLASFUNC(ztrsm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(xtrsm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-
-int BLASFUNC(strmm)(char *, char *, char *, char *, int *, int *,
-	   float *,  float *, int *, float *, int *);
-int BLASFUNC(dtrmm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(qtrmm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(ctrmm)(char *, char *, char *, char *, int *, int *,
-	   float *,  float *, int *, float *, int *);
-int BLASFUNC(ztrmm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(xtrmm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-
-int BLASFUNC(ssymm)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, int *, float  *, float  *, int *);
-int BLASFUNC(dsymm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(qsymm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(csymm)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zsymm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(xsymm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-
-int BLASFUNC(csymm3m)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zsymm3m)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(xsymm3m)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-
-int BLASFUNC(ssyrk)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, float  *, int *);
-int BLASFUNC(dsyrk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-int BLASFUNC(qsyrk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-int BLASFUNC(csyrk)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, float  *, int *);
-int BLASFUNC(zsyrk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-int BLASFUNC(xsyrk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-
-int BLASFUNC(ssyr2k)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float *, int *, float  *, float  *, int *);
-int BLASFUNC(dsyr2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(qsyr2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(csyr2k)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float *, int *, float  *, float  *, int *);
-int BLASFUNC(zsyr2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(xsyr2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-
-int BLASFUNC(chemm)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zhemm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(xhemm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
+int BLASFUNC(strsm)(const char *, const char *, const char *, const char *, const int *, const int *, const float *,  const float *,  const int *, float *,  const int *);
+int BLASFUNC(dtrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(qtrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(ctrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const float *,  const float *,  const int *, float *,  const int *);
+int BLASFUNC(ztrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(xtrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+
+int BLASFUNC(strmm)(const char *, const char *, const char *, const char *, const int *, const int *, const float *,  const float *,  const int *, float *,  const int *);
+int BLASFUNC(dtrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(qtrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(ctrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const float *,  const float *,  const int *, float *,  const int *);
+int BLASFUNC(ztrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(xtrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+
+int BLASFUNC(ssymm)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dsymm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qsymm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(csymm)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zsymm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xsymm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+
+int BLASFUNC(csymm3m)(char *, char *, int *, int *, float  *, float  *, int *, float  *, int *, float  *, float  *, int *);
+int BLASFUNC(zsymm3m)(char *, char *, int *, int *, double *, double *, int *, double *, int *, double *, double *, int *);
+int BLASFUNC(xsymm3m)(char *, char *, int *, int *, double *, double *, int *, double *, int *, double *, double *, int *);
+
+int BLASFUNC(ssyrk)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dsyrk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qsyrk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(csyrk)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zsyrk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xsyrk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+
+int BLASFUNC(ssyr2k)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dsyr2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
+int BLASFUNC(qsyr2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
+int BLASFUNC(csyr2k)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zsyr2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
+int BLASFUNC(xsyr2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
+
+int BLASFUNC(chemm)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zhemm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xhemm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
 int BLASFUNC(chemm3m)(char *, char *, int *, int *, float  *, float  *, int *,
 	   float  *, int *, float  *, float  *, int *);
@@ -520,136 +421,17 @@ int BLASFUNC(zhemm3m)(char *, char *, int *, int *, double *, double *, int *,
 int BLASFUNC(xhemm3m)(char *, char *, int *, int *, double *, double *, int *,
 	   double *, int *, double *, double *, int *);
 
-int BLASFUNC(cherk)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, float  *, int *);
-int BLASFUNC(zherk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-int BLASFUNC(xherk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-
-int BLASFUNC(cher2k)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float *, int *, float  *, float  *, int *);
-int BLASFUNC(zher2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(xher2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(cher2m)(char *, char *, char *, int *, int *, float  *, float  *, int *,
-	   float *, int *, float  *, float  *, int *);
-int BLASFUNC(zher2m)(char *, char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(xher2m)(char *, char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-
-int BLASFUNC(sgemt)(char *, int *, int *, float  *, float  *, int *,
-		    float  *, int *);
-int BLASFUNC(dgemt)(char *, int *, int *, double *, double *, int *,
-		    double *, int *);
-int BLASFUNC(cgemt)(char *, int *, int *, float  *, float  *, int *,
-		    float  *, int *);
-int BLASFUNC(zgemt)(char *, int *, int *, double *, double *, int *,
-		    double *, int *);
+int BLASFUNC(cherk)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zherk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xherk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+
+int BLASFUNC(cher2k)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zher2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xher2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(cher2m)(const char *, const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zher2m)(const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
+int BLASFUNC(xher2m)(const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
 
-int BLASFUNC(sgema)(char *, char *, int *, int *, float  *,
-		    float  *, int *, float *, float  *, int *, float *, int *);
-int BLASFUNC(dgema)(char *, char *, int *, int *, double *,
-		    double *, int *, double*, double *, int *, double*, int *);
-int BLASFUNC(cgema)(char *, char *, int *, int *, float  *,
-		    float  *, int *, float *, float  *, int *, float *, int *);
-int BLASFUNC(zgema)(char *, char *, int *, int *, double *,
-		    double *, int *, double*, double *, int *, double*, int *);
-
-int BLASFUNC(sgems)(char *, char *, int *, int *, float  *,
-		    float  *, int *, float *, float  *, int *, float *, int *);
-int BLASFUNC(dgems)(char *, char *, int *, int *, double *,
-		    double *, int *, double*, double *, int *, double*, int *);
-int BLASFUNC(cgems)(char *, char *, int *, int *, float  *,
-		    float  *, int *, float *, float  *, int *, float *, int *);
-int BLASFUNC(zgems)(char *, char *, int *, int *, double *,
-		    double *, int *, double*, double *, int *, double*, int *);
-
-int BLASFUNC(sgetf2)(int *, int *, float  *, int *, int *, int *);
-int BLASFUNC(dgetf2)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(qgetf2)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(cgetf2)(int *, int *, float  *, int *, int *, int *);
-int BLASFUNC(zgetf2)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(xgetf2)(int *, int *, double *, int *, int *, int *);
-
-int BLASFUNC(sgetrf)(int *, int *, float  *, int *, int *, int *);
-int BLASFUNC(dgetrf)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(qgetrf)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(cgetrf)(int *, int *, float  *, int *, int *, int *);
-int BLASFUNC(zgetrf)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(xgetrf)(int *, int *, double *, int *, int *, int *);
-
-int BLASFUNC(slaswp)(int *, float  *, int *, int *, int *, int *, int *);
-int BLASFUNC(dlaswp)(int *, double *, int *, int *, int *, int *, int *);
-int BLASFUNC(qlaswp)(int *, double *, int *, int *, int *, int *, int *);
-int BLASFUNC(claswp)(int *, float  *, int *, int *, int *, int *, int *);
-int BLASFUNC(zlaswp)(int *, double *, int *, int *, int *, int *, int *);
-int BLASFUNC(xlaswp)(int *, double *, int *, int *, int *, int *, int *);
-
-int BLASFUNC(sgetrs)(char *, int *, int *, float  *, int *, int *, float  *, int *, int *);
-int BLASFUNC(dgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
-int BLASFUNC(qgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
-int BLASFUNC(cgetrs)(char *, int *, int *, float  *, int *, int *, float  *, int *, int *);
-int BLASFUNC(zgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
-int BLASFUNC(xgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
-
-int BLASFUNC(sgesv)(int *, int *, float  *, int *, int *, float *, int *, int *);
-int BLASFUNC(dgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
-int BLASFUNC(qgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
-int BLASFUNC(cgesv)(int *, int *, float  *, int *, int *, float *, int *, int *);
-int BLASFUNC(zgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
-int BLASFUNC(xgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
-
-int BLASFUNC(spotf2)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dpotf2)(char *, int *, double *, int *, int *);
-int BLASFUNC(qpotf2)(char *, int *, double *, int *, int *);
-int BLASFUNC(cpotf2)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zpotf2)(char *, int *, double *, int *, int *);
-int BLASFUNC(xpotf2)(char *, int *, double *, int *, int *);
-
-int BLASFUNC(spotrf)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dpotrf)(char *, int *, double *, int *, int *);
-int BLASFUNC(qpotrf)(char *, int *, double *, int *, int *);
-int BLASFUNC(cpotrf)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zpotrf)(char *, int *, double *, int *, int *);
-int BLASFUNC(xpotrf)(char *, int *, double *, int *, int *);
-
-int BLASFUNC(slauu2)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dlauu2)(char *, int *, double *, int *, int *);
-int BLASFUNC(qlauu2)(char *, int *, double *, int *, int *);
-int BLASFUNC(clauu2)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zlauu2)(char *, int *, double *, int *, int *);
-int BLASFUNC(xlauu2)(char *, int *, double *, int *, int *);
-
-int BLASFUNC(slauum)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dlauum)(char *, int *, double *, int *, int *);
-int BLASFUNC(qlauum)(char *, int *, double *, int *, int *);
-int BLASFUNC(clauum)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zlauum)(char *, int *, double *, int *, int *);
-int BLASFUNC(xlauum)(char *, int *, double *, int *, int *);
-
-int BLASFUNC(strti2)(char *, char *, int *, float  *, int *, int *);
-int BLASFUNC(dtrti2)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(qtrti2)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(ctrti2)(char *, char *, int *, float  *, int *, int *);
-int BLASFUNC(ztrti2)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(xtrti2)(char *, char *, int *, double *, int *, int *);
-
-int BLASFUNC(strtri)(char *, char *, int *, float  *, int *, int *);
-int BLASFUNC(dtrtri)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(qtrtri)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(ctrtri)(char *, char *, int *, float  *, int *, int *);
-int BLASFUNC(ztrtri)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(xtrtri)(char *, char *, int *, double *, int *, int *);
-
-int BLASFUNC(spotri)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dpotri)(char *, int *, double *, int *, int *);
-int BLASFUNC(qpotri)(char *, int *, double *, int *, int *);
-int BLASFUNC(cpotri)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zpotri)(char *, int *, double *, int *, int *);
-int BLASFUNC(xpotri)(char *, int *, double *, int *, int *);
 
 #ifdef __cplusplus
 }
diff --git a/extern/Eigen3/Eigen/src/misc/lapack.h b/extern/Eigen3/Eigen/src/misc/lapack.h
new file mode 100644
index 00000000000..249f3575c6d
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/misc/lapack.h
@@ -0,0 +1,152 @@
+#ifndef LAPACK_H
+#define LAPACK_H
+
+#include "blas.h"
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+int BLASFUNC(csymv) (const char *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zsymv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xsymv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+
+
+int BLASFUNC(cspmv) (char *, int *, float  *, float *,
+         float  *, int *, float *, float *, int *);
+int BLASFUNC(zspmv) (char *, int *, double  *, double *,
+         double  *, int *, double *, double *, int *);
+int BLASFUNC(xspmv) (char *, int *, double  *, double *,
+         double  *, int *, double *, double *, int *);
+
+int BLASFUNC(csyr) (char *, int *, float   *, float  *, int *,
+        float  *, int *);
+int BLASFUNC(zsyr) (char *, int *, double  *, double *, int *,
+        double *, int *);
+int BLASFUNC(xsyr) (char *, int *, double  *, double *, int *,
+        double *, int *);
+
+int BLASFUNC(cspr) (char *, int *, float   *, float  *, int *,
+        float  *);
+int BLASFUNC(zspr) (char *, int *, double  *, double *, int *,
+        double *);
+int BLASFUNC(xspr) (char *, int *, double  *, double *, int *,
+        double *);
+
+int BLASFUNC(sgemt)(char *, int *, int *, float  *, float  *, int *,
+        float  *, int *);
+int BLASFUNC(dgemt)(char *, int *, int *, double *, double *, int *,
+        double *, int *);
+int BLASFUNC(cgemt)(char *, int *, int *, float  *, float  *, int *,
+        float  *, int *);
+int BLASFUNC(zgemt)(char *, int *, int *, double *, double *, int *,
+        double *, int *);
+
+int BLASFUNC(sgema)(char *, char *, int *, int *, float  *,
+        float  *, int *, float *, float  *, int *, float *, int *);
+int BLASFUNC(dgema)(char *, char *, int *, int *, double *,
+        double *, int *, double*, double *, int *, double*, int *);
+int BLASFUNC(cgema)(char *, char *, int *, int *, float  *,
+        float  *, int *, float *, float  *, int *, float *, int *);
+int BLASFUNC(zgema)(char *, char *, int *, int *, double *,
+        double *, int *, double*, double *, int *, double*, int *);
+
+int BLASFUNC(sgems)(char *, char *, int *, int *, float  *,
+        float  *, int *, float *, float  *, int *, float *, int *);
+int BLASFUNC(dgems)(char *, char *, int *, int *, double *,
+        double *, int *, double*, double *, int *, double*, int *);
+int BLASFUNC(cgems)(char *, char *, int *, int *, float  *,
+        float  *, int *, float *, float  *, int *, float *, int *);
+int BLASFUNC(zgems)(char *, char *, int *, int *, double *,
+        double *, int *, double*, double *, int *, double*, int *);
+
+int BLASFUNC(sgetf2)(int *, int *, float  *, int *, int *, int *);
+int BLASFUNC(dgetf2)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(qgetf2)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(cgetf2)(int *, int *, float  *, int *, int *, int *);
+int BLASFUNC(zgetf2)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(xgetf2)(int *, int *, double *, int *, int *, int *);
+
+int BLASFUNC(sgetrf)(int *, int *, float  *, int *, int *, int *);
+int BLASFUNC(dgetrf)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(qgetrf)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(cgetrf)(int *, int *, float  *, int *, int *, int *);
+int BLASFUNC(zgetrf)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(xgetrf)(int *, int *, double *, int *, int *, int *);
+
+int BLASFUNC(slaswp)(int *, float  *, int *, int *, int *, int *, int *);
+int BLASFUNC(dlaswp)(int *, double *, int *, int *, int *, int *, int *);
+int BLASFUNC(qlaswp)(int *, double *, int *, int *, int *, int *, int *);
+int BLASFUNC(claswp)(int *, float  *, int *, int *, int *, int *, int *);
+int BLASFUNC(zlaswp)(int *, double *, int *, int *, int *, int *, int *);
+int BLASFUNC(xlaswp)(int *, double *, int *, int *, int *, int *, int *);
+
+int BLASFUNC(sgetrs)(char *, int *, int *, float  *, int *, int *, float  *, int *, int *);
+int BLASFUNC(dgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
+int BLASFUNC(qgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
+int BLASFUNC(cgetrs)(char *, int *, int *, float  *, int *, int *, float  *, int *, int *);
+int BLASFUNC(zgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
+int BLASFUNC(xgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
+
+int BLASFUNC(sgesv)(int *, int *, float  *, int *, int *, float *, int *, int *);
+int BLASFUNC(dgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
+int BLASFUNC(qgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
+int BLASFUNC(cgesv)(int *, int *, float  *, int *, int *, float *, int *, int *);
+int BLASFUNC(zgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
+int BLASFUNC(xgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
+
+int BLASFUNC(spotf2)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dpotf2)(char *, int *, double *, int *, int *);
+int BLASFUNC(qpotf2)(char *, int *, double *, int *, int *);
+int BLASFUNC(cpotf2)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zpotf2)(char *, int *, double *, int *, int *);
+int BLASFUNC(xpotf2)(char *, int *, double *, int *, int *);
+
+int BLASFUNC(spotrf)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dpotrf)(char *, int *, double *, int *, int *);
+int BLASFUNC(qpotrf)(char *, int *, double *, int *, int *);
+int BLASFUNC(cpotrf)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zpotrf)(char *, int *, double *, int *, int *);
+int BLASFUNC(xpotrf)(char *, int *, double *, int *, int *);
+
+int BLASFUNC(slauu2)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dlauu2)(char *, int *, double *, int *, int *);
+int BLASFUNC(qlauu2)(char *, int *, double *, int *, int *);
+int BLASFUNC(clauu2)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zlauu2)(char *, int *, double *, int *, int *);
+int BLASFUNC(xlauu2)(char *, int *, double *, int *, int *);
+
+int BLASFUNC(slauum)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dlauum)(char *, int *, double *, int *, int *);
+int BLASFUNC(qlauum)(char *, int *, double *, int *, int *);
+int BLASFUNC(clauum)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zlauum)(char *, int *, double *, int *, int *);
+int BLASFUNC(xlauum)(char *, int *, double *, int *, int *);
+
+int BLASFUNC(strti2)(char *, char *, int *, float  *, int *, int *);
+int BLASFUNC(dtrti2)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(qtrti2)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(ctrti2)(char *, char *, int *, float  *, int *, int *);
+int BLASFUNC(ztrti2)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(xtrti2)(char *, char *, int *, double *, int *, int *);
+
+int BLASFUNC(strtri)(char *, char *, int *, float  *, int *, int *);
+int BLASFUNC(dtrtri)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(qtrtri)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(ctrtri)(char *, char *, int *, float  *, int *, int *);
+int BLASFUNC(ztrtri)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(xtrtri)(char *, char *, int *, double *, int *, int *);
+
+int BLASFUNC(spotri)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dpotri)(char *, int *, double *, int *, int *);
+int BLASFUNC(qpotri)(char *, int *, double *, int *, int *);
+int BLASFUNC(cpotri)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zpotri)(char *, int *, double *, int *, int *);
+int BLASFUNC(xpotri)(char *, int *, double *, int *, int *);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/extern/Eigen3/Eigen/src/misc/lapacke.h b/extern/Eigen3/Eigen/src/misc/lapacke.h
new file mode 100755
index 00000000000..8c7e79b034c
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/misc/lapacke.h
@@ -0,0 +1,16291 @@
+/*****************************************************************************
+  Copyright (c) 2010, Intel Corp.
+  All rights reserved.
+
+  Redistribution and use in source and binary forms, with or without
+  modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+      this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+    * Neither the name of Intel Corporation nor the names of its contributors
+      may be used to endorse or promote products derived from this software
+      without specific prior written permission.
+
+  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+  AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+  IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+  ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+  LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
+  THE POSSIBILITY OF SUCH DAMAGE.
+******************************************************************************
+* Contents: Native C interface to LAPACK
+* Author: Intel Corporation
+* Generated November, 2011
+*****************************************************************************/
+
+#ifndef _MKL_LAPACKE_H_
+
+#ifndef _LAPACKE_H_
+#define _LAPACKE_H_
+
+/*
+*  Turn on HAVE_LAPACK_CONFIG_H to redefine C-LAPACK datatypes
+*/
+#ifdef HAVE_LAPACK_CONFIG_H
+#include "lapacke_config.h"
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+
+#include <stdlib.h>
+
+#ifndef lapack_int
+#define lapack_int     int
+#endif
+
+#ifndef lapack_logical
+#define lapack_logical lapack_int
+#endif
+
+/* Complex types are structures equivalent to the
+* Fortran complex types COMPLEX(4) and COMPLEX(8).
+*
+* One can also redefine the types with his own types
+* for example by including in the code definitions like
+*
+* #define lapack_complex_float std::complex<float>
+* #define lapack_complex_double std::complex<double>
+*
+* or define these types in the command line:
+*
+* -Dlapack_complex_float="std::complex<float>"
+* -Dlapack_complex_double="std::complex<double>"
+*/
+
+#ifndef LAPACK_COMPLEX_CUSTOM
+
+/* Complex type (single precision) */
+#ifndef lapack_complex_float
+#include <complex.h>
+#define lapack_complex_float    float _Complex
+#endif
+
+#ifndef lapack_complex_float_real
+#define lapack_complex_float_real(z)       (creal(z))
+#endif
+
+#ifndef lapack_complex_float_imag
+#define lapack_complex_float_imag(z)       (cimag(z))
+#endif
+
+lapack_complex_float lapack_make_complex_float( float re, float im );
+
+/* Complex type (double precision) */
+#ifndef lapack_complex_double
+#include <complex.h>
+#define lapack_complex_double   double _Complex
+#endif
+
+#ifndef lapack_complex_double_real
+#define lapack_complex_double_real(z)      (creal(z))
+#endif
+
+#ifndef lapack_complex_double_imag
+#define lapack_complex_double_imag(z)       (cimag(z))
+#endif
+
+lapack_complex_double lapack_make_complex_double( double re, double im );
+
+#endif
+
+#ifndef LAPACKE_malloc
+#define LAPACKE_malloc( size ) malloc( size )
+#endif
+#ifndef LAPACKE_free
+#define LAPACKE_free( p )      free( p )
+#endif
+
+#define LAPACK_C2INT( x ) (lapack_int)(*((float*)&x ))
+#define LAPACK_Z2INT( x ) (lapack_int)(*((double*)&x ))
+
+#define LAPACK_ROW_MAJOR               101
+#define LAPACK_COL_MAJOR               102
+
+#define LAPACK_WORK_MEMORY_ERROR       -1010
+#define LAPACK_TRANSPOSE_MEMORY_ERROR  -1011
+
+/* Callback logical functions of one, two, or three arguments are used
+*  to select eigenvalues to sort to the top left of the Schur form.
+*  The value is selected if function returns TRUE (non-zero). */
+
+typedef lapack_logical (*LAPACK_S_SELECT2) ( const float*, const float* );
+typedef lapack_logical (*LAPACK_S_SELECT3)
+    ( const float*, const float*, const float* );
+typedef lapack_logical (*LAPACK_D_SELECT2) ( const double*, const double* );
+typedef lapack_logical (*LAPACK_D_SELECT3)
+    ( const double*, const double*, const double* );
+
+typedef lapack_logical (*LAPACK_C_SELECT1) ( const lapack_complex_float* );
+typedef lapack_logical (*LAPACK_C_SELECT2)
+    ( const lapack_complex_float*, const lapack_complex_float* );
+typedef lapack_logical (*LAPACK_Z_SELECT1) ( const lapack_complex_double* );
+typedef lapack_logical (*LAPACK_Z_SELECT2)
+    ( const lapack_complex_double*, const lapack_complex_double* );
+
+#include "lapacke_mangling.h"
+
+#define LAPACK_lsame LAPACK_GLOBAL(lsame,LSAME)
+lapack_logical LAPACK_lsame( char* ca,  char* cb,
+                              lapack_int lca, lapack_int lcb );
+
+/* C-LAPACK function prototypes */
+
+lapack_int LAPACKE_sbdsdc( int matrix_order, char uplo, char compq,
+                           lapack_int n, float* d, float* e, float* u,
+                           lapack_int ldu, float* vt, lapack_int ldvt, float* q,
+                           lapack_int* iq );
+lapack_int LAPACKE_dbdsdc( int matrix_order, char uplo, char compq,
+                           lapack_int n, double* d, double* e, double* u,
+                           lapack_int ldu, double* vt, lapack_int ldvt,
+                           double* q, lapack_int* iq );
+
+lapack_int LAPACKE_sbdsqr( int matrix_order, char uplo, lapack_int n,
+                           lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                           float* d, float* e, float* vt, lapack_int ldvt,
+                           float* u, lapack_int ldu, float* c, lapack_int ldc );
+lapack_int LAPACKE_dbdsqr( int matrix_order, char uplo, lapack_int n,
+                           lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                           double* d, double* e, double* vt, lapack_int ldvt,
+                           double* u, lapack_int ldu, double* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_cbdsqr( int matrix_order, char uplo, lapack_int n,
+                           lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                           float* d, float* e, lapack_complex_float* vt,
+                           lapack_int ldvt, lapack_complex_float* u,
+                           lapack_int ldu, lapack_complex_float* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_zbdsqr( int matrix_order, char uplo, lapack_int n,
+                           lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                           double* d, double* e, lapack_complex_double* vt,
+                           lapack_int ldvt, lapack_complex_double* u,
+                           lapack_int ldu, lapack_complex_double* c,
+                           lapack_int ldc );
+
+lapack_int LAPACKE_sdisna( char job, lapack_int m, lapack_int n, const float* d,
+                           float* sep );
+lapack_int LAPACKE_ddisna( char job, lapack_int m, lapack_int n,
+                           const double* d, double* sep );
+
+lapack_int LAPACKE_sgbbrd( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int ncc, lapack_int kl,
+                           lapack_int ku, float* ab, lapack_int ldab, float* d,
+                           float* e, float* q, lapack_int ldq, float* pt,
+                           lapack_int ldpt, float* c, lapack_int ldc );
+lapack_int LAPACKE_dgbbrd( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int ncc, lapack_int kl,
+                           lapack_int ku, double* ab, lapack_int ldab,
+                           double* d, double* e, double* q, lapack_int ldq,
+                           double* pt, lapack_int ldpt, double* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_cgbbrd( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int ncc, lapack_int kl,
+                           lapack_int ku, lapack_complex_float* ab,
+                           lapack_int ldab, float* d, float* e,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_complex_float* pt, lapack_int ldpt,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zgbbrd( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int ncc, lapack_int kl,
+                           lapack_int ku, lapack_complex_double* ab,
+                           lapack_int ldab, double* d, double* e,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* pt, lapack_int ldpt,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sgbcon( int matrix_order, char norm, lapack_int n,
+                           lapack_int kl, lapack_int ku, const float* ab,
+                           lapack_int ldab, const lapack_int* ipiv, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_dgbcon( int matrix_order, char norm, lapack_int n,
+                           lapack_int kl, lapack_int ku, const double* ab,
+                           lapack_int ldab, const lapack_int* ipiv,
+                           double anorm, double* rcond );
+lapack_int LAPACKE_cgbcon( int matrix_order, char norm, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zgbcon( int matrix_order, char norm, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_sgbequ( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const float* ab,
+                           lapack_int ldab, float* r, float* c, float* rowcnd,
+                           float* colcnd, float* amax );
+lapack_int LAPACKE_dgbequ( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const double* ab,
+                           lapack_int ldab, double* r, double* c,
+                           double* rowcnd, double* colcnd, double* amax );
+lapack_int LAPACKE_cgbequ( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           float* r, float* c, float* rowcnd, float* colcnd,
+                           float* amax );
+lapack_int LAPACKE_zgbequ( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           double* r, double* c, double* rowcnd, double* colcnd,
+                           double* amax );
+
+lapack_int LAPACKE_sgbequb( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_int kl, lapack_int ku, const float* ab,
+                            lapack_int ldab, float* r, float* c, float* rowcnd,
+                            float* colcnd, float* amax );
+lapack_int LAPACKE_dgbequb( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_int kl, lapack_int ku, const double* ab,
+                            lapack_int ldab, double* r, double* c,
+                            double* rowcnd, double* colcnd, double* amax );
+lapack_int LAPACKE_cgbequb( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_int kl, lapack_int ku,
+                            const lapack_complex_float* ab, lapack_int ldab,
+                            float* r, float* c, float* rowcnd, float* colcnd,
+                            float* amax );
+lapack_int LAPACKE_zgbequb( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_int kl, lapack_int ku,
+                            const lapack_complex_double* ab, lapack_int ldab,
+                            double* r, double* c, double* rowcnd,
+                            double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgbrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const float* ab, lapack_int ldab, const float* afb,
+                           lapack_int ldafb, const lapack_int* ipiv,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dgbrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const double* ab, lapack_int ldab, const double* afb,
+                           lapack_int ldafb, const lapack_int* ipiv,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_cgbrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_complex_float* afb, lapack_int ldafb,
+                           const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zgbrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_complex_double* afb, lapack_int ldafb,
+                           const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sgbrfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, const float* ab, lapack_int ldab,
+                            const float* afb, lapack_int ldafb,
+                            const lapack_int* ipiv, const float* r,
+                            const float* c, const float* b, lapack_int ldb,
+                            float* x, lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dgbrfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, const double* ab, lapack_int ldab,
+                            const double* afb, lapack_int ldafb,
+                            const lapack_int* ipiv, const double* r,
+                            const double* c, const double* b, lapack_int ldb,
+                            double* x, lapack_int ldx, double* rcond,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cgbrfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, const lapack_complex_float* ab,
+                            lapack_int ldab, const lapack_complex_float* afb,
+                            lapack_int ldafb, const lapack_int* ipiv,
+                            const float* r, const float* c,
+                            const lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zgbrfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, const lapack_complex_double* ab,
+                            lapack_int ldab, const lapack_complex_double* afb,
+                            lapack_int ldafb, const lapack_int* ipiv,
+                            const double* r, const double* c,
+                            const lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_sgbsv( int matrix_order, lapack_int n, lapack_int kl,
+                          lapack_int ku, lapack_int nrhs, float* ab,
+                          lapack_int ldab, lapack_int* ipiv, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dgbsv( int matrix_order, lapack_int n, lapack_int kl,
+                          lapack_int ku, lapack_int nrhs, double* ab,
+                          lapack_int ldab, lapack_int* ipiv, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_cgbsv( int matrix_order, lapack_int n, lapack_int kl,
+                          lapack_int ku, lapack_int nrhs,
+                          lapack_complex_float* ab, lapack_int ldab,
+                          lapack_int* ipiv, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zgbsv( int matrix_order, lapack_int n, lapack_int kl,
+                          lapack_int ku, lapack_int nrhs,
+                          lapack_complex_double* ab, lapack_int ldab,
+                          lapack_int* ipiv, lapack_complex_double* b,
+                          lapack_int ldb );
+
+lapack_int LAPACKE_sgbsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int kl, lapack_int ku,
+                           lapack_int nrhs, float* ab, lapack_int ldab,
+                           float* afb, lapack_int ldafb, lapack_int* ipiv,
+                           char* equed, float* r, float* c, float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr,
+                           float* rpivot );
+lapack_int LAPACKE_dgbsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int kl, lapack_int ku,
+                           lapack_int nrhs, double* ab, lapack_int ldab,
+                           double* afb, lapack_int ldafb, lapack_int* ipiv,
+                           char* equed, double* r, double* c, double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr,
+                           double* rpivot );
+lapack_int LAPACKE_cgbsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int kl, lapack_int ku,
+                           lapack_int nrhs, lapack_complex_float* ab,
+                           lapack_int ldab, lapack_complex_float* afb,
+                           lapack_int ldafb, lapack_int* ipiv, char* equed,
+                           float* r, float* c, lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr, float* rpivot );
+lapack_int LAPACKE_zgbsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int kl, lapack_int ku,
+                           lapack_int nrhs, lapack_complex_double* ab,
+                           lapack_int ldab, lapack_complex_double* afb,
+                           lapack_int ldafb, lapack_int* ipiv, char* equed,
+                           double* r, double* c, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, double* rcond, double* ferr,
+                           double* berr, double* rpivot );
+
+lapack_int LAPACKE_sgbsvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, float* ab, lapack_int ldab,
+                            float* afb, lapack_int ldafb, lapack_int* ipiv,
+                            char* equed, float* r, float* c, float* b,
+                            lapack_int ldb, float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dgbsvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, double* ab, lapack_int ldab,
+                            double* afb, lapack_int ldafb, lapack_int* ipiv,
+                            char* equed, double* r, double* c, double* b,
+                            lapack_int ldb, double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+lapack_int LAPACKE_cgbsvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, lapack_complex_float* ab,
+                            lapack_int ldab, lapack_complex_float* afb,
+                            lapack_int ldafb, lapack_int* ipiv, char* equed,
+                            float* r, float* c, lapack_complex_float* b,
+                            lapack_int ldb, lapack_complex_float* x,
+                            lapack_int ldx, float* rcond, float* rpvgrw,
+                            float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zgbsvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, lapack_complex_double* ab,
+                            lapack_int ldab, lapack_complex_double* afb,
+                            lapack_int ldafb, lapack_int* ipiv, char* equed,
+                            double* r, double* c, lapack_complex_double* b,
+                            lapack_int ldb, lapack_complex_double* x,
+                            lapack_int ldx, double* rcond, double* rpvgrw,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_sgbtrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, float* ab,
+                           lapack_int ldab, lapack_int* ipiv );
+lapack_int LAPACKE_dgbtrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, double* ab,
+                           lapack_int ldab, lapack_int* ipiv );
+lapack_int LAPACKE_cgbtrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zgbtrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_sgbtrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const float* ab, lapack_int ldab,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dgbtrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const double* ab, lapack_int ldab,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_cgbtrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_int* ipiv, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zgbtrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_sgebak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const float* scale,
+                           lapack_int m, float* v, lapack_int ldv );
+lapack_int LAPACKE_dgebak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const double* scale,
+                           lapack_int m, double* v, lapack_int ldv );
+lapack_int LAPACKE_cgebak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const float* scale,
+                           lapack_int m, lapack_complex_float* v,
+                           lapack_int ldv );
+lapack_int LAPACKE_zgebak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const double* scale,
+                           lapack_int m, lapack_complex_double* v,
+                           lapack_int ldv );
+
+lapack_int LAPACKE_sgebal( int matrix_order, char job, lapack_int n, float* a,
+                           lapack_int lda, lapack_int* ilo, lapack_int* ihi,
+                           float* scale );
+lapack_int LAPACKE_dgebal( int matrix_order, char job, lapack_int n, double* a,
+                           lapack_int lda, lapack_int* ilo, lapack_int* ihi,
+                           double* scale );
+lapack_int LAPACKE_cgebal( int matrix_order, char job, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ilo, lapack_int* ihi, float* scale );
+lapack_int LAPACKE_zgebal( int matrix_order, char job, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ilo, lapack_int* ihi, double* scale );
+
+lapack_int LAPACKE_sgebrd( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* d, float* e,
+                           float* tauq, float* taup );
+lapack_int LAPACKE_dgebrd( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* d, double* e,
+                           double* tauq, double* taup );
+lapack_int LAPACKE_cgebrd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda, float* d,
+                           float* e, lapack_complex_float* tauq,
+                           lapack_complex_float* taup );
+lapack_int LAPACKE_zgebrd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda, double* d,
+                           double* e, lapack_complex_double* tauq,
+                           lapack_complex_double* taup );
+
+lapack_int LAPACKE_sgecon( int matrix_order, char norm, lapack_int n,
+                           const float* a, lapack_int lda, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_dgecon( int matrix_order, char norm, lapack_int n,
+                           const double* a, lapack_int lda, double anorm,
+                           double* rcond );
+lapack_int LAPACKE_cgecon( int matrix_order, char norm, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           float anorm, float* rcond );
+lapack_int LAPACKE_zgecon( int matrix_order, char norm, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           double anorm, double* rcond );
+
+lapack_int LAPACKE_sgeequ( int matrix_order, lapack_int m, lapack_int n,
+                           const float* a, lapack_int lda, float* r, float* c,
+                           float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_dgeequ( int matrix_order, lapack_int m, lapack_int n,
+                           const double* a, lapack_int lda, double* r,
+                           double* c, double* rowcnd, double* colcnd,
+                           double* amax );
+lapack_int LAPACKE_cgeequ( int matrix_order, lapack_int m, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           float* r, float* c, float* rowcnd, float* colcnd,
+                           float* amax );
+lapack_int LAPACKE_zgeequ( int matrix_order, lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           double* r, double* c, double* rowcnd, double* colcnd,
+                           double* amax );
+
+lapack_int LAPACKE_sgeequb( int matrix_order, lapack_int m, lapack_int n,
+                            const float* a, lapack_int lda, float* r, float* c,
+                            float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_dgeequb( int matrix_order, lapack_int m, lapack_int n,
+                            const double* a, lapack_int lda, double* r,
+                            double* c, double* rowcnd, double* colcnd,
+                            double* amax );
+lapack_int LAPACKE_cgeequb( int matrix_order, lapack_int m, lapack_int n,
+                            const lapack_complex_float* a, lapack_int lda,
+                            float* r, float* c, float* rowcnd, float* colcnd,
+                            float* amax );
+lapack_int LAPACKE_zgeequb( int matrix_order, lapack_int m, lapack_int n,
+                            const lapack_complex_double* a, lapack_int lda,
+                            double* r, double* c, double* rowcnd,
+                            double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgees( int matrix_order, char jobvs, char sort,
+                          LAPACK_S_SELECT2 select, lapack_int n, float* a,
+                          lapack_int lda, lapack_int* sdim, float* wr,
+                          float* wi, float* vs, lapack_int ldvs );
+lapack_int LAPACKE_dgees( int matrix_order, char jobvs, char sort,
+                          LAPACK_D_SELECT2 select, lapack_int n, double* a,
+                          lapack_int lda, lapack_int* sdim, double* wr,
+                          double* wi, double* vs, lapack_int ldvs );
+lapack_int LAPACKE_cgees( int matrix_order, char jobvs, char sort,
+                          LAPACK_C_SELECT1 select, lapack_int n,
+                          lapack_complex_float* a, lapack_int lda,
+                          lapack_int* sdim, lapack_complex_float* w,
+                          lapack_complex_float* vs, lapack_int ldvs );
+lapack_int LAPACKE_zgees( int matrix_order, char jobvs, char sort,
+                          LAPACK_Z_SELECT1 select, lapack_int n,
+                          lapack_complex_double* a, lapack_int lda,
+                          lapack_int* sdim, lapack_complex_double* w,
+                          lapack_complex_double* vs, lapack_int ldvs );
+
+lapack_int LAPACKE_sgeesx( int matrix_order, char jobvs, char sort,
+                           LAPACK_S_SELECT2 select, char sense, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* sdim,
+                           float* wr, float* wi, float* vs, lapack_int ldvs,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_dgeesx( int matrix_order, char jobvs, char sort,
+                           LAPACK_D_SELECT2 select, char sense, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* sdim,
+                           double* wr, double* wi, double* vs, lapack_int ldvs,
+                           double* rconde, double* rcondv );
+lapack_int LAPACKE_cgeesx( int matrix_order, char jobvs, char sort,
+                           LAPACK_C_SELECT1 select, char sense, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* sdim, lapack_complex_float* w,
+                           lapack_complex_float* vs, lapack_int ldvs,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_zgeesx( int matrix_order, char jobvs, char sort,
+                           LAPACK_Z_SELECT1 select, char sense, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* sdim, lapack_complex_double* w,
+                           lapack_complex_double* vs, lapack_int ldvs,
+                           double* rconde, double* rcondv );
+
+lapack_int LAPACKE_sgeev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, float* a, lapack_int lda, float* wr,
+                          float* wi, float* vl, lapack_int ldvl, float* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_dgeev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, double* a, lapack_int lda, double* wr,
+                          double* wi, double* vl, lapack_int ldvl, double* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_cgeev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, lapack_complex_float* a, lapack_int lda,
+                          lapack_complex_float* w, lapack_complex_float* vl,
+                          lapack_int ldvl, lapack_complex_float* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_zgeev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, lapack_complex_double* a,
+                          lapack_int lda, lapack_complex_double* w,
+                          lapack_complex_double* vl, lapack_int ldvl,
+                          lapack_complex_double* vr, lapack_int ldvr );
+
+lapack_int LAPACKE_sgeevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n, float* a,
+                           lapack_int lda, float* wr, float* wi, float* vl,
+                           lapack_int ldvl, float* vr, lapack_int ldvr,
+                           lapack_int* ilo, lapack_int* ihi, float* scale,
+                           float* abnrm, float* rconde, float* rcondv );
+lapack_int LAPACKE_dgeevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n, double* a,
+                           lapack_int lda, double* wr, double* wi, double* vl,
+                           lapack_int ldvl, double* vr, lapack_int ldvr,
+                           lapack_int* ilo, lapack_int* ihi, double* scale,
+                           double* abnrm, double* rconde, double* rcondv );
+lapack_int LAPACKE_cgeevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* w, lapack_complex_float* vl,
+                           lapack_int ldvl, lapack_complex_float* vr,
+                           lapack_int ldvr, lapack_int* ilo, lapack_int* ihi,
+                           float* scale, float* abnrm, float* rconde,
+                           float* rcondv );
+lapack_int LAPACKE_zgeevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* w, lapack_complex_double* vl,
+                           lapack_int ldvl, lapack_complex_double* vr,
+                           lapack_int ldvr, lapack_int* ilo, lapack_int* ihi,
+                           double* scale, double* abnrm, double* rconde,
+                           double* rcondv );
+
+lapack_int LAPACKE_sgehrd( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, float* a, lapack_int lda,
+                           float* tau );
+lapack_int LAPACKE_dgehrd( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, double* a, lapack_int lda,
+                           double* tau );
+lapack_int LAPACKE_cgehrd( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* tau );
+lapack_int LAPACKE_zgehrd( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgejsv( int matrix_order, char joba, char jobu, char jobv,
+                           char jobr, char jobt, char jobp, lapack_int m,
+                           lapack_int n, float* a, lapack_int lda, float* sva,
+                           float* u, lapack_int ldu, float* v, lapack_int ldv,
+                           float* stat, lapack_int* istat );
+lapack_int LAPACKE_dgejsv( int matrix_order, char joba, char jobu, char jobv,
+                           char jobr, char jobt, char jobp, lapack_int m,
+                           lapack_int n, double* a, lapack_int lda, double* sva,
+                           double* u, lapack_int ldu, double* v, lapack_int ldv,
+                           double* stat, lapack_int* istat );
+
+lapack_int LAPACKE_sgelq2( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgelq2( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgelq2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgelq2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgelqf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgelqf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgelqf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgelqf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgels( int matrix_order, char trans, lapack_int m,
+                          lapack_int n, lapack_int nrhs, float* a,
+                          lapack_int lda, float* b, lapack_int ldb );
+lapack_int LAPACKE_dgels( int matrix_order, char trans, lapack_int m,
+                          lapack_int n, lapack_int nrhs, double* a,
+                          lapack_int lda, double* b, lapack_int ldb );
+lapack_int LAPACKE_cgels( int matrix_order, char trans, lapack_int m,
+                          lapack_int n, lapack_int nrhs,
+                          lapack_complex_float* a, lapack_int lda,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zgels( int matrix_order, char trans, lapack_int m,
+                          lapack_int n, lapack_int nrhs,
+                          lapack_complex_double* a, lapack_int lda,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sgelsd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, float* s, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_dgelsd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, double* s, double rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_cgelsd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, float* s, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_zgelsd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, double* s, double rcond,
+                           lapack_int* rank );
+
+lapack_int LAPACKE_sgelss( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, float* s, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_dgelss( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, double* s, double rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_cgelss( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, float* s, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_zgelss( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, double* s, double rcond,
+                           lapack_int* rank );
+
+lapack_int LAPACKE_sgelsy( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, lapack_int* jpvt, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_dgelsy( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, lapack_int* jpvt,
+                           double rcond, lapack_int* rank );
+lapack_int LAPACKE_cgelsy( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, lapack_int* jpvt, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_zgelsy( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_int* jpvt, double rcond,
+                           lapack_int* rank );
+
+lapack_int LAPACKE_sgeqlf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgeqlf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgeqlf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqlf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqp3( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* jpvt,
+                           float* tau );
+lapack_int LAPACKE_dgeqp3( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* jpvt,
+                           double* tau );
+lapack_int LAPACKE_cgeqp3( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* jpvt, lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqp3( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* jpvt, lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqpf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* jpvt,
+                           float* tau );
+lapack_int LAPACKE_dgeqpf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* jpvt,
+                           double* tau );
+lapack_int LAPACKE_cgeqpf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* jpvt, lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqpf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* jpvt, lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqr2( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgeqr2( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgeqr2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqr2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqrf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgeqrf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgeqrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqrfp( int matrix_order, lapack_int m, lapack_int n,
+                            float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgeqrfp( int matrix_order, lapack_int m, lapack_int n,
+                            double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgeqrfp( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqrfp( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgerfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const float* af, lapack_int ldaf,
+                           const lapack_int* ipiv, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dgerfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const double* af, lapack_int ldaf,
+                           const lapack_int* ipiv, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cgerfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zgerfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sgerfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int nrhs, const float* a,
+                            lapack_int lda, const float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* r,
+                            const float* c, const float* b, lapack_int ldb,
+                            float* x, lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dgerfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int nrhs, const double* a,
+                            lapack_int lda, const double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* r,
+                            const double* c, const double* b, lapack_int ldb,
+                            double* x, lapack_int ldx, double* rcond,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cgerfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_float* a, lapack_int lda,
+                            const lapack_complex_float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* r,
+                            const float* c, const lapack_complex_float* b,
+                            lapack_int ldb, lapack_complex_float* x,
+                            lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zgerfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_double* a, lapack_int lda,
+                            const lapack_complex_double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* r,
+                            const double* c, const lapack_complex_double* b,
+                            lapack_int ldb, lapack_complex_double* x,
+                            lapack_int ldx, double* rcond, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_sgerqf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgerqf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgerqf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgerqf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgesdd( int matrix_order, char jobz, lapack_int m,
+                           lapack_int n, float* a, lapack_int lda, float* s,
+                           float* u, lapack_int ldu, float* vt,
+                           lapack_int ldvt );
+lapack_int LAPACKE_dgesdd( int matrix_order, char jobz, lapack_int m,
+                           lapack_int n, double* a, lapack_int lda, double* s,
+                           double* u, lapack_int ldu, double* vt,
+                           lapack_int ldvt );
+lapack_int LAPACKE_cgesdd( int matrix_order, char jobz, lapack_int m,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, float* s, lapack_complex_float* u,
+                           lapack_int ldu, lapack_complex_float* vt,
+                           lapack_int ldvt );
+lapack_int LAPACKE_zgesdd( int matrix_order, char jobz, lapack_int m,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, double* s, lapack_complex_double* u,
+                           lapack_int ldu, lapack_complex_double* vt,
+                           lapack_int ldvt );
+
+lapack_int LAPACKE_sgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          float* a, lapack_int lda, lapack_int* ipiv, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          double* a, lapack_int lda, lapack_int* ipiv,
+                          double* b, lapack_int ldb );
+lapack_int LAPACKE_cgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          lapack_complex_float* a, lapack_int lda,
+                          lapack_int* ipiv, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          lapack_complex_double* a, lapack_int lda,
+                          lapack_int* ipiv, lapack_complex_double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dsgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                           double* a, lapack_int lda, lapack_int* ipiv,
+                           double* b, lapack_int ldb, double* x, lapack_int ldx,
+                           lapack_int* iter );
+lapack_int LAPACKE_zcgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, lapack_int* iter );
+
+lapack_int LAPACKE_sgesvd( int matrix_order, char jobu, char jobvt,
+                           lapack_int m, lapack_int n, float* a, lapack_int lda,
+                           float* s, float* u, lapack_int ldu, float* vt,
+                           lapack_int ldvt, float* superb );
+lapack_int LAPACKE_dgesvd( int matrix_order, char jobu, char jobvt,
+                           lapack_int m, lapack_int n, double* a,
+                           lapack_int lda, double* s, double* u, lapack_int ldu,
+                           double* vt, lapack_int ldvt, double* superb );
+lapack_int LAPACKE_cgesvd( int matrix_order, char jobu, char jobvt,
+                           lapack_int m, lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, float* s, lapack_complex_float* u,
+                           lapack_int ldu, lapack_complex_float* vt,
+                           lapack_int ldvt, float* superb );
+lapack_int LAPACKE_zgesvd( int matrix_order, char jobu, char jobvt,
+                           lapack_int m, lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, double* s, lapack_complex_double* u,
+                           lapack_int ldu, lapack_complex_double* vt,
+                           lapack_int ldvt, double* superb );
+
+lapack_int LAPACKE_sgesvj( int matrix_order, char joba, char jobu, char jobv,
+                           lapack_int m, lapack_int n, float* a, lapack_int lda,
+                           float* sva, lapack_int mv, float* v, lapack_int ldv,
+                           float* stat );
+lapack_int LAPACKE_dgesvj( int matrix_order, char joba, char jobu, char jobv,
+                           lapack_int m, lapack_int n, double* a,
+                           lapack_int lda, double* sva, lapack_int mv,
+                           double* v, lapack_int ldv, double* stat );
+
+lapack_int LAPACKE_sgesvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs, float* a,
+                           lapack_int lda, float* af, lapack_int ldaf,
+                           lapack_int* ipiv, char* equed, float* r, float* c,
+                           float* b, lapack_int ldb, float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr,
+                           float* rpivot );
+lapack_int LAPACKE_dgesvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs, double* a,
+                           lapack_int lda, double* af, lapack_int ldaf,
+                           lapack_int* ipiv, char* equed, double* r, double* c,
+                           double* b, lapack_int ldb, double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr,
+                           double* rpivot );
+lapack_int LAPACKE_cgesvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* af, lapack_int ldaf,
+                           lapack_int* ipiv, char* equed, float* r, float* c,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr,
+                           float* rpivot );
+lapack_int LAPACKE_zgesvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* af, lapack_int ldaf,
+                           lapack_int* ipiv, char* equed, double* r, double* c,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr,
+                           double* rpivot );
+
+lapack_int LAPACKE_sgesvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int nrhs, float* a,
+                            lapack_int lda, float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* r, float* c,
+                            float* b, lapack_int ldb, float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dgesvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int nrhs, double* a,
+                            lapack_int lda, double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* r, double* c,
+                            double* b, lapack_int ldb, double* x,
+                            lapack_int ldx, double* rcond, double* rpvgrw,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cgesvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* r, float* c,
+                            lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zgesvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* r, double* c,
+                            lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_sgetf2( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dgetf2( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_cgetf2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zgetf2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetrf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dgetrf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_cgetrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zgetrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetri( int matrix_order, lapack_int n, float* a,
+                           lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_dgetri( int matrix_order, lapack_int n, double* a,
+                           lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_cgetri( int matrix_order, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv );
+lapack_int LAPACKE_zgetri( int matrix_order, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dgetrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_cgetrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zgetrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sggbak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const float* lscale,
+                           const float* rscale, lapack_int m, float* v,
+                           lapack_int ldv );
+lapack_int LAPACKE_dggbak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const double* lscale,
+                           const double* rscale, lapack_int m, double* v,
+                           lapack_int ldv );
+lapack_int LAPACKE_cggbak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const float* lscale,
+                           const float* rscale, lapack_int m,
+                           lapack_complex_float* v, lapack_int ldv );
+lapack_int LAPACKE_zggbak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const double* lscale,
+                           const double* rscale, lapack_int m,
+                           lapack_complex_double* v, lapack_int ldv );
+
+lapack_int LAPACKE_sggbal( int matrix_order, char job, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb,
+                           lapack_int* ilo, lapack_int* ihi, float* lscale,
+                           float* rscale );
+lapack_int LAPACKE_dggbal( int matrix_order, char job, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           lapack_int* ilo, lapack_int* ihi, double* lscale,
+                           double* rscale );
+lapack_int LAPACKE_cggbal( int matrix_order, char job, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_int* ilo, lapack_int* ihi, float* lscale,
+                           float* rscale );
+lapack_int LAPACKE_zggbal( int matrix_order, char job, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_int* ilo, lapack_int* ihi, double* lscale,
+                           double* rscale );
+
+lapack_int LAPACKE_sgges( int matrix_order, char jobvsl, char jobvsr, char sort,
+                          LAPACK_S_SELECT3 selctg, lapack_int n, float* a,
+                          lapack_int lda, float* b, lapack_int ldb,
+                          lapack_int* sdim, float* alphar, float* alphai,
+                          float* beta, float* vsl, lapack_int ldvsl, float* vsr,
+                          lapack_int ldvsr );
+lapack_int LAPACKE_dgges( int matrix_order, char jobvsl, char jobvsr, char sort,
+                          LAPACK_D_SELECT3 selctg, lapack_int n, double* a,
+                          lapack_int lda, double* b, lapack_int ldb,
+                          lapack_int* sdim, double* alphar, double* alphai,
+                          double* beta, double* vsl, lapack_int ldvsl,
+                          double* vsr, lapack_int ldvsr );
+lapack_int LAPACKE_cgges( int matrix_order, char jobvsl, char jobvsr, char sort,
+                          LAPACK_C_SELECT2 selctg, lapack_int n,
+                          lapack_complex_float* a, lapack_int lda,
+                          lapack_complex_float* b, lapack_int ldb,
+                          lapack_int* sdim, lapack_complex_float* alpha,
+                          lapack_complex_float* beta, lapack_complex_float* vsl,
+                          lapack_int ldvsl, lapack_complex_float* vsr,
+                          lapack_int ldvsr );
+lapack_int LAPACKE_zgges( int matrix_order, char jobvsl, char jobvsr, char sort,
+                          LAPACK_Z_SELECT2 selctg, lapack_int n,
+                          lapack_complex_double* a, lapack_int lda,
+                          lapack_complex_double* b, lapack_int ldb,
+                          lapack_int* sdim, lapack_complex_double* alpha,
+                          lapack_complex_double* beta,
+                          lapack_complex_double* vsl, lapack_int ldvsl,
+                          lapack_complex_double* vsr, lapack_int ldvsr );
+
+lapack_int LAPACKE_sggesx( int matrix_order, char jobvsl, char jobvsr,
+                           char sort, LAPACK_S_SELECT3 selctg, char sense,
+                           lapack_int n, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, lapack_int* sdim, float* alphar,
+                           float* alphai, float* beta, float* vsl,
+                           lapack_int ldvsl, float* vsr, lapack_int ldvsr,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_dggesx( int matrix_order, char jobvsl, char jobvsr,
+                           char sort, LAPACK_D_SELECT3 selctg, char sense,
+                           lapack_int n, double* a, lapack_int lda, double* b,
+                           lapack_int ldb, lapack_int* sdim, double* alphar,
+                           double* alphai, double* beta, double* vsl,
+                           lapack_int ldvsl, double* vsr, lapack_int ldvsr,
+                           double* rconde, double* rcondv );
+lapack_int LAPACKE_cggesx( int matrix_order, char jobvsl, char jobvsr,
+                           char sort, LAPACK_C_SELECT2 selctg, char sense,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, lapack_int* sdim,
+                           lapack_complex_float* alpha,
+                           lapack_complex_float* beta,
+                           lapack_complex_float* vsl, lapack_int ldvsl,
+                           lapack_complex_float* vsr, lapack_int ldvsr,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_zggesx( int matrix_order, char jobvsl, char jobvsr,
+                           char sort, LAPACK_Z_SELECT2 selctg, char sense,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_int* sdim,
+                           lapack_complex_double* alpha,
+                           lapack_complex_double* beta,
+                           lapack_complex_double* vsl, lapack_int ldvsl,
+                           lapack_complex_double* vsr, lapack_int ldvsr,
+                           double* rconde, double* rcondv );
+
+lapack_int LAPACKE_sggev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, float* a, lapack_int lda, float* b,
+                          lapack_int ldb, float* alphar, float* alphai,
+                          float* beta, float* vl, lapack_int ldvl, float* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_dggev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, double* a, lapack_int lda, double* b,
+                          lapack_int ldb, double* alphar, double* alphai,
+                          double* beta, double* vl, lapack_int ldvl, double* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_cggev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, lapack_complex_float* a, lapack_int lda,
+                          lapack_complex_float* b, lapack_int ldb,
+                          lapack_complex_float* alpha,
+                          lapack_complex_float* beta, lapack_complex_float* vl,
+                          lapack_int ldvl, lapack_complex_float* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_zggev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, lapack_complex_double* a,
+                          lapack_int lda, lapack_complex_double* b,
+                          lapack_int ldb, lapack_complex_double* alpha,
+                          lapack_complex_double* beta,
+                          lapack_complex_double* vl, lapack_int ldvl,
+                          lapack_complex_double* vr, lapack_int ldvr );
+
+lapack_int LAPACKE_sggevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb,
+                           float* alphar, float* alphai, float* beta, float* vl,
+                           lapack_int ldvl, float* vr, lapack_int ldvr,
+                           lapack_int* ilo, lapack_int* ihi, float* lscale,
+                           float* rscale, float* abnrm, float* bbnrm,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_dggevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           double* alphar, double* alphai, double* beta,
+                           double* vl, lapack_int ldvl, double* vr,
+                           lapack_int ldvr, lapack_int* ilo, lapack_int* ihi,
+                           double* lscale, double* rscale, double* abnrm,
+                           double* bbnrm, double* rconde, double* rcondv );
+lapack_int LAPACKE_cggevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* alpha,
+                           lapack_complex_float* beta, lapack_complex_float* vl,
+                           lapack_int ldvl, lapack_complex_float* vr,
+                           lapack_int ldvr, lapack_int* ilo, lapack_int* ihi,
+                           float* lscale, float* rscale, float* abnrm,
+                           float* bbnrm, float* rconde, float* rcondv );
+lapack_int LAPACKE_zggevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* alpha,
+                           lapack_complex_double* beta,
+                           lapack_complex_double* vl, lapack_int ldvl,
+                           lapack_complex_double* vr, lapack_int ldvr,
+                           lapack_int* ilo, lapack_int* ihi, double* lscale,
+                           double* rscale, double* abnrm, double* bbnrm,
+                           double* rconde, double* rcondv );
+
+lapack_int LAPACKE_sggglm( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, float* d, float* x, float* y );
+lapack_int LAPACKE_dggglm( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, double* a, lapack_int lda, double* b,
+                           lapack_int ldb, double* d, double* x, double* y );
+lapack_int LAPACKE_cggglm( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* d,
+                           lapack_complex_float* x, lapack_complex_float* y );
+lapack_int LAPACKE_zggglm( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* d,
+                           lapack_complex_double* x, lapack_complex_double* y );
+
+lapack_int LAPACKE_sgghrd( int matrix_order, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           float* a, lapack_int lda, float* b, lapack_int ldb,
+                           float* q, lapack_int ldq, float* z, lapack_int ldz );
+lapack_int LAPACKE_dgghrd( int matrix_order, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           double* a, lapack_int lda, double* b, lapack_int ldb,
+                           double* q, lapack_int ldq, double* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_cgghrd( int matrix_order, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zgghrd( int matrix_order, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sgglse( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int p, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, float* c, float* d, float* x );
+lapack_int LAPACKE_dgglse( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int p, double* a, lapack_int lda, double* b,
+                           lapack_int ldb, double* c, double* d, double* x );
+lapack_int LAPACKE_cgglse( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int p, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* c,
+                           lapack_complex_float* d, lapack_complex_float* x );
+lapack_int LAPACKE_zgglse( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int p, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* c,
+                           lapack_complex_double* d, lapack_complex_double* x );
+
+lapack_int LAPACKE_sggqrf( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, float* a, lapack_int lda, float* taua,
+                           float* b, lapack_int ldb, float* taub );
+lapack_int LAPACKE_dggqrf( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, double* a, lapack_int lda,
+                           double* taua, double* b, lapack_int ldb,
+                           double* taub );
+lapack_int LAPACKE_cggqrf( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* taua,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* taub );
+lapack_int LAPACKE_zggqrf( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* taua,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* taub );
+
+lapack_int LAPACKE_sggrqf( int matrix_order, lapack_int m, lapack_int p,
+                           lapack_int n, float* a, lapack_int lda, float* taua,
+                           float* b, lapack_int ldb, float* taub );
+lapack_int LAPACKE_dggrqf( int matrix_order, lapack_int m, lapack_int p,
+                           lapack_int n, double* a, lapack_int lda,
+                           double* taua, double* b, lapack_int ldb,
+                           double* taub );
+lapack_int LAPACKE_cggrqf( int matrix_order, lapack_int m, lapack_int p,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* taua,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* taub );
+lapack_int LAPACKE_zggrqf( int matrix_order, lapack_int m, lapack_int p,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* taua,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* taub );
+
+lapack_int LAPACKE_sggsvd( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* k, lapack_int* l, float* a,
+                           lapack_int lda, float* b, lapack_int ldb,
+                           float* alpha, float* beta, float* u, lapack_int ldu,
+                           float* v, lapack_int ldv, float* q, lapack_int ldq,
+                           lapack_int* iwork );
+lapack_int LAPACKE_dggsvd( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* k, lapack_int* l, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           double* alpha, double* beta, double* u,
+                           lapack_int ldu, double* v, lapack_int ldv, double* q,
+                           lapack_int ldq, lapack_int* iwork );
+lapack_int LAPACKE_cggsvd( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* k, lapack_int* l,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           float* alpha, float* beta, lapack_complex_float* u,
+                           lapack_int ldu, lapack_complex_float* v,
+                           lapack_int ldv, lapack_complex_float* q,
+                           lapack_int ldq, lapack_int* iwork );
+lapack_int LAPACKE_zggsvd( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* k, lapack_int* l,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           double* alpha, double* beta,
+                           lapack_complex_double* u, lapack_int ldu,
+                           lapack_complex_double* v, lapack_int ldv,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_int* iwork );
+
+lapack_int LAPACKE_sggsvp( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb, float tola,
+                           float tolb, lapack_int* k, lapack_int* l, float* u,
+                           lapack_int ldu, float* v, lapack_int ldv, float* q,
+                           lapack_int ldq );
+lapack_int LAPACKE_dggsvp( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           double tola, double tolb, lapack_int* k,
+                           lapack_int* l, double* u, lapack_int ldu, double* v,
+                           lapack_int ldv, double* q, lapack_int ldq );
+lapack_int LAPACKE_cggsvp( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb, float tola,
+                           float tolb, lapack_int* k, lapack_int* l,
+                           lapack_complex_float* u, lapack_int ldu,
+                           lapack_complex_float* v, lapack_int ldv,
+                           lapack_complex_float* q, lapack_int ldq );
+lapack_int LAPACKE_zggsvp( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           double tola, double tolb, lapack_int* k,
+                           lapack_int* l, lapack_complex_double* u,
+                           lapack_int ldu, lapack_complex_double* v,
+                           lapack_int ldv, lapack_complex_double* q,
+                           lapack_int ldq );
+
+lapack_int LAPACKE_sgtcon( char norm, lapack_int n, const float* dl,
+                           const float* d, const float* du, const float* du2,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_dgtcon( char norm, lapack_int n, const double* dl,
+                           const double* d, const double* du, const double* du2,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+lapack_int LAPACKE_cgtcon( char norm, lapack_int n,
+                           const lapack_complex_float* dl,
+                           const lapack_complex_float* d,
+                           const lapack_complex_float* du,
+                           const lapack_complex_float* du2,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zgtcon( char norm, lapack_int n,
+                           const lapack_complex_double* dl,
+                           const lapack_complex_double* d,
+                           const lapack_complex_double* du,
+                           const lapack_complex_double* du2,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_sgtrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const float* dl, const float* d,
+                           const float* du, const float* dlf, const float* df,
+                           const float* duf, const float* du2,
+                           const lapack_int* ipiv, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dgtrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const double* dl, const double* d,
+                           const double* du, const double* dlf,
+                           const double* df, const double* duf,
+                           const double* du2, const lapack_int* ipiv,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_cgtrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* dl,
+                           const lapack_complex_float* d,
+                           const lapack_complex_float* du,
+                           const lapack_complex_float* dlf,
+                           const lapack_complex_float* df,
+                           const lapack_complex_float* duf,
+                           const lapack_complex_float* du2,
+                           const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zgtrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* dl,
+                           const lapack_complex_double* d,
+                           const lapack_complex_double* du,
+                           const lapack_complex_double* dlf,
+                           const lapack_complex_double* df,
+                           const lapack_complex_double* duf,
+                           const lapack_complex_double* du2,
+                           const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sgtsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          float* dl, float* d, float* du, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dgtsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          double* dl, double* d, double* du, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_cgtsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          lapack_complex_float* dl, lapack_complex_float* d,
+                          lapack_complex_float* du, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zgtsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          lapack_complex_double* dl, lapack_complex_double* d,
+                          lapack_complex_double* du, lapack_complex_double* b,
+                          lapack_int ldb );
+
+lapack_int LAPACKE_sgtsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs, const float* dl,
+                           const float* d, const float* du, float* dlf,
+                           float* df, float* duf, float* du2, lapack_int* ipiv,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dgtsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs, const double* dl,
+                           const double* d, const double* du, double* dlf,
+                           double* df, double* duf, double* du2,
+                           lapack_int* ipiv, const double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* rcond,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cgtsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* dl,
+                           const lapack_complex_float* d,
+                           const lapack_complex_float* du,
+                           lapack_complex_float* dlf, lapack_complex_float* df,
+                           lapack_complex_float* duf, lapack_complex_float* du2,
+                           lapack_int* ipiv, const lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zgtsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* dl,
+                           const lapack_complex_double* d,
+                           const lapack_complex_double* du,
+                           lapack_complex_double* dlf,
+                           lapack_complex_double* df,
+                           lapack_complex_double* duf,
+                           lapack_complex_double* du2, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_sgttrf( lapack_int n, float* dl, float* d, float* du,
+                           float* du2, lapack_int* ipiv );
+lapack_int LAPACKE_dgttrf( lapack_int n, double* dl, double* d, double* du,
+                           double* du2, lapack_int* ipiv );
+lapack_int LAPACKE_cgttrf( lapack_int n, lapack_complex_float* dl,
+                           lapack_complex_float* d, lapack_complex_float* du,
+                           lapack_complex_float* du2, lapack_int* ipiv );
+lapack_int LAPACKE_zgttrf( lapack_int n, lapack_complex_double* dl,
+                           lapack_complex_double* d, lapack_complex_double* du,
+                           lapack_complex_double* du2, lapack_int* ipiv );
+
+lapack_int LAPACKE_sgttrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const float* dl, const float* d,
+                           const float* du, const float* du2,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dgttrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const double* dl, const double* d,
+                           const double* du, const double* du2,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_cgttrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* dl,
+                           const lapack_complex_float* d,
+                           const lapack_complex_float* du,
+                           const lapack_complex_float* du2,
+                           const lapack_int* ipiv, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zgttrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* dl,
+                           const lapack_complex_double* d,
+                           const lapack_complex_double* du,
+                           const lapack_complex_double* du2,
+                           const lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_chbev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int kd, lapack_complex_float* ab,
+                          lapack_int ldab, float* w, lapack_complex_float* z,
+                          lapack_int ldz );
+lapack_int LAPACKE_zhbev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int kd, lapack_complex_double* ab,
+                          lapack_int ldab, double* w, lapack_complex_double* z,
+                          lapack_int ldz );
+
+lapack_int LAPACKE_chbevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_float* ab,
+                           lapack_int ldab, float* w, lapack_complex_float* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_zhbevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_double* ab,
+                           lapack_int ldab, double* w, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_chbevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int kd,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* q, lapack_int ldq, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_zhbevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int kd,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* q, lapack_int ldq, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chbgst( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_complex_float* bb, lapack_int ldbb,
+                           lapack_complex_float* x, lapack_int ldx );
+lapack_int LAPACKE_zhbgst( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_complex_double* bb, lapack_int ldbb,
+                           lapack_complex_double* x, lapack_int ldx );
+
+lapack_int LAPACKE_chbgv( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int ka, lapack_int kb,
+                          lapack_complex_float* ab, lapack_int ldab,
+                          lapack_complex_float* bb, lapack_int ldbb, float* w,
+                          lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhbgv( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int ka, lapack_int kb,
+                          lapack_complex_double* ab, lapack_int ldab,
+                          lapack_complex_double* bb, lapack_int ldbb, double* w,
+                          lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chbgvd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* bb, lapack_int ldbb, float* w,
+                           lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhbgvd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* bb, lapack_int ldbb,
+                           double* w, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_chbgvx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int ka, lapack_int kb,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* bb, lapack_int ldbb,
+                           lapack_complex_float* q, lapack_int ldq, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_zhbgvx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int ka, lapack_int kb,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* bb, lapack_int ldbb,
+                           lapack_complex_double* q, lapack_int ldq, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chbtrd( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_float* ab,
+                           lapack_int ldab, float* d, float* e,
+                           lapack_complex_float* q, lapack_int ldq );
+lapack_int LAPACKE_zhbtrd( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_double* ab,
+                           lapack_int ldab, double* d, double* e,
+                           lapack_complex_double* q, lapack_int ldq );
+
+lapack_int LAPACKE_checon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zhecon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_cheequb( int matrix_order, char uplo, lapack_int n,
+                            const lapack_complex_float* a, lapack_int lda,
+                            float* s, float* scond, float* amax );
+lapack_int LAPACKE_zheequb( int matrix_order, char uplo, lapack_int n,
+                            const lapack_complex_double* a, lapack_int lda,
+                            double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_cheev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_complex_float* a, lapack_int lda, float* w );
+lapack_int LAPACKE_zheev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_complex_double* a, lapack_int lda, double* w );
+
+lapack_int LAPACKE_cheevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda, float* w );
+lapack_int LAPACKE_zheevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           double* w );
+
+lapack_int LAPACKE_cheevr( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, float vl, float vu, lapack_int il,
+                           lapack_int iu, float abstol, lapack_int* m, float* w,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int* isuppz );
+lapack_int LAPACKE_zheevr( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, double vl, double vu, lapack_int il,
+                           lapack_int iu, double abstol, lapack_int* m,
+                           double* w, lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* isuppz );
+
+lapack_int LAPACKE_cheevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, float vl, float vu, lapack_int il,
+                           lapack_int iu, float abstol, lapack_int* m, float* w,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_zheevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, double vl, double vu, lapack_int il,
+                           lapack_int iu, double abstol, lapack_int* m,
+                           double* w, lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chegst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zhegst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_chegv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, lapack_complex_float* a,
+                          lapack_int lda, lapack_complex_float* b,
+                          lapack_int ldb, float* w );
+lapack_int LAPACKE_zhegv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, lapack_complex_double* a,
+                          lapack_int lda, lapack_complex_double* b,
+                          lapack_int ldb, double* w );
+
+lapack_int LAPACKE_chegvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, float* w );
+lapack_int LAPACKE_zhegvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, double* w );
+
+lapack_int LAPACKE_chegvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_zhegvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_cherfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zherfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_cherfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_float* a, lapack_int lda,
+                            const lapack_complex_float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* s,
+                            const lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zherfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_double* a, lapack_int lda,
+                            const lapack_complex_double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* s,
+                            const lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_chesv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* a,
+                          lapack_int lda, lapack_int* ipiv,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zhesv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* a,
+                          lapack_int lda, lapack_int* ipiv,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_chesvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* af,
+                           lapack_int ldaf, lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zhesvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* af,
+                           lapack_int ldaf, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_chesvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* s,
+                            lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zhesvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* s,
+                            lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_chetrd( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda, float* d,
+                           float* e, lapack_complex_float* tau );
+lapack_int LAPACKE_zhetrd( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda, double* d,
+                           double* e, lapack_complex_double* tau );
+
+lapack_int LAPACKE_chetrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zhetrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_chetri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv );
+lapack_int LAPACKE_zhetri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv );
+
+lapack_int LAPACKE_chetrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zhetrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_chfrk( int matrix_order, char transr, char uplo, char trans,
+                          lapack_int n, lapack_int k, float alpha,
+                          const lapack_complex_float* a, lapack_int lda,
+                          float beta, lapack_complex_float* c );
+lapack_int LAPACKE_zhfrk( int matrix_order, char transr, char uplo, char trans,
+                          lapack_int n, lapack_int k, double alpha,
+                          const lapack_complex_double* a, lapack_int lda,
+                          double beta, lapack_complex_double* c );
+
+lapack_int LAPACKE_shgeqz( int matrix_order, char job, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           float* h, lapack_int ldh, float* t, lapack_int ldt,
+                           float* alphar, float* alphai, float* beta, float* q,
+                           lapack_int ldq, float* z, lapack_int ldz );
+lapack_int LAPACKE_dhgeqz( int matrix_order, char job, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           double* h, lapack_int ldh, double* t, lapack_int ldt,
+                           double* alphar, double* alphai, double* beta,
+                           double* q, lapack_int ldq, double* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_chgeqz( int matrix_order, char job, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           lapack_complex_float* h, lapack_int ldh,
+                           lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* alpha,
+                           lapack_complex_float* beta, lapack_complex_float* q,
+                           lapack_int ldq, lapack_complex_float* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_zhgeqz( int matrix_order, char job, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           lapack_complex_double* h, lapack_int ldh,
+                           lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* alpha,
+                           lapack_complex_double* beta,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zhpcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_chpev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_complex_float* ap, float* w,
+                          lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhpev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_complex_double* ap, double* w,
+                          lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_complex_float* ap, float* w,
+                           lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhpevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_complex_double* ap, double* w,
+                           lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_float* ap, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_zhpevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_double* ap, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chpgst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, lapack_complex_float* ap,
+                           const lapack_complex_float* bp );
+lapack_int LAPACKE_zhpgst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, lapack_complex_double* ap,
+                           const lapack_complex_double* bp );
+
+lapack_int LAPACKE_chpgv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, lapack_complex_float* ap,
+                          lapack_complex_float* bp, float* w,
+                          lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhpgv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, lapack_complex_double* ap,
+                          lapack_complex_double* bp, double* w,
+                          lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpgvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, lapack_complex_float* ap,
+                           lapack_complex_float* bp, float* w,
+                           lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhpgvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, lapack_complex_double* ap,
+                           lapack_complex_double* bp, double* w,
+                           lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpgvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n,
+                           lapack_complex_float* ap, lapack_complex_float* bp,
+                           float vl, float vu, lapack_int il, lapack_int iu,
+                           float abstol, lapack_int* m, float* w,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_zhpgvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n,
+                           lapack_complex_double* ap, lapack_complex_double* bp,
+                           double vl, double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_complex_float* afp,
+                           const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zhprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_complex_double* afp,
+                           const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_chpsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* ap,
+                          lapack_int* ipiv, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zhpsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* ap,
+                          lapack_int* ipiv, lapack_complex_double* b,
+                          lapack_int ldb );
+
+lapack_int LAPACKE_chpsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           lapack_complex_float* afp, lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zhpsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           lapack_complex_double* afp, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_chptrd( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, float* d, float* e,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zhptrd( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, double* d, double* e,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_chptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_zhptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, lapack_int* ipiv );
+
+lapack_int LAPACKE_chptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, const lapack_int* ipiv );
+lapack_int LAPACKE_zhptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, const lapack_int* ipiv );
+
+lapack_int LAPACKE_chptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_int* ipiv, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zhptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_shsein( int matrix_order, char job, char eigsrc, char initv,
+                           lapack_logical* select, lapack_int n, const float* h,
+                           lapack_int ldh, float* wr, const float* wi,
+                           float* vl, lapack_int ldvl, float* vr,
+                           lapack_int ldvr, lapack_int mm, lapack_int* m,
+                           lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_dhsein( int matrix_order, char job, char eigsrc, char initv,
+                           lapack_logical* select, lapack_int n,
+                           const double* h, lapack_int ldh, double* wr,
+                           const double* wi, double* vl, lapack_int ldvl,
+                           double* vr, lapack_int ldvr, lapack_int mm,
+                           lapack_int* m, lapack_int* ifaill,
+                           lapack_int* ifailr );
+lapack_int LAPACKE_chsein( int matrix_order, char job, char eigsrc, char initv,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_float* h, lapack_int ldh,
+                           lapack_complex_float* w, lapack_complex_float* vl,
+                           lapack_int ldvl, lapack_complex_float* vr,
+                           lapack_int ldvr, lapack_int mm, lapack_int* m,
+                           lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_zhsein( int matrix_order, char job, char eigsrc, char initv,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_double* h, lapack_int ldh,
+                           lapack_complex_double* w, lapack_complex_double* vl,
+                           lapack_int ldvl, lapack_complex_double* vr,
+                           lapack_int ldvr, lapack_int mm, lapack_int* m,
+                           lapack_int* ifaill, lapack_int* ifailr );
+
+lapack_int LAPACKE_shseqr( int matrix_order, char job, char compz, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, float* h,
+                           lapack_int ldh, float* wr, float* wi, float* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_dhseqr( int matrix_order, char job, char compz, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, double* h,
+                           lapack_int ldh, double* wr, double* wi, double* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_chseqr( int matrix_order, char job, char compz, lapack_int n,
+                           lapack_int ilo, lapack_int ihi,
+                           lapack_complex_float* h, lapack_int ldh,
+                           lapack_complex_float* w, lapack_complex_float* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_zhseqr( int matrix_order, char job, char compz, lapack_int n,
+                           lapack_int ilo, lapack_int ihi,
+                           lapack_complex_double* h, lapack_int ldh,
+                           lapack_complex_double* w, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_clacgv( lapack_int n, lapack_complex_float* x,
+                           lapack_int incx );
+lapack_int LAPACKE_zlacgv( lapack_int n, lapack_complex_double* x,
+                           lapack_int incx );
+
+lapack_int LAPACKE_slacpy( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, const float* a, lapack_int lda, float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_dlacpy( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, const double* a, lapack_int lda, double* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_clacpy( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zlacpy( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_zlag2c( int matrix_order, lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_float* sa, lapack_int ldsa );
+
+lapack_int LAPACKE_slag2d( int matrix_order, lapack_int m, lapack_int n,
+                           const float* sa, lapack_int ldsa, double* a,
+                           lapack_int lda );
+
+lapack_int LAPACKE_dlag2s( int matrix_order, lapack_int m, lapack_int n,
+                           const double* a, lapack_int lda, float* sa,
+                           lapack_int ldsa );
+
+lapack_int LAPACKE_clag2z( int matrix_order, lapack_int m, lapack_int n,
+                           const lapack_complex_float* sa, lapack_int ldsa,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_slagge( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const float* d,
+                           float* a, lapack_int lda, lapack_int* iseed );
+lapack_int LAPACKE_dlagge( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const double* d,
+                           double* a, lapack_int lda, lapack_int* iseed );
+lapack_int LAPACKE_clagge( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const float* d,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* iseed );
+lapack_int LAPACKE_zlagge( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const double* d,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* iseed );
+
+float LAPACKE_slamch( char cmach );
+double LAPACKE_dlamch( char cmach );
+
+float LAPACKE_slange( int matrix_order, char norm, lapack_int m,
+                           lapack_int n, const float* a, lapack_int lda );
+double LAPACKE_dlange( int matrix_order, char norm, lapack_int m,
+                           lapack_int n, const double* a, lapack_int lda );
+float LAPACKE_clange( int matrix_order, char norm, lapack_int m,
+                           lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda );
+double LAPACKE_zlange( int matrix_order, char norm, lapack_int m,
+                           lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda );
+
+float LAPACKE_clanhe( int matrix_order, char norm, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda );
+double LAPACKE_zlanhe( int matrix_order, char norm, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda );
+
+float LAPACKE_slansy( int matrix_order, char norm, char uplo, lapack_int n,
+                           const float* a, lapack_int lda );
+double LAPACKE_dlansy( int matrix_order, char norm, char uplo, lapack_int n,
+                           const double* a, lapack_int lda );
+float LAPACKE_clansy( int matrix_order, char norm, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda );
+double LAPACKE_zlansy( int matrix_order, char norm, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda );
+
+float LAPACKE_slantr( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int m, lapack_int n, const float* a,
+                           lapack_int lda );
+double LAPACKE_dlantr( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int m, lapack_int n, const double* a,
+                           lapack_int lda );
+float LAPACKE_clantr( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int m, lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda );
+double LAPACKE_zlantr( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int m, lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda );
+
+
+lapack_int LAPACKE_slarfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, const float* v, lapack_int ldv,
+                           const float* t, lapack_int ldt, float* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_dlarfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, const double* v, lapack_int ldv,
+                           const double* t, lapack_int ldt, double* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_clarfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, const lapack_complex_float* v,
+                           lapack_int ldv, const lapack_complex_float* t,
+                           lapack_int ldt, lapack_complex_float* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_zlarfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, const lapack_complex_double* v,
+                           lapack_int ldv, const lapack_complex_double* t,
+                           lapack_int ldt, lapack_complex_double* c,
+                           lapack_int ldc );
+
+lapack_int LAPACKE_slarfg( lapack_int n, float* alpha, float* x,
+                           lapack_int incx, float* tau );
+lapack_int LAPACKE_dlarfg( lapack_int n, double* alpha, double* x,
+                           lapack_int incx, double* tau );
+lapack_int LAPACKE_clarfg( lapack_int n, lapack_complex_float* alpha,
+                           lapack_complex_float* x, lapack_int incx,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zlarfg( lapack_int n, lapack_complex_double* alpha,
+                           lapack_complex_double* x, lapack_int incx,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_slarft( int matrix_order, char direct, char storev,
+                           lapack_int n, lapack_int k, const float* v,
+                           lapack_int ldv, const float* tau, float* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_dlarft( int matrix_order, char direct, char storev,
+                           lapack_int n, lapack_int k, const double* v,
+                           lapack_int ldv, const double* tau, double* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_clarft( int matrix_order, char direct, char storev,
+                           lapack_int n, lapack_int k,
+                           const lapack_complex_float* v, lapack_int ldv,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zlarft( int matrix_order, char direct, char storev,
+                           lapack_int n, lapack_int k,
+                           const lapack_complex_double* v, lapack_int ldv,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_slarfx( int matrix_order, char side, lapack_int m,
+                           lapack_int n, const float* v, float tau, float* c,
+                           lapack_int ldc, float* work );
+lapack_int LAPACKE_dlarfx( int matrix_order, char side, lapack_int m,
+                           lapack_int n, const double* v, double tau, double* c,
+                           lapack_int ldc, double* work );
+lapack_int LAPACKE_clarfx( int matrix_order, char side, lapack_int m,
+                           lapack_int n, const lapack_complex_float* v,
+                           lapack_complex_float tau, lapack_complex_float* c,
+                           lapack_int ldc, lapack_complex_float* work );
+lapack_int LAPACKE_zlarfx( int matrix_order, char side, lapack_int m,
+                           lapack_int n, const lapack_complex_double* v,
+                           lapack_complex_double tau, lapack_complex_double* c,
+                           lapack_int ldc, lapack_complex_double* work );
+
+lapack_int LAPACKE_slarnv( lapack_int idist, lapack_int* iseed, lapack_int n,
+                           float* x );
+lapack_int LAPACKE_dlarnv( lapack_int idist, lapack_int* iseed, lapack_int n,
+                           double* x );
+lapack_int LAPACKE_clarnv( lapack_int idist, lapack_int* iseed, lapack_int n,
+                           lapack_complex_float* x );
+lapack_int LAPACKE_zlarnv( lapack_int idist, lapack_int* iseed, lapack_int n,
+                           lapack_complex_double* x );
+
+lapack_int LAPACKE_slaset( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, float alpha, float beta, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dlaset( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, double alpha, double beta, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_claset( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, lapack_complex_float alpha,
+                           lapack_complex_float beta, lapack_complex_float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_zlaset( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, lapack_complex_double alpha,
+                           lapack_complex_double beta, lapack_complex_double* a,
+                           lapack_int lda );
+
+lapack_int LAPACKE_slasrt( char id, lapack_int n, float* d );
+lapack_int LAPACKE_dlasrt( char id, lapack_int n, double* d );
+
+lapack_int LAPACKE_slaswp( int matrix_order, lapack_int n, float* a,
+                           lapack_int lda, lapack_int k1, lapack_int k2,
+                           const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_dlaswp( int matrix_order, lapack_int n, double* a,
+                           lapack_int lda, lapack_int k1, lapack_int k2,
+                           const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_claswp( int matrix_order, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int k1, lapack_int k2, const lapack_int* ipiv,
+                           lapack_int incx );
+lapack_int LAPACKE_zlaswp( int matrix_order, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int k1, lapack_int k2, const lapack_int* ipiv,
+                           lapack_int incx );
+
+lapack_int LAPACKE_slatms( int matrix_order, lapack_int m, lapack_int n,
+                           char dist, lapack_int* iseed, char sym, float* d,
+                           lapack_int mode, float cond, float dmax,
+                           lapack_int kl, lapack_int ku, char pack, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dlatms( int matrix_order, lapack_int m, lapack_int n,
+                           char dist, lapack_int* iseed, char sym, double* d,
+                           lapack_int mode, double cond, double dmax,
+                           lapack_int kl, lapack_int ku, char pack, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_clatms( int matrix_order, lapack_int m, lapack_int n,
+                           char dist, lapack_int* iseed, char sym, float* d,
+                           lapack_int mode, float cond, float dmax,
+                           lapack_int kl, lapack_int ku, char pack,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zlatms( int matrix_order, lapack_int m, lapack_int n,
+                           char dist, lapack_int* iseed, char sym, double* d,
+                           lapack_int mode, double cond, double dmax,
+                           lapack_int kl, lapack_int ku, char pack,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_slauum( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dlauum( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_clauum( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zlauum( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_sopgtr( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, const float* tau, float* q,
+                           lapack_int ldq );
+lapack_int LAPACKE_dopgtr( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, const double* tau, double* q,
+                           lapack_int ldq );
+
+lapack_int LAPACKE_sopmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n, const float* ap,
+                           const float* tau, float* c, lapack_int ldc );
+lapack_int LAPACKE_dopmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n, const double* ap,
+                           const double* tau, double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sorgbr( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorgbr( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int k, double* a,
+                           lapack_int lda, const double* tau );
+
+lapack_int LAPACKE_sorghr( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorghr( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorglq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorglq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorgql( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorgql( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorgqr( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorgqr( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorgrq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorgrq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorgtr( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, const float* tau );
+lapack_int LAPACKE_dorgtr( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, const double* tau );
+
+lapack_int LAPACKE_sormbr( int matrix_order, char vect, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormbr( int matrix_order, char vect, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormhr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, const float* a, lapack_int lda,
+                           const float* tau, float* c, lapack_int ldc );
+lapack_int LAPACKE_dormhr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, const double* a, lapack_int lda,
+                           const double* tau, double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormlq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormlq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormql( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormql( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormqr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormqr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormrq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormrq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormrz( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           lapack_int l, const float* a, lapack_int lda,
+                           const float* tau, float* c, lapack_int ldc );
+lapack_int LAPACKE_dormrz( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           lapack_int l, const double* a, lapack_int lda,
+                           const double* tau, double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n, const float* a,
+                           lapack_int lda, const float* tau, float* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_dormtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n, const double* a,
+                           lapack_int lda, const double* tau, double* c,
+                           lapack_int ldc );
+
+lapack_int LAPACKE_spbcon( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const float* ab, lapack_int ldab,
+                           float anorm, float* rcond );
+lapack_int LAPACKE_dpbcon( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const double* ab, lapack_int ldab,
+                           double anorm, double* rcond );
+lapack_int LAPACKE_cpbcon( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const lapack_complex_float* ab,
+                           lapack_int ldab, float anorm, float* rcond );
+lapack_int LAPACKE_zpbcon( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const lapack_complex_double* ab,
+                           lapack_int ldab, double anorm, double* rcond );
+
+lapack_int LAPACKE_spbequ( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const float* ab, lapack_int ldab,
+                           float* s, float* scond, float* amax );
+lapack_int LAPACKE_dpbequ( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const double* ab, lapack_int ldab,
+                           double* s, double* scond, double* amax );
+lapack_int LAPACKE_cpbequ( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const lapack_complex_float* ab,
+                           lapack_int ldab, float* s, float* scond,
+                           float* amax );
+lapack_int LAPACKE_zpbequ( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const lapack_complex_double* ab,
+                           lapack_int ldab, double* s, double* scond,
+                           double* amax );
+
+lapack_int LAPACKE_spbrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, const float* ab,
+                           lapack_int ldab, const float* afb, lapack_int ldafb,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dpbrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, const double* ab,
+                           lapack_int ldab, const double* afb, lapack_int ldafb,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_cpbrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_complex_float* afb, lapack_int ldafb,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zpbrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_complex_double* afb, lapack_int ldafb,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_spbstf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kb, float* bb, lapack_int ldbb );
+lapack_int LAPACKE_dpbstf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kb, double* bb, lapack_int ldbb );
+lapack_int LAPACKE_cpbstf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kb, lapack_complex_float* bb,
+                           lapack_int ldbb );
+lapack_int LAPACKE_zpbstf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kb, lapack_complex_double* bb,
+                           lapack_int ldbb );
+
+lapack_int LAPACKE_spbsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int kd, lapack_int nrhs, float* ab,
+                          lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dpbsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int kd, lapack_int nrhs, double* ab,
+                          lapack_int ldab, double* b, lapack_int ldb );
+lapack_int LAPACKE_cpbsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int kd, lapack_int nrhs,
+                          lapack_complex_float* ab, lapack_int ldab,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpbsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int kd, lapack_int nrhs,
+                          lapack_complex_double* ab, lapack_int ldab,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spbsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, float* ab,
+                           lapack_int ldab, float* afb, lapack_int ldafb,
+                           char* equed, float* s, float* b, lapack_int ldb,
+                           float* x, lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dpbsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, double* ab,
+                           lapack_int ldab, double* afb, lapack_int ldafb,
+                           char* equed, double* s, double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* rcond,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cpbsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* afb, lapack_int ldafb,
+                           char* equed, float* s, lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zpbsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* afb, lapack_int ldafb,
+                           char* equed, double* s, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, double* rcond, double* ferr,
+                           double* berr );
+
+lapack_int LAPACKE_spbtrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, float* ab, lapack_int ldab );
+lapack_int LAPACKE_dpbtrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, double* ab, lapack_int ldab );
+lapack_int LAPACKE_cpbtrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_float* ab,
+                           lapack_int ldab );
+lapack_int LAPACKE_zpbtrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_double* ab,
+                           lapack_int ldab );
+
+lapack_int LAPACKE_spbtrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, const float* ab,
+                           lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dpbtrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, const double* ab,
+                           lapack_int ldab, double* b, lapack_int ldb );
+lapack_int LAPACKE_cpbtrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpbtrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spftrf( int matrix_order, char transr, char uplo,
+                           lapack_int n, float* a );
+lapack_int LAPACKE_dpftrf( int matrix_order, char transr, char uplo,
+                           lapack_int n, double* a );
+lapack_int LAPACKE_cpftrf( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_zpftrf( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_spftri( int matrix_order, char transr, char uplo,
+                           lapack_int n, float* a );
+lapack_int LAPACKE_dpftri( int matrix_order, char transr, char uplo,
+                           lapack_int n, double* a );
+lapack_int LAPACKE_cpftri( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_zpftri( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_spftrs( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_int nrhs, const float* a,
+                           float* b, lapack_int ldb );
+lapack_int LAPACKE_dpftrs( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_int nrhs, const double* a,
+                           double* b, lapack_int ldb );
+lapack_int LAPACKE_cpftrs( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* a,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpftrs( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* a,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spocon( int matrix_order, char uplo, lapack_int n,
+                           const float* a, lapack_int lda, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_dpocon( int matrix_order, char uplo, lapack_int n,
+                           const double* a, lapack_int lda, double anorm,
+                           double* rcond );
+lapack_int LAPACKE_cpocon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           float anorm, float* rcond );
+lapack_int LAPACKE_zpocon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           double anorm, double* rcond );
+
+lapack_int LAPACKE_spoequ( int matrix_order, lapack_int n, const float* a,
+                           lapack_int lda, float* s, float* scond,
+                           float* amax );
+lapack_int LAPACKE_dpoequ( int matrix_order, lapack_int n, const double* a,
+                           lapack_int lda, double* s, double* scond,
+                           double* amax );
+lapack_int LAPACKE_cpoequ( int matrix_order, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           float* s, float* scond, float* amax );
+lapack_int LAPACKE_zpoequ( int matrix_order, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_spoequb( int matrix_order, lapack_int n, const float* a,
+                            lapack_int lda, float* s, float* scond,
+                            float* amax );
+lapack_int LAPACKE_dpoequb( int matrix_order, lapack_int n, const double* a,
+                            lapack_int lda, double* s, double* scond,
+                            double* amax );
+lapack_int LAPACKE_cpoequb( int matrix_order, lapack_int n,
+                            const lapack_complex_float* a, lapack_int lda,
+                            float* s, float* scond, float* amax );
+lapack_int LAPACKE_zpoequb( int matrix_order, lapack_int n,
+                            const lapack_complex_double* a, lapack_int lda,
+                            double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_sporfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const float* af, lapack_int ldaf, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dporfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const double* af, lapack_int ldaf, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cporfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* af,
+                           lapack_int ldaf, const lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_zporfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* af,
+                           lapack_int ldaf, const lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+
+lapack_int LAPACKE_sporfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs, const float* a,
+                            lapack_int lda, const float* af, lapack_int ldaf,
+                            const float* s, const float* b, lapack_int ldb,
+                            float* x, lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dporfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs, const double* a,
+                            lapack_int lda, const double* af, lapack_int ldaf,
+                            const double* s, const double* b, lapack_int ldb,
+                            double* x, lapack_int ldx, double* rcond,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cporfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_float* a, lapack_int lda,
+                            const lapack_complex_float* af, lapack_int ldaf,
+                            const float* s, const lapack_complex_float* b,
+                            lapack_int ldb, lapack_complex_float* x,
+                            lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zporfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_double* a, lapack_int lda,
+                            const lapack_complex_double* af, lapack_int ldaf,
+                            const double* s, const lapack_complex_double* b,
+                            lapack_int ldb, lapack_complex_double* x,
+                            lapack_int ldx, double* rcond, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_sposv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, float* a, lapack_int lda, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dposv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, double* a, lapack_int lda, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_cposv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* a,
+                          lapack_int lda, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zposv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* a,
+                          lapack_int lda, lapack_complex_double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dsposv( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, double* x, lapack_int ldx,
+                           lapack_int* iter );
+lapack_int LAPACKE_zcposv( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, lapack_int* iter );
+
+lapack_int LAPACKE_sposvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, float* a, lapack_int lda, float* af,
+                           lapack_int ldaf, char* equed, float* s, float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_dposvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* af, lapack_int ldaf, char* equed, double* s,
+                           double* b, lapack_int ldb, double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+lapack_int LAPACKE_cposvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* af,
+                           lapack_int ldaf, char* equed, float* s,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zposvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* af,
+                           lapack_int ldaf, char* equed, double* s,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_sposvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs, float* a,
+                            lapack_int lda, float* af, lapack_int ldaf,
+                            char* equed, float* s, float* b, lapack_int ldb,
+                            float* x, lapack_int ldx, float* rcond,
+                            float* rpvgrw, float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_dposvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs, double* a,
+                            lapack_int lda, double* af, lapack_int ldaf,
+                            char* equed, double* s, double* b, lapack_int ldb,
+                            double* x, lapack_int ldx, double* rcond,
+                            double* rpvgrw, double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cposvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* af, lapack_int ldaf,
+                            char* equed, float* s, lapack_complex_float* b,
+                            lapack_int ldb, lapack_complex_float* x,
+                            lapack_int ldx, float* rcond, float* rpvgrw,
+                            float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zposvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* af, lapack_int ldaf,
+                            char* equed, double* s, lapack_complex_double* b,
+                            lapack_int ldb, lapack_complex_double* x,
+                            lapack_int ldx, double* rcond, double* rpvgrw,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_spotrf( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dpotrf( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_cpotrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zpotrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_spotri( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dpotri( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_cpotri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zpotri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_spotrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           float* b, lapack_int ldb );
+lapack_int LAPACKE_dpotrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           double* b, lapack_int ldb );
+lapack_int LAPACKE_cpotrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zpotrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_sppcon( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, float anorm, float* rcond );
+lapack_int LAPACKE_dppcon( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, double anorm, double* rcond );
+lapack_int LAPACKE_cppcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_zppcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_sppequ( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, float* s, float* scond,
+                           float* amax );
+lapack_int LAPACKE_dppequ( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, double* s, double* scond,
+                           double* amax );
+lapack_int LAPACKE_cppequ( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap, float* s,
+                           float* scond, float* amax );
+lapack_int LAPACKE_zppequ( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap, double* s,
+                           double* scond, double* amax );
+
+lapack_int LAPACKE_spprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap, const float* afp,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dpprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap, const double* afp,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_cpprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_complex_float* afp,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zpprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_complex_double* afp,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sppsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, float* ap, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dppsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, double* ap, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_cppsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* ap,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zppsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* ap,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sppsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, float* ap, float* afp, char* equed,
+                           float* s, float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dppsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, double* ap, double* afp,
+                           char* equed, double* s, double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* rcond,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cppsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* ap,
+                           lapack_complex_float* afp, char* equed, float* s,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zppsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* ap,
+                           lapack_complex_double* afp, char* equed, double* s,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_spptrf( int matrix_order, char uplo, lapack_int n,
+                           float* ap );
+lapack_int LAPACKE_dpptrf( int matrix_order, char uplo, lapack_int n,
+                           double* ap );
+lapack_int LAPACKE_cpptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_zpptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_spptri( int matrix_order, char uplo, lapack_int n,
+                           float* ap );
+lapack_int LAPACKE_dpptri( int matrix_order, char uplo, lapack_int n,
+                           double* ap );
+lapack_int LAPACKE_cpptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_zpptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_spptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap, float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_dpptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap, double* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_cpptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spstrf( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, lapack_int* piv, lapack_int* rank,
+                           float tol );
+lapack_int LAPACKE_dpstrf( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, lapack_int* piv, lapack_int* rank,
+                           double tol );
+lapack_int LAPACKE_cpstrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* piv, lapack_int* rank, float tol );
+lapack_int LAPACKE_zpstrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* piv, lapack_int* rank, double tol );
+
+lapack_int LAPACKE_sptcon( lapack_int n, const float* d, const float* e,
+                           float anorm, float* rcond );
+lapack_int LAPACKE_dptcon( lapack_int n, const double* d, const double* e,
+                           double anorm, double* rcond );
+lapack_int LAPACKE_cptcon( lapack_int n, const float* d,
+                           const lapack_complex_float* e, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_zptcon( lapack_int n, const double* d,
+                           const lapack_complex_double* e, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_spteqr( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dpteqr( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, double* z, lapack_int ldz );
+lapack_int LAPACKE_cpteqr( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zpteqr( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_sptrfs( int matrix_order, lapack_int n, lapack_int nrhs,
+                           const float* d, const float* e, const float* df,
+                           const float* ef, const float* b, lapack_int ldb,
+                           float* x, lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dptrfs( int matrix_order, lapack_int n, lapack_int nrhs,
+                           const double* d, const double* e, const double* df,
+                           const double* ef, const double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* ferr,
+                           double* berr );
+lapack_int LAPACKE_cptrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* d,
+                           const lapack_complex_float* e, const float* df,
+                           const lapack_complex_float* ef,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zptrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* d,
+                           const lapack_complex_double* e, const double* df,
+                           const lapack_complex_double* ef,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sptsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          float* d, float* e, float* b, lapack_int ldb );
+lapack_int LAPACKE_dptsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          double* d, double* e, double* b, lapack_int ldb );
+lapack_int LAPACKE_cptsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          float* d, lapack_complex_float* e,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zptsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          double* d, lapack_complex_double* e,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sptsvx( int matrix_order, char fact, lapack_int n,
+                           lapack_int nrhs, const float* d, const float* e,
+                           float* df, float* ef, const float* b, lapack_int ldb,
+                           float* x, lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dptsvx( int matrix_order, char fact, lapack_int n,
+                           lapack_int nrhs, const double* d, const double* e,
+                           double* df, double* ef, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+lapack_int LAPACKE_cptsvx( int matrix_order, char fact, lapack_int n,
+                           lapack_int nrhs, const float* d,
+                           const lapack_complex_float* e, float* df,
+                           lapack_complex_float* ef,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zptsvx( int matrix_order, char fact, lapack_int n,
+                           lapack_int nrhs, const double* d,
+                           const lapack_complex_double* e, double* df,
+                           lapack_complex_double* ef,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_spttrf( lapack_int n, float* d, float* e );
+lapack_int LAPACKE_dpttrf( lapack_int n, double* d, double* e );
+lapack_int LAPACKE_cpttrf( lapack_int n, float* d, lapack_complex_float* e );
+lapack_int LAPACKE_zpttrf( lapack_int n, double* d, lapack_complex_double* e );
+
+lapack_int LAPACKE_spttrs( int matrix_order, lapack_int n, lapack_int nrhs,
+                           const float* d, const float* e, float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_dpttrs( int matrix_order, lapack_int n, lapack_int nrhs,
+                           const double* d, const double* e, double* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_cpttrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* d,
+                           const lapack_complex_float* e,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpttrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* d,
+                           const lapack_complex_double* e,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssbev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int kd, float* ab, lapack_int ldab, float* w,
+                          float* z, lapack_int ldz );
+lapack_int LAPACKE_dsbev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int kd, double* ab, lapack_int ldab, double* w,
+                          double* z, lapack_int ldz );
+
+lapack_int LAPACKE_ssbevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int kd, float* ab, lapack_int ldab, float* w,
+                           float* z, lapack_int ldz );
+lapack_int LAPACKE_dsbevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int kd, double* ab, lapack_int ldab,
+                           double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_ssbevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int kd, float* ab,
+                           lapack_int ldab, float* q, lapack_int ldq, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dsbevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int kd, double* ab,
+                           lapack_int ldab, double* q, lapack_int ldq,
+                           double vl, double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssbgst( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb, float* ab,
+                           lapack_int ldab, const float* bb, lapack_int ldbb,
+                           float* x, lapack_int ldx );
+lapack_int LAPACKE_dsbgst( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb, double* ab,
+                           lapack_int ldab, const double* bb, lapack_int ldbb,
+                           double* x, lapack_int ldx );
+
+lapack_int LAPACKE_ssbgv( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int ka, lapack_int kb, float* ab,
+                          lapack_int ldab, float* bb, lapack_int ldbb, float* w,
+                          float* z, lapack_int ldz );
+lapack_int LAPACKE_dsbgv( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int ka, lapack_int kb, double* ab,
+                          lapack_int ldab, double* bb, lapack_int ldbb,
+                          double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_ssbgvd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb, float* ab,
+                           lapack_int ldab, float* bb, lapack_int ldbb,
+                           float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dsbgvd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb, double* ab,
+                           lapack_int ldab, double* bb, lapack_int ldbb,
+                           double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_ssbgvx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int ka, lapack_int kb,
+                           float* ab, lapack_int ldab, float* bb,
+                           lapack_int ldbb, float* q, lapack_int ldq, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dsbgvx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int ka, lapack_int kb,
+                           double* ab, lapack_int ldab, double* bb,
+                           lapack_int ldbb, double* q, lapack_int ldq,
+                           double vl, double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssbtrd( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int kd, float* ab, lapack_int ldab, float* d,
+                           float* e, float* q, lapack_int ldq );
+lapack_int LAPACKE_dsbtrd( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int kd, double* ab, lapack_int ldab,
+                           double* d, double* e, double* q, lapack_int ldq );
+
+lapack_int LAPACKE_ssfrk( int matrix_order, char transr, char uplo, char trans,
+                          lapack_int n, lapack_int k, float alpha,
+                          const float* a, lapack_int lda, float beta,
+                          float* c );
+lapack_int LAPACKE_dsfrk( int matrix_order, char transr, char uplo, char trans,
+                          lapack_int n, lapack_int k, double alpha,
+                          const double* a, lapack_int lda, double beta,
+                          double* c );
+
+lapack_int LAPACKE_sspcon( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, const lapack_int* ipiv, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_dspcon( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, const lapack_int* ipiv,
+                           double anorm, double* rcond );
+lapack_int LAPACKE_cspcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zspcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_sspev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          float* ap, float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dspev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          double* ap, double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sspevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           float* ap, float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dspevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           double* ap, double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sspevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, float* ap, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dspevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, double* ap, double vl, double vu,
+                           lapack_int il, lapack_int iu, double abstol,
+                           lapack_int* m, double* w, double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_sspgst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, float* ap, const float* bp );
+lapack_int LAPACKE_dspgst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, double* ap, const double* bp );
+
+lapack_int LAPACKE_sspgv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, float* ap, float* bp,
+                          float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dspgv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, double* ap, double* bp,
+                          double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sspgvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, float* ap, float* bp,
+                           float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dspgvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, double* ap, double* bp,
+                           double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sspgvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n, float* ap,
+                           float* bp, float vl, float vu, lapack_int il,
+                           lapack_int iu, float abstol, lapack_int* m, float* w,
+                           float* z, lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_dspgvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n, double* ap,
+                           double* bp, double vl, double vu, lapack_int il,
+                           lapack_int iu, double abstol, lapack_int* m,
+                           double* w, double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_ssprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap, const float* afp,
+                           const lapack_int* ipiv, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dsprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap, const double* afp,
+                           const lapack_int* ipiv, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_csprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_complex_float* afp,
+                           const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zsprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_complex_double* afp,
+                           const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sspsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, float* ap, lapack_int* ipiv,
+                          float* b, lapack_int ldb );
+lapack_int LAPACKE_dspsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, double* ap, lapack_int* ipiv,
+                          double* b, lapack_int ldb );
+lapack_int LAPACKE_cspsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* ap,
+                          lapack_int* ipiv, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zspsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* ap,
+                          lapack_int* ipiv, lapack_complex_double* b,
+                          lapack_int ldb );
+
+lapack_int LAPACKE_sspsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap, float* afp,
+                           lapack_int* ipiv, const float* b, lapack_int ldb,
+                           float* x, lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dspsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap, double* afp,
+                           lapack_int* ipiv, const double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* rcond,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cspsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           lapack_complex_float* afp, lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zspsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           lapack_complex_double* afp, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_ssptrd( int matrix_order, char uplo, lapack_int n, float* ap,
+                           float* d, float* e, float* tau );
+lapack_int LAPACKE_dsptrd( int matrix_order, char uplo, lapack_int n,
+                           double* ap, double* d, double* e, double* tau );
+
+lapack_int LAPACKE_ssptrf( int matrix_order, char uplo, lapack_int n, float* ap,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_dsptrf( int matrix_order, char uplo, lapack_int n,
+                           double* ap, lapack_int* ipiv );
+lapack_int LAPACKE_csptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_zsptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, lapack_int* ipiv );
+
+lapack_int LAPACKE_ssptri( int matrix_order, char uplo, lapack_int n, float* ap,
+                           const lapack_int* ipiv );
+lapack_int LAPACKE_dsptri( int matrix_order, char uplo, lapack_int n,
+                           double* ap, const lapack_int* ipiv );
+lapack_int LAPACKE_csptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, const lapack_int* ipiv );
+lapack_int LAPACKE_zsptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, const lapack_int* ipiv );
+
+lapack_int LAPACKE_ssptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dsptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_csptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_int* ipiv, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zsptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_sstebz( char range, char order, lapack_int n, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           const float* d, const float* e, lapack_int* m,
+                           lapack_int* nsplit, float* w, lapack_int* iblock,
+                           lapack_int* isplit );
+lapack_int LAPACKE_dstebz( char range, char order, lapack_int n, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, const double* d, const double* e,
+                           lapack_int* m, lapack_int* nsplit, double* w,
+                           lapack_int* iblock, lapack_int* isplit );
+
+lapack_int LAPACKE_sstedc( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dstedc( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, double* z, lapack_int ldz );
+lapack_int LAPACKE_cstedc( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zstedc( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_sstegr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* isuppz );
+lapack_int LAPACKE_dstegr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* isuppz );
+lapack_int LAPACKE_cstegr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* isuppz );
+lapack_int LAPACKE_zstegr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* isuppz );
+
+lapack_int LAPACKE_sstein( int matrix_order, lapack_int n, const float* d,
+                           const float* e, lapack_int m, const float* w,
+                           const lapack_int* iblock, const lapack_int* isplit,
+                           float* z, lapack_int ldz, lapack_int* ifailv );
+lapack_int LAPACKE_dstein( int matrix_order, lapack_int n, const double* d,
+                           const double* e, lapack_int m, const double* w,
+                           const lapack_int* iblock, const lapack_int* isplit,
+                           double* z, lapack_int ldz, lapack_int* ifailv );
+lapack_int LAPACKE_cstein( int matrix_order, lapack_int n, const float* d,
+                           const float* e, lapack_int m, const float* w,
+                           const lapack_int* iblock, const lapack_int* isplit,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int* ifailv );
+lapack_int LAPACKE_zstein( int matrix_order, lapack_int n, const double* d,
+                           const double* e, lapack_int m, const double* w,
+                           const lapack_int* iblock, const lapack_int* isplit,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifailv );
+
+lapack_int LAPACKE_sstemr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, lapack_int* m,
+                           float* w, float* z, lapack_int ldz, lapack_int nzc,
+                           lapack_int* isuppz, lapack_logical* tryrac );
+lapack_int LAPACKE_dstemr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           lapack_int* m, double* w, double* z, lapack_int ldz,
+                           lapack_int nzc, lapack_int* isuppz,
+                           lapack_logical* tryrac );
+lapack_int LAPACKE_cstemr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, lapack_int* m,
+                           float* w, lapack_complex_float* z, lapack_int ldz,
+                           lapack_int nzc, lapack_int* isuppz,
+                           lapack_logical* tryrac );
+lapack_int LAPACKE_zstemr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           lapack_int* m, double* w, lapack_complex_double* z,
+                           lapack_int ldz, lapack_int nzc, lapack_int* isuppz,
+                           lapack_logical* tryrac );
+
+lapack_int LAPACKE_ssteqr( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dsteqr( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, double* z, lapack_int ldz );
+lapack_int LAPACKE_csteqr( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zsteqr( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_ssterf( lapack_int n, float* d, float* e );
+lapack_int LAPACKE_dsterf( lapack_int n, double* d, double* e );
+
+lapack_int LAPACKE_sstev( int matrix_order, char jobz, lapack_int n, float* d,
+                          float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dstev( int matrix_order, char jobz, lapack_int n, double* d,
+                          double* e, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sstevd( int matrix_order, char jobz, lapack_int n, float* d,
+                           float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dstevd( int matrix_order, char jobz, lapack_int n, double* d,
+                           double* e, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sstevr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* isuppz );
+lapack_int LAPACKE_dstevr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* isuppz );
+
+lapack_int LAPACKE_sstevx( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dstevx( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssycon( int matrix_order, char uplo, lapack_int n,
+                           const float* a, lapack_int lda,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_dsycon( int matrix_order, char uplo, lapack_int n,
+                           const double* a, lapack_int lda,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+lapack_int LAPACKE_csycon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zsycon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_ssyequb( int matrix_order, char uplo, lapack_int n,
+                            const float* a, lapack_int lda, float* s,
+                            float* scond, float* amax );
+lapack_int LAPACKE_dsyequb( int matrix_order, char uplo, lapack_int n,
+                            const double* a, lapack_int lda, double* s,
+                            double* scond, double* amax );
+lapack_int LAPACKE_csyequb( int matrix_order, char uplo, lapack_int n,
+                            const lapack_complex_float* a, lapack_int lda,
+                            float* s, float* scond, float* amax );
+lapack_int LAPACKE_zsyequb( int matrix_order, char uplo, lapack_int n,
+                            const lapack_complex_double* a, lapack_int lda,
+                            double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_ssyev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          float* a, lapack_int lda, float* w );
+lapack_int LAPACKE_dsyev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          double* a, lapack_int lda, double* w );
+
+lapack_int LAPACKE_ssyevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           float* a, lapack_int lda, float* w );
+lapack_int LAPACKE_dsyevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           double* a, lapack_int lda, double* w );
+
+lapack_int LAPACKE_ssyevr( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, float* a, lapack_int lda, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* isuppz );
+lapack_int LAPACKE_dsyevr( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, double* a, lapack_int lda, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* isuppz );
+
+lapack_int LAPACKE_ssyevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, float* a, lapack_int lda, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dsyevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, double* a, lapack_int lda, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssygst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, float* a, lapack_int lda,
+                           const float* b, lapack_int ldb );
+lapack_int LAPACKE_dsygst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, double* a, lapack_int lda,
+                           const double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssygv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, float* a, lapack_int lda,
+                          float* b, lapack_int ldb, float* w );
+lapack_int LAPACKE_dsygv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, double* a, lapack_int lda,
+                          double* b, lapack_int ldb, double* w );
+
+lapack_int LAPACKE_ssygvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, float* a, lapack_int lda,
+                           float* b, lapack_int ldb, float* w );
+lapack_int LAPACKE_dsygvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, double* w );
+
+lapack_int LAPACKE_ssygvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dsygvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssyrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const float* af, lapack_int ldaf,
+                           const lapack_int* ipiv, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dsyrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const double* af, lapack_int ldaf,
+                           const lapack_int* ipiv, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_csyrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zsyrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_ssyrfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs, const float* a,
+                            lapack_int lda, const float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* s,
+                            const float* b, lapack_int ldb, float* x,
+                            lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dsyrfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs, const double* a,
+                            lapack_int lda, const double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* s,
+                            const double* b, lapack_int ldb, double* x,
+                            lapack_int ldx, double* rcond, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+lapack_int LAPACKE_csyrfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_float* a, lapack_int lda,
+                            const lapack_complex_float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* s,
+                            const lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zsyrfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_double* a, lapack_int lda,
+                            const lapack_complex_double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* s,
+                            const lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_ssysv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, float* a, lapack_int lda,
+                          lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dsysv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, double* a, lapack_int lda,
+                          lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_csysv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* a,
+                          lapack_int lda, lapack_int* ipiv,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zsysv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* a,
+                          lapack_int lda, lapack_int* ipiv,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssysvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           float* af, lapack_int ldaf, lapack_int* ipiv,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dsysvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           double* af, lapack_int ldaf, lapack_int* ipiv,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* rcond, double* ferr,
+                           double* berr );
+lapack_int LAPACKE_csysvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* af,
+                           lapack_int ldaf, lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zsysvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* af,
+                           lapack_int ldaf, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_ssysvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs, float* a,
+                            lapack_int lda, float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* s, float* b,
+                            lapack_int ldb, float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dsysvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs, double* a,
+                            lapack_int lda, double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* s, double* b,
+                            lapack_int ldb, double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+lapack_int LAPACKE_csysvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* s,
+                            lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zsysvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* s,
+                            lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_ssytrd( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, float* d, float* e, float* tau );
+lapack_int LAPACKE_dsytrd( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, double* d, double* e, double* tau );
+
+lapack_int LAPACKE_ssytrf( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dsytrf( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_csytrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zsytrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_ssytri( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_dsytri( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_csytri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv );
+lapack_int LAPACKE_zsytri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv );
+
+lapack_int LAPACKE_ssytrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dsytrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_csytrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zsytrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stbcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, lapack_int kd, const float* ab,
+                           lapack_int ldab, float* rcond );
+lapack_int LAPACKE_dtbcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, lapack_int kd, const double* ab,
+                           lapack_int ldab, double* rcond );
+lapack_int LAPACKE_ctbcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, lapack_int kd,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           float* rcond );
+lapack_int LAPACKE_ztbcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, lapack_int kd,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           double* rcond );
+
+lapack_int LAPACKE_stbrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const float* ab, lapack_int ldab, const float* b,
+                           lapack_int ldb, const float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dtbrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const double* ab, lapack_int ldab, const double* b,
+                           lapack_int ldb, const double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_ctbrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           const lapack_complex_float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_ztbrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           const lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_stbtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const float* ab, lapack_int ldab, float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_dtbtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const double* ab, lapack_int ldab, double* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_ctbtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztbtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stfsm( int matrix_order, char transr, char side, char uplo,
+                          char trans, char diag, lapack_int m, lapack_int n,
+                          float alpha, const float* a, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dtfsm( int matrix_order, char transr, char side, char uplo,
+                          char trans, char diag, lapack_int m, lapack_int n,
+                          double alpha, const double* a, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_ctfsm( int matrix_order, char transr, char side, char uplo,
+                          char trans, char diag, lapack_int m, lapack_int n,
+                          lapack_complex_float alpha,
+                          const lapack_complex_float* a,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztfsm( int matrix_order, char transr, char side, char uplo,
+                          char trans, char diag, lapack_int m, lapack_int n,
+                          lapack_complex_double alpha,
+                          const lapack_complex_double* a,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stftri( int matrix_order, char transr, char uplo, char diag,
+                           lapack_int n, float* a );
+lapack_int LAPACKE_dtftri( int matrix_order, char transr, char uplo, char diag,
+                           lapack_int n, double* a );
+lapack_int LAPACKE_ctftri( int matrix_order, char transr, char uplo, char diag,
+                           lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_ztftri( int matrix_order, char transr, char uplo, char diag,
+                           lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_stfttp( int matrix_order, char transr, char uplo,
+                           lapack_int n, const float* arf, float* ap );
+lapack_int LAPACKE_dtfttp( int matrix_order, char transr, char uplo,
+                           lapack_int n, const double* arf, double* ap );
+lapack_int LAPACKE_ctfttp( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_float* arf,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_ztfttp( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_double* arf,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_stfttr( int matrix_order, char transr, char uplo,
+                           lapack_int n, const float* arf, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dtfttr( int matrix_order, char transr, char uplo,
+                           lapack_int n, const double* arf, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_ctfttr( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_float* arf,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztfttr( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_double* arf,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_stgevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const float* s, lapack_int lds, const float* p,
+                           lapack_int ldp, float* vl, lapack_int ldvl,
+                           float* vr, lapack_int ldvr, lapack_int mm,
+                           lapack_int* m );
+lapack_int LAPACKE_dtgevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const double* s, lapack_int lds, const double* p,
+                           lapack_int ldp, double* vl, lapack_int ldvl,
+                           double* vr, lapack_int ldvr, lapack_int mm,
+                           lapack_int* m );
+lapack_int LAPACKE_ctgevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_float* s, lapack_int lds,
+                           const lapack_complex_float* p, lapack_int ldp,
+                           lapack_complex_float* vl, lapack_int ldvl,
+                           lapack_complex_float* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ztgevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_double* s, lapack_int lds,
+                           const lapack_complex_double* p, lapack_int ldp,
+                           lapack_complex_double* vl, lapack_int ldvl,
+                           lapack_complex_double* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+
+lapack_int LAPACKE_stgexc( int matrix_order, lapack_logical wantq,
+                           lapack_logical wantz, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb, float* q,
+                           lapack_int ldq, float* z, lapack_int ldz,
+                           lapack_int* ifst, lapack_int* ilst );
+lapack_int LAPACKE_dtgexc( int matrix_order, lapack_logical wantq,
+                           lapack_logical wantz, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb, double* q,
+                           lapack_int ldq, double* z, lapack_int ldz,
+                           lapack_int* ifst, lapack_int* ilst );
+lapack_int LAPACKE_ctgexc( int matrix_order, lapack_logical wantq,
+                           lapack_logical wantz, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int ifst, lapack_int ilst );
+lapack_int LAPACKE_ztgexc( int matrix_order, lapack_logical wantq,
+                           lapack_logical wantz, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int ifst, lapack_int ilst );
+
+lapack_int LAPACKE_stgsen( int matrix_order, lapack_int ijob,
+                           lapack_logical wantq, lapack_logical wantz,
+                           const lapack_logical* select, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb,
+                           float* alphar, float* alphai, float* beta, float* q,
+                           lapack_int ldq, float* z, lapack_int ldz,
+                           lapack_int* m, float* pl, float* pr, float* dif );
+lapack_int LAPACKE_dtgsen( int matrix_order, lapack_int ijob,
+                           lapack_logical wantq, lapack_logical wantz,
+                           const lapack_logical* select, lapack_int n,
+                           double* a, lapack_int lda, double* b, lapack_int ldb,
+                           double* alphar, double* alphai, double* beta,
+                           double* q, lapack_int ldq, double* z, lapack_int ldz,
+                           lapack_int* m, double* pl, double* pr, double* dif );
+lapack_int LAPACKE_ctgsen( int matrix_order, lapack_int ijob,
+                           lapack_logical wantq, lapack_logical wantz,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* alpha,
+                           lapack_complex_float* beta, lapack_complex_float* q,
+                           lapack_int ldq, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* m, float* pl, float* pr,
+                           float* dif );
+lapack_int LAPACKE_ztgsen( int matrix_order, lapack_int ijob,
+                           lapack_logical wantq, lapack_logical wantz,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* alpha,
+                           lapack_complex_double* beta,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* m, double* pl, double* pr, double* dif );
+
+lapack_int LAPACKE_stgsja( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int k, lapack_int l, float* a, lapack_int lda,
+                           float* b, lapack_int ldb, float tola, float tolb,
+                           float* alpha, float* beta, float* u, lapack_int ldu,
+                           float* v, lapack_int ldv, float* q, lapack_int ldq,
+                           lapack_int* ncycle );
+lapack_int LAPACKE_dtgsja( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int k, lapack_int l, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           double tola, double tolb, double* alpha,
+                           double* beta, double* u, lapack_int ldu, double* v,
+                           lapack_int ldv, double* q, lapack_int ldq,
+                           lapack_int* ncycle );
+lapack_int LAPACKE_ctgsja( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int k, lapack_int l, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, float tola, float tolb, float* alpha,
+                           float* beta, lapack_complex_float* u, lapack_int ldu,
+                           lapack_complex_float* v, lapack_int ldv,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_int* ncycle );
+lapack_int LAPACKE_ztgsja( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int k, lapack_int l, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, double tola, double tolb,
+                           double* alpha, double* beta,
+                           lapack_complex_double* u, lapack_int ldu,
+                           lapack_complex_double* v, lapack_int ldv,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_int* ncycle );
+
+lapack_int LAPACKE_stgsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const float* a, lapack_int lda, const float* b,
+                           lapack_int ldb, const float* vl, lapack_int ldvl,
+                           const float* vr, lapack_int ldvr, float* s,
+                           float* dif, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_dtgsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const double* a, lapack_int lda, const double* b,
+                           lapack_int ldb, const double* vl, lapack_int ldvl,
+                           const double* vr, lapack_int ldvr, double* s,
+                           double* dif, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ctgsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           const lapack_complex_float* vl, lapack_int ldvl,
+                           const lapack_complex_float* vr, lapack_int ldvr,
+                           float* s, float* dif, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ztgsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           const lapack_complex_double* vl, lapack_int ldvl,
+                           const lapack_complex_double* vr, lapack_int ldvr,
+                           double* s, double* dif, lapack_int mm,
+                           lapack_int* m );
+
+lapack_int LAPACKE_stgsyl( int matrix_order, char trans, lapack_int ijob,
+                           lapack_int m, lapack_int n, const float* a,
+                           lapack_int lda, const float* b, lapack_int ldb,
+                           float* c, lapack_int ldc, const float* d,
+                           lapack_int ldd, const float* e, lapack_int lde,
+                           float* f, lapack_int ldf, float* scale, float* dif );
+lapack_int LAPACKE_dtgsyl( int matrix_order, char trans, lapack_int ijob,
+                           lapack_int m, lapack_int n, const double* a,
+                           lapack_int lda, const double* b, lapack_int ldb,
+                           double* c, lapack_int ldc, const double* d,
+                           lapack_int ldd, const double* e, lapack_int lde,
+                           double* f, lapack_int ldf, double* scale,
+                           double* dif );
+lapack_int LAPACKE_ctgsyl( int matrix_order, char trans, lapack_int ijob,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* c, lapack_int ldc,
+                           const lapack_complex_float* d, lapack_int ldd,
+                           const lapack_complex_float* e, lapack_int lde,
+                           lapack_complex_float* f, lapack_int ldf,
+                           float* scale, float* dif );
+lapack_int LAPACKE_ztgsyl( int matrix_order, char trans, lapack_int ijob,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* c, lapack_int ldc,
+                           const lapack_complex_double* d, lapack_int ldd,
+                           const lapack_complex_double* e, lapack_int lde,
+                           lapack_complex_double* f, lapack_int ldf,
+                           double* scale, double* dif );
+
+lapack_int LAPACKE_stpcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const float* ap, float* rcond );
+lapack_int LAPACKE_dtpcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const double* ap, double* rcond );
+lapack_int LAPACKE_ctpcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const lapack_complex_float* ap,
+                           float* rcond );
+lapack_int LAPACKE_ztpcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const lapack_complex_double* ap,
+                           double* rcond );
+
+lapack_int LAPACKE_stprfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const float* ap,
+                           const float* b, lapack_int ldb, const float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dtprfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const double* ap,
+                           const double* b, lapack_int ldb, const double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_ctprfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* ap,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           const lapack_complex_float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_ztprfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* ap,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           const lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_stptri( int matrix_order, char uplo, char diag, lapack_int n,
+                           float* ap );
+lapack_int LAPACKE_dtptri( int matrix_order, char uplo, char diag, lapack_int n,
+                           double* ap );
+lapack_int LAPACKE_ctptri( int matrix_order, char uplo, char diag, lapack_int n,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_ztptri( int matrix_order, char uplo, char diag, lapack_int n,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_stptrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const float* ap,
+                           float* b, lapack_int ldb );
+lapack_int LAPACKE_dtptrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const double* ap,
+                           double* b, lapack_int ldb );
+lapack_int LAPACKE_ctptrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* ap,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztptrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* ap,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stpttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const float* ap, float* arf );
+lapack_int LAPACKE_dtpttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const double* ap, double* arf );
+lapack_int LAPACKE_ctpttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_float* ap,
+                           lapack_complex_float* arf );
+lapack_int LAPACKE_ztpttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_double* ap,
+                           lapack_complex_double* arf );
+
+lapack_int LAPACKE_stpttr( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, float* a, lapack_int lda );
+lapack_int LAPACKE_dtpttr( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, double* a, lapack_int lda );
+lapack_int LAPACKE_ctpttr( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztpttr( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_strcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const float* a, lapack_int lda,
+                           float* rcond );
+lapack_int LAPACKE_dtrcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const double* a, lapack_int lda,
+                           double* rcond );
+lapack_int LAPACKE_ctrcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda, float* rcond );
+lapack_int LAPACKE_ztrcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda, double* rcond );
+
+lapack_int LAPACKE_strevc( int matrix_order, char side, char howmny,
+                           lapack_logical* select, lapack_int n, const float* t,
+                           lapack_int ldt, float* vl, lapack_int ldvl,
+                           float* vr, lapack_int ldvr, lapack_int mm,
+                           lapack_int* m );
+lapack_int LAPACKE_dtrevc( int matrix_order, char side, char howmny,
+                           lapack_logical* select, lapack_int n,
+                           const double* t, lapack_int ldt, double* vl,
+                           lapack_int ldvl, double* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ctrevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* vl, lapack_int ldvl,
+                           lapack_complex_float* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ztrevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* vl, lapack_int ldvl,
+                           lapack_complex_double* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+
+lapack_int LAPACKE_strexc( int matrix_order, char compq, lapack_int n, float* t,
+                           lapack_int ldt, float* q, lapack_int ldq,
+                           lapack_int* ifst, lapack_int* ilst );
+lapack_int LAPACKE_dtrexc( int matrix_order, char compq, lapack_int n,
+                           double* t, lapack_int ldt, double* q, lapack_int ldq,
+                           lapack_int* ifst, lapack_int* ilst );
+lapack_int LAPACKE_ctrexc( int matrix_order, char compq, lapack_int n,
+                           lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_int ifst, lapack_int ilst );
+lapack_int LAPACKE_ztrexc( int matrix_order, char compq, lapack_int n,
+                           lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_int ifst, lapack_int ilst );
+
+lapack_int LAPACKE_strrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const float* a,
+                           lapack_int lda, const float* b, lapack_int ldb,
+                           const float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dtrrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const double* a,
+                           lapack_int lda, const double* b, lapack_int ldb,
+                           const double* x, lapack_int ldx, double* ferr,
+                           double* berr );
+lapack_int LAPACKE_ctrrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           const lapack_complex_float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_ztrrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           const lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_strsen( int matrix_order, char job, char compq,
+                           const lapack_logical* select, lapack_int n, float* t,
+                           lapack_int ldt, float* q, lapack_int ldq, float* wr,
+                           float* wi, lapack_int* m, float* s, float* sep );
+lapack_int LAPACKE_dtrsen( int matrix_order, char job, char compq,
+                           const lapack_logical* select, lapack_int n,
+                           double* t, lapack_int ldt, double* q, lapack_int ldq,
+                           double* wr, double* wi, lapack_int* m, double* s,
+                           double* sep );
+lapack_int LAPACKE_ctrsen( int matrix_order, char job, char compq,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_complex_float* w, lapack_int* m, float* s,
+                           float* sep );
+lapack_int LAPACKE_ztrsen( int matrix_order, char job, char compq,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* w, lapack_int* m, double* s,
+                           double* sep );
+
+lapack_int LAPACKE_strsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const float* t, lapack_int ldt, const float* vl,
+                           lapack_int ldvl, const float* vr, lapack_int ldvr,
+                           float* s, float* sep, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_dtrsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const double* t, lapack_int ldt, const double* vl,
+                           lapack_int ldvl, const double* vr, lapack_int ldvr,
+                           double* s, double* sep, lapack_int mm,
+                           lapack_int* m );
+lapack_int LAPACKE_ctrsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_float* t, lapack_int ldt,
+                           const lapack_complex_float* vl, lapack_int ldvl,
+                           const lapack_complex_float* vr, lapack_int ldvr,
+                           float* s, float* sep, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ztrsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_double* t, lapack_int ldt,
+                           const lapack_complex_double* vl, lapack_int ldvl,
+                           const lapack_complex_double* vr, lapack_int ldvr,
+                           double* s, double* sep, lapack_int mm,
+                           lapack_int* m );
+
+lapack_int LAPACKE_strsyl( int matrix_order, char trana, char tranb,
+                           lapack_int isgn, lapack_int m, lapack_int n,
+                           const float* a, lapack_int lda, const float* b,
+                           lapack_int ldb, float* c, lapack_int ldc,
+                           float* scale );
+lapack_int LAPACKE_dtrsyl( int matrix_order, char trana, char tranb,
+                           lapack_int isgn, lapack_int m, lapack_int n,
+                           const double* a, lapack_int lda, const double* b,
+                           lapack_int ldb, double* c, lapack_int ldc,
+                           double* scale );
+lapack_int LAPACKE_ctrsyl( int matrix_order, char trana, char tranb,
+                           lapack_int isgn, lapack_int m, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* c, lapack_int ldc,
+                           float* scale );
+lapack_int LAPACKE_ztrsyl( int matrix_order, char trana, char tranb,
+                           lapack_int isgn, lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* c, lapack_int ldc,
+                           double* scale );
+
+lapack_int LAPACKE_strtri( int matrix_order, char uplo, char diag, lapack_int n,
+                           float* a, lapack_int lda );
+lapack_int LAPACKE_dtrtri( int matrix_order, char uplo, char diag, lapack_int n,
+                           double* a, lapack_int lda );
+lapack_int LAPACKE_ctrtri( int matrix_order, char uplo, char diag, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztrtri( int matrix_order, char uplo, char diag, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_strtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const float* a,
+                           lapack_int lda, float* b, lapack_int ldb );
+lapack_int LAPACKE_dtrtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const double* a,
+                           lapack_int lda, double* b, lapack_int ldb );
+lapack_int LAPACKE_ctrtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztrtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_strttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const float* a, lapack_int lda,
+                           float* arf );
+lapack_int LAPACKE_dtrttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const double* a, lapack_int lda,
+                           double* arf );
+lapack_int LAPACKE_ctrttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* arf );
+lapack_int LAPACKE_ztrttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* arf );
+
+lapack_int LAPACKE_strttp( int matrix_order, char uplo, lapack_int n,
+                           const float* a, lapack_int lda, float* ap );
+lapack_int LAPACKE_dtrttp( int matrix_order, char uplo, lapack_int n,
+                           const double* a, lapack_int lda, double* ap );
+lapack_int LAPACKE_ctrttp( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_ztrttp( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_stzrzf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dtzrzf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_ctzrzf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_ztzrzf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungbr( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zungbr( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cunghr( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zunghr( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cunglq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zunglq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungql( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zungql( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungqr( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zungqr( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungrq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zungrq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungtr( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau );
+lapack_int LAPACKE_zungtr( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cunmbr( int matrix_order, char vect, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmbr( int matrix_order, char vect, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmhr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmhr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmlq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmlq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmql( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmql( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmqr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmqr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmrq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmrq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmrz( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           lapack_int l, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmrz( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           lapack_int l, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cupgtr( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* q, lapack_int ldq );
+lapack_int LAPACKE_zupgtr( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* q, lapack_int ldq );
+
+lapack_int LAPACKE_cupmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_float* ap,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zupmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_double* ap,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sbdsdc_work( int matrix_order, char uplo, char compq,
+                                lapack_int n, float* d, float* e, float* u,
+                                lapack_int ldu, float* vt, lapack_int ldvt,
+                                float* q, lapack_int* iq, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dbdsdc_work( int matrix_order, char uplo, char compq,
+                                lapack_int n, double* d, double* e, double* u,
+                                lapack_int ldu, double* vt, lapack_int ldvt,
+                                double* q, lapack_int* iq, double* work,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_sbdsqr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                                float* d, float* e, float* vt, lapack_int ldvt,
+                                float* u, lapack_int ldu, float* c,
+                                lapack_int ldc, float* work );
+lapack_int LAPACKE_dbdsqr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                                double* d, double* e, double* vt,
+                                lapack_int ldvt, double* u, lapack_int ldu,
+                                double* c, lapack_int ldc, double* work );
+lapack_int LAPACKE_cbdsqr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                                float* d, float* e, lapack_complex_float* vt,
+                                lapack_int ldvt, lapack_complex_float* u,
+                                lapack_int ldu, lapack_complex_float* c,
+                                lapack_int ldc, float* work );
+lapack_int LAPACKE_zbdsqr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                                double* d, double* e, lapack_complex_double* vt,
+                                lapack_int ldvt, lapack_complex_double* u,
+                                lapack_int ldu, lapack_complex_double* c,
+                                lapack_int ldc, double* work );
+
+lapack_int LAPACKE_sdisna_work( char job, lapack_int m, lapack_int n,
+                                const float* d, float* sep );
+lapack_int LAPACKE_ddisna_work( char job, lapack_int m, lapack_int n,
+                                const double* d, double* sep );
+
+lapack_int LAPACKE_sgbbrd_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int ncc, lapack_int kl,
+                                lapack_int ku, float* ab, lapack_int ldab,
+                                float* d, float* e, float* q, lapack_int ldq,
+                                float* pt, lapack_int ldpt, float* c,
+                                lapack_int ldc, float* work );
+lapack_int LAPACKE_dgbbrd_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int ncc, lapack_int kl,
+                                lapack_int ku, double* ab, lapack_int ldab,
+                                double* d, double* e, double* q, lapack_int ldq,
+                                double* pt, lapack_int ldpt, double* c,
+                                lapack_int ldc, double* work );
+lapack_int LAPACKE_cgbbrd_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int ncc, lapack_int kl,
+                                lapack_int ku, lapack_complex_float* ab,
+                                lapack_int ldab, float* d, float* e,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* pt, lapack_int ldpt,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgbbrd_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int ncc, lapack_int kl,
+                                lapack_int ku, lapack_complex_double* ab,
+                                lapack_int ldab, double* d, double* e,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* pt, lapack_int ldpt,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgbcon_work( int matrix_order, char norm, lapack_int n,
+                                lapack_int kl, lapack_int ku, const float* ab,
+                                lapack_int ldab, const lapack_int* ipiv,
+                                float anorm, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgbcon_work( int matrix_order, char norm, lapack_int n,
+                                lapack_int kl, lapack_int ku, const double* ab,
+                                lapack_int ldab, const lapack_int* ipiv,
+                                double anorm, double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgbcon_work( int matrix_order, char norm, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zgbcon_work( int matrix_order, char norm, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, const lapack_int* ipiv,
+                                double anorm, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgbequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const float* ab,
+                                lapack_int ldab, float* r, float* c,
+                                float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_dgbequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const double* ab,
+                                lapack_int ldab, double* r, double* c,
+                                double* rowcnd, double* colcnd, double* amax );
+lapack_int LAPACKE_cgbequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                float* r, float* c, float* rowcnd,
+                                float* colcnd, float* amax );
+lapack_int LAPACKE_zgbequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, double* r, double* c,
+                                double* rowcnd, double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgbequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_int kl, lapack_int ku, const float* ab,
+                                 lapack_int ldab, float* r, float* c,
+                                 float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_dgbequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_int kl, lapack_int ku, const double* ab,
+                                 lapack_int ldab, double* r, double* c,
+                                 double* rowcnd, double* colcnd, double* amax );
+lapack_int LAPACKE_cgbequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_int kl, lapack_int ku,
+                                 const lapack_complex_float* ab,
+                                 lapack_int ldab, float* r, float* c,
+                                 float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_zgbequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_int kl, lapack_int ku,
+                                 const lapack_complex_double* ab,
+                                 lapack_int ldab, double* r, double* c,
+                                 double* rowcnd, double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgbrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const float* ab, lapack_int ldab,
+                                const float* afb, lapack_int ldafb,
+                                const lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgbrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const double* ab, lapack_int ldab,
+                                const double* afb, lapack_int ldafb,
+                                const lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgbrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_complex_float* afb,
+                                lapack_int ldafb, const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgbrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab,
+                                const lapack_complex_double* afb,
+                                lapack_int ldafb, const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgbrfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, const float* ab,
+                                 lapack_int ldab, const float* afb,
+                                 lapack_int ldafb, const lapack_int* ipiv,
+                                 const float* r, const float* c, const float* b,
+                                 lapack_int ldb, float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dgbrfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, const double* ab,
+                                 lapack_int ldab, const double* afb,
+                                 lapack_int ldafb, const lapack_int* ipiv,
+                                 const double* r, const double* c,
+                                 const double* b, lapack_int ldb, double* x,
+                                 lapack_int ldx, double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cgbrfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs,
+                                 const lapack_complex_float* ab,
+                                 lapack_int ldab,
+                                 const lapack_complex_float* afb,
+                                 lapack_int ldafb, const lapack_int* ipiv,
+                                 const float* r, const float* c,
+                                 const lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zgbrfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs,
+                                 const lapack_complex_double* ab,
+                                 lapack_int ldab,
+                                 const lapack_complex_double* afb,
+                                 lapack_int ldafb, const lapack_int* ipiv,
+                                 const double* r, const double* c,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_sgbsv_work( int matrix_order, lapack_int n, lapack_int kl,
+                               lapack_int ku, lapack_int nrhs, float* ab,
+                               lapack_int ldab, lapack_int* ipiv, float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dgbsv_work( int matrix_order, lapack_int n, lapack_int kl,
+                               lapack_int ku, lapack_int nrhs, double* ab,
+                               lapack_int ldab, lapack_int* ipiv, double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_cgbsv_work( int matrix_order, lapack_int n, lapack_int kl,
+                               lapack_int ku, lapack_int nrhs,
+                               lapack_complex_float* ab, lapack_int ldab,
+                               lapack_int* ipiv, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zgbsv_work( int matrix_order, lapack_int n, lapack_int kl,
+                               lapack_int ku, lapack_int nrhs,
+                               lapack_complex_double* ab, lapack_int ldab,
+                               lapack_int* ipiv, lapack_complex_double* b,
+                               lapack_int ldb );
+
+lapack_int LAPACKE_sgbsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int kl, lapack_int ku,
+                                lapack_int nrhs, float* ab, lapack_int ldab,
+                                float* afb, lapack_int ldafb, lapack_int* ipiv,
+                                char* equed, float* r, float* c, float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dgbsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int kl, lapack_int ku,
+                                lapack_int nrhs, double* ab, lapack_int ldab,
+                                double* afb, lapack_int ldafb, lapack_int* ipiv,
+                                char* equed, double* r, double* c, double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cgbsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int kl, lapack_int ku,
+                                lapack_int nrhs, lapack_complex_float* ab,
+                                lapack_int ldab, lapack_complex_float* afb,
+                                lapack_int ldafb, lapack_int* ipiv, char* equed,
+                                float* r, float* c, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zgbsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int kl, lapack_int ku,
+                                lapack_int nrhs, lapack_complex_double* ab,
+                                lapack_int ldab, lapack_complex_double* afb,
+                                lapack_int ldafb, lapack_int* ipiv, char* equed,
+                                double* r, double* c, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_sgbsvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, float* ab, lapack_int ldab,
+                                 float* afb, lapack_int ldafb, lapack_int* ipiv,
+                                 char* equed, float* r, float* c, float* b,
+                                 lapack_int ldb, float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dgbsvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, double* ab, lapack_int ldab,
+                                 double* afb, lapack_int ldafb,
+                                 lapack_int* ipiv, char* equed, double* r,
+                                 double* c, double* b, lapack_int ldb,
+                                 double* x, lapack_int ldx, double* rcond,
+                                 double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cgbsvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, lapack_complex_float* ab,
+                                 lapack_int ldab, lapack_complex_float* afb,
+                                 lapack_int ldafb, lapack_int* ipiv,
+                                 char* equed, float* r, float* c,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zgbsvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, lapack_complex_double* ab,
+                                 lapack_int ldab, lapack_complex_double* afb,
+                                 lapack_int ldafb, lapack_int* ipiv,
+                                 char* equed, double* r, double* c,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_sgbtrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, float* ab,
+                                lapack_int ldab, lapack_int* ipiv );
+lapack_int LAPACKE_dgbtrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, double* ab,
+                                lapack_int ldab, lapack_int* ipiv );
+lapack_int LAPACKE_cgbtrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                lapack_int* ipiv );
+lapack_int LAPACKE_zgbtrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                lapack_int* ipiv );
+
+lapack_int LAPACKE_sgbtrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const float* ab, lapack_int ldab,
+                                const lapack_int* ipiv, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dgbtrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const double* ab, lapack_int ldab,
+                                const lapack_int* ipiv, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_cgbtrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_int* ipiv, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zgbtrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sgebak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const float* scale, lapack_int m, float* v,
+                                lapack_int ldv );
+lapack_int LAPACKE_dgebak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const double* scale, lapack_int m, double* v,
+                                lapack_int ldv );
+lapack_int LAPACKE_cgebak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const float* scale, lapack_int m,
+                                lapack_complex_float* v, lapack_int ldv );
+lapack_int LAPACKE_zgebak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const double* scale, lapack_int m,
+                                lapack_complex_double* v, lapack_int ldv );
+
+lapack_int LAPACKE_sgebal_work( int matrix_order, char job, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* ilo,
+                                lapack_int* ihi, float* scale );
+lapack_int LAPACKE_dgebal_work( int matrix_order, char job, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* ilo,
+                                lapack_int* ihi, double* scale );
+lapack_int LAPACKE_cgebal_work( int matrix_order, char job, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ilo, lapack_int* ihi,
+                                float* scale );
+lapack_int LAPACKE_zgebal_work( int matrix_order, char job, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ilo, lapack_int* ihi,
+                                double* scale );
+
+lapack_int LAPACKE_sgebrd_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* d, float* e,
+                                float* tauq, float* taup, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dgebrd_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* d, double* e,
+                                double* tauq, double* taup, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_cgebrd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float* d, float* e, lapack_complex_float* tauq,
+                                lapack_complex_float* taup,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgebrd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double* d, double* e,
+                                lapack_complex_double* tauq,
+                                lapack_complex_double* taup,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgecon_work( int matrix_order, char norm, lapack_int n,
+                                const float* a, lapack_int lda, float anorm,
+                                float* rcond, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dgecon_work( int matrix_order, char norm, lapack_int n,
+                                const double* a, lapack_int lda, double anorm,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgecon_work( int matrix_order, char norm, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float anorm, float* rcond,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgecon_work( int matrix_order, char norm, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double anorm, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgeequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                const float* a, lapack_int lda, float* r,
+                                float* c, float* rowcnd, float* colcnd,
+                                float* amax );
+lapack_int LAPACKE_dgeequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda, double* r,
+                                double* c, double* rowcnd, double* colcnd,
+                                double* amax );
+lapack_int LAPACKE_cgeequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float* r, float* c, float* rowcnd,
+                                float* colcnd, float* amax );
+lapack_int LAPACKE_zgeequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double* r, double* c, double* rowcnd,
+                                double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgeequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 const float* a, lapack_int lda, float* r,
+                                 float* c, float* rowcnd, float* colcnd,
+                                 float* amax );
+lapack_int LAPACKE_dgeequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 const double* a, lapack_int lda, double* r,
+                                 double* c, double* rowcnd, double* colcnd,
+                                 double* amax );
+lapack_int LAPACKE_cgeequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 float* r, float* c, float* rowcnd,
+                                 float* colcnd, float* amax );
+lapack_int LAPACKE_zgeequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 double* r, double* c, double* rowcnd,
+                                 double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgees_work( int matrix_order, char jobvs, char sort,
+                               LAPACK_S_SELECT2 select, lapack_int n, float* a,
+                               lapack_int lda, lapack_int* sdim, float* wr,
+                               float* wi, float* vs, lapack_int ldvs,
+                               float* work, lapack_int lwork,
+                               lapack_logical* bwork );
+lapack_int LAPACKE_dgees_work( int matrix_order, char jobvs, char sort,
+                               LAPACK_D_SELECT2 select, lapack_int n, double* a,
+                               lapack_int lda, lapack_int* sdim, double* wr,
+                               double* wi, double* vs, lapack_int ldvs,
+                               double* work, lapack_int lwork,
+                               lapack_logical* bwork );
+lapack_int LAPACKE_cgees_work( int matrix_order, char jobvs, char sort,
+                               LAPACK_C_SELECT1 select, lapack_int n,
+                               lapack_complex_float* a, lapack_int lda,
+                               lapack_int* sdim, lapack_complex_float* w,
+                               lapack_complex_float* vs, lapack_int ldvs,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork, lapack_logical* bwork );
+lapack_int LAPACKE_zgees_work( int matrix_order, char jobvs, char sort,
+                               LAPACK_Z_SELECT1 select, lapack_int n,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_int* sdim, lapack_complex_double* w,
+                               lapack_complex_double* vs, lapack_int ldvs,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork, lapack_logical* bwork );
+
+lapack_int LAPACKE_sgeesx_work( int matrix_order, char jobvs, char sort,
+                                LAPACK_S_SELECT2 select, char sense,
+                                lapack_int n, float* a, lapack_int lda,
+                                lapack_int* sdim, float* wr, float* wi,
+                                float* vs, lapack_int ldvs, float* rconde,
+                                float* rcondv, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_dgeesx_work( int matrix_order, char jobvs, char sort,
+                                LAPACK_D_SELECT2 select, char sense,
+                                lapack_int n, double* a, lapack_int lda,
+                                lapack_int* sdim, double* wr, double* wi,
+                                double* vs, lapack_int ldvs, double* rconde,
+                                double* rcondv, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_cgeesx_work( int matrix_order, char jobvs, char sort,
+                                LAPACK_C_SELECT1 select, char sense,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, lapack_int* sdim,
+                                lapack_complex_float* w,
+                                lapack_complex_float* vs, lapack_int ldvs,
+                                float* rconde, float* rcondv,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_logical* bwork );
+lapack_int LAPACKE_zgeesx_work( int matrix_order, char jobvs, char sort,
+                                LAPACK_Z_SELECT1 select, char sense,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, lapack_int* sdim,
+                                lapack_complex_double* w,
+                                lapack_complex_double* vs, lapack_int ldvs,
+                                double* rconde, double* rcondv,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_logical* bwork );
+
+lapack_int LAPACKE_sgeev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, float* a, lapack_int lda,
+                               float* wr, float* wi, float* vl, lapack_int ldvl,
+                               float* vr, lapack_int ldvr, float* work,
+                               lapack_int lwork );
+lapack_int LAPACKE_dgeev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, double* a, lapack_int lda,
+                               double* wr, double* wi, double* vl,
+                               lapack_int ldvl, double* vr, lapack_int ldvr,
+                               double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, lapack_complex_float* a,
+                               lapack_int lda, lapack_complex_float* w,
+                               lapack_complex_float* vl, lapack_int ldvl,
+                               lapack_complex_float* vr, lapack_int ldvr,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork );
+lapack_int LAPACKE_zgeev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, lapack_complex_double* a,
+                               lapack_int lda, lapack_complex_double* w,
+                               lapack_complex_double* vl, lapack_int ldvl,
+                               lapack_complex_double* vr, lapack_int ldvr,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork );
+
+lapack_int LAPACKE_sgeevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n, float* a,
+                                lapack_int lda, float* wr, float* wi, float* vl,
+                                lapack_int ldvl, float* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi, float* scale,
+                                float* abnrm, float* rconde, float* rcondv,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgeevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n, double* a,
+                                lapack_int lda, double* wr, double* wi,
+                                double* vl, lapack_int ldvl, double* vr,
+                                lapack_int ldvr, lapack_int* ilo,
+                                lapack_int* ihi, double* scale, double* abnrm,
+                                double* rconde, double* rcondv, double* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_cgeevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* w,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi, float* scale,
+                                float* abnrm, float* rconde, float* rcondv,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork );
+lapack_int LAPACKE_zgeevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* w,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi, double* scale,
+                                double* abnrm, double* rconde, double* rcondv,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_sgehrd_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, float* a, lapack_int lda,
+                                float* tau, float* work, lapack_int lwork );
+lapack_int LAPACKE_dgehrd_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, double* a, lapack_int lda,
+                                double* tau, double* work, lapack_int lwork );
+lapack_int LAPACKE_cgehrd_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgehrd_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgejsv_work( int matrix_order, char joba, char jobu,
+                                char jobv, char jobr, char jobt, char jobp,
+                                lapack_int m, lapack_int n, float* a,
+                                lapack_int lda, float* sva, float* u,
+                                lapack_int ldu, float* v, lapack_int ldv,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgejsv_work( int matrix_order, char joba, char jobu,
+                                char jobv, char jobr, char jobt, char jobp,
+                                lapack_int m, lapack_int n, double* a,
+                                lapack_int lda, double* sva, double* u,
+                                lapack_int ldu, double* v, lapack_int ldv,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_sgelq2_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work );
+lapack_int LAPACKE_dgelq2_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work );
+lapack_int LAPACKE_cgelq2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zgelq2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_sgelqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgelqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgelqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgelqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgels_work( int matrix_order, char trans, lapack_int m,
+                               lapack_int n, lapack_int nrhs, float* a,
+                               lapack_int lda, float* b, lapack_int ldb,
+                               float* work, lapack_int lwork );
+lapack_int LAPACKE_dgels_work( int matrix_order, char trans, lapack_int m,
+                               lapack_int n, lapack_int nrhs, double* a,
+                               lapack_int lda, double* b, lapack_int ldb,
+                               double* work, lapack_int lwork );
+lapack_int LAPACKE_cgels_work( int matrix_order, char trans, lapack_int m,
+                               lapack_int n, lapack_int nrhs,
+                               lapack_complex_float* a, lapack_int lda,
+                               lapack_complex_float* b, lapack_int ldb,
+                               lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgels_work( int matrix_order, char trans, lapack_int m,
+                               lapack_int n, lapack_int nrhs,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_complex_double* b, lapack_int ldb,
+                               lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgelsd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float* s, float rcond,
+                                lapack_int* rank, float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgelsd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* s,
+                                double rcond, lapack_int* rank, double* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_cgelsd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, float* s, float rcond,
+                                lapack_int* rank, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_zgelsd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, double* s, double rcond,
+                                lapack_int* rank, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_sgelss_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float* s, float rcond,
+                                lapack_int* rank, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dgelss_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* s,
+                                double rcond, lapack_int* rank, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_cgelss_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, float* s, float rcond,
+                                lapack_int* rank, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zgelss_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, double* s, double rcond,
+                                lapack_int* rank, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork );
+
+lapack_int LAPACKE_sgelsy_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, lapack_int* jpvt,
+                                float rcond, lapack_int* rank, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dgelsy_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, lapack_int* jpvt,
+                                double rcond, lapack_int* rank, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_cgelsy_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, lapack_int* jpvt, float rcond,
+                                lapack_int* rank, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zgelsy_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_int* jpvt, double rcond,
+                                lapack_int* rank, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork );
+
+lapack_int LAPACKE_sgeqlf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgeqlf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeqlf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgeqlf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgeqp3_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* jpvt,
+                                float* tau, float* work, lapack_int lwork );
+lapack_int LAPACKE_dgeqp3_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* jpvt,
+                                double* tau, double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeqp3_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* jpvt, lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork );
+lapack_int LAPACKE_zgeqp3_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* jpvt, lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_sgeqpf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* jpvt,
+                                float* tau, float* work );
+lapack_int LAPACKE_dgeqpf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* jpvt,
+                                double* tau, double* work );
+lapack_int LAPACKE_cgeqpf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* jpvt, lapack_complex_float* tau,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgeqpf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* jpvt, lapack_complex_double* tau,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgeqr2_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work );
+lapack_int LAPACKE_dgeqr2_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work );
+lapack_int LAPACKE_cgeqr2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zgeqr2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_sgeqrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgeqrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeqrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgeqrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgeqrfp_work( int matrix_order, lapack_int m, lapack_int n,
+                                 float* a, lapack_int lda, float* tau,
+                                 float* work, lapack_int lwork );
+lapack_int LAPACKE_dgeqrfp_work( int matrix_order, lapack_int m, lapack_int n,
+                                 double* a, lapack_int lda, double* tau,
+                                 double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeqrfp_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* tau,
+                                 lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgeqrfp_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* tau,
+                                 lapack_complex_double* work,
+                                 lapack_int lwork );
+
+lapack_int LAPACKE_sgerfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const float* af, lapack_int ldaf,
+                                const lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgerfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cgerfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgerfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgerfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int nrhs, const float* a,
+                                 lapack_int lda, const float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* r, const float* c, const float* b,
+                                 lapack_int ldb, float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dgerfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int nrhs, const double* a,
+                                 lapack_int lda, const double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* r, const double* c,
+                                 const double* b, lapack_int ldb, double* x,
+                                 lapack_int ldx, double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cgerfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 const lapack_complex_float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* r, const float* c,
+                                 const lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zgerfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 const lapack_complex_double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* r, const double* c,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_sgerqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgerqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgerqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgerqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgesdd_work( int matrix_order, char jobz, lapack_int m,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* s, float* u, lapack_int ldu, float* vt,
+                                lapack_int ldvt, float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgesdd_work( int matrix_order, char jobz, lapack_int m,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* s, double* u, lapack_int ldu,
+                                double* vt, lapack_int ldvt, double* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_cgesdd_work( int matrix_order, char jobz, lapack_int m,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, float* s,
+                                lapack_complex_float* u, lapack_int ldu,
+                                lapack_complex_float* vt, lapack_int ldvt,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int* iwork );
+lapack_int LAPACKE_zgesdd_work( int matrix_order, char jobz, lapack_int m,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, double* s,
+                                lapack_complex_double* u, lapack_int ldu,
+                                lapack_complex_double* vt, lapack_int ldvt,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int* iwork );
+
+lapack_int LAPACKE_sgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               float* a, lapack_int lda, lapack_int* ipiv,
+                               float* b, lapack_int ldb );
+lapack_int LAPACKE_dgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               double* a, lapack_int lda, lapack_int* ipiv,
+                               double* b, lapack_int ldb );
+lapack_int LAPACKE_cgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               lapack_complex_float* a, lapack_int lda,
+                               lapack_int* ipiv, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_int* ipiv, lapack_complex_double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dsgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                double* a, lapack_int lda, lapack_int* ipiv,
+                                double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* work, float* swork,
+                                lapack_int* iter );
+lapack_int LAPACKE_zcgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, lapack_complex_double* work,
+                                lapack_complex_float* swork, double* rwork,
+                                lapack_int* iter );
+
+lapack_int LAPACKE_sgesvd_work( int matrix_order, char jobu, char jobvt,
+                                lapack_int m, lapack_int n, float* a,
+                                lapack_int lda, float* s, float* u,
+                                lapack_int ldu, float* vt, lapack_int ldvt,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgesvd_work( int matrix_order, char jobu, char jobvt,
+                                lapack_int m, lapack_int n, double* a,
+                                lapack_int lda, double* s, double* u,
+                                lapack_int ldu, double* vt, lapack_int ldvt,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgesvd_work( int matrix_order, char jobu, char jobvt,
+                                lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float* s, lapack_complex_float* u,
+                                lapack_int ldu, lapack_complex_float* vt,
+                                lapack_int ldvt, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zgesvd_work( int matrix_order, char jobu, char jobvt,
+                                lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double* s, lapack_complex_double* u,
+                                lapack_int ldu, lapack_complex_double* vt,
+                                lapack_int ldvt, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork );
+
+lapack_int LAPACKE_sgesvj_work( int matrix_order, char joba, char jobu,
+                                char jobv, lapack_int m, lapack_int n, float* a,
+                                lapack_int lda, float* sva, lapack_int mv,
+                                float* v, lapack_int ldv, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dgesvj_work( int matrix_order, char joba, char jobu,
+                                char jobv, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* sva,
+                                lapack_int mv, double* v, lapack_int ldv,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgesvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs, float* a,
+                                lapack_int lda, float* af, lapack_int ldaf,
+                                lapack_int* ipiv, char* equed, float* r,
+                                float* c, float* b, lapack_int ldb, float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dgesvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs, double* a,
+                                lapack_int lda, double* af, lapack_int ldaf,
+                                lapack_int* ipiv, char* equed, double* r,
+                                double* c, double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, double* work, lapack_int* iwork );
+lapack_int LAPACKE_cgesvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* af, lapack_int ldaf,
+                                lapack_int* ipiv, char* equed, float* r,
+                                float* c, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zgesvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* af, lapack_int ldaf,
+                                lapack_int* ipiv, char* equed, double* r,
+                                double* c, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_sgesvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int nrhs, float* a,
+                                 lapack_int lda, float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* r,
+                                 float* c, float* b, lapack_int ldb, float* x,
+                                 lapack_int ldx, float* rcond, float* rpvgrw,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dgesvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int nrhs, double* a,
+                                 lapack_int lda, double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* r,
+                                 double* c, double* b, lapack_int ldb,
+                                 double* x, lapack_int ldx, double* rcond,
+                                 double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cgesvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* r,
+                                 float* c, lapack_complex_float* b,
+                                 lapack_int ldb, lapack_complex_float* x,
+                                 lapack_int ldx, float* rcond, float* rpvgrw,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params,
+                                 lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgesvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* r,
+                                 double* c, lapack_complex_double* b,
+                                 lapack_int ldb, lapack_complex_double* x,
+                                 lapack_int ldx, double* rcond, double* rpvgrw,
+                                 double* berr, lapack_int n_err_bnds,
+                                 double* err_bnds_norm, double* err_bnds_comp,
+                                 lapack_int nparams, double* params,
+                                 lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgetf2_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dgetf2_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_cgetf2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ipiv );
+lapack_int LAPACKE_zgetf2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dgetrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_cgetrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ipiv );
+lapack_int LAPACKE_zgetrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetri_work( int matrix_order, lapack_int n, float* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgetri_work( int matrix_order, lapack_int n, double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgetri_work( int matrix_order, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgetri_work( int matrix_order, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgetrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const lapack_int* ipiv, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dgetrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_cgetrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zgetrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sggbak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const float* lscale, const float* rscale,
+                                lapack_int m, float* v, lapack_int ldv );
+lapack_int LAPACKE_dggbak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const double* lscale, const double* rscale,
+                                lapack_int m, double* v, lapack_int ldv );
+lapack_int LAPACKE_cggbak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const float* lscale, const float* rscale,
+                                lapack_int m, lapack_complex_float* v,
+                                lapack_int ldv );
+lapack_int LAPACKE_zggbak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const double* lscale, const double* rscale,
+                                lapack_int m, lapack_complex_double* v,
+                                lapack_int ldv );
+
+lapack_int LAPACKE_sggbal_work( int matrix_order, char job, lapack_int n,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, lapack_int* ilo,
+                                lapack_int* ihi, float* lscale, float* rscale,
+                                float* work );
+lapack_int LAPACKE_dggbal_work( int matrix_order, char job, lapack_int n,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, lapack_int* ilo,
+                                lapack_int* ihi, double* lscale, double* rscale,
+                                double* work );
+lapack_int LAPACKE_cggbal_work( int matrix_order, char job, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_int* ilo, lapack_int* ihi, float* lscale,
+                                float* rscale, float* work );
+lapack_int LAPACKE_zggbal_work( int matrix_order, char job, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_int* ilo, lapack_int* ihi,
+                                double* lscale, double* rscale, double* work );
+
+lapack_int LAPACKE_sgges_work( int matrix_order, char jobvsl, char jobvsr,
+                               char sort, LAPACK_S_SELECT3 selctg, lapack_int n,
+                               float* a, lapack_int lda, float* b,
+                               lapack_int ldb, lapack_int* sdim, float* alphar,
+                               float* alphai, float* beta, float* vsl,
+                               lapack_int ldvsl, float* vsr, lapack_int ldvsr,
+                               float* work, lapack_int lwork,
+                               lapack_logical* bwork );
+lapack_int LAPACKE_dgges_work( int matrix_order, char jobvsl, char jobvsr,
+                               char sort, LAPACK_D_SELECT3 selctg, lapack_int n,
+                               double* a, lapack_int lda, double* b,
+                               lapack_int ldb, lapack_int* sdim, double* alphar,
+                               double* alphai, double* beta, double* vsl,
+                               lapack_int ldvsl, double* vsr, lapack_int ldvsr,
+                               double* work, lapack_int lwork,
+                               lapack_logical* bwork );
+lapack_int LAPACKE_cgges_work( int matrix_order, char jobvsl, char jobvsr,
+                               char sort, LAPACK_C_SELECT2 selctg, lapack_int n,
+                               lapack_complex_float* a, lapack_int lda,
+                               lapack_complex_float* b, lapack_int ldb,
+                               lapack_int* sdim, lapack_complex_float* alpha,
+                               lapack_complex_float* beta,
+                               lapack_complex_float* vsl, lapack_int ldvsl,
+                               lapack_complex_float* vsr, lapack_int ldvsr,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork, lapack_logical* bwork );
+lapack_int LAPACKE_zgges_work( int matrix_order, char jobvsl, char jobvsr,
+                               char sort, LAPACK_Z_SELECT2 selctg, lapack_int n,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_complex_double* b, lapack_int ldb,
+                               lapack_int* sdim, lapack_complex_double* alpha,
+                               lapack_complex_double* beta,
+                               lapack_complex_double* vsl, lapack_int ldvsl,
+                               lapack_complex_double* vsr, lapack_int ldvsr,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork, lapack_logical* bwork );
+
+lapack_int LAPACKE_sggesx_work( int matrix_order, char jobvsl, char jobvsr,
+                                char sort, LAPACK_S_SELECT3 selctg, char sense,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, lapack_int* sdim,
+                                float* alphar, float* alphai, float* beta,
+                                float* vsl, lapack_int ldvsl, float* vsr,
+                                lapack_int ldvsr, float* rconde, float* rcondv,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_dggesx_work( int matrix_order, char jobvsl, char jobvsr,
+                                char sort, LAPACK_D_SELECT3 selctg, char sense,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, lapack_int* sdim,
+                                double* alphar, double* alphai, double* beta,
+                                double* vsl, lapack_int ldvsl, double* vsr,
+                                lapack_int ldvsr, double* rconde,
+                                double* rcondv, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_cggesx_work( int matrix_order, char jobvsl, char jobvsr,
+                                char sort, LAPACK_C_SELECT2 selctg, char sense,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, lapack_int* sdim,
+                                lapack_complex_float* alpha,
+                                lapack_complex_float* beta,
+                                lapack_complex_float* vsl, lapack_int ldvsl,
+                                lapack_complex_float* vsr, lapack_int ldvsr,
+                                float* rconde, float* rcondv,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int* iwork,
+                                lapack_int liwork, lapack_logical* bwork );
+lapack_int LAPACKE_zggesx_work( int matrix_order, char jobvsl, char jobvsr,
+                                char sort, LAPACK_Z_SELECT2 selctg, char sense,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_int* sdim,
+                                lapack_complex_double* alpha,
+                                lapack_complex_double* beta,
+                                lapack_complex_double* vsl, lapack_int ldvsl,
+                                lapack_complex_double* vsr, lapack_int ldvsr,
+                                double* rconde, double* rcondv,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int* iwork,
+                                lapack_int liwork, lapack_logical* bwork );
+
+lapack_int LAPACKE_sggev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, float* a, lapack_int lda, float* b,
+                               lapack_int ldb, float* alphar, float* alphai,
+                               float* beta, float* vl, lapack_int ldvl,
+                               float* vr, lapack_int ldvr, float* work,
+                               lapack_int lwork );
+lapack_int LAPACKE_dggev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, double* a, lapack_int lda,
+                               double* b, lapack_int ldb, double* alphar,
+                               double* alphai, double* beta, double* vl,
+                               lapack_int ldvl, double* vr, lapack_int ldvr,
+                               double* work, lapack_int lwork );
+lapack_int LAPACKE_cggev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, lapack_complex_float* a,
+                               lapack_int lda, lapack_complex_float* b,
+                               lapack_int ldb, lapack_complex_float* alpha,
+                               lapack_complex_float* beta,
+                               lapack_complex_float* vl, lapack_int ldvl,
+                               lapack_complex_float* vr, lapack_int ldvr,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork );
+lapack_int LAPACKE_zggev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, lapack_complex_double* a,
+                               lapack_int lda, lapack_complex_double* b,
+                               lapack_int ldb, lapack_complex_double* alpha,
+                               lapack_complex_double* beta,
+                               lapack_complex_double* vl, lapack_int ldvl,
+                               lapack_complex_double* vr, lapack_int ldvr,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork );
+
+lapack_int LAPACKE_sggevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n, float* a,
+                                lapack_int lda, float* b, lapack_int ldb,
+                                float* alphar, float* alphai, float* beta,
+                                float* vl, lapack_int ldvl, float* vr,
+                                lapack_int ldvr, lapack_int* ilo,
+                                lapack_int* ihi, float* lscale, float* rscale,
+                                float* abnrm, float* bbnrm, float* rconde,
+                                float* rcondv, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_logical* bwork );
+lapack_int LAPACKE_dggevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double* alphar, double* alphai, double* beta,
+                                double* vl, lapack_int ldvl, double* vr,
+                                lapack_int ldvr, lapack_int* ilo,
+                                lapack_int* ihi, double* lscale, double* rscale,
+                                double* abnrm, double* bbnrm, double* rconde,
+                                double* rcondv, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_logical* bwork );
+lapack_int LAPACKE_cggevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* alpha,
+                                lapack_complex_float* beta,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi, float* lscale,
+                                float* rscale, float* abnrm, float* bbnrm,
+                                float* rconde, float* rcondv,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int* iwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_zggevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* alpha,
+                                lapack_complex_double* beta,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi,
+                                double* lscale, double* rscale, double* abnrm,
+                                double* bbnrm, double* rconde, double* rcondv,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int* iwork,
+                                lapack_logical* bwork );
+
+lapack_int LAPACKE_sggglm_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float* d, float* x,
+                                float* y, float* work, lapack_int lwork );
+lapack_int LAPACKE_dggglm_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* d, double* x,
+                                double* y, double* work, lapack_int lwork );
+lapack_int LAPACKE_cggglm_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* d,
+                                lapack_complex_float* x,
+                                lapack_complex_float* y,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zggglm_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* d,
+                                lapack_complex_double* x,
+                                lapack_complex_double* y,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgghrd_work( int matrix_order, char compq, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, float* q, lapack_int ldq,
+                                float* z, lapack_int ldz );
+lapack_int LAPACKE_dgghrd_work( int matrix_order, char compq, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, double* q, lapack_int ldq,
+                                double* z, lapack_int ldz );
+lapack_int LAPACKE_cgghrd_work( int matrix_order, char compq, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zgghrd_work( int matrix_order, char compq, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sgglse_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int p, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float* c, float* d,
+                                float* x, float* work, lapack_int lwork );
+lapack_int LAPACKE_dgglse_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int p, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* c, double* d,
+                                double* x, double* work, lapack_int lwork );
+lapack_int LAPACKE_cgglse_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* c,
+                                lapack_complex_float* d,
+                                lapack_complex_float* x,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgglse_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* c,
+                                lapack_complex_double* d,
+                                lapack_complex_double* x,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sggqrf_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, float* a, lapack_int lda,
+                                float* taua, float* b, lapack_int ldb,
+                                float* taub, float* work, lapack_int lwork );
+lapack_int LAPACKE_dggqrf_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, double* a, lapack_int lda,
+                                double* taua, double* b, lapack_int ldb,
+                                double* taub, double* work, lapack_int lwork );
+lapack_int LAPACKE_cggqrf_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* taua,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* taub,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zggqrf_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* taua,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* taub,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sggrqf_work( int matrix_order, lapack_int m, lapack_int p,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* taua, float* b, lapack_int ldb,
+                                float* taub, float* work, lapack_int lwork );
+lapack_int LAPACKE_dggrqf_work( int matrix_order, lapack_int m, lapack_int p,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* taua, double* b, lapack_int ldb,
+                                double* taub, double* work, lapack_int lwork );
+lapack_int LAPACKE_cggrqf_work( int matrix_order, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* taua,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* taub,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zggrqf_work( int matrix_order, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* taua,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* taub,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sggsvd_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_int* k, lapack_int* l,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, float* alpha, float* beta,
+                                float* u, lapack_int ldu, float* v,
+                                lapack_int ldv, float* q, lapack_int ldq,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dggsvd_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_int* k, lapack_int* l,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, double* alpha, double* beta,
+                                double* u, lapack_int ldu, double* v,
+                                lapack_int ldv, double* q, lapack_int ldq,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cggsvd_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_int* k, lapack_int* l,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                float* alpha, float* beta,
+                                lapack_complex_float* u, lapack_int ldu,
+                                lapack_complex_float* v, lapack_int ldv,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_zggsvd_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_int* k, lapack_int* l,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                double* alpha, double* beta,
+                                lapack_complex_double* u, lapack_int ldu,
+                                lapack_complex_double* v, lapack_int ldv,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_sggsvp_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float tola,
+                                float tolb, lapack_int* k, lapack_int* l,
+                                float* u, lapack_int ldu, float* v,
+                                lapack_int ldv, float* q, lapack_int ldq,
+                                lapack_int* iwork, float* tau, float* work );
+lapack_int LAPACKE_dggsvp_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double tola,
+                                double tolb, lapack_int* k, lapack_int* l,
+                                double* u, lapack_int ldu, double* v,
+                                lapack_int ldv, double* q, lapack_int ldq,
+                                lapack_int* iwork, double* tau, double* work );
+lapack_int LAPACKE_cggsvp_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, float tola, float tolb,
+                                lapack_int* k, lapack_int* l,
+                                lapack_complex_float* u, lapack_int ldu,
+                                lapack_complex_float* v, lapack_int ldv,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_int* iwork, float* rwork,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zggsvp_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, double tola, double tolb,
+                                lapack_int* k, lapack_int* l,
+                                lapack_complex_double* u, lapack_int ldu,
+                                lapack_complex_double* v, lapack_int ldv,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_int* iwork, double* rwork,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_sgtcon_work( char norm, lapack_int n, const float* dl,
+                                const float* d, const float* du,
+                                const float* du2, const lapack_int* ipiv,
+                                float anorm, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgtcon_work( char norm, lapack_int n, const double* dl,
+                                const double* d, const double* du,
+                                const double* du2, const lapack_int* ipiv,
+                                double anorm, double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgtcon_work( char norm, lapack_int n,
+                                const lapack_complex_float* dl,
+                                const lapack_complex_float* d,
+                                const lapack_complex_float* du,
+                                const lapack_complex_float* du2,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zgtcon_work( char norm, lapack_int n,
+                                const lapack_complex_double* dl,
+                                const lapack_complex_double* d,
+                                const lapack_complex_double* du,
+                                const lapack_complex_double* du2,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_sgtrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const float* dl,
+                                const float* d, const float* du,
+                                const float* dlf, const float* df,
+                                const float* duf, const float* du2,
+                                const lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgtrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const double* dl,
+                                const double* d, const double* du,
+                                const double* dlf, const double* df,
+                                const double* duf, const double* du2,
+                                const lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgtrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* dl,
+                                const lapack_complex_float* d,
+                                const lapack_complex_float* du,
+                                const lapack_complex_float* dlf,
+                                const lapack_complex_float* df,
+                                const lapack_complex_float* duf,
+                                const lapack_complex_float* du2,
+                                const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgtrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* dl,
+                                const lapack_complex_double* d,
+                                const lapack_complex_double* du,
+                                const lapack_complex_double* dlf,
+                                const lapack_complex_double* df,
+                                const lapack_complex_double* duf,
+                                const lapack_complex_double* du2,
+                                const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgtsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               float* dl, float* d, float* du, float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dgtsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               double* dl, double* d, double* du, double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_cgtsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               lapack_complex_float* dl,
+                               lapack_complex_float* d,
+                               lapack_complex_float* du,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zgtsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               lapack_complex_double* dl,
+                               lapack_complex_double* d,
+                               lapack_complex_double* du,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sgtsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs, const float* dl,
+                                const float* d, const float* du, float* dlf,
+                                float* df, float* duf, float* du2,
+                                lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dgtsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs, const double* dl,
+                                const double* d, const double* du, double* dlf,
+                                double* df, double* duf, double* du2,
+                                lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cgtsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* dl,
+                                const lapack_complex_float* d,
+                                const lapack_complex_float* du,
+                                lapack_complex_float* dlf,
+                                lapack_complex_float* df,
+                                lapack_complex_float* duf,
+                                lapack_complex_float* du2, lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgtsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* dl,
+                                const lapack_complex_double* d,
+                                const lapack_complex_double* du,
+                                lapack_complex_double* dlf,
+                                lapack_complex_double* df,
+                                lapack_complex_double* duf,
+                                lapack_complex_double* du2, lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgttrf_work( lapack_int n, float* dl, float* d, float* du,
+                                float* du2, lapack_int* ipiv );
+lapack_int LAPACKE_dgttrf_work( lapack_int n, double* dl, double* d, double* du,
+                                double* du2, lapack_int* ipiv );
+lapack_int LAPACKE_cgttrf_work( lapack_int n, lapack_complex_float* dl,
+                                lapack_complex_float* d,
+                                lapack_complex_float* du,
+                                lapack_complex_float* du2, lapack_int* ipiv );
+lapack_int LAPACKE_zgttrf_work( lapack_int n, lapack_complex_double* dl,
+                                lapack_complex_double* d,
+                                lapack_complex_double* du,
+                                lapack_complex_double* du2, lapack_int* ipiv );
+
+lapack_int LAPACKE_sgttrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const float* dl,
+                                const float* d, const float* du,
+                                const float* du2, const lapack_int* ipiv,
+                                float* b, lapack_int ldb );
+lapack_int LAPACKE_dgttrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const double* dl,
+                                const double* d, const double* du,
+                                const double* du2, const lapack_int* ipiv,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_cgttrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* dl,
+                                const lapack_complex_float* d,
+                                const lapack_complex_float* du,
+                                const lapack_complex_float* du2,
+                                const lapack_int* ipiv, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zgttrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* dl,
+                                const lapack_complex_double* d,
+                                const lapack_complex_double* du,
+                                const lapack_complex_double* du2,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_chbev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int kd,
+                               lapack_complex_float* ab, lapack_int ldab,
+                               float* w, lapack_complex_float* z,
+                               lapack_int ldz, lapack_complex_float* work,
+                               float* rwork );
+lapack_int LAPACKE_zhbev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int kd,
+                               lapack_complex_double* ab, lapack_int ldab,
+                               double* w, lapack_complex_double* z,
+                               lapack_int ldz, lapack_complex_double* work,
+                               double* rwork );
+
+lapack_int LAPACKE_chbevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int kd,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zhbevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int kd,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_chbevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int kd,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                lapack_complex_float* q, lapack_int ldq,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                float* rwork, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_zhbevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int kd,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                lapack_complex_double* q, lapack_int ldq,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                double* rwork, lapack_int* iwork,
+                                lapack_int* ifail );
+
+lapack_int LAPACKE_chbgst_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_complex_float* bb, lapack_int ldbb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhbgst_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                const lapack_complex_double* bb,
+                                lapack_int ldbb, lapack_complex_double* x,
+                                lapack_int ldx, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_chbgv_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int ka, lapack_int kb,
+                               lapack_complex_float* ab, lapack_int ldab,
+                               lapack_complex_float* bb, lapack_int ldbb,
+                               float* w, lapack_complex_float* z,
+                               lapack_int ldz, lapack_complex_float* work,
+                               float* rwork );
+lapack_int LAPACKE_zhbgv_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int ka, lapack_int kb,
+                               lapack_complex_double* ab, lapack_int ldab,
+                               lapack_complex_double* bb, lapack_int ldbb,
+                               double* w, lapack_complex_double* z,
+                               lapack_int ldz, lapack_complex_double* work,
+                               double* rwork );
+
+lapack_int LAPACKE_chbgvd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                lapack_complex_float* bb, lapack_int ldbb,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zhbgvd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                lapack_complex_double* bb, lapack_int ldbb,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_chbgvx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int ka,
+                                lapack_int kb, lapack_complex_float* ab,
+                                lapack_int ldab, lapack_complex_float* bb,
+                                lapack_int ldbb, lapack_complex_float* q,
+                                lapack_int ldq, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zhbgvx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int ka,
+                                lapack_int kb, lapack_complex_double* ab,
+                                lapack_int ldab, lapack_complex_double* bb,
+                                lapack_int ldbb, lapack_complex_double* q,
+                                lapack_int ldq, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_chbtrd_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int kd,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                float* d, float* e, lapack_complex_float* q,
+                                lapack_int ldq, lapack_complex_float* work );
+lapack_int LAPACKE_zhbtrd_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int kd,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                double* d, double* e, lapack_complex_double* q,
+                                lapack_int ldq, lapack_complex_double* work );
+
+lapack_int LAPACKE_checon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zhecon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_cheequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 float* s, float* scond, float* amax,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_zheequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 double* s, double* scond, double* amax,
+                                 lapack_complex_double* work );
+
+lapack_int LAPACKE_cheev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_complex_float* a,
+                               lapack_int lda, float* w,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork );
+lapack_int LAPACKE_zheev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_complex_double* a,
+                               lapack_int lda, double* w,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork );
+
+lapack_int LAPACKE_cheevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, float* w,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_zheevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, double* w,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_cheevr_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_int* isuppz,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_zheevr_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_int* isuppz,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_cheevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zheevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_chegst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zhegst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_chegv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, lapack_complex_float* a,
+                               lapack_int lda, lapack_complex_float* b,
+                               lapack_int ldb, float* w,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork );
+lapack_int LAPACKE_zhegv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_complex_double* b, lapack_int ldb,
+                               double* w, lapack_complex_double* work,
+                               lapack_int lwork, double* rwork );
+
+lapack_int LAPACKE_chegvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                float* w, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zhegvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                double* w, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_chegvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zhegvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_cherfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zherfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_cherfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 const lapack_complex_float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* s, const lapack_complex_float* b,
+                                 lapack_int ldb, lapack_complex_float* x,
+                                 lapack_int ldx, float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zherfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 const lapack_complex_double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* s,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_chesv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* a,
+                               lapack_int lda, lapack_int* ipiv,
+                               lapack_complex_float* b, lapack_int ldb,
+                               lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zhesv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* a,
+                               lapack_int lda, lapack_int* ipiv,
+                               lapack_complex_double* b, lapack_int ldb,
+                               lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_chesvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* af, lapack_int ldaf,
+                                lapack_int* ipiv, const lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zhesvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* af, lapack_int ldaf,
+                                lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_chesvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* s,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zhesvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* s,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_chetrd_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float* d, float* e, lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zhetrd_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double* d, double* e,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_chetrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_zhetrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_chetri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zhetri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_chetrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zhetrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_chfrk_work( int matrix_order, char transr, char uplo,
+                               char trans, lapack_int n, lapack_int k,
+                               float alpha, const lapack_complex_float* a,
+                               lapack_int lda, float beta,
+                               lapack_complex_float* c );
+lapack_int LAPACKE_zhfrk_work( int matrix_order, char transr, char uplo,
+                               char trans, lapack_int n, lapack_int k,
+                               double alpha, const lapack_complex_double* a,
+                               lapack_int lda, double beta,
+                               lapack_complex_double* c );
+
+lapack_int LAPACKE_shgeqz_work( int matrix_order, char job, char compq,
+                                char compz, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, float* h, lapack_int ldh,
+                                float* t, lapack_int ldt, float* alphar,
+                                float* alphai, float* beta, float* q,
+                                lapack_int ldq, float* z, lapack_int ldz,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dhgeqz_work( int matrix_order, char job, char compq,
+                                char compz, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, double* h, lapack_int ldh,
+                                double* t, lapack_int ldt, double* alphar,
+                                double* alphai, double* beta, double* q,
+                                lapack_int ldq, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_chgeqz_work( int matrix_order, char job, char compq,
+                                char compz, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_float* h,
+                                lapack_int ldh, lapack_complex_float* t,
+                                lapack_int ldt, lapack_complex_float* alpha,
+                                lapack_complex_float* beta,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork );
+lapack_int LAPACKE_zhgeqz_work( int matrix_order, char job, char compq,
+                                char compz, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_double* h,
+                                lapack_int ldh, lapack_complex_double* t,
+                                lapack_int ldt, lapack_complex_double* alpha,
+                                lapack_complex_double* beta,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_chpcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zhpcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_chpev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_complex_float* ap, float* w,
+                               lapack_complex_float* z, lapack_int ldz,
+                               lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhpev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_complex_double* ap,
+                               double* w, lapack_complex_double* z,
+                               lapack_int ldz, lapack_complex_double* work,
+                               double* rwork );
+
+lapack_int LAPACKE_chpevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_complex_float* ap,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zhpevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_complex_double* ap,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_chpevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* ap, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zhpevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* ap, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_chpgst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, lapack_complex_float* ap,
+                                const lapack_complex_float* bp );
+lapack_int LAPACKE_zhpgst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, lapack_complex_double* ap,
+                                const lapack_complex_double* bp );
+
+lapack_int LAPACKE_chpgv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n,
+                               lapack_complex_float* ap,
+                               lapack_complex_float* bp, float* w,
+                               lapack_complex_float* z, lapack_int ldz,
+                               lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhpgv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n,
+                               lapack_complex_double* ap,
+                               lapack_complex_double* bp, double* w,
+                               lapack_complex_double* z, lapack_int ldz,
+                               lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_chpgvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* ap,
+                                lapack_complex_float* bp, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_zhpgvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* ap,
+                                lapack_complex_double* bp, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_chpgvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n,
+                                lapack_complex_float* ap,
+                                lapack_complex_float* bp, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zhpgvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n,
+                                lapack_complex_double* ap,
+                                lapack_complex_double* bp, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_chprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_complex_float* afp,
+                                const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* afp,
+                                const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_chpsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* ap,
+                               lapack_int* ipiv, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zhpsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* ap,
+                               lapack_int* ipiv, lapack_complex_double* b,
+                               lapack_int ldb );
+
+lapack_int LAPACKE_chpsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* ap,
+                                lapack_complex_float* afp, lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhpsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* afp, lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_chptrd_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap, float* d, float* e,
+                                lapack_complex_float* tau );
+lapack_int LAPACKE_zhptrd_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap, double* d, double* e,
+                                lapack_complex_double* tau );
+
+lapack_int LAPACKE_chptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_zhptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap, lapack_int* ipiv );
+
+lapack_int LAPACKE_chptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zhptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_chptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_int* ipiv, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zhptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_shsein_work( int matrix_order, char job, char eigsrc,
+                                char initv, lapack_logical* select,
+                                lapack_int n, const float* h, lapack_int ldh,
+                                float* wr, const float* wi, float* vl,
+                                lapack_int ldvl, float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, float* work,
+                                lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_dhsein_work( int matrix_order, char job, char eigsrc,
+                                char initv, lapack_logical* select,
+                                lapack_int n, const double* h, lapack_int ldh,
+                                double* wr, const double* wi, double* vl,
+                                lapack_int ldvl, double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, double* work,
+                                lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_chsein_work( int matrix_order, char job, char eigsrc,
+                                char initv, const lapack_logical* select,
+                                lapack_int n, const lapack_complex_float* h,
+                                lapack_int ldh, lapack_complex_float* w,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_zhsein_work( int matrix_order, char job, char eigsrc,
+                                char initv, const lapack_logical* select,
+                                lapack_int n, const lapack_complex_double* h,
+                                lapack_int ldh, lapack_complex_double* w,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* ifaill, lapack_int* ifailr );
+
+lapack_int LAPACKE_shseqr_work( int matrix_order, char job, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                float* h, lapack_int ldh, float* wr, float* wi,
+                                float* z, lapack_int ldz, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dhseqr_work( int matrix_order, char job, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                double* h, lapack_int ldh, double* wr,
+                                double* wi, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_chseqr_work( int matrix_order, char job, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                lapack_complex_float* h, lapack_int ldh,
+                                lapack_complex_float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zhseqr_work( int matrix_order, char job, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                lapack_complex_double* h, lapack_int ldh,
+                                lapack_complex_double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_clacgv_work( lapack_int n, lapack_complex_float* x,
+                                lapack_int incx );
+lapack_int LAPACKE_zlacgv_work( lapack_int n, lapack_complex_double* x,
+                                lapack_int incx );
+
+lapack_int LAPACKE_slacpy_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, const float* a, lapack_int lda,
+                                float* b, lapack_int ldb );
+lapack_int LAPACKE_dlacpy_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, const double* a, lapack_int lda,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_clacpy_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zlacpy_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_zlag2c_work( int matrix_order, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_float* sa, lapack_int ldsa );
+
+lapack_int LAPACKE_slag2d_work( int matrix_order, lapack_int m, lapack_int n,
+                                const float* sa, lapack_int ldsa, double* a,
+                                lapack_int lda );
+
+lapack_int LAPACKE_dlag2s_work( int matrix_order, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda, float* sa,
+                                lapack_int ldsa );
+
+lapack_int LAPACKE_clag2z_work( int matrix_order, lapack_int m, lapack_int n,
+                                const lapack_complex_float* sa, lapack_int ldsa,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_slagge_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const float* d,
+                                float* a, lapack_int lda, lapack_int* iseed,
+                                float* work );
+lapack_int LAPACKE_dlagge_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const double* d,
+                                double* a, lapack_int lda, lapack_int* iseed,
+                                double* work );
+lapack_int LAPACKE_clagge_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const float* d,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* iseed, lapack_complex_float* work );
+lapack_int LAPACKE_zlagge_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const double* d,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* iseed,
+                                lapack_complex_double* work );
+                                
+lapack_int LAPACKE_claghe_work( int matrix_order, lapack_int n, lapack_int k,
+                                const float* d, lapack_complex_float* a,
+                                lapack_int lda, lapack_int* iseed,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zlaghe_work( int matrix_order, lapack_int n, lapack_int k,
+                                const double* d, lapack_complex_double* a,
+                                lapack_int lda, lapack_int* iseed,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_slagsy_work( int matrix_order, lapack_int n, lapack_int k,
+                                const float* d, float* a, lapack_int lda,
+                                lapack_int* iseed, float* work );
+lapack_int LAPACKE_dlagsy_work( int matrix_order, lapack_int n, lapack_int k,
+                                const double* d, double* a, lapack_int lda,
+                                lapack_int* iseed, double* work );
+lapack_int LAPACKE_clagsy_work( int matrix_order, lapack_int n, lapack_int k,
+                                const float* d, lapack_complex_float* a,
+                                lapack_int lda, lapack_int* iseed,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zlagsy_work( int matrix_order, lapack_int n, lapack_int k,
+                                const double* d, lapack_complex_double* a,
+                                lapack_int lda, lapack_int* iseed,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_slapmr_work( int matrix_order, lapack_logical forwrd,
+                                lapack_int m, lapack_int n, float* x,
+                                lapack_int ldx, lapack_int* k );
+lapack_int LAPACKE_dlapmr_work( int matrix_order, lapack_logical forwrd,
+                                lapack_int m, lapack_int n, double* x,
+                                lapack_int ldx, lapack_int* k );
+lapack_int LAPACKE_clapmr_work( int matrix_order, lapack_logical forwrd,
+                                lapack_int m, lapack_int n,
+                                lapack_complex_float* x, lapack_int ldx,
+                                lapack_int* k );
+lapack_int LAPACKE_zlapmr_work( int matrix_order, lapack_logical forwrd,
+                                lapack_int m, lapack_int n,
+                                lapack_complex_double* x, lapack_int ldx,
+                                lapack_int* k );
+
+lapack_int LAPACKE_slartgp_work( float f, float g, float* cs, float* sn,
+                                 float* r );
+lapack_int LAPACKE_dlartgp_work( double f, double g, double* cs, double* sn,
+                                 double* r );
+
+lapack_int LAPACKE_slartgs_work( float x, float y, float sigma, float* cs,
+                                 float* sn );
+lapack_int LAPACKE_dlartgs_work( double x, double y, double sigma, double* cs,
+                                 double* sn );
+                                
+float LAPACKE_slapy2_work( float x, float y );
+double LAPACKE_dlapy2_work( double x, double y );
+
+float LAPACKE_slapy3_work( float x, float y, float z );
+double LAPACKE_dlapy3_work( double x, double y, double z );
+
+float LAPACKE_slamch_work( char cmach );
+double LAPACKE_dlamch_work( char cmach );
+
+float LAPACKE_slange_work( int matrix_order, char norm, lapack_int m,
+                                lapack_int n, const float* a, lapack_int lda,
+                                float* work );
+double LAPACKE_dlange_work( int matrix_order, char norm, lapack_int m,
+                                lapack_int n, const double* a, lapack_int lda,
+                                double* work );
+float LAPACKE_clange_work( int matrix_order, char norm, lapack_int m,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, float* work );
+double LAPACKE_zlange_work( int matrix_order, char norm, lapack_int m,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, double* work );
+
+float LAPACKE_clanhe_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, float* work );
+double LAPACKE_zlanhe_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, double* work );
+
+float LAPACKE_slansy_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const float* a, lapack_int lda,
+                                float* work );
+double LAPACKE_dlansy_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const double* a, lapack_int lda,
+                                double* work );
+float LAPACKE_clansy_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, float* work );
+double LAPACKE_zlansy_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, double* work );
+
+float LAPACKE_slantr_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int m, lapack_int n, const float* a,
+                                lapack_int lda, float* work );
+double LAPACKE_dlantr_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda, double* work );
+float LAPACKE_clantr_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float* work );
+double LAPACKE_zlantr_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double* work );
+
+lapack_int LAPACKE_slarfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, const float* v,
+                                lapack_int ldv, const float* t, lapack_int ldt,
+                                float* c, lapack_int ldc, float* work,
+                                lapack_int ldwork );
+lapack_int LAPACKE_dlarfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, const double* v,
+                                lapack_int ldv, const double* t, lapack_int ldt,
+                                double* c, lapack_int ldc, double* work,
+                                lapack_int ldwork );
+lapack_int LAPACKE_clarfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k,
+                                const lapack_complex_float* v, lapack_int ldv,
+                                const lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int ldwork );
+lapack_int LAPACKE_zlarfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k,
+                                const lapack_complex_double* v, lapack_int ldv,
+                                const lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work,
+                                lapack_int ldwork );
+
+lapack_int LAPACKE_slarfg_work( lapack_int n, float* alpha, float* x,
+                                lapack_int incx, float* tau );
+lapack_int LAPACKE_dlarfg_work( lapack_int n, double* alpha, double* x,
+                                lapack_int incx, double* tau );
+lapack_int LAPACKE_clarfg_work( lapack_int n, lapack_complex_float* alpha,
+                                lapack_complex_float* x, lapack_int incx,
+                                lapack_complex_float* tau );
+lapack_int LAPACKE_zlarfg_work( lapack_int n, lapack_complex_double* alpha,
+                                lapack_complex_double* x, lapack_int incx,
+                                lapack_complex_double* tau );
+
+lapack_int LAPACKE_slarft_work( int matrix_order, char direct, char storev,
+                                lapack_int n, lapack_int k, const float* v,
+                                lapack_int ldv, const float* tau, float* t,
+                                lapack_int ldt );
+lapack_int LAPACKE_dlarft_work( int matrix_order, char direct, char storev,
+                                lapack_int n, lapack_int k, const double* v,
+                                lapack_int ldv, const double* tau, double* t,
+                                lapack_int ldt );
+lapack_int LAPACKE_clarft_work( int matrix_order, char direct, char storev,
+                                lapack_int n, lapack_int k,
+                                const lapack_complex_float* v, lapack_int ldv,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zlarft_work( int matrix_order, char direct, char storev,
+                                lapack_int n, lapack_int k,
+                                const lapack_complex_double* v, lapack_int ldv,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_slarfx_work( int matrix_order, char side, lapack_int m,
+                                lapack_int n, const float* v, float tau,
+                                float* c, lapack_int ldc, float* work );
+lapack_int LAPACKE_dlarfx_work( int matrix_order, char side, lapack_int m,
+                                lapack_int n, const double* v, double tau,
+                                double* c, lapack_int ldc, double* work );
+lapack_int LAPACKE_clarfx_work( int matrix_order, char side, lapack_int m,
+                                lapack_int n, const lapack_complex_float* v,
+                                lapack_complex_float tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zlarfx_work( int matrix_order, char side, lapack_int m,
+                                lapack_int n, const lapack_complex_double* v,
+                                lapack_complex_double tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_slarnv_work( lapack_int idist, lapack_int* iseed,
+                                lapack_int n, float* x );
+lapack_int LAPACKE_dlarnv_work( lapack_int idist, lapack_int* iseed,
+                                lapack_int n, double* x );
+lapack_int LAPACKE_clarnv_work( lapack_int idist, lapack_int* iseed,
+                                lapack_int n, lapack_complex_float* x );
+lapack_int LAPACKE_zlarnv_work( lapack_int idist, lapack_int* iseed,
+                                lapack_int n, lapack_complex_double* x );
+
+lapack_int LAPACKE_slaset_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, float alpha, float beta, float* a,
+                                lapack_int lda );
+lapack_int LAPACKE_dlaset_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, double alpha, double beta,
+                                double* a, lapack_int lda );
+lapack_int LAPACKE_claset_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, lapack_complex_float alpha,
+                                lapack_complex_float beta,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zlaset_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, lapack_complex_double alpha,
+                                lapack_complex_double beta,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_slasrt_work( char id, lapack_int n, float* d );
+lapack_int LAPACKE_dlasrt_work( char id, lapack_int n, double* d );
+
+lapack_int LAPACKE_slaswp_work( int matrix_order, lapack_int n, float* a,
+                                lapack_int lda, lapack_int k1, lapack_int k2,
+                                const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_dlaswp_work( int matrix_order, lapack_int n, double* a,
+                                lapack_int lda, lapack_int k1, lapack_int k2,
+                                const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_claswp_work( int matrix_order, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int k1, lapack_int k2,
+                                const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_zlaswp_work( int matrix_order, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int k1, lapack_int k2,
+                                const lapack_int* ipiv, lapack_int incx );
+
+lapack_int LAPACKE_slatms_work( int matrix_order, lapack_int m, lapack_int n,
+                                char dist, lapack_int* iseed, char sym,
+                                float* d, lapack_int mode, float cond,
+                                float dmax, lapack_int kl, lapack_int ku,
+                                char pack, float* a, lapack_int lda,
+                                float* work );
+lapack_int LAPACKE_dlatms_work( int matrix_order, lapack_int m, lapack_int n,
+                                char dist, lapack_int* iseed, char sym,
+                                double* d, lapack_int mode, double cond,
+                                double dmax, lapack_int kl, lapack_int ku,
+                                char pack, double* a, lapack_int lda,
+                                double* work );
+lapack_int LAPACKE_clatms_work( int matrix_order, lapack_int m, lapack_int n,
+                                char dist, lapack_int* iseed, char sym,
+                                float* d, lapack_int mode, float cond,
+                                float dmax, lapack_int kl, lapack_int ku,
+                                char pack, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* work );
+lapack_int LAPACKE_zlatms_work( int matrix_order, lapack_int m, lapack_int n,
+                                char dist, lapack_int* iseed, char sym,
+                                double* d, lapack_int mode, double cond,
+                                double dmax, lapack_int kl, lapack_int ku,
+                                char pack, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* work );
+
+lapack_int LAPACKE_slauum_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda );
+lapack_int LAPACKE_dlauum_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda );
+lapack_int LAPACKE_clauum_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zlauum_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_sopgtr_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, const float* tau, float* q,
+                                lapack_int ldq, float* work );
+lapack_int LAPACKE_dopgtr_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, const double* tau, double* q,
+                                lapack_int ldq, double* work );
+
+lapack_int LAPACKE_sopmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const float* ap, const float* tau, float* c,
+                                lapack_int ldc, float* work );
+lapack_int LAPACKE_dopmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const double* ap, const double* tau, double* c,
+                                lapack_int ldc, double* work );
+
+lapack_int LAPACKE_sorgbr_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int k, float* a,
+                                lapack_int lda, const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorgbr_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int k, double* a,
+                                lapack_int lda, const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorghr_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorghr_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorglq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorglq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorgql_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorgql_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorgqr_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorgqr_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorgrq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorgrq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorgtr_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda, const float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dorgtr_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda, const double* tau,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormbr_work( int matrix_order, char vect, char side,
+                                char trans, lapack_int m, lapack_int n,
+                                lapack_int k, const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormbr_work( int matrix_order, char vect, char side,
+                                char trans, lapack_int m, lapack_int n,
+                                lapack_int k, const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormhr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormhr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormlq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormlq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormql_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormql_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormqr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormqr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormrq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormrq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormrz_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                lapack_int l, const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormrz_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                lapack_int l, const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_spbcon_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const float* ab, lapack_int ldab,
+                                float anorm, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dpbcon_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const double* ab,
+                                lapack_int ldab, double anorm, double* rcond,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cpbcon_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const lapack_complex_float* ab,
+                                lapack_int ldab, float anorm, float* rcond,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zpbcon_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const lapack_complex_double* ab,
+                                lapack_int ldab, double anorm, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spbequ_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const float* ab, lapack_int ldab,
+                                float* s, float* scond, float* amax );
+lapack_int LAPACKE_dpbequ_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const double* ab,
+                                lapack_int ldab, double* s, double* scond,
+                                double* amax );
+lapack_int LAPACKE_cpbequ_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const lapack_complex_float* ab,
+                                lapack_int ldab, float* s, float* scond,
+                                float* amax );
+lapack_int LAPACKE_zpbequ_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const lapack_complex_double* ab,
+                                lapack_int ldab, double* s, double* scond,
+                                double* amax );
+
+lapack_int LAPACKE_spbrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs, const float* ab,
+                                lapack_int ldab, const float* afb,
+                                lapack_int ldafb, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dpbrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const double* ab, lapack_int ldab,
+                                const double* afb, lapack_int ldafb,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cpbrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_complex_float* afb,
+                                lapack_int ldafb, const lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zpbrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab,
+                                const lapack_complex_double* afb,
+                                lapack_int ldafb,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spbstf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kb, float* bb, lapack_int ldbb );
+lapack_int LAPACKE_dpbstf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kb, double* bb, lapack_int ldbb );
+lapack_int LAPACKE_cpbstf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kb, lapack_complex_float* bb,
+                                lapack_int ldbb );
+lapack_int LAPACKE_zpbstf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kb, lapack_complex_double* bb,
+                                lapack_int ldbb );
+
+lapack_int LAPACKE_spbsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int kd, lapack_int nrhs, float* ab,
+                               lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dpbsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int kd, lapack_int nrhs, double* ab,
+                               lapack_int ldab, double* b, lapack_int ldb );
+lapack_int LAPACKE_cpbsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int kd, lapack_int nrhs,
+                               lapack_complex_float* ab, lapack_int ldab,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpbsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int kd, lapack_int nrhs,
+                               lapack_complex_double* ab, lapack_int ldab,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spbsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int kd, lapack_int nrhs,
+                                float* ab, lapack_int ldab, float* afb,
+                                lapack_int ldafb, char* equed, float* s,
+                                float* b, lapack_int ldb, float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dpbsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int kd, lapack_int nrhs,
+                                double* ab, lapack_int ldab, double* afb,
+                                lapack_int ldafb, char* equed, double* s,
+                                double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, double* work, lapack_int* iwork );
+lapack_int LAPACKE_cpbsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int kd, lapack_int nrhs,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                lapack_complex_float* afb, lapack_int ldafb,
+                                char* equed, float* s, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zpbsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int kd, lapack_int nrhs,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                lapack_complex_double* afb, lapack_int ldafb,
+                                char* equed, double* s,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spbtrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, float* ab, lapack_int ldab );
+lapack_int LAPACKE_dpbtrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, double* ab, lapack_int ldab );
+lapack_int LAPACKE_cpbtrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_complex_float* ab,
+                                lapack_int ldab );
+lapack_int LAPACKE_zpbtrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_complex_double* ab,
+                                lapack_int ldab );
+
+lapack_int LAPACKE_spbtrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs, const float* ab,
+                                lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dpbtrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const double* ab, lapack_int ldab, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_cpbtrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpbtrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_spftrf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, float* a );
+lapack_int LAPACKE_dpftrf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, double* a );
+lapack_int LAPACKE_cpftrf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_zpftrf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_spftri_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, float* a );
+lapack_int LAPACKE_dpftri_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, double* a );
+lapack_int LAPACKE_cpftri_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_zpftri_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_spftrs_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_int nrhs, const float* a,
+                                float* b, lapack_int ldb );
+lapack_int LAPACKE_dpftrs_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_int nrhs, const double* a,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_cpftrs_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpftrs_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spocon_work( int matrix_order, char uplo, lapack_int n,
+                                const float* a, lapack_int lda, float anorm,
+                                float* rcond, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dpocon_work( int matrix_order, char uplo, lapack_int n,
+                                const double* a, lapack_int lda, double anorm,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cpocon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float anorm, float* rcond,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zpocon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double anorm, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spoequ_work( int matrix_order, lapack_int n, const float* a,
+                                lapack_int lda, float* s, float* scond,
+                                float* amax );
+lapack_int LAPACKE_dpoequ_work( int matrix_order, lapack_int n, const double* a,
+                                lapack_int lda, double* s, double* scond,
+                                double* amax );
+lapack_int LAPACKE_cpoequ_work( int matrix_order, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float* s, float* scond, float* amax );
+lapack_int LAPACKE_zpoequ_work( int matrix_order, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_spoequb_work( int matrix_order, lapack_int n, const float* a,
+                                 lapack_int lda, float* s, float* scond,
+                                 float* amax );
+lapack_int LAPACKE_dpoequb_work( int matrix_order, lapack_int n,
+                                 const double* a, lapack_int lda, double* s,
+                                 double* scond, double* amax );
+lapack_int LAPACKE_cpoequb_work( int matrix_order, lapack_int n,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 float* s, float* scond, float* amax );
+lapack_int LAPACKE_zpoequb_work( int matrix_order, lapack_int n,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_sporfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const float* af, lapack_int ldaf,
+                                const float* b, lapack_int ldb, float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dporfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const double* af,
+                                lapack_int ldaf, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cporfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* af,
+                                lapack_int ldaf, const lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zporfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* af,
+                                lapack_int ldaf, const lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sporfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs, const float* a,
+                                 lapack_int lda, const float* af,
+                                 lapack_int ldaf, const float* s,
+                                 const float* b, lapack_int ldb, float* x,
+                                 lapack_int ldx, float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dporfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs, const double* a,
+                                 lapack_int lda, const double* af,
+                                 lapack_int ldaf, const double* s,
+                                 const double* b, lapack_int ldb, double* x,
+                                 lapack_int ldx, double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cporfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 const lapack_complex_float* af,
+                                 lapack_int ldaf, const float* s,
+                                 const lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zporfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 const lapack_complex_double* af,
+                                 lapack_int ldaf, const double* s,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_sposv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, float* a, lapack_int lda,
+                               float* b, lapack_int ldb );
+lapack_int LAPACKE_dposv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, double* a, lapack_int lda,
+                               double* b, lapack_int ldb );
+lapack_int LAPACKE_cposv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* a,
+                               lapack_int lda, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zposv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* a,
+                               lapack_int lda, lapack_complex_double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dsposv_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* work, float* swork,
+                                lapack_int* iter );
+lapack_int LAPACKE_zcposv_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, lapack_complex_double* work,
+                                lapack_complex_float* swork, double* rwork,
+                                lapack_int* iter );
+
+lapack_int LAPACKE_sposvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, float* a,
+                                lapack_int lda, float* af, lapack_int ldaf,
+                                char* equed, float* s, float* b, lapack_int ldb,
+                                float* x, lapack_int ldx, float* rcond,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dposvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, double* a,
+                                lapack_int lda, double* af, lapack_int ldaf,
+                                char* equed, double* s, double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cposvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* af, lapack_int ldaf,
+                                char* equed, float* s, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zposvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* af, lapack_int ldaf,
+                                char* equed, double* s,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sposvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs, float* a,
+                                 lapack_int lda, float* af, lapack_int ldaf,
+                                 char* equed, float* s, float* b,
+                                 lapack_int ldb, float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dposvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs, double* a,
+                                 lapack_int lda, double* af, lapack_int ldaf,
+                                 char* equed, double* s, double* b,
+                                 lapack_int ldb, double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cposvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* af, lapack_int ldaf,
+                                 char* equed, float* s, lapack_complex_float* b,
+                                 lapack_int ldb, lapack_complex_float* x,
+                                 lapack_int ldx, float* rcond, float* rpvgrw,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params,
+                                 lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zposvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* af, lapack_int ldaf,
+                                 char* equed, double* s,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_spotrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda );
+lapack_int LAPACKE_dpotrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda );
+lapack_int LAPACKE_cpotrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zpotrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_spotri_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda );
+lapack_int LAPACKE_dpotri_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda );
+lapack_int LAPACKE_cpotri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zpotri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_spotrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                float* b, lapack_int ldb );
+lapack_int LAPACKE_dpotrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, double* b, lapack_int ldb );
+lapack_int LAPACKE_cpotrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zpotrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_sppcon_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, float anorm, float* rcond,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dppcon_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, double anorm, double* rcond,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cppcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap, float anorm,
+                                float* rcond, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zppcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap, double anorm,
+                                double* rcond, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_sppequ_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, float* s, float* scond,
+                                float* amax );
+lapack_int LAPACKE_dppequ_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, double* s, double* scond,
+                                double* amax );
+lapack_int LAPACKE_cppequ_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap, float* s,
+                                float* scond, float* amax );
+lapack_int LAPACKE_zppequ_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap, double* s,
+                                double* scond, double* amax );
+
+lapack_int LAPACKE_spprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* ap,
+                                const float* afp, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dpprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* ap,
+                                const double* afp, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cpprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_complex_float* afp,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zpprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* afp,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sppsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, float* ap, float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dppsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, double* ap, double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_cppsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* ap,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zppsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* ap,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sppsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, float* ap,
+                                float* afp, char* equed, float* s, float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dppsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, double* ap,
+                                double* afp, char* equed, double* s, double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cppsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_float* ap,
+                                lapack_complex_float* afp, char* equed,
+                                float* s, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zppsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_double* ap,
+                                lapack_complex_double* afp, char* equed,
+                                double* s, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_spptrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap );
+lapack_int LAPACKE_dpptrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap );
+lapack_int LAPACKE_cpptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap );
+lapack_int LAPACKE_zpptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap );
+
+lapack_int LAPACKE_spptri_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap );
+lapack_int LAPACKE_dpptri_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap );
+lapack_int LAPACKE_cpptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap );
+lapack_int LAPACKE_zpptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap );
+
+lapack_int LAPACKE_spptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* ap, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dpptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* ap, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_cpptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spstrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* piv,
+                                lapack_int* rank, float tol, float* work );
+lapack_int LAPACKE_dpstrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* piv,
+                                lapack_int* rank, double tol, double* work );
+lapack_int LAPACKE_cpstrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* piv, lapack_int* rank, float tol,
+                                float* work );
+lapack_int LAPACKE_zpstrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* piv, lapack_int* rank, double tol,
+                                double* work );
+
+lapack_int LAPACKE_sptcon_work( lapack_int n, const float* d, const float* e,
+                                float anorm, float* rcond, float* work );
+lapack_int LAPACKE_dptcon_work( lapack_int n, const double* d, const double* e,
+                                double anorm, double* rcond, double* work );
+lapack_int LAPACKE_cptcon_work( lapack_int n, const float* d,
+                                const lapack_complex_float* e, float anorm,
+                                float* rcond, float* work );
+lapack_int LAPACKE_zptcon_work( lapack_int n, const double* d,
+                                const lapack_complex_double* e, double anorm,
+                                double* rcond, double* work );
+
+lapack_int LAPACKE_spteqr_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, float* z, lapack_int ldz,
+                                float* work );
+lapack_int LAPACKE_dpteqr_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, double* z, lapack_int ldz,
+                                double* work );
+lapack_int LAPACKE_cpteqr_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, lapack_complex_float* z,
+                                lapack_int ldz, float* work );
+lapack_int LAPACKE_zpteqr_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, lapack_complex_double* z,
+                                lapack_int ldz, double* work );
+
+lapack_int LAPACKE_sptrfs_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                const float* d, const float* e, const float* df,
+                                const float* ef, const float* b, lapack_int ldb,
+                                float* x, lapack_int ldx, float* ferr,
+                                float* berr, float* work );
+lapack_int LAPACKE_dptrfs_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                const double* d, const double* e,
+                                const double* df, const double* ef,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work );
+lapack_int LAPACKE_cptrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* d,
+                                const lapack_complex_float* e, const float* df,
+                                const lapack_complex_float* ef,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zptrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* d,
+                                const lapack_complex_double* e,
+                                const double* df,
+                                const lapack_complex_double* ef,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sptsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               float* d, float* e, float* b, lapack_int ldb );
+lapack_int LAPACKE_dptsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               double* d, double* e, double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_cptsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               float* d, lapack_complex_float* e,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zptsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               double* d, lapack_complex_double* e,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sptsvx_work( int matrix_order, char fact, lapack_int n,
+                                lapack_int nrhs, const float* d, const float* e,
+                                float* df, float* ef, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work );
+lapack_int LAPACKE_dptsvx_work( int matrix_order, char fact, lapack_int n,
+                                lapack_int nrhs, const double* d,
+                                const double* e, double* df, double* ef,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, double* work );
+lapack_int LAPACKE_cptsvx_work( int matrix_order, char fact, lapack_int n,
+                                lapack_int nrhs, const float* d,
+                                const lapack_complex_float* e, float* df,
+                                lapack_complex_float* ef,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zptsvx_work( int matrix_order, char fact, lapack_int n,
+                                lapack_int nrhs, const double* d,
+                                const lapack_complex_double* e, double* df,
+                                lapack_complex_double* ef,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spttrf_work( lapack_int n, float* d, float* e );
+lapack_int LAPACKE_dpttrf_work( lapack_int n, double* d, double* e );
+lapack_int LAPACKE_cpttrf_work( lapack_int n, float* d,
+                                lapack_complex_float* e );
+lapack_int LAPACKE_zpttrf_work( lapack_int n, double* d,
+                                lapack_complex_double* e );
+
+lapack_int LAPACKE_spttrs_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                const float* d, const float* e, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dpttrs_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                const double* d, const double* e, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_cpttrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* d,
+                                const lapack_complex_float* e,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpttrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* d,
+                                const lapack_complex_double* e,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssbev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int kd, float* ab,
+                               lapack_int ldab, float* w, float* z,
+                               lapack_int ldz, float* work );
+lapack_int LAPACKE_dsbev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int kd, double* ab,
+                               lapack_int ldab, double* w, double* z,
+                               lapack_int ldz, double* work );
+
+lapack_int LAPACKE_ssbevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int kd, float* ab,
+                                lapack_int ldab, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dsbevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int kd, double* ab,
+                                lapack_int ldab, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssbevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int kd,
+                                float* ab, lapack_int ldab, float* q,
+                                lapack_int ldq, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dsbevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int kd,
+                                double* ab, lapack_int ldab, double* q,
+                                lapack_int ldq, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int* iwork,
+                                lapack_int* ifail );
+
+lapack_int LAPACKE_ssbgst_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                float* ab, lapack_int ldab, const float* bb,
+                                lapack_int ldbb, float* x, lapack_int ldx,
+                                float* work );
+lapack_int LAPACKE_dsbgst_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                double* ab, lapack_int ldab, const double* bb,
+                                lapack_int ldbb, double* x, lapack_int ldx,
+                                double* work );
+
+lapack_int LAPACKE_ssbgv_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int ka, lapack_int kb,
+                               float* ab, lapack_int ldab, float* bb,
+                               lapack_int ldbb, float* w, float* z,
+                               lapack_int ldz, float* work );
+lapack_int LAPACKE_dsbgv_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int ka, lapack_int kb,
+                               double* ab, lapack_int ldab, double* bb,
+                               lapack_int ldbb, double* w, double* z,
+                               lapack_int ldz, double* work );
+
+lapack_int LAPACKE_ssbgvd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                float* ab, lapack_int ldab, float* bb,
+                                lapack_int ldbb, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dsbgvd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                double* ab, lapack_int ldab, double* bb,
+                                lapack_int ldbb, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssbgvx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int ka,
+                                lapack_int kb, float* ab, lapack_int ldab,
+                                float* bb, lapack_int ldbb, float* q,
+                                lapack_int ldq, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dsbgvx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int ka,
+                                lapack_int kb, double* ab, lapack_int ldab,
+                                double* bb, lapack_int ldbb, double* q,
+                                lapack_int ldq, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int* iwork,
+                                lapack_int* ifail );
+
+lapack_int LAPACKE_ssbtrd_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int kd, float* ab,
+                                lapack_int ldab, float* d, float* e, float* q,
+                                lapack_int ldq, float* work );
+lapack_int LAPACKE_dsbtrd_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int kd, double* ab,
+                                lapack_int ldab, double* d, double* e,
+                                double* q, lapack_int ldq, double* work );
+
+lapack_int LAPACKE_ssfrk_work( int matrix_order, char transr, char uplo,
+                               char trans, lapack_int n, lapack_int k,
+                               float alpha, const float* a, lapack_int lda,
+                               float beta, float* c );
+lapack_int LAPACKE_dsfrk_work( int matrix_order, char transr, char uplo,
+                               char trans, lapack_int n, lapack_int k,
+                               double alpha, const double* a, lapack_int lda,
+                               double beta, double* c );
+
+lapack_int LAPACKE_sspcon_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, const lapack_int* ipiv,
+                                float anorm, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dspcon_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, const lapack_int* ipiv,
+                                double anorm, double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cspcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zspcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_sspev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, float* ap, float* w, float* z,
+                               lapack_int ldz, float* work );
+lapack_int LAPACKE_dspev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, double* ap, double* w, double* z,
+                               lapack_int ldz, double* work );
+
+lapack_int LAPACKE_sspevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, float* ap, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dspevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, double* ap, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_sspevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, float* ap, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dspevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, double* ap, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                double* z, lapack_int ldz, double* work,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_sspgst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, float* ap, const float* bp );
+lapack_int LAPACKE_dspgst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, double* ap, const double* bp );
+
+lapack_int LAPACKE_sspgv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, float* ap, float* bp,
+                               float* w, float* z, lapack_int ldz,
+                               float* work );
+lapack_int LAPACKE_dspgv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, double* ap, double* bp,
+                               double* w, double* z, lapack_int ldz,
+                               double* work );
+
+lapack_int LAPACKE_sspgvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n, float* ap, float* bp,
+                                float* w, float* z, lapack_int ldz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dspgvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n, double* ap, double* bp,
+                                double* w, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_sspgvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n, float* ap,
+                                float* bp, float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, float* z, lapack_int ldz, float* work,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_dspgvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n, double* ap,
+                                double* bp, double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, double* z, lapack_int ldz,
+                                double* work, lapack_int* iwork,
+                                lapack_int* ifail );
+
+lapack_int LAPACKE_ssprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* ap,
+                                const float* afp, const lapack_int* ipiv,
+                                const float* b, lapack_int ldb, float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dsprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* ap,
+                                const double* afp, const lapack_int* ipiv,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_csprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_complex_float* afp,
+                                const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zsprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* afp,
+                                const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sspsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, float* ap, lapack_int* ipiv,
+                               float* b, lapack_int ldb );
+lapack_int LAPACKE_dspsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, double* ap, lapack_int* ipiv,
+                               double* b, lapack_int ldb );
+lapack_int LAPACKE_cspsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* ap,
+                               lapack_int* ipiv, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zspsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* ap,
+                               lapack_int* ipiv, lapack_complex_double* b,
+                               lapack_int ldb );
+
+lapack_int LAPACKE_sspsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, const float* ap,
+                                float* afp, lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dspsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, const double* ap,
+                                double* afp, lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cspsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* ap,
+                                lapack_complex_float* afp, lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zspsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* afp, lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_ssptrd_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap, float* d, float* e, float* tau );
+lapack_int LAPACKE_dsptrd_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap, double* d, double* e, double* tau );
+
+lapack_int LAPACKE_ssptrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_dsptrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap, lapack_int* ipiv );
+lapack_int LAPACKE_csptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_zsptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap, lapack_int* ipiv );
+
+lapack_int LAPACKE_ssptri_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap, const lapack_int* ipiv,
+                                float* work );
+lapack_int LAPACKE_dsptri_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap, const lapack_int* ipiv,
+                                double* work );
+lapack_int LAPACKE_csptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zsptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_ssptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* ap,
+                                const lapack_int* ipiv, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dsptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* ap,
+                                const lapack_int* ipiv, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_csptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_int* ipiv, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zsptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sstebz_work( char range, char order, lapack_int n, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, const float* d, const float* e,
+                                lapack_int* m, lapack_int* nsplit, float* w,
+                                lapack_int* iblock, lapack_int* isplit,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dstebz_work( char range, char order, lapack_int n, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, const double* d, const double* e,
+                                lapack_int* m, lapack_int* nsplit, double* w,
+                                lapack_int* iblock, lapack_int* isplit,
+                                double* work, lapack_int* iwork );
+
+lapack_int LAPACKE_sstedc_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, float* z, lapack_int ldz,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dstedc_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_cstedc_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zstedc_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_sstegr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w, float* z,
+                                lapack_int ldz, lapack_int* isuppz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dstegr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                double* z, lapack_int ldz, lapack_int* isuppz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_cstegr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_int* isuppz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zstegr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_int* isuppz, double* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_sstein_work( int matrix_order, lapack_int n, const float* d,
+                                const float* e, lapack_int m, const float* w,
+                                const lapack_int* iblock,
+                                const lapack_int* isplit, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifailv );
+lapack_int LAPACKE_dstein_work( int matrix_order, lapack_int n, const double* d,
+                                const double* e, lapack_int m, const double* w,
+                                const lapack_int* iblock,
+                                const lapack_int* isplit, double* z,
+                                lapack_int ldz, double* work, lapack_int* iwork,
+                                lapack_int* ifailv );
+lapack_int LAPACKE_cstein_work( int matrix_order, lapack_int n, const float* d,
+                                const float* e, lapack_int m, const float* w,
+                                const lapack_int* iblock,
+                                const lapack_int* isplit,
+                                lapack_complex_float* z, lapack_int ldz,
+                                float* work, lapack_int* iwork,
+                                lapack_int* ifailv );
+lapack_int LAPACKE_zstein_work( int matrix_order, lapack_int n, const double* d,
+                                const double* e, lapack_int m, const double* w,
+                                const lapack_int* iblock,
+                                const lapack_int* isplit,
+                                lapack_complex_double* z, lapack_int ldz,
+                                double* work, lapack_int* iwork,
+                                lapack_int* ifailv );
+
+lapack_int LAPACKE_sstemr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, lapack_int nzc,
+                                lapack_int* isuppz, lapack_logical* tryrac,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dstemr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, lapack_int nzc,
+                                lapack_int* isuppz, lapack_logical* tryrac,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_cstemr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_int nzc, lapack_int* isuppz,
+                                lapack_logical* tryrac, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zstemr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_int nzc, lapack_int* isuppz,
+                                lapack_logical* tryrac, double* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_ssteqr_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, float* z, lapack_int ldz,
+                                float* work );
+lapack_int LAPACKE_dsteqr_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, double* z, lapack_int ldz,
+                                double* work );
+lapack_int LAPACKE_csteqr_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, lapack_complex_float* z,
+                                lapack_int ldz, float* work );
+lapack_int LAPACKE_zsteqr_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, lapack_complex_double* z,
+                                lapack_int ldz, double* work );
+
+lapack_int LAPACKE_ssterf_work( lapack_int n, float* d, float* e );
+lapack_int LAPACKE_dsterf_work( lapack_int n, double* d, double* e );
+
+lapack_int LAPACKE_sstev_work( int matrix_order, char jobz, lapack_int n,
+                               float* d, float* e, float* z, lapack_int ldz,
+                               float* work );
+lapack_int LAPACKE_dstev_work( int matrix_order, char jobz, lapack_int n,
+                               double* d, double* e, double* z, lapack_int ldz,
+                               double* work );
+
+lapack_int LAPACKE_sstevd_work( int matrix_order, char jobz, lapack_int n,
+                                float* d, float* e, float* z, lapack_int ldz,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dstevd_work( int matrix_order, char jobz, lapack_int n,
+                                double* d, double* e, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_sstevr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w, float* z,
+                                lapack_int ldz, lapack_int* isuppz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dstevr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                double* z, lapack_int ldz, lapack_int* isuppz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_sstevx_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dstevx_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                double* z, lapack_int ldz, double* work,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_ssycon_work( int matrix_order, char uplo, lapack_int n,
+                                const float* a, lapack_int lda,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dsycon_work( int matrix_order, char uplo, lapack_int n,
+                                const double* a, lapack_int lda,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_csycon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zsycon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_ssyequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const float* a, lapack_int lda, float* s,
+                                 float* scond, float* amax, float* work );
+lapack_int LAPACKE_dsyequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const double* a, lapack_int lda, double* s,
+                                 double* scond, double* amax, double* work );
+lapack_int LAPACKE_csyequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 float* s, float* scond, float* amax,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_zsyequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 double* s, double* scond, double* amax,
+                                 lapack_complex_double* work );
+
+lapack_int LAPACKE_ssyev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, float* a, lapack_int lda, float* w,
+                               float* work, lapack_int lwork );
+lapack_int LAPACKE_dsyev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, double* a, lapack_int lda,
+                               double* w, double* work, lapack_int lwork );
+
+lapack_int LAPACKE_ssyevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* w, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dsyevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* w, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssyevr_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, float* a,
+                                lapack_int lda, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, lapack_int* isuppz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dsyevr_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, double* a,
+                                lapack_int lda, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, lapack_int* isuppz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssyevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, float* a,
+                                lapack_int lda, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_dsyevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, double* a,
+                                lapack_int lda, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_ssygst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, float* a, lapack_int lda,
+                                const float* b, lapack_int ldb );
+lapack_int LAPACKE_dsygst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, double* a, lapack_int lda,
+                                const double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssygv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, float* a,
+                               lapack_int lda, float* b, lapack_int ldb,
+                               float* w, float* work, lapack_int lwork );
+lapack_int LAPACKE_dsygv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, double* a,
+                               lapack_int lda, double* b, lapack_int ldb,
+                               double* w, double* work, lapack_int lwork );
+
+lapack_int LAPACKE_ssygvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n, float* a,
+                                lapack_int lda, float* b, lapack_int ldb,
+                                float* w, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dsygvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double* w, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssygvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n, float* a,
+                                lapack_int lda, float* b, lapack_int ldb,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, float* z, lapack_int ldz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dsygvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_ssyrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const float* af, lapack_int ldaf,
+                                const lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dsyrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_csyrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zsyrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_ssyrfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs, const float* a,
+                                 lapack_int lda, const float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* s, const float* b, lapack_int ldb,
+                                 float* x, lapack_int ldx, float* rcond,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dsyrfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs, const double* a,
+                                 lapack_int lda, const double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* s, const double* b,
+                                 lapack_int ldb, double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_csyrfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 const lapack_complex_float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* s, const lapack_complex_float* b,
+                                 lapack_int ldb, lapack_complex_float* x,
+                                 lapack_int ldx, float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zsyrfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 const lapack_complex_double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* s,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_ssysv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, float* a, lapack_int lda,
+                               lapack_int* ipiv, float* b, lapack_int ldb,
+                               float* work, lapack_int lwork );
+lapack_int LAPACKE_dsysv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, double* a, lapack_int lda,
+                               lapack_int* ipiv, double* b, lapack_int ldb,
+                               double* work, lapack_int lwork );
+lapack_int LAPACKE_csysv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* a,
+                               lapack_int lda, lapack_int* ipiv,
+                               lapack_complex_float* b, lapack_int ldb,
+                               lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zsysv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* a,
+                               lapack_int lda, lapack_int* ipiv,
+                               lapack_complex_double* b, lapack_int ldb,
+                               lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_ssysvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, const float* a,
+                                lapack_int lda, float* af, lapack_int ldaf,
+                                lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dsysvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, const double* a,
+                                lapack_int lda, double* af, lapack_int ldaf,
+                                lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_csysvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* af, lapack_int ldaf,
+                                lapack_int* ipiv, const lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zsysvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* af, lapack_int ldaf,
+                                lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_ssysvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs, float* a,
+                                 lapack_int lda, float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* s,
+                                 float* b, lapack_int ldb, float* x,
+                                 lapack_int ldx, float* rcond, float* rpvgrw,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dsysvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs, double* a,
+                                 lapack_int lda, double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* s,
+                                 double* b, lapack_int ldb, double* x,
+                                 lapack_int ldx, double* rcond, double* rpvgrw,
+                                 double* berr, lapack_int n_err_bnds,
+                                 double* err_bnds_norm, double* err_bnds_comp,
+                                 lapack_int nparams, double* params,
+                                 double* work, lapack_int* iwork );
+lapack_int LAPACKE_csysvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* s,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zsysvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* s,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_ssytrd_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda, float* d, float* e,
+                                float* tau, float* work, lapack_int lwork );
+lapack_int LAPACKE_dsytrd_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda, double* d, double* e,
+                                double* tau, double* work, lapack_int lwork );
+
+lapack_int LAPACKE_ssytrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* ipiv,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dsytrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* ipiv,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_csytrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_zsytrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_ssytri_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda,
+                                const lapack_int* ipiv, float* work );
+lapack_int LAPACKE_dsytri_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda,
+                                const lapack_int* ipiv, double* work );
+lapack_int LAPACKE_csytri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zsytri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_ssytrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const lapack_int* ipiv, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dsytrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_csytrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zsytrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stbcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, lapack_int kd,
+                                const float* ab, lapack_int ldab, float* rcond,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dtbcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, lapack_int kd,
+                                const double* ab, lapack_int ldab,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctbcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, lapack_int kd,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                float* rcond, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_ztbcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, lapack_int kd,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stbrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const float* ab,
+                                lapack_int ldab, const float* b, lapack_int ldb,
+                                const float* x, lapack_int ldx, float* ferr,
+                                float* berr, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dtbrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const double* ab,
+                                lapack_int ldab, const double* b,
+                                lapack_int ldb, const double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctbrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const lapack_complex_float* ab,
+                                lapack_int ldab, const lapack_complex_float* b,
+                                lapack_int ldb, const lapack_complex_float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztbrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, const lapack_complex_double* b,
+                                lapack_int ldb, const lapack_complex_double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stbtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const float* ab,
+                                lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dtbtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const double* ab,
+                                lapack_int ldab, double* b, lapack_int ldb );
+lapack_int LAPACKE_ctbtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const lapack_complex_float* ab,
+                                lapack_int ldab, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_ztbtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_stfsm_work( int matrix_order, char transr, char side,
+                               char uplo, char trans, char diag, lapack_int m,
+                               lapack_int n, float alpha, const float* a,
+                               float* b, lapack_int ldb );
+lapack_int LAPACKE_dtfsm_work( int matrix_order, char transr, char side,
+                               char uplo, char trans, char diag, lapack_int m,
+                               lapack_int n, double alpha, const double* a,
+                               double* b, lapack_int ldb );
+lapack_int LAPACKE_ctfsm_work( int matrix_order, char transr, char side,
+                               char uplo, char trans, char diag, lapack_int m,
+                               lapack_int n, lapack_complex_float alpha,
+                               const lapack_complex_float* a,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztfsm_work( int matrix_order, char transr, char side,
+                               char uplo, char trans, char diag, lapack_int m,
+                               lapack_int n, lapack_complex_double alpha,
+                               const lapack_complex_double* a,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stftri_work( int matrix_order, char transr, char uplo,
+                                char diag, lapack_int n, float* a );
+lapack_int LAPACKE_dtftri_work( int matrix_order, char transr, char uplo,
+                                char diag, lapack_int n, double* a );
+lapack_int LAPACKE_ctftri_work( int matrix_order, char transr, char uplo,
+                                char diag, lapack_int n,
+                                lapack_complex_float* a );
+lapack_int LAPACKE_ztftri_work( int matrix_order, char transr, char uplo,
+                                char diag, lapack_int n,
+                                lapack_complex_double* a );
+
+lapack_int LAPACKE_stfttp_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const float* arf, float* ap );
+lapack_int LAPACKE_dtfttp_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const double* arf, double* ap );
+lapack_int LAPACKE_ctfttp_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_float* arf,
+                                lapack_complex_float* ap );
+lapack_int LAPACKE_ztfttp_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_double* arf,
+                                lapack_complex_double* ap );
+
+lapack_int LAPACKE_stfttr_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const float* arf, float* a,
+                                lapack_int lda );
+lapack_int LAPACKE_dtfttr_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const double* arf, double* a,
+                                lapack_int lda );
+lapack_int LAPACKE_ctfttr_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_float* arf,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztfttr_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_double* arf,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_stgevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const float* s, lapack_int lds, const float* p,
+                                lapack_int ldp, float* vl, lapack_int ldvl,
+                                float* vr, lapack_int ldvr, lapack_int mm,
+                                lapack_int* m, float* work );
+lapack_int LAPACKE_dtgevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const double* s, lapack_int lds,
+                                const double* p, lapack_int ldp, double* vl,
+                                lapack_int ldvl, double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, double* work );
+lapack_int LAPACKE_ctgevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_float* s, lapack_int lds,
+                                const lapack_complex_float* p, lapack_int ldp,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztgevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_double* s, lapack_int lds,
+                                const lapack_complex_double* p, lapack_int ldp,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stgexc_work( int matrix_order, lapack_logical wantq,
+                                lapack_logical wantz, lapack_int n, float* a,
+                                lapack_int lda, float* b, lapack_int ldb,
+                                float* q, lapack_int ldq, float* z,
+                                lapack_int ldz, lapack_int* ifst,
+                                lapack_int* ilst, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dtgexc_work( int matrix_order, lapack_logical wantq,
+                                lapack_logical wantz, lapack_int n, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double* q, lapack_int ldq, double* z,
+                                lapack_int ldz, lapack_int* ifst,
+                                lapack_int* ilst, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_ctgexc_work( int matrix_order, lapack_logical wantq,
+                                lapack_logical wantz, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_int ifst, lapack_int ilst );
+lapack_int LAPACKE_ztgexc_work( int matrix_order, lapack_logical wantq,
+                                lapack_logical wantz, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_int ifst, lapack_int ilst );
+
+lapack_int LAPACKE_stgsen_work( int matrix_order, lapack_int ijob,
+                                lapack_logical wantq, lapack_logical wantz,
+                                const lapack_logical* select, lapack_int n,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, float* alphar, float* alphai,
+                                float* beta, float* q, lapack_int ldq, float* z,
+                                lapack_int ldz, lapack_int* m, float* pl,
+                                float* pr, float* dif, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dtgsen_work( int matrix_order, lapack_int ijob,
+                                lapack_logical wantq, lapack_logical wantz,
+                                const lapack_logical* select, lapack_int n,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, double* alphar, double* alphai,
+                                double* beta, double* q, lapack_int ldq,
+                                double* z, lapack_int ldz, lapack_int* m,
+                                double* pl, double* pr, double* dif,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_ctgsen_work( int matrix_order, lapack_int ijob,
+                                lapack_logical wantq, lapack_logical wantz,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* alpha,
+                                lapack_complex_float* beta,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_int* m, float* pl, float* pr, float* dif,
+                                lapack_complex_float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_ztgsen_work( int matrix_order, lapack_int ijob,
+                                lapack_logical wantq, lapack_logical wantz,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* alpha,
+                                lapack_complex_double* beta,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_int* m, double* pl, double* pr,
+                                double* dif, lapack_complex_double* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_stgsja_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, float tola, float tolb,
+                                float* alpha, float* beta, float* u,
+                                lapack_int ldu, float* v, lapack_int ldv,
+                                float* q, lapack_int ldq, float* work,
+                                lapack_int* ncycle );
+lapack_int LAPACKE_dtgsja_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, double tola, double tolb,
+                                double* alpha, double* beta, double* u,
+                                lapack_int ldu, double* v, lapack_int ldv,
+                                double* q, lapack_int ldq, double* work,
+                                lapack_int* ncycle );
+lapack_int LAPACKE_ctgsja_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                float tola, float tolb, float* alpha,
+                                float* beta, lapack_complex_float* u,
+                                lapack_int ldu, lapack_complex_float* v,
+                                lapack_int ldv, lapack_complex_float* q,
+                                lapack_int ldq, lapack_complex_float* work,
+                                lapack_int* ncycle );
+lapack_int LAPACKE_ztgsja_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                double tola, double tolb, double* alpha,
+                                double* beta, lapack_complex_double* u,
+                                lapack_int ldu, lapack_complex_double* v,
+                                lapack_int ldv, lapack_complex_double* q,
+                                lapack_int ldq, lapack_complex_double* work,
+                                lapack_int* ncycle );
+
+lapack_int LAPACKE_stgsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const float* a, lapack_int lda, const float* b,
+                                lapack_int ldb, const float* vl,
+                                lapack_int ldvl, const float* vr,
+                                lapack_int ldvr, float* s, float* dif,
+                                lapack_int mm, lapack_int* m, float* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_dtgsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const double* a, lapack_int lda,
+                                const double* b, lapack_int ldb,
+                                const double* vl, lapack_int ldvl,
+                                const double* vr, lapack_int ldvr, double* s,
+                                double* dif, lapack_int mm, lapack_int* m,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctgsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                const lapack_complex_float* vl, lapack_int ldvl,
+                                const lapack_complex_float* vr, lapack_int ldvr,
+                                float* s, float* dif, lapack_int mm,
+                                lapack_int* m, lapack_complex_float* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_ztgsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                const lapack_complex_double* vl,
+                                lapack_int ldvl,
+                                const lapack_complex_double* vr,
+                                lapack_int ldvr, double* s, double* dif,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, lapack_int lwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_stgsyl_work( int matrix_order, char trans, lapack_int ijob,
+                                lapack_int m, lapack_int n, const float* a,
+                                lapack_int lda, const float* b, lapack_int ldb,
+                                float* c, lapack_int ldc, const float* d,
+                                lapack_int ldd, const float* e, lapack_int lde,
+                                float* f, lapack_int ldf, float* scale,
+                                float* dif, float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dtgsyl_work( int matrix_order, char trans, lapack_int ijob,
+                                lapack_int m, lapack_int n, const double* a,
+                                lapack_int lda, const double* b, lapack_int ldb,
+                                double* c, lapack_int ldc, const double* d,
+                                lapack_int ldd, const double* e, lapack_int lde,
+                                double* f, lapack_int ldf, double* scale,
+                                double* dif, double* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctgsyl_work( int matrix_order, char trans, lapack_int ijob,
+                                lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* c, lapack_int ldc,
+                                const lapack_complex_float* d, lapack_int ldd,
+                                const lapack_complex_float* e, lapack_int lde,
+                                lapack_complex_float* f, lapack_int ldf,
+                                float* scale, float* dif,
+                                lapack_complex_float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ztgsyl_work( int matrix_order, char trans, lapack_int ijob,
+                                lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* c, lapack_int ldc,
+                                const lapack_complex_double* d, lapack_int ldd,
+                                const lapack_complex_double* e, lapack_int lde,
+                                lapack_complex_double* f, lapack_int ldf,
+                                double* scale, double* dif,
+                                lapack_complex_double* work, lapack_int lwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_stpcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, const float* ap,
+                                float* rcond, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dtpcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, const double* ap,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctpcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n,
+                                const lapack_complex_float* ap, float* rcond,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztpcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n,
+                                const lapack_complex_double* ap, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stprfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const float* ap, const float* b, lapack_int ldb,
+                                const float* x, lapack_int ldx, float* ferr,
+                                float* berr, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dtprfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const double* ap, const double* b,
+                                lapack_int ldb, const double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctprfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* ap,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                const lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztprfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                const lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stptri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, float* ap );
+lapack_int LAPACKE_dtptri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, double* ap );
+lapack_int LAPACKE_ctptri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, lapack_complex_float* ap );
+lapack_int LAPACKE_ztptri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, lapack_complex_double* ap );
+
+lapack_int LAPACKE_stptrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const float* ap, float* b, lapack_int ldb );
+lapack_int LAPACKE_dtptrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const double* ap, double* b, lapack_int ldb );
+lapack_int LAPACKE_ctptrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* ap,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztptrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stpttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const float* ap, float* arf );
+lapack_int LAPACKE_dtpttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const double* ap, double* arf );
+lapack_int LAPACKE_ctpttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_float* ap,
+                                lapack_complex_float* arf );
+lapack_int LAPACKE_ztpttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_double* ap,
+                                lapack_complex_double* arf );
+
+lapack_int LAPACKE_stpttr_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, float* a, lapack_int lda );
+lapack_int LAPACKE_dtpttr_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, double* a, lapack_int lda );
+lapack_int LAPACKE_ctpttr_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztpttr_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_strcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, const float* a,
+                                lapack_int lda, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dtrcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, const double* a,
+                                lapack_int lda, double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctrcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float* rcond, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_ztrcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double* rcond, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_strevc_work( int matrix_order, char side, char howmny,
+                                lapack_logical* select, lapack_int n,
+                                const float* t, lapack_int ldt, float* vl,
+                                lapack_int ldvl, float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, float* work );
+lapack_int LAPACKE_dtrevc_work( int matrix_order, char side, char howmny,
+                                lapack_logical* select, lapack_int n,
+                                const double* t, lapack_int ldt, double* vl,
+                                lapack_int ldvl, double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, double* work );
+lapack_int LAPACKE_ctrevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztrevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_strexc_work( int matrix_order, char compq, lapack_int n,
+                                float* t, lapack_int ldt, float* q,
+                                lapack_int ldq, lapack_int* ifst,
+                                lapack_int* ilst, float* work );
+lapack_int LAPACKE_dtrexc_work( int matrix_order, char compq, lapack_int n,
+                                double* t, lapack_int ldt, double* q,
+                                lapack_int ldq, lapack_int* ifst,
+                                lapack_int* ilst, double* work );
+lapack_int LAPACKE_ctrexc_work( int matrix_order, char compq, lapack_int n,
+                                lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_int ifst, lapack_int ilst );
+lapack_int LAPACKE_ztrexc_work( int matrix_order, char compq, lapack_int n,
+                                lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_int ifst, lapack_int ilst );
+
+lapack_int LAPACKE_strrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const float* a, lapack_int lda, const float* b,
+                                lapack_int ldb, const float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dtrrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const double* a, lapack_int lda,
+                                const double* b, lapack_int ldb,
+                                const double* x, lapack_int ldx, double* ferr,
+                                double* berr, double* work, lapack_int* iwork );
+lapack_int LAPACKE_ctrrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                const lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztrrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                const lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_strsen_work( int matrix_order, char job, char compq,
+                                const lapack_logical* select, lapack_int n,
+                                float* t, lapack_int ldt, float* q,
+                                lapack_int ldq, float* wr, float* wi,
+                                lapack_int* m, float* s, float* sep,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dtrsen_work( int matrix_order, char job, char compq,
+                                const lapack_logical* select, lapack_int n,
+                                double* t, lapack_int ldt, double* q,
+                                lapack_int ldq, double* wr, double* wi,
+                                lapack_int* m, double* s, double* sep,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_ctrsen_work( int matrix_order, char job, char compq,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* w, lapack_int* m,
+                                float* s, float* sep,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_ztrsen_work( int matrix_order, char job, char compq,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* w, lapack_int* m,
+                                double* s, double* sep,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_strsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const float* t, lapack_int ldt, const float* vl,
+                                lapack_int ldvl, const float* vr,
+                                lapack_int ldvr, float* s, float* sep,
+                                lapack_int mm, lapack_int* m, float* work,
+                                lapack_int ldwork, lapack_int* iwork );
+lapack_int LAPACKE_dtrsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const double* t, lapack_int ldt,
+                                const double* vl, lapack_int ldvl,
+                                const double* vr, lapack_int ldvr, double* s,
+                                double* sep, lapack_int mm, lapack_int* m,
+                                double* work, lapack_int ldwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctrsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_float* t, lapack_int ldt,
+                                const lapack_complex_float* vl, lapack_int ldvl,
+                                const lapack_complex_float* vr, lapack_int ldvr,
+                                float* s, float* sep, lapack_int mm,
+                                lapack_int* m, lapack_complex_float* work,
+                                lapack_int ldwork, float* rwork );
+lapack_int LAPACKE_ztrsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_double* t, lapack_int ldt,
+                                const lapack_complex_double* vl,
+                                lapack_int ldvl,
+                                const lapack_complex_double* vr,
+                                lapack_int ldvr, double* s, double* sep,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, lapack_int ldwork,
+                                double* rwork );
+
+lapack_int LAPACKE_strsyl_work( int matrix_order, char trana, char tranb,
+                                lapack_int isgn, lapack_int m, lapack_int n,
+                                const float* a, lapack_int lda, const float* b,
+                                lapack_int ldb, float* c, lapack_int ldc,
+                                float* scale );
+lapack_int LAPACKE_dtrsyl_work( int matrix_order, char trana, char tranb,
+                                lapack_int isgn, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda,
+                                const double* b, lapack_int ldb, double* c,
+                                lapack_int ldc, double* scale );
+lapack_int LAPACKE_ctrsyl_work( int matrix_order, char trana, char tranb,
+                                lapack_int isgn, lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* c, lapack_int ldc,
+                                float* scale );
+lapack_int LAPACKE_ztrsyl_work( int matrix_order, char trana, char tranb,
+                                lapack_int isgn, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* c, lapack_int ldc,
+                                double* scale );
+
+lapack_int LAPACKE_strtri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, float* a, lapack_int lda );
+lapack_int LAPACKE_dtrtri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, double* a, lapack_int lda );
+lapack_int LAPACKE_ctrtri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda );
+lapack_int LAPACKE_ztrtri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda );
+
+lapack_int LAPACKE_strtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const float* a, lapack_int lda, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dtrtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const double* a, lapack_int lda, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_ctrtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztrtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_strttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const float* a, lapack_int lda,
+                                float* arf );
+lapack_int LAPACKE_dtrttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const double* a, lapack_int lda,
+                                double* arf );
+lapack_int LAPACKE_ctrttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* arf );
+lapack_int LAPACKE_ztrttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* arf );
+
+lapack_int LAPACKE_strttp_work( int matrix_order, char uplo, lapack_int n,
+                                const float* a, lapack_int lda, float* ap );
+lapack_int LAPACKE_dtrttp_work( int matrix_order, char uplo, lapack_int n,
+                                const double* a, lapack_int lda, double* ap );
+lapack_int LAPACKE_ctrttp_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* ap );
+lapack_int LAPACKE_ztrttp_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* ap );
+
+lapack_int LAPACKE_stzrzf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dtzrzf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_ctzrzf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_ztzrzf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungbr_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int k,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungbr_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int k,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunghr_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunghr_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunglq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunglq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungql_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungql_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungqr_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungqr_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungrq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungrq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungtr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungtr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmbr_work( int matrix_order, char vect, char side,
+                                char trans, lapack_int m, lapack_int n,
+                                lapack_int k, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmbr_work( int matrix_order, char vect, char side,
+                                char trans, lapack_int m, lapack_int n,
+                                lapack_int k, const lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmhr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmhr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, const lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmlq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmlq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmql_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmql_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmqr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmqr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmrq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmrq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmrz_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                lapack_int l, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmrz_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                lapack_int l, const lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cupgtr_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zupgtr_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_cupmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const lapack_complex_float* ap,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zupmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_claghe( int matrix_order, lapack_int n, lapack_int k,
+                           const float* d, lapack_complex_float* a,
+                           lapack_int lda, lapack_int* iseed );
+lapack_int LAPACKE_zlaghe( int matrix_order, lapack_int n, lapack_int k,
+                           const double* d, lapack_complex_double* a,
+                           lapack_int lda, lapack_int* iseed );
+
+lapack_int LAPACKE_slagsy( int matrix_order, lapack_int n, lapack_int k,
+                           const float* d, float* a, lapack_int lda,
+                           lapack_int* iseed );
+lapack_int LAPACKE_dlagsy( int matrix_order, lapack_int n, lapack_int k,
+                           const double* d, double* a, lapack_int lda,
+                           lapack_int* iseed );
+lapack_int LAPACKE_clagsy( int matrix_order, lapack_int n, lapack_int k,
+                           const float* d, lapack_complex_float* a,
+                           lapack_int lda, lapack_int* iseed );
+lapack_int LAPACKE_zlagsy( int matrix_order, lapack_int n, lapack_int k,
+                           const double* d, lapack_complex_double* a,
+                           lapack_int lda, lapack_int* iseed );
+
+lapack_int LAPACKE_slapmr( int matrix_order, lapack_logical forwrd,
+                           lapack_int m, lapack_int n, float* x, lapack_int ldx,
+                           lapack_int* k );
+lapack_int LAPACKE_dlapmr( int matrix_order, lapack_logical forwrd,
+                           lapack_int m, lapack_int n, double* x,
+                           lapack_int ldx, lapack_int* k );
+lapack_int LAPACKE_clapmr( int matrix_order, lapack_logical forwrd,
+                           lapack_int m, lapack_int n, lapack_complex_float* x,
+                           lapack_int ldx, lapack_int* k );
+lapack_int LAPACKE_zlapmr( int matrix_order, lapack_logical forwrd,
+                           lapack_int m, lapack_int n, lapack_complex_double* x,
+                           lapack_int ldx, lapack_int* k );
+
+
+float LAPACKE_slapy2( float x, float y );
+double LAPACKE_dlapy2( double x, double y );
+
+float LAPACKE_slapy3( float x, float y, float z );
+double LAPACKE_dlapy3( double x, double y, double z );
+
+lapack_int LAPACKE_slartgp( float f, float g, float* cs, float* sn, float* r );
+lapack_int LAPACKE_dlartgp( double f, double g, double* cs, double* sn,
+                            double* r );
+
+lapack_int LAPACKE_slartgs( float x, float y, float sigma, float* cs,
+                            float* sn );
+lapack_int LAPACKE_dlartgs( double x, double y, double sigma, double* cs,
+                            double* sn );
+
+
+//LAPACK 3.3.0
+lapack_int LAPACKE_cbbcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, lapack_int m,
+                           lapack_int p, lapack_int q, float* theta, float* phi,
+                           lapack_complex_float* u1, lapack_int ldu1,
+                           lapack_complex_float* u2, lapack_int ldu2,
+                           lapack_complex_float* v1t, lapack_int ldv1t,
+                           lapack_complex_float* v2t, lapack_int ldv2t,
+                           float* b11d, float* b11e, float* b12d, float* b12e,
+                           float* b21d, float* b21e, float* b22d, float* b22e );
+lapack_int LAPACKE_cbbcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                float* theta, float* phi,
+                                lapack_complex_float* u1, lapack_int ldu1,
+                                lapack_complex_float* u2, lapack_int ldu2,
+                                lapack_complex_float* v1t, lapack_int ldv1t,
+                                lapack_complex_float* v2t, lapack_int ldv2t,
+                                float* b11d, float* b11e, float* b12d,
+                                float* b12e, float* b21d, float* b21e,
+                                float* b22d, float* b22e, float* rwork,
+                                lapack_int lrwork );
+lapack_int LAPACKE_cheswapr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float* a, lapack_int i1,
+                             lapack_int i2 );
+lapack_int LAPACKE_cheswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float* a, lapack_int i1,
+                                  lapack_int i2 );
+lapack_int LAPACKE_chetri2( int matrix_order, char uplo, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_chetri2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_chetri2x( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float* a, lapack_int lda,
+                             const lapack_int* ipiv, lapack_int nb );
+lapack_int LAPACKE_chetri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float* a, lapack_int lda,
+                                  const lapack_int* ipiv,
+                                  lapack_complex_float* work, lapack_int nb );
+lapack_int LAPACKE_chetrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const lapack_complex_float* a,
+                            lapack_int lda, const lapack_int* ipiv,
+                            lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_chetrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const lapack_complex_float* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_csyconv( int matrix_order, char uplo, char way, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_csyconv_work( int matrix_order, char uplo, char way,
+                                 lapack_int n, lapack_complex_float* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_csyswapr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float* a, lapack_int i1,
+                             lapack_int i2 );
+lapack_int LAPACKE_csyswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float* a, lapack_int i1,
+                                  lapack_int i2 );
+lapack_int LAPACKE_csytri2( int matrix_order, char uplo, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_csytri2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_csytri2x( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float* a, lapack_int lda,
+                             const lapack_int* ipiv, lapack_int nb );
+lapack_int LAPACKE_csytri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float* a, lapack_int lda,
+                                  const lapack_int* ipiv,
+                                  lapack_complex_float* work, lapack_int nb );
+lapack_int LAPACKE_csytrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const lapack_complex_float* a,
+                            lapack_int lda, const lapack_int* ipiv,
+                            lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_csytrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const lapack_complex_float* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_cunbdb( int matrix_order, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           lapack_complex_float* x11, lapack_int ldx11,
+                           lapack_complex_float* x12, lapack_int ldx12,
+                           lapack_complex_float* x21, lapack_int ldx21,
+                           lapack_complex_float* x22, lapack_int ldx22,
+                           float* theta, float* phi,
+                           lapack_complex_float* taup1,
+                           lapack_complex_float* taup2,
+                           lapack_complex_float* tauq1,
+                           lapack_complex_float* tauq2 );
+lapack_int LAPACKE_cunbdb_work( int matrix_order, char trans, char signs,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                lapack_complex_float* x11, lapack_int ldx11,
+                                lapack_complex_float* x12, lapack_int ldx12,
+                                lapack_complex_float* x21, lapack_int ldx21,
+                                lapack_complex_float* x22, lapack_int ldx22,
+                                float* theta, float* phi,
+                                lapack_complex_float* taup1,
+                                lapack_complex_float* taup2,
+                                lapack_complex_float* tauq1,
+                                lapack_complex_float* tauq2,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_cuncsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           lapack_complex_float* x11, lapack_int ldx11,
+                           lapack_complex_float* x12, lapack_int ldx12,
+                           lapack_complex_float* x21, lapack_int ldx21,
+                           lapack_complex_float* x22, lapack_int ldx22,
+                           float* theta, lapack_complex_float* u1,
+                           lapack_int ldu1, lapack_complex_float* u2,
+                           lapack_int ldu2, lapack_complex_float* v1t,
+                           lapack_int ldv1t, lapack_complex_float* v2t,
+                           lapack_int ldv2t );
+lapack_int LAPACKE_cuncsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                char signs, lapack_int m, lapack_int p,
+                                lapack_int q, lapack_complex_float* x11,
+                                lapack_int ldx11, lapack_complex_float* x12,
+                                lapack_int ldx12, lapack_complex_float* x21,
+                                lapack_int ldx21, lapack_complex_float* x22,
+                                lapack_int ldx22, float* theta,
+                                lapack_complex_float* u1, lapack_int ldu1,
+                                lapack_complex_float* u2, lapack_int ldu2,
+                                lapack_complex_float* v1t, lapack_int ldv1t,
+                                lapack_complex_float* v2t, lapack_int ldv2t,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int lrwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dbbcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, lapack_int m,
+                           lapack_int p, lapack_int q, double* theta,
+                           double* phi, double* u1, lapack_int ldu1, double* u2,
+                           lapack_int ldu2, double* v1t, lapack_int ldv1t,
+                           double* v2t, lapack_int ldv2t, double* b11d,
+                           double* b11e, double* b12d, double* b12e,
+                           double* b21d, double* b21e, double* b22d,
+                           double* b22e );
+lapack_int LAPACKE_dbbcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                double* theta, double* phi, double* u1,
+                                lapack_int ldu1, double* u2, lapack_int ldu2,
+                                double* v1t, lapack_int ldv1t, double* v2t,
+                                lapack_int ldv2t, double* b11d, double* b11e,
+                                double* b12d, double* b12e, double* b21d,
+                                double* b21e, double* b22d, double* b22e,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_dorbdb( int matrix_order, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           double* x11, lapack_int ldx11, double* x12,
+                           lapack_int ldx12, double* x21, lapack_int ldx21,
+                           double* x22, lapack_int ldx22, double* theta,
+                           double* phi, double* taup1, double* taup2,
+                           double* tauq1, double* tauq2 );
+lapack_int LAPACKE_dorbdb_work( int matrix_order, char trans, char signs,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                double* x11, lapack_int ldx11, double* x12,
+                                lapack_int ldx12, double* x21, lapack_int ldx21,
+                                double* x22, lapack_int ldx22, double* theta,
+                                double* phi, double* taup1, double* taup2,
+                                double* tauq1, double* tauq2, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           double* x11, lapack_int ldx11, double* x12,
+                           lapack_int ldx12, double* x21, lapack_int ldx21,
+                           double* x22, lapack_int ldx22, double* theta,
+                           double* u1, lapack_int ldu1, double* u2,
+                           lapack_int ldu2, double* v1t, lapack_int ldv1t,
+                           double* v2t, lapack_int ldv2t );
+lapack_int LAPACKE_dorcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                char signs, lapack_int m, lapack_int p,
+                                lapack_int q, double* x11, lapack_int ldx11,
+                                double* x12, lapack_int ldx12, double* x21,
+                                lapack_int ldx21, double* x22, lapack_int ldx22,
+                                double* theta, double* u1, lapack_int ldu1,
+                                double* u2, lapack_int ldu2, double* v1t,
+                                lapack_int ldv1t, double* v2t, lapack_int ldv2t,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dsyconv( int matrix_order, char uplo, char way, lapack_int n,
+                            double* a, lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_dsyconv_work( int matrix_order, char uplo, char way,
+                                 lapack_int n, double* a, lapack_int lda,
+                                 const lapack_int* ipiv, double* work );
+lapack_int LAPACKE_dsyswapr( int matrix_order, char uplo, lapack_int n,
+                             double* a, lapack_int i1, lapack_int i2 );
+lapack_int LAPACKE_dsyswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  double* a, lapack_int i1, lapack_int i2 );
+lapack_int LAPACKE_dsytri2( int matrix_order, char uplo, lapack_int n,
+                            double* a, lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_dsytri2_work( int matrix_order, char uplo, lapack_int n,
+                                 double* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_double* work, lapack_int lwork );
+lapack_int LAPACKE_dsytri2x( int matrix_order, char uplo, lapack_int n,
+                             double* a, lapack_int lda, const lapack_int* ipiv,
+                             lapack_int nb );
+lapack_int LAPACKE_dsytri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  double* a, lapack_int lda,
+                                  const lapack_int* ipiv, double* work,
+                                  lapack_int nb );
+lapack_int LAPACKE_dsytrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const double* a, lapack_int lda,
+                            const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_dsytrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const double* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 double* b, lapack_int ldb, double* work );
+lapack_int LAPACKE_sbbcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, lapack_int m,
+                           lapack_int p, lapack_int q, float* theta, float* phi,
+                           float* u1, lapack_int ldu1, float* u2,
+                           lapack_int ldu2, float* v1t, lapack_int ldv1t,
+                           float* v2t, lapack_int ldv2t, float* b11d,
+                           float* b11e, float* b12d, float* b12e, float* b21d,
+                           float* b21e, float* b22d, float* b22e );
+lapack_int LAPACKE_sbbcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                float* theta, float* phi, float* u1,
+                                lapack_int ldu1, float* u2, lapack_int ldu2,
+                                float* v1t, lapack_int ldv1t, float* v2t,
+                                lapack_int ldv2t, float* b11d, float* b11e,
+                                float* b12d, float* b12e, float* b21d,
+                                float* b21e, float* b22d, float* b22e,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_sorbdb( int matrix_order, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q, float* x11,
+                           lapack_int ldx11, float* x12, lapack_int ldx12,
+                           float* x21, lapack_int ldx21, float* x22,
+                           lapack_int ldx22, float* theta, float* phi,
+                           float* taup1, float* taup2, float* tauq1,
+                           float* tauq2 );
+lapack_int LAPACKE_sorbdb_work( int matrix_order, char trans, char signs,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                float* x11, lapack_int ldx11, float* x12,
+                                lapack_int ldx12, float* x21, lapack_int ldx21,
+                                float* x22, lapack_int ldx22, float* theta,
+                                float* phi, float* taup1, float* taup2,
+                                float* tauq1, float* tauq2, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_sorcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q, float* x11,
+                           lapack_int ldx11, float* x12, lapack_int ldx12,
+                           float* x21, lapack_int ldx21, float* x22,
+                           lapack_int ldx22, float* theta, float* u1,
+                           lapack_int ldu1, float* u2, lapack_int ldu2,
+                           float* v1t, lapack_int ldv1t, float* v2t,
+                           lapack_int ldv2t );
+lapack_int LAPACKE_sorcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                char signs, lapack_int m, lapack_int p,
+                                lapack_int q, float* x11, lapack_int ldx11,
+                                float* x12, lapack_int ldx12, float* x21,
+                                lapack_int ldx21, float* x22, lapack_int ldx22,
+                                float* theta, float* u1, lapack_int ldu1,
+                                float* u2, lapack_int ldu2, float* v1t,
+                                lapack_int ldv1t, float* v2t, lapack_int ldv2t,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ssyconv( int matrix_order, char uplo, char way, lapack_int n,
+                            float* a, lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_ssyconv_work( int matrix_order, char uplo, char way,
+                                 lapack_int n, float* a, lapack_int lda,
+                                 const lapack_int* ipiv, float* work );
+lapack_int LAPACKE_ssyswapr( int matrix_order, char uplo, lapack_int n,
+                             float* a, lapack_int i1, lapack_int i2 );
+lapack_int LAPACKE_ssyswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  float* a, lapack_int i1, lapack_int i2 );
+lapack_int LAPACKE_ssytri2( int matrix_order, char uplo, lapack_int n, float* a,
+                            lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_ssytri2_work( int matrix_order, char uplo, lapack_int n,
+                                 float* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_ssytri2x( int matrix_order, char uplo, lapack_int n,
+                             float* a, lapack_int lda, const lapack_int* ipiv,
+                             lapack_int nb );
+lapack_int LAPACKE_ssytri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  float* a, lapack_int lda,
+                                  const lapack_int* ipiv, float* work,
+                                  lapack_int nb );
+lapack_int LAPACKE_ssytrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const float* a, lapack_int lda,
+                            const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_ssytrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const float* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 float* b, lapack_int ldb, float* work );
+lapack_int LAPACKE_zbbcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, lapack_int m,
+                           lapack_int p, lapack_int q, double* theta,
+                           double* phi, lapack_complex_double* u1,
+                           lapack_int ldu1, lapack_complex_double* u2,
+                           lapack_int ldu2, lapack_complex_double* v1t,
+                           lapack_int ldv1t, lapack_complex_double* v2t,
+                           lapack_int ldv2t, double* b11d, double* b11e,
+                           double* b12d, double* b12e, double* b21d,
+                           double* b21e, double* b22d, double* b22e );
+lapack_int LAPACKE_zbbcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                double* theta, double* phi,
+                                lapack_complex_double* u1, lapack_int ldu1,
+                                lapack_complex_double* u2, lapack_int ldu2,
+                                lapack_complex_double* v1t, lapack_int ldv1t,
+                                lapack_complex_double* v2t, lapack_int ldv2t,
+                                double* b11d, double* b11e, double* b12d,
+                                double* b12e, double* b21d, double* b21e,
+                                double* b22d, double* b22e, double* rwork,
+                                lapack_int lrwork );
+lapack_int LAPACKE_zheswapr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double* a, lapack_int i1,
+                             lapack_int i2 );
+lapack_int LAPACKE_zheswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double* a, lapack_int i1,
+                                  lapack_int i2 );
+lapack_int LAPACKE_zhetri2( int matrix_order, char uplo, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_zhetri2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_double* work, lapack_int lwork );
+lapack_int LAPACKE_zhetri2x( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double* a, lapack_int lda,
+                             const lapack_int* ipiv, lapack_int nb );
+lapack_int LAPACKE_zhetri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double* a, lapack_int lda,
+                                  const lapack_int* ipiv,
+                                  lapack_complex_double* work, lapack_int nb );
+lapack_int LAPACKE_zhetrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const lapack_complex_double* a,
+                            lapack_int lda, const lapack_int* ipiv,
+                            lapack_complex_double* b, lapack_int ldb );
+lapack_int LAPACKE_zhetrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const lapack_complex_double* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* work );
+lapack_int LAPACKE_zsyconv( int matrix_order, char uplo, char way, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_zsyconv_work( int matrix_order, char uplo, char way,
+                                 lapack_int n, lapack_complex_double* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_double* work );
+lapack_int LAPACKE_zsyswapr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double* a, lapack_int i1,
+                             lapack_int i2 );
+lapack_int LAPACKE_zsyswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double* a, lapack_int i1,
+                                  lapack_int i2 );
+lapack_int LAPACKE_zsytri2( int matrix_order, char uplo, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_zsytri2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_double* work, lapack_int lwork );
+lapack_int LAPACKE_zsytri2x( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double* a, lapack_int lda,
+                             const lapack_int* ipiv, lapack_int nb );
+lapack_int LAPACKE_zsytri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double* a, lapack_int lda,
+                                  const lapack_int* ipiv,
+                                  lapack_complex_double* work, lapack_int nb );
+lapack_int LAPACKE_zsytrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const lapack_complex_double* a,
+                            lapack_int lda, const lapack_int* ipiv,
+                            lapack_complex_double* b, lapack_int ldb );
+lapack_int LAPACKE_zsytrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const lapack_complex_double* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* work );
+lapack_int LAPACKE_zunbdb( int matrix_order, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           lapack_complex_double* x11, lapack_int ldx11,
+                           lapack_complex_double* x12, lapack_int ldx12,
+                           lapack_complex_double* x21, lapack_int ldx21,
+                           lapack_complex_double* x22, lapack_int ldx22,
+                           double* theta, double* phi,
+                           lapack_complex_double* taup1,
+                           lapack_complex_double* taup2,
+                           lapack_complex_double* tauq1,
+                           lapack_complex_double* tauq2 );
+lapack_int LAPACKE_zunbdb_work( int matrix_order, char trans, char signs,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                lapack_complex_double* x11, lapack_int ldx11,
+                                lapack_complex_double* x12, lapack_int ldx12,
+                                lapack_complex_double* x21, lapack_int ldx21,
+                                lapack_complex_double* x22, lapack_int ldx22,
+                                double* theta, double* phi,
+                                lapack_complex_double* taup1,
+                                lapack_complex_double* taup2,
+                                lapack_complex_double* tauq1,
+                                lapack_complex_double* tauq2,
+                                lapack_complex_double* work, lapack_int lwork );
+lapack_int LAPACKE_zuncsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           lapack_complex_double* x11, lapack_int ldx11,
+                           lapack_complex_double* x12, lapack_int ldx12,
+                           lapack_complex_double* x21, lapack_int ldx21,
+                           lapack_complex_double* x22, lapack_int ldx22,
+                           double* theta, lapack_complex_double* u1,
+                           lapack_int ldu1, lapack_complex_double* u2,
+                           lapack_int ldu2, lapack_complex_double* v1t,
+                           lapack_int ldv1t, lapack_complex_double* v2t,
+                           lapack_int ldv2t );
+lapack_int LAPACKE_zuncsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                char signs, lapack_int m, lapack_int p,
+                                lapack_int q, lapack_complex_double* x11,
+                                lapack_int ldx11, lapack_complex_double* x12,
+                                lapack_int ldx12, lapack_complex_double* x21,
+                                lapack_int ldx21, lapack_complex_double* x22,
+                                lapack_int ldx22, double* theta,
+                                lapack_complex_double* u1, lapack_int ldu1,
+                                lapack_complex_double* u2, lapack_int ldu2,
+                                lapack_complex_double* v1t, lapack_int ldv1t,
+                                lapack_complex_double* v2t, lapack_int ldv2t,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int lrwork,
+                                lapack_int* iwork );
+//LAPACK 3.4.0
+lapack_int LAPACKE_sgemqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int nb, const float* v, lapack_int ldv,
+                            const float* t, lapack_int ldt, float* c,
+                            lapack_int ldc );
+lapack_int LAPACKE_dgemqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int nb, const double* v, lapack_int ldv,
+                            const double* t, lapack_int ldt, double* c,
+                            lapack_int ldc );
+lapack_int LAPACKE_cgemqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int nb, const lapack_complex_float* v,
+                            lapack_int ldv, const lapack_complex_float* t,
+                            lapack_int ldt, lapack_complex_float* c,
+                            lapack_int ldc );
+lapack_int LAPACKE_zgemqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int nb, const lapack_complex_double* v,
+                            lapack_int ldv, const lapack_complex_double* t,
+                            lapack_int ldt, lapack_complex_double* c,
+                            lapack_int ldc );
+
+lapack_int LAPACKE_sgeqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nb, float* a, lapack_int lda, float* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_dgeqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nb, double* a, lapack_int lda, double* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_cgeqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nb, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_zgeqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nb, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* t,
+                           lapack_int ldt );
+
+lapack_int LAPACKE_sgeqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            float* a, lapack_int lda, float* t,
+                            lapack_int ldt );
+lapack_int LAPACKE_dgeqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            double* a, lapack_int lda, double* t,
+                            lapack_int ldt );
+lapack_int LAPACKE_cgeqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zgeqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_sgeqrt3( int matrix_order, lapack_int m, lapack_int n,
+                            float* a, lapack_int lda, float* t,
+                            lapack_int ldt );
+lapack_int LAPACKE_dgeqrt3( int matrix_order, lapack_int m, lapack_int n,
+                            double* a, lapack_int lda, double* t,
+                            lapack_int ldt );
+lapack_int LAPACKE_cgeqrt3( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zgeqrt3( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stpmqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int l, lapack_int nb, const float* v,
+                            lapack_int ldv, const float* t, lapack_int ldt,
+                            float* a, lapack_int lda, float* b,
+                            lapack_int ldb );
+lapack_int LAPACKE_dtpmqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int l, lapack_int nb, const double* v,
+                            lapack_int ldv, const double* t, lapack_int ldt,
+                            double* a, lapack_int lda, double* b,
+                            lapack_int ldb );
+lapack_int LAPACKE_ctpmqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int l, lapack_int nb,
+                            const lapack_complex_float* v, lapack_int ldv,
+                            const lapack_complex_float* t, lapack_int ldt,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztpmqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int l, lapack_int nb,
+                            const lapack_complex_double* v, lapack_int ldv,
+                            const lapack_complex_double* t, lapack_int ldt,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_dtpqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int l, lapack_int nb, double* a,
+                           lapack_int lda, double* b, lapack_int ldb, double* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_ctpqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int l, lapack_int nb, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* t,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_int ldt );
+lapack_int LAPACKE_ztpqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int l, lapack_int nb,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stpqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            float* a, lapack_int lda, float* b, lapack_int ldb,
+                            float* t, lapack_int ldt );
+lapack_int LAPACKE_dtpqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            double* a, lapack_int lda, double* b,
+                            lapack_int ldb, double* t, lapack_int ldt );
+lapack_int LAPACKE_ctpqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_ztpqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stprfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_int l, const float* v,
+                           lapack_int ldv, const float* t, lapack_int ldt,
+                           float* a, lapack_int lda, float* b, lapack_int ldb,
+                           lapack_int myldwork );
+lapack_int LAPACKE_dtprfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_int l, const double* v,
+                           lapack_int ldv, const double* t, lapack_int ldt,
+                           double* a, lapack_int lda, double* b, lapack_int ldb,
+                           lapack_int myldwork );
+lapack_int LAPACKE_ctprfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_int l,
+                           const lapack_complex_float* v, lapack_int ldv,
+                           const lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_int myldwork );
+lapack_int LAPACKE_ztprfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_int l,
+                           const lapack_complex_double* v, lapack_int ldv,
+                           const lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_int myldwork );
+
+lapack_int LAPACKE_sgemqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int nb, const float* v, lapack_int ldv,
+                                 const float* t, lapack_int ldt, float* c,
+                                 lapack_int ldc, float* work );
+lapack_int LAPACKE_dgemqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int nb, const double* v, lapack_int ldv,
+                                 const double* t, lapack_int ldt, double* c,
+                                 lapack_int ldc, double* work );
+lapack_int LAPACKE_cgemqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int nb, const lapack_complex_float* v,
+                                 lapack_int ldv, const lapack_complex_float* t,
+                                 lapack_int ldt, lapack_complex_float* c,
+                                 lapack_int ldc, lapack_complex_float* work );
+lapack_int LAPACKE_zgemqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int nb, const lapack_complex_double* v,
+                                 lapack_int ldv, const lapack_complex_double* t,
+                                 lapack_int ldt, lapack_complex_double* c,
+                                 lapack_int ldc, lapack_complex_double* work );
+
+lapack_int LAPACKE_sgeqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nb, float* a, lapack_int lda,
+                                float* t, lapack_int ldt, float* work );
+lapack_int LAPACKE_dgeqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nb, double* a, lapack_int lda,
+                                double* t, lapack_int ldt, double* work );
+lapack_int LAPACKE_cgeqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nb, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* t,
+                                lapack_int ldt, lapack_complex_float* work );
+lapack_int LAPACKE_zgeqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nb, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* t,
+                                lapack_int ldt, lapack_complex_double* work );
+
+lapack_int LAPACKE_sgeqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 float* a, lapack_int lda, float* t,
+                                 lapack_int ldt );
+lapack_int LAPACKE_dgeqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 double* a, lapack_int lda, double* t,
+                                 lapack_int ldt );
+lapack_int LAPACKE_cgeqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zgeqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_sgeqrt3_work( int matrix_order, lapack_int m, lapack_int n,
+                                 float* a, lapack_int lda, float* t,
+                                 lapack_int ldt );
+lapack_int LAPACKE_dgeqrt3_work( int matrix_order, lapack_int m, lapack_int n,
+                                 double* a, lapack_int lda, double* t,
+                                 lapack_int ldt );
+lapack_int LAPACKE_cgeqrt3_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zgeqrt3_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stpmqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int l, lapack_int nb, const float* v,
+                                 lapack_int ldv, const float* t, lapack_int ldt,
+                                 float* a, lapack_int lda, float* b,
+                                 lapack_int ldb, float* work );
+lapack_int LAPACKE_dtpmqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int l, lapack_int nb, const double* v,
+                                 lapack_int ldv, const double* t,
+                                 lapack_int ldt, double* a, lapack_int lda,
+                                 double* b, lapack_int ldb, double* work );
+lapack_int LAPACKE_ctpmqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int l, lapack_int nb,
+                                 const lapack_complex_float* v, lapack_int ldv,
+                                 const lapack_complex_float* t, lapack_int ldt,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_ztpmqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int l, lapack_int nb,
+                                 const lapack_complex_double* v, lapack_int ldv,
+                                 const lapack_complex_double* t, lapack_int ldt,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* work );
+
+lapack_int LAPACKE_dtpqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int l, lapack_int nb, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double* t, lapack_int ldt, double* work );
+lapack_int LAPACKE_ctpqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int l, lapack_int nb,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* t,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_int ldt, lapack_complex_float* work );
+lapack_int LAPACKE_ztpqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int l, lapack_int nb,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_stpqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 float* a, lapack_int lda, float* b,
+                                 lapack_int ldb, float* t, lapack_int ldt );
+lapack_int LAPACKE_dtpqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 double* a, lapack_int lda, double* b,
+                                 lapack_int ldb, double* t, lapack_int ldt );
+lapack_int LAPACKE_ctpqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_ztpqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stprfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                const float* v, lapack_int ldv, const float* t,
+                                lapack_int ldt, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, const float* mywork,
+                                lapack_int myldwork );
+lapack_int LAPACKE_dtprfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                const double* v, lapack_int ldv,
+                                const double* t, lapack_int ldt, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                const double* mywork, lapack_int myldwork );
+lapack_int LAPACKE_ctprfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                const lapack_complex_float* v, lapack_int ldv,
+                                const lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                const float* mywork, lapack_int myldwork );
+lapack_int LAPACKE_ztprfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                const lapack_complex_double* v, lapack_int ldv,
+                                const lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                const double* mywork, lapack_int myldwork );
+//LAPACK 3.X.X
+lapack_int LAPACKE_csyr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float alpha,
+                             const lapack_complex_float* x, lapack_int incx,
+                             lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zsyr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double alpha,
+                             const lapack_complex_double* x, lapack_int incx,
+                             lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_csyr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float alpha,
+                                  const lapack_complex_float* x,
+                                  lapack_int incx, lapack_complex_float* a,
+                                  lapack_int lda );
+lapack_int LAPACKE_zsyr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double alpha,
+                                  const lapack_complex_double* x,
+                                  lapack_int incx, lapack_complex_double* a,
+                                  lapack_int lda );
+
+
+
+#define LAPACK_sgetrf LAPACK_GLOBAL(sgetrf,SGETRF)
+#define LAPACK_dgetrf LAPACK_GLOBAL(dgetrf,DGETRF)
+#define LAPACK_cgetrf LAPACK_GLOBAL(cgetrf,CGETRF)
+#define LAPACK_zgetrf LAPACK_GLOBAL(zgetrf,ZGETRF)
+#define LAPACK_sgbtrf LAPACK_GLOBAL(sgbtrf,SGBTRF)
+#define LAPACK_dgbtrf LAPACK_GLOBAL(dgbtrf,DGBTRF)
+#define LAPACK_cgbtrf LAPACK_GLOBAL(cgbtrf,CGBTRF)
+#define LAPACK_zgbtrf LAPACK_GLOBAL(zgbtrf,ZGBTRF)
+#define LAPACK_sgttrf LAPACK_GLOBAL(sgttrf,SGTTRF)
+#define LAPACK_dgttrf LAPACK_GLOBAL(dgttrf,DGTTRF)
+#define LAPACK_cgttrf LAPACK_GLOBAL(cgttrf,CGTTRF)
+#define LAPACK_zgttrf LAPACK_GLOBAL(zgttrf,ZGTTRF)
+#define LAPACK_spotrf LAPACK_GLOBAL(spotrf,SPOTRF)
+#define LAPACK_dpotrf LAPACK_GLOBAL(dpotrf,DPOTRF)
+#define LAPACK_cpotrf LAPACK_GLOBAL(cpotrf,CPOTRF)
+#define LAPACK_zpotrf LAPACK_GLOBAL(zpotrf,ZPOTRF)
+#define LAPACK_dpstrf LAPACK_GLOBAL(dpstrf,DPSTRF)
+#define LAPACK_spstrf LAPACK_GLOBAL(spstrf,SPSTRF)
+#define LAPACK_zpstrf LAPACK_GLOBAL(zpstrf,ZPSTRF)
+#define LAPACK_cpstrf LAPACK_GLOBAL(cpstrf,CPSTRF)
+#define LAPACK_dpftrf LAPACK_GLOBAL(dpftrf,DPFTRF)
+#define LAPACK_spftrf LAPACK_GLOBAL(spftrf,SPFTRF)
+#define LAPACK_zpftrf LAPACK_GLOBAL(zpftrf,ZPFTRF)
+#define LAPACK_cpftrf LAPACK_GLOBAL(cpftrf,CPFTRF)
+#define LAPACK_spptrf LAPACK_GLOBAL(spptrf,SPPTRF)
+#define LAPACK_dpptrf LAPACK_GLOBAL(dpptrf,DPPTRF)
+#define LAPACK_cpptrf LAPACK_GLOBAL(cpptrf,CPPTRF)
+#define LAPACK_zpptrf LAPACK_GLOBAL(zpptrf,ZPPTRF)
+#define LAPACK_spbtrf LAPACK_GLOBAL(spbtrf,SPBTRF)
+#define LAPACK_dpbtrf LAPACK_GLOBAL(dpbtrf,DPBTRF)
+#define LAPACK_cpbtrf LAPACK_GLOBAL(cpbtrf,CPBTRF)
+#define LAPACK_zpbtrf LAPACK_GLOBAL(zpbtrf,ZPBTRF)
+#define LAPACK_spttrf LAPACK_GLOBAL(spttrf,SPTTRF)
+#define LAPACK_dpttrf LAPACK_GLOBAL(dpttrf,DPTTRF)
+#define LAPACK_cpttrf LAPACK_GLOBAL(cpttrf,CPTTRF)
+#define LAPACK_zpttrf LAPACK_GLOBAL(zpttrf,ZPTTRF)
+#define LAPACK_ssytrf LAPACK_GLOBAL(ssytrf,SSYTRF)
+#define LAPACK_dsytrf LAPACK_GLOBAL(dsytrf,DSYTRF)
+#define LAPACK_csytrf LAPACK_GLOBAL(csytrf,CSYTRF)
+#define LAPACK_zsytrf LAPACK_GLOBAL(zsytrf,ZSYTRF)
+#define LAPACK_chetrf LAPACK_GLOBAL(chetrf,CHETRF)
+#define LAPACK_zhetrf LAPACK_GLOBAL(zhetrf,ZHETRF)
+#define LAPACK_ssptrf LAPACK_GLOBAL(ssptrf,SSPTRF)
+#define LAPACK_dsptrf LAPACK_GLOBAL(dsptrf,DSPTRF)
+#define LAPACK_csptrf LAPACK_GLOBAL(csptrf,CSPTRF)
+#define LAPACK_zsptrf LAPACK_GLOBAL(zsptrf,ZSPTRF)
+#define LAPACK_chptrf LAPACK_GLOBAL(chptrf,CHPTRF)
+#define LAPACK_zhptrf LAPACK_GLOBAL(zhptrf,ZHPTRF)
+#define LAPACK_sgetrs LAPACK_GLOBAL(sgetrs,SGETRS)
+#define LAPACK_dgetrs LAPACK_GLOBAL(dgetrs,DGETRS)
+#define LAPACK_cgetrs LAPACK_GLOBAL(cgetrs,CGETRS)
+#define LAPACK_zgetrs LAPACK_GLOBAL(zgetrs,ZGETRS)
+#define LAPACK_sgbtrs LAPACK_GLOBAL(sgbtrs,SGBTRS)
+#define LAPACK_dgbtrs LAPACK_GLOBAL(dgbtrs,DGBTRS)
+#define LAPACK_cgbtrs LAPACK_GLOBAL(cgbtrs,CGBTRS)
+#define LAPACK_zgbtrs LAPACK_GLOBAL(zgbtrs,ZGBTRS)
+#define LAPACK_sgttrs LAPACK_GLOBAL(sgttrs,SGTTRS)
+#define LAPACK_dgttrs LAPACK_GLOBAL(dgttrs,DGTTRS)
+#define LAPACK_cgttrs LAPACK_GLOBAL(cgttrs,CGTTRS)
+#define LAPACK_zgttrs LAPACK_GLOBAL(zgttrs,ZGTTRS)
+#define LAPACK_spotrs LAPACK_GLOBAL(spotrs,SPOTRS)
+#define LAPACK_dpotrs LAPACK_GLOBAL(dpotrs,DPOTRS)
+#define LAPACK_cpotrs LAPACK_GLOBAL(cpotrs,CPOTRS)
+#define LAPACK_zpotrs LAPACK_GLOBAL(zpotrs,ZPOTRS)
+#define LAPACK_dpftrs LAPACK_GLOBAL(dpftrs,DPFTRS)
+#define LAPACK_spftrs LAPACK_GLOBAL(spftrs,SPFTRS)
+#define LAPACK_zpftrs LAPACK_GLOBAL(zpftrs,ZPFTRS)
+#define LAPACK_cpftrs LAPACK_GLOBAL(cpftrs,CPFTRS)
+#define LAPACK_spptrs LAPACK_GLOBAL(spptrs,SPPTRS)
+#define LAPACK_dpptrs LAPACK_GLOBAL(dpptrs,DPPTRS)
+#define LAPACK_cpptrs LAPACK_GLOBAL(cpptrs,CPPTRS)
+#define LAPACK_zpptrs LAPACK_GLOBAL(zpptrs,ZPPTRS)
+#define LAPACK_spbtrs LAPACK_GLOBAL(spbtrs,SPBTRS)
+#define LAPACK_dpbtrs LAPACK_GLOBAL(dpbtrs,DPBTRS)
+#define LAPACK_cpbtrs LAPACK_GLOBAL(cpbtrs,CPBTRS)
+#define LAPACK_zpbtrs LAPACK_GLOBAL(zpbtrs,ZPBTRS)
+#define LAPACK_spttrs LAPACK_GLOBAL(spttrs,SPTTRS)
+#define LAPACK_dpttrs LAPACK_GLOBAL(dpttrs,DPTTRS)
+#define LAPACK_cpttrs LAPACK_GLOBAL(cpttrs,CPTTRS)
+#define LAPACK_zpttrs LAPACK_GLOBAL(zpttrs,ZPTTRS)
+#define LAPACK_ssytrs LAPACK_GLOBAL(ssytrs,SSYTRS)
+#define LAPACK_dsytrs LAPACK_GLOBAL(dsytrs,DSYTRS)
+#define LAPACK_csytrs LAPACK_GLOBAL(csytrs,CSYTRS)
+#define LAPACK_zsytrs LAPACK_GLOBAL(zsytrs,ZSYTRS)
+#define LAPACK_chetrs LAPACK_GLOBAL(chetrs,CHETRS)
+#define LAPACK_zhetrs LAPACK_GLOBAL(zhetrs,ZHETRS)
+#define LAPACK_ssptrs LAPACK_GLOBAL(ssptrs,SSPTRS)
+#define LAPACK_dsptrs LAPACK_GLOBAL(dsptrs,DSPTRS)
+#define LAPACK_csptrs LAPACK_GLOBAL(csptrs,CSPTRS)
+#define LAPACK_zsptrs LAPACK_GLOBAL(zsptrs,ZSPTRS)
+#define LAPACK_chptrs LAPACK_GLOBAL(chptrs,CHPTRS)
+#define LAPACK_zhptrs LAPACK_GLOBAL(zhptrs,ZHPTRS)
+#define LAPACK_strtrs LAPACK_GLOBAL(strtrs,STRTRS)
+#define LAPACK_dtrtrs LAPACK_GLOBAL(dtrtrs,DTRTRS)
+#define LAPACK_ctrtrs LAPACK_GLOBAL(ctrtrs,CTRTRS)
+#define LAPACK_ztrtrs LAPACK_GLOBAL(ztrtrs,ZTRTRS)
+#define LAPACK_stptrs LAPACK_GLOBAL(stptrs,STPTRS)
+#define LAPACK_dtptrs LAPACK_GLOBAL(dtptrs,DTPTRS)
+#define LAPACK_ctptrs LAPACK_GLOBAL(ctptrs,CTPTRS)
+#define LAPACK_ztptrs LAPACK_GLOBAL(ztptrs,ZTPTRS)
+#define LAPACK_stbtrs LAPACK_GLOBAL(stbtrs,STBTRS)
+#define LAPACK_dtbtrs LAPACK_GLOBAL(dtbtrs,DTBTRS)
+#define LAPACK_ctbtrs LAPACK_GLOBAL(ctbtrs,CTBTRS)
+#define LAPACK_ztbtrs LAPACK_GLOBAL(ztbtrs,ZTBTRS)
+#define LAPACK_sgecon LAPACK_GLOBAL(sgecon,SGECON)
+#define LAPACK_dgecon LAPACK_GLOBAL(dgecon,DGECON)
+#define LAPACK_cgecon LAPACK_GLOBAL(cgecon,CGECON)
+#define LAPACK_zgecon LAPACK_GLOBAL(zgecon,ZGECON)
+#define LAPACK_sgbcon LAPACK_GLOBAL(sgbcon,SGBCON)
+#define LAPACK_dgbcon LAPACK_GLOBAL(dgbcon,DGBCON)
+#define LAPACK_cgbcon LAPACK_GLOBAL(cgbcon,CGBCON)
+#define LAPACK_zgbcon LAPACK_GLOBAL(zgbcon,ZGBCON)
+#define LAPACK_sgtcon LAPACK_GLOBAL(sgtcon,SGTCON)
+#define LAPACK_dgtcon LAPACK_GLOBAL(dgtcon,DGTCON)
+#define LAPACK_cgtcon LAPACK_GLOBAL(cgtcon,CGTCON)
+#define LAPACK_zgtcon LAPACK_GLOBAL(zgtcon,ZGTCON)
+#define LAPACK_spocon LAPACK_GLOBAL(spocon,SPOCON)
+#define LAPACK_dpocon LAPACK_GLOBAL(dpocon,DPOCON)
+#define LAPACK_cpocon LAPACK_GLOBAL(cpocon,CPOCON)
+#define LAPACK_zpocon LAPACK_GLOBAL(zpocon,ZPOCON)
+#define LAPACK_sppcon LAPACK_GLOBAL(sppcon,SPPCON)
+#define LAPACK_dppcon LAPACK_GLOBAL(dppcon,DPPCON)
+#define LAPACK_cppcon LAPACK_GLOBAL(cppcon,CPPCON)
+#define LAPACK_zppcon LAPACK_GLOBAL(zppcon,ZPPCON)
+#define LAPACK_spbcon LAPACK_GLOBAL(spbcon,SPBCON)
+#define LAPACK_dpbcon LAPACK_GLOBAL(dpbcon,DPBCON)
+#define LAPACK_cpbcon LAPACK_GLOBAL(cpbcon,CPBCON)
+#define LAPACK_zpbcon LAPACK_GLOBAL(zpbcon,ZPBCON)
+#define LAPACK_sptcon LAPACK_GLOBAL(sptcon,SPTCON)
+#define LAPACK_dptcon LAPACK_GLOBAL(dptcon,DPTCON)
+#define LAPACK_cptcon LAPACK_GLOBAL(cptcon,CPTCON)
+#define LAPACK_zptcon LAPACK_GLOBAL(zptcon,ZPTCON)
+#define LAPACK_ssycon LAPACK_GLOBAL(ssycon,SSYCON)
+#define LAPACK_dsycon LAPACK_GLOBAL(dsycon,DSYCON)
+#define LAPACK_csycon LAPACK_GLOBAL(csycon,CSYCON)
+#define LAPACK_zsycon LAPACK_GLOBAL(zsycon,ZSYCON)
+#define LAPACK_checon LAPACK_GLOBAL(checon,CHECON)
+#define LAPACK_zhecon LAPACK_GLOBAL(zhecon,ZHECON)
+#define LAPACK_sspcon LAPACK_GLOBAL(sspcon,SSPCON)
+#define LAPACK_dspcon LAPACK_GLOBAL(dspcon,DSPCON)
+#define LAPACK_cspcon LAPACK_GLOBAL(cspcon,CSPCON)
+#define LAPACK_zspcon LAPACK_GLOBAL(zspcon,ZSPCON)
+#define LAPACK_chpcon LAPACK_GLOBAL(chpcon,CHPCON)
+#define LAPACK_zhpcon LAPACK_GLOBAL(zhpcon,ZHPCON)
+#define LAPACK_strcon LAPACK_GLOBAL(strcon,STRCON)
+#define LAPACK_dtrcon LAPACK_GLOBAL(dtrcon,DTRCON)
+#define LAPACK_ctrcon LAPACK_GLOBAL(ctrcon,CTRCON)
+#define LAPACK_ztrcon LAPACK_GLOBAL(ztrcon,ZTRCON)
+#define LAPACK_stpcon LAPACK_GLOBAL(stpcon,STPCON)
+#define LAPACK_dtpcon LAPACK_GLOBAL(dtpcon,DTPCON)
+#define LAPACK_ctpcon LAPACK_GLOBAL(ctpcon,CTPCON)
+#define LAPACK_ztpcon LAPACK_GLOBAL(ztpcon,ZTPCON)
+#define LAPACK_stbcon LAPACK_GLOBAL(stbcon,STBCON)
+#define LAPACK_dtbcon LAPACK_GLOBAL(dtbcon,DTBCON)
+#define LAPACK_ctbcon LAPACK_GLOBAL(ctbcon,CTBCON)
+#define LAPACK_ztbcon LAPACK_GLOBAL(ztbcon,ZTBCON)
+#define LAPACK_sgerfs LAPACK_GLOBAL(sgerfs,SGERFS)
+#define LAPACK_dgerfs LAPACK_GLOBAL(dgerfs,DGERFS)
+#define LAPACK_cgerfs LAPACK_GLOBAL(cgerfs,CGERFS)
+#define LAPACK_zgerfs LAPACK_GLOBAL(zgerfs,ZGERFS)
+#define LAPACK_dgerfsx LAPACK_GLOBAL(dgerfsx,DGERFSX)
+#define LAPACK_sgerfsx LAPACK_GLOBAL(sgerfsx,SGERFSX)
+#define LAPACK_zgerfsx LAPACK_GLOBAL(zgerfsx,ZGERFSX)
+#define LAPACK_cgerfsx LAPACK_GLOBAL(cgerfsx,CGERFSX)
+#define LAPACK_sgbrfs LAPACK_GLOBAL(sgbrfs,SGBRFS)
+#define LAPACK_dgbrfs LAPACK_GLOBAL(dgbrfs,DGBRFS)
+#define LAPACK_cgbrfs LAPACK_GLOBAL(cgbrfs,CGBRFS)
+#define LAPACK_zgbrfs LAPACK_GLOBAL(zgbrfs,ZGBRFS)
+#define LAPACK_dgbrfsx LAPACK_GLOBAL(dgbrfsx,DGBRFSX)
+#define LAPACK_sgbrfsx LAPACK_GLOBAL(sgbrfsx,SGBRFSX)
+#define LAPACK_zgbrfsx LAPACK_GLOBAL(zgbrfsx,ZGBRFSX)
+#define LAPACK_cgbrfsx LAPACK_GLOBAL(cgbrfsx,CGBRFSX)
+#define LAPACK_sgtrfs LAPACK_GLOBAL(sgtrfs,SGTRFS)
+#define LAPACK_dgtrfs LAPACK_GLOBAL(dgtrfs,DGTRFS)
+#define LAPACK_cgtrfs LAPACK_GLOBAL(cgtrfs,CGTRFS)
+#define LAPACK_zgtrfs LAPACK_GLOBAL(zgtrfs,ZGTRFS)
+#define LAPACK_sporfs LAPACK_GLOBAL(sporfs,SPORFS)
+#define LAPACK_dporfs LAPACK_GLOBAL(dporfs,DPORFS)
+#define LAPACK_cporfs LAPACK_GLOBAL(cporfs,CPORFS)
+#define LAPACK_zporfs LAPACK_GLOBAL(zporfs,ZPORFS)
+#define LAPACK_dporfsx LAPACK_GLOBAL(dporfsx,DPORFSX)
+#define LAPACK_sporfsx LAPACK_GLOBAL(sporfsx,SPORFSX)
+#define LAPACK_zporfsx LAPACK_GLOBAL(zporfsx,ZPORFSX)
+#define LAPACK_cporfsx LAPACK_GLOBAL(cporfsx,CPORFSX)
+#define LAPACK_spprfs LAPACK_GLOBAL(spprfs,SPPRFS)
+#define LAPACK_dpprfs LAPACK_GLOBAL(dpprfs,DPPRFS)
+#define LAPACK_cpprfs LAPACK_GLOBAL(cpprfs,CPPRFS)
+#define LAPACK_zpprfs LAPACK_GLOBAL(zpprfs,ZPPRFS)
+#define LAPACK_spbrfs LAPACK_GLOBAL(spbrfs,SPBRFS)
+#define LAPACK_dpbrfs LAPACK_GLOBAL(dpbrfs,DPBRFS)
+#define LAPACK_cpbrfs LAPACK_GLOBAL(cpbrfs,CPBRFS)
+#define LAPACK_zpbrfs LAPACK_GLOBAL(zpbrfs,ZPBRFS)
+#define LAPACK_sptrfs LAPACK_GLOBAL(sptrfs,SPTRFS)
+#define LAPACK_dptrfs LAPACK_GLOBAL(dptrfs,DPTRFS)
+#define LAPACK_cptrfs LAPACK_GLOBAL(cptrfs,CPTRFS)
+#define LAPACK_zptrfs LAPACK_GLOBAL(zptrfs,ZPTRFS)
+#define LAPACK_ssyrfs LAPACK_GLOBAL(ssyrfs,SSYRFS)
+#define LAPACK_dsyrfs LAPACK_GLOBAL(dsyrfs,DSYRFS)
+#define LAPACK_csyrfs LAPACK_GLOBAL(csyrfs,CSYRFS)
+#define LAPACK_zsyrfs LAPACK_GLOBAL(zsyrfs,ZSYRFS)
+#define LAPACK_dsyrfsx LAPACK_GLOBAL(dsyrfsx,DSYRFSX)
+#define LAPACK_ssyrfsx LAPACK_GLOBAL(ssyrfsx,SSYRFSX)
+#define LAPACK_zsyrfsx LAPACK_GLOBAL(zsyrfsx,ZSYRFSX)
+#define LAPACK_csyrfsx LAPACK_GLOBAL(csyrfsx,CSYRFSX)
+#define LAPACK_cherfs LAPACK_GLOBAL(cherfs,CHERFS)
+#define LAPACK_zherfs LAPACK_GLOBAL(zherfs,ZHERFS)
+#define LAPACK_zherfsx LAPACK_GLOBAL(zherfsx,ZHERFSX)
+#define LAPACK_cherfsx LAPACK_GLOBAL(cherfsx,CHERFSX)
+#define LAPACK_ssprfs LAPACK_GLOBAL(ssprfs,SSPRFS)
+#define LAPACK_dsprfs LAPACK_GLOBAL(dsprfs,DSPRFS)
+#define LAPACK_csprfs LAPACK_GLOBAL(csprfs,CSPRFS)
+#define LAPACK_zsprfs LAPACK_GLOBAL(zsprfs,ZSPRFS)
+#define LAPACK_chprfs LAPACK_GLOBAL(chprfs,CHPRFS)
+#define LAPACK_zhprfs LAPACK_GLOBAL(zhprfs,ZHPRFS)
+#define LAPACK_strrfs LAPACK_GLOBAL(strrfs,STRRFS)
+#define LAPACK_dtrrfs LAPACK_GLOBAL(dtrrfs,DTRRFS)
+#define LAPACK_ctrrfs LAPACK_GLOBAL(ctrrfs,CTRRFS)
+#define LAPACK_ztrrfs LAPACK_GLOBAL(ztrrfs,ZTRRFS)
+#define LAPACK_stprfs LAPACK_GLOBAL(stprfs,STPRFS)
+#define LAPACK_dtprfs LAPACK_GLOBAL(dtprfs,DTPRFS)
+#define LAPACK_ctprfs LAPACK_GLOBAL(ctprfs,CTPRFS)
+#define LAPACK_ztprfs LAPACK_GLOBAL(ztprfs,ZTPRFS)
+#define LAPACK_stbrfs LAPACK_GLOBAL(stbrfs,STBRFS)
+#define LAPACK_dtbrfs LAPACK_GLOBAL(dtbrfs,DTBRFS)
+#define LAPACK_ctbrfs LAPACK_GLOBAL(ctbrfs,CTBRFS)
+#define LAPACK_ztbrfs LAPACK_GLOBAL(ztbrfs,ZTBRFS)
+#define LAPACK_sgetri LAPACK_GLOBAL(sgetri,SGETRI)
+#define LAPACK_dgetri LAPACK_GLOBAL(dgetri,DGETRI)
+#define LAPACK_cgetri LAPACK_GLOBAL(cgetri,CGETRI)
+#define LAPACK_zgetri LAPACK_GLOBAL(zgetri,ZGETRI)
+#define LAPACK_spotri LAPACK_GLOBAL(spotri,SPOTRI)
+#define LAPACK_dpotri LAPACK_GLOBAL(dpotri,DPOTRI)
+#define LAPACK_cpotri LAPACK_GLOBAL(cpotri,CPOTRI)
+#define LAPACK_zpotri LAPACK_GLOBAL(zpotri,ZPOTRI)
+#define LAPACK_dpftri LAPACK_GLOBAL(dpftri,DPFTRI)
+#define LAPACK_spftri LAPACK_GLOBAL(spftri,SPFTRI)
+#define LAPACK_zpftri LAPACK_GLOBAL(zpftri,ZPFTRI)
+#define LAPACK_cpftri LAPACK_GLOBAL(cpftri,CPFTRI)
+#define LAPACK_spptri LAPACK_GLOBAL(spptri,SPPTRI)
+#define LAPACK_dpptri LAPACK_GLOBAL(dpptri,DPPTRI)
+#define LAPACK_cpptri LAPACK_GLOBAL(cpptri,CPPTRI)
+#define LAPACK_zpptri LAPACK_GLOBAL(zpptri,ZPPTRI)
+#define LAPACK_ssytri LAPACK_GLOBAL(ssytri,SSYTRI)
+#define LAPACK_dsytri LAPACK_GLOBAL(dsytri,DSYTRI)
+#define LAPACK_csytri LAPACK_GLOBAL(csytri,CSYTRI)
+#define LAPACK_zsytri LAPACK_GLOBAL(zsytri,ZSYTRI)
+#define LAPACK_chetri LAPACK_GLOBAL(chetri,CHETRI)
+#define LAPACK_zhetri LAPACK_GLOBAL(zhetri,ZHETRI)
+#define LAPACK_ssptri LAPACK_GLOBAL(ssptri,SSPTRI)
+#define LAPACK_dsptri LAPACK_GLOBAL(dsptri,DSPTRI)
+#define LAPACK_csptri LAPACK_GLOBAL(csptri,CSPTRI)
+#define LAPACK_zsptri LAPACK_GLOBAL(zsptri,ZSPTRI)
+#define LAPACK_chptri LAPACK_GLOBAL(chptri,CHPTRI)
+#define LAPACK_zhptri LAPACK_GLOBAL(zhptri,ZHPTRI)
+#define LAPACK_strtri LAPACK_GLOBAL(strtri,STRTRI)
+#define LAPACK_dtrtri LAPACK_GLOBAL(dtrtri,DTRTRI)
+#define LAPACK_ctrtri LAPACK_GLOBAL(ctrtri,CTRTRI)
+#define LAPACK_ztrtri LAPACK_GLOBAL(ztrtri,ZTRTRI)
+#define LAPACK_dtftri LAPACK_GLOBAL(dtftri,DTFTRI)
+#define LAPACK_stftri LAPACK_GLOBAL(stftri,STFTRI)
+#define LAPACK_ztftri LAPACK_GLOBAL(ztftri,ZTFTRI)
+#define LAPACK_ctftri LAPACK_GLOBAL(ctftri,CTFTRI)
+#define LAPACK_stptri LAPACK_GLOBAL(stptri,STPTRI)
+#define LAPACK_dtptri LAPACK_GLOBAL(dtptri,DTPTRI)
+#define LAPACK_ctptri LAPACK_GLOBAL(ctptri,CTPTRI)
+#define LAPACK_ztptri LAPACK_GLOBAL(ztptri,ZTPTRI)
+#define LAPACK_sgeequ LAPACK_GLOBAL(sgeequ,SGEEQU)
+#define LAPACK_dgeequ LAPACK_GLOBAL(dgeequ,DGEEQU)
+#define LAPACK_cgeequ LAPACK_GLOBAL(cgeequ,CGEEQU)
+#define LAPACK_zgeequ LAPACK_GLOBAL(zgeequ,ZGEEQU)
+#define LAPACK_dgeequb LAPACK_GLOBAL(dgeequb,DGEEQUB)
+#define LAPACK_sgeequb LAPACK_GLOBAL(sgeequb,SGEEQUB)
+#define LAPACK_zgeequb LAPACK_GLOBAL(zgeequb,ZGEEQUB)
+#define LAPACK_cgeequb LAPACK_GLOBAL(cgeequb,CGEEQUB)
+#define LAPACK_sgbequ LAPACK_GLOBAL(sgbequ,SGBEQU)
+#define LAPACK_dgbequ LAPACK_GLOBAL(dgbequ,DGBEQU)
+#define LAPACK_cgbequ LAPACK_GLOBAL(cgbequ,CGBEQU)
+#define LAPACK_zgbequ LAPACK_GLOBAL(zgbequ,ZGBEQU)
+#define LAPACK_dgbequb LAPACK_GLOBAL(dgbequb,DGBEQUB)
+#define LAPACK_sgbequb LAPACK_GLOBAL(sgbequb,SGBEQUB)
+#define LAPACK_zgbequb LAPACK_GLOBAL(zgbequb,ZGBEQUB)
+#define LAPACK_cgbequb LAPACK_GLOBAL(cgbequb,CGBEQUB)
+#define LAPACK_spoequ LAPACK_GLOBAL(spoequ,SPOEQU)
+#define LAPACK_dpoequ LAPACK_GLOBAL(dpoequ,DPOEQU)
+#define LAPACK_cpoequ LAPACK_GLOBAL(cpoequ,CPOEQU)
+#define LAPACK_zpoequ LAPACK_GLOBAL(zpoequ,ZPOEQU)
+#define LAPACK_dpoequb LAPACK_GLOBAL(dpoequb,DPOEQUB)
+#define LAPACK_spoequb LAPACK_GLOBAL(spoequb,SPOEQUB)
+#define LAPACK_zpoequb LAPACK_GLOBAL(zpoequb,ZPOEQUB)
+#define LAPACK_cpoequb LAPACK_GLOBAL(cpoequb,CPOEQUB)
+#define LAPACK_sppequ LAPACK_GLOBAL(sppequ,SPPEQU)
+#define LAPACK_dppequ LAPACK_GLOBAL(dppequ,DPPEQU)
+#define LAPACK_cppequ LAPACK_GLOBAL(cppequ,CPPEQU)
+#define LAPACK_zppequ LAPACK_GLOBAL(zppequ,ZPPEQU)
+#define LAPACK_spbequ LAPACK_GLOBAL(spbequ,SPBEQU)
+#define LAPACK_dpbequ LAPACK_GLOBAL(dpbequ,DPBEQU)
+#define LAPACK_cpbequ LAPACK_GLOBAL(cpbequ,CPBEQU)
+#define LAPACK_zpbequ LAPACK_GLOBAL(zpbequ,ZPBEQU)
+#define LAPACK_dsyequb LAPACK_GLOBAL(dsyequb,DSYEQUB)
+#define LAPACK_ssyequb LAPACK_GLOBAL(ssyequb,SSYEQUB)
+#define LAPACK_zsyequb LAPACK_GLOBAL(zsyequb,ZSYEQUB)
+#define LAPACK_csyequb LAPACK_GLOBAL(csyequb,CSYEQUB)
+#define LAPACK_zheequb LAPACK_GLOBAL(zheequb,ZHEEQUB)
+#define LAPACK_cheequb LAPACK_GLOBAL(cheequb,CHEEQUB)
+#define LAPACK_sgesv LAPACK_GLOBAL(sgesv,SGESV)
+#define LAPACK_dgesv LAPACK_GLOBAL(dgesv,DGESV)
+#define LAPACK_cgesv LAPACK_GLOBAL(cgesv,CGESV)
+#define LAPACK_zgesv LAPACK_GLOBAL(zgesv,ZGESV)
+#define LAPACK_dsgesv LAPACK_GLOBAL(dsgesv,DSGESV)
+#define LAPACK_zcgesv LAPACK_GLOBAL(zcgesv,ZCGESV)
+#define LAPACK_sgesvx LAPACK_GLOBAL(sgesvx,SGESVX)
+#define LAPACK_dgesvx LAPACK_GLOBAL(dgesvx,DGESVX)
+#define LAPACK_cgesvx LAPACK_GLOBAL(cgesvx,CGESVX)
+#define LAPACK_zgesvx LAPACK_GLOBAL(zgesvx,ZGESVX)
+#define LAPACK_dgesvxx LAPACK_GLOBAL(dgesvxx,DGESVXX)
+#define LAPACK_sgesvxx LAPACK_GLOBAL(sgesvxx,SGESVXX)
+#define LAPACK_zgesvxx LAPACK_GLOBAL(zgesvxx,ZGESVXX)
+#define LAPACK_cgesvxx LAPACK_GLOBAL(cgesvxx,CGESVXX)
+#define LAPACK_sgbsv LAPACK_GLOBAL(sgbsv,SGBSV)
+#define LAPACK_dgbsv LAPACK_GLOBAL(dgbsv,DGBSV)
+#define LAPACK_cgbsv LAPACK_GLOBAL(cgbsv,CGBSV)
+#define LAPACK_zgbsv LAPACK_GLOBAL(zgbsv,ZGBSV)
+#define LAPACK_sgbsvx LAPACK_GLOBAL(sgbsvx,SGBSVX)
+#define LAPACK_dgbsvx LAPACK_GLOBAL(dgbsvx,DGBSVX)
+#define LAPACK_cgbsvx LAPACK_GLOBAL(cgbsvx,CGBSVX)
+#define LAPACK_zgbsvx LAPACK_GLOBAL(zgbsvx,ZGBSVX)
+#define LAPACK_dgbsvxx LAPACK_GLOBAL(dgbsvxx,DGBSVXX)
+#define LAPACK_sgbsvxx LAPACK_GLOBAL(sgbsvxx,SGBSVXX)
+#define LAPACK_zgbsvxx LAPACK_GLOBAL(zgbsvxx,ZGBSVXX)
+#define LAPACK_cgbsvxx LAPACK_GLOBAL(cgbsvxx,CGBSVXX)
+#define LAPACK_sgtsv LAPACK_GLOBAL(sgtsv,SGTSV)
+#define LAPACK_dgtsv LAPACK_GLOBAL(dgtsv,DGTSV)
+#define LAPACK_cgtsv LAPACK_GLOBAL(cgtsv,CGTSV)
+#define LAPACK_zgtsv LAPACK_GLOBAL(zgtsv,ZGTSV)
+#define LAPACK_sgtsvx LAPACK_GLOBAL(sgtsvx,SGTSVX)
+#define LAPACK_dgtsvx LAPACK_GLOBAL(dgtsvx,DGTSVX)
+#define LAPACK_cgtsvx LAPACK_GLOBAL(cgtsvx,CGTSVX)
+#define LAPACK_zgtsvx LAPACK_GLOBAL(zgtsvx,ZGTSVX)
+#define LAPACK_sposv LAPACK_GLOBAL(sposv,SPOSV)
+#define LAPACK_dposv LAPACK_GLOBAL(dposv,DPOSV)
+#define LAPACK_cposv LAPACK_GLOBAL(cposv,CPOSV)
+#define LAPACK_zposv LAPACK_GLOBAL(zposv,ZPOSV)
+#define LAPACK_dsposv LAPACK_GLOBAL(dsposv,DSPOSV)
+#define LAPACK_zcposv LAPACK_GLOBAL(zcposv,ZCPOSV)
+#define LAPACK_sposvx LAPACK_GLOBAL(sposvx,SPOSVX)
+#define LAPACK_dposvx LAPACK_GLOBAL(dposvx,DPOSVX)
+#define LAPACK_cposvx LAPACK_GLOBAL(cposvx,CPOSVX)
+#define LAPACK_zposvx LAPACK_GLOBAL(zposvx,ZPOSVX)
+#define LAPACK_dposvxx LAPACK_GLOBAL(dposvxx,DPOSVXX)
+#define LAPACK_sposvxx LAPACK_GLOBAL(sposvxx,SPOSVXX)
+#define LAPACK_zposvxx LAPACK_GLOBAL(zposvxx,ZPOSVXX)
+#define LAPACK_cposvxx LAPACK_GLOBAL(cposvxx,CPOSVXX)
+#define LAPACK_sppsv LAPACK_GLOBAL(sppsv,SPPSV)
+#define LAPACK_dppsv LAPACK_GLOBAL(dppsv,DPPSV)
+#define LAPACK_cppsv LAPACK_GLOBAL(cppsv,CPPSV)
+#define LAPACK_zppsv LAPACK_GLOBAL(zppsv,ZPPSV)
+#define LAPACK_sppsvx LAPACK_GLOBAL(sppsvx,SPPSVX)
+#define LAPACK_dppsvx LAPACK_GLOBAL(dppsvx,DPPSVX)
+#define LAPACK_cppsvx LAPACK_GLOBAL(cppsvx,CPPSVX)
+#define LAPACK_zppsvx LAPACK_GLOBAL(zppsvx,ZPPSVX)
+#define LAPACK_spbsv LAPACK_GLOBAL(spbsv,SPBSV)
+#define LAPACK_dpbsv LAPACK_GLOBAL(dpbsv,DPBSV)
+#define LAPACK_cpbsv LAPACK_GLOBAL(cpbsv,CPBSV)
+#define LAPACK_zpbsv LAPACK_GLOBAL(zpbsv,ZPBSV)
+#define LAPACK_spbsvx LAPACK_GLOBAL(spbsvx,SPBSVX)
+#define LAPACK_dpbsvx LAPACK_GLOBAL(dpbsvx,DPBSVX)
+#define LAPACK_cpbsvx LAPACK_GLOBAL(cpbsvx,CPBSVX)
+#define LAPACK_zpbsvx LAPACK_GLOBAL(zpbsvx,ZPBSVX)
+#define LAPACK_sptsv LAPACK_GLOBAL(sptsv,SPTSV)
+#define LAPACK_dptsv LAPACK_GLOBAL(dptsv,DPTSV)
+#define LAPACK_cptsv LAPACK_GLOBAL(cptsv,CPTSV)
+#define LAPACK_zptsv LAPACK_GLOBAL(zptsv,ZPTSV)
+#define LAPACK_sptsvx LAPACK_GLOBAL(sptsvx,SPTSVX)
+#define LAPACK_dptsvx LAPACK_GLOBAL(dptsvx,DPTSVX)
+#define LAPACK_cptsvx LAPACK_GLOBAL(cptsvx,CPTSVX)
+#define LAPACK_zptsvx LAPACK_GLOBAL(zptsvx,ZPTSVX)
+#define LAPACK_ssysv LAPACK_GLOBAL(ssysv,SSYSV)
+#define LAPACK_dsysv LAPACK_GLOBAL(dsysv,DSYSV)
+#define LAPACK_csysv LAPACK_GLOBAL(csysv,CSYSV)
+#define LAPACK_zsysv LAPACK_GLOBAL(zsysv,ZSYSV)
+#define LAPACK_ssysvx LAPACK_GLOBAL(ssysvx,SSYSVX)
+#define LAPACK_dsysvx LAPACK_GLOBAL(dsysvx,DSYSVX)
+#define LAPACK_csysvx LAPACK_GLOBAL(csysvx,CSYSVX)
+#define LAPACK_zsysvx LAPACK_GLOBAL(zsysvx,ZSYSVX)
+#define LAPACK_dsysvxx LAPACK_GLOBAL(dsysvxx,DSYSVXX)
+#define LAPACK_ssysvxx LAPACK_GLOBAL(ssysvxx,SSYSVXX)
+#define LAPACK_zsysvxx LAPACK_GLOBAL(zsysvxx,ZSYSVXX)
+#define LAPACK_csysvxx LAPACK_GLOBAL(csysvxx,CSYSVXX)
+#define LAPACK_chesv LAPACK_GLOBAL(chesv,CHESV)
+#define LAPACK_zhesv LAPACK_GLOBAL(zhesv,ZHESV)
+#define LAPACK_chesvx LAPACK_GLOBAL(chesvx,CHESVX)
+#define LAPACK_zhesvx LAPACK_GLOBAL(zhesvx,ZHESVX)
+#define LAPACK_zhesvxx LAPACK_GLOBAL(zhesvxx,ZHESVXX)
+#define LAPACK_chesvxx LAPACK_GLOBAL(chesvxx,CHESVXX)
+#define LAPACK_sspsv LAPACK_GLOBAL(sspsv,SSPSV)
+#define LAPACK_dspsv LAPACK_GLOBAL(dspsv,DSPSV)
+#define LAPACK_cspsv LAPACK_GLOBAL(cspsv,CSPSV)
+#define LAPACK_zspsv LAPACK_GLOBAL(zspsv,ZSPSV)
+#define LAPACK_sspsvx LAPACK_GLOBAL(sspsvx,SSPSVX)
+#define LAPACK_dspsvx LAPACK_GLOBAL(dspsvx,DSPSVX)
+#define LAPACK_cspsvx LAPACK_GLOBAL(cspsvx,CSPSVX)
+#define LAPACK_zspsvx LAPACK_GLOBAL(zspsvx,ZSPSVX)
+#define LAPACK_chpsv LAPACK_GLOBAL(chpsv,CHPSV)
+#define LAPACK_zhpsv LAPACK_GLOBAL(zhpsv,ZHPSV)
+#define LAPACK_chpsvx LAPACK_GLOBAL(chpsvx,CHPSVX)
+#define LAPACK_zhpsvx LAPACK_GLOBAL(zhpsvx,ZHPSVX)
+#define LAPACK_sgeqrf LAPACK_GLOBAL(sgeqrf,SGEQRF)
+#define LAPACK_dgeqrf LAPACK_GLOBAL(dgeqrf,DGEQRF)
+#define LAPACK_cgeqrf LAPACK_GLOBAL(cgeqrf,CGEQRF)
+#define LAPACK_zgeqrf LAPACK_GLOBAL(zgeqrf,ZGEQRF)
+#define LAPACK_sgeqpf LAPACK_GLOBAL(sgeqpf,SGEQPF)
+#define LAPACK_dgeqpf LAPACK_GLOBAL(dgeqpf,DGEQPF)
+#define LAPACK_cgeqpf LAPACK_GLOBAL(cgeqpf,CGEQPF)
+#define LAPACK_zgeqpf LAPACK_GLOBAL(zgeqpf,ZGEQPF)
+#define LAPACK_sgeqp3 LAPACK_GLOBAL(sgeqp3,SGEQP3)
+#define LAPACK_dgeqp3 LAPACK_GLOBAL(dgeqp3,DGEQP3)
+#define LAPACK_cgeqp3 LAPACK_GLOBAL(cgeqp3,CGEQP3)
+#define LAPACK_zgeqp3 LAPACK_GLOBAL(zgeqp3,ZGEQP3)
+#define LAPACK_sorgqr LAPACK_GLOBAL(sorgqr,SORGQR)
+#define LAPACK_dorgqr LAPACK_GLOBAL(dorgqr,DORGQR)
+#define LAPACK_sormqr LAPACK_GLOBAL(sormqr,SORMQR)
+#define LAPACK_dormqr LAPACK_GLOBAL(dormqr,DORMQR)
+#define LAPACK_cungqr LAPACK_GLOBAL(cungqr,CUNGQR)
+#define LAPACK_zungqr LAPACK_GLOBAL(zungqr,ZUNGQR)
+#define LAPACK_cunmqr LAPACK_GLOBAL(cunmqr,CUNMQR)
+#define LAPACK_zunmqr LAPACK_GLOBAL(zunmqr,ZUNMQR)
+#define LAPACK_sgelqf LAPACK_GLOBAL(sgelqf,SGELQF)
+#define LAPACK_dgelqf LAPACK_GLOBAL(dgelqf,DGELQF)
+#define LAPACK_cgelqf LAPACK_GLOBAL(cgelqf,CGELQF)
+#define LAPACK_zgelqf LAPACK_GLOBAL(zgelqf,ZGELQF)
+#define LAPACK_sorglq LAPACK_GLOBAL(sorglq,SORGLQ)
+#define LAPACK_dorglq LAPACK_GLOBAL(dorglq,DORGLQ)
+#define LAPACK_sormlq LAPACK_GLOBAL(sormlq,SORMLQ)
+#define LAPACK_dormlq LAPACK_GLOBAL(dormlq,DORMLQ)
+#define LAPACK_cunglq LAPACK_GLOBAL(cunglq,CUNGLQ)
+#define LAPACK_zunglq LAPACK_GLOBAL(zunglq,ZUNGLQ)
+#define LAPACK_cunmlq LAPACK_GLOBAL(cunmlq,CUNMLQ)
+#define LAPACK_zunmlq LAPACK_GLOBAL(zunmlq,ZUNMLQ)
+#define LAPACK_sgeqlf LAPACK_GLOBAL(sgeqlf,SGEQLF)
+#define LAPACK_dgeqlf LAPACK_GLOBAL(dgeqlf,DGEQLF)
+#define LAPACK_cgeqlf LAPACK_GLOBAL(cgeqlf,CGEQLF)
+#define LAPACK_zgeqlf LAPACK_GLOBAL(zgeqlf,ZGEQLF)
+#define LAPACK_sorgql LAPACK_GLOBAL(sorgql,SORGQL)
+#define LAPACK_dorgql LAPACK_GLOBAL(dorgql,DORGQL)
+#define LAPACK_cungql LAPACK_GLOBAL(cungql,CUNGQL)
+#define LAPACK_zungql LAPACK_GLOBAL(zungql,ZUNGQL)
+#define LAPACK_sormql LAPACK_GLOBAL(sormql,SORMQL)
+#define LAPACK_dormql LAPACK_GLOBAL(dormql,DORMQL)
+#define LAPACK_cunmql LAPACK_GLOBAL(cunmql,CUNMQL)
+#define LAPACK_zunmql LAPACK_GLOBAL(zunmql,ZUNMQL)
+#define LAPACK_sgerqf LAPACK_GLOBAL(sgerqf,SGERQF)
+#define LAPACK_dgerqf LAPACK_GLOBAL(dgerqf,DGERQF)
+#define LAPACK_cgerqf LAPACK_GLOBAL(cgerqf,CGERQF)
+#define LAPACK_zgerqf LAPACK_GLOBAL(zgerqf,ZGERQF)
+#define LAPACK_sorgrq LAPACK_GLOBAL(sorgrq,SORGRQ)
+#define LAPACK_dorgrq LAPACK_GLOBAL(dorgrq,DORGRQ)
+#define LAPACK_cungrq LAPACK_GLOBAL(cungrq,CUNGRQ)
+#define LAPACK_zungrq LAPACK_GLOBAL(zungrq,ZUNGRQ)
+#define LAPACK_sormrq LAPACK_GLOBAL(sormrq,SORMRQ)
+#define LAPACK_dormrq LAPACK_GLOBAL(dormrq,DORMRQ)
+#define LAPACK_cunmrq LAPACK_GLOBAL(cunmrq,CUNMRQ)
+#define LAPACK_zunmrq LAPACK_GLOBAL(zunmrq,ZUNMRQ)
+#define LAPACK_stzrzf LAPACK_GLOBAL(stzrzf,STZRZF)
+#define LAPACK_dtzrzf LAPACK_GLOBAL(dtzrzf,DTZRZF)
+#define LAPACK_ctzrzf LAPACK_GLOBAL(ctzrzf,CTZRZF)
+#define LAPACK_ztzrzf LAPACK_GLOBAL(ztzrzf,ZTZRZF)
+#define LAPACK_sormrz LAPACK_GLOBAL(sormrz,SORMRZ)
+#define LAPACK_dormrz LAPACK_GLOBAL(dormrz,DORMRZ)
+#define LAPACK_cunmrz LAPACK_GLOBAL(cunmrz,CUNMRZ)
+#define LAPACK_zunmrz LAPACK_GLOBAL(zunmrz,ZUNMRZ)
+#define LAPACK_sggqrf LAPACK_GLOBAL(sggqrf,SGGQRF)
+#define LAPACK_dggqrf LAPACK_GLOBAL(dggqrf,DGGQRF)
+#define LAPACK_cggqrf LAPACK_GLOBAL(cggqrf,CGGQRF)
+#define LAPACK_zggqrf LAPACK_GLOBAL(zggqrf,ZGGQRF)
+#define LAPACK_sggrqf LAPACK_GLOBAL(sggrqf,SGGRQF)
+#define LAPACK_dggrqf LAPACK_GLOBAL(dggrqf,DGGRQF)
+#define LAPACK_cggrqf LAPACK_GLOBAL(cggrqf,CGGRQF)
+#define LAPACK_zggrqf LAPACK_GLOBAL(zggrqf,ZGGRQF)
+#define LAPACK_sgebrd LAPACK_GLOBAL(sgebrd,SGEBRD)
+#define LAPACK_dgebrd LAPACK_GLOBAL(dgebrd,DGEBRD)
+#define LAPACK_cgebrd LAPACK_GLOBAL(cgebrd,CGEBRD)
+#define LAPACK_zgebrd LAPACK_GLOBAL(zgebrd,ZGEBRD)
+#define LAPACK_sgbbrd LAPACK_GLOBAL(sgbbrd,SGBBRD)
+#define LAPACK_dgbbrd LAPACK_GLOBAL(dgbbrd,DGBBRD)
+#define LAPACK_cgbbrd LAPACK_GLOBAL(cgbbrd,CGBBRD)
+#define LAPACK_zgbbrd LAPACK_GLOBAL(zgbbrd,ZGBBRD)
+#define LAPACK_sorgbr LAPACK_GLOBAL(sorgbr,SORGBR)
+#define LAPACK_dorgbr LAPACK_GLOBAL(dorgbr,DORGBR)
+#define LAPACK_sormbr LAPACK_GLOBAL(sormbr,SORMBR)
+#define LAPACK_dormbr LAPACK_GLOBAL(dormbr,DORMBR)
+#define LAPACK_cungbr LAPACK_GLOBAL(cungbr,CUNGBR)
+#define LAPACK_zungbr LAPACK_GLOBAL(zungbr,ZUNGBR)
+#define LAPACK_cunmbr LAPACK_GLOBAL(cunmbr,CUNMBR)
+#define LAPACK_zunmbr LAPACK_GLOBAL(zunmbr,ZUNMBR)
+#define LAPACK_sbdsqr LAPACK_GLOBAL(sbdsqr,SBDSQR)
+#define LAPACK_dbdsqr LAPACK_GLOBAL(dbdsqr,DBDSQR)
+#define LAPACK_cbdsqr LAPACK_GLOBAL(cbdsqr,CBDSQR)
+#define LAPACK_zbdsqr LAPACK_GLOBAL(zbdsqr,ZBDSQR)
+#define LAPACK_sbdsdc LAPACK_GLOBAL(sbdsdc,SBDSDC)
+#define LAPACK_dbdsdc LAPACK_GLOBAL(dbdsdc,DBDSDC)
+#define LAPACK_ssytrd LAPACK_GLOBAL(ssytrd,SSYTRD)
+#define LAPACK_dsytrd LAPACK_GLOBAL(dsytrd,DSYTRD)
+#define LAPACK_sorgtr LAPACK_GLOBAL(sorgtr,SORGTR)
+#define LAPACK_dorgtr LAPACK_GLOBAL(dorgtr,DORGTR)
+#define LAPACK_sormtr LAPACK_GLOBAL(sormtr,SORMTR)
+#define LAPACK_dormtr LAPACK_GLOBAL(dormtr,DORMTR)
+#define LAPACK_chetrd LAPACK_GLOBAL(chetrd,CHETRD)
+#define LAPACK_zhetrd LAPACK_GLOBAL(zhetrd,ZHETRD)
+#define LAPACK_cungtr LAPACK_GLOBAL(cungtr,CUNGTR)
+#define LAPACK_zungtr LAPACK_GLOBAL(zungtr,ZUNGTR)
+#define LAPACK_cunmtr LAPACK_GLOBAL(cunmtr,CUNMTR)
+#define LAPACK_zunmtr LAPACK_GLOBAL(zunmtr,ZUNMTR)
+#define LAPACK_ssptrd LAPACK_GLOBAL(ssptrd,SSPTRD)
+#define LAPACK_dsptrd LAPACK_GLOBAL(dsptrd,DSPTRD)
+#define LAPACK_sopgtr LAPACK_GLOBAL(sopgtr,SOPGTR)
+#define LAPACK_dopgtr LAPACK_GLOBAL(dopgtr,DOPGTR)
+#define LAPACK_sopmtr LAPACK_GLOBAL(sopmtr,SOPMTR)
+#define LAPACK_dopmtr LAPACK_GLOBAL(dopmtr,DOPMTR)
+#define LAPACK_chptrd LAPACK_GLOBAL(chptrd,CHPTRD)
+#define LAPACK_zhptrd LAPACK_GLOBAL(zhptrd,ZHPTRD)
+#define LAPACK_cupgtr LAPACK_GLOBAL(cupgtr,CUPGTR)
+#define LAPACK_zupgtr LAPACK_GLOBAL(zupgtr,ZUPGTR)
+#define LAPACK_cupmtr LAPACK_GLOBAL(cupmtr,CUPMTR)
+#define LAPACK_zupmtr LAPACK_GLOBAL(zupmtr,ZUPMTR)
+#define LAPACK_ssbtrd LAPACK_GLOBAL(ssbtrd,SSBTRD)
+#define LAPACK_dsbtrd LAPACK_GLOBAL(dsbtrd,DSBTRD)
+#define LAPACK_chbtrd LAPACK_GLOBAL(chbtrd,CHBTRD)
+#define LAPACK_zhbtrd LAPACK_GLOBAL(zhbtrd,ZHBTRD)
+#define LAPACK_ssterf LAPACK_GLOBAL(ssterf,SSTERF)
+#define LAPACK_dsterf LAPACK_GLOBAL(dsterf,DSTERF)
+#define LAPACK_ssteqr LAPACK_GLOBAL(ssteqr,SSTEQR)
+#define LAPACK_dsteqr LAPACK_GLOBAL(dsteqr,DSTEQR)
+#define LAPACK_csteqr LAPACK_GLOBAL(csteqr,CSTEQR)
+#define LAPACK_zsteqr LAPACK_GLOBAL(zsteqr,ZSTEQR)
+#define LAPACK_sstemr LAPACK_GLOBAL(sstemr,SSTEMR)
+#define LAPACK_dstemr LAPACK_GLOBAL(dstemr,DSTEMR)
+#define LAPACK_cstemr LAPACK_GLOBAL(cstemr,CSTEMR)
+#define LAPACK_zstemr LAPACK_GLOBAL(zstemr,ZSTEMR)
+#define LAPACK_sstedc LAPACK_GLOBAL(sstedc,SSTEDC)
+#define LAPACK_dstedc LAPACK_GLOBAL(dstedc,DSTEDC)
+#define LAPACK_cstedc LAPACK_GLOBAL(cstedc,CSTEDC)
+#define LAPACK_zstedc LAPACK_GLOBAL(zstedc,ZSTEDC)
+#define LAPACK_sstegr LAPACK_GLOBAL(sstegr,SSTEGR)
+#define LAPACK_dstegr LAPACK_GLOBAL(dstegr,DSTEGR)
+#define LAPACK_cstegr LAPACK_GLOBAL(cstegr,CSTEGR)
+#define LAPACK_zstegr LAPACK_GLOBAL(zstegr,ZSTEGR)
+#define LAPACK_spteqr LAPACK_GLOBAL(spteqr,SPTEQR)
+#define LAPACK_dpteqr LAPACK_GLOBAL(dpteqr,DPTEQR)
+#define LAPACK_cpteqr LAPACK_GLOBAL(cpteqr,CPTEQR)
+#define LAPACK_zpteqr LAPACK_GLOBAL(zpteqr,ZPTEQR)
+#define LAPACK_sstebz LAPACK_GLOBAL(sstebz,SSTEBZ)
+#define LAPACK_dstebz LAPACK_GLOBAL(dstebz,DSTEBZ)
+#define LAPACK_sstein LAPACK_GLOBAL(sstein,SSTEIN)
+#define LAPACK_dstein LAPACK_GLOBAL(dstein,DSTEIN)
+#define LAPACK_cstein LAPACK_GLOBAL(cstein,CSTEIN)
+#define LAPACK_zstein LAPACK_GLOBAL(zstein,ZSTEIN)
+#define LAPACK_sdisna LAPACK_GLOBAL(sdisna,SDISNA)
+#define LAPACK_ddisna LAPACK_GLOBAL(ddisna,DDISNA)
+#define LAPACK_ssygst LAPACK_GLOBAL(ssygst,SSYGST)
+#define LAPACK_dsygst LAPACK_GLOBAL(dsygst,DSYGST)
+#define LAPACK_chegst LAPACK_GLOBAL(chegst,CHEGST)
+#define LAPACK_zhegst LAPACK_GLOBAL(zhegst,ZHEGST)
+#define LAPACK_sspgst LAPACK_GLOBAL(sspgst,SSPGST)
+#define LAPACK_dspgst LAPACK_GLOBAL(dspgst,DSPGST)
+#define LAPACK_chpgst LAPACK_GLOBAL(chpgst,CHPGST)
+#define LAPACK_zhpgst LAPACK_GLOBAL(zhpgst,ZHPGST)
+#define LAPACK_ssbgst LAPACK_GLOBAL(ssbgst,SSBGST)
+#define LAPACK_dsbgst LAPACK_GLOBAL(dsbgst,DSBGST)
+#define LAPACK_chbgst LAPACK_GLOBAL(chbgst,CHBGST)
+#define LAPACK_zhbgst LAPACK_GLOBAL(zhbgst,ZHBGST)
+#define LAPACK_spbstf LAPACK_GLOBAL(spbstf,SPBSTF)
+#define LAPACK_dpbstf LAPACK_GLOBAL(dpbstf,DPBSTF)
+#define LAPACK_cpbstf LAPACK_GLOBAL(cpbstf,CPBSTF)
+#define LAPACK_zpbstf LAPACK_GLOBAL(zpbstf,ZPBSTF)
+#define LAPACK_sgehrd LAPACK_GLOBAL(sgehrd,SGEHRD)
+#define LAPACK_dgehrd LAPACK_GLOBAL(dgehrd,DGEHRD)
+#define LAPACK_cgehrd LAPACK_GLOBAL(cgehrd,CGEHRD)
+#define LAPACK_zgehrd LAPACK_GLOBAL(zgehrd,ZGEHRD)
+#define LAPACK_sorghr LAPACK_GLOBAL(sorghr,SORGHR)
+#define LAPACK_dorghr LAPACK_GLOBAL(dorghr,DORGHR)
+#define LAPACK_sormhr LAPACK_GLOBAL(sormhr,SORMHR)
+#define LAPACK_dormhr LAPACK_GLOBAL(dormhr,DORMHR)
+#define LAPACK_cunghr LAPACK_GLOBAL(cunghr,CUNGHR)
+#define LAPACK_zunghr LAPACK_GLOBAL(zunghr,ZUNGHR)
+#define LAPACK_cunmhr LAPACK_GLOBAL(cunmhr,CUNMHR)
+#define LAPACK_zunmhr LAPACK_GLOBAL(zunmhr,ZUNMHR)
+#define LAPACK_sgebal LAPACK_GLOBAL(sgebal,SGEBAL)
+#define LAPACK_dgebal LAPACK_GLOBAL(dgebal,DGEBAL)
+#define LAPACK_cgebal LAPACK_GLOBAL(cgebal,CGEBAL)
+#define LAPACK_zgebal LAPACK_GLOBAL(zgebal,ZGEBAL)
+#define LAPACK_sgebak LAPACK_GLOBAL(sgebak,SGEBAK)
+#define LAPACK_dgebak LAPACK_GLOBAL(dgebak,DGEBAK)
+#define LAPACK_cgebak LAPACK_GLOBAL(cgebak,CGEBAK)
+#define LAPACK_zgebak LAPACK_GLOBAL(zgebak,ZGEBAK)
+#define LAPACK_shseqr LAPACK_GLOBAL(shseqr,SHSEQR)
+#define LAPACK_dhseqr LAPACK_GLOBAL(dhseqr,DHSEQR)
+#define LAPACK_chseqr LAPACK_GLOBAL(chseqr,CHSEQR)
+#define LAPACK_zhseqr LAPACK_GLOBAL(zhseqr,ZHSEQR)
+#define LAPACK_shsein LAPACK_GLOBAL(shsein,SHSEIN)
+#define LAPACK_dhsein LAPACK_GLOBAL(dhsein,DHSEIN)
+#define LAPACK_chsein LAPACK_GLOBAL(chsein,CHSEIN)
+#define LAPACK_zhsein LAPACK_GLOBAL(zhsein,ZHSEIN)
+#define LAPACK_strevc LAPACK_GLOBAL(strevc,STREVC)
+#define LAPACK_dtrevc LAPACK_GLOBAL(dtrevc,DTREVC)
+#define LAPACK_ctrevc LAPACK_GLOBAL(ctrevc,CTREVC)
+#define LAPACK_ztrevc LAPACK_GLOBAL(ztrevc,ZTREVC)
+#define LAPACK_strsna LAPACK_GLOBAL(strsna,STRSNA)
+#define LAPACK_dtrsna LAPACK_GLOBAL(dtrsna,DTRSNA)
+#define LAPACK_ctrsna LAPACK_GLOBAL(ctrsna,CTRSNA)
+#define LAPACK_ztrsna LAPACK_GLOBAL(ztrsna,ZTRSNA)
+#define LAPACK_strexc LAPACK_GLOBAL(strexc,STREXC)
+#define LAPACK_dtrexc LAPACK_GLOBAL(dtrexc,DTREXC)
+#define LAPACK_ctrexc LAPACK_GLOBAL(ctrexc,CTREXC)
+#define LAPACK_ztrexc LAPACK_GLOBAL(ztrexc,ZTREXC)
+#define LAPACK_strsen LAPACK_GLOBAL(strsen,STRSEN)
+#define LAPACK_dtrsen LAPACK_GLOBAL(dtrsen,DTRSEN)
+#define LAPACK_ctrsen LAPACK_GLOBAL(ctrsen,CTRSEN)
+#define LAPACK_ztrsen LAPACK_GLOBAL(ztrsen,ZTRSEN)
+#define LAPACK_strsyl LAPACK_GLOBAL(strsyl,STRSYL)
+#define LAPACK_dtrsyl LAPACK_GLOBAL(dtrsyl,DTRSYL)
+#define LAPACK_ctrsyl LAPACK_GLOBAL(ctrsyl,CTRSYL)
+#define LAPACK_ztrsyl LAPACK_GLOBAL(ztrsyl,ZTRSYL)
+#define LAPACK_sgghrd LAPACK_GLOBAL(sgghrd,SGGHRD)
+#define LAPACK_dgghrd LAPACK_GLOBAL(dgghrd,DGGHRD)
+#define LAPACK_cgghrd LAPACK_GLOBAL(cgghrd,CGGHRD)
+#define LAPACK_zgghrd LAPACK_GLOBAL(zgghrd,ZGGHRD)
+#define LAPACK_sggbal LAPACK_GLOBAL(sggbal,SGGBAL)
+#define LAPACK_dggbal LAPACK_GLOBAL(dggbal,DGGBAL)
+#define LAPACK_cggbal LAPACK_GLOBAL(cggbal,CGGBAL)
+#define LAPACK_zggbal LAPACK_GLOBAL(zggbal,ZGGBAL)
+#define LAPACK_sggbak LAPACK_GLOBAL(sggbak,SGGBAK)
+#define LAPACK_dggbak LAPACK_GLOBAL(dggbak,DGGBAK)
+#define LAPACK_cggbak LAPACK_GLOBAL(cggbak,CGGBAK)
+#define LAPACK_zggbak LAPACK_GLOBAL(zggbak,ZGGBAK)
+#define LAPACK_shgeqz LAPACK_GLOBAL(shgeqz,SHGEQZ)
+#define LAPACK_dhgeqz LAPACK_GLOBAL(dhgeqz,DHGEQZ)
+#define LAPACK_chgeqz LAPACK_GLOBAL(chgeqz,CHGEQZ)
+#define LAPACK_zhgeqz LAPACK_GLOBAL(zhgeqz,ZHGEQZ)
+#define LAPACK_stgevc LAPACK_GLOBAL(stgevc,STGEVC)
+#define LAPACK_dtgevc LAPACK_GLOBAL(dtgevc,DTGEVC)
+#define LAPACK_ctgevc LAPACK_GLOBAL(ctgevc,CTGEVC)
+#define LAPACK_ztgevc LAPACK_GLOBAL(ztgevc,ZTGEVC)
+#define LAPACK_stgexc LAPACK_GLOBAL(stgexc,STGEXC)
+#define LAPACK_dtgexc LAPACK_GLOBAL(dtgexc,DTGEXC)
+#define LAPACK_ctgexc LAPACK_GLOBAL(ctgexc,CTGEXC)
+#define LAPACK_ztgexc LAPACK_GLOBAL(ztgexc,ZTGEXC)
+#define LAPACK_stgsen LAPACK_GLOBAL(stgsen,STGSEN)
+#define LAPACK_dtgsen LAPACK_GLOBAL(dtgsen,DTGSEN)
+#define LAPACK_ctgsen LAPACK_GLOBAL(ctgsen,CTGSEN)
+#define LAPACK_ztgsen LAPACK_GLOBAL(ztgsen,ZTGSEN)
+#define LAPACK_stgsyl LAPACK_GLOBAL(stgsyl,STGSYL)
+#define LAPACK_dtgsyl LAPACK_GLOBAL(dtgsyl,DTGSYL)
+#define LAPACK_ctgsyl LAPACK_GLOBAL(ctgsyl,CTGSYL)
+#define LAPACK_ztgsyl LAPACK_GLOBAL(ztgsyl,ZTGSYL)
+#define LAPACK_stgsna LAPACK_GLOBAL(stgsna,STGSNA)
+#define LAPACK_dtgsna LAPACK_GLOBAL(dtgsna,DTGSNA)
+#define LAPACK_ctgsna LAPACK_GLOBAL(ctgsna,CTGSNA)
+#define LAPACK_ztgsna LAPACK_GLOBAL(ztgsna,ZTGSNA)
+#define LAPACK_sggsvp LAPACK_GLOBAL(sggsvp,SGGSVP)
+#define LAPACK_dggsvp LAPACK_GLOBAL(dggsvp,DGGSVP)
+#define LAPACK_cggsvp LAPACK_GLOBAL(cggsvp,CGGSVP)
+#define LAPACK_zggsvp LAPACK_GLOBAL(zggsvp,ZGGSVP)
+#define LAPACK_stgsja LAPACK_GLOBAL(stgsja,STGSJA)
+#define LAPACK_dtgsja LAPACK_GLOBAL(dtgsja,DTGSJA)
+#define LAPACK_ctgsja LAPACK_GLOBAL(ctgsja,CTGSJA)
+#define LAPACK_ztgsja LAPACK_GLOBAL(ztgsja,ZTGSJA)
+#define LAPACK_sgels LAPACK_GLOBAL(sgels,SGELS)
+#define LAPACK_dgels LAPACK_GLOBAL(dgels,DGELS)
+#define LAPACK_cgels LAPACK_GLOBAL(cgels,CGELS)
+#define LAPACK_zgels LAPACK_GLOBAL(zgels,ZGELS)
+#define LAPACK_sgelsy LAPACK_GLOBAL(sgelsy,SGELSY)
+#define LAPACK_dgelsy LAPACK_GLOBAL(dgelsy,DGELSY)
+#define LAPACK_cgelsy LAPACK_GLOBAL(cgelsy,CGELSY)
+#define LAPACK_zgelsy LAPACK_GLOBAL(zgelsy,ZGELSY)
+#define LAPACK_sgelss LAPACK_GLOBAL(sgelss,SGELSS)
+#define LAPACK_dgelss LAPACK_GLOBAL(dgelss,DGELSS)
+#define LAPACK_cgelss LAPACK_GLOBAL(cgelss,CGELSS)
+#define LAPACK_zgelss LAPACK_GLOBAL(zgelss,ZGELSS)
+#define LAPACK_sgelsd LAPACK_GLOBAL(sgelsd,SGELSD)
+#define LAPACK_dgelsd LAPACK_GLOBAL(dgelsd,DGELSD)
+#define LAPACK_cgelsd LAPACK_GLOBAL(cgelsd,CGELSD)
+#define LAPACK_zgelsd LAPACK_GLOBAL(zgelsd,ZGELSD)
+#define LAPACK_sgglse LAPACK_GLOBAL(sgglse,SGGLSE)
+#define LAPACK_dgglse LAPACK_GLOBAL(dgglse,DGGLSE)
+#define LAPACK_cgglse LAPACK_GLOBAL(cgglse,CGGLSE)
+#define LAPACK_zgglse LAPACK_GLOBAL(zgglse,ZGGLSE)
+#define LAPACK_sggglm LAPACK_GLOBAL(sggglm,SGGGLM)
+#define LAPACK_dggglm LAPACK_GLOBAL(dggglm,DGGGLM)
+#define LAPACK_cggglm LAPACK_GLOBAL(cggglm,CGGGLM)
+#define LAPACK_zggglm LAPACK_GLOBAL(zggglm,ZGGGLM)
+#define LAPACK_ssyev LAPACK_GLOBAL(ssyev,SSYEV)
+#define LAPACK_dsyev LAPACK_GLOBAL(dsyev,DSYEV)
+#define LAPACK_cheev LAPACK_GLOBAL(cheev,CHEEV)
+#define LAPACK_zheev LAPACK_GLOBAL(zheev,ZHEEV)
+#define LAPACK_ssyevd LAPACK_GLOBAL(ssyevd,SSYEVD)
+#define LAPACK_dsyevd LAPACK_GLOBAL(dsyevd,DSYEVD)
+#define LAPACK_cheevd LAPACK_GLOBAL(cheevd,CHEEVD)
+#define LAPACK_zheevd LAPACK_GLOBAL(zheevd,ZHEEVD)
+#define LAPACK_ssyevx LAPACK_GLOBAL(ssyevx,SSYEVX)
+#define LAPACK_dsyevx LAPACK_GLOBAL(dsyevx,DSYEVX)
+#define LAPACK_cheevx LAPACK_GLOBAL(cheevx,CHEEVX)
+#define LAPACK_zheevx LAPACK_GLOBAL(zheevx,ZHEEVX)
+#define LAPACK_ssyevr LAPACK_GLOBAL(ssyevr,SSYEVR)
+#define LAPACK_dsyevr LAPACK_GLOBAL(dsyevr,DSYEVR)
+#define LAPACK_cheevr LAPACK_GLOBAL(cheevr,CHEEVR)
+#define LAPACK_zheevr LAPACK_GLOBAL(zheevr,ZHEEVR)
+#define LAPACK_sspev LAPACK_GLOBAL(sspev,SSPEV)
+#define LAPACK_dspev LAPACK_GLOBAL(dspev,DSPEV)
+#define LAPACK_chpev LAPACK_GLOBAL(chpev,CHPEV)
+#define LAPACK_zhpev LAPACK_GLOBAL(zhpev,ZHPEV)
+#define LAPACK_sspevd LAPACK_GLOBAL(sspevd,SSPEVD)
+#define LAPACK_dspevd LAPACK_GLOBAL(dspevd,DSPEVD)
+#define LAPACK_chpevd LAPACK_GLOBAL(chpevd,CHPEVD)
+#define LAPACK_zhpevd LAPACK_GLOBAL(zhpevd,ZHPEVD)
+#define LAPACK_sspevx LAPACK_GLOBAL(sspevx,SSPEVX)
+#define LAPACK_dspevx LAPACK_GLOBAL(dspevx,DSPEVX)
+#define LAPACK_chpevx LAPACK_GLOBAL(chpevx,CHPEVX)
+#define LAPACK_zhpevx LAPACK_GLOBAL(zhpevx,ZHPEVX)
+#define LAPACK_ssbev LAPACK_GLOBAL(ssbev,SSBEV)
+#define LAPACK_dsbev LAPACK_GLOBAL(dsbev,DSBEV)
+#define LAPACK_chbev LAPACK_GLOBAL(chbev,CHBEV)
+#define LAPACK_zhbev LAPACK_GLOBAL(zhbev,ZHBEV)
+#define LAPACK_ssbevd LAPACK_GLOBAL(ssbevd,SSBEVD)
+#define LAPACK_dsbevd LAPACK_GLOBAL(dsbevd,DSBEVD)
+#define LAPACK_chbevd LAPACK_GLOBAL(chbevd,CHBEVD)
+#define LAPACK_zhbevd LAPACK_GLOBAL(zhbevd,ZHBEVD)
+#define LAPACK_ssbevx LAPACK_GLOBAL(ssbevx,SSBEVX)
+#define LAPACK_dsbevx LAPACK_GLOBAL(dsbevx,DSBEVX)
+#define LAPACK_chbevx LAPACK_GLOBAL(chbevx,CHBEVX)
+#define LAPACK_zhbevx LAPACK_GLOBAL(zhbevx,ZHBEVX)
+#define LAPACK_sstev LAPACK_GLOBAL(sstev,SSTEV)
+#define LAPACK_dstev LAPACK_GLOBAL(dstev,DSTEV)
+#define LAPACK_sstevd LAPACK_GLOBAL(sstevd,SSTEVD)
+#define LAPACK_dstevd LAPACK_GLOBAL(dstevd,DSTEVD)
+#define LAPACK_sstevx LAPACK_GLOBAL(sstevx,SSTEVX)
+#define LAPACK_dstevx LAPACK_GLOBAL(dstevx,DSTEVX)
+#define LAPACK_sstevr LAPACK_GLOBAL(sstevr,SSTEVR)
+#define LAPACK_dstevr LAPACK_GLOBAL(dstevr,DSTEVR)
+#define LAPACK_sgees LAPACK_GLOBAL(sgees,SGEES)
+#define LAPACK_dgees LAPACK_GLOBAL(dgees,DGEES)
+#define LAPACK_cgees LAPACK_GLOBAL(cgees,CGEES)
+#define LAPACK_zgees LAPACK_GLOBAL(zgees,ZGEES)
+#define LAPACK_sgeesx LAPACK_GLOBAL(sgeesx,SGEESX)
+#define LAPACK_dgeesx LAPACK_GLOBAL(dgeesx,DGEESX)
+#define LAPACK_cgeesx LAPACK_GLOBAL(cgeesx,CGEESX)
+#define LAPACK_zgeesx LAPACK_GLOBAL(zgeesx,ZGEESX)
+#define LAPACK_sgeev LAPACK_GLOBAL(sgeev,SGEEV)
+#define LAPACK_dgeev LAPACK_GLOBAL(dgeev,DGEEV)
+#define LAPACK_cgeev LAPACK_GLOBAL(cgeev,CGEEV)
+#define LAPACK_zgeev LAPACK_GLOBAL(zgeev,ZGEEV)
+#define LAPACK_sgeevx LAPACK_GLOBAL(sgeevx,SGEEVX)
+#define LAPACK_dgeevx LAPACK_GLOBAL(dgeevx,DGEEVX)
+#define LAPACK_cgeevx LAPACK_GLOBAL(cgeevx,CGEEVX)
+#define LAPACK_zgeevx LAPACK_GLOBAL(zgeevx,ZGEEVX)
+#define LAPACK_sgesvd LAPACK_GLOBAL(sgesvd,SGESVD)
+#define LAPACK_dgesvd LAPACK_GLOBAL(dgesvd,DGESVD)
+#define LAPACK_cgesvd LAPACK_GLOBAL(cgesvd,CGESVD)
+#define LAPACK_zgesvd LAPACK_GLOBAL(zgesvd,ZGESVD)
+#define LAPACK_sgesdd LAPACK_GLOBAL(sgesdd,SGESDD)
+#define LAPACK_dgesdd LAPACK_GLOBAL(dgesdd,DGESDD)
+#define LAPACK_cgesdd LAPACK_GLOBAL(cgesdd,CGESDD)
+#define LAPACK_zgesdd LAPACK_GLOBAL(zgesdd,ZGESDD)
+#define LAPACK_dgejsv LAPACK_GLOBAL(dgejsv,DGEJSV)
+#define LAPACK_sgejsv LAPACK_GLOBAL(sgejsv,SGEJSV)
+#define LAPACK_dgesvj LAPACK_GLOBAL(dgesvj,DGESVJ)
+#define LAPACK_sgesvj LAPACK_GLOBAL(sgesvj,SGESVJ)
+#define LAPACK_sggsvd LAPACK_GLOBAL(sggsvd,SGGSVD)
+#define LAPACK_dggsvd LAPACK_GLOBAL(dggsvd,DGGSVD)
+#define LAPACK_cggsvd LAPACK_GLOBAL(cggsvd,CGGSVD)
+#define LAPACK_zggsvd LAPACK_GLOBAL(zggsvd,ZGGSVD)
+#define LAPACK_ssygv LAPACK_GLOBAL(ssygv,SSYGV)
+#define LAPACK_dsygv LAPACK_GLOBAL(dsygv,DSYGV)
+#define LAPACK_chegv LAPACK_GLOBAL(chegv,CHEGV)
+#define LAPACK_zhegv LAPACK_GLOBAL(zhegv,ZHEGV)
+#define LAPACK_ssygvd LAPACK_GLOBAL(ssygvd,SSYGVD)
+#define LAPACK_dsygvd LAPACK_GLOBAL(dsygvd,DSYGVD)
+#define LAPACK_chegvd LAPACK_GLOBAL(chegvd,CHEGVD)
+#define LAPACK_zhegvd LAPACK_GLOBAL(zhegvd,ZHEGVD)
+#define LAPACK_ssygvx LAPACK_GLOBAL(ssygvx,SSYGVX)
+#define LAPACK_dsygvx LAPACK_GLOBAL(dsygvx,DSYGVX)
+#define LAPACK_chegvx LAPACK_GLOBAL(chegvx,CHEGVX)
+#define LAPACK_zhegvx LAPACK_GLOBAL(zhegvx,ZHEGVX)
+#define LAPACK_sspgv LAPACK_GLOBAL(sspgv,SSPGV)
+#define LAPACK_dspgv LAPACK_GLOBAL(dspgv,DSPGV)
+#define LAPACK_chpgv LAPACK_GLOBAL(chpgv,CHPGV)
+#define LAPACK_zhpgv LAPACK_GLOBAL(zhpgv,ZHPGV)
+#define LAPACK_sspgvd LAPACK_GLOBAL(sspgvd,SSPGVD)
+#define LAPACK_dspgvd LAPACK_GLOBAL(dspgvd,DSPGVD)
+#define LAPACK_chpgvd LAPACK_GLOBAL(chpgvd,CHPGVD)
+#define LAPACK_zhpgvd LAPACK_GLOBAL(zhpgvd,ZHPGVD)
+#define LAPACK_sspgvx LAPACK_GLOBAL(sspgvx,SSPGVX)
+#define LAPACK_dspgvx LAPACK_GLOBAL(dspgvx,DSPGVX)
+#define LAPACK_chpgvx LAPACK_GLOBAL(chpgvx,CHPGVX)
+#define LAPACK_zhpgvx LAPACK_GLOBAL(zhpgvx,ZHPGVX)
+#define LAPACK_ssbgv LAPACK_GLOBAL(ssbgv,SSBGV)
+#define LAPACK_dsbgv LAPACK_GLOBAL(dsbgv,DSBGV)
+#define LAPACK_chbgv LAPACK_GLOBAL(chbgv,CHBGV)
+#define LAPACK_zhbgv LAPACK_GLOBAL(zhbgv,ZHBGV)
+#define LAPACK_ssbgvd LAPACK_GLOBAL(ssbgvd,SSBGVD)
+#define LAPACK_dsbgvd LAPACK_GLOBAL(dsbgvd,DSBGVD)
+#define LAPACK_chbgvd LAPACK_GLOBAL(chbgvd,CHBGVD)
+#define LAPACK_zhbgvd LAPACK_GLOBAL(zhbgvd,ZHBGVD)
+#define LAPACK_ssbgvx LAPACK_GLOBAL(ssbgvx,SSBGVX)
+#define LAPACK_dsbgvx LAPACK_GLOBAL(dsbgvx,DSBGVX)
+#define LAPACK_chbgvx LAPACK_GLOBAL(chbgvx,CHBGVX)
+#define LAPACK_zhbgvx LAPACK_GLOBAL(zhbgvx,ZHBGVX)
+#define LAPACK_sgges LAPACK_GLOBAL(sgges,SGGES)
+#define LAPACK_dgges LAPACK_GLOBAL(dgges,DGGES)
+#define LAPACK_cgges LAPACK_GLOBAL(cgges,CGGES)
+#define LAPACK_zgges LAPACK_GLOBAL(zgges,ZGGES)
+#define LAPACK_sggesx LAPACK_GLOBAL(sggesx,SGGESX)
+#define LAPACK_dggesx LAPACK_GLOBAL(dggesx,DGGESX)
+#define LAPACK_cggesx LAPACK_GLOBAL(cggesx,CGGESX)
+#define LAPACK_zggesx LAPACK_GLOBAL(zggesx,ZGGESX)
+#define LAPACK_sggev LAPACK_GLOBAL(sggev,SGGEV)
+#define LAPACK_dggev LAPACK_GLOBAL(dggev,DGGEV)
+#define LAPACK_cggev LAPACK_GLOBAL(cggev,CGGEV)
+#define LAPACK_zggev LAPACK_GLOBAL(zggev,ZGGEV)
+#define LAPACK_sggevx LAPACK_GLOBAL(sggevx,SGGEVX)
+#define LAPACK_dggevx LAPACK_GLOBAL(dggevx,DGGEVX)
+#define LAPACK_cggevx LAPACK_GLOBAL(cggevx,CGGEVX)
+#define LAPACK_zggevx LAPACK_GLOBAL(zggevx,ZGGEVX)
+#define LAPACK_dsfrk LAPACK_GLOBAL(dsfrk,DSFRK)
+#define LAPACK_ssfrk LAPACK_GLOBAL(ssfrk,SSFRK)
+#define LAPACK_zhfrk LAPACK_GLOBAL(zhfrk,ZHFRK)
+#define LAPACK_chfrk LAPACK_GLOBAL(chfrk,CHFRK)
+#define LAPACK_dtfsm LAPACK_GLOBAL(dtfsm,DTFSM)
+#define LAPACK_stfsm LAPACK_GLOBAL(stfsm,STFSM)
+#define LAPACK_ztfsm LAPACK_GLOBAL(ztfsm,ZTFSM)
+#define LAPACK_ctfsm LAPACK_GLOBAL(ctfsm,CTFSM)
+#define LAPACK_dtfttp LAPACK_GLOBAL(dtfttp,DTFTTP)
+#define LAPACK_stfttp LAPACK_GLOBAL(stfttp,STFTTP)
+#define LAPACK_ztfttp LAPACK_GLOBAL(ztfttp,ZTFTTP)
+#define LAPACK_ctfttp LAPACK_GLOBAL(ctfttp,CTFTTP)
+#define LAPACK_dtfttr LAPACK_GLOBAL(dtfttr,DTFTTR)
+#define LAPACK_stfttr LAPACK_GLOBAL(stfttr,STFTTR)
+#define LAPACK_ztfttr LAPACK_GLOBAL(ztfttr,ZTFTTR)
+#define LAPACK_ctfttr LAPACK_GLOBAL(ctfttr,CTFTTR)
+#define LAPACK_dtpttf LAPACK_GLOBAL(dtpttf,DTPTTF)
+#define LAPACK_stpttf LAPACK_GLOBAL(stpttf,STPTTF)
+#define LAPACK_ztpttf LAPACK_GLOBAL(ztpttf,ZTPTTF)
+#define LAPACK_ctpttf LAPACK_GLOBAL(ctpttf,CTPTTF)
+#define LAPACK_dtpttr LAPACK_GLOBAL(dtpttr,DTPTTR)
+#define LAPACK_stpttr LAPACK_GLOBAL(stpttr,STPTTR)
+#define LAPACK_ztpttr LAPACK_GLOBAL(ztpttr,ZTPTTR)
+#define LAPACK_ctpttr LAPACK_GLOBAL(ctpttr,CTPTTR)
+#define LAPACK_dtrttf LAPACK_GLOBAL(dtrttf,DTRTTF)
+#define LAPACK_strttf LAPACK_GLOBAL(strttf,STRTTF)
+#define LAPACK_ztrttf LAPACK_GLOBAL(ztrttf,ZTRTTF)
+#define LAPACK_ctrttf LAPACK_GLOBAL(ctrttf,CTRTTF)
+#define LAPACK_dtrttp LAPACK_GLOBAL(dtrttp,DTRTTP)
+#define LAPACK_strttp LAPACK_GLOBAL(strttp,STRTTP)
+#define LAPACK_ztrttp LAPACK_GLOBAL(ztrttp,ZTRTTP)
+#define LAPACK_ctrttp LAPACK_GLOBAL(ctrttp,CTRTTP)
+#define LAPACK_sgeqrfp LAPACK_GLOBAL(sgeqrfp,SGEQRFP)
+#define LAPACK_dgeqrfp LAPACK_GLOBAL(dgeqrfp,DGEQRFP)
+#define LAPACK_cgeqrfp LAPACK_GLOBAL(cgeqrfp,CGEQRFP)
+#define LAPACK_zgeqrfp LAPACK_GLOBAL(zgeqrfp,ZGEQRFP)
+#define LAPACK_clacgv LAPACK_GLOBAL(clacgv,CLACGV)
+#define LAPACK_zlacgv LAPACK_GLOBAL(zlacgv,ZLACGV)
+#define LAPACK_slarnv LAPACK_GLOBAL(slarnv,SLARNV)
+#define LAPACK_dlarnv LAPACK_GLOBAL(dlarnv,DLARNV)
+#define LAPACK_clarnv LAPACK_GLOBAL(clarnv,CLARNV)
+#define LAPACK_zlarnv LAPACK_GLOBAL(zlarnv,ZLARNV)
+#define LAPACK_sgeqr2 LAPACK_GLOBAL(sgeqr2,SGEQR2)
+#define LAPACK_dgeqr2 LAPACK_GLOBAL(dgeqr2,DGEQR2)
+#define LAPACK_cgeqr2 LAPACK_GLOBAL(cgeqr2,CGEQR2)
+#define LAPACK_zgeqr2 LAPACK_GLOBAL(zgeqr2,ZGEQR2)
+#define LAPACK_slacpy LAPACK_GLOBAL(slacpy,SLACPY)
+#define LAPACK_dlacpy LAPACK_GLOBAL(dlacpy,DLACPY)
+#define LAPACK_clacpy LAPACK_GLOBAL(clacpy,CLACPY)
+#define LAPACK_zlacpy LAPACK_GLOBAL(zlacpy,ZLACPY)
+#define LAPACK_sgetf2 LAPACK_GLOBAL(sgetf2,SGETF2)
+#define LAPACK_dgetf2 LAPACK_GLOBAL(dgetf2,DGETF2)
+#define LAPACK_cgetf2 LAPACK_GLOBAL(cgetf2,CGETF2)
+#define LAPACK_zgetf2 LAPACK_GLOBAL(zgetf2,ZGETF2)
+#define LAPACK_slaswp LAPACK_GLOBAL(slaswp,SLASWP)
+#define LAPACK_dlaswp LAPACK_GLOBAL(dlaswp,DLASWP)
+#define LAPACK_claswp LAPACK_GLOBAL(claswp,CLASWP)
+#define LAPACK_zlaswp LAPACK_GLOBAL(zlaswp,ZLASWP)
+#define LAPACK_slange LAPACK_GLOBAL(slange,SLANGE)
+#define LAPACK_dlange LAPACK_GLOBAL(dlange,DLANGE)
+#define LAPACK_clange LAPACK_GLOBAL(clange,CLANGE)
+#define LAPACK_zlange LAPACK_GLOBAL(zlange,ZLANGE)
+#define LAPACK_clanhe LAPACK_GLOBAL(clanhe,CLANHE)
+#define LAPACK_zlanhe LAPACK_GLOBAL(zlanhe,ZLANHE)
+#define LAPACK_slansy LAPACK_GLOBAL(slansy,SLANSY)
+#define LAPACK_dlansy LAPACK_GLOBAL(dlansy,DLANSY)
+#define LAPACK_clansy LAPACK_GLOBAL(clansy,CLANSY)
+#define LAPACK_zlansy LAPACK_GLOBAL(zlansy,ZLANSY)
+#define LAPACK_slantr LAPACK_GLOBAL(slantr,SLANTR)
+#define LAPACK_dlantr LAPACK_GLOBAL(dlantr,DLANTR)
+#define LAPACK_clantr LAPACK_GLOBAL(clantr,CLANTR)
+#define LAPACK_zlantr LAPACK_GLOBAL(zlantr,ZLANTR)
+#define LAPACK_slamch LAPACK_GLOBAL(slamch,SLAMCH)
+#define LAPACK_dlamch LAPACK_GLOBAL(dlamch,DLAMCH)
+#define LAPACK_sgelq2 LAPACK_GLOBAL(sgelq2,SGELQ2)
+#define LAPACK_dgelq2 LAPACK_GLOBAL(dgelq2,DGELQ2)
+#define LAPACK_cgelq2 LAPACK_GLOBAL(cgelq2,CGELQ2)
+#define LAPACK_zgelq2 LAPACK_GLOBAL(zgelq2,ZGELQ2)
+#define LAPACK_slarfb LAPACK_GLOBAL(slarfb,SLARFB)
+#define LAPACK_dlarfb LAPACK_GLOBAL(dlarfb,DLARFB)
+#define LAPACK_clarfb LAPACK_GLOBAL(clarfb,CLARFB)
+#define LAPACK_zlarfb LAPACK_GLOBAL(zlarfb,ZLARFB)
+#define LAPACK_slarfg LAPACK_GLOBAL(slarfg,SLARFG)
+#define LAPACK_dlarfg LAPACK_GLOBAL(dlarfg,DLARFG)
+#define LAPACK_clarfg LAPACK_GLOBAL(clarfg,CLARFG)
+#define LAPACK_zlarfg LAPACK_GLOBAL(zlarfg,ZLARFG)
+#define LAPACK_slarft LAPACK_GLOBAL(slarft,SLARFT)
+#define LAPACK_dlarft LAPACK_GLOBAL(dlarft,DLARFT)
+#define LAPACK_clarft LAPACK_GLOBAL(clarft,CLARFT)
+#define LAPACK_zlarft LAPACK_GLOBAL(zlarft,ZLARFT)
+#define LAPACK_slarfx LAPACK_GLOBAL(slarfx,SLARFX)
+#define LAPACK_dlarfx LAPACK_GLOBAL(dlarfx,DLARFX)
+#define LAPACK_clarfx LAPACK_GLOBAL(clarfx,CLARFX)
+#define LAPACK_zlarfx LAPACK_GLOBAL(zlarfx,ZLARFX)
+#define LAPACK_slatms LAPACK_GLOBAL(slatms,SLATMS)
+#define LAPACK_dlatms LAPACK_GLOBAL(dlatms,DLATMS)
+#define LAPACK_clatms LAPACK_GLOBAL(clatms,CLATMS)
+#define LAPACK_zlatms LAPACK_GLOBAL(zlatms,ZLATMS)
+#define LAPACK_slag2d LAPACK_GLOBAL(slag2d,SLAG2D)
+#define LAPACK_dlag2s LAPACK_GLOBAL(dlag2s,DLAG2S)
+#define LAPACK_clag2z LAPACK_GLOBAL(clag2z,CLAG2Z)
+#define LAPACK_zlag2c LAPACK_GLOBAL(zlag2c,ZLAG2C)
+#define LAPACK_slauum LAPACK_GLOBAL(slauum,SLAUUM)
+#define LAPACK_dlauum LAPACK_GLOBAL(dlauum,DLAUUM)
+#define LAPACK_clauum LAPACK_GLOBAL(clauum,CLAUUM)
+#define LAPACK_zlauum LAPACK_GLOBAL(zlauum,ZLAUUM)
+#define LAPACK_slagge LAPACK_GLOBAL(slagge,SLAGGE)
+#define LAPACK_dlagge LAPACK_GLOBAL(dlagge,DLAGGE)
+#define LAPACK_clagge LAPACK_GLOBAL(clagge,CLAGGE)
+#define LAPACK_zlagge LAPACK_GLOBAL(zlagge,ZLAGGE)
+#define LAPACK_slaset LAPACK_GLOBAL(slaset,SLASET)
+#define LAPACK_dlaset LAPACK_GLOBAL(dlaset,DLASET)
+#define LAPACK_claset LAPACK_GLOBAL(claset,CLASET)
+#define LAPACK_zlaset LAPACK_GLOBAL(zlaset,ZLASET)
+#define LAPACK_slasrt LAPACK_GLOBAL(slasrt,SLASRT)
+#define LAPACK_dlasrt LAPACK_GLOBAL(dlasrt,DLASRT)
+#define LAPACK_slagsy LAPACK_GLOBAL(slagsy,SLAGSY)
+#define LAPACK_dlagsy LAPACK_GLOBAL(dlagsy,DLAGSY)
+#define LAPACK_clagsy LAPACK_GLOBAL(clagsy,CLAGSY)
+#define LAPACK_zlagsy LAPACK_GLOBAL(zlagsy,ZLAGSY)
+#define LAPACK_claghe LAPACK_GLOBAL(claghe,CLAGHE)
+#define LAPACK_zlaghe LAPACK_GLOBAL(zlaghe,ZLAGHE)
+#define LAPACK_slapmr LAPACK_GLOBAL(slapmr,SLAPMR)
+#define LAPACK_dlapmr LAPACK_GLOBAL(dlapmr,DLAPMR)
+#define LAPACK_clapmr LAPACK_GLOBAL(clapmr,CLAPMR)
+#define LAPACK_zlapmr LAPACK_GLOBAL(zlapmr,ZLAPMR)
+#define LAPACK_slapy2 LAPACK_GLOBAL(slapy2,SLAPY2)
+#define LAPACK_dlapy2 LAPACK_GLOBAL(dlapy2,DLAPY2)
+#define LAPACK_slapy3 LAPACK_GLOBAL(slapy3,SLAPY3)
+#define LAPACK_dlapy3 LAPACK_GLOBAL(dlapy3,DLAPY3)
+#define LAPACK_slartgp LAPACK_GLOBAL(slartgp,SLARTGP)
+#define LAPACK_dlartgp LAPACK_GLOBAL(dlartgp,DLARTGP)
+#define LAPACK_slartgs LAPACK_GLOBAL(slartgs,SLARTGS)
+#define LAPACK_dlartgs LAPACK_GLOBAL(dlartgs,DLARTGS)
+// LAPACK 3.3.0
+#define LAPACK_cbbcsd LAPACK_GLOBAL(cbbcsd,CBBCSD)
+#define LAPACK_cheswapr LAPACK_GLOBAL(cheswapr,CHESWAPR)
+#define LAPACK_chetri2 LAPACK_GLOBAL(chetri2,CHETRI2)
+#define LAPACK_chetri2x LAPACK_GLOBAL(chetri2x,CHETRI2X)
+#define LAPACK_chetrs2 LAPACK_GLOBAL(chetrs2,CHETRS2)
+#define LAPACK_csyconv LAPACK_GLOBAL(csyconv,CSYCONV)
+#define LAPACK_csyswapr LAPACK_GLOBAL(csyswapr,CSYSWAPR)
+#define LAPACK_csytri2 LAPACK_GLOBAL(csytri2,CSYTRI2)
+#define LAPACK_csytri2x LAPACK_GLOBAL(csytri2x,CSYTRI2X)
+#define LAPACK_csytrs2 LAPACK_GLOBAL(csytrs2,CSYTRS2)
+#define LAPACK_cunbdb LAPACK_GLOBAL(cunbdb,CUNBDB)
+#define LAPACK_cuncsd LAPACK_GLOBAL(cuncsd,CUNCSD)
+#define LAPACK_dbbcsd LAPACK_GLOBAL(dbbcsd,DBBCSD)
+#define LAPACK_dorbdb LAPACK_GLOBAL(dorbdb,DORBDB)
+#define LAPACK_dorcsd LAPACK_GLOBAL(dorcsd,DORCSD)
+#define LAPACK_dsyconv LAPACK_GLOBAL(dsyconv,DSYCONV)
+#define LAPACK_dsyswapr LAPACK_GLOBAL(dsyswapr,DSYSWAPR)
+#define LAPACK_dsytri2 LAPACK_GLOBAL(dsytri2,DSYTRI2)
+#define LAPACK_dsytri2x LAPACK_GLOBAL(dsytri2x,DSYTRI2X)
+#define LAPACK_dsytrs2 LAPACK_GLOBAL(dsytrs2,DSYTRS2)
+#define LAPACK_sbbcsd LAPACK_GLOBAL(sbbcsd,SBBCSD)
+#define LAPACK_sorbdb LAPACK_GLOBAL(sorbdb,SORBDB)
+#define LAPACK_sorcsd LAPACK_GLOBAL(sorcsd,SORCSD)
+#define LAPACK_ssyconv LAPACK_GLOBAL(ssyconv,SSYCONV)
+#define LAPACK_ssyswapr LAPACK_GLOBAL(ssyswapr,SSYSWAPR)
+#define LAPACK_ssytri2 LAPACK_GLOBAL(ssytri2,SSYTRI2)
+#define LAPACK_ssytri2x LAPACK_GLOBAL(ssytri2x,SSYTRI2X)
+#define LAPACK_ssytrs2 LAPACK_GLOBAL(ssytrs2,SSYTRS2)
+#define LAPACK_zbbcsd LAPACK_GLOBAL(zbbcsd,ZBBCSD)
+#define LAPACK_zheswapr LAPACK_GLOBAL(zheswapr,ZHESWAPR)
+#define LAPACK_zhetri2 LAPACK_GLOBAL(zhetri2,ZHETRI2)
+#define LAPACK_zhetri2x LAPACK_GLOBAL(zhetri2x,ZHETRI2X)
+#define LAPACK_zhetrs2 LAPACK_GLOBAL(zhetrs2,ZHETRS2)
+#define LAPACK_zsyconv LAPACK_GLOBAL(zsyconv,ZSYCONV)
+#define LAPACK_zsyswapr LAPACK_GLOBAL(zsyswapr,ZSYSWAPR)
+#define LAPACK_zsytri2 LAPACK_GLOBAL(zsytri2,ZSYTRI2)
+#define LAPACK_zsytri2x LAPACK_GLOBAL(zsytri2x,ZSYTRI2X)
+#define LAPACK_zsytrs2 LAPACK_GLOBAL(zsytrs2,ZSYTRS2)
+#define LAPACK_zunbdb LAPACK_GLOBAL(zunbdb,ZUNBDB)
+#define LAPACK_zuncsd LAPACK_GLOBAL(zuncsd,ZUNCSD)
+// LAPACK 3.4.0
+#define LAPACK_sgemqrt LAPACK_GLOBAL(sgemqrt,SGEMQRT)
+#define LAPACK_dgemqrt LAPACK_GLOBAL(dgemqrt,DGEMQRT)
+#define LAPACK_cgemqrt LAPACK_GLOBAL(cgemqrt,CGEMQRT)
+#define LAPACK_zgemqrt LAPACK_GLOBAL(zgemqrt,ZGEMQRT)
+#define LAPACK_sgeqrt LAPACK_GLOBAL(sgeqrt,SGEQRT)
+#define LAPACK_dgeqrt LAPACK_GLOBAL(dgeqrt,DGEQRT)
+#define LAPACK_cgeqrt LAPACK_GLOBAL(cgeqrt,CGEQRT)
+#define LAPACK_zgeqrt LAPACK_GLOBAL(zgeqrt,ZGEQRT)
+#define LAPACK_sgeqrt2 LAPACK_GLOBAL(sgeqrt2,SGEQRT2)
+#define LAPACK_dgeqrt2 LAPACK_GLOBAL(dgeqrt2,DGEQRT2)
+#define LAPACK_cgeqrt2 LAPACK_GLOBAL(cgeqrt2,CGEQRT2)
+#define LAPACK_zgeqrt2 LAPACK_GLOBAL(zgeqrt2,ZGEQRT2)
+#define LAPACK_sgeqrt3 LAPACK_GLOBAL(sgeqrt3,SGEQRT3)
+#define LAPACK_dgeqrt3 LAPACK_GLOBAL(dgeqrt3,DGEQRT3)
+#define LAPACK_cgeqrt3 LAPACK_GLOBAL(cgeqrt3,CGEQRT3)
+#define LAPACK_zgeqrt3 LAPACK_GLOBAL(zgeqrt3,ZGEQRT3)
+#define LAPACK_stpmqrt LAPACK_GLOBAL(stpmqrt,STPMQRT)
+#define LAPACK_dtpmqrt LAPACK_GLOBAL(dtpmqrt,DTPMQRT)
+#define LAPACK_ctpmqrt LAPACK_GLOBAL(ctpmqrt,CTPMQRT)
+#define LAPACK_ztpmqrt LAPACK_GLOBAL(ztpmqrt,ZTPMQRT)
+#define LAPACK_dtpqrt LAPACK_GLOBAL(dtpqrt,DTPQRT)
+#define LAPACK_ctpqrt LAPACK_GLOBAL(ctpqrt,CTPQRT)
+#define LAPACK_ztpqrt LAPACK_GLOBAL(ztpqrt,ZTPQRT)
+#define LAPACK_stpqrt2 LAPACK_GLOBAL(stpqrt2,STPQRT2)
+#define LAPACK_dtpqrt2 LAPACK_GLOBAL(dtpqrt2,DTPQRT2)
+#define LAPACK_ctpqrt2 LAPACK_GLOBAL(ctpqrt2,CTPQRT2)
+#define LAPACK_ztpqrt2 LAPACK_GLOBAL(ztpqrt2,ZTPQRT2)
+#define LAPACK_stprfb LAPACK_GLOBAL(stprfb,STPRFB)
+#define LAPACK_dtprfb LAPACK_GLOBAL(dtprfb,DTPRFB)
+#define LAPACK_ctprfb LAPACK_GLOBAL(ctprfb,CTPRFB)
+#define LAPACK_ztprfb LAPACK_GLOBAL(ztprfb,ZTPRFB)
+// LAPACK 3.X.X
+#define LAPACK_csyr LAPACK_GLOBAL(csyr,CSYR)
+#define LAPACK_zsyr LAPACK_GLOBAL(zsyr,ZSYR)
+
+
+void LAPACK_sgetrf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_dgetrf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_cgetrf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_int *info );
+void LAPACK_zgetrf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_int *info );
+void LAPACK_sgbtrf( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, float* ab, lapack_int* ldab,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_dgbtrf( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, double* ab, lapack_int* ldab,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_cgbtrf( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_zgbtrf( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_sgttrf( lapack_int* n, float* dl, float* d, float* du, float* du2,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_dgttrf( lapack_int* n, double* dl, double* d, double* du,
+                    double* du2, lapack_int* ipiv, lapack_int *info );
+void LAPACK_cgttrf( lapack_int* n, lapack_complex_float* dl,
+                    lapack_complex_float* d, lapack_complex_float* du,
+                    lapack_complex_float* du2, lapack_int* ipiv,
+                    lapack_int *info );
+void LAPACK_zgttrf( lapack_int* n, lapack_complex_double* dl,
+                    lapack_complex_double* d, lapack_complex_double* du,
+                    lapack_complex_double* du2, lapack_int* ipiv,
+                    lapack_int *info );
+void LAPACK_spotrf( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dpotrf( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_cpotrf( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_zpotrf( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_dpstrf( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* piv, lapack_int* rank, double* tol,
+                    double* work, lapack_int *info );
+void LAPACK_spstrf( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* piv, lapack_int* rank, float* tol, float* work,
+                    lapack_int *info );
+void LAPACK_zpstrf( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* piv, lapack_int* rank,
+                    double* tol, double* work, lapack_int *info );
+void LAPACK_cpstrf( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* piv, lapack_int* rank,
+                    float* tol, float* work, lapack_int *info );
+void LAPACK_dpftrf( char* transr, char* uplo, lapack_int* n, double* a,
+                    lapack_int *info );
+void LAPACK_spftrf( char* transr, char* uplo, lapack_int* n, float* a,
+                    lapack_int *info );
+void LAPACK_zpftrf( char* transr, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int *info );
+void LAPACK_cpftrf( char* transr, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int *info );
+void LAPACK_spptrf( char* uplo, lapack_int* n, float* ap, lapack_int *info );
+void LAPACK_dpptrf( char* uplo, lapack_int* n, double* ap, lapack_int *info );
+void LAPACK_cpptrf( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    lapack_int *info );
+void LAPACK_zpptrf( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    lapack_int *info );
+void LAPACK_spbtrf( char* uplo, lapack_int* n, lapack_int* kd, float* ab,
+                    lapack_int* ldab, lapack_int *info );
+void LAPACK_dpbtrf( char* uplo, lapack_int* n, lapack_int* kd, double* ab,
+                    lapack_int* ldab, lapack_int *info );
+void LAPACK_cpbtrf( char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_int *info );
+void LAPACK_zpbtrf( char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_int *info );
+void LAPACK_spttrf( lapack_int* n, float* d, float* e, lapack_int *info );
+void LAPACK_dpttrf( lapack_int* n, double* d, double* e, lapack_int *info );
+void LAPACK_cpttrf( lapack_int* n, float* d, lapack_complex_float* e,
+                    lapack_int *info );
+void LAPACK_zpttrf( lapack_int* n, double* d, lapack_complex_double* e,
+                    lapack_int *info );
+void LAPACK_ssytrf( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* ipiv, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dsytrf( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* ipiv, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_csytrf( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ipiv,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zsytrf( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_chetrf( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ipiv,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zhetrf( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ssptrf( char* uplo, lapack_int* n, float* ap, lapack_int* ipiv,
+                    lapack_int *info );
+void LAPACK_dsptrf( char* uplo, lapack_int* n, double* ap, lapack_int* ipiv,
+                    lapack_int *info );
+void LAPACK_csptrf( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_zsptrf( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_chptrf( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_zhptrf( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_sgetrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* a, lapack_int* lda, const lapack_int* ipiv,
+                    float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_dgetrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const lapack_int* ipiv,
+                    double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_cgetrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zgetrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_sgbtrs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const float* ab, lapack_int* ldab,
+                    const lapack_int* ipiv, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dgbtrs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const double* ab, lapack_int* ldab,
+                    const lapack_int* ipiv, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_cgbtrs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const lapack_complex_float* ab,
+                    lapack_int* ldab, const lapack_int* ipiv,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zgbtrs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const lapack_complex_double* ab,
+                    lapack_int* ldab, const lapack_int* ipiv,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_sgttrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* dl, const float* d, const float* du,
+                    const float* du2, const lapack_int* ipiv, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dgttrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* dl, const double* d, const double* du,
+                    const double* du2, const lapack_int* ipiv, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_cgttrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* dl,
+                    const lapack_complex_float* d,
+                    const lapack_complex_float* du,
+                    const lapack_complex_float* du2, const lapack_int* ipiv,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zgttrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* dl,
+                    const lapack_complex_double* d,
+                    const lapack_complex_double* du,
+                    const lapack_complex_double* du2, const lapack_int* ipiv,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_spotrs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dpotrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_cpotrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zpotrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dpftrs( char* transr, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_spftrs( char* transr, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* a, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zpftrs( char* transr, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_cpftrs( char* transr, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_spptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dpptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_cpptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zpptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_spbtrs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const float* ab, lapack_int* ldab, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dpbtrs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const double* ab, lapack_int* ldab, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_cpbtrs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zpbtrs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_spttrs( lapack_int* n, lapack_int* nrhs, const float* d,
+                    const float* e, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dpttrs( lapack_int* n, lapack_int* nrhs, const double* d,
+                    const double* e, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_cpttrs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* d,
+                    const lapack_complex_float* e, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zpttrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* d, const lapack_complex_double* e,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ssytrs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* a,
+                    lapack_int* lda, const lapack_int* ipiv, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dsytrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const lapack_int* ipiv,
+                    double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_csytrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zsytrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_chetrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zhetrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_ssptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, const lapack_int* ipiv, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dsptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, const lapack_int* ipiv, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_csptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, const lapack_int* ipiv,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zsptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, const lapack_int* ipiv,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_chptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, const lapack_int* ipiv,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zhptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, const lapack_int* ipiv,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_strtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dtrtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const double* a, lapack_int* lda,
+                    double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_ctrtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ztrtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_stptrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const float* ap, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dtptrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const double* ap, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_ctptrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_float* ap,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ztptrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_double* ap,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_stbtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs, const float* ab,
+                    lapack_int* ldab, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dtbtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs, const double* ab,
+                    lapack_int* ldab, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ctbtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ztbtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_sgecon( char* norm, lapack_int* n, const float* a, lapack_int* lda,
+                    float* anorm, float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgecon( char* norm, lapack_int* n, const double* a, lapack_int* lda,
+                    double* anorm, double* rcond, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgecon( char* norm, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, float* anorm, float* rcond,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgecon( char* norm, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, double* anorm, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sgbcon( char* norm, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    const float* ab, lapack_int* ldab, const lapack_int* ipiv,
+                    float* anorm, float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgbcon( char* norm, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    const double* ab, lapack_int* ldab, const lapack_int* ipiv,
+                    double* anorm, double* rcond, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgbcon( char* norm, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgbcon( char* norm, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sgtcon( char* norm, lapack_int* n, const float* dl, const float* d,
+                    const float* du, const float* du2, const lapack_int* ipiv,
+                    float* anorm, float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgtcon( char* norm, lapack_int* n, const double* dl,
+                    const double* d, const double* du, const double* du2,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cgtcon( char* norm, lapack_int* n, const lapack_complex_float* dl,
+                    const lapack_complex_float* d,
+                    const lapack_complex_float* du,
+                    const lapack_complex_float* du2, const lapack_int* ipiv,
+                    float* anorm, float* rcond, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zgtcon( char* norm, lapack_int* n, const lapack_complex_double* dl,
+                    const lapack_complex_double* d,
+                    const lapack_complex_double* du,
+                    const lapack_complex_double* du2, const lapack_int* ipiv,
+                    double* anorm, double* rcond, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_spocon( char* uplo, lapack_int* n, const float* a, lapack_int* lda,
+                    float* anorm, float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dpocon( char* uplo, lapack_int* n, const double* a, lapack_int* lda,
+                    double* anorm, double* rcond, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cpocon( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, float* anorm, float* rcond,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zpocon( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, double* anorm, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sppcon( char* uplo, lapack_int* n, const float* ap, float* anorm,
+                    float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dppcon( char* uplo, lapack_int* n, const double* ap, double* anorm,
+                    double* rcond, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cppcon( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    float* anorm, float* rcond, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zppcon( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    double* anorm, double* rcond, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_spbcon( char* uplo, lapack_int* n, lapack_int* kd, const float* ab,
+                    lapack_int* ldab, float* anorm, float* rcond, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dpbcon( char* uplo, lapack_int* n, lapack_int* kd, const double* ab,
+                    lapack_int* ldab, double* anorm, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cpbcon( char* uplo, lapack_int* n, lapack_int* kd,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    float* anorm, float* rcond, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zpbcon( char* uplo, lapack_int* n, lapack_int* kd,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    double* anorm, double* rcond, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sptcon( lapack_int* n, const float* d, const float* e, float* anorm,
+                    float* rcond, float* work, lapack_int *info );
+void LAPACK_dptcon( lapack_int* n, const double* d, const double* e,
+                    double* anorm, double* rcond, double* work,
+                    lapack_int *info );
+void LAPACK_cptcon( lapack_int* n, const float* d,
+                    const lapack_complex_float* e, float* anorm, float* rcond,
+                    float* work, lapack_int *info );
+void LAPACK_zptcon( lapack_int* n, const double* d,
+                    const lapack_complex_double* e, double* anorm,
+                    double* rcond, double* work, lapack_int *info );
+void LAPACK_ssycon( char* uplo, lapack_int* n, const float* a, lapack_int* lda,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dsycon( char* uplo, lapack_int* n, const double* a, lapack_int* lda,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_csycon( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_int* ipiv, float* anorm,
+                    float* rcond, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zsycon( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_int* ipiv, double* anorm,
+                    double* rcond, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_checon( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_int* ipiv, float* anorm,
+                    float* rcond, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zhecon( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_int* ipiv, double* anorm,
+                    double* rcond, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_sspcon( char* uplo, lapack_int* n, const float* ap,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dspcon( char* uplo, lapack_int* n, const double* ap,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cspcon( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zspcon( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_chpcon( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zhpcon( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_strcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const float* a, lapack_int* lda, float* rcond, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dtrcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const double* a, lapack_int* lda, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_ctrcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    float* rcond, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztrcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    double* rcond, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_stpcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const float* ap, float* rcond, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dtpcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const double* ap, double* rcond, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ctpcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const lapack_complex_float* ap, float* rcond,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztpcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const lapack_complex_double* ap, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_stbcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    lapack_int* kd, const float* ab, lapack_int* ldab,
+                    float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dtbcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    lapack_int* kd, const double* ab, lapack_int* ldab,
+                    double* rcond, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_ctbcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    lapack_int* kd, const lapack_complex_float* ab,
+                    lapack_int* ldab, float* rcond, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_ztbcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    lapack_int* kd, const lapack_complex_double* ab,
+                    lapack_int* ldab, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sgerfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* a, lapack_int* lda, const float* af,
+                    lapack_int* ldaf, const lapack_int* ipiv, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* ferr,
+                    float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgerfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const double* af,
+                    lapack_int* ldaf, const lapack_int* ipiv, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cgerfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgerfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dgerfsx( char* trans, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const double* a, lapack_int* lda, const double* af,
+                     lapack_int* ldaf, const lapack_int* ipiv, const double* r,
+                     const double* c, const double* b, lapack_int* ldb,
+                     double* x, lapack_int* ldx, double* rcond, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgerfsx( char* trans, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const float* a, lapack_int* lda, const float* af,
+                     lapack_int* ldaf, const lapack_int* ipiv, const float* r,
+                     const float* c, const float* b, lapack_int* ldb, float* x,
+                     lapack_int* ldx, float* rcond, float* berr,
+                     lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zgerfsx( char* trans, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_complex_double* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const double* r, const double* c,
+                     const lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cgerfsx( char* trans, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_float* a, lapack_int* lda,
+                     const lapack_complex_float* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const float* r, const float* c,
+                     const lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sgbrfs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const float* ab, lapack_int* ldab,
+                    const float* afb, lapack_int* ldafb, const lapack_int* ipiv,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgbrfs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const double* ab, lapack_int* ldab,
+                    const double* afb, lapack_int* ldafb,
+                    const lapack_int* ipiv, const double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* ferr, double* berr,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cgbrfs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const lapack_complex_float* ab,
+                    lapack_int* ldab, const lapack_complex_float* afb,
+                    lapack_int* ldafb, const lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgbrfs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const lapack_complex_double* ab,
+                    lapack_int* ldab, const lapack_complex_double* afb,
+                    lapack_int* ldafb, const lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_dgbrfsx( char* trans, char* equed, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, const double* ab,
+                     lapack_int* ldab, const double* afb, lapack_int* ldafb,
+                     const lapack_int* ipiv, const double* r, const double* c,
+                     const double* b, lapack_int* ldb, double* x,
+                     lapack_int* ldx, double* rcond, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgbrfsx( char* trans, char* equed, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, const float* ab,
+                     lapack_int* ldab, const float* afb, lapack_int* ldafb,
+                     const lapack_int* ipiv, const float* r, const float* c,
+                     const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                     float* rcond, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params, float* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_zgbrfsx( char* trans, char* equed, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs,
+                     const lapack_complex_double* ab, lapack_int* ldab,
+                     const lapack_complex_double* afb, lapack_int* ldafb,
+                     const lapack_int* ipiv, const double* r, const double* c,
+                     const lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cgbrfsx( char* trans, char* equed, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs,
+                     const lapack_complex_float* ab, lapack_int* ldab,
+                     const lapack_complex_float* afb, lapack_int* ldafb,
+                     const lapack_int* ipiv, const float* r, const float* c,
+                     const lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sgtrfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* dl, const float* d, const float* du,
+                    const float* dlf, const float* df, const float* duf,
+                    const float* du2, const lapack_int* ipiv, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* ferr,
+                    float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgtrfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* dl, const double* d, const double* du,
+                    const double* dlf, const double* df, const double* duf,
+                    const double* du2, const lapack_int* ipiv, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cgtrfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* dl,
+                    const lapack_complex_float* d,
+                    const lapack_complex_float* du,
+                    const lapack_complex_float* dlf,
+                    const lapack_complex_float* df,
+                    const lapack_complex_float* duf,
+                    const lapack_complex_float* du2, const lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgtrfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* dl,
+                    const lapack_complex_double* d,
+                    const lapack_complex_double* du,
+                    const lapack_complex_double* dlf,
+                    const lapack_complex_double* df,
+                    const lapack_complex_double* duf,
+                    const lapack_complex_double* du2, const lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sporfs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* a,
+                    lapack_int* lda, const float* af, lapack_int* ldaf,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dporfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const double* af,
+                    lapack_int* ldaf, const double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* ferr, double* berr,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cporfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* af, lapack_int* ldaf,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zporfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* af, lapack_int* ldaf,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_dporfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const double* a, lapack_int* lda, const double* af,
+                     lapack_int* ldaf, const double* s, const double* b,
+                     lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params, double* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_sporfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const float* a, lapack_int* lda, const float* af,
+                     lapack_int* ldaf, const float* s, const float* b,
+                     lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zporfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_complex_double* af, lapack_int* ldaf,
+                     const double* s, const lapack_complex_double* b,
+                     lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                     double* rcond, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cporfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_float* a, lapack_int* lda,
+                     const lapack_complex_float* af, lapack_int* ldaf,
+                     const float* s, const lapack_complex_float* b,
+                     lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                     float* rcond, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_spprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, const float* afp, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* ferr,
+                    float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dpprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, const double* afp, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cpprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap,
+                    const lapack_complex_float* afp,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zpprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap,
+                    const lapack_complex_double* afp,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_spbrfs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const float* ab, lapack_int* ldab, const float* afb,
+                    lapack_int* ldafb, const float* b, lapack_int* ldb,
+                    float* x, lapack_int* ldx, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dpbrfs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const double* ab, lapack_int* ldab, const double* afb,
+                    lapack_int* ldafb, const double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* ferr, double* berr,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cpbrfs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    const lapack_complex_float* afb, lapack_int* ldafb,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zpbrfs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    const lapack_complex_double* afb, lapack_int* ldafb,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sptrfs( lapack_int* n, lapack_int* nrhs, const float* d,
+                    const float* e, const float* df, const float* ef,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int *info );
+void LAPACK_dptrfs( lapack_int* n, lapack_int* nrhs, const double* d,
+                    const double* e, const double* df, const double* ef,
+                    const double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* ferr, double* berr, double* work,
+                    lapack_int *info );
+void LAPACK_cptrfs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* d,
+                    const lapack_complex_float* e, const float* df,
+                    const lapack_complex_float* ef,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zptrfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* d, const lapack_complex_double* e,
+                    const double* df, const lapack_complex_double* ef,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_ssyrfs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* a,
+                    lapack_int* lda, const float* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const float* b, lapack_int* ldb,
+                    float* x, lapack_int* ldx, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dsyrfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const double* af,
+                    lapack_int* ldaf, const lapack_int* ipiv, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_csyrfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zsyrfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dsyrfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const double* a, lapack_int* lda, const double* af,
+                     lapack_int* ldaf, const lapack_int* ipiv, const double* s,
+                     const double* b, lapack_int* ldb, double* x,
+                     lapack_int* ldx, double* rcond, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_ssyrfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const float* a, lapack_int* lda, const float* af,
+                     lapack_int* ldaf, const lapack_int* ipiv, const float* s,
+                     const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                     float* rcond, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params, float* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_zsyrfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_complex_double* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const double* s,
+                     const lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_csyrfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_float* a, lapack_int* lda,
+                     const lapack_complex_float* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const float* s,
+                     const lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_cherfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zherfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_zherfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_complex_double* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const double* s,
+                     const lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cherfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_float* a, lapack_int* lda,
+                     const lapack_complex_float* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const float* s,
+                     const lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_ssprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, const float* afp, const lapack_int* ipiv,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dsprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, const double* afp, const lapack_int* ipiv,
+                    const double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* ferr, double* berr, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_csprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap,
+                    const lapack_complex_float* afp, const lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zsprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap,
+                    const lapack_complex_double* afp, const lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_chprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap,
+                    const lapack_complex_float* afp, const lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap,
+                    const lapack_complex_double* afp, const lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_strrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const float* a, lapack_int* lda,
+                    const float* b, lapack_int* ldb, const float* x,
+                    lapack_int* ldx, float* ferr, float* berr, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dtrrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const double* a, lapack_int* lda,
+                    const double* b, lapack_int* ldb, const double* x,
+                    lapack_int* ldx, double* ferr, double* berr, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ctrrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* b,
+                    lapack_int* ldb, const lapack_complex_float* x,
+                    lapack_int* ldx, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztrrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* b,
+                    lapack_int* ldb, const lapack_complex_double* x,
+                    lapack_int* ldx, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_stprfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const float* ap, const float* b,
+                    lapack_int* ldb, const float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dtprfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const double* ap, const double* b,
+                    lapack_int* ldb, const double* x, lapack_int* ldx,
+                    double* ferr, double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_ctprfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_float* ap,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    const lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztprfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_double* ap,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    const lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_stbrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs, const float* ab,
+                    lapack_int* ldab, const float* b, lapack_int* ldb,
+                    const float* x, lapack_int* ldx, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dtbrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs, const double* ab,
+                    lapack_int* ldab, const double* b, lapack_int* ldb,
+                    const double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_ctbrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    const lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztbrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    const lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgetri( lapack_int* n, float* a, lapack_int* lda,
+                    const lapack_int* ipiv, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgetri( lapack_int* n, double* a, lapack_int* lda,
+                    const lapack_int* ipiv, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgetri( lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zgetri( lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_spotri( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dpotri( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_cpotri( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_zpotri( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_dpftri( char* transr, char* uplo, lapack_int* n, double* a,
+                    lapack_int *info );
+void LAPACK_spftri( char* transr, char* uplo, lapack_int* n, float* a,
+                    lapack_int *info );
+void LAPACK_zpftri( char* transr, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int *info );
+void LAPACK_cpftri( char* transr, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int *info );
+void LAPACK_spptri( char* uplo, lapack_int* n, float* ap, lapack_int *info );
+void LAPACK_dpptri( char* uplo, lapack_int* n, double* ap, lapack_int *info );
+void LAPACK_cpptri( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    lapack_int *info );
+void LAPACK_zpptri( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    lapack_int *info );
+void LAPACK_ssytri( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    const lapack_int* ipiv, float* work, lapack_int *info );
+void LAPACK_dsytri( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    const lapack_int* ipiv, double* work, lapack_int *info );
+void LAPACK_csytri( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, const lapack_int* ipiv,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zsytri( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, const lapack_int* ipiv,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_chetri( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, const lapack_int* ipiv,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zhetri( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, const lapack_int* ipiv,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_ssptri( char* uplo, lapack_int* n, float* ap,
+                    const lapack_int* ipiv, float* work, lapack_int *info );
+void LAPACK_dsptri( char* uplo, lapack_int* n, double* ap,
+                    const lapack_int* ipiv, double* work, lapack_int *info );
+void LAPACK_csptri( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    const lapack_int* ipiv, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zsptri( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    const lapack_int* ipiv, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_chptri( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    const lapack_int* ipiv, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zhptri( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    const lapack_int* ipiv, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_strtri( char* uplo, char* diag, lapack_int* n, float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_dtrtri( char* uplo, char* diag, lapack_int* n, double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_ctrtri( char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_ztrtri( char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dtftri( char* transr, char* uplo, char* diag, lapack_int* n,
+                    double* a, lapack_int *info );
+void LAPACK_stftri( char* transr, char* uplo, char* diag, lapack_int* n,
+                    float* a, lapack_int *info );
+void LAPACK_ztftri( char* transr, char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_double* a, lapack_int *info );
+void LAPACK_ctftri( char* transr, char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_float* a, lapack_int *info );
+void LAPACK_stptri( char* uplo, char* diag, lapack_int* n, float* ap,
+                    lapack_int *info );
+void LAPACK_dtptri( char* uplo, char* diag, lapack_int* n, double* ap,
+                    lapack_int *info );
+void LAPACK_ctptri( char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_float* ap, lapack_int *info );
+void LAPACK_ztptri( char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_double* ap, lapack_int *info );
+void LAPACK_sgeequ( lapack_int* m, lapack_int* n, const float* a,
+                    lapack_int* lda, float* r, float* c, float* rowcnd,
+                    float* colcnd, float* amax, lapack_int *info );
+void LAPACK_dgeequ( lapack_int* m, lapack_int* n, const double* a,
+                    lapack_int* lda, double* r, double* c, double* rowcnd,
+                    double* colcnd, double* amax, lapack_int *info );
+void LAPACK_cgeequ( lapack_int* m, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, float* r, float* c, float* rowcnd,
+                    float* colcnd, float* amax, lapack_int *info );
+void LAPACK_zgeequ( lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda, double* r,
+                    double* c, double* rowcnd, double* colcnd, double* amax,
+                    lapack_int *info );
+void LAPACK_dgeequb( lapack_int* m, lapack_int* n, const double* a,
+                     lapack_int* lda, double* r, double* c, double* rowcnd,
+                     double* colcnd, double* amax, lapack_int *info );
+void LAPACK_sgeequb( lapack_int* m, lapack_int* n, const float* a,
+                     lapack_int* lda, float* r, float* c, float* rowcnd,
+                     float* colcnd, float* amax, lapack_int *info );
+void LAPACK_zgeequb( lapack_int* m, lapack_int* n,
+                     const lapack_complex_double* a, lapack_int* lda, double* r,
+                     double* c, double* rowcnd, double* colcnd, double* amax,
+                     lapack_int *info );
+void LAPACK_cgeequb( lapack_int* m, lapack_int* n,
+                     const lapack_complex_float* a, lapack_int* lda, float* r,
+                     float* c, float* rowcnd, float* colcnd, float* amax,
+                     lapack_int *info );
+void LAPACK_sgbequ( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const float* ab, lapack_int* ldab, float* r,
+                    float* c, float* rowcnd, float* colcnd, float* amax,
+                    lapack_int *info );
+void LAPACK_dgbequ( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const double* ab, lapack_int* ldab,
+                    double* r, double* c, double* rowcnd, double* colcnd,
+                    double* amax, lapack_int *info );
+void LAPACK_cgbequ( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const lapack_complex_float* ab,
+                    lapack_int* ldab, float* r, float* c, float* rowcnd,
+                    float* colcnd, float* amax, lapack_int *info );
+void LAPACK_zgbequ( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const lapack_complex_double* ab,
+                    lapack_int* ldab, double* r, double* c, double* rowcnd,
+                    double* colcnd, double* amax, lapack_int *info );
+void LAPACK_dgbequb( lapack_int* m, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, const double* ab, lapack_int* ldab,
+                     double* r, double* c, double* rowcnd, double* colcnd,
+                     double* amax, lapack_int *info );
+void LAPACK_sgbequb( lapack_int* m, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, const float* ab, lapack_int* ldab,
+                     float* r, float* c, float* rowcnd, float* colcnd,
+                     float* amax, lapack_int *info );
+void LAPACK_zgbequb( lapack_int* m, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, const lapack_complex_double* ab,
+                     lapack_int* ldab, double* r, double* c, double* rowcnd,
+                     double* colcnd, double* amax, lapack_int *info );
+void LAPACK_cgbequb( lapack_int* m, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, const lapack_complex_float* ab,
+                     lapack_int* ldab, float* r, float* c, float* rowcnd,
+                     float* colcnd, float* amax, lapack_int *info );
+void LAPACK_spoequ( lapack_int* n, const float* a, lapack_int* lda, float* s,
+                    float* scond, float* amax, lapack_int *info );
+void LAPACK_dpoequ( lapack_int* n, const double* a, lapack_int* lda, double* s,
+                    double* scond, double* amax, lapack_int *info );
+void LAPACK_cpoequ( lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, float* s, float* scond, float* amax,
+                    lapack_int *info );
+void LAPACK_zpoequ( lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, double* s, double* scond, double* amax,
+                    lapack_int *info );
+void LAPACK_dpoequb( lapack_int* n, const double* a, lapack_int* lda, double* s,
+                     double* scond, double* amax, lapack_int *info );
+void LAPACK_spoequb( lapack_int* n, const float* a, lapack_int* lda, float* s,
+                     float* scond, float* amax, lapack_int *info );
+void LAPACK_zpoequb( lapack_int* n, const lapack_complex_double* a,
+                     lapack_int* lda, double* s, double* scond, double* amax,
+                     lapack_int *info );
+void LAPACK_cpoequb( lapack_int* n, const lapack_complex_float* a,
+                     lapack_int* lda, float* s, float* scond, float* amax,
+                     lapack_int *info );
+void LAPACK_sppequ( char* uplo, lapack_int* n, const float* ap, float* s,
+                    float* scond, float* amax, lapack_int *info );
+void LAPACK_dppequ( char* uplo, lapack_int* n, const double* ap, double* s,
+                    double* scond, double* amax, lapack_int *info );
+void LAPACK_cppequ( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    float* s, float* scond, float* amax, lapack_int *info );
+void LAPACK_zppequ( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    double* s, double* scond, double* amax, lapack_int *info );
+void LAPACK_spbequ( char* uplo, lapack_int* n, lapack_int* kd, const float* ab,
+                    lapack_int* ldab, float* s, float* scond, float* amax,
+                    lapack_int *info );
+void LAPACK_dpbequ( char* uplo, lapack_int* n, lapack_int* kd, const double* ab,
+                    lapack_int* ldab, double* s, double* scond, double* amax,
+                    lapack_int *info );
+void LAPACK_cpbequ( char* uplo, lapack_int* n, lapack_int* kd,
+                    const lapack_complex_float* ab, lapack_int* ldab, float* s,
+                    float* scond, float* amax, lapack_int *info );
+void LAPACK_zpbequ( char* uplo, lapack_int* n, lapack_int* kd,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    double* s, double* scond, double* amax, lapack_int *info );
+void LAPACK_dsyequb( char* uplo, lapack_int* n, const double* a,
+                     lapack_int* lda, double* s, double* scond, double* amax,
+                     double* work, lapack_int *info );
+void LAPACK_ssyequb( char* uplo, lapack_int* n, const float* a, lapack_int* lda,
+                     float* s, float* scond, float* amax, float* work,
+                     lapack_int *info );
+void LAPACK_zsyequb( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                     lapack_int* lda, double* s, double* scond, double* amax,
+                     lapack_complex_double* work, lapack_int *info );
+void LAPACK_csyequb( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                     lapack_int* lda, float* s, float* scond, float* amax,
+                     lapack_complex_float* work, lapack_int *info );
+void LAPACK_zheequb( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                     lapack_int* lda, double* s, double* scond, double* amax,
+                     lapack_complex_double* work, lapack_int *info );
+void LAPACK_cheequb( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                     lapack_int* lda, float* s, float* scond, float* amax,
+                     lapack_complex_float* work, lapack_int *info );
+void LAPACK_sgesv( lapack_int* n, lapack_int* nrhs, float* a, lapack_int* lda,
+                   lapack_int* ipiv, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dgesv( lapack_int* n, lapack_int* nrhs, double* a, lapack_int* lda,
+                   lapack_int* ipiv, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cgesv( lapack_int* n, lapack_int* nrhs, lapack_complex_float* a,
+                   lapack_int* lda, lapack_int* ipiv, lapack_complex_float* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_zgesv( lapack_int* n, lapack_int* nrhs, lapack_complex_double* a,
+                   lapack_int* lda, lapack_int* ipiv, lapack_complex_double* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_dsgesv( lapack_int* n, lapack_int* nrhs, double* a, lapack_int* lda,
+                    lapack_int* ipiv, double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* work, float* swork,
+                    lapack_int* iter, lapack_int *info );
+void LAPACK_zcgesv( lapack_int* n, lapack_int* nrhs, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    lapack_complex_double* work, lapack_complex_float* swork,
+                    double* rwork, lapack_int* iter, lapack_int *info );
+void LAPACK_sgesvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                    lapack_int* ipiv, char* equed, float* r, float* c, float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgesvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                    lapack_int* ipiv, char* equed, double* r, double* c,
+                    double* b, lapack_int* ldb, double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgesvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* af, lapack_int* ldaf,
+                    lapack_int* ipiv, char* equed, float* r, float* c,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgesvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* af, lapack_int* ldaf,
+                    lapack_int* ipiv, char* equed, double* r, double* c,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dgesvxx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                     double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* r, double* c,
+                     double* b, lapack_int* ldb, double* x, lapack_int* ldx,
+                     double* rcond, double* rpvgrw, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgesvxx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                     float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* r, float* c,
+                     float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                     float* rcond, float* rpvgrw, float* berr,
+                     lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zgesvxx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* r, double* c,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cgesvxx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* r, float* c,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sgbsv( lapack_int* n, lapack_int* kl, lapack_int* ku,
+                   lapack_int* nrhs, float* ab, lapack_int* ldab,
+                   lapack_int* ipiv, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dgbsv( lapack_int* n, lapack_int* kl, lapack_int* ku,
+                   lapack_int* nrhs, double* ab, lapack_int* ldab,
+                   lapack_int* ipiv, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cgbsv( lapack_int* n, lapack_int* kl, lapack_int* ku,
+                   lapack_int* nrhs, lapack_complex_float* ab, lapack_int* ldab,
+                   lapack_int* ipiv, lapack_complex_float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_zgbsv( lapack_int* n, lapack_int* kl, lapack_int* ku,
+                   lapack_int* nrhs, lapack_complex_double* ab,
+                   lapack_int* ldab, lapack_int* ipiv, lapack_complex_double* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_sgbsvx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_int* nrhs, float* ab,
+                    lapack_int* ldab, float* afb, lapack_int* ldafb,
+                    lapack_int* ipiv, char* equed, float* r, float* c, float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgbsvx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_int* nrhs, double* ab,
+                    lapack_int* ldab, double* afb, lapack_int* ldafb,
+                    lapack_int* ipiv, char* equed, double* r, double* c,
+                    double* b, lapack_int* ldb, double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgbsvx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_int* nrhs, lapack_complex_float* ab,
+                    lapack_int* ldab, lapack_complex_float* afb,
+                    lapack_int* ldafb, lapack_int* ipiv, char* equed, float* r,
+                    float* c, lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgbsvx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_int* nrhs, lapack_complex_double* ab,
+                    lapack_int* ldab, lapack_complex_double* afb,
+                    lapack_int* ldafb, lapack_int* ipiv, char* equed, double* r,
+                    double* c, lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dgbsvxx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, double* ab,
+                     lapack_int* ldab, double* afb, lapack_int* ldafb,
+                     lapack_int* ipiv, char* equed, double* r, double* c,
+                     double* b, lapack_int* ldb, double* x, lapack_int* ldx,
+                     double* rcond, double* rpvgrw, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgbsvxx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, float* ab,
+                     lapack_int* ldab, float* afb, lapack_int* ldafb,
+                     lapack_int* ipiv, char* equed, float* r, float* c,
+                     float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                     float* rcond, float* rpvgrw, float* berr,
+                     lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zgbsvxx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs,
+                     lapack_complex_double* ab, lapack_int* ldab,
+                     lapack_complex_double* afb, lapack_int* ldafb,
+                     lapack_int* ipiv, char* equed, double* r, double* c,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cgbsvxx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, lapack_complex_float* ab,
+                     lapack_int* ldab, lapack_complex_float* afb,
+                     lapack_int* ldafb, lapack_int* ipiv, char* equed, float* r,
+                     float* c, lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sgtsv( lapack_int* n, lapack_int* nrhs, float* dl, float* d,
+                   float* du, float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_dgtsv( lapack_int* n, lapack_int* nrhs, double* dl, double* d,
+                   double* du, double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_cgtsv( lapack_int* n, lapack_int* nrhs, lapack_complex_float* dl,
+                   lapack_complex_float* d, lapack_complex_float* du,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zgtsv( lapack_int* n, lapack_int* nrhs, lapack_complex_double* dl,
+                   lapack_complex_double* d, lapack_complex_double* du,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_sgtsvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* dl, const float* d, const float* du,
+                    float* dlf, float* df, float* duf, float* du2,
+                    lapack_int* ipiv, const float* b, lapack_int* ldb, float* x,
+                    lapack_int* ldx, float* rcond, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dgtsvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* dl, const double* d, const double* du,
+                    double* dlf, double* df, double* duf, double* du2,
+                    lapack_int* ipiv, const double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* rcond, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cgtsvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* dl,
+                    const lapack_complex_float* d,
+                    const lapack_complex_float* du, lapack_complex_float* dlf,
+                    lapack_complex_float* df, lapack_complex_float* duf,
+                    lapack_complex_float* du2, lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgtsvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* dl,
+                    const lapack_complex_double* d,
+                    const lapack_complex_double* du, lapack_complex_double* dlf,
+                    lapack_complex_double* df, lapack_complex_double* duf,
+                    lapack_complex_double* du2, lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sposv( char* uplo, lapack_int* n, lapack_int* nrhs, float* a,
+                   lapack_int* lda, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dposv( char* uplo, lapack_int* n, lapack_int* nrhs, double* a,
+                   lapack_int* lda, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cposv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zposv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dsposv( char* uplo, lapack_int* n, lapack_int* nrhs, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* work, float* swork,
+                    lapack_int* iter, lapack_int *info );
+void LAPACK_zcposv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx,
+                    lapack_complex_double* work, lapack_complex_float* swork,
+                    double* rwork, lapack_int* iter, lapack_int *info );
+void LAPACK_sposvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                    char* equed, float* s, float* b, lapack_int* ldb, float* x,
+                    lapack_int* ldx, float* rcond, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dposvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                    char* equed, double* s, double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* rcond, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cposvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* af, lapack_int* ldaf, char* equed,
+                    float* s, lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zposvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* af, lapack_int* ldaf, char* equed,
+                    double* s, lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dposvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                     char* equed, double* s, double* b, lapack_int* ldb,
+                     double* x, lapack_int* ldx, double* rcond, double* rpvgrw,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params, double* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_sposvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                     char* equed, float* s, float* b, lapack_int* ldb, float* x,
+                     lapack_int* ldx, float* rcond, float* rpvgrw, float* berr,
+                     lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zposvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* af, lapack_int* ldaf, char* equed,
+                     double* s, lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cposvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* af, lapack_int* ldaf, char* equed,
+                     float* s, lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sppsv( char* uplo, lapack_int* n, lapack_int* nrhs, float* ap,
+                   float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_dppsv( char* uplo, lapack_int* n, lapack_int* nrhs, double* ap,
+                   double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_cppsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* ap, lapack_complex_float* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_zppsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* ap, lapack_complex_double* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_sppsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    float* ap, float* afp, char* equed, float* s, float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dppsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    double* ap, double* afp, char* equed, double* s, double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cppsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* ap, lapack_complex_float* afp,
+                    char* equed, float* s, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zppsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* ap, lapack_complex_double* afp,
+                    char* equed, double* s, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_spbsv( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                   float* ab, lapack_int* ldab, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dpbsv( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                   double* ab, lapack_int* ldab, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cpbsv( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                   lapack_complex_float* ab, lapack_int* ldab,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zpbsv( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                   lapack_complex_double* ab, lapack_int* ldab,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_spbsvx( char* fact, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_int* nrhs, float* ab, lapack_int* ldab, float* afb,
+                    lapack_int* ldafb, char* equed, float* s, float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dpbsvx( char* fact, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_int* nrhs, double* ab, lapack_int* ldab, double* afb,
+                    lapack_int* ldafb, char* equed, double* s, double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cpbsvx( char* fact, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_int* nrhs, lapack_complex_float* ab,
+                    lapack_int* ldab, lapack_complex_float* afb,
+                    lapack_int* ldafb, char* equed, float* s,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zpbsvx( char* fact, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_int* nrhs, lapack_complex_double* ab,
+                    lapack_int* ldab, lapack_complex_double* afb,
+                    lapack_int* ldafb, char* equed, double* s,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sptsv( lapack_int* n, lapack_int* nrhs, float* d, float* e,
+                   float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_dptsv( lapack_int* n, lapack_int* nrhs, double* d, double* e,
+                   double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_cptsv( lapack_int* n, lapack_int* nrhs, float* d,
+                   lapack_complex_float* e, lapack_complex_float* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_zptsv( lapack_int* n, lapack_int* nrhs, double* d,
+                   lapack_complex_double* e, lapack_complex_double* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_sptsvx( char* fact, lapack_int* n, lapack_int* nrhs, const float* d,
+                    const float* e, float* df, float* ef, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int *info );
+void LAPACK_dptsvx( char* fact, lapack_int* n, lapack_int* nrhs,
+                    const double* d, const double* e, double* df, double* ef,
+                    const double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* rcond, double* ferr, double* berr,
+                    double* work, lapack_int *info );
+void LAPACK_cptsvx( char* fact, lapack_int* n, lapack_int* nrhs, const float* d,
+                    const lapack_complex_float* e, float* df,
+                    lapack_complex_float* ef, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zptsvx( char* fact, lapack_int* n, lapack_int* nrhs,
+                    const double* d, const lapack_complex_double* e, double* df,
+                    lapack_complex_double* ef, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_ssysv( char* uplo, lapack_int* n, lapack_int* nrhs, float* a,
+                   lapack_int* lda, lapack_int* ipiv, float* b, lapack_int* ldb,
+                   float* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_dsysv( char* uplo, lapack_int* n, lapack_int* nrhs, double* a,
+                   lapack_int* lda, lapack_int* ipiv, double* b,
+                   lapack_int* ldb, double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_csysv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* a, lapack_int* lda, lapack_int* ipiv,
+                   lapack_complex_float* b, lapack_int* ldb,
+                   lapack_complex_float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_zsysv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* a, lapack_int* lda, lapack_int* ipiv,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_ssysvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* a, lapack_int* lda, float* af,
+                    lapack_int* ldaf, lapack_int* ipiv, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dsysvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, double* af,
+                    lapack_int* ldaf, lapack_int* ipiv, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_csysvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* af, lapack_int* ldaf,
+                    lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zsysvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* af, lapack_int* ldaf,
+                    lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_dsysvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* s, double* b,
+                     lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params, double* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_ssysvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* s, float* b,
+                     lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params, float* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_zsysvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* s,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_csysvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* s,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_chesv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* a, lapack_int* lda, lapack_int* ipiv,
+                   lapack_complex_float* b, lapack_int* ldb,
+                   lapack_complex_float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_zhesv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* a, lapack_int* lda, lapack_int* ipiv,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_chesvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* af, lapack_int* ldaf,
+                    lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhesvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* af, lapack_int* ldaf,
+                    lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_zhesvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* s,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_chesvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* s,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sspsv( char* uplo, lapack_int* n, lapack_int* nrhs, float* ap,
+                   lapack_int* ipiv, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dspsv( char* uplo, lapack_int* n, lapack_int* nrhs, double* ap,
+                   lapack_int* ipiv, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cspsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* ap, lapack_int* ipiv,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zspsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* ap, lapack_int* ipiv,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_sspsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, float* afp, lapack_int* ipiv,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dspsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, double* afp, lapack_int* ipiv,
+                    const double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* rcond, double* ferr, double* berr,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cspsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, lapack_complex_float* afp,
+                    lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zspsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, lapack_complex_double* afp,
+                    lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_chpsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* ap, lapack_int* ipiv,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zhpsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* ap, lapack_int* ipiv,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_chpsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, lapack_complex_float* afp,
+                    lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhpsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, lapack_complex_double* afp,
+                    lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sgeqrf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgeqrf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgeqrf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgeqrf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgeqpf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* jpvt, float* tau, float* work,
+                    lapack_int *info );
+void LAPACK_dgeqpf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* jpvt, double* tau, double* work,
+                    lapack_int *info );
+void LAPACK_cgeqpf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* jpvt,
+                    lapack_complex_float* tau, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgeqpf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* jpvt,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgeqp3( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* jpvt, float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dgeqp3( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* jpvt, double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cgeqp3( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* jpvt,
+                    lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int *info );
+void LAPACK_zgeqp3( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* jpvt,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int *info );
+void LAPACK_sorgqr( lapack_int* m, lapack_int* n, lapack_int* k, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorgqr( lapack_int* m, lapack_int* n, lapack_int* k, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sormqr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormqr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cungqr( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zungqr( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmqr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmqr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgelqf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgelqf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgelqf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgelqf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sorglq( lapack_int* m, lapack_int* n, lapack_int* k, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorglq( lapack_int* m, lapack_int* n, lapack_int* k, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sormlq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormlq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cunglq( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zunglq( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmlq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmlq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgeqlf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgeqlf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgeqlf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgeqlf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sorgql( lapack_int* m, lapack_int* n, lapack_int* k, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorgql( lapack_int* m, lapack_int* n, lapack_int* k, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cungql( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zungql( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sormql( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormql( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cunmql( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmql( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgerqf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgerqf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgerqf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgerqf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sorgrq( lapack_int* m, lapack_int* n, lapack_int* k, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorgrq( lapack_int* m, lapack_int* n, lapack_int* k, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cungrq( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zungrq( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sormrq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormrq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cunmrq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmrq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_stzrzf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dtzrzf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ctzrzf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ztzrzf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sormrz( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, lapack_int* l, const float* a,
+                    lapack_int* lda, const float* tau, float* c,
+                    lapack_int* ldc, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dormrz( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, lapack_int* l, const double* a,
+                    lapack_int* lda, const double* tau, double* c,
+                    lapack_int* ldc, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmrz( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, lapack_int* l, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmrz( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, lapack_int* l,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau, lapack_complex_double* c,
+                    lapack_int* ldc, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sggqrf( lapack_int* n, lapack_int* m, lapack_int* p, float* a,
+                    lapack_int* lda, float* taua, float* b, lapack_int* ldb,
+                    float* taub, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dggqrf( lapack_int* n, lapack_int* m, lapack_int* p, double* a,
+                    lapack_int* lda, double* taua, double* b, lapack_int* ldb,
+                    double* taub, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cggqrf( lapack_int* n, lapack_int* m, lapack_int* p,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* taua, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* taub,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zggqrf( lapack_int* n, lapack_int* m, lapack_int* p,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* taua, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* taub,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sggrqf( lapack_int* m, lapack_int* p, lapack_int* n, float* a,
+                    lapack_int* lda, float* taua, float* b, lapack_int* ldb,
+                    float* taub, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dggrqf( lapack_int* m, lapack_int* p, lapack_int* n, double* a,
+                    lapack_int* lda, double* taua, double* b, lapack_int* ldb,
+                    double* taub, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cggrqf( lapack_int* m, lapack_int* p, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* taua, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* taub,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zggrqf( lapack_int* m, lapack_int* p, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* taua, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* taub,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgebrd( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* d, float* e, float* tauq, float* taup, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dgebrd( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* d, double* e, double* tauq, double* taup,
+                    double* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_cgebrd( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, float* d, float* e,
+                    lapack_complex_float* tauq, lapack_complex_float* taup,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgebrd( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, double* d, double* e,
+                    lapack_complex_double* tauq, lapack_complex_double* taup,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgbbrd( char* vect, lapack_int* m, lapack_int* n, lapack_int* ncc,
+                    lapack_int* kl, lapack_int* ku, float* ab, lapack_int* ldab,
+                    float* d, float* e, float* q, lapack_int* ldq, float* pt,
+                    lapack_int* ldpt, float* c, lapack_int* ldc, float* work,
+                    lapack_int *info );
+void LAPACK_dgbbrd( char* vect, lapack_int* m, lapack_int* n, lapack_int* ncc,
+                    lapack_int* kl, lapack_int* ku, double* ab,
+                    lapack_int* ldab, double* d, double* e, double* q,
+                    lapack_int* ldq, double* pt, lapack_int* ldpt, double* c,
+                    lapack_int* ldc, double* work, lapack_int *info );
+void LAPACK_cgbbrd( char* vect, lapack_int* m, lapack_int* n, lapack_int* ncc,
+                    lapack_int* kl, lapack_int* ku, lapack_complex_float* ab,
+                    lapack_int* ldab, float* d, float* e,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* pt, lapack_int* ldpt,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgbbrd( char* vect, lapack_int* m, lapack_int* n, lapack_int* ncc,
+                    lapack_int* kl, lapack_int* ku, lapack_complex_double* ab,
+                    lapack_int* ldab, double* d, double* e,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* pt, lapack_int* ldpt,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sorgbr( char* vect, lapack_int* m, lapack_int* n, lapack_int* k,
+                    float* a, lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorgbr( char* vect, lapack_int* m, lapack_int* n, lapack_int* k,
+                    double* a, lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sormbr( char* vect, char* side, char* trans, lapack_int* m,
+                    lapack_int* n, lapack_int* k, const float* a,
+                    lapack_int* lda, const float* tau, float* c,
+                    lapack_int* ldc, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dormbr( char* vect, char* side, char* trans, lapack_int* m,
+                    lapack_int* n, lapack_int* k, const double* a,
+                    lapack_int* lda, const double* tau, double* c,
+                    lapack_int* ldc, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cungbr( char* vect, lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zungbr( char* vect, lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmbr( char* vect, char* side, char* trans, lapack_int* m,
+                    lapack_int* n, lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmbr( char* vect, char* side, char* trans, lapack_int* m,
+                    lapack_int* n, lapack_int* k,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau, lapack_complex_double* c,
+                    lapack_int* ldc, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sbdsqr( char* uplo, lapack_int* n, lapack_int* ncvt,
+                    lapack_int* nru, lapack_int* ncc, float* d, float* e,
+                    float* vt, lapack_int* ldvt, float* u, lapack_int* ldu,
+                    float* c, lapack_int* ldc, float* work, lapack_int *info );
+void LAPACK_dbdsqr( char* uplo, lapack_int* n, lapack_int* ncvt,
+                    lapack_int* nru, lapack_int* ncc, double* d, double* e,
+                    double* vt, lapack_int* ldvt, double* u, lapack_int* ldu,
+                    double* c, lapack_int* ldc, double* work,
+                    lapack_int *info );
+void LAPACK_cbdsqr( char* uplo, lapack_int* n, lapack_int* ncvt,
+                    lapack_int* nru, lapack_int* ncc, float* d, float* e,
+                    lapack_complex_float* vt, lapack_int* ldvt,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* c, lapack_int* ldc, float* work,
+                    lapack_int *info );
+void LAPACK_zbdsqr( char* uplo, lapack_int* n, lapack_int* ncvt,
+                    lapack_int* nru, lapack_int* ncc, double* d, double* e,
+                    lapack_complex_double* vt, lapack_int* ldvt,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* c, lapack_int* ldc, double* work,
+                    lapack_int *info );
+void LAPACK_sbdsdc( char* uplo, char* compq, lapack_int* n, float* d, float* e,
+                    float* u, lapack_int* ldu, float* vt, lapack_int* ldvt,
+                    float* q, lapack_int* iq, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dbdsdc( char* uplo, char* compq, lapack_int* n, double* d,
+                    double* e, double* u, lapack_int* ldu, double* vt,
+                    lapack_int* ldvt, double* q, lapack_int* iq, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ssytrd( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    float* d, float* e, float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dsytrd( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    double* d, double* e, double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sorgtr( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    const float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dorgtr( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    const double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sormtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_chetrd( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, float* d, float* e,
+                    lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zhetrd( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, double* d, double* e,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cungtr( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zungtr( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ssptrd( char* uplo, lapack_int* n, float* ap, float* d, float* e,
+                    float* tau, lapack_int *info );
+void LAPACK_dsptrd( char* uplo, lapack_int* n, double* ap, double* d, double* e,
+                    double* tau, lapack_int *info );
+void LAPACK_sopgtr( char* uplo, lapack_int* n, const float* ap,
+                    const float* tau, float* q, lapack_int* ldq, float* work,
+                    lapack_int *info );
+void LAPACK_dopgtr( char* uplo, lapack_int* n, const double* ap,
+                    const double* tau, double* q, lapack_int* ldq, double* work,
+                    lapack_int *info );
+void LAPACK_sopmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const float* ap, const float* tau, float* c,
+                    lapack_int* ldc, float* work, lapack_int *info );
+void LAPACK_dopmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const double* ap, const double* tau,
+                    double* c, lapack_int* ldc, double* work,
+                    lapack_int *info );
+void LAPACK_chptrd( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    float* d, float* e, lapack_complex_float* tau,
+                    lapack_int *info );
+void LAPACK_zhptrd( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    double* d, double* e, lapack_complex_double* tau,
+                    lapack_int *info );
+void LAPACK_cupgtr( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    const lapack_complex_float* tau, lapack_complex_float* q,
+                    lapack_int* ldq, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zupgtr( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    const lapack_complex_double* tau, lapack_complex_double* q,
+                    lapack_int* ldq, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_cupmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const lapack_complex_float* ap,
+                    const lapack_complex_float* tau, lapack_complex_float* c,
+                    lapack_int* ldc, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zupmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const lapack_complex_double* ap,
+                    const lapack_complex_double* tau, lapack_complex_double* c,
+                    lapack_int* ldc, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_ssbtrd( char* vect, char* uplo, lapack_int* n, lapack_int* kd,
+                    float* ab, lapack_int* ldab, float* d, float* e, float* q,
+                    lapack_int* ldq, float* work, lapack_int *info );
+void LAPACK_dsbtrd( char* vect, char* uplo, lapack_int* n, lapack_int* kd,
+                    double* ab, lapack_int* ldab, double* d, double* e,
+                    double* q, lapack_int* ldq, double* work,
+                    lapack_int *info );
+void LAPACK_chbtrd( char* vect, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_float* ab, lapack_int* ldab, float* d,
+                    float* e, lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zhbtrd( char* vect, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_double* ab, lapack_int* ldab, double* d,
+                    double* e, lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_ssterf( lapack_int* n, float* d, float* e, lapack_int *info );
+void LAPACK_dsterf( lapack_int* n, double* d, double* e, lapack_int *info );
+void LAPACK_ssteqr( char* compz, lapack_int* n, float* d, float* e, float* z,
+                    lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dsteqr( char* compz, lapack_int* n, double* d, double* e, double* z,
+                    lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_csteqr( char* compz, lapack_int* n, float* d, float* e,
+                    lapack_complex_float* z, lapack_int* ldz, float* work,
+                    lapack_int *info );
+void LAPACK_zsteqr( char* compz, lapack_int* n, double* d, double* e,
+                    lapack_complex_double* z, lapack_int* ldz, double* work,
+                    lapack_int *info );
+void LAPACK_sstemr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    lapack_int* m, float* w, float* z, lapack_int* ldz,
+                    lapack_int* nzc, lapack_int* isuppz, lapack_logical* tryrac,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dstemr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, lapack_int* m, double* w, double* z,
+                    lapack_int* ldz, lapack_int* nzc, lapack_int* isuppz,
+                    lapack_logical* tryrac, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_cstemr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_int* nzc, lapack_int* isuppz,
+                    lapack_logical* tryrac, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_zstemr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, lapack_int* m, double* w,
+                    lapack_complex_double* z, lapack_int* ldz, lapack_int* nzc,
+                    lapack_int* isuppz, lapack_logical* tryrac, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_sstedc( char* compz, lapack_int* n, float* d, float* e, float* z,
+                    lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dstedc( char* compz, lapack_int* n, double* d, double* e, double* z,
+                    lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_cstedc( char* compz, lapack_int* n, float* d, float* e,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zstedc( char* compz, lapack_int* n, double* d, double* e,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sstegr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w, float* z,
+                    lapack_int* ldz, lapack_int* isuppz, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_dstegr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, lapack_int* isuppz,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_cstegr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_int* isuppz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_zstegr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_int* isuppz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_spteqr( char* compz, lapack_int* n, float* d, float* e, float* z,
+                    lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dpteqr( char* compz, lapack_int* n, double* d, double* e, double* z,
+                    lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_cpteqr( char* compz, lapack_int* n, float* d, float* e,
+                    lapack_complex_float* z, lapack_int* ldz, float* work,
+                    lapack_int *info );
+void LAPACK_zpteqr( char* compz, lapack_int* n, double* d, double* e,
+                    lapack_complex_double* z, lapack_int* ldz, double* work,
+                    lapack_int *info );
+void LAPACK_sstebz( char* range, char* order, lapack_int* n, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    const float* d, const float* e, lapack_int* m,
+                    lapack_int* nsplit, float* w, lapack_int* iblock,
+                    lapack_int* isplit, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dstebz( char* range, char* order, lapack_int* n, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    const double* d, const double* e, lapack_int* m,
+                    lapack_int* nsplit, double* w, lapack_int* iblock,
+                    lapack_int* isplit, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_sstein( lapack_int* n, const float* d, const float* e,
+                    lapack_int* m, const float* w, const lapack_int* iblock,
+                    const lapack_int* isplit, float* z, lapack_int* ldz,
+                    float* work, lapack_int* iwork, lapack_int* ifailv,
+                    lapack_int *info );
+void LAPACK_dstein( lapack_int* n, const double* d, const double* e,
+                    lapack_int* m, const double* w, const lapack_int* iblock,
+                    const lapack_int* isplit, double* z, lapack_int* ldz,
+                    double* work, lapack_int* iwork, lapack_int* ifailv,
+                    lapack_int *info );
+void LAPACK_cstein( lapack_int* n, const float* d, const float* e,
+                    lapack_int* m, const float* w, const lapack_int* iblock,
+                    const lapack_int* isplit, lapack_complex_float* z,
+                    lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifailv, lapack_int *info );
+void LAPACK_zstein( lapack_int* n, const double* d, const double* e,
+                    lapack_int* m, const double* w, const lapack_int* iblock,
+                    const lapack_int* isplit, lapack_complex_double* z,
+                    lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifailv, lapack_int *info );
+void LAPACK_sdisna( char* job, lapack_int* m, lapack_int* n, const float* d,
+                    float* sep, lapack_int *info );
+void LAPACK_ddisna( char* job, lapack_int* m, lapack_int* n, const double* d,
+                    double* sep, lapack_int *info );
+void LAPACK_ssygst( lapack_int* itype, char* uplo, lapack_int* n, float* a,
+                    lapack_int* lda, const float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dsygst( lapack_int* itype, char* uplo, lapack_int* n, double* a,
+                    lapack_int* lda, const double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_chegst( lapack_int* itype, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zhegst( lapack_int* itype, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_sspgst( lapack_int* itype, char* uplo, lapack_int* n, float* ap,
+                    const float* bp, lapack_int *info );
+void LAPACK_dspgst( lapack_int* itype, char* uplo, lapack_int* n, double* ap,
+                    const double* bp, lapack_int *info );
+void LAPACK_chpgst( lapack_int* itype, char* uplo, lapack_int* n,
+                    lapack_complex_float* ap, const lapack_complex_float* bp,
+                    lapack_int *info );
+void LAPACK_zhpgst( lapack_int* itype, char* uplo, lapack_int* n,
+                    lapack_complex_double* ap, const lapack_complex_double* bp,
+                    lapack_int *info );
+void LAPACK_ssbgst( char* vect, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, float* ab, lapack_int* ldab,
+                    const float* bb, lapack_int* ldbb, float* x,
+                    lapack_int* ldx, float* work, lapack_int *info );
+void LAPACK_dsbgst( char* vect, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, double* ab, lapack_int* ldab,
+                    const double* bb, lapack_int* ldbb, double* x,
+                    lapack_int* ldx, double* work, lapack_int *info );
+void LAPACK_chbgst( char* vect, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, lapack_complex_float* ab, lapack_int* ldab,
+                    const lapack_complex_float* bb, lapack_int* ldbb,
+                    lapack_complex_float* x, lapack_int* ldx,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhbgst( char* vect, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, lapack_complex_double* ab, lapack_int* ldab,
+                    const lapack_complex_double* bb, lapack_int* ldbb,
+                    lapack_complex_double* x, lapack_int* ldx,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_spbstf( char* uplo, lapack_int* n, lapack_int* kb, float* bb,
+                    lapack_int* ldbb, lapack_int *info );
+void LAPACK_dpbstf( char* uplo, lapack_int* n, lapack_int* kb, double* bb,
+                    lapack_int* ldbb, lapack_int *info );
+void LAPACK_cpbstf( char* uplo, lapack_int* n, lapack_int* kb,
+                    lapack_complex_float* bb, lapack_int* ldbb,
+                    lapack_int *info );
+void LAPACK_zpbstf( char* uplo, lapack_int* n, lapack_int* kb,
+                    lapack_complex_double* bb, lapack_int* ldbb,
+                    lapack_int *info );
+void LAPACK_sgehrd( lapack_int* n, lapack_int* ilo, lapack_int* ihi, float* a,
+                    lapack_int* lda, float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgehrd( lapack_int* n, lapack_int* ilo, lapack_int* ihi, double* a,
+                    lapack_int* lda, double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cgehrd( lapack_int* n, lapack_int* ilo, lapack_int* ihi,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zgehrd( lapack_int* n, lapack_int* ilo, lapack_int* ihi,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sorghr( lapack_int* n, lapack_int* ilo, lapack_int* ihi, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorghr( lapack_int* n, lapack_int* ilo, lapack_int* ihi, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sormhr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, const float* a,
+                    lapack_int* lda, const float* tau, float* c,
+                    lapack_int* ldc, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dormhr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, const double* a,
+                    lapack_int* lda, const double* tau, double* c,
+                    lapack_int* ldc, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunghr( lapack_int* n, lapack_int* ilo, lapack_int* ihi,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zunghr( lapack_int* n, lapack_int* ilo, lapack_int* ihi,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmhr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* c,
+                    lapack_int* ldc, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zunmhr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau, lapack_complex_double* c,
+                    lapack_int* ldc, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sgebal( char* job, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* ilo, lapack_int* ihi, float* scale,
+                    lapack_int *info );
+void LAPACK_dgebal( char* job, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* ilo, lapack_int* ihi, double* scale,
+                    lapack_int *info );
+void LAPACK_cgebal( char* job, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ilo, lapack_int* ihi,
+                    float* scale, lapack_int *info );
+void LAPACK_zgebal( char* job, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ilo, lapack_int* ihi,
+                    double* scale, lapack_int *info );
+void LAPACK_sgebak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const float* scale, lapack_int* m,
+                    float* v, lapack_int* ldv, lapack_int *info );
+void LAPACK_dgebak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const double* scale, lapack_int* m,
+                    double* v, lapack_int* ldv, lapack_int *info );
+void LAPACK_cgebak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const float* scale, lapack_int* m,
+                    lapack_complex_float* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_zgebak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const double* scale, lapack_int* m,
+                    lapack_complex_double* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_shseqr( char* job, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, float* h, lapack_int* ldh, float* wr,
+                    float* wi, float* z, lapack_int* ldz, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dhseqr( char* job, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, double* h, lapack_int* ldh, double* wr,
+                    double* wi, double* z, lapack_int* ldz, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_chseqr( char* job, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, lapack_complex_float* h, lapack_int* ldh,
+                    lapack_complex_float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zhseqr( char* job, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, lapack_complex_double* h, lapack_int* ldh,
+                    lapack_complex_double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_shsein( char* job, char* eigsrc, char* initv,
+                    lapack_logical* select, lapack_int* n, const float* h,
+                    lapack_int* ldh, float* wr, const float* wi, float* vl,
+                    lapack_int* ldvl, float* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, float* work,
+                    lapack_int* ifaill, lapack_int* ifailr, lapack_int *info );
+void LAPACK_dhsein( char* job, char* eigsrc, char* initv,
+                    lapack_logical* select, lapack_int* n, const double* h,
+                    lapack_int* ldh, double* wr, const double* wi, double* vl,
+                    lapack_int* ldvl, double* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, double* work,
+                    lapack_int* ifaill, lapack_int* ifailr, lapack_int *info );
+void LAPACK_chsein( char* job, char* eigsrc, char* initv,
+                    const lapack_logical* select, lapack_int* n,
+                    const lapack_complex_float* h, lapack_int* ldh,
+                    lapack_complex_float* w, lapack_complex_float* vl,
+                    lapack_int* ldvl, lapack_complex_float* vr,
+                    lapack_int* ldvr, lapack_int* mm, lapack_int* m,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int* ifaill, lapack_int* ifailr, lapack_int *info );
+void LAPACK_zhsein( char* job, char* eigsrc, char* initv,
+                    const lapack_logical* select, lapack_int* n,
+                    const lapack_complex_double* h, lapack_int* ldh,
+                    lapack_complex_double* w, lapack_complex_double* vl,
+                    lapack_int* ldvl, lapack_complex_double* vr,
+                    lapack_int* ldvr, lapack_int* mm, lapack_int* m,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int* ifaill, lapack_int* ifailr, lapack_int *info );
+void LAPACK_strevc( char* side, char* howmny, lapack_logical* select,
+                    lapack_int* n, const float* t, lapack_int* ldt, float* vl,
+                    lapack_int* ldvl, float* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, float* work,
+                    lapack_int *info );
+void LAPACK_dtrevc( char* side, char* howmny, lapack_logical* select,
+                    lapack_int* n, const double* t, lapack_int* ldt, double* vl,
+                    lapack_int* ldvl, double* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, double* work,
+                    lapack_int *info );
+void LAPACK_ctrevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* vl, lapack_int* ldvl,
+                    lapack_complex_float* vr, lapack_int* ldvr, lapack_int* mm,
+                    lapack_int* m, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztrevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* vl, lapack_int* ldvl,
+                    lapack_complex_double* vr, lapack_int* ldvr, lapack_int* mm,
+                    lapack_int* m, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_strsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const float* t, lapack_int* ldt,
+                    const float* vl, lapack_int* ldvl, const float* vr,
+                    lapack_int* ldvr, float* s, float* sep, lapack_int* mm,
+                    lapack_int* m, float* work, lapack_int* ldwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dtrsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const double* t, lapack_int* ldt,
+                    const double* vl, lapack_int* ldvl, const double* vr,
+                    lapack_int* ldvr, double* s, double* sep, lapack_int* mm,
+                    lapack_int* m, double* work, lapack_int* ldwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ctrsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_float* t,
+                    lapack_int* ldt, const lapack_complex_float* vl,
+                    lapack_int* ldvl, const lapack_complex_float* vr,
+                    lapack_int* ldvr, float* s, float* sep, lapack_int* mm,
+                    lapack_int* m, lapack_complex_float* work,
+                    lapack_int* ldwork, float* rwork, lapack_int *info );
+void LAPACK_ztrsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_double* t,
+                    lapack_int* ldt, const lapack_complex_double* vl,
+                    lapack_int* ldvl, const lapack_complex_double* vr,
+                    lapack_int* ldvr, double* s, double* sep, lapack_int* mm,
+                    lapack_int* m, lapack_complex_double* work,
+                    lapack_int* ldwork, double* rwork, lapack_int *info );
+void LAPACK_strexc( char* compq, lapack_int* n, float* t, lapack_int* ldt,
+                    float* q, lapack_int* ldq, lapack_int* ifst,
+                    lapack_int* ilst, float* work, lapack_int *info );
+void LAPACK_dtrexc( char* compq, lapack_int* n, double* t, lapack_int* ldt,
+                    double* q, lapack_int* ldq, lapack_int* ifst,
+                    lapack_int* ilst, double* work, lapack_int *info );
+void LAPACK_ctrexc( char* compq, lapack_int* n, lapack_complex_float* t,
+                    lapack_int* ldt, lapack_complex_float* q, lapack_int* ldq,
+                    lapack_int* ifst, lapack_int* ilst, lapack_int *info );
+void LAPACK_ztrexc( char* compq, lapack_int* n, lapack_complex_double* t,
+                    lapack_int* ldt, lapack_complex_double* q, lapack_int* ldq,
+                    lapack_int* ifst, lapack_int* ilst, lapack_int *info );
+void LAPACK_strsen( char* job, char* compq, const lapack_logical* select,
+                    lapack_int* n, float* t, lapack_int* ldt, float* q,
+                    lapack_int* ldq, float* wr, float* wi, lapack_int* m,
+                    float* s, float* sep, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dtrsen( char* job, char* compq, const lapack_logical* select,
+                    lapack_int* n, double* t, lapack_int* ldt, double* q,
+                    lapack_int* ldq, double* wr, double* wi, lapack_int* m,
+                    double* s, double* sep, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_ctrsen( char* job, char* compq, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* w, lapack_int* m, float* s,
+                    float* sep, lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ztrsen( char* job, char* compq, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* w, lapack_int* m, double* s,
+                    double* sep, lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_strsyl( char* trana, char* tranb, lapack_int* isgn, lapack_int* m,
+                    lapack_int* n, const float* a, lapack_int* lda,
+                    const float* b, lapack_int* ldb, float* c, lapack_int* ldc,
+                    float* scale, lapack_int *info );
+void LAPACK_dtrsyl( char* trana, char* tranb, lapack_int* isgn, lapack_int* m,
+                    lapack_int* n, const double* a, lapack_int* lda,
+                    const double* b, lapack_int* ldb, double* c,
+                    lapack_int* ldc, double* scale, lapack_int *info );
+void LAPACK_ctrsyl( char* trana, char* tranb, lapack_int* isgn, lapack_int* m,
+                    lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* c, lapack_int* ldc,
+                    float* scale, lapack_int *info );
+void LAPACK_ztrsyl( char* trana, char* tranb, lapack_int* isgn, lapack_int* m,
+                    lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* c, lapack_int* ldc,
+                    double* scale, lapack_int *info );
+void LAPACK_sgghrd( char* compq, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, float* q, lapack_int* ldq, float* z,
+                    lapack_int* ldz, lapack_int *info );
+void LAPACK_dgghrd( char* compq, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, double* q, lapack_int* ldq, double* z,
+                    lapack_int* ldz, lapack_int *info );
+void LAPACK_cgghrd( char* compq, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_int *info );
+void LAPACK_zgghrd( char* compq, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_int *info );
+void LAPACK_sggbal( char* job, lapack_int* n, float* a, lapack_int* lda,
+                    float* b, lapack_int* ldb, lapack_int* ilo, lapack_int* ihi,
+                    float* lscale, float* rscale, float* work,
+                    lapack_int *info );
+void LAPACK_dggbal( char* job, lapack_int* n, double* a, lapack_int* lda,
+                    double* b, lapack_int* ldb, lapack_int* ilo,
+                    lapack_int* ihi, double* lscale, double* rscale,
+                    double* work, lapack_int *info );
+void LAPACK_cggbal( char* job, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int* ilo, lapack_int* ihi, float* lscale,
+                    float* rscale, float* work, lapack_int *info );
+void LAPACK_zggbal( char* job, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int* ilo, lapack_int* ihi, double* lscale,
+                    double* rscale, double* work, lapack_int *info );
+void LAPACK_sggbak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const float* lscale, const float* rscale,
+                    lapack_int* m, float* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_dggbak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const double* lscale, const double* rscale,
+                    lapack_int* m, double* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_cggbak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const float* lscale, const float* rscale,
+                    lapack_int* m, lapack_complex_float* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_zggbak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const double* lscale, const double* rscale,
+                    lapack_int* m, lapack_complex_double* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_shgeqz( char* job, char* compq, char* compz, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, float* h, lapack_int* ldh,
+                    float* t, lapack_int* ldt, float* alphar, float* alphai,
+                    float* beta, float* q, lapack_int* ldq, float* z,
+                    lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dhgeqz( char* job, char* compq, char* compz, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, double* h,
+                    lapack_int* ldh, double* t, lapack_int* ldt, double* alphar,
+                    double* alphai, double* beta, double* q, lapack_int* ldq,
+                    double* z, lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_chgeqz( char* job, char* compq, char* compz, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, lapack_complex_float* h,
+                    lapack_int* ldh, lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhgeqz( char* job, char* compq, char* compz, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, lapack_complex_double* h,
+                    lapack_int* ldh, lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_stgevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const float* s, lapack_int* lds,
+                    const float* p, lapack_int* ldp, float* vl,
+                    lapack_int* ldvl, float* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, float* work,
+                    lapack_int *info );
+void LAPACK_dtgevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const double* s, lapack_int* lds,
+                    const double* p, lapack_int* ldp, double* vl,
+                    lapack_int* ldvl, double* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, double* work,
+                    lapack_int *info );
+void LAPACK_ctgevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_float* s,
+                    lapack_int* lds, const lapack_complex_float* p,
+                    lapack_int* ldp, lapack_complex_float* vl, lapack_int* ldvl,
+                    lapack_complex_float* vr, lapack_int* ldvr, lapack_int* mm,
+                    lapack_int* m, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztgevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_double* s,
+                    lapack_int* lds, const lapack_complex_double* p,
+                    lapack_int* ldp, lapack_complex_double* vl,
+                    lapack_int* ldvl, lapack_complex_double* vr,
+                    lapack_int* ldvr, lapack_int* mm, lapack_int* m,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_stgexc( lapack_logical* wantq, lapack_logical* wantz, lapack_int* n,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    float* q, lapack_int* ldq, float* z, lapack_int* ldz,
+                    lapack_int* ifst, lapack_int* ilst, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dtgexc( lapack_logical* wantq, lapack_logical* wantz, lapack_int* n,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* q, lapack_int* ldq, double* z, lapack_int* ldz,
+                    lapack_int* ifst, lapack_int* ilst, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_ctgexc( lapack_logical* wantq, lapack_logical* wantz, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* z, lapack_int* ldz, lapack_int* ifst,
+                    lapack_int* ilst, lapack_int *info );
+void LAPACK_ztgexc( lapack_logical* wantq, lapack_logical* wantz, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* z, lapack_int* ldz, lapack_int* ifst,
+                    lapack_int* ilst, lapack_int *info );
+void LAPACK_stgsen( lapack_int* ijob, lapack_logical* wantq,
+                    lapack_logical* wantz, const lapack_logical* select,
+                    lapack_int* n, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, float* alphar, float* alphai, float* beta,
+                    float* q, lapack_int* ldq, float* z, lapack_int* ldz,
+                    lapack_int* m, float* pl, float* pr, float* dif,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dtgsen( lapack_int* ijob, lapack_logical* wantq,
+                    lapack_logical* wantz, const lapack_logical* select,
+                    lapack_int* n, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, double* alphar, double* alphai,
+                    double* beta, double* q, lapack_int* ldq, double* z,
+                    lapack_int* ldz, lapack_int* m, double* pl, double* pr,
+                    double* dif, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_ctgsen( lapack_int* ijob, lapack_logical* wantq,
+                    lapack_logical* wantz, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* z, lapack_int* ldz, lapack_int* m,
+                    float* pl, float* pr, float* dif,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_ztgsen( lapack_int* ijob, lapack_logical* wantq,
+                    lapack_logical* wantz, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* z, lapack_int* ldz, lapack_int* m,
+                    double* pl, double* pr, double* dif,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_stgsyl( char* trans, lapack_int* ijob, lapack_int* m, lapack_int* n,
+                    const float* a, lapack_int* lda, const float* b,
+                    lapack_int* ldb, float* c, lapack_int* ldc, const float* d,
+                    lapack_int* ldd, const float* e, lapack_int* lde, float* f,
+                    lapack_int* ldf, float* scale, float* dif, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_dtgsyl( char* trans, lapack_int* ijob, lapack_int* m, lapack_int* n,
+                    const double* a, lapack_int* lda, const double* b,
+                    lapack_int* ldb, double* c, lapack_int* ldc,
+                    const double* d, lapack_int* ldd, const double* e,
+                    lapack_int* lde, double* f, lapack_int* ldf, double* scale,
+                    double* dif, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ctgsyl( char* trans, lapack_int* ijob, lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    const lapack_complex_float* d, lapack_int* ldd,
+                    const lapack_complex_float* e, lapack_int* lde,
+                    lapack_complex_float* f, lapack_int* ldf, float* scale,
+                    float* dif, lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ztgsyl( char* trans, lapack_int* ijob, lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    const lapack_complex_double* d, lapack_int* ldd,
+                    const lapack_complex_double* e, lapack_int* lde,
+                    lapack_complex_double* f, lapack_int* ldf, double* scale,
+                    double* dif, lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_stgsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const float* a, lapack_int* lda,
+                    const float* b, lapack_int* ldb, const float* vl,
+                    lapack_int* ldvl, const float* vr, lapack_int* ldvr,
+                    float* s, float* dif, lapack_int* mm, lapack_int* m,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dtgsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const double* a, lapack_int* lda,
+                    const double* b, lapack_int* ldb, const double* vl,
+                    lapack_int* ldvl, const double* vr, lapack_int* ldvr,
+                    double* s, double* dif, lapack_int* mm, lapack_int* m,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_ctgsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* b,
+                    lapack_int* ldb, const lapack_complex_float* vl,
+                    lapack_int* ldvl, const lapack_complex_float* vr,
+                    lapack_int* ldvr, float* s, float* dif, lapack_int* mm,
+                    lapack_int* m, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_ztgsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* b,
+                    lapack_int* ldb, const lapack_complex_double* vl,
+                    lapack_int* ldvl, const lapack_complex_double* vr,
+                    lapack_int* ldvr, double* s, double* dif, lapack_int* mm,
+                    lapack_int* m, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_sggsvp( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, float* a, lapack_int* lda,
+                    float* b, lapack_int* ldb, float* tola, float* tolb,
+                    lapack_int* k, lapack_int* l, float* u, lapack_int* ldu,
+                    float* v, lapack_int* ldv, float* q, lapack_int* ldq,
+                    lapack_int* iwork, float* tau, float* work,
+                    lapack_int *info );
+void LAPACK_dggsvp( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, double* a, lapack_int* lda,
+                    double* b, lapack_int* ldb, double* tola, double* tolb,
+                    lapack_int* k, lapack_int* l, double* u, lapack_int* ldu,
+                    double* v, lapack_int* ldv, double* q, lapack_int* ldq,
+                    lapack_int* iwork, double* tau, double* work,
+                    lapack_int *info );
+void LAPACK_cggsvp( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* b, lapack_int* ldb,
+                    float* tola, float* tolb, lapack_int* k, lapack_int* l,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* v, lapack_int* ldv,
+                    lapack_complex_float* q, lapack_int* ldq, lapack_int* iwork,
+                    float* rwork, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zggsvp( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* b, lapack_int* ldb,
+                    double* tola, double* tolb, lapack_int* k, lapack_int* l,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* v, lapack_int* ldv,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_int* iwork, double* rwork,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_stgsja( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_int* k, lapack_int* l,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    float* tola, float* tolb, float* alpha, float* beta,
+                    float* u, lapack_int* ldu, float* v, lapack_int* ldv,
+                    float* q, lapack_int* ldq, float* work, lapack_int* ncycle,
+                    lapack_int *info );
+void LAPACK_dtgsja( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_int* k, lapack_int* l,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* tola, double* tolb, double* alpha, double* beta,
+                    double* u, lapack_int* ldu, double* v, lapack_int* ldv,
+                    double* q, lapack_int* ldq, double* work,
+                    lapack_int* ncycle, lapack_int *info );
+void LAPACK_ctgsja( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_int* k, lapack_int* l,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* tola,
+                    float* tolb, float* alpha, float* beta,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* v, lapack_int* ldv,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* work, lapack_int* ncycle,
+                    lapack_int *info );
+void LAPACK_ztgsja( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_int* k, lapack_int* l,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* tola,
+                    double* tolb, double* alpha, double* beta,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* v, lapack_int* ldv,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* work, lapack_int* ncycle,
+                    lapack_int *info );
+void LAPACK_sgels( char* trans, lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                   float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                   float* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_dgels( char* trans, lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                   double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                   double* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_cgels( char* trans, lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb,
+                   lapack_complex_float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_zgels( char* trans, lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_sgelsy( lapack_int* m, lapack_int* n, lapack_int* nrhs, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb,
+                    lapack_int* jpvt, float* rcond, lapack_int* rank,
+                    float* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_dgelsy( lapack_int* m, lapack_int* n, lapack_int* nrhs, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb,
+                    lapack_int* jpvt, double* rcond, lapack_int* rank,
+                    double* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_cgelsy( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, lapack_int* jpvt,
+                    float* rcond, lapack_int* rank, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int *info );
+void LAPACK_zgelsy( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, lapack_int* jpvt,
+                    double* rcond, lapack_int* rank,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgelss( lapack_int* m, lapack_int* n, lapack_int* nrhs, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb, float* s,
+                    float* rcond, lapack_int* rank, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dgelss( lapack_int* m, lapack_int* n, lapack_int* nrhs, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* s,
+                    double* rcond, lapack_int* rank, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cgelss( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* s,
+                    float* rcond, lapack_int* rank, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int *info );
+void LAPACK_zgelss( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* s,
+                    double* rcond, lapack_int* rank,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgelsd( lapack_int* m, lapack_int* n, lapack_int* nrhs, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb, float* s,
+                    float* rcond, lapack_int* rank, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_dgelsd( lapack_int* m, lapack_int* n, lapack_int* nrhs, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* s,
+                    double* rcond, lapack_int* rank, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_cgelsd( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* s,
+                    float* rcond, lapack_int* rank, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_zgelsd( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* s,
+                    double* rcond, lapack_int* rank,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgglse( lapack_int* m, lapack_int* n, lapack_int* p, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb, float* c,
+                    float* d, float* x, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgglse( lapack_int* m, lapack_int* n, lapack_int* p, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* c,
+                    double* d, double* x, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgglse( lapack_int* m, lapack_int* n, lapack_int* p,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* c, lapack_complex_float* d,
+                    lapack_complex_float* x, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zgglse( lapack_int* m, lapack_int* n, lapack_int* p,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* c, lapack_complex_double* d,
+                    lapack_complex_double* x, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sggglm( lapack_int* n, lapack_int* m, lapack_int* p, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb, float* d,
+                    float* x, float* y, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dggglm( lapack_int* n, lapack_int* m, lapack_int* p, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* d,
+                    double* x, double* y, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cggglm( lapack_int* n, lapack_int* m, lapack_int* p,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* d, lapack_complex_float* x,
+                    lapack_complex_float* y, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zggglm( lapack_int* n, lapack_int* m, lapack_int* p,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* d, lapack_complex_double* x,
+                    lapack_complex_double* y, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_ssyev( char* jobz, char* uplo, lapack_int* n, float* a,
+                   lapack_int* lda, float* w, float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_dsyev( char* jobz, char* uplo, lapack_int* n, double* a,
+                   lapack_int* lda, double* w, double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_cheev( char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda, float* w,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_int *info );
+void LAPACK_zheev( char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda, double* w,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_int *info );
+void LAPACK_ssyevd( char* jobz, char* uplo, lapack_int* n, float* a,
+                    lapack_int* lda, float* w, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dsyevd( char* jobz, char* uplo, lapack_int* n, double* a,
+                    lapack_int* lda, double* w, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_cheevd( char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* w,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zheevd( char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* w,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_ssyevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    float* a, lapack_int* lda, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, float* z, lapack_int* ldz,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dsyevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    double* a, lapack_int* lda, double* vl, double* vu,
+                    lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, double* z, lapack_int* ldz,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_cheevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_zheevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_ssyevr( char* jobz, char* range, char* uplo, lapack_int* n,
+                    float* a, lapack_int* lda, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, float* z, lapack_int* ldz,
+                    lapack_int* isuppz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dsyevr( char* jobz, char* range, char* uplo, lapack_int* n,
+                    double* a, lapack_int* lda, double* vl, double* vu,
+                    lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, double* z, lapack_int* ldz,
+                    lapack_int* isuppz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_cheevr( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_int* isuppz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zheevr( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_int* isuppz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sspev( char* jobz, char* uplo, lapack_int* n, float* ap, float* w,
+                   float* z, lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dspev( char* jobz, char* uplo, lapack_int* n, double* ap, double* w,
+                   double* z, lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_chpev( char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_float* ap, float* w, lapack_complex_float* z,
+                   lapack_int* ldz, lapack_complex_float* work, float* rwork,
+                   lapack_int *info );
+void LAPACK_zhpev( char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_double* ap, double* w,
+                   lapack_complex_double* z, lapack_int* ldz,
+                   lapack_complex_double* work, double* rwork,
+                   lapack_int *info );
+void LAPACK_sspevd( char* jobz, char* uplo, lapack_int* n, float* ap, float* w,
+                    float* z, lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dspevd( char* jobz, char* uplo, lapack_int* n, double* ap,
+                    double* w, double* z, lapack_int* ldz, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_chpevd( char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_float* ap, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* lrwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_zhpevd( char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_double* ap, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sspevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    float* ap, float* vl, float* vu, lapack_int* il,
+                    lapack_int* iu, float* abstol, lapack_int* m, float* w,
+                    float* z, lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dspevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    double* ap, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_chpevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_float* ap, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work, float* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_zhpevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_double* ap, double* vl, double* vu,
+                    lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_ssbev( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                   float* ab, lapack_int* ldab, float* w, float* z,
+                   lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dsbev( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                   double* ab, lapack_int* ldab, double* w, double* z,
+                   lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_chbev( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                   lapack_complex_float* ab, lapack_int* ldab, float* w,
+                   lapack_complex_float* z, lapack_int* ldz,
+                   lapack_complex_float* work, float* rwork, lapack_int *info );
+void LAPACK_zhbev( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                   lapack_complex_double* ab, lapack_int* ldab, double* w,
+                   lapack_complex_double* z, lapack_int* ldz,
+                   lapack_complex_double* work, double* rwork,
+                   lapack_int *info );
+void LAPACK_ssbevd( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                    float* ab, lapack_int* ldab, float* w, float* z,
+                    lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dsbevd( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                    double* ab, lapack_int* ldab, double* w, double* z,
+                    lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_chbevd( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_float* ab, lapack_int* ldab, float* w,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zhbevd( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_double* ab, lapack_int* ldab, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_ssbevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* kd, float* ab, lapack_int* ldab, float* q,
+                    lapack_int* ldq, float* vl, float* vu, lapack_int* il,
+                    lapack_int* iu, float* abstol, lapack_int* m, float* w,
+                    float* z, lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dsbevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* kd, double* ab, lapack_int* ldab, double* q,
+                    lapack_int* ldq, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_chbevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* kd, lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_complex_float* q, lapack_int* ldq, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work, float* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_zhbevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* kd, lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_complex_double* q, lapack_int* ldq, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_sstev( char* jobz, lapack_int* n, float* d, float* e, float* z,
+                   lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dstev( char* jobz, lapack_int* n, double* d, double* e, double* z,
+                   lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_sstevd( char* jobz, lapack_int* n, float* d, float* e, float* z,
+                    lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dstevd( char* jobz, lapack_int* n, double* d, double* e, double* z,
+                    lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_sstevx( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w, float* z,
+                    lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dstevx( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_sstevr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w, float* z,
+                    lapack_int* ldz, lapack_int* isuppz, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_dstevr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, lapack_int* isuppz,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sgees( char* jobvs, char* sort, LAPACK_S_SELECT2 select,
+                   lapack_int* n, float* a, lapack_int* lda, lapack_int* sdim,
+                   float* wr, float* wi, float* vs, lapack_int* ldvs,
+                   float* work, lapack_int* lwork, lapack_logical* bwork,
+                   lapack_int *info );
+void LAPACK_dgees( char* jobvs, char* sort, LAPACK_D_SELECT2 select,
+                   lapack_int* n, double* a, lapack_int* lda, lapack_int* sdim,
+                   double* wr, double* wi, double* vs, lapack_int* ldvs,
+                   double* work, lapack_int* lwork, lapack_logical* bwork,
+                   lapack_int *info );
+void LAPACK_cgees( char* jobvs, char* sort, LAPACK_C_SELECT1 select,
+                   lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                   lapack_int* sdim, lapack_complex_float* w,
+                   lapack_complex_float* vs, lapack_int* ldvs,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_logical* bwork, lapack_int *info );
+void LAPACK_zgees( char* jobvs, char* sort, LAPACK_Z_SELECT1 select,
+                   lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                   lapack_int* sdim, lapack_complex_double* w,
+                   lapack_complex_double* vs, lapack_int* ldvs,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_logical* bwork, lapack_int *info );
+void LAPACK_sgeesx( char* jobvs, char* sort, LAPACK_S_SELECT2 select,
+                    char* sense, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* sdim, float* wr, float* wi, float* vs,
+                    lapack_int* ldvs, float* rconde, float* rcondv, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_logical* bwork, lapack_int *info );
+void LAPACK_dgeesx( char* jobvs, char* sort, LAPACK_D_SELECT2 select,
+                    char* sense, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* sdim, double* wr, double* wi, double* vs,
+                    lapack_int* ldvs, double* rconde, double* rcondv,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_cgeesx( char* jobvs, char* sort, LAPACK_C_SELECT1 select,
+                    char* sense, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* sdim, lapack_complex_float* w,
+                    lapack_complex_float* vs, lapack_int* ldvs, float* rconde,
+                    float* rcondv, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_zgeesx( char* jobvs, char* sort, LAPACK_Z_SELECT1 select,
+                    char* sense, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* sdim, lapack_complex_double* w,
+                    lapack_complex_double* vs, lapack_int* ldvs, double* rconde,
+                    double* rcondv, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_sgeev( char* jobvl, char* jobvr, lapack_int* n, float* a,
+                   lapack_int* lda, float* wr, float* wi, float* vl,
+                   lapack_int* ldvl, float* vr, lapack_int* ldvr, float* work,
+                   lapack_int* lwork, lapack_int *info );
+void LAPACK_dgeev( char* jobvl, char* jobvr, lapack_int* n, double* a,
+                   lapack_int* lda, double* wr, double* wi, double* vl,
+                   lapack_int* ldvl, double* vr, lapack_int* ldvr, double* work,
+                   lapack_int* lwork, lapack_int *info );
+void LAPACK_cgeev( char* jobvl, char* jobvr, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* w, lapack_complex_float* vl,
+                   lapack_int* ldvl, lapack_complex_float* vr, lapack_int* ldvr,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_int *info );
+void LAPACK_zgeev( char* jobvl, char* jobvr, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* w, lapack_complex_double* vl,
+                   lapack_int* ldvl, lapack_complex_double* vr,
+                   lapack_int* ldvr, lapack_complex_double* work,
+                   lapack_int* lwork, double* rwork, lapack_int *info );
+void LAPACK_sgeevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, float* a, lapack_int* lda, float* wr,
+                    float* wi, float* vl, lapack_int* ldvl, float* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    float* scale, float* abnrm, float* rconde, float* rcondv,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgeevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, double* a, lapack_int* lda, double* wr,
+                    double* wi, double* vl, lapack_int* ldvl, double* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    double* scale, double* abnrm, double* rconde,
+                    double* rcondv, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgeevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* w, lapack_complex_float* vl,
+                    lapack_int* ldvl, lapack_complex_float* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    float* scale, float* abnrm, float* rconde, float* rcondv,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgeevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* w, lapack_complex_double* vl,
+                    lapack_int* ldvl, lapack_complex_double* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    double* scale, double* abnrm, double* rconde,
+                    double* rcondv, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int *info );
+void LAPACK_sgesvd( char* jobu, char* jobvt, lapack_int* m, lapack_int* n,
+                    float* a, lapack_int* lda, float* s, float* u,
+                    lapack_int* ldu, float* vt, lapack_int* ldvt, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dgesvd( char* jobu, char* jobvt, lapack_int* m, lapack_int* n,
+                    double* a, lapack_int* lda, double* s, double* u,
+                    lapack_int* ldu, double* vt, lapack_int* ldvt, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cgesvd( char* jobu, char* jobvt, lapack_int* m, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* s,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* vt, lapack_int* ldvt,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgesvd( char* jobu, char* jobvt, lapack_int* m, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* s,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* vt, lapack_int* ldvt,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgesdd( char* jobz, lapack_int* m, lapack_int* n, float* a,
+                    lapack_int* lda, float* s, float* u, lapack_int* ldu,
+                    float* vt, lapack_int* ldvt, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dgesdd( char* jobz, lapack_int* m, lapack_int* n, double* a,
+                    lapack_int* lda, double* s, double* u, lapack_int* ldu,
+                    double* vt, lapack_int* ldvt, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_cgesdd( char* jobz, lapack_int* m, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* s,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* vt, lapack_int* ldvt,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_zgesdd( char* jobz, lapack_int* m, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* s,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* vt, lapack_int* ldvt,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_dgejsv( char* joba, char* jobu, char* jobv, char* jobr, char* jobt,
+                    char* jobp, lapack_int* m, lapack_int* n, double* a,
+                    lapack_int* lda, double* sva, double* u, lapack_int* ldu,
+                    double* v, lapack_int* ldv, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_sgejsv( char* joba, char* jobu, char* jobv, char* jobr, char* jobt,
+                    char* jobp, lapack_int* m, lapack_int* n, float* a,
+                    lapack_int* lda, float* sva, float* u, lapack_int* ldu,
+                    float* v, lapack_int* ldv, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dgesvj( char* joba, char* jobu, char* jobv, lapack_int* m,
+                    lapack_int* n, double* a, lapack_int* lda, double* sva,
+                    lapack_int* mv, double* v, lapack_int* ldv, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sgesvj( char* joba, char* jobu, char* jobv, lapack_int* m,
+                    lapack_int* n, float* a, lapack_int* lda, float* sva,
+                    lapack_int* mv, float* v, lapack_int* ldv, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sggsvd( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* n, lapack_int* p, lapack_int* k, lapack_int* l,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    float* alpha, float* beta, float* u, lapack_int* ldu,
+                    float* v, lapack_int* ldv, float* q, lapack_int* ldq,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dggsvd( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* n, lapack_int* p, lapack_int* k, lapack_int* l,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* alpha, double* beta, double* u, lapack_int* ldu,
+                    double* v, lapack_int* ldv, double* q, lapack_int* ldq,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cggsvd( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* n, lapack_int* p, lapack_int* k, lapack_int* l,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* alpha,
+                    float* beta, lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* v, lapack_int* ldv,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* work, float* rwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_zggsvd( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* n, lapack_int* p, lapack_int* k, lapack_int* l,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* alpha,
+                    double* beta, lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* v, lapack_int* ldv,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ssygv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                   float* w, float* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_dsygv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                   double* w, double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_chegv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb, float* w,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_int *info );
+void LAPACK_zhegv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb, double* w,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_int *info );
+void LAPACK_ssygvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    float* w, float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dsygvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* w, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_chegvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* w,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zhegvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* w,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_ssygvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, float* vl, float* vu, lapack_int* il,
+                    lapack_int* iu, float* abstol, lapack_int* m, float* w,
+                    float* z, lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_dsygvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_chegvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_zhegvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_sspgv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   float* ap, float* bp, float* w, float* z, lapack_int* ldz,
+                   float* work, lapack_int *info );
+void LAPACK_dspgv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   double* ap, double* bp, double* w, double* z,
+                   lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_chpgv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_float* ap, lapack_complex_float* bp, float* w,
+                   lapack_complex_float* z, lapack_int* ldz,
+                   lapack_complex_float* work, float* rwork, lapack_int *info );
+void LAPACK_zhpgv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_double* ap, lapack_complex_double* bp,
+                   double* w, lapack_complex_double* z, lapack_int* ldz,
+                   lapack_complex_double* work, double* rwork,
+                   lapack_int *info );
+void LAPACK_sspgvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    float* ap, float* bp, float* w, float* z, lapack_int* ldz,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dspgvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    double* ap, double* bp, double* w, double* z,
+                    lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_chpgvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_float* ap, lapack_complex_float* bp,
+                    float* w, lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zhpgvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_double* ap, lapack_complex_double* bp,
+                    double* w, lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sspgvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, float* ap, float* bp, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, float* z, lapack_int* ldz,
+                    float* work, lapack_int* iwork, lapack_int* ifail,
+                    lapack_int *info );
+void LAPACK_dspgvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, double* ap, double* bp, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, double* z, lapack_int* ldz,
+                    double* work, lapack_int* iwork, lapack_int* ifail,
+                    lapack_int *info );
+void LAPACK_chpgvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, lapack_complex_float* ap,
+                    lapack_complex_float* bp, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work, float* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_zhpgvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, lapack_complex_double* ap,
+                    lapack_complex_double* bp, double* vl, double* vu,
+                    lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_ssbgv( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                   lapack_int* kb, float* ab, lapack_int* ldab, float* bb,
+                   lapack_int* ldbb, float* w, float* z, lapack_int* ldz,
+                   float* work, lapack_int *info );
+void LAPACK_dsbgv( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                   lapack_int* kb, double* ab, lapack_int* ldab, double* bb,
+                   lapack_int* ldbb, double* w, double* z, lapack_int* ldz,
+                   double* work, lapack_int *info );
+void LAPACK_chbgv( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                   lapack_int* kb, lapack_complex_float* ab, lapack_int* ldab,
+                   lapack_complex_float* bb, lapack_int* ldbb, float* w,
+                   lapack_complex_float* z, lapack_int* ldz,
+                   lapack_complex_float* work, float* rwork, lapack_int *info );
+void LAPACK_zhbgv( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                   lapack_int* kb, lapack_complex_double* ab, lapack_int* ldab,
+                   lapack_complex_double* bb, lapack_int* ldbb, double* w,
+                   lapack_complex_double* z, lapack_int* ldz,
+                   lapack_complex_double* work, double* rwork,
+                   lapack_int *info );
+void LAPACK_ssbgvd( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, float* ab, lapack_int* ldab, float* bb,
+                    lapack_int* ldbb, float* w, float* z, lapack_int* ldz,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dsbgvd( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, double* ab, lapack_int* ldab, double* bb,
+                    lapack_int* ldbb, double* w, double* z, lapack_int* ldz,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_chbgvd( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_complex_float* bb, lapack_int* ldbb, float* w,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zhbgvd( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_complex_double* bb, lapack_int* ldbb, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_ssbgvx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* ka, lapack_int* kb, float* ab, lapack_int* ldab,
+                    float* bb, lapack_int* ldbb, float* q, lapack_int* ldq,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w, float* z,
+                    lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dsbgvx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* ka, lapack_int* kb, double* ab,
+                    lapack_int* ldab, double* bb, lapack_int* ldbb, double* q,
+                    lapack_int* ldq, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_chbgvx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* ka, lapack_int* kb, lapack_complex_float* ab,
+                    lapack_int* ldab, lapack_complex_float* bb,
+                    lapack_int* ldbb, lapack_complex_float* q, lapack_int* ldq,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, float* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_zhbgvx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* ka, lapack_int* kb, lapack_complex_double* ab,
+                    lapack_int* ldab, lapack_complex_double* bb,
+                    lapack_int* ldbb, lapack_complex_double* q, lapack_int* ldq,
+                    double* vl, double* vu, lapack_int* il, lapack_int* iu,
+                    double* abstol, lapack_int* m, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_sgges( char* jobvsl, char* jobvsr, char* sort,
+                   LAPACK_S_SELECT3 selctg, lapack_int* n, float* a,
+                   lapack_int* lda, float* b, lapack_int* ldb, lapack_int* sdim,
+                   float* alphar, float* alphai, float* beta, float* vsl,
+                   lapack_int* ldvsl, float* vsr, lapack_int* ldvsr,
+                   float* work, lapack_int* lwork, lapack_logical* bwork,
+                   lapack_int *info );
+void LAPACK_dgges( char* jobvsl, char* jobvsr, char* sort,
+                   LAPACK_D_SELECT3 selctg, lapack_int* n, double* a,
+                   lapack_int* lda, double* b, lapack_int* ldb,
+                   lapack_int* sdim, double* alphar, double* alphai,
+                   double* beta, double* vsl, lapack_int* ldvsl, double* vsr,
+                   lapack_int* ldvsr, double* work, lapack_int* lwork,
+                   lapack_logical* bwork, lapack_int *info );
+void LAPACK_cgges( char* jobvsl, char* jobvsr, char* sort,
+                   LAPACK_C_SELECT2 selctg, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int* sdim,
+                   lapack_complex_float* alpha, lapack_complex_float* beta,
+                   lapack_complex_float* vsl, lapack_int* ldvsl,
+                   lapack_complex_float* vsr, lapack_int* ldvsr,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_logical* bwork, lapack_int *info );
+void LAPACK_zgges( char* jobvsl, char* jobvsr, char* sort,
+                   LAPACK_Z_SELECT2 selctg, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb, lapack_int* sdim,
+                   lapack_complex_double* alpha, lapack_complex_double* beta,
+                   lapack_complex_double* vsl, lapack_int* ldvsl,
+                   lapack_complex_double* vsr, lapack_int* ldvsr,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_logical* bwork, lapack_int *info );
+void LAPACK_sggesx( char* jobvsl, char* jobvsr, char* sort,
+                    LAPACK_S_SELECT3 selctg, char* sense, lapack_int* n,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    lapack_int* sdim, float* alphar, float* alphai, float* beta,
+                    float* vsl, lapack_int* ldvsl, float* vsr,
+                    lapack_int* ldvsr, float* rconde, float* rcondv,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_dggesx( char* jobvsl, char* jobvsr, char* sort,
+                    LAPACK_D_SELECT3 selctg, char* sense, lapack_int* n,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    lapack_int* sdim, double* alphar, double* alphai,
+                    double* beta, double* vsl, lapack_int* ldvsl, double* vsr,
+                    lapack_int* ldvsr, double* rconde, double* rcondv,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_cggesx( char* jobvsl, char* jobvsr, char* sort,
+                    LAPACK_C_SELECT2 selctg, char* sense, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, lapack_int* sdim,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* vsl, lapack_int* ldvsl,
+                    lapack_complex_float* vsr, lapack_int* ldvsr, float* rconde,
+                    float* rcondv, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_zggesx( char* jobvsl, char* jobvsr, char* sort,
+                    LAPACK_Z_SELECT2 selctg, char* sense, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, lapack_int* sdim,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* vsl, lapack_int* ldvsl,
+                    lapack_complex_double* vsr, lapack_int* ldvsr,
+                    double* rconde, double* rcondv, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_sggev( char* jobvl, char* jobvr, lapack_int* n, float* a,
+                   lapack_int* lda, float* b, lapack_int* ldb, float* alphar,
+                   float* alphai, float* beta, float* vl, lapack_int* ldvl,
+                   float* vr, lapack_int* ldvr, float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_dggev( char* jobvl, char* jobvr, lapack_int* n, double* a,
+                   lapack_int* lda, double* b, lapack_int* ldb, double* alphar,
+                   double* alphai, double* beta, double* vl, lapack_int* ldvl,
+                   double* vr, lapack_int* ldvr, double* work,
+                   lapack_int* lwork, lapack_int *info );
+void LAPACK_cggev( char* jobvl, char* jobvr, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb,
+                   lapack_complex_float* alpha, lapack_complex_float* beta,
+                   lapack_complex_float* vl, lapack_int* ldvl,
+                   lapack_complex_float* vr, lapack_int* ldvr,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_int *info );
+void LAPACK_zggev( char* jobvl, char* jobvr, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_complex_double* alpha, lapack_complex_double* beta,
+                   lapack_complex_double* vl, lapack_int* ldvl,
+                   lapack_complex_double* vr, lapack_int* ldvr,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_int *info );
+void LAPACK_sggevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, float* alphar, float* alphai, float* beta,
+                    float* vl, lapack_int* ldvl, float* vr, lapack_int* ldvr,
+                    lapack_int* ilo, lapack_int* ihi, float* lscale,
+                    float* rscale, float* abnrm, float* bbnrm, float* rconde,
+                    float* rcondv, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_dggevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, double* alphar, double* alphai,
+                    double* beta, double* vl, lapack_int* ldvl, double* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    double* lscale, double* rscale, double* abnrm,
+                    double* bbnrm, double* rconde, double* rcondv, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_cggevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* vl, lapack_int* ldvl,
+                    lapack_complex_float* vr, lapack_int* ldvr, lapack_int* ilo,
+                    lapack_int* ihi, float* lscale, float* rscale, float* abnrm,
+                    float* bbnrm, float* rconde, float* rcondv,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* iwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_zggevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* vl, lapack_int* ldvl,
+                    lapack_complex_double* vr, lapack_int* ldvr,
+                    lapack_int* ilo, lapack_int* ihi, double* lscale,
+                    double* rscale, double* abnrm, double* bbnrm,
+                    double* rconde, double* rcondv, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int* iwork,
+                    lapack_logical* bwork, lapack_int *info );
+void LAPACK_dsfrk( char* transr, char* uplo, char* trans, lapack_int* n,
+                   lapack_int* k, double* alpha, const double* a,
+                   lapack_int* lda, double* beta, double* c );
+void LAPACK_ssfrk( char* transr, char* uplo, char* trans, lapack_int* n,
+                   lapack_int* k, float* alpha, const float* a, lapack_int* lda,
+                   float* beta, float* c );
+void LAPACK_zhfrk( char* transr, char* uplo, char* trans, lapack_int* n,
+                   lapack_int* k, double* alpha, const lapack_complex_double* a,
+                   lapack_int* lda, double* beta, lapack_complex_double* c );
+void LAPACK_chfrk( char* transr, char* uplo, char* trans, lapack_int* n,
+                   lapack_int* k, float* alpha, const lapack_complex_float* a,
+                   lapack_int* lda, float* beta, lapack_complex_float* c );
+void LAPACK_dtfsm( char* transr, char* side, char* uplo, char* trans,
+                   char* diag, lapack_int* m, lapack_int* n, double* alpha,
+                   const double* a, double* b, lapack_int* ldb );
+void LAPACK_stfsm( char* transr, char* side, char* uplo, char* trans,
+                   char* diag, lapack_int* m, lapack_int* n, float* alpha,
+                   const float* a, float* b, lapack_int* ldb );
+void LAPACK_ztfsm( char* transr, char* side, char* uplo, char* trans,
+                   char* diag, lapack_int* m, lapack_int* n,
+                   lapack_complex_double* alpha, const lapack_complex_double* a,
+                   lapack_complex_double* b, lapack_int* ldb );
+void LAPACK_ctfsm( char* transr, char* side, char* uplo, char* trans,
+                   char* diag, lapack_int* m, lapack_int* n,
+                   lapack_complex_float* alpha, const lapack_complex_float* a,
+                   lapack_complex_float* b, lapack_int* ldb );
+void LAPACK_dtfttp( char* transr, char* uplo, lapack_int* n, const double* arf,
+                    double* ap, lapack_int *info );
+void LAPACK_stfttp( char* transr, char* uplo, lapack_int* n, const float* arf,
+                    float* ap, lapack_int *info );
+void LAPACK_ztfttp( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_double* arf, lapack_complex_double* ap,
+                    lapack_int *info );
+void LAPACK_ctfttp( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_float* arf, lapack_complex_float* ap,
+                    lapack_int *info );
+void LAPACK_dtfttr( char* transr, char* uplo, lapack_int* n, const double* arf,
+                    double* a, lapack_int* lda, lapack_int *info );
+void LAPACK_stfttr( char* transr, char* uplo, lapack_int* n, const float* arf,
+                    float* a, lapack_int* lda, lapack_int *info );
+void LAPACK_ztfttr( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_double* arf, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_ctfttr( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_float* arf, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_dtpttf( char* transr, char* uplo, lapack_int* n, const double* ap,
+                    double* arf, lapack_int *info );
+void LAPACK_stpttf( char* transr, char* uplo, lapack_int* n, const float* ap,
+                    float* arf, lapack_int *info );
+void LAPACK_ztpttf( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_double* ap, lapack_complex_double* arf,
+                    lapack_int *info );
+void LAPACK_ctpttf( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_float* ap, lapack_complex_float* arf,
+                    lapack_int *info );
+void LAPACK_dtpttr( char* uplo, lapack_int* n, const double* ap, double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_stpttr( char* uplo, lapack_int* n, const float* ap, float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_ztpttr( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_ctpttr( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dtrttf( char* transr, char* uplo, lapack_int* n, const double* a,
+                    lapack_int* lda, double* arf, lapack_int *info );
+void LAPACK_strttf( char* transr, char* uplo, lapack_int* n, const float* a,
+                    lapack_int* lda, float* arf, lapack_int *info );
+void LAPACK_ztrttf( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* arf, lapack_int *info );
+void LAPACK_ctrttf( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* arf, lapack_int *info );
+void LAPACK_dtrttp( char* uplo, lapack_int* n, const double* a, lapack_int* lda,
+                    double* ap, lapack_int *info );
+void LAPACK_strttp( char* uplo, lapack_int* n, const float* a, lapack_int* lda,
+                    float* ap, lapack_int *info );
+void LAPACK_ztrttp( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* ap,
+                    lapack_int *info );
+void LAPACK_ctrttp( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* ap,
+                    lapack_int *info );
+void LAPACK_sgeqrfp( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                     float* tau, float* work, lapack_int* lwork,
+                     lapack_int *info );
+void LAPACK_dgeqrfp( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                     double* tau, double* work, lapack_int* lwork,
+                     lapack_int *info );
+void LAPACK_cgeqrfp( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, lapack_complex_float* tau,
+                     lapack_complex_float* work, lapack_int* lwork,
+                     lapack_int *info );
+void LAPACK_zgeqrfp( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, lapack_complex_double* tau,
+                     lapack_complex_double* work, lapack_int* lwork,
+                     lapack_int *info );
+void LAPACK_clacgv( lapack_int* n, lapack_complex_float* x, lapack_int* incx );
+void LAPACK_zlacgv( lapack_int* n, lapack_complex_double* x, lapack_int* incx );
+void LAPACK_slarnv( lapack_int* idist, lapack_int* iseed, lapack_int* n,
+                    float* x );
+void LAPACK_dlarnv( lapack_int* idist, lapack_int* iseed, lapack_int* n,
+                    double* x );
+void LAPACK_clarnv( lapack_int* idist, lapack_int* iseed, lapack_int* n,
+                    lapack_complex_float* x );
+void LAPACK_zlarnv( lapack_int* idist, lapack_int* iseed, lapack_int* n,
+                    lapack_complex_double* x );
+void LAPACK_sgeqr2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int *info );
+void LAPACK_dgeqr2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int *info );
+void LAPACK_cgeqr2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zgeqr2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_slacpy( char* uplo, lapack_int* m, lapack_int* n, const float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb );
+void LAPACK_dlacpy( char* uplo, lapack_int* m, lapack_int* n, const double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb );
+void LAPACK_clacpy( char* uplo, lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb );
+void LAPACK_zlacpy( char* uplo, lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb );
+void LAPACK_sgetf2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_dgetf2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_cgetf2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_int *info );
+void LAPACK_zgetf2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_int *info );
+void LAPACK_slaswp( lapack_int* n, float* a, lapack_int* lda, lapack_int* k1,
+                    lapack_int* k2, const lapack_int* ipiv, lapack_int* incx );
+void LAPACK_dlaswp( lapack_int* n, double* a, lapack_int* lda, lapack_int* k1,
+                    lapack_int* k2, const lapack_int* ipiv, lapack_int* incx );
+void LAPACK_claswp( lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_int* k1, lapack_int* k2, const lapack_int* ipiv,
+                    lapack_int* incx );
+void LAPACK_zlaswp( lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_int* k1, lapack_int* k2, const lapack_int* ipiv,
+                    lapack_int* incx );
+float LAPACK_slange( char* norm, lapack_int* m, lapack_int* n, const float* a,
+                    lapack_int* lda, float* work );
+double LAPACK_dlange( char* norm, lapack_int* m, lapack_int* n, const double* a,
+                    lapack_int* lda, double* work );
+float LAPACK_clange( char* norm, lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda, float* work );
+double LAPACK_zlange( char* norm, lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda, double* work );
+float LAPACK_clanhe( char* norm, char* uplo, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda, float* work );
+double LAPACK_zlanhe( char* norm, char* uplo, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda, double* work );
+float LAPACK_slansy( char* norm, char* uplo, lapack_int* n, const float* a,
+                    lapack_int* lda, float* work );
+double LAPACK_dlansy( char* norm, char* uplo, lapack_int* n, const double* a,
+                    lapack_int* lda, double* work );
+float LAPACK_clansy( char* norm, char* uplo, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda, float* work );
+double LAPACK_zlansy( char* norm, char* uplo, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda, double* work );
+float LAPACK_slantr( char* norm, char* uplo, char* diag, lapack_int* m,
+                    lapack_int* n, const float* a, lapack_int* lda, float* work );
+double LAPACK_dlantr( char* norm, char* uplo, char* diag, lapack_int* m,
+                    lapack_int* n, const double* a, lapack_int* lda, double* work );
+float LAPACK_clantr( char* norm, char* uplo, char* diag, lapack_int* m,
+                    lapack_int* n, const lapack_complex_float* a, lapack_int* lda,
+                    float* work );
+double LAPACK_zlantr( char* norm, char* uplo, char* diag, lapack_int* m,
+                    lapack_int* n, const lapack_complex_double* a, lapack_int* lda,
+                    double* work );
+float LAPACK_slamch( char* cmach );
+double LAPACK_dlamch( char* cmach );
+void LAPACK_sgelq2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int *info );
+void LAPACK_dgelq2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int *info );
+void LAPACK_cgelq2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zgelq2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_slarfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, const float* v,
+                    lapack_int* ldv, const float* t, lapack_int* ldt, float* c,
+                    lapack_int* ldc, float* work, lapack_int* ldwork );
+void LAPACK_dlarfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k,
+                    const double* v, lapack_int* ldv, const double* t,
+                    lapack_int* ldt, double* c, lapack_int* ldc, double* work,
+                    lapack_int* ldwork );
+void LAPACK_clarfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k,
+                    const lapack_complex_float* v, lapack_int* ldv,
+                    const lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* ldwork );
+void LAPACK_zlarfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k,
+                    const lapack_complex_double* v, lapack_int* ldv,
+                    const lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* ldwork );
+void LAPACK_slarfg( lapack_int* n, float* alpha, float* x, lapack_int* incx,
+                    float* tau );
+void LAPACK_dlarfg( lapack_int* n, double* alpha, double* x, lapack_int* incx,
+                    double* tau );
+void LAPACK_clarfg( lapack_int* n, lapack_complex_float* alpha,
+                    lapack_complex_float* x, lapack_int* incx,
+                    lapack_complex_float* tau );
+void LAPACK_zlarfg( lapack_int* n, lapack_complex_double* alpha,
+                    lapack_complex_double* x, lapack_int* incx,
+                    lapack_complex_double* tau );
+void LAPACK_slarft( char* direct, char* storev, lapack_int* n, lapack_int* k,
+                    const float* v, lapack_int* ldv, const float* tau, float* t,
+                    lapack_int* ldt );
+void LAPACK_dlarft( char* direct, char* storev, lapack_int* n, lapack_int* k,
+                    const double* v, lapack_int* ldv, const double* tau,
+                    double* t, lapack_int* ldt );
+void LAPACK_clarft( char* direct, char* storev, lapack_int* n, lapack_int* k,
+                    const lapack_complex_float* v, lapack_int* ldv,
+                    const lapack_complex_float* tau, lapack_complex_float* t,
+                    lapack_int* ldt );
+void LAPACK_zlarft( char* direct, char* storev, lapack_int* n, lapack_int* k,
+                    const lapack_complex_double* v, lapack_int* ldv,
+                    const lapack_complex_double* tau, lapack_complex_double* t,
+                    lapack_int* ldt );
+void LAPACK_slarfx( char* side, lapack_int* m, lapack_int* n, const float* v,
+                    float* tau, float* c, lapack_int* ldc, float* work );
+void LAPACK_dlarfx( char* side, lapack_int* m, lapack_int* n, const double* v,
+                    double* tau, double* c, lapack_int* ldc, double* work );
+void LAPACK_clarfx( char* side, lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* v, lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work );
+void LAPACK_zlarfx( char* side, lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* v, lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work );
+void LAPACK_slatms( lapack_int* m, lapack_int* n, char* dist, lapack_int* iseed,
+                    char* sym, float* d, lapack_int* mode, float* cond,
+                    float* dmax, lapack_int* kl, lapack_int* ku, char* pack,
+                    float* a, lapack_int* lda, float* work, lapack_int *info );
+void LAPACK_dlatms( lapack_int* m, lapack_int* n, char* dist, lapack_int* iseed,
+                    char* sym, double* d, lapack_int* mode, double* cond,
+                    double* dmax, lapack_int* kl, lapack_int* ku, char* pack,
+                    double* a, lapack_int* lda, double* work,
+                    lapack_int *info );
+void LAPACK_clatms( lapack_int* m, lapack_int* n, char* dist, lapack_int* iseed,
+                    char* sym, float* d, lapack_int* mode, float* cond,
+                    float* dmax, lapack_int* kl, lapack_int* ku, char* pack,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zlatms( lapack_int* m, lapack_int* n, char* dist, lapack_int* iseed,
+                    char* sym, double* d, lapack_int* mode, double* cond,
+                    double* dmax, lapack_int* kl, lapack_int* ku, char* pack,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_slag2d( lapack_int* m, lapack_int* n, const float* sa,
+                    lapack_int* ldsa, double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dlag2s( lapack_int* m, lapack_int* n, const double* a,
+                    lapack_int* lda, float* sa, lapack_int* ldsa,
+                    lapack_int *info );
+void LAPACK_clag2z( lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* sa, lapack_int* ldsa,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_zlag2c( lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_float* sa, lapack_int* ldsa,
+                    lapack_int *info );
+void LAPACK_slauum( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dlauum( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_clauum( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_zlauum( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_slagge( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const float* d, float* a, lapack_int* lda,
+                    lapack_int* iseed, float* work, lapack_int *info );
+void LAPACK_dlagge( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const double* d, double* a, lapack_int* lda,
+                    lapack_int* iseed, double* work, lapack_int *info );
+void LAPACK_clagge( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const float* d, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* iseed,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zlagge( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const double* d, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* iseed,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_slaset( char* uplo, lapack_int* m, lapack_int* n, float* alpha,
+                    float* beta, float* a, lapack_int* lda );
+void LAPACK_dlaset( char* uplo, lapack_int* m, lapack_int* n, double* alpha,
+                    double* beta, double* a, lapack_int* lda );
+void LAPACK_claset( char* uplo, lapack_int* m, lapack_int* n,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* a, lapack_int* lda );
+void LAPACK_zlaset( char* uplo, lapack_int* m, lapack_int* n,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* a, lapack_int* lda );
+void LAPACK_slasrt( char* id, lapack_int* n, float* d, lapack_int *info );
+void LAPACK_dlasrt( char* id, lapack_int* n, double* d, lapack_int *info );
+void LAPACK_claghe( lapack_int* n, lapack_int* k, const float* d,
+                    lapack_complex_float* a, lapack_int* lda, lapack_int* iseed,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zlaghe( lapack_int* n, lapack_int* k, const double* d,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int* iseed, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_slagsy( lapack_int* n, lapack_int* k, const float* d, float* a,
+                    lapack_int* lda, lapack_int* iseed, float* work,
+                    lapack_int *info );
+void LAPACK_dlagsy( lapack_int* n, lapack_int* k, const double* d, double* a,
+                    lapack_int* lda, lapack_int* iseed, double* work,
+                    lapack_int *info );
+void LAPACK_clagsy( lapack_int* n, lapack_int* k, const float* d,
+                    lapack_complex_float* a, lapack_int* lda, lapack_int* iseed,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zlagsy( lapack_int* n, lapack_int* k, const double* d,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int* iseed, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_slapmr( lapack_logical* forwrd, lapack_int* m, lapack_int* n,
+                    float* x, lapack_int* ldx, lapack_int* k );
+void LAPACK_dlapmr( lapack_logical* forwrd, lapack_int* m, lapack_int* n,
+                    double* x, lapack_int* ldx, lapack_int* k );
+void LAPACK_clapmr( lapack_logical* forwrd, lapack_int* m, lapack_int* n,
+                    lapack_complex_float* x, lapack_int* ldx, lapack_int* k );
+void LAPACK_zlapmr( lapack_logical* forwrd, lapack_int* m, lapack_int* n,
+                    lapack_complex_double* x, lapack_int* ldx, lapack_int* k );
+float LAPACK_slapy2( float* x, float* y );
+double LAPACK_dlapy2( double* x, double* y );
+float LAPACK_slapy3( float* x, float* y, float* z );
+double LAPACK_dlapy3( double* x, double* y, double* z );
+void LAPACK_slartgp( float* f, float* g, float* cs, float* sn, float* r );
+void LAPACK_dlartgp( double* f, double* g, double* cs, double* sn, double* r );
+void LAPACK_slartgs( float* x, float* y, float* sigma, float* cs, float* sn );
+void LAPACK_dlartgs( double* x, double* y, double* sigma, double* cs,
+                     double* sn );
+// LAPACK 3.3.0
+void LAPACK_cbbcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    float* theta, float* phi,
+                    lapack_complex_float* u1, lapack_int* ldu1,
+                    lapack_complex_float* u2, lapack_int* ldu2,
+                    lapack_complex_float* v1t, lapack_int* ldv1t,
+                    lapack_complex_float* v2t, lapack_int* ldv2t,
+                    float* b11d, float* b11e, float* b12d,
+                    float* b12e, float* b21d, float* b21e,
+                    float* b22d, float* b22e, float* rwork,
+                    lapack_int* lrwork , lapack_int *info );
+void LAPACK_cheswapr( char* uplo, lapack_int* n,
+                      lapack_complex_float* a, lapack_int* i1,
+                      lapack_int* i2 );
+void LAPACK_chetri2( char* uplo, lapack_int* n,
+                     lapack_complex_float* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_chetri2x( char* uplo, lapack_int* n,
+                      lapack_complex_float* a, lapack_int* lda,
+                      const lapack_int* ipiv,
+                      lapack_complex_float* work, lapack_int* nb , lapack_int *info );
+void LAPACK_chetrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs, const lapack_complex_float* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* work , lapack_int *info );
+void LAPACK_csyconv( char* uplo, char* way,
+                     lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     lapack_complex_float* work , lapack_int *info );
+void LAPACK_csyswapr( char* uplo, lapack_int* n,
+                      lapack_complex_float* a, lapack_int* i1,
+                      lapack_int* i2 );
+void LAPACK_csytri2( char* uplo, lapack_int* n,
+                     lapack_complex_float* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_csytri2x( char* uplo, lapack_int* n,
+                      lapack_complex_float* a, lapack_int* lda,
+                      const lapack_int* ipiv,
+                      lapack_complex_float* work, lapack_int* nb , lapack_int *info );
+void LAPACK_csytrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs, const lapack_complex_float* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* work , lapack_int *info );
+void LAPACK_cunbdb( char* trans, char* signs,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    lapack_complex_float* x11, lapack_int* ldx11,
+                    lapack_complex_float* x12, lapack_int* ldx12,
+                    lapack_complex_float* x21, lapack_int* ldx21,
+                    lapack_complex_float* x22, lapack_int* ldx22,
+                    float* theta, float* phi,
+                    lapack_complex_float* taup1,
+                    lapack_complex_float* taup2,
+                    lapack_complex_float* tauq1,
+                    lapack_complex_float* tauq2,
+                    lapack_complex_float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_cuncsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    char* signs, lapack_int* m, lapack_int* p,
+                    lapack_int* q, lapack_complex_float* x11,
+                    lapack_int* ldx11, lapack_complex_float* x12,
+                    lapack_int* ldx12, lapack_complex_float* x21,
+                    lapack_int* ldx21, lapack_complex_float* x22,
+                    lapack_int* ldx22, float* theta,
+                    lapack_complex_float* u1, lapack_int* ldu1,
+                    lapack_complex_float* u2, lapack_int* ldu2,
+                    lapack_complex_float* v1t, lapack_int* ldv1t,
+                    lapack_complex_float* v2t, lapack_int* ldv2t,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    float* rwork, lapack_int* lrwork,
+                    lapack_int* iwork , lapack_int *info );
+void LAPACK_dbbcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    double* theta, double* phi, double* u1,
+                    lapack_int* ldu1, double* u2, lapack_int* ldu2,
+                    double* v1t, lapack_int* ldv1t, double* v2t,
+                    lapack_int* ldv2t, double* b11d, double* b11e,
+                    double* b12d, double* b12e, double* b21d,
+                    double* b21e, double* b22d, double* b22e,
+                    double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_dorbdb( char* trans, char* signs,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    double* x11, lapack_int* ldx11, double* x12,
+                    lapack_int* ldx12, double* x21, lapack_int* ldx21,
+                    double* x22, lapack_int* ldx22, double* theta,
+                    double* phi, double* taup1, double* taup2,
+                    double* tauq1, double* tauq2, double* work,
+                    lapack_int* lwork , lapack_int *info );
+void LAPACK_dorcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    char* signs, lapack_int* m, lapack_int* p,
+                    lapack_int* q, double* x11, lapack_int* ldx11,
+                    double* x12, lapack_int* ldx12, double* x21,
+                    lapack_int* ldx21, double* x22, lapack_int* ldx22,
+                    double* theta, double* u1, lapack_int* ldu1,
+                    double* u2, lapack_int* ldu2, double* v1t,
+                    lapack_int* ldv1t, double* v2t, lapack_int* ldv2t,
+                    double* work, lapack_int* lwork,
+                    lapack_int* iwork , lapack_int *info );
+void LAPACK_dsyconv( char* uplo, char* way,
+                     lapack_int* n, double* a, lapack_int* lda,
+                     const lapack_int* ipiv, double* work , lapack_int *info );
+void LAPACK_dsyswapr( char* uplo, lapack_int* n,
+                      double* a, lapack_int* i1, lapack_int* i2 );
+void LAPACK_dsytri2( char* uplo, lapack_int* n,
+                     double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_dsytri2x( char* uplo, lapack_int* n,
+                      double* a, lapack_int* lda,
+                      const lapack_int* ipiv, double* work,
+                      lapack_int* nb , lapack_int *info );
+void LAPACK_dsytrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs, const double* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     double* b, lapack_int* ldb, double* work , lapack_int *info );
+void LAPACK_sbbcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    float* theta, float* phi, float* u1,
+                    lapack_int* ldu1, float* u2, lapack_int* ldu2,
+                    float* v1t, lapack_int* ldv1t, float* v2t,
+                    lapack_int* ldv2t, float* b11d, float* b11e,
+                    float* b12d, float* b12e, float* b21d,
+                    float* b21e, float* b22d, float* b22e,
+                    float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_sorbdb( char* trans, char* signs,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    float* x11, lapack_int* ldx11, float* x12,
+                    lapack_int* ldx12, float* x21, lapack_int* ldx21,
+                    float* x22, lapack_int* ldx22, float* theta,
+                    float* phi, float* taup1, float* taup2,
+                    float* tauq1, float* tauq2, float* work,
+                    lapack_int* lwork , lapack_int *info );
+void LAPACK_sorcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    char* signs, lapack_int* m, lapack_int* p,
+                    lapack_int* q, float* x11, lapack_int* ldx11,
+                    float* x12, lapack_int* ldx12, float* x21,
+                    lapack_int* ldx21, float* x22, lapack_int* ldx22,
+                    float* theta, float* u1, lapack_int* ldu1,
+                    float* u2, lapack_int* ldu2, float* v1t,
+                    lapack_int* ldv1t, float* v2t, lapack_int* ldv2t,
+                    float* work, lapack_int* lwork,
+                    lapack_int* iwork , lapack_int *info );
+void LAPACK_ssyconv( char* uplo, char* way,
+                     lapack_int* n, float* a, lapack_int* lda,
+                     const lapack_int* ipiv, float* work , lapack_int *info );
+void LAPACK_ssyswapr( char* uplo, lapack_int* n,
+                      float* a, lapack_int* i1, lapack_int* i2 );
+void LAPACK_ssytri2( char* uplo, lapack_int* n,
+                     float* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_ssytri2x( char* uplo, lapack_int* n,
+                      float* a, lapack_int* lda,
+                      const lapack_int* ipiv, float* work,
+                      lapack_int* nb , lapack_int *info );
+void LAPACK_ssytrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs, const float* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     float* b, lapack_int* ldb, float* work , lapack_int *info );
+void LAPACK_zbbcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    double* theta, double* phi,
+                    lapack_complex_double* u1, lapack_int* ldu1,
+                    lapack_complex_double* u2, lapack_int* ldu2,
+                    lapack_complex_double* v1t, lapack_int* ldv1t,
+                    lapack_complex_double* v2t, lapack_int* ldv2t,
+                    double* b11d, double* b11e, double* b12d,
+                    double* b12e, double* b21d, double* b21e,
+                    double* b22d, double* b22e, double* rwork,
+                    lapack_int* lrwork , lapack_int *info );
+void LAPACK_zheswapr( char* uplo, lapack_int* n,
+                      lapack_complex_double* a, lapack_int* i1,
+                      lapack_int* i2 );
+void LAPACK_zhetri2( char* uplo, lapack_int* n,
+                     lapack_complex_double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_zhetri2x( char* uplo, lapack_int* n,
+                      lapack_complex_double* a, lapack_int* lda,
+                      const lapack_int* ipiv,
+                      lapack_complex_double* work, lapack_int* nb , lapack_int *info );
+void LAPACK_zhetrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* work , lapack_int *info );
+void LAPACK_zsyconv( char* uplo, char* way,
+                     lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     lapack_complex_double* work , lapack_int *info );
+void LAPACK_zsyswapr( char* uplo, lapack_int* n,
+                      lapack_complex_double* a, lapack_int* i1,
+                      lapack_int* i2 );
+void LAPACK_zsytri2( char* uplo, lapack_int* n,
+                     lapack_complex_double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_zsytri2x( char* uplo, lapack_int* n,
+                      lapack_complex_double* a, lapack_int* lda,
+                      const lapack_int* ipiv,
+                      lapack_complex_double* work, lapack_int* nb , lapack_int *info );
+void LAPACK_zsytrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* work , lapack_int *info );
+void LAPACK_zunbdb( char* trans, char* signs,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    lapack_complex_double* x11, lapack_int* ldx11,
+                    lapack_complex_double* x12, lapack_int* ldx12,
+                    lapack_complex_double* x21, lapack_int* ldx21,
+                    lapack_complex_double* x22, lapack_int* ldx22,
+                    double* theta, double* phi,
+                    lapack_complex_double* taup1,
+                    lapack_complex_double* taup2,
+                    lapack_complex_double* tauq1,
+                    lapack_complex_double* tauq2,
+                    lapack_complex_double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_zuncsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    char* signs, lapack_int* m, lapack_int* p,
+                    lapack_int* q, lapack_complex_double* x11,
+                    lapack_int* ldx11, lapack_complex_double* x12,
+                    lapack_int* ldx12, lapack_complex_double* x21,
+                    lapack_int* ldx21, lapack_complex_double* x22,
+                    lapack_int* ldx22, double* theta,
+                    lapack_complex_double* u1, lapack_int* ldu1,
+                    lapack_complex_double* u2, lapack_int* ldu2,
+                    lapack_complex_double* v1t, lapack_int* ldv1t,
+                    lapack_complex_double* v2t, lapack_int* ldv2t,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork,
+                    lapack_int* iwork , lapack_int *info );
+// LAPACK 3.4.0
+void LAPACK_sgemqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* nb, const float* v,
+                     lapack_int* ldv, const float* t, lapack_int* ldt, float* c,
+                     lapack_int* ldc, float* work, lapack_int *info );
+void LAPACK_dgemqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* nb, const double* v,
+                     lapack_int* ldv, const double* t, lapack_int* ldt,
+                     double* c, lapack_int* ldc, double* work,
+                     lapack_int *info );
+void LAPACK_cgemqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* nb,
+                     const lapack_complex_float* v, lapack_int* ldv,
+                     const lapack_complex_float* t, lapack_int* ldt,
+                     lapack_complex_float* c, lapack_int* ldc,
+                     lapack_complex_float* work, lapack_int *info );
+void LAPACK_zgemqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* nb,
+                     const lapack_complex_double* v, lapack_int* ldv,
+                     const lapack_complex_double* t, lapack_int* ldt,
+                     lapack_complex_double* c, lapack_int* ldc,
+                     lapack_complex_double* work, lapack_int *info );
+void LAPACK_sgeqrt( lapack_int* m, lapack_int* n, lapack_int* nb, float* a,
+                    lapack_int* lda, float* t, lapack_int* ldt, float* work,
+                    lapack_int *info );
+void LAPACK_dgeqrt( lapack_int* m, lapack_int* n, lapack_int* nb, double* a,
+                    lapack_int* lda, double* t, lapack_int* ldt, double* work,
+                    lapack_int *info );
+void LAPACK_cgeqrt( lapack_int* m, lapack_int* n, lapack_int* nb,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zgeqrt( lapack_int* m, lapack_int* n, lapack_int* nb,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_sgeqrt2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                     float* t, lapack_int* ldt, lapack_int *info );
+void LAPACK_dgeqrt2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                     double* t, lapack_int* ldt, lapack_int *info );
+void LAPACK_cgeqrt2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, lapack_complex_float* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_zgeqrt2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, lapack_complex_double* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_sgeqrt3( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                     float* t, lapack_int* ldt, lapack_int *info );
+void LAPACK_dgeqrt3( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                     double* t, lapack_int* ldt, lapack_int *info );
+void LAPACK_cgeqrt3( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, lapack_complex_float* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_zgeqrt3( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, lapack_complex_double* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_stpmqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* l, lapack_int* nb,
+                     const float* v, lapack_int* ldv, const float* t,
+                     lapack_int* ldt, float* a, lapack_int* lda, float* b,
+                     lapack_int* ldb, float* work, lapack_int *info );
+void LAPACK_dtpmqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* l, lapack_int* nb,
+                     const double* v, lapack_int* ldv, const double* t,
+                     lapack_int* ldt, double* a, lapack_int* lda, double* b,
+                     lapack_int* ldb, double* work, lapack_int *info );
+void LAPACK_ctpmqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* l, lapack_int* nb,
+                     const lapack_complex_float* v, lapack_int* ldv,
+                     const lapack_complex_float* t, lapack_int* ldt,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* work, lapack_int *info );
+void LAPACK_ztpmqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* l, lapack_int* nb,
+                     const lapack_complex_double* v, lapack_int* ldv,
+                     const lapack_complex_double* t, lapack_int* ldt,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* work, lapack_int *info );
+void LAPACK_dtpqrt( lapack_int* m, lapack_int* n, lapack_int* l, lapack_int* nb,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* t, lapack_int* ldt, double* work,
+                    lapack_int *info );
+void LAPACK_ctpqrt( lapack_int* m, lapack_int* n, lapack_int* l, lapack_int* nb,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* t, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int* ldt,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_ztpqrt( lapack_int* m, lapack_int* n, lapack_int* l, lapack_int* nb,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_stpqrt2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                     float* b, lapack_int* ldb, float* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_dtpqrt2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                     double* b, lapack_int* ldb, double* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_ctpqrt2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_ztpqrt2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_stprfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, lapack_int* l,
+                    const float* v, lapack_int* ldv, const float* t,
+                    lapack_int* ldt, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, const float* mywork,
+                    lapack_int* myldwork );
+void LAPACK_dtprfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, lapack_int* l,
+                    const double* v, lapack_int* ldv, const double* t,
+                    lapack_int* ldt, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, const double* mywork,
+                    lapack_int* myldwork );
+void LAPACK_ctprfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, lapack_int* l,
+                    const lapack_complex_float* v, lapack_int* ldv,
+                    const lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    const float* mywork, lapack_int* myldwork );
+void LAPACK_ztprfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, lapack_int* l,
+                    const lapack_complex_double* v, lapack_int* ldv,
+                    const lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    const double* mywork, lapack_int* myldwork );
+// LAPACK 3.X.X
+void LAPACK_csyr( char* uplo, lapack_int* n, lapack_complex_float* alpha,
+                      const lapack_complex_float* x, lapack_int* incx,
+                      lapack_complex_float* a, lapack_int* lda );
+void LAPACK_zsyr( char* uplo, lapack_int* n, lapack_complex_double* alpha,
+                      const lapack_complex_double* x, lapack_int* incx,
+                      lapack_complex_double* a, lapack_int* lda );
+
+#ifdef __cplusplus
+}
+#endif /* __cplusplus */
+
+#endif /* _LAPACKE_H_ */
+
+#endif /* _MKL_LAPACKE_H_ */
diff --git a/extern/Eigen3/Eigen/src/misc/lapacke_mangling.h b/extern/Eigen3/Eigen/src/misc/lapacke_mangling.h
new file mode 100644
index 00000000000..6211fd144d3
--- /dev/null
+++ b/extern/Eigen3/Eigen/src/misc/lapacke_mangling.h
@@ -0,0 +1,17 @@
+#ifndef LAPACK_HEADER_INCLUDED
+#define LAPACK_HEADER_INCLUDED
+
+#ifndef LAPACK_GLOBAL
+#if defined(LAPACK_GLOBAL_PATTERN_LC) || defined(ADD_)
+#define LAPACK_GLOBAL(lcname,UCNAME)  lcname##_
+#elif defined(LAPACK_GLOBAL_PATTERN_UC) || defined(UPPER)
+#define LAPACK_GLOBAL(lcname,UCNAME)  UCNAME
+#elif defined(LAPACK_GLOBAL_PATTERN_MC) || defined(NOCHANGE)
+#define LAPACK_GLOBAL(lcname,UCNAME)  lcname
+#else
+#define LAPACK_GLOBAL(lcname,UCNAME)  lcname##_
+#endif
+#endif
+
+#endif
+
diff --git a/extern/Eigen3/Eigen/src/plugins/ArrayCwiseBinaryOps.h b/extern/Eigen3/Eigen/src/plugins/ArrayCwiseBinaryOps.h
index 1951286f3ae..1f8a531af59 100644
--- a/extern/Eigen3/Eigen/src/plugins/ArrayCwiseBinaryOps.h
+++ b/extern/Eigen3/Eigen/src/plugins/ArrayCwiseBinaryOps.h
@@ -1,12 +1,14 @@
+
 /** \returns an expression of the coefficient wise product of \c *this and \a other
   *
   * \sa MatrixBase::cwiseProduct
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_PRODUCT_RETURN_TYPE(Derived,OtherDerived)
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,product)
 operator*(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return EIGEN_CWISE_PRODUCT_RETURN_TYPE(Derived,OtherDerived)(derived(), other.derived());
+  return EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,product)(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient wise quotient of \c *this and \a other
@@ -14,10 +16,11 @@ operator*(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
   * \sa MatrixBase::cwiseQuotient
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_quotient_op<Scalar>, const Derived, const OtherDerived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_quotient_op<Scalar,typename OtherDerived::Scalar>, const Derived, const OtherDerived>
 operator/(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return CwiseBinaryOp<internal::scalar_quotient_op<Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
+  return CwiseBinaryOp<internal::scalar_quotient_op<Scalar,typename OtherDerived::Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient-wise min of \c *this and \a other
@@ -27,13 +30,14 @@ operator/(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
   *
   * \sa max()
   */
-EIGEN_MAKE_CWISE_BINARY_OP(min,internal::scalar_min_op)
+EIGEN_MAKE_CWISE_BINARY_OP(min,min)
 
 /** \returns an expression of the coefficient-wise min of \c *this and scalar \a other
   *
   * \sa max()
   */
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar>, const Derived,
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar,Scalar>, const Derived,
                                         const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject> >
 #ifdef EIGEN_PARSED_BY_DOXYGEN
 min
@@ -52,13 +56,14 @@ min
   *
   * \sa min()
   */
-EIGEN_MAKE_CWISE_BINARY_OP(max,internal::scalar_max_op)
+EIGEN_MAKE_CWISE_BINARY_OP(max,max)
 
 /** \returns an expression of the coefficient-wise max of \c *this and scalar \a other
   *
   * \sa min()
   */
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar>, const Derived,
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar,Scalar>, const Derived,
                                         const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject> >
 #ifdef EIGEN_PARSED_BY_DOXYGEN
 max
@@ -70,33 +75,62 @@ max
   return (max)(Derived::PlainObject::Constant(rows(), cols(), other));
 }
 
+/** \returns an expression of the coefficient-wise power of \c *this to the given array of \a exponents.
+  *
+  * This function computes the coefficient-wise power.
+  *
+  * Example: \include Cwise_array_power_array.cpp
+  * Output: \verbinclude Cwise_array_power_array.out
+  */
+EIGEN_MAKE_CWISE_BINARY_OP(pow,pow)
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(pow,pow)
+#else
+/** \returns an expression of the coefficients of \c *this rasied to the constant power \a exponent
+  *
+  * \tparam T is the scalar type of \a exponent. It must be compatible with the scalar type of the given expression.
+  *
+  * This function computes the coefficient-wise power. The function MatrixBase::pow() in the
+  * unsupported module MatrixFunctions computes the matrix power.
+  *
+  * Example: \include Cwise_pow.cpp
+  * Output: \verbinclude Cwise_pow.out
+  *
+  * \sa ArrayBase::pow(ArrayBase), square(), cube(), exp(), log()
+  */
+template<typename T>
+const CwiseBinaryOp<internal::scalar_pow_op<Scalar,T>,Derived,Constant<T> > pow(const T& exponent) const;
+#endif
+
 
+// TODO code generating macros could be moved to Macros.h and could include generation of documentation
 #define EIGEN_MAKE_CWISE_COMP_OP(OP, COMPARATOR) \
 template<typename OtherDerived> \
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const OtherDerived> \
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_cmp_op<Scalar, typename OtherDerived::Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const OtherDerived> \
 OP(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \
 { \
-  return CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const OtherDerived>(derived(), other.derived()); \
+  return CwiseBinaryOp<internal::scalar_cmp_op<Scalar, typename OtherDerived::Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const OtherDerived>(derived(), other.derived()); \
 }\
-typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject> > Cmp ## COMPARATOR ## ReturnType; \
-typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_ ## COMPARATOR>, const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject>, const Derived > RCmp ## COMPARATOR ## ReturnType; \
-EIGEN_STRONG_INLINE const Cmp ## COMPARATOR ## ReturnType \
+typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar,Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject> > Cmp ## COMPARATOR ## ReturnType; \
+typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar,Scalar, internal::cmp_ ## COMPARATOR>, const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject>, const Derived > RCmp ## COMPARATOR ## ReturnType; \
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Cmp ## COMPARATOR ## ReturnType \
 OP(const Scalar& s) const { \
   return this->OP(Derived::PlainObject::Constant(rows(), cols(), s)); \
 } \
-friend EIGEN_STRONG_INLINE const RCmp ## COMPARATOR ## ReturnType \
+EIGEN_DEVICE_FUNC friend EIGEN_STRONG_INLINE const RCmp ## COMPARATOR ## ReturnType \
 OP(const Scalar& s, const Derived& d) { \
   return Derived::PlainObject::Constant(d.rows(), d.cols(), s).OP(d); \
 }
 
 #define EIGEN_MAKE_CWISE_COMP_R_OP(OP, R_OP, RCOMPARATOR) \
 template<typename OtherDerived> \
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_##RCOMPARATOR>, const OtherDerived, const Derived> \
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_cmp_op<typename OtherDerived::Scalar, Scalar, internal::cmp_##RCOMPARATOR>, const OtherDerived, const Derived> \
 OP(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \
 { \
-  return CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_##RCOMPARATOR>, const OtherDerived, const Derived>(other.derived(), derived()); \
+  return CwiseBinaryOp<internal::scalar_cmp_op<typename OtherDerived::Scalar, Scalar, internal::cmp_##RCOMPARATOR>, const OtherDerived, const Derived>(other.derived(), derived()); \
 } \
-\
+EIGEN_DEVICE_FUNC \
 inline const RCmp ## RCOMPARATOR ## ReturnType \
 OP(const Scalar& s) const { \
   return Derived::PlainObject::Constant(rows(), cols(), s).R_OP(*this); \
@@ -107,6 +141,7 @@ OP(const Scalar& s, const Derived& d) { \
 }
 
 
+
 /** \returns an expression of the coefficient-wise \< operator of *this and \a other
   *
   * Example: \include Cwise_less.cpp
@@ -171,83 +206,127 @@ EIGEN_MAKE_CWISE_COMP_OP(operator==, EQ)
   */
 EIGEN_MAKE_CWISE_COMP_OP(operator!=, NEQ)
 
+
 #undef EIGEN_MAKE_CWISE_COMP_OP
 #undef EIGEN_MAKE_CWISE_COMP_R_OP
 
 // scalar addition
-
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP(operator+,sum)
+#else
 /** \returns an expression of \c *this with each coeff incremented by the constant \a scalar
   *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  *
   * Example: \include Cwise_plus.cpp
   * Output: \verbinclude Cwise_plus.out
   *
   * \sa operator+=(), operator-()
   */
-inline const CwiseUnaryOp<internal::scalar_add_op<Scalar>, const Derived>
-operator+(const Scalar& scalar) const
-{
-  return CwiseUnaryOp<internal::scalar_add_op<Scalar>, const Derived>(derived(), internal::scalar_add_op<Scalar>(scalar));
-}
-
-friend inline const CwiseUnaryOp<internal::scalar_add_op<Scalar>, const Derived>
-operator+(const Scalar& scalar,const EIGEN_CURRENT_STORAGE_BASE_CLASS<Derived>& other)
-{
-  return other + scalar;
-}
+template<typename T>
+const CwiseBinaryOp<internal::scalar_sum_op<Scalar,T>,Derived,Constant<T> > operator+(const T& scalar) const;
+/** \returns an expression of \a expr with each coeff incremented by the constant \a scalar
+  *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  */
+template<typename T> friend
+const CwiseBinaryOp<internal::scalar_sum_op<T,Scalar>,Constant<T>,Derived> operator+(const T& scalar, const StorageBaseType& expr);
+#endif
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP(operator-,difference)
+#else
 /** \returns an expression of \c *this with each coeff decremented by the constant \a scalar
   *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  *
   * Example: \include Cwise_minus.cpp
   * Output: \verbinclude Cwise_minus.out
   *
-  * \sa operator+(), operator-=()
+  * \sa operator+=(), operator-()
   */
-inline const CwiseUnaryOp<internal::scalar_add_op<Scalar>, const Derived>
-operator-(const Scalar& scalar) const
-{
-  return *this + (-scalar);
-}
+template<typename T>
+const CwiseBinaryOp<internal::scalar_difference_op<Scalar,T>,Derived,Constant<T> > operator-(const T& scalar) const;
+/** \returns an expression of the constant matrix of value \a scalar decremented by the coefficients of \a expr
+  *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  */
+template<typename T> friend
+const CwiseBinaryOp<internal::scalar_difference_op<T,Scalar>,Constant<T>,Derived> operator-(const T& scalar, const StorageBaseType& expr);
+#endif
 
-friend inline const CwiseUnaryOp<internal::scalar_add_op<Scalar>, const CwiseUnaryOp<internal::scalar_opposite_op<Scalar>, const Derived> >
-operator-(const Scalar& scalar,const EIGEN_CURRENT_STORAGE_BASE_CLASS<Derived>& other)
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+  EIGEN_MAKE_SCALAR_BINARY_OP_ONTHELEFT(operator/,quotient)
+#else
+  /**
+    * \brief Component-wise division of the scalar \a s by array elements of \a a.
+    *
+    * \tparam Scalar is the scalar type of \a x. It must be compatible with the scalar type of the given array expression (\c Derived::Scalar).
+    */
+  template<typename T> friend
+  inline const CwiseBinaryOp<internal::scalar_quotient_op<T,Scalar>,Constant<T>,Derived>
+  operator/(const T& s,const StorageBaseType& a);
+#endif
+
+/** \returns an expression of the coefficient-wise ^ operator of *this and \a other
+ *
+ * \warning this operator is for expression of bool only.
+ *
+ * Example: \include Cwise_boolean_xor.cpp
+ * Output: \verbinclude Cwise_boolean_xor.out
+ *
+ * \sa operator&&(), select()
+ */
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
+inline const CwiseBinaryOp<internal::scalar_boolean_xor_op, const Derived, const OtherDerived>
+operator^(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return (-other) + scalar;
+  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
+                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
+  return CwiseBinaryOp<internal::scalar_boolean_xor_op, const Derived, const OtherDerived>(derived(),other.derived());
 }
 
-/** \returns an expression of the coefficient-wise && operator of *this and \a other
+// NOTE disabled until we agree on argument order
+#if 0
+/** \cpp11 \returns an expression of the coefficient-wise polygamma function.
   *
-  * \warning this operator is for expression of bool only.
+  * \specialfunctions_module
   *
-  * Example: \include Cwise_boolean_and.cpp
-  * Output: \verbinclude Cwise_boolean_and.out
+  * It returns the \a n -th derivative of the digamma(psi) evaluated at \c *this.
   *
-  * \sa operator||(), select()
+  * \warning Be careful with the order of the parameters: x.polygamma(n) is equivalent to polygamma(n,x)
+  *
+  * \sa Eigen::polygamma()
   */
-template<typename OtherDerived>
-inline const CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>
-operator&&(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
+template<typename DerivedN>
+inline const CwiseBinaryOp<internal::scalar_polygamma_op<Scalar>, const DerivedN, const Derived>
+polygamma(const EIGEN_CURRENT_STORAGE_BASE_CLASS<DerivedN> &n) const
 {
-  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
-                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
-  return CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>(derived(),other.derived());
+  return CwiseBinaryOp<internal::scalar_polygamma_op<Scalar>, const DerivedN, const Derived>(n.derived(), this->derived());
 }
+#endif
 
-/** \returns an expression of the coefficient-wise || operator of *this and \a other
+/** \returns an expression of the coefficient-wise zeta function.
+  *
+  * \specialfunctions_module
+  *
+  * It returns the Riemann zeta function of two arguments \c *this and \a q:
   *
-  * \warning this operator is for expression of bool only.
+  * \param *this is the exposent, it must be > 1
+  * \param q is the shift, it must be > 0
   *
-  * Example: \include Cwise_boolean_or.cpp
-  * Output: \verbinclude Cwise_boolean_or.out
+  * \note This function supports only float and double scalar types. To support other scalar types, the user has
+  * to provide implementations of zeta(T,T) for any scalar type T to be supported.
   *
-  * \sa operator&&(), select()
+  * This method is an alias for zeta(*this,q);
+  *
+  * \sa Eigen::zeta()
   */
-template<typename OtherDerived>
-inline const CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>
-operator||(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
+template<typename DerivedQ>
+inline const CwiseBinaryOp<internal::scalar_zeta_op<Scalar>, const Derived, const DerivedQ>
+zeta(const EIGEN_CURRENT_STORAGE_BASE_CLASS<DerivedQ> &q) const
 {
-  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
-                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
-  return CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>(derived(),other.derived());
+  return CwiseBinaryOp<internal::scalar_zeta_op<Scalar>, const Derived, const DerivedQ>(this->derived(), q.derived());
 }
-
-
diff --git a/extern/Eigen3/Eigen/src/plugins/ArrayCwiseUnaryOps.h b/extern/Eigen3/Eigen/src/plugins/ArrayCwiseUnaryOps.h
index 1c3ed3fcd70..ebaa3f192b1 100644
--- a/extern/Eigen3/Eigen/src/plugins/ArrayCwiseUnaryOps.h
+++ b/extern/Eigen3/Eigen/src/plugins/ArrayCwiseUnaryOps.h
@@ -1,16 +1,62 @@
 
 
+typedef CwiseUnaryOp<internal::scalar_abs_op<Scalar>, const Derived> AbsReturnType;
+typedef CwiseUnaryOp<internal::scalar_arg_op<Scalar>, const Derived> ArgReturnType;
+typedef CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const Derived> Abs2ReturnType;
+typedef CwiseUnaryOp<internal::scalar_sqrt_op<Scalar>, const Derived> SqrtReturnType;
+typedef CwiseUnaryOp<internal::scalar_rsqrt_op<Scalar>, const Derived> RsqrtReturnType;
+typedef CwiseUnaryOp<internal::scalar_sign_op<Scalar>, const Derived> SignReturnType;
+typedef CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const Derived> InverseReturnType;
+typedef CwiseUnaryOp<internal::scalar_boolean_not_op<Scalar>, const Derived> BooleanNotReturnType;
+
+typedef CwiseUnaryOp<internal::scalar_exp_op<Scalar>, const Derived> ExpReturnType;
+typedef CwiseUnaryOp<internal::scalar_log_op<Scalar>, const Derived> LogReturnType;
+typedef CwiseUnaryOp<internal::scalar_log1p_op<Scalar>, const Derived> Log1pReturnType;
+typedef CwiseUnaryOp<internal::scalar_log10_op<Scalar>, const Derived> Log10ReturnType;
+typedef CwiseUnaryOp<internal::scalar_cos_op<Scalar>, const Derived> CosReturnType;
+typedef CwiseUnaryOp<internal::scalar_sin_op<Scalar>, const Derived> SinReturnType;
+typedef CwiseUnaryOp<internal::scalar_tan_op<Scalar>, const Derived> TanReturnType;
+typedef CwiseUnaryOp<internal::scalar_acos_op<Scalar>, const Derived> AcosReturnType;
+typedef CwiseUnaryOp<internal::scalar_asin_op<Scalar>, const Derived> AsinReturnType;
+typedef CwiseUnaryOp<internal::scalar_atan_op<Scalar>, const Derived> AtanReturnType;
+typedef CwiseUnaryOp<internal::scalar_tanh_op<Scalar>, const Derived> TanhReturnType;
+typedef CwiseUnaryOp<internal::scalar_sinh_op<Scalar>, const Derived> SinhReturnType;
+typedef CwiseUnaryOp<internal::scalar_cosh_op<Scalar>, const Derived> CoshReturnType;
+typedef CwiseUnaryOp<internal::scalar_square_op<Scalar>, const Derived> SquareReturnType;
+typedef CwiseUnaryOp<internal::scalar_cube_op<Scalar>, const Derived> CubeReturnType;
+typedef CwiseUnaryOp<internal::scalar_round_op<Scalar>, const Derived> RoundReturnType;
+typedef CwiseUnaryOp<internal::scalar_floor_op<Scalar>, const Derived> FloorReturnType;
+typedef CwiseUnaryOp<internal::scalar_ceil_op<Scalar>, const Derived> CeilReturnType;
+typedef CwiseUnaryOp<internal::scalar_isnan_op<Scalar>, const Derived> IsNaNReturnType;
+typedef CwiseUnaryOp<internal::scalar_isinf_op<Scalar>, const Derived> IsInfReturnType;
+typedef CwiseUnaryOp<internal::scalar_isfinite_op<Scalar>, const Derived> IsFiniteReturnType;
+
 /** \returns an expression of the coefficient-wise absolute value of \c *this
   *
   * Example: \include Cwise_abs.cpp
   * Output: \verbinclude Cwise_abs.out
   *
-  * \sa abs2()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_abs">Math functions</a>, abs2()
   */
-EIGEN_STRONG_INLINE const CwiseUnaryOp<internal::scalar_abs_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const AbsReturnType
 abs() const
 {
-  return derived();
+  return AbsReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise phase angle of \c *this
+  *
+  * Example: \include Cwise_arg.cpp
+  * Output: \verbinclude Cwise_arg.out
+  *
+  * \sa abs()
+  */
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const ArgReturnType
+arg() const
+{
+  return ArgReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise squared absolute value of \c *this
@@ -18,78 +64,190 @@ abs() const
   * Example: \include Cwise_abs2.cpp
   * Output: \verbinclude Cwise_abs2.out
   *
-  * \sa abs(), square()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_abs2">Math functions</a>, abs(), square()
   */
-EIGEN_STRONG_INLINE const CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const Abs2ReturnType
 abs2() const
 {
-  return derived();
+  return Abs2ReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise exponential of *this.
   *
+  * This function computes the coefficient-wise exponential. The function MatrixBase::exp() in the
+  * unsupported module MatrixFunctions computes the matrix exponential.
+  *
   * Example: \include Cwise_exp.cpp
   * Output: \verbinclude Cwise_exp.out
   *
-  * \sa pow(), log(), sin(), cos()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_exp">Math functions</a>, pow(), log(), sin(), cos()
   */
-inline const CwiseUnaryOp<internal::scalar_exp_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const ExpReturnType
 exp() const
 {
-  return derived();
+  return ExpReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise logarithm of *this.
   *
+  * This function computes the coefficient-wise logarithm. The function MatrixBase::log() in the
+  * unsupported module MatrixFunctions computes the matrix logarithm.
+  *
   * Example: \include Cwise_log.cpp
   * Output: \verbinclude Cwise_log.out
   *
-  * \sa exp()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_log">Math functions</a>, exp()
   */
-inline const CwiseUnaryOp<internal::scalar_log_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const LogReturnType
 log() const
 {
-  return derived();
+  return LogReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise logarithm of 1 plus \c *this.
+  *
+  * In exact arithmetic, \c x.log() is equivalent to \c (x+1).log(),
+  * however, with finite precision, this function is much more accurate when \c x is close to zero.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_log1p">Math functions</a>, log()
+  */
+EIGEN_DEVICE_FUNC
+inline const Log1pReturnType
+log1p() const
+{
+  return Log1pReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise base-10 logarithm of *this.
+  *
+  * This function computes the coefficient-wise base-10 logarithm.
+  *
+  * Example: \include Cwise_log10.cpp
+  * Output: \verbinclude Cwise_log10.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_log10">Math functions</a>, log()
+  */
+EIGEN_DEVICE_FUNC
+inline const Log10ReturnType
+log10() const
+{
+  return Log10ReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise square root of *this.
   *
+  * This function computes the coefficient-wise square root. The function MatrixBase::sqrt() in the
+  * unsupported module MatrixFunctions computes the matrix square root.
+  *
   * Example: \include Cwise_sqrt.cpp
   * Output: \verbinclude Cwise_sqrt.out
   *
-  * \sa pow(), square()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_sqrt">Math functions</a>, pow(), square()
   */
-inline const CwiseUnaryOp<internal::scalar_sqrt_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const SqrtReturnType
 sqrt() const
 {
-  return derived();
+  return SqrtReturnType(derived());
 }
 
+/** \returns an expression of the coefficient-wise inverse square root of *this.
+  *
+  * This function computes the coefficient-wise inverse square root.
+  *
+  * Example: \include Cwise_sqrt.cpp
+  * Output: \verbinclude Cwise_sqrt.out
+  *
+  * \sa pow(), square()
+  */
+EIGEN_DEVICE_FUNC
+inline const RsqrtReturnType
+rsqrt() const
+{
+  return RsqrtReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise signum of *this.
+  *
+  * This function computes the coefficient-wise signum.
+  *
+  * Example: \include Cwise_sign.cpp
+  * Output: \verbinclude Cwise_sign.out
+  *
+  * \sa pow(), square()
+  */
+EIGEN_DEVICE_FUNC
+inline const SignReturnType
+sign() const
+{
+  return SignReturnType(derived());
+}
+
+
 /** \returns an expression of the coefficient-wise cosine of *this.
   *
+  * This function computes the coefficient-wise cosine. The function MatrixBase::cos() in the
+  * unsupported module MatrixFunctions computes the matrix cosine.
+  *
   * Example: \include Cwise_cos.cpp
   * Output: \verbinclude Cwise_cos.out
   *
-  * \sa sin(), acos()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_cos">Math functions</a>, sin(), acos()
   */
-inline const CwiseUnaryOp<internal::scalar_cos_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const CosReturnType
 cos() const
 {
-  return derived();
+  return CosReturnType(derived());
 }
 
 
 /** \returns an expression of the coefficient-wise sine of *this.
   *
+  * This function computes the coefficient-wise sine. The function MatrixBase::sin() in the
+  * unsupported module MatrixFunctions computes the matrix sine.
+  *
   * Example: \include Cwise_sin.cpp
   * Output: \verbinclude Cwise_sin.out
   *
-  * \sa cos(), asin()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_sin">Math functions</a>, cos(), asin()
   */
-inline const CwiseUnaryOp<internal::scalar_sin_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const SinReturnType
 sin() const
 {
-  return derived();
+  return SinReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise tan of *this.
+  *
+  * Example: \include Cwise_tan.cpp
+  * Output: \verbinclude Cwise_tan.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_tan">Math functions</a>, cos(), sin()
+  */
+EIGEN_DEVICE_FUNC
+inline const TanReturnType
+tan() const
+{
+  return TanReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise arc tan of *this.
+  *
+  * Example: \include Cwise_atan.cpp
+  * Output: \verbinclude Cwise_atan.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_atan">Math functions</a>, tan(), asin(), acos()
+  */
+EIGEN_DEVICE_FUNC
+inline const AtanReturnType
+atan() const
+{
+  return AtanReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise arc cosine of *this.
@@ -97,12 +255,13 @@ sin() const
   * Example: \include Cwise_acos.cpp
   * Output: \verbinclude Cwise_acos.out
   *
-  * \sa cos(), asin()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_acos">Math functions</a>, cos(), asin()
   */
-inline const CwiseUnaryOp<internal::scalar_acos_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const AcosReturnType
 acos() const
 {
-  return derived();
+  return AcosReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise arc sine of *this.
@@ -110,42 +269,56 @@ acos() const
   * Example: \include Cwise_asin.cpp
   * Output: \verbinclude Cwise_asin.out
   *
-  * \sa sin(), acos()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_asin">Math functions</a>, sin(), acos()
   */
-inline const CwiseUnaryOp<internal::scalar_asin_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const AsinReturnType
 asin() const
 {
-  return derived();
+  return AsinReturnType(derived());
 }
 
-/** \returns an expression of the coefficient-wise tan of *this.
+/** \returns an expression of the coefficient-wise hyperbolic tan of *this.
   *
-  * Example: \include Cwise_tan.cpp
-  * Output: \verbinclude Cwise_tan.out
+  * Example: \include Cwise_tanh.cpp
+  * Output: \verbinclude Cwise_tanh.out
   *
-  * \sa cos(), sin()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_tanh">Math functions</a>, tan(), sinh(), cosh()
   */
-inline const CwiseUnaryOp<internal::scalar_tan_op<Scalar>, Derived>
-tan() const
+EIGEN_DEVICE_FUNC
+inline const TanhReturnType
+tanh() const
 {
-  return derived();
+  return TanhReturnType(derived());
 }
 
-
-/** \returns an expression of the coefficient-wise power of *this to the given exponent.
+/** \returns an expression of the coefficient-wise hyperbolic sin of *this.
   *
-  * Example: \include Cwise_pow.cpp
-  * Output: \verbinclude Cwise_pow.out
+  * Example: \include Cwise_sinh.cpp
+  * Output: \verbinclude Cwise_sinh.out
   *
-  * \sa exp(), log()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_sinh">Math functions</a>, sin(), tanh(), cosh()
   */
-inline const CwiseUnaryOp<internal::scalar_pow_op<Scalar>, const Derived>
-pow(const Scalar& exponent) const
+EIGEN_DEVICE_FUNC
+inline const SinhReturnType
+sinh() const
 {
-  return CwiseUnaryOp<internal::scalar_pow_op<Scalar>, const Derived>
-          (derived(), internal::scalar_pow_op<Scalar>(exponent));
+  return SinhReturnType(derived());
 }
 
+/** \returns an expression of the coefficient-wise hyperbolic cos of *this.
+  *
+  * Example: \include Cwise_cosh.cpp
+  * Output: \verbinclude Cwise_cosh.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_cosh">Math functions</a>, tan(), sinh(), cosh()
+  */
+EIGEN_DEVICE_FUNC
+inline const CoshReturnType
+cosh() const
+{
+  return CoshReturnType(derived());
+}
 
 /** \returns an expression of the coefficient-wise inverse of *this.
   *
@@ -154,10 +327,11 @@ pow(const Scalar& exponent) const
   *
   * \sa operator/(), operator*()
   */
-inline const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const InverseReturnType
 inverse() const
 {
-  return derived();
+  return InverseReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise square of *this.
@@ -165,12 +339,13 @@ inverse() const
   * Example: \include Cwise_square.cpp
   * Output: \verbinclude Cwise_square.out
   *
-  * \sa operator/(), operator*(), abs2()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_squareE">Math functions</a>, abs2(), cube(), pow()
   */
-inline const CwiseUnaryOp<internal::scalar_square_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const SquareReturnType
 square() const
 {
-  return derived();
+  return SquareReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise cube of *this.
@@ -178,10 +353,200 @@ square() const
   * Example: \include Cwise_cube.cpp
   * Output: \verbinclude Cwise_cube.out
   *
-  * \sa square(), pow()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_cube">Math functions</a>, square(), pow()
   */
-inline const CwiseUnaryOp<internal::scalar_cube_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const CubeReturnType
 cube() const
 {
-  return derived();
+  return CubeReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise round of *this.
+  *
+  * Example: \include Cwise_round.cpp
+  * Output: \verbinclude Cwise_round.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_round">Math functions</a>, ceil(), floor()
+  */
+EIGEN_DEVICE_FUNC
+inline const RoundReturnType
+round() const
+{
+  return RoundReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise floor of *this.
+  *
+  * Example: \include Cwise_floor.cpp
+  * Output: \verbinclude Cwise_floor.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_floor">Math functions</a>, ceil(), round()
+  */
+EIGEN_DEVICE_FUNC
+inline const FloorReturnType
+floor() const
+{
+  return FloorReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise ceil of *this.
+  *
+  * Example: \include Cwise_ceil.cpp
+  * Output: \verbinclude Cwise_ceil.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_ceil">Math functions</a>, floor(), round()
+  */
+EIGEN_DEVICE_FUNC
+inline const CeilReturnType
+ceil() const
+{
+  return CeilReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise isnan of *this.
+  *
+  * Example: \include Cwise_isNaN.cpp
+  * Output: \verbinclude Cwise_isNaN.out
+  *
+  * \sa isfinite(), isinf()
+  */
+EIGEN_DEVICE_FUNC
+inline const IsNaNReturnType
+isNaN() const
+{
+  return IsNaNReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise isinf of *this.
+  *
+  * Example: \include Cwise_isInf.cpp
+  * Output: \verbinclude Cwise_isInf.out
+  *
+  * \sa isnan(), isfinite()
+  */
+EIGEN_DEVICE_FUNC
+inline const IsInfReturnType
+isInf() const
+{
+  return IsInfReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise isfinite of *this.
+  *
+  * Example: \include Cwise_isFinite.cpp
+  * Output: \verbinclude Cwise_isFinite.out
+  *
+  * \sa isnan(), isinf()
+  */
+EIGEN_DEVICE_FUNC
+inline const IsFiniteReturnType
+isFinite() const
+{
+  return IsFiniteReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise ! operator of *this
+  *
+  * \warning this operator is for expression of bool only.
+  *
+  * Example: \include Cwise_boolean_not.cpp
+  * Output: \verbinclude Cwise_boolean_not.out
+  *
+  * \sa operator!=()
+  */
+EIGEN_DEVICE_FUNC
+inline const BooleanNotReturnType
+operator!() const
+{
+  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value),
+                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
+  return BooleanNotReturnType(derived());
+}
+
+
+// --- SpecialFunctions module ---
+
+typedef CwiseUnaryOp<internal::scalar_lgamma_op<Scalar>, const Derived> LgammaReturnType;
+typedef CwiseUnaryOp<internal::scalar_digamma_op<Scalar>, const Derived> DigammaReturnType;
+typedef CwiseUnaryOp<internal::scalar_erf_op<Scalar>, const Derived> ErfReturnType;
+typedef CwiseUnaryOp<internal::scalar_erfc_op<Scalar>, const Derived> ErfcReturnType;
+
+/** \cpp11 \returns an expression of the coefficient-wise ln(|gamma(*this)|).
+  *
+  * \specialfunctions_module
+  *
+  * Example: \include Cwise_lgamma.cpp
+  * Output: \verbinclude Cwise_lgamma.out
+  *
+  * \note This function supports only float and double scalar types in c++11 mode. To support other scalar types,
+  * or float/double in non c++11 mode, the user has to provide implementations of lgamma(T) for any scalar
+  * type T to be supported.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_lgamma">Math functions</a>, digamma()
+  */
+EIGEN_DEVICE_FUNC
+inline const LgammaReturnType
+lgamma() const
+{
+  return LgammaReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise digamma (psi, derivative of lgamma).
+  *
+  * \specialfunctions_module
+  *
+  * \note This function supports only float and double scalar types. To support other scalar types,
+  * the user has to provide implementations of digamma(T) for any scalar
+  * type T to be supported.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_digamma">Math functions</a>, Eigen::digamma(), Eigen::polygamma(), lgamma()
+  */
+EIGEN_DEVICE_FUNC
+inline const DigammaReturnType
+digamma() const
+{
+  return DigammaReturnType(derived());
+}
+
+/** \cpp11 \returns an expression of the coefficient-wise Gauss error
+  * function of *this.
+  *
+  * \specialfunctions_module
+  *
+  * Example: \include Cwise_erf.cpp
+  * Output: \verbinclude Cwise_erf.out
+  *
+  * \note This function supports only float and double scalar types in c++11 mode. To support other scalar types,
+  * or float/double in non c++11 mode, the user has to provide implementations of erf(T) for any scalar
+  * type T to be supported.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_erf">Math functions</a>, erfc()
+  */
+EIGEN_DEVICE_FUNC
+inline const ErfReturnType
+erf() const
+{
+  return ErfReturnType(derived());
+}
+
+/** \cpp11 \returns an expression of the coefficient-wise Complementary error
+  * function of *this.
+  *
+  * \specialfunctions_module
+  *
+  * Example: \include Cwise_erfc.cpp
+  * Output: \verbinclude Cwise_erfc.out
+  *
+  * \note This function supports only float and double scalar types in c++11 mode. To support other scalar types,
+  * or float/double in non c++11 mode, the user has to provide implementations of erfc(T) for any scalar
+  * type T to be supported.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_erfc">Math functions</a>, erf()
+  */
+EIGEN_DEVICE_FUNC
+inline const ErfcReturnType
+erfc() const
+{
+  return ErfcReturnType(derived());
 }
diff --git a/extern/Eigen3/Eigen/src/plugins/BlockMethods.h b/extern/Eigen3/Eigen/src/plugins/BlockMethods.h
index 2788251e0cd..ac35a0086cf 100644
--- a/extern/Eigen3/Eigen/src/plugins/BlockMethods.h
+++ b/extern/Eigen3/Eigen/src/plugins/BlockMethods.h
@@ -8,27 +8,32 @@
 // Public License v. 2.0. If a copy of the MPL was not distributed
 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
-
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 
-/** \internal expression type of a column */
+/// \internal expression type of a column */
 typedef Block<Derived, internal::traits<Derived>::RowsAtCompileTime, 1, !IsRowMajor> ColXpr;
 typedef const Block<const Derived, internal::traits<Derived>::RowsAtCompileTime, 1, !IsRowMajor> ConstColXpr;
-/** \internal expression type of a row */
+/// \internal expression type of a row */
 typedef Block<Derived, 1, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> RowXpr;
 typedef const Block<const Derived, 1, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> ConstRowXpr;
-/** \internal expression type of a block of whole columns */
+/// \internal expression type of a block of whole columns */
 typedef Block<Derived, internal::traits<Derived>::RowsAtCompileTime, Dynamic, !IsRowMajor> ColsBlockXpr;
 typedef const Block<const Derived, internal::traits<Derived>::RowsAtCompileTime, Dynamic, !IsRowMajor> ConstColsBlockXpr;
-/** \internal expression type of a block of whole rows */
+/// \internal expression type of a block of whole rows */
 typedef Block<Derived, Dynamic, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> RowsBlockXpr;
 typedef const Block<const Derived, Dynamic, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> ConstRowsBlockXpr;
-/** \internal expression type of a block of whole columns */
+/// \internal expression type of a block of whole columns */
 template<int N> struct NColsBlockXpr { typedef Block<Derived, internal::traits<Derived>::RowsAtCompileTime, N, !IsRowMajor> Type; };
 template<int N> struct ConstNColsBlockXpr { typedef const Block<const Derived, internal::traits<Derived>::RowsAtCompileTime, N, !IsRowMajor> Type; };
-/** \internal expression type of a block of whole rows */
+/// \internal expression type of a block of whole rows */
 template<int N> struct NRowsBlockXpr { typedef Block<Derived, N, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> Type; };
 template<int N> struct ConstNRowsBlockXpr { typedef const Block<const Derived, N, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> Type; };
+/// \internal expression of a block */
+typedef Block<Derived> BlockXpr;
+typedef const Block<const Derived> ConstBlockXpr;
+/// \internal expression of a block of fixed sizes */
+template<int Rows, int Cols> struct FixedBlockXpr { typedef Block<Derived,Rows,Cols> Type; };
+template<int Rows, int Cols> struct ConstFixedBlockXpr { typedef Block<const Derived,Rows,Cols> Type; };
 
 typedef VectorBlock<Derived> SegmentReturnType;
 typedef const VectorBlock<const Derived> ConstSegmentReturnType;
@@ -37,378 +42,430 @@ template<int Size> struct ConstFixedSegmentReturnType { typedef const VectorBloc
 
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
-/** \returns a dynamic-size expression of a block in *this.
-  *
-  * \param startRow the first row in the block
-  * \param startCol the first column in the block
-  * \param blockRows the number of rows in the block
-  * \param blockCols the number of columns in the block
-  *
-  * Example: \include MatrixBase_block_int_int_int_int.cpp
-  * Output: \verbinclude MatrixBase_block_int_int_int_int.out
-  *
-  * \note Even though the returned expression has dynamic size, in the case
-  * when it is applied to a fixed-size matrix, it inherits a fixed maximal size,
-  * which means that evaluating it does not cause a dynamic memory allocation.
-  *
-  * \sa class Block, block(Index,Index)
-  */
-inline Block<Derived> block(Index startRow, Index startCol, Index blockRows, Index blockCols)
-{
-  return Block<Derived>(derived(), startRow, startCol, blockRows, blockCols);
-}
-
-/** This is the const version of block(Index,Index,Index,Index). */
-inline const Block<const Derived> block(Index startRow, Index startCol, Index blockRows, Index blockCols) const
-{
-  return Block<const Derived>(derived(), startRow, startCol, blockRows, blockCols);
-}
-
-
-
-
-/** \returns a dynamic-size expression of a top-right corner of *this.
-  *
-  * \param cRows the number of rows in the corner
-  * \param cCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_topRightCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_topRightCorner_int_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-inline Block<Derived> topRightCorner(Index cRows, Index cCols)
-{
-  return Block<Derived>(derived(), 0, cols() - cCols, cRows, cCols);
-}
-
-/** This is the const version of topRightCorner(Index, Index).*/
-inline const Block<const Derived> topRightCorner(Index cRows, Index cCols) const
-{
-  return Block<const Derived>(derived(), 0, cols() - cCols, cRows, cCols);
-}
-
-/** \returns an expression of a fixed-size top-right corner of *this.
-  *
-  * \tparam CRows the number of rows in the corner
-  * \tparam CCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_template_int_int_topRightCorner.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_topRightCorner.out
-  *
-  * \sa class Block, block<int,int>(Index,Index)
-  */
+/// \returns a dynamic-size expression of a block in *this.
+///
+/// \param startRow the first row in the block
+/// \param startCol the first column in the block
+/// \param blockRows the number of rows in the block
+/// \param blockCols the number of columns in the block
+///
+/// Example: \include MatrixBase_block_int_int_int_int.cpp
+/// Output: \verbinclude MatrixBase_block_int_int_int_int.out
+///
+/// \note Even though the returned expression has dynamic size, in the case
+/// when it is applied to a fixed-size matrix, it inherits a fixed maximal size,
+/// which means that evaluating it does not cause a dynamic memory allocation.
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr block(Index startRow, Index startCol, Index blockRows, Index blockCols)
+{
+  return BlockXpr(derived(), startRow, startCol, blockRows, blockCols);
+}
+
+/// This is the const version of block(Index,Index,Index,Index). */
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr block(Index startRow, Index startCol, Index blockRows, Index blockCols) const
+{
+  return ConstBlockXpr(derived(), startRow, startCol, blockRows, blockCols);
+}
+
+
+
+
+/// \returns a dynamic-size expression of a top-right corner of *this.
+///
+/// \param cRows the number of rows in the corner
+/// \param cCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_topRightCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_topRightCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr topRightCorner(Index cRows, Index cCols)
+{
+  return BlockXpr(derived(), 0, cols() - cCols, cRows, cCols);
+}
+
+/// This is the const version of topRightCorner(Index, Index).
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr topRightCorner(Index cRows, Index cCols) const
+{
+  return ConstBlockXpr(derived(), 0, cols() - cCols, cRows, cCols);
+}
+
+/// \returns an expression of a fixed-size top-right corner of *this.
+///
+/// \tparam CRows the number of rows in the corner
+/// \tparam CCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_template_int_int_topRightCorner.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_topRightCorner.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block<int,int>(Index,Index)
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> topRightCorner()
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<CRows,CCols>::Type topRightCorner()
 {
-  return Block<Derived, CRows, CCols>(derived(), 0, cols() - CCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), 0, cols() - CCols);
 }
 
-/** This is the const version of topRightCorner<int, int>().*/
+/// This is the const version of topRightCorner<int, int>().
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> topRightCorner() const
-{
-  return Block<const Derived, CRows, CCols>(derived(), 0, cols() - CCols);
-}
-
-/** \returns an expression of a top-right corner of *this.
-  *
-  * \tparam CRows number of rows in corner as specified at compile-time
-  * \tparam CCols number of columns in corner as specified at compile-time
-  * \param  cRows number of rows in corner as specified at run-time
-  * \param  cCols number of columns in corner as specified at run-time
-  *
-  * This function is mainly useful for corners where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a cRows should equal \a CRows unless
-  * \a CRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_topRightCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_topRightCorner_int_int.out
-  *
-  * \sa class Block
-  */
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type topRightCorner() const
+{
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), 0, cols() - CCols);
+}
+
+/// \returns an expression of a top-right corner of *this.
+///
+/// \tparam CRows number of rows in corner as specified at compile-time
+/// \tparam CCols number of columns in corner as specified at compile-time
+/// \param  cRows number of rows in corner as specified at run-time
+/// \param  cCols number of columns in corner as specified at run-time
+///
+/// This function is mainly useful for corners where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a cRows should equal \a CRows unless
+/// \a CRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_topRightCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_topRightCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> topRightCorner(Index cRows, Index cCols)
+inline typename FixedBlockXpr<CRows,CCols>::Type topRightCorner(Index cRows, Index cCols)
 {
-  return Block<Derived, CRows, CCols>(derived(), 0, cols() - cCols, cRows, cCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), 0, cols() - cCols, cRows, cCols);
 }
 
-/** This is the const version of topRightCorner<int, int>(Index, Index).*/
+/// This is the const version of topRightCorner<int, int>(Index, Index).
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> topRightCorner(Index cRows, Index cCols) const
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type topRightCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived, CRows, CCols>(derived(), 0, cols() - cCols, cRows, cCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), 0, cols() - cCols, cRows, cCols);
 }
 
 
 
-/** \returns a dynamic-size expression of a top-left corner of *this.
-  *
-  * \param cRows the number of rows in the corner
-  * \param cCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_topLeftCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_topLeftCorner_int_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-inline Block<Derived> topLeftCorner(Index cRows, Index cCols)
+/// \returns a dynamic-size expression of a top-left corner of *this.
+///
+/// \param cRows the number of rows in the corner
+/// \param cCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_topLeftCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_topLeftCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr topLeftCorner(Index cRows, Index cCols)
 {
-  return Block<Derived>(derived(), 0, 0, cRows, cCols);
+  return BlockXpr(derived(), 0, 0, cRows, cCols);
 }
 
-/** This is the const version of topLeftCorner(Index, Index).*/
-inline const Block<const Derived> topLeftCorner(Index cRows, Index cCols) const
+/// This is the const version of topLeftCorner(Index, Index).
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr topLeftCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived>(derived(), 0, 0, cRows, cCols);
+  return ConstBlockXpr(derived(), 0, 0, cRows, cCols);
 }
 
-/** \returns an expression of a fixed-size top-left corner of *this.
-  *
-  * The template parameters CRows and CCols are the number of rows and columns in the corner.
-  *
-  * Example: \include MatrixBase_template_int_int_topLeftCorner.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_topLeftCorner.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns an expression of a fixed-size top-left corner of *this.
+///
+/// The template parameters CRows and CCols are the number of rows and columns in the corner.
+///
+/// Example: \include MatrixBase_template_int_int_topLeftCorner.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_topLeftCorner.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> topLeftCorner()
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<CRows,CCols>::Type topLeftCorner()
 {
-  return Block<Derived, CRows, CCols>(derived(), 0, 0);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), 0, 0);
 }
 
-/** This is the const version of topLeftCorner<int, int>().*/
+/// This is the const version of topLeftCorner<int, int>().
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> topLeftCorner() const
-{
-  return Block<const Derived, CRows, CCols>(derived(), 0, 0);
-}
-
-/** \returns an expression of a top-left corner of *this.
-  *
-  * \tparam CRows number of rows in corner as specified at compile-time
-  * \tparam CCols number of columns in corner as specified at compile-time
-  * \param  cRows number of rows in corner as specified at run-time
-  * \param  cCols number of columns in corner as specified at run-time
-  *
-  * This function is mainly useful for corners where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a cRows should equal \a CRows unless
-  * \a CRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_topLeftCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_topLeftCorner_int_int.out
-  *
-  * \sa class Block
-  */
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type topLeftCorner() const
+{
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), 0, 0);
+}
+
+/// \returns an expression of a top-left corner of *this.
+///
+/// \tparam CRows number of rows in corner as specified at compile-time
+/// \tparam CCols number of columns in corner as specified at compile-time
+/// \param  cRows number of rows in corner as specified at run-time
+/// \param  cCols number of columns in corner as specified at run-time
+///
+/// This function is mainly useful for corners where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a cRows should equal \a CRows unless
+/// \a CRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_topLeftCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_topLeftCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> topLeftCorner(Index cRows, Index cCols)
+inline typename FixedBlockXpr<CRows,CCols>::Type topLeftCorner(Index cRows, Index cCols)
 {
-  return Block<Derived, CRows, CCols>(derived(), 0, 0, cRows, cCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), 0, 0, cRows, cCols);
 }
 
-/** This is the const version of topLeftCorner<int, int>(Index, Index).*/
+/// This is the const version of topLeftCorner<int, int>(Index, Index).
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> topLeftCorner(Index cRows, Index cCols) const
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type topLeftCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived, CRows, CCols>(derived(), 0, 0, cRows, cCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), 0, 0, cRows, cCols);
 }
 
 
 
-/** \returns a dynamic-size expression of a bottom-right corner of *this.
-  *
-  * \param cRows the number of rows in the corner
-  * \param cCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_bottomRightCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_bottomRightCorner_int_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-inline Block<Derived> bottomRightCorner(Index cRows, Index cCols)
+/// \returns a dynamic-size expression of a bottom-right corner of *this.
+///
+/// \param cRows the number of rows in the corner
+/// \param cCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_bottomRightCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_bottomRightCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr bottomRightCorner(Index cRows, Index cCols)
 {
-  return Block<Derived>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
+  return BlockXpr(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
 }
 
-/** This is the const version of bottomRightCorner(Index, Index).*/
-inline const Block<const Derived> bottomRightCorner(Index cRows, Index cCols) const
+/// This is the const version of bottomRightCorner(Index, Index).
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr bottomRightCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
+  return ConstBlockXpr(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
 }
 
-/** \returns an expression of a fixed-size bottom-right corner of *this.
-  *
-  * The template parameters CRows and CCols are the number of rows and columns in the corner.
-  *
-  * Example: \include MatrixBase_template_int_int_bottomRightCorner.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_bottomRightCorner.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns an expression of a fixed-size bottom-right corner of *this.
+///
+/// The template parameters CRows and CCols are the number of rows and columns in the corner.
+///
+/// Example: \include MatrixBase_template_int_int_bottomRightCorner.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_bottomRightCorner.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> bottomRightCorner()
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<CRows,CCols>::Type bottomRightCorner()
 {
-  return Block<Derived, CRows, CCols>(derived(), rows() - CRows, cols() - CCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), rows() - CRows, cols() - CCols);
 }
 
-/** This is the const version of bottomRightCorner<int, int>().*/
+/// This is the const version of bottomRightCorner<int, int>().
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> bottomRightCorner() const
-{
-  return Block<const Derived, CRows, CCols>(derived(), rows() - CRows, cols() - CCols);
-}
-
-/** \returns an expression of a bottom-right corner of *this.
-  *
-  * \tparam CRows number of rows in corner as specified at compile-time
-  * \tparam CCols number of columns in corner as specified at compile-time
-  * \param  cRows number of rows in corner as specified at run-time
-  * \param  cCols number of columns in corner as specified at run-time
-  *
-  * This function is mainly useful for corners where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a cRows should equal \a CRows unless
-  * \a CRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_bottomRightCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_bottomRightCorner_int_int.out
-  *
-  * \sa class Block
-  */
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type bottomRightCorner() const
+{
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), rows() - CRows, cols() - CCols);
+}
+
+/// \returns an expression of a bottom-right corner of *this.
+///
+/// \tparam CRows number of rows in corner as specified at compile-time
+/// \tparam CCols number of columns in corner as specified at compile-time
+/// \param  cRows number of rows in corner as specified at run-time
+/// \param  cCols number of columns in corner as specified at run-time
+///
+/// This function is mainly useful for corners where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a cRows should equal \a CRows unless
+/// \a CRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_bottomRightCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_bottomRightCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> bottomRightCorner(Index cRows, Index cCols)
+inline typename FixedBlockXpr<CRows,CCols>::Type bottomRightCorner(Index cRows, Index cCols)
 {
-  return Block<Derived, CRows, CCols>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
 }
 
-/** This is the const version of bottomRightCorner<int, int>(Index, Index).*/
+/// This is the const version of bottomRightCorner<int, int>(Index, Index).
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> bottomRightCorner(Index cRows, Index cCols) const
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type bottomRightCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived, CRows, CCols>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
 }
 
 
 
-/** \returns a dynamic-size expression of a bottom-left corner of *this.
-  *
-  * \param cRows the number of rows in the corner
-  * \param cCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_bottomLeftCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_bottomLeftCorner_int_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-inline Block<Derived> bottomLeftCorner(Index cRows, Index cCols)
+/// \returns a dynamic-size expression of a bottom-left corner of *this.
+///
+/// \param cRows the number of rows in the corner
+/// \param cCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_bottomLeftCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_bottomLeftCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr bottomLeftCorner(Index cRows, Index cCols)
 {
-  return Block<Derived>(derived(), rows() - cRows, 0, cRows, cCols);
+  return BlockXpr(derived(), rows() - cRows, 0, cRows, cCols);
 }
 
-/** This is the const version of bottomLeftCorner(Index, Index).*/
-inline const Block<const Derived> bottomLeftCorner(Index cRows, Index cCols) const
+/// This is the const version of bottomLeftCorner(Index, Index).
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr bottomLeftCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived>(derived(), rows() - cRows, 0, cRows, cCols);
+  return ConstBlockXpr(derived(), rows() - cRows, 0, cRows, cCols);
 }
 
-/** \returns an expression of a fixed-size bottom-left corner of *this.
-  *
-  * The template parameters CRows and CCols are the number of rows and columns in the corner.
-  *
-  * Example: \include MatrixBase_template_int_int_bottomLeftCorner.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_bottomLeftCorner.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns an expression of a fixed-size bottom-left corner of *this.
+///
+/// The template parameters CRows and CCols are the number of rows and columns in the corner.
+///
+/// Example: \include MatrixBase_template_int_int_bottomLeftCorner.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_bottomLeftCorner.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> bottomLeftCorner()
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<CRows,CCols>::Type bottomLeftCorner()
 {
-  return Block<Derived, CRows, CCols>(derived(), rows() - CRows, 0);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), rows() - CRows, 0);
 }
 
-/** This is the const version of bottomLeftCorner<int, int>().*/
+/// This is the const version of bottomLeftCorner<int, int>().
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> bottomLeftCorner() const
-{
-  return Block<const Derived, CRows, CCols>(derived(), rows() - CRows, 0);
-}
-
-/** \returns an expression of a bottom-left corner of *this.
-  *
-  * \tparam CRows number of rows in corner as specified at compile-time
-  * \tparam CCols number of columns in corner as specified at compile-time
-  * \param  cRows number of rows in corner as specified at run-time
-  * \param  cCols number of columns in corner as specified at run-time
-  *
-  * This function is mainly useful for corners where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a cRows should equal \a CRows unless
-  * \a CRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_bottomLeftCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_bottomLeftCorner_int_int.out
-  *
-  * \sa class Block
-  */
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type bottomLeftCorner() const
+{
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), rows() - CRows, 0);
+}
+
+/// \returns an expression of a bottom-left corner of *this.
+///
+/// \tparam CRows number of rows in corner as specified at compile-time
+/// \tparam CCols number of columns in corner as specified at compile-time
+/// \param  cRows number of rows in corner as specified at run-time
+/// \param  cCols number of columns in corner as specified at run-time
+///
+/// This function is mainly useful for corners where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a cRows should equal \a CRows unless
+/// \a CRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_bottomLeftCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_bottomLeftCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> bottomLeftCorner(Index cRows, Index cCols)
+inline typename FixedBlockXpr<CRows,CCols>::Type bottomLeftCorner(Index cRows, Index cCols)
 {
-  return Block<Derived, CRows, CCols>(derived(), rows() - cRows, 0, cRows, cCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), rows() - cRows, 0, cRows, cCols);
 }
 
-/** This is the const version of bottomLeftCorner<int, int>(Index, Index).*/
+/// This is the const version of bottomLeftCorner<int, int>(Index, Index).
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> bottomLeftCorner(Index cRows, Index cCols) const
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type bottomLeftCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived, CRows, CCols>(derived(), rows() - cRows, 0, cRows, cCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), rows() - cRows, 0, cRows, cCols);
 }
 
 
 
-/** \returns a block consisting of the top rows of *this.
-  *
-  * \param n the number of rows in the block
-  *
-  * Example: \include MatrixBase_topRows_int.cpp
-  * Output: \verbinclude MatrixBase_topRows_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the top rows of *this.
+///
+/// \param n the number of rows in the block
+///
+/// Example: \include MatrixBase_topRows_int.cpp
+/// Output: \verbinclude MatrixBase_topRows_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline RowsBlockXpr topRows(Index n)
 {
   return RowsBlockXpr(derived(), 0, 0, n, cols());
 }
 
-/** This is the const version of topRows(Index).*/
+/// This is the const version of topRows(Index).
+EIGEN_DEVICE_FUNC
 inline ConstRowsBlockXpr topRows(Index n) const
 {
   return ConstRowsBlockXpr(derived(), 0, 0, n, cols());
 }
 
-/** \returns a block consisting of the top rows of *this.
-  *
-  * \tparam N the number of rows in the block as specified at compile-time
-  * \param n the number of rows in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_topRows.cpp
-  * Output: \verbinclude MatrixBase_template_int_topRows.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the top rows of *this.
+///
+/// \tparam N the number of rows in the block as specified at compile-time
+/// \param n the number of rows in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_topRows.cpp
+/// Output: \verbinclude MatrixBase_template_int_topRows.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NRowsBlockXpr<N>::Type topRows(Index n = N)
 {
   return typename NRowsBlockXpr<N>::Type(derived(), 0, 0, n, cols());
 }
 
-/** This is the const version of topRows<int>().*/
+/// This is the const version of topRows<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNRowsBlockXpr<N>::Type topRows(Index n = N) const
 {
   return typename ConstNRowsBlockXpr<N>::Type(derived(), 0, 0, n, cols());
@@ -416,47 +473,55 @@ inline typename ConstNRowsBlockXpr<N>::Type topRows(Index n = N) const
 
 
 
-/** \returns a block consisting of the bottom rows of *this.
-  *
-  * \param n the number of rows in the block
-  *
-  * Example: \include MatrixBase_bottomRows_int.cpp
-  * Output: \verbinclude MatrixBase_bottomRows_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the bottom rows of *this.
+///
+/// \param n the number of rows in the block
+///
+/// Example: \include MatrixBase_bottomRows_int.cpp
+/// Output: \verbinclude MatrixBase_bottomRows_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline RowsBlockXpr bottomRows(Index n)
 {
   return RowsBlockXpr(derived(), rows() - n, 0, n, cols());
 }
 
-/** This is the const version of bottomRows(Index).*/
+/// This is the const version of bottomRows(Index).
+EIGEN_DEVICE_FUNC
 inline ConstRowsBlockXpr bottomRows(Index n) const
 {
   return ConstRowsBlockXpr(derived(), rows() - n, 0, n, cols());
 }
 
-/** \returns a block consisting of the bottom rows of *this.
-  *
-  * \tparam N the number of rows in the block as specified at compile-time
-  * \param n the number of rows in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_bottomRows.cpp
-  * Output: \verbinclude MatrixBase_template_int_bottomRows.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the bottom rows of *this.
+///
+/// \tparam N the number of rows in the block as specified at compile-time
+/// \param n the number of rows in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_bottomRows.cpp
+/// Output: \verbinclude MatrixBase_template_int_bottomRows.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NRowsBlockXpr<N>::Type bottomRows(Index n = N)
 {
   return typename NRowsBlockXpr<N>::Type(derived(), rows() - n, 0, n, cols());
 }
 
-/** This is the const version of bottomRows<int>().*/
+/// This is the const version of bottomRows<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNRowsBlockXpr<N>::Type bottomRows(Index n = N) const
 {
   return typename ConstNRowsBlockXpr<N>::Type(derived(), rows() - n, 0, n, cols());
@@ -464,49 +529,57 @@ inline typename ConstNRowsBlockXpr<N>::Type bottomRows(Index n = N) const
 
 
 
-/** \returns a block consisting of a range of rows of *this.
-  *
-  * \param startRow the index of the first row in the block
-  * \param n the number of rows in the block
-  *
-  * Example: \include DenseBase_middleRows_int.cpp
-  * Output: \verbinclude DenseBase_middleRows_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of a range of rows of *this.
+///
+/// \param startRow the index of the first row in the block
+/// \param n the number of rows in the block
+///
+/// Example: \include DenseBase_middleRows_int.cpp
+/// Output: \verbinclude DenseBase_middleRows_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline RowsBlockXpr middleRows(Index startRow, Index n)
 {
   return RowsBlockXpr(derived(), startRow, 0, n, cols());
 }
 
-/** This is the const version of middleRows(Index,Index).*/
+/// This is the const version of middleRows(Index,Index).
+EIGEN_DEVICE_FUNC
 inline ConstRowsBlockXpr middleRows(Index startRow, Index n) const
 {
   return ConstRowsBlockXpr(derived(), startRow, 0, n, cols());
 }
 
-/** \returns a block consisting of a range of rows of *this.
-  *
-  * \tparam N the number of rows in the block as specified at compile-time
-  * \param startRow the index of the first row in the block
-  * \param n the number of rows in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include DenseBase_template_int_middleRows.cpp
-  * Output: \verbinclude DenseBase_template_int_middleRows.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of a range of rows of *this.
+///
+/// \tparam N the number of rows in the block as specified at compile-time
+/// \param startRow the index of the first row in the block
+/// \param n the number of rows in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include DenseBase_template_int_middleRows.cpp
+/// Output: \verbinclude DenseBase_template_int_middleRows.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NRowsBlockXpr<N>::Type middleRows(Index startRow, Index n = N)
 {
   return typename NRowsBlockXpr<N>::Type(derived(), startRow, 0, n, cols());
 }
 
-/** This is the const version of middleRows<int>().*/
+/// This is the const version of middleRows<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNRowsBlockXpr<N>::Type middleRows(Index startRow, Index n = N) const
 {
   return typename ConstNRowsBlockXpr<N>::Type(derived(), startRow, 0, n, cols());
@@ -514,47 +587,55 @@ inline typename ConstNRowsBlockXpr<N>::Type middleRows(Index startRow, Index n =
 
 
 
-/** \returns a block consisting of the left columns of *this.
-  *
-  * \param n the number of columns in the block
-  *
-  * Example: \include MatrixBase_leftCols_int.cpp
-  * Output: \verbinclude MatrixBase_leftCols_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the left columns of *this.
+///
+/// \param n the number of columns in the block
+///
+/// Example: \include MatrixBase_leftCols_int.cpp
+/// Output: \verbinclude MatrixBase_leftCols_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline ColsBlockXpr leftCols(Index n)
 {
   return ColsBlockXpr(derived(), 0, 0, rows(), n);
 }
 
-/** This is the const version of leftCols(Index).*/
+/// This is the const version of leftCols(Index).
+EIGEN_DEVICE_FUNC
 inline ConstColsBlockXpr leftCols(Index n) const
 {
   return ConstColsBlockXpr(derived(), 0, 0, rows(), n);
 }
 
-/** \returns a block consisting of the left columns of *this.
-  *
-  * \tparam N the number of columns in the block as specified at compile-time
-  * \param n the number of columns in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_leftCols.cpp
-  * Output: \verbinclude MatrixBase_template_int_leftCols.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the left columns of *this.
+///
+/// \tparam N the number of columns in the block as specified at compile-time
+/// \param n the number of columns in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_leftCols.cpp
+/// Output: \verbinclude MatrixBase_template_int_leftCols.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NColsBlockXpr<N>::Type leftCols(Index n = N)
 {
   return typename NColsBlockXpr<N>::Type(derived(), 0, 0, rows(), n);
 }
 
-/** This is the const version of leftCols<int>().*/
+/// This is the const version of leftCols<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNColsBlockXpr<N>::Type leftCols(Index n = N) const
 {
   return typename ConstNColsBlockXpr<N>::Type(derived(), 0, 0, rows(), n);
@@ -562,47 +643,55 @@ inline typename ConstNColsBlockXpr<N>::Type leftCols(Index n = N) const
 
 
 
-/** \returns a block consisting of the right columns of *this.
-  *
-  * \param n the number of columns in the block
-  *
-  * Example: \include MatrixBase_rightCols_int.cpp
-  * Output: \verbinclude MatrixBase_rightCols_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the right columns of *this.
+///
+/// \param n the number of columns in the block
+///
+/// Example: \include MatrixBase_rightCols_int.cpp
+/// Output: \verbinclude MatrixBase_rightCols_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline ColsBlockXpr rightCols(Index n)
 {
   return ColsBlockXpr(derived(), 0, cols() - n, rows(), n);
 }
 
-/** This is the const version of rightCols(Index).*/
+/// This is the const version of rightCols(Index).
+EIGEN_DEVICE_FUNC
 inline ConstColsBlockXpr rightCols(Index n) const
 {
   return ConstColsBlockXpr(derived(), 0, cols() - n, rows(), n);
 }
 
-/** \returns a block consisting of the right columns of *this.
-  *
-  * \tparam N the number of columns in the block as specified at compile-time
-  * \param n the number of columns in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_rightCols.cpp
-  * Output: \verbinclude MatrixBase_template_int_rightCols.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the right columns of *this.
+///
+/// \tparam N the number of columns in the block as specified at compile-time
+/// \param n the number of columns in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_rightCols.cpp
+/// Output: \verbinclude MatrixBase_template_int_rightCols.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NColsBlockXpr<N>::Type rightCols(Index n = N)
 {
   return typename NColsBlockXpr<N>::Type(derived(), 0, cols() - n, rows(), n);
 }
 
-/** This is the const version of rightCols<int>().*/
+/// This is the const version of rightCols<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNColsBlockXpr<N>::Type rightCols(Index n = N) const
 {
   return typename ConstNColsBlockXpr<N>::Type(derived(), 0, cols() - n, rows(), n);
@@ -610,49 +699,57 @@ inline typename ConstNColsBlockXpr<N>::Type rightCols(Index n = N) const
 
 
 
-/** \returns a block consisting of a range of columns of *this.
-  *
-  * \param startCol the index of the first column in the block
-  * \param numCols the number of columns in the block
-  *
-  * Example: \include DenseBase_middleCols_int.cpp
-  * Output: \verbinclude DenseBase_middleCols_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of a range of columns of *this.
+///
+/// \param startCol the index of the first column in the block
+/// \param numCols the number of columns in the block
+///
+/// Example: \include DenseBase_middleCols_int.cpp
+/// Output: \verbinclude DenseBase_middleCols_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline ColsBlockXpr middleCols(Index startCol, Index numCols)
 {
   return ColsBlockXpr(derived(), 0, startCol, rows(), numCols);
 }
 
-/** This is the const version of middleCols(Index,Index).*/
+/// This is the const version of middleCols(Index,Index).
+EIGEN_DEVICE_FUNC
 inline ConstColsBlockXpr middleCols(Index startCol, Index numCols) const
 {
   return ConstColsBlockXpr(derived(), 0, startCol, rows(), numCols);
 }
 
-/** \returns a block consisting of a range of columns of *this.
-  *
-  * \tparam N the number of columns in the block as specified at compile-time
-  * \param startCol the index of the first column in the block
-  * \param n the number of columns in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include DenseBase_template_int_middleCols.cpp
-  * Output: \verbinclude DenseBase_template_int_middleCols.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of a range of columns of *this.
+///
+/// \tparam N the number of columns in the block as specified at compile-time
+/// \param startCol the index of the first column in the block
+/// \param n the number of columns in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include DenseBase_template_int_middleCols.cpp
+/// Output: \verbinclude DenseBase_template_int_middleCols.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NColsBlockXpr<N>::Type middleCols(Index startCol, Index n = N)
 {
   return typename NColsBlockXpr<N>::Type(derived(), 0, startCol, rows(), n);
 }
 
-/** This is the const version of middleCols<int>().*/
+/// This is the const version of middleCols<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNColsBlockXpr<N>::Type middleCols(Index startCol, Index n = N) const
 {
   return typename ConstNColsBlockXpr<N>::Type(derived(), 0, startCol, rows(), n);
@@ -660,119 +757,134 @@ inline typename ConstNColsBlockXpr<N>::Type middleCols(Index startCol, Index n =
 
 
 
-/** \returns a fixed-size expression of a block in *this.
-  *
-  * The template parameters \a BlockRows and \a BlockCols are the number of
-  * rows and columns in the block.
-  *
-  * \param startRow the first row in the block
-  * \param startCol the first column in the block
-  *
-  * Example: \include MatrixBase_block_int_int.cpp
-  * Output: \verbinclude MatrixBase_block_int_int.out
-  *
-  * \note since block is a templated member, the keyword template has to be used
-  * if the matrix type is also a template parameter: \code m.template block<3,3>(1,1); \endcode
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-template<int BlockRows, int BlockCols>
-inline Block<Derived, BlockRows, BlockCols> block(Index startRow, Index startCol)
-{
-  return Block<Derived, BlockRows, BlockCols>(derived(), startRow, startCol);
-}
-
-/** This is the const version of block<>(Index, Index). */
-template<int BlockRows, int BlockCols>
-inline const Block<const Derived, BlockRows, BlockCols> block(Index startRow, Index startCol) const
-{
-  return Block<const Derived, BlockRows, BlockCols>(derived(), startRow, startCol);
-}
-
-/** \returns an expression of a block in *this.
-  *
-  * \tparam BlockRows number of rows in block as specified at compile-time
-  * \tparam BlockCols number of columns in block as specified at compile-time
-  * \param  startRow  the first row in the block
-  * \param  startCol  the first column in the block
-  * \param  blockRows number of rows in block as specified at run-time
-  * \param  blockCols number of columns in block as specified at run-time
-  *
-  * This function is mainly useful for blocks where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a blockRows should equal \a BlockRows unless
-  * \a BlockRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_block_int_int_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_block_int_int_int_int.cpp
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-template<int BlockRows, int BlockCols>
-inline Block<Derived, BlockRows, BlockCols> block(Index startRow, Index startCol, 
+/// \returns a fixed-size expression of a block in *this.
+///
+/// The template parameters \a NRows and \a NCols are the number of
+/// rows and columns in the block.
+///
+/// \param startRow the first row in the block
+/// \param startCol the first column in the block
+///
+/// Example: \include MatrixBase_block_int_int.cpp
+/// Output: \verbinclude MatrixBase_block_int_int.out
+///
+/// \note since block is a templated member, the keyword template has to be used
+/// if the matrix type is also a template parameter: \code m.template block<3,3>(1,1); \endcode
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+template<int NRows, int NCols>
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index startCol)
+{
+  return typename FixedBlockXpr<NRows,NCols>::Type(derived(), startRow, startCol);
+}
+
+/// This is the const version of block<>(Index, Index). */
+template<int NRows, int NCols>
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index startCol) const
+{
+  return typename ConstFixedBlockXpr<NRows,NCols>::Type(derived(), startRow, startCol);
+}
+
+/// \returns an expression of a block in *this.
+///
+/// \tparam NRows number of rows in block as specified at compile-time
+/// \tparam NCols number of columns in block as specified at compile-time
+/// \param  startRow  the first row in the block
+/// \param  startCol  the first column in the block
+/// \param  blockRows number of rows in block as specified at run-time
+/// \param  blockCols number of columns in block as specified at run-time
+///
+/// This function is mainly useful for blocks where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a blockRows should equal \a NRows unless
+/// \a NRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_block_int_int_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_block_int_int_int_int.cpp
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+template<int NRows, int NCols>
+inline typename FixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index startCol,
                                                   Index blockRows, Index blockCols)
 {
-  return Block<Derived, BlockRows, BlockCols>(derived(), startRow, startCol, blockRows, blockCols);
+  return typename FixedBlockXpr<NRows,NCols>::Type(derived(), startRow, startCol, blockRows, blockCols);
 }
 
-/** This is the const version of block<>(Index, Index, Index, Index). */
-template<int BlockRows, int BlockCols>
-inline const Block<const Derived, BlockRows, BlockCols> block(Index startRow, Index startCol,
+/// This is the const version of block<>(Index, Index, Index, Index).
+template<int NRows, int NCols>
+inline const typename ConstFixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index startCol,
                                                               Index blockRows, Index blockCols) const
 {
-  return Block<const Derived, BlockRows, BlockCols>(derived(), startRow, startCol, blockRows, blockCols);
+  return typename ConstFixedBlockXpr<NRows,NCols>::Type(derived(), startRow, startCol, blockRows, blockCols);
 }
 
-/** \returns an expression of the \a i-th column of *this. Note that the numbering starts at 0.
-  *
-  * Example: \include MatrixBase_col.cpp
-  * Output: \verbinclude MatrixBase_col.out
-  *
+/// \returns an expression of the \a i-th column of *this. Note that the numbering starts at 0.
+///
+/// Example: \include MatrixBase_col.cpp
+/// Output: \verbinclude MatrixBase_col.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+/**
   * \sa row(), class Block */
+EIGEN_DEVICE_FUNC
 inline ColXpr col(Index i)
 {
   return ColXpr(derived(), i);
 }
 
-/** This is the const version of col(). */
+/// This is the const version of col().
+EIGEN_DEVICE_FUNC
 inline ConstColXpr col(Index i) const
 {
   return ConstColXpr(derived(), i);
 }
 
-/** \returns an expression of the \a i-th row of *this. Note that the numbering starts at 0.
-  *
-  * Example: \include MatrixBase_row.cpp
-  * Output: \verbinclude MatrixBase_row.out
-  *
+/// \returns an expression of the \a i-th row of *this. Note that the numbering starts at 0.
+///
+/// Example: \include MatrixBase_row.cpp
+/// Output: \verbinclude MatrixBase_row.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+/**
   * \sa col(), class Block */
+EIGEN_DEVICE_FUNC
 inline RowXpr row(Index i)
 {
   return RowXpr(derived(), i);
 }
 
-/** This is the const version of row(). */
+/// This is the const version of row(). */
+EIGEN_DEVICE_FUNC
 inline ConstRowXpr row(Index i) const
 {
   return ConstRowXpr(derived(), i);
 }
 
-/** \returns a dynamic-size expression of a segment (i.e. a vector block) in *this.
-  *
-  * \only_for_vectors
-  *
-  * \param start the first coefficient in the segment
-  * \param n the number of coefficients in the segment
-  *
-  * Example: \include MatrixBase_segment_int_int.cpp
-  * Output: \verbinclude MatrixBase_segment_int_int.out
-  *
-  * \note Even though the returned expression has dynamic size, in the case
-  * when it is applied to a fixed-size vector, it inherits a fixed maximal size,
-  * which means that evaluating it does not cause a dynamic memory allocation.
-  *
-  * \sa class Block, segment(Index)
-  */
+/// \returns a dynamic-size expression of a segment (i.e. a vector block) in *this.
+///
+/// \only_for_vectors
+///
+/// \param start the first coefficient in the segment
+/// \param n the number of coefficients in the segment
+///
+/// Example: \include MatrixBase_segment_int_int.cpp
+/// Output: \verbinclude MatrixBase_segment_int_int.out
+///
+/// \note Even though the returned expression has dynamic size, in the case
+/// when it is applied to a fixed-size vector, it inherits a fixed maximal size,
+/// which means that evaluating it does not cause a dynamic memory allocation.
+///
+/// \sa class Block, segment(Index)
+///
+EIGEN_DEVICE_FUNC
 inline SegmentReturnType segment(Index start, Index n)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
@@ -780,154 +892,165 @@ inline SegmentReturnType segment(Index start, Index n)
 }
 
 
-/** This is the const version of segment(Index,Index).*/
+/// This is the const version of segment(Index,Index).
+EIGEN_DEVICE_FUNC
 inline ConstSegmentReturnType segment(Index start, Index n) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return ConstSegmentReturnType(derived(), start, n);
 }
 
-/** \returns a dynamic-size expression of the first coefficients of *this.
-  *
-  * \only_for_vectors
-  *
-  * \param n the number of coefficients in the segment
-  *
-  * Example: \include MatrixBase_start_int.cpp
-  * Output: \verbinclude MatrixBase_start_int.out
-  *
-  * \note Even though the returned expression has dynamic size, in the case
-  * when it is applied to a fixed-size vector, it inherits a fixed maximal size,
-  * which means that evaluating it does not cause a dynamic memory allocation.
-  *
-  * \sa class Block, block(Index,Index)
-  */
+/// \returns a dynamic-size expression of the first coefficients of *this.
+///
+/// \only_for_vectors
+///
+/// \param n the number of coefficients in the segment
+///
+/// Example: \include MatrixBase_start_int.cpp
+/// Output: \verbinclude MatrixBase_start_int.out
+///
+/// \note Even though the returned expression has dynamic size, in the case
+/// when it is applied to a fixed-size vector, it inherits a fixed maximal size,
+/// which means that evaluating it does not cause a dynamic memory allocation.
+///
+/// \sa class Block, block(Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline SegmentReturnType head(Index n)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return SegmentReturnType(derived(), 0, n);
 }
 
-/** This is the const version of head(Index).*/
+/// This is the const version of head(Index).
+EIGEN_DEVICE_FUNC
 inline ConstSegmentReturnType head(Index n) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return ConstSegmentReturnType(derived(), 0, n);
 }
 
-/** \returns a dynamic-size expression of the last coefficients of *this.
-  *
-  * \only_for_vectors
-  *
-  * \param n the number of coefficients in the segment
-  *
-  * Example: \include MatrixBase_end_int.cpp
-  * Output: \verbinclude MatrixBase_end_int.out
-  *
-  * \note Even though the returned expression has dynamic size, in the case
-  * when it is applied to a fixed-size vector, it inherits a fixed maximal size,
-  * which means that evaluating it does not cause a dynamic memory allocation.
-  *
-  * \sa class Block, block(Index,Index)
-  */
+/// \returns a dynamic-size expression of the last coefficients of *this.
+///
+/// \only_for_vectors
+///
+/// \param n the number of coefficients in the segment
+///
+/// Example: \include MatrixBase_end_int.cpp
+/// Output: \verbinclude MatrixBase_end_int.out
+///
+/// \note Even though the returned expression has dynamic size, in the case
+/// when it is applied to a fixed-size vector, it inherits a fixed maximal size,
+/// which means that evaluating it does not cause a dynamic memory allocation.
+///
+/// \sa class Block, block(Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline SegmentReturnType tail(Index n)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return SegmentReturnType(derived(), this->size() - n, n);
 }
 
-/** This is the const version of tail(Index).*/
+/// This is the const version of tail(Index).
+EIGEN_DEVICE_FUNC
 inline ConstSegmentReturnType tail(Index n) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return ConstSegmentReturnType(derived(), this->size() - n, n);
 }
 
-/** \returns a fixed-size expression of a segment (i.e. a vector block) in \c *this
-  *
-  * \only_for_vectors
-  *
-  * \tparam N the number of coefficients in the segment as specified at compile-time
-  * \param start the index of the first element in the segment
-  * \param n the number of coefficients in the segment as specified at compile-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_segment.cpp
-  * Output: \verbinclude MatrixBase_template_int_segment.out
-  *
-  * \sa class Block
-  */
+/// \returns a fixed-size expression of a segment (i.e. a vector block) in \c *this
+///
+/// \only_for_vectors
+///
+/// \tparam N the number of coefficients in the segment as specified at compile-time
+/// \param start the index of the first element in the segment
+/// \param n the number of coefficients in the segment as specified at compile-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_segment.cpp
+/// Output: \verbinclude MatrixBase_template_int_segment.out
+///
+/// \sa class Block
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename FixedSegmentReturnType<N>::Type segment(Index start, Index n = N)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename FixedSegmentReturnType<N>::Type(derived(), start, n);
 }
 
-/** This is the const version of segment<int>(Index).*/
+/// This is the const version of segment<int>(Index).
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstFixedSegmentReturnType<N>::Type segment(Index start, Index n = N) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename ConstFixedSegmentReturnType<N>::Type(derived(), start, n);
 }
 
-/** \returns a fixed-size expression of the first coefficients of *this.
-  *
-  * \only_for_vectors
-  *
-  * \tparam N the number of coefficients in the segment as specified at compile-time
-  * \param  n the number of coefficients in the segment as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_start.cpp
-  * Output: \verbinclude MatrixBase_template_int_start.out
-  *
-  * \sa class Block
-  */
+/// \returns a fixed-size expression of the first coefficients of *this.
+///
+/// \only_for_vectors
+///
+/// \tparam N the number of coefficients in the segment as specified at compile-time
+/// \param  n the number of coefficients in the segment as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_start.cpp
+/// Output: \verbinclude MatrixBase_template_int_start.out
+///
+/// \sa class Block
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename FixedSegmentReturnType<N>::Type head(Index n = N)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename FixedSegmentReturnType<N>::Type(derived(), 0, n);
 }
 
-/** This is the const version of head<int>().*/
+/// This is the const version of head<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstFixedSegmentReturnType<N>::Type head(Index n = N) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename ConstFixedSegmentReturnType<N>::Type(derived(), 0, n);
 }
 
-/** \returns a fixed-size expression of the last coefficients of *this.
-  *
-  * \only_for_vectors
-  *
-  * \tparam N the number of coefficients in the segment as specified at compile-time
-  * \param  n the number of coefficients in the segment as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_end.cpp
-  * Output: \verbinclude MatrixBase_template_int_end.out
-  *
-  * \sa class Block
-  */
+/// \returns a fixed-size expression of the last coefficients of *this.
+///
+/// \only_for_vectors
+///
+/// \tparam N the number of coefficients in the segment as specified at compile-time
+/// \param  n the number of coefficients in the segment as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_end.cpp
+/// Output: \verbinclude MatrixBase_template_int_end.out
+///
+/// \sa class Block
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename FixedSegmentReturnType<N>::Type tail(Index n = N)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename FixedSegmentReturnType<N>::Type(derived(), size() - n);
 }
 
-/** This is the const version of tail<int>.*/
+/// This is the const version of tail<int>.
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstFixedSegmentReturnType<N>::Type tail(Index n = N) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
diff --git a/extern/Eigen3/Eigen/src/plugins/CommonCwiseBinaryOps.h b/extern/Eigen3/Eigen/src/plugins/CommonCwiseBinaryOps.h
index 688d2244088..8b6730ede02 100644
--- a/extern/Eigen3/Eigen/src/plugins/CommonCwiseBinaryOps.h
+++ b/extern/Eigen3/Eigen/src/plugins/CommonCwiseBinaryOps.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -16,7 +16,7 @@
   *
   * \sa class CwiseBinaryOp, operator-=()
   */
-EIGEN_MAKE_CWISE_BINARY_OP(operator-,internal::scalar_difference_op)
+EIGEN_MAKE_CWISE_BINARY_OP(operator-,difference)
 
 /** \returns an expression of the sum of \c *this and \a other
   *
@@ -24,7 +24,7 @@ EIGEN_MAKE_CWISE_BINARY_OP(operator-,internal::scalar_difference_op)
   *
   * \sa class CwiseBinaryOp, operator+=()
   */
-EIGEN_MAKE_CWISE_BINARY_OP(operator+,internal::scalar_sum_op)
+EIGEN_MAKE_CWISE_BINARY_OP(operator+,sum)
 
 /** \returns an expression of a custom coefficient-wise operator \a func of *this and \a other
   *
@@ -38,9 +38,78 @@ EIGEN_MAKE_CWISE_BINARY_OP(operator+,internal::scalar_sum_op)
   * \sa class CwiseBinaryOp, operator+(), operator-(), cwiseProduct()
   */
 template<typename CustomBinaryOp, typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE const CwiseBinaryOp<CustomBinaryOp, const Derived, const OtherDerived>
 binaryExpr(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other, const CustomBinaryOp& func = CustomBinaryOp()) const
 {
   return CwiseBinaryOp<CustomBinaryOp, const Derived, const OtherDerived>(derived(), other.derived(), func);
 }
 
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP(operator*,product)
+#else
+/** \returns an expression of \c *this scaled by the scalar factor \a scalar
+  *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  */
+template<typename T>
+const CwiseBinaryOp<internal::scalar_product_op<Scalar,T>,Derived,Constant<T> > operator*(const T& scalar) const;
+/** \returns an expression of \a expr scaled by the scalar factor \a scalar
+  *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  */
+template<typename T> friend
+const CwiseBinaryOp<internal::scalar_product_op<T,Scalar>,Constant<T>,Derived> operator*(const T& scalar, const StorageBaseType& expr);
+#endif
+
+
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(operator/,quotient)
+#else
+/** \returns an expression of \c *this divided by the scalar value \a scalar
+  *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  */
+template<typename T>
+const CwiseBinaryOp<internal::scalar_quotient_op<Scalar,T>,Derived,Constant<T> > operator/(const T& scalar) const;
+#endif
+
+/** \returns an expression of the coefficient-wise boolean \b and operator of \c *this and \a other
+  *
+  * \warning this operator is for expression of bool only.
+  *
+  * Example: \include Cwise_boolean_and.cpp
+  * Output: \verbinclude Cwise_boolean_and.out
+  *
+  * \sa operator||(), select()
+  */
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
+inline const CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>
+operator&&(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
+{
+  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
+                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
+  return CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>(derived(),other.derived());
+}
+
+/** \returns an expression of the coefficient-wise boolean \b or operator of \c *this and \a other
+  *
+  * \warning this operator is for expression of bool only.
+  *
+  * Example: \include Cwise_boolean_or.cpp
+  * Output: \verbinclude Cwise_boolean_or.out
+  *
+  * \sa operator&&(), select()
+  */
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
+inline const CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>
+operator||(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
+{
+  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
+                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
+  return CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>(derived(),other.derived());
+}
diff --git a/extern/Eigen3/Eigen/src/plugins/CommonCwiseUnaryOps.h b/extern/Eigen3/Eigen/src/plugins/CommonCwiseUnaryOps.h
index 08e931aaddd..89f4faaac6b 100644
--- a/extern/Eigen3/Eigen/src/plugins/CommonCwiseUnaryOps.h
+++ b/extern/Eigen3/Eigen/src/plugins/CommonCwiseUnaryOps.h
@@ -12,10 +12,6 @@
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 
-/** \internal Represents a scalar multiple of an expression */
-typedef CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const Derived> ScalarMultipleReturnType;
-/** \internal Represents a quotient of an expression by a scalar*/
-typedef CwiseUnaryOp<internal::scalar_quotient1_op<Scalar>, const Derived> ScalarQuotient1ReturnType;
 /** \internal the return type of conjugate() */
 typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
                     const CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, const Derived>,
@@ -36,137 +32,132 @@ typedef CwiseUnaryOp<internal::scalar_imag_op<Scalar>, const Derived> ImagReturn
 /** \internal the return type of imag() */
 typedef CwiseUnaryView<internal::scalar_imag_ref_op<Scalar>, Derived> NonConstImagReturnType;
 
-#endif // not EIGEN_PARSED_BY_DOXYGEN
-
-/** \returns an expression of the opposite of \c *this
-  */
-inline const CwiseUnaryOp<internal::scalar_opposite_op<typename internal::traits<Derived>::Scalar>, const Derived>
-operator-() const { return derived(); }
-
-
-/** \returns an expression of \c *this scaled by the scalar factor \a scalar */
-inline const ScalarMultipleReturnType
-operator*(const Scalar& scalar) const
-{
-  return CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const Derived>
-    (derived(), internal::scalar_multiple_op<Scalar>(scalar));
-}
+typedef CwiseUnaryOp<internal::scalar_opposite_op<Scalar>, const Derived> NegativeReturnType;
 
-#ifdef EIGEN_PARSED_BY_DOXYGEN
-const ScalarMultipleReturnType operator*(const RealScalar& scalar) const;
-#endif
-
-/** \returns an expression of \c *this divided by the scalar value \a scalar */
-inline const CwiseUnaryOp<internal::scalar_quotient1_op<typename internal::traits<Derived>::Scalar>, const Derived>
-operator/(const Scalar& scalar) const
-{
-  return CwiseUnaryOp<internal::scalar_quotient1_op<Scalar>, const Derived>
-    (derived(), internal::scalar_quotient1_op<Scalar>(scalar));
-}
-
-/** Overloaded for efficient real matrix times complex scalar value */
-inline const CwiseUnaryOp<internal::scalar_multiple2_op<Scalar,std::complex<Scalar> >, const Derived>
-operator*(const std::complex<Scalar>& scalar) const
-{
-  return CwiseUnaryOp<internal::scalar_multiple2_op<Scalar,std::complex<Scalar> >, const Derived>
-    (*static_cast<const Derived*>(this), internal::scalar_multiple2_op<Scalar,std::complex<Scalar> >(scalar));
-}
+#endif // not EIGEN_PARSED_BY_DOXYGEN
 
-inline friend const ScalarMultipleReturnType
-operator*(const Scalar& scalar, const StorageBaseType& matrix)
-{ return matrix*scalar; }
-
-inline friend const CwiseUnaryOp<internal::scalar_multiple2_op<Scalar,std::complex<Scalar> >, const Derived>
-operator*(const std::complex<Scalar>& scalar, const StorageBaseType& matrix)
-{ return matrix*scalar; }
-
-/** \returns an expression of *this with the \a Scalar type casted to
-  * \a NewScalar.
-  *
-  * The template parameter \a NewScalar is the type we are casting the scalars to.
-  *
-  * \sa class CwiseUnaryOp
-  */
+/// \returns an expression of the opposite of \c *this
+///
+EIGEN_DOC_UNARY_ADDONS(operator-,opposite)
+///
+EIGEN_DEVICE_FUNC
+inline const NegativeReturnType
+operator-() const { return NegativeReturnType(derived()); }
+
+
+template<class NewType> struct CastXpr { typedef typename internal::cast_return_type<Derived,const CwiseUnaryOp<internal::scalar_cast_op<Scalar, NewType>, const Derived> >::type Type; };
+
+/// \returns an expression of \c *this with the \a Scalar type casted to
+/// \a NewScalar.
+///
+/// The template parameter \a NewScalar is the type we are casting the scalars to.
+///
+EIGEN_DOC_UNARY_ADDONS(cast,conversion function)
+///
+/// \sa class CwiseUnaryOp
+///
 template<typename NewType>
-typename internal::cast_return_type<Derived,const CwiseUnaryOp<internal::scalar_cast_op<typename internal::traits<Derived>::Scalar, NewType>, const Derived> >::type
+EIGEN_DEVICE_FUNC
+typename CastXpr<NewType>::Type
 cast() const
 {
-  return derived();
+  return typename CastXpr<NewType>::Type(derived());
 }
 
-/** \returns an expression of the complex conjugate of \c *this.
-  *
-  * \sa adjoint() */
+/// \returns an expression of the complex conjugate of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(conjugate,complex conjugate)
+///
+/// \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_conj">Math functions</a>, MatrixBase::adjoint()
+EIGEN_DEVICE_FUNC
 inline ConjugateReturnType
 conjugate() const
 {
   return ConjugateReturnType(derived());
 }
 
-/** \returns a read-only expression of the real part of \c *this.
-  *
-  * \sa imag() */
+/// \returns a read-only expression of the real part of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(real,real part function)
+///
+/// \sa imag()
+EIGEN_DEVICE_FUNC
 inline RealReturnType
-real() const { return derived(); }
-
-/** \returns an read-only expression of the imaginary part of \c *this.
-  *
-  * \sa real() */
+real() const { return RealReturnType(derived()); }
+
+/// \returns an read-only expression of the imaginary part of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(imag,imaginary part function)
+///
+/// \sa real()
+EIGEN_DEVICE_FUNC
 inline const ImagReturnType
-imag() const { return derived(); }
-
-/** \brief Apply a unary operator coefficient-wise
-  * \param[in]  func  Functor implementing the unary operator
-  * \tparam  CustomUnaryOp Type of \a func  
-  * \returns An expression of a custom coefficient-wise unary operator \a func of *this
-  *
-  * The function \c ptr_fun() from the C++ standard library can be used to make functors out of normal functions.
-  *
-  * Example:
-  * \include class_CwiseUnaryOp_ptrfun.cpp
-  * Output: \verbinclude class_CwiseUnaryOp_ptrfun.out
-  *
-  * Genuine functors allow for more possibilities, for instance it may contain a state.
-  *
-  * Example:
-  * \include class_CwiseUnaryOp.cpp
-  * Output: \verbinclude class_CwiseUnaryOp.out
-  *
-  * \sa class CwiseUnaryOp, class CwiseBinaryOp
-  */
+imag() const { return ImagReturnType(derived()); }
+
+/// \brief Apply a unary operator coefficient-wise
+/// \param[in]  func  Functor implementing the unary operator
+/// \tparam  CustomUnaryOp Type of \a func
+/// \returns An expression of a custom coefficient-wise unary operator \a func of *this
+///
+/// The function \c ptr_fun() from the C++ standard library can be used to make functors out of normal functions.
+///
+/// Example:
+/// \include class_CwiseUnaryOp_ptrfun.cpp
+/// Output: \verbinclude class_CwiseUnaryOp_ptrfun.out
+///
+/// Genuine functors allow for more possibilities, for instance it may contain a state.
+///
+/// Example:
+/// \include class_CwiseUnaryOp.cpp
+/// Output: \verbinclude class_CwiseUnaryOp.out
+///
+EIGEN_DOC_UNARY_ADDONS(unaryExpr,unary function)
+///
+/// \sa unaryViewExpr, binaryExpr, class CwiseUnaryOp
+///
 template<typename CustomUnaryOp>
+EIGEN_DEVICE_FUNC
 inline const CwiseUnaryOp<CustomUnaryOp, const Derived>
 unaryExpr(const CustomUnaryOp& func = CustomUnaryOp()) const
 {
   return CwiseUnaryOp<CustomUnaryOp, const Derived>(derived(), func);
 }
 
-/** \returns an expression of a custom coefficient-wise unary operator \a func of *this
-  *
-  * The template parameter \a CustomUnaryOp is the type of the functor
-  * of the custom unary operator.
-  *
-  * Example:
-  * \include class_CwiseUnaryOp.cpp
-  * Output: \verbinclude class_CwiseUnaryOp.out
-  *
-  * \sa class CwiseUnaryOp, class CwiseBinaryOp
-  */
+/// \returns an expression of a custom coefficient-wise unary operator \a func of *this
+///
+/// The template parameter \a CustomUnaryOp is the type of the functor
+/// of the custom unary operator.
+///
+/// Example:
+/// \include class_CwiseUnaryOp.cpp
+/// Output: \verbinclude class_CwiseUnaryOp.out
+///
+EIGEN_DOC_UNARY_ADDONS(unaryViewExpr,unary function)
+///
+/// \sa unaryExpr, binaryExpr class CwiseUnaryOp
+///
 template<typename CustomViewOp>
+EIGEN_DEVICE_FUNC
 inline const CwiseUnaryView<CustomViewOp, const Derived>
 unaryViewExpr(const CustomViewOp& func = CustomViewOp()) const
 {
   return CwiseUnaryView<CustomViewOp, const Derived>(derived(), func);
 }
 
-/** \returns a non const expression of the real part of \c *this.
-  *
-  * \sa imag() */
+/// \returns a non const expression of the real part of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(real,real part function)
+///
+/// \sa imag()
+EIGEN_DEVICE_FUNC
 inline NonConstRealReturnType
-real() { return derived(); }
-
-/** \returns a non const expression of the imaginary part of \c *this.
-  *
-  * \sa real() */
+real() { return NonConstRealReturnType(derived()); }
+
+/// \returns a non const expression of the imaginary part of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(imag,imaginary part function)
+///
+/// \sa real()
+EIGEN_DEVICE_FUNC
 inline NonConstImagReturnType
-imag() { return derived(); }
+imag() { return NonConstImagReturnType(derived()); }
diff --git a/extern/Eigen3/Eigen/src/plugins/MatrixCwiseBinaryOps.h b/extern/Eigen3/Eigen/src/plugins/MatrixCwiseBinaryOps.h
index c4a042b7027..f1084abefbc 100644
--- a/extern/Eigen3/Eigen/src/plugins/MatrixCwiseBinaryOps.h
+++ b/extern/Eigen3/Eigen/src/plugins/MatrixCwiseBinaryOps.h
@@ -18,10 +18,11 @@
   * \sa class CwiseBinaryOp, cwiseAbs2
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_PRODUCT_RETURN_TYPE(Derived,OtherDerived)
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,product)
 cwiseProduct(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return EIGEN_CWISE_PRODUCT_RETURN_TYPE(Derived,OtherDerived)(derived(), other.derived());
+  return EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,product)(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient-wise == operator of *this and \a other
@@ -37,6 +38,7 @@ cwiseProduct(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
   * \sa cwiseNotEqual(), isApprox(), isMuchSmallerThan()
   */
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 inline const CwiseBinaryOp<std::equal_to<Scalar>, const Derived, const OtherDerived>
 cwiseEqual(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
@@ -56,6 +58,7 @@ cwiseEqual(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
   * \sa cwiseEqual(), isApprox(), isMuchSmallerThan()
   */
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 inline const CwiseBinaryOp<std::not_equal_to<Scalar>, const Derived, const OtherDerived>
 cwiseNotEqual(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
@@ -70,17 +73,19 @@ cwiseNotEqual(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
   * \sa class CwiseBinaryOp, max()
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar>, const Derived, const OtherDerived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar,Scalar>, const Derived, const OtherDerived>
 cwiseMin(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return CwiseBinaryOp<internal::scalar_min_op<Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
+  return CwiseBinaryOp<internal::scalar_min_op<Scalar,Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient-wise min of *this and scalar \a other
   *
   * \sa class CwiseBinaryOp, min()
   */
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar>, const Derived, const ConstantReturnType>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar,Scalar>, const Derived, const ConstantReturnType>
 cwiseMin(const Scalar &other) const
 {
   return cwiseMin(Derived::Constant(rows(), cols(), other));
@@ -94,17 +99,19 @@ cwiseMin(const Scalar &other) const
   * \sa class CwiseBinaryOp, min()
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar>, const Derived, const OtherDerived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar,Scalar>, const Derived, const OtherDerived>
 cwiseMax(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return CwiseBinaryOp<internal::scalar_max_op<Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
+  return CwiseBinaryOp<internal::scalar_max_op<Scalar,Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient-wise max of *this and scalar \a other
   *
   * \sa class CwiseBinaryOp, min()
   */
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar>, const Derived, const ConstantReturnType>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar,Scalar>, const Derived, const ConstantReturnType>
 cwiseMax(const Scalar &other) const
 {
   return cwiseMax(Derived::Constant(rows(), cols(), other));
@@ -119,13 +126,14 @@ cwiseMax(const Scalar &other) const
   * \sa class CwiseBinaryOp, cwiseProduct(), cwiseInverse()
   */
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_quotient_op<Scalar>, const Derived, const OtherDerived>
 cwiseQuotient(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
   return CwiseBinaryOp<internal::scalar_quotient_op<Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
 }
 
-typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar,internal::cmp_EQ>, const Derived, const ConstantReturnType> CwiseScalarEqualReturnType;
+typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar,Scalar,internal::cmp_EQ>, const Derived, const ConstantReturnType> CwiseScalarEqualReturnType;
 
 /** \returns an expression of the coefficient-wise == operator of \c *this and a scalar \a s
   *
@@ -136,8 +144,9 @@ typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar,internal::cmp_EQ>, const De
   *
   * \sa cwiseEqual(const MatrixBase<OtherDerived> &) const
   */
+EIGEN_DEVICE_FUNC
 inline const CwiseScalarEqualReturnType
 cwiseEqual(const Scalar& s) const
 {
-  return CwiseScalarEqualReturnType(derived(), Derived::Constant(rows(), cols(), s), internal::scalar_cmp_op<Scalar,internal::cmp_EQ>());
+  return CwiseScalarEqualReturnType(derived(), Derived::Constant(rows(), cols(), s), internal::scalar_cmp_op<Scalar,Scalar,internal::cmp_EQ>());
 }
diff --git a/extern/Eigen3/Eigen/src/plugins/MatrixCwiseUnaryOps.h b/extern/Eigen3/Eigen/src/plugins/MatrixCwiseUnaryOps.h
index 8de10935d55..b1be3d566c6 100644
--- a/extern/Eigen3/Eigen/src/plugins/MatrixCwiseUnaryOps.h
+++ b/extern/Eigen3/Eigen/src/plugins/MatrixCwiseUnaryOps.h
@@ -8,45 +8,78 @@
 // Public License v. 2.0. If a copy of the MPL was not distributed
 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
-// This file is a base class plugin containing matrix specifics coefficient wise functions.
-
-/** \returns an expression of the coefficient-wise absolute value of \c *this
-  *
-  * Example: \include MatrixBase_cwiseAbs.cpp
-  * Output: \verbinclude MatrixBase_cwiseAbs.out
-  *
-  * \sa cwiseAbs2()
-  */
-EIGEN_STRONG_INLINE const CwiseUnaryOp<internal::scalar_abs_op<Scalar>, const Derived>
-cwiseAbs() const { return derived(); }
-
-/** \returns an expression of the coefficient-wise squared absolute value of \c *this
-  *
-  * Example: \include MatrixBase_cwiseAbs2.cpp
-  * Output: \verbinclude MatrixBase_cwiseAbs2.out
-  *
-  * \sa cwiseAbs()
-  */
-EIGEN_STRONG_INLINE const CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const Derived>
-cwiseAbs2() const { return derived(); }
-
-/** \returns an expression of the coefficient-wise square root of *this.
-  *
-  * Example: \include MatrixBase_cwiseSqrt.cpp
-  * Output: \verbinclude MatrixBase_cwiseSqrt.out
-  *
-  * \sa cwisePow(), cwiseSquare()
-  */
-inline const CwiseUnaryOp<internal::scalar_sqrt_op<Scalar>, const Derived>
-cwiseSqrt() const { return derived(); }
-
-/** \returns an expression of the coefficient-wise inverse of *this.
-  *
-  * Example: \include MatrixBase_cwiseInverse.cpp
-  * Output: \verbinclude MatrixBase_cwiseInverse.out
-  *
-  * \sa cwiseProduct()
-  */
-inline const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const Derived>
-cwiseInverse() const { return derived(); }
+// This file is included into the body of the base classes supporting matrix specific coefficient-wise functions.
+// This include MatrixBase and SparseMatrixBase.
+
+
+typedef CwiseUnaryOp<internal::scalar_abs_op<Scalar>, const Derived> CwiseAbsReturnType;
+typedef CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const Derived> CwiseAbs2ReturnType;
+typedef CwiseUnaryOp<internal::scalar_sqrt_op<Scalar>, const Derived> CwiseSqrtReturnType;
+typedef CwiseUnaryOp<internal::scalar_sign_op<Scalar>, const Derived> CwiseSignReturnType;
+typedef CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const Derived> CwiseInverseReturnType;
+
+/// \returns an expression of the coefficient-wise absolute value of \c *this
+///
+/// Example: \include MatrixBase_cwiseAbs.cpp
+/// Output: \verbinclude MatrixBase_cwiseAbs.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseAbs,absolute value)
+///
+/// \sa cwiseAbs2()
+///
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseAbsReturnType
+cwiseAbs() const { return CwiseAbsReturnType(derived()); }
+
+/// \returns an expression of the coefficient-wise squared absolute value of \c *this
+///
+/// Example: \include MatrixBase_cwiseAbs2.cpp
+/// Output: \verbinclude MatrixBase_cwiseAbs2.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseAbs2,squared absolute value)
+///
+/// \sa cwiseAbs()
+///
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseAbs2ReturnType
+cwiseAbs2() const { return CwiseAbs2ReturnType(derived()); }
+
+/// \returns an expression of the coefficient-wise square root of *this.
+///
+/// Example: \include MatrixBase_cwiseSqrt.cpp
+/// Output: \verbinclude MatrixBase_cwiseSqrt.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseSqrt,square-root)
+///
+/// \sa cwisePow(), cwiseSquare()
+///
+EIGEN_DEVICE_FUNC
+inline const CwiseSqrtReturnType
+cwiseSqrt() const { return CwiseSqrtReturnType(derived()); }
+
+/// \returns an expression of the coefficient-wise signum of *this.
+///
+/// Example: \include MatrixBase_cwiseSign.cpp
+/// Output: \verbinclude MatrixBase_cwiseSign.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseSign,sign function)
+///
+EIGEN_DEVICE_FUNC
+inline const CwiseSignReturnType
+cwiseSign() const { return CwiseSignReturnType(derived()); }
+
+
+/// \returns an expression of the coefficient-wise inverse of *this.
+///
+/// Example: \include MatrixBase_cwiseInverse.cpp
+/// Output: \verbinclude MatrixBase_cwiseInverse.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseInverse,inverse)
+///
+/// \sa cwiseProduct()
+///
+EIGEN_DEVICE_FUNC
+inline const CwiseInverseReturnType
+cwiseInverse() const { return CwiseInverseReturnType(derived()); }
+
 
diff --git a/extern/Eigen3/eigen-update.sh b/extern/Eigen3/eigen-update.sh
index 9d22098c4fd..4ea6d56277d 100755
--- a/extern/Eigen3/eigen-update.sh
+++ b/extern/Eigen3/eigen-update.sh
@@ -17,7 +17,7 @@ if [ -d eigen ]
 then
     cd eigen
     # put here the version you want to use
-    hg up 3.2.7
+    hg up 3.3.7
     rm -f `find Eigen/ -type f -name "CMakeLists.txt"`
     cp -r Eigen ..
     cd ..
diff --git a/extern/cuew/include/cuew.h b/extern/cuew/include/cuew.h
index fa334678e54..0fa0f1291fa 100644
--- a/extern/cuew/include/cuew.h
+++ b/extern/cuew/include/cuew.h
@@ -27,7 +27,7 @@ extern "C" {
 #define CUEW_VERSION_MAJOR 2
 #define CUEW_VERSION_MINOR 0
 
-#define CUDA_VERSION 9010
+#define CUDA_VERSION 9020
 #define CU_IPC_HANDLE_SIZE 64
 #define CU_STREAM_LEGACY ((CUstream)0x1)
 #define CU_STREAM_PER_THREAD ((CUstream)0x2)
@@ -383,6 +383,10 @@ typedef enum CUdevice_attribute_enum {
   CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH = 95,
   CU_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH = 96,
   CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN = 97,
+  CU_DEVICE_ATTRIBUTE_CAN_FLUSH_REMOTE_WRITES = 98,
+  CU_DEVICE_ATTRIBUTE_HOST_REGISTER_SUPPORTED = 99,
+  CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES = 100,
+  CU_DEVICE_ATTRIBUTE_DIRECT_MANAGED_MEM_ACCESS_FROM_HOST = 101,
   CU_DEVICE_ATTRIBUTE_MAX,
 } CUdevice_attribute;
 
@@ -408,6 +412,7 @@ typedef enum CUpointer_attribute_enum {
   CU_POINTER_ATTRIBUTE_SYNC_MEMOPS = 6,
   CU_POINTER_ATTRIBUTE_BUFFER_ID = 7,
   CU_POINTER_ATTRIBUTE_IS_MANAGED = 8,
+  CU_POINTER_ATTRIBUTE_DEVICE_ORDINAL = 9,
 } CUpointer_attribute;
 
 typedef enum CUfunction_attribute_enum {
@@ -640,6 +645,7 @@ typedef enum CUdevice_P2PAttribute_enum {
   CU_DEVICE_P2P_ATTRIBUTE_PERFORMANCE_RANK = 0x01,
   CU_DEVICE_P2P_ATTRIBUTE_ACCESS_SUPPORTED = 0x02,
   CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED = 0x03,
+  CU_DEVICE_P2P_ATTRIBUTE_ARRAY_ACCESS_ACCESS_SUPPORTED = 0x04,
 } CUdevice_P2PAttribute;
 
 typedef void (CUDA_CB *CUstreamCallback)(CUstream hStream, CUresult status, void* userData);
@@ -888,6 +894,7 @@ typedef CUresult CUDAAPI tcuDriverGetVersion(int* driverVersion);
 typedef CUresult CUDAAPI tcuDeviceGet(CUdevice* device, int ordinal);
 typedef CUresult CUDAAPI tcuDeviceGetCount(int* count);
 typedef CUresult CUDAAPI tcuDeviceGetName(char* name, int len, CUdevice dev);
+typedef CUresult CUDAAPI tcuDeviceGetUuid(CUuuid* uuid, CUdevice dev);
 typedef CUresult CUDAAPI tcuDeviceTotalMem_v2(size_t* bytes, CUdevice dev);
 typedef CUresult CUDAAPI tcuDeviceGetAttribute(int* pi, CUdevice_attribute attrib, CUdevice dev);
 typedef CUresult CUDAAPI tcuDeviceGetProperties(CUdevprop* prop, CUdevice dev);
@@ -1005,6 +1012,7 @@ typedef CUresult CUDAAPI tcuStreamCreate(CUstream* phStream, unsigned int Flags)
 typedef CUresult CUDAAPI tcuStreamCreateWithPriority(CUstream* phStream, unsigned int flags, int priority);
 typedef CUresult CUDAAPI tcuStreamGetPriority(CUstream hStream, int* priority);
 typedef CUresult CUDAAPI tcuStreamGetFlags(CUstream hStream, unsigned int* flags);
+typedef CUresult CUDAAPI tcuStreamGetCtx(CUstream hStream, CUcontext* pctx);
 typedef CUresult CUDAAPI tcuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags);
 typedef CUresult CUDAAPI tcuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void* userData, unsigned int flags);
 typedef CUresult CUDAAPI tcuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags);
@@ -1127,6 +1135,7 @@ extern tcuDriverGetVersion *cuDriverGetVersion;
 extern tcuDeviceGet *cuDeviceGet;
 extern tcuDeviceGetCount *cuDeviceGetCount;
 extern tcuDeviceGetName *cuDeviceGetName;
+extern tcuDeviceGetUuid *cuDeviceGetUuid;
 extern tcuDeviceTotalMem_v2 *cuDeviceTotalMem_v2;
 extern tcuDeviceGetAttribute *cuDeviceGetAttribute;
 extern tcuDeviceGetProperties *cuDeviceGetProperties;
@@ -1244,6 +1253,7 @@ extern tcuStreamCreate *cuStreamCreate;
 extern tcuStreamCreateWithPriority *cuStreamCreateWithPriority;
 extern tcuStreamGetPriority *cuStreamGetPriority;
 extern tcuStreamGetFlags *cuStreamGetFlags;
+extern tcuStreamGetCtx *cuStreamGetCtx;
 extern tcuStreamWaitEvent *cuStreamWaitEvent;
 extern tcuStreamAddCallback *cuStreamAddCallback;
 extern tcuStreamAttachMemAsync *cuStreamAttachMemAsync;
diff --git a/extern/cuew/src/cuew.c b/extern/cuew/src/cuew.c
index a22f6fda570..1f386c4e5d3 100644
--- a/extern/cuew/src/cuew.c
+++ b/extern/cuew/src/cuew.c
@@ -77,6 +77,7 @@ tcuDriverGetVersion *cuDriverGetVersion;
 tcuDeviceGet *cuDeviceGet;
 tcuDeviceGetCount *cuDeviceGetCount;
 tcuDeviceGetName *cuDeviceGetName;
+tcuDeviceGetUuid *cuDeviceGetUuid;
 tcuDeviceTotalMem_v2 *cuDeviceTotalMem_v2;
 tcuDeviceGetAttribute *cuDeviceGetAttribute;
 tcuDeviceGetProperties *cuDeviceGetProperties;
@@ -194,6 +195,7 @@ tcuStreamCreate *cuStreamCreate;
 tcuStreamCreateWithPriority *cuStreamCreateWithPriority;
 tcuStreamGetPriority *cuStreamGetPriority;
 tcuStreamGetFlags *cuStreamGetFlags;
+tcuStreamGetCtx *cuStreamGetCtx;
 tcuStreamWaitEvent *cuStreamWaitEvent;
 tcuStreamAddCallback *cuStreamAddCallback;
 tcuStreamAttachMemAsync *cuStreamAttachMemAsync;
@@ -385,6 +387,7 @@ static int cuewCudaInit(void) {
   CUDA_LIBRARY_FIND(cuDeviceGet);
   CUDA_LIBRARY_FIND(cuDeviceGetCount);
   CUDA_LIBRARY_FIND(cuDeviceGetName);
+  CUDA_LIBRARY_FIND(cuDeviceGetUuid);
   CUDA_LIBRARY_FIND(cuDeviceTotalMem_v2);
   CUDA_LIBRARY_FIND(cuDeviceGetAttribute);
   CUDA_LIBRARY_FIND(cuDeviceGetProperties);
@@ -502,6 +505,7 @@ static int cuewCudaInit(void) {
   CUDA_LIBRARY_FIND(cuStreamCreateWithPriority);
   CUDA_LIBRARY_FIND(cuStreamGetPriority);
   CUDA_LIBRARY_FIND(cuStreamGetFlags);
+  CUDA_LIBRARY_FIND(cuStreamGetCtx);
   CUDA_LIBRARY_FIND(cuStreamWaitEvent);
   CUDA_LIBRARY_FIND(cuStreamAddCallback);
   CUDA_LIBRARY_FIND(cuStreamAttachMemAsync);
@@ -619,10 +623,11 @@ static int cuewNvrtcInit(void) {
   /* Library paths. */
 #ifdef _WIN32
   /* Expected in c:/windows/system or similar, no path needed. */
-  const char *nvrtc_paths[] = {"nvrtc64_80.dll",
-                               "nvrtc64_90.dll",
+  const char *nvrtc_paths[] = {"nvrtc64_101_0.dll",
+                               "nvrtc64_100_0.dll",
                                "nvrtc64_91.dll",
-                               "nvrtc64_10_0.dll",
+                               "nvrtc64_90.dll",
+                               "nvrtc64_80.dll",
                                NULL};
 #elif defined(__APPLE__)
   /* Default installation path. */
diff --git a/extern/glew/CMakeLists.txt b/extern/glew/CMakeLists.txt
index e0773d828ec..18ca88e6fec 100644
--- a/extern/glew/CMakeLists.txt
+++ b/extern/glew/CMakeLists.txt
@@ -18,6 +18,13 @@
 # All rights reserved.
 # ***** END GPL LICENSE BLOCK *****
 
+# avoid noisy warnings
+if(CMAKE_COMPILER_IS_GNUCC OR CMAKE_C_COMPILER_ID MATCHES "Clang")
+  add_c_flag(
+    "-Wno-strict-prototypes"
+  )
+endif()
+
 set(INC
   include
 )
diff --git a/extern/lzma/CMakeLists.txt b/extern/lzma/CMakeLists.txt
index 8bdddf0aced..d76b418fd74 100644
--- a/extern/lzma/CMakeLists.txt
+++ b/extern/lzma/CMakeLists.txt
@@ -18,6 +18,13 @@
 # All rights reserved.
 # ***** END GPL LICENSE BLOCK *****
 
+# avoid noisy warnings
+if(CMAKE_COMPILER_IS_GNUCC OR CMAKE_C_COMPILER_ID MATCHES "Clang")
+  add_c_flag(
+    "-Wno-self-assign"
+  )
+endif()
+
 set(INC
   .
 )
diff --git a/intern/cycles/blender/CMakeLists.txt b/intern/cycles/blender/CMakeLists.txt
index 7354b1e615e..9658c7b4a62 100644
--- a/intern/cycles/blender/CMakeLists.txt
+++ b/intern/cycles/blender/CMakeLists.txt
@@ -68,7 +68,7 @@ set(ADDON_FILES
 add_definitions(${GL_DEFINITIONS})
 
 if(WITH_CYCLES_DEVICE_OPENCL)
-    add_definitions(-DWITH_OPENCL)
+  add_definitions(-DWITH_OPENCL)
 endif()
 
 if(WITH_CYCLES_NETWORK)
diff --git a/intern/cycles/blender/blender_curves.cpp b/intern/cycles/blender/blender_curves.cpp
index 29a1408d85d..4dba8ffbe0e 100644
--- a/intern/cycles/blender/blender_curves.cpp
+++ b/intern/cycles/blender/blender_curves.cpp
@@ -656,7 +656,7 @@ static void ExportCurveSegments(Scene *scene, Mesh *mesh, ParticleCurveData *CDa
       }
 
       if (attr_random != NULL) {
-        attr_random->add(hash_int_01(num_curves));
+        attr_random->add(hash_uint2_to_float(num_curves, 0));
       }
 
       mesh->add_curve(num_keys, CData->psys_shader[sys]);
diff --git a/intern/cycles/blender/blender_object.cpp b/intern/cycles/blender/blender_object.cpp
index 7ccf8226e5b..b670922ac88 100644
--- a/intern/cycles/blender/blender_object.cpp
+++ b/intern/cycles/blender/blender_object.cpp
@@ -217,7 +217,7 @@ void BlenderSync::sync_light(BL::Object &b_parent,
     light->random_id = random_id;
   }
   else {
-    light->random_id = hash_int_2d(hash_string(b_ob.name().c_str()), 0);
+    light->random_id = hash_uint2(hash_string(b_ob.name().c_str()), 0);
   }
 
   if (light->type == LIGHT_AREA)
@@ -444,6 +444,7 @@ Object *BlenderSync::sync_object(BL::Depsgraph &b_depsgraph,
   if (object_updated || (object->mesh && object->mesh->need_update) || tfm != object->tfm) {
     object->name = b_ob.name().c_str();
     object->pass_id = b_ob.pass_index();
+    object->color = get_float3(b_ob.color());
     object->tfm = tfm;
     object->motion.clear();
 
@@ -490,7 +491,7 @@ Object *BlenderSync::sync_object(BL::Depsgraph &b_depsgraph,
     else {
       object->dupli_generated = make_float3(0.0f, 0.0f, 0.0f);
       object->dupli_uv = make_float2(0.0f, 0.0f);
-      object->random_id = hash_int_2d(hash_string(object->name.c_str()), 0);
+      object->random_id = hash_uint2(hash_string(object->name.c_str()), 0);
     }
 
     object->tag_update(scene);
diff --git a/intern/cycles/blender/blender_python.cpp b/intern/cycles/blender/blender_python.cpp
index 80c6fee9723..9753da71eb4 100644
--- a/intern/cycles/blender/blender_python.cpp
+++ b/intern/cycles/blender/blender_python.cpp
@@ -177,6 +177,8 @@ static PyObject *init_func(PyObject * /*self*/, PyObject *args)
 
   BlenderSession::headless = headless;
 
+  DebugFlags().running_inside_blender = true;
+
   VLOG(2) << "Debug flags initialized to:\n" << DebugFlags();
 
   Py_RETURN_NONE;
diff --git a/intern/cycles/blender/blender_session.cpp b/intern/cycles/blender/blender_session.cpp
index d38a97dc4ea..8923450c469 100644
--- a/intern/cycles/blender/blender_session.cpp
+++ b/intern/cycles/blender/blender_session.cpp
@@ -114,11 +114,6 @@ BlenderSession::~BlenderSession()
   free_session();
 }
 
-void BlenderSession::create()
-{
-  create_session();
-}
-
 void BlenderSession::create_session()
 {
   SessionParams session_params = BlenderSync::get_session_params(
@@ -199,8 +194,12 @@ void BlenderSession::reset_session(BL::BlendData &b_data, BL::Depsgraph &b_depsg
     height = render_resolution_y(b_render);
   }
 
-  if (session == NULL) {
-    create();
+  bool is_new_session = (session == NULL);
+  if (is_new_session) {
+    /* Initialize session and remember it was just created so not to
+     * re-create it below.
+     */
+    create_session();
   }
 
   if (b_v3d) {
@@ -219,8 +218,10 @@ void BlenderSession::reset_session(BL::BlendData &b_data, BL::Depsgraph &b_depsg
     /* if scene or session parameters changed, it's easier to simply re-create
      * them rather than trying to distinguish which settings need to be updated
      */
-    free_session();
-    create_session();
+    if (!is_new_session) {
+      free_session();
+      create_session();
+    }
     return;
   }
 
@@ -539,8 +540,8 @@ void BlenderSession::render(BL::Depsgraph &b_depsgraph_)
     /* Make sure all views have different noise patterns. - hardcoded value just to make it random
      */
     if (view_index != 0) {
-      scene->integrator->seed += hash_int_2d(scene->integrator->seed,
-                                             hash_int(view_index * 0xdeadbeef));
+      scene->integrator->seed += hash_uint2(scene->integrator->seed,
+                                            hash_uint2(view_index * 0xdeadbeef, 0));
       scene->integrator->tag_update(scene);
     }
 
@@ -1481,8 +1482,8 @@ void BlenderSession::update_resumable_tile_manager(int num_samples)
 
   /* Round after doing the multiplications with num_chunks and num_samples_per_chunk
    * to allow for many small chunks. */
-  int rounded_range_start_sample = (int)floor(range_start_sample + 0.5f);
-  int rounded_range_num_samples = max((int)floor(range_num_samples + 0.5f), 1);
+  int rounded_range_start_sample = (int)floorf(range_start_sample + 0.5f);
+  int rounded_range_num_samples = max((int)floorf(range_num_samples + 0.5f), 1);
 
   /* Make sure we don't overshoot. */
   if (rounded_range_start_sample + rounded_range_num_samples > num_samples) {
diff --git a/intern/cycles/blender/blender_session.h b/intern/cycles/blender/blender_session.h
index d57ebe76b82..7445fb53458 100644
--- a/intern/cycles/blender/blender_session.h
+++ b/intern/cycles/blender/blender_session.h
@@ -49,8 +49,6 @@ class BlenderSession {
 
   ~BlenderSession();
 
-  void create();
-
   /* session */
   void create_session();
   void free_session();
diff --git a/intern/cycles/blender/blender_shader.cpp b/intern/cycles/blender/blender_shader.cpp
index b95e88573fa..720f521c589 100644
--- a/intern/cycles/blender/blender_shader.cpp
+++ b/intern/cycles/blender/blender_shader.cpp
@@ -326,16 +326,16 @@ static ShaderNode *add_node(Scene *scene,
   }
   else if (b_node.is_a(&RNA_ShaderNodeMath)) {
     BL::ShaderNodeMath b_math_node(b_node);
-    MathNode *math = new MathNode();
-    math->type = (NodeMath)b_math_node.operation();
-    math->use_clamp = b_math_node.use_clamp();
-    node = math;
+    MathNode *math_node = new MathNode();
+    math_node->type = (NodeMathType)b_math_node.operation();
+    math_node->use_clamp = b_math_node.use_clamp();
+    node = math_node;
   }
   else if (b_node.is_a(&RNA_ShaderNodeVectorMath)) {
     BL::ShaderNodeVectorMath b_vector_math_node(b_node);
-    VectorMathNode *vmath = new VectorMathNode();
-    vmath->type = (NodeVectorMath)b_vector_math_node.operation();
-    node = vmath;
+    VectorMathNode *vector_math_node = new VectorMathNode();
+    vector_math_node->type = (NodeVectorMathType)b_vector_math_node.operation();
+    node = vector_math_node;
   }
   else if (b_node.is_a(&RNA_ShaderNodeVectorTransform)) {
     BL::ShaderNodeVectorTransform b_vector_transform_node(b_node);
@@ -607,6 +607,9 @@ static ShaderNode *add_node(Scene *scene,
   else if (b_node.is_a(&RNA_ShaderNodeHairInfo)) {
     node = new HairInfoNode();
   }
+  else if (b_node.is_a(&RNA_ShaderNodeVolumeInfo)) {
+    node = new VolumeInfoNode();
+  }
   else if (b_node.is_a(&RNA_ShaderNodeBump)) {
     BL::ShaderNodeBump b_bump_node(b_node);
     BumpNode *bump = new BumpNode();
@@ -844,6 +847,12 @@ static ShaderNode *add_node(Scene *scene,
     }
     node = ies;
   }
+  else if (b_node.is_a(&RNA_ShaderNodeTexWhiteNoise)) {
+    BL::ShaderNodeTexWhiteNoise b_tex_white_noise_node(b_node);
+    WhiteNoiseTextureNode *white_noise_node = new WhiteNoiseTextureNode();
+    white_noise_node->dimensions = b_tex_white_noise_node.dimensions();
+    node = white_noise_node;
+  }
   else if (b_node.is_a(&RNA_ShaderNodeNormalMap)) {
     BL::ShaderNodeNormalMap b_normal_map_node(b_node);
     NormalMapNode *nmap = new NormalMapNode();
diff --git a/intern/cycles/blender/blender_sync.cpp b/intern/cycles/blender/blender_sync.cpp
index bac571b02ce..aec21887088 100644
--- a/intern/cycles/blender/blender_sync.cpp
+++ b/intern/cycles/blender/blender_sync.cpp
@@ -255,13 +255,13 @@ void BlenderSync::sync_integrator()
 
   integrator->seed = get_int(cscene, "seed");
   if (get_boolean(cscene, "use_animated_seed")) {
-    integrator->seed = hash_int_2d(b_scene.frame_current(), get_int(cscene, "seed"));
+    integrator->seed = hash_uint2(b_scene.frame_current(), get_int(cscene, "seed"));
     if (b_scene.frame_subframe() != 0.0f) {
       /* TODO(sergey): Ideally should be some sort of hash_merge,
        * but this is good enough for now.
        */
-      integrator->seed += hash_int_2d((int)(b_scene.frame_subframe() * (float)INT_MAX),
-                                      get_int(cscene, "seed"));
+      integrator->seed += hash_uint2((int)(b_scene.frame_subframe() * (float)INT_MAX),
+                                     get_int(cscene, "seed"));
     }
   }
 
diff --git a/intern/cycles/blender/blender_util.h b/intern/cycles/blender/blender_util.h
index c9d1dc67e54..3625dd45ae2 100644
--- a/intern/cycles/blender/blender_util.h
+++ b/intern/cycles/blender/blender_util.h
@@ -594,7 +594,7 @@ template<typename K, typename T> class id_map {
 
   T *find(const BL::ID &id)
   {
-    return find(id.ptr.id.data);
+    return find(id.ptr.owner_id);
   }
 
   T *find(const K &key)
@@ -629,7 +629,7 @@ template<typename K, typename T> class id_map {
 
   bool sync(T **r_data, const BL::ID &id)
   {
-    return sync(r_data, id, id, id.ptr.id.data);
+    return sync(r_data, id, id, id.ptr.owner_id);
   }
 
   bool sync(T **r_data, const BL::ID &id, const BL::ID &parent, const K &key)
diff --git a/intern/cycles/bvh/bvh.cpp b/intern/cycles/bvh/bvh.cpp
index 53c66777928..b6a4aba74b5 100644
--- a/intern/cycles/bvh/bvh.cpp
+++ b/intern/cycles/bvh/bvh.cpp
@@ -71,7 +71,11 @@ BVHLayout BVHParams::best_bvh_layout(BVHLayout requested_layout, BVHLayoutMask s
   /* This is a mask of supported BVH layouts which are narrower than the
    * requested one.
    */
-  const BVHLayoutMask allowed_layouts_mask = (supported_layouts & (requested_layout_mask - 1));
+  BVHLayoutMask allowed_layouts_mask = (supported_layouts & (requested_layout_mask - 1));
+  /* If the requested layout is not supported, choose from the supported layouts instead. */
+  if (allowed_layouts_mask == 0) {
+    allowed_layouts_mask = supported_layouts;
+  }
   /* We get widest from allowed ones and convert mask to actual layout. */
   const BVHLayoutMask widest_allowed_layout_mask = __bsr(allowed_layouts_mask);
   return (BVHLayout)(1 << widest_allowed_layout_mask);
@@ -90,23 +94,27 @@ int BVHStackEntry::encodeIdx() const
 
 /* BVH */
 
-BVH::BVH(const BVHParams &params_, const vector<Object *> &objects_)
-    : params(params_), objects(objects_)
+BVH::BVH(const BVHParams &params_, const vector<Mesh *> &meshes_, const vector<Object *> &objects_)
+    : params(params_), meshes(meshes_), objects(objects_)
 {
 }
 
-BVH *BVH::create(const BVHParams &params, const vector<Object *> &objects)
+BVH *BVH::create(const BVHParams &params,
+                 const vector<Mesh *> &meshes,
+                 const vector<Object *> &objects)
 {
   switch (params.bvh_layout) {
     case BVH_LAYOUT_BVH2:
-      return new BVH2(params, objects);
+      return new BVH2(params, meshes, objects);
     case BVH_LAYOUT_BVH4:
-      return new BVH4(params, objects);
+      return new BVH4(params, meshes, objects);
     case BVH_LAYOUT_BVH8:
-      return new BVH8(params, objects);
+      return new BVH8(params, meshes, objects);
     case BVH_LAYOUT_EMBREE:
 #ifdef WITH_EMBREE
-      return new BVHEmbree(params, objects);
+      return new BVHEmbree(params, meshes, objects);
+#else
+      break;
 #endif
     case BVH_LAYOUT_NONE:
     case BVH_LAYOUT_ALL:
@@ -356,26 +364,17 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
   size_t pack_leaf_nodes_offset = leaf_nodes_size;
   size_t object_offset = 0;
 
-  map<Mesh *, int> mesh_map;
-
-  foreach (Object *ob, objects) {
-    Mesh *mesh = ob->mesh;
+  foreach (Mesh *mesh, meshes) {
     BVH *bvh = mesh->bvh;
 
-    if (mesh->need_build_bvh()) {
-      if (mesh_map.find(mesh) == mesh_map.end()) {
-        prim_index_size += bvh->pack.prim_index.size();
-        prim_tri_verts_size += bvh->pack.prim_tri_verts.size();
-        nodes_size += bvh->pack.nodes.size();
-        leaf_nodes_size += bvh->pack.leaf_nodes.size();
-
-        mesh_map[mesh] = 1;
-      }
+    if (mesh->need_build_bvh(params.bvh_layout)) {
+      prim_index_size += bvh->pack.prim_index.size();
+      prim_tri_verts_size += bvh->pack.prim_tri_verts.size();
+      nodes_size += bvh->pack.nodes.size();
+      leaf_nodes_size += bvh->pack.leaf_nodes.size();
     }
   }
 
-  mesh_map.clear();
-
   pack.prim_index.resize(prim_index_size);
   pack.prim_type.resize(prim_index_size);
   pack.prim_object.resize(prim_index_size);
@@ -400,6 +399,8 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
   int4 *pack_leaf_nodes = (pack.leaf_nodes.size()) ? &pack.leaf_nodes[0] : NULL;
   float2 *pack_prim_time = (pack.prim_time.size()) ? &pack.prim_time[0] : NULL;
 
+  map<Mesh *, int> mesh_map;
+
   /* merge */
   foreach (Object *ob, objects) {
     Mesh *mesh = ob->mesh;
@@ -407,7 +408,7 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
     /* We assume that if mesh doesn't need own BVH it was already included
      * into a top-level BVH and no packing here is needed.
      */
-    if (!mesh->need_build_bvh()) {
+    if (!mesh->need_build_bvh(params.bvh_layout)) {
       pack.object_node[object_offset++] = 0;
       continue;
     }
diff --git a/intern/cycles/bvh/bvh.h b/intern/cycles/bvh/bvh.h
index edce3ca6f2a..92082e4de86 100644
--- a/intern/cycles/bvh/bvh.h
+++ b/intern/cycles/bvh/bvh.h
@@ -26,11 +26,14 @@
 CCL_NAMESPACE_BEGIN
 
 class Stats;
+class Device;
+class DeviceScene;
 class BVHNode;
 struct BVHStackEntry;
 class BVHParams;
 class BoundBox;
 class LeafNode;
+class Mesh;
 class Object;
 class Progress;
 
@@ -81,18 +84,25 @@ class BVH {
  public:
   PackedBVH pack;
   BVHParams params;
+  vector<Mesh *> meshes;
   vector<Object *> objects;
 
-  static BVH *create(const BVHParams &params, const vector<Object *> &objects);
+  static BVH *create(const BVHParams &params,
+                     const vector<Mesh *> &meshes,
+                     const vector<Object *> &objects);
   virtual ~BVH()
   {
   }
 
   virtual void build(Progress &progress, Stats *stats = NULL);
+  virtual void copy_to_device(Progress & /*progress*/, DeviceScene * /*dscene*/)
+  {
+  }
+
   void refit(Progress &progress);
 
  protected:
-  BVH(const BVHParams &params, const vector<Object *> &objects);
+  BVH(const BVHParams &params, const vector<Mesh *> &meshes, const vector<Object *> &objects);
 
   /* Refit range of primitives. */
   void refit_primitives(int start, int end, BoundBox &bbox, uint &visibility);
diff --git a/intern/cycles/bvh/bvh2.cpp b/intern/cycles/bvh/bvh2.cpp
index f419d413ef6..b1a9148c297 100644
--- a/intern/cycles/bvh/bvh2.cpp
+++ b/intern/cycles/bvh/bvh2.cpp
@@ -25,7 +25,10 @@
 
 CCL_NAMESPACE_BEGIN
 
-BVH2::BVH2(const BVHParams &params_, const vector<Object *> &objects_) : BVH(params_, objects_)
+BVH2::BVH2(const BVHParams &params_,
+           const vector<Mesh *> &meshes_,
+           const vector<Object *> &objects_)
+    : BVH(params_, meshes_, objects_)
 {
 }
 
diff --git a/intern/cycles/bvh/bvh2.h b/intern/cycles/bvh/bvh2.h
index c6a4e6fa73a..a3eaff9cf65 100644
--- a/intern/cycles/bvh/bvh2.h
+++ b/intern/cycles/bvh/bvh2.h
@@ -46,7 +46,7 @@ class BVH2 : public BVH {
  protected:
   /* constructor */
   friend class BVH;
-  BVH2(const BVHParams &params, const vector<Object *> &objects);
+  BVH2(const BVHParams &params, const vector<Mesh *> &meshes, const vector<Object *> &objects);
 
   /* Building process. */
   virtual BVHNode *widen_children_nodes(const BVHNode *root) override;
diff --git a/intern/cycles/bvh/bvh4.cpp b/intern/cycles/bvh/bvh4.cpp
index 850bdf5b8b4..89b42ee1d21 100644
--- a/intern/cycles/bvh/bvh4.cpp
+++ b/intern/cycles/bvh/bvh4.cpp
@@ -31,7 +31,10 @@ CCL_NAMESPACE_BEGIN
  * life easier all over the place.
  */
 
-BVH4::BVH4(const BVHParams &params_, const vector<Object *> &objects_) : BVH(params_, objects_)
+BVH4::BVH4(const BVHParams &params_,
+           const vector<Mesh *> &meshes_,
+           const vector<Object *> &objects_)
+    : BVH(params_, meshes_, objects_)
 {
   params.bvh_layout = BVH_LAYOUT_BVH4;
 }
@@ -43,8 +46,7 @@ BVHNode *bvh_node_merge_children_recursively(const BVHNode *node)
   if (node->is_leaf()) {
     return new LeafNode(*reinterpret_cast<const LeafNode *>(node));
   }
-  /* Collect nodes of one layer deeper, allowing us to have more childrem in
-   * an inner layer. */
+  /* Collect nodes of one layer deeper, allowing us to have more children in an inner layer. */
   assert(node->num_children() <= 2);
   const BVHNode *children[4];
   const BVHNode *child0 = node->get_child(0);
diff --git a/intern/cycles/bvh/bvh4.h b/intern/cycles/bvh/bvh4.h
index 38b0961d3df..c44f2833c84 100644
--- a/intern/cycles/bvh/bvh4.h
+++ b/intern/cycles/bvh/bvh4.h
@@ -46,7 +46,7 @@ class BVH4 : public BVH {
  protected:
   /* constructor */
   friend class BVH;
-  BVH4(const BVHParams &params, const vector<Object *> &objects);
+  BVH4(const BVHParams &params, const vector<Mesh *> &meshes, const vector<Object *> &objects);
 
   /* Building process. */
   virtual BVHNode *widen_children_nodes(const BVHNode *root) override;
diff --git a/intern/cycles/bvh/bvh8.cpp b/intern/cycles/bvh/bvh8.cpp
index 10fd01dd8d0..d3516525f78 100644
--- a/intern/cycles/bvh/bvh8.cpp
+++ b/intern/cycles/bvh/bvh8.cpp
@@ -36,7 +36,10 @@
 
 CCL_NAMESPACE_BEGIN
 
-BVH8::BVH8(const BVHParams &params_, const vector<Object *> &objects_) : BVH(params_, objects_)
+BVH8::BVH8(const BVHParams &params_,
+           const vector<Mesh *> &meshes_,
+           const vector<Object *> &objects_)
+    : BVH(params_, meshes_, objects_)
 {
 }
 
diff --git a/intern/cycles/bvh/bvh8.h b/intern/cycles/bvh/bvh8.h
index 6292353c7d4..5f26fd423e1 100644
--- a/intern/cycles/bvh/bvh8.h
+++ b/intern/cycles/bvh/bvh8.h
@@ -57,7 +57,7 @@ class BVH8 : public BVH {
  protected:
   /* constructor */
   friend class BVH;
-  BVH8(const BVHParams &params, const vector<Object *> &objects);
+  BVH8(const BVHParams &params, const vector<Mesh *> &meshes, const vector<Object *> &objects);
 
   /* Building process. */
   virtual BVHNode *widen_children_nodes(const BVHNode *root) override;
diff --git a/intern/cycles/bvh/bvh_embree.cpp b/intern/cycles/bvh/bvh_embree.cpp
index b011bc63dbd..d12a0c137ea 100644
--- a/intern/cycles/bvh/bvh_embree.cpp
+++ b/intern/cycles/bvh/bvh_embree.cpp
@@ -285,8 +285,10 @@ RTCDevice BVHEmbree::rtc_shared_device = NULL;
 int BVHEmbree::rtc_shared_users = 0;
 thread_mutex BVHEmbree::rtc_shared_mutex;
 
-BVHEmbree::BVHEmbree(const BVHParams &params_, const vector<Object *> &objects_)
-    : BVH(params_, objects_),
+BVHEmbree::BVHEmbree(const BVHParams &params_,
+                     const vector<Mesh *> &meshes_,
+                     const vector<Object *> &objects_)
+    : BVH(params_, meshes_, objects_),
       scene(NULL),
       mem_used(0),
       top_level(NULL),
@@ -497,6 +499,11 @@ void BVHEmbree::build(Progress &progress, Stats *stats_)
   stats = NULL;
 }
 
+void BVHEmbree::copy_to_device(Progress & /*progress*/, DeviceScene *dscene)
+{
+  dscene->data.bvh.scene = scene;
+}
+
 BVHNode *BVHEmbree::widen_children_nodes(const BVHNode * /*root*/)
 {
   assert(!"Must not be called.");
@@ -840,7 +847,7 @@ void BVHEmbree::pack_nodes(const BVHNode *)
     Mesh *mesh = ob->mesh;
     BVH *bvh = mesh->bvh;
 
-    if (mesh->need_build_bvh()) {
+    if (mesh->need_build_bvh(BVH_LAYOUT_EMBREE)) {
       if (mesh_map.find(mesh) == mesh_map.end()) {
         prim_index_size += bvh->pack.prim_index.size();
         prim_tri_verts_size += bvh->pack.prim_tri_verts.size();
@@ -872,7 +879,7 @@ void BVHEmbree::pack_nodes(const BVHNode *)
     /* We assume that if mesh doesn't need own BVH it was already included
      * into a top-level BVH and no packing here is needed.
      */
-    if (!mesh->need_build_bvh()) {
+    if (!mesh->need_build_bvh(BVH_LAYOUT_EMBREE)) {
       pack.object_node[object_offset++] = prim_offset;
       continue;
     }
diff --git a/intern/cycles/bvh/bvh_embree.h b/intern/cycles/bvh/bvh_embree.h
index 60702713583..123e87dd9b0 100644
--- a/intern/cycles/bvh/bvh_embree.h
+++ b/intern/cycles/bvh/bvh_embree.h
@@ -36,6 +36,7 @@ class Mesh;
 class BVHEmbree : public BVH {
  public:
   virtual void build(Progress &progress, Stats *stats) override;
+  virtual void copy_to_device(Progress &progress, DeviceScene *dscene) override;
   virtual ~BVHEmbree();
   RTCScene scene;
   static void destroy(RTCScene);
@@ -45,7 +46,9 @@ class BVHEmbree : public BVH {
 
  protected:
   friend class BVH;
-  BVHEmbree(const BVHParams &params, const vector<Object *> &objects);
+  BVHEmbree(const BVHParams &params,
+            const vector<Mesh *> &meshes,
+            const vector<Object *> &objects);
 
   virtual void pack_nodes(const BVHNode *) override;
   virtual void refit_nodes() override;
diff --git a/intern/cycles/device/device_memory.cpp b/intern/cycles/device/device_memory.cpp
index 859535307f4..c8a71bf4b3b 100644
--- a/intern/cycles/device/device_memory.cpp
+++ b/intern/cycles/device/device_memory.cpp
@@ -44,6 +44,29 @@ device_memory::~device_memory()
 {
 }
 
+device_memory::device_memory(device_memory &&other)
+    : data_type(other.data_type),
+      data_elements(other.data_elements),
+      data_size(other.data_size),
+      device_size(other.device_size),
+      data_width(other.data_width),
+      data_height(other.data_height),
+      data_depth(other.data_depth),
+      type(other.type),
+      name(other.name),
+      interpolation(other.interpolation),
+      extension(other.extension),
+      device(other.device),
+      device_pointer(other.device_pointer),
+      host_pointer(other.host_pointer),
+      shared_pointer(other.shared_pointer)
+{
+  other.device_size = 0;
+  other.device_pointer = 0;
+  other.host_pointer = 0;
+  other.shared_pointer = 0;
+}
+
 void *device_memory::host_alloc(size_t size)
 {
   if (!size) {
diff --git a/intern/cycles/device/device_memory.h b/intern/cycles/device/device_memory.h
index f50184efba7..5b43ce8b0bc 100644
--- a/intern/cycles/device/device_memory.h
+++ b/intern/cycles/device/device_memory.h
@@ -229,8 +229,11 @@ class device_memory {
   device_memory(Device *device, const char *name, MemoryType type);
 
   /* No copying allowed. */
-  device_memory(const device_memory &);
-  device_memory &operator=(const device_memory &);
+  device_memory(const device_memory &) = delete;
+  device_memory &operator=(const device_memory &) = delete;
+
+  /* But moving is possible. */
+  device_memory(device_memory &&);
 
   /* Host allocation on the device. All host_pointer memory should be
    * allocated with these functions, for devices that support using
@@ -269,6 +272,11 @@ template<typename T> class device_only_memory : public device_memory {
     free();
   }
 
+  device_only_memory(device_only_memory &&other)
+      : device_memory(static_cast<device_memory &&>(other))
+  {
+  }
+
   void alloc_to_device(size_t num, bool shrink_to_fit = true)
   {
     size_t new_size = num;
@@ -327,6 +335,10 @@ template<typename T> class device_vector : public device_memory {
     free();
   }
 
+  device_vector(device_vector &&other) : device_memory(static_cast<device_memory &&>(other))
+  {
+  }
+
   /* Host memory allocation. */
   T *alloc(size_t width, size_t height = 0, size_t depth = 0)
   {
diff --git a/intern/cycles/device/opencl/opencl_split.cpp b/intern/cycles/device/opencl/opencl_split.cpp
index 79474fb0814..eea5cf2fb8a 100644
--- a/intern/cycles/device/opencl/opencl_split.cpp
+++ b/intern/cycles/device/opencl/opencl_split.cpp
@@ -676,7 +676,7 @@ OpenCLDevice::OpenCLDevice(DeviceInfo &info, Stats &stats, Profiler &profiler, b
   device_initialized = true;
 
   split_kernel = new OpenCLSplitKernel(this);
-  if (!background) {
+  if (use_preview_kernels) {
     load_preview_kernels();
   }
 }
diff --git a/intern/cycles/device/opencl/opencl_util.cpp b/intern/cycles/device/opencl/opencl_util.cpp
index dc9b4072841..3eeff31f8c2 100644
--- a/intern/cycles/device/opencl/opencl_util.cpp
+++ b/intern/cycles/device/opencl/opencl_util.cpp
@@ -619,15 +619,16 @@ void OpenCLDevice::OpenCLProgram::compile()
       debug_src = &clsrc;
     }
 
-    /* If binary kernel exists already, try use it. */
-    if (compile_separate(clbin)) {
+    if (DebugFlags().running_inside_blender && compile_separate(clbin)) {
       add_log(string("Built and loaded program from ") + clbin + ".", true);
       loaded = true;
     }
     else {
-      add_log(string("Separate-process building of ") + clbin +
-                  " failed, will fall back to regular building.",
-              true);
+      if (DebugFlags().running_inside_blender) {
+        add_log(string("Separate-process building of ") + clbin +
+                    " failed, will fall back to regular building.",
+                true);
+      }
 
       /* If does not exist or loading binary failed, compile kernel. */
       if (!compile_kernel(debug_src)) {
diff --git a/intern/cycles/graph/node.h b/intern/cycles/graph/node.h
index 226c49b387a..d35a1bb489c 100644
--- a/intern/cycles/graph/node.h
+++ b/intern/cycles/graph/node.h
@@ -33,7 +33,7 @@ struct Transform;
 
 struct Node {
   explicit Node(const NodeType *type, ustring name = ustring());
-  virtual ~Node();
+  virtual ~Node() = 0;
 
   /* set values */
   void set(const SocketType &input, bool value);
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index ab05d6ee1e9..48439a8b68f 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -221,6 +221,7 @@ set(SRC_SVM_HEADERS
   svm/svm_voronoi.h
   svm/svm_voxel.h
   svm/svm_wave.h
+  svm/svm_white_noise.h
 )
 
 set(SRC_GEOM_HEADERS
diff --git a/intern/cycles/kernel/bvh/bvh.h b/intern/cycles/kernel/bvh/bvh.h
index be0f05285e8..162b2fb5cdb 100644
--- a/intern/cycles/kernel/bvh/bvh.h
+++ b/intern/cycles/kernel/bvh/bvh.h
@@ -177,24 +177,23 @@ ccl_device_inline bool scene_intersect_valid(const Ray *ray)
   return isfinite_safe(ray->P.x) && isfinite_safe(ray->D.x);
 }
 
-/* Note: ray is passed by value to work around a possible CUDA compiler bug. */
 ccl_device_intersect bool scene_intersect(KernelGlobals *kg,
-                                          const Ray ray,
+                                          const Ray *ray,
                                           const uint visibility,
                                           Intersection *isect)
 {
   PROFILING_INIT(kg, PROFILING_INTERSECT);
 
-  if (!scene_intersect_valid(&ray)) {
+  if (!scene_intersect_valid(ray)) {
     return false;
   }
 #ifdef __EMBREE__
   if (kernel_data.bvh.scene) {
-    isect->t = ray.t;
+    isect->t = ray->t;
     CCLIntersectContext ctx(kg, CCLIntersectContext::RAY_REGULAR);
     IntersectContext rtc_ctx(&ctx);
     RTCRayHit ray_hit;
-    kernel_embree_setup_rayhit(ray, ray_hit, visibility);
+    kernel_embree_setup_rayhit(*ray, ray_hit, visibility);
     rtcIntersect1(kernel_data.bvh.scene, &rtc_ctx.context, &ray_hit);
     if (ray_hit.hit.geomID != RTC_INVALID_GEOMETRY_ID &&
         ray_hit.hit.primID != RTC_INVALID_GEOMETRY_ID) {
@@ -207,42 +206,43 @@ ccl_device_intersect bool scene_intersect(KernelGlobals *kg,
 #ifdef __OBJECT_MOTION__
   if (kernel_data.bvh.have_motion) {
 #  ifdef __HAIR__
-    if (kernel_data.bvh.have_curves)
-      return bvh_intersect_hair_motion(kg, &ray, isect, visibility);
+    if (kernel_data.bvh.have_curves) {
+      return bvh_intersect_hair_motion(kg, ray, isect, visibility);
+    }
 #  endif /* __HAIR__ */
 
-    return bvh_intersect_motion(kg, &ray, isect, visibility);
+    return bvh_intersect_motion(kg, ray, isect, visibility);
   }
 #endif /* __OBJECT_MOTION__ */
 
 #ifdef __HAIR__
-  if (kernel_data.bvh.have_curves)
-    return bvh_intersect_hair(kg, &ray, isect, visibility);
+  if (kernel_data.bvh.have_curves) {
+    return bvh_intersect_hair(kg, ray, isect, visibility);
+  }
 #endif /* __HAIR__ */
 
 #ifdef __KERNEL_CPU__
 
 #  ifdef __INSTANCING__
-  if (kernel_data.bvh.have_instancing)
-    return bvh_intersect_instancing(kg, &ray, isect, visibility);
+  if (kernel_data.bvh.have_instancing) {
+    return bvh_intersect_instancing(kg, ray, isect, visibility);
+  }
 #  endif /* __INSTANCING__ */
-
-  return bvh_intersect(kg, &ray, isect, visibility);
+  return bvh_intersect(kg, ray, isect, visibility);
 #else /* __KERNEL_CPU__ */
 
 #  ifdef __INSTANCING__
-  return bvh_intersect_instancing(kg, &ray, isect, visibility);
+  return bvh_intersect_instancing(kg, ray, isect, visibility);
 #  else
-  return bvh_intersect(kg, &ray, isect, visibility);
+  return bvh_intersect(kg, ray, isect, visibility);
 #  endif /* __INSTANCING__ */
 
 #endif /* __KERNEL_CPU__ */
 }
 
 #ifdef __BVH_LOCAL__
-/* Note: ray is passed by value to work around a possible CUDA compiler bug. */
 ccl_device_intersect bool scene_intersect_local(KernelGlobals *kg,
-                                                const Ray ray,
+                                                const Ray *ray,
                                                 LocalIntersection *local_isect,
                                                 int local_object,
                                                 uint *lcg_state,
@@ -250,7 +250,7 @@ ccl_device_intersect bool scene_intersect_local(KernelGlobals *kg,
 {
   PROFILING_INIT(kg, PROFILING_INTERSECT_LOCAL);
 
-  if (!scene_intersect_valid(&ray)) {
+  if (!scene_intersect_valid(ray)) {
     local_isect->num_hits = 0;
     return false;
   }
@@ -264,19 +264,19 @@ ccl_device_intersect bool scene_intersect_local(KernelGlobals *kg,
     ctx.sss_object_id = local_object;
     IntersectContext rtc_ctx(&ctx);
     RTCRay rtc_ray;
-    kernel_embree_setup_ray(ray, rtc_ray, PATH_RAY_ALL_VISIBILITY);
+    kernel_embree_setup_ray(*ray, rtc_ray, PATH_RAY_ALL_VISIBILITY);
 
     /* Get the Embree scene for this intersection. */
     RTCGeometry geom = rtcGetGeometry(kernel_data.bvh.scene, local_object * 2);
     if (geom) {
-      float3 P = ray.P;
-      float3 dir = ray.D;
-      float3 idir = ray.D;
+      float3 P = ray->P;
+      float3 dir = ray->D;
+      float3 idir = ray->D;
       const int object_flag = kernel_tex_fetch(__object_flag, local_object);
       if (!(object_flag & SD_OBJECT_TRANSFORM_APPLIED)) {
         Transform ob_itfm;
         rtc_ray.tfar = bvh_instance_motion_push(
-            kg, local_object, &ray, &P, &dir, &idir, ray.t, &ob_itfm);
+            kg, local_object, ray, &P, &dir, &idir, ray->t, &ob_itfm);
         /* bvh_instance_motion_push() returns the inverse transform but
          * it's not needed here. */
         (void)ob_itfm;
@@ -299,10 +299,10 @@ ccl_device_intersect bool scene_intersect_local(KernelGlobals *kg,
 #  endif /* __EMBREE__ */
 #  ifdef __OBJECT_MOTION__
   if (kernel_data.bvh.have_motion) {
-    return bvh_intersect_local_motion(kg, &ray, local_isect, local_object, lcg_state, max_hits);
+    return bvh_intersect_local_motion(kg, ray, local_isect, local_object, lcg_state, max_hits);
   }
 #  endif /* __OBJECT_MOTION__ */
-  return bvh_intersect_local(kg, &ray, local_isect, local_object, lcg_state, max_hits);
+  return bvh_intersect_local(kg, ray, local_isect, local_object, lcg_state, max_hits);
 }
 #endif
 
@@ -377,15 +377,18 @@ ccl_device_intersect bool scene_intersect_volume(KernelGlobals *kg,
   if (!scene_intersect_valid(ray)) {
     return false;
   }
+
 #  ifdef __OBJECT_MOTION__
   if (kernel_data.bvh.have_motion) {
     return bvh_intersect_volume_motion(kg, ray, isect, visibility);
   }
 #  endif /* __OBJECT_MOTION__ */
+
 #  ifdef __KERNEL_CPU__
 #    ifdef __INSTANCING__
-  if (kernel_data.bvh.have_instancing)
+  if (kernel_data.bvh.have_instancing) {
     return bvh_intersect_volume_instancing(kg, ray, isect, visibility);
+  }
 #    endif /* __INSTANCING__ */
   return bvh_intersect_volume(kg, ray, isect, visibility);
 #  else /* __KERNEL_CPU__ */
@@ -422,15 +425,18 @@ ccl_device_intersect uint scene_intersect_volume_all(KernelGlobals *kg,
     rtcOccluded1(kernel_data.bvh.scene, &rtc_ctx.context, &rtc_ray);
     return rtc_ray.tfar == -INFINITY;
   }
-#  endif
+#  endif /* __EMBREE__ */
+
 #  ifdef __OBJECT_MOTION__
   if (kernel_data.bvh.have_motion) {
     return bvh_intersect_volume_all_motion(kg, ray, isect, max_hits, visibility);
   }
 #  endif /* __OBJECT_MOTION__ */
+
 #  ifdef __INSTANCING__
-  if (kernel_data.bvh.have_instancing)
+  if (kernel_data.bvh.have_instancing) {
     return bvh_intersect_volume_all_instancing(kg, ray, isect, max_hits, visibility);
+  }
 #  endif /* __INSTANCING__ */
   return bvh_intersect_volume_all(kg, ray, isect, max_hits, visibility);
 }
diff --git a/intern/cycles/kernel/bvh/bvh_nodes.h b/intern/cycles/kernel/bvh/bvh_nodes.h
index a33bc73e25b..db598d1c7fa 100644
--- a/intern/cycles/kernel/bvh/bvh_nodes.h
+++ b/intern/cycles/kernel/bvh/bvh_nodes.h
@@ -39,7 +39,9 @@ ccl_device_forceinline int bvh_aligned_node_intersect(KernelGlobals *kg,
 {
 
   /* fetch node data */
+#  ifdef __VISIBILITY_FLAG__
   float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0);
+#  endif
   float4 node0 = kernel_tex_fetch(__bvh_nodes, node_addr + 1);
   float4 node1 = kernel_tex_fetch(__bvh_nodes, node_addr + 2);
   float4 node2 = kernel_tex_fetch(__bvh_nodes, node_addr + 3);
@@ -111,7 +113,9 @@ ccl_device_forceinline int bvh_unaligned_node_intersect(KernelGlobals *kg,
                                                         float dist[2])
 {
   int mask = 0;
+#  ifdef __VISIBILITY_FLAG__
   float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0);
+#  endif
   if (bvh_unaligned_node_intersect_child(kg, P, dir, t, node_addr, 0, &dist[0])) {
 #  ifdef __VISIBILITY_FLAG__
     if ((__float_as_uint(cnodes.x) & visibility))
diff --git a/intern/cycles/kernel/closure/bssrdf.h b/intern/cycles/kernel/closure/bssrdf.h
index 0da4d6bd060..4d88a822821 100644
--- a/intern/cycles/kernel/closure/bssrdf.h
+++ b/intern/cycles/kernel/closure/bssrdf.h
@@ -231,7 +231,7 @@ ccl_device float bssrdf_burley_eval(const float d, float r)
    * NOTES:
    * - Surface albedo is already included into sc->weight, no need to
    *   multiply by this term here.
-   * - This is normalized diffuse model, so the equation is mutliplied
+   * - This is normalized diffuse model, so the equation is multiplied
    *   by 2*pi, which also matches cdf().
    */
   float exp_r_3_d = expf(-r / (3.0f * d));
diff --git a/intern/cycles/kernel/filter/filter_transform.h b/intern/cycles/kernel/filter/filter_transform.h
index 585c4b33787..880a661214e 100644
--- a/intern/cycles/kernel/filter/filter_transform.h
+++ b/intern/cycles/kernel/filter/filter_transform.h
@@ -70,9 +70,9 @@ ccl_device void kernel_filter_construct_transform(const float *ccl_restrict buff
   filter_calculate_scale(feature_scale, use_time);
 
   /* === Generate the feature transformation. ===
-   * This transformation maps the num_features-dimentional feature space to a reduced feature
-   * (r-feature) space which generally has fewer dimensions. This mainly helps to prevent
-   * overfitting. */
+   * This transformation maps the num_features-dimensional feature space to a reduced feature
+   * (r-feature) space which generally has fewer dimensions.
+   * This mainly helps to prevent over-fitting. */
   float feature_matrix[DENOISE_FEATURES * DENOISE_FEATURES];
   math_matrix_zero(feature_matrix, num_features);
   FOR_PIXEL_WINDOW
@@ -85,7 +85,7 @@ ccl_device void kernel_filter_construct_transform(const float *ccl_restrict buff
 
   math_matrix_jacobi_eigendecomposition(feature_matrix, transform, num_features, 1);
   *rank = 0;
-  /* Prevent overfitting when a small window is used. */
+  /* Prevent over-fitting when a small window is used. */
   int max_rank = min(num_features, num_pixels / 3);
   if (pca_threshold < 0.0f) {
     float threshold_energy = 0.0f;
diff --git a/intern/cycles/kernel/geom/geom_object.h b/intern/cycles/kernel/geom/geom_object.h
index f410e6e27e2..af4e6fbd89b 100644
--- a/intern/cycles/kernel/geom/geom_object.h
+++ b/intern/cycles/kernel/geom/geom_object.h
@@ -227,6 +227,17 @@ ccl_device_inline float object_surface_area(KernelGlobals *kg, int object)
   return kernel_tex_fetch(__objects, object).surface_area;
 }
 
+/* Color of the object */
+
+ccl_device_inline float3 object_color(KernelGlobals *kg, int object)
+{
+  if (object == OBJECT_NONE)
+    return make_float3(0.0f, 0.0f, 0.0f);
+
+  const ccl_global KernelObject *kobject = &kernel_tex_fetch(__objects, object);
+  return make_float3(kobject->color[0], kobject->color[1], kobject->color[2]);
+}
+
 /* Pass ID number of object */
 
 ccl_device_inline float object_pass_id(KernelGlobals *kg, int object)
diff --git a/intern/cycles/kernel/kernel_accumulate.h b/intern/cycles/kernel/kernel_accumulate.h
index b9d723222a1..46a51f5a560 100644
--- a/intern/cycles/kernel/kernel_accumulate.h
+++ b/intern/cycles/kernel/kernel_accumulate.h
@@ -319,10 +319,12 @@ ccl_device_inline void path_radiance_accum_ao(PathRadiance *L,
                                               float3 bsdf,
                                               float3 ao)
 {
+#ifdef __PASSES__
   /* Store AO pass. */
   if (L->use_light_pass && state->bounce == 0) {
     L->ao += alpha * throughput * ao;
   }
+#endif
 
 #ifdef __SHADOW_TRICKS__
   /* For shadow catcher, accumulate ratio. */
diff --git a/intern/cycles/kernel/kernel_camera.h b/intern/cycles/kernel/kernel_camera.h
index 97cd3a3320c..62ce04ba48f 100644
--- a/intern/cycles/kernel/kernel_camera.h
+++ b/intern/cycles/kernel/kernel_camera.h
@@ -237,7 +237,9 @@ ccl_device void camera_sample_orthographic(KernelGlobals *kg,
 /* Panorama Camera */
 
 ccl_device_inline void camera_sample_panorama(ccl_constant KernelCamera *cam,
+#ifdef __CAMERA_MOTION__
                                               const ccl_global DecomposedTransform *cam_motion,
+#endif
                                               float raster_x,
                                               float raster_y,
                                               float lens_u,
@@ -413,8 +415,12 @@ ccl_device_inline void camera_sample(KernelGlobals *kg,
     camera_sample_orthographic(kg, raster_x, raster_y, lens_u, lens_v, ray);
   }
   else {
+#ifdef __CAMERA_MOTION__
     const ccl_global DecomposedTransform *cam_motion = kernel_tex_array(__camera_motion);
     camera_sample_panorama(&kernel_data.cam, cam_motion, raster_x, raster_y, lens_u, lens_v, ray);
+#else
+    camera_sample_panorama(&kernel_data.cam, raster_x, raster_y, lens_u, lens_v, ray);
+#endif
   }
 }
 
diff --git a/intern/cycles/kernel/kernel_compat_cuda.h b/intern/cycles/kernel/kernel_compat_cuda.h
index 469b81d120b..4f508d7cdaa 100644
--- a/intern/cycles/kernel/kernel_compat_cuda.h
+++ b/intern/cycles/kernel/kernel_compat_cuda.h
@@ -58,6 +58,7 @@ __device__ half __float2half(const float f)
 #  define ccl_device_forceinline __device__ __forceinline__
 #endif
 #define ccl_device_noinline __device__ __noinline__
+#define ccl_device_noinline_cpu ccl_device
 #define ccl_global
 #define ccl_static_constant __constant__
 #define ccl_constant const
@@ -72,6 +73,7 @@ __device__ half __float2half(const float f)
  */
 #define ccl_ref
 #define ccl_align(n) __align__(n)
+#define ccl_optional_struct_init
 
 #define ATTR_FALLTHROUGH
 
diff --git a/intern/cycles/kernel/kernel_compat_opencl.h b/intern/cycles/kernel/kernel_compat_opencl.h
index e040ea88d7c..552734cc361 100644
--- a/intern/cycles/kernel/kernel_compat_opencl.h
+++ b/intern/cycles/kernel/kernel_compat_opencl.h
@@ -35,6 +35,7 @@
 #define ccl_device_inline ccl_device
 #define ccl_device_forceinline ccl_device
 #define ccl_device_noinline ccl_device ccl_noinline
+#define ccl_device_noinline_cpu ccl_device
 #define ccl_may_alias
 #define ccl_static_constant static __constant
 #define ccl_constant __constant
@@ -45,6 +46,7 @@
 #define ccl_restrict restrict
 #define ccl_ref
 #define ccl_align(n) __attribute__((aligned(n)))
+#define ccl_optional_struct_init
 
 #ifdef __SPLIT_KERNEL__
 #  define ccl_addr_space __global
diff --git a/intern/cycles/kernel/kernel_emission.h b/intern/cycles/kernel/kernel_emission.h
index 34300543f91..459280cf433 100644
--- a/intern/cycles/kernel/kernel_emission.h
+++ b/intern/cycles/kernel/kernel_emission.h
@@ -17,17 +17,17 @@
 CCL_NAMESPACE_BEGIN
 
 /* Direction Emission */
-ccl_device_noinline float3 direct_emissive_eval(KernelGlobals *kg,
-                                                ShaderData *emission_sd,
-                                                LightSample *ls,
-                                                ccl_addr_space PathState *state,
-                                                float3 I,
-                                                differential3 dI,
-                                                float t,
-                                                float time)
+ccl_device_noinline_cpu float3 direct_emissive_eval(KernelGlobals *kg,
+                                                    ShaderData *emission_sd,
+                                                    LightSample *ls,
+                                                    ccl_addr_space PathState *state,
+                                                    float3 I,
+                                                    differential3 dI,
+                                                    float t,
+                                                    float time)
 {
   /* setup shading at emitter */
-  float3 eval;
+  float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
   if (shader_constant_emission_eval(kg, ls->shader, &eval)) {
     if ((ls->prim != PRIM_NONE) && dot(ls->Ng, I) < 0.0f) {
@@ -98,15 +98,15 @@ ccl_device_noinline float3 direct_emissive_eval(KernelGlobals *kg,
   return eval;
 }
 
-ccl_device_noinline bool direct_emission(KernelGlobals *kg,
-                                         ShaderData *sd,
-                                         ShaderData *emission_sd,
-                                         LightSample *ls,
-                                         ccl_addr_space PathState *state,
-                                         Ray *ray,
-                                         BsdfEval *eval,
-                                         bool *is_lamp,
-                                         float rand_terminate)
+ccl_device_noinline_cpu bool direct_emission(KernelGlobals *kg,
+                                             ShaderData *sd,
+                                             ShaderData *emission_sd,
+                                             LightSample *ls,
+                                             ccl_addr_space PathState *state,
+                                             Ray *ray,
+                                             BsdfEval *eval,
+                                             bool *is_lamp,
+                                             float rand_terminate)
 {
   if (ls->pdf == 0.0f)
     return false;
@@ -208,7 +208,7 @@ ccl_device_noinline bool direct_emission(KernelGlobals *kg,
 
 /* Indirect Primitive Emission */
 
-ccl_device_noinline float3 indirect_primitive_emission(
+ccl_device_noinline_cpu float3 indirect_primitive_emission(
     KernelGlobals *kg, ShaderData *sd, float t, int path_flag, float bsdf_pdf)
 {
   /* evaluate emissive closure */
@@ -234,18 +234,16 @@ ccl_device_noinline float3 indirect_primitive_emission(
 
 /* Indirect Lamp Emission */
 
-ccl_device_noinline bool indirect_lamp_emission(KernelGlobals *kg,
-                                                ShaderData *emission_sd,
-                                                ccl_addr_space PathState *state,
-                                                Ray *ray,
-                                                float3 *emission)
+ccl_device_noinline_cpu bool indirect_lamp_emission(KernelGlobals *kg,
+                                                    ShaderData *emission_sd,
+                                                    ccl_addr_space PathState *state,
+                                                    Ray *ray,
+                                                    float3 *emission)
 {
   bool hit_lamp = false;
 
-  *emission = make_float3(0.0f, 0.0f, 0.0f);
-
   for (int lamp = 0; lamp < kernel_data.integrator.num_all_lights; lamp++) {
-    LightSample ls;
+    LightSample ls ccl_optional_struct_init;
 
     if (!lamp_light_eval(kg, lamp, ray->P, ray->D, ray->t, &ls))
       continue;
@@ -293,10 +291,10 @@ ccl_device_noinline bool indirect_lamp_emission(KernelGlobals *kg,
 
 /* Indirect Background */
 
-ccl_device_noinline float3 indirect_background(KernelGlobals *kg,
-                                               ShaderData *emission_sd,
-                                               ccl_addr_space PathState *state,
-                                               ccl_addr_space Ray *ray)
+ccl_device_noinline_cpu float3 indirect_background(KernelGlobals *kg,
+                                                   ShaderData *emission_sd,
+                                                   ccl_addr_space PathState *state,
+                                                   ccl_addr_space Ray *ray)
 {
 #ifdef __BACKGROUND__
   int shader = kernel_data.background.surface_shader;
@@ -314,7 +312,7 @@ ccl_device_noinline float3 indirect_background(KernelGlobals *kg,
   }
 
   /* Evaluate background shader. */
-  float3 L;
+  float3 L = make_float3(0.0f, 0.0f, 0.0f);
   if (!shader_constant_emission_eval(kg, shader, &L)) {
 #  ifdef __SPLIT_KERNEL__
     Ray priv_ray = *ray;
diff --git a/intern/cycles/kernel/kernel_jitter.h b/intern/cycles/kernel/kernel_jitter.h
index f7270a14940..e59d8946950 100644
--- a/intern/cycles/kernel/kernel_jitter.h
+++ b/intern/cycles/kernel/kernel_jitter.h
@@ -38,43 +38,13 @@ ccl_device_inline int cmj_fast_mod_pow2(int a, int b)
 ccl_device_inline int cmj_fast_div_pow2(int a, int b)
 {
   kernel_assert(b > 1);
-#if defined(__KERNEL_SSE2__)
-#  ifdef _MSC_VER
-  unsigned long ctz;
-  _BitScanForward(&ctz, b);
-  return a >> ctz;
-#  else
-  return a >> __builtin_ctz(b);
-#  endif
-#elif defined(__KERNEL_CUDA__)
-  return a >> (__ffs(b) - 1);
-#else
-  return a / b;
-#endif
+  return a >> count_trailing_zeros(b);
 }
 
 ccl_device_inline uint cmj_w_mask(uint w)
 {
   kernel_assert(w > 1);
-#if defined(__KERNEL_SSE2__)
-#  ifdef _MSC_VER
-  unsigned long leading_zero;
-  _BitScanReverse(&leading_zero, w);
-  return ((1 << (1 + leading_zero)) - 1);
-#  else
-  return ((1 << (32 - __builtin_clz(w))) - 1);
-#  endif
-#elif defined(__KERNEL_CUDA__)
-  return ((1 << (32 - __clz(w))) - 1);
-#else
-  w |= w >> 1;
-  w |= w >> 2;
-  w |= w >> 4;
-  w |= w >> 8;
-  w |= w >> 16;
-
-  return w;
-#endif
+  return ((1 << (32 - count_leading_zeros(w))) - 1);
 }
 
 ccl_device_inline uint cmj_permute(uint i, uint l, uint p)
diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index 9128bfa9d95..ce908ce0fe2 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -182,17 +182,7 @@ ccl_device float lamp_light_pdf(KernelGlobals *kg, const float3 Ng, const float3
 
 #ifdef __BACKGROUND_MIS__
 
-/* TODO(sergey): In theory it should be all fine to use noinline for all
- * devices, but we're so close to the release so better not screw things
- * up for CPU at least.
- */
-#  ifdef __KERNEL_GPU__
-ccl_device_noinline
-#  else
-ccl_device
-#  endif
-    float3
-    background_map_sample(KernelGlobals *kg, float randu, float randv, float *pdf)
+ccl_device float3 background_map_sample(KernelGlobals *kg, float randu, float randv, float *pdf)
 {
   /* for the following, the CDF values are actually a pair of floats, with the
    * function value as X and the actual CDF as Y.  The last entry's function
@@ -274,13 +264,7 @@ ccl_device
 /* TODO(sergey): Same as above, after the release we should consider using
  * 'noinline' for all devices.
  */
-#  ifdef __KERNEL_GPU__
-ccl_device_noinline
-#  else
-ccl_device
-#  endif
-    float
-    background_map_pdf(KernelGlobals *kg, float3 direction)
+ccl_device float background_map_pdf(KernelGlobals *kg, float3 direction)
 {
   float2 uv = direction_to_equirectangular(direction);
   int res_x = kernel_data.integrator.pdf_background_res_x;
@@ -1092,7 +1076,7 @@ ccl_device int light_distribution_sample(KernelGlobals *kg, float *randu)
   int len = kernel_data.integrator.num_distribution + 1;
   float r = *randu;
 
-  while (len > 0) {
+  do {
     int half_len = len >> 1;
     int middle = first + half_len;
 
@@ -1103,7 +1087,7 @@ ccl_device int light_distribution_sample(KernelGlobals *kg, float *randu)
       first = middle + 1;
       len = len - half_len - 1;
     }
-  }
+  } while (len > 0);
 
   /* Clamping should not be needed but float rounding errors seem to
    * make this fail on rare occasions. */
@@ -1120,42 +1104,49 @@ ccl_device int light_distribution_sample(KernelGlobals *kg, float *randu)
 
 /* Generic Light */
 
-ccl_device bool light_select_reached_max_bounces(KernelGlobals *kg, int index, int bounce)
+ccl_device_inline bool light_select_reached_max_bounces(KernelGlobals *kg, int index, int bounce)
 {
   return (bounce > kernel_tex_fetch(__lights, index).max_bounces);
 }
 
-ccl_device_noinline bool light_sample(
-    KernelGlobals *kg, float randu, float randv, float time, float3 P, int bounce, LightSample *ls)
+ccl_device_noinline bool light_sample(KernelGlobals *kg,
+                                      int lamp,
+                                      float randu,
+                                      float randv,
+                                      float time,
+                                      float3 P,
+                                      int bounce,
+                                      LightSample *ls)
 {
-  /* sample index */
-  int index = light_distribution_sample(kg, &randu);
-
-  /* fetch light data */
-  const ccl_global KernelLightDistribution *kdistribution = &kernel_tex_fetch(__light_distribution,
-                                                                              index);
-  int prim = kdistribution->prim;
-
-  if (prim >= 0) {
-    int object = kdistribution->mesh_light.object_id;
-    int shader_flag = kdistribution->mesh_light.shader_flag;
+  if (lamp < 0) {
+    /* sample index */
+    int index = light_distribution_sample(kg, &randu);
+
+    /* fetch light data */
+    const ccl_global KernelLightDistribution *kdistribution = &kernel_tex_fetch(
+        __light_distribution, index);
+    int prim = kdistribution->prim;
+
+    if (prim >= 0) {
+      int object = kdistribution->mesh_light.object_id;
+      int shader_flag = kdistribution->mesh_light.shader_flag;
+
+      triangle_light_sample(kg, prim, object, randu, randv, time, ls, P);
+      ls->shader |= shader_flag;
+      return (ls->pdf > 0.0f);
+    }
 
-    triangle_light_sample(kg, prim, object, randu, randv, time, ls, P);
-    ls->shader |= shader_flag;
-    return (ls->pdf > 0.0f);
+    lamp = -prim - 1;
   }
-  else {
-    int lamp = -prim - 1;
 
-    if (UNLIKELY(light_select_reached_max_bounces(kg, lamp, bounce))) {
-      return false;
-    }
-
-    return lamp_light_sample(kg, lamp, randu, randv, P, ls);
+  if (UNLIKELY(light_select_reached_max_bounces(kg, lamp, bounce))) {
+    return false;
   }
+
+  return lamp_light_sample(kg, lamp, randu, randv, P, ls);
 }
 
-ccl_device int light_select_num_samples(KernelGlobals *kg, int index)
+ccl_device_inline int light_select_num_samples(KernelGlobals *kg, int index)
 {
   return kernel_tex_fetch(__lights, index).samples;
 }
diff --git a/intern/cycles/kernel/kernel_passes.h b/intern/cycles/kernel/kernel_passes.h
index e84937e2fd9..4a424866efe 100644
--- a/intern/cycles/kernel/kernel_passes.h
+++ b/intern/cycles/kernel/kernel_passes.h
@@ -114,14 +114,12 @@ ccl_device_inline void kernel_write_denoising_shadow(KernelGlobals *kg,
   float value = path_total_shaded / max(path_total, 1e-7f);
   kernel_write_pass_float(buffer + 2, value * value);
 }
-#endif /* __DENOISING_FEATURES__ */
 
 ccl_device_inline void kernel_update_denoising_features(KernelGlobals *kg,
                                                         ShaderData *sd,
                                                         ccl_addr_space PathState *state,
                                                         PathRadiance *L)
 {
-#ifdef __DENOISING_FEATURES__
   if (state->denoising_feature_weight == 0.0f) {
     return;
   }
@@ -162,13 +160,8 @@ ccl_device_inline void kernel_update_denoising_features(KernelGlobals *kg,
 
     state->denoising_feature_weight = 0.0f;
   }
-#else
-  (void)kg;
-  (void)sd;
-  (void)state;
-  (void)L;
-#endif /* __DENOISING_FEATURES__ */
 }
+#endif /* __DENOISING_FEATURES__ */
 
 #ifdef __KERNEL_DEBUG__
 ccl_device_inline void kernel_write_debug_passes(KernelGlobals *kg,
diff --git a/intern/cycles/kernel/kernel_path.h b/intern/cycles/kernel/kernel_path.h
index f3e2a8a234a..0dc55eba14a 100644
--- a/intern/cycles/kernel/kernel_path.h
+++ b/intern/cycles/kernel/kernel_path.h
@@ -65,7 +65,7 @@ ccl_device_forceinline bool kernel_path_scene_intersect(KernelGlobals *kg,
     ray->t = kernel_data.background.ao_distance;
   }
 
-  bool hit = scene_intersect(kg, *ray, visibility, isect);
+  bool hit = scene_intersect(kg, ray, visibility, isect);
 
 #ifdef __KERNEL_DEBUG__
   if (state->flag & PATH_RAY_CAMERA) {
@@ -92,7 +92,7 @@ ccl_device_forceinline void kernel_path_lamp_emission(KernelGlobals *kg,
 #ifdef __LAMP_MIS__
   if (kernel_data.integrator.use_lamp_mis && !(state->flag & PATH_RAY_CAMERA)) {
     /* ray starting from previous non-transparent bounce */
-    Ray light_ray;
+    Ray light_ray ccl_optional_struct_init;
 
     light_ray.P = ray->P - state->ray_t * ray->D;
     state->ray_t += isect->t;
@@ -103,7 +103,7 @@ ccl_device_forceinline void kernel_path_lamp_emission(KernelGlobals *kg,
     light_ray.dP = ray->dP;
 
     /* intersect with lamp */
-    float3 emission;
+    float3 emission = make_float3(0.0f, 0.0f, 0.0f);
 
     if (indirect_lamp_emission(kg, emission_sd, state, &light_ray, &emission))
       path_radiance_accum_emission(L, state, throughput, emission);
@@ -474,12 +474,10 @@ ccl_device void kernel_path_indirect(KernelGlobals *kg,
 #    endif /* __SUBSURFACE__ */
 
 #    if defined(__EMISSION__)
-        if (kernel_data.integrator.use_direct_light) {
-          int all = (kernel_data.integrator.sample_all_lights_indirect) ||
-                    (state->flag & PATH_RAY_SHADOW_CATCHER);
-          kernel_branched_path_surface_connect_light(
-              kg, sd, emission_sd, state, throughput, 1.0f, L, all);
-        }
+        int all = (kernel_data.integrator.sample_all_lights_indirect) ||
+                  (state->flag & PATH_RAY_SHADOW_CATCHER);
+        kernel_branched_path_surface_connect_light(
+            kg, sd, emission_sd, state, throughput, 1.0f, L, all);
 #    endif /* defined(__EMISSION__) */
 
 #    ifdef __VOLUME__
@@ -590,7 +588,9 @@ ccl_device_forceinline void kernel_path_integrate(KernelGlobals *kg,
           throughput /= probability;
         }
 
+#  ifdef __DENOISING_FEATURES__
         kernel_update_denoising_features(kg, &sd, state, L);
+#  endif
 
 #  ifdef __AO__
         /* ambient occlusion */
@@ -610,8 +610,10 @@ ccl_device_forceinline void kernel_path_integrate(KernelGlobals *kg,
         }
 #  endif /* __SUBSURFACE__ */
 
+#  ifdef __EMISSION__
         /* direct lighting */
         kernel_path_surface_connect_light(kg, &sd, emission_sd, throughput, state, L);
+#  endif /* __EMISSION__ */
 
 #  ifdef __VOLUME__
       }
diff --git a/intern/cycles/kernel/kernel_path_branched.h b/intern/cycles/kernel/kernel_path_branched.h
index f3a1ea3f4fd..ea6b23e7eb4 100644
--- a/intern/cycles/kernel/kernel_path_branched.h
+++ b/intern/cycles/kernel/kernel_path_branched.h
@@ -198,14 +198,14 @@ ccl_device_forceinline void kernel_branched_path_volume(KernelGlobals *kg,
 #    endif /* __VOLUME__ */
 
 /* bounce off surface and integrate indirect light */
-ccl_device_noinline void kernel_branched_path_surface_indirect_light(KernelGlobals *kg,
-                                                                     ShaderData *sd,
-                                                                     ShaderData *indirect_sd,
-                                                                     ShaderData *emission_sd,
-                                                                     float3 throughput,
-                                                                     float num_samples_adjust,
-                                                                     PathState *state,
-                                                                     PathRadiance *L)
+ccl_device_noinline_cpu void kernel_branched_path_surface_indirect_light(KernelGlobals *kg,
+                                                                         ShaderData *sd,
+                                                                         ShaderData *indirect_sd,
+                                                                         ShaderData *emission_sd,
+                                                                         float3 throughput,
+                                                                         float num_samples_adjust,
+                                                                         PathState *state,
+                                                                         PathRadiance *L)
 {
   float sum_sample_weight = 0.0f;
 #    ifdef __DENOISING_FEATURES__
@@ -445,7 +445,9 @@ ccl_device void kernel_branched_path_integrate(KernelGlobals *kg,
         }
       }
 
+#    ifdef __DENOISING_FEATURES__
       kernel_update_denoising_features(kg, &sd, &state, L);
+#    endif
 
 #    ifdef __AO__
       /* ambient occlusion */
diff --git a/intern/cycles/kernel/kernel_path_surface.h b/intern/cycles/kernel/kernel_path_surface.h
index a1ab4951565..a32690d51eb 100644
--- a/intern/cycles/kernel/kernel_path_surface.h
+++ b/intern/cycles/kernel/kernel_path_surface.h
@@ -20,7 +20,7 @@ CCL_NAMESPACE_BEGIN
     defined(__BAKING__)
 /* branched path tracing: connect path directly to position on one or more lights and add it to L
  */
-ccl_device_noinline void kernel_branched_path_surface_connect_light(
+ccl_device_noinline_cpu void kernel_branched_path_surface_connect_light(
     KernelGlobals *kg,
     ShaderData *sd,
     ShaderData *emission_sd,
@@ -32,140 +32,100 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(
 {
 #  ifdef __EMISSION__
   /* sample illumination from lights to find path contribution */
-  if (!(sd->flag & SD_BSDF_HAS_EVAL))
-    return;
+  BsdfEval L_light ccl_optional_struct_init;
+
+  int num_lights = 0;
+  if (kernel_data.integrator.use_direct_light) {
+    if (sample_all_lights) {
+      num_lights = kernel_data.integrator.num_all_lights;
+      if (kernel_data.integrator.pdf_triangles != 0.0f) {
+        num_lights += 1;
+      }
+    }
+    else {
+      num_lights = 1;
+    }
+  }
+
+  for (int i = 0; i < num_lights; i++) {
+    /* sample one light at random */
+    int num_samples = 1;
+    int num_all_lights = 1;
+    uint lamp_rng_hash = state->rng_hash;
+    bool double_pdf = false;
+    bool is_mesh_light = false;
+    bool is_lamp = false;
+
+    if (sample_all_lights) {
+      /* lamp sampling */
+      is_lamp = i < kernel_data.integrator.num_all_lights;
+      if (is_lamp) {
+        if (UNLIKELY(light_select_reached_max_bounces(kg, i, state->bounce))) {
+          continue;
+        }
+        num_samples = ceil_to_int(num_samples_adjust * light_select_num_samples(kg, i));
+        num_all_lights = kernel_data.integrator.num_all_lights;
+        lamp_rng_hash = cmj_hash(state->rng_hash, i);
+        double_pdf = kernel_data.integrator.pdf_triangles != 0.0f;
+      }
+      /* mesh light sampling */
+      else {
+        num_samples = ceil_to_int(num_samples_adjust * kernel_data.integrator.mesh_light_samples);
+        double_pdf = kernel_data.integrator.num_all_lights != 0;
+        is_mesh_light = true;
+      }
+    }
 
-  Ray light_ray;
-  BsdfEval L_light;
-  bool is_lamp;
+    float num_samples_inv = num_samples_adjust / (num_samples * num_all_lights);
 
+    for (int j = 0; j < num_samples; j++) {
+      Ray light_ray ccl_optional_struct_init;
+      light_ray.t = 0.0f; /* reset ray */
 #    ifdef __OBJECT_MOTION__
-  light_ray.time = sd->time;
+      light_ray.time = sd->time;
 #    endif
+      bool has_emission = false;
 
-  if (sample_all_lights) {
-    /* lamp sampling */
-    for (int i = 0; i < kernel_data.integrator.num_all_lights; i++) {
-      if (UNLIKELY(light_select_reached_max_bounces(kg, i, state->bounce)))
-        continue;
-
-      int num_samples = ceil_to_int(num_samples_adjust * light_select_num_samples(kg, i));
-      float num_samples_inv = num_samples_adjust /
-                              (num_samples * kernel_data.integrator.num_all_lights);
-      uint lamp_rng_hash = cmj_hash(state->rng_hash, i);
-
-      for (int j = 0; j < num_samples; j++) {
+      if (kernel_data.integrator.use_direct_light && (sd->flag & SD_BSDF_HAS_EVAL)) {
         float light_u, light_v;
         path_branched_rng_2D(
             kg, lamp_rng_hash, state, j, num_samples, PRNG_LIGHT_U, &light_u, &light_v);
         float terminate = path_branched_rng_light_termination(
             kg, lamp_rng_hash, state, j, num_samples);
 
-        LightSample ls;
-        if (lamp_light_sample(kg, i, light_u, light_v, sd->P, &ls)) {
-          /* The sampling probability returned by lamp_light_sample assumes that all lights were
-           * sampled.
-           * However, this code only samples lamps, so if the scene also had mesh lights, the real
-           * probability is twice as high. */
-          if (kernel_data.integrator.pdf_triangles != 0.0f)
-            ls.pdf *= 2.0f;
-
-          if (direct_emission(
-                  kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
-            /* trace shadow ray */
-            float3 shadow;
-
-            if (!shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow)) {
-              /* accumulate */
-              path_radiance_accum_light(L,
-                                        state,
-                                        throughput * num_samples_inv,
-                                        &L_light,
-                                        shadow,
-                                        num_samples_inv,
-                                        is_lamp);
-            }
-            else {
-              path_radiance_accum_total_light(L, state, throughput * num_samples_inv, &L_light);
-            }
-          }
-        }
-      }
-    }
-
-    /* mesh light sampling */
-    if (kernel_data.integrator.pdf_triangles != 0.0f) {
-      int num_samples = ceil_to_int(num_samples_adjust *
-                                    kernel_data.integrator.mesh_light_samples);
-      float num_samples_inv = num_samples_adjust / num_samples;
-
-      for (int j = 0; j < num_samples; j++) {
-        float light_u, light_v;
-        path_branched_rng_2D(
-            kg, state->rng_hash, state, j, num_samples, PRNG_LIGHT_U, &light_u, &light_v);
-        float terminate = path_branched_rng_light_termination(
-            kg, state->rng_hash, state, j, num_samples);
-
         /* only sample triangle lights */
-        if (kernel_data.integrator.num_all_lights)
+        if (is_mesh_light && double_pdf) {
           light_u = 0.5f * light_u;
+        }
 
-        LightSample ls;
-        if (light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
-          /* Same as above, probability needs to be corrected since the sampling was forced to
-           * select a mesh light. */
-          if (kernel_data.integrator.num_all_lights)
+        LightSample ls ccl_optional_struct_init;
+        const int lamp = is_lamp ? i : -1;
+        if (light_sample(kg, lamp, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+          /* The sampling probability returned by lamp_light_sample assumes that all lights were
+           * sampled. However, this code only samples lamps, so if the scene also had mesh lights,
+           * the real probability is twice as high. */
+          if (double_pdf) {
             ls.pdf *= 2.0f;
-
-          if (direct_emission(
-                  kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
-            /* trace shadow ray */
-            float3 shadow;
-
-            if (!shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow)) {
-              /* accumulate */
-              path_radiance_accum_light(L,
-                                        state,
-                                        throughput * num_samples_inv,
-                                        &L_light,
-                                        shadow,
-                                        num_samples_inv,
-                                        is_lamp);
-            }
-            else {
-              path_radiance_accum_total_light(L, state, throughput * num_samples_inv, &L_light);
-            }
           }
+
+          has_emission = direct_emission(
+              kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate);
         }
       }
-    }
-  }
-  else {
-    /* sample one light at random */
-    float light_u, light_v;
-    path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
-    float terminate = path_state_rng_light_termination(kg, state);
 
-    LightSample ls;
-    if (light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
-      /* sample random light */
-      if (direct_emission(
-              kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
-        /* trace shadow ray */
-        float3 shadow;
+      /* trace shadow ray */
+      float3 shadow;
+
+      const bool blocked = shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow);
 
-        if (!shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow)) {
+      if (has_emission) {
+        if (!blocked) {
           /* accumulate */
-          path_radiance_accum_light(L,
-                                    state,
-                                    throughput * num_samples_adjust,
-                                    &L_light,
-                                    shadow,
-                                    num_samples_adjust,
-                                    is_lamp);
+          path_radiance_accum_light(
+              L, state, throughput * num_samples_inv, &L_light, shadow, num_samples_inv, is_lamp);
         }
         else {
-          path_radiance_accum_total_light(L, state, throughput * num_samples_adjust, &L_light);
+          path_radiance_accum_total_light(L, state, throughput * num_samples_inv, &L_light);
         }
       }
     }
@@ -187,9 +147,9 @@ ccl_device bool kernel_branched_path_surface_bounce(KernelGlobals *kg,
 {
   /* sample BSDF */
   float bsdf_pdf;
-  BsdfEval bsdf_eval;
-  float3 bsdf_omega_in;
-  differential3 bsdf_domega_in;
+  BsdfEval bsdf_eval ccl_optional_struct_init;
+  float3 bsdf_omega_in ccl_optional_struct_init;
+  differential3 bsdf_domega_in ccl_optional_struct_init;
   float bsdf_u, bsdf_v;
   path_branched_rng_2D(
       kg, state->rng_hash, state, sample, num_samples, PRNG_BSDF_U, &bsdf_u, &bsdf_v);
@@ -255,45 +215,48 @@ ccl_device_inline void kernel_path_surface_connect_light(KernelGlobals *kg,
   PROFILING_INIT(kg, PROFILING_CONNECT_LIGHT);
 
 #ifdef __EMISSION__
-  if (!(kernel_data.integrator.use_direct_light && (sd->flag & SD_BSDF_HAS_EVAL)))
-    return;
-
 #  ifdef __SHADOW_TRICKS__
-  if (state->flag & PATH_RAY_SHADOW_CATCHER) {
-    kernel_branched_path_surface_connect_light(kg, sd, emission_sd, state, throughput, 1.0f, L, 1);
-    return;
-  }
-#  endif
-
+  int all = (state->flag & PATH_RAY_SHADOW_CATCHER);
+  kernel_branched_path_surface_connect_light(kg, sd, emission_sd, state, throughput, 1.0f, L, all);
+#  else
   /* sample illumination from lights to find path contribution */
-  float light_u, light_v;
-  path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
+  Ray light_ray ccl_optional_struct_init;
+  BsdfEval L_light ccl_optional_struct_init;
+  bool is_lamp = false;
+  bool has_emission = false;
 
-  Ray light_ray;
-  BsdfEval L_light;
-  bool is_lamp;
-
-#  ifdef __OBJECT_MOTION__
+  light_ray.t = 0.0f;
+#    ifdef __OBJECT_MOTION__
   light_ray.time = sd->time;
-#  endif
+#    endif
 
-  LightSample ls;
-  if (light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
-    float terminate = path_state_rng_light_termination(kg, state);
-    if (direct_emission(
-            kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
-      /* trace shadow ray */
-      float3 shadow;
+  if (kernel_data.integrator.use_direct_light && (sd->flag & SD_BSDF_HAS_EVAL)) {
+    float light_u, light_v;
+    path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
 
-      if (!shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow)) {
-        /* accumulate */
-        path_radiance_accum_light(L, state, throughput, &L_light, shadow, 1.0f, is_lamp);
-      }
-      else {
-        path_radiance_accum_total_light(L, state, throughput, &L_light);
-      }
+    LightSample ls ccl_optional_struct_init;
+    if (light_sample(kg, -1, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+      float terminate = path_state_rng_light_termination(kg, state);
+      has_emission = direct_emission(
+          kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate);
     }
   }
+
+  /* trace shadow ray */
+  float3 shadow;
+
+  const bool blocked = shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow);
+
+  if (has_emission) {
+    if (!blocked) {
+      /* accumulate */
+      path_radiance_accum_light(L, state, throughput, &L_light, shadow, 1.0f, is_lamp);
+    }
+    else {
+      path_radiance_accum_total_light(L, state, throughput, &L_light);
+    }
+  }
+#  endif
 #endif
 }
 
@@ -311,9 +274,9 @@ ccl_device bool kernel_path_surface_bounce(KernelGlobals *kg,
   if (sd->flag & SD_BSDF) {
     /* sample BSDF */
     float bsdf_pdf;
-    BsdfEval bsdf_eval;
-    float3 bsdf_omega_in;
-    differential3 bsdf_domega_in;
+    BsdfEval bsdf_eval ccl_optional_struct_init;
+    float3 bsdf_omega_in ccl_optional_struct_init;
+    differential3 bsdf_domega_in ccl_optional_struct_init;
     float bsdf_u, bsdf_v;
     path_state_rng_2D(kg, state, PRNG_BSDF_U, &bsdf_u, &bsdf_v);
     int label;
diff --git a/intern/cycles/kernel/kernel_path_volume.h b/intern/cycles/kernel/kernel_path_volume.h
index fea4dfc159d..db10629ee9f 100644
--- a/intern/cycles/kernel/kernel_path_volume.h
+++ b/intern/cycles/kernel/kernel_path_volume.h
@@ -26,55 +26,54 @@ ccl_device_inline void kernel_path_volume_connect_light(KernelGlobals *kg,
                                                         PathRadiance *L)
 {
 #  ifdef __EMISSION__
-  if (!kernel_data.integrator.use_direct_light)
-    return;
-
   /* sample illumination from lights to find path contribution */
-  float light_u, light_v;
-  path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
-
-  Ray light_ray;
-  BsdfEval L_light;
-  LightSample ls;
-  bool is_lamp;
+  Ray light_ray ccl_optional_struct_init;
+  BsdfEval L_light ccl_optional_struct_init;
+  bool is_lamp = false;
+  bool has_emission = false;
 
+  light_ray.t = 0.0f;
+#    ifdef __OBJECT_MOTION__
   /* connect to light from given point where shader has been evaluated */
   light_ray.time = sd->time;
+#    endif
 
-  if (light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
-    float terminate = path_state_rng_light_termination(kg, state);
-    if (direct_emission(
-            kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
-      /* trace shadow ray */
-      float3 shadow;
+  if (kernel_data.integrator.use_direct_light) {
+    float light_u, light_v;
+    path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
 
-      if (!shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow)) {
-        /* accumulate */
-        path_radiance_accum_light(L, state, throughput, &L_light, shadow, 1.0f, is_lamp);
-      }
+    LightSample ls ccl_optional_struct_init;
+    if (light_sample(kg, -1, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+      float terminate = path_state_rng_light_termination(kg, state);
+      has_emission = direct_emission(
+          kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate);
     }
   }
+
+  /* trace shadow ray */
+  float3 shadow;
+
+  const bool blocked = shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow);
+
+  if (has_emission && !blocked) {
+    /* accumulate */
+    path_radiance_accum_light(L, state, throughput, &L_light, shadow, 1.0f, is_lamp);
+  }
 #  endif /* __EMISSION__ */
 }
 
-#  ifdef __KERNEL_GPU__
-ccl_device_noinline
-#  else
-ccl_device
-#  endif
-    bool
-    kernel_path_volume_bounce(KernelGlobals *kg,
-                              ShaderData *sd,
-                              ccl_addr_space float3 *throughput,
-                              ccl_addr_space PathState *state,
-                              PathRadianceState *L_state,
-                              ccl_addr_space Ray *ray)
+ccl_device_noinline_cpu bool kernel_path_volume_bounce(KernelGlobals *kg,
+                                                       ShaderData *sd,
+                                                       ccl_addr_space float3 *throughput,
+                                                       ccl_addr_space PathState *state,
+                                                       PathRadianceState *L_state,
+                                                       ccl_addr_space Ray *ray)
 {
   /* sample phase function */
   float phase_pdf;
-  BsdfEval phase_eval;
-  float3 phase_omega_in;
-  differential3 phase_domega_in;
+  BsdfEval phase_eval ccl_optional_struct_init;
+  float3 phase_omega_in ccl_optional_struct_init;
+  differential3 phase_domega_in ccl_optional_struct_init;
   float phase_u, phase_v;
   path_state_rng_2D(kg, state, PRNG_BSDF_U, &phase_u, &phase_v);
   int label;
@@ -128,7 +127,7 @@ ccl_device
   return true;
 }
 
-#  ifndef __SPLIT_KERNEL__
+#  if !defined(__SPLIT_KERNEL__) && (defined(__BRANCHED_PATH__) || defined(__VOLUME_DECOUPLED__))
 ccl_device void kernel_branched_path_volume_connect_light(KernelGlobals *kg,
                                                           ShaderData *sd,
                                                           ShaderData *emission_sd,
@@ -140,96 +139,71 @@ ccl_device void kernel_branched_path_volume_connect_light(KernelGlobals *kg,
                                                           const VolumeSegment *segment)
 {
 #    ifdef __EMISSION__
-  if (!kernel_data.integrator.use_direct_light)
-    return;
-
-  Ray light_ray;
-  BsdfEval L_light;
-  bool is_lamp;
-
-  light_ray.time = sd->time;
+  BsdfEval L_light ccl_optional_struct_init;
 
+  int num_lights = 1;
   if (sample_all_lights) {
-    /* lamp sampling */
-    for (int i = 0; i < kernel_data.integrator.num_all_lights; i++) {
-      if (UNLIKELY(light_select_reached_max_bounces(kg, i, state->bounce)))
-        continue;
-
-      int num_samples = light_select_num_samples(kg, i);
-      float num_samples_inv = 1.0f / (num_samples * kernel_data.integrator.num_all_lights);
-      uint lamp_rng_hash = cmj_hash(state->rng_hash, i);
-
-      for (int j = 0; j < num_samples; j++) {
-        /* sample random position on given light */
-        float light_u, light_v;
-        path_branched_rng_2D(
-            kg, lamp_rng_hash, state, j, num_samples, PRNG_LIGHT_U, &light_u, &light_v);
-
-        LightSample ls;
-        lamp_light_sample(kg, i, light_u, light_v, ray->P, &ls);
-
-        float3 tp = throughput;
-
-        /* sample position on volume segment */
-        float rphase = path_branched_rng_1D(
-            kg, state->rng_hash, state, j, num_samples, PRNG_PHASE_CHANNEL);
-        float rscatter = path_branched_rng_1D(
-            kg, state->rng_hash, state, j, num_samples, PRNG_SCATTER_DISTANCE);
-
-        VolumeIntegrateResult result = kernel_volume_decoupled_scatter(kg,
-                                                                       state,
-                                                                       ray,
-                                                                       sd,
-                                                                       &tp,
-                                                                       rphase,
-                                                                       rscatter,
-                                                                       segment,
-                                                                       (ls.t != FLT_MAX) ? &ls.P :
-                                                                                           NULL,
-                                                                       false);
+    num_lights = kernel_data.integrator.num_all_lights;
+    if (kernel_data.integrator.pdf_triangles != 0.0f) {
+      num_lights += 1;
+    }
+  }
 
-        /* todo: split up light_sample so we don't have to call it again with new position */
-        if (result == VOLUME_PATH_SCATTERED &&
-            lamp_light_sample(kg, i, light_u, light_v, sd->P, &ls)) {
-          if (kernel_data.integrator.pdf_triangles != 0.0f)
-            ls.pdf *= 2.0f;
-
-          float terminate = path_branched_rng_light_termination(
-              kg, state->rng_hash, state, j, num_samples);
-          if (direct_emission(
-                  kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
-            /* trace shadow ray */
-            float3 shadow;
-
-            if (!shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow)) {
-              /* accumulate */
-              path_radiance_accum_light(
-                  L, state, tp * num_samples_inv, &L_light, shadow, num_samples_inv, is_lamp);
-            }
-          }
+  for (int i = 0; i < num_lights; ++i) {
+    /* sample one light at random */
+    int num_samples = 1;
+    int num_all_lights = 1;
+    uint lamp_rng_hash = state->rng_hash;
+    bool double_pdf = false;
+    bool is_mesh_light = false;
+    bool is_lamp = false;
+
+    if (sample_all_lights) {
+      /* lamp sampling */
+      is_lamp = i < kernel_data.integrator.num_all_lights;
+      if (is_lamp) {
+        if (UNLIKELY(light_select_reached_max_bounces(kg, i, state->bounce))) {
+          continue;
         }
+        num_samples = light_select_num_samples(kg, i);
+        num_all_lights = kernel_data.integrator.num_all_lights;
+        lamp_rng_hash = cmj_hash(state->rng_hash, i);
+        double_pdf = kernel_data.integrator.pdf_triangles != 0.0f;
+      }
+      /* mesh light sampling */
+      else {
+        num_samples = kernel_data.integrator.mesh_light_samples;
+        double_pdf = kernel_data.integrator.num_all_lights != 0;
+        is_mesh_light = true;
       }
     }
 
-    /* mesh light sampling */
-    if (kernel_data.integrator.pdf_triangles != 0.0f) {
-      int num_samples = kernel_data.integrator.mesh_light_samples;
-      float num_samples_inv = 1.0f / num_samples;
+    float num_samples_inv = 1.0f / (num_samples * num_all_lights);
+
+    for (int j = 0; j < num_samples; j++) {
+      Ray light_ray ccl_optional_struct_init;
+      light_ray.t = 0.0f; /* reset ray */
+#      ifdef __OBJECT_MOTION__
+      light_ray.time = sd->time;
+#      endif
+      bool has_emission = false;
 
-      for (int j = 0; j < num_samples; j++) {
-        /* sample random position on random triangle */
+      float3 tp = throughput;
+
+      if (kernel_data.integrator.use_direct_light) {
+        /* sample random position on random light/triangle */
         float light_u, light_v;
         path_branched_rng_2D(
-            kg, state->rng_hash, state, j, num_samples, PRNG_LIGHT_U, &light_u, &light_v);
+            kg, lamp_rng_hash, state, j, num_samples, PRNG_LIGHT_U, &light_u, &light_v);
 
         /* only sample triangle lights */
-        if (kernel_data.integrator.num_all_lights)
+        if (is_mesh_light && double_pdf) {
           light_u = 0.5f * light_u;
+        }
 
-        LightSample ls;
-        light_sample(kg, light_u, light_v, sd->time, ray->P, state->bounce, &ls);
-
-        float3 tp = throughput;
+        LightSample ls ccl_optional_struct_init;
+        const int lamp = is_lamp ? i : -1;
+        light_sample(kg, lamp, light_u, light_v, sd->time, ray->P, state->bounce, &ls);
 
         /* sample position on volume segment */
         float rphase = path_branched_rng_1D(
@@ -249,69 +223,31 @@ ccl_device void kernel_branched_path_volume_connect_light(KernelGlobals *kg,
                                                                                            NULL,
                                                                        false);
 
-        /* todo: split up light_sample so we don't have to call it again with new position */
-        if (result == VOLUME_PATH_SCATTERED &&
-            light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
-          if (kernel_data.integrator.num_all_lights)
-            ls.pdf *= 2.0f;
-
-          float terminate = path_branched_rng_light_termination(
-              kg, state->rng_hash, state, j, num_samples);
-          if (direct_emission(
-                  kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
-            /* trace shadow ray */
-            float3 shadow;
-
-            if (!shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow)) {
-              /* accumulate */
-              path_radiance_accum_light(
-                  L, state, tp * num_samples_inv, &L_light, shadow, num_samples_inv, is_lamp);
+        if (result == VOLUME_PATH_SCATTERED) {
+          /* todo: split up light_sample so we don't have to call it again with new position */
+          if (light_sample(kg, lamp, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
+            if (double_pdf) {
+              ls.pdf *= 2.0f;
             }
+
+            /* sample random light */
+            float terminate = path_branched_rng_light_termination(
+                kg, state->rng_hash, state, j, num_samples);
+            has_emission = direct_emission(
+                kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate);
           }
         }
       }
-    }
-  }
-  else {
-    /* sample random position on random light */
-    float light_u, light_v;
-    path_state_rng_2D(kg, state, PRNG_LIGHT_U, &light_u, &light_v);
 
-    LightSample ls;
-    light_sample(kg, light_u, light_v, sd->time, ray->P, state->bounce, &ls);
-
-    float3 tp = throughput;
-
-    /* sample position on volume segment */
-    float rphase = path_state_rng_1D(kg, state, PRNG_PHASE_CHANNEL);
-    float rscatter = path_state_rng_1D(kg, state, PRNG_SCATTER_DISTANCE);
-
-    VolumeIntegrateResult result = kernel_volume_decoupled_scatter(kg,
-                                                                   state,
-                                                                   ray,
-                                                                   sd,
-                                                                   &tp,
-                                                                   rphase,
-                                                                   rscatter,
-                                                                   segment,
-                                                                   (ls.t != FLT_MAX) ? &ls.P :
-                                                                                       NULL,
-                                                                   false);
-
-    /* todo: split up light_sample so we don't have to call it again with new position */
-    if (result == VOLUME_PATH_SCATTERED &&
-        light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
-      /* sample random light */
-      float terminate = path_state_rng_light_termination(kg, state);
-      if (direct_emission(
-              kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
-        /* trace shadow ray */
-        float3 shadow;
-
-        if (!shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow)) {
-          /* accumulate */
-          path_radiance_accum_light(L, state, tp, &L_light, shadow, 1.0f, is_lamp);
-        }
+      /* trace shadow ray */
+      float3 shadow;
+
+      const bool blocked = shadow_blocked(kg, sd, emission_sd, state, &light_ray, &shadow);
+
+      if (has_emission && !blocked) {
+        /* accumulate */
+        path_radiance_accum_light(
+            L, state, tp * num_samples_inv, &L_light, shadow, num_samples_inv, is_lamp);
       }
     }
   }
diff --git a/intern/cycles/kernel/kernel_random.h b/intern/cycles/kernel/kernel_random.h
index 78eafbfe3cb..80738213d2a 100644
--- a/intern/cycles/kernel/kernel_random.h
+++ b/intern/cycles/kernel/kernel_random.h
@@ -41,10 +41,9 @@ ccl_device uint sobol_dimension(KernelGlobals *kg, int index, int dimension)
 {
   uint result = 0;
   uint i = index + SOBOL_SKIP;
-  for (uint j = 0; i; i >>= 1, j++) {
-    if (i & 1) {
-      result ^= kernel_tex_fetch(__sobol_directions, 32 * dimension + j);
-    }
+  for (int j = 0, x; (x = find_first_set(i)); i >>= x) {
+    j += x;
+    result ^= kernel_tex_fetch(__sobol_directions, 32 * dimension + j - 1);
   }
   return result;
 }
@@ -130,7 +129,7 @@ ccl_device_inline void path_rng_init(KernelGlobals *kg,
                                      float *fy)
 {
   /* load state */
-  *rng_hash = hash_int_2d(x, y);
+  *rng_hash = hash_uint2(x, y);
   *rng_hash ^= kernel_data.integrator.seed;
 
 #ifdef __DEBUG_CORRELATION__
diff --git a/intern/cycles/kernel/kernel_shader.h b/intern/cycles/kernel/kernel_shader.h
index 4963e012e15..b8202326cdf 100644
--- a/intern/cycles/kernel/kernel_shader.h
+++ b/intern/cycles/kernel/kernel_shader.h
@@ -686,8 +686,7 @@ ccl_device_inline const ShaderClosure *shader_bsdf_pick(ShaderData *sd, float *r
         if (r < next_sum) {
           sampled = i;
 
-          /* Rescale to reuse for direction sample, to better
-           * preserve stratification. */
+          /* Rescale to reuse for direction sample, to better preserve stratification. */
           *randu = (r - partial_sum) / sc->sample_weight;
           break;
         }
@@ -743,8 +742,7 @@ ccl_device_inline const ShaderClosure *shader_bssrdf_pick(ShaderData *sd,
             *throughput *= (sum_bsdf + sum_bssrdf) / sum_bssrdf;
             sampled = i;
 
-            /* Rescale to reuse for direction sample, to better
-             * preserve stratifaction. */
+            /* Rescale to reuse for direction sample, to better preserve stratification. */
             *randu = (r - partial_sum) / sc->sample_weight;
             break;
           }
@@ -780,7 +778,7 @@ ccl_device_inline int shader_bsdf_sample(KernelGlobals *kg,
   kernel_assert(CLOSURE_IS_BSDF(sc->type));
 
   int label;
-  float3 eval;
+  float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
   *pdf = 0.0f;
   label = bsdf_sample(kg, sd, sc, randu, randv, &eval, omega_in, domega_in, pdf);
@@ -810,7 +808,7 @@ ccl_device int shader_bsdf_sample_closure(KernelGlobals *kg,
   PROFILING_INIT(kg, PROFILING_CLOSURE_SAMPLE);
 
   int label;
-  float3 eval;
+  float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
   *pdf = 0.0f;
   label = bsdf_sample(kg, sd, sc, randu, randv, &eval, omega_in, domega_in, pdf);
@@ -1223,7 +1221,7 @@ ccl_device int shader_volume_phase_sample(KernelGlobals *kg,
    * depending on color channels, even if this is perhaps not a common case */
   const ShaderClosure *sc = &sd->closure[sampled];
   int label;
-  float3 eval;
+  float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
   *pdf = 0.0f;
   label = volume_phase_sample(sd, sc, randu, randv, &eval, omega_in, domega_in, pdf);
@@ -1248,7 +1246,7 @@ ccl_device int shader_phase_sample_closure(KernelGlobals *kg,
   PROFILING_INIT(kg, PROFILING_CLOSURE_VOLUME_SAMPLE);
 
   int label;
-  float3 eval;
+  float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
   *pdf = 0.0f;
   label = volume_phase_sample(sd, sc, randu, randv, &eval, omega_in, domega_in, pdf);
@@ -1358,7 +1356,7 @@ ccl_device bool shader_transparent_shadow(KernelGlobals *kg, Intersection *isect
   int shader = 0;
 
 #  ifdef __HAIR__
-  if (kernel_tex_fetch(__prim_type, isect->prim) & PRIMITIVE_ALL_TRIANGLE) {
+  if (isect->type & PRIMITIVE_ALL_TRIANGLE) {
 #  endif
     shader = kernel_tex_fetch(__tri_shader, prim);
 #  ifdef __HAIR__
diff --git a/intern/cycles/kernel/kernel_shadow.h b/intern/cycles/kernel/kernel_shadow.h
index 6640f64518a..c02d7d77faf 100644
--- a/intern/cycles/kernel/kernel_shadow.h
+++ b/intern/cycles/kernel/kernel_shadow.h
@@ -103,7 +103,7 @@ ccl_device bool shadow_blocked_opaque(KernelGlobals *kg,
                                       Intersection *isect,
                                       float3 *shadow)
 {
-  const bool blocked = scene_intersect(kg, *ray, visibility & PATH_RAY_SHADOW_OPAQUE, isect);
+  const bool blocked = scene_intersect(kg, ray, visibility & PATH_RAY_SHADOW_OPAQUE, isect);
 #ifdef __VOLUME__
   if (!blocked && state->volume_stack[0].shader != SHADER_NONE) {
     /* Apply attenuation from current volume shader. */
@@ -318,7 +318,7 @@ ccl_device bool shadow_blocked_transparent_stepped_loop(KernelGlobals *kg,
       if (bounce >= kernel_data.integrator.transparent_max_bounce) {
         return true;
       }
-      if (!scene_intersect(kg, *ray, visibility & PATH_RAY_SHADOW_TRANSPARENT, isect)) {
+      if (!scene_intersect(kg, ray, visibility & PATH_RAY_SHADOW_TRANSPARENT, isect)) {
         break;
       }
       if (!shader_transparent_shadow(kg, isect)) {
@@ -374,7 +374,7 @@ ccl_device bool shadow_blocked_transparent_stepped(KernelGlobals *kg,
                                                    Intersection *isect,
                                                    float3 *shadow)
 {
-  bool blocked = scene_intersect(kg, *ray, visibility & PATH_RAY_SHADOW_OPAQUE, isect);
+  bool blocked = scene_intersect(kg, ray, visibility & PATH_RAY_SHADOW_OPAQUE, isect);
   bool is_transparent_isect = blocked ? shader_transparent_shadow(kg, isect) : false;
   return shadow_blocked_transparent_stepped_loop(
       kg, sd, shadow_sd, state, visibility, ray, isect, blocked, is_transparent_isect, shadow);
@@ -433,7 +433,7 @@ ccl_device_inline bool shadow_blocked(KernelGlobals *kg,
    * TODO(sergey): Check why using record-all behavior causes slowdown in such
    * cases. Could that be caused by a higher spill pressure?
    */
-  const bool blocked = scene_intersect(kg, *ray, visibility & PATH_RAY_SHADOW_OPAQUE, &isect);
+  const bool blocked = scene_intersect(kg, ray, visibility & PATH_RAY_SHADOW_OPAQUE, &isect);
   const bool is_transparent_isect = blocked ? shader_transparent_shadow(kg, &isect) : false;
   if (!blocked || !is_transparent_isect || max_hits + 1 >= SHADOW_STACK_MAX_HITS) {
     return shadow_blocked_transparent_stepped_loop(
diff --git a/intern/cycles/kernel/kernel_subsurface.h b/intern/cycles/kernel/kernel_subsurface.h
index 7510e50a962..8dc1904058d 100644
--- a/intern/cycles/kernel/kernel_subsurface.h
+++ b/intern/cycles/kernel/kernel_subsurface.h
@@ -222,7 +222,7 @@ ccl_device_inline int subsurface_scatter_disk(KernelGlobals *kg,
 
   /* intersect with the same object. if multiple intersections are found it
    * will use at most BSSRDF_MAX_HITS hits, a random subset of all hits */
-  scene_intersect_local(kg, *ray, ss_isect, sd->object, lcg_state, BSSRDF_MAX_HITS);
+  scene_intersect_local(kg, ray, ss_isect, sd->object, lcg_state, BSSRDF_MAX_HITS);
   int num_eval_hits = min(ss_isect->num_hits, BSSRDF_MAX_HITS);
 
   for (int hit = 0; hit < num_eval_hits; hit++) {
@@ -418,7 +418,7 @@ ccl_device_noinline bool subsurface_random_walk(KernelGlobals *kg,
     float t = -logf(1.0f - rdist) / sample_sigma_t;
 
     ray->t = t;
-    scene_intersect_local(kg, *ray, ss_isect, sd->object, NULL, 1);
+    scene_intersect_local(kg, ray, ss_isect, sd->object, NULL, 1);
     hit = (ss_isect->num_hits > 0);
 
     if (hit) {
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index a1d950bbc70..b3cb6ca7c19 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -115,7 +115,6 @@ CCL_NAMESPACE_BEGIN
 #  define __LAMP_MIS__
 #  define __CAMERA_MOTION__
 #  define __OBJECT_MOTION__
-#  define __HAIR__
 #  define __BAKING__
 #  define __PRINCIPLED__
 #  define __SUBSURFACE__
@@ -650,9 +649,8 @@ typedef struct Ray {
  * is fixed.
  */
 #ifndef __KERNEL_OPENCL_AMD__
-  float3 P; /* origin */
-  float3 D; /* direction */
-
+  float3 P;   /* origin */
+  float3 D;   /* direction */
   float t;    /* length of the ray */
   float time; /* time (for motion blur) */
 #else
@@ -1408,6 +1406,7 @@ typedef struct KernelObject {
   float surface_area;
   float pass_id;
   float random_number;
+  float color[3];
   int particle_index;
 
   float dupli_generated[3];
@@ -1420,11 +1419,9 @@ typedef struct KernelObject {
   uint patch_map_offset;
   uint attribute_map_offset;
   uint motion_offset;
-  uint pad1;
 
   float cryptomatte_object;
   float cryptomatte_asset;
-  float pad2, pad3;
 } KernelObject;
 static_assert_align(KernelObject, 16);
 
@@ -1523,7 +1520,7 @@ static_assert_align(KernelShader, 16);
  * Queue 1 - Active rays
  * Queue 2 - Background queue
  * Queue 3 - Shadow ray cast kernel - AO
- * Queeu 4 - Shadow ray cast kernel - direct lighting
+ * Queue 4 - Shadow ray cast kernel - direct lighting
  */
 
 /* Queue names */
diff --git a/intern/cycles/kernel/kernel_volume.h b/intern/cycles/kernel/kernel_volume.h
index cc85110bdd8..4ddcbec0fef 100644
--- a/intern/cycles/kernel/kernel_volume.h
+++ b/intern/cycles/kernel/kernel_volume.h
@@ -187,7 +187,7 @@ ccl_device void kernel_volume_shadow_homogeneous(KernelGlobals *kg,
                                                  ShaderData *sd,
                                                  float3 *throughput)
 {
-  float3 sigma_t;
+  float3 sigma_t = make_float3(0.0f, 0.0f, 0.0f);
 
   if (volume_shader_extinction_sample(kg, sd, state, ray->P, &sigma_t))
     *throughput *= volume_color_transmittance(sigma_t, ray->t);
@@ -225,7 +225,7 @@ ccl_device void kernel_volume_shadow_heterogeneous(KernelGlobals *kg,
     }
 
     float3 new_P = ray->P + ray->D * (t + step_offset);
-    float3 sigma_t;
+    float3 sigma_t = make_float3(0.0f, 0.0f, 0.0f);
 
     /* compute attenuation over segment */
     if (volume_shader_extinction_sample(kg, sd, state, new_P, &sigma_t)) {
@@ -428,7 +428,7 @@ kernel_volume_integrate_homogeneous(KernelGlobals *kg,
                                     ccl_addr_space float3 *throughput,
                                     bool probalistic_scatter)
 {
-  VolumeShaderCoefficients coeff;
+  VolumeShaderCoefficients coeff ccl_optional_struct_init;
 
   if (!volume_shader_sample(kg, sd, state, ray->P, &coeff))
     return VOLUME_PATH_MISSED;
@@ -565,7 +565,7 @@ kernel_volume_integrate_heterogeneous_distance(KernelGlobals *kg,
     }
 
     float3 new_P = ray->P + ray->D * (t + step_offset);
-    VolumeShaderCoefficients coeff;
+    VolumeShaderCoefficients coeff ccl_optional_struct_init;
 
     /* compute segment */
     if (volume_shader_sample(kg, sd, state, new_P, &coeff)) {
@@ -621,6 +621,7 @@ kernel_volume_integrate_heterogeneous_distance(KernelGlobals *kg,
         new_tp = tp * transmittance;
       }
       else {
+        transmittance = make_float3(0.0f, 0.0f, 0.0f);
         new_tp = tp;
       }
 
@@ -671,7 +672,7 @@ kernel_volume_integrate_heterogeneous_distance(KernelGlobals *kg,
  * ray, with the assumption that there are no surfaces blocking light
  * between the endpoints. distance sampling is used to decide if we will
  * scatter or not. */
-ccl_device_noinline VolumeIntegrateResult
+ccl_device_noinline_cpu VolumeIntegrateResult
 kernel_volume_integrate(KernelGlobals *kg,
                         ccl_addr_space PathState *state,
                         ShaderData *sd,
@@ -800,7 +801,7 @@ ccl_device void kernel_volume_decoupled_record(KernelGlobals *kg,
     }
 
     float3 new_P = ray->P + ray->D * (t + step_offset);
-    VolumeShaderCoefficients coeff;
+    VolumeShaderCoefficients coeff ccl_optional_struct_init;
 
     /* compute segment */
     if (volume_shader_sample(kg, sd, state, new_P, &coeff)) {
@@ -1275,7 +1276,7 @@ ccl_device void kernel_volume_stack_init(KernelGlobals *kg,
    */
   if (stack_index == 0 && kernel_data.background.volume_shader == SHADER_NONE) {
     stack[0].shader = kernel_data.background.volume_shader;
-    stack[0].object = PRIM_NONE;
+    stack[0].object = OBJECT_NONE;
     stack[1].shader = SHADER_NONE;
   }
   else {
diff --git a/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h b/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
index 4289e2bbb85..8f311baf010 100644
--- a/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
+++ b/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
@@ -19,6 +19,10 @@
 
 CCL_NAMESPACE_BEGIN
 
+/* Make template functions private so symbols don't conflict between kernels with different
+ * instruction sets. */
+namespace {
+
 template<typename T> struct TextureInterpolator {
 #define SET_CUBIC_SPLINE_WEIGHTS(u, t) \
   { \
@@ -523,6 +527,8 @@ ccl_device float4 kernel_tex_image_interp_3d(
   }
 }
 
+} /* Namespace. */
+
 CCL_NAMESPACE_END
 
 #endif  // __KERNEL_CPU_IMAGE_H__
diff --git a/intern/cycles/kernel/osl/osl_services.cpp b/intern/cycles/kernel/osl/osl_services.cpp
index 316d24b0954..415de9cd66b 100644
--- a/intern/cycles/kernel/osl/osl_services.cpp
+++ b/intern/cycles/kernel/osl/osl_services.cpp
@@ -81,6 +81,7 @@ ustring OSLRenderServices::u_screen("screen");
 ustring OSLRenderServices::u_raster("raster");
 ustring OSLRenderServices::u_ndc("NDC");
 ustring OSLRenderServices::u_object_location("object:location");
+ustring OSLRenderServices::u_object_color("object:color");
 ustring OSLRenderServices::u_object_index("object:index");
 ustring OSLRenderServices::u_geom_dupli_generated("geom:dupli_generated");
 ustring OSLRenderServices::u_geom_dupli_uv("geom:dupli_uv");
@@ -668,6 +669,10 @@ bool OSLRenderServices::get_object_standard_attribute(
     float3 f = object_location(kg, sd);
     return set_attribute_float3(f, type, derivatives, val);
   }
+  else if (name == u_object_color) {
+    float3 f = object_color(kg, sd->object);
+    return set_attribute_float3(f, type, derivatives, val);
+  }
   else if (name == u_object_index) {
     float f = object_pass_id(kg, sd->object);
     return set_attribute_float(f, type, derivatives, val);
@@ -697,7 +702,7 @@ bool OSLRenderServices::get_object_standard_attribute(
   }
   else if (name == u_particle_random) {
     int particle_id = object_particle_id(kg, sd->object);
-    float f = hash_int_01(particle_index(kg, particle_id));
+    float f = hash_uint2_to_float(particle_index(kg, particle_id), 0);
     return set_attribute_float(f, type, derivatives, val);
   }
 
@@ -1401,7 +1406,7 @@ bool OSLRenderServices::trace(TraceOpt &options,
 
   /* Raytrace, leaving out shadow opaque to avoid early exit. */
   uint visibility = PATH_RAY_ALL_VISIBILITY - PATH_RAY_SHADOW_OPAQUE;
-  return scene_intersect(kg, ray, visibility, &tracedata->isect);
+  return scene_intersect(kg, &ray, visibility, &tracedata->isect);
 }
 
 bool OSLRenderServices::getmessage(OSL::ShaderGlobals *sg,
diff --git a/intern/cycles/kernel/osl/osl_services.h b/intern/cycles/kernel/osl/osl_services.h
index 024ef656be1..469c5188730 100644
--- a/intern/cycles/kernel/osl/osl_services.h
+++ b/intern/cycles/kernel/osl/osl_services.h
@@ -245,6 +245,7 @@ class OSLRenderServices : public OSL::RendererServices {
   static ustring u_raster;
   static ustring u_ndc;
   static ustring u_object_location;
+  static ustring u_object_color;
   static ustring u_object_index;
   static ustring u_geom_dupli_generated;
   static ustring u_geom_dupli_uv;
diff --git a/intern/cycles/kernel/shaders/CMakeLists.txt b/intern/cycles/kernel/shaders/CMakeLists.txt
index 63cef6e841b..c50bffe27b2 100644
--- a/intern/cycles/kernel/shaders/CMakeLists.txt
+++ b/intern/cycles/kernel/shaders/CMakeLists.txt
@@ -85,6 +85,7 @@ set(SRC_OSL
   node_wavelength.osl
   node_blackbody.osl
   node_wave_texture.osl
+  node_white_noise_texture.osl
   node_wireframe.osl
   node_hair_bsdf.osl
   node_principled_hair_bsdf.osl
diff --git a/intern/cycles/kernel/shaders/node_math.osl b/intern/cycles/kernel/shaders/node_math.osl
index 8830339e05f..13e4c91ba10 100644
--- a/intern/cycles/kernel/shaders/node_math.osl
+++ b/intern/cycles/kernel/shaders/node_math.osl
@@ -18,56 +18,30 @@
 
 float safe_divide(float a, float b)
 {
-  float result;
-
-  if (b == 0.0)
-    result = 0.0;
-  else
-    result = a / b;
-
-  return result;
+  return (b != 0.0) ? a / b : 0.0;
 }
 
 float safe_modulo(float a, float b)
 {
-  float result;
-
-  if (b == 0.0)
-    result = 0.0;
-  else
-    result = fmod(a, b);
-
-  return result;
+  return (b != 0.0) ? fmod(a, b) : 0.0;
 }
 
 float safe_sqrt(float a)
 {
-  float result;
-
-  if (a > 0.0)
-    result = sqrt(a);
-  else
-    result = 0.0;
-
-  return result;
+  return (a > 0.0) ? sqrt(a) : 0.0;
 }
 
 float safe_log(float a, float b)
 {
-  if (a < 0.0 || b < 0.0)
-    return 0.0;
-
-  return log(a) / log(b);
+  return (a > 0.0 && b > 0.0) ? log(a) / log(b) : 0.0;
 }
 
+/* OSL asin, acos, and pow functions are safe by default. */
 shader node_math(string type = "add",
-                 int use_clamp = 0,
-                 float Value1 = 0.0,
-                 float Value2 = 0.0,
+                 float Value1 = 0.5,
+                 float Value2 = 0.5,
                  output float Value = 0.0)
 {
-  /* OSL asin, acos, pow check for values that could give rise to nan */
-
   if (type == "add")
     Value = Value1 + Value2;
   else if (type == "subtract")
@@ -76,47 +50,46 @@ shader node_math(string type = "add",
     Value = Value1 * Value2;
   else if (type == "divide")
     Value = safe_divide(Value1, Value2);
-  else if (type == "sine")
-    Value = sin(Value1);
-  else if (type == "cosine")
-    Value = cos(Value1);
-  else if (type == "tangent")
-    Value = tan(Value1);
-  else if (type == "arcsine")
-    Value = asin(Value1);
-  else if (type == "arccosine")
-    Value = acos(Value1);
-  else if (type == "arctangent")
-    Value = atan(Value1);
   else if (type == "power")
     Value = pow(Value1, Value2);
   else if (type == "logarithm")
     Value = safe_log(Value1, Value2);
+  else if (type == "sqrt")
+    Value = safe_sqrt(Value1);
+  else if (type == "absolute")
+    Value = fabs(Value1);
   else if (type == "minimum")
     Value = min(Value1, Value2);
   else if (type == "maximum")
     Value = max(Value1, Value2);
-  else if (type == "round")
-    Value = floor(Value1 + 0.5);
   else if (type == "less_than")
     Value = Value1 < Value2;
   else if (type == "greater_than")
     Value = Value1 > Value2;
-  else if (type == "modulo")
-    Value = safe_modulo(Value1, Value2);
-  else if (type == "absolute")
-    Value = fabs(Value1);
-  else if (type == "arctan2")
-    Value = atan2(Value1, Value2);
+  else if (type == "round")
+    Value = floor(Value1 + 0.5);
   else if (type == "floor")
     Value = floor(Value1);
   else if (type == "ceil")
     Value = ceil(Value1);
-  else if (type == "fract")
+  else if (type == "fraction")
     Value = Value1 - floor(Value1);
-  else if (type == "sqrt")
-    Value = safe_sqrt(Value1);
-
-  if (use_clamp)
-    Value = clamp(Value, 0.0, 1.0);
+  else if (type == "modulo")
+    Value = safe_modulo(Value1, Value2);
+  else if (type == "sine")
+    Value = sin(Value1);
+  else if (type == "cosine")
+    Value = cos(Value1);
+  else if (type == "tangent")
+    Value = tan(Value1);
+  else if (type == "arcsine")
+    Value = asin(Value1);
+  else if (type == "arccosine")
+    Value = acos(Value1);
+  else if (type == "arctangent")
+    Value = atan(Value1);
+  else if (type == "arctan2")
+    Value = atan2(Value1, Value2);
+  else
+    warning("%s", "Unknown math operator!");
 }
diff --git a/intern/cycles/kernel/shaders/node_object_info.osl b/intern/cycles/kernel/shaders/node_object_info.osl
index 0904a30a53f..350404bb747 100644
--- a/intern/cycles/kernel/shaders/node_object_info.osl
+++ b/intern/cycles/kernel/shaders/node_object_info.osl
@@ -17,11 +17,13 @@
 #include "stdosl.h"
 
 shader node_object_info(output point Location = point(0.0, 0.0, 0.0),
+                        output color Color = color(1.0, 1.0, 1.0),
                         output float ObjectIndex = 0.0,
                         output float MaterialIndex = 0.0,
                         output float Random = 0.0)
 {
   getattribute("object:location", Location);
+  getattribute("object:color", Color);
   getattribute("object:index", ObjectIndex);
   getattribute("material:index", MaterialIndex);
   getattribute("object:random", Random);
diff --git a/intern/cycles/kernel/shaders/node_vector_math.osl b/intern/cycles/kernel/shaders/node_vector_math.osl
index 10bb0c7283c..fd5e27aa144 100644
--- a/intern/cycles/kernel/shaders/node_vector_math.osl
+++ b/intern/cycles/kernel/shaders/node_vector_math.osl
@@ -16,34 +16,97 @@
 
 #include "stdosl.h"
 
+float safe_divide(float a, float b)
+{
+  return (b != 0.0) ? a / b : 0.0;
+}
+
+vector safe_divide(vector a, vector b)
+{
+  return vector((b[0] != 0.0) ? a[0] / b[0] : 0.0,
+                (b[1] != 0.0) ? a[1] / b[1] : 0.0,
+                (b[2] != 0.0) ? a[2] / b[2] : 0.0);
+}
+
+vector project(vector v, vector v_proj)
+{
+  float lenSquared = dot(v_proj, v_proj);
+  return (lenSquared != 0.0) ? (dot(v, v_proj) / lenSquared) * v_proj : vector(0.0);
+}
+
+vector snap(vector a, vector b)
+{
+  return floor(safe_divide(a, b)) * b;
+}
+
 shader node_vector_math(string type = "add",
                         vector Vector1 = vector(0.0, 0.0, 0.0),
                         vector Vector2 = vector(0.0, 0.0, 0.0),
+                        float Scale = 1.0,
                         output float Value = 0.0,
                         output vector Vector = vector(0.0, 0.0, 0.0))
 {
   if (type == "add") {
     Vector = Vector1 + Vector2;
-    Value = (abs(Vector[0]) + abs(Vector[1]) + abs(Vector[2])) / 3.0;
   }
   else if (type == "subtract") {
     Vector = Vector1 - Vector2;
-    Value = (abs(Vector[0]) + abs(Vector[1]) + abs(Vector[2])) / 3.0;
   }
-  else if (type == "average") {
-    Value = length(Vector1 + Vector2);
-    Vector = normalize(Vector1 + Vector2);
+  else if (type == "multiply") {
+    Vector = Vector1 * Vector2;
+  }
+  else if (type == "divide") {
+    Vector = safe_divide(Vector1, Vector2);
+  }
+  else if (type == "cross_product") {
+    Vector = cross(Vector1, Vector2);
+  }
+  else if (type == "project") {
+    Vector = project(Vector1, Vector2);
+  }
+  else if (type == "reflect") {
+    Vector = reflect(Vector1, normalize(Vector2));
   }
   else if (type == "dot_product") {
     Value = dot(Vector1, Vector2);
   }
-  else if (type == "cross_product") {
-    vector c = cross(Vector1, Vector2);
-    Value = length(c);
-    Vector = normalize(c);
+  else if (type == "distance") {
+    Value = distance(Vector1, Vector2);
   }
-  else if (type == "normalize") {
+  else if (type == "length") {
     Value = length(Vector1);
+  }
+  else if (type == "scale") {
+    Vector = Vector1 * Scale;
+  }
+  else if (type == "normalize") {
     Vector = normalize(Vector1);
   }
+  else if (type == "snap") {
+    Vector = snap(Vector1, Vector2);
+  }
+  else if (type == "floor") {
+    Vector = floor(Vector1);
+  }
+  else if (type == "ceil") {
+    Vector = ceil(Vector1);
+  }
+  else if (type == "modulo") {
+    Vector = mod(Vector1, Vector2);
+  }
+  else if (type == "fraction") {
+    Vector = Vector1 - floor(Vector1);
+  }
+  else if (type == "absolute") {
+    Vector = abs(Vector1);
+  }
+  else if (type == "minimum") {
+    Vector = min(Vector1, Vector2);
+  }
+  else if (type == "maximum") {
+    Vector = max(Vector1, Vector2);
+  }
+  else {
+    warning("%s", "Unknown vector math operator!");
+  }
 }
diff --git a/intern/cycles/kernel/shaders/node_white_noise_texture.osl b/intern/cycles/kernel/shaders/node_white_noise_texture.osl
new file mode 100644
index 00000000000..f026fb4ab39
--- /dev/null
+++ b/intern/cycles/kernel/shaders/node_white_noise_texture.osl
@@ -0,0 +1,39 @@
+/*
+ * Copyright 2011-2013 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "stdosl.h"
+
+shader node_white_noise_texture(string dimensions = "3D",
+                                point Vector = point(0.0, 0.0, 0.0),
+                                float W = 0.0,
+                                output float Value = 0.0)
+{
+  if (dimensions == "1D") {
+    Value = noise("hash", W);
+  }
+  else if (dimensions == "2D") {
+    Value = noise("hash", Vector[0], Vector[1]);
+  }
+  else if (dimensions == "3D") {
+    Value = noise("hash", Vector);
+  }
+  else if (dimensions == "4D") {
+    Value = noise("hash", Vector, W);
+  }
+  else {
+    warning("%s", "Unknown dimension!");
+  }
+}
diff --git a/intern/cycles/kernel/split/kernel_direct_lighting.h b/intern/cycles/kernel/split/kernel_direct_lighting.h
index b2ca59d60cc..3be2b35812f 100644
--- a/intern/cycles/kernel/split/kernel_direct_lighting.h
+++ b/intern/cycles/kernel/split/kernel_direct_lighting.h
@@ -86,8 +86,7 @@ ccl_device void kernel_direct_lighting(KernelGlobals *kg,
       float terminate = path_state_rng_light_termination(kg, state);
 
       LightSample ls;
-      if (light_sample(kg, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
-
+      if (light_sample(kg, -1, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
         Ray light_ray;
         light_ray.time = sd->time;
 
diff --git a/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h b/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
index 5cd4131e2ae..56cdb22bba3 100644
--- a/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
+++ b/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
@@ -132,10 +132,12 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(
         }
       }
 
+#ifdef __DENOISING_FEATURES__
       if (IS_STATE(ray_state, ray_index, RAY_ACTIVE)) {
         PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
         kernel_update_denoising_features(kg, sd, state, L);
       }
+#endif
     }
 
 #ifdef __AO__
diff --git a/intern/cycles/kernel/svm/svm.h b/intern/cycles/kernel/svm/svm.h
index 0d731d62e94..ab8570618ab 100644
--- a/intern/cycles/kernel/svm/svm.h
+++ b/intern/cycles/kernel/svm/svm.h
@@ -132,16 +132,25 @@ ccl_device_inline float4 fetch_node_float(KernelGlobals *kg, int offset)
                      __uint_as_float(node.w));
 }
 
-ccl_device_inline void decode_node_uchar4(uint i, uint *x, uint *y, uint *z, uint *w)
+ccl_device_forceinline void svm_unpack_node_uchar2(uint i, uint *x, uint *y)
 {
-  if (x)
-    *x = (i & 0xFF);
-  if (y)
-    *y = ((i >> 8) & 0xFF);
-  if (z)
-    *z = ((i >> 16) & 0xFF);
-  if (w)
-    *w = ((i >> 24) & 0xFF);
+  *x = (i & 0xFF);
+  *y = ((i >> 8) & 0xFF);
+}
+
+ccl_device_forceinline void svm_unpack_node_uchar3(uint i, uint *x, uint *y, uint *z)
+{
+  *x = (i & 0xFF);
+  *y = ((i >> 8) & 0xFF);
+  *z = ((i >> 16) & 0xFF);
+}
+
+ccl_device_forceinline void svm_unpack_node_uchar4(uint i, uint *x, uint *y, uint *z, uint *w)
+{
+  *x = (i & 0xFF);
+  *y = ((i >> 8) & 0xFF);
+  *z = ((i >> 16) & 0xFF);
+  *w = ((i >> 24) & 0xFF);
 }
 
 CCL_NAMESPACE_END
@@ -194,6 +203,7 @@ CCL_NAMESPACE_END
 #include "kernel/svm/svm_bump.h"
 #include "kernel/svm/svm_map_range.h"
 #include "kernel/svm/svm_clamp.h"
+#include "kernel/svm/svm_white_noise.h"
 
 #ifdef __SHADER_RAYTRACE__
 #  include "kernel/svm/svm_ao.h"
@@ -432,6 +442,9 @@ ccl_device_noinline void svm_eval_nodes(KernelGlobals *kg,
       case NODE_TEX_BRICK:
         svm_node_tex_brick(kg, sd, stack, node, &offset);
         break;
+      case NODE_TEX_WHITE_NOISE:
+        svm_node_tex_white_noise(kg, sd, stack, node.y, node.z, node.w, &offset);
+        break;
 #  endif /* __TEXTURES__ */
 #  ifdef __EXTRA_NODES__
       case NODE_NORMAL:
diff --git a/intern/cycles/kernel/svm/svm_ao.h b/intern/cycles/kernel/svm/svm_ao.h
index 62413979201..4cb986b897a 100644
--- a/intern/cycles/kernel/svm/svm_ao.h
+++ b/intern/cycles/kernel/svm/svm_ao.h
@@ -16,6 +16,8 @@
 
 CCL_NAMESPACE_BEGIN
 
+#ifdef __SHADER_RAYTRACE__
+
 ccl_device_noinline float svm_ao(KernelGlobals *kg,
                                  ShaderData *sd,
                                  float3 N,
@@ -64,13 +66,13 @@ ccl_device_noinline float svm_ao(KernelGlobals *kg,
     ray.dD = differential3_zero();
 
     if (flags & NODE_AO_ONLY_LOCAL) {
-      if (!scene_intersect_local(kg, ray, NULL, sd->object, NULL, 0)) {
+      if (!scene_intersect_local(kg, &ray, NULL, sd->object, NULL, 0)) {
         unoccluded++;
       }
     }
     else {
       Intersection isect;
-      if (!scene_intersect(kg, ray, PATH_RAY_SHADOW_OPAQUE, &isect)) {
+      if (!scene_intersect(kg, &ray, PATH_RAY_SHADOW_OPAQUE, &isect)) {
         unoccluded++;
       }
     }
@@ -83,10 +85,10 @@ ccl_device void svm_node_ao(
     KernelGlobals *kg, ShaderData *sd, ccl_addr_space PathState *state, float *stack, uint4 node)
 {
   uint flags, dist_offset, normal_offset, out_ao_offset;
-  decode_node_uchar4(node.y, &flags, &dist_offset, &normal_offset, &out_ao_offset);
+  svm_unpack_node_uchar4(node.y, &flags, &dist_offset, &normal_offset, &out_ao_offset);
 
   uint color_offset, out_color_offset, samples;
-  decode_node_uchar4(node.z, &color_offset, &out_color_offset, &samples, NULL);
+  svm_unpack_node_uchar3(node.z, &color_offset, &out_color_offset, &samples);
 
   float dist = stack_load_float_default(stack, dist_offset, node.w);
   float3 normal = stack_valid(normal_offset) ? stack_load_float3(stack, normal_offset) : sd->N;
@@ -102,4 +104,6 @@ ccl_device void svm_node_ao(
   }
 }
 
+#endif /* __SHADER_RAYTRACE__ */
+
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/svm/svm_attribute.h b/intern/cycles/kernel/svm/svm_attribute.h
index a67cfe91a30..eaee0f9e4ee 100644
--- a/intern/cycles/kernel/svm/svm_attribute.h
+++ b/intern/cycles/kernel/svm/svm_attribute.h
@@ -46,8 +46,8 @@ ccl_device AttributeDescriptor svm_node_attr_init(
 
 ccl_device void svm_node_attr(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
-  NodeAttributeType type;
-  uint out_offset;
+  NodeAttributeType type = NODE_ATTR_FLOAT;
+  uint out_offset = 0;
   AttributeDescriptor desc = svm_node_attr_init(kg, sd, node, &type, &out_offset);
 
   /* fetch and store attribute */
@@ -80,16 +80,10 @@ ccl_device void svm_node_attr(KernelGlobals *kg, ShaderData *sd, float *stack, u
   }
 }
 
-#ifndef __KERNEL_CUDA__
-ccl_device
-#else
-ccl_device_noinline
-#endif
-    void
-    svm_node_attr_bump_dx(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
+ccl_device void svm_node_attr_bump_dx(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
-  NodeAttributeType type;
-  uint out_offset;
+  NodeAttributeType type = NODE_ATTR_FLOAT;
+  uint out_offset = 0;
   AttributeDescriptor desc = svm_node_attr_init(kg, sd, node, &type, &out_offset);
 
   /* fetch and store attribute */
@@ -125,16 +119,10 @@ ccl_device_noinline
   }
 }
 
-#ifndef __KERNEL_CUDA__
-ccl_device
-#else
-ccl_device_noinline
-#endif
-    void
-    svm_node_attr_bump_dy(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
+ccl_device void svm_node_attr_bump_dy(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
-  NodeAttributeType type;
-  uint out_offset;
+  NodeAttributeType type = NODE_ATTR_FLOAT;
+  uint out_offset = 0;
   AttributeDescriptor desc = svm_node_attr_init(kg, sd, node, &type, &out_offset);
 
   /* fetch and store attribute */
diff --git a/intern/cycles/kernel/svm/svm_bevel.h b/intern/cycles/kernel/svm/svm_bevel.h
index fcf28e96e98..434502f31f9 100644
--- a/intern/cycles/kernel/svm/svm_bevel.h
+++ b/intern/cycles/kernel/svm/svm_bevel.h
@@ -16,6 +16,8 @@
 
 CCL_NAMESPACE_BEGIN
 
+#ifdef __SHADER_RAYTRACE__
+
 /* Bevel shader averaging normals from nearby surfaces.
  *
  * Sampling strategy from: BSSRDF Importance Sampling, SIGGRAPH 2013
@@ -51,7 +53,7 @@ ccl_device_noinline float3 svm_bevel(KernelGlobals *kg,
   float3 sum_N = make_float3(0.0f, 0.0f, 0.0f);
 
   for (int sample = 0; sample < num_samples; sample++) {
-    float disk_u, disk_v;
+    float disk_u = 0.0f, disk_v = 0.0f;
     path_branched_rng_2D(
         kg, state->rng_hash, state, sample, num_samples, PRNG_BEVEL_U, &disk_u, &disk_v);
 
@@ -110,7 +112,7 @@ ccl_device_noinline float3 svm_bevel(KernelGlobals *kg,
 
     /* Intersect with the same object. if multiple intersections are found it
      * will use at most LOCAL_MAX_HITS hits, a random subset of all hits. */
-    scene_intersect_local(kg, *ray, &isect, sd->object, &lcg_state, LOCAL_MAX_HITS);
+    scene_intersect_local(kg, ray, &isect, sd->object, &lcg_state, LOCAL_MAX_HITS);
 
     int num_eval_hits = min(isect.num_hits, LOCAL_MAX_HITS);
 
@@ -120,14 +122,14 @@ ccl_device_noinline float3 svm_bevel(KernelGlobals *kg,
       if (sd->type & PRIMITIVE_TRIANGLE) {
         hit_P = triangle_refine_local(kg, sd, &isect.hits[hit], ray);
       }
-#ifdef __OBJECT_MOTION__
+#  ifdef __OBJECT_MOTION__
       else if (sd->type & PRIMITIVE_MOTION_TRIANGLE) {
         float3 verts[3];
         motion_triangle_vertices(
             kg, sd->object, kernel_tex_fetch(__prim_index, isect.hits[hit].prim), sd->time, verts);
         hit_P = motion_triangle_refine_local(kg, sd, &isect.hits[hit], ray, verts);
       }
-#endif /* __OBJECT_MOTION__ */
+#  endif /* __OBJECT_MOTION__ */
 
       /* Get geometric normal. */
       float3 hit_Ng = isect.Ng[hit];
@@ -151,11 +153,11 @@ ccl_device_noinline float3 svm_bevel(KernelGlobals *kg,
         if (sd->type & PRIMITIVE_TRIANGLE) {
           N = triangle_smooth_normal(kg, N, prim, u, v);
         }
-#ifdef __OBJECT_MOTION__
+#  ifdef __OBJECT_MOTION__
         else if (sd->type & PRIMITIVE_MOTION_TRIANGLE) {
           N = motion_triangle_smooth_normal(kg, N, sd->object, prim, u, v, sd->time);
         }
-#endif /* __OBJECT_MOTION__ */
+#  endif /* __OBJECT_MOTION__ */
       }
 
       /* Transform normals to world space. */
@@ -200,7 +202,7 @@ ccl_device void svm_node_bevel(
     KernelGlobals *kg, ShaderData *sd, ccl_addr_space PathState *state, float *stack, uint4 node)
 {
   uint num_samples, radius_offset, normal_offset, out_offset;
-  decode_node_uchar4(node.y, &num_samples, &radius_offset, &normal_offset, &out_offset);
+  svm_unpack_node_uchar4(node.y, &num_samples, &radius_offset, &normal_offset, &out_offset);
 
   float radius = stack_load_float(stack, radius_offset);
   float3 bevel_N = svm_bevel(kg, sd, state, radius, num_samples);
@@ -214,4 +216,6 @@ ccl_device void svm_node_bevel(
   stack_store_float3(stack, out_offset, bevel_N);
 }
 
+#endif /* __SHADER_RAYTRACE__ */
+
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/svm/svm_brick.h b/intern/cycles/kernel/svm/svm_brick.h
index b5cbfcc72df..6984afa30a5 100644
--- a/intern/cycles/kernel/svm/svm_brick.h
+++ b/intern/cycles/kernel/svm/svm_brick.h
@@ -18,7 +18,7 @@ CCL_NAMESPACE_BEGIN
 
 /* Brick */
 
-ccl_device_noinline float brick_noise(uint n) /* fast integer noise */
+ccl_device_inline float brick_noise(uint n) /* fast integer noise */
 {
   uint nn;
   n = (n + 1013) & 0x7fffffff;
@@ -27,16 +27,16 @@ ccl_device_noinline float brick_noise(uint n) /* fast integer noise */
   return 0.5f * ((float)nn / 1073741824.0f);
 }
 
-ccl_device_noinline float2 svm_brick(float3 p,
-                                     float mortar_size,
-                                     float mortar_smooth,
-                                     float bias,
-                                     float brick_width,
-                                     float row_height,
-                                     float offset_amount,
-                                     int offset_frequency,
-                                     float squash_amount,
-                                     int squash_frequency)
+ccl_device_noinline_cpu float2 svm_brick(float3 p,
+                                         float mortar_size,
+                                         float mortar_smooth,
+                                         float bias,
+                                         float brick_width,
+                                         float row_height,
+                                         float offset_amount,
+                                         int offset_frequency,
+                                         float squash_amount,
+                                         int squash_frequency)
 {
   int bricknum, rownum;
   float offset = 0.0f;
@@ -87,13 +87,13 @@ ccl_device void svm_node_tex_brick(
   /* RNA properties */
   uint offset_frequency, squash_frequency;
 
-  decode_node_uchar4(node.y, &co_offset, &color1_offset, &color2_offset, &mortar_offset);
-  decode_node_uchar4(
+  svm_unpack_node_uchar4(node.y, &co_offset, &color1_offset, &color2_offset, &mortar_offset);
+  svm_unpack_node_uchar4(
       node.z, &scale_offset, &mortar_size_offset, &bias_offset, &brick_width_offset);
-  decode_node_uchar4(
+  svm_unpack_node_uchar4(
       node.w, &row_height_offset, &color_offset, &fac_offset, &mortar_smooth_offset);
 
-  decode_node_uchar4(node2.x, &offset_frequency, &squash_frequency, NULL, NULL);
+  svm_unpack_node_uchar2(node2.x, &offset_frequency, &squash_frequency);
 
   float3 co = stack_load_float3(stack, co_offset);
 
diff --git a/intern/cycles/kernel/svm/svm_brightness.h b/intern/cycles/kernel/svm/svm_brightness.h
index dcd75a2fe8f..9554b5946fb 100644
--- a/intern/cycles/kernel/svm/svm_brightness.h
+++ b/intern/cycles/kernel/svm/svm_brightness.h
@@ -22,7 +22,7 @@ ccl_device void svm_node_brightness(
   uint bright_offset, contrast_offset;
   float3 color = stack_load_float3(stack, in_color);
 
-  decode_node_uchar4(node, &bright_offset, &contrast_offset, NULL, NULL);
+  svm_unpack_node_uchar2(node, &bright_offset, &contrast_offset);
   float brightness = stack_load_float(stack, bright_offset);
   float contrast = stack_load_float(stack, contrast_offset);
 
diff --git a/intern/cycles/kernel/svm/svm_checker.h b/intern/cycles/kernel/svm/svm_checker.h
index 63b4d1e149b..d54cb73df91 100644
--- a/intern/cycles/kernel/svm/svm_checker.h
+++ b/intern/cycles/kernel/svm/svm_checker.h
@@ -18,7 +18,7 @@ CCL_NAMESPACE_BEGIN
 
 /* Checker */
 
-ccl_device_noinline float svm_checker(float3 p)
+ccl_device float svm_checker(float3 p)
 {
   /* avoid precision issues on unit coordinates */
   p.x = (p.x + 0.000001f) * 0.999999f;
@@ -37,8 +37,8 @@ ccl_device void svm_node_tex_checker(KernelGlobals *kg, ShaderData *sd, float *s
   uint co_offset, color1_offset, color2_offset, scale_offset;
   uint color_offset, fac_offset;
 
-  decode_node_uchar4(node.y, &co_offset, &color1_offset, &color2_offset, &scale_offset);
-  decode_node_uchar4(node.z, &color_offset, &fac_offset, NULL, NULL);
+  svm_unpack_node_uchar4(node.y, &co_offset, &color1_offset, &color2_offset, &scale_offset);
+  svm_unpack_node_uchar2(node.z, &color_offset, &fac_offset);
 
   float3 co = stack_load_float3(stack, co_offset);
   float3 color1 = stack_load_float3(stack, color1_offset);
diff --git a/intern/cycles/kernel/svm/svm_clamp.h b/intern/cycles/kernel/svm/svm_clamp.h
index 5ff4a599028..a45e70a3f15 100644
--- a/intern/cycles/kernel/svm/svm_clamp.h
+++ b/intern/cycles/kernel/svm/svm_clamp.h
@@ -27,7 +27,7 @@ ccl_device void svm_node_clamp(KernelGlobals *kg,
                                int *offset)
 {
   uint min_stack_offset, max_stack_offset;
-  decode_node_uchar4(parameters_stack_offsets, &min_stack_offset, &max_stack_offset, NULL, NULL);
+  svm_unpack_node_uchar2(parameters_stack_offsets, &min_stack_offset, &max_stack_offset);
 
   uint4 defaults = read_node(kg, offset);
 
diff --git a/intern/cycles/kernel/svm/svm_closure.h b/intern/cycles/kernel/svm/svm_closure.h
index 270fe4c8615..1511fc65835 100644
--- a/intern/cycles/kernel/svm/svm_closure.h
+++ b/intern/cycles/kernel/svm/svm_closure.h
@@ -85,7 +85,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg,
   uint type, param1_offset, param2_offset;
 
   uint mix_weight_offset;
-  decode_node_uchar4(node.y, &type, &param1_offset, &param2_offset, &mix_weight_offset);
+  svm_unpack_node_uchar4(node.y, &type, &param1_offset, &param2_offset, &mix_weight_offset);
   float mix_weight = (stack_valid(mix_weight_offset) ? stack_load_float(stack, mix_weight_offset) :
                                                        1.0f);
 
@@ -122,21 +122,21 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg,
       uint4 data_node2 = read_node(kg, offset);
 
       float3 T = stack_load_float3(stack, data_node.y);
-      decode_node_uchar4(data_node.z,
-                         &specular_offset,
-                         &roughness_offset,
-                         &specular_tint_offset,
-                         &anisotropic_offset);
-      decode_node_uchar4(data_node.w,
-                         &sheen_offset,
-                         &sheen_tint_offset,
-                         &clearcoat_offset,
-                         &clearcoat_roughness_offset);
-      decode_node_uchar4(data_node2.x,
-                         &eta_offset,
-                         &transmission_offset,
-                         &anisotropic_rotation_offset,
-                         &transmission_roughness_offset);
+      svm_unpack_node_uchar4(data_node.z,
+                             &specular_offset,
+                             &roughness_offset,
+                             &specular_tint_offset,
+                             &anisotropic_offset);
+      svm_unpack_node_uchar4(data_node.w,
+                             &sheen_offset,
+                             &sheen_tint_offset,
+                             &clearcoat_offset,
+                             &clearcoat_roughness_offset);
+      svm_unpack_node_uchar4(data_node2.x,
+                             &eta_offset,
+                             &transmission_offset,
+                             &anisotropic_rotation_offset,
+                             &transmission_roughness_offset);
 
       // get Disney principled parameters
       float metallic = param1;
@@ -793,19 +793,19 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg,
       float3 weight = sd->svm_closure_weight * mix_weight;
 
       uint offset_ofs, ior_ofs, color_ofs, parametrization;
-      decode_node_uchar4(data_node.y, &offset_ofs, &ior_ofs, &color_ofs, &parametrization);
+      svm_unpack_node_uchar4(data_node.y, &offset_ofs, &ior_ofs, &color_ofs, &parametrization);
       float alpha = stack_load_float_default(stack, offset_ofs, data_node.z);
       float ior = stack_load_float_default(stack, ior_ofs, data_node.w);
 
       uint coat_ofs, melanin_ofs, melanin_redness_ofs, absorption_coefficient_ofs;
-      decode_node_uchar4(data_node2.x,
-                         &coat_ofs,
-                         &melanin_ofs,
-                         &melanin_redness_ofs,
-                         &absorption_coefficient_ofs);
+      svm_unpack_node_uchar4(data_node2.x,
+                             &coat_ofs,
+                             &melanin_ofs,
+                             &melanin_redness_ofs,
+                             &absorption_coefficient_ofs);
 
       uint tint_ofs, random_ofs, random_color_ofs, random_roughness_ofs;
-      decode_node_uchar4(
+      svm_unpack_node_uchar4(
           data_node3.x, &tint_ofs, &random_ofs, &random_color_ofs, &random_roughness_ofs);
 
       const AttributeDescriptor attr_descr_random = find_attribute(kg, sd, data_node4.y);
@@ -982,7 +982,7 @@ ccl_device void svm_node_closure_volume(
   uint type, density_offset, anisotropy_offset;
 
   uint mix_weight_offset;
-  decode_node_uchar4(node.y, &type, &density_offset, &anisotropy_offset, &mix_weight_offset);
+  svm_unpack_node_uchar4(node.y, &type, &density_offset, &anisotropy_offset, &mix_weight_offset);
   float mix_weight = (stack_valid(mix_weight_offset) ? stack_load_float(stack, mix_weight_offset) :
                                                        1.0f);
 
@@ -1040,7 +1040,7 @@ ccl_device void svm_node_principled_volume(KernelGlobals *kg,
   }
 
   uint density_offset, anisotropy_offset, absorption_color_offset, mix_weight_offset;
-  decode_node_uchar4(
+  svm_unpack_node_uchar4(
       node.y, &density_offset, &anisotropy_offset, &absorption_color_offset, &mix_weight_offset);
   float mix_weight = (stack_valid(mix_weight_offset) ? stack_load_float(stack, mix_weight_offset) :
                                                        1.0f);
@@ -1099,7 +1099,7 @@ ccl_device void svm_node_principled_volume(KernelGlobals *kg,
   }
 
   uint emission_offset, emission_color_offset, blackbody_offset, temperature_offset;
-  decode_node_uchar4(
+  svm_unpack_node_uchar4(
       node.z, &emission_offset, &emission_color_offset, &blackbody_offset, &temperature_offset);
   float emission = (stack_valid(emission_offset)) ? stack_load_float(stack, emission_offset) :
                                                     __uint_as_float(value_node.z);
@@ -1229,7 +1229,8 @@ ccl_device void svm_node_mix_closure(ShaderData *sd, float *stack, uint4 node)
   /* fetch weight from blend input, previous mix closures,
    * and write to stack to be used by closure nodes later */
   uint weight_offset, in_weight_offset, weight1_offset, weight2_offset;
-  decode_node_uchar4(node.y, &weight_offset, &in_weight_offset, &weight1_offset, &weight2_offset);
+  svm_unpack_node_uchar4(
+      node.y, &weight_offset, &in_weight_offset, &weight1_offset, &weight2_offset);
 
   float weight = stack_load_float(stack, weight_offset);
   weight = saturate(weight);
diff --git a/intern/cycles/kernel/svm/svm_color_util.h b/intern/cycles/kernel/svm/svm_color_util.h
index 12b59d2616b..3a6a5ba782f 100644
--- a/intern/cycles/kernel/svm/svm_color_util.h
+++ b/intern/cycles/kernel/svm/svm_color_util.h
@@ -264,7 +264,7 @@ ccl_device float3 svm_mix_clamp(float3 col)
   return outcol;
 }
 
-ccl_device_noinline float3 svm_mix(NodeMix type, float fac, float3 c1, float3 c2)
+ccl_device_noinline_cpu float3 svm_mix(NodeMix type, float fac, float3 c1, float3 c2)
 {
   float t = saturate(fac);
 
diff --git a/intern/cycles/kernel/svm/svm_displace.h b/intern/cycles/kernel/svm/svm_displace.h
index f16664a684c..250fac6bcb8 100644
--- a/intern/cycles/kernel/svm/svm_displace.h
+++ b/intern/cycles/kernel/svm/svm_displace.h
@@ -23,7 +23,7 @@ ccl_device void svm_node_set_bump(KernelGlobals *kg, ShaderData *sd, float *stac
 #ifdef __RAY_DIFFERENTIALS__
   /* get normal input */
   uint normal_offset, scale_offset, invert, use_object_space;
-  decode_node_uchar4(node.y, &normal_offset, &scale_offset, &invert, &use_object_space);
+  svm_unpack_node_uchar4(node.y, &normal_offset, &scale_offset, &invert, &use_object_space);
 
   float3 normal_in = stack_valid(normal_offset) ? stack_load_float3(stack, normal_offset) : sd->N;
 
@@ -42,7 +42,7 @@ ccl_device void svm_node_set_bump(KernelGlobals *kg, ShaderData *sd, float *stac
 
   /* get bump values */
   uint c_offset, x_offset, y_offset, strength_offset;
-  decode_node_uchar4(node.z, &c_offset, &x_offset, &y_offset, &strength_offset);
+  svm_unpack_node_uchar4(node.z, &c_offset, &x_offset, &y_offset, &strength_offset);
 
   float h_c = stack_load_float(stack, c_offset);
   float h_x = stack_load_float(stack, x_offset);
@@ -95,7 +95,7 @@ ccl_device void svm_node_set_displacement(KernelGlobals *kg,
 ccl_device void svm_node_displacement(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
   uint height_offset, midlevel_offset, scale_offset, normal_offset;
-  decode_node_uchar4(node.y, &height_offset, &midlevel_offset, &scale_offset, &normal_offset);
+  svm_unpack_node_uchar4(node.y, &height_offset, &midlevel_offset, &scale_offset, &normal_offset);
 
   float height = stack_load_float(stack, height_offset);
   float midlevel = stack_load_float(stack, midlevel_offset);
@@ -126,7 +126,7 @@ ccl_device void svm_node_vector_displacement(
   uint space = data_node.x;
 
   uint vector_offset, midlevel_offset, scale_offset, displacement_offset;
-  decode_node_uchar4(
+  svm_unpack_node_uchar4(
       node.y, &vector_offset, &midlevel_offset, &scale_offset, &displacement_offset);
 
   float3 vector = stack_load_float3(stack, vector_offset);
diff --git a/intern/cycles/kernel/svm/svm_fresnel.h b/intern/cycles/kernel/svm/svm_fresnel.h
index 03119991597..96d602e35bf 100644
--- a/intern/cycles/kernel/svm/svm_fresnel.h
+++ b/intern/cycles/kernel/svm/svm_fresnel.h
@@ -22,7 +22,7 @@ ccl_device void svm_node_fresnel(
     ShaderData *sd, float *stack, uint ior_offset, uint ior_value, uint node)
 {
   uint normal_offset, out_offset;
-  decode_node_uchar4(node, &normal_offset, &out_offset, NULL, NULL);
+  svm_unpack_node_uchar2(node, &normal_offset, &out_offset);
   float eta = (stack_valid(ior_offset)) ? stack_load_float(stack, ior_offset) :
                                           __uint_as_float(ior_value);
   float3 normal_in = stack_valid(normal_offset) ? stack_load_float3(stack, normal_offset) : sd->N;
@@ -43,7 +43,7 @@ ccl_device void svm_node_layer_weight(ShaderData *sd, float *stack, uint4 node)
   uint blend_value = node.z;
 
   uint type, normal_offset, out_offset;
-  decode_node_uchar4(node.w, &type, &normal_offset, &out_offset, NULL);
+  svm_unpack_node_uchar3(node.w, &type, &normal_offset, &out_offset);
 
   float blend = (stack_valid(blend_offset)) ? stack_load_float(stack, blend_offset) :
                                               __uint_as_float(blend_value);
diff --git a/intern/cycles/kernel/svm/svm_geometry.h b/intern/cycles/kernel/svm/svm_geometry.h
index a9104643299..019c6294082 100644
--- a/intern/cycles/kernel/svm/svm_geometry.h
+++ b/intern/cycles/kernel/svm/svm_geometry.h
@@ -113,6 +113,10 @@ ccl_device void svm_node_object_info(
       stack_store_float3(stack, out_offset, object_location(kg, sd));
       return;
     }
+    case NODE_INFO_OB_COLOR: {
+      stack_store_float3(stack, out_offset, object_color(kg, sd->object));
+      return;
+    }
     case NODE_INFO_OB_INDEX:
       data = object_pass_id(kg, sd->object);
       break;
@@ -149,7 +153,7 @@ ccl_device void svm_node_particle_info(
     }
     case NODE_INFO_PAR_RANDOM: {
       int particle_id = object_particle_id(kg, sd->object);
-      float random = hash_int_01(particle_index(kg, particle_id));
+      float random = hash_uint2_to_float(particle_index(kg, particle_id), 0);
       stack_store_float(stack, out_offset, random);
       break;
     }
diff --git a/intern/cycles/kernel/svm/svm_gradient.h b/intern/cycles/kernel/svm/svm_gradient.h
index c315564fbc2..08304bc47e8 100644
--- a/intern/cycles/kernel/svm/svm_gradient.h
+++ b/intern/cycles/kernel/svm/svm_gradient.h
@@ -64,7 +64,7 @@ ccl_device void svm_node_tex_gradient(ShaderData *sd, float *stack, uint4 node)
 {
   uint type, co_offset, color_offset, fac_offset;
 
-  decode_node_uchar4(node.y, &type, &co_offset, &fac_offset, &color_offset);
+  svm_unpack_node_uchar4(node.y, &type, &co_offset, &fac_offset, &color_offset);
 
   float3 co = stack_load_float3(stack, co_offset);
 
diff --git a/intern/cycles/kernel/svm/svm_hsv.h b/intern/cycles/kernel/svm/svm_hsv.h
index 72379fba870..1f7bd421869 100644
--- a/intern/cycles/kernel/svm/svm_hsv.h
+++ b/intern/cycles/kernel/svm/svm_hsv.h
@@ -24,8 +24,8 @@ ccl_device void svm_node_hsv(
 {
   uint in_color_offset, fac_offset, out_color_offset;
   uint hue_offset, sat_offset, val_offset;
-  decode_node_uchar4(node.y, &in_color_offset, &fac_offset, &out_color_offset, NULL);
-  decode_node_uchar4(node.z, &hue_offset, &sat_offset, &val_offset, NULL);
+  svm_unpack_node_uchar3(node.y, &in_color_offset, &fac_offset, &out_color_offset);
+  svm_unpack_node_uchar3(node.z, &hue_offset, &sat_offset, &val_offset);
 
   float fac = stack_load_float(stack, fac_offset);
   float3 in_color = stack_load_float3(stack, in_color_offset);
diff --git a/intern/cycles/kernel/svm/svm_ies.h b/intern/cycles/kernel/svm/svm_ies.h
index e57e54ef123..56c804b44d0 100644
--- a/intern/cycles/kernel/svm/svm_ies.h
+++ b/intern/cycles/kernel/svm/svm_ies.h
@@ -101,8 +101,8 @@ ccl_device_inline float kernel_ies_interp(KernelGlobals *kg,
 ccl_device void svm_node_ies(
     KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node, int *offset)
 {
-  uint vector_offset, strength_offset, fac_offset, dummy, slot = node.z;
-  decode_node_uchar4(node.y, &strength_offset, &vector_offset, &fac_offset, &dummy);
+  uint vector_offset, strength_offset, fac_offset, slot = node.z;
+  svm_unpack_node_uchar3(node.y, &strength_offset, &vector_offset, &fac_offset);
 
   float3 vector = stack_load_float3(stack, vector_offset);
   float strength = stack_load_float_default(stack, strength_offset, node.w);
diff --git a/intern/cycles/kernel/svm/svm_image.h b/intern/cycles/kernel/svm/svm_image.h
index 2ef64662d0e..64abdd2d8b3 100644
--- a/intern/cycles/kernel/svm/svm_image.h
+++ b/intern/cycles/kernel/svm/svm_image.h
@@ -16,6 +16,8 @@
 
 CCL_NAMESPACE_BEGIN
 
+#ifdef __TEXTURES__
+
 ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, uint flags)
 {
   float4 r = kernel_tex_image_interp(kg, id, x, y);
@@ -48,7 +50,7 @@ ccl_device void svm_node_tex_image(KernelGlobals *kg, ShaderData *sd, float *sta
   uint id = node.y;
   uint co_offset, out_offset, alpha_offset, flags;
 
-  decode_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &flags);
+  svm_unpack_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &flags);
 
   float3 co = stack_load_float3(stack, co_offset);
   float2 tex_co;
@@ -143,7 +145,7 @@ ccl_device void svm_node_tex_image_box(KernelGlobals *kg, ShaderData *sd, float
 
   /* now fetch textures */
   uint co_offset, out_offset, alpha_offset, flags;
-  decode_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &flags);
+  svm_unpack_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &flags);
 
   float3 co = stack_load_float3(stack, co_offset);
   uint id = node.y;
@@ -179,7 +181,7 @@ ccl_device void svm_node_tex_environment(KernelGlobals *kg,
   uint co_offset, out_offset, alpha_offset, flags;
   uint projection = node.w;
 
-  decode_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &flags);
+  svm_unpack_node_uchar4(node.z, &co_offset, &out_offset, &alpha_offset, &flags);
 
   float3 co = stack_load_float3(stack, co_offset);
   float2 uv;
@@ -199,4 +201,6 @@ ccl_device void svm_node_tex_environment(KernelGlobals *kg,
     stack_store_float(stack, alpha_offset, f.w);
 }
 
+#endif /* __TEXTURES__ */
+
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/svm/svm_light_path.h b/intern/cycles/kernel/svm/svm_light_path.h
index 65a9a284a17..768c65918cd 100644
--- a/intern/cycles/kernel/svm/svm_light_path.h
+++ b/intern/cycles/kernel/svm/svm_light_path.h
@@ -84,7 +84,7 @@ ccl_device void svm_node_light_falloff(ShaderData *sd, float *stack, uint4 node)
 {
   uint strength_offset, out_offset, smooth_offset;
 
-  decode_node_uchar4(node.z, &strength_offset, &smooth_offset, &out_offset, NULL);
+  svm_unpack_node_uchar3(node.z, &strength_offset, &smooth_offset, &out_offset);
 
   float strength = stack_load_float(stack, strength_offset);
   uint type = node.y;
diff --git a/intern/cycles/kernel/svm/svm_magic.h b/intern/cycles/kernel/svm/svm_magic.h
index 115d2e2fe4b..9c160e6d8cc 100644
--- a/intern/cycles/kernel/svm/svm_magic.h
+++ b/intern/cycles/kernel/svm/svm_magic.h
@@ -18,7 +18,7 @@ CCL_NAMESPACE_BEGIN
 
 /* Magic */
 
-ccl_device_noinline float3 svm_magic(float3 p, int n, float distortion)
+ccl_device_noinline_cpu float3 svm_magic(float3 p, int n, float distortion)
 {
   float x = sinf((p.x + p.y + p.z) * 5.0f);
   float y = cosf((-p.x + p.y - p.z) * 5.0f);
@@ -93,8 +93,8 @@ ccl_device void svm_node_tex_magic(
   uint depth;
   uint scale_offset, distortion_offset, co_offset, fac_offset, color_offset;
 
-  decode_node_uchar4(node.y, &depth, &color_offset, &fac_offset, NULL);
-  decode_node_uchar4(node.z, &co_offset, &scale_offset, &distortion_offset, NULL);
+  svm_unpack_node_uchar3(node.y, &depth, &color_offset, &fac_offset);
+  svm_unpack_node_uchar3(node.z, &co_offset, &scale_offset, &distortion_offset);
 
   uint4 node2 = read_node(kg, offset);
   float3 co = stack_load_float3(stack, co_offset);
diff --git a/intern/cycles/kernel/svm/svm_map_range.h b/intern/cycles/kernel/svm/svm_map_range.h
index 5d542e959d1..f2a68adbe61 100644
--- a/intern/cycles/kernel/svm/svm_map_range.h
+++ b/intern/cycles/kernel/svm/svm_map_range.h
@@ -27,11 +27,11 @@ ccl_device void svm_node_map_range(KernelGlobals *kg,
                                    int *offset)
 {
   uint from_min_stack_offset, from_max_stack_offset, to_min_stack_offset, to_max_stack_offset;
-  decode_node_uchar4(parameters_stack_offsets,
-                     &from_min_stack_offset,
-                     &from_max_stack_offset,
-                     &to_min_stack_offset,
-                     &to_max_stack_offset);
+  svm_unpack_node_uchar4(parameters_stack_offsets,
+                         &from_min_stack_offset,
+                         &from_max_stack_offset,
+                         &to_min_stack_offset,
+                         &to_max_stack_offset);
 
   uint4 defaults = read_node(kg, offset);
 
diff --git a/intern/cycles/kernel/svm/svm_math.h b/intern/cycles/kernel/svm/svm_math.h
index 5920913825b..d156dec497c 100644
--- a/intern/cycles/kernel/svm/svm_math.h
+++ b/intern/cycles/kernel/svm/svm_math.h
@@ -16,48 +16,50 @@
 
 CCL_NAMESPACE_BEGIN
 
-/* Nodes */
-
 ccl_device void svm_node_math(KernelGlobals *kg,
                               ShaderData *sd,
                               float *stack,
-                              uint itype,
-                              uint f1_offset,
-                              uint f2_offset,
+                              uint type,
+                              uint inputs_stack_offsets,
+                              uint result_stack_offset,
                               int *offset)
 {
-  NodeMath type = (NodeMath)itype;
-  float f1 = stack_load_float(stack, f1_offset);
-  float f2 = stack_load_float(stack, f2_offset);
-  float f = svm_math(type, f1, f2);
+  uint a_stack_offset, b_stack_offset;
+  svm_unpack_node_uchar2(inputs_stack_offsets, &a_stack_offset, &b_stack_offset);
 
-  uint4 node1 = read_node(kg, offset);
+  float a = stack_load_float(stack, a_stack_offset);
+  float b = stack_load_float(stack, b_stack_offset);
+  float result = svm_math((NodeMathType)type, a, b);
 
-  stack_store_float(stack, node1.y, f);
+  stack_store_float(stack, result_stack_offset, result);
 }
 
 ccl_device void svm_node_vector_math(KernelGlobals *kg,
                                      ShaderData *sd,
                                      float *stack,
-                                     uint itype,
-                                     uint v1_offset,
-                                     uint v2_offset,
+                                     uint type,
+                                     uint inputs_stack_offsets,
+                                     uint outputs_stack_offsets,
                                      int *offset)
 {
-  NodeVectorMath type = (NodeVectorMath)itype;
-  float3 v1 = stack_load_float3(stack, v1_offset);
-  float3 v2 = stack_load_float3(stack, v2_offset);
-  float f;
-  float3 v;
+  uint value_stack_offset, vector_stack_offset;
+  uint a_stack_offset, b_stack_offset, scale_stack_offset;
+  svm_unpack_node_uchar3(
+      inputs_stack_offsets, &a_stack_offset, &b_stack_offset, &scale_stack_offset);
+  svm_unpack_node_uchar2(outputs_stack_offsets, &value_stack_offset, &vector_stack_offset);
 
-  svm_vector_math(&f, &v, type, v1, v2);
+  float3 a = stack_load_float3(stack, a_stack_offset);
+  float3 b = stack_load_float3(stack, b_stack_offset);
+  float scale = stack_load_float(stack, scale_stack_offset);
 
-  uint4 node1 = read_node(kg, offset);
+  float value;
+  float3 vector;
+  svm_vector_math(&value, &vector, (NodeVectorMathType)type, a, b, scale);
 
-  if (stack_valid(node1.y))
-    stack_store_float(stack, node1.y, f);
-  if (stack_valid(node1.z))
-    stack_store_float3(stack, node1.z, v);
+  if (stack_valid(value_stack_offset))
+    stack_store_float(stack, value_stack_offset, value);
+  if (stack_valid(vector_stack_offset))
+    stack_store_float3(stack, vector_stack_offset, vector);
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/svm/svm_math_util.h b/intern/cycles/kernel/svm/svm_math_util.h
index e3544515f1b..c07a1e4ed98 100644
--- a/intern/cycles/kernel/svm/svm_math_util.h
+++ b/intern/cycles/kernel/svm/svm_math_util.h
@@ -16,99 +16,130 @@
 
 CCL_NAMESPACE_BEGIN
 
-ccl_device float average_fac(float3 v)
-{
-  return (fabsf(v.x) + fabsf(v.y) + fabsf(v.z)) / 3.0f;
-}
-
 ccl_device void svm_vector_math(
-    float *Fac, float3 *Vector, NodeVectorMath type, float3 Vector1, float3 Vector2)
+    float *value, float3 *vector, NodeVectorMathType type, float3 a, float3 b, float scale)
 {
-  if (type == NODE_VECTOR_MATH_ADD) {
-    *Vector = Vector1 + Vector2;
-    *Fac = average_fac(*Vector);
-  }
-  else if (type == NODE_VECTOR_MATH_SUBTRACT) {
-    *Vector = Vector1 - Vector2;
-    *Fac = average_fac(*Vector);
-  }
-  else if (type == NODE_VECTOR_MATH_AVERAGE) {
-    *Vector = safe_normalize_len(Vector1 + Vector2, Fac);
-  }
-  else if (type == NODE_VECTOR_MATH_DOT_PRODUCT) {
-    *Fac = dot(Vector1, Vector2);
-    *Vector = make_float3(0.0f, 0.0f, 0.0f);
-  }
-  else if (type == NODE_VECTOR_MATH_CROSS_PRODUCT) {
-    *Vector = safe_normalize_len(cross(Vector1, Vector2), Fac);
-  }
-  else if (type == NODE_VECTOR_MATH_NORMALIZE) {
-    *Vector = safe_normalize_len(Vector1, Fac);
-  }
-  else {
-    *Fac = 0.0f;
-    *Vector = make_float3(0.0f, 0.0f, 0.0f);
+  switch (type) {
+    case NODE_VECTOR_MATH_ADD:
+      *vector = a + b;
+      break;
+    case NODE_VECTOR_MATH_SUBTRACT:
+      *vector = a - b;
+      break;
+    case NODE_VECTOR_MATH_MULTIPLY:
+      *vector = a * b;
+      break;
+    case NODE_VECTOR_MATH_DIVIDE:
+      *vector = safe_divide_float3_float3(a, b);
+      break;
+    case NODE_VECTOR_MATH_CROSS_PRODUCT:
+      *vector = cross(a, b);
+      break;
+    case NODE_VECTOR_MATH_PROJECT:
+      *vector = project(a, b);
+      break;
+    case NODE_VECTOR_MATH_REFLECT:
+      *vector = reflect(a, b);
+      break;
+    case NODE_VECTOR_MATH_DOT_PRODUCT:
+      *value = dot(a, b);
+      break;
+    case NODE_VECTOR_MATH_DISTANCE:
+      *value = distance(a, b);
+      break;
+    case NODE_VECTOR_MATH_LENGTH:
+      *value = len(a);
+      break;
+    case NODE_VECTOR_MATH_SCALE:
+      *vector = a * scale;
+      break;
+    case NODE_VECTOR_MATH_NORMALIZE:
+      *vector = safe_normalize(a);
+      break;
+    case NODE_VECTOR_MATH_SNAP:
+      *vector = floor(safe_divide_float3_float3(a, b)) * b;
+      break;
+    case NODE_VECTOR_MATH_FLOOR:
+      *vector = floor(a);
+      break;
+    case NODE_VECTOR_MATH_CEIL:
+      *vector = ceil(a);
+      break;
+    case NODE_VECTOR_MATH_MODULO:
+      *vector = make_float3(safe_modulo(a.x, b.x), safe_modulo(a.y, b.y), safe_modulo(a.z, b.z));
+      break;
+    case NODE_VECTOR_MATH_FRACTION:
+      *vector = a - floor(a);
+      break;
+    case NODE_VECTOR_MATH_ABSOLUTE:
+      *vector = fabs(a);
+      break;
+    case NODE_VECTOR_MATH_MINIMUM:
+      *vector = min(a, b);
+      break;
+    case NODE_VECTOR_MATH_MAXIMUM:
+      *vector = max(a, b);
+      break;
+    default:
+      *vector = make_float3(0.0f, 0.0f, 0.0f);
+      *value = 0.0f;
   }
 }
 
-ccl_device float svm_math(NodeMath type, float Fac1, float Fac2)
+ccl_device float svm_math(NodeMathType type, float a, float b)
 {
-  float Fac;
-
-  if (type == NODE_MATH_ADD)
-    Fac = Fac1 + Fac2;
-  else if (type == NODE_MATH_SUBTRACT)
-    Fac = Fac1 - Fac2;
-  else if (type == NODE_MATH_MULTIPLY)
-    Fac = Fac1 * Fac2;
-  else if (type == NODE_MATH_DIVIDE)
-    Fac = safe_divide(Fac1, Fac2);
-  else if (type == NODE_MATH_SINE)
-    Fac = sinf(Fac1);
-  else if (type == NODE_MATH_COSINE)
-    Fac = cosf(Fac1);
-  else if (type == NODE_MATH_TANGENT)
-    Fac = tanf(Fac1);
-  else if (type == NODE_MATH_ARCSINE)
-    Fac = safe_asinf(Fac1);
-  else if (type == NODE_MATH_ARCCOSINE)
-    Fac = safe_acosf(Fac1);
-  else if (type == NODE_MATH_ARCTANGENT)
-    Fac = atanf(Fac1);
-  else if (type == NODE_MATH_POWER)
-    Fac = safe_powf(Fac1, Fac2);
-  else if (type == NODE_MATH_LOGARITHM)
-    Fac = safe_logf(Fac1, Fac2);
-  else if (type == NODE_MATH_MINIMUM)
-    Fac = fminf(Fac1, Fac2);
-  else if (type == NODE_MATH_MAXIMUM)
-    Fac = fmaxf(Fac1, Fac2);
-  else if (type == NODE_MATH_ROUND)
-    Fac = floorf(Fac1 + 0.5f);
-  else if (type == NODE_MATH_LESS_THAN)
-    Fac = Fac1 < Fac2;
-  else if (type == NODE_MATH_GREATER_THAN)
-    Fac = Fac1 > Fac2;
-  else if (type == NODE_MATH_MODULO)
-    Fac = safe_modulo(Fac1, Fac2);
-  else if (type == NODE_MATH_ABSOLUTE)
-    Fac = fabsf(Fac1);
-  else if (type == NODE_MATH_ARCTAN2)
-    Fac = atan2f(Fac1, Fac2);
-  else if (type == NODE_MATH_FLOOR)
-    Fac = floorf(Fac1);
-  else if (type == NODE_MATH_CEIL)
-    Fac = ceilf(Fac1);
-  else if (type == NODE_MATH_FRACT)
-    Fac = Fac1 - floorf(Fac1);
-  else if (type == NODE_MATH_SQRT)
-    Fac = safe_sqrtf(Fac1);
-  else if (type == NODE_MATH_CLAMP)
-    Fac = saturate(Fac1);
-  else
-    Fac = 0.0f;
-
-  return Fac;
+  switch (type) {
+    case NODE_MATH_ADD:
+      return a + b;
+    case NODE_MATH_SUBTRACT:
+      return a - b;
+    case NODE_MATH_MULTIPLY:
+      return a * b;
+    case NODE_MATH_DIVIDE:
+      return safe_divide(a, b);
+    case NODE_MATH_POWER:
+      return safe_powf(a, b);
+    case NODE_MATH_LOGARITHM:
+      return safe_logf(a, b);
+    case NODE_MATH_SQRT:
+      return safe_sqrtf(a);
+    case NODE_MATH_ABSOLUTE:
+      return fabsf(a);
+    case NODE_MATH_MINIMUM:
+      return fminf(a, b);
+    case NODE_MATH_MAXIMUM:
+      return fmaxf(a, b);
+    case NODE_MATH_LESS_THAN:
+      return a < b;
+    case NODE_MATH_GREATER_THAN:
+      return a > b;
+    case NODE_MATH_ROUND:
+      return floorf(a + 0.5f);
+    case NODE_MATH_FLOOR:
+      return floorf(a);
+    case NODE_MATH_CEIL:
+      return ceilf(a);
+    case NODE_MATH_FRACTION:
+      return a - floorf(a);
+    case NODE_MATH_MODULO:
+      return safe_modulo(a, b);
+    case NODE_MATH_SINE:
+      return sinf(a);
+    case NODE_MATH_COSINE:
+      return cosf(a);
+    case NODE_MATH_TANGENT:
+      return tanf(a);
+    case NODE_MATH_ARCSINE:
+      return safe_asinf(a);
+    case NODE_MATH_ARCCOSINE:
+      return safe_acosf(a);
+    case NODE_MATH_ARCTANGENT:
+      return atanf(a);
+    case NODE_MATH_ARCTAN2:
+      return atan2f(a, b);
+    default:
+      return 0.0f;
+  }
 }
 
 /* Calculate color in range 800..12000 using an approximation
diff --git a/intern/cycles/kernel/svm/svm_musgrave.h b/intern/cycles/kernel/svm/svm_musgrave.h
index 67fb5ca6241..9291c7e7295 100644
--- a/intern/cycles/kernel/svm/svm_musgrave.h
+++ b/intern/cycles/kernel/svm/svm_musgrave.h
@@ -25,7 +25,10 @@ CCL_NAMESPACE_BEGIN
  * from "Texturing and Modelling: A procedural approach"
  */
 
-ccl_device_noinline float noise_musgrave_fBm(float3 p, float H, float lacunarity, float octaves)
+ccl_device_noinline_cpu float noise_musgrave_fBm(float3 p,
+                                                 float H,
+                                                 float lacunarity,
+                                                 float octaves)
 {
   float rmd;
   float value = 0.0f;
@@ -53,10 +56,10 @@ ccl_device_noinline float noise_musgrave_fBm(float3 p, float H, float lacunarity
  * octaves: number of frequencies in the fBm
  */
 
-ccl_device_noinline float noise_musgrave_multi_fractal(float3 p,
-                                                       float H,
-                                                       float lacunarity,
-                                                       float octaves)
+ccl_device_noinline_cpu float noise_musgrave_multi_fractal(float3 p,
+                                                           float H,
+                                                           float lacunarity,
+                                                           float octaves)
 {
   float rmd;
   float value = 1.0f;
@@ -85,7 +88,7 @@ ccl_device_noinline float noise_musgrave_multi_fractal(float3 p,
  * offset: raises the terrain from `sea level'
  */
 
-ccl_device_noinline float noise_musgrave_hetero_terrain(
+ccl_device_noinline_cpu float noise_musgrave_hetero_terrain(
     float3 p, float H, float lacunarity, float octaves, float offset)
 {
   float value, increment, rmd;
@@ -121,7 +124,7 @@ ccl_device_noinline float noise_musgrave_hetero_terrain(
  * offset: raises the terrain from `sea level'
  */
 
-ccl_device_noinline float noise_musgrave_hybrid_multi_fractal(
+ccl_device_noinline_cpu float noise_musgrave_hybrid_multi_fractal(
     float3 p, float H, float lacunarity, float octaves, float offset, float gain)
 {
   float result, signal, weight, rmd;
@@ -159,7 +162,7 @@ ccl_device_noinline float noise_musgrave_hybrid_multi_fractal(
  * offset: raises the terrain from `sea level'
  */
 
-ccl_device_noinline float noise_musgrave_ridged_multi_fractal(
+ccl_device_noinline_cpu float noise_musgrave_ridged_multi_fractal(
     float3 p, float H, float lacunarity, float octaves, float offset, float gain)
 {
   float result, signal, weight;
@@ -222,10 +225,10 @@ ccl_device void svm_node_tex_musgrave(
   uint dimension_offset, lacunarity_offset, detail_offset, offset_offset;
   uint gain_offset, scale_offset;
 
-  decode_node_uchar4(node.y, &type, &co_offset, &color_offset, &fac_offset);
-  decode_node_uchar4(
+  svm_unpack_node_uchar4(node.y, &type, &co_offset, &color_offset, &fac_offset);
+  svm_unpack_node_uchar4(
       node.z, &dimension_offset, &lacunarity_offset, &detail_offset, &offset_offset);
-  decode_node_uchar4(node.w, &gain_offset, &scale_offset, NULL, NULL);
+  svm_unpack_node_uchar2(node.w, &gain_offset, &scale_offset);
 
   float3 co = stack_load_float3(stack, co_offset);
   float dimension = stack_load_float_default(stack, dimension_offset, node2.x);
diff --git a/intern/cycles/kernel/svm/svm_noise.h b/intern/cycles/kernel/svm/svm_noise.h
index 322579ccfe3..dd375af27e5 100644
--- a/intern/cycles/kernel/svm/svm_noise.h
+++ b/intern/cycles/kernel/svm/svm_noise.h
@@ -41,42 +41,6 @@ ccl_device_inline ssei quick_floor_sse(const ssef &x)
 }
 #endif
 
-ccl_device uint hash(uint kx, uint ky, uint kz)
-{
-  // define some handy macros
-#define rot(x, k) (((x) << (k)) | ((x) >> (32 - (k))))
-#define final(a, b, c) \
-  { \
-    c ^= b; \
-    c -= rot(b, 14); \
-    a ^= c; \
-    a -= rot(c, 11); \
-    b ^= a; \
-    b -= rot(a, 25); \
-    c ^= b; \
-    c -= rot(b, 16); \
-    a ^= c; \
-    a -= rot(c, 4); \
-    b ^= a; \
-    b -= rot(a, 14); \
-    c ^= b; \
-    c -= rot(b, 24); \
-  }
-  // now hash the data!
-  uint a, b, c, len = 3;
-  a = b = c = 0xdeadbeef + (len << 2) + 13;
-
-  c += kz;
-  b += ky;
-  a += kx;
-  final(a, b, c);
-
-  return c;
-  // macros not needed anymore
-#undef rot
-#undef final
-}
-
 #ifdef __KERNEL_SSE2__
 ccl_device_inline ssei hash_sse(const ssei &kx, const ssei &ky, const ssei &kz)
 {
@@ -218,7 +182,7 @@ ccl_device_inline ssef scale3_sse(const ssef &result)
 #endif
 
 #ifndef __KERNEL_SSE2__
-ccl_device_noinline float perlin(float x, float y, float z)
+ccl_device_noinline_cpu float perlin(float x, float y, float z)
 {
   int X;
   float fx = floorfrac(x, &X);
@@ -236,17 +200,19 @@ ccl_device_noinline float perlin(float x, float y, float z)
   result = nerp(
       w,
       nerp(v,
-           nerp(u, grad(hash(X, Y, Z), fx, fy, fz), grad(hash(X + 1, Y, Z), fx - 1.0f, fy, fz)),
            nerp(u,
-                grad(hash(X, Y + 1, Z), fx, fy - 1.0f, fz),
-                grad(hash(X + 1, Y + 1, Z), fx - 1.0f, fy - 1.0f, fz))),
+                grad(hash_uint3(X, Y, Z), fx, fy, fz),
+                grad(hash_uint3(X + 1, Y, Z), fx - 1.0f, fy, fz)),
+           nerp(u,
+                grad(hash_uint3(X, Y + 1, Z), fx, fy - 1.0f, fz),
+                grad(hash_uint3(X + 1, Y + 1, Z), fx - 1.0f, fy - 1.0f, fz))),
       nerp(v,
            nerp(u,
-                grad(hash(X, Y, Z + 1), fx, fy, fz - 1.0f),
-                grad(hash(X + 1, Y, Z + 1), fx - 1.0f, fy, fz - 1.0f)),
+                grad(hash_uint3(X, Y, Z + 1), fx, fy, fz - 1.0f),
+                grad(hash_uint3(X + 1, Y, Z + 1), fx - 1.0f, fy, fz - 1.0f)),
            nerp(u,
-                grad(hash(X, Y + 1, Z + 1), fx, fy - 1.0f, fz - 1.0f),
-                grad(hash(X + 1, Y + 1, Z + 1), fx - 1.0f, fy - 1.0f, fz - 1.0f))));
+                grad(hash_uint3(X, Y + 1, Z + 1), fx, fy - 1.0f, fz - 1.0f),
+                grad(hash_uint3(X + 1, Y + 1, Z + 1), fx - 1.0f, fy - 1.0f, fz - 1.0f))));
   float r = scale3(result);
 
   /* can happen for big coordinates, things even out to 0.0 then anyway */
@@ -312,16 +278,16 @@ ccl_device float snoise(float3 p)
 ccl_device float cellnoise(float3 p)
 {
   int3 ip = quick_floor_to_int3(p);
-  return bits_to_01(hash(ip.x, ip.y, ip.z));
+  return hash_uint3_to_float(ip.x, ip.y, ip.z);
 }
 
 ccl_device float3 cellnoise3(float3 p)
 {
   int3 ip = quick_floor_to_int3(p);
 #ifndef __KERNEL_SSE__
-  float r = bits_to_01(hash(ip.x, ip.y, ip.z));
-  float g = bits_to_01(hash(ip.y, ip.x, ip.z));
-  float b = bits_to_01(hash(ip.y, ip.z, ip.x));
+  float r = hash_uint3_to_float(ip.x, ip.y, ip.z);
+  float g = hash_uint3_to_float(ip.y, ip.x, ip.z);
+  float b = hash_uint3_to_float(ip.y, ip.z, ip.x);
   return make_float3(r, g, b);
 #else
   ssei ip_yxz = shuffle<1, 0, 2, 3>(ssei(ip.m128));
diff --git a/intern/cycles/kernel/svm/svm_noisetex.h b/intern/cycles/kernel/svm/svm_noisetex.h
index 3324e86fcd8..91dc11691e6 100644
--- a/intern/cycles/kernel/svm/svm_noisetex.h
+++ b/intern/cycles/kernel/svm/svm_noisetex.h
@@ -23,8 +23,8 @@ ccl_device void svm_node_tex_noise(
 {
   uint co_offset, scale_offset, detail_offset, distortion_offset, fac_offset, color_offset;
 
-  decode_node_uchar4(node.y, &co_offset, &scale_offset, &detail_offset, &distortion_offset);
-  decode_node_uchar4(node.z, &color_offset, &fac_offset, NULL, NULL);
+  svm_unpack_node_uchar4(node.y, &co_offset, &scale_offset, &detail_offset, &distortion_offset);
+  svm_unpack_node_uchar2(node.z, &color_offset, &fac_offset);
 
   uint4 node2 = read_node(kg, offset);
 
diff --git a/intern/cycles/kernel/svm/svm_ramp.h b/intern/cycles/kernel/svm/svm_ramp.h
index 6084ee35a1f..85ccf39144b 100644
--- a/intern/cycles/kernel/svm/svm_ramp.h
+++ b/intern/cycles/kernel/svm/svm_ramp.h
@@ -59,7 +59,7 @@ ccl_device void svm_node_rgb_ramp(
   uint fac_offset, color_offset, alpha_offset;
   uint interpolate = node.z;
 
-  decode_node_uchar4(node.y, &fac_offset, &color_offset, &alpha_offset, NULL);
+  svm_unpack_node_uchar3(node.y, &fac_offset, &color_offset, &alpha_offset);
 
   uint table_size = read_node(kg, offset).x;
 
@@ -78,7 +78,7 @@ ccl_device void svm_node_curves(
     KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node, int *offset)
 {
   uint fac_offset, color_offset, out_offset;
-  decode_node_uchar4(node.y, &fac_offset, &color_offset, &out_offset, NULL);
+  svm_unpack_node_uchar3(node.y, &fac_offset, &color_offset, &out_offset);
 
   uint table_size = read_node(kg, offset).x;
 
diff --git a/intern/cycles/kernel/svm/svm_tex_coord.h b/intern/cycles/kernel/svm/svm_tex_coord.h
index 1fb3e20f9e0..a876d6bc916 100644
--- a/intern/cycles/kernel/svm/svm_tex_coord.h
+++ b/intern/cycles/kernel/svm/svm_tex_coord.h
@@ -257,7 +257,7 @@ ccl_device void svm_node_tex_coord_bump_dy(
 ccl_device void svm_node_normal_map(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
   uint color_offset, strength_offset, normal_offset, space;
-  decode_node_uchar4(node.y, &color_offset, &strength_offset, &normal_offset, &space);
+  svm_unpack_node_uchar4(node.y, &color_offset, &strength_offset, &normal_offset, &space);
 
   float3 color = stack_load_float3(stack, color_offset);
   color = 2.0f * make_float3(color.x - 0.5f, color.y - 0.5f, color.z - 0.5f);
@@ -349,7 +349,7 @@ ccl_device void svm_node_normal_map(KernelGlobals *kg, ShaderData *sd, float *st
 ccl_device void svm_node_tangent(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
   uint tangent_offset, direction_type, axis;
-  decode_node_uchar4(node.y, &tangent_offset, &direction_type, &axis, NULL);
+  svm_unpack_node_uchar3(node.y, &tangent_offset, &direction_type, &axis);
 
   float3 tangent;
   float3 attribute_value;
diff --git a/intern/cycles/kernel/svm/svm_types.h b/intern/cycles/kernel/svm/svm_types.h
index 9a8a5401297..a3caa1ab68d 100644
--- a/intern/cycles/kernel/svm/svm_types.h
+++ b/intern/cycles/kernel/svm/svm_types.h
@@ -140,6 +140,7 @@ typedef enum ShaderNodeType {
   NODE_IES,
   NODE_MAP_RANGE,
   NODE_CLAMP,
+  NODE_TEX_WHITE_NOISE,
 } ShaderNodeType;
 
 typedef enum NodeAttributeType {
@@ -160,6 +161,7 @@ typedef enum NodeGeometry {
 
 typedef enum NodeObjectInfo {
   NODE_INFO_OB_LOCATION,
+  NODE_INFO_OB_COLOR,
   NODE_INFO_OB_INDEX,
   NODE_INFO_MAT_INDEX,
   NODE_INFO_OB_RANDOM
@@ -244,7 +246,7 @@ typedef enum NodeMix {
   NODE_MIX_CLAMP /* used for the clamp UI option */
 } NodeMix;
 
-typedef enum NodeMath {
+typedef enum NodeMathType {
   NODE_MATH_ADD,
   NODE_MATH_SUBTRACT,
   NODE_MATH_MULTIPLY,
@@ -267,19 +269,35 @@ typedef enum NodeMath {
   NODE_MATH_ARCTAN2,
   NODE_MATH_FLOOR,
   NODE_MATH_CEIL,
-  NODE_MATH_FRACT,
+  NODE_MATH_FRACTION,
   NODE_MATH_SQRT,
-  NODE_MATH_CLAMP /* used for the clamp UI option */
-} NodeMath;
+} NodeMathType;
 
-typedef enum NodeVectorMath {
+typedef enum NodeVectorMathType {
   NODE_VECTOR_MATH_ADD,
   NODE_VECTOR_MATH_SUBTRACT,
-  NODE_VECTOR_MATH_AVERAGE,
-  NODE_VECTOR_MATH_DOT_PRODUCT,
+  NODE_VECTOR_MATH_MULTIPLY,
+  NODE_VECTOR_MATH_DIVIDE,
+
   NODE_VECTOR_MATH_CROSS_PRODUCT,
-  NODE_VECTOR_MATH_NORMALIZE
-} NodeVectorMath;
+  NODE_VECTOR_MATH_PROJECT,
+  NODE_VECTOR_MATH_REFLECT,
+  NODE_VECTOR_MATH_DOT_PRODUCT,
+
+  NODE_VECTOR_MATH_DISTANCE,
+  NODE_VECTOR_MATH_LENGTH,
+  NODE_VECTOR_MATH_SCALE,
+  NODE_VECTOR_MATH_NORMALIZE,
+
+  NODE_VECTOR_MATH_SNAP,
+  NODE_VECTOR_MATH_FLOOR,
+  NODE_VECTOR_MATH_CEIL,
+  NODE_VECTOR_MATH_MODULO,
+  NODE_VECTOR_MATH_FRACTION,
+  NODE_VECTOR_MATH_ABSOLUTE,
+  NODE_VECTOR_MATH_MINIMUM,
+  NODE_VECTOR_MATH_MAXIMUM,
+} NodeVectorMathType;
 
 typedef enum NodeVectorTransformType {
   NODE_VECTOR_TRANSFORM_TYPE_VECTOR,
diff --git a/intern/cycles/kernel/svm/svm_vector_transform.h b/intern/cycles/kernel/svm/svm_vector_transform.h
index 7ec0f07f2e4..1e95492cf1b 100644
--- a/intern/cycles/kernel/svm/svm_vector_transform.h
+++ b/intern/cycles/kernel/svm/svm_vector_transform.h
@@ -26,8 +26,8 @@ ccl_device void svm_node_vector_transform(KernelGlobals *kg,
   uint itype, ifrom, ito;
   uint vector_in, vector_out;
 
-  decode_node_uchar4(node.y, &itype, &ifrom, &ito, NULL);
-  decode_node_uchar4(node.z, &vector_in, &vector_out, NULL, NULL);
+  svm_unpack_node_uchar3(node.y, &itype, &ifrom, &ito);
+  svm_unpack_node_uchar2(node.z, &vector_in, &vector_out);
 
   float3 in = stack_load_float3(stack, vector_in);
 
diff --git a/intern/cycles/kernel/svm/svm_voronoi.h b/intern/cycles/kernel/svm/svm_voronoi.h
index 3e28a316169..3d7fa523968 100644
--- a/intern/cycles/kernel/svm/svm_voronoi.h
+++ b/intern/cycles/kernel/svm/svm_voronoi.h
@@ -116,8 +116,8 @@ ccl_device void svm_node_tex_voronoi(
   uint co_offset, coloring, distance, feature;
   uint scale_offset, e_offset, fac_offset, color_offset;
 
-  decode_node_uchar4(node.y, &co_offset, &coloring, &distance, &feature);
-  decode_node_uchar4(node.z, &scale_offset, &e_offset, &fac_offset, &color_offset);
+  svm_unpack_node_uchar4(node.y, &co_offset, &coloring, &distance, &feature);
+  svm_unpack_node_uchar4(node.z, &scale_offset, &e_offset, &fac_offset, &color_offset);
 
   float3 co = stack_load_float3(stack, co_offset);
   float scale = stack_load_float_default(stack, scale_offset, node2.x);
diff --git a/intern/cycles/kernel/svm/svm_voxel.h b/intern/cycles/kernel/svm/svm_voxel.h
index 26d8cc71d3b..b79be8e5bde 100644
--- a/intern/cycles/kernel/svm/svm_voxel.h
+++ b/intern/cycles/kernel/svm/svm_voxel.h
@@ -23,7 +23,7 @@ ccl_device void svm_node_tex_voxel(
     KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node, int *offset)
 {
   uint co_offset, density_out_offset, color_out_offset, space;
-  decode_node_uchar4(node.z, &co_offset, &density_out_offset, &color_out_offset, &space);
+  svm_unpack_node_uchar4(node.z, &co_offset, &density_out_offset, &color_out_offset, &space);
 #ifdef __VOLUME__
   int id = node.y;
   float3 co = stack_load_float3(stack, co_offset);
diff --git a/intern/cycles/kernel/svm/svm_wave.h b/intern/cycles/kernel/svm/svm_wave.h
index 003ad7dc63a..baaa89ab0cb 100644
--- a/intern/cycles/kernel/svm/svm_wave.h
+++ b/intern/cycles/kernel/svm/svm_wave.h
@@ -18,12 +18,12 @@ CCL_NAMESPACE_BEGIN
 
 /* Wave */
 
-ccl_device_noinline float svm_wave(NodeWaveType type,
-                                   NodeWaveProfile profile,
-                                   float3 p,
-                                   float detail,
-                                   float distortion,
-                                   float dscale)
+ccl_device_noinline_cpu float svm_wave(NodeWaveType type,
+                                       NodeWaveProfile profile,
+                                       float3 p,
+                                       float detail,
+                                       float distortion,
+                                       float dscale)
 {
   float n;
 
@@ -54,8 +54,8 @@ ccl_device void svm_node_tex_wave(
   uint co_offset, scale_offset, detail_offset, dscale_offset, distortion_offset, color_offset,
       fac_offset;
 
-  decode_node_uchar4(node.y, &type, &color_offset, &fac_offset, &dscale_offset);
-  decode_node_uchar4(node.z, &co_offset, &scale_offset, &detail_offset, &distortion_offset);
+  svm_unpack_node_uchar4(node.y, &type, &color_offset, &fac_offset, &dscale_offset);
+  svm_unpack_node_uchar4(node.z, &co_offset, &scale_offset, &detail_offset, &distortion_offset);
 
   float3 co = stack_load_float3(stack, co_offset);
   float scale = stack_load_float_default(stack, scale_offset, node2.x);
diff --git a/intern/cycles/kernel/svm/svm_white_noise.h b/intern/cycles/kernel/svm/svm_white_noise.h
new file mode 100644
index 00000000000..71d4591d25d
--- /dev/null
+++ b/intern/cycles/kernel/svm/svm_white_noise.h
@@ -0,0 +1,55 @@
+/*
+ * Copyright 2011-2013 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+CCL_NAMESPACE_BEGIN
+
+ccl_device void svm_node_tex_white_noise(KernelGlobals *kg,
+                                         ShaderData *sd,
+                                         float *stack,
+                                         uint dimensions,
+                                         uint inputs_stack_offsets,
+                                         uint value_stack_offset,
+                                         int *offset)
+{
+  uint vector_stack_offset, w_stack_offset;
+  svm_unpack_node_uchar2(inputs_stack_offsets, &vector_stack_offset, &w_stack_offset);
+
+  float3 vector = stack_load_float3(stack, vector_stack_offset);
+  float w = stack_load_float(stack, w_stack_offset);
+
+  float value;
+  switch (dimensions) {
+    case 1:
+      value = hash_float_to_float(w);
+      break;
+    case 2:
+      value = hash_float2_to_float(make_float2(vector.x, vector.y));
+      break;
+    case 3:
+      value = hash_float3_to_float(vector);
+      break;
+    case 4:
+      value = hash_float4_to_float(make_float4(vector.x, vector.y, vector.z, w));
+      break;
+    default:
+      value = 0.0f;
+      kernel_assert(0);
+      break;
+  }
+  stack_store_float(stack, value_stack_offset, value);
+}
+
+CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/svm/svm_wireframe.h b/intern/cycles/kernel/svm/svm_wireframe.h
index 55e61d0e8c7..49158bd86d5 100644
--- a/intern/cycles/kernel/svm/svm_wireframe.h
+++ b/intern/cycles/kernel/svm/svm_wireframe.h
@@ -93,7 +93,7 @@ ccl_device void svm_node_wireframe(KernelGlobals *kg, ShaderData *sd, float *sta
   uint in_size = node.y;
   uint out_fac = node.z;
   uint use_pixel_size, bump_offset;
-  decode_node_uchar4(node.w, &use_pixel_size, &bump_offset, NULL, NULL);
+  svm_unpack_node_uchar2(node.w, &use_pixel_size, &bump_offset);
 
   /* Input Data */
   float size = stack_load_float(stack, in_size);
diff --git a/intern/cycles/render/bake.cpp b/intern/cycles/render/bake.cpp
index 73893921500..b906357b7b5 100644
--- a/intern/cycles/render/bake.cpp
+++ b/intern/cycles/render/bake.cpp
@@ -124,7 +124,7 @@ BakeData *BakeManager::init(const int object, const size_t tri_offset, const siz
 void BakeManager::set_shader_limit(const size_t x, const size_t y)
 {
   m_shader_limit = x * y;
-  m_shader_limit = (size_t)pow(2, ceil(log(m_shader_limit) / log(2)));
+  m_shader_limit = (size_t)pow(2, std::ceil(log(m_shader_limit) / log(2)));
 }
 
 bool BakeManager::bake(Device *device,
diff --git a/intern/cycles/render/constant_fold.cpp b/intern/cycles/render/constant_fold.cpp
index e475ff60eef..851d4b71df8 100644
--- a/intern/cycles/render/constant_fold.cpp
+++ b/intern/cycles/render/constant_fold.cpp
@@ -301,7 +301,7 @@ void ConstantFolder::fold_mix(NodeMix type, bool clamp) const
   }
 }
 
-void ConstantFolder::fold_math(NodeMath type, bool clamp) const
+void ConstantFolder::fold_math(NodeMathType type) const
 {
   ShaderInput *value1_in = node->input("Value1");
   ShaderInput *value2_in = node->input("Value2");
@@ -310,25 +310,25 @@ void ConstantFolder::fold_math(NodeMath type, bool clamp) const
     case NODE_MATH_ADD:
       /* X + 0 == 0 + X == X */
       if (is_zero(value1_in)) {
-        try_bypass_or_make_constant(value2_in, clamp);
+        try_bypass_or_make_constant(value2_in);
       }
       else if (is_zero(value2_in)) {
-        try_bypass_or_make_constant(value1_in, clamp);
+        try_bypass_or_make_constant(value1_in);
       }
       break;
     case NODE_MATH_SUBTRACT:
       /* X - 0 == X */
       if (is_zero(value2_in)) {
-        try_bypass_or_make_constant(value1_in, clamp);
+        try_bypass_or_make_constant(value1_in);
       }
       break;
     case NODE_MATH_MULTIPLY:
       /* X * 1 == 1 * X == X */
       if (is_one(value1_in)) {
-        try_bypass_or_make_constant(value2_in, clamp);
+        try_bypass_or_make_constant(value2_in);
       }
       else if (is_one(value2_in)) {
-        try_bypass_or_make_constant(value1_in, clamp);
+        try_bypass_or_make_constant(value1_in);
       }
       /* X * 0 == 0 * X == 0 */
       else if (is_zero(value1_in) || is_zero(value2_in)) {
@@ -338,7 +338,7 @@ void ConstantFolder::fold_math(NodeMath type, bool clamp) const
     case NODE_MATH_DIVIDE:
       /* X / 1 == X */
       if (is_one(value2_in)) {
-        try_bypass_or_make_constant(value1_in, clamp);
+        try_bypass_or_make_constant(value1_in);
       }
       /* 0 / X == 0 */
       else if (is_zero(value1_in)) {
@@ -352,17 +352,18 @@ void ConstantFolder::fold_math(NodeMath type, bool clamp) const
       }
       /* X ^ 1 == X */
       else if (is_one(value2_in)) {
-        try_bypass_or_make_constant(value1_in, clamp);
+        try_bypass_or_make_constant(value1_in);
       }
     default:
       break;
   }
 }
 
-void ConstantFolder::fold_vector_math(NodeVectorMath type) const
+void ConstantFolder::fold_vector_math(NodeVectorMathType type) const
 {
   ShaderInput *vector1_in = node->input("Vector1");
   ShaderInput *vector2_in = node->input("Vector2");
+  ShaderInput *scale_in = node->input("Scale");
 
   switch (type) {
     case NODE_VECTOR_MATH_ADD:
@@ -380,6 +381,27 @@ void ConstantFolder::fold_vector_math(NodeVectorMath type) const
         try_bypass_or_make_constant(vector1_in);
       }
       break;
+    case NODE_VECTOR_MATH_MULTIPLY:
+      /* X * 0 == 0 * X == 0 */
+      if (is_zero(vector1_in) || is_zero(vector2_in)) {
+        make_zero();
+      } /* X * 1 == 1 * X == X */
+      else if (is_one(vector1_in)) {
+        try_bypass_or_make_constant(vector2_in);
+      }
+      else if (is_one(vector2_in)) {
+        try_bypass_or_make_constant(vector1_in);
+      }
+      break;
+    case NODE_VECTOR_MATH_DIVIDE:
+      /* X / 0 == 0 / X == 0 */
+      if (is_zero(vector1_in) || is_zero(vector2_in)) {
+        make_zero();
+      } /* X / 1 == X */
+      else if (is_one(vector2_in)) {
+        try_bypass_or_make_constant(vector1_in);
+      }
+      break;
     case NODE_VECTOR_MATH_DOT_PRODUCT:
     case NODE_VECTOR_MATH_CROSS_PRODUCT:
       /* X * 0 == 0 * X == 0 */
@@ -387,6 +409,21 @@ void ConstantFolder::fold_vector_math(NodeVectorMath type) const
         make_zero();
       }
       break;
+    case NODE_VECTOR_MATH_LENGTH:
+    case NODE_VECTOR_MATH_ABSOLUTE:
+      if (is_zero(vector1_in)) {
+        make_zero();
+      }
+      break;
+    case NODE_VECTOR_MATH_SCALE:
+      /* X * 0 == 0 * X == 0 */
+      if (is_zero(vector1_in) || is_zero(scale_in)) {
+        make_zero();
+      } /* X * 1 == X */
+      else if (is_one(scale_in)) {
+        try_bypass_or_make_constant(vector1_in);
+      }
+      break;
     default:
       break;
   }
diff --git a/intern/cycles/render/constant_fold.h b/intern/cycles/render/constant_fold.h
index c14b94868dc..881636a9fe1 100644
--- a/intern/cycles/render/constant_fold.h
+++ b/intern/cycles/render/constant_fold.h
@@ -64,8 +64,8 @@ class ConstantFolder {
 
   /* Specific nodes. */
   void fold_mix(NodeMix type, bool clamp) const;
-  void fold_math(NodeMath type, bool clamp) const;
-  void fold_vector_math(NodeVectorMath type) const;
+  void fold_math(NodeMathType type) const;
+  void fold_vector_math(NodeVectorMathType type) const;
 };
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/render/integrator.cpp b/intern/cycles/render/integrator.cpp
index 76258a292e8..b41b0b7b260 100644
--- a/intern/cycles/render/integrator.cpp
+++ b/intern/cycles/render/integrator.cpp
@@ -141,7 +141,7 @@ void Integrator::device_update(Device *device, DeviceScene *dscene, Scene *scene
   kintegrator->caustics_refractive = caustics_refractive;
   kintegrator->filter_glossy = (filter_glossy == 0.0f) ? FLT_MAX : 1.0f / filter_glossy;
 
-  kintegrator->seed = hash_int(seed);
+  kintegrator->seed = hash_uint2(seed, 0);
 
   kintegrator->use_ambient_occlusion = ((Pass::contains(scene->film->passes, PASS_AO)) ||
                                         dscene->data.background.ao_factor != 0.0f);
diff --git a/intern/cycles/render/mesh.cpp b/intern/cycles/render/mesh.cpp
index 6ac66661859..3d2cc98d38b 100644
--- a/intern/cycles/render/mesh.cpp
+++ b/intern/cycles/render/mesh.cpp
@@ -433,6 +433,8 @@ Mesh::Mesh() : Node(node_type)
 
   attr_map_offset = 0;
 
+  prim_offset = 0;
+
   num_subd_verts = 0;
 
   attributes.triangle_mesh = this;
@@ -558,6 +560,9 @@ void Mesh::clear(bool preserve_voxel_data)
 
   used_shaders.clear();
 
+  vert_to_stitching_key_map.clear();
+  vert_stitching_map.clear();
+
   if (!preserve_voxel_data) {
     geometry_flags = GEOMETRY_NONE;
   }
@@ -1013,9 +1018,11 @@ void Mesh::compute_bvh(
 
   compute_bounds();
 
-  if (need_build_bvh()) {
+  const BVHLayout bvh_layout = BVHParams::best_bvh_layout(params->bvh_layout,
+                                                          device->get_bvh_layout_mask());
+  if (need_build_bvh(bvh_layout)) {
     string msg = "Updating Mesh BVH ";
-    if (name == "")
+    if (name.empty())
       msg += string_printf("%u/%u", (uint)(n + 1), (uint)total);
     else
       msg += string_printf("%s %u/%u", name.c_str(), (uint)(n + 1), (uint)total);
@@ -1023,12 +1030,17 @@ void Mesh::compute_bvh(
     Object object;
     object.mesh = this;
 
+    vector<Mesh *> meshes;
+    meshes.push_back(this);
     vector<Object *> objects;
     objects.push_back(&object);
 
     if (bvh && !need_update_rebuild) {
       progress->set_status(msg, "Refitting BVH");
+
+      bvh->meshes = meshes;
       bvh->objects = objects;
+
       bvh->refit(*progress);
     }
     else {
@@ -1036,8 +1048,7 @@ void Mesh::compute_bvh(
 
       BVHParams bparams;
       bparams.use_spatial_split = params->use_bvh_spatial_split;
-      bparams.bvh_layout = BVHParams::best_bvh_layout(params->bvh_layout,
-                                                      device->get_bvh_layout_mask());
+      bparams.bvh_layout = bvh_layout;
       bparams.use_unaligned_nodes = dscene->data.bvh.have_curves &&
                                     params->use_bvh_unaligned_nodes;
       bparams.num_motion_triangle_steps = params->num_bvh_time_steps;
@@ -1047,7 +1058,7 @@ void Mesh::compute_bvh(
       bparams.curve_subdivisions = dscene->data.curve.subdivisions;
 
       delete bvh;
-      bvh = BVH::create(bparams, objects);
+      bvh = BVH::create(bparams, meshes, objects);
       MEM_GUARDED_CALL(progress, bvh->build, *progress);
     }
   }
@@ -1128,7 +1139,7 @@ int Mesh::motion_step(float time) const
   return -1;
 }
 
-bool Mesh::need_build_bvh() const
+bool Mesh::need_build_bvh(BVHLayout) const
 {
   return !transform_applied || has_surface_bssrdf;
 }
@@ -1722,6 +1733,8 @@ void MeshManager::mesh_calc_offset(Scene *scene)
   size_t face_size = 0;
   size_t corner_size = 0;
 
+  size_t prim_size = 0;
+
   foreach (Mesh *mesh, scene->meshes) {
     mesh->vert_offset = vert_size;
     mesh->tri_offset = tri_size;
@@ -1751,6 +1764,9 @@ void MeshManager::mesh_calc_offset(Scene *scene)
     }
     face_size += mesh->subd_faces.size();
     corner_size += mesh->subd_face_corners.size();
+
+    mesh->prim_offset = prim_size;
+    prim_size += mesh->num_primitives();
   }
 }
 
@@ -1929,7 +1945,7 @@ void MeshManager::device_update_bvh(Device *device,
   }
 #endif
 
-  BVH *bvh = BVH::create(bparams, scene->objects);
+  BVH *bvh = BVH::create(bparams, scene->meshes, scene->objects);
   bvh->build(progress, &device->stats);
 
   if (progress.get_cancel()) {
@@ -1994,14 +2010,7 @@ void MeshManager::device_update_bvh(Device *device,
   dscene->data.bvh.bvh_layout = bparams.bvh_layout;
   dscene->data.bvh.use_bvh_steps = (scene->params.num_bvh_time_steps != 0);
 
-#ifdef WITH_EMBREE
-  if (bparams.bvh_layout == BVH_LAYOUT_EMBREE) {
-    dscene->data.bvh.scene = ((BVHEmbree *)bvh)->scene;
-  }
-  else {
-    dscene->data.bvh.scene = NULL;
-  }
-#endif
+  bvh->copy_to_device(progress, dscene);
 
   delete bvh;
 }
@@ -2213,6 +2222,8 @@ void MeshManager::device_update(Device *device,
   /* Update displacement. */
   bool displacement_done = false;
   size_t num_bvh = 0;
+  BVHLayout bvh_layout = BVHParams::best_bvh_layout(scene->params.bvh_layout,
+                                                    device->get_bvh_layout_mask());
 
   foreach (Mesh *mesh, scene->meshes) {
     if (mesh->need_update) {
@@ -2220,7 +2231,7 @@ void MeshManager::device_update(Device *device,
         displacement_done = true;
       }
 
-      if (mesh->need_build_bvh()) {
+      if (mesh->need_build_bvh(bvh_layout)) {
         num_bvh++;
       }
     }
@@ -2245,7 +2256,7 @@ void MeshManager::device_update(Device *device,
     if (mesh->need_update) {
       pool.push(function_bind(
           &Mesh::compute_bvh, mesh, device, dscene, &scene->params, &progress, i, num_bvh));
-      if (mesh->need_build_bvh()) {
+      if (mesh->need_build_bvh(bvh_layout)) {
         i++;
       }
     }
diff --git a/intern/cycles/render/mesh.h b/intern/cycles/render/mesh.h
index 5bb6ab328b7..4a24a9c2656 100644
--- a/intern/cycles/render/mesh.h
+++ b/intern/cycles/render/mesh.h
@@ -19,6 +19,7 @@
 
 #include "graph/node.h"
 
+#include "bvh/bvh_params.h"
 #include "render/attribute.h"
 #include "render/shader.h"
 
@@ -94,7 +95,7 @@ class Mesh : public Node {
     int first_key;
     int num_keys;
 
-    int num_segments()
+    int num_segments() const
     {
       return num_keys - 1;
     }
@@ -167,6 +168,16 @@ class Mesh : public Node {
     return curve_first_key.size();
   }
 
+  size_t num_segments() const
+  {
+    return curve_keys.size() - curve_first_key.size();
+  }
+
+  size_t num_primitives() const
+  {
+    return num_triangles() + num_segments();
+  }
+
   /* Mesh SubdFace */
   struct SubdFace {
     int start_corner;
@@ -268,8 +279,18 @@ class Mesh : public Node {
 
   size_t attr_map_offset;
 
+  size_t prim_offset;
+
   size_t num_subd_verts;
 
+ private:
+  unordered_map<int, int> vert_to_stitching_key_map; /* real vert index -> stitching index */
+  unordered_multimap<int, int>
+      vert_stitching_map; /* stitching index -> multiple real vert indices */
+  friend class DiagSplit;
+  friend class MeshManager;
+
+ public:
   /* Functions */
   Mesh();
   ~Mesh();
@@ -332,8 +353,9 @@ class Mesh : public Node {
    *   same BVH tree.
    * - Special ray intersection is needed, for example to limit subsurface rays
    *   to only the mesh itself.
+   * - The BVH layout requires the top level to only contain instances.
    */
-  bool need_build_bvh() const;
+  bool need_build_bvh(BVHLayout layout) const;
 
   /* Check if the mesh should be treated as instanced. */
   bool is_instanced() const;
diff --git a/intern/cycles/render/mesh_displace.cpp b/intern/cycles/render/mesh_displace.cpp
index 5ae9348d83e..6a6c2fbb3eb 100644
--- a/intern/cycles/render/mesh_displace.cpp
+++ b/intern/cycles/render/mesh_displace.cpp
@@ -22,7 +22,9 @@
 #include "render/shader.h"
 
 #include "util/util_foreach.h"
+#include "util/util_map.h"
 #include "util/util_progress.h"
+#include "util/util_set.h"
 
 CCL_NAMESPACE_BEGIN
 
@@ -184,6 +186,38 @@ bool MeshManager::displace(
 
   d_output.free();
 
+  /* stitch */
+  unordered_set<int> stitch_keys;
+  for (pair<int, int> i : mesh->vert_to_stitching_key_map) {
+    stitch_keys.insert(i.second); /* stitching index */
+  }
+
+  typedef unordered_multimap<int, int>::iterator map_it_t;
+
+  for (int key : stitch_keys) {
+    pair<map_it_t, map_it_t> verts = mesh->vert_stitching_map.equal_range(key);
+
+    float3 pos = make_float3(0.0f, 0.0f, 0.0f);
+    int num = 0;
+
+    for (map_it_t v = verts.first; v != verts.second; ++v) {
+      int vert = v->second;
+
+      pos += mesh->verts[vert];
+      num++;
+    }
+
+    if (num <= 1) {
+      continue;
+    }
+
+    pos *= 1.0f / num;
+
+    for (map_it_t v = verts.first; v != verts.second; ++v) {
+      mesh->verts[v->second] = pos;
+    }
+  }
+
   /* for displacement method both, we only need to recompute the face
    * normals, as bump mapping in the shader will already alter the
    * vertex normal, so we start from the non-displaced vertex normals
@@ -238,7 +272,25 @@ bool MeshManager::displace(
     for (size_t i = 0; i < num_triangles; i++) {
       if (tri_has_true_disp[i]) {
         for (size_t j = 0; j < 3; j++) {
-          vN[mesh->get_triangle(i).v[j]] += fN[i];
+          int vert = mesh->get_triangle(i).v[j];
+          vN[vert] += fN[i];
+
+          /* add face normals to stitched vertices */
+          if (stitch_keys.size()) {
+            map_it_t key = mesh->vert_to_stitching_key_map.find(vert);
+
+            if (key != mesh->vert_to_stitching_key_map.end()) {
+              pair<map_it_t, map_it_t> verts = mesh->vert_stitching_map.equal_range(key->second);
+
+              for (map_it_t v = verts.first; v != verts.second; ++v) {
+                if (v->second == vert) {
+                  continue;
+                }
+
+                vN[v->second] += fN[i];
+              }
+            }
+          }
         }
       }
     }
@@ -289,8 +341,27 @@ bool MeshManager::displace(
         for (size_t i = 0; i < num_triangles; i++) {
           if (tri_has_true_disp[i]) {
             for (size_t j = 0; j < 3; j++) {
+              int vert = mesh->get_triangle(i).v[j];
               float3 fN = compute_face_normal(mesh->get_triangle(i), mP);
-              mN[mesh->get_triangle(i).v[j]] += fN;
+              mN[vert] += fN;
+
+              /* add face normals to stitched vertices */
+              if (stitch_keys.size()) {
+                map_it_t key = mesh->vert_to_stitching_key_map.find(vert);
+
+                if (key != mesh->vert_to_stitching_key_map.end()) {
+                  pair<map_it_t, map_it_t> verts = mesh->vert_stitching_map.equal_range(
+                      key->second);
+
+                  for (map_it_t v = verts.first; v != verts.second; ++v) {
+                    if (v->second == vert) {
+                      continue;
+                    }
+
+                    mN[v->second] += fN;
+                  }
+                }
+              }
             }
           }
         }
diff --git a/intern/cycles/render/mesh_subdivision.cpp b/intern/cycles/render/mesh_subdivision.cpp
index 46c8240fb71..a5a35fc049e 100644
--- a/intern/cycles/render/mesh_subdivision.cpp
+++ b/intern/cycles/render/mesh_subdivision.cpp
@@ -24,6 +24,7 @@
 
 #include "util/util_foreach.h"
 #include "util/util_algorithm.h"
+#include "util/util_hash.h"
 
 CCL_NAMESPACE_BEGIN
 
@@ -318,6 +319,9 @@ class OsdData {
 struct OsdPatch : Patch {
   OsdData *osd_data;
 
+  OsdPatch()
+  {
+  }
   OsdPatch(OsdData *data) : osd_data(data)
   {
   }
@@ -358,11 +362,6 @@ struct OsdPatch : Patch {
       *N = (t != 0.0f) ? *N / t : make_float3(0.0f, 0.0f, 1.0f);
     }
   }
-
-  BoundBox bound()
-  {
-    return BoundBox::empty;
-  }
 };
 
 #endif
@@ -397,29 +396,62 @@ void Mesh::tessellate(DiagSplit *split)
   Attribute *attr_vN = subd_attributes.find(ATTR_STD_VERTEX_NORMAL);
   float3 *vN = attr_vN->data_float3();
 
+  /* count patches */
+  int num_patches = 0;
   for (int f = 0; f < num_faces; f++) {
     SubdFace &face = subd_faces[f];
 
     if (face.is_quad()) {
-      /* quad */
-      QuadDice::SubPatch subpatch;
+      num_patches++;
+    }
+    else {
+      num_patches += face.num_corners;
+    }
+  }
 
-      LinearQuadPatch quad_patch;
+  /* build patches from faces */
 #ifdef WITH_OPENSUBDIV
-      OsdPatch osd_patch(&osd_data);
+  if (subdivision_type == SUBDIVISION_CATMULL_CLARK) {
+    vector<OsdPatch> osd_patches(num_patches, &osd_data);
+    OsdPatch *patch = osd_patches.data();
 
-      if (subdivision_type == SUBDIVISION_CATMULL_CLARK) {
-        osd_patch.patch_index = face.ptex_offset;
+    for (int f = 0; f < num_faces; f++) {
+      SubdFace &face = subd_faces[f];
 
-        subpatch.patch = &osd_patch;
+      if (face.is_quad()) {
+        patch->patch_index = face.ptex_offset;
+        patch->from_ngon = false;
+        patch->shader = face.shader;
+        patch++;
+      }
+      else {
+        for (int corner = 0; corner < face.num_corners; corner++) {
+          patch->patch_index = face.ptex_offset + corner;
+          patch->from_ngon = true;
+          patch->shader = face.shader;
+          patch++;
+        }
       }
-      else
+    }
+
+    /* split patches */
+    split->split_patches(osd_patches.data(), sizeof(OsdPatch));
+  }
+  else
 #endif
-      {
-        float3 *hull = quad_patch.hull;
-        float3 *normals = quad_patch.normals;
+  {
+    vector<LinearQuadPatch> linear_patches(num_patches);
+    LinearQuadPatch *patch = linear_patches.data();
+
+    for (int f = 0; f < num_faces; f++) {
+      SubdFace &face = subd_faces[f];
 
-        quad_patch.patch_index = face.ptex_offset;
+      if (face.is_quad()) {
+        float3 *hull = patch->hull;
+        float3 *normals = patch->normals;
+
+        patch->patch_index = face.ptex_offset;
+        patch->from_ngon = false;
 
         for (int i = 0; i < 4; i++) {
           hull[i] = verts[subd_face_corners[face.start_corner + i]];
@@ -440,55 +472,11 @@ void Mesh::tessellate(DiagSplit *split)
         swap(hull[2], hull[3]);
         swap(normals[2], normals[3]);
 
-        subpatch.patch = &quad_patch;
+        patch->shader = face.shader;
+        patch++;
       }
-
-      subpatch.patch->shader = face.shader;
-
-      /* Quad faces need to be split at least once to line up with split ngons, we do this
-       * here in this manner because if we do it later edge factors may end up slightly off.
-       */
-      subpatch.P00 = make_float2(0.0f, 0.0f);
-      subpatch.P10 = make_float2(0.5f, 0.0f);
-      subpatch.P01 = make_float2(0.0f, 0.5f);
-      subpatch.P11 = make_float2(0.5f, 0.5f);
-      split->split_quad(subpatch.patch, &subpatch);
-
-      subpatch.P00 = make_float2(0.5f, 0.0f);
-      subpatch.P10 = make_float2(1.0f, 0.0f);
-      subpatch.P01 = make_float2(0.5f, 0.5f);
-      subpatch.P11 = make_float2(1.0f, 0.5f);
-      split->split_quad(subpatch.patch, &subpatch);
-
-      subpatch.P00 = make_float2(0.0f, 0.5f);
-      subpatch.P10 = make_float2(0.5f, 0.5f);
-      subpatch.P01 = make_float2(0.0f, 1.0f);
-      subpatch.P11 = make_float2(0.5f, 1.0f);
-      split->split_quad(subpatch.patch, &subpatch);
-
-      subpatch.P00 = make_float2(0.5f, 0.5f);
-      subpatch.P10 = make_float2(1.0f, 0.5f);
-      subpatch.P01 = make_float2(0.5f, 1.0f);
-      subpatch.P11 = make_float2(1.0f, 1.0f);
-      split->split_quad(subpatch.patch, &subpatch);
-    }
-    else {
-      /* ngon */
-#ifdef WITH_OPENSUBDIV
-      if (subdivision_type == SUBDIVISION_CATMULL_CLARK) {
-        OsdPatch patch(&osd_data);
-
-        patch.shader = face.shader;
-
-        for (int corner = 0; corner < face.num_corners; corner++) {
-          patch.patch_index = face.ptex_offset + corner;
-
-          split->split_quad(&patch);
-        }
-      }
-      else
-#endif
-      {
+      else {
+        /* ngon */
         float3 center_vert = make_float3(0.0f, 0.0f, 0.0f);
         float3 center_normal = make_float3(0.0f, 0.0f, 0.0f);
 
@@ -499,13 +487,13 @@ void Mesh::tessellate(DiagSplit *split)
         }
 
         for (int corner = 0; corner < face.num_corners; corner++) {
-          LinearQuadPatch patch;
-          float3 *hull = patch.hull;
-          float3 *normals = patch.normals;
+          float3 *hull = patch->hull;
+          float3 *normals = patch->normals;
 
-          patch.patch_index = face.ptex_offset + corner;
+          patch->patch_index = face.ptex_offset + corner;
+          patch->from_ngon = true;
 
-          patch.shader = face.shader;
+          patch->shader = face.shader;
 
           hull[0] =
               verts[subd_face_corners[face.start_corner + mod(corner + 0, face.num_corners)]];
@@ -537,10 +525,13 @@ void Mesh::tessellate(DiagSplit *split)
             }
           }
 
-          split->split_quad(&patch);
+          patch++;
         }
       }
     }
+
+    /* split patches */
+    split->split_patches(linear_patches.data(), sizeof(LinearQuadPatch));
   }
 
   /* interpolate center points for attributes */
diff --git a/intern/cycles/render/nodes.cpp b/intern/cycles/render/nodes.cpp
index 9255181b421..bf8185d2b38 100644
--- a/intern/cycles/render/nodes.cpp
+++ b/intern/cycles/render/nodes.cpp
@@ -163,8 +163,10 @@ void TextureMapping::compile(SVMCompiler &compiler, int offset_in, int offset_ou
   }
 
   if (type == NORMAL) {
-    compiler.add_node(NODE_VECTOR_MATH, NODE_VECTOR_MATH_NORMALIZE, offset_out, offset_out);
-    compiler.add_node(NODE_VECTOR_MATH, SVM_STACK_INVALID, offset_out);
+    compiler.add_node(NODE_VECTOR_MATH,
+                      NODE_VECTOR_MATH_NORMALIZE,
+                      compiler.encode_uchar4(offset_out, offset_out, offset_out),
+                      compiler.encode_uchar4(SVM_STACK_INVALID, offset_out));
   }
 }
 
@@ -1108,6 +1110,53 @@ void IESLightNode::compile(OSLCompiler &compiler)
   compiler.add(this, "node_ies_light");
 }
 
+/* White Noise Texture */
+
+NODE_DEFINE(WhiteNoiseTextureNode)
+{
+  NodeType *type = NodeType::add("white_noise_texture", create, NodeType::SHADER);
+
+  static NodeEnum dimensions_enum;
+  dimensions_enum.insert("1D", 1);
+  dimensions_enum.insert("2D", 2);
+  dimensions_enum.insert("3D", 3);
+  dimensions_enum.insert("4D", 4);
+  SOCKET_ENUM(dimensions, "Dimensions", dimensions_enum, 3);
+
+  SOCKET_IN_POINT(vector, "Vector", make_float3(0.0f, 0.0f, 0.0f));
+  SOCKET_IN_FLOAT(w, "W", 0.0f);
+
+  SOCKET_OUT_FLOAT(value, "Value");
+
+  return type;
+}
+
+WhiteNoiseTextureNode::WhiteNoiseTextureNode() : ShaderNode(node_type)
+{
+}
+
+void WhiteNoiseTextureNode::compile(SVMCompiler &compiler)
+{
+  ShaderInput *vector_in = input("Vector");
+  ShaderInput *w_in = input("W");
+  ShaderOutput *value_out = output("Value");
+
+  int vector_stack_offset = compiler.stack_assign(vector_in);
+  int w_stack_offset = compiler.stack_assign(w_in);
+  int value_stack_offset = compiler.stack_assign(value_out);
+
+  compiler.add_node(NODE_TEX_WHITE_NOISE,
+                    dimensions,
+                    compiler.encode_uchar4(vector_stack_offset, w_stack_offset),
+                    value_stack_offset);
+}
+
+void WhiteNoiseTextureNode::compile(OSLCompiler &compiler)
+{
+  compiler.parameter(this, "dimensions");
+  compiler.add(this, "node_white_noise_texture");
+}
+
 /* Musgrave Texture */
 
 NODE_DEFINE(MusgraveTextureNode)
@@ -3891,6 +3940,7 @@ NODE_DEFINE(ObjectInfoNode)
   NodeType *type = NodeType::add("object_info", create, NodeType::SHADER);
 
   SOCKET_OUT_VECTOR(location, "Location");
+  SOCKET_OUT_COLOR(color, "Color");
   SOCKET_OUT_FLOAT(object_index, "Object Index");
   SOCKET_OUT_FLOAT(material_index, "Material Index");
   SOCKET_OUT_FLOAT(random, "Random");
@@ -3909,6 +3959,11 @@ void ObjectInfoNode::compile(SVMCompiler &compiler)
     compiler.add_node(NODE_OBJECT_INFO, NODE_INFO_OB_LOCATION, compiler.stack_assign(out));
   }
 
+  out = output("Color");
+  if (!out->links.empty()) {
+    compiler.add_node(NODE_OBJECT_INFO, NODE_INFO_OB_COLOR, compiler.stack_assign(out));
+  }
+
   out = output("Object Index");
   if (!out->links.empty()) {
     compiler.add_node(NODE_OBJECT_INFO, NODE_INFO_OB_INDEX, compiler.stack_assign(out));
@@ -4119,6 +4174,90 @@ void HairInfoNode::compile(OSLCompiler &compiler)
   compiler.add(this, "node_hair_info");
 }
 
+/* Volume Info */
+
+NODE_DEFINE(VolumeInfoNode)
+{
+  NodeType *type = NodeType::add("volume_info", create, NodeType::SHADER);
+
+  SOCKET_OUT_COLOR(color, "Color");
+  SOCKET_OUT_FLOAT(density, "Density");
+  SOCKET_OUT_FLOAT(flame, "Flame");
+  SOCKET_OUT_FLOAT(temperature, "Temperature");
+
+  return type;
+}
+
+VolumeInfoNode::VolumeInfoNode() : ShaderNode(node_type)
+{
+}
+
+/* The requested attributes are not updated after node expansion.
+ * So we explicitly request the required attributes.
+ */
+void VolumeInfoNode::attributes(Shader *shader, AttributeRequestSet *attributes)
+{
+  if (shader->has_volume) {
+    if (!output("Color")->links.empty()) {
+      attributes->add(ATTR_STD_VOLUME_COLOR);
+    }
+    if (!output("Density")->links.empty()) {
+      attributes->add(ATTR_STD_VOLUME_DENSITY);
+    }
+    if (!output("Flame")->links.empty()) {
+      attributes->add(ATTR_STD_VOLUME_FLAME);
+    }
+    if (!output("Temperature")->links.empty()) {
+      attributes->add(ATTR_STD_VOLUME_TEMPERATURE);
+    }
+    attributes->add(ATTR_STD_GENERATED_TRANSFORM);
+  }
+  ShaderNode::attributes(shader, attributes);
+}
+
+void VolumeInfoNode::expand(ShaderGraph *graph)
+{
+  ShaderOutput *color_out = output("Color");
+  if (!color_out->links.empty()) {
+    AttributeNode *attr = new AttributeNode();
+    attr->attribute = "color";
+    graph->add(attr);
+    graph->relink(color_out, attr->output("Color"));
+  }
+
+  ShaderOutput *density_out = output("Density");
+  if (!density_out->links.empty()) {
+    AttributeNode *attr = new AttributeNode();
+    attr->attribute = "density";
+    graph->add(attr);
+    graph->relink(density_out, attr->output("Fac"));
+  }
+
+  ShaderOutput *flame_out = output("Flame");
+  if (!flame_out->links.empty()) {
+    AttributeNode *attr = new AttributeNode();
+    attr->attribute = "flame";
+    graph->add(attr);
+    graph->relink(flame_out, attr->output("Fac"));
+  }
+
+  ShaderOutput *temperature_out = output("Temperature");
+  if (!temperature_out->links.empty()) {
+    AttributeNode *attr = new AttributeNode();
+    attr->attribute = "temperature";
+    graph->add(attr);
+    graph->relink(temperature_out, attr->output("Fac"));
+  }
+}
+
+void VolumeInfoNode::compile(SVMCompiler &)
+{
+}
+
+void VolumeInfoNode::compile(OSLCompiler &)
+{
+}
+
 /* Value */
 
 NODE_DEFINE(ValueNode)
@@ -5422,14 +5561,14 @@ NODE_DEFINE(MathNode)
   type_enum.insert("arctan2", NODE_MATH_ARCTAN2);
   type_enum.insert("floor", NODE_MATH_FLOOR);
   type_enum.insert("ceil", NODE_MATH_CEIL);
-  type_enum.insert("fract", NODE_MATH_FRACT);
+  type_enum.insert("fraction", NODE_MATH_FRACTION);
   type_enum.insert("sqrt", NODE_MATH_SQRT);
   SOCKET_ENUM(type, "Type", type_enum, NODE_MATH_ADD);
 
   SOCKET_BOOLEAN(use_clamp, "Use Clamp", false);
 
-  SOCKET_IN_FLOAT(value1, "Value1", 0.0f);
-  SOCKET_IN_FLOAT(value2, "Value2", 0.0f);
+  SOCKET_IN_FLOAT(value1, "Value1", 0.5f);
+  SOCKET_IN_FLOAT(value2, "Value2", 0.5f);
 
   SOCKET_OUT_FLOAT(value, "Value");
 
@@ -5440,13 +5579,28 @@ MathNode::MathNode() : ShaderNode(node_type)
 {
 }
 
+void MathNode::expand(ShaderGraph *graph)
+{
+  if (use_clamp) {
+    ShaderOutput *result_out = output("Value");
+    if (!result_out->links.empty()) {
+      ClampNode *clamp_node = new ClampNode();
+      clamp_node->min = 0.0f;
+      clamp_node->max = 1.0f;
+      graph->add(clamp_node);
+      graph->relink(result_out, clamp_node->output("Result"));
+      graph->connect(result_out, clamp_node->input("Value"));
+    }
+  }
+}
+
 void MathNode::constant_fold(const ConstantFolder &folder)
 {
   if (folder.all_inputs_constant()) {
-    folder.make_constant_clamp(svm_math(type, value1, value2), use_clamp);
+    folder.make_constant(svm_math(type, value1, value2));
   }
   else {
-    folder.fold_math(type, use_clamp);
+    folder.fold_math(type);
   }
 }
 
@@ -5456,20 +5610,19 @@ void MathNode::compile(SVMCompiler &compiler)
   ShaderInput *value2_in = input("Value2");
   ShaderOutput *value_out = output("Value");
 
-  compiler.add_node(
-      NODE_MATH, type, compiler.stack_assign(value1_in), compiler.stack_assign(value2_in));
-  compiler.add_node(NODE_MATH, compiler.stack_assign(value_out));
+  int value1_stack_offset = compiler.stack_assign(value1_in);
+  int value2_stack_offset = compiler.stack_assign(value2_in);
+  int value_stack_offset = compiler.stack_assign(value_out);
 
-  if (use_clamp) {
-    compiler.add_node(NODE_MATH, NODE_MATH_CLAMP, compiler.stack_assign(value_out));
-    compiler.add_node(NODE_MATH, compiler.stack_assign(value_out));
-  }
+  compiler.add_node(NODE_MATH,
+                    type,
+                    compiler.encode_uchar4(value1_stack_offset, value2_stack_offset),
+                    value_stack_offset);
 }
 
 void MathNode::compile(OSLCompiler &compiler)
 {
   compiler.parameter(this, "type");
-  compiler.parameter(this, "use_clamp");
   compiler.add(this, "node_math");
 }
 
@@ -5482,14 +5635,32 @@ NODE_DEFINE(VectorMathNode)
   static NodeEnum type_enum;
   type_enum.insert("add", NODE_VECTOR_MATH_ADD);
   type_enum.insert("subtract", NODE_VECTOR_MATH_SUBTRACT);
-  type_enum.insert("average", NODE_VECTOR_MATH_AVERAGE);
-  type_enum.insert("dot_product", NODE_VECTOR_MATH_DOT_PRODUCT);
+  type_enum.insert("multiply", NODE_VECTOR_MATH_MULTIPLY);
+  type_enum.insert("divide", NODE_VECTOR_MATH_DIVIDE);
+
   type_enum.insert("cross_product", NODE_VECTOR_MATH_CROSS_PRODUCT);
+  type_enum.insert("project", NODE_VECTOR_MATH_PROJECT);
+  type_enum.insert("reflect", NODE_VECTOR_MATH_REFLECT);
+  type_enum.insert("dot_product", NODE_VECTOR_MATH_DOT_PRODUCT);
+
+  type_enum.insert("distance", NODE_VECTOR_MATH_DISTANCE);
+  type_enum.insert("length", NODE_VECTOR_MATH_LENGTH);
+  type_enum.insert("scale", NODE_VECTOR_MATH_SCALE);
   type_enum.insert("normalize", NODE_VECTOR_MATH_NORMALIZE);
+
+  type_enum.insert("snap", NODE_VECTOR_MATH_SNAP);
+  type_enum.insert("floor", NODE_VECTOR_MATH_FLOOR);
+  type_enum.insert("ceil", NODE_VECTOR_MATH_CEIL);
+  type_enum.insert("modulo", NODE_VECTOR_MATH_MODULO);
+  type_enum.insert("fraction", NODE_VECTOR_MATH_FRACTION);
+  type_enum.insert("absolute", NODE_VECTOR_MATH_ABSOLUTE);
+  type_enum.insert("minimum", NODE_VECTOR_MATH_MINIMUM);
+  type_enum.insert("maximum", NODE_VECTOR_MATH_MAXIMUM);
   SOCKET_ENUM(type, "Type", type_enum, NODE_VECTOR_MATH_ADD);
 
   SOCKET_IN_VECTOR(vector1, "Vector1", make_float3(0.0f, 0.0f, 0.0f));
   SOCKET_IN_VECTOR(vector2, "Vector2", make_float3(0.0f, 0.0f, 0.0f));
+  SOCKET_IN_FLOAT(scale, "Scale", 1.0f);
 
   SOCKET_OUT_FLOAT(value, "Value");
   SOCKET_OUT_VECTOR(vector, "Vector");
@@ -5503,12 +5674,11 @@ VectorMathNode::VectorMathNode() : ShaderNode(node_type)
 
 void VectorMathNode::constant_fold(const ConstantFolder &folder)
 {
-  float value;
-  float3 vector;
+  float value = 0.0f;
+  float3 vector = make_float3(0.0f, 0.0f, 0.0f);
 
   if (folder.all_inputs_constant()) {
-    svm_vector_math(&value, &vector, type, vector1, vector2);
-
+    svm_vector_math(&value, &vector, type, vector1, vector2, scale);
     if (folder.output == output("Value")) {
       folder.make_constant(value);
     }
@@ -5525,15 +5695,21 @@ void VectorMathNode::compile(SVMCompiler &compiler)
 {
   ShaderInput *vector1_in = input("Vector1");
   ShaderInput *vector2_in = input("Vector2");
+  ShaderInput *scale_in = input("Scale");
   ShaderOutput *value_out = output("Value");
   ShaderOutput *vector_out = output("Vector");
 
-  compiler.add_node(NODE_VECTOR_MATH,
-                    type,
-                    compiler.stack_assign(vector1_in),
-                    compiler.stack_assign(vector2_in));
+  int vector1_stack_offset = compiler.stack_assign(vector1_in);
+  int vector2_stack_offset = compiler.stack_assign(vector2_in);
+  int scale_stack_offset = compiler.stack_assign(scale_in);
+  int value_stack_offset = compiler.stack_assign_if_linked(value_out);
+  int vector_stack_offset = compiler.stack_assign_if_linked(vector_out);
+
   compiler.add_node(
-      NODE_VECTOR_MATH, compiler.stack_assign(value_out), compiler.stack_assign(vector_out));
+      NODE_VECTOR_MATH,
+      type,
+      compiler.encode_uchar4(vector1_stack_offset, vector2_stack_offset, scale_stack_offset),
+      compiler.encode_uchar4(value_stack_offset, vector_stack_offset));
 }
 
 void VectorMathNode::compile(OSLCompiler &compiler)
diff --git a/intern/cycles/render/nodes.h b/intern/cycles/render/nodes.h
index d4708adf9af..fbed2ff0ef6 100644
--- a/intern/cycles/render/nodes.h
+++ b/intern/cycles/render/nodes.h
@@ -370,6 +370,19 @@ class IESLightNode : public TextureNode {
   void get_slot();
 };
 
+class WhiteNoiseTextureNode : public ShaderNode {
+ public:
+  SHADER_NODE_CLASS(WhiteNoiseTextureNode)
+  virtual int get_group()
+  {
+    return NODE_GROUP_LEVEL_2;
+  }
+
+  int dimensions;
+  float3 vector;
+  float w;
+};
+
 class MappingNode : public ShaderNode {
  public:
   SHADER_NODE_CLASS(MappingNode)
@@ -948,6 +961,21 @@ class HairInfoNode : public ShaderNode {
   }
 };
 
+class VolumeInfoNode : public ShaderNode {
+ public:
+  SHADER_NODE_CLASS(VolumeInfoNode)
+  void attributes(Shader *shader, AttributeRequestSet *attributes);
+  bool has_attribute_dependency()
+  {
+    return true;
+  }
+  bool has_spatial_varying()
+  {
+    return true;
+  }
+  void expand(ShaderGraph *graph);
+};
+
 class ValueNode : public ShaderNode {
  public:
   SHADER_NODE_CLASS(ValueNode)
@@ -1260,11 +1288,12 @@ class MathNode : public ShaderNode {
   {
     return NODE_GROUP_LEVEL_1;
   }
+  void expand(ShaderGraph *graph);
   void constant_fold(const ConstantFolder &folder);
 
   float value1;
   float value2;
-  NodeMath type;
+  NodeMathType type;
   bool use_clamp;
 };
 
@@ -1291,7 +1320,8 @@ class VectorMathNode : public ShaderNode {
 
   float3 vector1;
   float3 vector2;
-  NodeVectorMath type;
+  float scale;
+  NodeVectorMathType type;
 };
 
 class VectorTransformNode : public ShaderNode {
diff --git a/intern/cycles/render/object.cpp b/intern/cycles/render/object.cpp
index 6c6f8810412..849329a086d 100644
--- a/intern/cycles/render/object.cpp
+++ b/intern/cycles/render/object.cpp
@@ -90,6 +90,7 @@ NODE_DEFINE(Object)
   SOCKET_NODE(mesh, "Mesh", &Mesh::node_type);
   SOCKET_TRANSFORM(tfm, "Transform", transform_identity());
   SOCKET_UINT(visibility, "Visibility", ~0);
+  SOCKET_COLOR(color, "Color", make_float3(0.0f, 0.0f, 0.0f));
   SOCKET_UINT(random_id, "Random ID", 0);
   SOCKET_INT(pass_id, "Pass ID", 0);
   SOCKET_BOOLEAN(use_holdout, "Use Holdout", false);
@@ -371,6 +372,7 @@ void ObjectManager::device_update_object_transform(UpdateObjectTransformState *s
    */
   float uniform_scale;
   float surface_area = 0.0f;
+  float3 color = ob->color;
   float pass_id = ob->pass_id;
   float random_number = (float)ob->random_id * (1.0f / (float)0xFFFFFFFF);
   int particle_index = (ob->particle_system) ?
@@ -425,6 +427,9 @@ void ObjectManager::device_update_object_transform(UpdateObjectTransformState *s
   kobject.tfm = tfm;
   kobject.itfm = itfm;
   kobject.surface_area = surface_area;
+  kobject.color[0] = color.x;
+  kobject.color[1] = color.y;
+  kobject.color[2] = color.z;
   kobject.pass_id = pass_id;
   kobject.random_number = random_number;
   kobject.particle_index = particle_index;
diff --git a/intern/cycles/render/object.h b/intern/cycles/render/object.h
index 2fd43900da1..cbbff0d4c6d 100644
--- a/intern/cycles/render/object.h
+++ b/intern/cycles/render/object.h
@@ -51,6 +51,7 @@ class Object : public Node {
   BoundBox bounds;
   uint random_id;
   int pass_id;
+  float3 color;
   ustring asset_name;
   vector<ParamValue> attributes;
   uint visibility;
diff --git a/intern/cycles/render/session.cpp b/intern/cycles/render/session.cpp
index 157bb6d63d3..332f1fb6167 100644
--- a/intern/cycles/render/session.cpp
+++ b/intern/cycles/render/session.cpp
@@ -232,7 +232,7 @@ void Session::run_gpu()
     }
 
     /* Don't go in pause mode when image was rendered with preview kernels
-     * When feature kernels become available the session will be resetted. */
+     * When feature kernels become available the session will be reset. */
     else if (no_tiles && kernel_state == DEVICE_KERNEL_WAITING_FOR_FEATURE_KERNEL) {
       time_sleep(0.1);
     }
diff --git a/intern/cycles/render/stats.h b/intern/cycles/render/stats.h
index f1bf1903483..e45403a3754 100644
--- a/intern/cycles/render/stats.h
+++ b/intern/cycles/render/stats.h
@@ -29,7 +29,7 @@ CCL_NAMESPACE_BEGIN
  * semantic around the units of size, it just should be the same for all
  * entries.
  *
- * This is a generic entry foi all size-related statistics, which helps
+ * This is a generic entry for all size-related statistics, which helps
  * avoiding duplicating code for things like sorting.
  */
 class NamedSizeEntry {
diff --git a/intern/cycles/subd/CMakeLists.txt b/intern/cycles/subd/CMakeLists.txt
index f5ceaa0436d..b874c5c3c2d 100644
--- a/intern/cycles/subd/CMakeLists.txt
+++ b/intern/cycles/subd/CMakeLists.txt
@@ -19,6 +19,7 @@ set(SRC_HEADERS
   subd_patch.h
   subd_patch_table.h
   subd_split.h
+  subd_subpatch.h
 )
 
 set(LIB
diff --git a/intern/cycles/subd/subd_dice.cpp b/intern/cycles/subd/subd_dice.cpp
index fb96be5065b..714ff6ed629 100644
--- a/intern/cycles/subd/subd_dice.cpp
+++ b/intern/cycles/subd/subd_dice.cpp
@@ -38,107 +38,91 @@ EdgeDice::EdgeDice(const SubdParams &params_) : params(params_)
   }
 }
 
-void EdgeDice::reserve(int num_verts)
+void EdgeDice::reserve(int num_verts, int num_triangles)
 {
   Mesh *mesh = params.mesh;
 
   vert_offset = mesh->verts.size();
   tri_offset = mesh->num_triangles();
 
-  /* todo: optimize so we can reserve in advance, this is like push_back_slow() */
-  if (vert_offset + num_verts > mesh->verts.capacity()) {
-    mesh->reserve_mesh(size_t((vert_offset + num_verts) * 1.2), mesh->num_triangles());
-  }
-
-  mesh->resize_mesh(vert_offset + num_verts, tri_offset);
+  mesh->resize_mesh(mesh->verts.size() + num_verts, mesh->num_triangles());
+  mesh->reserve_mesh(mesh->verts.size() + num_verts, mesh->num_triangles() + num_triangles);
 
   Attribute *attr_vN = mesh->attributes.add(ATTR_STD_VERTEX_NORMAL);
 
-  mesh_P = mesh->verts.data();
-  mesh_N = attr_vN->data_float3();
+  mesh_P = mesh->verts.data() + vert_offset;
+  mesh_N = attr_vN->data_float3() + vert_offset;
+
+  params.mesh->num_subd_verts += num_verts;
 }
 
-int EdgeDice::add_vert(Patch *patch, float2 uv)
+void EdgeDice::set_vert(Patch *patch, int index, float2 uv)
 {
   float3 P, N;
 
   patch->eval(&P, NULL, NULL, &N, uv.x, uv.y);
 
-  assert(vert_offset < params.mesh->verts.size());
-
-  mesh_P[vert_offset] = P;
-  mesh_N[vert_offset] = N;
-  params.mesh->vert_patch_uv[vert_offset] = make_float2(uv.x, uv.y);
-
-  if (params.ptex) {
-    Attribute *attr_ptex_uv = params.mesh->attributes.add(ATTR_STD_PTEX_UV);
-    params.mesh->attributes.resize();
+  assert(index < params.mesh->verts.size());
 
-    float3 *ptex_uv = attr_ptex_uv->data_float3();
-    ptex_uv[vert_offset] = make_float3(uv.x, uv.y, 0.0f);
-  }
-
-  params.mesh->num_subd_verts++;
-
-  return vert_offset++;
+  mesh_P[index] = P;
+  mesh_N[index] = N;
+  params.mesh->vert_patch_uv[index + vert_offset] = make_float2(uv.x, uv.y);
 }
 
 void EdgeDice::add_triangle(Patch *patch, int v0, int v1, int v2)
 {
   Mesh *mesh = params.mesh;
 
-  /* todo: optimize so we can reserve in advance, this is like push_back_slow() */
-  if (mesh->triangles.size() == mesh->triangles.capacity())
-    mesh->reserve_mesh(mesh->verts.size(), size_t(max(mesh->num_triangles() + 1, 1) * 1.2));
-
-  mesh->add_triangle(v0, v1, v2, patch->shader, true);
+  mesh->add_triangle(v0 + vert_offset, v1 + vert_offset, v2 + vert_offset, patch->shader, true);
   params.mesh->triangle_patch[params.mesh->num_triangles() - 1] = patch->patch_index;
 
-  if (params.ptex) {
-    Attribute *attr_ptex_face_id = params.mesh->attributes.add(ATTR_STD_PTEX_FACE_ID);
-    params.mesh->attributes.resize();
-
-    float *ptex_face_id = attr_ptex_face_id->data_float();
-    ptex_face_id[tri_offset] = (float)patch->ptex_face_id();
-  }
-
   tri_offset++;
 }
 
-void EdgeDice::stitch_triangles(Patch *patch, vector<int> &outer, vector<int> &inner)
+void EdgeDice::stitch_triangles(Subpatch &sub, int edge)
 {
-  if (inner.size() == 0 || outer.size() == 0)
+  int Mu = max(sub.edge_u0.T, sub.edge_u1.T);
+  int Mv = max(sub.edge_v0.T, sub.edge_v1.T);
+  Mu = max(Mu, 2);
+  Mv = max(Mv, 2);
+
+  int outer_T = sub.edges[edge].T;
+  int inner_T = ((edge % 2) == 0) ? Mv - 2 : Mu - 2;
+
+  if (inner_T < 0 || outer_T < 0)
     return;  // XXX avoid crashes for Mu or Mv == 1, missing polygons
 
   /* stitch together two arrays of verts with triangles. at each step,
    * we compare using the next verts on both sides, to find the split
    * direction with the smallest diagonal, and use that in order to keep
    * the triangle shape reasonable. */
-  for (size_t i = 0, j = 0; i + 1 < inner.size() || j + 1 < outer.size();) {
+  for (size_t i = 0, j = 0; i < inner_T || j < outer_T;) {
     int v0, v1, v2;
 
-    v0 = inner[i];
-    v1 = outer[j];
+    v0 = sub.get_vert_along_grid_edge(edge, i);
+    v1 = sub.get_vert_along_edge(edge, j);
 
-    if (j + 1 == outer.size()) {
-      v2 = inner[++i];
+    if (j == outer_T) {
+      v2 = sub.get_vert_along_grid_edge(edge, ++i);
     }
-    else if (i + 1 == inner.size()) {
-      v2 = outer[++j];
+    else if (i == inner_T) {
+      v2 = sub.get_vert_along_edge(edge, ++j);
     }
     else {
       /* length of diagonals */
-      float len1 = len_squared(mesh_P[inner[i]] - mesh_P[outer[j + 1]]);
-      float len2 = len_squared(mesh_P[outer[j]] - mesh_P[inner[i + 1]]);
+      float len1 = len_squared(mesh_P[sub.get_vert_along_grid_edge(edge, i)] -
+                               mesh_P[sub.get_vert_along_edge(edge, j + 1)]);
+      float len2 = len_squared(mesh_P[sub.get_vert_along_edge(edge, j)] -
+                               mesh_P[sub.get_vert_along_grid_edge(edge, i + 1)]);
 
       /* use smallest diagonal */
       if (len1 < len2)
-        v2 = outer[++j];
+        v2 = sub.get_vert_along_edge(edge, ++j);
       else
-        v2 = inner[++i];
+        v2 = sub.get_vert_along_grid_edge(edge, ++i);
     }
 
-    add_triangle(patch, v0, v1, v2);
+    add_triangle(sub.patch, v1, v0, v2);
   }
 }
 
@@ -148,22 +132,15 @@ QuadDice::QuadDice(const SubdParams &params_) : EdgeDice(params_)
 {
 }
 
-void QuadDice::reserve(EdgeFactors &ef, int Mu, int Mv)
-{
-  /* XXX need to make this also work for edge factor 0 and 1 */
-  int num_verts = (ef.tu0 + ef.tu1 + ef.tv0 + ef.tv1) + (Mu - 1) * (Mv - 1);
-  EdgeDice::reserve(num_verts);
-}
-
-float2 QuadDice::map_uv(SubPatch &sub, float u, float v)
+float2 QuadDice::map_uv(Subpatch &sub, float u, float v)
 {
   /* map UV from subpatch to patch parametric coordinates */
-  float2 d0 = interp(sub.P00, sub.P01, v);
-  float2 d1 = interp(sub.P10, sub.P11, v);
+  float2 d0 = interp(sub.c00, sub.c01, v);
+  float2 d1 = interp(sub.c10, sub.c11, v);
   return interp(d0, d1, u);
 }
 
-float3 QuadDice::eval_projected(SubPatch &sub, float u, float v)
+float3 QuadDice::eval_projected(Subpatch &sub, float u, float v)
 {
   float2 uv = map_uv(sub, u, v);
   float3 P;
@@ -175,70 +152,40 @@ float3 QuadDice::eval_projected(SubPatch &sub, float u, float v)
   return P;
 }
 
-int QuadDice::add_vert(SubPatch &sub, float u, float v)
+void QuadDice::set_vert(Subpatch &sub, int index, float u, float v)
 {
-  return EdgeDice::add_vert(sub.patch, map_uv(sub, u, v));
+  EdgeDice::set_vert(sub.patch, index, map_uv(sub, u, v));
 }
 
-void QuadDice::add_side_u(SubPatch &sub,
-                          vector<int> &outer,
-                          vector<int> &inner,
-                          int Mu,
-                          int Mv,
-                          int tu,
-                          int side,
-                          int offset)
+void QuadDice::set_side(Subpatch &sub, int edge)
 {
-  outer.clear();
-  inner.clear();
+  int t = sub.edges[edge].T;
 
   /* set verts on the edge of the patch */
-  outer.push_back(offset + ((side) ? 2 : 0));
-
-  for (int i = 1; i < tu; i++) {
-    float u = i / (float)tu;
-    float v = (side) ? 1.0f : 0.0f;
-
-    outer.push_back(add_vert(sub, u, v));
-  }
-
-  outer.push_back(offset + ((side) ? 3 : 1));
-
-  /* set verts on the edge of the inner grid */
-  for (int i = 0; i < Mu - 1; i++) {
-    int j = (side) ? Mv - 1 - 1 : 0;
-    inner.push_back(offset + 4 + i + j * (Mu - 1));
-  }
-}
-
-void QuadDice::add_side_v(SubPatch &sub,
-                          vector<int> &outer,
-                          vector<int> &inner,
-                          int Mu,
-                          int Mv,
-                          int tv,
-                          int side,
-                          int offset)
-{
-  outer.clear();
-  inner.clear();
-
-  /* set verts on the edge of the patch */
-  outer.push_back(offset + ((side) ? 1 : 0));
-
-  for (int j = 1; j < tv; j++) {
-    float u = (side) ? 1.0f : 0.0f;
-    float v = j / (float)tv;
-
-    outer.push_back(add_vert(sub, u, v));
-  }
-
-  outer.push_back(offset + ((side) ? 3 : 2));
+  for (int i = 0; i < t; i++) {
+    float f = i / (float)t;
+
+    float u, v;
+    switch (edge) {
+      case 0:
+        u = 0;
+        v = f;
+        break;
+      case 1:
+        u = f;
+        v = 1;
+        break;
+      case 2:
+        u = 1;
+        v = 1.0f - f;
+        break;
+      case 3:
+        u = 1.0f - f;
+        v = 0;
+        break;
+    }
 
-  /* set verts on the edge of the inner grid */
-  for (int j = 0; j < Mv - 1; j++) {
-    int i = (side) ? Mu - 1 - 1 : 0;
-    inner.push_back(offset + 4 + i + j * (Mu - 1));
+    set_vert(sub, sub.get_vert_along_edge(edge, i), u, v);
   }
 }
 
@@ -247,7 +194,7 @@ float QuadDice::quad_area(const float3 &a, const float3 &b, const float3 &c, con
   return triangle_area(a, b, d) + triangle_area(a, d, c);
 }
 
-float QuadDice::scale_factor(SubPatch &sub, EdgeFactors &ef, int Mu, int Mv)
+float QuadDice::scale_factor(Subpatch &sub, int Mu, int Mv)
 {
   /* estimate area as 4x largest of 4 quads */
   float3 P[3][3];
@@ -269,23 +216,14 @@ float QuadDice::scale_factor(SubPatch &sub, EdgeFactors &ef, int Mu, int Mv)
   // XXX does the -sqrt solution matter
   // XXX max(D, 0.0) is highly suspicious, need to test cases
   // where D goes negative
-  float N = 0.5f * (Ntris - (ef.tu0 + ef.tu1 + ef.tv0 + ef.tv1));
+  float N = 0.5f * (Ntris - (sub.edge_u0.T + sub.edge_u1.T + sub.edge_v0.T + sub.edge_v1.T));
   float D = 4.0f * N * Mu * Mv + (Mu + Mv) * (Mu + Mv);
   float S = (Mu + Mv + sqrtf(max(D, 0.0f))) / (2 * Mu * Mv);
 
   return S;
 }
 
-void QuadDice::add_corners(SubPatch &sub)
-{
-  /* add verts for patch corners */
-  add_vert(sub, 0.0f, 0.0f);
-  add_vert(sub, 1.0f, 0.0f);
-  add_vert(sub, 0.0f, 1.0f);
-  add_vert(sub, 1.0f, 1.0f);
-}
-
-void QuadDice::add_grid(SubPatch &sub, int Mu, int Mv, int offset)
+void QuadDice::add_grid(Subpatch &sub, int Mu, int Mv, int offset)
 {
   /* create inner grid */
   float du = 1.0f / (float)Mu;
@@ -296,13 +234,13 @@ void QuadDice::add_grid(SubPatch &sub, int Mu, int Mv, int offset)
       float u = i * du;
       float v = j * dv;
 
-      add_vert(sub, u, v);
+      set_vert(sub, offset + (i - 1) + (j - 1) * (Mu - 1), u, v);
 
       if (i < Mu - 1 && j < Mv - 1) {
-        int i1 = offset + 4 + (i - 1) + (j - 1) * (Mu - 1);
-        int i2 = offset + 4 + i + (j - 1) * (Mu - 1);
-        int i3 = offset + 4 + i + j * (Mu - 1);
-        int i4 = offset + 4 + (i - 1) + j * (Mu - 1);
+        int i1 = offset + (i - 1) + (j - 1) * (Mu - 1);
+        int i2 = offset + i + (j - 1) * (Mu - 1);
+        int i3 = offset + i + j * (Mu - 1);
+        int i4 = offset + (i - 1) + j * (Mu - 1);
 
         add_triangle(sub.patch, i1, i2, i3);
         add_triangle(sub.patch, i1, i3, i4);
@@ -311,11 +249,11 @@ void QuadDice::add_grid(SubPatch &sub, int Mu, int Mv, int offset)
   }
 }
 
-void QuadDice::dice(SubPatch &sub, EdgeFactors &ef)
+void QuadDice::dice(Subpatch &sub)
 {
   /* compute inner grid size with scale factor */
-  int Mu = max(ef.tu0, ef.tu1);
-  int Mv = max(ef.tv0, ef.tv1);
+  int Mu = max(sub.edge_u0.T, sub.edge_u1.T);
+  int Mv = max(sub.edge_v0.T, sub.edge_v1.T);
 
 #if 0 /* Doesn't work very well, especially at grazing angles. */
   float S = scale_factor(sub, ef, Mu, Mv);
@@ -323,36 +261,22 @@ void QuadDice::dice(SubPatch &sub, EdgeFactors &ef)
   float S = 1.0f;
 #endif
 
-  Mu = max((int)ceil(S * Mu), 2);  // XXX handle 0 & 1?
-  Mv = max((int)ceil(S * Mv), 2);  // XXX handle 0 & 1?
-
-  /* reserve space for new verts */
-  int offset = params.mesh->verts.size();
-  reserve(ef, Mu, Mv);
-
-  /* corners and inner grid */
-  add_corners(sub);
-  add_grid(sub, Mu, Mv, offset);
-
-  /* bottom side */
-  vector<int> outer, inner;
-
-  add_side_u(sub, outer, inner, Mu, Mv, ef.tu0, 0, offset);
-  stitch_triangles(sub.patch, outer, inner);
-
-  /* top side */
-  add_side_u(sub, outer, inner, Mu, Mv, ef.tu1, 1, offset);
-  stitch_triangles(sub.patch, inner, outer);
+  Mu = max((int)ceilf(S * Mu), 2);  // XXX handle 0 & 1?
+  Mv = max((int)ceilf(S * Mv), 2);  // XXX handle 0 & 1?
 
-  /* left side */
-  add_side_v(sub, outer, inner, Mu, Mv, ef.tv0, 0, offset);
-  stitch_triangles(sub.patch, inner, outer);
+  /* inner grid */
+  add_grid(sub, Mu, Mv, sub.inner_grid_vert_offset);
 
-  /* right side */
-  add_side_v(sub, outer, inner, Mu, Mv, ef.tv1, 1, offset);
-  stitch_triangles(sub.patch, outer, inner);
+  /* sides */
+  set_side(sub, 0);
+  set_side(sub, 1);
+  set_side(sub, 2);
+  set_side(sub, 3);
 
-  assert(vert_offset == params.mesh->verts.size());
+  stitch_triangles(sub, 0);
+  stitch_triangles(sub, 1);
+  stitch_triangles(sub, 2);
+  stitch_triangles(sub, 3);
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/subd/subd_dice.h b/intern/cycles/subd/subd_dice.h
index eee54e01861..ee63403d40c 100644
--- a/intern/cycles/subd/subd_dice.h
+++ b/intern/cycles/subd/subd_dice.h
@@ -25,6 +25,8 @@
 #include "util/util_types.h"
 #include "util/util_vector.h"
 
+#include "subd/subd_subpatch.h"
+
 CCL_NAMESPACE_BEGIN
 
 class Camera;
@@ -67,78 +69,33 @@ class EdgeDice {
 
   explicit EdgeDice(const SubdParams &params);
 
-  void reserve(int num_verts);
+  void reserve(int num_verts, int num_triangles);
 
-  int add_vert(Patch *patch, float2 uv);
+  void set_vert(Patch *patch, int index, float2 uv);
   void add_triangle(Patch *patch, int v0, int v1, int v2);
 
-  void stitch_triangles(Patch *patch, vector<int> &outer, vector<int> &inner);
+  void stitch_triangles(Subpatch &sub, int edge);
 };
 
-/* Quad EdgeDice
- *
- * Edge tessellation factors and subpatch coordinates are as follows:
- *
- *            tu1
- *     P01 --------- P11
- *     |               |
- * tv0 |               | tv1
- *     |               |
- *     P00 --------- P10
- *            tu0
- */
+/* Quad EdgeDice */
 
 class QuadDice : public EdgeDice {
  public:
-  struct SubPatch {
-    Patch *patch;
-
-    float2 P00;
-    float2 P10;
-    float2 P01;
-    float2 P11;
-  };
-
-  struct EdgeFactors {
-    int tu0;
-    int tu1;
-    int tv0;
-    int tv1;
-  };
-
   explicit QuadDice(const SubdParams &params);
 
-  void reserve(EdgeFactors &ef, int Mu, int Mv);
-  float3 eval_projected(SubPatch &sub, float u, float v);
-
-  float2 map_uv(SubPatch &sub, float u, float v);
-  int add_vert(SubPatch &sub, float u, float v);
-
-  void add_corners(SubPatch &sub);
-  void add_grid(SubPatch &sub, int Mu, int Mv, int offset);
-
-  void add_side_u(SubPatch &sub,
-                  vector<int> &outer,
-                  vector<int> &inner,
-                  int Mu,
-                  int Mv,
-                  int tu,
-                  int side,
-                  int offset);
-
-  void add_side_v(SubPatch &sub,
-                  vector<int> &outer,
-                  vector<int> &inner,
-                  int Mu,
-                  int Mv,
-                  int tv,
-                  int side,
-                  int offset);
+  float3 eval_projected(Subpatch &sub, float u, float v);
+
+  float2 map_uv(Subpatch &sub, float u, float v);
+  void set_vert(Subpatch &sub, int index, float u, float v);
+
+  void add_grid(Subpatch &sub, int Mu, int Mv, int offset);
+
+  void set_side(Subpatch &sub, int edge);
 
   float quad_area(const float3 &a, const float3 &b, const float3 &c, const float3 &d);
-  float scale_factor(SubPatch &sub, EdgeFactors &ef, int Mu, int Mv);
+  float scale_factor(Subpatch &sub, int Mu, int Mv);
 
-  void dice(SubPatch &sub, EdgeFactors &ef);
+  void dice(Subpatch &sub);
 };
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/subd/subd_patch.h b/intern/cycles/subd/subd_patch.h
index 5209d4d0b07..2e3484f95e3 100644
--- a/intern/cycles/subd/subd_patch.h
+++ b/intern/cycles/subd/subd_patch.h
@@ -24,18 +24,13 @@ CCL_NAMESPACE_BEGIN
 
 class Patch {
  public:
-  virtual ~Patch()
-  {
-  }
+  virtual ~Patch() = default;
+
   virtual void eval(float3 *P, float3 *dPdu, float3 *dPdv, float3 *N, float u, float v) = 0;
-  virtual BoundBox bound() = 0;
-  virtual int ptex_face_id()
-  {
-    return -1;
-  }
 
   int patch_index;
   int shader;
+  bool from_ngon;
 };
 
 /* Linear Quad Patch */
diff --git a/intern/cycles/subd/subd_split.cpp b/intern/cycles/subd/subd_split.cpp
index e6603632ba7..1a8c182510c 100644
--- a/intern/cycles/subd/subd_split.cpp
+++ b/intern/cycles/subd/subd_split.cpp
@@ -21,6 +21,9 @@
 #include "subd/subd_patch.h"
 #include "subd/subd_split.h"
 
+#include "util/util_algorithm.h"
+#include "util/util_foreach.h"
+#include "util/util_hash.h"
 #include "util/util_math.h"
 #include "util/util_types.h"
 
@@ -28,14 +31,12 @@ CCL_NAMESPACE_BEGIN
 
 /* DiagSplit */
 
-DiagSplit::DiagSplit(const SubdParams &params_) : params(params_)
-{
-}
+#define DSPLIT_NON_UNIFORM -1
+#define STITCH_NGON_CENTER_VERT_INDEX_OFFSET 0x60000000
+#define STITCH_NGON_SPLIT_EDGE_CENTER_VERT_TAG (0x60000000 - 1)
 
-void DiagSplit::dispatch(QuadDice::SubPatch &sub, QuadDice::EdgeFactors &ef)
+DiagSplit::DiagSplit(const SubdParams &params_) : params(params_)
 {
-  subpatches_quad.push_back(sub);
-  edgefactors_quad.push_back(ef);
 }
 
 float3 DiagSplit::to_world(Patch *patch, float2 uv)
@@ -49,45 +50,62 @@ float3 DiagSplit::to_world(Patch *patch, float2 uv)
   return P;
 }
 
-int DiagSplit::T(Patch *patch, float2 Pstart, float2 Pend)
+static void order_float2(float2 &a, float2 &b)
 {
-  float3 Plast = make_float3(0.0f, 0.0f, 0.0f);
+  if (b.x < a.x || b.y < a.y) {
+    swap(a, b);
+  }
+}
+
+int DiagSplit::T(Patch *patch, float2 Pstart, float2 Pend, bool recursive_resolve)
+{
+  order_float2(Pstart, Pend); /* May not be necessary, but better to be safe. */
+
   float Lsum = 0.0f;
   float Lmax = 0.0f;
 
-  for (int i = 0; i < params.test_steps; i++) {
+  float3 Plast = to_world(patch, Pstart);
+
+  for (int i = 1; i < params.test_steps; i++) {
     float t = i / (float)(params.test_steps - 1);
 
     float3 P = to_world(patch, Pstart + t * (Pend - Pstart));
 
-    if (i > 0) {
-      float L;
+    float L;
 
-      if (!params.camera) {
-        L = len(P - Plast);
-      }
-      else {
-        Camera *cam = params.camera;
-
-        float pixel_width = cam->world_to_raster_size((P + Plast) * 0.5f);
-        L = len(P - Plast) / pixel_width;
-      }
+    if (!params.camera) {
+      L = len(P - Plast);
+    }
+    else {
+      Camera *cam = params.camera;
 
-      Lsum += L;
-      Lmax = max(L, Lmax);
+      float pixel_width = cam->world_to_raster_size((P + Plast) * 0.5f);
+      L = len(P - Plast) / pixel_width;
     }
 
+    Lsum += L;
+    Lmax = max(L, Lmax);
+
     Plast = P;
   }
 
-  int tmin = (int)ceil(Lsum / params.dicing_rate);
-  int tmax = (int)ceil((params.test_steps - 1) * Lmax /
-                       params.dicing_rate);  // XXX paper says N instead of N-1, seems wrong?
+  int tmin = (int)ceilf(Lsum / params.dicing_rate);
+  int tmax = (int)ceilf((params.test_steps - 1) * Lmax /
+                        params.dicing_rate);  // XXX paper says N instead of N-1, seems wrong?
+  int res = max(tmax, 1);
 
-  if (tmax - tmin > params.split_threshold)
-    return DSPLIT_NON_UNIFORM;
+  if (tmax - tmin > params.split_threshold) {
+    if (!recursive_resolve) {
+      res = DSPLIT_NON_UNIFORM;
+    }
+    else {
+      float2 P = (Pstart + Pend) * 0.5f;
+      res = T(patch, Pstart, P, true) + T(patch, P, Pend, true);
+    }
+  }
 
-  return tmax;
+  limit_edge_factor(res, patch, Pstart, Pend);
+  return res;
 }
 
 void DiagSplit::partition_edge(
@@ -99,159 +117,632 @@ void DiagSplit::partition_edge(
     *t1 = T(patch, *P, Pend);
   }
   else {
-    int I = (int)floor((float)t * 0.5f);
-    *P = interp(Pstart, Pend, (t == 0) ? 0 : I / (float)t); /* XXX is t faces or verts */
+    assert(t >= 2); /* Need at least two segments to partition into. */
+
+    int I = (int)floorf((float)t * 0.5f);
+    *P = interp(Pstart, Pend, I / (float)t);
     *t0 = I;
     *t1 = t - I;
   }
 }
 
-static void limit_edge_factors(const QuadDice::SubPatch &sub, QuadDice::EdgeFactors &ef, int max_t)
+void DiagSplit::limit_edge_factor(int &T, Patch *patch, float2 Pstart, float2 Pend)
+{
+  int max_t = 1 << params.max_level;
+  int max_t_for_edge = int(max_t * len(Pstart - Pend));
+
+  if (patch->from_ngon) {
+    max_t_for_edge >>= 1; /* Initial split of ngon causes edges to extend half the distance. */
+  }
+
+  T = (max_t_for_edge <= 1) ? 1 : min(T, max_t_for_edge);
+
+  assert(T >= 1 || T == DSPLIT_NON_UNIFORM);
+}
+
+void DiagSplit::resolve_edge_factors(Subpatch &sub)
 {
-  float2 P00 = sub.P00;
-  float2 P01 = sub.P01;
-  float2 P10 = sub.P10;
-  float2 P11 = sub.P11;
-
-  int tu0 = int(max_t * len(P10 - P00));
-  int tu1 = int(max_t * len(P11 - P01));
-  int tv0 = int(max_t * len(P01 - P00));
-  int tv1 = int(max_t * len(P11 - P10));
-
-  ef.tu0 = tu0 <= 1 ? 1 : min(ef.tu0, tu0);
-  ef.tu1 = tu1 <= 1 ? 1 : min(ef.tu1, tu1);
-  ef.tv0 = tv0 <= 1 ? 1 : min(ef.tv0, tv0);
-  ef.tv1 = tv1 <= 1 ? 1 : min(ef.tv1, tv1);
+  /* Resolve DSPLIT_NON_UNIFORM to actual T value if splitting is no longer possible. */
+  if (sub.edge_u0.T == 1 && sub.edge_u1.T == DSPLIT_NON_UNIFORM) {
+    sub.edge_u1.T = T(sub.patch, sub.c01, sub.c11, true);
+  }
+  if (sub.edge_u1.T == 1 && sub.edge_u0.T == DSPLIT_NON_UNIFORM) {
+    sub.edge_u0.T = T(sub.patch, sub.c00, sub.c10, true);
+  }
+  if (sub.edge_v0.T == 1 && sub.edge_v1.T == DSPLIT_NON_UNIFORM) {
+    sub.edge_v1.T = T(sub.patch, sub.c11, sub.c10, true);
+  }
+  if (sub.edge_v1.T == 1 && sub.edge_v0.T == DSPLIT_NON_UNIFORM) {
+    sub.edge_v0.T = T(sub.patch, sub.c01, sub.c00, true);
+  }
 }
 
-void DiagSplit::split(QuadDice::SubPatch &sub, QuadDice::EdgeFactors &ef, int depth)
+void DiagSplit::split(Subpatch &sub, int depth)
 {
   if (depth > 32) {
     /* We should never get here, but just in case end recursion safely. */
-    ef.tu0 = 1;
-    ef.tu1 = 1;
-    ef.tv0 = 1;
-    ef.tv1 = 1;
+    assert(!"diagsplit recursion limit reached");
+
+    sub.edge_u0.T = 1;
+    sub.edge_u1.T = 1;
+    sub.edge_v0.T = 1;
+    sub.edge_v1.T = 1;
 
-    dispatch(sub, ef);
+    subpatches.push_back(sub);
     return;
   }
 
-  bool split_u = (ef.tu0 == DSPLIT_NON_UNIFORM || ef.tu1 == DSPLIT_NON_UNIFORM);
-  bool split_v = (ef.tv0 == DSPLIT_NON_UNIFORM || ef.tv1 == DSPLIT_NON_UNIFORM);
+  bool split_u = (sub.edge_u0.T == DSPLIT_NON_UNIFORM || sub.edge_u1.T == DSPLIT_NON_UNIFORM);
+  bool split_v = (sub.edge_v0.T == DSPLIT_NON_UNIFORM || sub.edge_v1.T == DSPLIT_NON_UNIFORM);
 
   /* Split subpatches such that the ratio of T for opposite edges doesn't
    * exceed 1.5, this reduces over tessellation for some patches
    */
-  bool tmp_split_v = split_v;
-  if (!split_u && min(ef.tu0, ef.tu1) > 8 && min(ef.tu0, ef.tu1) * 1.5f < max(ef.tu0, ef.tu1))
+  /* clang-format off */
+  if (min(sub.edge_u0.T, sub.edge_u1.T) > 8 && /* Must be uniform and preferably greater than 8 to split. */
+      min(sub.edge_v0.T, sub.edge_v1.T) >= 2 && /* Must be uniform and at least 2 to split. */
+      max(sub.edge_u0.T, sub.edge_u1.T) / min(sub.edge_u0.T, sub.edge_u1.T) > 1.5f)
+  {
     split_v = true;
-  if (!tmp_split_v && min(ef.tu0, ef.tu1) > 8 && min(ef.tv0, ef.tv1) * 1.5f < max(ef.tv0, ef.tv1))
+  }
+  if (min(sub.edge_v0.T, sub.edge_v1.T) > 8 &&
+      min(sub.edge_u0.T, sub.edge_u1.T) >= 2 &&
+      max(sub.edge_v0.T, sub.edge_v1.T) / min(sub.edge_v0.T, sub.edge_v1.T) > 1.5f)
+  {
     split_u = true;
+  }
+  /* clang-format on */
 
-  /* alternate axis */
+  /* Alternate axis. */
   if (split_u && split_v) {
     split_u = depth % 2;
   }
 
-  if (split_u) {
-    /* partition edges */
-    QuadDice::EdgeFactors ef0, ef1;
-    float2 Pu0, Pu1;
+  if (!split_u && !split_v) {
+    /* Add the unsplit subpatch. */
+    subpatches.push_back(sub);
+    Subpatch &subpatch = subpatches[subpatches.size() - 1];
+
+    /* Update T values and offsets. */
+    for (int i = 0; i < 4; i++) {
+      Subpatch::edge_t &edge = subpatch.edges[i];
+
+      edge.offset = edge.edge->T;
+      edge.edge->T += edge.T;
+    }
+  }
+  else {
+    /* Copy into new subpatches. */
+    Subpatch sub_a = sub;
+    Subpatch sub_b = sub;
+
+    /* Pointers to various subpatch elements. */
+    Subpatch::edge_t *sub_across_0, *sub_across_1;
+    Subpatch::edge_t *sub_a_across_0, *sub_a_across_1;
+    Subpatch::edge_t *sub_b_across_0, *sub_b_across_1;
+
+    Subpatch::edge_t *sub_a_split, *sub_b_split;
+
+    float2 *Pa, *Pb, *Pc, *Pd;
+
+    /* Set pointers based on split axis. */
+    if (split_u) {
+      sub_across_0 = &sub.edge_u0;
+      sub_across_1 = &sub.edge_u1;
+      sub_a_across_0 = &sub_a.edge_u0;
+      sub_a_across_1 = &sub_a.edge_u1;
+      sub_b_across_0 = &sub_b.edge_u0;
+      sub_b_across_1 = &sub_b.edge_u1;
+
+      sub_a_split = &sub_a.edge_v1;
+      sub_b_split = &sub_b.edge_v0;
+
+      Pa = &sub_a.c11;
+      Pb = &sub_a.c10;
+      Pc = &sub_b.c01;
+      Pd = &sub_b.c00;
+    }
+    else {
+      sub_across_0 = &sub.edge_v0;
+      sub_across_1 = &sub.edge_v1;
+      sub_a_across_0 = &sub_a.edge_v0;
+      sub_a_across_1 = &sub_a.edge_v1;
+      sub_b_across_0 = &sub_b.edge_v0;
+      sub_b_across_1 = &sub_b.edge_v1;
+
+      sub_a_split = &sub_a.edge_u0;
+      sub_b_split = &sub_b.edge_u1;
+
+      Pa = &sub_a.c10;
+      Pb = &sub_a.c00;
+      Pc = &sub_b.c11;
+      Pd = &sub_b.c01;
+    }
 
-    partition_edge(sub.patch, &Pu0, &ef0.tu0, &ef1.tu0, sub.P00, sub.P10, ef.tu0);
-    partition_edge(sub.patch, &Pu1, &ef0.tu1, &ef1.tu1, sub.P01, sub.P11, ef.tu1);
+    /* Partition edges */
+    float2 P0, P1;
 
-    /* split */
-    int tsplit = T(sub.patch, Pu0, Pu1);
-    ef0.tv0 = ef.tv0;
-    ef0.tv1 = tsplit;
+    partition_edge(
+        sub.patch, &P0, &sub_a_across_0->T, &sub_b_across_0->T, *Pd, *Pb, sub_across_0->T);
+    partition_edge(
+        sub.patch, &P1, &sub_a_across_1->T, &sub_b_across_1->T, *Pc, *Pa, sub_across_1->T);
 
-    ef1.tv0 = tsplit;
-    ef1.tv1 = ef.tv1;
+    /* Split */
+    *Pa = P1;
+    *Pb = P0;
 
-    /* create subpatches */
-    QuadDice::SubPatch sub0 = {sub.patch, sub.P00, Pu0, sub.P01, Pu1};
-    QuadDice::SubPatch sub1 = {sub.patch, Pu0, sub.P10, Pu1, sub.P11};
+    *Pc = P1;
+    *Pd = P0;
 
-    limit_edge_factors(sub0, ef0, 1 << params.max_level);
-    limit_edge_factors(sub1, ef1, 1 << params.max_level);
+    int tsplit = T(sub.patch, P0, P1);
 
-    split(sub0, ef0, depth + 1);
-    split(sub1, ef1, depth + 1);
+    if (depth == -2 && tsplit == 1) {
+      tsplit = 2; /* Ensure we can always split at depth -1. */
+    }
+
+    sub_a_split->T = tsplit;
+    sub_b_split->T = tsplit;
+
+    resolve_edge_factors(sub_a);
+    resolve_edge_factors(sub_b);
+
+    /* Create new edge */
+    Edge &edge = *alloc_edge();
+
+    sub_a_split->edge = &edge;
+    sub_b_split->edge = &edge;
+
+    sub_a_split->offset = 0;
+    sub_b_split->offset = 0;
+
+    sub_a_split->indices_decrease_along_edge = false;
+    sub_b_split->indices_decrease_along_edge = true;
+
+    sub_a_split->sub_edges_created_in_reverse_order = !split_u;
+    sub_b_split->sub_edges_created_in_reverse_order = !split_u;
+
+    edge.top_indices_decrease = sub_across_1->sub_edges_created_in_reverse_order;
+    edge.bottom_indices_decrease = sub_across_0->sub_edges_created_in_reverse_order;
+
+    /* Recurse */
+    edge.T = 0;
+    split(sub_a, depth + 1);
+
+    int edge_t = edge.T;
+    (void)edge_t;
+
+    edge.top_offset = sub_across_1->edge->T;
+    edge.bottom_offset = sub_across_0->edge->T;
+
+    edge.T = 0; /* We calculate T twice along each edge. :/ */
+    split(sub_b, depth + 1);
+
+    assert(edge.T == edge_t); /* If this fails we will crash at some later point! */
+
+    edge.top = sub_across_1->edge;
+    edge.bottom = sub_across_0->edge;
   }
-  else if (split_v) {
-    /* partition edges */
-    QuadDice::EdgeFactors ef0, ef1;
-    float2 Pv0, Pv1;
+}
+
+int DiagSplit::alloc_verts(int n)
+{
+  int a = num_alloced_verts;
+  num_alloced_verts += n;
+  return a;
+}
 
-    partition_edge(sub.patch, &Pv0, &ef0.tv0, &ef1.tv0, sub.P00, sub.P01, ef.tv0);
-    partition_edge(sub.patch, &Pv1, &ef0.tv1, &ef1.tv1, sub.P10, sub.P11, ef.tv1);
+Edge *DiagSplit::alloc_edge()
+{
+  edges.emplace_back();
+  return &edges.back();
+}
 
-    /* split */
-    int tsplit = T(sub.patch, Pv0, Pv1);
-    ef0.tu0 = ef.tu0;
-    ef0.tu1 = tsplit;
+void DiagSplit::split_patches(Patch *patches, size_t patches_byte_stride)
+{
+  int patch_index = 0;
 
-    ef1.tu0 = tsplit;
-    ef1.tu1 = ef.tu1;
+  for (int f = 0; f < params.mesh->subd_faces.size(); f++) {
+    Mesh::SubdFace &face = params.mesh->subd_faces[f];
 
-    /* create subpatches */
-    QuadDice::SubPatch sub0 = {sub.patch, sub.P00, sub.P10, Pv0, Pv1};
-    QuadDice::SubPatch sub1 = {sub.patch, Pv0, Pv1, sub.P01, sub.P11};
+    Patch *patch = (Patch *)(((char *)patches) + patch_index * patches_byte_stride);
 
-    limit_edge_factors(sub0, ef0, 1 << params.max_level);
-    limit_edge_factors(sub1, ef1, 1 << params.max_level);
+    if (face.is_quad()) {
+      patch_index++;
 
-    split(sub0, ef0, depth + 1);
-    split(sub1, ef1, depth + 1);
+      split_quad(face, patch);
+    }
+    else {
+      patch_index += face.num_corners;
+
+      split_ngon(face, patch, patches_byte_stride);
+    }
   }
-  else {
-    dispatch(sub, ef);
+
+  params.mesh->vert_to_stitching_key_map.clear();
+  params.mesh->vert_stitching_map.clear();
+
+  post_split();
+}
+
+static Edge *create_edge_from_corner(DiagSplit *split,
+                                     const Mesh *mesh,
+                                     const Mesh::SubdFace &face,
+                                     int corner,
+                                     bool &reversed,
+                                     int v0,
+                                     int v1)
+{
+  int a = mesh->subd_face_corners[face.start_corner + mod(corner + 0, face.num_corners)];
+  int b = mesh->subd_face_corners[face.start_corner + mod(corner + 1, face.num_corners)];
+
+  reversed = !(b < a);
+
+  if (b < a) {
+    swap(a, b);
+    swap(v0, v1);
   }
+
+  Edge *edge = split->alloc_edge();
+
+  edge->is_stitch_edge = true;
+  edge->stitch_start_vert_index = a;
+  edge->stitch_end_vert_index = b;
+
+  edge->start_vert_index = v0;
+  edge->end_vert_index = v1;
+
+  edge->stitch_edge_key = {a, b};
+
+  return edge;
 }
 
-void DiagSplit::split_quad(Patch *patch, QuadDice::SubPatch *subpatch)
+void DiagSplit::split_quad(const Mesh::SubdFace &face, Patch *patch)
 {
-  QuadDice::SubPatch sub_split;
-  QuadDice::EdgeFactors ef_split;
+  Subpatch subpatch(patch);
+
+  int v = alloc_verts(4);
+
+  bool v0_reversed, u1_reversed, v1_reversed, u0_reversed;
+  subpatch.edge_v0.edge = create_edge_from_corner(
+      this, params.mesh, face, 3, v0_reversed, v + 3, v + 0);
+  subpatch.edge_u1.edge = create_edge_from_corner(
+      this, params.mesh, face, 2, u1_reversed, v + 2, v + 3);
+  subpatch.edge_v1.edge = create_edge_from_corner(
+      this, params.mesh, face, 1, v1_reversed, v + 1, v + 2);
+  subpatch.edge_u0.edge = create_edge_from_corner(
+      this, params.mesh, face, 0, u0_reversed, v + 0, v + 1);
+
+  subpatch.edge_v0.sub_edges_created_in_reverse_order = !v0_reversed;
+  subpatch.edge_u1.sub_edges_created_in_reverse_order = u1_reversed;
+  subpatch.edge_v1.sub_edges_created_in_reverse_order = v1_reversed;
+  subpatch.edge_u0.sub_edges_created_in_reverse_order = !u0_reversed;
+
+  subpatch.edge_v0.indices_decrease_along_edge = v0_reversed;
+  subpatch.edge_u1.indices_decrease_along_edge = u1_reversed;
+  subpatch.edge_v1.indices_decrease_along_edge = v1_reversed;
+  subpatch.edge_u0.indices_decrease_along_edge = u0_reversed;
+
+  /* Forces a split in both axis for quads, needed to match split of ngons into quads. */
+  subpatch.edge_u0.T = DSPLIT_NON_UNIFORM;
+  subpatch.edge_u1.T = DSPLIT_NON_UNIFORM;
+  subpatch.edge_v0.T = DSPLIT_NON_UNIFORM;
+  subpatch.edge_v1.T = DSPLIT_NON_UNIFORM;
+
+  split(subpatch, -2);
+}
+
+static Edge *create_split_edge_from_corner(DiagSplit *split,
+                                           const Mesh *mesh,
+                                           const Mesh::SubdFace &face,
+                                           int corner,
+                                           int side,
+                                           bool &reversed,
+                                           int v0,
+                                           int v1,
+                                           int vc)
+{
+  Edge *edge = split->alloc_edge();
+
+  int a = mesh->subd_face_corners[face.start_corner + mod(corner + 0, face.num_corners)];
+  int b = mesh->subd_face_corners[face.start_corner + mod(corner + 1, face.num_corners)];
 
-  if (subpatch) {
-    sub_split = *subpatch;
+  if (b < a) {
+    edge->stitch_edge_key = {b, a};
   }
   else {
-    sub_split.patch = patch;
-    sub_split.P00 = make_float2(0.0f, 0.0f);
-    sub_split.P10 = make_float2(1.0f, 0.0f);
-    sub_split.P01 = make_float2(0.0f, 1.0f);
-    sub_split.P11 = make_float2(1.0f, 1.0f);
+    edge->stitch_edge_key = {a, b};
   }
 
-  ef_split.tu0 = T(patch, sub_split.P00, sub_split.P10);
-  ef_split.tu1 = T(patch, sub_split.P01, sub_split.P11);
-  ef_split.tv0 = T(patch, sub_split.P00, sub_split.P01);
-  ef_split.tv1 = T(patch, sub_split.P10, sub_split.P11);
+  reversed = !(b < a);
 
-  limit_edge_factors(sub_split, ef_split, 1 << params.max_level);
+  if (side == 0) {
+    a = vc;
+  }
+  else {
+    b = vc;
+  }
+
+  if (!reversed) {
+    swap(a, b);
+    swap(v0, v1);
+  }
+
+  edge->is_stitch_edge = true;
+  edge->stitch_start_vert_index = a;
+  edge->stitch_end_vert_index = b;
+
+  edge->start_vert_index = v0;
+  edge->end_vert_index = v1;
+
+  return edge;
+}
+
+void DiagSplit::split_ngon(const Mesh::SubdFace &face, Patch *patches, size_t patches_byte_stride)
+{
+  Edge *prev_edge_u0 = nullptr;
+  Edge *first_edge_v0 = nullptr;
+
+  for (int corner = 0; corner < face.num_corners; corner++) {
+    Patch *patch = (Patch *)(((char *)patches) + corner * patches_byte_stride);
+
+    Subpatch subpatch(patch);
+
+    int v = alloc_verts(4);
+
+    /* Setup edges. */
+    Edge *edge_u1 = alloc_edge();
+    Edge *edge_v1 = alloc_edge();
+
+    edge_v1->is_stitch_edge = true;
+    edge_u1->is_stitch_edge = true;
+
+    edge_u1->stitch_start_vert_index = -(face.start_corner + mod(corner + 0, face.num_corners)) -
+                                       1;
+    edge_u1->stitch_end_vert_index = STITCH_NGON_CENTER_VERT_INDEX_OFFSET + face.ptex_offset;
+
+    edge_u1->start_vert_index = v + 3;
+    edge_u1->end_vert_index = v + 2;
+
+    edge_u1->stitch_edge_key = {edge_u1->stitch_start_vert_index, edge_u1->stitch_end_vert_index};
+
+    edge_v1->stitch_start_vert_index = -(face.start_corner + mod(corner + 1, face.num_corners)) -
+                                       1;
+    edge_v1->stitch_end_vert_index = STITCH_NGON_CENTER_VERT_INDEX_OFFSET + face.ptex_offset;
+
+    edge_v1->start_vert_index = v + 1;
+    edge_v1->end_vert_index = v + 2;
+
+    edge_v1->stitch_edge_key = {edge_v1->stitch_start_vert_index, edge_v1->stitch_end_vert_index};
+
+    bool v0_reversed, u0_reversed;
+
+    subpatch.edge_v0.edge = create_split_edge_from_corner(this,
+                                                          params.mesh,
+                                                          face,
+                                                          corner - 1,
+                                                          0,
+                                                          v0_reversed,
+                                                          v + 3,
+                                                          v + 0,
+                                                          STITCH_NGON_SPLIT_EDGE_CENTER_VERT_TAG);
+
+    subpatch.edge_u1.edge = edge_u1;
+    subpatch.edge_v1.edge = edge_v1;
+
+    subpatch.edge_u0.edge = create_split_edge_from_corner(this,
+                                                          params.mesh,
+                                                          face,
+                                                          corner + 0,
+                                                          1,
+                                                          u0_reversed,
+                                                          v + 0,
+                                                          v + 1,
+                                                          STITCH_NGON_SPLIT_EDGE_CENTER_VERT_TAG);
+
+    subpatch.edge_v0.sub_edges_created_in_reverse_order = !v0_reversed;
+    subpatch.edge_u1.sub_edges_created_in_reverse_order = false;
+    subpatch.edge_v1.sub_edges_created_in_reverse_order = true;
+    subpatch.edge_u0.sub_edges_created_in_reverse_order = !u0_reversed;
+
+    subpatch.edge_v0.indices_decrease_along_edge = v0_reversed;
+    subpatch.edge_u1.indices_decrease_along_edge = false;
+    subpatch.edge_v1.indices_decrease_along_edge = true;
+    subpatch.edge_u0.indices_decrease_along_edge = u0_reversed;
+
+    /* Perfrom split. */
+    {
+      subpatch.edge_u0.T = T(subpatch.patch, subpatch.c00, subpatch.c10);
+      subpatch.edge_u1.T = T(subpatch.patch, subpatch.c01, subpatch.c11);
+      subpatch.edge_v0.T = T(subpatch.patch, subpatch.c00, subpatch.c01);
+      subpatch.edge_v1.T = T(subpatch.patch, subpatch.c10, subpatch.c11);
+
+      resolve_edge_factors(subpatch);
+
+      split(subpatch, 0);
+    }
+
+    /* Update offsets after T is known from split. */
+    edge_u1->top = subpatch.edge_v0.edge;
+    edge_u1->stitch_top_offset = edge_u1->top->T * (v0_reversed ? -1 : 1);
+    edge_v1->top = subpatch.edge_u0.edge;
+    edge_v1->stitch_top_offset = edge_v1->top->T * (!u0_reversed ? -1 : 1);
+
+    if (corner == 0) {
+      first_edge_v0 = subpatch.edge_v0.edge;
+    }
+
+    if (prev_edge_u0) {
+      if (v0_reversed) {
+        subpatch.edge_v0.edge->stitch_offset = prev_edge_u0->T;
+      }
+      else {
+        prev_edge_u0->stitch_offset = subpatch.edge_v0.edge->T;
+      }
+
+      int T = subpatch.edge_v0.edge->T + prev_edge_u0->T;
+      subpatch.edge_v0.edge->stitch_edge_T = T;
+      prev_edge_u0->stitch_edge_T = T;
+    }
 
-  split(sub_split, ef_split);
+    if (corner == face.num_corners - 1) {
+      if (v0_reversed) {
+        subpatch.edge_u0.edge->stitch_offset = first_edge_v0->T;
+      }
+      else {
+        first_edge_v0->stitch_offset = subpatch.edge_u0.edge->T;
+      }
+
+      int T = first_edge_v0->T + subpatch.edge_u0.edge->T;
+      first_edge_v0->stitch_edge_T = T;
+      subpatch.edge_u0.edge->stitch_edge_T = T;
+    }
+
+    prev_edge_u0 = subpatch.edge_u0.edge;
+  }
+}
+
+void DiagSplit::post_split()
+{
+  int num_stitch_verts = 0;
+
+  /* All patches are now split, and all T values known. */
 
+  foreach (Edge &edge, edges) {
+    if (edge.second_vert_index < 0) {
+      edge.second_vert_index = alloc_verts(edge.T - 1);
+    }
+
+    if (edge.is_stitch_edge) {
+      num_stitch_verts = max(num_stitch_verts,
+                             max(edge.stitch_start_vert_index, edge.stitch_end_vert_index));
+    }
+  }
+
+  num_stitch_verts += 1;
+
+  /* Map of edge key to edge stitching vert offset. */
+  struct pair_hasher {
+    size_t operator()(const pair<int, int> &k) const
+    {
+      return hash_uint2(k.first, k.second);
+    }
+  };
+  typedef unordered_map<pair<int, int>, int, pair_hasher> edge_stitch_verts_map_t;
+  edge_stitch_verts_map_t edge_stitch_verts_map;
+
+  foreach (Edge &edge, edges) {
+    if (edge.is_stitch_edge) {
+      if (edge.stitch_edge_T == 0) {
+        edge.stitch_edge_T = edge.T;
+      }
+
+      if (edge_stitch_verts_map.find(edge.stitch_edge_key) == edge_stitch_verts_map.end()) {
+        edge_stitch_verts_map[edge.stitch_edge_key] = num_stitch_verts;
+        num_stitch_verts += edge.stitch_edge_T - 1;
+      }
+    }
+  }
+
+  /* Set start and end indices for edges generated from a split. */
+  foreach (Edge &edge, edges) {
+    if (edge.start_vert_index < 0) {
+      /* Fixup offsets. */
+      if (edge.top_indices_decrease) {
+        edge.top_offset = edge.top->T - edge.top_offset;
+      }
+
+      edge.start_vert_index = edge.top->get_vert_along_edge(edge.top_offset);
+    }
+
+    if (edge.end_vert_index < 0) {
+      if (edge.bottom_indices_decrease) {
+        edge.bottom_offset = edge.bottom->T - edge.bottom_offset;
+      }
+
+      edge.end_vert_index = edge.bottom->get_vert_along_edge(edge.bottom_offset);
+    }
+  }
+
+  int vert_offset = params.mesh->verts.size();
+
+  /* Add verts to stitching map. */
+  foreach (const Edge &edge, edges) {
+    if (edge.is_stitch_edge) {
+      int second_stitch_vert_index = edge_stitch_verts_map[edge.stitch_edge_key];
+
+      for (int i = 0; i <= edge.T; i++) {
+        /* Get proper stitching key. */
+        int key;
+
+        if (i == 0) {
+          key = edge.stitch_start_vert_index;
+        }
+        else if (i == edge.T) {
+          key = edge.stitch_end_vert_index;
+        }
+        else {
+          key = second_stitch_vert_index + i - 1 + edge.stitch_offset;
+        }
+
+        if (key == STITCH_NGON_SPLIT_EDGE_CENTER_VERT_TAG) {
+          if (i == 0) {
+            key = second_stitch_vert_index - 1 + edge.stitch_offset;
+          }
+          else if (i == edge.T) {
+            key = second_stitch_vert_index - 1 + edge.T;
+          }
+        }
+        else if (key < 0 && edge.top) { /* ngon spoke edge */
+          int s = edge_stitch_verts_map[edge.top->stitch_edge_key];
+          if (edge.stitch_top_offset >= 0) {
+            key = s - 1 + edge.stitch_top_offset;
+          }
+          else {
+            key = s - 1 + edge.top->stitch_edge_T + edge.stitch_top_offset;
+          }
+        }
+
+        /* Get real vert index. */
+        int vert = edge.get_vert_along_edge(i) + vert_offset;
+
+        /* Add to map */
+        if (params.mesh->vert_to_stitching_key_map.find(vert) ==
+            params.mesh->vert_to_stitching_key_map.end()) {
+          params.mesh->vert_to_stitching_key_map[vert] = key;
+          params.mesh->vert_stitching_map.insert({key, vert});
+        }
+      }
+    }
+  }
+
+  /* Dice; TODO(mai): Move this out of split. */
   QuadDice dice(params);
 
-  for (size_t i = 0; i < subpatches_quad.size(); i++) {
-    QuadDice::SubPatch &sub = subpatches_quad[i];
-    QuadDice::EdgeFactors &ef = edgefactors_quad[i];
+  int num_verts = num_alloced_verts;
+  int num_triangles = 0;
+
+  for (size_t i = 0; i < subpatches.size(); i++) {
+    subpatches[i].inner_grid_vert_offset = num_verts;
+    num_verts += subpatches[i].calc_num_inner_verts();
+    num_triangles += subpatches[i].calc_num_triangles();
+  }
+
+  dice.reserve(num_verts, num_triangles);
+
+  for (size_t i = 0; i < subpatches.size(); i++) {
+    Subpatch &sub = subpatches[i];
 
-    ef.tu0 = max(ef.tu0, 1);
-    ef.tu1 = max(ef.tu1, 1);
-    ef.tv0 = max(ef.tv0, 1);
-    ef.tv1 = max(ef.tv1, 1);
+    sub.edge_u0.T = max(sub.edge_u0.T, 1);
+    sub.edge_u1.T = max(sub.edge_u1.T, 1);
+    sub.edge_v0.T = max(sub.edge_v0.T, 1);
+    sub.edge_v1.T = max(sub.edge_v1.T, 1);
 
-    dice.dice(sub, ef);
+    dice.dice(sub);
   }
 
-  subpatches_quad.clear();
-  edgefactors_quad.clear();
+  /* Cleanup */
+  subpatches.clear();
+  edges.clear();
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/subd/subd_split.h b/intern/cycles/subd/subd_split.h
index 6e68d8ee598..773f4ddf120 100644
--- a/intern/cycles/subd/subd_split.h
+++ b/intern/cycles/subd/subd_split.h
@@ -23,35 +23,51 @@
  * for more details. */
 
 #include "subd/subd_dice.h"
+#include "subd/subd_subpatch.h"
 
+#include "util/util_deque.h"
 #include "util/util_types.h"
 #include "util/util_vector.h"
 
+#include <deque>
+
 CCL_NAMESPACE_BEGIN
 
 class Mesh;
 class Patch;
 
-#define DSPLIT_NON_UNIFORM -1
-
 class DiagSplit {
- public:
-  vector<QuadDice::SubPatch> subpatches_quad;
-  vector<QuadDice::EdgeFactors> edgefactors_quad;
-
   SubdParams params;
 
-  explicit DiagSplit(const SubdParams &params);
+  vector<Subpatch> subpatches;
+  /* deque is used so that element pointers remain vaild when size is changed. */
+  deque<Edge> edges;
 
   float3 to_world(Patch *patch, float2 uv);
-  int T(Patch *patch, float2 Pstart, float2 Pend);
+  int T(Patch *patch, float2 Pstart, float2 Pend, bool recursive_resolve = false);
+
+  void limit_edge_factor(int &T, Patch *patch, float2 Pstart, float2 Pend);
+  void resolve_edge_factors(Subpatch &sub);
+
   void partition_edge(
       Patch *patch, float2 *P, int *t0, int *t1, float2 Pstart, float2 Pend, int t);
 
-  void dispatch(QuadDice::SubPatch &sub, QuadDice::EdgeFactors &ef);
-  void split(QuadDice::SubPatch &sub, QuadDice::EdgeFactors &ef, int depth = 0);
+  void split(Subpatch &sub, int depth = 0);
+
+  int num_alloced_verts = 0;
+  int alloc_verts(int n); /* Returns start index of new verts. */
+
+ public:
+  Edge *alloc_edge();
+
+  explicit DiagSplit(const SubdParams &params);
+
+  void split_patches(Patch *patches, size_t patches_byte_stride);
+
+  void split_quad(const Mesh::SubdFace &face, Patch *patch);
+  void split_ngon(const Mesh::SubdFace &face, Patch *patches, size_t patches_byte_stride);
 
-  void split_quad(Patch *patch, QuadDice::SubPatch *subpatch = NULL);
+  void post_split();
 };
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/subd/subd_subpatch.h b/intern/cycles/subd/subd_subpatch.h
new file mode 100644
index 00000000000..1a32b763cb8
--- /dev/null
+++ b/intern/cycles/subd/subd_subpatch.h
@@ -0,0 +1,219 @@
+/*
+ * Copyright 2011-2018 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef __SUBD_SUBPATCH_H__
+#define __SUBD_SUBPATCH_H__
+
+#include "util/util_map.h"
+#include "util/util_types.h"
+
+CCL_NAMESPACE_BEGIN
+
+/* Subpatch */
+
+class Subpatch {
+ public:
+  class Patch *patch; /* Patch this is a subpatch of. */
+  int inner_grid_vert_offset;
+
+  struct edge_t {
+    int T;
+    int offset; /* Offset along main edge, interpretation depends on the two flags below. */
+
+    bool indices_decrease_along_edge;
+    bool sub_edges_created_in_reverse_order;
+
+    struct Edge *edge;
+
+    int get_vert_along_edge(int n) const;
+  };
+
+  /*
+   *            eu1
+   *     c01 --------- c11
+   *     |               |
+   * ev0 |               | ev1
+   *     |               |
+   *     c00 --------- c10
+   *            eu0
+   */
+
+  union {
+    float2 corners[4]; /* UV within patch, clockwise starting from uv (0, 0) towards (0, 1) etc. */
+    struct {
+      float2 c00, c01, c11, c10;
+    };
+  };
+
+  union {
+    edge_t
+        edges[4]; /* Edges of this subpatch, each edge starts at the corner of the same index. */
+    struct {
+      edge_t edge_v0, edge_u1, edge_v1, edge_u0;
+    };
+  };
+
+  explicit Subpatch(Patch *patch = nullptr)
+      : patch(patch),
+        c00(make_float2(0.0f, 0.0f)),
+        c01(make_float2(0.0f, 1.0f)),
+        c11(make_float2(1.0f, 1.0f)),
+        c10(make_float2(1.0f, 0.0f))
+  {
+  }
+
+  Subpatch(Patch *patch, float2 c00, float2 c01, float2 c11, float2 c10)
+      : patch(patch), c00(c00), c01(c01), c11(c11), c10(c10)
+  {
+  }
+
+  int calc_num_inner_verts() const
+  {
+    int Mu = max(edge_u0.T, edge_u1.T);
+    int Mv = max(edge_v0.T, edge_v1.T);
+    Mu = max(Mu, 2);
+    Mv = max(Mv, 2);
+    return (Mu - 1) * (Mv - 1);
+  }
+
+  int calc_num_triangles() const
+  {
+    int Mu = max(edge_u0.T, edge_u1.T);
+    int Mv = max(edge_v0.T, edge_v1.T);
+    Mu = max(Mu, 2);
+    Mv = max(Mv, 2);
+
+    int inner_triangles = (Mu - 2) * (Mv - 2) * 2;
+    int edge_triangles = edge_u0.T + edge_u1.T + edge_v0.T + edge_v1.T + (Mu - 2) * 2 +
+                         (Mv - 2) * 2;
+
+    return inner_triangles + edge_triangles;
+  }
+
+  int get_vert_along_edge(int e, int n) const;
+
+  int get_vert_along_grid_edge(int edge, int n) const
+  {
+    int Mu = max(edge_u0.T, edge_u1.T);
+    int Mv = max(edge_v0.T, edge_v1.T);
+    Mu = max(Mu, 2);
+    Mv = max(Mv, 2);
+
+    switch (edge) {
+      case 0:
+        return inner_grid_vert_offset + n * (Mu - 1);
+      case 1:
+        return inner_grid_vert_offset + (Mu - 1) * (Mv - 2) + n;
+      case 2:
+        return inner_grid_vert_offset + ((Mu - 1) * (Mv - 1) - 1) - n * (Mu - 1);
+      case 3:
+        return inner_grid_vert_offset + (Mu - 2) - n;
+    }
+
+    return -1;
+  }
+};
+
+struct Edge {
+  /* Number of segments the edge will be diced into, see DiagSplit paper. */
+  int T;
+
+  /* top is edge adjacent to start, bottom is adjacent to end. */
+  Edge *top, *bottom;
+
+  int top_offset, bottom_offset;
+  bool top_indices_decrease, bottom_indices_decrease;
+
+  int start_vert_index;
+  int end_vert_index;
+
+  /* Index of the second vert from this edges corner along the edge towards the next corner. */
+  int second_vert_index;
+
+  /* Vertices on edge are to be stitched. */
+  bool is_stitch_edge;
+
+  /* Key to match this edge with others to be stitched with.
+   * The ints in the pair are ordered stitching indices */
+  pair<int, int> stitch_edge_key;
+
+  /* Full T along edge (may be larger than T for edges split from ngon edges) */
+  int stitch_edge_T;
+  int stitch_offset;
+  int stitch_top_offset;
+  int stitch_start_vert_index;
+  int stitch_end_vert_index;
+
+  Edge()
+      : T(0),
+        top(nullptr),
+        bottom(nullptr),
+        top_offset(-1),
+        bottom_offset(-1),
+        top_indices_decrease(false),
+        bottom_indices_decrease(false),
+        start_vert_index(-1),
+        end_vert_index(-1),
+        second_vert_index(-1),
+        is_stitch_edge(false),
+        stitch_edge_T(0),
+        stitch_offset(0)
+  {
+  }
+
+  int get_vert_along_edge(int n) const
+  {
+    assert(n >= 0 && n <= T);
+
+    if (n == 0) {
+      return start_vert_index;
+    }
+    else if (n == T) {
+      return end_vert_index;
+    }
+
+    return second_vert_index + n - 1;
+  }
+};
+
+inline int Subpatch::edge_t::get_vert_along_edge(int n) const
+{
+  assert(n >= 0 && n <= T);
+
+  if (!indices_decrease_along_edge && !sub_edges_created_in_reverse_order) {
+    n = offset + n;
+  }
+  else if (!indices_decrease_along_edge && sub_edges_created_in_reverse_order) {
+    n = edge->T - offset - T + n;
+  }
+  else if (indices_decrease_along_edge && !sub_edges_created_in_reverse_order) {
+    n = offset + T - n;
+  }
+  else if (indices_decrease_along_edge && sub_edges_created_in_reverse_order) {
+    n = edge->T - offset - n;
+  }
+
+  return edge->get_vert_along_edge(n);
+}
+
+inline int Subpatch::get_vert_along_edge(int edge, int n) const
+{
+  return edges[edge].get_vert_along_edge(n);
+}
+
+CCL_NAMESPACE_END
+
+#endif /* __SUBD_SUBPATCH_H__ */
diff --git a/intern/cycles/test/render_graph_finalize_test.cpp b/intern/cycles/test/render_graph_finalize_test.cpp
index 7fb92bfb862..ca93f8b02d0 100644
--- a/intern/cycles/test/render_graph_finalize_test.cpp
+++ b/intern/cycles/test/render_graph_finalize_test.cpp
@@ -960,6 +960,13 @@ TEST_F(RenderGraph, constant_fold_blackbody)
   graph.finalize(scene);
 }
 
+/* A Note About The Math Node
+ *
+ * The clamp option is implemented using graph expansion, where a
+ * Clamp node named "clamp" is added and connected to the output.
+ * So the final result is actually from the node "clamp".
+ */
+
 /*
  * Tests: Math with all constant inputs (clamp false).
  */
@@ -985,7 +992,7 @@ TEST_F(RenderGraph, constant_fold_math)
 TEST_F(RenderGraph, constant_fold_math_clamp)
 {
   EXPECT_ANY_MESSAGE(log);
-  CORRECT_INFO_MESSAGE(log, "Folding Math::Value to constant (1).");
+  CORRECT_INFO_MESSAGE(log, "Folding clamp::Result to constant (1).");
 
   builder
       .add_node(ShaderNodeBuilder<MathNode>("Math")
@@ -1003,7 +1010,7 @@ TEST_F(RenderGraph, constant_fold_math_clamp)
  * Includes 2 tests: constant on each side.
  */
 static void build_math_partial_test_graph(ShaderGraphBuilder &builder,
-                                          NodeMath type,
+                                          NodeMathType type,
                                           float constval)
 {
   builder
@@ -1038,7 +1045,7 @@ TEST_F(RenderGraph, constant_fold_part_math_add_0)
   /* X + 0 == 0 + X == X */
   CORRECT_INFO_MESSAGE(log, "Folding Math_Cx::Value to socket Attribute::Fac.");
   CORRECT_INFO_MESSAGE(log, "Folding Math_xC::Value to socket Attribute::Fac.");
-  INVALID_INFO_MESSAGE(log, "Folding Out::");
+  INVALID_INFO_MESSAGE(log, "Folding clamp::");
 
   build_math_partial_test_graph(builder, NODE_MATH_ADD, 0.0f);
   graph.finalize(scene);
@@ -1053,7 +1060,7 @@ TEST_F(RenderGraph, constant_fold_part_math_sub_0)
   /* X - 0 == X */
   INVALID_INFO_MESSAGE(log, "Folding Math_Cx::");
   CORRECT_INFO_MESSAGE(log, "Folding Math_xC::Value to socket Attribute::Fac.");
-  INVALID_INFO_MESSAGE(log, "Folding Out::");
+  INVALID_INFO_MESSAGE(log, "Folding clamp::");
 
   build_math_partial_test_graph(builder, NODE_MATH_SUBTRACT, 0.0f);
   graph.finalize(scene);
@@ -1068,7 +1075,7 @@ TEST_F(RenderGraph, constant_fold_part_math_mul_1)
   /* X * 1 == 1 * X == X */
   CORRECT_INFO_MESSAGE(log, "Folding Math_Cx::Value to socket Attribute::Fac.");
   CORRECT_INFO_MESSAGE(log, "Folding Math_xC::Value to socket Attribute::Fac.");
-  INVALID_INFO_MESSAGE(log, "Folding Out::");
+  INVALID_INFO_MESSAGE(log, "Folding clamp::");
 
   build_math_partial_test_graph(builder, NODE_MATH_MULTIPLY, 1.0f);
   graph.finalize(scene);
@@ -1083,7 +1090,7 @@ TEST_F(RenderGraph, constant_fold_part_math_div_1)
   /* X / 1 == X */
   INVALID_INFO_MESSAGE(log, "Folding Math_Cx::");
   CORRECT_INFO_MESSAGE(log, "Folding Math_xC::Value to socket Attribute::Fac.");
-  INVALID_INFO_MESSAGE(log, "Folding Out::");
+  INVALID_INFO_MESSAGE(log, "Folding clamp::");
 
   build_math_partial_test_graph(builder, NODE_MATH_DIVIDE, 1.0f);
   graph.finalize(scene);
@@ -1098,7 +1105,7 @@ TEST_F(RenderGraph, constant_fold_part_math_mul_0)
   /* X * 0 == 0 * X == 0 */
   CORRECT_INFO_MESSAGE(log, "Folding Math_Cx::Value to constant (0).");
   CORRECT_INFO_MESSAGE(log, "Folding Math_xC::Value to constant (0).");
-  CORRECT_INFO_MESSAGE(log, "Folding Out::Value to constant (0)");
+  CORRECT_INFO_MESSAGE(log, "Folding clamp::Result to constant (0)");
   CORRECT_INFO_MESSAGE(log, "Discarding closure EmissionNode.");
 
   build_math_partial_test_graph(builder, NODE_MATH_MULTIPLY, 0.0f);
@@ -1114,7 +1121,7 @@ TEST_F(RenderGraph, constant_fold_part_math_div_0)
   /* 0 / X == 0 */
   CORRECT_INFO_MESSAGE(log, "Folding Math_Cx::Value to constant (0).");
   INVALID_INFO_MESSAGE(log, "Folding Math_xC::");
-  INVALID_INFO_MESSAGE(log, "Folding Out::");
+  INVALID_INFO_MESSAGE(log, "Folding clamp::");
 
   build_math_partial_test_graph(builder, NODE_MATH_DIVIDE, 0.0f);
   graph.finalize(scene);
@@ -1129,7 +1136,7 @@ TEST_F(RenderGraph, constant_fold_part_math_pow_0)
   /* X ^ 0 == 1 */
   INVALID_INFO_MESSAGE(log, "Folding Math_Cx::");
   CORRECT_INFO_MESSAGE(log, "Folding Math_xC::Value to constant (1).");
-  INVALID_INFO_MESSAGE(log, "Folding Out::");
+  INVALID_INFO_MESSAGE(log, "Folding clamp::");
 
   build_math_partial_test_graph(builder, NODE_MATH_POWER, 0.0f);
   graph.finalize(scene);
@@ -1144,7 +1151,7 @@ TEST_F(RenderGraph, constant_fold_part_math_pow_1)
   /* 1 ^ X == 1; X ^ 1 == X */
   CORRECT_INFO_MESSAGE(log, "Folding Math_Cx::Value to constant (1)");
   CORRECT_INFO_MESSAGE(log, "Folding Math_xC::Value to socket Attribute::Fac.");
-  INVALID_INFO_MESSAGE(log, "Folding Out::");
+  INVALID_INFO_MESSAGE(log, "Folding clamp::");
 
   build_math_partial_test_graph(builder, NODE_MATH_POWER, 1.0f);
   graph.finalize(scene);
@@ -1156,21 +1163,14 @@ TEST_F(RenderGraph, constant_fold_part_math_pow_1)
 TEST_F(RenderGraph, constant_fold_vector_math)
 {
   EXPECT_ANY_MESSAGE(log);
-  CORRECT_INFO_MESSAGE(log, "Folding VectorMath::Value to constant (1).");
   CORRECT_INFO_MESSAGE(log, "Folding VectorMath::Vector to constant (3, 0, 0).");
-  CORRECT_INFO_MESSAGE(log, "Folding convert_vector_to_float::value_float to constant (1).");
-  CORRECT_INFO_MESSAGE(log, "Folding Math::Value to constant (2).");
-  CORRECT_INFO_MESSAGE(log, "Folding convert_float_to_color::value_color to constant (2, 2, 2).");
 
   builder
       .add_node(ShaderNodeBuilder<VectorMathNode>("VectorMath")
                     .set(&VectorMathNode::type, NODE_VECTOR_MATH_SUBTRACT)
                     .set("Vector1", make_float3(1.3f, 0.5f, 0.7f))
                     .set("Vector2", make_float3(-1.7f, 0.5f, 0.7f)))
-      .add_node(ShaderNodeBuilder<MathNode>("Math").set(&MathNode::type, NODE_MATH_ADD))
-      .add_connection("VectorMath::Vector", "Math::Value1")
-      .add_connection("VectorMath::Value", "Math::Value2")
-      .output_color("Math::Value");
+      .output_color("VectorMath::Vector");
 
   graph.finalize(scene);
 }
@@ -1180,7 +1180,7 @@ TEST_F(RenderGraph, constant_fold_vector_math)
  * Includes 2 tests: constant on each side.
  */
 static void build_vecmath_partial_test_graph(ShaderGraphBuilder &builder,
-                                             NodeVectorMath type,
+                                             NodeVectorMathType type,
                                              float3 constval)
 {
   builder
@@ -1234,22 +1234,6 @@ TEST_F(RenderGraph, constant_fold_part_vecmath_sub_0)
 }
 
 /*
- * Tests: partial folding for Vector Math Dot Product with known 0.
- */
-TEST_F(RenderGraph, constant_fold_part_vecmath_dot_0)
-{
-  EXPECT_ANY_MESSAGE(log);
-  /* X * 0 == 0 * X == X */
-  CORRECT_INFO_MESSAGE(log, "Folding Math_Cx::Vector to constant (0, 0, 0).");
-  CORRECT_INFO_MESSAGE(log, "Folding Math_xC::Vector to constant (0, 0, 0).");
-  CORRECT_INFO_MESSAGE(log, "Folding Out::Vector to constant (0, 0, 0).");
-  CORRECT_INFO_MESSAGE(log, "Discarding closure EmissionNode.");
-
-  build_vecmath_partial_test_graph(builder, NODE_VECTOR_MATH_DOT_PRODUCT, make_float3(0, 0, 0));
-  graph.finalize(scene);
-}
-
-/*
  * Tests: partial folding for Vector Math Cross Product with known 0.
  */
 TEST_F(RenderGraph, constant_fold_part_vecmath_cross_0)
diff --git a/intern/cycles/util/CMakeLists.txt b/intern/cycles/util/CMakeLists.txt
index 4f1c82897cc..a6a6b6704f0 100644
--- a/intern/cycles/util/CMakeLists.txt
+++ b/intern/cycles/util/CMakeLists.txt
@@ -55,6 +55,7 @@ set(SRC_HEADERS
   util_boundbox.h
   util_debug.h
   util_defines.h
+  util_deque.h
   util_guarded_allocator.cpp
   util_foreach.h
   util_function.h
diff --git a/intern/cycles/util/util_debug.cpp b/intern/cycles/util/util_debug.cpp
index aabfea7fc49..e52fda26845 100644
--- a/intern/cycles/util/util_debug.cpp
+++ b/intern/cycles/util/util_debug.cpp
@@ -120,7 +120,7 @@ void DebugFlags::OpenCL::reset()
   debug = (getenv("CYCLES_OPENCL_DEBUG") != NULL);
 }
 
-DebugFlags::DebugFlags() : viewport_static_bvh(false)
+DebugFlags::DebugFlags() : viewport_static_bvh(false), running_inside_blender(false)
 {
   /* Nothing for now. */
 }
diff --git a/intern/cycles/util/util_debug.h b/intern/cycles/util/util_debug.h
index 83d9e96ffa5..5b0004ea768 100644
--- a/intern/cycles/util/util_debug.h
+++ b/intern/cycles/util/util_debug.h
@@ -33,6 +33,8 @@ class DebugFlags {
   /* Use static BVH in viewport, to match final render exactly. */
   bool viewport_static_bvh;
 
+  bool running_inside_blender;
+
   /* Descriptor of CPU feature-set to be used. */
   struct CPU {
     CPU();
diff --git a/intern/cycles/util/util_defines.h b/intern/cycles/util/util_defines.h
index 7f3bead0a18..2778cffba3a 100644
--- a/intern/cycles/util/util_defines.h
+++ b/intern/cycles/util/util_defines.h
@@ -30,6 +30,7 @@
 #  ifndef __KERNEL_GPU__
 #    define ccl_device static inline
 #    define ccl_device_noinline static
+#    define ccl_device_noinline_cpu ccl_device_noinline
 #    define ccl_global
 #    define ccl_static_constant static const
 #    define ccl_constant const
@@ -38,6 +39,7 @@
 #    define ccl_private
 #    define ccl_restrict __restrict
 #    define ccl_ref &
+#    define ccl_optional_struct_init
 #    define __KERNEL_WITH_SSE_ALIGN__
 
 #    if defined(_WIN32) && !defined(FREE_WINDOWS)
diff --git a/intern/cycles/util/util_deque.h b/intern/cycles/util/util_deque.h
new file mode 100644
index 00000000000..ccac961aa7d
--- /dev/null
+++ b/intern/cycles/util/util_deque.h
@@ -0,0 +1,28 @@
+/*
+ * Copyright 2011-2018 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef __UTIL_DEQUE_H__
+#define __UTIL_DEQUE_H__
+
+#include <deque>
+
+CCL_NAMESPACE_BEGIN
+
+using std::deque;
+
+CCL_NAMESPACE_END
+
+#endif /* __UTIL_DEQUE_H__ */
diff --git a/intern/cycles/util/util_hash.h b/intern/cycles/util/util_hash.h
index 785482967db..2b1f26de03d 100644
--- a/intern/cycles/util/util_hash.h
+++ b/intern/cycles/util/util_hash.h
@@ -21,39 +21,196 @@
 
 CCL_NAMESPACE_BEGIN
 
-ccl_device_inline uint hash_int_2d(uint kx, uint ky)
-{
+/* ***** Jenkins Lookup3 Hash Functions ***** */
+
+/* Source: http://burtleburtle.net/bob/c/lookup3.c */
+
 #define rot(x, k) (((x) << (k)) | ((x) >> (32 - (k))))
 
+#define mix(a, b, c) \
+  { \
+    a -= c; \
+    a ^= rot(c, 4); \
+    c += b; \
+    b -= a; \
+    b ^= rot(a, 6); \
+    a += c; \
+    c -= b; \
+    c ^= rot(b, 8); \
+    b += a; \
+    a -= c; \
+    a ^= rot(c, 16); \
+    c += b; \
+    b -= a; \
+    b ^= rot(a, 19); \
+    a += c; \
+    c -= b; \
+    c ^= rot(b, 4); \
+    b += a; \
+  }
+
+#define final(a, b, c) \
+  { \
+    c ^= b; \
+    c -= rot(b, 14); \
+    a ^= c; \
+    a -= rot(c, 11); \
+    b ^= a; \
+    b -= rot(a, 25); \
+    c ^= b; \
+    c -= rot(b, 16); \
+    a ^= c; \
+    a -= rot(c, 4); \
+    b ^= a; \
+    b -= rot(a, 14); \
+    c ^= b; \
+    c -= rot(b, 24); \
+  }
+
+ccl_device_inline uint hash_uint(uint kx)
+{
   uint a, b, c;
+  a = b = c = 0xdeadbeef + (1 << 2) + 13;
+
+  a += kx;
+  final(a, b, c);
 
+  return c;
+}
+
+ccl_device_inline uint hash_uint2(uint kx, uint ky)
+{
+  uint a, b, c;
   a = b = c = 0xdeadbeef + (2 << 2) + 13;
+
+  b += ky;
+  a += kx;
+  final(a, b, c);
+
+  return c;
+}
+
+ccl_device_inline uint hash_uint3(uint kx, uint ky, uint kz)
+{
+  uint a, b, c;
+  a = b = c = 0xdeadbeef + (3 << 2) + 13;
+
+  c += kz;
+  b += ky;
+  a += kx;
+  final(a, b, c);
+
+  return c;
+}
+
+ccl_device_inline uint hash_uint4(uint kx, uint ky, uint kz, uint kw)
+{
+  uint a, b, c;
+  a = b = c = 0xdeadbeef + (4 << 2) + 13;
+
   a += kx;
   b += ky;
+  c += kz;
+  mix(a, b, c);
 
-  c ^= b;
-  c -= rot(b, 14);
-  a ^= c;
-  a -= rot(c, 11);
-  b ^= a;
-  b -= rot(a, 25);
-  c ^= b;
-  c -= rot(b, 16);
-  a ^= c;
-  a -= rot(c, 4);
-  b ^= a;
-  b -= rot(a, 14);
-  c ^= b;
-  c -= rot(b, 24);
+  a += kw;
+  final(a, b, c);
 
   return c;
+}
 
 #undef rot
+#undef final
+#undef mix
+
+/* Hashing uint or uint[234] into a float in the range [0, 1]. */
+
+ccl_device_inline float hash_uint_to_float(uint kx)
+{
+  return (float)hash_uint(kx) / (float)0xFFFFFFFFu;
+}
+
+ccl_device_inline float hash_uint2_to_float(uint kx, uint ky)
+{
+  return (float)hash_uint2(kx, ky) / (float)0xFFFFFFFFu;
+}
+
+ccl_device_inline float hash_uint3_to_float(uint kx, uint ky, uint kz)
+{
+  return (float)hash_uint3(kx, ky, kz) / (float)0xFFFFFFFFu;
+}
+
+ccl_device_inline float hash_uint4_to_float(uint kx, uint ky, uint kz, uint kw)
+{
+  return (float)hash_uint4(kx, ky, kz, kw) / (float)0xFFFFFFFFu;
+}
+
+/* Hashing float or float[234] into a float in the range [0, 1]. */
+
+ccl_device_inline float hash_float_to_float(float k)
+{
+  return hash_uint_to_float(__float_as_uint(k));
+}
+
+ccl_device_inline float hash_float2_to_float(float2 k)
+{
+  return hash_uint2_to_float(__float_as_uint(k.x), __float_as_uint(k.y));
+}
+
+ccl_device_inline float hash_float3_to_float(float3 k)
+{
+  return hash_uint3_to_float(__float_as_uint(k.x), __float_as_uint(k.y), __float_as_uint(k.z));
+}
+
+ccl_device_inline float hash_float4_to_float(float4 k)
+{
+  return hash_uint4_to_float(
+      __float_as_uint(k.x), __float_as_uint(k.y), __float_as_uint(k.z), __float_as_uint(k.w));
+}
+
+/* Hashing float[234] into float[234] of components in the range [0, 1]. */
+
+ccl_device_inline float2 hash_float2_to_float2(float2 k)
+{
+  return make_float2(hash_float2_to_float(k), hash_float3_to_float(make_float3(k.x, k.y, 1.0)));
+}
+
+ccl_device_inline float3 hash_float3_to_float3(float3 k)
+{
+  return make_float3(hash_float3_to_float(k),
+                     hash_float4_to_float(make_float4(k.x, k.y, k.z, 1.0)),
+                     hash_float4_to_float(make_float4(k.x, k.y, k.z, 2.0)));
 }
 
-ccl_device_inline uint hash_int(uint k)
+ccl_device_inline float4 hash_float4_to_float4(float4 k)
+{
+  return make_float4(hash_float4_to_float(k),
+                     hash_float4_to_float(make_float4(k.w, k.x, k.y, k.z)),
+                     hash_float4_to_float(make_float4(k.z, k.w, k.x, k.y)),
+                     hash_float4_to_float(make_float4(k.y, k.z, k.w, k.x)));
+}
+
+/* Hashing float or float[234] into float3 of components in range [0, 1]. */
+
+ccl_device_inline float3 hash_float_to_float3(float k)
 {
-  return hash_int_2d(k, 0);
+  return make_float3(hash_float_to_float(k),
+                     hash_float2_to_float(make_float2(k, 1.0)),
+                     hash_float2_to_float(make_float2(k, 2.0)));
+}
+
+ccl_device_inline float3 hash_float2_to_float3(float2 k)
+{
+  return make_float3(hash_float2_to_float(k),
+                     hash_float3_to_float(make_float3(k.x, k.y, 1.0)),
+                     hash_float3_to_float(make_float3(k.x, k.y, 2.0)));
+}
+
+ccl_device_inline float3 hash_float4_to_float3(float4 k)
+{
+  return make_float3(hash_float4_to_float(k),
+                     hash_float4_to_float(make_float4(k.z, k.x, k.w, k.y)),
+                     hash_float4_to_float(make_float4(k.w, k.z, k.y, k.x)));
 }
 
 #ifndef __KERNEL_GPU__
@@ -68,11 +225,6 @@ static inline uint hash_string(const char *str)
 }
 #endif
 
-ccl_device_inline float hash_int_01(uint k)
-{
-  return (float)hash_int(k) * (1.0f / (float)0xFFFFFFFF);
-}
-
 CCL_NAMESPACE_END
 
 #endif /* __UTIL_HASH_H__ */
diff --git a/intern/cycles/util/util_map.h b/intern/cycles/util/util_map.h
index 8385b08dd5a..f1b2522362f 100644
--- a/intern/cycles/util/util_map.h
+++ b/intern/cycles/util/util_map.h
@@ -25,6 +25,7 @@ CCL_NAMESPACE_BEGIN
 using std::map;
 using std::pair;
 using std::unordered_map;
+using std::unordered_multimap;
 
 template<typename T> static void map_free_memory(T &data)
 {
diff --git a/intern/cycles/util/util_math.h b/intern/cycles/util/util_math.h
index dde0d31f467..ebddd56bd40 100644
--- a/intern/cycles/util/util_math.h
+++ b/intern/cycles/util/util_math.h
@@ -617,6 +617,57 @@ ccl_device float bits_to_01(uint bits)
   return bits * (1.0f / (float)0xFFFFFFFF);
 }
 
+ccl_device_inline uint count_leading_zeros(uint x)
+{
+#if defined(__KERNEL_CUDA__) || defined(__KERNEL_OPTIX__)
+  return __clz(x);
+#elif defined(__KERNEL_OPENCL__)
+  return clz(x);
+#else
+  assert(x != 0);
+#  ifdef _MSC_VER
+  unsigned long leading_zero = 0;
+  _BitScanReverse(&leading_zero, x);
+  return (31 - leading_zero);
+#  else
+  return __builtin_clz(x);
+#  endif
+#endif
+}
+
+ccl_device_inline uint count_trailing_zeros(uint x)
+{
+#if defined(__KERNEL_CUDA__) || defined(__KERNEL_OPTIX__)
+  return (__ffs(x) - 1);
+#elif defined(__KERNEL_OPENCL__)
+  return (31 - count_leading_zeros(x & -x));
+#else
+  assert(x != 0);
+#  ifdef _MSC_VER
+  unsigned long ctz = 0;
+  _BitScanForward(&ctz, x);
+  return ctz;
+#  else
+  return __builtin_ctz(x);
+#  endif
+#endif
+}
+
+ccl_device_inline uint find_first_set(uint x)
+{
+#if defined(__KERNEL_CUDA__) || defined(__KERNEL_OPTIX__)
+  return __ffs(x);
+#elif defined(__KERNEL_OPENCL__)
+  return (x != 0) ? (32 - count_leading_zeros(x & (-x))) : 0;
+#else
+#  ifdef _MSC_VER
+  return (x != 0) ? (32 - count_leading_zeros(x & (-x))) : 0;
+#  else
+  return __builtin_ffs(x);
+#  endif
+#endif
+}
+
 /* projections */
 ccl_device_inline float2 map_to_tube(const float3 co)
 {
diff --git a/intern/cycles/util/util_math_float3.h b/intern/cycles/util/util_math_float3.h
index 554b7408148..c9a5b34aa58 100644
--- a/intern/cycles/util/util_math_float3.h
+++ b/intern/cycles/util/util_math_float3.h
@@ -47,6 +47,7 @@ ccl_device_inline float3 operator/=(float3 &a, float f);
 ccl_device_inline bool operator==(const float3 &a, const float3 &b);
 ccl_device_inline bool operator!=(const float3 &a, const float3 &b);
 
+ccl_device_inline float distance(const float3 &a, const float3 &b);
 ccl_device_inline float dot(const float3 &a, const float3 &b);
 ccl_device_inline float dot_xy(const float3 &a, const float3 &b);
 ccl_device_inline float3 cross(const float3 &a, const float3 &b);
@@ -58,6 +59,8 @@ ccl_device_inline float3 fabs(const float3 &a);
 ccl_device_inline float3 mix(const float3 &a, const float3 &b, float t);
 ccl_device_inline float3 rcp(const float3 &a);
 ccl_device_inline float3 sqrt(const float3 &a);
+ccl_device_inline float3 floor(const float3 &a);
+ccl_device_inline float3 ceil(const float3 &a);
 #endif /* !__KERNEL_OPENCL__ */
 
 ccl_device_inline float min3(float3 a);
@@ -65,10 +68,15 @@ ccl_device_inline float max3(float3 a);
 ccl_device_inline float len(const float3 a);
 ccl_device_inline float len_squared(const float3 a);
 
+ccl_device_inline float3 reflect(const float3 incident, const float3 normal);
+ccl_device_inline float3 project(const float3 v, const float3 v_proj);
+
 ccl_device_inline float3 saturate3(float3 a);
 ccl_device_inline float3 safe_normalize(const float3 a);
 ccl_device_inline float3 normalize_len(const float3 a, float *t);
 ccl_device_inline float3 safe_normalize_len(const float3 a, float *t);
+ccl_device_inline float3 safe_divide_float3_float3(const float3 a, const float3 b);
+ccl_device_inline float3 safe_divide_float3_float(const float3 a, const float b);
 ccl_device_inline float3 interp(float3 a, float3 b, float t);
 ccl_device_inline float3 sqr3(float3 a);
 
@@ -205,6 +213,11 @@ ccl_device_inline bool operator!=(const float3 &a, const float3 &b)
   return !(a == b);
 }
 
+ccl_device_inline float distance(const float3 &a, const float3 &b)
+{
+  return len(a - b);
+}
+
 ccl_device_inline float dot(const float3 &a, const float3 &b)
 {
 #  if defined(__KERNEL_SSE41__) && defined(__KERNEL_SSE__)
@@ -281,6 +294,24 @@ ccl_device_inline float3 sqrt(const float3 &a)
 #  endif
 }
 
+ccl_device_inline float3 floor(const float3 &a)
+{
+#  ifdef __KERNEL_SSE__
+  return float3(_mm_floor_ps(a));
+#  else
+  return make_float3(floorf(a.x), floorf(a.y), floorf(a.z));
+#  endif
+}
+
+ccl_device_inline float3 ceil(const float3 &a)
+{
+#  ifdef __KERNEL_SSE__
+  return float3(_mm_ceil_ps(a));
+#  else
+  return make_float3(ceilf(a.x), ceilf(a.y), ceilf(a.z));
+#  endif
+}
+
 ccl_device_inline float3 mix(const float3 &a, const float3 &b, float t)
 {
   return a + t * (b - a);
@@ -321,6 +352,19 @@ ccl_device_inline float len_squared(const float3 a)
   return dot(a, a);
 }
 
+ccl_device_inline float3 reflect(const float3 incident, const float3 normal)
+{
+  float3 unit_normal = normalize(normal);
+  return incident - 2.0f * unit_normal * dot(incident, unit_normal);
+}
+
+ccl_device_inline float3 project(const float3 v, const float3 v_proj)
+{
+  float len_squared = dot(v_proj, v_proj);
+  return (len_squared != 0.0f) ? (dot(v, v_proj) / len_squared) * v_proj :
+                                 make_float3(0.0f, 0.0f, 0.0f);
+}
+
 ccl_device_inline float3 saturate3(float3 a)
 {
   return make_float3(saturate(a.x), saturate(a.y), saturate(a.z));
@@ -345,6 +389,18 @@ ccl_device_inline float3 safe_normalize_len(const float3 a, float *t)
   return (*t != 0.0f) ? a / (*t) : a;
 }
 
+ccl_device_inline float3 safe_divide_float3_float3(const float3 a, const float3 b)
+{
+  return make_float3((b.x != 0.0f) ? a.x / b.x : 0.0f,
+                     (b.y != 0.0f) ? a.y / b.y : 0.0f,
+                     (b.z != 0.0f) ? a.z / b.z : 0.0f);
+}
+
+ccl_device_inline float3 safe_divide_float3_float(const float3 a, const float b)
+{
+  return (b != 0.0f) ? a / b : make_float3(0.0f, 0.0f, 0.0f);
+}
+
 ccl_device_inline float3 interp(float3 a, float3 b, float t)
 {
   return a + t * (b - a);
diff --git a/intern/cycles/util/util_system.cpp b/intern/cycles/util/util_system.cpp
index 0cd991c6231..f700f9bd277 100644
--- a/intern/cycles/util/util_system.cpp
+++ b/intern/cycles/util/util_system.cpp
@@ -357,7 +357,7 @@ size_t system_physical_ram()
   MEMORYSTATUSEX ram;
   ram.dwLength = sizeof(ram);
   GlobalMemoryStatusEx(&ram);
-  return ram.ullTotalPhys * 1024;
+  return ram.ullTotalPhys;
 #elif defined(__APPLE__)
   uint64_t ram = 0;
   size_t len = sizeof(ram);
diff --git a/intern/cycles/util/util_vector.h b/intern/cycles/util/util_vector.h
index 437478d64d3..04fb33368d9 100644
--- a/intern/cycles/util/util_vector.h
+++ b/intern/cycles/util/util_vector.h
@@ -27,7 +27,7 @@
 
 CCL_NAMESPACE_BEGIN
 
-/* Own subclass-ed vestion of std::vector. Subclass is needed because:
+/* Own subclass-ed version of std::vector. Subclass is needed because:
  *
  * - Use own allocator which keeps track of used/peak memory.
  * - Have method to ensure capacity is re-set to 0.
diff --git a/intern/ffmpeg/ffmpeg_compat.h b/intern/ffmpeg/ffmpeg_compat.h
index 863fbd40bb0..d254fa76ba9 100644
--- a/intern/ffmpeg/ffmpeg_compat.h
+++ b/intern/ffmpeg/ffmpeg_compat.h
@@ -364,7 +364,7 @@ void avcodec_free_frame(AVFrame **frame)
 #  define avio_size url_fsize
 #endif
 
-/* there are some version inbetween, which have avio_... functions but no
+/* There are some version in between, which have avio_... functions but no
  * AVIO_FLAG_... */
 #ifndef AVIO_FLAG_WRITE
 #  define AVIO_FLAG_WRITE URL_WRONLY
diff --git a/intern/ghost/GHOST_C-api.h b/intern/ghost/GHOST_C-api.h
index 0f735711b24..9c6a0861280 100644
--- a/intern/ghost/GHOST_C-api.h
+++ b/intern/ghost/GHOST_C-api.h
@@ -477,7 +477,7 @@ extern GHOST_TEventDataPtr GHOST_GetEventData(GHOST_EventHandle eventhandle);
 
 /**
  * Returns the timer callback.
- * \param timertaskhandle The handle to the timertask
+ * \param timertaskhandle The handle to the timer-task.
  * \return The timer callback.
  */
 extern GHOST_TimerProcPtr GHOST_GetTimerProc(GHOST_TimerTaskHandle timertaskhandle);
diff --git a/intern/ghost/GHOST_ISystem.h b/intern/ghost/GHOST_ISystem.h
index f0ceb7fb8ba..27be80a2f20 100644
--- a/intern/ghost/GHOST_ISystem.h
+++ b/intern/ghost/GHOST_ISystem.h
@@ -402,7 +402,7 @@ class GHOST_ISystem {
 #ifdef WITH_INPUT_NDOF
   /**
    * Sets 3D mouse deadzone
-   * \param deadzone: Deadzone of the 3D mouse (both for rotation and pan) relative to full range
+   * \param deadzone: Dead-zone of the 3D mouse (both for rotation and pan) relative to full range
    */
   virtual void setNDOFDeadZone(float deadzone) = 0;
 #endif
diff --git a/intern/guardedalloc/intern/mallocn_guarded_impl.c b/intern/guardedalloc/intern/mallocn_guarded_impl.c
index 8ba14b3887b..ef74f394fb1 100644
--- a/intern/guardedalloc/intern/mallocn_guarded_impl.c
+++ b/intern/guardedalloc/intern/mallocn_guarded_impl.c
@@ -552,7 +552,13 @@ void *MEM_guarded_malloc_arrayN(size_t len, size_t size, const char *str)
 
 void *MEM_guarded_mallocN_aligned(size_t len, size_t alignment, const char *str)
 {
-  MemHead *memh;
+  /* We only support alignment to a power of two. */
+  assert(IS_POW2(alignment));
+
+  /* Use a minimal alignment of 8. Otherwise MEM_guarded_freeN thinks it is an illegal pointer. */
+  if (alignment < 8) {
+    alignment = 8;
+  }
 
   /* It's possible that MemHead's size is not properly aligned,
    * do extra padding to deal with this.
@@ -567,13 +573,10 @@ void *MEM_guarded_mallocN_aligned(size_t len, size_t alignment, const char *str)
    */
   assert(alignment < 1024);
 
-  /* We only support alignment to a power of two. */
-  assert(IS_POW2(alignment));
-
   len = SIZET_ALIGN_4(len);
 
-  memh = (MemHead *)aligned_malloc(len + extra_padding + sizeof(MemHead) + sizeof(MemTail),
-                                   alignment);
+  MemHead *memh = (MemHead *)aligned_malloc(
+      len + extra_padding + sizeof(MemHead) + sizeof(MemTail), alignment);
 
   if (LIKELY(memh)) {
     /* We keep padding in the beginning of MemHead,
diff --git a/intern/iksolver/CMakeLists.txt b/intern/iksolver/CMakeLists.txt
index 5e12cdbcc2f..a7a346ccddd 100644
--- a/intern/iksolver/CMakeLists.txt
+++ b/intern/iksolver/CMakeLists.txt
@@ -27,6 +27,10 @@ set(INC_SYS
   ${EIGEN3_INCLUDE_DIRS}
 )
 
+if(WIN32)
+  add_definitions(-D_USE_MATH_DEFINES)
+endif()
+
 set(SRC
   intern/IK_QJacobian.cpp
   intern/IK_QJacobianSolver.cpp
diff --git a/intern/libmv/CMakeLists.txt b/intern/libmv/CMakeLists.txt
index 019364d84df..e16e27368d0 100644
--- a/intern/libmv/CMakeLists.txt
+++ b/intern/libmv/CMakeLists.txt
@@ -38,6 +38,9 @@ set(LIB
 )
 
 if(WITH_LIBMV)
+  if(WIN32)
+    add_definitions(-D_USE_MATH_DEFINES)
+  endif()
   add_definitions(${GFLAGS_DEFINES})
   add_definitions(${GLOG_DEFINES})
   add_definitions(${CERES_DEFINES})
diff --git a/intern/libmv/libmv/tracking/track_region.cc b/intern/libmv/libmv/tracking/track_region.cc
index ef6dac65236..895c9a1e23d 100644
--- a/intern/libmv/libmv/tracking/track_region.cc
+++ b/intern/libmv/libmv/tracking/track_region.cc
@@ -25,9 +25,6 @@
 // smart coder went through the TODO's and made the suggested performance
 // enhancements.
 
-// Necessary for M_E when building with MSVC.
-#define _USE_MATH_DEFINES
-
 #include "libmv/tracking/track_region.h"
 
 #include <Eigen/SVD>
diff --git a/intern/libmv/libmv/tracking/track_region.h b/intern/libmv/libmv/tracking/track_region.h
index be1d8ef3e03..61dce22bcb8 100644
--- a/intern/libmv/libmv/tracking/track_region.h
+++ b/intern/libmv/libmv/tracking/track_region.h
@@ -20,9 +20,6 @@
 
 #ifndef LIBMV_TRACKING_TRACK_REGION_H_
 
-// Necessary for M_E when building with MSVC.
-#define _USE_MATH_DEFINES
-
 #include "libmv/image/image.h"
 #include "libmv/image/sample.h"
 #include "libmv/numeric/numeric.h"
diff --git a/make.bat b/make.bat
index b0323cde1bb..3f09718ab4f 100644
--- a/make.bat
+++ b/make.bat
@@ -26,12 +26,6 @@ if "%SHOW_HELP%" == "1" (
 	goto EOF
 )
 
-if "%BUILD_UPDATE%" == "1" (
-	call "%BLENDER_DIR%\build_files\windows\update_sources.cmd"
-	goto EOF
-)
-
-
 if "%FORMAT%" == "1" (
 	call "%BLENDER_DIR%\build_files\windows\format.cmd"
 	goto EOF
@@ -54,6 +48,14 @@ if "%BUILD_VS_YEAR%" == "" (
 	)
 )
 
+if "%BUILD_UPDATE%" == "1" (
+	call "%BLENDER_DIR%\build_files\windows\check_libraries.cmd"
+	if errorlevel 1 goto EOF
+
+	call "%BLENDER_DIR%\build_files\windows\update_sources.cmd"
+	goto EOF
+)
+
 call "%BLENDER_DIR%\build_files\windows\set_build_dir.cmd"
 
 echo Building blender with VS%BUILD_VS_YEAR% for %BUILD_ARCH% in %BUILD_DIR%
diff --git a/release/darwin/buildbot/background.tif b/release/darwin/buildbot/background.tif
new file mode 100644
index 00000000000..5253a6bf439
--- /dev/null
+++ b/release/darwin/buildbot/background.tif
diff --git a/release/darwin/bundle.sh b/release/darwin/bundle.sh
index b0f489e6fc2..91ce4f61d37 100755
--- a/release/darwin/bundle.sh
+++ b/release/darwin/bundle.sh
@@ -54,6 +54,11 @@ while [[ $# -gt 0 ]]; do
             shift
             shift
             ;;
+        --background-image)
+            _background_image="$2"
+            shift
+            shift
+            ;;
         -h|--help)
             echo "Usage:"
             echo " $(basename "$0") --source DIR --dmg IMAGENAME "
diff --git a/release/datafiles/blender_icons.svg b/release/datafiles/blender_icons.svg
index 3ed2da387ac..2633ece87fc 100644
--- a/release/datafiles/blender_icons.svg
+++ b/release/datafiles/blender_icons.svg
@@ -1,6 +1,6 @@
 <?xml version="1.0" encoding="UTF-8" standalone="no"?>
 <!-- Created with Inkscape (http://www.inkscape.org/) -->
-<svg xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:cc="http://creativecommons.org/ns#" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:svg="http://www.w3.org/2000/svg" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd" xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape" width="602" height="640" id="svg2" sodipodi:version="0.32" inkscape:version="0.92.3 (2405546, 2018-03-11)" version="1.0" sodipodi:docname="blender_icons 2.8 - SMALL.svg" inkscape:output_extension="org.inkscape.output.svg.inkscape" style="display:inline;enable-background:new" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\ICONS - no backdrop.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
+<svg xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:cc="http://creativecommons.org/ns#" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:svg="http://www.w3.org/2000/svg" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd" xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape" width="602" height="640" id="svg2" sodipodi:version="0.32" inkscape:version="0.92.3 (2405546, 2018-03-11)" version="1.0" sodipodi:docname="blender_icons 2.81 - SMALL.svg" inkscape:output_extension="org.inkscape.output.svg.inkscape" style="display:inline;enable-background:new" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\ICONS - no backdrop.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
   
   <defs id="defs4">
     
@@ -3016,7 +3016,7 @@
       <path id="path6071" d="m 480.45673,673.52118 v 0.75 h -1 v 3.5 h 1 v 0.75 h 2 v -1 h -2 v -3 h 2 v -1 z" style="display:inline;overflow:visible;visibility:visible;opacity:0.6;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate" inkscape:connector-curvature="0" sodipodi:nodetypes="ccccccccccccc"/>
     </g>
   </g>
-  <g inkscape:groupmode="layer" id="layer8" inkscape:label="2.8" style="display:inline">
+  <g inkscape:groupmode="layer" id="layer8" inkscape:label="2.81" style="display:inline">
     <path inkscape:connector-curvature="0" id="path6587" d="M 61.503909,123 A 6.5039066,6.5039066 0 0 1 55.000003,129.50391 6.5039066,6.5039066 0 0 1 48.496097,123 6.5039066,6.5039066 0 0 1 55.000003,116.4961 6.5039066,6.5039066 0 0 1 61.503909,123 Z" style="color:#000000;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;fill:#ffffff;fill-opacity:1;fill-rule:evenodd;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"/>
     <g id="g10975" transform="translate(-378.00707,-293.99994)" style="display:inline;opacity:1;fill:#ffffff;enable-background:new" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96"/>
     <g id="g10009" transform="translate(5.3904633e-4,-20.999933)" style="display:inline;opacity:1;fill:#ffffff;stroke-width:1;stroke-miterlimit:4;stroke-dasharray:none;enable-background:new" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96"/>
@@ -3295,10 +3295,6 @@
     <g style="display:inline;fill:#ffffff;enable-background:new" id="g23466" transform="translate(-1.8536743e-6,-0.999995)">
       <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" d="m 10.482422,494 c -0.189602,0.007 -0.359053,0.12023 -0.4375,0.29297 l -3.7734376,8.25 c -0.00134,0.003 -0.00264,0.007 -0.00391,0.01 C 6.1587525,502.80633 6,503.10119 6,503.5 c 0,0.83813 0.5261723,1.53471 1.296875,1.92188 0.7707027,0.38715 1.7901254,0.55582 3.023437,0.57226 0.295371,0.003 0.52928,-0.24867 0.503907,-0.54297 9.47e-4,0.0108 -0.01109,-0.37852 -0.01563,-0.70898 -0.0045,-0.33047 -0.0078,-0.6698 -0.0078,-0.74219 0,-1.86768 1.015625,-3.51914 2.519531,-4.4375 0.228366,-0.13928 0.306682,-0.43361 0.177735,-0.66797 L 10.9375,494.25781 C 10.846425,494.09307 10.670549,493.99343 10.482422,494 Z M 18,494 c -1.098638,0 -1.999999,0.90136 -2,2 1e-6,1.09864 0.901362,2 2,2 1.098638,0 1.999999,-0.90136 2,-2 -1e-6,-1.09864 -0.901362,-2 -2,-2 z m -2,6 c -2.203217,0 -3.999999,1.79678 -4,4 10e-7,2.20322 1.796783,4 4,4 2.203217,0 3.999999,-1.79678 4,-4 -1e-6,-2.20322 -1.796783,-4 -4,-4 z m 0,1 c 1.662777,0 2.999999,1.33722 3,3 -1e-6,1.66278 -1.337223,3 -3,3 -1.662777,0 -2.999999,-1.33722 -3,-3 10e-7,-1.66278 1.337223,-3 3,-3 z m -0.541016,1.0332 C 14.658737,502.0332 14,502.69389 14,503.49414 a 0.50005,0.50005 0 1 0 1,0 c 0,-0.25981 0.199172,-0.46094 0.458984,-0.46094 a 0.50005,0.50005 0 1 0 0,-1 z" transform="translate(1.8536743e-6,0.999995)" id="path10645" inkscape:connector-curvature="0"/>
     </g>
-    <g style="display:inline;opacity:1;fill:#ffffff;enable-background:new" transform="translate(-84.000002,-273)" id="g12159" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
-      <rect ry="0" rx="0" y="346" x="446" height="16" width="16" id="rect12155" style="display:inline;opacity:0;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;enable-background:new"/>
-      <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1px;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="m 447.5,348 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 10 1 c 0,0.41667 0.19292,0.77495 0.45898,1.04102 C 447.72505,360.80708 448.08333,361 448.5,361 h 0.5 10.5 c 0.41667,0 0.77495,-0.19292 1.04102,-0.45898 C 460.80708,360.27495 461,359.91667 461,359.5 v -6 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 h -3 -7 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 5.5 c 0,0.20833 -0.11958,0.47505 -0.32227,0.67773 C 449.47505,359.88042 449.20833,360 449,360 h -0.5 c -0.0833,0 -0.22505,-0.0571 -0.33398,-0.16602 C 448.05708,359.72505 448,359.58333 448,359.5 V 359 h 0.5 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 V 352 h 8.5 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 v -2 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 L 452,348.992 V 348.5 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 z" id="path12157" inkscape:connector-curvature="0" sodipodi:nodetypes="cccscscscscccccscsssscccccccccccc"/>
-    </g>
     <rect style="display:inline;overflow:visible;visibility:visible;opacity:0;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:2.70000005;marker:none;enable-background:accumulate" id="rect12082" width="16" height="16" x="26" y="136" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96"/>
     <g transform="translate(-1.8536743e-6,4.4999696e-6)" style="display:inline;fill:#ffffff;enable-background:new" id="g14497">
       <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:butt;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="M 324.60938,536 C 323.75288,536 323,536.6483 323,537.5 v 3 c 0,0.8517 0.75288,1.5 1.60938,1.5 h 4.78124 C 330.24712,542 331,541.3517 331,540.5 v -3 c 0,-0.8517 -0.75287,-1.5 -1.60938,-1.5 z M 332,537 v 3 c 0.56267,0 1,-0.50398 1,-1.03125 v -0.9375 C 333,537.50408 332.56269,537 332,537 Z m -7.53516,3 a 0.50005,0.50005 0 0 1 0.0352,0 h 5 a 0.50005,0.50005 0 1 1 0,1 h -5 a 0.50030966,0.50030966 0 0 1 -0.0352,-1 z m -1.85546,4 C 321.75288,544 321,544.6483 321,545.5 v 3 c 0,0.8517 0.75288,1.5 1.60938,1.5 h 4.78124 C 328.24712,550 329,549.3517 329,548.5 v -3 c 0,-0.8517 -0.75287,-1.5 -1.60938,-1.5 z M 330,546 v 3 c 0.56267,0 1,-0.50398 1,-1.03125 v -0.9375 C 331,546.50408 330.56269,546 330,546 Z m -7.53516,2 a 0.50005,0.50005 0 0 1 0.0352,0 h 5 a 0.50005,0.50005 0 1 1 0,1 h -5 a 0.50030966,0.50030966 0 0 1 -0.0352,-1 z" transform="translate(1.8536743e-6,-4.4999696e-6)" id="rect12085" inkscape:connector-curvature="0"/>
@@ -3415,9 +3411,6 @@
         <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="M 746.49219,-91.007812 A 0.50005,0.50005 0 0 0 746,-90.5 v 3 a 0.50005,0.50005 0 1 0 1,0 v -3 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z m 0,5 A 0.50005,0.50005 0 0 0 746,-85.5 v 3 a 0.50005,0.50005 0 1 0 1,0 v -3 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z m 0,5 A 0.50005,0.50005 0 0 0 746,-80.5 v 3 a 0.50005,0.50005 0 1 0 1,0 v -3 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z" id="path8336-9" inkscape:connector-curvature="0"/>
       </g>
     </g>
-    <g transform="translate(-1.8536743e-6,4.4999696e-6)" style="display:inline;opacity:1;fill:#ffffff;enable-background:new" id="g10098-7" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
-      <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:0.99999994;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" d="M 533.53711,53 C 532.69436,53 532,53.694362 532,54.537109 V 65.462891 C 532,66.305638 532.69436,67 533.53711,67 h 8.92578 C 543.30564,67 544,66.305638 544,65.462891 V 54.537109 C 544,53.694362 543.30564,53 542.46289,53 Z m 0,1 H 535 v 3 h 1 v -3 h 1 v 3 h 1 v -3 h 1 v 3 h 1 v -3 h 1 v 3 h 1 v -3 h 0.46289 C 542.76894,54 543,54.231065 543,54.537109 V 65.462891 C 543,65.768935 542.76894,66 542.46289,66 h -8.92578 C 533.23106,66 533,65.768935 533,65.462891 V 54.537109 C 533,54.231065 533.23106,54 533.53711,54 Z M 534,64 v 1 h 1 v -1 z m 7,0 v 1 h 1 v -1 z" transform="translate(1.8536743e-6,-4.4999696e-6)" id="rect10082-5" inkscape:connector-curvature="0"/>
-    </g>
     <g style="display:inline;opacity:1;fill:#ffffff;enable-background:new" id="g4847-8" transform="translate(-230.99288,-230.99288)" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
       <g transform="translate(126)" style="opacity:0.6;fill:#ffffff" id="g4287-2">
         <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1px;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="m 156.5,158 c -0.27614,0.004 -0.49651,0.23167 -0.49219,0.50781 v 0.75 h 1 v -0.75 C 157.01223,158.22555 156.78226,157.99558 156.5,158 Z m 7,0 c -0.27614,0.004 -0.49651,0.23167 -0.49219,0.50781 v 2.4961 L 160.00195,161 160,162 l 3.00781,0.004 v 6.0039 h -6.0039 L 157,165 l -1,0.002 0.004,3.00586 h -2.4961 c -0.67616,-0.01 -0.67616,1.00957 0,1 h 2.4961 l 0.004,2.5 c -0.01,0.67616 1.00957,0.67616 1,0 l -0.004,-2.5 h 6.0039 v 2.5 c -0.01,0.67616 1.00957,0.67616 1,0 v -2.5 h 2.5 c 0.67616,0.01 0.67616,-1.00956 0,-1 h -2.5 v -6.0039 l 2.5,0.004 c 0.67616,0.01 0.67616,-1.00956 0,-1 l -2.5,-0.004 v -2.4961 C 164.0119,158.22555 163.78226,157.99558 163.5,158 Z m -9.99414,3 c -0.6733,-0.0135 -0.6832,1 -0.01,1 l 0.75782,0.008 0.008,-1 z" transform="translate(104.99288,230.99288)" id="path4273-3" inkscape:connector-curvature="0"/>
@@ -3491,7 +3484,6 @@
     </g>
     <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:3;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" d="m 11.455078,430.98242 a 1.50015,1.50015 0 0 0 -1.4765624,1.52149 v 6.3789 l -3.5605468,3.56055 a 1.50015,1.50015 0 1 0 2.1210937,2.12109 l 3.5605465,-3.56054 h 6.378907 a 1.50015,1.50015 0 1 0 0,-3 h -5.5 v -5.5 a 1.50015,1.50015 0 0 0 -1.523438,-1.52149 z" id="path12350-2" inkscape:connector-curvature="0"/>
     <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" d="m 199.50586,431 c -0.31034,2.7e-4 -0.5455,0.28021 -0.49219,0.58594 0.36378,2.06995 0.21088,3.24033 -0.13086,3.98828 -0.34174,0.74794 -0.89741,1.14621 -1.5625,1.65234 -0.66509,0.50614 -1.42742,1.12518 -1.87304,2.24414 -0.44562,1.11896 -0.57595,2.68203 -0.19141,5.10742 0.0385,0.24306 0.24806,0.422 0.49414,0.42188 h 8.75586 c 0.30739,1.4e-4 0.54213,-0.27449 0.49414,-0.57812 -0.36857,-2.32471 -0.21903,-3.70089 0.13086,-4.58204 0.34989,-0.88114 0.89751,-1.32255 1.54492,-1.8164 0.64741,-0.49385 1.4068,-1.03318 1.86328,-2.03516 0.45648,-1.00198 0.58623,-2.39428 0.20313,-4.57422 C 208.70043,431.17484 208.49284,431.0002 208.25,431 Z" id="path8644-7" inkscape:connector-curvature="0" sodipodi:nodetypes="ccccscccccssccc"/>
-    <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="m 515.5,452 c -0.8225,0 -1.5,0.6775 -1.5,1.5 0,0.64684 0.42097,1.19777 1,1.40625 v 3.38672 l -3.85352,3.85351 c -0.0938,0.0938 -0.14646,0.22092 -0.14648,0.35352 v 0.59375 c -0.57903,0.20848 -1,0.75941 -1,1.40625 0,0.8225 0.6775,1.5 1.5,1.5 0.8225,0 1.5,-0.6775 1.5,-1.5 0,-0.64684 -0.42097,-1.19777 -1,-1.40625 v -0.38672 l 3.85352,-3.85351 c 0.0938,-0.0938 0.14646,-0.22092 0.14648,-0.35352 v -3.59375 c 0.57903,-0.20848 1,-0.75941 1,-1.40625 0,-0.8225 -0.6775,-1.5 -1.5,-1.5 z m 7,0 c -0.64684,0 -1.19777,0.42097 -1.40625,1 H 519.5 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 4.79297 l -3.85352,3.85351 c -0.0938,0.0938 -0.14646,0.22092 -0.14648,0.35352 v 0.59375 c -0.57903,0.20848 -1,0.75941 -1,1.40625 0,0.8225 0.6775,1.5 1.5,1.5 0.8225,0 1.5,-0.6775 1.5,-1.5 0,-0.64684 -0.42097,-1.19777 -1,-1.40625 v -0.38672 l 3.85352,-3.85351 C 519.94729,458.75976 519.99998,458.6326 520,458.5 V 454 h 1.09375 c 0.20848,0.57903 0.75941,1 1.40625,1 0.8225,0 1.5,-0.6775 1.5,-1.5 0,-0.8225 -0.6775,-1.5 -1.5,-1.5 z m -7,1 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28206 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21794 -0.5,-0.5 0,-0.28206 0.21794,-0.5 0.5,-0.5 z m 7,0 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28206 -0.21794,0.5 -0.5,0.5 -0.27091,0 -0.47782,-0.20294 -0.49414,-0.46875 10e-4,-0.0208 10e-4,-0.0417 0,-0.0625 C 522.02218,453.20294 522.22909,453 522.5,453 Z m -11,3 c -0.8225,0 -1.5,0.6775 -1.5,1.5 0,0.8225 0.6775,1.5 1.5,1.5 0.8225,0 1.5,-0.6775 1.5,-1.5 0,-0.8225 -0.6775,-1.5 -1.5,-1.5 z m 0,1 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28206 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21794 -0.5,-0.5 0,-0.28206 0.21794,-0.5 0.5,-0.5 z m 11,0 c -0.8225,0 -1.5,0.6775 -1.5,1.5 0,0.64684 0.42097,1.19777 1,1.40625 v 1.38672 l -0.85352,0.85351 -1,1 c -0.002,0.002 -0.004,0.004 -0.006,0.006 C 519.94524,463.05837 519.72956,463 519.5,463 c -0.8225,0 -1.5,0.6775 -1.5,1.5 0,0.8225 0.6775,1.5 1.5,1.5 0.8225,0 1.5,-0.6775 1.5,-1.5 0,-0.23016 -0.0579,-0.44679 -0.15234,-0.64258 0.002,-10e-4 0.004,-0.003 0.006,-0.004 l 1,-1 1,-1 c 0.0937,-0.0938 0.14633,-0.22087 0.14634,-0.35342 v -1.59375 c 0.57903,-0.20848 1,-0.75941 1,-1.40625 0,-0.8225 -0.6775,-1.5 -1.5,-1.5 z m 0,1 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.27091 -0.20294,0.47782 -0.46875,0.49414 -0.013,-0.001 -0.0261,-0.002 -0.0391,-0.002 -0.008,4.8e-4 -0.0156,10e-4 -0.0234,0.002 C 522.20294,458.97782 522,458.77091 522,458.5 c 0,-0.28206 0.21794,-0.5 0.5,-0.5 z m -3,6 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28206 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21794 -0.5,-0.5 0,-0.28206 0.21794,-0.5 0.5,-0.5 z m -8.03125,0.006 c 0.0208,0.001 0.0417,0.001 0.0625,0 0.26581,0.0162 0.46875,0.22309 0.46875,0.494 0,0.28206 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21794 -0.5,-0.5 0,-0.27091 0.20294,-0.47782 0.46875,-0.49414 z m 4,0 c 0.0208,0.001 0.0417,0.001 0.0625,0 0.26581,0.0162 0.46875,0.22309 0.46875,0.494 0,0.28206 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21794 -0.5,-0.5 0,-0.27091 0.20294,-0.47782 0.46875,-0.49414 z" id="circle14423" inkscape:connector-curvature="0" sodipodi:nodetypes="sscccccssscccccssscccccccsssccccccsssssssssssccssssssssssssscccccssssccccccsssscccsssssssccsssccccssscc"/>
     <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1px;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" d="m 96.5,473 c -2.087116,0 -3.478164,1.04607 -4.390625,2.1875 a 0.50024408,0.50024408 0 0 0 0.78125,0.625 C 93.690488,474.81193 94.734388,474 96.5,474 c 2.098393,0 3.134109,0.87695 3.75195,1.94531 C 100.8698,477.01367 101,478.31651 101,479 h -5.25 c -1.273438,0 -2.443243,0.36418 -3.314453,1.05469 -0.87121,0.69051 -1.427734,1.72791 -1.427735,2.94531 10e-7,1.2174 0.556525,2.2548 1.427735,2.94531 C 93.306757,486.63582 94.476562,487 95.75,487 h 0.75 c 1.755585,0 3.387714,-0.96684 4.53711,-2.0332 0.0781,0.53303 0.28951,1.00434 0.64062,1.35547 C 102.12862,486.77315 102.775,487 103.5,487 a 0.50005,0.50005 0 1 0 0,-1 c -0.525,0 -0.87862,-0.14815 -1.11523,-0.38477 C 102.14815,485.37862 102,485.025 102,484.5 V 479 c 0,-0.79721 -0.12242,-2.24322 -0.88086,-3.55469 C 100.3607,474.13385 98.896459,473 96.5,473 Z m -0.75,7 H 101 v 3.56055 C 100.04582,484.68473 98.18875,486 96.5,486 h -0.75 c -1.081007,0 -2.03187,-0.31555 -2.693359,-0.83984 -0.66149,-0.52429 -1.048828,-1.23676 -1.048829,-2.16016 10e-7,-0.9234 0.387339,-1.63587 1.048829,-2.16016 C 93.718131,480.31555 94.668993,480 95.75,480 Z" id="path10278-2" inkscape:connector-curvature="0"/>
     <g style="display:inline;opacity:1;fill:#ffffff;enable-background:new" id="g12310-0" transform="translate(62.999998,210)" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
       <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" d="m 222,431 c -0.1326,2e-5 -0.25976,0.0527 -0.35352,0.14648 L 218.79297,434 H 216.5 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 7 c 3e-5,0.27613 0.22387,0.49997 0.5,0.5 h 2.29297 l 2.85351,2.85352 c 0.0938,0.0938 0.22092,0.14646 0.35352,0.14648 h 0.5 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 V 431.78125 431.5 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 z m 6.4668,2.01562 a 0.50005,0.50005 0 0 0 -0.42188,0.72071 c 1.27607,2.70648 1.27328,5.84072 -0.008,8.54492 a 0.50017783,0.50017783 0 1 0 0.9043,0.42773 c 1.40867,-2.97397 1.41099,-6.42391 0.008,-9.40039 a 0.50005,0.50005 0 0 0 -0.48242,-0.29297 z m -2.86133,0.9375 a 0.50005,0.50005 0 0 0 -0.43359,0.74219 c 1.10286,2.06013 1.1045,4.5334 0.006,6.59571 a 0.50023236,0.50023236 0 1 0 0.88282,0.4707 c 1.25518,-2.35583 1.25221,-5.18378 -0.008,-7.53711 a 0.50005,0.50005 0 0 0 -0.44726,-0.27149 z" transform="translate(-62.999998,-210)" id="path12304-9" inkscape:connector-curvature="0"/>
@@ -3543,19 +3535,9 @@
     <g style="display:inline;fill:#ffffff;enable-background:new" id="g13226" transform="translate(-189,84.000005)">
       <path style="color:#000000;display:inline;overflow:visible;visibility:visible;opacity:1;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;marker:none;enable-background:accumulate" d="m 285,536 -0.15625,1.90625 c -0.6231,0.14227 -1.07677,0.25145 -1.59375,0.59375 l -1.75,-1.5 -1.5,1.5 1.5,1.75 c -0.34229,0.51699 -0.45148,0.97065 -0.59375,1.59375 L 279,542 v 1 1 l 1.90625,0.15625 c 0.14227,0.6231 0.25145,1.07677 0.59375,1.59375 l -1.5,1.75 1.5,1.5 1.75,-1.5 c 0.51699,0.34229 0.97065,0.45148 1.59375,0.59375 L 285,550 h 1 1 l 0.15625,-1.90625 c 0.6231,-0.14227 1.07677,-0.25145 1.59375,-0.59375 l 1.75,1.5 1.5,-1.5 -1.5,-1.75 c 0.34229,-0.51699 0.45148,-0.97065 0.59375,-1.59375 L 293,544 v -1 -1 l -1.90625,-0.15625 C 290.95148,541.22065 290.8423,540.76698 290.5,540.25 l 1.5,-1.75 -1.5,-1.5 -1.75,1.5 c -0.51699,-0.34229 -0.97065,-0.45148 -1.59375,-0.59375 L 287,536 h -1 z m 1,5 c 1.11641,0 2,0.88359 2,2 0,1.11641 -0.88359,2 -2,2 -1.11641,0 -2,-0.88359 -2,-2 0,-1.11641 0.88359,-2 2,-2 z" transform="translate(189,-84.000005)" id="path52982-3" inkscape:connector-curvature="0"/>
     </g>
-    <g id="g6507" style="fill:#ffffff">
-      <g inkscape:export-ydpi="96" inkscape:export-xdpi="96" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" style="display:inline;opacity:1;fill:#ffffff;enable-background:new" id="g13642" transform="translate(-405.99706,-183.50454)">
-        <path sodipodi:nodetypes="sssssccccccccccccc" inkscape:connector-curvature="0" id="path13640" d="m 695.49706,256.50454 c -1.933,0 -3.5,1.567 -3.5,3.5 0,1.933 1.567,3.5 3.5,3.5 1.933,0 3.5,-1.567 3.5,-3.5 0,-1.933 -1.567,-3.5 -3.5,-3.5 z m -0.5,1 h 1 v 2 h 2 v 1 h -2 v 2 h -1 v -2 h -2 v -1 h 2 z" style="display:inline;opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:0.80000001;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;enable-background:new"/>
-      </g>
-      <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:0.6;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1px;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="m 279.5,76 c -0.27613,2.8e-5 -0.49997,0.223869 -0.5,0.5 v 10 c 0,0.41667 0.19292,0.774946 0.45898,1.041016 C 279.72505,87.807076 280.08333,88 280.5,88 h 11 c 0.41667,0 0.77495,-0.192924 1.04102,-0.458984 C 292.80708,87.274946 293,86.91667 293,86.5 v -5 c -3e-5,-0.276131 -0.22387,-0.499972 -0.5,-0.5 h -11 c -0.27613,2.8e-5 -0.49997,0.223869 -0.5,0.5 v 4 c -0.01,0.676161 1.00956,0.676161 1,0 V 82 h 10 v 4.5 c 0,0.0833 -0.0571,0.225054 -0.16602,0.333984 C 291.72505,86.942924 291.58333,87 291.5,87 h -11 c -0.0833,0 -0.22505,-0.0571 -0.33398,-0.166016 C 280.05708,86.725054 280,86.58333 280,86.5 V 78 h 4.49609 c 0.67451,0.01322 0.68245,-1.002674 0.008,-1 L 284,76.996094 V 76.5 c -3e-5,-0.276131 -0.22387,-0.499972 -0.5,-0.5 z" id="path13699" inkscape:connector-curvature="0" sodipodi:nodetypes="ccscsscsccccccccssssssccccccc"/>
-    </g>
     <g style="display:inline;opacity:1;fill:#ffffff;stroke-width:1.14285719;enable-background:new" id="g13938" transform="matrix(0.875,0,0,0.875,-254.625,-304.87495)" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
       <rect style="display:inline;overflow:visible;visibility:visible;opacity:0;fill:#ffffff;stroke:none;stroke-width:3.42857146;marker:none;enable-background:accumulate" id="rect13853" width="16" height="16" x="299" y="408.99994"/>
     </g>
-    <g id="g13932" transform="translate(-357,189)" style="display:inline;opacity:1;fill:#ffffff;enable-background:new" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
-      <rect style="display:inline;opacity:0;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;enable-background:new" id="rect13928" width="16" height="16" x="446" y="346" rx="0" ry="0"/>
-      <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1px;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="m 447.5,348 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 10 1 c 0,0.41667 0.19292,0.77495 0.45898,1.04102 C 447.72505,360.80708 448.08333,361 448.5,361 h 0.5 10.5 c 0.41667,0 0.77495,-0.19292 1.04102,-0.45898 C 460.80708,360.27495 461,359.91667 461,359.5 v -6 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 h -3 -7 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 5.5 c 0,0.20833 -0.11958,0.47505 -0.32227,0.67773 C 449.47505,359.88042 449.20833,360 449,360 h -0.5 c -0.0833,0 -0.22505,-0.0571 -0.33398,-0.16602 C 448.05708,359.72505 448,359.58333 448,359.5 V 359 h 0.5 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 V 352 h 8.5 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 v -2 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 L 452,348.992 V 348.5 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 z" id="path13930" inkscape:connector-curvature="0" sodipodi:nodetypes="cccscscscscccccscsssscccccccccccc"/>
-    </g>
     <g transform="translate(-1.8536743e-6,4.4999696e-6)" style="display:inline;fill:#ffffff;enable-background:new" id="g14058" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
       <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:evenodd;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="m 451.5,536 a 0.50005,0.50005 0 0 0 -0.5,0.5 v 9 a 0.50005,0.50005 0 0 0 0.5,0.5 h 9 a 0.50005,0.50005 0 0 0 0.5,-0.5 v -9 a 0.50005,0.50005 0 0 0 -0.5,-0.5 z m 0.5,1 h 8 v 8 h -8 z" id="path14208" inkscape:connector-curvature="0"/>
       <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:0.6;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:evenodd;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="M 449.49219,537.99219 A 0.50005,0.50005 0 0 0 449,538.5 v 9 a 0.50005,0.50005 0 0 0 0.5,0.5 h 9 a 0.50005,0.50005 0 1 0 0,-1 H 450 v -8.5 a 0.50005,0.50005 0 0 0 -0.50781,-0.50781 z m -2,2 A 0.50005,0.50005 0 0 0 447,540.5 v 9 a 0.50005,0.50005 0 0 0 0.5,0.5 h 9 a 0.50005,0.50005 0 1 0 0,-1 H 448 v -8.5 a 0.50005,0.50005 0 0 0 -0.50781,-0.50781 z" id="path14212" inkscape:connector-curvature="0"/>
@@ -5284,12 +5266,64 @@
         <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;paint-order:markers stroke fill;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" d="m 226.5,-289 a 0.50005,0.50005 0 0 0 -0.5,0.5 v 13 a 0.50005,0.50005 0 0 0 0.5,0.5 h 3 a 0.50005,0.50005 0 0 0 0.5,-0.5 v -13 a 0.50005,0.50005 0 0 0 -0.5,-0.5 z m 0.5,1 h 2 v 12 h -2 z" id="path10532" inkscape:connector-curvature="0"/>
       </g>
     </g>
-    <path
-       inkscape:connector-curvature="0"
-       id="path5001"
-       d="m 408.48148,452.0625 c -1.71014,0.0728 -3.32227,1.4838 -3.32227,3.9375 0,2.05278 1.07076,4.01178 2.38672,5.71289 1.31595,1.70111 2.90024,3.15481 4.03125,4.16016 0.0914,0.0816 0.20954,0.12673 0.33203,0.12695 h 0.5 c 0.12249,-2.2e-4 0.24064,-0.0454 0.33203,-0.12695 1.13101,-1.00535 2.7153,-2.45905 4.03125,-4.16016 1.31596,-1.70111 2.38672,-3.66011 2.38672,-5.71289 0,-2.4537 -1.61213,-3.86473 -3.32227,-3.9375 -1.52139,-0.0647 -3.0532,0.927 -3.67773,2.69727 -0.62453,-1.77027 -2.15634,-2.76201 -3.67773,-2.69727 z"
-       style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"
-       sodipodi:nodetypes="csccccccsccc" />
+    <g style="display:inline;enable-background:new" id="g36486" transform="translate(-63,-21)">
+      <g transform="translate(-342.99706,-162.50454)" id="g36382" style="display:inline;opacity:1;enable-background:new" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
+        <path style="display:inline;opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:0.80000001;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;enable-background:new" d="m 695.49706,256.50454 c -1.933,0 -3.5,1.567 -3.5,3.5 0,1.933 1.567,3.5 3.5,3.5 1.933,0 3.5,-1.567 3.5,-3.5 0,-1.933 -1.567,-3.5 -3.5,-3.5 z m -0.5,1 h 1 v 2 h 2 v 1 h -2 v 2 h -1 v -2 h -2 v -1 h 2 z" id="path36380" inkscape:connector-curvature="0" sodipodi:nodetypes="sssssccccccccccccc"/>
+      </g>
+      <g id="g36393" style="display:inline;enable-background:new" transform="translate(-587.00005,105)">
+        <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:0.6;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="m 279.5,77 c -0.27613,2.8e-5 -0.49997,0.223869 -0.5,0.5 v 10 c 3e-5,0.27613 0.22387,0.49997 0.5,0.5 h 11 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 v -6 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 H 280 v -3 h 2 v 0.5 c 3e-5,0.276131 0.22387,0.499972 0.5,0.5 h 3.26562 c -0.20665,-0.306775 -0.36146,-0.646717 -0.48828,-1 H 283 v -0.5 c -3e-5,-0.276131 -0.22387,-0.499972 -0.5,-0.5 z m 0.5,5 h 10 v 5 h -10 z" transform="translate(650.00005,-84)" id="path36378" inkscape:connector-curvature="0" sodipodi:nodetypes="cccccccccccccccccccccccc"/>
+      </g>
+    </g>
+    <g style="display:inline;enable-background:new" id="g36374-6" transform="translate(-881.00006,546)">
+      <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="m 90.5,536 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 13 c 2.8e-5,0.27613 0.223869,0.49997 0.5,0.5 h 13 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 v -11 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 H 95 v -1.5 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 z m 0.5,5 h 12 v 8 H 91 Z" transform="translate(881.00006,-546)" id="rect36366-6" inkscape:connector-curvature="0" sodipodi:nodetypes="ccccccccccccccccc"/>
+    </g>
+    <path style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" d="m 363.5,74 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 13 c 3e-5,0.27613 0.22387,0.49997 0.5,0.5 h 13 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 v -11 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 H 368 v -1.5 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 z m 0.5,5 h 12 v 8 h -12 z" id="rect36366-6-8" inkscape:connector-curvature="0" sodipodi:nodetypes="ccccccccccccccccc"/>
+    <g style="display:inline;enable-background:new" id="g28096">
+      <path sodipodi:nodetypes="cccccccccccccccccccccccccccccccccccccccccccccccccccccccssccccccsssc" inkscape:connector-curvature="0" id="path27877" d="m 406.5,53 c -0.27613,2.8e-5 -0.49997,0.223869 -0.5,0.5 v 11 c 3e-5,0.276131 0.22387,0.499972 0.5,0.5 h 4.5 v -1 h -4 V 54 h 5 v 1 h 2 v -1 h 5 v 10 h -2 v 1 h 2.5 c 0.27613,-2.8e-5 0.49997,-0.223869 0.5,-0.5 v -11 c -3e-5,-0.276131 -0.22387,-0.499972 -0.5,-0.5 z m 7.5,2 v 1 h 2 v -1 z m 0,1 h -2 v 1 h 2 z m 0,1 v 1 h 2 v -1 z m 0,1 h -2 v 1 h 2 z m 0,1 v 1 h 2 v -1 z m 0,1 h -2 v 1 h 2 z m 0,1 v 1 h -1.5 c -0.27613,2.8e-5 -0.49997,0.223869 -0.5,0.5 V 65 c 0,1.099351 0.90065,2 2,2 1.09935,0 2,-0.900649 2,-2 v -4 z m -1,2 h 1 1 v 2 c 0,0.562649 -0.43735,1 -1,1 -0.56265,0 -1,-0.437351 -1,-1 z" style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"/>
+      <g transform="translate(-80)" id="g27713" style="display:inline;enable-background:new">
+        <path inkscape:connector-curvature="0" d="m 618,63 h 1 v 1.999999 h -1 z m -2,0 h 1 v 1.999999 h -1 z m -2,0 h 1 v 1.999999 h -1 z M 613.53711,53 C 612.69436,53 612,53.694362 612,54.537109 v 8 2.925782 C 612,66.305638 612.69436,67 613.53711,67 h 8.92578 c 0.73741,0 1.36143,-0.531036 1.50586,-1.228516 0.0206,-0.09964 0.0312,-0.20325 0.0312,-0.308593 v -2.925782 -8 C 624,53.694362 623.30564,53 622.46289,53 Z m 0,9 h 8.92578 C 622.76894,62 623,62.231065 623,62.537109 v 2.925782 C 623,65.768935 622.76894,66 622.46289,66 h -8.92578 C 613.23106,66 613,65.768935 613,65.462891 V 62.537109 C 613,62.231065 613.23106,62 613.53711,62 Z" style="display:inline;opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new" id="rect27779"/>
+      </g>
+      <path inkscape:connector-curvature="0" d="m 489.5,45 a 0.50004994,0.50004994 0 0 0 -0.0996,0.01172 0.50004994,0.50004994 0 0 0 -0.0937,0.03125 0.50004994,0.50004994 0 0 0 -0.0879,0.04883 0.50004994,0.50004994 0 0 0 -0.0391,0.03125 0.50004994,0.50004994 0 0 0 -0.0371,0.0332 0.50004994,0.50004994 0 0 0 -0.0312,0.03906 0.50004994,0.50004994 0 0 0 -0.0293,0.04102 0.50004994,0.50004994 0 0 0 -0.0234,0.04297 0.50004994,0.50004994 0 0 0 -0.0195,0.04687 0.50004994,0.50004994 0 0 0 -0.0156,0.04687 0.50004994,0.50004994 0 0 0 -0.01,0.04883 0.50004994,0.50004994 0 0 0 -0.002,0.150391 0.50004994,0.50004994 0 0 0 0.01,0.04883 0.50004994,0.50004994 0 0 0 0.0156,0.04687 0.50004994,0.50004994 0 0 0 0.0176,0.04687 0.50004994,0.50004994 0 0 0 0.0234,0.04297 0.50004994,0.50004994 0 0 0 0.0293,0.04297 0.50004994,0.50004994 0 0 0 0.0664,0.07227 0.50004994,0.50004994 0 0 0 0.26953,0.123047 A 0.50004994,0.50004994 0 0 0 489.5,46 h 13 a 0.50004994,0.50004994 0 0 0 0.0996,-0.01172 0.50004994,0.50004994 0 0 0 0.0469,-0.01367 0.50004994,0.50004994 0 0 0 0.0469,-0.01758 0.50004994,0.50004994 0 0 0 0.0449,-0.02148 0.50004994,0.50004994 0 0 0 0.082,-0.05859 0.50004994,0.50004994 0 0 0 0.0371,-0.0332 0.50004994,0.50004994 0 0 0 0.0312,-0.03906 0.50004994,0.50004994 0 0 0 0.0723,-0.13086 0.50004994,0.50004994 0 0 0 0.0156,-0.04687 0.50004994,0.50004994 0 0 0 0.0156,-0.09961 0.50004994,0.50004994 0 0 0 0,-0.04883 0.50004994,0.50004994 0 0 0 -0.0137,-0.09961 0.50004994,0.50004994 0 0 0 -0.0156,-0.04687 0.50004994,0.50004994 0 0 0 -0.0176,-0.04687 0.50004994,0.50004994 0 0 0 -0.0527,-0.08594 0.50004994,0.50004994 0 0 0 -0.0312,-0.03711 0.50004994,0.50004994 0 0 0 -0.0742,-0.06641 0.50004994,0.50004994 0 0 0 -0.043,-0.02734 0.50004994,0.50004994 0 0 0 -0.043,-0.02344 0.50004994,0.50004994 0 0 0 -0.0469,-0.01758 0.50004994,0.50004994 0 0 0 -0.0488,-0.01367 0.50004994,0.50004994 0 0 0 -0.0488,-0.0098 A 0.50004994,0.50004994 0 0 0 502.5,45 Z m 4.5,-3.5 c -3e-5,0.276131 -0.22387,0.499972 -0.5,0.5 h -2 c -0.27613,-2.8e-5 -0.49997,-0.223869 -0.5,-0.5 v -2 c 3e-5,-0.276131 0.22387,-0.499972 0.5,-0.5 h 2 c 0.27613,2.8e-5 0.49997,0.223869 0.5,0.5 z M 493,41 v -1 h -1 v 1 z m 5,-3.5 c -3e-5,0.276131 -0.22387,0.499972 -0.5,0.5 h -2 c -0.27613,-2.8e-5 -0.49997,-0.223869 -0.5,-0.5 v -2 c 3e-5,-0.276131 0.22387,-0.499972 0.5,-0.5 h 2 c 0.27613,2.8e-5 0.49997,0.223869 0.5,0.5 z m -4,0 c -3e-5,0.276131 -0.22387,0.499972 -0.5,0.5 h -2 c -0.27613,-2.8e-5 -0.49997,-0.223869 -0.5,-0.5 v -2 c 3e-5,-0.276131 0.22387,-0.499972 0.5,-0.5 h 2 c 0.27613,2.8e-5 0.49997,0.223869 0.5,0.5 z M 489.5,33 a 0.50004997,0.50004997 0 0 0 -0.5,0.5 v 10 a 0.50004997,0.50004997 0 0 0 0.5,0.5 h 13 a 0.50004997,0.50004997 0 0 0 0.5,-0.5 v -10 A 0.50004997,0.50004997 0 0 0 502.5,33 Z m 0.5,1 h 12 v 9 h -12 z" style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:0.99999988;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" id="path27891"/>
+      <g transform="translate(-123,-21)" id="g27886">
+        <g transform="translate(21,-21)" id="g28033" style="display:inline;enable-background:new">
+          <path inkscape:connector-curvature="0" id="path28027" transform="translate(102,41)" d="M 516.03711,33 C 514.91694,33 514,33.916938 514,35.037109 V 36 c -0.0409,0.706384 1.0409,0.706384 1,0 V 35.037109 C 515,34.454485 515.45448,34 516.03711,34 h 4.92578 C 521.54552,34 522,34.454485 522,35.037109 v 5.058594 c -0.58287,0.214917 -1,0.784943 -1,1.441406 v 1.925782 c 3e-5,0.276131 0.22387,0.499972 0.5,0.5 H 522 V 46 h 1 v -2.037109 h 0.5 c 0.27613,-2.8e-5 0.49997,-0.223869 0.5,-0.5 v -1.925782 c 0,-0.656467 -0.41713,-1.22649 -1,-1.441406 V 35.037109 C 523,33.916938 522.08306,33 520.96289,33 Z" style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1.00157475;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"/>
+        </g>
+        <g id="g28053-2" style="display:inline;enable-background:new">
+          <path d="m 616,85 h 1 v 0.999999 h -1 z m -2,0 h 1 v 0.999999 h -1 z m -0.46289,-8 C 612.69436,77 612,77.694362 612,78.537109 v 7.925782 C 612,87.305638 612.69436,88 613.53711,88 h 5.92578 c 0.73741,0 1.36143,-0.531036 1.50586,-1.228516 0.0206,-0.09964 0.0312,-0.20325 0.0312,-0.308593 V 78.537109 C 621,77.694362 620.30564,77 619.46289,77 Z m 0,7 h 5.92578 C 619.76894,84 620,84.231065 620,84.537109 v 1.925782 C 620,86.768935 619.76894,87 619.46289,87 h -5.92578 C 613.23106,87 613,86.768935 613,86.462891 V 84.537109 C 613,84.231065 613.23106,84 613.53711,84 Z" style="display:inline;opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new" transform="translate(21,-21)" id="rect28051-4" inkscape:connector-curvature="0"/>
+        </g>
+      </g>
+      <g transform="translate(-123,-21)" id="g27900">
+        <g id="g28156" transform="translate(0,-21)">
+          <path inkscape:connector-curvature="0" id="circle28098" transform="translate(123,41)" d="m 543.5,33 c -0.8225,0 -1.5,0.677492 -1.5,1.5 0,0.230167 0.0579,0.446791 0.15234,0.642578 a 0.50005,0.50005 0 0 0 -0.006,0.0039 L 540.29297,37 H 539.5 a 0.50005,0.50005 0 1 0 0,1 h 1 a 0.50005,0.50005 0 0 0 0.35352,-0.146484 l 2,-2 a 0.50005,0.50005 0 0 0 0.004,-0.0059 C 543.05321,35.942094 543.26984,36 543.5,36 c 0.8225,0 1.5,-0.677492 1.5,-1.5 0,-0.822508 -0.6775,-1.5 -1.5,-1.5 z m 0,1 c 0.28206,0 0.5,0.217938 0.5,0.5 0,0.282062 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.217938 -0.5,-0.5 0,-0.282062 0.21794,-0.5 0.5,-0.5 z m 0,5 c -0.64684,0 -1.19777,0.420971 -1.40625,1 H 539.5 a 0.50005,0.50005 0 1 0 0,1 h 2.59375 c 0.20848,0.579029 0.75941,1 1.40625,1 0.8225,0 1.5,-0.677492 1.5,-1.5 0,-0.822508 -0.6775,-1.5 -1.5,-1.5 z m 0,1 c 0.28206,0 0.5,0.217938 0.5,0.5 0,0.282062 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.217938 -0.5,-0.5 0,-0.282062 0.21794,-0.5 0.5,-0.5 z m -4,2.992188 a 0.50005,0.50005 0 1 0 0,1 h 0.79297 l 1.85351,1.853515 a 0.50005,0.50005 0 0 0 0.008,0.0078 C 542.05866,46.050309 542,46.268282 542,46.5 c 0,0.822508 0.6775,1.5 1.5,1.5 0.8225,0 1.5,-0.677492 1.5,-1.5 0,-0.822508 -0.6775,-1.5 -1.5,-1.5 -0.22767,0 -0.44256,0.05591 -0.63672,0.148438 a 0.50005,0.50005 0 0 0 -0.01,-0.0098 l -2,-2 A 0.50005,0.50005 0 0 0 540.5,42.992188 Z M 543.5,46 c 0.28206,0 0.5,0.217938 0.5,0.5 0,0.282062 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.217938 -0.5,-0.5 0,-0.282062 0.21794,-0.5 0.5,-0.5 z" style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"/>
+        </g>
+        <g transform="translate(21)" id="g28053">
+          <path inkscape:connector-curvature="0" d="m 616,85 h 1 v 0.999999 h -1 z m -2,0 h 1 v 0.999999 h -1 z m -0.46289,-8 C 612.69436,77 612,77.694362 612,78.537109 v 7.925782 C 612,87.305638 612.69436,88 613.53711,88 h 5.92578 c 0.73741,0 1.36143,-0.531036 1.50586,-1.228516 0.0206,-0.09964 0.0312,-0.20325 0.0312,-0.308593 V 78.537109 C 621,77.694362 620.30564,77 619.46289,77 Z m 0,7 h 5.92578 C 619.76894,84 620,84.231065 620,84.537109 v 1.925782 C 620,86.768935 619.76894,87 619.46289,87 h -5.92578 C 613.23106,87 613,86.768935 613,86.462891 V 84.537109 C 613,84.231065 613.23106,84 613.53711,84 Z" style="display:inline;opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new" transform="translate(21,-21)" id="rect28051"/>
+        </g>
+      </g>
+      <path id="circle28282" d="m 511.5,452 c -0.8225,0 -1.5,0.6775 -1.5,1.5 0,0.82251 0.6775,1.5 1.5,1.5 0.64684,0 1.19777,-0.42097 1.40625,-1 h 1.1875 c 0.20848,0.57903 0.75941,1 1.40625,1 0.8225,0 1.5,-0.67749 1.5,-1.5 0,-0.8225 -0.6775,-1.5 -1.5,-1.5 -0.64637,0 -1.19742,0.42162 -1.40625,1 h -1.1875 c -0.20883,-0.57838 -0.75988,-1 -1.40625,-1 z m 11,0 c -0.64637,0 -1.19742,0.42162 -1.40625,1 H 520.5 a 0.50005,0.50005 0 0 0 -0.35352,0.14648 L 516.29297,457 h -3.38672 c -0.20883,-0.57838 -0.75988,-1 -1.40625,-1 -0.8225,0 -1.5,0.6775 -1.5,1.5 0,0.82251 0.6775,1.5 1.5,1.5 0.64684,0 1.19777,-0.42097 1.40625,-1 H 516.5 a 0.50005,0.50005 0 0 0 0.35352,-0.14648 L 520.70703,454 h 0.38672 c 0.20848,0.57903 0.75941,1 1.40625,1 0.8225,0 1.5,-0.67749 1.5,-1.5 0,-0.8225 -0.6775,-1.5 -1.5,-1.5 z m -11,1 c 0.27091,0 0.47782,0.20294 0.49414,0.46875 a 0.50005,0.50005 0 0 0 0,0.0625 C 511.97782,453.79707 511.77091,454 511.5,454 c -0.28206,0 -0.5,-0.21793 -0.5,-0.5 0,-0.28206 0.21794,-0.5 0.5,-0.5 z m 4,0 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28207 -0.21794,0.5 -0.5,0.5 -0.27091,0 -0.47782,-0.20293 -0.49414,-0.46875 a 0.50005,0.50005 0 0 0 0,-0.0625 C 515.02218,453.20294 515.22909,453 515.5,453 Z m 7,0 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28207 -0.21794,0.5 -0.5,0.5 -0.27091,0 -0.47782,-0.20293 -0.49414,-0.46875 a 0.50005,0.50005 0 0 0 0,-0.0625 C 522.02218,453.20294 522.22909,453 522.5,453 Z m 0,3 c -0.64637,0 -1.19742,0.42162 -1.40625,1 H 520.5 a 0.50005,0.50005 0 0 0 -0.35352,0.14648 l -5,5 A 0.50005,0.50005 0 0 0 515,462.5 v 0.59375 c -0.57903,0.20848 -1,0.75941 -1,1.40625 0,0.82251 0.6775,1.5 1.5,1.5 0.8225,0 1.5,-0.67749 1.5,-1.5 0,-0.64684 -0.42097,-1.19777 -1,-1.40625 v -0.38672 L 520.70703,458 h 0.38672 c 0.20848,0.57903 0.75941,1 1.40625,1 0.8225,0 1.5,-0.67749 1.5,-1.5 0,-0.8225 -0.6775,-1.5 -1.5,-1.5 z m -11,1 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28207 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21793 -0.5,-0.5 0,-0.28206 0.21794,-0.5 0.5,-0.5 z m 11,0 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28207 -0.21794,0.5 -0.5,0.5 -0.27091,0 -0.47782,-0.20293 -0.49414,-0.46875 a 0.50005,0.50005 0 0 0 0,-0.0625 C 522.02218,457.20294 522.22909,457 522.5,457 Z m -11,3 c -0.8225,0 -1.5,0.6775 -1.5,1.5 0,0.82251 0.6775,1.5 1.5,1.5 0.8225,0 1.5,-0.67749 1.5,-1.5 0,-0.8225 -0.6775,-1.5 -1.5,-1.5 z m 11,0 c -0.8225,0 -1.5,0.6775 -1.5,1.5 0,0.23017 0.0579,0.44679 0.15234,0.64258 a 0.50005,0.50005 0 0 0 -0.006,0.004 l -1,1 a 0.50005,0.50005 0 0 0 -0.004,0.006 C 519.94679,463.05791 519.73016,463 519.5,463 c -0.8225,0 -1.5,0.6775 -1.5,1.5 0,0.82251 0.6775,1.5 1.5,1.5 0.8225,0 1.5,-0.67749 1.5,-1.5 0,-0.23002 -0.058,-0.44493 -0.15234,-0.64062 a 0.50005,0.50005 0 0 0 0.006,-0.006 l 1,-1 a 0.50005,0.50005 0 0 0 0.004,-0.004 c 0.1955,0.0946 0.41225,0.15062 0.64234,0.15062 0.8225,0 1.5,-0.67749 1.5,-1.5 0,-0.8225 -0.6775,-1.5 -1.5,-1.5 z m -11,1 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28207 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21793 -0.5,-0.5 0,-0.28206 0.21794,-0.5 0.5,-0.5 z m 11,0 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28207 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21793 -0.5,-0.5 0,-0.28206 0.21794,-0.5 0.5,-0.5 z m -3,3 c 0.28206,0 0.5,0.21794 0.5,0.5 0,0.28207 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21793 -0.5,-0.5 0,-0.28206 0.21794,-0.5 0.5,-0.5 z m -4.0293,0.006 a 0.50005,0.50005 0 0 0 0.0371,0 0.50005,0.50005 0 0 0 0.0176,0 c 0.26874,0.0134 0.4746,0.22107 0.4746,0.494 0,0.28207 -0.21794,0.5 -0.5,0.5 -0.28206,0 -0.5,-0.21793 -0.5,-0.5 0,-0.27158 0.20391,-0.47876 0.4707,-0.49414 z" style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" inkscape:connector-curvature="0"/>
+    </g>
+    <g transform="translate(-4.4569653e-6,1.6485533e-6)" style="display:inline;enable-background:new" id="g27998">
+      <g transform="translate(42,-84)" id="g27867">
+        <path sodipodi:nodetypes="cccccccccccccccccc" inkscape:connector-curvature="0" id="path27865" d="m 114.5,97 c -0.27613,2.8e-5 -0.49997,0.223869 -0.5,0.5 v 3 c 3e-5,0.27613 0.22387,0.49997 0.5,0.5 h 10 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 v -3 c -3e-5,-0.276131 -0.22387,-0.499972 -0.5,-0.5 z m -3,6 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 3 c 3e-5,0.27613 0.22387,0.49997 0.5,0.5 h 10 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 v -3 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 z" style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"/>
+      </g>
+      <g transform="translate(-42,-84)" id="g28017">
+        <g transform="translate(-357.00711,41.992878)" id="g13408-5" style="display:inline;opacity:1;enable-background:new" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
+          <g id="g12509-8" transform="translate(378.99999,-439.99995)" style="display:inline;opacity:0.98999999;enable-background:new"/>
+          <path inkscape:connector-curvature="0" id="path27987" d="m 531.49219,52.992188 a 0.50005,0.50005 0 0 0 -0.31641,0.121093 0.50005,0.50005 0 0 0 -0.0195,0.01563 0.50005,0.50005 0 0 0 -0.0176,0.01758 0.50005,0.50005 0 0 0 -0.002,0.0039 0.50005,0.50005 0 0 0 -0.0312,0.0332 0.50005,0.50005 0 0 0 -0.002,0.0039 0.50005,0.50005 0 0 0 -0.0273,0.03711 0.50005,0.50005 0 0 0 -0.002,0.0039 A 0.50005,0.50005 0 0 0 531,53.582031 V 54.5 a 0.50005,0.50005 0 1 0 1,0 V 54 h 0.5 a 0.50005,0.50005 0 1 0 0,-1 h -0.91992 a 0.50005,0.50005 0 0 0 -0.0879,-0.0078 z m 13,0 A 0.50005,0.50005 0 0 0 544.41797,53 H 543.5 a 0.50005,0.50005 0 1 0 0,1 h 0.5 v 0.5 a 0.50005,0.50005 0 1 0 1,0 v -0.919922 a 0.50005,0.50005 0 0 0 -0.11328,-0.404297 0.50005,0.50005 0 0 0 -0.002,-0.0039 0.50005,0.50005 0 0 0 -0.0312,-0.0332 0.50005,0.50005 0 0 0 -0.004,-0.002 0.50005,0.50005 0 0 0 -0.0332,-0.03125 0.50005,0.50005 0 0 0 -0.004,-0.002 0.50005,0.50005 0 0 0 -0.32031,-0.111328 z M 534.5,53 a 0.50005,0.50005 0 1 0 0,1 h 1 a 0.50005,0.50005 0 1 0 0,-1 z m 3,0 a 0.50005,0.50005 0 1 0 0,1 h 1 a 0.50005,0.50005 0 1 0 0,-1 z m 3,0 a 0.50005,0.50005 0 1 0 0,1 h 1 a 0.50005,0.50005 0 1 0 0,-1 z m -9.00781,2.992188 A 0.50005,0.50005 0 0 0 531,56.5 v 1 a 0.50005,0.50005 0 1 0 1,0 v -1 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z m 13,0 A 0.50005,0.50005 0 0 0 544,56.5 v 1 a 0.50005,0.50005 0 1 0 1,0 v -1 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z m -13,3 A 0.50005,0.50005 0 0 0 531,59.5 v 1 a 0.50005,0.50005 0 1 0 1,0 v -1 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z m 13,0 A 0.50005,0.50005 0 0 0 544,59.5 v 1 a 0.50005,0.50005 0 1 0 1,0 v -1 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z m -13,3 A 0.50005,0.50005 0 0 0 531,62.5 v 1 a 0.50005,0.50005 0 1 0 1,0 v -1 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z m 13,0 A 0.50005,0.50005 0 0 0 544,62.5 v 1 a 0.50005,0.50005 0 1 0 1,0 v -1 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z m -13,3 A 0.50005,0.50005 0 0 0 531,65.5 v 0.919922 a 0.50005,0.50005 0 0 0 0.11328,0.404297 0.50005,0.50005 0 0 0 0.0156,0.01953 0.50005,0.50005 0 0 0 0.0176,0.01758 0.50005,0.50005 0 0 0 0.004,0.002 0.50005,0.50005 0 0 0 0.0332,0.03125 0.50005,0.50005 0 0 0 0.004,0.002 A 0.50005,0.50005 0 0 0 531.58203,67 H 532.5 a 0.50005,0.50005 0 1 0 0,-1 H 532 v -0.5 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z m 13,0 A 0.50005,0.50005 0 0 0 544,65.5 V 66 h -0.5 a 0.50005,0.50005 0 1 0 0,1 h 0.91992 a 0.50005,0.50005 0 0 0 0.4043,-0.113281 0.50005,0.50005 0 0 0 0.004,-0.002 0.50005,0.50005 0 0 0 0.0332,-0.03125 0.50005,0.50005 0 0 0 0.002,-0.0039 0.50005,0.50005 0 0 0 0.0312,-0.0332 0.50005,0.50005 0 0 0 0.002,-0.0039 A 0.50005,0.50005 0 0 0 545,66.417969 V 65.5 a 0.50005,0.50005 0 0 0 -0.50781,-0.507812 z M 534.5,66 a 0.50005,0.50005 0 1 0 0,1 h 1 a 0.50005,0.50005 0 1 0 0,-1 z m 3,0 a 0.50005,0.50005 0 1 0 0,1 h 1 a 0.50005,0.50005 0 1 0 0,-1 z m 3,0 a 0.50005,0.50005 0 1 0 0,1 h 1 a 0.50005,0.50005 0 1 0 0,-1 z" style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:evenodd;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"/>
+        </g>
+        <g style="display:inline;enable-background:new" id="g27919-7" transform="rotate(-180,149.5,102.00001)">
+          <path sodipodi:nodetypes="cccccccccccccccccc" inkscape:connector-curvature="0" id="path27916-5" d="m 117.5,98 c -0.27613,2.8e-5 -0.49997,0.223869 -0.5,0.5 v 2 c 3e-5,0.27613 0.22387,0.49997 0.5,0.5 h 5 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 v -2 c -3e-5,-0.276131 -0.22387,-0.499972 -0.5,-0.5 z m -4,5 c -0.27613,3e-5 -0.49997,0.22387 -0.5,0.5 v 2 c 3e-5,0.27613 0.22387,0.49997 0.5,0.5 h 7 c 0.27613,-3e-5 0.49997,-0.22387 0.5,-0.5 v -2 c -3e-5,-0.27613 -0.22387,-0.49997 -0.5,-0.5 z" style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"/>
+        </g>
+      </g>
+      <g style="display:inline;opacity:1;enable-background:new" transform="translate(63,-20)" id="g27948" inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png" inkscape:export-xdpi="96" inkscape:export-ydpi="96">
+        <g transform="translate(1,-1)" id="g27945">
+          <g id="g27954">
+            <path inkscape:connector-curvature="0" d="m 237.49414,326.05078 c -0.11724,0.001 -0.22936,0.0482 -0.3125,0.13086 l -2,2 c -0.17593,0.17578 -0.17593,0.46094 0,0.63672 l 2,2 c 0.17578,0.17593 0.46094,0.17593 0.63672,0 l 2,-2 c 0.17593,-0.17578 0.17593,-0.46094 0,-0.63672 l -2,-2 c -0.086,-0.0855 -0.20293,-0.13272 -0.32422,-0.13086 z m 11,0 c -0.11724,0.001 -0.22936,0.0482 -0.3125,0.13086 l -2,2 c -0.17593,0.17578 -0.17593,0.46094 0,0.63672 l 2,2 c 0.17578,0.17593 0.46094,0.17593 0.63672,0 l 2,-2 c 0.17593,-0.17578 0.17593,-0.46094 0,-0.63672 l -2,-2 c -0.086,-0.0855 -0.20293,-0.13272 -0.32422,-0.13086 z m -6.02734,6.75 a 0.45004499,0.45004499 0 0 0 -0.29688,0.14453 l -3,3.25 a 0.45004499,0.45004499 0 0 0 0.0117,0.62305 l 3,3 a 0.45004499,0.45004499 0 0 0 0.63672,0 l 3,-3 a 0.45004499,0.45004499 0 0 0 0.0117,-0.62305 l -3,-3.25 a 0.45004499,0.45004499 0 0 0 -0.36328,-0.14453 z m 0.0332,1.11133 2.37695,2.57617 -2.37695,2.375 -2.37695,-2.37695 z" style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:0.89999998;stroke-linecap:butt;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new" id="path27939"/>
+            <path style="opacity:0.5;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke" d="m 239.12109,326 1.43946,1.43945 c 0.58556,0.5858 0.58556,1.5353 0,2.1211 L 239.12109,331 h 7.75782 l -1.43946,-1.43945 c -0.58556,-0.5858 -0.58556,-1.5353 0,-2.1211 L 246.87891,326 Z" id="path27943" inkscape:connector-curvature="0" sodipodi:nodetypes="ccccccccc"/>
+          </g>
+        </g>
+      </g>
+    </g>
   </g>
   <g inkscape:groupmode="layer" id="layer2" inkscape:label="EMPTY ICON TRACKING" style="display:none">
     <g id="g16331" style="fill:#ffcc00">
diff --git a/release/datafiles/blender_icons16/icon16_con_action.dat b/release/datafiles/blender_icons16/icon16_con_action.dat
new file mode 100644
index 00000000000..48484dea517
--- /dev/null
+++ b/release/datafiles/blender_icons16/icon16_con_action.dat
diff --git a/release/datafiles/blender_icons16/icon16_seq_strip_duplicate.dat b/release/datafiles/blender_icons16/icon16_seq_strip_duplicate.dat
new file mode 100644
index 00000000000..4b1da378bd8
--- /dev/null
+++ b/release/datafiles/blender_icons16/icon16_seq_strip_duplicate.dat
diff --git a/release/datafiles/blender_icons16/icon16_seq_strip_meta.dat b/release/datafiles/blender_icons16/icon16_seq_strip_meta.dat
new file mode 100644
index 00000000000..0202584a6d5
--- /dev/null
+++ b/release/datafiles/blender_icons16/icon16_seq_strip_meta.dat
diff --git a/release/datafiles/blender_icons16/icon16_snap_midpoint.dat b/release/datafiles/blender_icons16/icon16_snap_midpoint.dat
new file mode 100644
index 00000000000..903ab37dc24
--- /dev/null
+++ b/release/datafiles/blender_icons16/icon16_snap_midpoint.dat
diff --git a/release/datafiles/blender_icons16/icon16_snap_perpendicular.dat b/release/datafiles/blender_icons16/icon16_snap_perpendicular.dat
new file mode 100644
index 00000000000..02b86db4396
--- /dev/null
+++ b/release/datafiles/blender_icons16/icon16_snap_perpendicular.dat
diff --git a/release/datafiles/blender_icons32/icon32_con_action.dat b/release/datafiles/blender_icons32/icon32_con_action.dat
new file mode 100644
index 00000000000..3166e5a7f37
--- /dev/null
+++ b/release/datafiles/blender_icons32/icon32_con_action.dat
diff --git a/release/datafiles/blender_icons32/icon32_seq_strip_duplicate.dat b/release/datafiles/blender_icons32/icon32_seq_strip_duplicate.dat
new file mode 100644
index 00000000000..4322d33f19e
--- /dev/null
+++ b/release/datafiles/blender_icons32/icon32_seq_strip_duplicate.dat
diff --git a/release/datafiles/blender_icons32/icon32_seq_strip_meta.dat b/release/datafiles/blender_icons32/icon32_seq_strip_meta.dat
new file mode 100644
index 00000000000..1bb6b3170cd
--- /dev/null
+++ b/release/datafiles/blender_icons32/icon32_seq_strip_meta.dat
diff --git a/release/datafiles/blender_icons32/icon32_snap_midpoint.dat b/release/datafiles/blender_icons32/icon32_snap_midpoint.dat
new file mode 100644
index 00000000000..c22b196035a
--- /dev/null
+++ b/release/datafiles/blender_icons32/icon32_snap_midpoint.dat
diff --git a/release/datafiles/blender_icons32/icon32_snap_perpendicular.dat b/release/datafiles/blender_icons32/icon32_snap_perpendicular.dat
new file mode 100644
index 00000000000..2460b0150ab
--- /dev/null
+++ b/release/datafiles/blender_icons32/icon32_snap_perpendicular.dat
diff --git a/release/datafiles/prvicons.png b/release/datafiles/prvicons.png
index edbffae420b..dbc0f11124e 100644
--- a/release/datafiles/prvicons.png
+++ b/release/datafiles/prvicons.png
diff --git a/release/datafiles/prvicons.svg b/release/datafiles/prvicons.svg
index 63cd8dc1954..80929124251 100644
--- a/release/datafiles/prvicons.svg
+++ b/release/datafiles/prvicons.svg
@@ -10,20 +10,20 @@
    xmlns:xlink="http://www.w3.org/1999/xlink"
    xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
    xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
-   width="192"
-   height="192"
+   width="1536"
+   height="256"
    id="svg2"
    sodipodi:version="0.32"
-   inkscape:version="0.92.2 2405546, 2018-03-11"
+   inkscape:version="0.92.3 (2405546, 2018-03-11)"
    version="1.0"
    sodipodi:docname="prvicons.svg"
    inkscape:output_extension="org.inkscape.output.svg.inkscape"
    style="display:inline;enable-background:new"
-   inkscape:export-filename="blender_icons.png"
-   inkscape:export-xdpi="90"
-   inkscape:export-ydpi="90">
+   inkscape:export-filename="prvicons.png"
+   inkscape:export-xdpi="96"
+   inkscape:export-ydpi="96">
   <metadata
-     id="metadata2373">
+     id="metadata2699">
     <rdf:RDF>
       <cc:Work
          rdf:about="">
@@ -35,7 +35,7 @@
     </rdf:RDF>
   </metadata>
   <sodipodi:namedview
-     pagecolor="#ffffff"
+     pagecolor="#5b5b5b"
      bordercolor="#666666"
      borderopacity="1"
      objecttolerance="10"
@@ -43,17 +43,93 @@
      guidetolerance="10"
      inkscape:pageopacity="0"
      inkscape:pageshadow="2"
-     inkscape:window-width="1074"
-     inkscape:window-height="1896"
-     id="namedview2371"
-     showgrid="false"
-     inkscape:zoom="1.2291667"
-     inkscape:cx="60.20339"
-     inkscape:cy="13.830508"
-     inkscape:window-x="1080"
-     inkscape:window-y="0"
-     inkscape:window-maximized="0"
-     inkscape:current-layer="layer1" />
+     inkscape:window-width="1920"
+     inkscape:window-height="1017"
+     id="namedview2697"
+     showgrid="true"
+     inkscape:snap-bbox="true"
+     inkscape:bbox-paths="true"
+     inkscape:bbox-nodes="true"
+     inkscape:snap-bbox-edge-midpoints="true"
+     inkscape:snap-bbox-midpoints="true"
+     inkscape:zoom="0.70710678"
+     inkscape:cx="752.76338"
+     inkscape:cy="111.39186"
+     inkscape:window-x="1912"
+     inkscape:window-y="-8"
+     inkscape:window-maximized="1"
+     inkscape:current-layer="layer1"
+     inkscape:snap-page="true"
+     inkscape:object-paths="true"
+     showguides="true"
+     inkscape:guide-bbox="true"
+     inkscape:snap-grids="true"
+     inkscape:snap-intersection-paths="true"
+     inkscape:snap-smooth-nodes="true"
+     inkscape:snap-nodes="true"
+     inkscape:snap-global="true"
+     showborder="true"
+     inkscape:showpageshadow="true"
+     inkscape:pagecheckerboard="true">
+    <inkscape:grid
+       type="xygrid"
+       id="grid6640"
+       color="#d3d3d3"
+       opacity="0.1254902"
+       empcolor="#ffffff"
+       empopacity="0.25098039"
+       spacingx="0.5"
+       spacingy="0.5"
+       empspacing="2"
+       originx="116"
+       originy="-514"
+       visible="true"
+       enabled="true" />
+    <inkscape:grid
+       type="xygrid"
+       id="grid5463"
+       enabled="false"
+       empspacing="0"
+       spacingx="256"
+       spacingy="256" />
+    <sodipodi:guide
+       position="256,68"
+       orientation="1,0"
+       id="guide5465"
+       inkscape:locked="false" />
+    <sodipodi:guide
+       position="512,77"
+       orientation="1,0"
+       id="guide5467"
+       inkscape:locked="false" />
+    <sodipodi:guide
+       position="768,27"
+       orientation="1,0"
+       id="guide5469"
+       inkscape:locked="false" />
+    <sodipodi:guide
+       position="1024,5.75"
+       orientation="1,0"
+       id="guide5471"
+       inkscape:locked="false" />
+    <sodipodi:guide
+       position="1280,-2.5"
+       orientation="1,0"
+       id="guide5473"
+       inkscape:locked="false" />
+    <inkscape:grid
+       type="xygrid"
+       id="grid5475"
+       spacingx="32"
+       spacingy="32"
+       enabled="false" />
+    <inkscape:grid
+       type="xygrid"
+       id="grid5500"
+       spacingx="8"
+       spacingy="8"
+       enabled="false" />
+  </sodipodi:namedview>
   <defs
      id="defs4">
     <linearGradient
@@ -703,7 +779,7 @@
        x2="24.25"
        y1="245"
        x1="22.75"
-       gradientTransform="translate(-61,0)"
+       gradientTransform="translate(-61)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23201"
        xlink:href="#linearGradient1610"
@@ -713,7 +789,7 @@
        x2="24.5"
        y1="245"
        x1="22.75"
-       gradientTransform="translate(-1,0)"
+       gradientTransform="translate(-1)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23199"
        xlink:href="#linearGradient1610"
@@ -957,7 +1033,7 @@
        x2="424.75217"
        y1="77.44017"
        x1="441.98615"
-       gradientTransform="translate(0.01387,0)"
+       gradientTransform="translate(0.01387)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient22896"
        xlink:href="#linearGradient319"
@@ -967,7 +1043,7 @@
        x2="424.75217"
        y1="78"
        x1="438.61115"
-       gradientTransform="translate(0.01387,0)"
+       gradientTransform="translate(0.01387)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient22846"
        xlink:href="#linearGradient319"
@@ -977,7 +1053,7 @@
        x2="424.75217"
        y1="77"
        x1="437.98615"
-       gradientTransform="translate(0.01387,0)"
+       gradientTransform="translate(0.01387)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient22844"
        xlink:href="#linearGradient319"
@@ -2342,7 +2418,7 @@
        x2="336.72485"
        y1="143.70836"
        x1="349.04059"
-       gradientTransform="matrix(0.707107,-0.707107,0.707107,0.707107,-241.7085,428.4841)"
+       gradientTransform="rotate(-45,396.37194,506.00979)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient15748"
        xlink:href="#linearGradient37542"
@@ -3952,7 +4028,7 @@
        x2="54.8125"
        y1="500"
        x1="108"
-       gradientTransform="translate(42,0)"
+       gradientTransform="translate(42)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient16638"
        xlink:href="#linearGradient1610"
@@ -3962,7 +4038,7 @@
        x2="26.561054"
        y1="501.74075"
        x1="85.874489"
-       gradientTransform="translate(42,0)"
+       gradientTransform="translate(42)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient16640"
        xlink:href="#linearGradient15809"
@@ -4088,7 +4164,7 @@
        x2="415.94211"
        y1="-34"
        x1="416.00461"
-       gradientTransform="translate(-0.004608,0)"
+       gradientTransform="translate(-0.004608)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient30243"
        xlink:href="#linearGradient24081"
@@ -4098,7 +4174,7 @@
        x2="416.5"
        y1="-29.933779"
        x1="416.5"
-       gradientTransform="translate(-0.004608,0)"
+       gradientTransform="translate(-0.004608)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient30245"
        xlink:href="#linearGradient16500"
@@ -4108,7 +4184,7 @@
        x2="416.46497"
        y1="-34.342831"
        x1="416.41162"
-       gradientTransform="translate(-1,0)"
+       gradientTransform="translate(-1)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient30247"
        xlink:href="#linearGradient20324"
@@ -4190,7 +4266,7 @@
        x2="415.45193"
        y1="-31.506163"
        x1="415.41223"
-       gradientTransform="translate(-0.004608,0)"
+       gradientTransform="translate(-0.004608)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient30340"
        xlink:href="#linearGradient1610"
@@ -4200,7 +4276,7 @@
        x2="416.5"
        y1="-29.933779"
        x1="416.5"
-       gradientTransform="translate(-0.004608,0)"
+       gradientTransform="translate(-0.004608)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient30342"
        xlink:href="#linearGradient16500"
@@ -4210,7 +4286,7 @@
        x2="416.46497"
        y1="-33.8125"
        x1="416.5"
-       gradientTransform="translate(-1,0)"
+       gradientTransform="translate(-1)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient30344"
        xlink:href="#linearGradient16500"
@@ -4301,7 +4377,7 @@
        x2="415.45193"
        y1="-31.506163"
        x1="415.41223"
-       gradientTransform="translate(-0.004608,0)"
+       gradientTransform="translate(-0.004608)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient30408"
        xlink:href="#linearGradient1610"
@@ -4311,7 +4387,7 @@
        x2="416.5"
        y1="-29.933779"
        x1="416.5"
-       gradientTransform="translate(-0.004608,0)"
+       gradientTransform="translate(-0.004608)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient30410"
        xlink:href="#linearGradient16500"
@@ -4321,7 +4397,7 @@
        x2="416.46497"
        y1="-33.8125"
        x1="416.5"
-       gradientTransform="translate(-1,0)"
+       gradientTransform="translate(-1)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient30412"
        xlink:href="#linearGradient16500"
@@ -4558,7 +4634,7 @@
        x2="-42.377892"
        y1="442.6875"
        x1="-51.6875"
-       gradientTransform="translate(18,0)"
+       gradientTransform="translate(18)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient17135"
        xlink:href="#linearGradient1610"
@@ -4883,7 +4959,7 @@
        inkscape:collect="always" />
     <linearGradient
        y2="-32.351803"
-       x2="4.8398785e-16"
+       x2="4.8398785e-016"
        y1="-29.678047"
        x1="4.9341426"
        gradientUnits="userSpaceOnUse"
@@ -5632,7 +5708,7 @@
        x2="476.76578"
        y1="163.08553"
        x1="480.09564"
-       gradientTransform="translate(-6,0)"
+       gradientTransform="translate(-6)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23241"
        xlink:href="#linearGradient19625"
@@ -5681,7 +5757,7 @@
        x2="475.50031"
        y1="163.08553"
        x1="480.09564"
-       gradientTransform="translate(-6,0)"
+       gradientTransform="translate(-6)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23445"
        xlink:href="#linearGradient23974"
@@ -5759,7 +5835,7 @@
        x2="60.155113"
        y1="253.5"
        x1="79.04213"
-       gradientTransform="translate(-21,0)"
+       gradientTransform="translate(-21)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient40967"
        xlink:href="#linearGradient32842"
@@ -6107,7 +6183,7 @@
        x2="476.68781"
        y1="163"
        x1="483.00034"
-       gradientTransform="translate(-6,0)"
+       gradientTransform="translate(-6)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient24543"
        xlink:href="#linearGradient1610"
@@ -6309,7 +6385,7 @@
        x2="284.9375"
        y1="200.66605"
        x1="268.21783"
-       gradientTransform="translate(0.01612278,0)"
+       gradientTransform="translate(0.01612278)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient41172"
        xlink:href="#linearGradient319"
@@ -6361,7 +6437,7 @@
        x2="265.70886"
        y1="215.3125"
        x1="276.39999"
-       gradientTransform="translate(0.01612278,0)"
+       gradientTransform="translate(0.01612278)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient42093"
        xlink:href="#linearGradient23974"
@@ -6402,7 +6478,7 @@
        x2="290.98389"
        y1="195"
        x1="265.98389"
-       gradientTransform="translate(0.01612278,0)"
+       gradientTransform="translate(0.01612278)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient42290"
        xlink:href="#linearGradient319"
@@ -6581,7 +6657,7 @@
        x2="388.78125"
        y1="410"
        x1="387"
-       gradientTransform="translate(-1,0)"
+       gradientTransform="translate(-1)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient22249"
        xlink:href="#linearGradient319"
@@ -6591,7 +6667,7 @@
        x2="389.14081"
        y1="409.84152"
        x1="386.88852"
-       gradientTransform="translate(-1,0)"
+       gradientTransform="translate(-1)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient22251"
        xlink:href="#linearGradient319"
@@ -6601,7 +6677,7 @@
        x2="388.86676"
        y1="409.86362"
        x1="387"
-       gradientTransform="translate(-1,0)"
+       gradientTransform="translate(-1)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient22253"
        xlink:href="#linearGradient319"
@@ -6839,7 +6915,7 @@
        x2="145.5"
        y1="627.5"
        x1="145.5"
-       gradientTransform="matrix(0,1,-1,0,850,297)"
+       gradientTransform="rotate(90,276.5,573.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23177"
        xlink:href="#linearGradient20324"
@@ -6849,7 +6925,7 @@
        x2="145.5"
        y1="627.5"
        x1="145.5"
-       gradientTransform="matrix(-1,0,0,-1,366,1072)"
+       gradientTransform="rotate(180,183,536)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23179"
        xlink:href="#linearGradient20324"
@@ -6859,7 +6935,7 @@
        x2="145.5"
        y1="627.5"
        x1="145.5"
-       gradientTransform="matrix(0,-1,1,0,-409,588)"
+       gradientTransform="rotate(-90,89.5,498.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23181"
        xlink:href="#linearGradient20324"
@@ -6889,7 +6965,7 @@
        x2="150.5"
        y1="647.75"
        x1="150.5"
-       gradientTransform="matrix(0,1,-1,0,870,292)"
+       gradientTransform="rotate(90,289,581)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23187"
        xlink:href="#linearGradient37623"
@@ -6909,7 +6985,7 @@
        x2="150.5"
        y1="647.75"
        x1="150.5"
-       gradientTransform="matrix(0,-1,1,0,-429,593)"
+       gradientTransform="rotate(-90,82,511)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23191"
        xlink:href="#linearGradient37623"
@@ -7079,7 +7155,7 @@
        x2="145.5"
        y1="627.5"
        x1="145.5"
-       gradientTransform="matrix(0,1,-1,0,850,297)"
+       gradientTransform="rotate(90,276.5,573.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23580"
        xlink:href="#linearGradient20324"
@@ -7089,7 +7165,7 @@
        x2="145.5"
        y1="627.5"
        x1="145.5"
-       gradientTransform="matrix(-1,0,0,-1,366,1072)"
+       gradientTransform="rotate(180,183,536)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23582"
        xlink:href="#linearGradient20324"
@@ -7099,7 +7175,7 @@
        x2="145.5"
        y1="627.5"
        x1="145.5"
-       gradientTransform="matrix(0,-1,1,0,-409,588)"
+       gradientTransform="rotate(-90,89.5,498.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23587"
        xlink:href="#linearGradient20324"
@@ -7129,7 +7205,7 @@
        x2="150.5"
        y1="647.75"
        x1="150.5"
-       gradientTransform="matrix(0,1,-1,0,870,292)"
+       gradientTransform="rotate(90,289,581)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23593"
        xlink:href="#linearGradient37623"
@@ -7149,7 +7225,7 @@
        x2="150.5"
        y1="647.75"
        x1="150.5"
-       gradientTransform="matrix(0,-1,1,0,-429,593)"
+       gradientTransform="rotate(-90,82,511)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23600"
        xlink:href="#linearGradient37623"
@@ -7399,7 +7475,7 @@
        x2="145.5"
        y1="627.5"
        x1="145.5"
-       gradientTransform="matrix(0,1,-1,0,850,297)"
+       gradientTransform="rotate(90,276.5,573.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23813"
        xlink:href="#linearGradient24632"
@@ -7409,7 +7485,7 @@
        x2="145.5"
        y1="627.5"
        x1="145.5"
-       gradientTransform="matrix(-1,0,0,-1,366,1072)"
+       gradientTransform="rotate(180,183,536)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23815"
        xlink:href="#linearGradient24632"
@@ -7419,7 +7495,7 @@
        x2="145.5"
        y1="627.5"
        x1="145.5"
-       gradientTransform="matrix(0,-1,1,0,-409,588)"
+       gradientTransform="rotate(-90,89.5,498.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23817"
        xlink:href="#linearGradient24632"
@@ -7449,7 +7525,7 @@
        x2="150.5"
        y1="647.75"
        x1="150.5"
-       gradientTransform="matrix(0,1,-1,0,870,292)"
+       gradientTransform="rotate(90,289,581)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23823"
        xlink:href="#linearGradient24642"
@@ -7469,7 +7545,7 @@
        x2="150.5"
        y1="647.75"
        x1="150.5"
-       gradientTransform="matrix(0,-1,1,0,-429,593)"
+       gradientTransform="rotate(-90,82,511)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient23827"
        xlink:href="#linearGradient24642"
@@ -8133,7 +8209,7 @@
        x2="269.3085"
        y1="279.72827"
        x1="248.69196"
-       gradientTransform="translate(-1,0)"
+       gradientTransform="translate(-1)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient62436"
        xlink:href="#linearGradient319"
@@ -8153,7 +8229,7 @@
        x2="327.92044"
        y1="283.61511"
        x1="354.50601"
-       gradientTransform="matrix(0,-1,1,0,59,638)"
+       gradientTransform="rotate(-90,348.5,289.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient62560"
        xlink:href="#linearGradient319"
@@ -8173,7 +8249,7 @@
        x2="344.5"
        y1="288.5"
        x1="342"
-       gradientTransform="matrix(0,1,-1,0,638,-40)"
+       gradientTransform="rotate(90,339,299)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient25383"
        xlink:href="#linearGradient1610"
@@ -8183,7 +8259,7 @@
        x2="344.5"
        y1="288.5"
        x1="342"
-       gradientTransform="matrix(-1,0,0,-1,699,599)"
+       gradientTransform="rotate(180,349.5,299.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient25385"
        xlink:href="#linearGradient1610"
@@ -8193,7 +8269,7 @@
        x2="344.5"
        y1="288.5"
        x1="342"
-       gradientTransform="matrix(0,-1,1,0,60,660)"
+       gradientTransform="rotate(-90,360,300)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient25387"
        xlink:href="#linearGradient1610"
@@ -8213,7 +8289,7 @@
        x2="344.5"
        y1="288.5"
        x1="342"
-       gradientTransform="matrix(0,1,-1,0,638,-40)"
+       gradientTransform="rotate(90,339,299)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient25575"
        xlink:href="#linearGradient1610"
@@ -8223,7 +8299,7 @@
        x2="344.5"
        y1="288.5"
        x1="342"
-       gradientTransform="matrix(-1,0,0,-1,699,599)"
+       gradientTransform="rotate(180,349.5,299.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient25577"
        xlink:href="#linearGradient1610"
@@ -8233,7 +8309,7 @@
        x2="344.5"
        y1="288.5"
        x1="342"
-       gradientTransform="matrix(0,-1,1,0,60,660)"
+       gradientTransform="rotate(-90,360,300)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient25579"
        xlink:href="#linearGradient1610"
@@ -8821,7 +8897,7 @@
        x2="99.628899"
        y1="114.71685"
        x1="112.18942"
-       gradientTransform="translate(19.367382,0)"
+       gradientTransform="translate(19.367382)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient27771"
        xlink:href="#linearGradient13973"
@@ -8923,7 +8999,7 @@
        x2="353.72073"
        y1="314.11398"
        x1="352.98236"
-       gradientTransform="matrix(0.707107,-0.707107,0.707107,0.707107,-346.7085,428.4841)"
+       gradientTransform="rotate(-45,343.872,632.75601)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient53127"
        xlink:href="#linearGradient37542"
@@ -9157,7 +9233,7 @@
        x2="-14.000002"
        y1="448"
        x1="-22.902081"
-       gradientTransform="translate(2e-6,0)"
+       gradientTransform="translate(2e-6)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient50870"
        xlink:href="#linearGradient23178"
@@ -9185,7 +9261,7 @@
        x2="452.63602"
        y1="89.363937"
        x1="458.99997"
-       gradientTransform="matrix(0.7071068,-0.7071068,0.7071067,0.7071067,-140.04288,401.30258)"
+       gradientTransform="rotate(-45,414.39357,369.69801)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient33585"
        xlink:href="#linearGradient9030"
@@ -9824,7 +9900,7 @@
        x2="332.88193"
        y1="126.23978"
        x1="387.30396"
-       gradientTransform="matrix(0.707107,-0.707107,0.707107,0.707107,-241.7085,428.4841)"
+       gradientTransform="rotate(-45,396.37194,506.00979)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient37636"
        xlink:href="#linearGradient37542"
@@ -9834,7 +9910,7 @@
        x2="353.72073"
        y1="314.11398"
        x1="352.98236"
-       gradientTransform="matrix(0.707107,-0.707107,0.707107,0.707107,-346.7085,428.4841)"
+       gradientTransform="rotate(-45,343.872,632.75601)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient37638"
        xlink:href="#linearGradient37542"
@@ -11307,7 +11383,7 @@
        x2="-20"
        y1="283"
        x1="-20"
-       gradientTransform="translate(-2,0)"
+       gradientTransform="translate(-2)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient38570"
        xlink:href="#linearGradient1610"
@@ -11473,7 +11549,7 @@
        x2="97.75"
        y1="84.25"
        x1="101"
-       gradientTransform="translate(21,0)"
+       gradientTransform="translate(21)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient39843-3"
        xlink:href="#linearGradient1610-8"
@@ -11494,7 +11570,7 @@
        x2="96.592278"
        y1="81.439644"
        x1="87.44548"
-       gradientTransform="translate(21,0)"
+       gradientTransform="translate(21)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient39845-2"
        xlink:href="#linearGradient319-77"
@@ -11606,7 +11682,7 @@
        x2="424.75217"
        y1="77"
        x1="437.98615"
-       gradientTransform="translate(0.01387,0)"
+       gradientTransform="translate(0.01387)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient41672"
        xlink:href="#linearGradient319"
@@ -11616,7 +11692,7 @@
        x2="424.75217"
        y1="78"
        x1="438.61115"
-       gradientTransform="translate(0.01387,0)"
+       gradientTransform="translate(0.01387)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient41674"
        xlink:href="#linearGradient319"
@@ -11626,7 +11702,7 @@
        x2="424.75217"
        y1="77.44017"
        x1="441.98615"
-       gradientTransform="translate(0.01387,0)"
+       gradientTransform="translate(0.01387)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient41676"
        xlink:href="#linearGradient319"
@@ -11718,7 +11794,7 @@
        x2="99.628899"
        y1="113.45913"
        x1="111.46314"
-       gradientTransform="translate(19.367382,0)"
+       gradientTransform="translate(19.367382)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient38724"
        xlink:href="#linearGradient13973"
@@ -11864,7 +11940,7 @@
        x2="-28.177105"
        y1="-121.58411"
        x1="-70.605209"
-       gradientTransform="translate(22,0)"
+       gradientTransform="translate(22)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient39686-1"
        xlink:href="#linearGradient1610-3-7"
@@ -11885,7 +11961,7 @@
        x2="-55.5975"
        y1="-124"
        x1="-74"
-       gradientTransform="translate(22,0)"
+       gradientTransform="translate(22)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient39688-9"
        xlink:href="#linearGradient319-5-9-4"
@@ -13353,7 +13429,7 @@
        x2="-88"
        y1="408"
        x1="-97"
-       gradientTransform="translate(6,0)"
+       gradientTransform="translate(6)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient38695"
        xlink:href="#linearGradient16500"
@@ -13393,7 +13469,7 @@
        x2="-88"
        y1="413.51562"
        x1="-90.5"
-       gradientTransform="matrix(-1,0,0,-1,-157,821.03125)"
+       gradientTransform="rotate(180,-78.5,410.51562)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient38706"
        xlink:href="#linearGradient319"
@@ -13413,7 +13489,7 @@
        x2="-88"
        y1="408"
        x1="-97"
-       gradientTransform="translate(8,0)"
+       gradientTransform="translate(8)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient38723"
        xlink:href="#linearGradient16500"
@@ -13443,7 +13519,7 @@
        x2="-86.75"
        y1="413.98114"
        x1="-89.75"
-       gradientTransform="matrix(-1,0,0,-1,-156,821.03125)"
+       gradientTransform="rotate(180,-78,410.51562)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient38729"
        xlink:href="#linearGradient319"
@@ -13463,7 +13539,7 @@
        x2="-86.75"
        y1="413.98114"
        x1="-89.75"
-       gradientTransform="matrix(-1,0,0,-1,-156,824.54874)"
+       gradientTransform="rotate(180,-78,412.27437)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient38736"
        xlink:href="#linearGradient319"
@@ -13492,7 +13568,7 @@
        x2="-88"
        y1="408"
        x1="-97"
-       gradientTransform="translate(6,0)"
+       gradientTransform="translate(6)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient39203"
        xlink:href="#linearGradient16500"
@@ -13541,7 +13617,7 @@
        x2="-88"
        y1="408"
        x1="-97"
-       gradientTransform="translate(6,0)"
+       gradientTransform="translate(6)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient39252"
        xlink:href="#linearGradient16500"
@@ -13654,7 +13730,7 @@
        x2="58.680996"
        y1="641.73358"
        x1="39.102718"
-       gradientTransform="translate(-20,0)"
+       gradientTransform="translate(-20)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient39508"
        xlink:href="#linearGradient319-46"
@@ -13885,7 +13961,7 @@
        x2="99.628899"
        y1="112.62726"
        x1="111.2239"
-       gradientTransform="translate(19.367382,0)"
+       gradientTransform="translate(19.367382)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient40639-1-2-1"
        xlink:href="#linearGradient13973-3-7-8"
@@ -14295,7 +14371,7 @@
        x2="336.40625"
        y1="108.35222"
        x1="351.15625"
-       gradientTransform="matrix(-1,0,0,-1,698,183)"
+       gradientTransform="rotate(180,349,91.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient44944"
        xlink:href="#linearGradient44939-8"
@@ -14563,7 +14639,7 @@
        x2="353.72073"
        y1="314.11398"
        x1="352.98236"
-       gradientTransform="matrix(0.707107,-0.707107,0.707107,0.707107,-346.7085,428.4841)"
+       gradientTransform="rotate(-45,343.872,632.75601)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient25449"
        xlink:href="#linearGradient37542"
@@ -14818,7 +14894,7 @@
          id="stop37546-94" />
     </linearGradient>
     <linearGradient
-       gradientTransform="matrix(0,1,-1,0,634.98585,-146.00607)"
+       gradientTransform="rotate(90,390.49596,244.48989)"
        y2="247.99998"
        x2="392.0101"
        y1="222.99998"
@@ -15112,13 +15188,13 @@
        x2="752"
        y1="402"
        x1="743"
-       gradientTransform="matrix(0,-1,1,0,-239.00001,1286)"
+       gradientTransform="rotate(-90,523.5,762.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient28583"
        xlink:href="#linearGradient16500"
        inkscape:collect="always" />
     <linearGradient
-       gradientTransform="matrix(0,1,-1,0,574.99991,384.00001)"
+       gradientTransform="rotate(90,95.49995,479.49996)"
        y2="406.47784"
        x2="161.83331"
        y1="413.87982"
@@ -15142,7 +15218,7 @@
        x2="163.11441"
        y1="544.21143"
        x1="148.56801"
-       gradientTransform="matrix(0,1,-1,0,719.99999,383.00001)"
+       gradientTransform="rotate(90,168.49999,551.5)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient28600"
        xlink:href="#linearGradient319"
@@ -15685,7 +15761,7 @@
        x2="85.347076"
        y1="100.22395"
        x1="85.548706"
-       gradientTransform="matrix(0.707107,-0.707107,0.707107,0.707107,257.1483,118.6716)"
+       gradientTransform="rotate(-45,271.82331,-251.06972)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient31019"
        xlink:href="#linearGradient319"
@@ -15705,7 +15781,7 @@
        x2="85.347076"
        y1="100.22395"
        x1="85.548706"
-       gradientTransform="matrix(0.707107,-0.707107,0.707107,0.707107,257.1483,118.6716)"
+       gradientTransform="rotate(-45,271.82331,-251.06972)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient31055"
        xlink:href="#linearGradient319"
@@ -15725,7 +15801,7 @@
        x2="85.347076"
        y1="100.22395"
        x1="85.548706"
-       gradientTransform="matrix(0.707107,-0.707107,0.707107,0.707107,257.1483,118.6716)"
+       gradientTransform="rotate(-45,271.82331,-251.06972)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient31151"
        xlink:href="#linearGradient319"
@@ -15745,7 +15821,7 @@
        x2="85.60022"
        y1="99.348953"
        x1="85.861206"
-       gradientTransform="matrix(0.707107,-0.707107,0.707107,0.707107,257.14826,118.6716)"
+       gradientTransform="rotate(-45,271.82329,-251.06967)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient31155"
        xlink:href="#linearGradient30958"
@@ -15850,7 +15926,7 @@
        x2="-303.00003"
        y1="475"
        x1="-307"
-       gradientTransform="translate(-42,0)"
+       gradientTransform="translate(-42)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient52998-5-5"
        xlink:href="#linearGradient319-31-8-9-1"
@@ -15882,7 +15958,7 @@
        x2="-304.76843"
        y1="476.0105"
        x1="-308.7684"
-       gradientTransform="translate(-42,0)"
+       gradientTransform="translate(-42)"
        gradientUnits="userSpaceOnUse"
        id="linearGradient53000-3-9"
        xlink:href="#linearGradient319-31-8-9-1"
@@ -16560,25 +16636,30 @@
      d=""
      sodipodi:nodetypes="cc"
      id="path23417"
-     style="fill:#000000;fill-opacity:1;fill-rule:evenodd;stroke:#1a1a1a;stroke-width:3;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0;marker:none;visibility:visible;display:inline;overflow:visible;enable-background:accumulate" />
+     style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:evenodd;stroke:#1a1a1a;stroke-width:3;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;enable-background:accumulate" />
   <path
      inkscape:connector-curvature="0"
      d=""
      sodipodi:nodetypes="cc"
      id="path23347"
-     style="fill:none;stroke:#ffffff;stroke-width:1.79999995;stroke-linecap:round;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none" />
+     style="fill:none;stroke:#ffffff;stroke-width:1.79999995;stroke-linecap:round;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1" />
   <g
      inkscape:groupmode="layer"
      id="layer4"
      inkscape:label="sheet_layout"
-     style="display:inline" />
+     style="display:none"
+     transform="translate(116,578)" />
   <g
-     transform="translate(-872,-180)"
+     transform="translate(-756,398)"
      style="display:inline"
      id="layer1"
      inkscape:groupmode="layer"
      inkscape:label="ICONS">
     <g
+       style="opacity:0.5"
+       id="g7488"
+       transform="translate(218)" />
+    <g
        transform="translate(-211.20006,170)"
        id="g10203" />
     <g
@@ -16748,7 +16829,7 @@
     <g
        id="g17117" />
     <g
-       transform="translate(9,0)"
+       transform="translate(9)"
        id="g17121" />
     <g
        id="g17128"
@@ -16797,7 +16878,7 @@
        transform="translate(67,200.06499)"
        id="g22051" />
     <g
-       transform="translate(-23,0)"
+       transform="translate(-23)"
        id="g40816">
       <g
          id="g40830" />
@@ -16855,7 +16936,7 @@
        inkscape:export-filename="C:\Documents and Settings\Tata\Pulpit\BLENDER ICONSET\blender-cvs-windows\.blender\.blender\icons\jendrzych's iconset.png"
        inkscape:export-xdpi="90"
        inkscape:export-ydpi="90"
-       transform="matrix(-1,0,0,-1,104.1613,262.99999)"
+       transform="rotate(180,52.08065,131.5)"
        style="display:inline;stroke:#ffffff">
       <path
          inkscape:connector-curvature="0"
@@ -16918,13 +16999,6 @@
        inkscape:export-filename="C:\Documents and Settings\Tata\Pulpit\Kopia blender\.blender\icons\blender's iconset.png"
        inkscape:export-xdpi="90"
        inkscape:export-ydpi="90" />
-    <rect
-       y="180"
-       x="872"
-       height="192"
-       width="192"
-       id="rect30285"
-       style="display:inline;overflow:visible;visibility:visible;opacity:0;fill:#000000;fill-opacity:1;fill-rule:evenodd;stroke:none;stroke-width:5.39191818;marker:none;enable-background:accumulate" />
     <g
        transform="translate(0,12)"
        id="g46790" />
@@ -16937,714 +17011,233 @@
        d=""
        inkscape:connector-curvature="0" />
     <g
-       id="g15770-4"
-       transform="translate(-1653.0047,902.99644)"
-       style="display:inline;opacity:0.6;enable-background:new">
-      <path
-         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"
-         d="m 2691.4961,-577 c -1.4585,10e-6 -2.8611,0.57961 -3.877,1.63477 L 2673.293,-561 h 14.705 c 0.5523,-6e-5 0.9999,-0.44774 1,-1 v -10.5 c 0,-1.94484 1.5551,-3.5 3.5,-3.5 0.6573,-0.009 0.9864,-0.50465 0.9864,-1 z"
-         id="path15766-8"
-         inkscape:connector-curvature="0" />
-      <path
-         sodipodi:nodetypes="cccccccc"
-         inkscape:connector-curvature="0"
-         d="m 2691.4963,-576.50002 h 19.9969 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 l 0.01,41.00357 c 0,0.55581 -0.4477,1.00353 -1,1 h -37.0035 c -0.5559,0.004 -1.0036,-0.44419 -1,-1 v -24.00355"
-         style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1;stroke-linecap:square;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-         id="path15768-2" />
-    </g>
-    <path
-       sodipodi:nodetypes="ccssssssccccscssscccccsscscsccccssccc"
-       style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1px;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"
-       d="m 875.49998,329 c -0.4167,0 -0.7749,0.19292 -1.041,0.45898 -0.2661,0.26607 -0.459,0.62435 -0.459,1.04102 v 32 c 0,3.03164 2.4684,5.5 5.5,5.5 H 912.5 c 3.0316,0 5.5,-2.46836 5.5,-5.5 v -20 c 0,-0.42425 -0.1991,-0.77237 -0.4648,-1.04883 -0.2658,-0.27646 -0.6395,-0.51191 -1.1055,-0.44531 l 0.07,-0.006 h -6.5 -27.50002 c -0.095,0 -0.1807,0.0361 -0.3223,0.17773 -0.1417,0.14177 -0.1777,0.22697 -0.1777,0.32227 v 21 c 0,1.20549 -0.8756,2.25014 -2.0625,2.46094 -1.1869,0.2108 -2.3701,-0.46784 -2.7852,-1.59961 -0.041,-0.1158 -0.061,-0.23824 -0.058,-0.36119 h 2.4062 v -0.008 c 0.03,-2.1e-4 0.058,0.005 0.088,0 0.2451,-0.0435 0.4121,-0.2432 0.4121,-0.49219 v -21 c 0,-0.74923 0.3782,-1.35083 0.7637,-1.73633 0.3856,-0.38565 0.9871,-0.76367 1.7363,-0.76367 h 27.50002 v -4.5 c 0,-0.41667 -0.1929,-0.77495 -0.459,-1.04102 -0.2661,-0.26606 -0.6243,-0.45898 -1.041,-0.45898 l -18.02732,0.0449 c -0.055,1.2e-4 -0.1892,-0.0544 -0.2989,-0.17383 -0.1114,-0.1193 -0.1738,-0.28349 -0.1738,-0.37109 v -2 c 0,-0.41667 -0.1929,-0.77495 -0.459,-1.04102 -0.2661,-0.26606 -0.6243,-0.45898 -1.041,-0.45898 z"
-       id="path15794"
-       inkscape:connector-curvature="0" />
-    <g
-       transform="translate(-1139.094,997)"
+       transform="translate(810,316.5)"
        style="display:inline;enable-background:new"
-       id="g16516">
-      <g
-         id="g15792"
-         transform="translate(-500,-182)"
-         style="opacity:0.6">
-        <path
-           sodipodi:nodetypes="cscssscssssssssssccsc"
-           inkscape:connector-curvature="0"
-           d="m 2564.8618,-451.50005 c 0.3211,0.5561 0.8787,0.91004 1.4878,0.98518 0.2496,0.0308 0.5078,0.0148 0.7619,-0.0533 0.8744,-0.23427 1.4824,-1.02663 1.4824,-1.93185 v -22 c 0,-0.4714 0.5286,-1 1,-1 l 34,2e-5 c 0.4667,-0.0667 1,0.52863 1,1.00003 v 22 c 0,2.76142 -2.2386,5 -5,5 h -33 c -2.7614,0 -5,-2.23858 -5,-5 v -33 c 0,-0.5 0.5,-1 1,-1 h 13 c 0.5,0 1,0.5 1,1 v 2 c 0,0.5 0.4872,1.0452 0.9723,1.0442 l 18.0277,-0.0442 c 0.5,0 1,0.5 1,1 v 2"
-           style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1px;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-           id="path15787" />
-        <path
-           sodipodi:nodetypes="cc"
-           inkscape:connector-curvature="0"
-           id="path15790"
-           d="m 2596.5,-479.5 v 1.5"
-           style="fill:none;stroke:#ffffff;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1" />
-      </g>
-      <g
-         id="g15799"
-         transform="matrix(0,1,1,0,2028.5,-908.5)"
-         style="display:inline;opacity:1;stroke:#ffffff;stroke-width:2;stroke-miterlimit:4;stroke-dasharray:none;enable-background:new"
-         inkscape:export-filename="C:\Users\Andrzej Ambroż\Desktop\mtrx.png"
-         inkscape:export-xdpi="96"
-         inkscape:export-ydpi="96">
-        <path
-           id="path15797"
-           style="opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:2;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke"
-           d="m 271.5,63.5 v -5 c 0,-1.104569 -0.89543,-2 -2,-2 h -11 m 4,5 -5,-5 5,-5"
-           inkscape:connector-curvature="0"
-           sodipodi:nodetypes="ccccccc" />
-      </g>
-    </g>
+       id="g72880" />
+    <image
+       sodipodi:absref="C:\Users\Andrzej Ambroż\Desktop\a230341b4bf9a9d653c3e8062c4702e859d6d32f.png"
+       y="143.4964"
+       x="8460.25"
+       id="image7203"
+       xlink:href="file:///C:/Users/Andrzej%20Ambro%C5%BC/Desktop/a230341b4bf9a9d653c3e8062c4702e859d6d32f.png"
+       style="stroke-width:0.69403505;image-rendering:optimizeSpeed"
+       preserveAspectRatio="none"
+       height="487.00717"
+       width="1377.452" />
     <g
-       style="display:inline;opacity:0.6;enable-background:new"
-       transform="translate(-1653.0001,855.00002)"
-       id="g6937">
+       transform="translate(-30,-450)"
+       id="g7205" />
+    <g
+       id="g7480"
+       transform="matrix(8.4250411,0,0,8.2967687,-7333.563,3000.3193)"
+       style="stroke-width:0.1196078" />
+    <g
+       style="stroke-width:0.1196078"
+       transform="matrix(8.4250411,0,0,-8.2967687,-7336.563,-3470.4526)"
+       id="g7569" />
+    <g
+       id="g7599"
+       style="display:inline;opacity:1;enable-background:new"
+       transform="matrix(1.4999383,0,0,1.4999989,-371.4454,70.499748)">
       <path
+         sodipodi:nodetypes="ccccccccccc"
          inkscape:connector-curvature="0"
-         id="path6933"
-         d="m 2691.4961,-577 c -1.4585,10e-6 -2.8611,0.57961 -3.877,1.63477 L 2673.293,-561 h 14.705 c 0.5523,-6e-5 0.9999,-0.44774 1,-1 v -10.5 c 0,-1.94484 1.5551,-3.5 3.5,-3.5 0.6573,-0.009 0.9864,-0.50465 0.9864,-1 z"
-         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" />
+         d="m 884,-165 c 0,0.55581 -0.4477,1.00353 -1,1 h -92.0036 c -0.5558,0.004 -1.0036,-0.44419 -1,-1 v -86 c 0,-1.32609 0.5269,-2.59785 1.4824,-3.51777 l 34,-34 c 0.9198,-0.95544 2.1916,-1.48222 3.5176,-1.48223 l 53.9911,2e-5 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 z"
+         style="display:inline;opacity:0.2;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:2;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
+         id="path7410" />
       <path
-         id="path6935"
-         style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1;stroke-linecap:square;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-         d="m 2691.4963,-576.50002 h 19.9969 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 l 0.01,41.00357 c 0,0.55581 -0.4477,1.00353 -1,1 h -37.0035 c -0.5559,0.004 -1.0036,-0.44419 -1,-1 v -24.00355"
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:2;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+         d="m 76.996094,21 c -1.59084,10e-6 -3.126431,0.634133 -4.238282,1.789062 L 38.785156,56.761719 C 37.630236,57.873629 36.996094,59.40902 36.996094,61 v 85.99414 c -0.007,1.09637 0.910592,2.01376 2.007812,2.00586 H 130.99414 C 132.09029,149.007 133,148.0928 133,147 L 132.9883,23 v -0.002 c -0.003,-1.09124 -0.90724,-1.998107 -2,-1.998047 z M 130.33323,146.33338 H 39.66283 V 62.999985 c 0.0019,-1.44306 0.588363,-2.666668 2.000083,-2.666668 h 32.001316 c 1.09262,-1.2e-4 2.666576,-0.914632 2.666776,-2.000002 v -32.66669 c 0,-1.411771 0.588253,-2.000002 2.000083,-2.000002 h 52.002142 z"
+         transform="translate(752,-312)"
+         id="path7412"
          inkscape:connector-curvature="0"
-         sodipodi:nodetypes="cccccccc" />
+         sodipodi:nodetypes="cccscccccccccccsccsscc" />
     </g>
     <g
-       id="g6943"
-       transform="translate(-1701.0001,855.00002)"
-       style="display:inline;opacity:0.6;enable-background:new">
+       style="display:inline;stroke-width:0.10177074;enable-background:new"
+       id="g7452"
+       transform="matrix(14.821991,0,0,14.654739,-7507.9905,-1149.2844)">
+      <path
+         style="display:inline;opacity:0.2;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:0.20354147;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
+         d="m 613.67414,61.535437 h 8.65173 c 0.57439,0 1.03681,0.46242 1.03681,1.036816 v 2.839759 c 0,0.574396 -0.46242,1.036815 -1.03681,1.036815 h -8.65173 c -0.57439,0 -1.03681,-0.462419 -1.03681,-1.036815 v -2.839759 c 0,-0.574396 0.46242,-1.036816 1.03681,-1.036816 z"
+         id="rect7450"
+         inkscape:connector-curvature="0" />
       <path
-         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"
-         d="m 2691.4961,-577 c -1.4585,10e-6 -2.8611,0.57961 -3.877,1.63477 L 2673.293,-561 h 14.705 c 0.5523,-6e-5 0.9999,-0.44774 1,-1 v -10.5 c 0,-1.94484 1.5551,-3.5 3.5,-3.5 0.6573,-0.009 0.9864,-0.50465 0.9864,-1 z"
-         id="path6939"
+         id="path7424"
+         transform="matrix(0.10118267,0,0,0.10236222,598.97766,126.33073)"
+         d="M 145.24609,-711 C 139.56927,-711 135,-706.4825 135,-700.87109 v 78 C 135,-628.4825 139.56927,-633 145.24609,-633 h 85.50782 c 5.67682,0 10.24609,4.5175 10.24609,10.12891 v -78 C 241,-706.4825 236.43073,-711 230.75391,-711 Z"
+         style="display:inline;opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:2;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
          inkscape:connector-curvature="0" />
       <path
-         sodipodi:nodetypes="cccccccc"
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:2;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+         d="m 553.24414,21.003906 c -6.21668,0 -11.25391,4.978036 -11.25391,11.123047 v 78.001957 27.74414 c 0,6.14501 5.03723,11.12304 11.25391,11.12304 h 85.51172 c 2.7198,0 5.21303,-0.95235 7.1582,-2.53906 2.50094,-2.04005 4.09571,-5.12741 4.09571,-8.58398 V 110.12891 32.126953 c 0,-6.145011 -5.03723,-11.123047 -11.25391,-11.123047 z m 0,1.984375 h 85.51172 c 5.13704,0 9.24609,4.060861 9.24609,9.138672 v 71.667967 c -2.03176,-2.89305 -5.41213,-4.789061 -9.24609,-4.789061 h -85.51172 c -3.83396,0 -7.21433,1.896011 -9.24609,4.789061 V 32.126953 c 0,-5.077811 4.10905,-9.138672 9.24609,-9.138672 z m 0,78.001949 h 85.51172 c 5.13704,0 8.58679,4.26374 8.58679,9.34155 v 27.33166 c 0,5.07781 -3.53119,8.66614 -8.66823,8.66614 h -85.34878 c -5.13704,0 -8.66823,-3.58833 -8.66823,-8.66614 v -27.33166 c 0,-5.07781 3.44969,-9.34155 8.58673,-9.34155 z"
+         id="rect7426"
+         transform="matrix(0.10118267,0,0,0.10236222,557.69513,51.299216)"
          inkscape:connector-curvature="0"
-         d="m 2691.4963,-576.50002 h 19.9969 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 l 0.01,41.00357 c 0,0.55581 -0.4477,1.00353 -1,1 h -37.0035 c -0.5559,0.004 -1.0036,-0.44419 -1,-1 v -24.00355"
-         style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1;stroke-linecap:square;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-         id="path6941" />
-    </g>
-    <g
-       style="display:inline;opacity:0.6;enable-background:new"
-       transform="translate(-1749.0047,854.99644)"
-       id="g6949">
+         sodipodi:nodetypes="sscssscscsssssscsscsssssssssss" />
       <path
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:0.10177073;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+         d="m 621.91311,62.457001 c 0.22334,0 0.40429,0.180956 0.40429,0.404297 v 0.0098 c 0,0.22334 -0.18095,0.404297 -0.40429,0.404297 -0.22335,0 -0.40625,-0.180957 -0.40625,-0.404297 v -0.0098 c 0,-0.223341 0.1829,-0.404297 0.40625,-0.404297 z"
+         id="rect7434"
          inkscape:connector-curvature="0"
-         id="path6945"
-         d="m 2691.4961,-577 c -1.4585,10e-6 -2.8611,0.57961 -3.877,1.63477 L 2673.293,-561 h 14.705 c 0.5523,-6e-5 0.9999,-0.44774 1,-1 v -10.5 c 0,-1.94484 1.5551,-3.5 3.5,-3.5 0.6573,-0.009 0.9864,-0.50465 0.9864,-1 z"
-         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" />
+         sodipodi:nodetypes="sssssss" />
       <path
-         id="path6947"
-         style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1;stroke-linecap:square;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-         d="m 2691.4963,-576.50002 h 19.9969 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 l 0.01,41.00357 c 0,0.55581 -0.4477,1.00353 -1,1 h -37.0035 c -0.5559,0.004 -1.0036,-0.44419 -1,-1 v -24.00355"
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:0.20354147;stroke-linecap:round;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+         d="m 613.57935,62.45579 c -0.056,0.0012 -0.10057,0.04747 -0.0996,0.103516 v 2.867187 c -0.01,0.144644 0.21266,0.144644 0.20312,0 v -2.867187 c 9.9e-4,-0.05757 -0.0459,-0.104503 -0.10351,-0.103516 z m 0.60742,0 c -0.056,0.0012 -0.10057,0.04747 -0.0996,0.103516 v 2.867187 c -0.01,0.144644 0.21266,0.144644 0.20312,0 v -2.867187 c 9.9e-4,-0.05757 -0.0459,-0.104503 -0.10351,-0.103516 z m 0.60742,0 c -0.0568,1.15e-4 -0.10253,0.04671 -0.10156,0.103516 v 2.867187 c 0,0.136713 0.20507,0.136713 0.20507,0 v -2.867187 c 9.9e-4,-0.05757 -0.0459,-0.104503 -0.10351,-0.103516 z m 0.60742,0 c -0.0568,1.15e-4 -0.10253,0.04671 -0.10156,0.103516 v 2.867187 c -0.01,0.144644 0.21266,0.144644 0.20312,0 v -2.867187 c 9.7e-4,-0.0568 -0.0447,-0.103401 -0.10156,-0.103516 z m 0.60547,0 c -0.056,0.0012 -0.10057,0.04747 -0.0996,0.103516 v 2.867187 c -0.01,0.144644 0.21266,0.144644 0.20312,0 v -2.867187 c 9.9e-4,-0.05757 -0.0459,-0.104503 -0.10351,-0.103516 z"
+         id="path7444"
          inkscape:connector-curvature="0"
-         sodipodi:nodetypes="cccccccc" />
+         sodipodi:nodetypes="ccccccccccccccccssccccccccccccccc" />
     </g>
     <g
-       id="g6955"
-       transform="translate(-1797.0001,855.00002)"
-       style="display:inline;opacity:0.6;enable-background:new">
+       transform="matrix(12.319682,0,0,12.172353,-5727.9739,-984.34032)"
+       id="g7322"
+       style="display:inline;stroke-width:0.12309717;enable-background:new">
       <path
-         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"
-         d="m 2691.4961,-577 c -1.4585,10e-6 -2.8611,0.57961 -3.877,1.63477 L 2673.293,-561 h 14.705 c 0.5523,-6e-5 0.9999,-0.44774 1,-1 v -10.5 c 0,-1.94484 1.5551,-3.5 3.5,-3.5 0.6573,-0.009 0.9864,-0.50465 0.9864,-1 z"
-         id="path6951"
+         style="display:inline;opacity:0.2;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:0.24619435;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
+         d="m 613.67414,61.612633 h 8.65173 c 0.57439,0 1.03681,0.46242 1.03681,1.036815 v 2.762566 c 0,0.574395 -0.46242,1.036815 -1.03681,1.036815 h -8.65173 c -0.57439,0 -1.03681,-0.46242 -1.03681,-1.036815 v -2.762566 c 0,-0.574395 0.46242,-1.036815 1.03681,-1.036815 z"
+         id="rect7418"
          inkscape:connector-curvature="0" />
       <path
-         sodipodi:nodetypes="cccccccc"
+         style="display:inline;opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:2.41910744;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
+         d="M 145.24609,-711 C 139.56928,-711.01185 135,-706.4825 135,-700.87109 v 78 c 0,-5.61141 4.56928,-10.14076 10.24609,-10.12891 l 85.17749,0.17786 c 5.67681,0.0118 10.24609,4.5175 10.24609,10.12891 v -78 c 0,-5.61141 -4.56928,-10.11706 -10.24609,-10.12891 l -42.58875,-0.0889 z"
+         transform="matrix(0.10118267,0,0,0.10236222,598.97766,126.33073)"
+         id="rect7413"
          inkscape:connector-curvature="0"
-         d="m 2691.4963,-576.50002 h 19.9969 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 l 0.01,41.00357 c 0,0.55581 -0.4477,1.00353 -1,1 h -37.0035 c -0.5559,0.004 -1.0036,-0.44419 -1,-1 v -24.00355"
-         style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1;stroke-linecap:square;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-         id="path6953" />
-    </g>
-    <g
-       id="g6961"
-       transform="translate(-1653.0001,807.00002)"
-       style="display:inline;opacity:0.6;enable-background:new">
-      <path
-         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"
-         d="m 2691.4961,-577 c -1.4585,10e-6 -2.8611,0.57961 -3.877,1.63477 L 2673.293,-561 h 14.705 c 0.5523,-6e-5 0.9999,-0.44774 1,-1 v -10.5 c 0,-1.94484 1.5551,-3.5 3.5,-3.5 0.6573,-0.009 0.9864,-0.50465 0.9864,-1 z"
-         id="path6957"
-         inkscape:connector-curvature="0" />
+         sodipodi:nodetypes="sscsscsscs" />
       <path
-         sodipodi:nodetypes="cccccccc"
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:2;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+         d="m 717.48242,42.996094 c -5.24135,0 -9.49414,4.190752 -9.49414,9.355468 v 65.017578 22.26953 c 0,5.16472 4.25279,9.35547 9.49414,9.35547 h 70.02539 c 5.24136,0 9.49414,-4.19075 9.49414,-9.35547 V 117.36914 52.351562 c 0,-5.164716 -4.25278,-9.355468 -9.49414,-9.355468 z m 0,1.984375 h 70.02539 c 4.15996,0 7.48047,3.271903 7.48047,7.371093 v 59.269528 c -1.73801,-2.19271 -4.43984,-3.60742 -7.48047,-3.60742 h -70.02539 c -3.04062,0 -5.74244,1.41471 -7.48047,3.60742 V 52.351562 c 0,-4.09919 3.32052,-7.371093 7.48047,-7.371093 z m 0,65.017581 h 70.02539 c 4.15996,0 6.83699,3.09749 6.83699,7.19667 v 22.52459 c 0,4.09919 -2.48286,6.62487 -6.64282,6.62487 h -70.41379 c -4.15995,0 -6.64281,-2.52568 -6.64281,-6.62487 v -22.52459 c 0,-4.09918 2.67709,-7.19667 6.83704,-7.19667 z"
+         transform="matrix(0.12219364,0,0,0.1240074,526.00203,48.095825)"
+         id="rect7312"
          inkscape:connector-curvature="0"
-         d="m 2691.4963,-576.50002 h 19.9969 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 l 0.01,41.00357 c 0,0.55581 -0.4477,1.00353 -1,1 h -37.0035 c -0.5559,0.004 -1.0036,-0.44419 -1,-1 v -24.00355"
-         style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1;stroke-linecap:square;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-         id="path6959" />
+         sodipodi:nodetypes="sscsssscsssssscsscssssssssscs" />
+      <g
+         id="g7411"
+         transform="matrix(-1,0,0,1,1235.8994,0)"
+         style="stroke-width:0.12309717">
+        <path
+           style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:0.12309717;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+           d="m 613.88871,62.499044 c -0.23599,0 -0.42773,0.193143 -0.42773,0.429687 v 0.0098 c 0,0.236544 0.19174,0.429687 0.42773,0.429687 0.23599,0 0.42774,-0.193143 0.42774,-0.429687 v -0.0098 c 0,-0.236544 -0.19175,-0.429687 -0.42774,-0.429687 z"
+           id="rect7316"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="sssssss" />
+        <g
+           transform="translate(6.2733256)"
+           id="g7368"
+           style="display:inline;stroke-width:0.12309717;enable-background:new">
+          <path
+             style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:0.24619435;stroke-linecap:round;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+             d="m 613.40672,62.503838 c -0.0633,0.0071 -0.11074,0.06131 -0.10937,0.125 V 65.4804 c 0,0.165367 0.24805,0.165367 0.24805,0 v -2.851562 c 0.002,-0.07532 -0.0639,-0.13455 -0.13868,-0.125 z m 0.64679,10e-7 c -0.0633,0.0071 -0.11074,0.06131 -0.10937,0.125 V 65.4804 c -0.005,0.169458 0.25158,0.169458 0.24609,0 v -2.851562 c 0.002,-0.07444 -0.0627,-0.133249 -0.13672,-0.124999 z m 0.64937,10e-7 c -0.0633,0.0071 -0.11074,0.06131 -0.10937,0.125 v 2.85156 c -0.005,0.169458 0.25158,0.169458 0.24609,0 v -2.851562 c 0.002,-0.07444 -0.0627,-0.133249 -0.13672,-0.124998 z m 0.65195,1e-6 c -0.0633,0.0071 -0.11075,0.06131 -0.10938,0.125 v 2.85156 c -0.005,0.169247 0.25137,0.169247 0.2461,0 v -2.851562 c 0.002,-0.07444 -0.0627,-0.133249 -0.13672,-0.124998 z m 0.64937,-3e-6 c -0.0633,0.0071 -0.11075,0.06131 -0.10938,0.125 V 65.4804 c -0.005,0.169248 0.25136,0.169248 0.2461,0 v -2.851563 c 0.002,-0.07444 -0.0627,-0.133249 -0.13672,-0.124999 z"
+             id="path7359"
+             inkscape:connector-curvature="0"
+             sodipodi:nodetypes="ccsscccccccccccccccccccccccccc" />
+        </g>
+      </g>
     </g>
     <g
-       style="display:inline;opacity:0.6;enable-background:new"
-       transform="translate(-1701.0001,807.00002)"
-       id="g6967">
+       id="g7565"
+       transform="matrix(1.4994746,0,0,1.4999989,314.92923,431.89982)">
       <path
+         id="path28027-5"
+         transform="translate(294.00001,-553.00001)"
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:2.00098109;stroke-linecap:butt;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+         d="m 816.43612,111.73314 c 0,0.5515 0.7823,1.33328 1.3338,1.33333 h 10.0035 c 0.5538,-5e-5 1.3359,-0.77955 1.3338,-1.33333 v -8.00001 c -0.0108,-2.92338 -2.53774,-5.20733 -5.3352,-5.666667 l 3e-5,-63.33338 c -9e-4,-7.597621 -6.34645,-13.676694 -14.00685,-13.666674 l -44.01737,-3e-6 c -7.9872,0.0106 -13.99805,6.01116 -13.99805,14 v 9 l 2.07069,-0.06546 v -9 c 0,-7.01107 4.91866,-11.925191 11.93126,-11.934541 l 44.01542,3e-6 c 6.5948,-0.009 12.0041,5.153493 12.0049,11.666674 l -3e-5,63.331427 c -2.80192,0.45829 -5.33501,2.738541 -5.3359,5.668621 z m 2.0007,2.66667 v 5.33334 c 6e-5,0.55226 0.78364,1.33327 1.3359,1.33333 h 6 c 0.55226,-6e-5 1.33374,-0.78107 1.3338,-1.33333 v -5.33334 h -2.0007 v 4.66667 h -4.6683 v -4.66667 z"
          inkscape:connector-curvature="0"
-         id="path6963"
-         d="m 2691.4961,-577 c -1.4585,10e-6 -2.8611,0.57961 -3.877,1.63477 L 2673.293,-561 h 14.705 c 0.5523,-6e-5 0.9999,-0.44774 1,-1 v -10.5 c 0,-1.94484 1.5551,-3.5 3.5,-3.5 0.6573,-0.009 0.9864,-0.50465 0.9864,-1 z"
-         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" />
-      <path
-         id="path6965"
-         style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1;stroke-linecap:square;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-         d="m 2691.4963,-576.50002 h 19.9969 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 l 0.01,41.00357 c 0,0.55581 -0.4477,1.00353 -1,1 h -37.0035 c -0.5559,0.004 -1.0036,-0.44419 -1,-1 v -24.00355"
-         inkscape:connector-curvature="0"
-         sodipodi:nodetypes="cccccccc" />
+         sodipodi:nodetypes="cccccccccsccsccccccccccccccccc" />
     </g>
     <g
-       id="g6973"
-       transform="translate(-1749.0047,806.99644)"
-       style="display:inline;opacity:0.6;enable-background:new">
+       id="g7749"
+       transform="matrix(1.4998278,0,0,1.4999989,-448.32862,-37.50017)"
+       style="display:inline;opacity:1;enable-background:new">
       <path
-         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"
-         d="m 2691.4961,-577 c -1.4585,10e-6 -2.8611,0.57961 -3.877,1.63477 L 2673.293,-561 h 14.705 c 0.5523,-6e-5 0.9999,-0.44774 1,-1 v -10.5 c 0,-1.94484 1.5551,-3.5 3.5,-3.5 0.6573,-0.009 0.9864,-0.50465 0.9864,-1 z"
-         id="path6969"
+         sodipodi:nodetypes="sssssssss"
+         id="path7747"
+         d="m 998.5,-179 h 121 c 1.385,0 2.5,1.115 2.5,2.5 v 66 c 0,1.385 -1.115,2.5 -2.5,2.5 h -121 c -1.385,0 -2.5,-1.115 -2.5,-2.5 v -66 c 0,-1.385 1.115,-2.5 2.5,-2.5 z"
+         style="opacity:0.2;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:1.99999988;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:105.70774078;stroke-opacity:1;marker:none;paint-order:fill markers stroke"
          inkscape:connector-curvature="0" />
       <path
-         sodipodi:nodetypes="cccccccc"
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1.99999988;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:105.70774078;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"
+         d="m 195.5,38 c -1.94642,0 -3.5,1.60898 -3.5,3.5 v 89 c 0,1.9217 1.5783,3.5 3.5,3.5 h 121 c 1.9217,0 3.5,-1.5783 3.5,-3.5 l 0.0147,-78.333361 c 0,-1.9217 -1.5783,-3.5 -3.5,-3.5 h -77 c -0.3983,10e-6 -0.77905,-0.157923 -1.06055,-0.439453 -0.2816,-0.2816 -0.43841,-0.662348 -0.43945,-1.060547 L 238,41.5 C 237.995,39.578306 236.4217,38 234.5,38 Z m -0.83305,27.333318 c 0,-0.8551 0.48518,-1.333334 1.33348,-1.333334 h 120.01378 c 0.8483,0 1.33349,0.485033 1.33349,1.333334 v 64.666712 c 0,0.84831 -0.48519,1.33334 -1.33349,1.33334 l -120.01378,0 c -0.8483,0 -1.33783,-0.48505 -1.33348,-1.33334"
+         transform="translate(803,-241)"
+         id="path7743"
          inkscape:connector-curvature="0"
-         d="m 2691.4963,-576.50002 h 19.9969 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 l 0.01,41.00357 c 0,0.55581 -0.4477,1.00353 -1,1 h -37.0035 c -0.5559,0.004 -1.0036,-0.44419 -1,-1 v -24.00355"
-         style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1;stroke-linecap:square;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-         id="path6971" />
+         sodipodi:nodetypes="sccssccsccsssscssssssc" />
     </g>
     <g
        style="display:inline;opacity:0.6;enable-background:new"
-       transform="translate(-1797.0001,807.00002)"
-       id="g6979">
-      <path
-         inkscape:connector-curvature="0"
-         id="path6975"
-         d="m 2691.4961,-577 c -1.4585,10e-6 -2.8611,0.57961 -3.877,1.63477 L 2673.293,-561 h 14.705 c 0.5523,-6e-5 0.9999,-0.44774 1,-1 v -10.5 c 0,-1.94484 1.5551,-3.5 3.5,-3.5 0.6573,-0.009 0.9864,-0.50465 0.9864,-1 z"
-         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" />
-      <path
-         id="path6977"
-         style="display:inline;opacity:1;vector-effect:none;fill:none;fill-opacity:1;stroke:#ffffff;stroke-width:1;stroke-linecap:square;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
-         d="m 2691.4963,-576.50002 h 19.9969 c 0.5523,-3e-5 1,0.44769 1.0015,0.99998 l 0.01,41.00357 c 0,0.55581 -0.4477,1.00353 -1,1 h -37.0035 c -0.5559,0.004 -1.0036,-0.44419 -1,-1 v -24.00355"
-         inkscape:connector-curvature="0"
-         sodipodi:nodetypes="cccccccc" />
-    </g>
-    <g
-       transform="translate(810,316.5)"
-       style="display:inline;enable-background:new"
-       id="g72880" />
-    <g
-       id="g3789">
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 996,189 v 0.75 h -1 v 3.5 h 1 V 194 h 2 v -1 -1.25 h -1 V 193 h -1 v -3 h 2 v -1 z"
-         id="path72434" />
-      <rect
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         id="rect72436"
-         width="1"
-         height="5"
-         x="1029"
-         y="189" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 954,189 v 4 h -1 v -1 h -1 v 1 h 0.75 v 1 h 1.5 v -1 H 955 v -4 z"
-         id="path72438" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 999,189 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v 2 h 1 v -5 h -1 v 2 z"
-         id="path72440" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 975,189 v 1 h 2 v -1 z m 2,1 v 3 h 1 v -3 z m 0,3 h -2 v 1 h 2 z m -2,0 v -3 h -1 v 3 z"
-         id="path72442" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 940,189 v 0.75 h -1 v 1.5 h 1 V 192 h 1 v 1 h -2 v 1 h 2 v -0.75 h 1 v -1.5 h -1 V 191 h -1 v -1 h 1.5 v -1 z"
-         id="path72446" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 964,189 v 1 h 1 v 4 h 1 v -4 h 1 v -1 z"
-         id="path72448" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 943,189 v 4.25 h 1 V 194 h 2 v -0.75 h 1 V 189 h -1 v 4 h -2 v -4 z"
-         id="path72450" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 948,189 v 5 h 1 1 v -0.5 h 1 V 192 h -0.75 v -1 H 951 v -1.25 h -1 V 189 h -1 z m 1,1 h 1 v 1 h -1 z m 0,2 h 1 v 1 h -1 z"
-         id="path72452" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 961,189 v 0.75 h -1 v 3.5 h 1 V 194 h 2 v -1 h -2 v -3 h 2 v -1 z"
-         id="path72454" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 956,189 v 5 h 1 2 v -1 h -2 v -1 h 1 v -1 h -1 v -1 h 2 v -1 h -2 z"
-         id="path72456" />
-      <path
-         inkscape:connector-curvature="0"
-         id="path72458"
-         d="m 1019,189 v 5 h 1 2 v -1 h -2 v -4 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6" />
-      <path
-         id="path72460"
-         d="m 970,189 v 1 h 1 v 4 h 1 v -4 h 1 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 981,189 v 1 h 1 v 4 h 1 v -4 h 1 v -1 z"
-         id="path72462" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 985,189 v 5 h 1 v -2 h 1 v 2 h 1 v -5 h -1 v 2 h -1 v -2 z"
-         id="path72464" />
-      <path
-         id="path72466"
-         d="m 989,189 v 5 h 1 2 v -1 h -2 v -1 h 1 v -1 h -1 v -1 h 2 v -1 h -2 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72468"
-         d="m 1004,189 v 4.25 h 1 V 194 h 2 v -0.75 h 1 V 189 h -1 v 4 h -2 v -4 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72470"
-         d="m 1012,189 v 0.75 h -1 v 3.5 h 1 V 194 h 2 v -1 -1.25 h -1 V 193 h -1 v -3 h 2 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1015,189 v 5 h 1 v -2 h 1.25 v -1 h 0.75 v -1.25 h -1 V 189 h -1 z m 1,1 h 1 v 1 h -1 z"
-         id="path72472" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1025,189 v 5 h 1 2 v -1 h -2 v -4 z"
-         id="path72474"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72476"
-         d="m 1032,189 v 0.75 h -1 v 3.5 h 1 V 194 h 2 v -1 h -2 v -3 h 2 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1035,189 v 5 h 1 2 v -1 h -2 v -1 h 1 v -1 h -1 v -1 h 2 v -1 h -2 z"
-         id="path72478" />
-      <path
-         id="path72480"
-         d="m 1039,189 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v 2 h 1 v -5 h -1 v 2 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72482"
-         d="m 1045,189 v 0.75 h -1 v 3.5 h 1 V 194 h 2 v -1 h -2 v -3 h 2 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72484"
-         d="m 1048,189 v 5 h 1 2 v -1 h -2 v -1 h 1 v -1 h -1 v -1 h 2 v -1 h -2 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1052.75,193 v 1 H 1052 v 1 h 1 v -1 h 1 v -1 z"
-         id="path72486" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 882,189 v 1 h 1 v 4 h 1 v -4 h 1 v -1 z"
-         id="path32008-7"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 886,189 v 5 h 1 v -2 h 1 v 2 h 1 v -5 h -1 v 2 h -1 v -2 z"
-         id="path32010-4"
-         inkscape:connector-curvature="0" />
-      <rect
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         id="rect32012-0"
-         width="1"
-         height="5"
-         x="890"
-         y="189" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 893,189 v 0.75 h -1 v 1.5 h 1 V 192 h 1 v 1 h -2 v 1 h 2 v -0.75 h 1 v -1.5 h -1 V 191 h -1 v -1 h 1.5 v -1 z"
-         id="path32015-9"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 899,189 v 0.75 h -1 v 3.5 h 1 V 194 h 2 v -1 h -2 v -3 h 2 v -1 z"
-         id="path32017-4"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 907,189 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v 2 h 1 v -5 h -1 v 2 z"
-         id="path32019-8"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 903,189 v 1 h 2 v -1 z m 2,1 v 3 h 1 v -3 z m 0,3 h -2 v 1 h 2 z m -2,0 v -3 h -1 v 3 z"
-         id="path32021-8"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 912,189 v 1 h 1 v 4 h 1 v -4 h 1 v -1 z"
-         id="path32023-2"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 916,189 v 5 h 1 2 v -1 h -2 v -1 h 1 v -1 h -1 v -1 h 2 v -1 h -2 z"
-         id="path32026-4"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path32029-5"
-         d="m 920,189 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v 2 h 1 v -5 h -1 v 2 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path32031-5"
-         d="m 925,189 v 1 h 1 v 4 h 1 v -4 h 1 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <rect
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         id="rect32033-1"
-         width="1"
-         height="5"
-         x="931"
-         y="189" />
-      <path
-         id="path32035-7"
-         d="m 934,189 v 0.75 h -1 v 1.5 h 1 V 192 h 1 v 1 h -2 v 1 h 2 v -0.75 h 1 v -1.5 h -1 V 191 h -1 v -1 h 1.5 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 888,197 v 5 h 1 v -2 h 1 v 2 h 1 v -5 h -1 v 2 h -1 v -2 z"
-         id="path72562" />
-      <rect
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         id="rect72564"
-         width="1"
-         height="5"
-         x="892"
-         y="197" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 882,197 v 3.25 h 1 V 202 h 1 v -1.75 h 1 V 202 h 1 v -1.75 h 1 V 197 h -1 v 3 h -1 v -2 h -1 v 2 h -1 v -3 z"
-         id="path72576" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 940,197 v 2.25 h 1 V 202 h 1 v -2.75 h 1 V 197 h -1 v 2 h -1 v -2 z"
-         id="path72578" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 904.75,197 v 1 H 904 v 4 h 1 v -2 h 1 v 2 h 1 v -4 h -0.75 v -1 z m 0.25,1 h 1 v 1 h -1 z"
-         id="path72580" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 895,197 v 0.75 h -1 v 3.5 h 1 V 202 h 2 v -1 h -2 v -3 h 2 v -1 z"
-         id="path72582" />
-      <path
-         id="path72588"
-         d="m 898,197 v 5 h 1 v -2 h 1 v 2 h 1 v -5 h -1 v 2 h -1 v -2 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72590"
-         d="m 909,197 v 0.75 h -1 v 3.5 h 1 V 202 h 2 v -1 h -2 v -3 h 2 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 913,197 v 0.75 h -1 v 3.5 h 1 V 202 h 2 v -1 h -2 v -3 h 2 v -1 z"
-         id="path72592" />
-      <path
-         id="path72594"
-         d="m 917,197 v 1 h 2 v -1 z m 2,1 v 3 h 1 v -3 z m 0,3 h -2 v 1 h 2 z m -2,0 v -3 h -1 v 3 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 921,197 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v -1 z m 1,0 h 1 v 3 h 1 v -5 h -1 v 1 h -1 z"
-         id="path72596" />
-      <path
-         id="path72598"
-         d="m 927,197 v 5 h 1 v -2 h 1.25 v -1 H 930 v -1.25 h -1 V 197 h -1 z m 1,1 h 1 v 1 h -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72600"
-         d="m 931.75,197 v 1 H 931 v 4 h 1 v -2 h 1 v 2 h 1 v -4 h -0.75 v -1 z m 0.25,1 h 1 v 1 h -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72602"
-         d="m 935,197 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v 2 h 1 v -5 h -1 v 2 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <rect
-         y="197"
-         x="946"
-         height="5"
-         width="1"
-         id="rect72604"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6" />
-      <path
-         id="path72606"
-         d="m 948,197 v 1 h 1 v 4 h 1 v -4 h 1 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72608"
-         d="m 954.75,197 v 1 H 954 v 4 h 1 v -2 h 1 v 2 h 1 v -4 h -0.75 v -1 z m 0.25,1 h 1 v 1 h -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 958,197 v 5 h 1 2 v -1 h -2 v -4 z"
-         id="path72610"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72612"
-         d="m 963,197 v 1 h 2 v -1 z m 2,1 v 3 h 1 v -3 z m 0,3 h -2 v 1 h 2 z m -2,0 v -3 h -1 v 3 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72614"
-         d="m 967,197 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v 2 h 1 v -5 h -1 v 2 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72616"
-         d="m 973,197 v 0.75 h -1 v 3.5 h 1 V 202 h 2 v -1 -1.25 h -1 V 201 h -1 v -3 h 2 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <rect
-         y="197"
-         x="984"
-         height="5"
-         width="1"
-         id="rect72618"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6" />
-      <path
-         id="path72620"
-         d="m 978,197 v 3.25 h 1 V 202 h 1 v -1.75 h 1 V 202 h 1 v -1.75 h 1 V 197 h -1 v 3 h -1 v -2 h -1 v 2 h -1 v -3 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 990,197 v 5 h 1 v -2 h 1 v 2 h 1 v -5 h -1 v 2 h -1 v -2 z"
-         id="path72622" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 986,197 v 1 h 1 v 4 h 1 v -4 h 1 v -1 z"
-         id="path72624" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 997,197 v 0.75 h -1 v 3.5 h 1 V 202 h 2 v -1 h -2 v -3 h 2 v -1 z"
-         id="path32207-9-7"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1001,197 v 1 h 2 v -1 z m 2,1 v 3 h 1 v -3 z m 0,3 h -2 v 1 h 2 z m -2,0 v -3 h -1 v 3 z"
-         id="path32209-6-7"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1005,197 v 5 h 1 v -2 h 1.25 v -1 h 0.75 v -1.25 h -1 V 197 h -1 z m 1,1 h 1 v 1 h -1 z"
-         id="path32211-3-7"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1009,197 v 2.25 h 1 V 202 h 1 v -2.75 h 1 V 197 h -1 v 2 h -1 v -2 z"
-         id="path32213-7-3"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1013,197 v 5 h 1 v -2 h 1 v 2 h 1 v -2 h -0.75 v -1 h 0.75 v -1.25 h -1 V 197 h -1 z m 1,1 h 1 v 1 h -1 z"
-         id="path32216-8-3"
-         inkscape:connector-curvature="0" />
-      <rect
-         y="197"
-         x="1017"
-         height="5"
-         width="1"
-         id="rect32218-8-5"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1020,197 v 0.75 h -1 v 3.5 h 1 V 202 h 2 v -1 -1.25 h -1 V 201 h -1 v -3 h 2 v -1 z"
-         id="path32220-2-9"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1023,197 v 5 h 1 v -2 h 1 v 2 h 1 v -5 h -1 v 2 h -1 v -2 z"
-         id="path32222-9-9"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1027,197 v 1 h 1 v 4 h 1 v -4 h 1 v -1 z"
-         id="path32224-1-8"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1035,197 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v 2 h 1 v -5 h -1 v 2 z"
-         id="path72678" />
-      <path
-         inkscape:connector-curvature="0"
-         id="path72680"
-         d="m 1045,197 v 1 h 2 v -1 z m 2,1 v 3 h 1 v -3 z m 0,3 h -2 v 1 h 2 z m -2,0 v -3 h -1 v 3 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6" />
-      <rect
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         id="rect72682"
-         width="1"
-         height="5"
-         x="1033"
-         y="197" />
-      <rect
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         id="rect72684"
-         width="1"
-         height="1"
-         x="1049"
-         y="201" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate;opacity:0.6"
-         d="m 1040,197 v 5 h 1 v -2 h 1 v -1 h -1 v -1 h 2 v -1 h -2 z"
-         id="path72686" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 1028.75,209 v 1 H 1028 v 4 h 1 v -2 h 1 v 2 h 1 v -4 h -0.75 v -1 z m 0.25,1 h 1 v 1 h -1 z"
-         id="path32244-2" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 1032,209 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v -1 z m 1,0 h 1 v 3 h 1 v -5 h -1 v 1 h -1 z"
-         id="path32246-0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 1038,209 v 5 h 1 1 v -0.5 h 1 V 212 h -0.75 v -1 h 0.75 v -1.25 h -1 V 209 h -1 z m 1,1 h 1 v 1 h -1 z m 0,2 h 1 v 1 h -1 z"
-         id="path32248-9" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 1042,209 v 5 h 1 v -2 h 1 v 2 h 1 v -2 h -0.75 v -1 h 0.75 v -1.25 h -1 V 209 h -1 z m 1,1 h 1 v 1 h -1 z"
-         id="path32250-7" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 1047,209 v 1 h 2 v -1 z m 2,1 v 3 h 1 v -3 z m 0,3 h -2 v 1 h 2 z m -2,0 v -3 h -1 v 3 z"
-         id="path32252-3" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 1051,209 v 1 h 2 v 1 h 1 v -1 -1 z m 2,2 h -1 v 1 h 1 z m -1,1 h -1 v 1 1 h 3 v -1 h -2 z"
-         id="path32254-7" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 997.75,209 v 1 H 997 v 4 h 1 v -2 h 1 v 2 h 1 v -4 h -0.75 v -1 z m 0.25,1 h 1 v 1 h -1 z"
-         id="path32256-2" />
-      <path
-         inkscape:connector-curvature="0"
-         id="path32258-6"
-         d="m 1001,209 v 5 h 1 v -3 h 1 v -1 h -1 v -1 z m 2,2 v 1 h 1 v 2 h 1 v -5 h -1 v 2 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate" />
+       transform="matrix(1.5,0,0,1.4999989,-192.5,-37.50017)"
+       id="g7067">
       <path
          inkscape:connector-curvature="0"
-         id="path32260-0"
-         d="m 1006,209 v 1 3 1 h 2 v -0.75 h 1 v -3.5 h -1 V 209 Z m 1,1 h 1 v 3 h -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate" />
+         style="opacity:0.2;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:1.99999988;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:105.70774078;stroke-opacity:1;marker:none;paint-order:fill markers stroke"
+         d="m 998.5,-179 h 121 c 1.385,0 2.5,1.115 2.5,2.5 v 66 c 0,1.385 -1.115,2.5 -2.5,2.5 h -121 c -1.385,0 -2.5,-1.115 -2.5,-2.5 v -66 c 0,-1.385 1.115,-2.5 2.5,-2.5 z"
+         id="path7065"
+         sodipodi:nodetypes="sssssssss" />
       <path
+         sodipodi:nodetypes="sccssccsccsssscssssssc"
          inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 1010,209 v 5 h 1 v -2 h 1 v 2 h 1 v -2 h -0.75 v -1 h 0.75 v -1.25 h -1 V 209 h -1 z m 1,1 h 1 v 1 h -1 z"
-         id="path32262-1" />
+         id="path7063"
+         transform="translate(803,-241)"
+         d="m 195.5,38 c -1.94642,0 -3.5,1.60898 -3.5,3.5 v 89 c 0,1.9217 1.5783,3.5 3.5,3.5 h 121 c 1.9217,0 3.5,-1.5783 3.5,-3.5 l -0.0147,-78.333361 c 0,-1.9217 -1.5783,-3.5 -3.5,-3.5 h -77 c -0.3983,10e-6 -0.77905,-0.157923 -1.06055,-0.439453 -0.2816,-0.2816 -0.44049,-0.662348 -0.43945,-1.060547 L 238,41.5 c 0.005,-1.921694 -1.5783,-3.5 -3.5,-3.5 z m -0.83333,27.333318 c 0,-0.8551 0.48503,-1.333334 1.33333,-1.333334 l 120,0 c 0.8483,0 1.33333,0.485034 1.33333,1.333334 v 64.666712 c 0,0.8483 -0.48503,1.33334 -1.33333,1.33334 l -120,0 c -0.8483,0 -1.33771,-0.48505 -1.33333,-1.33334"
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1.99999988;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:105.70774078;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate" />
       <path
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:5.98361349;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+         d="m 1059.6667,-155 7.2318,7.13082 c 2.8203,2.82167 7.0528,-1.4088 4.2325,-4.23047 l -12.0157,-12.01562 c -0.5714,-0.57187 -1.3499,-0.88819 -2.1582,-0.87696 l -10e-5,4e-5 c -0.7802,0.0111 -1.5251,0.32646 -2.0761,0.87891 l -11.9844,12.01758 c -2.9422,2.82368 1.4208,7.17459 4.2363,4.22461 L 1054.3333,-155 v 22.00002 c 0,4.42612 3.2913,8.00779 7.6802,8.00779 h 12 c 4.0871,0.10004 4.4069,-5.44117 0.3198,-5.34113 h -12 c -0.9323,0 -2.6654,-1.70279 -2.6666,-2.66666 z"
+         id="path15797-7"
          inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 1014,209 v 1 h 2 v 1 h 1 v -1 -1 z m 2,2 h -1 v 1 h 1 z m -1,1 h -1 v 1 1 h 3 v -1 h -2 z"
-         id="path32264-6" />
-      <path
-         inkscape:connector-curvature="0"
-         id="path32266-5"
-         d="m 1018,209 v 5 h 1 2 v -1 h -2 v -1 h 1 v -1 h -1 v -1 h 2 v -1 h -2 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate" />
+         sodipodi:nodetypes="cccccccccccsccscc" />
+    </g>
+    <path
+       style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:2.99947333;stroke-linecap:round;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:accumulate"
+       d="m 2238.2815,-365.92073 c -5.52,0.36326 -8.7584,3.47967 -8.7584,8.43246 0,4.43345 3.2458,8.13054 7.4769,8.85937 V -335 c 0,9.91268 -7.251,17.99291 -17,18 l -15.3901,0.0264 v 2.96484 H 2220 c 11.4103,-0.008 20.0341,-9.52115 20.0341,-20.99121 v -13.6289 c 4.2339,-0.72648 7.4827,-4.42393 7.4827,-8.85937 0,-4.95279 -3.4428,-8.31926 -9.2353,-8.43249 z m 0.406,2.98323 c 3.3303,0 5.8304,2.11776 5.8304,5.44923 0,3.33147 -2.6676,6 -5.9979,6 -3.3304,0 -5.9979,-2.66853 -5.9979,-6 0,-3.33147 2.835,-5.44923 6.1654,-5.44923 z M 2247,-291 c -4.4432,-10e-6 -8.6177,3.4839 -9,7 h -34 l 1,4 h 33 c 0.7202,4.24251 4.561,7.00001 9,7 4.9511,10e-6 9,-4.04722 9,-9 0,-4.95279 -4.0489,-9.00001 -9,-9 z m 0,3 c 3.3304,0 6,2.66853 6,6 0,3.33146 -2.6696,6 -6,6 -3.3303,0 -6,-2.66854 -6,-6 0,-3.33147 2.6697,-6 6,-6 z m -42,37 v 4 h 15 c 9.749,0.007 17,8.08732 17,18 v 13 c -4.2311,0.72882 -7.0144,4.57086 -7.0144,9.00433 0,4.9528 4.0457,9.00001 8.9969,8.99999 4.9511,2e-5 8.9968,-4.04719 8.9968,-8.99999 0,-4.43545 -2.7454,-8.27785 -6.9793,-9.00433 v -14 c 0,-11.47007 -9.5897,-20.99166 -21,-21 z m 33.9825,38.00433 c 3.3303,0 5.9979,2.66852 5.9979,6 0,3.33148 -2.6676,6 -5.9979,6 -3.3304,0 -5.9979,-2.66852 -5.9979,-6 0,-3.33148 2.6675,-6 5.9979,-6 z"
+       id="circle7536"
+       inkscape:connector-curvature="0"
+       sodipodi:nodetypes="cscsccccscscscsssccccccscscssscscscscscsccsssss" />
+    <g
+       transform="matrix(12.319682,0,0,12.172353,-5471.9739,-984.3403)"
+       id="g7322-0"
+       style="display:inline;stroke-width:0.12309717;enable-background:new">
       <path
          inkscape:connector-curvature="0"
-         id="path32268-7"
-         d="m 1024,209 v 4 h -1 v -1 h -1 v 1 h 0.75 v 1 h 1.5 v -1 h 0.75 v -4 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate" />
-      <rect
-         y="207"
-         x="1052"
-         height="1"
-         width="1"
-         id="rect32270-5"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate" />
+         style="display:inline;opacity:0.2;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:0.24619435;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
+         d="m 613.67414,61.612633 h 8.65173 c 0.57439,0 1.03681,0.46242 1.03681,1.036815 v 2.762566 c 0,0.574395 -0.46242,1.036815 -1.03681,1.036815 h -8.65173 c -0.57439,0 -1.03681,-0.46242 -1.03681,-1.036815 v -2.762566 c 0,-0.574395 0.46242,-1.036815 1.03681,-1.036815 z"
+         id="rect7418-6" />
       <path
+         style="display:inline;opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;stroke:none;stroke-width:2.41910744;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;enable-background:new"
+         d="M 145.24609,-711 C 139.56928,-711.01185 135,-706.4825 135,-700.87109 v 78 c 0,-5.61141 4.56928,-10.14076 10.24609,-10.12891 l 85.17749,0.17786 c 5.67681,0.0118 10.24609,4.5175 10.24609,10.12891 v -78 c 0,-5.61141 -4.56928,-10.11706 -10.24609,-10.12891 l -42.58875,-0.0889 z"
+         transform="matrix(0.10118267,0,0,0.10236222,598.97766,126.33073)"
+         id="rect7413-9"
          inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 969.96875,208 v 0.75 h -1 V 209 h -0.25 v 1 h -0.75 v 3 h 0.75 v 1 h 0.25 v 0.25 h 1 V 215 h 3 v -0.75 h 1 V 214 h 0.25 v -1 h 0.75 v -3 h -0.75 v -1 h -0.25 v -0.25 h -1 V 208 Z m 0,1 h 3 v 1 h 1 v 3 h -1 v 1 h -3 v -1 h -1 v -3 h 1 z"
-         id="path32272-1"
-         sodipodi:nodetypes="cccccccccccccccccccccccccccccccccccccccccc" />
+         sodipodi:nodetypes="sscsscsscs" />
       <path
          inkscape:connector-curvature="0"
-         id="path32274-2"
-         d="m 970.96875,210 v 0.75 h -1 v 1.5 h 1 V 213 h 1.25 v -1 h -1.25 v -1 h 1.25 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         sodipodi:nodetypes="ccccccccccccc" />
-      <path
-         id="path72688"
-         d="m 981.71875,209 v 1 h -0.75 v 3 h 0.75 v 1 h 1.5 v -1 h 0.75 v -3 h -0.75 v -1 z m 0.25,1 h 1 v 3 h -1 z m 0.25,1 v 1 h 0.5 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         inkscape:connector-curvature="0" />
-      <path
-         id="path72690"
-         d="m 985.96875,209 v 1 h -1 v 1 h 1 v 3 h 1 v -5 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         inkscape:connector-curvature="0" />
-      <path
-         inkscape:connector-curvature="0"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 976.96875,209 v 1 h 2 v 1 h -1 v 1 h 1 v -0.75 h 1 v -1.5 h -1 V 209 Z m 1,3 h -1 v 1 1 h 3 v -1 h -2 z"
-         id="path72694" />
-      <path
-         id="path72696"
-         d="m 992.96875,210 v 1 h 1 v -1 z m -0.21875,3 v 1 H 992 v 1 h 1 v -1 h 1 v -1 z"
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         inkscape:connector-curvature="0" />
-      <path
-         style="display:inline;overflow:visible;visibility:visible;fill:#000000;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:1;marker:none;enable-background:accumulate"
-         d="m 988.75,209 v 1 h -0.5 v 1 h 0.5 v 1 H 988 v 1 h 0.75 v 1 h 1.5 v -1 H 991 v -1 h -0.75 v -1 h 0.5 v -1 h -0.5 v -1 z m 0.25,1 h 1 v 1 h -1 z m 0,2 h 1 v 1 h -1 z"
-         id="path19551-18-1"
-         inkscape:connector-curvature="0" />
+         style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:2.00080741;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+         d="m 717.48242,42.996094 c -5.24135,0 -9.49414,4.190752 -9.49414,9.355468 v 65.017578 22.26953 c 0,5.16472 4.25279,9.35547 9.49414,9.35547 h 70.02539 c 5.24136,0 9.49414,-4.19075 9.49414,-9.35547 V 117.36914 52.351562 c 0,-5.164716 -4.25278,-9.355468 -9.49414,-9.355468 z m 0,1.984375 h 70.02539 c 4.15996,0 7.48047,3.271903 7.48047,7.371093 v 59.269528 c -1.73801,-2.19271 -4.43984,-3.60742 -7.48047,-3.60742 h -70.02539 c -3.04062,0 -5.74244,1.41471 -7.48047,3.60742 V 52.351562 c 0,-4.09919 3.32052,-7.371093 7.48047,-7.371093 z m 0,65.017581 h 70.02539 c 4.15996,0 6.83699,3.09748 6.83699,7.19666 v 22.52458 c 0,4.09919 -3.14714,6.62488 -7.3071,6.62488 h -69.74951 c -4.15995,0 -6.64281,-2.52569 -6.64281,-6.62488 v -22.52458 c 0,-4.09918 2.67709,-7.19666 6.83704,-7.19666 z"
+         transform="matrix(0.12219364,0,0,0.1240074,526.00203,48.095825)"
+         id="rect7312-4"
+         sodipodi:nodetypes="sscsssscsssssscsscssssssssscs" />
+      <g
+         id="g7411-1"
+         transform="matrix(-1,0,0,1,1235.8994,0)"
+         style="stroke-width:0.12309717">
+        <path
+           style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:0.12309717;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;marker:none;paint-order:fill markers stroke;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+           d="m 613.88871,62.499042 c -0.23599,0 -0.42773,0.193143 -0.42773,0.429687 v 0.0098 c 0,0.236544 0.19174,0.429687 0.42773,0.429687 0.23599,0 0.42774,-0.193143 0.42774,-0.429687 v -0.0098 c 0,-0.236544 -0.19175,-0.429687 -0.42774,-0.429687 z"
+           id="rect7316-4"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="sssssss" />
+        <g
+           transform="translate(6.2733256)"
+           id="g7368-5"
+           style="display:inline;stroke-width:0.12309717;enable-background:new">
+          <path
+             style="color:#000000;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-size:medium;line-height:normal;font-family:sans-serif;font-variant-ligatures:normal;font-variant-position:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-alternates:normal;font-feature-settings:normal;text-indent:0;text-align:start;text-decoration:none;text-decoration-line:none;text-decoration-style:solid;text-decoration-color:#000000;letter-spacing:normal;word-spacing:normal;text-transform:none;writing-mode:lr-tb;direction:ltr;text-orientation:mixed;dominant-baseline:auto;baseline-shift:baseline;text-anchor:start;white-space:normal;shape-padding:0;clip-rule:nonzero;display:inline;overflow:visible;visibility:visible;opacity:1;isolation:auto;mix-blend-mode:normal;color-interpolation:sRGB;color-interpolation-filters:linearRGB;solid-color:#000000;solid-opacity:1;vector-effect:none;fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:0.24619435;stroke-linecap:round;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1;color-rendering:auto;image-rendering:auto;shape-rendering:auto;text-rendering:auto;enable-background:new"
+             d="m 613.40672,62.503836 c -0.0633,0.0071 -0.11074,0.06131 -0.10937,0.125 v 2.851562 c 0,0.165367 0.24805,0.165367 0.24805,0 v -2.851562 c 0.002,-0.07532 -0.0639,-0.13455 -0.13868,-0.125 z m 0.64937,3e-6 c -0.0633,0.0071 -0.11074,0.06131 -0.10937,0.125 v 2.851562 c -0.005,0.169458 0.25158,0.169458 0.24609,0 v -2.851562 c 0.002,-0.07444 -0.0627,-0.133249 -0.13672,-0.125 z m 0.64936,3e-6 c -0.0633,0.0071 -0.11074,0.06131 -0.10937,0.125 v 2.851562 c -0.005,0.169458 0.25158,0.169458 0.24609,0 v -2.851562 c 0.002,-0.07444 -0.0627,-0.133249 -0.13672,-0.125 z m 0.64938,3e-6 c -0.0633,0.0071 -0.11075,0.06131 -0.10938,0.125 v 2.851562 c -0.005,0.169248 0.25136,0.169248 0.2461,0 v -2.851562 c 0.002,-0.07444 -0.0627,-0.133249 -0.13672,-0.125 z m 0.64937,-9e-6 c -0.0633,0.0071 -0.11075,0.06131 -0.10938,0.125 v 2.851562 c -0.005,0.169248 0.25136,0.169248 0.2461,0 v -2.851562 c 0.002,-0.07444 -0.0627,-0.133249 -0.13672,-0.125 z"
+             id="path7359-7"
+             inkscape:connector-curvature="0"
+             sodipodi:nodetypes="ccsscccccccccccccccccccccccccc" />
+        </g>
+      </g>
     </g>
   </g>
 </svg>
diff --git a/release/datafiles/prvicons_update.py b/release/datafiles/prvicons_update.py
index bc170b98545..96306f0cea9 100755..100644
--- a/release/datafiles/prvicons_update.py
+++ b/release/datafiles/prvicons_update.py
@@ -1,23 +1,25 @@
-#!/usr/bin/env python3
-
-# This script updates icons from the SVG file
-import os
-import subprocess
-import sys
-
-BASEDIR = os.path.abspath(os.path.dirname(__file__))
-
-inkscape_path = 'inkscape'
-
-if sys.platform == 'darwin':
-    inkscape_app_path = '/Applications/Inkscape.app/Contents/Resources/script'
-    if os.path.exists(inkscape_app_path):
-        inkscape_path = inkscape_app_path
-
-cmd = (
-    inkscape_path,
-    os.path.join(BASEDIR, "prvicons.svg"),
-    "--without-gui",
-    "--export-png=" + os.path.join(BASEDIR, "prvicons.png"),
-)
-subprocess.check_call(cmd)
+#!/usr/bin/env python3

+

+# This script updates icons from the SVG file

+import os

+import subprocess

+import sys

+

+BASEDIR = os.path.abspath(os.path.dirname(__file__))

+

+inkscape_path = 'inkscape'

+

+if sys.platform == 'darwin':

+    inkscape_app_path = '/Applications/Inkscape.app/Contents/Resources/script'

+    if os.path.exists(inkscape_app_path):

+        inkscape_path = inkscape_app_path

+

+cmd = (

+    inkscape_path,

+    os.path.join(BASEDIR, "prvicons.svg"),

+    "--export-width=1536",

+    "--export-height=256",

+    "--without-gui",

+    "--export-png=" + os.path.join(BASEDIR, "prvicons.png"),

+)

+subprocess.check_call(cmd)

diff --git a/release/datafiles/userdef/userdef_default.c b/release/datafiles/userdef/userdef_default.c
index 99c8cd72dba..4ad99993ada 100644
--- a/release/datafiles/userdef/userdef_default.c
+++ b/release/datafiles/userdef/userdef_default.c
@@ -26,6 +26,8 @@
 
 #include "BKE_blender_version.h"
 
+#include "BLO_readfile.h" /* own include */
+
 const UserDef U_default = {
     .versionfile = BLENDER_VERSION,
     .subversionfile = BLENDER_SUBVERSION,
diff --git a/release/datafiles/userdef/userdef_default_theme.c b/release/datafiles/userdef/userdef_default_theme.c
index 8611a9ae881..e9a85517647 100644
--- a/release/datafiles/userdef/userdef_default_theme.c
+++ b/release/datafiles/userdef/userdef_default_theme.c
@@ -370,7 +370,7 @@ const bTheme U_theme_default = {
     .paint_curve_handle = RGBA(0x7fff7f7f),
   },
   .space_file = {
-    .back = RGBA(0x35353500),
+    .back = RGBA(0x33333300),
     .title = RGBA(0xffffffff),
     .text = RGBA(0xe6e6e6ff),
     .text_hi = RGBA(0xffffffff),
@@ -381,13 +381,14 @@ const bTheme U_theme_default = {
     .tab_inactive = RGBA(0x2b2b2bff),
     .tab_back = RGBA(0x232323ff),
     .tab_outline = RGBA(0x232323ff),
-    .button = RGBA(0x424242ff),
+    .button = RGBA(0x4b4b4bff),
     .button_title = RGBA(0xffffffff),
     .button_text = RGBA(0xe5e5e5ff),
     .button_text_hi = RGBA(0xffffffff),
+    .execution_buts = RGBA(0x444444ff),
     .panelcolors = {
-      .header = RGBA(0x424242cc),
-      .back = RGBA(0x333333b3),
+      .header = RGBA(0x4b4b4bff),
+      .back = RGBA(0x404040ff),
       .sub_back = RGBA(0x0000003e),
     },
     .hilite = RGBA(0x4f76b3ff),
@@ -748,7 +749,8 @@ const bTheme U_theme_default = {
     .outline_width = 1,
     .facedot_size = 4,
     .match = RGBA(0x337f334c),
-    .selected_highlight = RGBA(0x314e784c),
+    .selected_highlight = RGBA(0x223a5bff),
+    .active = RGBA(0x3b5689ff),
     .selected_object = RGBA(0xe96a00ff),
     .active_object = RGBA(0xffaf29ff),
     .edited_object = RGBA(0x00806266),
diff --git a/release/scripts/modules/addon_utils.py b/release/scripts/modules/addon_utils.py
index 376193f73a5..e212df17f60 100644
--- a/release/scripts/modules/addon_utils.py
+++ b/release/scripts/modules/addon_utils.py
@@ -467,7 +467,7 @@ def reset_all(*, reload_scripts=False):
     paths_list = paths()
 
     for path in paths_list:
-        _bpy.utils._sys_path_ensure(path)
+        _bpy.utils._sys_path_ensure_append(path)
         for mod_name, _mod_path in _bpy.path.module_names(path):
             is_enabled, is_loaded = check(mod_name)
 
diff --git a/release/scripts/modules/bl_i18n_utils/utils.py b/release/scripts/modules/bl_i18n_utils/utils.py
index bbc0c5c8405..2cca4171193 100644
--- a/release/scripts/modules/bl_i18n_utils/utils.py
+++ b/release/scripts/modules/bl_i18n_utils/utils.py
@@ -834,7 +834,7 @@ class I18nMessages:
     def parse_messages_from_po(self, src, key=None):
         """
         Parse a po file.
-        Note: This function will silently "arrange" mis-formated entries, thus using afterward write_messages() should
+        Note: This function will silently "arrange" mis-formatted entries, thus using afterward write_messages() should
               always produce a po-valid file, though not correct!
         """
         reading_msgid = False
diff --git a/release/scripts/modules/bl_ui_utils/bug_report_url.py b/release/scripts/modules/bl_ui_utils/bug_report_url.py
index be94b45c8ac..008eafc2c46 100644
--- a/release/scripts/modules/bl_ui_utils/bug_report_url.py
+++ b/release/scripts/modules/bl_ui_utils/bug_report_url.py
@@ -64,10 +64,7 @@ def url_prefill_from_blender(addon_info=None):
             "\n"
             "**Addon Information**\n"
         )
-        fh.write((
-            "Name: {name} {version}\n"
-            "Author: {author}\n").format(**addon_info)
-        )
+        fh.write(addon_info)
 
     fh.write(
         "\n"
diff --git a/release/scripts/modules/bpy_extras/mesh_utils.py b/release/scripts/modules/bpy_extras/mesh_utils.py
index 41727565cfa..1576947b8b4 100644
--- a/release/scripts/modules/bpy_extras/mesh_utils.py
+++ b/release/scripts/modules/bpy_extras/mesh_utils.py
@@ -241,9 +241,9 @@ def edge_loops_from_edges(mesh, edges=None):
 
 def ngon_tessellate(from_data, indices, fix_loops=True, debug_print=True):
     """
-    Takes a polyline of indices (fgon) and returns a list of face
+    Takes a polyline of indices (ngon) and returns a list of face
     index lists. Designed to be used for importers that need indices for an
-    fgon to create from existing verts.
+    ngon to create from existing verts.
 
     :arg from_data: either a mesh, or a list/tuple of vectors.
     :type from_data: list or :class:`bpy.types.Mesh`
diff --git a/release/scripts/modules/console/complete_import.py b/release/scripts/modules/console/complete_import.py
index ff3099d7285..6f51f6b691b 100644
--- a/release/scripts/modules/console/complete_import.py
+++ b/release/scripts/modules/console/complete_import.py
@@ -178,7 +178,7 @@ def complete(line):
     words = line.split(' ')
     if len(words) == 3 and words[0] == 'from':
         return ['import ']
-    if len(words) < 3 and (words[0] in ['import', 'from']):
+    if len(words) < 3 and (words[0] in {'import', 'from'}):
         if len(words) == 1:
             return get_root_modules()
         mod = words[1].split('.')
diff --git a/release/scripts/modules/rna_manual_reference.py b/release/scripts/modules/rna_manual_reference.py
new file mode 100644
index 00000000000..a09f1f2285f
--- /dev/null
+++ b/release/scripts/modules/rna_manual_reference.py
@@ -0,0 +1,856 @@
+# Do not edit this file. This file is auto generated from rna_manual_reference_updater.py
+
+import bpy
+
+if bpy.app.version_cycle in {'rc', 'release'}:
+    manual_version = '%d.%d' % bpy.app.version[:2]
+else:
+    manual_version = 'dev'
+
+url_manual_prefix = "https://docs.blender.org/manual/en/" + manual_version + "/"
+
+language = ""
+if bpy.context.preferences.view.use_international_fonts:
+    language = bpy.context.preferences.view.language
+    if language == 'DEFAULT':
+        import os
+        language = os.getenv('LANG', '').split('.')[0]
+
+LANG = {
+    "de_DE":        "de",
+    "ru_RU":        "ru",
+    "uk_UA":        "uk",
+    "es":           "es",
+    "fr_FR":        "fr",
+    "it_IT":        "it",
+    "ja_JP":        "ja",
+    "ko_KR":        "ko",
+    "pt_PT":        "pt",
+    "pt_BR":        "pt",
+    "vi_VN":        "vi",
+    "zh_CN":        "zh-hans",
+    "zh_TW":        "zh-hant",
+}.get(language)
+
+if LANG is not None:
+    url_manual_prefix = url_manual_prefix.replace("manual/en", "manual/" + LANG)
+
+url_manual_mapping = (
+	("bpy.types.cyclesobjectsettings.use_adaptive_subdivision*", "render/cycles/object_settings/adaptive_subdiv.html#bpy-types-cyclesobjectsettings-use-adaptive-subdivision"),
+	("bpy.types.cyclesrendersettings.offscreen_dicing_scale*", "render/cycles/render_settings/subdivision.html#bpy-types-cyclesrendersettings-offscreen-dicing-scale"),
+	("bpy.types.linestylegeometrymodifier_backbonestretcher*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/backbone_stretcher.html#bpy-types-linestylegeometrymodifier-backbonestretcher"),
+	("bpy.types.linestylegeometrymodifier_sinusdisplacement*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/sinus_displacement.html#bpy-types-linestylegeometrymodifier-sinusdisplacement"),
+	("bpy.types.linestylegeometrymodifier_polygonalization*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/polygonization.html#bpy-types-linestylegeometrymodifier-polygonalization"),
+	("bpy.types.cyclesrendersettings.distance_cull_margin*", "render/cycles/render_settings/simplify.html#bpy-types-cyclesrendersettings-distance-cull-margin"),
+	("bpy.types.linestylegeometrymodifier_simplification*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/simplification.html#bpy-types-linestylegeometrymodifier-simplification"),
+	("bpy.types.cyclesrendersettings.camera_cull_margin*", "render/cycles/render_settings/simplify.html#bpy-types-cyclesrendersettings-camera-cull-margin"),
+	("bpy.types.linestylegeometrymodifier_perlinnoise1d*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/perlin_noise_1d.html#bpy-types-linestylegeometrymodifier-perlinnoise1d"),
+	("bpy.types.linestylegeometrymodifier_perlinnoise2d*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/perlin_noise_2d.html#bpy-types-linestylegeometrymodifier-perlinnoise2d"),
+	("bpy.types.smokedomainsettings.use_high_resolution*", "render/cycles/render_settings/simplify.html#bpy-types-smokedomainsettings-use-high-resolution"),
+	("bpy.types.cyclesrendersettings.use_distance_cull*", "render/cycles/render_settings/simplify.html#bpy-types-cyclesrendersettings-use-distance-cull"),
+	("bpy.types.linestylegeometrymodifier_guidinglines*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/guiding_lines.html#bpy-types-linestylegeometrymodifier-guidinglines"),
+	("bpy.types.linestylegeometrymodifier_spatialnoise*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/spatial_noise.html#bpy-types-linestylegeometrymodifier-spatialnoise"),
+	("bpy.types.linestylethicknessmodifier_calligraphy*", "render/freestyle/parameter_editor/line_style/modifiers/thickness/calligraphy.html#bpy-types-linestylethicknessmodifier-calligraphy"),
+	("bpy.types.cyclesrendersettings.max_subdivisions*", "render/cycles/render_settings/subdivision.html#bpy-types-cyclesrendersettings-max-subdivisions"),
+	("bpy.types.linestyle*modifier_distancefromcamera*", "render/freestyle/parameter_editor/line_style/modifiers/color/distance_from_camera.html#bpy-types-linestyle-modifier-distancefromcamera"),
+	("bpy.types.linestyle*modifier_distancefromobject*", "render/freestyle/parameter_editor/line_style/modifiers/color/distance_from_object.html#bpy-types-linestyle-modifier-distancefromobject"),
+	("bpy.types.linestylegeometrymodifier_2dtransform*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/2d_transform.html#bpy-types-linestylegeometrymodifier-2dtransform"),
+	("bpy.types.linestylegeometrymodifier_beziercurve*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/bezier_curve.html#bpy-types-linestylegeometrymodifier-beziercurve"),
+	("bpy.types.cyclesrendersettings.use_camera_cull*", "render/cycles/render_settings/simplify.html#bpy-types-cyclesrendersettings-use-camera-cull"),
+	("bpy.types.linestylegeometrymodifier_tipremover*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/tip_remover.html#bpy-types-linestylegeometrymodifier-tipremover"),
+	("bpy.types.cyclesmaterialsettings.displacement*", "render/cycles/material_settings.html#bpy-types-cyclesmaterialsettings-displacement"),
+	("bpy.types.linestylegeometrymodifier_blueprint*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/blueprint.html#bpy-types-linestylegeometrymodifier-blueprint"),
+	("bpy.types.rendersettings.simplify_subdivision*", "render/cycles/render_settings/simplify.html#bpy-types-rendersettings-simplify-subdivision"),
+	("bpy.types.cyclesrendersettings.dicing_camera*", "render/cycles/render_settings/subdivision.html#bpy-types-cyclesrendersettings-dicing-camera"),
+	("bpy.types.cyclesrendersettings.texture_limit*", "render/cycles/render_settings/simplify.html#bpy-types-cyclesrendersettings-texture-limit"),
+	("bpy.types.linestylegeometrymodifier_2doffset*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/2d_offset.html#bpy-types-linestylegeometrymodifier-2doffset"),
+	("bpy.types.linestylegeometrymodifier_sampling*", "render/freestyle/parameter_editor/line_style/modifiers/geometry/sampling.html#bpy-types-linestylegeometrymodifier-sampling"),
+	("bpy.types.cyclesrendersettings.*dicing_rate*", "render/cycles/render_settings/subdivision.html#bpy-types-cyclesrendersettings-dicing-rate"),
+	("bpy.types.rendersettings.use_file_extension*", "render/output/settings.html#bpy-types-rendersettings-use-file-extension"),
+	("bpy.types.spaceview3d.transform_orientation*", "scene_layout/object/editing/transform/control/orientations.html#bpy-types-spaceview3d-transform-orientation"),
+	("bpy.ops.object.constraint_add_with_targets*", "animation/constraints/interface/adding_removing.html#bpy-ops-object-constraint-add-with-targets"),
+	("bpy.types.cyclesobjectsettings.dicing_rate*", "render/cycles/object_settings/adaptive_subdiv.html#bpy-types-cyclesobjectsettings-dicing-rate"),
+	("bpy.types.posebone.use_ik_rotation_control*", "animation/armatures/posing/bone_constraints/inverse_kinematics/introduction.html#bpy-types-posebone-use-ik-rotation-control"),
+	("bpy.ops.constraint.disable_keep_transform*", "animation/constraints/interface/common.html#bpy-ops-constraint-disable-keep-transform"),
+	("bpy.types.imagepaint.use_backface_culling*", "sculpt_paint/texture_paint/tool_settings/options.html#bpy-types-imagepaint-use-backface-culling"),
+	("bpy.types.linestyle*modifier_curvature_3d*", "render/freestyle/parameter_editor/line_style/modifiers/color/curvature_3d.html#bpy-types-linestyle-modifier-curvature-3d"),
+	("bpy.types.rendersettings.use_render_cache*", "render/output/settings.html#bpy-types-rendersettings-use-render-cache"),
+	("bpy.types.sceneeevee.use_taa_reprojection*", "render/eevee/render_settings/sampling.html#bpy-types-sceneeevee-use-taa-reprojection"),
+	("bpy.ops.object.anim_transforms_to_deltas*", "scene_layout/object/editing/transform/clear_apply.html#bpy-ops-object-anim-transforms-to-deltas"),
+	("bpy.types.compositornodeplanetrackdeform*", "compositing/types/distort/plane_track_deform.html#bpy-types-compositornodeplanetrackdeform"),
+	("bpy.types.linestyle*modifier_alongstroke*", "render/freestyle/parameter_editor/line_style/modifiers/color/along_stroke.html#bpy-types-linestyle-modifier-alongstroke"),
+	("bpy.types.linestyle*modifier_creaseangle*", "render/freestyle/parameter_editor/line_style/modifiers/color/crease_angle.html#bpy-types-linestyle-modifier-creaseangle"),
+	("bpy.types.linestylecolormodifier_tangent*", "render/freestyle/parameter_editor/line_style/modifiers/color/tangent.html#bpy-types-linestylecolormodifier-tangent"),
+	("bpy.types.rendersettings.use_placeholder*", "render/output/settings.html#bpy-types-rendersettings-use-placeholder"),
+	("bpy.types.shadernodesubsurfacescattering*", "render/shader_nodes/shader/sss.html#bpy-types-shadernodesubsurfacescattering"),
+	("bpy.types.compositornodecolorcorrection*", "compositing/types/color/color_correction.html#bpy-types-compositornodecolorcorrection"),
+	("bpy.types.compositornodemoviedistortion*", "compositing/types/distort/movie_distortion.html#bpy-types-compositornodemoviedistortion"),
+	("bpy.types.ffmpegsettings.audio_channels*", "scene_layout/scene/properties.html#bpy-types-ffmpegsettings-audio-channels"),
+	("bpy.types.fmodifierenvelopecontrolpoint*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fmodifierenvelopecontrolpoint"),
+	("bpy.types.material.use_sss_translucency*", "render/eevee/materials/settings.html#bpy-types-material-use-sss-translucency"),
+	("bpy.types.sceneeevee.taa_render_samples*", "render/eevee/render_settings/sampling.html#bpy-types-sceneeevee-taa-render-samples"),
+	("bpy.types.spaceuveditor.pixel_snap_mode*", "modeling/meshes/editing/uv/layout.html#bpy-types-spaceuveditor-pixel-snap-mode"),
+	("bpy.types.spaceuveditor.use_live_unwrap*", "modeling/meshes/editing/uv/layout.html#bpy-types-spaceuveditor-use-live-unwrap"),
+	("bpy.types.vertexweightproximitymodifier*", "modeling/modifiers/modify/weight_proximity.html#bpy-types-vertexweightproximitymodifier"),
+	("bpy.types.compositornodebrightcontrast*", "compositing/types/color/bright_contrast.html#bpy-types-compositornodebrightcontrast"),
+	("bpy.types.compositornodedoubleedgemask*", "compositing/types/matte/double_edge_mask.html#bpy-types-compositornodedoubleedgemask"),
+	("bpy.types.ffmpegsettings.audio_mixrate*", "scene_layout/scene/properties.html#bpy-types-ffmpegsettings-audio-mixrate"),
+	("bpy.types.material.preview_render_type*", "render/materials/preview.html#bpy-types-material-preview-render-type"),
+	("bpy.types.rendersettings.use_overwrite*", "render/output/settings.html#bpy-types-rendersettings-use-overwrite"),
+	("bpy.types.sceneeevee.volumetric_shadow*", "render/eevee/render_settings/volumetrics.html#bpy-types-sceneeevee-volumetric-shadow"),
+	("bpy.types.shadernodebsdfhairprincipled*", "render/shader_nodes/shader/hair_principled.html#bpy-types-shadernodebsdfhairprincipled"),
+	("bpy.types.shadernodevectordisplacement*", "render/shader_nodes/vector/vector_displacement.html#bpy-types-shadernodevectordisplacement"),
+	("bpy.types.spacegrapheditor.show_cursor*", "editors/graph_editor/fcurves/properties.html#bpy-types-spacegrapheditor-show-cursor"),
+	("bpy.types.spaceimageeditor.show_repeat*", "editors/image/view_tab.html#bpy-types-spaceimageeditor-show-repeat"),
+	("bpy.ops.object.visual_transform_apply*", "scene_layout/object/editing/transform/clear_apply.html#bpy-ops-object-visual-transform-apply"),
+	("bpy.types.brush.texture_overlay_alpha*", "sculpt_paint/brush/display.html#bpy-types-brush-texture-overlay-alpha"),
+	("bpy.types.compositornodebilateralblur*", "compositing/types/filter/bilateral_blur.html#bpy-types-compositornodebilateralblur"),
+	("bpy.types.compositornodedistancematte*", "compositing/types/matte/distance_key.html#bpy-types-compositornodedistancematte"),
+	("bpy.types.imagepaint.screen_grab_size*", "sculpt_paint/texture_paint/tool_settings/options.html#bpy-types-imagepaint-screen-grab-size"),
+	("bpy.types.linestyle*modifier_material*", "render/freestyle/parameter_editor/line_style/modifiers/color/material.html#bpy-types-linestyle-modifier-material"),
+	("bpy.types.particlesettingstextureslot*", "physics/particles/texture_influence.html#bpy-types-particlesettingstextureslot"),
+	("bpy.types.posebone.ik_rotation_weight*", "animation/armatures/posing/bone_constraints/inverse_kinematics/introduction.html#bpy-types-posebone-ik-rotation-weight"),
+	("bpy.types.sceneeevee.volumetric_light*", "render/eevee/render_settings/volumetrics.html#bpy-types-sceneeevee-volumetric-light"),
+	("bpy.ops.mesh.normals_make_consistent*", "modeling/meshes/editing/normals.html#bpy-ops-mesh-normals-make-consistent"),
+	("bpy.ops.object.duplicate_move_linked*", "scene_layout/object/editing/duplication.html#bpy-ops-object-duplicate-move-linked"),
+	("bpy.ops.view3d.localview_remove_from*", "editors/3dview/navigate/views.html#bpy-ops-view3d-localview-remove-from"),
+	("bpy.types.brush.cursor_overlay_alpha*", "sculpt_paint/brush/display.html#bpy-types-brush-cursor-overlay-alpha"),
+	("bpy.types.brush.topology_rake_factor*", "sculpt_paint/sculpting/tool_settings/dyntopo.html#bpy-types-brush-topology-rake-factor"),
+	("bpy.types.compositornodechannelmatte*", "compositing/types/matte/channel_key.html#bpy-types-compositornodechannelmatte"),
+	("bpy.types.compositornodecolorbalance*", "compositing/types/color/color_balance.html#bpy-types-compositornodecolorbalance"),
+	("bpy.types.compositornodekeyingscreen*", "compositing/types/matte/keying_screen.html#bpy-types-compositornodekeyingscreen"),
+	("bpy.types.dynamicpaintcanvassettings*", "physics/dynamic_paint/canvas.html#bpy-types-dynamicpaintcanvassettings"),
+	("bpy.types.fmodifierfunctiongenerator*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fmodifierfunctiongenerator"),
+	("bpy.types.movietrackingstabilization*", "movie_clip/tracking/clip/properties/stabilization/index.html#bpy-types-movietrackingstabilization"),
+	("bpy.types.object.display_bounds_type*", "scene_layout/object/properties/display.html#bpy-types-object-display-bounds-type"),
+	("bpy.types.shadernodeambientocclusion*", "render/shader_nodes/input/ao.html#bpy-types-shadernodeambientocclusion"),
+	("bpy.types.shadernodevolumeabsorption*", "render/shader_nodes/shader/volume_absorption.html#bpy-types-shadernodevolumeabsorption"),
+	("bpy.types.shadernodevolumeprincipled*", "render/shader_nodes/shader/volume_principled.html#bpy-types-shadernodevolumeprincipled"),
+	("bpy.types.toolsettings.use_uv_sculpt*", "modeling/meshes/editing/uv/uv_sculpt.html#bpy-types-toolsettings-use-uv-sculpt"),
+	("bpy.ops.mesh.set_normals_from_faces*", "modeling/meshes/editing/normals.html#bpy-ops-mesh-set-normals-from-faces"),
+	("bpy.ops.object.duplicates_make_real*", "scene_layout/object/editing/transform/clear_apply.html#bpy-ops-object-duplicates-make-real"),
+	("bpy.ops.object.transforms_to_deltas*", "scene_layout/object/editing/transform/clear_apply.html#bpy-ops-object-transforms-to-deltas"),
+	("bpy.types.brush.use_primary_overlay*", "sculpt_paint/brush/display.html#bpy-types-brush-use-primary-overlay"),
+	("bpy.types.compositornodechromamatte*", "compositing/types/matte/chroma_key.html#bpy-types-compositornodechromamatte"),
+	("bpy.types.compositornodedilateerode*", "compositing/types/filter/dilate_erode.html#bpy-types-compositornodedilateerode"),
+	("bpy.types.compositornodeellipsemask*", "compositing/types/matte/ellipse_mask.html#bpy-types-compositornodeellipsemask"),
+	("bpy.types.compositornodesplitviewer*", "compositing/types/output/split_viewer.html#bpy-types-compositornodesplitviewer"),
+	("bpy.types.curve.use_uv_as_generated*", "editors/uv/generated_uvs.html#bpy-types-curve-use-uv-as-generated"),
+	("bpy.types.dynamicpaintbrushsettings*", "physics/dynamic_paint/brush.html#bpy-types-dynamicpaintbrushsettings"),
+	("bpy.types.rendersettings.use_border*", "render/output/settings.html#bpy-types-rendersettings-use-border"),
+	("bpy.types.sceneeevee.bokeh_max_size*", "render/eevee/render_settings/depth_of_field.html#bpy-types-sceneeevee-bokeh-max-size"),
+	("bpy.types.shadernodebsdfanisotropic*", "render/shader_nodes/shader/anisotropic.html#bpy-types-shadernodebsdfanisotropic"),
+	("bpy.types.shadernodebsdftranslucent*", "render/shader_nodes/shader/translucent.html#bpy-types-shadernodebsdftranslucent"),
+	("bpy.types.shadernodebsdftransparent*", "render/shader_nodes/shader/transparent.html#bpy-types-shadernodebsdftransparent"),
+	("bpy.types.shadernodevectortransform*", "render/shader_nodes/vector/transform.html#bpy-types-shadernodevectortransform"),
+	("bpy.types.spaceuveditor.lock_bounds*", "modeling/meshes/editing/uv/layout.html#bpy-types-spaceuveditor-lock-bounds"),
+	("bpy.types.spline.tilt_interpolation*", "modeling/curves/properties/active_spline.html#bpy-types-spline-tilt-interpolation"),
+	("bpy.ops.mesh.quads_convert_to_tris*", "modeling/meshes/editing/faces.html#bpy-ops-mesh-quads-convert-to-tris"),
+	("bpy.ops.node.read_fullsamplelayers*", "interface/controls/nodes/editing.html#bpy-ops-node-read-fullsamplelayers"),
+	("bpy.ops.object.datalayout_transfer*", "modeling/meshes/editing/data_transfer.html#bpy-ops-object-datalayout-transfer"),
+	("bpy.ops.object.randomize_transform*", "scene_layout/object/editing/transform/tools.html#bpy-ops-object-randomize-transform"),
+	("bpy.ops.object.vertex_group_levels*", "sculpt_paint/weight_paint/editing.html#bpy-ops-object-vertex-group-levels"),
+	("bpy.ops.sequencer.crossfade_sounds*", "video_editing/sequencer/strips/transitions/cross.html#bpy-ops-sequencer-crossfade-sounds"),
+	("bpy.ops.transform.edge_bevelweight*", "modeling/meshes/editing/edges.html#bpy-ops-transform-edge-bevelweight"),
+	("bpy.ops.wm.previews_batch_generate*", "files/blend/previews.html#bpy-ops-wm-previews-batch-generate"),
+	("bpy.types.brush.use_cursor_overlay*", "sculpt_paint/brush/display.html#bpy-types-brush-use-cursor-overlay"),
+	("bpy.types.compositornodebokehimage*", "compositing/types/input/bokeh_image.html#bpy-types-compositornodebokehimage"),
+	("bpy.types.compositornodecolormatte*", "compositing/types/matte/color_key.html#bpy-types-compositornodecolormatte"),
+	("bpy.types.compositornodecolorspill*", "compositing/types/matte/color_spill.html#bpy-types-compositornodecolorspill"),
+	("bpy.types.compositornodehuecorrect*", "compositing/types/color/hue_correct.html#bpy-types-compositornodehuecorrect"),
+	("bpy.types.compositornodeoutputfile*", "compositing/types/output/file.html#bpy-types-compositornodeoutputfile"),
+	("bpy.types.compositornodeswitchview*", "compositing/types/converter/switch_view.html#bpy-types-compositornodeswitchview"),
+	("bpy.types.copytransformsconstraint*", "animation/constraints/transform/copy_transforms.html#bpy-types-copytransformsconstraint"),
+	("bpy.types.correctivesmoothmodifier*", "modeling/modifiers/deform/corrective_smooth.html#bpy-types-correctivesmoothmodifier"),
+	("bpy.types.cyclesvisibilitysettings*", "render/cycles/object_settings/object_data.html#bpy-types-cyclesvisibilitysettings"),
+	("bpy.types.imagepaint.interpolation*", "sculpt_paint/texture_paint/tool_settings/texture_slots.html#bpy-types-imagepaint-interpolation"),
+	("bpy.types.linestyle*modifier_noise*", "render/freestyle/parameter_editor/line_style/modifiers/color/noise.html#bpy-types-linestyle-modifier-noise"),
+	("bpy.types.maintainvolumeconstraint*", "animation/constraints/transform/maintain_volume.html#bpy-types-maintainvolumeconstraint"),
+	("bpy.types.mesh.use_mirror_topology*", "modeling/meshes/editing/mesh_options.html#bpy-types-mesh-use-mirror-topology"),
+	("bpy.types.particleinstancemodifier*", "modeling/modifiers/simulate/particle_instance.html#bpy-types-particleinstancemodifier"),
+	("bpy.types.shadernodebrightcontrast*", "render/shader_nodes/color/bright_contrast.html#bpy-types-shadernodebrightcontrast"),
+	("bpy.types.shadernodebsdfprincipled*", "render/shader_nodes/shader/principled.html#bpy-types-shadernodebsdfprincipled"),
+	("bpy.types.shadernodebsdfrefraction*", "render/shader_nodes/shader/refraction.html#bpy-types-shadernodebsdfrefraction"),
+	("bpy.types.shadernodeoutputmaterial*", "render/shader_nodes/output/material.html#bpy-types-shadernodeoutputmaterial"),
+	("bpy.types.shadernodetexenvironment*", "render/shader_nodes/textures/environment.html#bpy-types-shadernodetexenvironment"),
+	("bpy.types.transformcacheconstraint*", "animation/constraints/transform/transform_cache.html#bpy-types-transformcacheconstraint"),
+	("bpy.types.userpreferencesfilepaths*", "editors/preferences/file_paths.html#bpy-types-userpreferencesfilepaths"),
+	("bpy.types.vertexweighteditmodifier*", "modeling/modifiers/modify/weight_edit.html#bpy-types-vertexweighteditmodifier"),
+	("bpy.ops.curve.match_texture_space*", "editors/uv/generated_uvs.html#bpy-ops-curve-match-texture-space"),
+	("bpy.ops.font.text_paste_from_file*", "modeling/texts/selecting_editing.html#bpy-ops-font-text-paste-from-file"),
+	("bpy.ops.render.play-rendered-anim*", "render/output/animation_player.html#bpy-ops-render-play-rendered-anim"),
+	("bpy.types.camera.show_composition*", "render/cameras.html#bpy-types-camera-show-composition"),
+	("bpy.types.compositornodealphaover*", "compositing/types/color/alpha_over.html#bpy-types-compositornodealphaover"),
+	("bpy.types.compositornodebokehblur*", "compositing/types/filter/bokeh_blur.html#bpy-types-compositornodebokehblur"),
+	("bpy.types.compositornodecomposite*", "compositing/types/output/composite.html#bpy-types-compositornodecomposite"),
+	("bpy.types.compositornodedespeckle*", "compositing/types/filter/despeckle.html#bpy-types-compositornodedespeckle"),
+	("bpy.types.compositornodediffmatte*", "compositing/types/matte/difference_key.html#bpy-types-compositornodediffmatte"),
+	("bpy.types.compositornodelumamatte*", "compositing/types/matte/luminance_key.html#bpy-types-compositornodelumamatte"),
+	("bpy.types.compositornodemovieclip*", "compositing/types/input/movie_clip.html#bpy-types-compositornodemovieclip"),
+	("bpy.types.compositornodenormalize*", "compositing/types/vector/normalize.html#bpy-types-compositornodenormalize"),
+	("bpy.types.compositornodepremulkey*", "compositing/types/converter/alpha_convert.html#bpy-types-compositornodepremulkey"),
+	("bpy.types.compositornodestabilize*", "compositing/types/distort/stabilize_2d.html#bpy-types-compositornodestabilize"),
+	("bpy.types.compositornodetransform*", "compositing/types/distort/transform.html#bpy-types-compositornodetransform"),
+	("bpy.types.compositornodetranslate*", "compositing/types/distort/translate.html#bpy-types-compositornodetranslate"),
+	("bpy.types.fluidsimulationmodifier*", "physics/fluid/index.html#bpy-types-fluidsimulationmodifier"),
+	("bpy.types.freestylemodulesettings*", "render/freestyle/python.html#bpy-types-freestylemodulesettings"),
+	("bpy.types.laplaciandeformmodifier*", "modeling/modifiers/deform/laplacian_deform.html#bpy-types-laplaciandeformmodifier"),
+	("bpy.types.laplaciansmoothmodifier*", "modeling/modifiers/deform/laplacian_smooth.html#bpy-types-laplaciansmoothmodifier"),
+	("bpy.types.limitdistanceconstraint*", "animation/constraints/transform/limit_distance.html#bpy-types-limitdistanceconstraint"),
+	("bpy.types.limitlocationconstraint*", "animation/constraints/transform/limit_location.html#bpy-types-limitlocationconstraint"),
+	("bpy.types.limitrotationconstraint*", "animation/constraints/transform/limit_rotation.html#bpy-types-limitrotationconstraint"),
+	("bpy.types.rendersettings.filepath*", "render/output/settings.html#bpy-types-rendersettings-filepath"),
+	("bpy.types.sceneeevee.use_overscan*", "render/eevee/render_settings/film.html#bpy-types-sceneeevee-use-overscan"),
+	("bpy.types.shadernodeeeveespecular*", "render/shader_nodes/shader/specular_bsdf.html#bpy-types-shadernodeeeveespecular"),
+	("bpy.types.shadernodehuesaturation*", "render/shader_nodes/color/hue_saturation.html#bpy-types-shadernodehuesaturation"),
+	("bpy.types.shadernodevolumescatter*", "render/shader_nodes/shader/volume_scatter.html#bpy-types-shadernodevolumescatter"),
+	("bpy.types.spacegrapheditor.cursor*", "editors/graph_editor/introduction.html#bpy-types-spacegrapheditor-cursor"),
+	("bpy.types.vertexweightmixmodifier*", "modeling/modifiers/modify/weight_mix.html#bpy-types-vertexweightmixmodifier"),
+	("bpy.ops.object.constraints_clear*", "animation/constraints/interface/adding_removing.html#bpy-ops-object-constraints-clear"),
+	("bpy.ops.uv.average_islands_scale*", "modeling/meshes/editing/uv/layout.html#bpy-ops-uv-average-islands-scale"),
+	("bpy.ops.view3d.edit_mesh_extrude*", "modeling/meshes/editing/duplicating/extrude.html#bpy-ops-view3d-edit-mesh-extrude"),
+	("bpy.types.brightcontrastmodifier*", "video_editing/sequencer/properties/modifiers.html#bpy-types-brightcontrastmodifier"),
+	("bpy.types.brush.cursor_color_add*", "sculpt_paint/brush/display.html#bpy-types-brush-cursor-color-add"),
+	("bpy.types.camerasolverconstraint*", "animation/constraints/motion_tracking/camera_solver.html#bpy-types-camerasolverconstraint"),
+	("bpy.types.clothcollisionsettings*", "physics/cloth/settings/collisions.html#bpy-types-clothcollisionsettings"),
+	("bpy.types.compositornodecurvergb*", "compositing/types/color/rgb_curves.html#bpy-types-compositornodecurvergb"),
+	("bpy.types.compositornodecurvevec*", "compositing/types/vector/vector_curves.html#bpy-types-compositornodecurvevec"),
+	("bpy.types.compositornodedisplace*", "compositing/types/distort/displace.html#bpy-types-compositornodedisplace"),
+	("bpy.types.compositornodelensdist*", "compositing/types/distort/lens_distortion.html#bpy-types-compositornodelensdist"),
+	("bpy.types.compositornodemaprange*", "compositing/types/vector/map_range.html#bpy-types-compositornodemaprange"),
+	("bpy.types.compositornodemapvalue*", "compositing/types/vector/map_value.html#bpy-types-compositornodemapvalue"),
+	("bpy.types.compositornodepixelate*", "compositing/types/filter/pixelate.html#bpy-types-compositornodepixelate"),
+	("bpy.types.compositornodesetalpha*", "compositing/types/converter/set_alpha.html#bpy-types-compositornodesetalpha"),
+	("bpy.types.compositornodesunbeams*", "compositing/types/filter/sun_beams.html#bpy-types-compositornodesunbeams"),
+	("bpy.types.compositornodetrackpos*", "compositing/types/input/track_position.html#bpy-types-compositornodetrackpos"),
+	("bpy.types.compositornodezcombine*", "compositing/types/color/z_combine.html#bpy-types-compositornodezcombine"),
+	("bpy.types.copylocationconstraint*", "animation/constraints/transform/copy_location.html#bpy-types-copylocationconstraint"),
+	("bpy.types.copyrotationconstraint*", "animation/constraints/transform/copy_rotation.html#bpy-types-copyrotationconstraint"),
+	("bpy.types.cyclesmaterialsettings*", "render/cycles/material_settings.html#bpy-types-cyclesmaterialsettings"),
+	("bpy.types.imagepaint.use_occlude*", "sculpt_paint/texture_paint/tool_settings/options.html#bpy-types-imagepaint-use-occlude"),
+	("bpy.types.mesh.auto_smooth_angle*", "modeling/meshes/structure.html#bpy-types-mesh-auto-smooth-angle"),
+	("bpy.types.objectsolverconstraint*", "animation/constraints/motion_tracking/object_solver.html#bpy-types-objectsolverconstraint"),
+	("bpy.types.particlesystemmodifier*", "physics/particles/index.html#bpy-types-particlesystemmodifier"),
+	("bpy.types.sceneeevee.motion_blur*", "render/eevee/render_settings/motion_blur.html#bpy-types-sceneeevee-motion-blur"),
+	("bpy.types.sceneeevee.taa_samples*", "render/eevee/render_settings/sampling.html#bpy-types-sceneeevee-taa-samples"),
+	("bpy.types.shadernodedisplacement*", "render/shader_nodes/vector/displacement.html#bpy-types-shadernodedisplacement"),
+	("bpy.types.shadernodelightfalloff*", "render/shader_nodes/color/light_falloff.html#bpy-types-shadernodelightfalloff"),
+	("bpy.types.shadernodeparticleinfo*", "render/shader_nodes/input/particle_info.html#bpy-types-shadernodeparticleinfo"),
+	("bpy.types.weightednormalmodifier*", "modeling/modifiers/modify/weighted_normal.html#bpy-types-weightednormalmodifier"),
+	("bpy.ops.object.constraints_copy*", "animation/constraints/interface/adding_removing.html#bpy-ops-object-constraints-copy"),
+	("bpy.ops.object.make_single_user*", "scene_layout/object/editing/duplication.html#bpy-ops-object-make-single-user"),
+	("bpy.ops.object.select_hierarchy*", "scene_layout/object/selecting.html#bpy-ops-object-select-hierarchy"),
+	("bpy.ops.screen.screen_full_area*", "interface/window_system/areas.html#bpy-ops-screen-screen-full-area"),
+	("bpy.ops.transform.rotate_normal*", "modeling/meshes/editing/normals.html#bpy-ops-transform-rotate-normal"),
+	("bpy.ops.transform.shrink_fatten*", "modeling/meshes/editing/transform/shrink-fatten.html#bpy-ops-transform-shrink-fatten"),
+	("bpy.ops.wm.dependency_relations*", "advanced/operators.html#bpy-ops-wm-dependency-relations"),
+	("bpy.ops.wm.previews_batch_clear*", "files/blend/previews.html#bpy-ops-wm-previews-batch-clear"),
+	("bpy.types.brush.use_custom_icon*", "sculpt_paint/brush/display.html#bpy-types-brush-use-custom-icon"),
+	("bpy.types.camerabackgroundimage*", "render/cameras.html#bpy-types-camerabackgroundimage"),
+	("bpy.types.compositornodeboxmask*", "compositing/types/matte/box_mask.html#bpy-types-compositornodeboxmask"),
+	("bpy.types.compositornodedefocus*", "compositing/types/filter/defocus.html#bpy-types-compositornodedefocus"),
+	("bpy.types.compositornodeinpaint*", "compositing/types/filter/inpaint.html#bpy-types-compositornodeinpaint"),
+	("bpy.types.compositornodergbtobw*", "compositing/types/converter/rgb_to_bw.html#bpy-types-compositornodergbtobw"),
+	("bpy.types.compositornoderlayers*", "compositing/types/input/render_layers.html#bpy-types-compositornoderlayers"),
+	("bpy.types.compositornodetexture*", "compositing/types/input/texture.html#bpy-types-compositornodetexture"),
+	("bpy.types.compositornodetonemap*", "compositing/types/color/tone_map.html#bpy-types-compositornodetonemap"),
+	("bpy.types.compositornodevecblur*", "compositing/types/filter/vector_blur.html#bpy-types-compositornodevecblur"),
+	("bpy.types.dampedtrackconstraint*", "animation/constraints/tracking/damped_track.html#bpy-types-dampedtrackconstraint"),
+	("bpy.types.distortednoisetexture*", "render/materials/legacy_textures/types/distorted_noise.html#bpy-types-distortednoisetexture"),
+	("bpy.types.followtrackconstraint*", "animation/constraints/motion_tracking/follow_track.html#bpy-types-followtrackconstraint"),
+	("bpy.types.lockedtrackconstraint*", "animation/constraints/tracking/locked_track.html#bpy-types-lockedtrackconstraint"),
+	("bpy.types.obstaclefluidsettings*", "physics/fluid/types/obstacle.html#bpy-types-obstaclefluidsettings"),
+	("bpy.types.particlefluidsettings*", "physics/particles/emitter/physics/fluid.html#bpy-types-particlefluidsettings"),
+	("bpy.types.sceneeevee.volumetric*", "render/eevee/render_settings/volumetrics.html#bpy-types-sceneeevee-volumetric"),
+	("bpy.types.shadernodebsdfdiffuse*", "render/shader_nodes/shader/diffuse.html#bpy-types-shadernodebsdfdiffuse"),
+	("bpy.types.shadernodelayerweight*", "render/shader_nodes/input/layer_weight.html#bpy-types-shadernodelayerweight"),
+	("bpy.types.shadernodeoutputlight*", "render/shader_nodes/output/light.html#bpy-types-shadernodeoutputlight"),
+	("bpy.types.shadernodeoutputworld*", "render/shader_nodes/output/world.html#bpy-types-shadernodeoutputworld"),
+	("bpy.types.shadernodetexgradient*", "render/shader_nodes/textures/gradient.html#bpy-types-shadernodetexgradient"),
+	("bpy.types.shadernodetexmusgrave*", "render/shader_nodes/textures/musgrave.html#bpy-types-shadernodetexmusgrave"),
+	("bpy.types.shadernodevectorcurve*", "render/shader_nodes/vector/curves.html#bpy-types-shadernodevectorcurve"),
+	("bpy.types.userpreferencessystem*", "editors/preferences/system.html#bpy-types-userpreferencessystem"),
+	("bpy.ops.mesh.bridge-edge-loops*", "modeling/meshes/editing/edges.html#bpy-ops-mesh-bridge-edge-loops"),
+	("bpy.ops.object.paths_calculate*", "animation/motion_paths.html#bpy-ops-object-paths-calculate"),
+	("bpy.ops.object.transform_apply*", "scene_layout/object/editing/transform/clear_apply.html#bpy-ops-object-transform-apply"),
+	("bpy.ops.outliner.lib_operation*", "files/linked_libraries.html#bpy-ops-outliner-lib-operation"),
+	("bpy.ops.screen.region_quadview*", "editors/3dview/navigate/views.html#bpy-ops-screen-region-quadview"),
+	("bpy.ops.uv.follow_active_quads*", "modeling/meshes/editing/uv/unwrapping/mapping_types.html#bpy-ops-uv-follow-active-quads"),
+	("bpy.types.colorbalancemodifier*", "video_editing/sequencer/properties/modifiers.html#bpy-types-colorbalancemodifier"),
+	("bpy.types.compositornodefilter*", "compositing/types/filter/filter_node.html#bpy-types-compositornodefilter"),
+	("bpy.types.compositornodehuesat*", "compositing/types/color/hue_saturation.html#bpy-types-compositornodehuesat"),
+	("bpy.types.compositornodeidmask*", "compositing/types/converter/id_mask.html#bpy-types-compositornodeidmask"),
+	("bpy.types.compositornodeinvert*", "compositing/types/color/invert.html#bpy-types-compositornodeinvert"),
+	("bpy.types.compositornodekeying*", "compositing/types/matte/keying.html#bpy-types-compositornodekeying"),
+	("bpy.types.compositornodelevels*", "compositing/types/output/levels.html#bpy-types-compositornodelevels"),
+	("bpy.types.compositornodemixrgb*", "compositing/types/color/mix.html#bpy-types-compositornodemixrgb"),
+	("bpy.types.compositornodenormal*", "compositing/types/vector/normal.html#bpy-types-compositornodenormal"),
+	("bpy.types.compositornoderotate*", "compositing/types/distort/rotate.html#bpy-types-compositornoderotate"),
+	("bpy.types.compositornodeviewer*", "compositing/types/output/viewer.html#bpy-types-compositornodeviewer"),
+	("bpy.types.constraint.influence*", "animation/constraints/interface/common.html#bpy-types-constraint-influence"),
+	("bpy.types.controlfluidsettings*", "physics/fluid/types/control.html#bpy-types-controlfluidsettings"),
+	("bpy.types.datatransfermodifier*", "modeling/modifiers/modify/data_transfer.html#bpy-types-datatransfermodifier"),
+	("bpy.types.dynamicpaintmodifier*", "physics/dynamic_paint/index.html#bpy-types-dynamicpaintmodifier"),
+	("bpy.types.ffmpegsettings.audio*", "render/output/file_formats.html#bpy-types-ffmpegsettings-audio"),
+	("bpy.types.followpathconstraint*", "animation/constraints/relationship/follow_path.html#bpy-types-followpathconstraint"),
+	("bpy.types.gaussianblursequence*", "video_editing/sequencer/strips/effects/blur.html#bpy-types-gaussianblursequence"),
+	("bpy.types.image.display_aspect*", "editors/image/view_tab.html#bpy-types-image-display-aspect"),
+	("bpy.types.limitscaleconstraint*", "animation/constraints/transform/limit_scale.html#bpy-types-limitscaleconstraint"),
+	("bpy.types.mesh.use_auto_smooth*", "modeling/meshes/structure.html#bpy-types-mesh-use-auto-smooth"),
+	("bpy.types.outflowfluidsettings*", "physics/fluid/types/flow.html#bpy-types-outflowfluidsettings"),
+	("bpy.types.scene.background_set*", "scene_layout/scene/properties.html#bpy-types-scene-background-set"),
+	("bpy.types.shadernodebackground*", "render/shader_nodes/shader/background.html#bpy-types-shadernodebackground"),
+	("bpy.types.shadernodebsdfglossy*", "render/shader_nodes/shader/glossy.html#bpy-types-shadernodebsdfglossy"),
+	("bpy.types.shadernodebsdfvelvet*", "render/shader_nodes/shader/velvet.html#bpy-types-shadernodebsdfvelvet"),
+	("bpy.types.shadernodecameradata*", "render/shader_nodes/input/camera_data.html#bpy-types-shadernodecameradata"),
+	("bpy.types.shadernodeobjectinfo*", "render/shader_nodes/input/object_info.html#bpy-types-shadernodeobjectinfo"),
+	("bpy.types.shadernodetexchecker*", "render/shader_nodes/textures/checker.html#bpy-types-shadernodetexchecker"),
+	("bpy.types.shadernodetexvoronoi*", "render/shader_nodes/textures/voronoi.html#bpy-types-shadernodetexvoronoi"),
+	("bpy.types.shadernodevectormath*", "render/shader_nodes/converter/vector_math.html#bpy-types-shadernodevectormath"),
+	("bpy.types.shadernodewavelength*", "render/shader_nodes/converter/wavelength.html#bpy-types-shadernodewavelength"),
+	("bpy.types.shrinkwrapconstraint*", "animation/constraints/relationship/shrinkwrap.html#bpy-types-shrinkwrapconstraint"),
+	("bpy.types.simpledeformmodifier*", "modeling/modifiers/deform/simple_deform.html#bpy-types-simpledeformmodifier"),
+	("bpy.types.spacedopesheeteditor*", "editors/dope_sheet/index.html#bpy-types-spacedopesheeteditor"),
+	("bpy.types.spaceuserpreferences*", "editors/preferences/index.html#bpy-types-spaceuserpreferences"),
+	("bpy.types.speedcontrolsequence*", "video_editing/sequencer/strips/effects/speed_control.html#bpy-types-speedcontrolsequence"),
+	("bpy.types.texturenodecurvetime*", "editors/texture_node/types/input/time.html#bpy-types-texturenodecurvetime"),
+	("bpy.types.transformorientation*", "scene_layout/object/editing/transform/control/orientations.html#bpy-types-transformorientation"),
+	("bpy.types.unifiedpaintsettings*", "sculpt_paint/texture_paint/tool_settings/options.html#bpy-types-unifiedpaintsettings"),
+	("bpy.types.userpreferencesinput*", "editors/preferences/input.html#bpy-types-userpreferencesinput"),
+	("bpy.types.whitebalancemodifier*", "video_editing/sequencer/properties/modifiers.html#bpy-types-whitebalancemodifier"),
+	("bpy.ops.mesh.smoothen_normals*", "modeling/meshes/editing/normals.html#bpy-ops-mesh-smoothen-normals"),
+	("bpy.ops.object.duplicate_move*", "scene_layout/object/editing/duplication.html#bpy-ops-object-duplicate-move"),
+	("bpy.ops.object.select_by_type*", "scene_layout/object/selecting.html#bpy-ops-object-select-by-type"),
+	("bpy.ops.object.select_grouped*", "scene_layout/object/selecting.html#bpy-ops-object-select-grouped"),
+	("bpy.ops.object.select_pattern*", "scene_layout/object/selecting.html#bpy-ops-object-select-pattern"),
+	("bpy.ops.screen.repeat_history*", "interface/undo_redo.html#bpy-ops-screen-repeat-history"),
+	("bpy.ops.sequencer.refresh_all*", "video_editing/sequencer/navigating.html#bpy-ops-sequencer-refresh-all"),
+	("bpy.ops.transform.edge_crease*", "modeling/meshes/editing/edges.html#bpy-ops-transform-edge-crease"),
+	("bpy.ops.uv.seams_from_islands*", "modeling/meshes/editing/uv/unwrapping/seams.html#bpy-ops-uv-seams-from-islands"),
+	("bpy.types.brush.icon_filepath*", "sculpt_paint/brush/display.html#bpy-types-brush-icon-filepath"),
+	("bpy.types.camera.display_size*", "render/cameras.html#bpy-types-camera-display-size"),
+	("bpy.types.compositornodedblur*", "compositing/types/filter/directional_blur.html#bpy-types-compositornodedblur"),
+	("bpy.types.compositornodegamma*", "compositing/types/color/gamma.html#bpy-types-compositornodegamma"),
+	("bpy.types.compositornodeglare*", "compositing/types/filter/glare.html#bpy-types-compositornodeglare"),
+	("bpy.types.compositornodegroup*", "compositing/types/groups.html#bpy-types-compositornodegroup"),
+	("bpy.types.compositornodeimage*", "compositing/types/input/image.html#bpy-types-compositornodeimage"),
+	("bpy.types.compositornodemapuv*", "compositing/types/distort/map_uv.html#bpy-types-compositornodemapuv"),
+	("bpy.types.compositornodescale*", "compositing/types/distort/scale.html#bpy-types-compositornodescale"),
+	("bpy.types.compositornodevalue*", "compositing/types/input/value.html#bpy-types-compositornodevalue"),
+	("bpy.types.copyscaleconstraint*", "animation/constraints/transform/copy_scale.html#bpy-types-copyscaleconstraint"),
+	("bpy.types.cyclesworldsettings*", "render/cycles/world_settings.html#bpy-types-cyclesworldsettings"),
+	("bpy.types.domainfluidsettings*", "physics/fluid/types/domain.html#bpy-types-domainfluidsettings"),
+	("bpy.types.imageformatsettings*", "files/media/image_formats.html#bpy-types-imageformatsettings"),
+	("bpy.types.inflowfluidsettings*", "physics/fluid/types/flow.html#bpy-types-inflowfluidsettings"),
+	("bpy.types.kinematicconstraint*", "animation/constraints/tracking/ik_solver.html#bpy-types-kinematicconstraint"),
+	("bpy.types.mesh.use_paint_mask*", "sculpt_paint/brush/introduction.html#bpy-types-mesh-use-paint-mask"),
+	("bpy.types.movietrackingcamera*", "movie_clip/tracking/clip/properties/camera_data.html#bpy-types-movietrackingcamera"),
+	("bpy.types.object.display_type*", "scene_layout/object/properties/display.html#bpy-types-object-display-type"),
+	("bpy.types.particledupliweight*", "physics/particles/emitter/vertex_groups.html#bpy-types-particledupliweight"),
+	("bpy.types.poseboneconstraints*", "animation/armatures/posing/bone_constraints/index.html#bpy-types-poseboneconstraints"),
+	("bpy.types.rigidbodyconstraint*", "physics/rigid_body/constraints/index.html#bpy-types-rigidbodyconstraint"),
+	("bpy.types.shadernodeaddshader*", "render/shader_nodes/shader/add.html#bpy-types-shadernodeaddshader"),
+	("bpy.types.shadernodeattribute*", "render/shader_nodes/input/attribute.html#bpy-types-shadernodeattribute"),
+	("bpy.types.shadernodeblackbody*", "render/shader_nodes/converter/blackbody.html#bpy-types-shadernodeblackbody"),
+	("bpy.types.shadernodebsdfglass*", "render/shader_nodes/shader/glass.html#bpy-types-shadernodebsdfglass"),
+	("bpy.types.shadernodelightpath*", "render/shader_nodes/input/light_path.html#bpy-types-shadernodelightpath"),
+	("bpy.types.shadernodemixshader*", "render/shader_nodes/shader/mix.html#bpy-types-shadernodemixshader"),
+	("bpy.types.shadernodenormalmap*", "render/shader_nodes/vector/normal_map.html#bpy-types-shadernodenormalmap"),
+	("bpy.types.shadernodewireframe*", "render/shader_nodes/input/wireframe.html#bpy-types-shadernodewireframe"),
+	("bpy.types.smokedomainsettings*", "physics/smoke/types/domain.html#bpy-types-smokedomainsettings"),
+	("bpy.types.spacesequenceeditor*", "video_editing/index.html#bpy-types-spacesequenceeditor"),
+	("bpy.types.stretchtoconstraint*", "animation/constraints/tracking/stretch_to.html#bpy-types-stretchtoconstraint"),
+	("bpy.types.texturenodecurvergb*", "editors/texture_node/types/color/rgb_curves.html#bpy-types-texturenodecurvergb"),
+	("bpy.types.texturenodevaltorgb*", "editors/texture_node/types/converter/rgb_to_bw.html#bpy-types-texturenodevaltorgb"),
+	("bpy.types.transformconstraint*", "animation/constraints/transform/transformation.html#bpy-types-transformconstraint"),
+	("bpy.types.triangulatemodifier*", "modeling/modifiers/generate/triangulate.html#bpy-types-triangulatemodifier"),
+	("bpy.types.userpreferencesedit*", "editors/preferences/editing.html#bpy-types-userpreferencesedit"),
+	("bpy.types.userpreferencesview*", "editors/preferences/interface.html#bpy-types-userpreferencesview"),
+	("bpy.types.windowmanager.addon*", "editors/preferences/addons.html#bpy-types-windowmanager-addon"),
+	("bpy.ops.anim.keyframe_delete*", "animation/keyframes/editing.html#bpy-ops-anim-keyframe-delete"),
+	("bpy.ops.anim.keyframe_insert*", "animation/keyframes/editing.html#bpy-ops-anim-keyframe-insert"),
+	("bpy.ops.mesh.average_normals*", "modeling/meshes/editing/normals.html#bpy-ops-mesh-average-normals"),
+	("bpy.ops.mesh.vertices_smooth*", "modeling/meshes/editing/transform/smooth.html#bpy-ops-mesh-vertices-smooth"),
+	("bpy.ops.node.read_viewlayers*", "interface/controls/nodes/editing.html#bpy-ops-node-read-viewlayers"),
+	("bpy.ops.object.data_transfer*", "modeling/meshes/editing/data_transfer.html#bpy-ops-object-data-transfer"),
+	("bpy.ops.object.select_camera*", "scene_layout/object/selecting.html#bpy-ops-object-select-camera"),
+	("bpy.ops.object.select_linked*", "scene_layout/object/selecting.html#bpy-ops-object-select-linked"),
+	("bpy.ops.object.select_mirror*", "scene_layout/object/selecting.html#bpy-ops-object-select-mirror"),
+	("bpy.ops.object.select_random*", "scene_layout/object/selecting.html#bpy-ops-object-select-random"),
+	("bpy.ops.sound.bake_animation*", "scene_layout/scene/properties.html#bpy-ops-sound-bake-animation"),
+	("bpy.ops.transform.edge_slide*", "modeling/meshes/editing/edges.html#bpy-ops-transform-edge-slide"),
+	("bpy.ops.transform.vert_slide*", "modeling/meshes/editing/vertices.html#bpy-ops-transform-vert-slide"),
+	("bpy.ops.uv.project_from_view*", "modeling/meshes/editing/uv/unwrapping/mapping_types.html#bpy-ops-uv-project-from-view"),
+	("bpy.ops.wm.memory_statistics*", "advanced/operators.html#bpy-ops-wm-memory-statistics"),
+	("bpy.types.adjustmentsequence*", "video_editing/sequencer/strips/adjustment.html#bpy-types-adjustmentsequence"),
+	("bpy.types.alphaundersequence*", "video_editing/sequencer/strips/effects/alpha_over_under_overdrop.html#bpy-types-alphaundersequence"),
+	("bpy.types.armatureconstraint*", "animation/constraints/relationship/armature.html#bpy-types-armatureconstraint"),
+	("bpy.types.compositornodeblur*", "compositing/types/filter/blur_node.html#bpy-types-compositornodeblur"),
+	("bpy.types.compositornodecomb*", "editors/texture_node/types/color/combine_separate.html#bpy-types-compositornodecomb"),
+	("bpy.types.compositornodecrop*", "compositing/types/distort/crop.html#bpy-types-compositornodecrop"),
+	("bpy.types.compositornodeflip*", "compositing/types/distort/flip.html#bpy-types-compositornodeflip"),
+	("bpy.types.compositornodemask*", "compositing/types/input/mask.html#bpy-types-compositornodemask"),
+	("bpy.types.compositornodemath*", "compositing/types/converter/math.html#bpy-types-compositornodemath"),
+	("bpy.types.compositornodetime*", "compositing/types/input/time.html#bpy-types-compositornodetime"),
+	("bpy.types.curvepaintsettings*", "modeling/curves/editing/other.html#bpy-types-curvepaintsettings"),
+	("bpy.types.fluidfluidsettings*", "physics/fluid/types/fluid_object.html#bpy-types-fluidfluidsettings"),
+	("bpy.types.fmodifiergenerator*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fmodifiergenerator"),
+	("bpy.types.freestylelinestyle*", "render/freestyle/parameter_editor/line_style/index.html#bpy-types-freestylelinestyle"),
+	("bpy.types.gammacrosssequence*", "video_editing/sequencer/strips/transitions/cross.html#bpy-types-gammacrosssequence"),
+	("bpy.types.huecorrectmodifier*", "video_editing/sequencer/properties/modifiers.html#bpy-types-huecorrectmodifier"),
+	("bpy.types.imagepaint.stencil*", "sculpt_paint/texture_paint/tool_settings/mask.html#bpy-types-imagepaint-stencil"),
+	("bpy.types.meshdeformmodifier*", "modeling/modifiers/deform/mesh_deform.html#bpy-types-meshdeformmodifier"),
+	("bpy.types.movietrackingtrack*", "movie_clip/tracking/clip/properties/introduction.html#bpy-types-movietrackingtrack"),
+	("bpy.types.nodeoutputfileslot*", "compositing/types/output/file.html#bpy-types-nodeoutputfileslot"),
+	("bpy.types.normaleditmodifier*", "modeling/modifiers/modify/normal_edit.html#bpy-types-normaleditmodifier"),
+	("bpy.types.object.show_bounds*", "scene_layout/object/properties/display.html#bpy-types-object-show-bounds"),
+	("bpy.types.scene.audio_volume*", "scene_layout/scene/properties.html#bpy-types-scene-audio-volume"),
+	("bpy.types.shadernodebsdfhair*", "render/shader_nodes/shader/hair.html#bpy-types-shadernodebsdfhair"),
+	("bpy.types.shadernodebsdftoon*", "render/shader_nodes/shader/toon.html#bpy-types-shadernodebsdftoon"),
+	("bpy.types.shadernodeemission*", "render/shader_nodes/shader/emission.html#bpy-types-shadernodeemission"),
+	("bpy.types.shadernodegeometry*", "render/shader_nodes/input/geometry.html#bpy-types-shadernodegeometry"),
+	("bpy.types.shadernodehairinfo*", "render/shader_nodes/input/hair_info.html#bpy-types-shadernodehairinfo"),
+	("bpy.types.shadernodergbcurve*", "render/shader_nodes/color/rgb_curves.html#bpy-types-shadernodergbcurve"),
+	("bpy.types.shadernodeseparate*", "render/shader_nodes/converter/combine_separate.html#bpy-types-shadernodeseparate"),
+	("bpy.types.shadernodetexbrick*", "render/shader_nodes/textures/brick.html#bpy-types-shadernodetexbrick"),
+	("bpy.types.shadernodetexcoord*", "render/shader_nodes/input/texture_coordinate.html#bpy-types-shadernodetexcoord"),
+	("bpy.types.shadernodeteximage*", "render/shader_nodes/textures/image.html#bpy-types-shadernodeteximage"),
+	("bpy.types.shadernodetexmagic*", "render/shader_nodes/textures/magic.html#bpy-types-shadernodetexmagic"),
+	("bpy.types.shadernodetexnoise*", "render/shader_nodes/textures/noise.html#bpy-types-shadernodetexnoise"),
+	("bpy.types.shrinkwrapmodifier*", "modeling/modifiers/deform/shrinkwrap.html#bpy-types-shrinkwrapmodifier"),
+	("bpy.types.splineikconstraint*", "animation/constraints/tracking/spline_ik.html#bpy-types-splineikconstraint"),
+	("bpy.types.texturenodetexture*", "editors/texture_node/types/input/texture.html#bpy-types-texturenodetexture"),
+	("bpy.ops.anim.keyframe_clear*", "animation/keyframes/editing.html#bpy-ops-anim-keyframe-clear"),
+	("bpy.ops.curve.primitive*add*", "modeling/curves/primitives.html#bpy-ops-curve-primitive-add"),
+	("bpy.ops.mesh.duplicate_move*", "modeling/meshes/editing/duplicating/duplicate.html#bpy-ops-mesh-duplicate-move"),
+	("bpy.ops.mesh.extrude_region*", "modeling/meshes/editing/duplicating/extrude.html#bpy-ops-mesh-extrude-region"),
+	("bpy.ops.object.parent_clear*", "scene_layout/object/properties/relations/parents.html#bpy-ops-object-parent-clear"),
+	("bpy.ops.object.shade_smooth*", "modeling/meshes/editing/normals.html#bpy-ops-object-shade-smooth"),
+	("bpy.ops.transform.push_pull*", "modeling/meshes/editing/transform/push_pull.html#bpy-ops-transform-push-pull"),
+	("bpy.ops.transform.trackball*", "scene_layout/object/editing/transform/basics.html#bpy-ops-transform-trackball"),
+	("bpy.ops.transform.transform*", "scene_layout/object/editing/transform/control/orientations.html#bpy-ops-transform-transform"),
+	("bpy.ops.transform.translate*", "scene_layout/object/editing/transform/basics.html#bpy-ops-transform-translate"),
+	("bpy.ops.uv.cylinder_project*", "modeling/meshes/editing/uv/unwrapping/mapping_types.html#bpy-ops-uv-cylinder-project"),
+	("bpy.ops.uv.minimize_stretch*", "modeling/meshes/editing/uv/layout.html#bpy-ops-uv-minimize-stretch"),
+	("bpy.types.alphaoversequence*", "video_editing/sequencer/strips/effects/alpha_over_under_overdrop.html#bpy-types-alphaoversequence"),
+	("bpy.types.armatureeditbones*", "animation/armatures/bones/editing/index.html#bpy-types-armatureeditbones"),
+	("bpy.types.cameradofsettings*", "render/cameras.html#bpy-types-cameradofsettings"),
+	("bpy.types.childofconstraint*", "animation/constraints/relationship/child_of.html#bpy-types-childofconstraint"),
+	("bpy.types.clamptoconstraint*", "animation/constraints/tracking/clamp_to.html#bpy-types-clamptoconstraint"),
+	("bpy.types.collisionmodifier*", "physics/collision.html#bpy-types-collisionmodifier"),
+	("bpy.types.collisionsettings*", "physics/collision.html#bpy-types-collisionsettings"),
+	("bpy.types.compositornodergb*", "compositing/types/input/rgb.html#bpy-types-compositornodergb"),
+	("bpy.types.compositornodesep*", "editors/texture_node/types/color/combine_separate.html#bpy-types-compositornodesep"),
+	("bpy.types.edgesplitmodifier*", "modeling/modifiers/generate/edge_split.html#bpy-types-edgesplitmodifier"),
+	("bpy.types.fmodifierenvelope*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fmodifierenvelope"),
+	("bpy.types.freestylesettings*", "render/freestyle/view_layer.html#bpy-types-freestylesettings"),
+	("bpy.types.imagepaint.dither*", "sculpt_paint/texture_paint/tool_settings/options.html#bpy-types-imagepaint-dither"),
+	("bpy.types.mesh.texture_mesh*", "editors/uv/generated_uvs.html#bpy-types-mesh-texture-mesh"),
+	("bpy.types.mesh.use_mirror_x*", "modeling/meshes/editing/mesh_options.html#bpy-types-mesh-use-mirror-x"),
+	("bpy.types.meshcachemodifier*", "modeling/modifiers/modify/mesh_cache.html#bpy-types-meshcachemodifier"),
+	("bpy.types.movieclipsequence*", "video_editing/sequencer/strips/clip_mask.html#bpy-types-movieclipsequence"),
+	("bpy.types.object.dimensions*", "scene_layout/object/properties/transforms.html#bpy-types-object-dimensions"),
+	("bpy.types.object.track_axis*", "scene_layout/object/properties/relations/extras.html#bpy-types-object-track-axis"),
+	("bpy.types.scene.active_clip*", "scene_layout/scene/properties.html#bpy-types-scene-active-clip"),
+	("bpy.types.sceneeevee.shadow*", "render/eevee/render_settings/shadows.html#bpy-types-sceneeevee-shadow"),
+	("bpy.types.shadernodecombine*", "render/shader_nodes/converter/combine_separate.html#bpy-types-shadernodecombine"),
+	("bpy.types.shadernodefresnel*", "render/shader_nodes/input/fresnel.html#bpy-types-shadernodefresnel"),
+	("bpy.types.shadernodeholdout*", "render/shader_nodes/shader/holdout.html#bpy-types-shadernodeholdout"),
+	("bpy.types.shadernodemapping*", "render/shader_nodes/vector/mapping.html#bpy-types-shadernodemapping"),
+	("bpy.types.shadernodergbtobw*", "render/shader_nodes/converter/rgb_to_bw.html#bpy-types-shadernodergbtobw"),
+	("bpy.types.shadernodetangent*", "render/shader_nodes/input/tangent.html#bpy-types-shadernodetangent"),
+	("bpy.types.shadernodetexwave*", "render/shader_nodes/textures/wave.html#bpy-types-shadernodetexwave"),
+	("bpy.types.smokecollsettings*", "physics/smoke/types/collision.html#bpy-types-smokecollsettings"),
+	("bpy.types.smokeflowsettings*", "physics/smoke/types/flow_object.html#bpy-types-smokeflowsettings"),
+	("bpy.types.texturenodebricks*", "editors/texture_node/types/patterns/bricks.html#bpy-types-texturenodebricks"),
+	("bpy.types.texturenodemixrgb*", "editors/texture_node/types/color/mix_rgb.html#bpy-types-texturenodemixrgb"),
+	("bpy.types.texturenodeoutput*", "editors/texture_node/types/output/output.html#bpy-types-texturenodeoutput"),
+	("bpy.types.tracktoconstraint*", "animation/constraints/tracking/track_to.html#bpy-types-tracktoconstraint"),
+	("bpy.types.transformsequence*", "video_editing/sequencer/strips/effects/transform.html#bpy-types-transformsequence"),
+	("bpy.types.uvprojectmodifier*", "modeling/modifiers/modify/uv_project.html#bpy-types-uvprojectmodifier"),
+	("bpy.types.wireframemodifier*", "modeling/modifiers/generate/wireframe.html#bpy-types-wireframemodifier"),
+	("bpy.types.worldmistsettings*", "render/cycles/world_settings.html#bpy-types-worldmistsettings"),
+	("bpy.ops.mesh.loopcut_slide*", "modeling/meshes/editing/subdividing/loop.html#bpy-ops-mesh-loopcut-slide"),
+	("bpy.ops.mesh.merge_normals*", "modeling/meshes/editing/normals.html#bpy-ops-mesh-merge-normals"),
+	("bpy.ops.mesh.normals_tools*", "modeling/meshes/editing/normals.html#bpy-ops-mesh-normals-tools"),
+	("bpy.ops.mesh.point_normals*", "modeling/meshes/editing/normals.html#bpy-ops-mesh-point-normals"),
+	("bpy.ops.mesh.primitive*add*", "modeling/meshes/primitives.html#bpy-ops-mesh-primitive-add"),
+	("bpy.ops.mesh.split_normals*", "modeling/meshes/editing/normals.html#bpy-ops-mesh-split-normals"),
+	("bpy.ops.object.gpencil_add*", "grease_pencil/primitives.html#bpy-ops-object-gpencil-add"),
+	("bpy.ops.object.select_less*", "scene_layout/object/selecting.html#bpy-ops-object-select-less"),
+	("bpy.ops.object.select_more*", "scene_layout/object/selecting.html#bpy-ops-object-select-more"),
+	("bpy.ops.object.track_clear*", "animation/constraints/interface/adding_removing.html#bpy-ops-object-track-clear"),
+	("bpy.ops.screen.repeat_last*", "interface/undo_redo.html#bpy-ops-screen-repeat-last"),
+	("bpy.ops.transform.tosphere*", "modeling/meshes/editing/transform/to_sphere.html#bpy-ops-transform-tosphere"),
+	("bpy.ops.wm.previews_ensure*", "files/blend/previews.html#bpy-ops-wm-previews-ensure"),
+	("bpy.types.actionconstraint*", "animation/constraints/relationship/action.html#bpy-types-actionconstraint"),
+	("bpy.types.addonpreferences*", "editors/preferences/addons.html#bpy-types-addonpreferences"),
+	("bpy.types.armaturemodifier*", "modeling/modifiers/deform/armature.html#bpy-types-armaturemodifier"),
+	("bpy.types.colormixsequence*", "video_editing/sequencer/strips/effects/color_mix.html#bpy-types-colormixsequence"),
+	("bpy.types.decimatemodifier*", "modeling/modifiers/generate/decimate.html#bpy-types-decimatemodifier"),
+	("bpy.types.displacemodifier*", "modeling/modifiers/deform/displace.html#bpy-types-displacemodifier"),
+	("bpy.types.displaysafeareas*", "render/cameras.html#bpy-types-displaysafeareas"),
+	("bpy.types.fmodifierstepped*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fmodifierstepped"),
+	("bpy.types.freestylelineset*", "render/freestyle/parameter_editor/line_set.html#bpy-types-freestylelineset"),
+	("bpy.types.mesh.*customdata*", "modeling/meshes/properties/custom_data.html#bpy-types-mesh-customdata"),
+	("bpy.types.multicamsequence*", "video_editing/sequencer/strips/effects/multicam.html#bpy-types-multicamsequence"),
+	("bpy.types.multiplysequence*", "video_editing/sequencer/strips/effects/multiply.html#bpy-types-multiplysequence"),
+	("bpy.types.multiresmodifier*", "modeling/modifiers/generate/multiresolution.html#bpy-types-multiresmodifier"),
+	("bpy.types.object.use_extra*", "scene_layout/object/properties/relations/extras.html#bpy-types-object-use-extra"),
+	("bpy.types.overdropsequence*", "video_editing/sequencer/strips/effects/alpha_over_under_overdrop.html#bpy-types-overdropsequence"),
+	("bpy.types.paint.show_brush*", "sculpt_paint/brush/display.html#bpy-types-paint-show-brush"),
+	("bpy.types.paint.use_cavity*", "sculpt_paint/texture_paint/tool_settings/options.html#bpy-types-paint-use-cavity"),
+	("bpy.types.particlesettings*", "physics/particles/index.html#bpy-types-particlesettings"),
+	("bpy.types.sceneeevee.bloom*", "render/eevee/render_settings/bloom.html#bpy-types-sceneeevee-bloom"),
+	("bpy.types.scenerenderlayer*", "render/layers/layers.html#bpy-types-scenerenderlayer"),
+	("bpy.types.sequencemodifier*", "video_editing/sequencer/properties/modifiers.html#bpy-types-sequencemodifier"),
+	("bpy.types.shadernodeinvert*", "render/shader_nodes/color/invert.html#bpy-types-shadernodeinvert"),
+	("bpy.types.shadernodemixrgb*", "render/shader_nodes/color/mix.html#bpy-types-shadernodemixrgb"),
+	("bpy.types.shadernodenormal*", "render/shader_nodes/vector/normal.html#bpy-types-shadernodenormal"),
+	("bpy.types.shadernodescript*", "render/shader_nodes/osl.html#bpy-types-shadernodescript"),
+	("bpy.types.shadernodetexies*", "render/shader_nodes/textures/ies.html#bpy-types-shadernodetexies"),
+	("bpy.types.shadernodetexsky*", "render/shader_nodes/textures/sky.html#bpy-types-shadernodetexsky"),
+	("bpy.types.softbodymodifier*", "physics/soft_body/index.html#bpy-types-softbodymodifier"),
+	("bpy.types.softbodysettings*", "physics/soft_body/settings.html#bpy-types-softbodysettings"),
+	("bpy.types.solidifymodifier*", "modeling/modifiers/generate/solidify.html#bpy-types-solidifymodifier"),
+	("bpy.types.spacegrapheditor*", "editors/graph_editor/index.html#bpy-types-spacegrapheditor"),
+	("bpy.types.spaceview3d.lock*", "editors/3dview/properties/sidebar.html#bpy-types-spaceview3d-lock"),
+	("bpy.types.sphfluidsettings*", "physics/fluid/index.html#bpy-types-sphfluidsettings"),
+	("bpy.types.subtractsequence*", "video_editing/sequencer/strips/effects/subtract.html#bpy-types-subtractsequence"),
+	("bpy.types.texturenodegroup*", "editors/texture_node/types/groups.html#bpy-types-texturenodegroup"),
+	("bpy.types.texturenodeimage*", "editors/texture_node/types/input/image.html#bpy-types-texturenodeimage"),
+	("bpy.ops.mesh.flip_normals*", "modeling/meshes/editing/normals.html#bpy-ops-mesh-flip-normals"),
+	("bpy.ops.object.lightprobe*", "render/eevee/lightprobes/index.html#bpy-ops-object-lightprobe"),
+	("bpy.ops.object.make_links*", "scene_layout/object/editing/duplication.html#bpy-ops-object-make-links"),
+	("bpy.ops.object.make_local*", "files/linked_libraries.html#bpy-ops-object-make-local"),
+	("bpy.ops.object.origin_set*", "scene_layout/object/origin.html#bpy-ops-object-origin-set"),
+	("bpy.ops.object.parent_set*", "scene_layout/object/properties/relations/parents.html#bpy-ops-object-parent-set"),
+	("bpy.ops.object.proxy_make*", "files/linked_libraries.html#bpy-ops-object-proxy-make"),
+	("bpy.ops.object.select_all*", "scene_layout/object/selecting.html#bpy-ops-object-select-all"),
+	("bpy.ops.object.shade_flat*", "modeling/meshes/editing/normals.html#bpy-ops-object-shade-flat"),
+	("bpy.ops.preferences.addon*", "editors/preferences/addons.html#bpy-ops-preferences-addon"),
+	("bpy.ops.scene.light_cache*", "render/eevee/render_settings/indirect_lighting.html#bpy-ops-scene-light-cache"),
+	("bpy.ops.screen.area_dupli*", "interface/window_system/areas.html#bpy-ops-screen-area-dupli"),
+	("bpy.ops.uv.remove_doubles*", "modeling/meshes/editing/uv/layout.html#bpy-ops-uv-remove-doubles"),
+	("bpy.ops.uv.sphere_project*", "modeling/meshes/editing/uv/unwrapping/mapping_types.html#bpy-ops-uv-sphere-project"),
+	("bpy.ops.wm.previews_clear*", "files/blend/previews.html#bpy-ops-wm-previews-clear"),
+	("bpy.types.booleanmodifier*", "modeling/modifiers/generate/booleans.html#bpy-types-booleanmodifier"),
+	("bpy.types.constraint.mute*", "animation/constraints/interface/header.html#bpy-types-constraint-mute"),
+	("bpy.types.explodemodifier*", "modeling/modifiers/simulate/explode.html#bpy-types-explodemodifier"),
+	("bpy.types.fcurvemodifiers*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fcurvemodifiers"),
+	("bpy.types.floorconstraint*", "animation/constraints/relationship/floor.html#bpy-types-floorconstraint"),
+	("bpy.types.fluidmeshvertex*", "physics/fluid/index.html#bpy-types-fluidmeshvertex"),
+	("bpy.types.fmodifiercycles*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fmodifiercycles"),
+	("bpy.types.fmodifierlimits*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fmodifierlimits"),
+	("bpy.types.imagepaint.mode*", "sculpt_paint/texture_paint/tool_settings/texture_slots.html#bpy-types-imagepaint-mode"),
+	("bpy.types.latticemodifier*", "modeling/modifiers/deform/lattice.html#bpy-types-latticemodifier"),
+	("bpy.types.musgravetexture*", "render/materials/legacy_textures/types/musgrave.html#bpy-types-musgravetexture"),
+	("bpy.types.object.location*", "scene_layout/object/properties/transforms.html#bpy-types-object-location"),
+	("bpy.types.object.rotation*", "scene_layout/object/properties/transforms.html#bpy-types-object-rotation"),
+	("bpy.types.particlehairkey*", "physics/particles/emitter/physics/keyed.html#bpy-types-particlehairkey"),
+	("bpy.types.pivotconstraint*", "animation/constraints/relationship/pivot.html#bpy-types-pivotconstraint"),
+	("bpy.types.rigidbodyobject*", "physics/rigid_body/index.html#bpy-types-rigidbodyobject"),
+	("bpy.types.sceneeevee.gtao*", "render/eevee/render_settings/ambient_occlusion.html#bpy-types-sceneeevee-gtao"),
+	("bpy.types.shadernodebevel*", "render/shader_nodes/input/bevel.html#bpy-types-shadernodebevel"),
+	("bpy.types.shadernodegamma*", "render/shader_nodes/color/gamma.html#bpy-types-shadernodegamma"),
+	("bpy.types.shadernodegroup*", "render/shader_nodes/groups.html#bpy-types-shadernodegroup"),
+	("bpy.types.shadernodeuvmap*", "render/shader_nodes/input/uv_map.html#bpy-types-shadernodeuvmap"),
+	("bpy.types.shadernodevalue*", "render/shader_nodes/input/value.html#bpy-types-shadernodevalue"),
+	("bpy.types.spaceclipeditor*", "movie_clip/index.html#bpy-types-spaceclipeditor"),
+	("bpy.types.spaceproperties*", "editors/properties_editor.html#bpy-types-spaceproperties"),
+	("bpy.types.spacetexteditor*", "editors/text_editor.html#bpy-types-spacetexteditor"),
+	("bpy.types.subsurfmodifier*", "modeling/modifiers/generate/subdivision_surface.html#bpy-types-subsurfmodifier"),
+	("bpy.types.texturenodemath*", "editors/texture_node/types/converter/math.html#bpy-types-texturenodemath"),
+	("bpy.types.userpreferences*", "editors/preferences/index.html#bpy-types-userpreferences"),
+	("bpy.ops.graph.frame_jump*", "editors/graph_editor/fcurves/properties.html#bpy-ops-graph-frame-jump"),
+	("bpy.ops.mesh.edge_rotate*", "modeling/meshes/editing/edges.html#bpy-ops-mesh-edge-rotate"),
+	("bpy.ops.object.hide_view*", "scene_layout/object/editing/introduction.html#bpy-ops-object-hide-view"),
+	("bpy.ops.object.track_set*", "animation/constraints/interface/adding_removing.html#bpy-ops-object-track-set"),
+	("bpy.ops.transform.mirror*", "scene_layout/object/editing/transform/mirror.html#bpy-ops-transform-mirror"),
+	("bpy.ops.transform.resize*", "scene_layout/object/editing/transform/basics.html#bpy-ops-transform-resize"),
+	("bpy.ops.transform.rotate*", "scene_layout/object/editing/transform/basics.html#bpy-ops-transform-rotate"),
+	("bpy.ops.uv.lightmap_pack*", "modeling/meshes/editing/uv/unwrapping/mapping_types.html#bpy-ops-uv-lightmap-pack"),
+	("bpy.ops.uv.smart_project*", "modeling/meshes/editing/uv/unwrapping/mapping_types.html#bpy-ops-uv-smart-project"),
+	("bpy.ops.view3d.localview*", "editors/3dview/navigate/views.html#bpy-ops-view3d-localview"),
+	("bpy.types.curvesmodifier*", "video_editing/sequencer/properties/modifiers.html#bpy-types-curvesmodifier"),
+	("bpy.types.ffmpegsettings*", "render/output/file_formats.html#bpy-types-ffmpegsettings"),
+	("bpy.types.fmodifiernoise*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fmodifiernoise"),
+	("bpy.types.material.paint*", "sculpt_paint/texture_paint/index.html#bpy-types-material-paint"),
+	("bpy.types.mirrormodifier*", "modeling/modifiers/generate/mirror.html#bpy-types-mirrormodifier"),
+	("bpy.types.movieclipproxy*", "editors/clip/sidebar.html#bpy-types-movieclipproxy"),
+	("bpy.types.object.up_axis*", "scene_layout/object/properties/relations/extras.html#bpy-types-object-up-axis"),
+	("bpy.types.particlesystem*", "physics/particles/index.html#bpy-types-particlesystem"),
+	("bpy.types.particletarget*", "physics/particles/emitter/physics/keyed.html#bpy-types-particletarget"),
+	("bpy.types.remeshmodifier*", "modeling/modifiers/generate/remesh.html#bpy-types-remeshmodifier"),
+	("bpy.types.rendersettings*", "render/index.html#bpy-types-rendersettings"),
+	("bpy.types.rigidbodyworld*", "physics/rigid_body/world.html#bpy-types-rigidbodyworld"),
+	("bpy.types.sceneeevee.ssr*", "render/eevee/render_settings/screen_space_reflections.html#bpy-types-sceneeevee-ssr"),
+	("bpy.types.sceneeevee.sss*", "render/eevee/render_settings/subsurface_scattering.html#bpy-types-sceneeevee-sss"),
+	("bpy.types.shadernodebump*", "render/shader_nodes/vector/bump.html#bpy-types-shadernodebump"),
+	("bpy.types.shadernodemath*", "render/shader_nodes/converter/math.html#bpy-types-shadernodemath"),
+	("bpy.types.smoothmodifier*", "modeling/modifiers/deform/smooth.html#bpy-types-smoothmodifier"),
+	("bpy.types.timelinemarker*", "animation/markers.html#bpy-types-timelinemarker"),
+	("bpy.types.usersolidlight*", "editors/preferences/lights.html#bpy-types-usersolidlight"),
+	("bpy.types.uvwarpmodifier*", "modeling/modifiers/modify/uv_warp.html#bpy-types-uvwarpmodifier"),
+	("bpy.types.voronoitexture*", "render/materials/legacy_textures/types/voronoi.html#bpy-types-voronoitexture"),
+	("bpy.types.walknavigation*", "editors/3dview/navigate/walk_fly.html#bpy-types-walknavigation"),
+	("bpy.ops.*.select_circle*", "interface/selecting.html#bpy-ops-select-circle"),
+	("bpy.ops.anim.keying_set*", "animation/keyframes/keying_sets.html#bpy-ops-anim-keying-set"),
+	("bpy.ops.ed.undo_history*", "interface/undo_redo.html#bpy-ops-ed-undo-history"),
+	("bpy.ops.mesh.edge_split*", "modeling/meshes/editing/edges.html#bpy-ops-mesh-edge-split"),
+	("bpy.ops.mesh.mark_sharp*", "modeling/meshes/editing/edges.html#bpy-ops-mesh-mark-sharp"),
+	("bpy.ops.object.armature*", "animation/armatures/index.html#bpy-ops-object-armature"),
+	("bpy.ops.object.face_map*", "modeling/meshes/properties/object_data.html#bpy-ops-object-face-map"),
+	("bpy.ops.rigidbody.world*", "physics/rigid_body/world.html#bpy-ops-rigidbody-world"),
+	("bpy.ops.transform.shear*", "modeling/meshes/editing/transform/shear.html#bpy-ops-transform-shear"),
+	("bpy.ops.uv.cube_project*", "modeling/meshes/editing/uv/unwrapping/mapping_types.html#bpy-ops-uv-cube-project"),
+	("bpy.ops.uv.pack_islands*", "modeling/meshes/editing/uv/layout.html#bpy-ops-uv-pack-islands"),
+	("bpy.ops.wm.app_template*", "advanced/app_templates.html#bpy-ops-wm-app-template"),
+	("bpy.ops.wm.redraw_timer*", "advanced/operators.html#bpy-ops-wm-redraw-timer"),
+	("bpy.types.*light.shadow*", "render/eevee/lighting.html#bpy-types-light-shadow"),
+	("bpy.types.armaturebones*", "animation/armatures/bones/index.html#bpy-types-armaturebones"),
+	("bpy.types.arraymodifier*", "modeling/modifiers/generate/array.html#bpy-types-arraymodifier"),
+	("bpy.types.bevelmodifier*", "modeling/modifiers/generate/bevel.html#bpy-types-bevelmodifier"),
+	("bpy.types.buildmodifier*", "modeling/modifiers/generate/build.html#bpy-types-buildmodifier"),
+	("bpy.types.camera.sensor*", "render/cameras.html#bpy-types-camera-sensor"),
+	("bpy.types.clothmodifier*", "physics/cloth/index.html#bpy-types-clothmodifier"),
+	("bpy.types.clothsettings*", "physics/cloth/settings/index.html#bpy-types-clothsettings"),
+	("bpy.types.cloudstexture*", "render/materials/legacy_textures/types/clouds.html#bpy-types-cloudstexture"),
+	("bpy.types.colorsequence*", "video_editing/sequencer/strips/color.html#bpy-types-colorsequence"),
+	("bpy.types.crosssequence*", "video_editing/sequencer/strips/transitions/cross.html#bpy-types-crosssequence"),
+	("bpy.types.curvemodifier*", "modeling/modifiers/deform/curve.html#bpy-types-curvemodifier"),
+	("bpy.types.fieldsettings*", "physics/forces/force_fields/index.html#bpy-types-fieldsettings"),
+	("bpy.types.fluidsettings*", "physics/fluid/index.html#bpy-types-fluidsettings"),
+	("bpy.types.imagesequence*", "video_editing/sequencer/strips/movie_image.html#bpy-types-imagesequence"),
+	("bpy.types.marbletexture*", "render/materials/legacy_textures/types/marble.html#bpy-types-marbletexture"),
+	("bpy.types.modifier.show*", "modeling/modifiers/introduction.html#bpy-types-modifier-show"),
+	("bpy.types.moviesequence*", "video_editing/sequencer/strips/movie_image.html#bpy-types-moviesequence"),
+	("bpy.types.movietracking*", "movie_clip/tracking/index.html#bpy-types-movietracking"),
+	("bpy.types.object.parent*", "scene_layout/object/properties/relations/parents.html#bpy-types-object-parent"),
+	("bpy.types.oceanmodifier*", "modeling/modifiers/simulate/ocean.html#bpy-types-oceanmodifier"),
+	("bpy.types.particlebrush*", "physics/particles/mode.html#bpy-types-particlebrush"),
+	("bpy.types.scene.gravity*", "physics/forces/gravity.html#bpy-types-scene-gravity"),
+	("bpy.types.sceneeevee.gi*", "render/eevee/render_settings/indirect_lighting.html#bpy-types-sceneeevee-gi"),
+	("bpy.types.scenesequence*", "video_editing/sequencer/strips/scene.html#bpy-types-scenesequence"),
+	("bpy.types.screwmodifier*", "modeling/modifiers/generate/screw.html#bpy-types-screwmodifier"),
+	("bpy.types.sequenceproxy*", "video_editing/sequencer/properties/proxy_cache.html#bpy-types-sequenceproxy"),
+	("bpy.types.shadernodergb*", "render/shader_nodes/input/rgb.html#bpy-types-shadernodergb"),
+	("bpy.types.smokemodifier*", "physics/smoke/index.html#bpy-types-smokemodifier"),
+	("bpy.types.soundsequence*", "video_editing/sequencer/strips/sound.html#bpy-types-soundsequence"),
+	("bpy.types.spaceoutliner*", "editors/outliner.html#bpy-types-spaceoutliner"),
+	("bpy.types.spacetimeline*", "editors/timeline.html#bpy-types-spacetimeline"),
+	("bpy.types.stuccitexture*", "render/materials/legacy_textures/types/stucci.html#bpy-types-stuccitexture"),
+	("bpy.types.windowmanager*", "interface/index.html#bpy-types-windowmanager"),
+	("bpy.ops.*.select_lasso*", "interface/selecting.html#bpy-ops-select-lasso"),
+	("bpy.ops.mesh.mark_seam*", "modeling/meshes/editing/edges.html#bpy-ops-mesh-mark-seam"),
+	("bpy.ops.mesh.subdivide*", "modeling/meshes/editing/subdividing/subdivide.html#bpy-ops-mesh-subdivide"),
+	("bpy.ops.object.convert*", "scene_layout/object/editing/introduction.html#bpy-ops-object-convert"),
+	("bpy.ops.object.speaker*", "render/output/audio/speaker.html#bpy-ops-object-speaker"),
+	("bpy.ops.transform.bend*", "modeling/meshes/editing/transform/bend.html#bpy-ops-transform-bend"),
+	("bpy.types.bakesettings*", "render/cycles/baking.html#bpy-types-bakesettings"),
+	("bpy.types.blendtexture*", "render/materials/legacy_textures/types/blend.html#bpy-types-blendtexture"),
+	("bpy.types.castmodifier*", "modeling/modifiers/deform/cast.html#bpy-types-castmodifier"),
+	("bpy.types.colormanaged*", "render/color_management.html#bpy-types-colormanaged"),
+	("bpy.types.glowsequence*", "video_editing/sequencer/strips/effects/glow.html#bpy-types-glowsequence"),
+	("bpy.types.hookmodifier*", "modeling/modifiers/deform/hooks.html#bpy-types-hookmodifier"),
+	("bpy.types.latticepoint*", "animation/lattice.html#bpy-types-latticepoint"),
+	("bpy.types.magictexture*", "render/materials/legacy_textures/types/magic.html#bpy-types-magictexture"),
+	("bpy.types.maskmodifier*", "modeling/modifiers/generate/mask.html#bpy-types-maskmodifier"),
+	("bpy.types.masksequence*", "video_editing/sequencer/strips/clip_mask.html#bpy-types-masksequence"),
+	("bpy.types.metasequence*", "video_editing/sequencer/meta.html#bpy-types-metasequence"),
+	("bpy.types.object.color*", "scene_layout/object/properties/display.html#bpy-types-object-color"),
+	("bpy.types.object.delta*", "scene_layout/object/properties/transforms.html#bpy-types-object-delta"),
+	("bpy.types.object.empty*", "modeling/empties.html#bpy-types-object-empty"),
+	("bpy.types.object.scale*", "scene_layout/object/properties/transforms.html#bpy-types-object-scale"),
+	("bpy.types.particleedit*", "physics/particles/mode.html#bpy-types-particleedit"),
+	("bpy.types.scene.camera*", "scene_layout/scene/properties.html#bpy-types-scene-camera"),
+	("bpy.types.skinmodifier*", "modeling/modifiers/generate/skin.html#bpy-types-skinmodifier"),
+	("bpy.types.spaceconsole*", "editors/python_console.html#bpy-types-spaceconsole"),
+	("bpy.types.textsequence*", "video_editing/sequencer/strips/text.html#bpy-types-textsequence"),
+	("bpy.types.unitsettings*", "scene_layout/scene/properties.html#bpy-types-unitsettings"),
+	("bpy.types.vertexcolors*", "sculpt_paint/vertex_paint/index.html#bpy-types-vertexcolors"),
+	("bpy.types.view3dcursor*", "editors/3dview/3d_cursor.html#bpy-types-view3dcursor"),
+	("bpy.types.warpmodifier*", "modeling/modifiers/deform/warp.html#bpy-types-warpmodifier"),
+	("bpy.types.wavemodifier*", "modeling/modifiers/deform/wave.html#bpy-types-wavemodifier"),
+	("bpy.types.wipesequence*", "video_editing/sequencer/strips/transitions/wipe.html#bpy-types-wipesequence"),
+	("bpy.ops.image.project*", "sculpt_paint/texture_paint/tool_settings/options.html#bpy-ops-image-project"),
+	("bpy.ops.object.*clear*", "scene_layout/object/editing/transform/clear_apply.html#bpy-ops-object-clear"),
+	("bpy.ops.object.delete*", "scene_layout/object/editing/introduction.html#bpy-ops-object-delete"),
+	("bpy.ops.screen.header*", "interface/window_system/regions.html#bpy-ops-screen-header"),
+	("bpy.ops.view3d.select*", "editors/3dview/selecting.html#bpy-ops-view3d-select"),
+	("bpy.ops.wm.debug_menu*", "advanced/operators.html#bpy-ops-wm-debug-menu"),
+	("bpy.ops.wm.properties*", "files/data_blocks.html#bpy-ops-wm-properties"),
+	("bpy.types.addsequence*", "video_editing/sequencer/strips/effects/add.html#bpy-types-addsequence"),
+	("bpy.types.camera.show*", "render/cameras.html#bpy-types-camera-show"),
+	("bpy.types.consoleline*", "editors/python_console.html#bpy-types-consoleline"),
+	("bpy.types.meshstatvis*", "modeling/meshes/mesh_analysis.html#bpy-types-meshstatvis"),
+	("bpy.types.nodesetting*", "interface/controls/nodes/parts.html#bpy-types-nodesetting"),
+	("bpy.types.object.lock*", "scene_layout/object/properties/transforms.html#bpy-types-object-lock"),
+	("bpy.types.object.show*", "scene_layout/object/properties/display.html#bpy-types-object-show"),
+	("bpy.types.particlekey*", "physics/particles/emitter/physics/keyed.html#bpy-types-particlekey"),
+	("bpy.types.posebone.ik*", "animation/armatures/posing/bone_constraints/inverse_kinematics/introduction.html#bpy-types-posebone-ik"),
+	("bpy.types.renderlayer*", "render/layers/layers.html#bpy-types-renderlayer"),
+	("bpy.types.spaceview3d*", "editors/3dview/index.html#bpy-types-spaceview3d"),
+	("bpy.types.uipopupmenu*", "interface/controls/buttons/menus.html#bpy-types-uipopupmenu"),
+	("bpy.types.vertexpaint*", "sculpt_paint/vertex_paint/index.html#bpy-types-vertexpaint"),
+	("bpy.types.woodtexture*", "render/materials/legacy_textures/types/wood.html#bpy-types-woodtexture"),
+	("bpy.ops.*.select_box*", "interface/selecting.html#bpy-ops-select-box"),
+	("bpy.ops.object.align*", "scene_layout/object/editing/transform/tools.html#bpy-ops-object-align"),
+	("bpy.ops.object.empty*", "modeling/empties.html#bpy-ops-object-empty"),
+	("bpy.ops.object.quick*", "physics/introduction.html#bpy-ops-object-quick"),
+	("bpy.ops.uv.mark_seam*", "modeling/meshes/editing/uv/unwrapping/seams.html#bpy-ops-uv-mark-seam"),
+	("bpy.ops.view3d.ruler*", "editors/3dview/toolbar/measure.html#bpy-ops-view3d-ruler"),
+	("bpy.types.areaspaces*", "interface/window_system/areas.html#bpy-types-areaspaces"),
+	("bpy.types.bpy_struct*", "files/data_blocks.html#bpy-types-bpy-struct"),
+	("bpy.types.collection*", "scene_layout/collections/collections.html#bpy-types-collection"),
+	("bpy.types.compositor*", "compositing/index.html#bpy-types-compositor"),
+	("bpy.types.constraint*", "animation/constraints/index.html#bpy-types-constraint"),
+	("bpy.types.imagepaint*", "sculpt_paint/texture_paint/index.html#bpy-types-imagepaint"),
+	("bpy.types.lightprobe*", "render/eevee/lightprobes/index.html#bpy-types-lightprobe"),
+	("bpy.types.nodesocket*", "interface/controls/nodes/parts.html#bpy-types-nodesocket"),
+	("bpy.types.pointcache*", "physics/baking.html#bpy-types-pointcache"),
+	("bpy.types.pointlight*", "render/lights/light_object.html#bpy-types-pointlight"),
+	("bpy.types.renderview*", "render/output/multiview/index.html#bpy-types-renderview"),
+	("bpy.types.sceneeevee*", "render/eevee/index.html#bpy-types-sceneeevee"),
+	("bpy.types.vectorfont*", "modeling/texts/index.html#bpy-types-vectorfont"),
+	("bpy.ops.mesh.bisect*", "modeling/meshes/editing/subdividing/bisect.html#bpy-ops-mesh-bisect"),
+	("bpy.ops.object.join*", "scene_layout/object/editing/introduction.html#bpy-ops-object-join"),
+	("bpy.ops.object.text*", "modeling/texts/index.html#bpy-ops-object-text"),
+	("bpy.ops.view3d.snap*", "scene_layout/object/editing/transform/control/snap.html#bpy-ops-view3d-snap"),
+	("bpy.types.*texspace*", "editors/uv/generated_uvs.html#bpy-types-texspace"),
+	("bpy.types.arealight*", "render/lights/light_object.html#bpy-types-arealight"),
+	("bpy.types.blenddata*", "files/data_blocks.html#bpy-types-blenddata"),
+	("bpy.types.colorramp*", "interface/controls/templates/color_ramp.html#bpy-types-colorramp"),
+	("bpy.types.dopesheet*", "editors/dope_sheet/index.html#bpy-types-dopesheet"),
+	("bpy.types.fmodifier*", "editors/graph_editor/fcurves/modifiers.html#bpy-types-fmodifier"),
+	("bpy.types.freestyle*", "render/freestyle/index.html#bpy-types-freestyle"),
+	("bpy.types.movieclip*", "movie_clip/index.html#bpy-types-movieclip"),
+	("bpy.types.nodeframe*", "interface/controls/nodes/frame.html#bpy-types-nodeframe"),
+	("bpy.types.nodegroup*", "interface/controls/nodes/groups.html#bpy-types-nodegroup"),
+	("bpy.types.spotlight*", "render/lights/light_object.html#bpy-types-spotlight"),
+	("bpy.types.textcurve*", "modeling/texts/index.html#bpy-types-textcurve"),
+	("bpy.types.uipiemenu*", "interface/controls/buttons/menus.html#bpy-types-uipiemenu"),
+	("bpy.types.uipopover*", "interface/controls/buttons/menus.html#bpy-types-uipopover"),
+	("bpy.ops.collection*", "scene_layout/collections/collections.html#bpy-ops-collection"),
+	("bpy.ops.constraint*", "animation/constraints/index.html#bpy-ops-constraint"),
+	("bpy.ops.curve.draw*", "modeling/curves/editing/other.html#bpy-ops-curve-draw"),
+	("bpy.ops.mesh.knife*", "modeling/meshes/editing/subdividing/knife.html#bpy-ops-mesh-knife"),
+	("bpy.ops.mesh.noise*", "modeling/meshes/editing/transform/noise.html#bpy-ops-mesh-noise"),
+	("bpy.ops.mesh.screw*", "modeling/meshes/editing/duplicating/screw.html#bpy-ops-mesh-screw"),
+	("bpy.ops.safe_areas*", "render/cameras.html#bpy-ops-safe-areas"),
+	("bpy.types.armature*", "animation/armatures/index.html#bpy-types-armature"),
+	("bpy.types.editbone*", "animation/armatures/bones/editing/index.html#bpy-types-editbone"),
+	("bpy.types.facemaps*", "modeling/meshes/properties/object_data.html#bpy-types-facemaps"),
+	("bpy.types.keyframe*", "animation/keyframes/index.html#bpy-types-keyframe"),
+	("bpy.types.linesets*", "render/freestyle/parameter_editor/line_set.html#bpy-types-linesets"),
+	("bpy.types.metaball*", "modeling/metas/index.html#bpy-types-metaball"),
+	("bpy.types.modifier*", "modeling/modifiers/index.html#bpy-types-modifier"),
+	("bpy.types.nlastrip*", "editors/nla/strips.html#bpy-types-nlastrip"),
+	("bpy.types.nlatrack*", "editors/nla/tracks.html#bpy-types-nlatrack"),
+	("bpy.types.nodelink*", "interface/controls/nodes/parts.html#bpy-types-nodelink"),
+	("bpy.types.nodetree*", "interface/controls/nodes/parts.html#bpy-types-nodetree"),
+	("bpy.types.particle*", "physics/particles/index.html#bpy-types-particle"),
+	("bpy.types.sequence*", "video_editing/index.html#bpy-types-sequence"),
+	("bpy.types.shapekey*", "animation/shape_keys/index.html#bpy-types-shapekey"),
+	("bpy.types.spacenla*", "editors/nla/index.html#bpy-types-spacenla"),
+	("bpy.types.sunlight*", "render/lights/light_object.html#bpy-types-sunlight"),
+	("bpy.ops.mesh.spin*", "modeling/meshes/editing/duplicating/spin.html#bpy-ops-mesh-spin"),
+	("bpy.ops.rigidbody*", "physics/rigid_body/index.html#bpy-ops-rigidbody"),
+	("bpy.ops.sequencer*", "video_editing/index.html#bpy-ops-sequencer"),
+	("bpy.ops.transform*", "scene_layout/object/editing/transform/index.html#bpy-ops-transform"),
+	("bpy.ops.uv.stitch*", "modeling/meshes/editing/uv/layout.html#bpy-ops-uv-stitch"),
+	("bpy.ops.uv.unwrap*", "modeling/meshes/editing/uv/unwrapping/mapping_types.html#bpy-ops-uv-unwrap"),
+	("bpy.types.animviz*", "animation/motion_paths.html#bpy-types-animviz"),
+	("bpy.types.lattice*", "animation/lattice.html#bpy-types-lattice"),
+	("bpy.types.library*", "files/linked_libraries.html#bpy-types-library"),
+	("bpy.types.speaker*", "render/output/audio/speaker.html#bpy-types-speaker"),
+	("bpy.types.textbox*", "modeling/texts/layout.html#bpy-types-textbox"),
+	("bpy.types.texture*", "render/materials/legacy_textures/index.html#bpy-types-texture"),
+	("bpy.ops.armature*", "animation/armatures/index.html#bpy-ops-armature"),
+	("bpy.ops.nla.bake*", "animation/actions.html#bpy-ops-nla-bake"),
+	("bpy.ops.outliner*", "editors/outliner.html#bpy-ops-outliner"),
+	("bpy.ops.particle*", "physics/particles/index.html#bpy-ops-particle"),
+	("bpy.ops.uv.align*", "modeling/meshes/editing/uv/layout.html#bpy-ops-uv-align"),
+	("bpy.ops.uv.reset*", "modeling/meshes/editing/uv/unwrapping/mapping_types.html#bpy-ops-uv-reset"),
+	("bpy.ops.wm.addon*", "editors/preferences/addons.html#bpy-ops-wm-addon"),
+	("bpy.types.action*", "animation/actions.html#bpy-types-action"),
+	("bpy.types.cycles*", "render/cycles/index.html#bpy-types-cycles"),
+	("bpy.types.driver*", "animation/drivers/index.html#bpy-types-driver"),
+	("bpy.types.fcurve*", "editors/graph_editor/fcurves/index.html#bpy-types-fcurve"),
+	("bpy.types.header*", "interface/window_system/regions.html#bpy-types-header"),
+	("bpy.types.object*", "scene_layout/object/index.html#bpy-types-object"),
+	("bpy.types.region*", "interface/window_system/regions.html#bpy-types-region"),
+	("bpy.types.render*", "render/index.html#bpy-types-render"),
+	("bpy.types.sculpt*", "sculpt_paint/sculpting/index.html#bpy-types-sculpt"),
+	("bpy.types.shader*", "render/shader_nodes/shader/index.html#bpy-types-shader"),
+	("bpy.types.window*", "interface/index.html#bpy-types-window"),
+	("bpy.ops.buttons*", "interface/index.html#bpy-ops-buttons"),
+	("bpy.ops.console*", "editors/python_console.html#bpy-ops-console"),
+	("bpy.ops.ed.redo*", "interface/undo_redo.html#bpy-ops-ed-redo"),
+	("bpy.ops.ed.undo*", "interface/undo_redo.html#bpy-ops-ed-undo"),
+	("bpy.ops.gpencil*", "grease_pencil/index.html#bpy-ops-gpencil"),
+	("bpy.ops.lattice*", "animation/lattice.html#bpy-ops-lattice"),
+	("bpy.ops.poselib*", "animation/armatures/properties/pose_library.html#bpy-ops-poselib"),
+	("bpy.ops.ptcache*", "physics/baking.html#bpy-ops-ptcache"),
+	("bpy.ops.surface*", "modeling/surfaces/index.html#bpy-ops-surface"),
+	("bpy.ops.texture*", "render/materials/legacy_textures/index.html#bpy-ops-texture"),
+	("bpy.ops.uv.weld*", "modeling/meshes/editing/uv/layout.html#bpy-ops-uv-weld"),
+	("bpy.types.addon*", "editors/preferences/addons.html#bpy-types-addon"),
+	("bpy.types.brush*", "sculpt_paint/brush/brush.html#bpy-types-brush"),
+	("bpy.types.curve*", "modeling/curves/index.html#bpy-types-curve"),
+	("bpy.types.image*", "files/media/image_formats.html#bpy-types-image"),
+	("bpy.types.itasc*", "animation/armatures/posing/bone_constraints/inverse_kinematics/introduction.html#bpy-types-itasc"),
+	("bpy.types.nodes*", "interface/controls/nodes/index.html#bpy-types-nodes"),
+	("bpy.types.paint*", "sculpt_paint/index.html#bpy-types-paint"),
+	("bpy.types.panel*", "interface/window_system/tabs_panels.html#bpy-types-panel"),
+	("bpy.types.scene*", "scene_layout/scene/index.html#bpy-types-scene"),
+	("bpy.types.sound*", "render/output/audio/index.html#bpy-types-sound"),
+	("bpy.types.space*", "editors/index.html#bpy-types-space"),
+	("bpy.types.theme*", "editors/preferences/themes.html#bpy-types-theme"),
+	("bpy.ops.action*", "animation/actions.html#bpy-ops-action"),
+	("bpy.ops.cycles*", "render/cycles/index.html#bpy-ops-cycles"),
+	("bpy.ops.dpaint*", "physics/dynamic_paint/index.html#bpy-ops-dpaint"),
+	("bpy.ops.export*", "files/import_export.html#bpy-ops-export"),
+	("bpy.ops.import*", "files/import_export.html#bpy-ops-import"),
+	("bpy.ops.marker*", "animation/markers.html#bpy-ops-marker"),
+	("bpy.ops.object*", "scene_layout/object/index.html#bpy-ops-object"),
+	("bpy.ops.render*", "render/index.html#bpy-ops-render"),
+	("bpy.ops.script*", "advanced/scripting/index.html#bpy-ops-script"),
+	("bpy.ops.sculpt*", "sculpt_paint/sculpting/index.html#bpy-ops-sculpt"),
+	("bpy.ops.uv.pin*", "modeling/meshes/editing/uv/layout.html#bpy-ops-uv-pin"),
+	("bpy.ops.view3d*", "editors/3dview/index.html#bpy-ops-view3d"),
+	("bpy.types.area*", "interface/window_system/areas.html#bpy-types-area"),
+	("bpy.types.boid*", "physics/particles/emitter/physics/boids.html#bpy-types-boid"),
+	("bpy.types.bone*", "animation/armatures/bones/index.html#bpy-types-bone"),
+	("bpy.types.mask*", "movie_clip/masking/index.html#bpy-types-mask"),
+	("bpy.types.menu*", "interface/controls/buttons/menus.html#bpy-types-menu"),
+	("bpy.types.mesh*", "modeling/meshes/index.html#bpy-types-mesh"),
+	("bpy.types.pose*", "animation/armatures/posing/index.html#bpy-types-pose"),
+	("bpy.types.text*", "modeling/texts/index.html#bpy-types-text"),
+	("bpy.ops.brush*", "sculpt_paint/brush/brush.html#bpy-ops-brush"),
+	("bpy.ops.cloth*", "physics/cloth/index.html#bpy-ops-cloth"),
+	("bpy.ops.curve*", "modeling/curves/index.html#bpy-ops-curve"),
+	("bpy.ops.fluid*", "physics/fluid/index.html#bpy-ops-fluid"),
+	("bpy.ops.graph*", "editors/graph_editor/index.html#bpy-ops-graph"),
+	("bpy.ops.image*", "files/media/image_formats.html#bpy-ops-image"),
+	("bpy.ops.mball*", "modeling/metas/index.html#bpy-ops-mball"),
+	("bpy.ops.paint*", "sculpt_paint/index.html#bpy-ops-paint"),
+	("bpy.ops.scene*", "scene_layout/scene/index.html#bpy-ops-scene"),
+	("bpy.ops.sound*", "render/output/audio/index.html#bpy-ops-sound"),
+	("bpy.types.key*", "animation/shape_keys/index.html#bpy-types-key"),
+	("bpy.ops.anim*", "animation/index.html#bpy-ops-anim"),
+	("bpy.ops.boid*", "physics/particles/emitter/physics/boids.html#bpy-ops-boid"),
+	("bpy.ops.clip*", "movie_clip/index.html#bpy-ops-clip"),
+	("bpy.ops.font*", "modeling/texts/index.html#bpy-ops-font"),
+	("bpy.ops.mask*", "movie_clip/masking/index.html#bpy-ops-mask"),
+	("bpy.ops.mesh*", "modeling/meshes/index.html#bpy-ops-mesh"),
+	("bpy.ops.node*", "interface/controls/nodes/index.html#bpy-ops-node"),
+	("bpy.ops.pose*", "animation/armatures/posing/index.html#bpy-ops-pose"),
+	("bpy.ops.text*", "editors/text_editor.html#bpy-ops-text"),
+	("bpy.ops.time*", "editors/timeline.html#bpy-ops-time"),
+	("bpy.types.id*", "files/data_blocks.html#bpy-types-id"),
+	("bpy.ops.nla*", "editors/nla/index.html#bpy-ops-nla"),
+	("bpy.ops.ed*", "interface/undo_redo.html#bpy-ops-ed"),
+	("bpy.ops.ui*", "interface/index.html#bpy-ops-ui"),
+	("bpy.ops.wm*", "interface/index.html#bpy-ops-wm"),
+)
diff --git a/release/scripts/presets/interface_theme/blender_light.xml b/release/scripts/presets/interface_theme/blender_light.xml
index dc0bb6b629e..49b01ec3309 100644
--- a/release/scripts/presets/interface_theme/blender_light.xml
+++ b/release/scripts/presets/interface_theme/blender_light.xml
@@ -974,6 +974,7 @@
       <ThemeOutliner
         match="#337f33"
         selected_highlight="#7a8e99"
+        active="#92aab7"
         selected_object="#ffddb3"
         active_object="#ffffff"
         edited_object="#0080624d"
diff --git a/release/scripts/presets/keyconfig/keymap_data/blender_default.py b/release/scripts/presets/keyconfig/keymap_data/blender_default.py
index 3f6a5f0a514..7cbea843ae8 100644
--- a/release/scripts/presets/keyconfig/keymap_data/blender_default.py
+++ b/release/scripts/presets/keyconfig/keymap_data/blender_default.py
@@ -409,6 +409,7 @@ def km_window(params):
         items.extend([
             ("wm.doc_view_manual_ui_context", {"type": 'F1', "value": 'PRESS'}, None),
             op_panel("TOPBAR_PT_name", {"type": 'F2', "value": 'PRESS'}, [("keep_open", False)]),
+            ("wm.batch_rename", {"type": 'F2', "value": 'PRESS', "alt": True}, None),
             ("wm.search_menu", {"type": 'F3', "value": 'PRESS'}, None),
             op_menu("TOPBAR_MT_file_context_menu", {"type": 'F4', "value": 'PRESS'}),
         ])
@@ -700,20 +701,39 @@ def km_outliner(params):
     items.extend([
         ("outliner.highlight_update", {"type": 'MOUSEMOVE', "value": 'ANY', "any": True}, None),
         ("outliner.item_rename", {"type": 'LEFTMOUSE', "value": 'DOUBLE_CLICK'}, None),
+        ("outliner.item_rename", {"type": 'F2', "value": 'PRESS'}, None),
         ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK'},
-         {"properties": [("extend", False), ("recursive", False), ("deselect_all", not params.legacy)]}),
-        ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True},
-         {"properties": [("extend", True), ("recursive", False)]}),
+         {"properties": [("extend", False), ("deselect_all", not params.legacy)]}),
         ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "ctrl": True},
-         {"properties": [("extend", False), ("recursive", True)]}),
-        ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True, "ctrl": True},
-         {"properties": [("extend", True), ("recursive", True)]}),
+         {"properties": [("extend", True), ("deselect_all", not params.legacy)]}),
+        ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True},
+         {"properties": [("extend", False), ("extend_range", True), ("deselect_all", not params.legacy)]}),
+        ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "ctrl": True, "shift": True},
+         {"properties": [("extend", True), ("extend_range", True), ("deselect_all", not params.legacy)]}),
         ("outliner.select_box", {"type": 'B', "value": 'PRESS'}, None),
-        ("outliner.item_openclose", {"type": 'RET', "value": 'PRESS'},
+        ("outliner.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY'}, {"properties": [("tweak", True)]}),
+        ("outliner.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY', "shift": True},
+         {"properties": [("tweak", True), ("mode", "ADD")]}),
+        ("outliner.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY', "ctrl": True},
+         {"properties": [("tweak", True), ("mode", "SUB")]}),
+        ("outliner.select_walk", {"type": 'UP_ARROW', "value": 'PRESS'}, {"properties": [("direction", 'UP')]}),
+        ("outliner.select_walk", {"type": 'UP_ARROW', "value": 'PRESS', "shift": True},
+         {"properties": [("direction", 'UP'), ("extend", True)]}),
+        ("outliner.select_walk", {"type": 'DOWN_ARROW', "value": 'PRESS'}, {"properties": [("direction", 'DOWN')]}),
+        ("outliner.select_walk", {"type": 'DOWN_ARROW', "value": 'PRESS', "shift": True},
+         {"properties": [("direction", 'DOWN'), ("extend", True)]}),
+        ("outliner.select_walk", {"type": 'LEFT_ARROW', "value": 'PRESS'}, {"properties": [("direction", 'LEFT')]}),
+        ("outliner.select_walk", {"type": 'LEFT_ARROW', "value": 'PRESS', "shift": True},
+         {"properties": [("direction", 'LEFT'), ("toggle_all", True)]}),
+        ("outliner.select_walk", {"type": 'RIGHT_ARROW', "value": 'PRESS'}, {"properties": [("direction", 'RIGHT')]}),
+        ("outliner.select_walk", {"type": 'RIGHT_ARROW', "value": 'PRESS', "shift": True},
+         {"properties": [("direction", 'RIGHT'), ("toggle_all", True)]}),
+        ("outliner.item_openclose", {"type": 'LEFTMOUSE', "value": 'CLICK'},
          {"properties": [("all", False)]}),
-        ("outliner.item_openclose", {"type": 'RET', "value": 'PRESS', "shift": True},
+        ("outliner.item_openclose", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True},
          {"properties": [("all", True)]}),
-        ("outliner.item_rename", {"type": 'LEFTMOUSE', "value": 'PRESS', "ctrl": True}, None),
+        ("outliner.item_openclose", {"type": 'EVT_TWEAK_L', "value": 'ANY'},
+         {"properties": [("all", False)]}),
         ("outliner.operation", {"type": 'RIGHTMOUSE', "value": 'PRESS'}, None),
         ("outliner.item_drag_drop", {"type": 'EVT_TWEAK_L', "value": 'ANY'}, None),
         ("outliner.item_drag_drop", {"type": 'EVT_TWEAK_L', "value": 'ANY', "shift": True}, None),
@@ -1039,18 +1059,18 @@ def km_view3d(params):
         ("view3d.view_axis", {"type": 'NDOF_BUTTON_TOP', "value": 'PRESS', "shift": True},
          {"properties": [("type", 'TOP'), ("align_active", True)]}),
         # Selection.
-        *((operator,
+        *(("view3d.select",
            {"type": params.select_mouse, "value": params.select_mouse_value, **{m: True for m in mods}},
            {"properties": [(c, True) for c in props]},
-        ) for operator, props, mods in (
-            ("view3d.select", ("deselect_all",) if not params.legacy else (), ()),
-            ("view3d.select", ("toggle",), ("shift",)),
-            ("view3d.select", ("center", "object"), ("ctrl",)),
-            ("view3d.select", ("enumerate",), ("alt",)),
-            ("view3d.select", ("extend", "toggle", "center"), ("shift", "ctrl")),
-            ("view3d.select", ("center", "enumerate"), ("ctrl", "alt")),
-            ("view3d.select", ("toggle", "enumerate"), ("shift", "alt")),
-            ("view3d.select", ("toggle", "center", "enumerate"), ("shift", "ctrl", "alt")),
+        ) for props, mods in (
+            (("deselect_all",) if not params.legacy else (), ()),
+            (("toggle",), ("shift",)),
+            (("center", "object"), ("ctrl",)),
+            (("enumerate",), ("alt",)),
+            (("toggle", "center"), ("shift", "ctrl")),
+            (("center", "enumerate"), ("ctrl", "alt")),
+            (("toggle", "enumerate"), ("shift", "alt")),
+            (("toggle", "center", "enumerate"), ("shift", "ctrl", "alt")),
         )),
         ("view3d.select_box", {"type": 'B', "value": 'PRESS'}, None),
         ("view3d.select_lasso", {"type": params.action_tweak, "value": 'ANY', "ctrl": True},
@@ -1751,6 +1771,11 @@ def km_file_browser(_params):
     )
 
     items.extend([
+        *_template_space_region_type_toggle(
+            toolbar_key={"type": 'T', "value": 'PRESS'},
+        ),
+        ("screen.region_toggle", {"type": 'N', "value": 'PRESS'},
+         {"properties": [("region_type", 'TOOL_PROPS')]}),
         ("file.parent", {"type": 'UP_ARROW', "value": 'PRESS', "alt": True}, None),
         ("file.previous", {"type": 'LEFT_ARROW', "value": 'PRESS', "alt": True}, None),
         ("file.next", {"type": 'RIGHT_ARROW', "value": 'PRESS', "alt": True}, None),
@@ -1762,7 +1787,6 @@ def km_file_browser(_params):
          {"properties": [("data_path", 'space_data.params.show_hidden')]}),
         ("file.directory_new", {"type": 'I', "value": 'PRESS'}, None),
         ("file.smoothscroll", {"type": 'TIMER1', "value": 'ANY', "any": True}, None),
-        ("file.bookmark_toggle", {"type": 'T', "value": 'PRESS'}, None),
         ("file.bookmark_add", {"type": 'B', "value": 'PRESS', "ctrl": True}, None),
     ])
 
@@ -1781,17 +1805,13 @@ def km_file_browser_main(params):
         ("file.execute", {"type": 'LEFTMOUSE', "value": 'DOUBLE_CLICK'},
          {"properties": [("need_active", True)]}),
         ("file.refresh", {"type": 'NUMPAD_PERIOD', "value": 'PRESS'}, None),
-        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK'}, None),
-        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True},
+        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK'},
+         {"properties": [("open", False)]}),
+        ("file.select", {"type": 'LEFTMOUSE', "value": 'DOUBLE_CLICK'}, None),
+        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK', "ctrl": True},
          {"properties": [("extend", True)]}),
-        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True, "ctrl": True},
+        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True},
          {"properties": [("extend", True), ("fill", True)]}),
-        ("file.select", {"type": 'RIGHTMOUSE', "value": 'CLICK'},
-         {"properties": [("open", False)]}),
-        ("file.select", {"type": 'RIGHTMOUSE', "value": 'CLICK', "shift": True},
-         {"properties": [("extend", True), ("open", False)]}),
-        ("file.select", {"type": 'RIGHTMOUSE', "value": 'CLICK', "alt": True},
-         {"properties": [("extend", True), ("fill", True), ("open", False)]}),
         ("file.select_walk", {"type": 'UP_ARROW', "value": 'PRESS'},
          {"properties": [("direction", 'UP')]}),
         ("file.select_walk", {"type": 'UP_ARROW', "value": 'PRESS', "shift": True},
@@ -1823,8 +1843,11 @@ def km_file_browser_main(params):
         ("file.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY'}, None),
         ("file.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY', "shift": True},
          {"properties": [("mode", 'ADD')]}),
-        ("file.rename", {"type": 'LEFTMOUSE', "value": 'PRESS', "ctrl": True}, None),
+        ("file.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY', "ctrl": True},
+         {"properties": [("mode", 'SUB')]}),
         ("file.highlight", {"type": 'MOUSEMOVE', "value": 'ANY', "any": True}, None),
+        ("file.sort_column_ui_context", {"type": 'LEFTMOUSE', "value": 'PRESS', "any": True}, None),
+        op_menu("FILEBROWSER_MT_context_menu", params.context_menu_event),
         ("file.filenum", {"type": 'NUMPAD_PLUS', "value": 'PRESS'},
          {"properties": [("increment", 1)]}),
         ("file.filenum", {"type": 'NUMPAD_PLUS', "value": 'PRESS', "shift": True},
@@ -2929,13 +2952,6 @@ def km_grease_pencil(_params):
         # Erase
         ("gpencil.annotate", {"type": 'RIGHTMOUSE', "value": 'PRESS', "key_modifier": 'D'},
          {"properties": [("mode", 'ERASER'), ("wait_for_input", False)]}),
-
-        # Add blank frame (B because it's easy to reach from D).
-        ("gpencil.blank_frame_add", {"type": 'B', "value": 'PRESS', "key_modifier": 'D'}, None),
-        # Delete active frame - for easier video tutorials/review sessions.
-        # This works even when not in edit mode.
-        ("gpencil.active_frames_delete_all", {"type": 'X', "value": 'PRESS', "key_modifier": 'D'}, None),
-        ("gpencil.active_frames_delete_all", {"type": 'DEL', "value": 'PRESS', "key_modifier": 'D'}, None),
     ])
 
     return keymap
@@ -3018,6 +3034,8 @@ def km_grease_pencil_stroke_edit_mode(params):
         op_menu("VIEW3D_MT_edit_gpencil_delete", {"type": 'DEL', "value": 'PRESS'}),
         ("gpencil.dissolve", {"type": 'X', "value": 'PRESS', "ctrl": True}, None),
         ("gpencil.dissolve", {"type": 'DEL', "value": 'PRESS', "ctrl": True}, None),
+        # Animation menu
+        ("gpencil.blank_frame_add", {"type": 'I', "value": 'PRESS', "shift": True}, None),
         ("gpencil.active_frames_delete_all", {"type": 'X', "value": 'PRESS', "shift": True}, None),
         ("gpencil.active_frames_delete_all", {"type": 'DEL', "value": 'PRESS', "shift": True}, None),
         # Context menu
@@ -3049,7 +3067,7 @@ def km_grease_pencil_stroke_edit_mode(params):
         # Isolate layer
         ("gpencil.layer_isolate", {"type": 'NUMPAD_ASTERIX', "value": 'PRESS'}, None),
         # Move to layer
-        ("gpencil.move_to_layer", {"type": 'M', "value": 'PRESS'}, None),
+        op_menu("GPENCIL_MT_move_to_layer", {"type": 'M', "value": 'PRESS'}),
         # Transform tools
         ("transform.translate", {"type": 'G', "value": 'PRESS'}, None),
         ("transform.translate", {"type": params.select_tweak, "value": 'ANY'}, None),
@@ -3106,6 +3124,19 @@ def km_grease_pencil_stroke_paint_mode(params):
         op_panel("VIEW3D_PT_gpencil_draw_context_menu", params.context_menu_event),
         # Draw delete menu
         op_menu("GPENCIL_MT_gpencil_draw_delete", {"type": 'X', "value": 'PRESS'}),
+        # Animation menu
+        ("gpencil.blank_frame_add", {"type": 'I', "value": 'PRESS', "shift": True}, None),
+        ("gpencil.active_frames_delete_all", {"type": 'X', "value": 'PRESS', "shift": True}, None),
+        ("gpencil.active_frames_delete_all", {"type": 'DEL', "value": 'PRESS', "shift": True}, None),
+        # Interpolation
+        ("gpencil.interpolate", {"type": 'E', "value": 'PRESS', "ctrl": True, "alt": True}, None),
+        ("gpencil.interpolate_sequence", {"type": 'E', "value": 'PRESS', "shift": True, "ctrl": True}, None),
+        # Show/hide
+        ("gpencil.reveal", {"type": 'H', "value": 'PRESS', "alt": True}, None),
+        ("gpencil.hide", {"type": 'H', "value": 'PRESS'},
+         {"properties": [("unselected", False)]}),
+        ("gpencil.hide", {"type": 'H', "value": 'PRESS', "shift": True},
+         {"properties": [("unselected", True)]}),
     ])
 
     return keymap
@@ -3444,6 +3475,8 @@ def km_object_mode(params):
         ("object.duplicate_move", {"type": 'D', "value": 'PRESS', "shift": True}, None),
         ("object.duplicate_move_linked", {"type": 'D', "value": 'PRESS', "alt": True}, None),
         ("object.join", {"type": 'J', "value": 'PRESS', "ctrl": True}, None),
+        ("wm.context_toggle", {"type": 'PERIOD', "value": 'PRESS', "ctrl": True},
+         {"properties": [("data_path", 'tool_settings.use_transform_data_origin')]}),
         ("anim.keyframe_insert_menu", {"type": 'I', "value": 'PRESS'}, None),
         ("anim.keyframe_delete_v3d", {"type": 'I', "value": 'PRESS', "alt": True}, None),
         ("anim.keying_set_active_set", {"type": 'I', "value": 'PRESS', "shift": True, "ctrl": True, "alt": True}, None),
@@ -5433,11 +5466,11 @@ def km_3d_view_tool_edit_mesh_poly_build(params):
         "3D View Tool: Edit Mesh, Poly Build",
         {"space_type": 'VIEW_3D', "region_type": 'WINDOW'},
         {"items": [
-            ("mesh.polybuild_face_at_cursor_move", {"type": params.tool_mouse, "value": 'PRESS'},
+            ("mesh.polybuild_extrude_at_cursor_move", {"type": params.tool_mouse, "value": 'PRESS'},
              {"properties": [("TRANSFORM_OT_translate", [("release_confirm", True)])]}),
-            ("mesh.polybuild_split_at_cursor_move", {"type": params.tool_mouse, "value": 'PRESS', "ctrl": True},
+            ("mesh.polybuild_face_at_cursor_move", {"type": params.tool_mouse, "value": 'PRESS', "ctrl": True},
              {"properties": [("TRANSFORM_OT_translate", [("release_confirm", True)])]}),
-            ("mesh.polybuild_dissolve_at_cursor", {"type": params.tool_mouse, "value": 'CLICK', "alt": True}, None),
+            ("mesh.polybuild_delete_at_cursor", {"type": params.tool_mouse, "value": 'CLICK', "shift": True}, None),
         ]},
     )
 
diff --git a/release/scripts/presets/keyconfig/keymap_data/industry_compatible_data.py b/release/scripts/presets/keyconfig/keymap_data/industry_compatible_data.py
index d41899b7b5b..b23b98f8e6e 100644
--- a/release/scripts/presets/keyconfig/keymap_data/industry_compatible_data.py
+++ b/release/scripts/presets/keyconfig/keymap_data/industry_compatible_data.py
@@ -440,24 +440,41 @@ def km_outliner(params):
     )
 
     items.extend([
-        op_panel("TOPBAR_PT_name", {"type": 'RET', "value": 'PRESS'}, [("keep_open", False)]),
+        ("outliner.item_rename", {"type": 'LEFTMOUSE', "value": 'DOUBLE_CLICK'}, None),
+        ("outliner.item_rename", {"type": 'RET', "value": 'PRESS'}, None),
         ("wm.search_menu", {"type": 'TAB', "value": 'PRESS'}, None),
         ("outliner.highlight_update", {"type": 'MOUSEMOVE', "value": 'ANY', "any": True}, None),
-        ("outliner.item_rename", {"type": 'LEFTMOUSE', "value": 'DOUBLE_CLICK'}, None),
         ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK'},
-         {"properties": [("extend", False), ("recursive", False), ("deselect_all", True)]}),
-        ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True},
-         {"properties": [("extend", True), ("recursive", False)]}),
+         {"properties": [("extend", False), ("deselect_all", True)]}),
         ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "ctrl": True},
-         {"properties": [("extend", False), ("recursive", True)]}),
-        ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True, "ctrl": True},
-         {"properties": [("extend", True), ("recursive", True)]}),
-        ("outliner.select_box", {"type": 'Q', "value": 'PRESS'}, None),
-        ("outliner.item_openclose", {"type": 'RIGHT_ARROW', "value": 'PRESS'},
+         {"properties": [("extend", True), ("deselect_all", True)]}),
+        ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True},
+         {"properties": [("extend", False), ("extend_range", True), ("deselect_all", True)]}),
+        ("outliner.item_activate", {"type": 'LEFTMOUSE', "value": 'CLICK', "ctrl": True, "shift": True},
+         {"properties": [("extend", True), ("extend_range", True), ("deselect_all", True)]}),
+        ("outliner.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY'}, {"properties": [("tweak", True)]}),
+        ("outliner.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY', "shift": True},
+         {"properties": [("tweak", True), ("mode", "ADD")]}),
+        ("outliner.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY', "ctrl": True},
+         {"properties": [("tweak", True), ("mode", "SUB")]}),
+        ("outliner.select_walk", {"type": 'UP_ARROW', "value": 'PRESS'}, {"properties": [("direction", 'UP')]}),
+        ("outliner.select_walk", {"type": 'UP_ARROW', "value": 'PRESS', "shift": True},
+         {"properties": [("direction", 'UP'), ("extend", True)]}),
+        ("outliner.select_walk", {"type": 'DOWN_ARROW', "value": 'PRESS'}, {"properties": [("direction", 'DOWN')]}),
+        ("outliner.select_walk", {"type": 'DOWN_ARROW', "value": 'PRESS', "shift": True},
+         {"properties": [("direction", 'DOWN'), ("extend", True)]}),
+        ("outliner.select_walk", {"type": 'LEFT_ARROW', "value": 'PRESS'}, {"properties": [("direction", 'LEFT')]}),
+        ("outliner.select_walk", {"type": 'LEFT_ARROW', "value": 'PRESS', "shift": True},
+         {"properties": [("direction", 'LEFT'), ("toggle_all", True)]}),
+        ("outliner.select_walk", {"type": 'RIGHT_ARROW', "value": 'PRESS'}, {"properties": [("direction", 'RIGHT')]}),
+        ("outliner.select_walk", {"type": 'RIGHT_ARROW', "value": 'PRESS', "shift": True},
+         {"properties": [("direction", 'RIGHT'), ("toggle_all", True)]}),
+        ("outliner.item_openclose", {"type": 'LEFTMOUSE', "value": 'CLICK'},
          {"properties": [("all", False)]}),
-        ("outliner.item_openclose", {"type": 'LEFT_ARROW', "value": 'PRESS'},
+        ("outliner.item_openclose", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True},
          {"properties": [("all", True)]}),
-        ("outliner.item_rename", {"type": 'RET', "value": 'PRESS'}, None),
+        ("outliner.item_openclose", {"type": 'EVT_TWEAK_L', "value": 'ANY'},
+         {"properties": [("all", False)]}),
         ("outliner.operation", {"type": 'RIGHTMOUSE', "value": 'PRESS'}, None),
         ("outliner.item_drag_drop", {"type": 'EVT_TWEAK_L', "value": 'ANY'}, None),
         ("outliner.item_drag_drop", {"type": 'EVT_TWEAK_L', "value": 'ANY', "shift": True}, None),
@@ -1179,13 +1196,13 @@ def km_file_browser_main(params):
         ("file.execute", {"type": 'LEFTMOUSE', "value": 'DOUBLE_CLICK'},
          {"properties": [("need_active", True)]}),
         ("file.refresh", {"type": 'R', "value": 'PRESS', "ctrl": True}, None),
-        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK'}, None),
-        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True},
+        ("file.select", {"type": 'LEFTMOUSE', "value": 'DOUBLE_CLICK'}, None),
+        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK'},
+         {"properties": [("open", False)]}),
+        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK', "ctrl": True},
          {"properties": [("extend", True)]}),
-        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True, "ctrl": True},
+        ("file.select", {"type": 'LEFTMOUSE', "value": 'CLICK', "shift": True,},
          {"properties": [("extend", True), ("fill", True)]}),
-        ("file.select", {"type": 'RIGHTMOUSE', "value": 'CLICK'},
-         {"properties": [("open", False)]}),
         ("file.select", {"type": 'RIGHTMOUSE', "value": 'CLICK', "shift": True},
          {"properties": [("extend", True), ("open", False)]}),
         ("file.select", {"type": 'RIGHTMOUSE', "value": 'CLICK', "alt": True},
@@ -1222,6 +1239,8 @@ def km_file_browser_main(params):
         ("file.select_box", {"type": 'EVT_TWEAK_L', "value": 'ANY', "shift": True},
          {"properties": [("mode", 'ADD')]}),
         ("file.highlight", {"type": 'MOUSEMOVE', "value": 'ANY', "any": True}, None),
+        ("file.sort_column_ui_context", {"type": 'LEFTMOUSE', "value": 'PRESS', "any": True}, None),
+        op_menu("FILEBROWSER_MT_context_menu", {"type": 'RIGHTMOUSE', "value": 'PRESS'}),
         ("file.filenum", {"type": 'NUMPAD_PLUS', "value": 'PRESS'},
          {"properties": [("increment", 1)]}),
         ("file.filenum", {"type": 'NUMPAD_PLUS', "value": 'PRESS', "shift": True},
@@ -2190,22 +2209,6 @@ def km_grease_pencil(_params):
     )
 
     items.extend([
-        # Draw
-        ("gpencil.annotate", {"type": 'LEFTMOUSE', "value": 'PRESS', "key_modifier": 'D'},
-         {"properties": [("mode", 'DRAW'), ("wait_for_input", False)]}),
-        # Draw - straight lines
-        ("gpencil.annotate", {"type": 'LEFTMOUSE', "value": 'PRESS', "alt": True, "key_modifier": 'D'},
-         {"properties": [("mode", 'DRAW_STRAIGHT'), ("wait_for_input", False)]}),
-        # Draw - poly lines
-        ("gpencil.annotate", {"type": 'LEFTMOUSE', "value": 'PRESS', "shift": True, "alt": True, "key_modifier": 'D'},
-         {"properties": [("mode", 'DRAW_POLY'), ("wait_for_input", False)]}),
-        # Erase
-        ("gpencil.annotate", {"type": 'RIGHTMOUSE', "value": 'PRESS', "key_modifier": 'D'},
-         {"properties": [("mode", 'ERASER'), ("wait_for_input", False)]}),
-
-        # Add blank frame (B because it's easy to reach from D).
-        ("gpencil.blank_frame_add", {"type": 'B', "value": 'PRESS', "key_modifier": 'D'}, None),
-        # Delete active frame - for easier video tutorials/review sessions.
         # This works even when not in edit mode.
         ("gpencil.active_frames_delete_all", {"type": 'BACK_SPACE', "value": 'PRESS', "key_modifier": 'D'}, None),
         ("gpencil.active_frames_delete_all", {"type": 'DEL', "value": 'PRESS', "key_modifier": 'D'}, None),
@@ -2855,6 +2858,14 @@ def km_image_paint(params):
          {"properties": [("data_path", 'tool_settings.image_paint.brush.use_smooth_stroke')]}),
         op_menu("VIEW3D_MT_angle_control", {"type": 'R', "value": 'PRESS'}),
         op_panel("VIEW3D_PT_paint_texture_context_menu", {"type": 'RIGHTMOUSE', "value": 'PRESS'}),
+        # Tools
+        ("paint.brush_select", {"type": 'D', "value": 'PRESS'},
+         {"properties": [("image_tool", 'DRAW')]}),
+        ("paint.brush_select", {"type": 'B', "value": 'PRESS'},
+         {"properties": [("image_tool", 'SOFTEN')]}),
+        ("paint.brush_select", {"type": 'G', "value": 'PRESS'},
+         {"properties": [("image_tool", 'FILL')]}),
+        op_tool("builtin.select_box", {"type": 'Q', "value": 'PRESS'}),
     ])
 
     return keymap
@@ -2897,6 +2908,12 @@ def km_vertex_paint(params):
          {"properties": [("data_path", 'tool_settings.vertex_paint.brush.use_smooth_stroke')]}),
         op_menu("VIEW3D_MT_angle_control", {"type": 'R', "value": 'PRESS'}),
         op_panel("VIEW3D_PT_paint_vertex_context_menu", {"type": 'RIGHTMOUSE', "value": 'PRESS'}),
+        # Tools
+        ("paint.brush_select", {"type": 'D', "value": 'PRESS'},
+         {"properties": [("vertex_tool", 'DRAW')]}),
+        ("paint.brush_select", {"type": 'B', "value": 'PRESS'},
+         {"properties": [("vertex_tool", 'BLUR')]}),
+        op_tool("builtin.select_box", {"type": 'Q', "value": 'PRESS'}),
     ])
 
     return keymap
@@ -2912,7 +2929,6 @@ def km_weight_paint(params):
 
     items.extend([
         ("paint.weight_paint", {"type": 'LEFTMOUSE', "value": 'PRESS'}, None),
-        ("paint.weight_sample", {"type": 'I', "value": 'PRESS'}, None),
         ("paint.weight_sample_group", {"type": 'LEFTMOUSE', "value": 'PRESS', "shift": True}, None),
         ("brush.scale_size", {"type": 'LEFT_BRACKET', "value": 'PRESS'},
          {"properties": [("scalar", 0.9)]}),
@@ -2928,6 +2944,13 @@ def km_weight_paint(params):
         op_panel("VIEW3D_PT_paint_weight_context_menu", {"type": 'RIGHTMOUSE', "value": 'PRESS'}),
         # Bone selection for combined weight paint + pose mode.
         ("view3d.select", {"type": 'LEFTMOUSE', "value": 'PRESS', "ctrl": True}, None),
+         # Tools
+        ("paint.brush_select", {"type": 'D', "value": 'PRESS'},
+         {"properties": [("weight_tool", 'DRAW')]}),
+        ("paint.brush_select", {"type": 'B', "value": 'PRESS'},
+         {"properties": [("weight_tool", 'BLUR')]}),
+        op_tool("builtin.sample_weight", {"type": 'I', "value": 'PRESS'}),
+        op_tool("builtin.select_box", {"type": 'Q', "value": 'PRESS'}),
     ])
 
     return keymap
diff --git a/release/scripts/startup/bl_operators/anim.py b/release/scripts/startup/bl_operators/anim.py
index 1470aed4d55..d55644f19c7 100644
--- a/release/scripts/startup/bl_operators/anim.py
+++ b/release/scripts/startup/bl_operators/anim.py
@@ -105,7 +105,7 @@ class ANIM_OT_keying_set_export(Operator):
             # - idtype_list is used to get the list of id-datablocks from
             #   bpy.data.* since this info isn't available elsewhere
             # - id.bl_rna.name gives a name suitable for UI,
-            #   with a capitalised first letter, but we need
+            #   with a capitalized first letter, but we need
             #   the plural form that's all lower case
             # - special handling is needed for "nested" ID-blocks
             #   (e.g. nodetree in Material)
diff --git a/release/scripts/startup/bl_operators/userpref.py b/release/scripts/startup/bl_operators/userpref.py
index 034aa9fff6c..6ec6855296c 100644
--- a/release/scripts/startup/bl_operators/userpref.py
+++ b/release/scripts/startup/bl_operators/userpref.py
@@ -667,6 +667,7 @@ class PREFERENCES_OT_addon_install(Operator):
                 info = addon_utils.module_bl_info(mod)
 
                 # show the newly installed addon.
+                context.preferences.view.show_addons_enabled_only = False
                 context.window_manager.addon_filter = 'All'
                 context.window_manager.addon_search = info["name"]
                 break
@@ -796,6 +797,7 @@ class PREFERENCES_OT_addon_show(Operator):
             info["show_expanded"] = True
 
             context.preferences.active_section = 'ADDONS'
+            context.preferences.view.show_addons_enabled_only = False
             context.window_manager.addon_filter = 'All'
             context.window_manager.addon_search = info["name"]
             bpy.ops.screen.userpref_show('INVOKE_DEFAULT')
diff --git a/release/scripts/startup/bl_operators/wm.py b/release/scripts/startup/bl_operators/wm.py
index d59ad0694db..f7b4a4e80e6 100644
--- a/release/scripts/startup/bl_operators/wm.py
+++ b/release/scripts/startup/bl_operators/wm.py
@@ -25,6 +25,7 @@ from bpy.types import (
 )
 from bpy.props import (
     BoolProperty,
+    CollectionProperty,
     EnumProperty,
     FloatProperty,
     IntProperty,
@@ -846,6 +847,86 @@ class WM_OT_url_open(Operator):
         return {'FINISHED'}
 
 
+class WM_OT_url_open_preset(Operator):
+    """Open a preset website in the web-browser"""
+    bl_idname = "wm.url_open_preset"
+    bl_label = "Open Preset Website"
+    bl_options = {'INTERNAL'}
+
+    type: EnumProperty(
+        name="Site",
+        items=lambda self, _context: (
+            item for (item, _) in WM_OT_url_open_preset.preset_items
+        ),
+    )
+
+    id: StringProperty(
+        name="Identifier",
+        description="Optional identifier",
+    )
+
+    def _url_from_bug(self, _context):
+        from bl_ui_utils.bug_report_url import url_prefill_from_blender
+        return url_prefill_from_blender()
+
+    def _url_from_bug_addon(self, _context):
+        from bl_ui_utils.bug_report_url import url_prefill_from_blender
+        return url_prefill_from_blender(addon_info=self.id)
+
+    def _url_from_release_notes(self, _context):
+        return "https://www.blender.org/download/releases/%d-%d/" % bpy.app.version[:2]
+
+    def _url_from_manual(self, _context):
+        if bpy.app.version_cycle in {"rc", "release"}:
+            manual_version = "%d.%d" % bpy.app.version[:2]
+        else:
+            manual_version = "dev"
+        return "https://docs.blender.org/manual/en/" + manual_version + "/"
+
+    # This list is: (enum_item, url) pairs.
+    # Allow dynamically extending.
+    preset_items = [
+        # Dynamic URL's.
+        (('BUG', "Bug",
+          "Report a bug with pre-filled version information"),
+         _url_from_bug),
+        (('BUG_ADDON', "Add-On Bug",
+          "Report a bug in an add-on"),
+         _url_from_bug_addon),
+        (('RELEASE_NOTES', "Release Notes",
+          "Read about whats new in this version of Blender"),
+         _url_from_release_notes),
+        (('MANUAL', "Manual",
+          "The reference manual for this version of Blender"),
+         _url_from_manual),
+
+        # Static URL's.
+        (('FUND', "Development Fund",
+          "The donation program to support maintenance and improvements"),
+         "https://fund.blender.org"),
+        (('BLENDER', "blender.org",
+          "Blender's official web-site"),
+         "https://www.blender.org"),
+        (('CREDITS', "Credits",
+          "Lists committers to Blender's source code"),
+         "https://www.blender.org/about/credits/"),
+    ]
+
+    def execute(self, context):
+        url = None
+        type = self.type
+        for (item_id, _, _), url in self.preset_items:
+            if item_id == type:
+                if callable(url):
+                    url = url(self, context)
+                break
+
+        import webbrowser
+        webbrowser.open(url)
+
+        return {'FINISHED'}
+
+
 class WM_OT_path_open(Operator):
     """Open a path in a file browser"""
     bl_idname = "wm.path_open"
@@ -1094,6 +1175,7 @@ rna_vector_subtype_items = (
     ('QUATERNION', "Quaternion Rotation", "Quaternion rotation (affects NLA blending)"),
 )
 
+
 class WM_OT_properties_edit(Operator):
     bl_idname = "wm.properties_edit"
     bl_label = "Edit Property"
@@ -1683,6 +1765,423 @@ class WM_OT_toolbar(Operator):
         return {'FINISHED'}
 
 
+class BatchRenameAction(bpy.types.PropertyGroup):
+    # category: StringProperty()
+    type: EnumProperty(
+        name="Operation",
+        items=(
+            ('REPLACE', "Find/Replace", "Replace text in the name"),
+            ('SET', "Set Name", "Set a new name or prefix/suffix the existing one"),
+            ('STRIP', "Strip Characters", "Strip leading/trailing text from the name"),
+            ('CASE', "Change Case", "Change case of each name"),
+        ),
+    )
+
+    # We could split these into sub-properties, however it's not so important.
+
+    # type: 'SET'.
+    set_name: StringProperty(name="Name")
+    set_method: EnumProperty(
+        name="Method",
+        items=(
+            ('NEW', "New", ""),
+            ('PREFIX', "Prefix", ""),
+            ('SUFFIX', "Suffix", ""),
+        ),
+        default='SUFFIX',
+    )
+
+    # type: 'STRIP'.
+    strip_chars: EnumProperty(
+        name="Strip Characters",
+        options={'ENUM_FLAG'},
+        items=(
+            ('SPACE', "Spaces", ""),
+            ('DIGIT', "Digits", ""),
+            ('PUNCT', "Punctuation", ""),
+        ),
+    )
+
+    # type: 'STRIP'.
+    strip_part: EnumProperty(
+        name="Strip Part",
+        options={'ENUM_FLAG'},
+        items=(
+            ('START', "Start", ""),
+            ('END', "End", ""),
+        ),
+    )
+
+    # type: 'REPLACE'.
+    replace_src: StringProperty(name="Find")
+    replace_dst: StringProperty(name="Replace")
+    replace_match_case: BoolProperty(name="Case Sensitive")
+    replace_regex: BoolProperty(
+        name="Regular Expression",
+        description="Use regular expressions to match text in the 'Find' field"
+    )
+
+    # type: 'CASE'.
+    case_method: EnumProperty(
+        name="Case",
+        items=(
+            ('UPPER', "Upper Case", ""),
+            ('LOWER', "Lower Case", ""),
+            ('TITLE', "Title Caps", ""),
+        ),
+    )
+
+    # Weak, add/remove as properties.
+    op_add: BoolProperty()
+    op_remove: BoolProperty()
+
+
+class WM_OT_batch_rename(Operator):
+    bl_idname = "wm.batch_rename"
+    bl_label = "Batch Rename"
+
+    bl_options = {'UNDO', 'INTERNAL'}
+
+    data_type: EnumProperty(
+        name="Type",
+        items=(
+            ('OBJECT', "Objects", ""),
+            ('MATERIAL', "Materials", ""),
+            None,
+            # Enum identifiers are compared with 'object.type'.
+            ('MESH', "Meshes", ""),
+            ('CURVE', "Curves", ""),
+            ('META', "Meta Balls", ""),
+            ('ARMATURE', "Armatures", ""),
+            ('LATTICE', "Lattices", ""),
+            ('GPENCIL', "Grease Pencils", ""),
+            ('CAMERA', "Cameras", ""),
+            ('SPEAKER', "Speakers", ""),
+            ('LIGHT_PROBE', "Light Probes", ""),
+            None,
+            ('BONE', "Bones", ""),
+            ('NODE', "Nodes", ""),
+            ('SEQUENCE_STRIP', "Sequence Strips", ""),
+        ),
+        description="Type of data to rename",
+    )
+
+    data_source: EnumProperty(
+        name="Source",
+        items=(
+            ('SELECT', "Selected", ""),
+            ('ALL', "All", ""),
+        ),
+    )
+
+    actions: CollectionProperty(type=BatchRenameAction)
+
+    @staticmethod
+    def _data_from_context(context, data_type, only_selected, check_context=False):
+
+        mode = context.mode
+        scene = context.scene
+        space = context.space_data
+        space_type = None if (space is None) else space.type
+
+        data = None
+        if space_type == 'SEQUENCE_EDITOR':
+            data_type_test = 'SEQUENCE_STRIP'
+            if check_context:
+                return data_type_test
+            if data_type == data_type_test:
+                data = (
+                    # TODO, we don't have access to seqbasep, this won't work when inside metas.
+                    [seq for seq in context.scene.sequence_editor.sequences_all if seq.select]
+                    if only_selected else
+                    context.scene.sequence_editor.sequences_all,
+                    "name",
+                    "Strip(s)",
+                )
+        elif space_type == 'NODE_EDITOR':
+            data_type_test = 'NODE'
+            if check_context:
+                return data_type_test
+            if data_type == data_type_test:
+                data = (
+                    context.selected_nodes
+                    if only_selected else
+                    list(space.node_tree.nodes),
+                    "name",
+                    "Node(s)",
+                )
+        else:
+            if mode == 'POSE' or (mode == 'WEIGHT_PAINT' and context.pose_object):
+                data_type_test = 'BONE'
+                if check_context:
+                    return data_type_test
+                if data_type == data_type_test:
+                    data = (
+                        [pchan.bone for pchan in context.selected_pose_bones]
+                        if only_selected else
+                        [pchan.bone for ob in context.objects_in_mode_unique_data for pbone in ob.pose.bones],
+                        "name",
+                        "Bone(s)",
+                    )
+            elif mode == 'EDIT_ARMATURE':
+                data_type_test = 'BONE'
+                if check_context:
+                    return data_type_test
+                if data_type == data_type_test:
+                    data = (
+                        context.selected_editable_bones
+                        if only_selected else
+                        [ebone for ob in context.objects_in_mode_unique_data for ebone in ob.data.edit_bones],
+                        "name",
+                        "Edit Bone(s)",
+                    )
+
+        if check_context:
+            return 'OBJECT'
+
+        object_data_type_attrs_map = {
+            'MESH': ("meshes", "Mesh(es)"),
+            'CURVE': ("curves", "Curve(s)"),
+            'META': ("metaballs", "MetaBall(s)"),
+            'ARMATURE': ("armatures", "Armature(s)"),
+            'LATTICE': ("lattices", "Lattice(s)"),
+            'GPENCIL': ("grease_pencils", "Grease Pencil(s)"),
+            'CAMERA': ("cameras", "Camera(s)"),
+            'SPEAKER': ("speakers", "Speaker(s)"),
+            'LIGHT_PROBE': ("light_probes", "LightProbe(s)"),
+        }
+
+        # Finish with space types.
+        if data is None:
+
+            if data_type == 'OBJECT':
+                data = (
+                    context.selected_editable_objects
+                    if only_selected else
+                    [id for id in bpy.data.objects if id.library is None],
+                    "name",
+                    "Object(s)",
+                )
+            elif data_type == 'MATERIAL':
+                data = (
+                    tuple(set(
+                        slot.material
+                        for ob in context.selected_objects
+                        for slot in ob.material_slots
+                        if slot.material is not None
+                    ))
+                    if only_selected else
+                    [id for id in bpy.data.materials if id.library is None],
+                    "name",
+                    "Material(s)",
+                )
+            elif data_type in object_data_type_attrs_map.keys():
+                attr, descr = object_data_type_attrs_map[data_type]
+                data = (
+                    tuple(set(
+                        id
+                        for ob in context.selected_objects
+                        if ob.type == data_type
+                        for id in (ob.data,)
+                        if id is not None and id.library is None
+                    ))
+                    if only_selected else
+                    [id for id in getattr(bpy.data, attr) if id.library is None],
+                    "name",
+                    descr,
+                )
+
+
+        return data
+
+    @staticmethod
+    def _apply_actions(actions, name):
+        import string
+        import re
+
+        for action in actions:
+            ty = action.type
+            if ty == 'SET':
+                text = action.set_name
+                method = action.set_method
+                if method == 'NEW':
+                    name = text
+                elif method == 'PREFIX':
+                    name = text + name
+                elif method == 'SUFFIX':
+                    name = name + text
+                else:
+                    assert(0)
+
+            elif ty == 'STRIP':
+                chars = action.strip_chars
+                chars_strip = (
+                    "{:s}{:s}{:s}"
+                ).format(
+                    string.punctuation if 'PUNCT' in chars else "",
+                    string.digits if 'DIGIT' in chars else "",
+                    " " if 'SPACE' in chars else "",
+                )
+                part = action.strip_part
+                if 'START' in part:
+                    name = name.lstrip(chars_strip)
+                if 'END' in part:
+                    name = name.rstrip(chars_strip)
+
+            elif ty == 'REPLACE':
+                if action.replace_regex:
+                    replace_src = action.replace_src
+                else:
+                    replace_src = re.escape(action.replace_src)
+                name = re.sub(
+                    replace_src,
+                    re.escape(action.replace_dst),
+                    name,
+                    flags=(
+                        0 if action.replace_match_case else
+                        re.IGNORECASE
+                    ),
+                )
+            elif ty == 'CASE':
+                method = action.case_method
+                if method == 'UPPER':
+                    name = name.upper()
+                elif method == 'LOWER':
+                    name = name.lower()
+                elif method == 'TITLE':
+                    name = name.title()
+                else:
+                    assert(0)
+            else:
+                assert(0)
+        return name
+
+    def _data_update(self, context):
+        only_selected = self.data_source == 'SELECT'
+
+        self._data = self._data_from_context(context, self.data_type, only_selected)
+        if self._data is None:
+            self.data_type = self._data_from_context(context, None, False, check_context=True)
+            self._data = self._data_from_context(context, self.data_type, only_selected)
+
+        self._data_source_prev = self.data_source
+        self._data_type_prev = self.data_type
+
+    def draw(self, context):
+        import re
+
+        layout = self.layout
+
+        split = layout.split(factor=0.5)
+        split.label(text="Data Type:")
+        split.prop(self, "data_type", text="")
+
+        split = layout.split(factor=0.5)
+        split.label(text="Rename {:d} {:s}:".format(len(self._data[0]), self._data[2]))
+        split.row().prop(self, "data_source", expand=True)
+
+        for action in self.actions:
+            box = layout.box()
+
+            row = box.row(align=True)
+            row.prop(action, "type", text="")
+            row.prop(action, "op_add", text="", icon='ADD')
+            row.prop(action, "op_remove", text="", icon='REMOVE')
+
+            ty = action.type
+            if ty == 'SET':
+                box.prop(action, "set_method")
+                box.prop(action, "set_name")
+            elif ty == 'STRIP':
+                box.row().prop(action, "strip_chars")
+                box.row().prop(action, "strip_part")
+            elif ty == 'REPLACE':
+
+                row = box.row()
+                re_error = None
+                if action.replace_regex:
+                    try:
+                        re.compile(action.replace_src)
+                    except Exception as ex:
+                        row.alert = True
+                        re_error = str(ex)
+                row.prop(action, "replace_src")
+                if re_error is not None:
+                    box.label(text=re_error)
+
+                box.row().prop(action, "replace_dst")
+                row = box.row()
+                row.prop(action, "replace_match_case")
+                row.prop(action, "replace_regex")
+            elif ty == 'CASE':
+                box.row().prop(action, "case_method", expand=True)
+
+    def check(self, context):
+        changed = False
+        for i, action in enumerate(self.actions):
+            if action.op_add:
+                action.op_add = False
+                self.actions.add()
+                if i + 2 != len(self.actions):
+                    self.actions.move(len(self.actions) - 1, i + 1)
+                changed = True
+                break
+            if action.op_remove:
+                action.op_remove = False
+                if len(self.actions) > 1:
+                    self.actions.remove(i)
+                changed = True
+                break
+
+        if (
+                (self._data_source_prev != self.data_source) or
+                (self._data_type_prev != self.data_type)
+        ):
+            self._data_update(context)
+            changed = True
+
+        return changed
+
+    def execute(self, context):
+        import re
+
+        seq, attr, descr = self._data
+
+        actions = self.actions
+
+        # Sanitize actions.
+        for action in actions:
+            if action.replace_regex:
+                try:
+                    re.compile(action.replace_src)
+                except Exception as ex:
+                    self.report({'ERROR'}, "Invalid regular expression: " + str(ex))
+                    return {'CANCELLED'}
+
+        total_len = 0
+        change_len = 0
+        for item in seq:
+            name_src = getattr(item, attr)
+            name_dst = self._apply_actions(actions, name_src)
+            if name_src != name_dst:
+                setattr(item, attr, name_dst)
+                change_len += 1
+            total_len += 1
+
+        self.report({'INFO'}, "Renamed {:d} of {:d} {:s}".format(total_len, change_len, descr))
+
+        return {'FINISHED'}
+
+    def invoke(self, context, event):
+
+        self._data_update(context)
+
+        if not self.actions:
+            self.actions.add()
+        wm = context.window_manager
+        return wm.invoke_props_dialog(self, width=400)
+
+
 class WM_MT_splash(Menu):
     bl_label = "Splash"
 
@@ -1808,26 +2307,13 @@ class WM_MT_splash(Menu):
         if found_recent:
             col2_title.label(text="Recent Files")
         else:
-            if bpy.app.version_cycle in {'rc', 'release'}:
-                manual_version = '%d.%d' % bpy.app.version[:2]
-            else:
-                manual_version = 'dev'
 
             # Links if no recent files
             col2_title.label(text="Getting Started")
 
-            col2.operator(
-                "wm.url_open", text="Manual", icon='URL'
-            ).url = "https://docs.blender.org/manual/en/" + manual_version + "/"
-            col2.operator(
-                "wm.url_open", text="Release Notes", icon='URL',
-            ).url = "https://www.blender.org/download/releases/%d-%d/" % bpy.app.version[:2]
-            col2.operator(
-                "wm.url_open", text="Blender Website", icon='URL',
-            ).url = "https://www.blender.org"
-            col2.operator(
-                "wm.url_open", text="Credits", icon='URL',
-            ).url = "https://www.blender.org/about/credits/"
+            col2.operator("wm.url_open_preset", text="Manual", icon='URL').type = 'MANUAL'
+            col2.operator("wm.url_open_preset", text="Blender Website", icon='URL').type = 'BLENDER'
+            col2.operator("wm.url_open_preset", text="Credits", icon='URL').type = 'CREDITS'
 
         layout.separator()
 
@@ -1840,12 +2326,9 @@ class WM_MT_splash(Menu):
         col1.operator("wm.recover_last_session", icon='RECOVER_LAST')
 
         col2 = split.column()
-        col2.operator(
-            "wm.url_open", text="Release Notes", icon='URL',
-        ).url = "https://www.blender.org/download/releases/%d-%d/" % bpy.app.version[:2]
-        col2.operator(
-            "wm.url_open", text="Development Fund", icon='FUND'
-        ).url = "https://fund.blender.org"
+
+        col2.operator("wm.url_open_preset", text="Release Notes", icon='URL').type = 'RELEASE_NOTES'
+        col2.operator("wm.url_open_preset", text="Development Fund", icon='FUND').type = 'FUND'
 
         layout.separator()
         layout.separator()
@@ -1911,8 +2394,11 @@ classes = (
     WM_OT_owner_disable,
     WM_OT_owner_enable,
     WM_OT_url_open,
+    WM_OT_url_open_preset,
     WM_OT_tool_set_by_id,
     WM_OT_tool_set_by_index,
     WM_OT_toolbar,
+    BatchRenameAction,
+    WM_OT_batch_rename,
     WM_MT_splash,
 )
diff --git a/release/scripts/startup/bl_ui/__init__.py b/release/scripts/startup/bl_ui/__init__.py
index 0bb96d49d32..a4096741b4c 100644
--- a/release/scripts/startup/bl_ui/__init__.py
+++ b/release/scripts/startup/bl_ui/__init__.py
@@ -128,8 +128,6 @@ def register():
         items = [
             ('All', "All", "All Add-ons"),
             ('User', "User", "All Add-ons Installed by User"),
-            ('Enabled', "Enabled", "All Enabled Add-ons"),
-            ('Disabled', "Disabled", "All Disabled Add-ons"),
         ]
 
         items_unique = set()
diff --git a/release/scripts/startup/bl_ui/properties_constraint.py b/release/scripts/startup/bl_ui/properties_constraint.py
index 6fa953574cb..0ba9fd0630c 100644
--- a/release/scripts/startup/bl_ui/properties_constraint.py
+++ b/release/scripts/startup/bl_ui/properties_constraint.py
@@ -370,6 +370,8 @@ class ConstraintButtonsPanel:
     def COPY_ROTATION(self, _context, layout, con):
         self.target_template(layout, con)
 
+        layout.prop(con, "euler_order", text="Order")
+
         split = layout.split()
 
         col = split.column()
@@ -430,6 +432,7 @@ class ConstraintButtonsPanel:
         row.prop(con, "use_z", text="Z")
 
         layout.prop(con, "power")
+        layout.prop(con, "use_make_uniform")
 
         row = layout.row()
         row.prop(con, "use_offset")
@@ -684,6 +687,9 @@ class ConstraintButtonsPanel:
         col.row().label(text="Source:")
         col.row().prop(con, "map_from", expand=True)
 
+        if con.map_from == 'ROTATION':
+            layout.prop(con, "from_rotation_mode", text="Mode")
+
         split = layout.split()
         ext = "" if con.map_from == 'LOCATION' else "_rot" if con.map_from == 'ROTATION' else "_scale"
 
@@ -726,6 +732,9 @@ class ConstraintButtonsPanel:
         col.label(text="Destination:")
         col.row().prop(con, "map_to", expand=True)
 
+        if con.map_to == 'ROTATION':
+            layout.prop(con, "to_euler_order", text="Order")
+
         split = layout.split()
         ext = "" if con.map_to == 'LOCATION' else "_rot" if con.map_to == 'ROTATION' else "_scale"
 
diff --git a/release/scripts/startup/bl_ui/properties_data_empty.py b/release/scripts/startup/bl_ui/properties_data_empty.py
index 6650aef61aa..4ce87b85410 100644
--- a/release/scripts/startup/bl_ui/properties_data_empty.py
+++ b/release/scripts/startup/bl_ui/properties_data_empty.py
@@ -60,6 +60,7 @@ class DATA_PT_empty(DataButtonsPanel, Panel):
             col.row().prop(ob, "empty_image_side", text="Side", expand=True)
             col.prop(ob, "show_empty_image_orthographic", text="Display Orthographic")
             col.prop(ob, "show_empty_image_perspective", text="Display Perspective")
+            col.prop(ob, "show_empty_image_only_axis_aligned")
 
 
 class DATA_PT_empty_image(DataButtonsPanel, Panel):
diff --git a/release/scripts/startup/bl_ui/properties_data_gpencil.py b/release/scripts/startup/bl_ui/properties_data_gpencil.py
index 04424b2a97a..62cde078a2d 100644
--- a/release/scripts/startup/bl_ui/properties_data_gpencil.py
+++ b/release/scripts/startup/bl_ui/properties_data_gpencil.py
@@ -84,7 +84,7 @@ class GPENCIL_MT_layer_context_menu(Menu):
         ob = context.object
         gpd = ob.data
 
-        layout.operator("gpencil.layer_duplicate", icon='ADD')  # XXX: needs a dedicated icon
+        layout.operator("gpencil.layer_duplicate", icon='DUPLICATE')
 
         layout.separator()
 
@@ -123,29 +123,16 @@ class DATA_PT_gpencil_layers(DataButtonsPanel, Panel):
 
     def draw_layers(self, _context, layout, gpd):
 
-        row = layout.row()
+        gpl = gpd.layers.active
 
-        col = row.column()
+        row = layout.row()
         layer_rows = 7
+        
+        col = row.column()
         col.template_list("GPENCIL_UL_layer", "", gpd, "layers", gpd.layers, "active_index",
                           rows=layer_rows, sort_reverse=True, sort_lock=True)
 
-        gpl = gpd.layers.active
-
-        if gpl:
-            srow = col.row(align=True)
-            srow.prop(gpl, "blend_mode", text="Blend")
-
-            srow = col.row(align=True)
-            srow.prop(gpl, "opacity", text="Opacity", slider=True)
-            srow.prop(gpl, "mask_layer", text="",
-                      icon='MOD_MASK' if gpl.mask_layer else 'LAYER_ACTIVE')
-
-            srow = col.row(align=True)
-            srow.prop(gpl, "use_solo_mode", text="Show Only On Keyframed")
-
         col = row.column()
-
         sub = col.column(align=True)
         sub.operator("gpencil.layer_add", icon='ADD', text="")
         sub.operator("gpencil.layer_remove", icon='REMOVE', text="")
@@ -165,6 +152,23 @@ class DATA_PT_gpencil_layers(DataButtonsPanel, Panel):
                 sub = col.column(align=True)
                 sub.operator("gpencil.layer_isolate", icon='LOCKED', text="").affect_visibility = False
                 sub.operator("gpencil.layer_isolate", icon='RESTRICT_VIEW_ON', text="").affect_visibility = True
+        
+        # Layer main properties
+        row = layout.row()        
+        col = layout.column(align=True)
+
+        if gpl:
+            
+            layout = self.layout
+            layout.use_property_split = True
+            layout.use_property_decorate = True
+            col = layout.column(align=True)
+
+            col = layout.row(align=True)
+            col.prop(gpl, "blend_mode", text="Blend")
+
+            col = layout.row(align=True)
+            col.prop(gpl, "opacity", text="Opacity", slider=True)
 
 
 class DATA_PT_gpencil_layer_adjustments(LayerDataButtonsPanel, Panel):
@@ -241,6 +245,9 @@ class DATA_PT_gpencil_layer_display(LayerDataButtonsPanel, Panel):
         col = layout.row(align=True)
         col.prop(gpl, "channel_color")
 
+        col = layout.row(align=True)
+        col.prop(gpl, "use_solo_mode", text="Show Only On Keyframed")
+
 
 class DATA_PT_gpencil_onion_skinning(DataButtonsPanel, Panel):
     bl_label = "Onion Skinning"
diff --git a/release/scripts/startup/bl_ui/properties_data_modifier.py b/release/scripts/startup/bl_ui/properties_data_modifier.py
index ae813c06511..d52ed7dec4e 100644
--- a/release/scripts/startup/bl_ui/properties_data_modifier.py
+++ b/release/scripts/startup/bl_ui/properties_data_modifier.py
@@ -1743,11 +1743,20 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
         row.prop_search(md, "vertex_group", ob, "vertex_groups", text="")
         row.prop(md, "invert_vertex", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -1775,11 +1784,20 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
         row.prop_search(md, "vertex_group", ob, "vertex_groups", text="")
         row.prop(md, "invert_vertex", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -1799,11 +1817,19 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
 
         col = layout.column()
         col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -1834,11 +1860,18 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
 
         col = layout.column()
         col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -1871,11 +1904,20 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
         row.prop_search(md, "vertex_group", ob, "vertex_groups", text="")
         row.prop(md, "invert_vertex", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -1898,11 +1940,18 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
 
         col = layout.column()
         col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -1970,11 +2019,18 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
 
         col = layout.column()
         col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -1998,19 +2054,32 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
         row.prop(md, "create_materials")
         if md.opacity_mode == 'COLOR':
             row.prop(md, "modify_color", text="Change")
+        if md.opacity_mode == 'MATERIAL':
+            row = layout.row()
+            row.prop(md, "create_materials")
+            row.prop(md, "modify_color", text="Change")
+        else:
+            col = layout.column()
+            col.separator()
+            col.label(text="Vertex Group:")
+            row = col.row(align=True)
+            row.prop_search(md, "vertex_group", ob, "vertex_groups", text="")
+            row.prop(md, "invert_vertex", text="", icon='ARROW_LEFTRIGHT')
 
         col = layout.column()
         col.separator()
-        col.label(text="Vertex Group:")
-        row = col.row(align=True)
-        row.prop_search(md, "vertex_group", ob, "vertex_groups", text="")
-        row.prop(md, "invert_vertex", text="", icon='ARROW_LEFTRIGHT')
 
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -2058,11 +2127,18 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
 
         col = layout.column()
         col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -2144,11 +2220,20 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
         row.prop_search(md, "vertex_group", ob, "vertex_groups", text="")
         row.prop(md, "invert_vertex", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -2170,11 +2255,18 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
 
         col = layout.column()
         col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -2221,11 +2313,20 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
         row.prop_search(md, "vertex_group", ob, "vertex_groups", text="")
         row.prop(md, "invert_vertex", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
@@ -2249,11 +2350,20 @@ class DATA_PT_gpencil_modifiers(ModifierButtonsPanel, Panel):
         row.prop_search(md, "vertex_group", ob, "vertex_groups", text="")
         row.prop(md, "invert_vertex", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Material:")
         row = col.row(align=True)
+        row.prop_search(md, "material", gpd, "materials", text="", icon='SHADING_TEXTURE')
+        row.prop(md, "invert_materials", text="", icon='ARROW_LEFTRIGHT')
+        row = layout.row(align=True)
         row.prop(md, "pass_index", text="Pass")
         row.prop(md, "invert_material_pass", text="", icon='ARROW_LEFTRIGHT')
 
+        col = layout.column()
+        col.separator()
+
         col.label(text="Layer:")
         row = col.row(align=True)
         row.prop_search(md, "layer", gpd, "layers", text="", icon='GREASEPENCIL')
diff --git a/release/scripts/startup/bl_ui/properties_grease_pencil_common.py b/release/scripts/startup/bl_ui/properties_grease_pencil_common.py
index 6f089f93f00..b41be13555c 100644
--- a/release/scripts/startup/bl_ui/properties_grease_pencil_common.py
+++ b/release/scripts/startup/bl_ui/properties_grease_pencil_common.py
@@ -169,8 +169,8 @@ class GreasePencilStrokeEditPanel:
         layout.label(text="Edit:")
         row = layout.row(align=True)
         row.operator("gpencil.copy", text="Copy")
-        row.operator("gpencil.paste", text="Paste").type = 'COPY'
-        row.operator("gpencil.paste", text="Paste & Merge").type = 'MERGE'
+        row.operator("gpencil.paste", text="Paste").type = 'ACTIVE'
+        row.operator("gpencil.paste", text="Paste by Layer").type = 'LAYER'
 
         col = layout.column(align=True)
         col.operator("gpencil.delete")
@@ -507,8 +507,9 @@ class GPENCIL_MT_pie_tools_more(Menu):
         # gpd = context.gpencil_data
 
         col = pie.column(align=True)
-        col.operator("gpencil.copy", icon='COPYDOWN', text="Copy")
-        col.operator("gpencil.paste", icon='PASTEDOWN', text="Paste")
+        col.operator("gpencil.copy", text="Copy", icon='COPYDOWN')
+        col.operator("gpencil.paste", text="Paste", icon='PASTEDOWN').type = 'ACTIVE'
+        col.operator("gpencil.paste", text="Paste by Layer").type = 'LAYER'
 
         col = pie.column(align=True)
         col.operator("gpencil.select_more", icon='ADD')
@@ -590,31 +591,65 @@ class GPENCIL_MT_snap(Menu):
         layout.operator("view3d.snap_cursor_to_grid", text="Cursor to Grid")
 
 
+class GPENCIL_MT_move_to_layer(Menu):
+    bl_label = "Move to Layer"
+
+    def draw(self, context):
+        layout = self.layout
+        gpd = context.gpencil_data
+        if gpd:
+            gpl_active = context.active_gpencil_layer
+            tot_layers = len(gpd.layers)
+            i = tot_layers - 1
+            while(i >= 0):
+                gpl = gpd.layers[i]
+                if gpl.info == gpl_active.info:
+                    icon='GREASEPENCIL'
+                else:
+                    icon = 'NONE'
+                layout.operator("gpencil.move_to_layer", text=gpl.info, icon=icon).layer=i
+                i -= 1
+
+            layout.separator()
+
+        layout.operator("gpencil.layer_add", text="New Layer", icon='ADD')
+
+
 class GPENCIL_MT_gpencil_draw_delete(Menu):
-    bl_label = "GPencil Draw Delete"
+    bl_label = "Delete"
 
     def draw(self, _context):
         layout = self.layout
 
         layout.operator_context = 'INVOKE_REGION_WIN'
 
-        layout.operator("gpencil.active_frames_delete_all", text="Delete Frame")
+        layout.operator("gpencil.delete", text="Delete Active Keyframe (Active Layer)").type = 'FRAME'
+        layout.operator("gpencil.active_frames_delete_all", text="Delete Active Keyframes (All Layers)")
 
 
 class GPENCIL_MT_cleanup(Menu):
     bl_label = "Clean Up"
 
     def draw(self, _context):
+
+        ob = _context.active_object
+
         layout = self.layout
+
         layout.operator("gpencil.frame_clean_loose", text="Delete Loose Points")
-        layout.operator("gpencil.stroke_merge_by_distance", text="Merge by Distance")
+
+        if ob.mode != 'PAINT_GPENCIL':
+            layout.operator("gpencil.stroke_merge_by_distance", text="Merge by Distance")
+        
         layout.separator()
 
         layout.operator("gpencil.frame_clean_fill", text="Boundary Strokes").mode = 'ACTIVE'
         layout.operator("gpencil.frame_clean_fill", text="Boundary Strokes all Frames").mode = 'ALL'
-        layout.separator()
 
-        layout.operator("gpencil.reproject")
+        if ob.mode != 'PAINT_GPENCIL':
+            layout.separator()
+
+            layout.operator("gpencil.reproject")
 
 
 class GPENCIL_UL_annotation_layer(UIList):
@@ -857,7 +892,8 @@ class GreasePencilMaterialsPanel:
             if is_view3d and brush is not None:
                 gp_settings = brush.gpencil_settings
                 if gp_settings.use_material_pin is False:
-                    ma = ob.material_slots[ob.active_material_index].material
+                    if ob.active_material_index > 0:
+                        ma = ob.material_slots[ob.active_material_index].material
                 else:
                     ma = gp_settings.material
 
@@ -923,6 +959,7 @@ classes = (
 
     GPENCIL_MT_snap,
     GPENCIL_MT_cleanup,
+    GPENCIL_MT_move_to_layer,
 
     GPENCIL_MT_gpencil_draw_delete,
 
diff --git a/release/scripts/startup/bl_ui/properties_material_gpencil.py b/release/scripts/startup/bl_ui/properties_material_gpencil.py
index 685e4e2f7c8..10b10005664 100644
--- a/release/scripts/startup/bl_ui/properties_material_gpencil.py
+++ b/release/scripts/startup/bl_ui/properties_material_gpencil.py
@@ -44,6 +44,9 @@ class GPENCIL_MT_color_context_menu(Menu):
         layout.operator("gpencil.stroke_lock_color", text="Lock Unselected")
         layout.operator("gpencil.lock_layer", text="Lock Unused")
 
+        layout.separator()
+
+        layout.operator("object.material_slot_remove_unused")
 
 class GPENCIL_UL_matslots(UIList):
     def draw_item(self, _context, layout, _data, item, icon, _active_data, _active_propname, _index):
@@ -152,17 +155,13 @@ class MATERIAL_PT_gpencil_strokecolor(GPMaterialButtonsPanel, Panel):
                     if gpcolor.use_stroke_texture_mix is True:
                         col.prop(gpcolor, "mix_stroke_factor", text="Factor")
 
-            if (
-                    gpcolor.stroke_style == 'SOLID' or
-                    gpcolor.use_stroke_pattern or
-                    gpcolor.use_stroke_texture_mix
-            ):
+            if (gpcolor.stroke_style == 'SOLID' or gpcolor.use_stroke_pattern or gpcolor.use_stroke_texture_mix):
                 col.prop(gpcolor, "color", text="Color")
 
             if gpcolor.mode in {'DOTS', 'BOX'}:
                 col.prop(gpcolor, "alignment_mode")
 
-            if gpcolor.mode == 'LINE':
+            if gpcolor.mode == 'LINE' and gpcolor.stroke_style != 'TEXTURE':
                 col.prop(gpcolor, "use_overlap_strokes")
 
 class MATERIAL_PT_gpencil_fillcolor(GPMaterialButtonsPanel, Panel):
@@ -275,8 +274,7 @@ class MATERIAL_PT_gpencil_material_presets(PresetPanel, Panel):
     preset_add_operator = "scene.gpencil_material_preset_add"
 
 
-classes = (
-    GPENCIL_UL_matslots,
+classes = (GPENCIL_UL_matslots,
     GPENCIL_MT_color_context_menu,
     MATERIAL_PT_gpencil_slots,
     MATERIAL_PT_gpencil_preview,
@@ -285,8 +283,7 @@ classes = (
     MATERIAL_PT_gpencil_strokecolor,
     MATERIAL_PT_gpencil_fillcolor,
     MATERIAL_PT_gpencil_options,
-    MATERIAL_PT_gpencil_custom_props,
-)
+    MATERIAL_PT_gpencil_custom_props,)
 
 if __name__ == "__main__":  # only for live edit.
     from bpy.utils import register_class
diff --git a/release/scripts/startup/bl_ui/properties_object.py b/release/scripts/startup/bl_ui/properties_object.py
index 1d98f0a7609..63d0804b232 100644
--- a/release/scripts/startup/bl_ui/properties_object.py
+++ b/release/scripts/startup/bl_ui/properties_object.py
@@ -109,20 +109,18 @@ class OBJECT_PT_delta_transform(ObjectButtonsPanel, Panel):
 
         ob = context.object
 
-        col = flow.column()
-        col.prop(ob, "delta_location")
+        col = layout.column()
+        col.prop(ob, "delta_location", text="Location")
 
-        col = flow.column()
         rotation_mode = ob.rotation_mode
         if rotation_mode == 'QUATERNION':
             col.prop(ob, "delta_rotation_quaternion", text="Rotation")
         elif rotation_mode == 'AXIS_ANGLE':
-            col.label(text="Not for Axis-Angle")
+            pass
         else:
-            col.prop(ob, "delta_rotation_euler", text="Delta Rotation")
+            col.prop(ob, "delta_rotation_euler", text="Rotation")
 
-        col = flow.column()
-        col.prop(ob, "delta_scale")
+        col.prop(ob, "delta_scale", text="Scale")
 
 
 class OBJECT_PT_relations(ObjectButtonsPanel, Panel):
diff --git a/release/scripts/startup/bl_ui/properties_output.py b/release/scripts/startup/bl_ui/properties_output.py
index 93a41c892f1..3d62191ea1a 100644
--- a/release/scripts/startup/bl_ui/properties_output.py
+++ b/release/scripts/startup/bl_ui/properties_output.py
@@ -479,9 +479,7 @@ class RENDER_PT_stereoscopy(RenderOutputButtonsPanel, Panel):
         basic_stereo = rd.views_format == 'STEREO_3D'
 
         row = layout.row()
-        row.use_property_split = True
-        row.use_property_decorate = False
-        row.prop(rd, "views_format")
+        layout.row().prop(rd, "views_format", expand=True)
 
         if basic_stereo:
             row = layout.row()
diff --git a/release/scripts/startup/bl_ui/properties_paint_common.py b/release/scripts/startup/bl_ui/properties_paint_common.py
index 9e1a642d891..23b2556eddb 100644
--- a/release/scripts/startup/bl_ui/properties_paint_common.py
+++ b/release/scripts/startup/bl_ui/properties_paint_common.py
@@ -113,10 +113,8 @@ def brush_texpaint_common(panel, context, layout, brush, _settings, projpaint=Fa
 
     col = layout.column()
 
-    if capabilities.has_color:
-        if brush.blend in {'ERASE_ALPHA', 'ADD_ALPHA'}:
-            if brush.image_tool == 'FILL' and not projpaint:
-                col.prop(brush, "fill_threshold")
+    if brush.image_tool == 'FILL' and not projpaint:
+        col.prop(brush, "fill_threshold", text="Gradient Type", slider=True)
 
     elif brush.image_tool == 'SOFTEN':
         col.row().prop(brush, "direction", expand=True)
@@ -183,7 +181,7 @@ def brush_texpaint_common_gradient(_panel, context, layout, brush, _settings, pr
 
     if brush.image_tool == 'DRAW':
         UnifiedPaintPanel.prop_unified_color(col, context, brush, "secondary_color", text="Background Color")
-        col.prop(brush, "gradient_stroke_mode", text="Mode")
+        col.prop(brush, "gradient_stroke_mode", text="Gradient Mapping")
         if brush.gradient_stroke_mode in {'SPACING_REPEAT', 'SPACING_CLAMP'}:
             col.prop(brush, "grad_spacing")
     else:  # if brush.image_tool == 'FILL':
diff --git a/release/scripts/startup/bl_ui/properties_particle.py b/release/scripts/startup/bl_ui/properties_particle.py
index 10e4d528913..5f15f9b6e83 100644
--- a/release/scripts/startup/bl_ui/properties_particle.py
+++ b/release/scripts/startup/bl_ui/properties_particle.py
@@ -1196,7 +1196,7 @@ class PARTICLE_PT_boidbrain(ParticleButtonsPanel, Panel):
                 row = layout.row()
                 row.prop(rule, "use_line")
                 sub = row.row()
-                sub.active = rule.line
+                sub.active = rule.use_line
                 sub.prop(rule, "queue_count")
             elif rule.type == 'AVERAGE_SPEED':
                 row.prop(rule, "speed", slider=True)
diff --git a/release/scripts/startup/bl_ui/space_clip.py b/release/scripts/startup/bl_ui/space_clip.py
index 37c76bf327c..4edda9b2460 100644
--- a/release/scripts/startup/bl_ui/space_clip.py
+++ b/release/scripts/startup/bl_ui/space_clip.py
@@ -1218,6 +1218,25 @@ class CLIP_PT_tools_grease_pencil_draw(AnnotationDrawingToolsPanel, Panel):
     bl_region_type = 'TOOLS'
 
 
+class CLIP_MT_view_zoom(Menu):
+    bl_label = "Fractional Zoom"
+
+    def draw(self, context):
+        layout = self.layout
+
+        ratios = ((1, 8), (1, 4), (1, 2), (1, 1), (2, 1), (4, 1), (8, 1))
+
+        for i, (a, b) in enumerate(ratios):
+            if i in {3, 4}:  # Draw separators around Zoom 1:1.
+                layout.separator()
+
+            layout.operator(
+                "clip.view_zoom_ratio",
+                text=iface_(f"Zoom {a:d}:{b:d}"),
+                translate=False,
+            ).ratio = a / b
+
+
 class CLIP_MT_view(Menu):
     bl_label = "View"
 
@@ -1238,20 +1257,17 @@ class CLIP_MT_view(Menu):
             layout.operator("clip.view_all", text="View Fit").fit_view = True
 
             layout.separator()
+            
             layout.operator("clip.view_zoom_in")
             layout.operator("clip.view_zoom_out")
 
             layout.separator()
+            
             layout.prop(sc, "show_metadata")
+            
             layout.separator()
 
-            ratios = ((1, 8), (1, 4), (1, 2), (1, 1), (2, 1), (4, 1), (8, 1))
-
-            text = iface_("Zoom %d:%d")
-            for a, b in ratios:
-                layout.operator("clip.view_zoom_ratio",
-                                text=text % (a, b),
-                                translate=False).ratio = a / b
+            layout.menu("CLIP_MT_view_zoom")
         else:
             if sc.view == 'GRAPH':
                 layout.operator_context = 'INVOKE_REGION_PREVIEW'
@@ -1762,6 +1778,7 @@ classes = (
     CLIP_PT_tools_scenesetup,
     CLIP_PT_annotation,
     CLIP_PT_tools_grease_pencil_draw,
+    CLIP_MT_view_zoom,
     CLIP_MT_view,
     CLIP_MT_clip,
     CLIP_MT_proxy,
diff --git a/release/scripts/startup/bl_ui/space_dopesheet.py b/release/scripts/startup/bl_ui/space_dopesheet.py
index 3b1b33b26e9..cb0dcb86ce4 100644
--- a/release/scripts/startup/bl_ui/space_dopesheet.py
+++ b/release/scripts/startup/bl_ui/space_dopesheet.py
@@ -318,7 +318,6 @@ class DOPESHEET_MT_view(Menu):
         layout.separator()
 
         layout.prop(st, "use_realtime_update")
-        layout.prop(st, "show_frame_indicator")
         layout.prop(st, "show_sliders")
         layout.prop(st, "show_group_colors")
         layout.prop(st, "show_interpolation")
diff --git a/release/scripts/startup/bl_ui/space_filebrowser.py b/release/scripts/startup/bl_ui/space_filebrowser.py
index b939586ff81..7a803906369 100644
--- a/release/scripts/startup/bl_ui/space_filebrowser.py
+++ b/release/scripts/startup/bl_ui/space_filebrowser.py
@@ -34,58 +34,131 @@ class FILEBROWSER_HT_header(Header):
 
         layout.menu("FILEBROWSER_MT_view")
 
-        row = layout.row(align=True)
-        row.operator("file.previous", text="", icon='BACK')
-        row.operator("file.next", text="", icon='FORWARD')
-        row.operator("file.parent", text="", icon='FILE_PARENT')
-        row.operator("file.refresh", text="", icon='FILE_REFRESH')
+        # can be None when save/reload with a file selector open
 
-        layout.operator_context = 'EXEC_DEFAULT'
-        layout.operator("file.directory_new", icon='NEWFOLDER', text="")
+        layout.separator_spacer()
 
-        layout.operator_context = 'INVOKE_DEFAULT'
+        layout.template_running_jobs()
+
+
+class FILEBROWSER_PT_display(Panel):
+    bl_space_type = 'FILE_BROWSER'
+    bl_region_type = 'HEADER'
+    bl_label = "Display"
 
+    @classmethod
+    def poll(cls, context):
         # can be None when save/reload with a file selector open
-        if params:
-            is_lib_browser = params.use_library_browsing
+        return context.space_data.params is not None
 
-            layout.prop(params, "display_type", expand=True, text="")
-            layout.prop(params, "sort_method", expand=True, text="")
-            layout.prop(params, "show_hidden", text="", icon='FILE_HIDDEN')
+    def draw(self, context):
+        layout = self.layout
 
-        layout.separator_spacer()
+        space = context.space_data
+        params = space.params
+        is_lib_browser = params.use_library_browsing
 
-        layout.template_running_jobs()
+        layout.label(text="Display as")
+        layout.column().prop(params, "display_type", expand=True)
 
-        if params:
-            layout.prop(params, "use_filter", text="", icon='FILTER')
+        layout.use_property_split = True
+        layout.use_property_decorate = False  # No animation.
 
-            row = layout.row(align=True)
-            row.active = params.use_filter
-            row.prop(params, "use_filter_folder", text="")
+        if params.display_type == 'THUMBNAIL':
+            layout.prop(params, "display_size", text="Size")
+        else:
+            layout.prop(params, "show_details_size", text="Size")
+            layout.prop(params, "show_details_datetime", text="Date")
 
-            if params.filter_glob:
-                # if st.active_operator and hasattr(st.active_operator, "filter_glob"):
-                #     row.prop(params, "filter_glob", text="")
-                row.label(text=params.filter_glob)
-            else:
-                row.prop(params, "use_filter_blender", text="")
-                row.prop(params, "use_filter_backup", text="")
-                row.prop(params, "use_filter_image", text="")
-                row.prop(params, "use_filter_movie", text="")
-                row.prop(params, "use_filter_script", text="")
-                row.prop(params, "use_filter_font", text="")
-                row.prop(params, "use_filter_sound", text="")
-                row.prop(params, "use_filter_text", text="")
+        layout.prop(params, "recursion_level", text="Recursions")
 
-            if is_lib_browser:
-                row.prop(params, "use_filter_blendid", text="")
-                if params.use_filter_blendid:
-                    row.separator()
-                    row.prop(params, "filter_id_category", text="")
+        layout.use_property_split = False
+        layout.separator()
+
+        layout.label(text="Sort by")
+        layout.column().prop(params, "sort_method", expand=True)
+        layout.prop(params, "use_sort_invert")
 
-            row.separator()
-            row.prop(params, "filter_search", text="", icon='VIEWZOOM')
+
+class FILEBROWSER_PT_filter(Panel):
+    bl_space_type = 'FILE_BROWSER'
+    bl_region_type = 'HEADER'
+    bl_label = "Filter"
+
+    @classmethod
+    def poll(cls, context):
+        # can be None when save/reload with a file selector open
+        return context.space_data.params is not None
+
+    def draw(self, context):
+        layout = self.layout
+
+        space = context.space_data
+        params = space.params
+        is_lib_browser = params.use_library_browsing
+
+        row = layout.row(align=True)
+        row.prop(params, "use_filter", text="", toggle=0)
+        row.label(text="Filter")
+
+        col = layout.column()
+        col.active = params.use_filter
+
+        row = col.row()
+        row.label(icon='FILE_FOLDER')
+        row.prop(params, "use_filter_folder", text="Folders", toggle=0)
+
+        if params.filter_glob:
+            col.label(text=params.filter_glob)
+        else:
+            row = col.row()
+            row.label(icon='FILE_BLEND')
+            row.prop(params, "use_filter_blender",
+                     text=".blend Files", toggle=0)
+            row = col.row()
+            row.label(icon='FILE_BACKUP')
+            row.prop(params, "use_filter_backup",
+                     text="Backup .blend Files", toggle=0)
+            row = col.row()
+            row.label(icon='FILE_IMAGE')
+            row.prop(params, "use_filter_image", text="Image Files", toggle=0)
+            row = col.row()
+            row.label(icon='FILE_MOVIE')
+            row.prop(params, "use_filter_movie", text="Movie Files", toggle=0)
+            row = col.row()
+            row.label(icon='FILE_SCRIPT')
+            row.prop(params, "use_filter_script",
+                     text="Script Files", toggle=0)
+            row = col.row()
+            row.label(icon='FILE_FONT')
+            row.prop(params, "use_filter_font", text="Font Files", toggle=0)
+            row = col.row()
+            row.label(icon='FILE_SOUND')
+            row.prop(params, "use_filter_sound", text="Sound Files", toggle=0)
+            row = col.row()
+            row.label(icon='FILE_TEXT')
+            row.prop(params, "use_filter_text", text="Text Files", toggle=0)
+
+        col.separator()
+
+        if is_lib_browser:
+            row = col.row()
+            row.label(icon='BLANK1')  # Indentation
+            row.prop(params, "use_filter_blendid",
+                     text="Blender IDs", toggle=0)
+            if params.use_filter_blendid:
+                row = col.row()
+                row.label(icon='BLANK1')  # Indentation
+                row.prop(params, "filter_id_category", text="")
+
+                col.separator()
+
+        layout.prop(params, "show_hidden")
+
+
+def panel_poll_is_upper_region(region):
+    # The upper region is left-aligned, the lower is split into it then.
+    return region.alignment == 'LEFT'
 
 
 class FILEBROWSER_UL_dir(UIList):
@@ -119,10 +192,13 @@ class FILEBROWSER_UL_dir(UIList):
 class FILEBROWSER_PT_bookmarks_volumes(Panel):
     bl_space_type = 'FILE_BROWSER'
     bl_region_type = 'TOOLS'
-    bl_options = {'DEFAULT_CLOSED'}
     bl_category = "Bookmarks"
     bl_label = "Volumes"
 
+    @classmethod
+    def poll(cls, context):
+        return panel_poll_is_upper_region(context.region)
+
     def draw(self, context):
         layout = self.layout
         space = context.space_data
@@ -141,7 +217,7 @@ class FILEBROWSER_PT_bookmarks_system(Panel):
 
     @classmethod
     def poll(cls, context):
-        return not context.preferences.filepaths.hide_system_bookmarks
+        return not context.preferences.filepaths.hide_system_bookmarks and panel_poll_is_upper_region(context.region)
 
     def draw(self, context):
         layout = self.layout
@@ -161,8 +237,10 @@ class FILEBROWSER_MT_bookmarks_context_menu(Menu):
         layout.operator("file.bookmark_cleanup", icon='X', text="Cleanup")
 
         layout.separator()
-        layout.operator("file.bookmark_move", icon='TRIA_UP_BAR', text="Move To Top").direction = 'TOP'
-        layout.operator("file.bookmark_move", icon='TRIA_DOWN_BAR', text="Move To Bottom").direction = 'BOTTOM'
+        layout.operator("file.bookmark_move", icon='TRIA_UP_BAR',
+                        text="Move To Top").direction = 'TOP'
+        layout.operator("file.bookmark_move", icon='TRIA_DOWN_BAR',
+                        text="Move To Bottom").direction = 'BOTTOM'
 
 
 class FILEBROWSER_PT_bookmarks_favorites(Panel):
@@ -171,6 +249,10 @@ class FILEBROWSER_PT_bookmarks_favorites(Panel):
     bl_category = "Bookmarks"
     bl_label = "Favorites"
 
+    @classmethod
+    def poll(cls, context):
+        return panel_poll_is_upper_region(context.region)
+
     def draw(self, context):
         layout = self.layout
         space = context.space_data
@@ -185,12 +267,15 @@ class FILEBROWSER_PT_bookmarks_favorites(Panel):
             col = row.column(align=True)
             col.operator("file.bookmark_add", icon='ADD', text="")
             col.operator("file.bookmark_delete", icon='REMOVE', text="")
-            col.menu("FILEBROWSER_MT_bookmarks_context_menu", icon='DOWNARROW_HLT', text="")
+            col.menu("FILEBROWSER_MT_bookmarks_context_menu",
+                     icon='DOWNARROW_HLT', text="")
 
             if num_rows > 1:
                 col.separator()
-                col.operator("file.bookmark_move", icon='TRIA_UP', text="").direction = 'UP'
-                col.operator("file.bookmark_move", icon='TRIA_DOWN', text="").direction = 'DOWN'
+                col.operator("file.bookmark_move", icon='TRIA_UP',
+                             text="").direction = 'UP'
+                col.operator("file.bookmark_move", icon='TRIA_DOWN',
+                             text="").direction = 'DOWN'
         else:
             layout.operator("file.bookmark_add", icon='ADD')
 
@@ -203,7 +288,7 @@ class FILEBROWSER_PT_bookmarks_recents(Panel):
 
     @classmethod
     def poll(cls, context):
-        return not context.preferences.filepaths.hide_recent_locations
+        return not context.preferences.filepaths.hide_recent_locations and panel_poll_is_upper_region(context.region)
 
     def draw(self, context):
         layout = self.layout
@@ -227,7 +312,7 @@ class FILEBROWSER_PT_advanced_filter(Panel):
     @classmethod
     def poll(cls, context):
         # only useful in append/link (library) context currently...
-        return context.space_data.params.use_library_browsing
+        return context.space_data.params.use_library_browsing and panel_poll_is_upper_region(context.region)
 
     def draw(self, context):
         layout = self.layout
@@ -242,6 +327,133 @@ class FILEBROWSER_PT_advanced_filter(Panel):
                 col.prop(params, "filter_id")
 
 
+class FILEBROWSER_PT_options_toggle(Panel):
+    bl_space_type = 'FILE_BROWSER'
+    bl_region_type = 'TOOLS'
+    bl_label = "Options Toggle"
+    bl_options = {'HIDE_HEADER'}
+
+    @classmethod
+    def poll(cls, context):
+        sfile = context.space_data
+        return context.region.alignment == 'BOTTOM' and sfile.active_operator
+
+    def is_option_region_visible(self, context):
+        for region in context.area.regions:
+            if region.type == 'TOOL_PROPS' and region.width <= 1:
+                return False
+
+        return True
+
+    def draw(self, context):
+        layout = self.layout
+        label = "Hide Options" if self.is_option_region_visible(
+            context) else "Options"
+
+        layout.scale_x = 1.3
+        layout.scale_y = 1.3
+
+        layout.operator("screen.region_toggle",
+                        text=label).region_type = 'TOOL_PROPS'
+
+
+class FILEBROWSER_PT_directory_path(Panel):
+    bl_space_type = 'FILE_BROWSER'
+    bl_region_type = 'UI'
+    bl_label = "Directory Path"
+    bl_category = "Attributes"
+    bl_options = {'HIDE_HEADER'}
+
+    def is_header_visible(self, context):
+        for region in context.area.regions:
+            if region.type == 'HEADER' and region.height <= 1:
+                return False
+
+        return True
+
+    def draw(self, context):
+        layout = self.layout
+        space = context.space_data
+        params = space.params
+
+        layout.scale_x = 1.3
+        layout.scale_y = 1.3
+
+        row = layout.row()
+
+        subrow = row.row(align=True)
+        subrow.operator("file.previous", text="", icon='BACK')
+        subrow.operator("file.next", text="", icon='FORWARD')
+        subrow.operator("file.parent", text="", icon='FILE_PARENT')
+        subrow.operator("file.refresh", text="", icon='FILE_REFRESH')
+
+        row.operator("file.directory_new", icon='NEWFOLDER', text="")
+
+        subrow = row.row()
+        subrow.prop(params, "directory", text="")
+
+        subrow = row.row()
+        subrow.scale_x = 0.5
+        subrow.prop(params, "filter_search", text="", icon='VIEWZOOM')
+
+        # Uses prop_with_popover() as popover() only adds the triangle icon in headers.
+        row.prop_with_popover(
+            params,
+            "display_type",
+            panel="FILEBROWSER_PT_display",
+            text="",
+            icon_only=True,
+        )
+        row.prop_with_popover(
+            params,
+            "display_type",
+            panel="FILEBROWSER_PT_filter",
+            text="",
+            icon='FILTER',
+            icon_only=True,
+        )
+
+
+class FILEBROWSER_PT_file_operation(Panel):
+    bl_space_type = 'FILE_BROWSER'
+    bl_region_type = 'EXECUTE'
+    bl_label = "Execute File Operation"
+    bl_options = {'HIDE_HEADER'}
+
+    @classmethod
+    def poll(cls, context):
+        return context.space_data.active_operator
+
+    def draw(self, context):
+        import sys
+
+        layout = self.layout
+        space = context.space_data
+        params = space.params
+
+        layout.scale_x = 1.3
+        layout.scale_y = 1.3
+
+        row = layout.row()
+        sub = row.row()
+        sub.prop(params, "filename", text="")
+        sub = row.row()
+        sub.ui_units_x = 5
+
+        # subsub = sub.row(align=True)
+        # subsub.operator("file.filenum", text="", icon='ADD').increment = 1
+        # subsub.operator("file.filenum", text="", icon='REMOVE').increment = -1
+
+        # organize buttons according to the OS standard
+        if sys.platform[:3] != "win":
+            sub.operator("FILE_OT_cancel", text="Cancel")
+        subsub = sub.row()
+        subsub.active_default = True
+        subsub.operator("FILE_OT_execute", text=params.title)
+        if sys.platform[:3] == "win":
+            sub.operator("FILE_OT_cancel", text="Cancel")
+
+
 class FILEBROWSER_MT_view(Menu):
     bl_label = "View"
 
@@ -250,7 +462,7 @@ class FILEBROWSER_MT_view(Menu):
         st = context.space_data
         params = st.params
 
-        layout.prop(st, "show_region_toolbar")
+        layout.prop(st, "show_region_toolbar", text="Source List")
         layout.prop(st, "show_region_ui", text="File Path")
 
         layout.separator()
@@ -263,8 +475,46 @@ class FILEBROWSER_MT_view(Menu):
         layout.menu("INFO_MT_area")
 
 
+class FILEBROWSER_MT_context_menu(Menu):
+    bl_label = "Files Context Menu"
+
+    def draw(self, context):
+        layout = self.layout
+        st = context.space_data
+        params = st.params
+
+        layout.operator("file.previous", text="Back")
+        layout.operator("file.next", text="Forward")
+        layout.operator("file.parent", text="Go to Parent")
+        layout.operator("file.refresh", text="Refresh")
+
+        layout.separator()
+
+        layout.operator("file.filenum", text="Increase Number",
+                        icon='ADD').increment = 1
+        layout.operator("file.filenum", text="Decrease Number",
+                        icon='REMOVE').increment = -1
+
+        layout.separator()
+
+        layout.operator("file.rename", text="Rename")
+        # layout.operator("file.delete")
+        layout.operator("file.directory_new", text="New Folder")
+        layout.operator("file.bookmark_add", text="Add Bookmark")
+
+        layout.separator()
+
+        layout.prop_menu_enum(params, "display_type")
+        if params.display_type == 'THUMBNAIL':
+            layout.prop_menu_enum(params, "display_size")
+        layout.prop_menu_enum(params, "recursion_level", text="Recursions")
+        layout.prop_menu_enum(params, "sort_method")
+
+
 classes = (
     FILEBROWSER_HT_header,
+    FILEBROWSER_PT_display,
+    FILEBROWSER_PT_filter,
     FILEBROWSER_UL_dir,
     FILEBROWSER_PT_bookmarks_volumes,
     FILEBROWSER_PT_bookmarks_system,
@@ -272,7 +522,11 @@ classes = (
     FILEBROWSER_PT_bookmarks_favorites,
     FILEBROWSER_PT_bookmarks_recents,
     FILEBROWSER_PT_advanced_filter,
+    FILEBROWSER_PT_directory_path,
+    FILEBROWSER_PT_file_operation,
+    FILEBROWSER_PT_options_toggle,
     FILEBROWSER_MT_view,
+    FILEBROWSER_MT_context_menu,
 )
 
 if __name__ == "__main__":  # only for live edit.
diff --git a/release/scripts/startup/bl_ui/space_graph.py b/release/scripts/startup/bl_ui/space_graph.py
index b6195d1626a..1a3ef82efa1 100644
--- a/release/scripts/startup/bl_ui/space_graph.py
+++ b/release/scripts/startup/bl_ui/space_graph.py
@@ -114,7 +114,6 @@ class GRAPH_MT_view(Menu):
         layout.separator()
 
         layout.prop(st, "use_realtime_update")
-        layout.prop(st, "show_frame_indicator")
         layout.prop(st, "show_cursor")
         layout.prop(st, "show_sliders")
         layout.prop(st, "show_group_colors")
diff --git a/release/scripts/startup/bl_ui/space_image.py b/release/scripts/startup/bl_ui/space_image.py
index 06505c54c9c..5e05c49b567 100644
--- a/release/scripts/startup/bl_ui/space_image.py
+++ b/release/scripts/startup/bl_ui/space_image.py
@@ -142,7 +142,7 @@ class IMAGE_MT_view_zoom(Menu):
 
             layout.operator(
                 "image.view_zoom_ratio",
-                text=iface_(f"Zoom {a:d}:{b:d}"),
+                text=iface_("Zoom %d:%d") % (a, b),
                 translate=False,
             ).ratio = a / b
 
@@ -961,7 +961,6 @@ class IMAGE_PT_view_display_uv_edit_overlays(Panel):
         col = layout.column()
 
         col.prop(uvedit, "edge_display_type", text="Display As")
-        col.prop(uvedit, "show_edges", text="Edges")
         col.prop(uvedit, "show_faces", text="Faces")
 
         col = layout.column()
diff --git a/release/scripts/startup/bl_ui/space_nla.py b/release/scripts/startup/bl_ui/space_nla.py
index aaa971cfa55..825e4b41609 100644
--- a/release/scripts/startup/bl_ui/space_nla.py
+++ b/release/scripts/startup/bl_ui/space_nla.py
@@ -91,7 +91,6 @@ class NLA_MT_view(Menu):
         layout.separator()
 
         layout.prop(st, "use_realtime_update")
-        layout.prop(st, "show_frame_indicator")
 
         layout.prop(st, "show_seconds")
         layout.prop(st, "show_locked_time")
diff --git a/release/scripts/startup/bl_ui/space_outliner.py b/release/scripts/startup/bl_ui/space_outliner.py
index 6832e3d463b..7bf203d8e39 100644
--- a/release/scripts/startup/bl_ui/space_outliner.py
+++ b/release/scripts/startup/bl_ui/space_outliner.py
@@ -46,6 +46,10 @@ class OUTLINER_HT_header(Header):
 
         layout.separator_spacer()
 
+        if display_mode == 'SEQUENCE':
+            row = layout.row(align=True)
+            row.prop(space, "use_sync_select", icon="UV_SYNC_SELECT", text="")
+
         row = layout.row(align=True)
         if display_mode in {'SCENES', 'VIEW_LAYER'}:
             row.popover(
@@ -328,6 +332,10 @@ class OUTLINER_PT_filter(Panel):
             col.prop(space, "use_sort_alpha")
             layout.separator()
 
+        row = layout.row(align=True)
+        row.prop(space, "use_sync_select", text="Sync Selection")
+        layout.separator()
+
         col = layout.column(align=True)
         col.label(text="Search:")
         col.prop(space, "use_filter_complete", text="Exact Match")
diff --git a/release/scripts/startup/bl_ui/space_sequencer.py b/release/scripts/startup/bl_ui/space_sequencer.py
index 9df37158c07..5a678854a3f 100644
--- a/release/scripts/startup/bl_ui/space_sequencer.py
+++ b/release/scripts/startup/bl_ui/space_sequencer.py
@@ -103,15 +103,14 @@ class SEQUENCER_HT_header(Header):
 
         SEQUENCER_MT_editor_menus.draw_collapsible(context, layout)
 
-        layout.separator_spacer()
-
         if st.view_type in {'PREVIEW', 'SEQUENCER_PREVIEW'}:
+
+            layout.separator_spacer()
+
             layout.prop(st, "display_mode", text="", icon_only=True)
 
-        if st.view_type != 'SEQUENCER':
             layout.prop(st, "preview_channels", text="", icon_only=True)
 
-        if st.view_type in {'PREVIEW', 'SEQUENCER_PREVIEW'}:
             gpd = context.gpencil_data
             tool_settings = context.tool_settings
 
@@ -200,7 +199,7 @@ class SEQUENCER_MT_preview_zoom(Menu):
 
             layout.operator(
                 "sequencer.view_zoom_ratio",
-                text=iface_(f"Zoom {a:d}:{b:d}"),
+                text=iface_("Zoom %d:%d") % (a, b),
                 translate=False,
             ).ratio = a / b
         layout.operator_context = 'INVOKE_DEFAULT'
@@ -265,7 +264,6 @@ class SEQUENCER_MT_view(Menu):
             layout.separator()
             layout.operator_context = 'INVOKE_DEFAULT'
 
-#           layout.prop(st, "show_frame_indicator") #Do not have any function and do not work.
             layout.prop(st, "show_strip_offset")
             layout.prop(st, "show_marker_lines")
 
@@ -869,6 +867,8 @@ class SEQUENCER_PT_strip(SequencerButtonsPanel, Panel):
             icon_header = 'FONT_DATA'
         elif strip_type == 'ADJUSTMENT':
             icon_header = 'COLOR'
+        elif strip_type == 'META':
+            icon_header = 'SEQ_STRIP_META'
         else:
             icon_header = 'SEQ_SEQUENCER'
 
@@ -1467,7 +1467,7 @@ class SEQUENCER_PT_time(SequencerButtonsPanel, Panel):
         split.alignment = 'RIGHT'
         split.label(text="End")
         split = split.split(factor=0.8 + max_factor, align=True)
-        split.label(text="{:>14}".format(smpte_from_frame(frame_final_end) + ":"))
+        split.label(text="{:>14}".format(smpte_from_frame(frame_final_end)))
         split.alignment = 'RIGHT'
         split.label(text=str(frame_final_end) + " ")
 
@@ -1511,7 +1511,7 @@ class SEQUENCER_PT_time(SequencerButtonsPanel, Panel):
         split.label(text="Playhead")
         split = split.split(factor=0.8 + max_factor, align=True)
         playhead = frame_current - frame_final_start
-        split.label(text="{:>14}".format(smpte_from_frame(playhead) + ":"))
+        split.label(text="{:>14}".format(smpte_from_frame(playhead)))
         split.alignment = 'RIGHT'
         split.label(text=str(playhead) + " ")
 
@@ -1880,7 +1880,7 @@ class SEQUENCER_PT_preview(SequencerButtonsPanel_Output, Panel):
         col = layout.column()
         col.prop(render, "sequencer_gl_preview", text="Preview Shading")
 
-        if render.sequencer_gl_preview in ['SOLID', 'WIREFRAME']:
+        if render.sequencer_gl_preview in {'SOLID', 'WIREFRAME'}:
             col.prop(render, "use_sequencer_override_scene_strip")
 
 
@@ -2128,6 +2128,7 @@ classes = (
 
     SEQUENCER_PT_strip,
 
+    SEQUENCER_PT_effect,
     SEQUENCER_PT_adjust,
     SEQUENCER_PT_adjust_comp,
     SEQUENCER_PT_adjust_transform,
@@ -2137,7 +2138,6 @@ classes = (
     SEQUENCER_PT_adjust_color,
     SEQUENCER_PT_adjust_sound,
 
-    SEQUENCER_PT_effect,
     SEQUENCER_PT_scene,
     SEQUENCER_PT_mask,
     SEQUENCER_PT_effect_text_style,
diff --git a/release/scripts/startup/bl_ui/space_time.py b/release/scripts/startup/bl_ui/space_time.py
index d2c41f13ee3..04a904edde3 100644
--- a/release/scripts/startup/bl_ui/space_time.py
+++ b/release/scripts/startup/bl_ui/space_time.py
@@ -48,26 +48,27 @@ class TIME_HT_editor_buttons(Header):
             #   hide the play-reversed button
             #   since JACK transport doesn't support reversed playback
             if scene.sync_mode == 'AUDIO_SYNC' and context.preferences.system.audio_device == 'JACK':
-                sub = row.row(align=True)
-                sub.scale_x = 1.4
-                sub.operator("screen.animation_play", text="", icon='PLAY')
+                row.scale_x = 2
+                row.operator("screen.animation_play", text="", icon='PLAY')
+                row.scale_x = 1
             else:
                 row.operator("screen.animation_play", text="", icon='PLAY_REVERSE').reverse = True
                 row.operator("screen.animation_play", text="", icon='PLAY')
         else:
-            sub = row.row(align=True)
-            sub.scale_x = 1.4
-            sub.operator("screen.animation_play", text="", icon='PAUSE')
+            row.scale_x = 2
+            row.operator("screen.animation_play", text="", icon='PAUSE')
+            row.scale_x = 1
         row.operator("screen.keyframe_jump", text="", icon='NEXT_KEYFRAME').next = True
         row.operator("screen.frame_jump", text="", icon='FF').end = True
 
         layout.separator_spacer()
 
         row = layout.row()
-        row.scale_x = 0.95
         if scene.show_subframe:
+            row.scale_x = 1.15
             row.prop(scene, "frame_float", text="")
         else:
+            row.scale_x = 0.95
             row.prop(scene, "frame_current", text="")
 
         row = layout.row(align=True)
@@ -135,7 +136,6 @@ class TIME_MT_view(Menu):
         layout.separator()
 
         layout.prop(st, "show_marker_lines")
-        layout.prop(st, "show_frame_indicator")
         layout.prop(scene, "show_keys_from_selected_only")
 
         layout.separator()
diff --git a/release/scripts/startup/bl_ui/space_toolsystem_toolbar.py b/release/scripts/startup/bl_ui/space_toolsystem_toolbar.py
index 17e753d531f..0af9b51597d 100644
--- a/release/scripts/startup/bl_ui/space_toolsystem_toolbar.py
+++ b/release/scripts/startup/bl_ui/space_toolsystem_toolbar.py
@@ -490,12 +490,17 @@ class _defs_edit_mesh:
 
     @ToolDef.from_fn
     def poly_build():
+        def draw_settings(context, layout, tool):
+            props = tool.operator_properties("mesh.polybuild_face_at_cursor_move")
+            props_macro = props.MESH_OT_polybuild_face_at_cursor
+            layout.prop(props_macro, "create_quads")
         return dict(
             idname="builtin.poly_build",
             label="Poly Build",
             icon="ops.mesh.polybuild_hover",
             widget="VIEW3D_GGT_mesh_preselect_elem",
             keymap=(),
+            draw_settings=draw_settings,
         )
 
     @ToolDef.from_fn
@@ -1508,8 +1513,9 @@ class _defs_gpencil_sculpt:
             return True
         ob = context.active_object
         ts = context.scene.tool_settings
-        return (ob and ob.type == 'GPENCIL' and
-                (ts.gpencil_sculpt.use_select_mask))
+        return ob and ob.type == 'GPENCIL' and (ts.use_gpencil_select_mask_point or
+                                                ts.use_gpencil_select_mask_stroke or
+                                                ts.use_gpencil_select_mask_segment)
 
     @staticmethod
     def generate_from_brushes(context):
diff --git a/release/scripts/startup/bl_ui/space_topbar.py b/release/scripts/startup/bl_ui/space_topbar.py
index f735e7b770f..f66a363e6ca 100644
--- a/release/scripts/startup/bl_ui/space_topbar.py
+++ b/release/scripts/startup/bl_ui/space_topbar.py
@@ -344,18 +344,12 @@ class TOPBAR_MT_app_about(Menu):
     def draw(self, _context):
         layout = self.layout
 
-        layout.operator(
-            "wm.url_open", text="Release Notes", icon='URL',
-        ).url = "https://www.blender.org/download/releases/%d-%d/" % bpy.app.version[:2]
+        layout.operator("wm.url_open_preset", text="Release Notes", icon='URL').type = 'RELEASE_NOTES'
 
         layout.separator()
 
-        layout.operator(
-            "wm.url_open", text="Blender Website", icon='URL',
-        ).url = "https://www.blender.org/"
-        layout.operator(
-            "wm.url_open", text="Credits", icon='URL',
-        ).url = "https://www.blender.org/about/credits/"
+        layout.operator("wm.url_open_preset", text="Blender Website", icon='URL').type = 'BLENDER'
+        layout.operator("wm.url_open_preset", text="Credits", icon='URL').type = 'CREDITS'
 
         layout.separator()
 
@@ -370,9 +364,7 @@ class TOPBAR_MT_app_support(Menu):
     def draw(self, _context):
         layout = self.layout
 
-        layout.operator(
-            "wm.url_open", text="Development Fund", icon='FUND',
-        ).url = "https://fund.blender.org"
+        layout.operator("wm.url_open_preset", text="Development Fund", icon='FUND').type = 'FUND'
 
         layout.separator()
 
@@ -513,6 +505,8 @@ class TOPBAR_MT_edit(Menu):
         props.name = "TOPBAR_PT_name"
         props.keep_open = False
 
+        layout.operator("wm.batch_rename")
+
         layout.separator()
 
         # Should move elsewhere (impacts outliner & 3D view).
@@ -565,23 +559,13 @@ class TOPBAR_MT_help(Menu):
     bl_label = "Help"
 
     def draw(self, context):
-        # If 'url_prefill_from_blender' becomes slow it could be made into a separate operator
-        # to avoid constructing the bug report just to show this menu.
-        from bl_ui_utils.bug_report_url import url_prefill_from_blender
-
         layout = self.layout
 
         show_developer = context.preferences.view.show_developer_ui
 
-        if bpy.app.version_cycle in {'rc', 'release'}:
-            manual_version = '%d.%d' % bpy.app.version[:2]
-        else:
-            manual_version = 'dev'
+        layout.operator("wm.url_open_preset", text="Manual", icon='HELP',).type = 'MANUAL'
 
         layout.operator(
-            "wm.url_open", text="Manual", icon='HELP',
-        ).url = "https://docs.blender.org/manual/en/" + manual_version + "/"
-        layout.operator(
             "wm.url_open", text="Tutorials", icon='URL',
         ).url = "https://www.blender.org/tutorials"
         layout.operator(
@@ -612,9 +596,7 @@ class TOPBAR_MT_help(Menu):
 
         layout.separator()
 
-        layout.operator(
-            "wm.url_open", text="Report a Bug", icon='URL',
-        ).url = url_prefill_from_blender()
+        layout.operator("wm.url_open_preset", text="Report a Bug", icon='URL').type = 'BUG'
 
         layout.separator()
 
diff --git a/release/scripts/startup/bl_ui/space_userpref.py b/release/scripts/startup/bl_ui/space_userpref.py
index 87baea9cdd8..8ed8eb04898 100644
--- a/release/scripts/startup/bl_ui/space_userpref.py
+++ b/release/scripts/startup/bl_ui/space_userpref.py
@@ -1708,7 +1708,6 @@ class USERPREF_PT_addons(Panel):
     def draw(self, context):
         import os
         import addon_utils
-        from bl_ui_utils.bug_report_url import url_prefill_from_blender
 
         layout = self.layout
 
@@ -1744,6 +1743,7 @@ class USERPREF_PT_addons(Panel):
         row.operator("preferences.addon_refresh", icon='FILE_REFRESH', text="Refresh")
 
         row = layout.row()
+        row.prop(context.preferences.view, "show_addons_enabled_only")
         row.prop(context.window_manager, "addon_filter", text="")
         row.prop(context.window_manager, "addon_search", text="", icon='VIEWZOOM')
 
@@ -1770,6 +1770,7 @@ class USERPREF_PT_addons(Panel):
                 "(see console for details)",
             )
 
+        show_enabled_only = context.preferences.view.show_addons_enabled_only
         filter = context.window_manager.addon_filter
         search = context.window_manager.addon_search.lower()
         support = context.window_manager.addon_support
@@ -1786,13 +1787,15 @@ class USERPREF_PT_addons(Panel):
                 continue
 
             # check if addon should be visible with current filters
-            if (
-                    (filter == "All") or
-                    (filter == info["category"]) or
-                    (filter == "Enabled" and is_enabled) or
-                    (filter == "Disabled" and not is_enabled) or
-                    (filter == "User" and (mod.__file__.startswith(addon_user_dirs)))
-            ):
+            is_visible = (
+                (filter == "All") or
+                (filter == info["category"]) or
+                (filter == "User" and (mod.__file__.startswith(addon_user_dirs)))
+            )
+            if show_enabled_only:
+                is_visible = is_visible and is_enabled
+
+            if is_visible:
                 if search and search not in info["name"].lower():
                     if info["author"]:
                         if search not in info["author"].lower():
@@ -1883,13 +1886,18 @@ class USERPREF_PT_addons(Panel):
                             ).url = info["wiki_url"]
                         # Only add "Report a Bug" button if tracker_url is set
                         # or the add-on is bundled (use official tracker then).
-                        if info.get("tracker_url") or not user_addon:
+                        if info.get("tracker_url"):
                             sub.operator(
                                 "wm.url_open", text="Report a Bug", icon='URL',
-                            ).url = info.get(
-                                "tracker_url",
-                                url_prefill_from_blender(info),
+                            ).url = info["tracker_url"]
+                        elif not user_addon:
+                            addon_info = ("Name: {} {}\nAuthor: {}\n").format(
+                                info["name"], info["version"], info["author"])
+                            props = sub.operator(
+                                "wm.url_open_preset", text="Report a Bug", icon='URL',
                             )
+                            props.type = 'BUG_ADDON'
+                            props.id = addon_info
                         if user_addon:
                             sub.operator(
                                 "preferences.addon_remove", text="Remove", icon='CANCEL',
diff --git a/release/scripts/startup/bl_ui/space_view3d.py b/release/scripts/startup/bl_ui/space_view3d.py
index cdedf3012ef..d83f4cfc129 100644
--- a/release/scripts/startup/bl_ui/space_view3d.py
+++ b/release/scripts/startup/bl_ui/space_view3d.py
@@ -599,17 +599,20 @@ class VIEW3D_HT_header(Header):
                 sub.separator(factor=0.4)
                 sub.prop(tool_settings, "use_gpencil_draw_additive", text="", icon='FREEZE')
 
+            # Select mode for Editing
             if gpd.use_stroke_edit_mode:
                 row = layout.row(align=True)
-                row.prop(tool_settings, "gpencil_selectmode", text="", expand=True)
+                row.prop(tool_settings, "gpencil_selectmode_edit", text="", expand=True)
 
-            if gpd.use_stroke_edit_mode or gpd.is_stroke_sculpt_mode or gpd.is_stroke_weight_mode:
+            # Select mode for Sculpt
+            if gpd.is_stroke_sculpt_mode :
                 row = layout.row(align=True)
+                row.prop(tool_settings, "use_gpencil_select_mask_point", text="")
+                row.prop(tool_settings, "use_gpencil_select_mask_stroke", text="")
+                row.prop(tool_settings, "use_gpencil_select_mask_segment", text="")
 
-                if gpd.is_stroke_sculpt_mode:
-                    row.prop(tool_settings.gpencil_sculpt, "use_select_mask", text="")
-                    row.separator()
-
+            if gpd.use_stroke_edit_mode or gpd.is_stroke_sculpt_mode or gpd.is_stroke_weight_mode:
+                row = layout.row(align=True)
                 row.prop(gpd, "use_multiedit", text="", icon='GP_MULTIFRAME_EDITING')
 
                 sub = row.row(align=True)
@@ -621,8 +624,6 @@ class VIEW3D_HT_header(Header):
 
             if gpd.use_stroke_edit_mode:
                 row = layout.row(align=True)
-                row.prop(tool_settings.gpencil_sculpt, "use_select_mask", text="")
-
                 row.popover(
                     panel="VIEW3D_PT_tools_grease_pencil_interpolate",
                     text="Interpolate",
@@ -732,7 +733,10 @@ class VIEW3D_MT_editor_menus(Menu):
         # Select Menu
         if gp_edit:
             if mode_string not in {'PAINT_GPENCIL', 'WEIGHT_GPENCIL'}:
-                if mode_string == 'SCULPT_GPENCIL' and ts.gpencil_sculpt.use_select_mask:
+                if mode_string == 'SCULPT_GPENCIL' and \
+                    (ts.use_gpencil_select_mask_point or
+                     ts.use_gpencil_select_mask_stroke or
+                     ts.use_gpencil_select_mask_segment):
                     layout.menu("VIEW3D_MT_select_gpencil")
                 elif mode_string == 'EDIT_GPENCIL':
                     layout.menu("VIEW3D_MT_select_gpencil")
@@ -2095,6 +2099,9 @@ class VIEW3D_MT_object_relations(Menu):
         layout = self.layout
 
         layout.operator("object.proxy_make", text="Make Proxy...")
+        
+        if bpy.app.use_override_library:
+            layout.operator("object.make_override_library", text="Make Library Override...")
 
         layout.operator("object.make_dupli_face")
 
@@ -2326,6 +2333,7 @@ class VIEW3D_MT_object_context_menu(Menu):
             layout.separator()
 
             layout.operator("object.convert", text="Convert to Mesh").target = 'MESH'
+            layout.operator("object.convert", text="Convert to Grease Pencil").target = 'GPENCIL'
             layout.operator_menu_enum("object.origin_set", text="Set Origin", property="type")
 
             layout.separator()
@@ -2355,10 +2363,10 @@ class VIEW3D_MT_object_context_menu(Menu):
 
             layout.operator_context = 'INVOKE_REGION_WIN'
 
-            props = layout.operator("wm.context_modal_mouse", text="Energy")
+            props = layout.operator("wm.context_modal_mouse", text="Power")
             props.data_path_iter = "selected_editable_objects"
             props.data_path_item = "data.energy"
-            props.header_text = "Light Energy: %.3f"
+            props.header_text = "Light Power: %.3f"
 
             if light.type == 'AREA':
                 props = layout.operator("wm.context_modal_mouse", text="Size X")
@@ -3798,7 +3806,7 @@ class VIEW3D_MT_edit_mesh_normals(Menu):
 
         layout.operator("mesh.set_normals_from_faces", text="Set From Faces")
 
-        layout.operator_context = 'INVOKE_DEFAULT'
+        layout.operator_context = 'INVOKE_REGION_WIN'
         layout.operator("transform.rotate_normal", text="Rotate...")
         layout.operator("mesh.point_normals", text="Point to Target...")
         layout.operator_context = 'EXEC_DEFAULT'
@@ -3929,7 +3937,9 @@ class VIEW3D_MT_edit_gpencil_delete(Menu):
 
         layout.separator()
 
-        layout.operator("gpencil.active_frames_delete_all")
+        layout.operator("gpencil.delete", text="Delete Active Keyframe (Active Layer)").type = 'FRAME'
+        layout.operator("gpencil.active_frames_delete_all", text="Delete Active Keyframes (All Layers)")
+
 # Edit Curve
 # draw_curve is used by VIEW3D_MT_edit_curve and VIEW3D_MT_edit_surface
 
@@ -4385,7 +4395,7 @@ class VIEW3D_MT_gpencil_simplify(Menu):
 
 
 class VIEW3D_MT_paint_gpencil(Menu):
-    bl_label = "Strokes"
+    bl_label = "Draw"
 
     def draw(self, _context):
 
@@ -4396,8 +4406,8 @@ class VIEW3D_MT_paint_gpencil(Menu):
 
         layout.separator()
 
-        layout.operator("gpencil.delete", text="Delete Frame").type = 'FRAME'
-        layout.operator("gpencil.active_frames_delete_all")
+        layout.menu("VIEW3D_MT_edit_gpencil_showhide")
+        layout.menu("GPENCIL_MT_cleanup")
 
 
 class VIEW3D_MT_assign_material(Menu):
@@ -4456,8 +4466,8 @@ class VIEW3D_MT_edit_gpencil(Menu):
         layout.operator("gpencil.duplicate_move", text="Duplicate")
         layout.operator("gpencil.stroke_split", text="Split")
         layout.operator("gpencil.copy", text="Copy", icon='COPYDOWN')
-        layout.operator("gpencil.paste", text="Paste", icon='PASTEDOWN').type = 'COPY'
-        layout.operator("gpencil.paste", text="Paste & Merge").type = 'MERGE'
+        layout.operator("gpencil.paste", text="Paste", icon='PASTEDOWN').type = 'ACTIVE'
+        layout.operator("gpencil.paste", text="Paste by Layer").type = 'LAYER'
 
         layout.separator()
 
@@ -4488,13 +4498,13 @@ class VIEW3D_MT_edit_gpencil_stroke(Menu):
 
         layout.separator()
 
-        layout.operator_menu_enum("gpencil.stroke_join", "type", text="Join...")
+        layout.operator_menu_enum("gpencil.stroke_join", "type", text="Join")
 
         layout.separator()
 
-        layout.operator_menu_enum("gpencil.move_to_layer", "layer", text="Move to Layer")
+        layout.menu("GPENCIL_MT_move_to_layer")
         layout.menu("VIEW3D_MT_assign_material")
-        layout.operator_menu_enum("gpencil.stroke_arrange", "direction", text="Arrange Strokes...")
+        layout.operator_menu_enum("gpencil.stroke_arrange", "direction", text="Arrange Strokes")
 
         layout.separator()
 
@@ -4503,7 +4513,7 @@ class VIEW3D_MT_edit_gpencil_stroke(Menu):
         op.type = 'CLOSE'
         op.geometry = True
         layout.operator("gpencil.stroke_cyclical_set", text="Toggle Cyclic").type = 'TOGGLE'
-        layout.operator_menu_enum("gpencil.stroke_caps_set", text="Toggle Caps...", property="type")
+        layout.operator_menu_enum("gpencil.stroke_caps_set", text="Toggle Caps", property="type")
         layout.operator("gpencil.stroke_flip", text="Switch Direction")
 
 
@@ -4560,17 +4570,18 @@ class VIEW3D_MT_gpencil_animation(Menu):
     def draw(self, _context):
         layout = self.layout
 
-        layout.operator("gpencil.blank_frame_add")
+        layout.operator("gpencil.blank_frame_add", text="Insert Blank Keyframe (Active Layer)")
+        layout.operator("gpencil.blank_frame_add", text="Insert Blank Keyframe (All Layers)").all_layers = True
 
         layout.separator()
 
-        layout.operator("gpencil.frame_duplicate", text="Duplicate Active Frame")
-        layout.operator("gpencil.frame_duplicate", text="Duplicate All Layers").mode = 'ALL'
+        layout.operator("gpencil.frame_duplicate", text="Duplicate Active Keyframe (Active Layer)")
+        layout.operator("gpencil.frame_duplicate", text="Duplicate Active Keyframe (All Layers)").mode = 'ALL'
 
         layout.separator()
 
-        layout.operator("gpencil.delete", text="Delete Active Frame").type = 'FRAME'
-        layout.operator("gpencil.active_frames_delete_all", text="Delete All Active Frames")
+        layout.operator("gpencil.delete", text="Delete Active Keyframe (Active Layer)").type = 'FRAME'
+        layout.operator("gpencil.active_frames_delete_all", text="Delete Active Keyframes (All Layers)")
 
 
 class VIEW3D_MT_edit_gpencil_transform(Menu):
@@ -4591,18 +4602,19 @@ class VIEW3D_MT_edit_gpencil_transform(Menu):
         layout.operator("transform.transform", text="Shrink Fatten").mode = 'GPENCIL_SHRINKFATTEN'
 
 
-        layout.operator("gpencil.interpolate", text="Interpolate")
-        layout.operator("gpencil.interpolate_sequence", text="Sequence")
-
 class VIEW3D_MT_edit_gpencil_showhide(Menu):
     bl_label = "Show/hide"
 
     def draw(self, _context):
         layout = self.layout
 
-        layout.operator("gpencil.hide", text="Hide Active Layer")
         layout.operator("gpencil.reveal", text="Show All Layers")
 
+        layout.separator()
+
+        layout.operator("gpencil.hide", text="Hide Active Layer").unselected = False
+        layout.operator("gpencil.hide", text="Hide Inactive Layers").unselected = True
+
 
 class VIEW3D_MT_edit_gpencil_interpolate(Menu):
     bl_label = "Interpolate"
@@ -5247,8 +5259,10 @@ class VIEW3D_PT_shading_options(Panel):
             sub.active = shading.show_object_outline
             sub.prop(shading, "object_outline_color", text="")
 
+        if shading.type == 'SOLID':
             col = layout.column()
-            if (shading.light == 'STUDIO') and (shading.type != 'WIREFRAME'):
+            if shading.light in {'STUDIO', 'MATCAP'}:
+                col.active = shading.selected_studio_light.has_specular_highlight_pass
                 col.prop(shading, "show_specular_highlight", text="Specular Lighting")
 
 
@@ -5350,6 +5364,8 @@ class VIEW3D_PT_overlay_guides(Panel):
         layout = self.layout
 
         view = context.space_data
+        scene = context.scene
+
         overlay = view.overlay
         shading = view.shading
         display_all = overlay.show_overlays
@@ -5377,6 +5393,8 @@ class VIEW3D_PT_overlay_guides(Panel):
                 (overlay.show_ortho_grid and grid_active)
             )
             sub.prop(overlay, "grid_scale", text="Scale")
+            sub = sub.row(align=True)
+            sub.active = scene.unit_settings.system == 'NONE'
             sub.prop(overlay, "grid_subdivisions", text="Subdivisions")
 
         sub = split.column()
@@ -5882,7 +5900,10 @@ class VIEW3D_PT_pivot_point(Panel):
         if (obj is None) or (mode in {'OBJECT', 'POSE', 'WEIGHT_PAINT'}):
             col.separator()
 
-            col.prop(tool_settings, "use_transform_pivot_point_align")
+            col.label(text="Affect Only")
+            col.prop(tool_settings, "use_transform_data_origin", text="Origins")
+            col.prop(tool_settings, "use_transform_pivot_point_align", text="Locations")
+            col.prop(tool_settings, "use_transform_skip_children", text="Parents (Skip Children)")
 
 
 class VIEW3D_PT_snapping(Panel):
@@ -6266,66 +6287,123 @@ class VIEW3D_PT_gpencil_multi_frame(Panel):
 
 
 class VIEW3D_MT_gpencil_edit_context_menu(Menu):
-    bl_label = "Edit Context Menu"
+    bl_label = ""
 
     def draw(self, context):
-        layout = self.layout
+
+        is_point_mode = context.tool_settings.gpencil_selectmode_edit == 'POINT'
+        is_stroke_mode = context.tool_settings.gpencil_selectmode_edit == 'STROKE'
+        is_segment_mode = context.tool_settings.gpencil_selectmode_edit == 'SEGMENT'
+
         is_3d_view = context.space_data.type == 'VIEW_3D'
 
+        layout = self.layout
+
         layout.operator_context = 'INVOKE_REGION_WIN'
 
-        # Add
-        layout.operator("gpencil.stroke_subdivide", text="Subdivide")
+        row = layout.row()
 
-        layout.separator()
+        if is_point_mode or is_segment_mode:
+            col = row.column()
 
-        # Transform
-        layout.operator("transform.transform", text="Shrink/Fatten").mode = 'GPENCIL_SHRINKFATTEN'
-        layout.operator("gpencil.stroke_smooth", text="Smooth")
-        layout.operator("gpencil.stroke_trim", text="Trim")
+            col.label(text="Point Context Menu", icon='GP_SELECT_POINTS')
+            col.separator()
 
-        layout.separator()
+            # Additive Operators
+            col.operator("gpencil.stroke_subdivide", text="Subdivide").only_selected = True
 
-        # Modify
-        layout.menu("VIEW3D_MT_assign_material")
-        layout.operator_menu_enum("gpencil.stroke_arrange", "direction", text="Arrange Strokes")
-        layout.operator("gpencil.stroke_flip", text="Flip Direction")
-        layout.operator_menu_enum("gpencil.stroke_caps_set", text="Toggle Caps", property="type")
+            col.separator()
 
-        layout.separator()
+            col.operator("gpencil.extrude_move", text="Extrude Points")
 
-        layout.operator("gpencil.duplicate_move", text="Duplicate")
-        layout.operator("gpencil.copy", text="Copy", icon='COPYDOWN')
-        layout.operator("gpencil.paste", text="Paste", icon='PASTEDOWN').type = 'COPY'
-        layout.operator("gpencil.paste", text="Paste & Merge").type = 'MERGE'
-        layout.menu("VIEW3D_MT_gpencil_copy_layer")
-        layout.operator("gpencil.frame_duplicate", text="Duplicate Active Frame")
-        layout.operator("gpencil.frame_duplicate", text="Duplicate Active Frame All Layers").mode = 'ALL'
+            col.separator()
 
-        layout.separator()
+            # Deform Operators
+            col.operator("gpencil.stroke_smooth", text="Smooth Points").only_selected = True
+            col.operator("transform.bend", text="Bend")
+            col.operator("transform.shear", text="Shear")
+            col.operator("transform.tosphere", text="To Sphere")
+            col.operator("transform.transform", text="Shrink Fatten").mode = 'GPENCIL_SHRINKFATTEN'
 
-        layout.operator("gpencil.stroke_join", text="Join").type = 'JOIN'
-        layout.operator("gpencil.stroke_join", text="Join & Copy").type = 'JOINCOPY'
-        layout.operator_menu_enum("gpencil.stroke_separate", "mode")
-        layout.operator("gpencil.stroke_split", text="Split")
-        op = layout.operator("gpencil.stroke_cyclical_set", text="Close")
-        op.type = 'CLOSE'
-        op.geometry = True
+            col.separator()
 
-        layout.separator()
+            col.menu("VIEW3D_MT_mirror", text="Mirror Points")
+            col.menu("VIEW3D_MT_snap", text="Snap Points")
 
-        layout.menu("VIEW3D_MT_mirror")
-        layout.menu("VIEW3D_MT_snap")
+            col.separator()
 
-        layout.separator()
+            # Duplicate operators
+            col.operator("gpencil.duplicate_move", text="Duplicate")
+            col.operator("gpencil.copy", text="Copy", icon='COPYDOWN')
+            col.operator("gpencil.paste", text="Paste", icon='PASTEDOWN').type = 'ACTIVE'
+            col.operator("gpencil.paste", text="Paste by Layer").type = 'LAYER'
 
-        # Remove
-        if is_3d_view:
-            layout.menu("GPENCIL_MT_cleanup")
+            col.separator()
 
-        layout.menu("VIEW3D_MT_gpencil_simplify")
-        layout.operator("gpencil.stroke_merge", text="Merge")
-        layout.menu("VIEW3D_MT_edit_gpencil_delete")
+            # Removal Operators
+            col.operator("gpencil.stroke_merge", text="Merge Points")
+            col.operator("gpencil.stroke_merge_by_distance").use_unselected = False
+            col.operator("gpencil.stroke_split", text="Split")
+            col.operator("gpencil.stroke_separate", text="Separate").mode = 'POINT'
+
+            col.separator()
+
+            col.operator("gpencil.delete", text="Delete Points").type = 'POINTS'
+            col.operator("gpencil.dissolve", text="Dissolve Points").type = 'POINTS'
+            col.operator("gpencil.dissolve", text="Dissolve Between").type = 'BETWEEN'
+            col.operator("gpencil.dissolve", text="Dissolve Unselected").type = 'UNSELECT'
+
+        if is_stroke_mode:
+
+            col = row.column()
+            col.label(text="Stroke Context Menu", icon='GP_SELECT_STROKES')
+            col.separator()
+
+            # Main Strokes Operators
+            col.operator("gpencil.stroke_subdivide", text="Subdivide").only_selected = False
+            col.menu("VIEW3D_MT_gpencil_simplify")
+            col.operator("gpencil.stroke_trim", text="Trim")
+
+            col.separator()
+
+            col.operator("gpencil.stroke_smooth", text="Smooth Stroke").only_selected = False
+            col.operator("transform.transform", text="Shrink Fatten").mode = 'GPENCIL_SHRINKFATTEN'
+
+            col.separator()
+
+            # Layer and Materials operators
+            col.menu("GPENCIL_MT_move_to_layer")            
+            col.menu("VIEW3D_MT_assign_material")
+            col.operator_menu_enum("gpencil.stroke_arrange", "direction", text="Arrange Strokes")
+
+            col.separator()
+
+            col.menu("VIEW3D_MT_mirror", text="Mirror Stroke")
+            col.menu("VIEW3D_MT_snap", text="Snap Stroke")
+
+            col.separator()
+
+            # Duplicate operators
+            col.operator("gpencil.duplicate_move", text="Duplicate")
+            col.operator("gpencil.copy", text="Copy", icon='COPYDOWN')
+            col.operator("gpencil.paste", text="Paste", icon='PASTEDOWN').type = 'ACTIVE'
+            col.operator("gpencil.paste", text="Paste by Layer").type = 'LAYER'
+
+            col.separator()
+
+            # Removal Operators
+            col.operator("gpencil.stroke_merge_by_distance").use_unselected = True
+            col.operator_menu_enum("gpencil.stroke_join", "type", text="Join")
+            col.operator("gpencil.stroke_split", text="Split")
+            col.operator("gpencil.stroke_separate", text="Separate").mode = 'STROKE'
+
+            col.separator()
+
+            col.operator("gpencil.delete", text="Delete Strokes").type = 'STROKES'
+
+            col.separator()
+
+            col.operator("gpencil.reproject", text="Reproject Strokes")
 
 
 class VIEW3D_PT_gpencil_sculpt_context_menu(Panel):
@@ -6350,14 +6428,18 @@ class VIEW3D_PT_gpencil_sculpt_context_menu(Panel):
 
         layout.operator_context = 'INVOKE_REGION_WIN'
 
-        layout.operator("gpencil.blank_frame_add", text="Insert Blank", icon='ADD')
-        layout.operator("gpencil.frame_duplicate", text="Duplicate Active", icon='DUPLICATE')
-        layout.operator("gpencil.frame_duplicate", text="Duplicate for All Layers", icon='DUPLICATE').mode = 'ALL'
+        layout.operator("gpencil.blank_frame_add", text="Insert Blank in Active Layer", icon='ADD')
+        layout.operator("gpencil.blank_frame_add", text="Insert Blank in All Layers", icon='ADD').all_layers = True
 
         layout.separator()
 
-        layout.operator("gpencil.delete", text="Delete Active", icon='REMOVE').type = 'FRAME'
-        layout.operator("gpencil.active_frames_delete_all", text="Delete All Active Layers", icon='REMOVE')
+        layout.operator("gpencil.frame_duplicate", text="Duplicate Active Layer", icon='DUPLICATE')
+        layout.operator("gpencil.frame_duplicate", text="Duplicate All Layers", icon='DUPLICATE').mode = 'ALL'
+
+        layout.separator()
+
+        layout.operator("gpencil.delete", text="Delete Active Layer", icon='REMOVE').type = 'FRAME'
+        layout.operator("gpencil.active_frames_delete_all", text="Delete All Layers", icon='REMOVE')
 
 
 class VIEW3D_PT_gpencil_draw_context_menu(Panel):
@@ -6383,14 +6465,18 @@ class VIEW3D_PT_gpencil_draw_context_menu(Panel):
 
         layout.operator_context = 'INVOKE_REGION_WIN'
 
-        layout.operator("gpencil.blank_frame_add", text="Insert Blank", icon='ADD')
-        layout.operator("gpencil.frame_duplicate", text="Duplicate Active", icon='DUPLICATE')
-        layout.operator("gpencil.frame_duplicate", text="Duplicate for All Layers", icon='DUPLICATE').mode = 'ALL'
+        layout.operator("gpencil.blank_frame_add", text="Insert Blank in Active Layer", icon='ADD')
+        layout.operator("gpencil.blank_frame_add", text="Insert Blank in All Layers", icon='ADD').all_layers = True
+
+        layout.separator()
+
+        layout.operator("gpencil.frame_duplicate", text="Duplicate Active Layer", icon='DUPLICATE')
+        layout.operator("gpencil.frame_duplicate", text="Duplicate All Layers", icon='DUPLICATE').mode = 'ALL'
 
         layout.separator()
 
-        layout.operator("gpencil.delete", text="Delete Active", icon='REMOVE').type = 'FRAME'
-        layout.operator("gpencil.active_frames_delete_all", text="Delete All Active Layers", icon='REMOVE')
+        layout.operator("gpencil.delete", text="Delete Active Layer", icon='REMOVE').type = 'FRAME'
+        layout.operator("gpencil.active_frames_delete_all", text="Delete All Layers", icon='REMOVE')
 
 
 class VIEW3D_PT_paint_vertex_context_menu(Panel):
diff --git a/release/scripts/startup/bl_ui/space_view3d_toolbar.py b/release/scripts/startup/bl_ui/space_view3d_toolbar.py
index 7f1d4e513cf..3982b0fe128 100644
--- a/release/scripts/startup/bl_ui/space_view3d_toolbar.py
+++ b/release/scripts/startup/bl_ui/space_view3d_toolbar.py
@@ -177,6 +177,7 @@ class VIEW3D_PT_tools_meshedit_options_automerge(View3DPanel, Panel):
         layout.use_property_decorate = False
 
         layout.active = tool_settings.use_mesh_automerge
+        layout.prop(tool_settings, "use_mesh_automerge_and_split", toggle=False)
         layout.prop(tool_settings, "double_threshold", text="Threshold")
 
 # ********** default tools for editmode_curve ****************
@@ -363,6 +364,11 @@ class VIEW3D_PT_tools_brush(Panel, View3DPaintPanel):
             if not self.is_popover:
                 brush_basic_sculpt_settings(col, context, brush)
 
+            # normal_radius_factor
+            col.separator()
+            row = col.row()
+            row.prop(brush, "normal_radius_factor", slider=True)
+
             # topology_rake_factor
             if (
                     capabilities.has_topology_rake and
@@ -912,6 +918,8 @@ class VIEW3D_PT_tools_brush_stroke(Panel, View3DPaintPanel):
             if brush.sculpt_capabilities.has_space_attenuation:
                 col.prop(brush, "use_space_attenuation")
 
+            col.prop(brush, "use_scene_spacing")
+
             if brush.sculpt_capabilities.has_jitter:
 
                 row = col.row(align=True)
diff --git a/release/scripts/startup/nodeitems_builtins.py b/release/scripts/startup/nodeitems_builtins.py
index c6e83a646e5..5dc2cf8d106 100644
--- a/release/scripts/startup/nodeitems_builtins.py
+++ b/release/scripts/startup/nodeitems_builtins.py
@@ -186,6 +186,7 @@ shader_node_categories = [
         NodeItem("ShaderNodeAmbientOcclusion"),
         NodeItem("ShaderNodeObjectInfo"),
         NodeItem("ShaderNodeHairInfo"),
+        NodeItem("ShaderNodeVolumeInfo"),
         NodeItem("ShaderNodeParticleInfo"),
         NodeItem("ShaderNodeCameraData"),
         NodeItem("ShaderNodeUVMap"),
@@ -237,6 +238,7 @@ shader_node_categories = [
         NodeItem("ShaderNodeTexBrick"),
         NodeItem("ShaderNodeTexPointDensity"),
         NodeItem("ShaderNodeTexIES"),
+        NodeItem("ShaderNodeTexWhiteNoise"),
     ]),
     ShaderNodeCategory("SH_NEW_OP_COLOR", "Color", items=[
         NodeItem("ShaderNodeMixRGB"),
diff --git a/release/windows/blendthumb/src/BlenderThumb.cpp b/release/windows/blendthumb/src/BlenderThumb.cpp
deleted file mode 100644
index 553428d5b5d..00000000000
--- a/release/windows/blendthumb/src/BlenderThumb.cpp
+++ /dev/null
@@ -1,320 +0,0 @@
-/*
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public License
- * as published by the Free Software Foundation; either version 2
- * of the License, or (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
- */
-
-#include <shlwapi.h>
-#include <thumbcache.h> // For IThumbnailProvider.
-#include <new>
-
-#pragma comment(lib, "shlwapi.lib")
-
-// this thumbnail provider implements IInitializeWithStream to enable being hosted
-// in an isolated process for robustness
-
-class CBlendThumb : public IInitializeWithStream, public IThumbnailProvider
-{
-public:
-	CBlendThumb() : _cRef(1), _pStream(NULL) {}
-
-	virtual ~CBlendThumb()
-	{
-		if (_pStream)
-		{
-			_pStream->Release();
-		}
-	}
-
-	// IUnknown
-	IFACEMETHODIMP QueryInterface(REFIID riid, void **ppv)
-	{
-		static const QITAB qit[] =
-		{
-			QITABENT(CBlendThumb, IInitializeWithStream),
-			QITABENT(CBlendThumb, IThumbnailProvider),
-			{ 0 },
-		};
-		return QISearch(this, qit, riid, ppv);
-	}
-
-	IFACEMETHODIMP_(ULONG) AddRef()
-	{
-		return InterlockedIncrement(&_cRef);
-	}
-
-	IFACEMETHODIMP_(ULONG) Release()
-	{
-		ULONG cRef = InterlockedDecrement(&_cRef);
-		if (!cRef)
-		{
-			delete this;
-		}
-		return cRef;
-	}
-
-	// IInitializeWithStream
-	IFACEMETHODIMP Initialize(IStream *pStream, DWORD grfMode);
-
-	// IThumbnailProvider
-	IFACEMETHODIMP GetThumbnail(UINT cx, HBITMAP *phbmp, WTS_ALPHATYPE *pdwAlpha);
-
-private:
-	long _cRef;
-	IStream *_pStream;	 // provided during initialization.
-};
-
-HRESULT CBlendThumb_CreateInstance(REFIID riid, void **ppv)
-{
-	CBlendThumb *pNew = new (std::nothrow) CBlendThumb();
-	HRESULT hr = pNew ? S_OK : E_OUTOFMEMORY;
-	if (SUCCEEDED(hr))
-	{
-		hr = pNew->QueryInterface(riid, ppv);
-		pNew->Release();
-	}
-	return hr;
-}
-
-// IInitializeWithStream
-IFACEMETHODIMP CBlendThumb::Initialize(IStream *pStream, DWORD)
-{
-	HRESULT hr = E_UNEXPECTED;  // can only be inited once
-	if (_pStream == NULL)
-	{
-		// take a reference to the stream if we have not been inited yet
-		hr = pStream->QueryInterface(&_pStream);
-	}
-	return hr;
-}
-
-#include <math.h>
-#include <zlib.h>
-#include "Wincodec.h"
-const unsigned char gzip_magic[3] = { 0x1f, 0x8b, 0x08 };
-
-// IThumbnailProvider
-IFACEMETHODIMP CBlendThumb::GetThumbnail(UINT cx, HBITMAP *phbmp, WTS_ALPHATYPE *pdwAlpha)
-{
-	ULONG BytesRead;
-	HRESULT hr = S_FALSE;
-	LARGE_INTEGER SeekPos;
-
-	// Compressed?	
-	unsigned char in_magic[3];
-	_pStream->Read(&in_magic,3,&BytesRead);
-	bool gzipped = true;
-	for ( int i=0; i < 3; i++ )
-		if ( in_magic[i] != gzip_magic[i] )
-		{
-			gzipped = false;
-			break;
-		}
-
-	if (gzipped)
-	{
-		// Zlib inflate
-		z_stream stream;
-		stream.zalloc = Z_NULL;
-		stream.zfree = Z_NULL;
-		stream.opaque = Z_NULL;
-
-		// Get compressed file length
-		SeekPos.QuadPart = 0;
-		_pStream->Seek(SeekPos,STREAM_SEEK_END,NULL);
-
-		// Get compressed and uncompressed size
-		uLong source_size;
-		uLongf dest_size;
-		//SeekPos.QuadPart = -4; // last 4 bytes define size of uncompressed file
-		//ULARGE_INTEGER Tell;
-		//_pStream->Seek(SeekPos,STREAM_SEEK_END,&Tell);
-		//source_size = (uLong)Tell.QuadPart + 4; // src
-		//_pStream->Read(&dest_size,4,&BytesRead); // dest
-		dest_size = 1024*70; // thumbnail is currently always inside the first 65KB...if it moves or enlargens this line will have to change or go!
-		source_size = (uLong)max(SeekPos.QuadPart,dest_size); // for safety, assume no compression
-		
-		// Input
-		Bytef* src = new Bytef[source_size];
-		stream.next_in = (Bytef*)src;
-		stream.avail_in = (uInt)source_size;
-		
-		// Output
-		Bytef* dest = new Bytef[dest_size];
-		stream.next_out = (Bytef*)dest;
-		stream.avail_out = dest_size; 
-
-		// IStream to src
-		SeekPos.QuadPart = 0;
-		_pStream->Seek(SeekPos,STREAM_SEEK_SET,NULL);
-		_pStream->Read(src,source_size,&BytesRead);
-		
-		// Do the inflation
-		int err;
-		err = inflateInit2(&stream,16); // 16 means "gzip"...nice!
-		err = inflate(&stream, Z_FINISH);		
-		err = inflateEnd(&stream);
-				
-		// Replace the IStream, which is read-only
-		_pStream->Release();
-		_pStream = SHCreateMemStream(dest,dest_size);
-		
-		delete[] src;
-		delete[] dest;
-	}
-
-	// Blender version, early out if sub 2.5
-	SeekPos.QuadPart = 9;
-	_pStream->Seek(SeekPos,STREAM_SEEK_SET,NULL);
-	char version[4];
-	version[3] = '\0';
-	_pStream->Read(&version,3,&BytesRead);
-	if ( BytesRead != 3)
-		return E_UNEXPECTED;
-	int iVersion = atoi(version);
-	if ( iVersion < 250 )
-		return S_FALSE;
-	
-	// 32 or 64 bit blend?
-	SeekPos.QuadPart = 7;
-	_pStream->Seek(SeekPos,STREAM_SEEK_SET,NULL);
-
-	char _PointerSize;
-	_pStream->Read(&_PointerSize,1,&BytesRead);
-
-	int PointerSize	= _PointerSize == '_' ? 4 : 8;
-	int HeaderSize	= 16 + PointerSize;
-
-	// Find and read thumbnail ("TEST") block
-	SeekPos.QuadPart = 12;
-	_pStream->Seek(SeekPos,STREAM_SEEK_SET,NULL);
-	int BlockOffset = 12;
-	while ( _pStream )
-	{
-		// Scan current block
-		char BlockName[5];
-		BlockName[4] = '\0';
-		int	BlockSize = 0;
-
-		if (_pStream->Read(BlockName,4,&BytesRead) == S_OK && _pStream->Read((void*)&BlockSize,4,&BytesRead) == S_OK)
-		{
-			if ( strcmp (BlockName,"TEST") != 0 )
-			{
-				SeekPos.QuadPart = BlockOffset += HeaderSize + BlockSize;
-				_pStream->Seek(SeekPos,STREAM_SEEK_SET,NULL);
-				continue;
-			}
-		}
-		else break; // eof
-
-		// Found the block
-		SeekPos.QuadPart = BlockOffset + HeaderSize;
-		_pStream->Seek(SeekPos,STREAM_SEEK_SET,NULL);
-
-		int width, height;
-		_pStream->Read((char*)&width,4,&BytesRead);
-		_pStream->Read((char*)&height,4,&BytesRead);
-		BlockSize -= 8;
-
-		// Isolate RGBA data
-		char* pRGBA = new char[BlockSize];
-		_pStream->Read(pRGBA,BlockSize,&BytesRead);
-
-		if (BytesRead != (ULONG)BlockSize)
-			return E_UNEXPECTED;
-
-		// Convert to BGRA for Windows
-		for (int i=0; i < BlockSize; i+=4 )
-		{
-			#define RED_BYTE pRGBA[i]
-			#define BLUE_BYTE pRGBA[i+2]
-
-			char red = RED_BYTE;
-			RED_BYTE = BLUE_BYTE;
-			BLUE_BYTE = red;
-		}
-		
-		// Flip vertically (Blender stores it upside-down)
-		unsigned int LineSize = width*4;
-		char* FlippedImage = new char[BlockSize];
-		for (int i=0; i<height; i++)
-		{
-			if ( 0 != memcpy_s(&FlippedImage[ (height - i - 1)*LineSize ],LineSize,&pRGBA[ i*LineSize ],LineSize))
-				return E_UNEXPECTED;
-		}
-		delete[] pRGBA;
-		pRGBA = FlippedImage;
-
-		// Create image
-		*phbmp = CreateBitmap(width,height,1,32,pRGBA);
-		if (!*phbmp)
-			return E_FAIL;
-		*pdwAlpha = WTSAT_ARGB; // it's actually BGRA, not sure why this works
-
-		// Scale down if required
-		if ( (unsigned)width > cx || (unsigned)height > cx )
-		{
-			float scale = 1.0f / (max(width,height) / (float)cx);
-			LONG NewWidth = (LONG)(width *scale);
-			LONG NewHeight = (LONG)(height *scale);
-
-#ifdef _DEBUG
-#if 1
-			MessageBox(0,L"Attach now",L"Debugging",MB_OK);
-#endif
-#endif
-			IWICImagingFactory *pImgFac;
-			hr = CoCreateInstance(CLSID_WICImagingFactory, NULL, CLSCTX_INPROC_SERVER, IID_PPV_ARGS(&pImgFac));
-			
-			IWICBitmap* WICBmp;
-			hr = pImgFac->CreateBitmapFromHBITMAP(*phbmp,0,WICBitmapUseAlpha,&WICBmp);
-			
-			BITMAPINFO bmi = {};
-			bmi.bmiHeader.biSize = sizeof(bmi.bmiHeader);
-			bmi.bmiHeader.biWidth = NewWidth;
-			bmi.bmiHeader.biHeight = -NewHeight;
-			bmi.bmiHeader.biPlanes = 1;
-			bmi.bmiHeader.biBitCount = 32;
-			bmi.bmiHeader.biCompression = BI_RGB;
-
-			BYTE *pBits;
-			HBITMAP ResizedHBmp = CreateDIBSection(NULL, &bmi, DIB_RGB_COLORS, (void**)&pBits, NULL, 0);
-			hr = ResizedHBmp ? S_OK : E_OUTOFMEMORY;
-			if (SUCCEEDED(hr))
-			{				
-				IWICBitmapScaler* pIScaler;
-				hr = pImgFac->CreateBitmapScaler(&pIScaler);
-				hr = pIScaler->Initialize(WICBmp,NewWidth,NewHeight,WICBitmapInterpolationModeFant);
-								
-				WICRect rect = {0, 0, NewWidth, NewHeight};
-				hr = pIScaler->CopyPixels(&rect, NewWidth * 4, NewWidth * NewHeight * 4, pBits);
-
-				if (SUCCEEDED(hr))
-				{
-					DeleteObject(*phbmp);
-					*phbmp = ResizedHBmp;
-				}
-				else
-					DeleteObject(ResizedHBmp);
-
-				pIScaler->Release();
-			}
-			WICBmp->Release();
-			pImgFac->Release();
-		}
-		else
-			hr = S_OK;
-		break;
-	}
-	return hr;
-}
diff --git a/release/windows/blendthumb/src/Dll.cpp b/release/windows/blendthumb/src/Dll.cpp
deleted file mode 100644
index 7b0521cd5a8..00000000000
--- a/release/windows/blendthumb/src/Dll.cpp
+++ /dev/null
@@ -1,273 +0,0 @@
-/*
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public License
- * as published by the Free Software Foundation; either version 2
- * of the License, or (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
- */
-
-#include <objbase.h>
-#include <shlwapi.h>
-#include <thumbcache.h> // For IThumbnailProvider.
-#include <shlobj.h>     // For SHChangeNotify
-#include <new>
-
-extern HRESULT CBlendThumb_CreateInstance(REFIID riid, void **ppv);
-
-#define SZ_CLSID_BLENDTHUMBHANDLER     L"{D45F043D-F17F-4e8a-8435-70971D9FA46D}"
-#define SZ_BLENDTHUMBHANDLER           L"Blender Thumbnail Handler"
-const CLSID CLSID_BlendThumbHandler	= { 0xd45f043d, 0xf17f, 0x4e8a, { 0x84, 0x35, 0x70, 0x97, 0x1d, 0x9f, 0xa4, 0x6d } };
-
-typedef HRESULT (*PFNCREATEINSTANCE)(REFIID riid, void **ppvObject);
-struct CLASS_OBJECT_INIT
-{
-    const CLSID *pClsid;
-    PFNCREATEINSTANCE pfnCreate;
-};
-
-// add classes supported by this module here
-const CLASS_OBJECT_INIT c_rgClassObjectInit[] =
-{
-    { &CLSID_BlendThumbHandler, CBlendThumb_CreateInstance }
-};
-
-
-long g_cRefModule = 0;
-
-// Handle the DLL's module
-HINSTANCE g_hInst = NULL;
-
-// Standard DLL functions
-STDAPI_(BOOL) DllMain(HINSTANCE hInstance, DWORD dwReason, void *)
-{
-    if (dwReason == DLL_PROCESS_ATTACH)
-    {
-        g_hInst = hInstance;
-        DisableThreadLibraryCalls(hInstance);
-    }
-    return TRUE;
-}
-
-STDAPI DllCanUnloadNow()
-{
-    // Only allow the DLL to be unloaded after all outstanding references have been released
-    return (g_cRefModule == 0) ? S_OK : S_FALSE;
-}
-
-void DllAddRef()
-{
-    InterlockedIncrement(&g_cRefModule);
-}
-
-void DllRelease()
-{
-    InterlockedDecrement(&g_cRefModule);
-}
-
-class CClassFactory : public IClassFactory
-{
-public:
-    static HRESULT CreateInstance(REFCLSID clsid, const CLASS_OBJECT_INIT *pClassObjectInits, size_t cClassObjectInits, REFIID riid, void **ppv)
-    {
-        *ppv = NULL;
-        HRESULT hr = CLASS_E_CLASSNOTAVAILABLE;
-        for (size_t i = 0; i < cClassObjectInits; i++)
-        {
-            if (clsid == *pClassObjectInits[i].pClsid)
-            {
-                IClassFactory *pClassFactory = new (std::nothrow) CClassFactory(pClassObjectInits[i].pfnCreate);
-                hr = pClassFactory ? S_OK : E_OUTOFMEMORY;
-                if (SUCCEEDED(hr))
-                {
-                    hr = pClassFactory->QueryInterface(riid, ppv);
-                    pClassFactory->Release();
-                }
-                break; // match found
-            }
-        }
-        return hr;
-    }
-
-    CClassFactory(PFNCREATEINSTANCE pfnCreate) : _cRef(1), _pfnCreate(pfnCreate)
-    {
-        DllAddRef();
-    }
-
-    // IUnknown
-    IFACEMETHODIMP QueryInterface(REFIID riid, void ** ppv)
-    {
-        static const QITAB qit[] =
-        {
-            QITABENT(CClassFactory, IClassFactory),
-            { 0 }
-        };
-        return QISearch(this, qit, riid, ppv);
-    }
-
-    IFACEMETHODIMP_(ULONG) AddRef()
-    {
-        return InterlockedIncrement(&_cRef);
-    }
-
-    IFACEMETHODIMP_(ULONG) Release()
-    {
-        long cRef = InterlockedDecrement(&_cRef);
-        if (cRef == 0)
-        {
-            delete this;
-        }
-        return cRef;
-    }
-
-    // IClassFactory
-    IFACEMETHODIMP CreateInstance(IUnknown *punkOuter, REFIID riid, void **ppv)
-    {
-        return punkOuter ? CLASS_E_NOAGGREGATION : _pfnCreate(riid, ppv);
-    }
-
-    IFACEMETHODIMP LockServer(BOOL fLock)
-    {
-        if (fLock)
-        {
-            DllAddRef();
-        }
-        else
-        {
-            DllRelease();
-        }
-        return S_OK;
-    }
-
-private:
-    ~CClassFactory()
-    {
-        DllRelease();
-    }
-
-    long _cRef;
-    PFNCREATEINSTANCE _pfnCreate;
-};
-
-STDAPI DllGetClassObject(REFCLSID clsid, REFIID riid, void **ppv)
-{
-    return CClassFactory::CreateInstance(clsid, c_rgClassObjectInit, ARRAYSIZE(c_rgClassObjectInit), riid, ppv);
-}
-
-// A struct to hold the information required for a registry entry
-
-struct REGISTRY_ENTRY
-{
-    HKEY   hkeyRoot;
-    PCWSTR pszKeyName;
-    PCWSTR pszValueName;
-	DWORD  dwValueType;
-    PCWSTR pszData;
-};
-
-// Creates a registry key (if needed) and sets the default value of the key
-
-HRESULT CreateRegKeyAndSetValue(const REGISTRY_ENTRY *pRegistryEntry)
-{
-    HKEY hKey;
-    HRESULT hr = HRESULT_FROM_WIN32(RegCreateKeyExW(pRegistryEntry->hkeyRoot, pRegistryEntry->pszKeyName,
-                                0, NULL, REG_OPTION_NON_VOLATILE, KEY_SET_VALUE, NULL, &hKey, NULL));
-    if (SUCCEEDED(hr))
-    {
-		// All this just to support REG_DWORD...
-		DWORD size;
-		DWORD data;
-		BYTE* lpData = (LPBYTE) pRegistryEntry->pszData;
-		switch (pRegistryEntry->dwValueType)
-		{
-		case REG_SZ:
-			size = ((DWORD) wcslen(pRegistryEntry->pszData) + 1) * sizeof(WCHAR);
-			break;
-		case REG_DWORD:
-			size = sizeof(DWORD);
-			data = (DWORD)pRegistryEntry->pszData;
-			lpData = (BYTE*)&data;
-			break;
-		default:
-			return E_INVALIDARG;
-		}
-
-        hr = HRESULT_FROM_WIN32(RegSetValueExW(hKey, pRegistryEntry->pszValueName, 0, pRegistryEntry->dwValueType,
-                            lpData, size ));
-        RegCloseKey(hKey);
-    }
-    return hr;
-}
-
-//
-// Registers this COM server
-//
-STDAPI DllRegisterServer()
-{
-    HRESULT hr;
-
-    WCHAR szModuleName[MAX_PATH];
-
-    if (!GetModuleFileNameW(g_hInst, szModuleName, ARRAYSIZE(szModuleName)))
-    {
-        hr = HRESULT_FROM_WIN32(GetLastError());
-    }
-    else
-    {
-        const REGISTRY_ENTRY rgRegistryEntries[] =
-        {
-            // RootKey			KeyName																			ValueName			ValueType		Data
-            {HKEY_CURRENT_USER,	L"Software\\Classes\\CLSID\\" SZ_CLSID_BLENDTHUMBHANDLER,						NULL,				REG_SZ,			SZ_BLENDTHUMBHANDLER},
-            {HKEY_CURRENT_USER,	L"Software\\Classes\\CLSID\\" SZ_CLSID_BLENDTHUMBHANDLER L"\\InProcServer32",	NULL,				REG_SZ,			szModuleName},
-            {HKEY_CURRENT_USER,	L"Software\\Classes\\CLSID\\" SZ_CLSID_BLENDTHUMBHANDLER L"\\InProcServer32",	L"ThreadingModel",	REG_SZ,			L"Apartment"},
-            {HKEY_CURRENT_USER,	L"Software\\Classes\\.blend\\",													L"Treatment",		REG_DWORD,		0}, // doesn't appear to do anything...
-            {HKEY_CURRENT_USER,	L"Software\\Classes\\.blend\\ShellEx\\{e357fccd-a995-4576-b01f-234630154e96}",	NULL,				REG_SZ,			SZ_CLSID_BLENDTHUMBHANDLER},
-        };
-
-        hr = S_OK;
-        for (int i = 0; i < ARRAYSIZE(rgRegistryEntries) && SUCCEEDED(hr); i++)
-        {
-            hr = CreateRegKeyAndSetValue(&rgRegistryEntries[i]);
-        }
-    }
-    if (SUCCEEDED(hr))
-    {
-        // This tells the shell to invalidate the thumbnail cache.  This is important because any .blend files
-        // viewed before registering this handler would otherwise show cached blank thumbnails.
-        SHChangeNotify(SHCNE_ASSOCCHANGED, SHCNF_IDLIST, NULL, NULL);
-    }
-    return hr;
-}
-
-//
-// Unregisters this COM server
-//
-STDAPI DllUnregisterServer()
-{
-    HRESULT hr = S_OK;
-
-    const PCWSTR rgpszKeys[] =
-    {
-        L"Software\\Classes\\CLSID\\" SZ_CLSID_BLENDTHUMBHANDLER,
-        L"Software\\Classes\\.blend\\ShellEx\\{e357fccd-a995-4576-b01f-234630154e96}"
-    };
-
-    // Delete the registry entries
-    for (int i = 0; i < ARRAYSIZE(rgpszKeys) && SUCCEEDED(hr); i++)
-    {
-        hr = HRESULT_FROM_WIN32(RegDeleteTreeW(HKEY_CURRENT_USER, rgpszKeys[i]));
-        if (hr == HRESULT_FROM_WIN32(ERROR_FILE_NOT_FOUND))
-        {
-            // If the registry entry has already been deleted, say S_OK.
-            hr = S_OK;
-        }
-    }
-    return hr;
-}
diff --git a/release/windows/installer/00.checked.bmp b/release/windows/installer/00.checked.bmp
deleted file mode 100644
index 6c2e98d361c..00000000000
--- a/release/windows/installer/00.checked.bmp
+++ /dev/null
diff --git a/release/windows/installer/00.header.bmp b/release/windows/installer/00.header.bmp
deleted file mode 100644
index b631ba73933..00000000000
--- a/release/windows/installer/00.header.bmp
+++ /dev/null
diff --git a/release/windows/installer/00.installer.ico b/release/windows/installer/00.installer.ico
deleted file mode 100644
index 922c9d472d9..00000000000
--- a/release/windows/installer/00.installer.ico
+++ /dev/null
diff --git a/release/windows/installer/00.sconsblender.nsi b/release/windows/installer/00.sconsblender.nsi
deleted file mode 100644
index 2d91b9da7f1..00000000000
--- a/release/windows/installer/00.sconsblender.nsi
+++ /dev/null
@@ -1,240 +0,0 @@
-;
-;
-; Blender Self-Installer for Windows (NSIS - http://nsis.sourceforge.net)
-;
-
-SetCompressor /SOLID lzma
-
-Name "Blender [VERSION]" 
-
-RequestExecutionLevel admin
-
-!include "MUI.nsh"
-!include "WinVer.nsh"
-!include "FileFunc.nsh"
-!include "WordFunc.nsh"
-!include "nsDialogs.nsh"
-!include "x64.nsh"
-
-!define MUI_ABORTWARNING
-
-!define MUI_WELCOMEPAGE_TEXT  "This wizard will guide you through the installation of Blender. It is recommended that you close all other applications before starting Setup."
-!define MUI_WELCOMEFINISHPAGE_BITMAP "[RELDIR]\01.installer.bmp"
-!define MUI_HEADERIMAGE
-!define MUI_HEADERIMAGE_BITMAP  "[RELDIR]\00.header.bmp"
-!define MUI_COMPONENTSPAGE_SMALLDESC
-!define MUI_FINISHPAGE_RUN "$INSTDIR\blender.exe"
-!define MUI_CHECKBITMAP "[RELDIR]\00.checked.bmp"
-!define MUI_UNWELCOMEFINISHPAGE_BITMAP "[RELDIR]\01.installer.bmp"
-
-!insertmacro MUI_PAGE_WELCOME
-!insertmacro MUI_PAGE_LICENSE "[DISTDIR]\Copyright.txt"
-!insertmacro MUI_PAGE_COMPONENTS
-    
-!insertmacro MUI_PAGE_DIRECTORY
-!insertmacro MUI_PAGE_INSTFILES
-!insertmacro MUI_PAGE_FINISH
-  
-!insertmacro MUI_UNPAGE_WELCOME
-UninstPage custom un.OptionalRemoveConfig un.OptionalRemoveConfigOnLeave
-!insertmacro MUI_UNPAGE_CONFIRM
-!insertmacro MUI_UNPAGE_INSTFILES
-!insertmacro MUI_UNPAGE_FINISH
-
-!insertmacro Locate
-!insertmacro VersionCompare
-
-
-Icon "[RELDIR]\00.installer.ico"
-UninstallIcon "[RELDIR]\00.installer.ico"
-
-;--------------------------------
-;Languages
- 
-  !insertmacro MUI_LANGUAGE "English"
-    
-;--------------------------------
-;Language Strings
-
-  ;Description
-  LangString DESC_InstallFiles ${LANG_ENGLISH} "Copy all required files to the application folder."
-  LangString DESC_StartMenu ${LANG_ENGLISH} "Add shortcut items to the Start Menu. (Recommended)"
-  LangString DESC_DesktopShortcut ${LANG_ENGLISH} "Add a shortcut to Blender on your desktop."
-  LangString DESC_BlendRegister ${LANG_ENGLISH} "Blender can register itself with .blend files to allow double-clicking from Windows Explorer, etc."
-;--------------------------------
-;Data
-
-Caption "Blender [VERSION] Installer"
-OutFile "[DISTDIR]\..\blender-[VERSION]-windows[BITNESS].exe"
-InstallDir $INSTDIR ; $INSTDIR is set inside .onInit
-BrandingText "Blender Foundation | http://www.blender.org"
-ComponentText "This will install Blender [VERSION] on your computer."
-
-VIAddVersionKey "ProductName" "Blender"
-VIAddVersionKey "CompanyName" "http://www.blender.org"
-VIAddVersionKey "FileDescription" "Free open source 3D content creation suite."
-VIAddVersionKey "FileVersion" "[SHORTVERSION].0.0"
-
-VIProductVersion "[SHORTVERSION].0.0"
-
-DirText "Use the field below to specify the folder where you want Blender to be copied to. To specify a different folder, type a new name or use the Browse button to select an existing folder."
-
-SilentUnInstall normal
-
-Var SHORTVERSION ; This is blender_version_decimal() from path_util.c
-Var BLENDERCONFIG
-Var REMOVECONFIG
-
-; Custom controls
-Var HWND
-
-Var HWND_KEEPCONFIG
-Var HWND_REMOVECONFIG
-
-Function .onInit
-  ClearErrors
-  StrCpy $SHORTVERSION "[SHORTVERSION]"
-
-  ${If} ${RunningX64}
-    ${If} "[BITNESS]" == "32"
-      StrCpy $INSTDIR "$PROGRAMFILES32\Blender Foundation\Blender" ; Can't use InstallDir inside Section
-    ${ElseIf} "[BITNESS]" == "64"
-      StrCpy $INSTDIR "$PROGRAMFILES64\Blender Foundation\Blender"
-    ${EndIf}
-  ${Else}
-    StrCpy $INSTDIR "$PROGRAMFILES\Blender Foundation\Blender"
-  ${EndIf}
-FunctionEnd
-
-Function un.onInit
-  SetShellVarContext current
-  StrCpy $BLENDERCONFIG "$APPDATA\Blender Foundation\Blender"
-  SetShellVarContext all
-FunctionEnd
-
-Function un.OptionalRemoveConfig
-  nsDialogs::Create /NOUNLOAD 1018
-  Pop $HWND
-  
-  ${If} $HWND == error
-    Abort
-  ${EndIf}
-  
-  ${NSD_CreateRadioButton} 0 50 100% 12u "Keep configuration files, autosaved .blend files and installed addons (recommended)"
-  Pop $HWND_KEEPCONFIG
-  ${NSD_CreateRadioButton} 0 80 100% 12u "Remove all files, including configuration files, autosaved .blend files and installed addons"
-  Pop $HWND_REMOVECONFIG
-
-  SendMessage $HWND_KEEPCONFIG ${BM_SETCHECK} 1 0
-  
-  nsDialogs::Show
-  
-FunctionEnd
-
-Function un.OptionalRemoveConfigOnLeave
-  ${NSD_GetState} $HWND_REMOVECONFIG $R0
-  ${If} $R0 == "1"
-    StrCpy $REMOVECONFIG "1"
-  ${Else}
-    StrCpy $REMOVECONFIG "0"
-  ${EndIf}
-FunctionEnd
-
-
-Section "Blender [VERSION] (required)" InstallFiles
-  SectionIn RO
-
-  ; Set output path to the installation directory.
-  SetOutPath $INSTDIR
-  ; The contents of Blender installation root dir
-  [ROOTDIRCONTS]
-  
-  ; All datafiles (python, scripts, datafiles)
-  [DODATAFILES]
-  
-  SetOutPath $INSTDIR
-  ${If} ${RunningX64}
-    SetRegView 64
-  ${EndIf}
-  ; Write the installation path into the registry
-  WriteRegStr HKLM "SOFTWARE\BlenderFoundation" "Install_Dir" "$INSTDIR"
-  WriteRegStr HKLM "SOFTWARE\BlenderFoundation" "ShortVersion" "[SHORTVERSION]"
-  ; Write the uninstall keys for Windows
-  WriteRegStr   HKLM "Software\Microsoft\Windows\CurrentVersion\Uninstall\Blender" "DisplayName" "Blender"
-  WriteRegStr   HKLM "Software\Microsoft\Windows\CurrentVersion\Uninstall\Blender" "Publisher" "Blender Foundation"
-  WriteRegStr   HKLM "Software\Microsoft\Windows\CurrentVersion\Uninstall\Blender" "URLInfoAbout" "http://www.blender.org/"
-  WriteRegStr   HKLM "Software\Microsoft\Windows\CurrentVersion\Uninstall\Blender" "DisplayVersion" "[VERSION]"
-  WriteRegStr   HKLM "Software\Microsoft\Windows\CurrentVersion\Uninstall\Blender" "DisplayIcon" "$INSTDIR\blender.exe"
-  WriteRegStr   HKLM "Software\Microsoft\Windows\CurrentVersion\Uninstall\Blender" "UninstallString" '"$INSTDIR\uninstall.exe"'
-  WriteRegDWORD HKLM "Software\Microsoft\Windows\CurrentVersion\Uninstall\Blender" "NoModify" 1
-  WriteRegDWORD HKLM "Software\Microsoft\Windows\CurrentVersion\Uninstall\Blender" "NoRepair " 1
-  WriteUninstaller "uninstall.exe"
-
-SectionEnd
-
-Section "Add Start Menu Shortcuts" StartMenu
-  SetShellVarContext all
-  CreateDirectory "$SMPROGRAMS\Blender Foundation\Blender\"
-  CreateShortCut "$SMPROGRAMS\Blender Foundation\Blender\Uninstall.lnk" "$INSTDIR\uninstall.exe" "" "$INSTDIR\uninstall.exe" 0
-  CreateShortCut "$SMPROGRAMS\Blender Foundation\Blender\Blender.lnk" "$INSTDIR\Blender.exe" "" "$INSTDIR\blender.exe" 0
-  CreateShortCut "$SMPROGRAMS\Blender Foundation\Blender\Readme.lnk" "$INSTDIR\readme.html" "" "" 0
-  CreateShortCut "$SMPROGRAMS\Blender Foundation\Blender\Copyright.lnk" "$INSTDIR\Copyright.txt" "" "$INSTDIR\copyright.txt" 0
-  CreateShortCut "$SMPROGRAMS\Blender Foundation\Blender\GPL-license.lnk" "$INSTDIR\GPL-license.txt" "" "$INSTDIR\GPL-license.txt" 0
-  System::Call 'shell32.dll::SHChangeNotify(i, i, i, i) v (0x08000000, 0, 0, 0)' ; refresh icons
-SectionEnd
-
-Section "Add Desktop Shortcut" DesktopShortcut
-  CreateShortCut "$DESKTOP\Blender.lnk" "$INSTDIR\blender.exe" "" "$INSTDIR\blender.exe" 0
-  System::Call 'shell32.dll::SHChangeNotify(i, i, i, i) v (0x08000000, 0, 0, 0)' ; refresh icons
-SectionEnd
-
-Section "Open .blend files with Blender" BlendRegister
-  ExecWait '"$INSTDIR\blender.exe" -r'
-SectionEnd
-
-UninstallText "This will uninstall Blender [VERSION], and all installed files. Hit 'Uninstall' to continue."
-
-Section "Uninstall"
-  ; Remove registry keys
-  ${If} ${RunningX64}
-    SetRegView 64
-  ${EndIf}
-  
-  ReadRegStr $SHORTVERSION HKLM "SOFTWARE\BlenderFoundation" "ShortVersion"
-  DeleteRegKey HKLM "SOFTWARE\Microsoft\Windows\CurrentVersion\Uninstall\Blender"
-  DeleteRegKey HKLM "SOFTWARE\BlenderFoundation"
-  DeleteRegKey HKCR ".blend"
-  DeleteRegKey HKCR "blendfile"
-  DeleteRegKey HKCR "CLSID\{D45F043D-F17F-4e8a-8435-70971D9FA46D}"
-  SetShellVarContext all
- 
-  ; Remove files
-  [DELROOTDIRCONTS]
-  [DELDATAFILES]
-  [DELDATADIRS]
-
-  Delete "$INSTDIR\uninstall.exe"
-
-  ${If} $REMOVECONFIG == "1"
-    RMDir /r "$BLENDERCONFIG\$SHORTVERSION"
-  ${Endif}
-
-  ; Remove install directory if it's empty
-  RMDir $INSTDIR
-  ; Remove shortcuts
-  Delete "$SMPROGRAMS\Blender Foundation\Blender\*.*"
-  Delete "$DESKTOP\Blender.lnk"
-  ; Remove all link related directories and files
-  RMDir "$SMPROGRAMS\Blender Foundation\Blender"
-  RMDir "$SMPROGRAMS\Blender Foundation"
-  
-  System::Call 'shell32.dll::SHChangeNotify(i, i, i, i) v (0x08000000, 0, 0, 0)' ; Refresh icons
-SectionEnd
-
-!insertmacro MUI_FUNCTION_DESCRIPTION_BEGIN
-  !insertmacro MUI_DESCRIPTION_TEXT ${InstallFiles} $(DESC_InstallFiles)
-  !insertmacro MUI_DESCRIPTION_TEXT ${StartMenu} $(DESC_StartMenu)
-  !insertmacro MUI_DESCRIPTION_TEXT ${DesktopShortcut} $(DESC_DesktopShortcut)
-  !insertmacro MUI_DESCRIPTION_TEXT ${BlendRegister} $(DESC_BlendRegister)
-!insertmacro MUI_FUNCTION_DESCRIPTION_END
-
diff --git a/release/windows/installer/01.installer.bmp b/release/windows/installer/01.installer.bmp
deleted file mode 100644
index 10fb01454a4..00000000000
--- a/release/windows/installer/01.installer.bmp
+++ /dev/null
diff --git a/source/blender/CMakeLists.txt b/source/blender/CMakeLists.txt
index 76442048594..203543b0ef0 100644
--- a/source/blender/CMakeLists.txt
+++ b/source/blender/CMakeLists.txt
@@ -153,3 +153,7 @@ endif()
 if(WITH_ALEMBIC)
   add_subdirectory(alembic)
 endif()
+
+if(WIN32)
+  add_subdirectory(blendthumb)
+endif()
diff --git a/source/blender/alembic/ABC_alembic.h b/source/blender/alembic/ABC_alembic.h
index 653382017d6..696e0ff1810 100644
--- a/source/blender/alembic/ABC_alembic.h
+++ b/source/blender/alembic/ABC_alembic.h
@@ -117,6 +117,12 @@ struct Mesh *ABC_read_mesh(struct CacheReader *reader,
                            const char **err_str,
                            int flags);
 
+bool ABC_mesh_topology_changed(struct CacheReader *reader,
+                               struct Object *ob,
+                               struct Mesh *existing_mesh,
+                               const float time,
+                               const char **err_str);
+
 void CacheReader_incref(struct CacheReader *reader);
 void CacheReader_free(struct CacheReader *reader);
 
diff --git a/source/blender/alembic/intern/abc_mball.h b/source/blender/alembic/intern/abc_mball.h
index 770fa6fdf8c..4be977fa8f5 100644
--- a/source/blender/alembic/intern/abc_mball.h
+++ b/source/blender/alembic/intern/abc_mball.h
@@ -24,9 +24,7 @@
 #include "abc_object.h"
 #include "abc_mesh.h"
 
-struct EvaluationContext;
 struct Main;
-struct MetaBall;
 struct Object;
 
 /* AbcMBallWriter converts the metaballs to meshes at every frame,
diff --git a/source/blender/alembic/intern/abc_mesh.cc b/source/blender/alembic/intern/abc_mesh.cc
index 6647ca83bd6..8728303923a 100644
--- a/source/blender/alembic/intern/abc_mesh.cc
+++ b/source/blender/alembic/intern/abc_mesh.cc
@@ -1131,6 +1131,31 @@ bool AbcMeshReader::accepts_object_type(
   return true;
 }
 
+bool AbcMeshReader::topology_changed(Mesh *existing_mesh, const ISampleSelector &sample_sel)
+{
+  IPolyMeshSchema::Sample sample;
+  try {
+    sample = m_schema.getValue(sample_sel);
+  }
+  catch (Alembic::Util::Exception &ex) {
+    printf("Alembic: error reading mesh sample for '%s/%s' at time %f: %s\n",
+           m_iobject.getFullName().c_str(),
+           m_schema.getName().c_str(),
+           sample_sel.getRequestedTime(),
+           ex.what());
+    // A similar error in read_mesh() would just return existing_mesh.
+    return false;
+  }
+
+  const P3fArraySamplePtr &positions = sample.getPositions();
+  const Alembic::Abc::Int32ArraySamplePtr &face_indices = sample.getFaceIndices();
+  const Alembic::Abc::Int32ArraySamplePtr &face_counts = sample.getFaceCounts();
+
+  return positions->size() != existing_mesh->totvert ||
+         face_counts->size() != existing_mesh->totpoly ||
+         face_indices->size() != existing_mesh->totloop;
+}
+
 Mesh *AbcMeshReader::read_mesh(Mesh *existing_mesh,
                                const ISampleSelector &sample_sel,
                                int read_flag,
@@ -1162,10 +1187,7 @@ Mesh *AbcMeshReader::read_mesh(Mesh *existing_mesh,
   ImportSettings settings;
   settings.read_flag |= read_flag;
 
-  bool topology_changed = positions->size() != existing_mesh->totvert ||
-                          face_counts->size() != existing_mesh->totpoly ||
-                          face_indices->size() != existing_mesh->totloop;
-  if (topology_changed) {
+  if (topology_changed(existing_mesh, sample_sel)) {
     new_mesh = BKE_mesh_new_nomain_from_template(
         existing_mesh, positions->size(), 0, 0, face_indices->size(), face_counts->size());
 
@@ -1314,7 +1336,7 @@ static void read_subd_sample(const std::string &iobject_full_name,
 
   if ((settings->read_flag & MOD_MESHSEQ_READ_POLY) != 0) {
     /* Alembic's 'SubD' scheme is used to store subdivision surfaces, i.e. the pre-subdivision
-     * mesh. Currently we don't add a subdivison modifier when we load such data. This code is
+     * mesh. Currently we don't add a subdivision modifier when we load such data. This code is
      * assuming that the subdivided surface should be smooth, and sets a flag that will eventually
      * mark all polygons as such. */
     abc_mesh_data.poly_flag_smooth = true;
diff --git a/source/blender/alembic/intern/abc_mesh.h b/source/blender/alembic/intern/abc_mesh.h
index 859ab121eb6..c15fb948e03 100644
--- a/source/blender/alembic/intern/abc_mesh.h
+++ b/source/blender/alembic/intern/abc_mesh.h
@@ -100,16 +100,18 @@ class AbcMeshReader : public AbcObjectReader {
  public:
   AbcMeshReader(const Alembic::Abc::IObject &object, ImportSettings &settings);
 
-  bool valid() const;
+  bool valid() const override;
   bool accepts_object_type(const Alembic::AbcCoreAbstract::ObjectHeader &alembic_header,
                            const Object *const ob,
-                           const char **err_str) const;
-  void readObjectData(Main *bmain, const Alembic::Abc::ISampleSelector &sample_sel);
+                           const char **err_str) const override;
+  void readObjectData(Main *bmain, const Alembic::Abc::ISampleSelector &sample_sel) override;
 
   struct Mesh *read_mesh(struct Mesh *existing_mesh,
                          const Alembic::Abc::ISampleSelector &sample_sel,
                          int read_flag,
-                         const char **err_str);
+                         const char **err_str) override;
+  bool topology_changed(Mesh *existing_mesh,
+                        const Alembic::Abc::ISampleSelector &sample_sel) override;
 
  private:
   void readFaceSetsSample(Main *bmain,
diff --git a/source/blender/alembic/intern/abc_object.cc b/source/blender/alembic/intern/abc_object.cc
index f863fe4fee7..ebebbc0da1e 100644
--- a/source/blender/alembic/intern/abc_object.cc
+++ b/source/blender/alembic/intern/abc_object.cc
@@ -246,6 +246,14 @@ struct Mesh *AbcObjectReader::read_mesh(struct Mesh *existing_mesh,
   return existing_mesh;
 }
 
+bool AbcObjectReader::topology_changed(Mesh * /*existing_mesh*/,
+                                       const Alembic::Abc::ISampleSelector & /*sample_sel*/)
+{
+  /* The default implementation of read_mesh() just returns the original mesh, so never changes the
+   * topology. */
+  return false;
+}
+
 void AbcObjectReader::setupObjectTransform(const float time)
 {
   bool is_constant = false;
diff --git a/source/blender/alembic/intern/abc_object.h b/source/blender/alembic/intern/abc_object.h
index 537f24ab7d6..efe78b9017c 100644
--- a/source/blender/alembic/intern/abc_object.h
+++ b/source/blender/alembic/intern/abc_object.h
@@ -190,6 +190,8 @@ class AbcObjectReader {
                                  const Alembic::Abc::ISampleSelector &sample_sel,
                                  int read_flag,
                                  const char **err_str);
+  virtual bool topology_changed(Mesh *existing_mesh,
+                                const Alembic::Abc::ISampleSelector &sample_sel);
 
   /** Reads the object matrix and sets up an object transform if animated. */
   void setupObjectTransform(const float time);
diff --git a/source/blender/alembic/intern/abc_util.h b/source/blender/alembic/intern/abc_util.h
index f7f662212fd..5eb0ed70599 100644
--- a/source/blender/alembic/intern/abc_util.h
+++ b/source/blender/alembic/intern/abc_util.h
@@ -44,7 +44,6 @@ using Alembic::Abc::chrono_t;
 class AbcObjectReader;
 struct ImportSettings;
 
-struct Base;
 struct ID;
 struct Object;
 
diff --git a/source/blender/alembic/intern/alembic_capi.cc b/source/blender/alembic/intern/alembic_capi.cc
index d63b310fef7..dc525e0c46c 100644
--- a/source/blender/alembic/intern/alembic_capi.cc
+++ b/source/blender/alembic/intern/alembic_capi.cc
@@ -937,12 +937,7 @@ void ABC_get_transform(CacheReader *reader, float r_mat[4][4], float time, float
 
 /* ************************************************************************** */
 
-Mesh *ABC_read_mesh(CacheReader *reader,
-                    Object *ob,
-                    Mesh *existing_mesh,
-                    const float time,
-                    const char **err_str,
-                    int read_flag)
+static AbcObjectReader *get_abc_reader(CacheReader *reader, Object *ob, const char **err_str)
 {
   AbcObjectReader *abc_reader = reinterpret_cast<AbcObjectReader *>(reader);
   IObject iobject = abc_reader->iobject();
@@ -958,12 +953,44 @@ Mesh *ABC_read_mesh(CacheReader *reader,
     return NULL;
   }
 
+  return abc_reader;
+}
+
+static ISampleSelector sample_selector_for_time(float time)
+{
   /* kFloorIndex is used to be compatible with non-interpolating
    * properties; they use the floor. */
-  ISampleSelector sample_sel(time, ISampleSelector::kFloorIndex);
+  return ISampleSelector(time, ISampleSelector::kFloorIndex);
+}
+
+Mesh *ABC_read_mesh(CacheReader *reader,
+                    Object *ob,
+                    Mesh *existing_mesh,
+                    const float time,
+                    const char **err_str,
+                    int read_flag)
+{
+  AbcObjectReader *abc_reader = get_abc_reader(reader, ob, err_str);
+  if (abc_reader == NULL) {
+    return NULL;
+  }
+
+  ISampleSelector sample_sel = sample_selector_for_time(time);
   return abc_reader->read_mesh(existing_mesh, sample_sel, read_flag, err_str);
 }
 
+bool ABC_mesh_topology_changed(
+    CacheReader *reader, Object *ob, Mesh *existing_mesh, const float time, const char **err_str)
+{
+  AbcObjectReader *abc_reader = get_abc_reader(reader, ob, err_str);
+  if (abc_reader == NULL) {
+    return false;
+  }
+
+  ISampleSelector sample_sel = sample_selector_for_time(time);
+  return abc_reader->topology_changed(existing_mesh, sample_sel);
+}
+
 /* ************************************************************************** */
 
 void CacheReader_free(CacheReader *reader)
diff --git a/release/windows/blendthumb/CMakeLists.txt b/source/blender/blendthumb/CMakeLists.txt
index 45c583d1421..6c786dfcc7a 100644
--- a/release/windows/blendthumb/CMakeLists.txt
+++ b/source/blender/blendthumb/CMakeLists.txt
@@ -19,13 +19,8 @@
 # ***** END GPL LICENSE BLOCK *****
 
 #-----------------------------------------------------------------------------
-cmake_minimum_required(VERSION 2.8)
-project(BlendThumb)
+include_directories(${ZLIB_INCLUDE_DIRS})
 
-#Bring the headers, such as Student.h into the project
-include_directories(${ZLIB_INCLUDE})
-
-#Can manually add the sources using the set command as follows:
 set(SRC
   src/BlenderThumb.cpp
   src/BlendThumb.def
@@ -34,5 +29,10 @@ set(SRC
 )
 
 add_library(BlendThumb SHARED ${SRC})
-target_link_libraries(BlendThumb ${ZLIB_LIBS})
-install (TARGETS BlendThumb DESTINATION bin)
+target_link_libraries(BlendThumb ${ZLIB_LIBRARIES})
+
+install(
+  FILES $<TARGET_FILE:BlendThumb>
+  COMPONENT Blender
+  DESTINATION "."
+)
diff --git a/release/windows/blendthumb/src/BlendThumb.def b/source/blender/blendthumb/src/BlendThumb.def
index 71f9236735f..71f9236735f 100644
--- a/release/windows/blendthumb/src/BlendThumb.def
+++ b/source/blender/blendthumb/src/BlendThumb.def
diff --git a/release/windows/blendthumb/src/BlendThumb.rc b/source/blender/blendthumb/src/BlendThumb.rc
index 5dfd416b0c5..5dfd416b0c5 100644
--- a/release/windows/blendthumb/src/BlendThumb.rc
+++ b/source/blender/blendthumb/src/BlendThumb.rc
diff --git a/source/blender/blendthumb/src/BlenderThumb.cpp b/source/blender/blendthumb/src/BlenderThumb.cpp
new file mode 100644
index 00000000000..94dae437260
--- /dev/null
+++ b/source/blender/blendthumb/src/BlenderThumb.cpp
@@ -0,0 +1,313 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+#include <shlwapi.h>
+#include <thumbcache.h>  // For IThumbnailProvider.
+#include <new>
+
+#pragma comment(lib, "shlwapi.lib")
+
+// this thumbnail provider implements IInitializeWithStream to enable being hosted
+// in an isolated process for robustness
+
+class CBlendThumb : public IInitializeWithStream, public IThumbnailProvider {
+ public:
+  CBlendThumb() : _cRef(1), _pStream(NULL)
+  {
+  }
+
+  virtual ~CBlendThumb()
+  {
+    if (_pStream) {
+      _pStream->Release();
+    }
+  }
+
+  // IUnknown
+  IFACEMETHODIMP QueryInterface(REFIID riid, void **ppv)
+  {
+    static const QITAB qit[] = {
+        QITABENT(CBlendThumb, IInitializeWithStream),
+        QITABENT(CBlendThumb, IThumbnailProvider),
+        {0},
+    };
+    return QISearch(this, qit, riid, ppv);
+  }
+
+  IFACEMETHODIMP_(ULONG) AddRef()
+  {
+    return InterlockedIncrement(&_cRef);
+  }
+
+  IFACEMETHODIMP_(ULONG) Release()
+  {
+    ULONG cRef = InterlockedDecrement(&_cRef);
+    if (!cRef) {
+      delete this;
+    }
+    return cRef;
+  }
+
+  // IInitializeWithStream
+  IFACEMETHODIMP Initialize(IStream *pStream, DWORD grfMode);
+
+  // IThumbnailProvider
+  IFACEMETHODIMP GetThumbnail(UINT cx, HBITMAP *phbmp, WTS_ALPHATYPE *pdwAlpha);
+
+ private:
+  long _cRef;
+  IStream *_pStream;  // provided during initialization.
+};
+
+HRESULT CBlendThumb_CreateInstance(REFIID riid, void **ppv)
+{
+  CBlendThumb *pNew = new (std::nothrow) CBlendThumb();
+  HRESULT hr = pNew ? S_OK : E_OUTOFMEMORY;
+  if (SUCCEEDED(hr)) {
+    hr = pNew->QueryInterface(riid, ppv);
+    pNew->Release();
+  }
+  return hr;
+}
+
+// IInitializeWithStream
+IFACEMETHODIMP CBlendThumb::Initialize(IStream *pStream, DWORD)
+{
+  HRESULT hr = E_UNEXPECTED;  // can only be inited once
+  if (_pStream == NULL) {
+    // take a reference to the stream if we have not been inited yet
+    hr = pStream->QueryInterface(&_pStream);
+  }
+  return hr;
+}
+
+#include <math.h>
+#include <zlib.h>
+#include "Wincodec.h"
+const unsigned char gzip_magic[3] = {0x1f, 0x8b, 0x08};
+
+// IThumbnailProvider
+IFACEMETHODIMP CBlendThumb::GetThumbnail(UINT cx, HBITMAP *phbmp, WTS_ALPHATYPE *pdwAlpha)
+{
+  ULONG BytesRead;
+  HRESULT hr = S_FALSE;
+  LARGE_INTEGER SeekPos;
+
+  // Compressed?
+  unsigned char in_magic[3];
+  _pStream->Read(&in_magic, 3, &BytesRead);
+  bool gzipped = true;
+  for (int i = 0; i < 3; i++)
+    if (in_magic[i] != gzip_magic[i]) {
+      gzipped = false;
+      break;
+    }
+
+  if (gzipped) {
+    // Zlib inflate
+    z_stream stream;
+    stream.zalloc = Z_NULL;
+    stream.zfree = Z_NULL;
+    stream.opaque = Z_NULL;
+
+    // Get compressed file length
+    SeekPos.QuadPart = 0;
+    _pStream->Seek(SeekPos, STREAM_SEEK_END, NULL);
+
+    // Get compressed and uncompressed size
+    uLong source_size;
+    uLongf dest_size;
+    // SeekPos.QuadPart = -4; // last 4 bytes define size of uncompressed file
+    // ULARGE_INTEGER Tell;
+    //_pStream->Seek(SeekPos,STREAM_SEEK_END,&Tell);
+    // source_size = (uLong)Tell.QuadPart + 4; // src
+    //_pStream->Read(&dest_size,4,&BytesRead); // dest
+    dest_size = 1024 * 70;  // thumbnail is currently always inside the first 65KB...if it moves or
+                            // enlargens this line will have to change or go!
+    source_size = (uLong)max(SeekPos.QuadPart, dest_size);  // for safety, assume no compression
+
+    // Input
+    Bytef *src = new Bytef[source_size];
+    stream.next_in = (Bytef *)src;
+    stream.avail_in = (uInt)source_size;
+
+    // Output
+    Bytef *dest = new Bytef[dest_size];
+    stream.next_out = (Bytef *)dest;
+    stream.avail_out = dest_size;
+
+    // IStream to src
+    SeekPos.QuadPart = 0;
+    _pStream->Seek(SeekPos, STREAM_SEEK_SET, NULL);
+    _pStream->Read(src, source_size, &BytesRead);
+
+    // Do the inflation
+    int err;
+    err = inflateInit2(&stream, 16);  // 16 means "gzip"...nice!
+    err = inflate(&stream, Z_FINISH);
+    err = inflateEnd(&stream);
+
+    // Replace the IStream, which is read-only
+    _pStream->Release();
+    _pStream = SHCreateMemStream(dest, dest_size);
+
+    delete[] src;
+    delete[] dest;
+  }
+
+  // Blender version, early out if sub 2.5
+  SeekPos.QuadPart = 9;
+  _pStream->Seek(SeekPos, STREAM_SEEK_SET, NULL);
+  char version[4];
+  version[3] = '\0';
+  _pStream->Read(&version, 3, &BytesRead);
+  if (BytesRead != 3)
+    return E_UNEXPECTED;
+  int iVersion = atoi(version);
+  if (iVersion < 250)
+    return S_FALSE;
+
+  // 32 or 64 bit blend?
+  SeekPos.QuadPart = 7;
+  _pStream->Seek(SeekPos, STREAM_SEEK_SET, NULL);
+
+  char _PointerSize;
+  _pStream->Read(&_PointerSize, 1, &BytesRead);
+
+  int PointerSize = _PointerSize == '_' ? 4 : 8;
+  int HeaderSize = 16 + PointerSize;
+
+  // Find and read thumbnail ("TEST") block
+  SeekPos.QuadPart = 12;
+  _pStream->Seek(SeekPos, STREAM_SEEK_SET, NULL);
+  int BlockOffset = 12;
+  while (_pStream) {
+    // Scan current block
+    char BlockName[5];
+    BlockName[4] = '\0';
+    int BlockSize = 0;
+
+    if (_pStream->Read(BlockName, 4, &BytesRead) == S_OK &&
+        _pStream->Read((void *)&BlockSize, 4, &BytesRead) == S_OK) {
+      if (strcmp(BlockName, "TEST") != 0) {
+        SeekPos.QuadPart = BlockOffset += HeaderSize + BlockSize;
+        _pStream->Seek(SeekPos, STREAM_SEEK_SET, NULL);
+        continue;
+      }
+    }
+    else
+      break;  // eof
+
+    // Found the block
+    SeekPos.QuadPart = BlockOffset + HeaderSize;
+    _pStream->Seek(SeekPos, STREAM_SEEK_SET, NULL);
+
+    int width, height;
+    _pStream->Read((char *)&width, 4, &BytesRead);
+    _pStream->Read((char *)&height, 4, &BytesRead);
+    BlockSize -= 8;
+
+    // Isolate RGBA data
+    char *pRGBA = new char[BlockSize];
+    _pStream->Read(pRGBA, BlockSize, &BytesRead);
+
+    if (BytesRead != (ULONG)BlockSize)
+      return E_UNEXPECTED;
+
+    // Convert to BGRA for Windows
+    for (int i = 0; i < BlockSize; i += 4) {
+#define RED_BYTE pRGBA[i]
+#define BLUE_BYTE pRGBA[i + 2]
+
+      char red = RED_BYTE;
+      RED_BYTE = BLUE_BYTE;
+      BLUE_BYTE = red;
+    }
+
+    // Flip vertically (Blender stores it upside-down)
+    unsigned int LineSize = width * 4;
+    char *FlippedImage = new char[BlockSize];
+    for (int i = 0; i < height; i++) {
+      if (0 != memcpy_s(&FlippedImage[(height - i - 1) * LineSize],
+                        LineSize,
+                        &pRGBA[i * LineSize],
+                        LineSize))
+        return E_UNEXPECTED;
+    }
+    delete[] pRGBA;
+    pRGBA = FlippedImage;
+
+    // Create image
+    *phbmp = CreateBitmap(width, height, 1, 32, pRGBA);
+    if (!*phbmp)
+      return E_FAIL;
+    *pdwAlpha = WTSAT_ARGB;  // it's actually BGRA, not sure why this works
+
+    // Scale down if required
+    if ((unsigned)width > cx || (unsigned)height > cx) {
+      float scale = 1.0f / (max(width, height) / (float)cx);
+      LONG NewWidth = (LONG)(width * scale);
+      LONG NewHeight = (LONG)(height * scale);
+
+#ifdef _DEBUG
+#  if 0
+			MessageBox(0,"Attach now","Debugging",MB_OK);
+#  endif
+#endif
+      IWICImagingFactory *pImgFac;
+      hr = CoCreateInstance(
+          CLSID_WICImagingFactory, NULL, CLSCTX_INPROC_SERVER, IID_PPV_ARGS(&pImgFac));
+
+      IWICBitmap *WICBmp;
+      hr = pImgFac->CreateBitmapFromHBITMAP(*phbmp, 0, WICBitmapUseAlpha, &WICBmp);
+
+      BITMAPINFO bmi = {};
+      bmi.bmiHeader.biSize = sizeof(bmi.bmiHeader);
+      bmi.bmiHeader.biWidth = NewWidth;
+      bmi.bmiHeader.biHeight = -NewHeight;
+      bmi.bmiHeader.biPlanes = 1;
+      bmi.bmiHeader.biBitCount = 32;
+      bmi.bmiHeader.biCompression = BI_RGB;
+
+      BYTE *pBits;
+      HBITMAP ResizedHBmp = CreateDIBSection(NULL, &bmi, DIB_RGB_COLORS, (void **)&pBits, NULL, 0);
+      hr = ResizedHBmp ? S_OK : E_OUTOFMEMORY;
+      if (SUCCEEDED(hr)) {
+        IWICBitmapScaler *pIScaler;
+        hr = pImgFac->CreateBitmapScaler(&pIScaler);
+        hr = pIScaler->Initialize(WICBmp, NewWidth, NewHeight, WICBitmapInterpolationModeFant);
+
+        WICRect rect = {0, 0, NewWidth, NewHeight};
+        hr = pIScaler->CopyPixels(&rect, NewWidth * 4, NewWidth * NewHeight * 4, pBits);
+
+        if (SUCCEEDED(hr)) {
+          DeleteObject(*phbmp);
+          *phbmp = ResizedHBmp;
+        }
+        else
+          DeleteObject(ResizedHBmp);
+
+        pIScaler->Release();
+      }
+      WICBmp->Release();
+      pImgFac->Release();
+    }
+    else
+      hr = S_OK;
+    break;
+  }
+  return hr;
+}
diff --git a/source/blender/blendthumb/src/Dll.cpp b/source/blender/blendthumb/src/Dll.cpp
new file mode 100644
index 00000000000..5a6fb4e692c
--- /dev/null
+++ b/source/blender/blendthumb/src/Dll.cpp
@@ -0,0 +1,280 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+#include <objbase.h>
+#include <shlwapi.h>
+#include <thumbcache.h>  // For IThumbnailProvider.
+#include <shlobj.h>      // For SHChangeNotify
+#include <new>
+
+extern HRESULT CBlendThumb_CreateInstance(REFIID riid, void **ppv);
+
+#define SZ_CLSID_BLENDTHUMBHANDLER L"{D45F043D-F17F-4e8a-8435-70971D9FA46D}"
+#define SZ_BLENDTHUMBHANDLER L"Blender Thumbnail Handler"
+const CLSID CLSID_BlendThumbHandler = {
+    0xd45f043d, 0xf17f, 0x4e8a, {0x84, 0x35, 0x70, 0x97, 0x1d, 0x9f, 0xa4, 0x6d}};
+
+typedef HRESULT (*PFNCREATEINSTANCE)(REFIID riid, void **ppvObject);
+struct CLASS_OBJECT_INIT {
+  const CLSID *pClsid;
+  PFNCREATEINSTANCE pfnCreate;
+};
+
+// add classes supported by this module here
+const CLASS_OBJECT_INIT c_rgClassObjectInit[] = {
+    {&CLSID_BlendThumbHandler, CBlendThumb_CreateInstance}};
+
+long g_cRefModule = 0;
+
+// Handle the DLL's module
+HINSTANCE g_hInst = NULL;
+
+// Standard DLL functions
+STDAPI_(BOOL) DllMain(HINSTANCE hInstance, DWORD dwReason, void *)
+{
+  if (dwReason == DLL_PROCESS_ATTACH) {
+    g_hInst = hInstance;
+    DisableThreadLibraryCalls(hInstance);
+  }
+  return TRUE;
+}
+
+STDAPI DllCanUnloadNow()
+{
+  // Only allow the DLL to be unloaded after all outstanding references have been released
+  return (g_cRefModule == 0) ? S_OK : S_FALSE;
+}
+
+void DllAddRef()
+{
+  InterlockedIncrement(&g_cRefModule);
+}
+
+void DllRelease()
+{
+  InterlockedDecrement(&g_cRefModule);
+}
+
+class CClassFactory : public IClassFactory {
+ public:
+  static HRESULT CreateInstance(REFCLSID clsid,
+                                const CLASS_OBJECT_INIT *pClassObjectInits,
+                                size_t cClassObjectInits,
+                                REFIID riid,
+                                void **ppv)
+  {
+    *ppv = NULL;
+    HRESULT hr = CLASS_E_CLASSNOTAVAILABLE;
+    for (size_t i = 0; i < cClassObjectInits; i++) {
+      if (clsid == *pClassObjectInits[i].pClsid) {
+        IClassFactory *pClassFactory = new (std::nothrow)
+            CClassFactory(pClassObjectInits[i].pfnCreate);
+        hr = pClassFactory ? S_OK : E_OUTOFMEMORY;
+        if (SUCCEEDED(hr)) {
+          hr = pClassFactory->QueryInterface(riid, ppv);
+          pClassFactory->Release();
+        }
+        break;  // match found
+      }
+    }
+    return hr;
+  }
+
+  CClassFactory(PFNCREATEINSTANCE pfnCreate) : _cRef(1), _pfnCreate(pfnCreate)
+  {
+    DllAddRef();
+  }
+
+  // IUnknown
+  IFACEMETHODIMP QueryInterface(REFIID riid, void **ppv)
+  {
+    static const QITAB qit[] = {QITABENT(CClassFactory, IClassFactory), {0}};
+    return QISearch(this, qit, riid, ppv);
+  }
+
+  IFACEMETHODIMP_(ULONG) AddRef()
+  {
+    return InterlockedIncrement(&_cRef);
+  }
+
+  IFACEMETHODIMP_(ULONG) Release()
+  {
+    long cRef = InterlockedDecrement(&_cRef);
+    if (cRef == 0) {
+      delete this;
+    }
+    return cRef;
+  }
+
+  // IClassFactory
+  IFACEMETHODIMP CreateInstance(IUnknown *punkOuter, REFIID riid, void **ppv)
+  {
+    return punkOuter ? CLASS_E_NOAGGREGATION : _pfnCreate(riid, ppv);
+  }
+
+  IFACEMETHODIMP LockServer(BOOL fLock)
+  {
+    if (fLock) {
+      DllAddRef();
+    }
+    else {
+      DllRelease();
+    }
+    return S_OK;
+  }
+
+ private:
+  ~CClassFactory()
+  {
+    DllRelease();
+  }
+
+  long _cRef;
+  PFNCREATEINSTANCE _pfnCreate;
+};
+
+STDAPI DllGetClassObject(REFCLSID clsid, REFIID riid, void **ppv)
+{
+  return CClassFactory::CreateInstance(
+      clsid, c_rgClassObjectInit, ARRAYSIZE(c_rgClassObjectInit), riid, ppv);
+}
+
+// A struct to hold the information required for a registry entry
+
+struct REGISTRY_ENTRY {
+  HKEY hkeyRoot;
+  PCWSTR pszKeyName;
+  PCWSTR pszValueName;
+  DWORD dwValueType;
+  PCWSTR pszData;
+};
+
+// Creates a registry key (if needed) and sets the default value of the key
+
+HRESULT CreateRegKeyAndSetValue(const REGISTRY_ENTRY *pRegistryEntry)
+{
+  HKEY hKey;
+  HRESULT hr = HRESULT_FROM_WIN32(RegCreateKeyExW(pRegistryEntry->hkeyRoot,
+                                                  pRegistryEntry->pszKeyName,
+                                                  0,
+                                                  NULL,
+                                                  REG_OPTION_NON_VOLATILE,
+                                                  KEY_SET_VALUE,
+                                                  NULL,
+                                                  &hKey,
+                                                  NULL));
+  if (SUCCEEDED(hr)) {
+    // All this just to support REG_DWORD...
+    DWORD size;
+    DWORD data;
+    BYTE *lpData = (LPBYTE)pRegistryEntry->pszData;
+    switch (pRegistryEntry->dwValueType) {
+      case REG_SZ:
+        size = ((DWORD)wcslen(pRegistryEntry->pszData) + 1) * sizeof(WCHAR);
+        break;
+      case REG_DWORD:
+        size = sizeof(DWORD);
+        data = (DWORD)pRegistryEntry->pszData;
+        lpData = (BYTE *)&data;
+        break;
+      default:
+        return E_INVALIDARG;
+    }
+
+    hr = HRESULT_FROM_WIN32(RegSetValueExW(
+        hKey, pRegistryEntry->pszValueName, 0, pRegistryEntry->dwValueType, lpData, size));
+    RegCloseKey(hKey);
+  }
+  return hr;
+}
+
+//
+// Registers this COM server
+//
+STDAPI DllRegisterServer()
+{
+  HRESULT hr;
+
+  WCHAR szModuleName[MAX_PATH];
+
+  if (!GetModuleFileNameW(g_hInst, szModuleName, ARRAYSIZE(szModuleName))) {
+    hr = HRESULT_FROM_WIN32(GetLastError());
+  }
+  else {
+    const REGISTRY_ENTRY rgRegistryEntries[] = {
+        // RootKey			KeyName																			ValueName			ValueType		Data
+        {HKEY_CURRENT_USER,
+         L"Software\\Classes\\CLSID\\" SZ_CLSID_BLENDTHUMBHANDLER,
+         NULL,
+         REG_SZ,
+         SZ_BLENDTHUMBHANDLER},
+        {HKEY_CURRENT_USER,
+         L"Software\\Classes\\CLSID\\" SZ_CLSID_BLENDTHUMBHANDLER L"\\InProcServer32",
+         NULL,
+         REG_SZ,
+         szModuleName},
+        {HKEY_CURRENT_USER,
+         L"Software\\Classes\\CLSID\\" SZ_CLSID_BLENDTHUMBHANDLER L"\\InProcServer32",
+         L"ThreadingModel",
+         REG_SZ,
+         L"Apartment"},
+        {HKEY_CURRENT_USER,
+         L"Software\\Classes\\.blend\\",
+         L"Treatment",
+         REG_DWORD,
+         0},  // doesn't appear to do anything...
+        {HKEY_CURRENT_USER,
+         L"Software\\Classes\\.blend\\ShellEx\\{e357fccd-a995-4576-b01f-234630154e96}",
+         NULL,
+         REG_SZ,
+         SZ_CLSID_BLENDTHUMBHANDLER},
+    };
+
+    hr = S_OK;
+    for (int i = 0; i < ARRAYSIZE(rgRegistryEntries) && SUCCEEDED(hr); i++) {
+      hr = CreateRegKeyAndSetValue(&rgRegistryEntries[i]);
+    }
+  }
+  if (SUCCEEDED(hr)) {
+    // This tells the shell to invalidate the thumbnail cache.  This is important because any
+    // .blend files viewed before registering this handler would otherwise show cached blank
+    // thumbnails.
+    SHChangeNotify(SHCNE_ASSOCCHANGED, SHCNF_IDLIST, NULL, NULL);
+  }
+  return hr;
+}
+
+//
+// Unregisters this COM server
+//
+STDAPI DllUnregisterServer()
+{
+  HRESULT hr = S_OK;
+
+  const PCWSTR rgpszKeys[] = {
+      L"Software\\Classes\\CLSID\\" SZ_CLSID_BLENDTHUMBHANDLER,
+      L"Software\\Classes\\.blend\\ShellEx\\{e357fccd-a995-4576-b01f-234630154e96}"};
+
+  // Delete the registry entries
+  for (int i = 0; i < ARRAYSIZE(rgpszKeys) && SUCCEEDED(hr); i++) {
+    hr = HRESULT_FROM_WIN32(RegDeleteTreeW(HKEY_CURRENT_USER, rgpszKeys[i]));
+    if (hr == HRESULT_FROM_WIN32(ERROR_FILE_NOT_FOUND)) {
+      // If the registry entry has already been deleted, say S_OK.
+      hr = S_OK;
+    }
+  }
+  return hr;
+}
diff --git a/source/blender/blenfont/intern/blf_thumbs.c b/source/blender/blenfont/intern/blf_thumbs.c
index d6710b91539..4e44a85443d 100644
--- a/source/blender/blenfont/intern/blf_thumbs.c
+++ b/source/blender/blenfont/intern/blf_thumbs.c
@@ -109,7 +109,7 @@ void BLF_thumb_preview(const char *filename,
     font->pos[1] -= font->glyph_cache->ascender * 1.1f;
 
     /* We fallback to default english strings in case not enough chars are available in current
-     * font for given translated string (useful in non-latin i18n context, like chinese,
+     * font for given translated string (useful in non-latin i18n context, like Chinese,
      * since many fonts will then show nothing but ugly 'missing char' in their preview).
      * Does not handle all cases, but much better than nothing.
      */
diff --git a/source/blender/blenkernel/BKE_DerivedMesh.h b/source/blender/blenkernel/BKE_DerivedMesh.h
index d215f214a6d..2145a673e45 100644
--- a/source/blender/blenkernel/BKE_DerivedMesh.h
+++ b/source/blender/blenkernel/BKE_DerivedMesh.h
@@ -79,12 +79,10 @@ struct CustomData_MeshMasks;
 struct Depsgraph;
 struct MEdge;
 struct MFace;
-struct MLoopNorSpaceArray;
 struct MVert;
 struct Mesh;
 struct ModifierData;
 struct Object;
-struct PBVH;
 struct Scene;
 
 /*
@@ -150,23 +148,6 @@ struct DerivedMesh {
   /** Calculate vert and face normals */
   void (*calcNormals)(DerivedMesh *dm);
 
-  /** Calculate loop (split) normals */
-  void (*calcLoopNormals)(DerivedMesh *dm, const bool use_split_normals, const float split_angle);
-
-  /** Calculate loop (split) normals, and returns split loop normal spacearr. */
-  void (*calcLoopNormalsSpaceArray)(DerivedMesh *dm,
-                                    const bool use_split_normals,
-                                    const float split_angle,
-                                    struct MLoopNorSpaceArray *r_lnors_spacearr);
-
-  void (*calcLoopTangents)(DerivedMesh *dm,
-                           bool calc_active_tangent,
-                           const char (*tangent_names)[MAX_NAME],
-                           int tangent_names_count);
-
-  /** Recalculates mesh tessellation */
-  void (*recalcTessellation)(DerivedMesh *dm);
-
   /** Loop tessellation cache (WARNING! Only call inside threading-protected code!) */
   void (*recalcLoopTri)(DerivedMesh *dm);
   /** accessor functions */
@@ -261,51 +242,6 @@ struct DerivedMesh {
   DMFlagMat *(*getGridFlagMats)(DerivedMesh *dm);
   unsigned int **(*getGridHidden)(DerivedMesh *dm);
 
-  /** Iterate over each mapped vertex in the derived mesh, calling the
-   * given function with the original vert and the mapped vert's new
-   * coordinate and normal. For historical reasons the normal can be
-   * passed as a float or short array, only one should be non-NULL.
-   */
-  void (*foreachMappedVert)(DerivedMesh *dm,
-                            void (*func)(void *userData,
-                                         int index,
-                                         const float co[3],
-                                         const float no_f[3],
-                                         const short no_s[3]),
-                            void *userData,
-                            DMForeachFlag flag);
-
-  /** Iterate over each mapped edge in the derived mesh, calling the
-   * given function with the original edge and the mapped edge's new
-   * coordinates.
-   */
-  void (*foreachMappedEdge)(
-      DerivedMesh *dm,
-      void (*func)(void *userData, int index, const float v0co[3], const float v1co[3]),
-      void *userData);
-
-  /** Iterate over each mapped loop in the derived mesh, calling the given function
-   * with the original loop index and the mapped loops's new coordinate and normal.
-   */
-  void (*foreachMappedLoop)(DerivedMesh *dm,
-                            void (*func)(void *userData,
-                                         int vertex_index,
-                                         int face_index,
-                                         const float co[3],
-                                         const float no[3]),
-                            void *userData,
-                            DMForeachFlag flag);
-
-  /** Iterate over each mapped face in the derived mesh, calling the
-   * given function with the original face and the mapped face's (or
-   * faces') center and normal.
-   */
-  void (*foreachMappedFaceCenter)(
-      DerivedMesh *dm,
-      void (*func)(void *userData, int index, const float cent[3], const float no[3]),
-      void *userData,
-      DMForeachFlag flag);
-
   /** Iterate over all vertex points, calling DO_MINMAX with given args.
    *
    * Also called in Editmode
@@ -319,9 +255,6 @@ struct DerivedMesh {
   /** Get vertex location, undefined if index is not valid */
   void (*getVertCo)(DerivedMesh *dm, int index, float r_co[3]);
 
-  /** Fill the array (of length .getNumVerts()) with all vertex locations */
-  void (*getVertCos)(DerivedMesh *dm, float (*r_cos)[3]);
-
   /** Get smooth vertex normal, undefined if index is not valid */
   void (*getVertNo)(DerivedMesh *dm, int index, float r_no[3]);
   void (*getPolyNo)(DerivedMesh *dm, int index, float r_no[3]);
@@ -330,10 +263,6 @@ struct DerivedMesh {
    */
   const struct MeshElemMap *(*getPolyMap)(struct Object *ob, DerivedMesh *dm);
 
-  /** Get the BVH used for paint modes
-   */
-  struct PBVH *(*getPBVH)(struct Object *ob, DerivedMesh *dm);
-
   /** Release reference to the DerivedMesh. This function decides internally
    * if the DerivedMesh will be freed, or cached for later use. */
   void (*release)(DerivedMesh *dm);
@@ -372,14 +301,6 @@ void DM_from_template(DerivedMesh *dm,
  */
 int DM_release(DerivedMesh *dm);
 
-/** utility function to convert a DerivedMesh to a Mesh
- */
-void DM_to_mesh(DerivedMesh *dm,
-                struct Mesh *me,
-                struct Object *ob,
-                const struct CustomData_MeshMasks *mask,
-                bool take_ownership);
-
 void DM_set_only_copy(DerivedMesh *dm, const struct CustomData_MeshMasks *mask);
 
 /* adds a vertex/edge/face custom data layer to a DerivedMesh, optionally
@@ -446,8 +367,6 @@ DerivedMesh *mesh_create_derived_render(struct Depsgraph *depsgraph,
                                         const struct CustomData_MeshMasks *dataMask);
 
 /* same as above but wont use render settings */
-DerivedMesh *mesh_create_derived(struct Mesh *me, float (*vertCos)[3]);
-
 struct Mesh *editbmesh_get_eval_cage(struct Depsgraph *depsgraph,
                                      struct Scene *scene,
                                      struct Object *,
@@ -464,7 +383,7 @@ struct Mesh *editbmesh_get_eval_cage_and_final(struct Depsgraph *depsgraph,
                                                const struct CustomData_MeshMasks *dataMask,
                                                struct Mesh **r_final);
 
-float (*editbmesh_get_vertex_cos(struct BMEditMesh *em, int *r_numVerts))[3];
+float (*editbmesh_vert_coords_alloc(struct BMEditMesh *em, int *r_vert_len))[3];
 bool editbmesh_modifier_is_enabled(struct Scene *scene,
                                    struct ModifierData *md,
                                    bool has_prev_mesh);
diff --git a/source/blender/blenkernel/BKE_anim.h b/source/blender/blenkernel/BKE_anim.h
index dab76fb79f1..dce6f337885 100644
--- a/source/blender/blenkernel/BKE_anim.h
+++ b/source/blender/blenkernel/BKE_anim.h
@@ -24,7 +24,6 @@
  */
 struct Depsgraph;
 struct ListBase;
-struct Main;
 struct Object;
 struct ParticleSystem;
 struct Path;
diff --git a/source/blender/blenkernel/BKE_animsys.h b/source/blender/blenkernel/BKE_animsys.h
index fa8e294bdc1..a2a14011595 100644
--- a/source/blender/blenkernel/BKE_animsys.h
+++ b/source/blender/blenkernel/BKE_animsys.h
@@ -34,7 +34,6 @@ struct KeyingSet;
 struct ListBase;
 struct Main;
 struct NlaKeyframingContext;
-struct PathResolvedRNA;
 struct PointerRNA;
 struct PropertyRNA;
 struct ReportList;
diff --git a/source/blender/blenkernel/BKE_armature.h b/source/blender/blenkernel/BKE_armature.h
index 99c6bb405b5..3beae7a1f9d 100644
--- a/source/blender/blenkernel/BKE_armature.h
+++ b/source/blender/blenkernel/BKE_armature.h
@@ -25,7 +25,6 @@
 
 struct Bone;
 struct Depsgraph;
-struct GHash;
 struct ListBase;
 struct Main;
 struct Object;
@@ -78,6 +77,8 @@ struct bArmature *BKE_armature_copy(struct Main *bmain, const struct bArmature *
 void BKE_armature_copy_bone_transforms(struct bArmature *armature_dst,
                                        const struct bArmature *armature_src);
 
+void BKE_armature_transform(struct bArmature *arm, const float mat[4][4], const bool do_props);
+
 /* Bounding box. */
 struct BoundBox *BKE_armature_boundbox_get(struct Object *ob);
 
@@ -100,7 +101,7 @@ float distfactor_to_bone(
 
 void BKE_armature_where_is(struct bArmature *arm);
 void BKE_armature_where_is_bone(struct Bone *bone,
-                                struct Bone *prevbone,
+                                const struct Bone *bone_parent,
                                 const bool use_recursion);
 void BKE_pose_clear_pointers(struct bPose *pose);
 void BKE_pose_remap_bone_pointers(struct bArmature *armature, struct bPose *pose);
diff --git a/source/blender/blenkernel/BKE_blender_undo.h b/source/blender/blenkernel/BKE_blender_undo.h
index 6c2221326e9..7392d3947a2 100644
--- a/source/blender/blenkernel/BKE_blender_undo.h
+++ b/source/blender/blenkernel/BKE_blender_undo.h
@@ -26,7 +26,6 @@ extern "C" {
 
 struct Main;
 struct MemFileUndoData;
-struct Scene;
 struct bContext;
 
 #define BKE_UNDO_STR_MAX 64
diff --git a/source/blender/blenkernel/BKE_blender_version.h b/source/blender/blenkernel/BKE_blender_version.h
index 0c55ae8ee83..65086f57616 100644
--- a/source/blender/blenkernel/BKE_blender_version.h
+++ b/source/blender/blenkernel/BKE_blender_version.h
@@ -27,7 +27,7 @@
  * \note Use #STRINGIFY() rather than defining with quotes.
  */
 #define BLENDER_VERSION 281
-#define BLENDER_SUBVERSION 2
+#define BLENDER_SUBVERSION 5
 /** Several breakages with 280, e.g. collections vs layers. */
 #define BLENDER_MINVERSION 280
 #define BLENDER_MINSUBVERSION 0
@@ -37,6 +37,8 @@
 #define BLENDER_VERSION_CHAR
 /** alpha/beta/rc/release, docs use this. */
 #define BLENDER_VERSION_CYCLE alpha
+/** Optionally set to 1,2,... for example to to get alpha1 or rc2. */
+#define BLENDER_VERSION_CYCLE_NUMBER
 
 /** Defined in from blender.c */
 extern char versionstr[];
diff --git a/source/blender/blenkernel/BKE_brush.h b/source/blender/blenkernel/BKE_brush.h
index eeebfe1f2f6..26ab7f8ba0c 100644
--- a/source/blender/blenkernel/BKE_brush.h
+++ b/source/blender/blenkernel/BKE_brush.h
@@ -28,8 +28,6 @@ struct Brush;
 struct ImBuf;
 struct ImagePool;
 struct Main;
-struct Material;
-struct Paint;
 struct Scene;
 struct ToolSettings;
 struct UnifiedPaintSettings;
diff --git a/source/blender/blenkernel/BKE_bvhutils.h b/source/blender/blenkernel/BKE_bvhutils.h
index c88a64097bb..fdcb4066482 100644
--- a/source/blender/blenkernel/BKE_bvhutils.h
+++ b/source/blender/blenkernel/BKE_bvhutils.h
@@ -31,7 +31,6 @@
  */
 
 struct BMEditMesh;
-struct DerivedMesh;
 struct MFace;
 struct MVert;
 struct Mesh;
@@ -93,19 +92,18 @@ typedef struct BVHTreeFromMesh {
  *
  * free_bvhtree_from_mesh should be called when the tree is no longer needed.
  */
-BVHTree *bvhtree_from_editmesh_verts(BVHTreeFromEditMesh *data,
-                                     struct BMEditMesh *em,
-                                     float epsilon,
-                                     int tree_type,
-                                     int axis,
-                                     BVHCache **bvh_cache);
+BVHTree *bvhtree_from_editmesh_verts(
+    BVHTreeFromEditMesh *data, struct BMEditMesh *em, float epsilon, int tree_type, int axis);
+
 BVHTree *bvhtree_from_editmesh_verts_ex(BVHTreeFromEditMesh *data,
                                         struct BMEditMesh *em,
                                         const BLI_bitmap *mask,
                                         int verts_num_active,
                                         float epsilon,
                                         int tree_type,
-                                        int axis);
+                                        int axis,
+                                        const int bvh_cache_type,
+                                        BVHCache **bvh_cache);
 
 BVHTree *bvhtree_from_mesh_verts_ex(struct BVHTreeFromMesh *data,
                                     const struct MVert *vert,
@@ -115,21 +113,22 @@ BVHTree *bvhtree_from_mesh_verts_ex(struct BVHTreeFromMesh *data,
                                     int verts_num_active,
                                     float epsilon,
                                     int tree_type,
-                                    int axis);
-
-BVHTree *bvhtree_from_editmesh_edges(BVHTreeFromEditMesh *data,
-                                     struct BMEditMesh *em,
-                                     float epsilon,
-                                     int tree_type,
-                                     int axis,
-                                     BVHCache **bvh_cache);
+                                    int axis,
+                                    const int bvh_cache_type,
+                                    BVHCache **bvh_cache);
+
+BVHTree *bvhtree_from_editmesh_edges(
+    BVHTreeFromEditMesh *data, struct BMEditMesh *em, float epsilon, int tree_type, int axis);
+
 BVHTree *bvhtree_from_editmesh_edges_ex(BVHTreeFromEditMesh *data,
                                         struct BMEditMesh *em,
                                         const BLI_bitmap *edges_mask,
                                         int edges_num_active,
                                         float epsilon,
                                         int tree_type,
-                                        int axis);
+                                        int axis,
+                                        const int bvh_cache_type,
+                                        BVHCache **bvh_cache);
 
 BVHTree *bvhtree_from_mesh_edges_ex(struct BVHTreeFromMesh *data,
                                     const struct MVert *vert,
@@ -141,7 +140,9 @@ BVHTree *bvhtree_from_mesh_edges_ex(struct BVHTreeFromMesh *data,
                                     int edges_num_active,
                                     float epsilon,
                                     int tree_type,
-                                    int axis);
+                                    int axis,
+                                    const int bvh_cache_type,
+                                    BVHCache **bvh_cache);
 
 BVHTree *bvhtree_from_mesh_faces_ex(struct BVHTreeFromMesh *data,
                                     const struct MVert *vert,
@@ -153,14 +154,13 @@ BVHTree *bvhtree_from_mesh_faces_ex(struct BVHTreeFromMesh *data,
                                     int numFaces_active,
                                     float epsilon,
                                     int tree_type,
-                                    int axis);
+                                    int axis,
+                                    const int bvh_cache_type,
+                                    BVHCache **bvh_cache);
+
+BVHTree *bvhtree_from_editmesh_looptri(
+    BVHTreeFromEditMesh *data, struct BMEditMesh *em, float epsilon, int tree_type, int axis);
 
-BVHTree *bvhtree_from_editmesh_looptri(BVHTreeFromEditMesh *data,
-                                       struct BMEditMesh *em,
-                                       float epsilon,
-                                       int tree_type,
-                                       int axis,
-                                       BVHCache **bvhCache);
 BVHTree *bvhtree_from_editmesh_looptri_ex(BVHTreeFromEditMesh *data,
                                           struct BMEditMesh *em,
                                           const BLI_bitmap *mask,
@@ -168,7 +168,8 @@ BVHTree *bvhtree_from_editmesh_looptri_ex(BVHTreeFromEditMesh *data,
                                           float epsilon,
                                           int tree_type,
                                           int axis,
-                                          BVHCache **bvhCache);
+                                          const int bvh_cache_type,
+                                          BVHCache **bvh_cache);
 
 BVHTree *bvhtree_from_mesh_looptri_ex(struct BVHTreeFromMesh *data,
                                       const struct MVert *vert,
@@ -182,13 +183,21 @@ BVHTree *bvhtree_from_mesh_looptri_ex(struct BVHTreeFromMesh *data,
                                       int looptri_num_active,
                                       float epsilon,
                                       int tree_type,
-                                      int axis);
+                                      int axis,
+                                      const int bvh_cache_type,
+                                      BVHCache **bvh_cache);
 
 BVHTree *BKE_bvhtree_from_mesh_get(struct BVHTreeFromMesh *data,
                                    struct Mesh *mesh,
                                    const int type,
                                    const int tree_type);
 
+BVHTree *BKE_bvhtree_from_editmesh_get(BVHTreeFromEditMesh *data,
+                                       struct BMEditMesh *em,
+                                       const int tree_type,
+                                       const int bvh_cache_type,
+                                       BVHCache **bvh_cache);
+
 /**
  * Frees data allocated by a call to bvhtree_from_mesh_*.
  */
diff --git a/source/blender/blenkernel/BKE_cachefile.h b/source/blender/blenkernel/BKE_cachefile.h
index 257975e3c17..c8393fd3666 100644
--- a/source/blender/blenkernel/BKE_cachefile.h
+++ b/source/blender/blenkernel/BKE_cachefile.h
@@ -33,7 +33,6 @@ struct CacheReader;
 struct Depsgraph;
 struct Main;
 struct Object;
-struct Scene;
 
 void BKE_cachefiles_init(void);
 void BKE_cachefiles_exit(void);
diff --git a/source/blender/blenkernel/BKE_camera.h b/source/blender/blenkernel/BKE_camera.h
index caed4959eff..da80b3d5032 100644
--- a/source/blender/blenkernel/BKE_camera.h
+++ b/source/blender/blenkernel/BKE_camera.h
@@ -32,14 +32,12 @@ extern "C" {
 
 struct Camera;
 struct Depsgraph;
-struct GPUFXSettings;
 struct Main;
 struct Object;
 struct RegionView3D;
 struct RenderData;
 struct Scene;
 struct View3D;
-struct ViewLayer;
 struct rctf;
 
 /* Camera Datablock */
diff --git a/source/blender/blenkernel/BKE_cdderivedmesh.h b/source/blender/blenkernel/BKE_cdderivedmesh.h
index 08d53bf0e21..50af6c876b4 100644
--- a/source/blender/blenkernel/BKE_cdderivedmesh.h
+++ b/source/blender/blenkernel/BKE_cdderivedmesh.h
@@ -33,9 +33,7 @@
 struct BMEditMesh;
 struct CustomData_MeshMasks;
 struct DerivedMesh;
-struct MLoopNorSpaceArray;
 struct Mesh;
-struct Object;
 
 /* creates a new CDDerivedMesh */
 struct DerivedMesh *CDDM_new(int numVerts, int numEdges, int numFaces, int numLoops, int numPolys);
@@ -51,79 +49,16 @@ struct DerivedMesh *CDDM_from_mesh_ex(struct Mesh *mesh,
                                       eCDAllocType alloctype,
                                       const struct CustomData_MeshMasks *mask);
 
-struct DerivedMesh *CDDM_from_bmesh(struct BMesh *bm, const bool use_mdisps);
-
 /* creates a CDDerivedMesh from the given BMEditMesh */
-DerivedMesh *CDDM_from_editbmesh(struct BMEditMesh *em,
-                                 const bool use_mdisps,
-                                 const bool use_tessface);
-
-/* creates a CDDerivedMesh from the given curve object */
-struct DerivedMesh *CDDM_from_curve(struct Object *ob);
-
-/* creates a CDDerivedMesh from the given curve object and specified dispbase */
-/* useful for OrcoDM creation for curves with constructive modifiers */
-DerivedMesh *CDDM_from_curve_displist(struct Object *ob, struct ListBase *dispbase);
+DerivedMesh *CDDM_from_editbmesh(struct BMEditMesh *em, const bool use_mdisps);
 
 /* Copies the given DerivedMesh with verts, faces & edges stored as
  * custom element data.
  */
 struct DerivedMesh *CDDM_copy(struct DerivedMesh *dm);
 
-/* creates a CDDerivedMesh with the same layer stack configuration as the
- * given DerivedMesh and containing the requested numbers of elements.
- * elements are initialized to all zeros
- */
-struct DerivedMesh *CDDM_from_template_ex(struct DerivedMesh *source,
-                                          int numVerts,
-                                          int numEdges,
-                                          int numFaces,
-                                          int numLoops,
-                                          int numPolys,
-                                          const struct CustomData_MeshMasks *mask);
-struct DerivedMesh *CDDM_from_template(struct DerivedMesh *source,
-                                       int numVerts,
-                                       int numEdges,
-                                       int numFaces,
-                                       int numLoops,
-                                       int numPolys);
-
-/* applies vertex coordinates or normals to a CDDerivedMesh. if the MVert
- * layer is a referenced layer, it will be duplicate to not overwrite the
- * original
- */
-void CDDM_apply_vert_coords(struct DerivedMesh *cddm, float (*vertCoords)[3]);
-void CDDM_apply_vert_normals(struct DerivedMesh *cddm, short (*vertNormals)[3]);
-
-/* recalculates vertex and face normals for a CDDerivedMesh
- */
-void CDDM_calc_normals_mapping_ex(struct DerivedMesh *dm, const bool only_face_normals);
-void CDDM_calc_normals_mapping(struct DerivedMesh *dm);
-void CDDM_calc_normals(struct DerivedMesh *dm);
-
-void CDDM_calc_loop_normals(struct DerivedMesh *dm,
-                            const bool use_split_normals,
-                            const float split_angle);
-void CDDM_calc_loop_normals_spacearr(struct DerivedMesh *dm,
-                                     const bool use_split_normals,
-                                     const float split_angle,
-                                     struct MLoopNorSpaceArray *r_lnors_spacearr);
-
-/* reconstitute face triangulation */
-void CDDM_recalc_tessellation(struct DerivedMesh *dm);
-void CDDM_recalc_tessellation_ex(struct DerivedMesh *dm, const bool do_face_nor_cpy);
-
 void CDDM_recalc_looptri(struct DerivedMesh *dm);
 
-/* lowers the number of vertices/edges/faces in a CDDerivedMesh
- * the layer data stays the same size
- */
-void CDDM_lower_num_verts(struct DerivedMesh *dm, int numVerts);
-void CDDM_lower_num_edges(struct DerivedMesh *dm, int numEdges);
-void CDDM_lower_num_loops(struct DerivedMesh *dm, int numLoops);
-void CDDM_lower_num_polys(struct DerivedMesh *dm, int numPolys);
-void CDDM_lower_num_tessfaces(DerivedMesh *dm, int numTessFaces);
-
 /* vertex/edge/face access functions
  * should always succeed if index is within bounds
  * note these return pointers - any change modifies the internals of the mesh
@@ -145,13 +80,4 @@ struct MFace *CDDM_get_tessfaces(struct DerivedMesh *dm);
 struct MLoop *CDDM_get_loops(struct DerivedMesh *dm);
 struct MPoly *CDDM_get_polys(struct DerivedMesh *dm);
 
-/* Assigns news m*** layers to the cddm.  Note that you must handle
- * freeing the old ones yourself.  Also you must ensure dm->num****Data
- * is correct.*/
-void CDDM_set_mvert(struct DerivedMesh *dm, struct MVert *mvert);
-void CDDM_set_medge(struct DerivedMesh *dm, struct MEdge *medge);
-void CDDM_set_mface(struct DerivedMesh *dm, struct MFace *mface);
-void CDDM_set_mloop(struct DerivedMesh *dm, struct MLoop *mloop);
-void CDDM_set_mpoly(struct DerivedMesh *dm, struct MPoly *mpoly);
-
 #endif
diff --git a/source/blender/blenkernel/BKE_cloth.h b/source/blender/blenkernel/BKE_cloth.h
index 43df78c33e4..01f94c39215 100644
--- a/source/blender/blenkernel/BKE_cloth.h
+++ b/source/blender/blenkernel/BKE_cloth.h
@@ -29,7 +29,6 @@
 struct ClothModifierData;
 struct CollisionModifierData;
 struct Depsgraph;
-struct MFace;
 struct Mesh;
 struct Object;
 struct Scene;
diff --git a/source/blender/blenkernel/BKE_collection.h b/source/blender/blenkernel/BKE_collection.h
index 0ab101f065d..0d33d86ec16 100644
--- a/source/blender/blenkernel/BKE_collection.h
+++ b/source/blender/blenkernel/BKE_collection.h
@@ -35,8 +35,6 @@ extern "C" {
 struct BLI_Iterator;
 struct Base;
 struct Collection;
-struct Depsgraph;
-struct ID;
 struct Main;
 struct Object;
 struct Scene;
@@ -72,13 +70,9 @@ struct Collection *BKE_collection_duplicate(struct Main *bmain,
                                             const bool do_hierarchy,
                                             const bool do_objects,
                                             const bool do_obdata);
-struct Collection *BKE_collection_copy_master(struct Main *bmain,
-                                              struct Collection *collection,
-                                              const int flag);
 
 /* Master Collection for Scene */
 
-struct Collection *BKE_collection_master(const struct Scene *scene);
 struct Collection *BKE_collection_master_add(void);
 struct Scene *BKE_collection_master_scene_search(const struct Main *bmain,
                                                  const struct Collection *master_collection);
@@ -227,7 +221,7 @@ void BKE_scene_objects_iterator_end(struct BLI_Iterator *iter);
     bool is_scene_collection = (_scene) != NULL; \
 \
     if (_scene) { \
-      _instance_next = BKE_collection_master(_scene); \
+      _instance_next = _scene->master_collection; \
     } \
     else { \
       _instance_next = (_bmain)->collections.first; \
diff --git a/source/blender/blenkernel/BKE_collision.h b/source/blender/blenkernel/BKE_collision.h
index 291502a0c28..5d7a5094eb5 100644
--- a/source/blender/blenkernel/BKE_collision.h
+++ b/source/blender/blenkernel/BKE_collision.h
@@ -37,11 +37,9 @@
 struct Collection;
 struct CollisionModifierData;
 struct Depsgraph;
-struct MFace;
 struct MVert;
 struct MVertTri;
 struct Object;
-struct Scene;
 
 ////////////////////////////////////////
 // used for collisions in collision.c
diff --git a/source/blender/blenkernel/BKE_context.h b/source/blender/blenkernel/BKE_context.h
index 755a155653b..105f8e82343 100644
--- a/source/blender/blenkernel/BKE_context.h
+++ b/source/blender/blenkernel/BKE_context.h
@@ -33,7 +33,6 @@ extern "C" {
 
 struct ARegion;
 struct Base;
-struct Brush;
 struct CacheFile;
 struct Collection;
 struct Depsgraph;
@@ -59,7 +58,6 @@ struct Text;
 struct ToolSettings;
 struct View3D;
 struct ViewLayer;
-struct ViewRender;
 struct bGPDframe;
 struct bGPDlayer;
 struct bGPdata;
@@ -98,7 +96,7 @@ typedef struct bContextStore {
 
 /* for the context's rna mode enum
  * keep aligned with data_mode_strings in context.c */
-enum eContextObjectMode {
+typedef enum eContextObjectMode {
   CTX_MODE_EDIT_MESH = 0,
   CTX_MODE_EDIT_CURVE,
   CTX_MODE_EDIT_SURFACE,
@@ -117,7 +115,7 @@ enum eContextObjectMode {
   CTX_MODE_EDIT_GPENCIL,
   CTX_MODE_SCULPT_GPENCIL,
   CTX_MODE_WEIGHT_GPENCIL,
-};
+} eContextObjectMode;
 #define CTX_MODE_NUM (CTX_MODE_WEIGHT_GPENCIL + 1)
 
 /* Context */
@@ -239,7 +237,7 @@ bool CTX_data_dir(const char *member);
 
 #define CTX_DATA_BEGIN_WITH_ID(C, Type, instance, member, Type_id, instance_id) \
   CTX_DATA_BEGIN (C, Type, instance, member) \
-    Type_id instance_id = ctx_link->ptr.id.data;
+    Type_id instance_id = (Type_id)ctx_link->ptr.owner_id;
 
 int ctx_data_list_count(const bContext *C, int (*func)(const bContext *, ListBase *));
 
diff --git a/source/blender/blenkernel/BKE_curve.h b/source/blender/blenkernel/BKE_curve.h
index 51ff4673aba..be49ca150b2 100644
--- a/source/blender/blenkernel/BKE_curve.h
+++ b/source/blender/blenkernel/BKE_curve.h
@@ -26,15 +26,12 @@
 struct BezTriple;
 struct Curve;
 struct Depsgraph;
-struct EditNurb;
 struct GHash;
-struct LinkNode;
 struct ListBase;
 struct Main;
 struct Nurb;
 struct Object;
 struct Path;
-struct Scene;
 struct TextBox;
 struct rctf;
 
@@ -124,11 +121,22 @@ void BKE_curve_nurb_vert_active_set(struct Curve *cu, const struct Nurb *nu, con
 bool BKE_curve_nurb_vert_active_get(struct Curve *cu, struct Nurb **r_nu, void **r_vert);
 void BKE_curve_nurb_vert_active_validate(struct Curve *cu);
 
-float (*BKE_curve_nurbs_vertexCos_get(struct ListBase *lb, int *r_numVerts))[3];
-void BK_curve_nurbs_vertexCos_apply(struct ListBase *lb, float (*vertexCos)[3]);
+float (*BKE_curve_nurbs_vert_coords_alloc(struct ListBase *lb, int *r_vert_len))[3];
+void BKE_curve_nurbs_vert_coords_get(struct ListBase *lb, float (*vert_coords)[3], int vert_len);
 
-float (*BKE_curve_nurbs_keyVertexCos_get(struct ListBase *lb, float *key))[3];
-void BKE_curve_nurbs_keyVertexTilts_apply(struct ListBase *lb, float *key);
+void BKE_curve_nurbs_vert_coords_apply_with_mat4(struct ListBase *lb,
+                                                 const float (*vert_coords)[3],
+                                                 const float mat[4][4],
+                                                 const bool constrain_2d);
+
+void BKE_curve_nurbs_vert_coords_apply(struct ListBase *lb,
+                                       const float (*vert_coords)[3],
+                                       const bool constrain_2d);
+
+float (*BKE_curve_nurbs_key_vert_coords_alloc(struct ListBase *lb,
+                                              float *key,
+                                              int *r_vert_len))[3];
+void BKE_curve_nurbs_key_vert_tilts_apply(struct ListBase *lb, float *key);
 
 void BKE_curve_editNurb_keyIndex_delCV(struct GHash *keyindex, const void *cv);
 void BKE_curve_editNurb_keyIndex_free(struct GHash **keyindex);
diff --git a/source/blender/blenkernel/BKE_displist.h b/source/blender/blenkernel/BKE_displist.h
index c57639a8193..cfacbfe0a21 100644
--- a/source/blender/blenkernel/BKE_displist.h
+++ b/source/blender/blenkernel/BKE_displist.h
@@ -49,9 +49,7 @@ enum {
 /* prototypes */
 
 struct Depsgraph;
-struct LinkNode;
 struct ListBase;
-struct Main;
 struct Mesh;
 struct Object;
 struct Scene;
diff --git a/source/blender/blenkernel/BKE_dynamicpaint.h b/source/blender/blenkernel/BKE_dynamicpaint.h
index c4f05d404ce..734bfbc019a 100644
--- a/source/blender/blenkernel/BKE_dynamicpaint.h
+++ b/source/blender/blenkernel/BKE_dynamicpaint.h
@@ -25,9 +25,7 @@ struct Depsgraph;
 struct DynamicPaintCanvasSettings;
 struct DynamicPaintModifierData;
 struct DynamicPaintRuntime;
-struct Main;
 struct Scene;
-struct ViewLayer;
 
 /* Actual surface point */
 typedef struct PaintSurfaceData {
diff --git a/source/blender/blenkernel/BKE_editmesh.h b/source/blender/blenkernel/BKE_editmesh.h
index b7280c702d2..062968eddfc 100644
--- a/source/blender/blenkernel/BKE_editmesh.h
+++ b/source/blender/blenkernel/BKE_editmesh.h
@@ -31,7 +31,6 @@
 struct BMLoop;
 struct BMesh;
 struct Depsgraph;
-struct DerivedMesh;
 struct EditMeshData;
 struct Mesh;
 struct MeshStatVis;
@@ -79,7 +78,7 @@ typedef struct BMEditMesh {
 } BMEditMesh;
 
 /* editmesh.c */
-void BKE_editmesh_tessface_calc(BMEditMesh *em);
+void BKE_editmesh_looptri_calc(BMEditMesh *em);
 BMEditMesh *BKE_editmesh_create(BMesh *bm, const bool do_tessellate);
 BMEditMesh *BKE_editmesh_copy(BMEditMesh *em);
 BMEditMesh *BKE_editmesh_from_object(struct Object *ob);
@@ -88,7 +87,7 @@ void BKE_editmesh_free(BMEditMesh *em);
 
 void BKE_editmesh_color_free(BMEditMesh *em);
 void BKE_editmesh_color_ensure(BMEditMesh *em, const char htype);
-float (*BKE_editmesh_vertexCos_get_orco(BMEditMesh *em, int *r_numVerts))[3];
+float (*BKE_editmesh_vert_coords_alloc_orco(BMEditMesh *em, int *r_vert_len))[3];
 void BKE_editmesh_lnorspace_update(BMEditMesh *em);
 void BKE_editmesh_ensure_autosmooth(BMEditMesh *em);
 
@@ -98,7 +97,7 @@ void BKE_editmesh_statvis_calc(BMEditMesh *em,
                                struct EditMeshData *emd,
                                const struct MeshStatVis *statvis);
 
-float (*BKE_editmesh_vertexCos_get(
-    struct Depsgraph *depsgraph, struct BMEditMesh *em, struct Scene *scene, int *r_numVerts))[3];
+float (*BKE_editmesh_vert_coords_alloc(
+    struct Depsgraph *depsgraph, struct BMEditMesh *em, struct Scene *scene, int *r_vert_len))[3];
 
 #endif /* __BKE_EDITMESH_H__ */
diff --git a/source/blender/blenkernel/BKE_fcurve.h b/source/blender/blenkernel/BKE_fcurve.h
index 4c1a115eb23..5be9a35b168 100644
--- a/source/blender/blenkernel/BKE_fcurve.h
+++ b/source/blender/blenkernel/BKE_fcurve.h
@@ -89,6 +89,9 @@ struct ChannelDriver *fcurve_copy_driver(const struct ChannelDriver *driver);
 
 void driver_variables_copy(struct ListBase *dst_list, const struct ListBase *src_list);
 
+void BKE_driver_target_matrix_to_rot_channels(
+    float mat[4][4], int auto_order, int rotation_mode, int channel, bool angles, float r_buf[4]);
+
 void driver_free_variable(struct ListBase *variables, struct DriverVar *dvar);
 void driver_free_variable_ex(struct ChannelDriver *driver, struct DriverVar *dvar);
 
diff --git a/source/blender/blenkernel/BKE_gpencil.h b/source/blender/blenkernel/BKE_gpencil.h
index c544c065a70..b4f636199f8 100644
--- a/source/blender/blenkernel/BKE_gpencil.h
+++ b/source/blender/blenkernel/BKE_gpencil.h
@@ -24,31 +24,40 @@
  * \ingroup bke
  */
 
-struct Scene;
-struct ArrayGpencilModifierData;
 struct BoundBox;
 struct Brush;
 struct CurveMapping;
 struct Depsgraph;
-struct GpencilModifierData;
-struct LatticeGpencilModifierData;
 struct ListBase;
 struct Main;
 struct Material;
 struct Object;
-struct SimplifyGpencilModifierData;
 struct ToolSettings;
 struct bDeformGroup;
 struct bGPDframe;
 struct bGPDlayer;
-struct bGPDpalette;
-struct bGPDpalettecolor;
 struct bGPDspoint;
 struct bGPDstroke;
 struct bGPdata;
 
 struct MDeformVert;
-struct MDeformWeight;
+
+#define GPENCIL_SIMPLIFY(scene) ((scene->r.simplify_gpencil & SIMPLIFY_GPENCIL_ENABLE))
+#define GPENCIL_SIMPLIFY_ONPLAY(playing) \
+  (((playing == true) && (scene->r.simplify_gpencil & SIMPLIFY_GPENCIL_ON_PLAY)) || \
+   ((scene->r.simplify_gpencil & SIMPLIFY_GPENCIL_ON_PLAY) == 0))
+#define GPENCIL_SIMPLIFY_FILL(scene, playing) \
+  ((GPENCIL_SIMPLIFY_ONPLAY(playing) && (GPENCIL_SIMPLIFY(scene)) && \
+    (scene->r.simplify_gpencil & SIMPLIFY_GPENCIL_FILL)))
+#define GPENCIL_SIMPLIFY_MODIF(scene, playing) \
+  ((GPENCIL_SIMPLIFY_ONPLAY(playing) && (GPENCIL_SIMPLIFY(scene)) && \
+    (scene->r.simplify_gpencil & SIMPLIFY_GPENCIL_MODIFIER)))
+#define GPENCIL_SIMPLIFY_FX(scene, playing) \
+  ((GPENCIL_SIMPLIFY_ONPLAY(playing) && (GPENCIL_SIMPLIFY(scene)) && \
+    (scene->r.simplify_gpencil & SIMPLIFY_GPENCIL_FX)))
+#define GPENCIL_SIMPLIFY_BLEND(scene, playing) \
+  ((GPENCIL_SIMPLIFY_ONPLAY(playing) && (GPENCIL_SIMPLIFY(scene)) && \
+    (scene->r.simplify_gpencil & SIMPLIFY_GPENCIL_BLEND)))
 
 struct GPUBatch;
 struct GPUVertBuf;
@@ -134,7 +143,7 @@ void BKE_gpencil_free_stroke(struct bGPDstroke *gps);
 bool BKE_gpencil_free_strokes(struct bGPDframe *gpf);
 void BKE_gpencil_free_frames(struct bGPDlayer *gpl);
 void BKE_gpencil_free_layers(struct ListBase *list);
-bool BKE_gpencil_free_frame_runtime_data(struct bGPDframe *derived_gpf);
+bool BKE_gpencil_free_frame_runtime_data(struct bGPDframe *gpf_eval);
 void BKE_gpencil_free(struct bGPdata *gpd, bool free_all);
 
 void BKE_gpencil_batch_cache_dirty_tag(struct bGPdata *gpd);
@@ -220,13 +229,14 @@ struct bGPDlayer *BKE_gpencil_layer_get_index(struct bGPdata *gpd,
 struct bGPDlayer *BKE_gpencil_layer_getactive(struct bGPdata *gpd);
 void BKE_gpencil_layer_setactive(struct bGPdata *gpd, struct bGPDlayer *active);
 void BKE_gpencil_layer_delete(struct bGPdata *gpd, struct bGPDlayer *gpl);
+void BKE_gpencil_layer_autolock_set(struct bGPdata *gpd);
 
 /* Brush */
 struct Material *BKE_gpencil_brush_material_get(struct Brush *brush);
 void BKE_gpencil_brush_material_set(struct Brush *brush, struct Material *material);
 
 /* Object */
-struct Material *BKE_gpencil_object_material_ensure_active(struct Main *bmain, struct Object *ob);
+struct Material *BKE_gpencil_object_material_ensure_active(struct Object *ob);
 struct Material *BKE_gpencil_object_material_ensure_from_brush(struct Main *bmain,
                                                                struct Object *ob,
                                                                struct Brush *brush);
@@ -250,8 +260,7 @@ struct Material *BKE_gpencil_object_material_ensure_from_active_input_toolsettin
 struct Material *BKE_gpencil_object_material_ensure_from_active_input_brush(struct Main *bmain,
                                                                             struct Object *ob,
                                                                             struct Brush *brush);
-struct Material *BKE_gpencil_object_material_ensure_from_active_input_material(struct Main *bmain,
-                                                                               struct Object *ob);
+struct Material *BKE_gpencil_object_material_ensure_from_active_input_material(struct Object *ob);
 
 /* object boundbox */
 bool BKE_gpencil_data_minmax(const struct bGPdata *gpd, float r_min[3], float r_max[3]);
@@ -324,7 +333,8 @@ void BKE_gpencil_convert_curve(struct Main *bmain,
                                struct Object *ob_gp,
                                struct Object *ob_cu,
                                const bool gpencil_lines,
-                               const bool use_collections);
+                               const bool use_collections,
+                               const bool only_stroke);
 
 extern void (*BKE_gpencil_batch_cache_dirty_tag_cb)(struct bGPdata *gpd);
 extern void (*BKE_gpencil_batch_cache_free_cb)(struct bGPdata *gpd);
diff --git a/source/blender/blenkernel/BKE_gpencil_modifier.h b/source/blender/blenkernel/BKE_gpencil_modifier.h
index 373efadad55..36f38996c36 100644
--- a/source/blender/blenkernel/BKE_gpencil_modifier.h
+++ b/source/blender/blenkernel/BKE_gpencil_modifier.h
@@ -24,23 +24,16 @@
 #include "BLI_compiler_attrs.h"
 #include "BKE_customdata.h"
 
-struct BMEditMesh;
-struct DepsNodeHandle;
 struct Depsgraph;
-struct DerivedMesh;
 struct GpencilModifierData;
 struct ID;
 struct ListBase;
 struct Main;
-struct Mesh;
 struct ModifierUpdateDepsgraphContext;
 struct Object;
 struct Scene;
-struct ViewLayer;
-struct bArmature;
 /* NOTE: bakeModifier() called from UI:
  * needs to create new databloc-ks, hence the need for this. */
-struct bContext;
 struct bGPDframe;
 struct bGPDlayer;
 struct bGPDstroke;
diff --git a/source/blender/blenkernel/BKE_image_save.h b/source/blender/blenkernel/BKE_image_save.h
index 26c377a2574..8dfece944ff 100644
--- a/source/blender/blenkernel/BKE_image_save.h
+++ b/source/blender/blenkernel/BKE_image_save.h
@@ -31,7 +31,6 @@ extern "C" {
 
 struct Image;
 struct Main;
-struct RenderResult;
 struct ReportList;
 struct Scene;
 
diff --git a/source/blender/blenkernel/BKE_key.h b/source/blender/blenkernel/BKE_key.h
index 58ad3f6230d..477a22b07bb 100644
--- a/source/blender/blenkernel/BKE_key.h
+++ b/source/blender/blenkernel/BKE_key.h
@@ -31,7 +31,6 @@ struct ListBase;
 struct Main;
 struct Mesh;
 struct Object;
-struct WeightsArrayCache;
 
 /* Kernel prototypes */
 #ifdef __cplusplus
@@ -92,13 +91,17 @@ void BKE_keyblock_mesh_calc_normals(struct KeyBlock *kb,
                                     float (*r_polynors)[3],
                                     float (*r_loopnors)[3]);
 
-void BKE_keyblock_update_from_vertcos(struct Object *ob, struct KeyBlock *kb, float (*vertCos)[3]);
+void BKE_keyblock_update_from_vertcos(struct Object *ob,
+                                      struct KeyBlock *kb,
+                                      const float (*vertCos)[3]);
 void BKE_keyblock_convert_from_vertcos(struct Object *ob,
                                        struct KeyBlock *kb,
-                                       float (*vertCos)[3]);
+                                       const float (*vertCos)[3]);
 float (*BKE_keyblock_convert_to_vertcos(struct Object *ob, struct KeyBlock *kb))[3];
 
-void BKE_keyblock_update_from_offset(struct Object *ob, struct KeyBlock *kb, float (*ofs)[3]);
+void BKE_keyblock_update_from_offset(struct Object *ob,
+                                     struct KeyBlock *kb,
+                                     const float (*ofs)[3]);
 
 /* other management */
 bool BKE_keyblock_move(struct Object *ob, int org_index, int new_index);
diff --git a/source/blender/blenkernel/BKE_lattice.h b/source/blender/blenkernel/BKE_lattice.h
index c4173cd867e..a3befa6f3bb 100644
--- a/source/blender/blenkernel/BKE_lattice.h
+++ b/source/blender/blenkernel/BKE_lattice.h
@@ -58,7 +58,7 @@ void outside_lattice(struct Lattice *lt);
 
 void curve_deform_verts(struct Object *cuOb,
                         struct Object *target,
-                        float (*vertexCos)[3],
+                        float (*vert_coords)[3],
                         int numVerts,
                         struct MDeformVert *dvert,
                         const int defgrp_index,
@@ -73,14 +73,14 @@ void curve_deform_vector(struct Object *cuOb,
 void lattice_deform_verts(struct Object *laOb,
                           struct Object *target,
                           struct Mesh *mesh,
-                          float (*vertexCos)[3],
+                          float (*vert_coords)[3],
                           int numVerts,
                           const char *vgroup,
                           float influence);
 void armature_deform_verts(struct Object *armOb,
                            struct Object *target,
                            const struct Mesh *mesh,
-                           float (*vertexCos)[3],
+                           float (*vert_coords)[3],
                            float (*defMats)[3][3],
                            int numVerts,
                            int deformflag,
@@ -88,8 +88,12 @@ void armature_deform_verts(struct Object *armOb,
                            const char *defgrp_name,
                            struct bGPDstroke *gps);
 
-float (*BKE_lattice_vertexcos_get(struct Object *ob, int *r_numVerts))[3];
-void BKE_lattice_vertexcos_apply(struct Object *ob, float (*vertexCos)[3]);
+float (*BKE_lattice_vert_coords_alloc(const struct Lattice *lt, int *r_vert_len))[3];
+void BKE_lattice_vert_coords_get(const struct Lattice *lt, float (*vert_coords)[3]);
+void BKE_lattice_vert_coords_apply_with_mat4(struct Lattice *lt,
+                                             const float (*vert_coords)[3],
+                                             const float mat[4][4]);
+void BKE_lattice_vert_coords_apply(struct Lattice *lt, const float (*vert_coords)[3]);
 void BKE_lattice_modifiers_calc(struct Depsgraph *depsgraph,
                                 struct Scene *scene,
                                 struct Object *ob);
diff --git a/source/blender/blenkernel/BKE_layer.h b/source/blender/blenkernel/BKE_layer.h
index daac35a1196..eb65b7641e1 100644
--- a/source/blender/blenkernel/BKE_layer.h
+++ b/source/blender/blenkernel/BKE_layer.h
@@ -37,17 +37,12 @@ extern "C" {
 struct Base;
 struct Collection;
 struct Depsgraph;
-struct ID;
-struct IDProperty;
 struct LayerCollection;
-struct ListBase;
 struct Main;
 struct Object;
-struct RenderEngine;
 struct Scene;
 struct View3D;
 struct ViewLayer;
-struct WorkSpace;
 
 struct ViewLayer *BKE_view_layer_default_view(const struct Scene *scene);
 struct ViewLayer *BKE_view_layer_default_render(const struct Scene *scene);
diff --git a/source/blender/blenkernel/BKE_library.h b/source/blender/blenkernel/BKE_library.h
index c0ac71f9c96..261416dc025 100644
--- a/source/blender/blenkernel/BKE_library.h
+++ b/source/blender/blenkernel/BKE_library.h
@@ -46,17 +46,14 @@ extern "C" {
  * except in some specific cases requiring advanced (and potentially dangerous) handling.
  */
 
-struct BlendThumbnail;
 struct GHash;
 struct ID;
-struct ImBuf;
 struct Library;
 struct ListBase;
 struct Main;
 struct PointerRNA;
 struct PropertyRNA;
 struct bContext;
-struct wmWindowManager;
 
 size_t BKE_libblock_get_alloc_info(short type, const char **name);
 void *BKE_libblock_alloc_notest(short type) ATTR_WARN_UNUSED_RESULT;
@@ -117,10 +114,7 @@ enum {
                          LIB_ID_CREATE_NO_DEG_TAG | LIB_ID_COPY_NO_PREVIEW | LIB_ID_COPY_CACHES,
 };
 
-void BKE_libblock_copy_ex(struct Main *bmain,
-                          const struct ID *id,
-                          struct ID **r_newid,
-                          const int flag);
+void BKE_libblock_copy_ex(struct Main *bmain, const struct ID *id, struct ID **r_newid, int flag);
 void *BKE_libblock_copy(struct Main *bmain, const struct ID *id) ATTR_WARN_UNUSED_RESULT
     ATTR_NONNULL();
 /* Special version. sued by datablock localization. */
diff --git a/source/blender/blenkernel/BKE_library_override.h b/source/blender/blenkernel/BKE_library_override.h
index 5440b0ebe63..93b2355ce55 100644
--- a/source/blender/blenkernel/BKE_library_override.h
+++ b/source/blender/blenkernel/BKE_library_override.h
@@ -35,8 +35,8 @@ bool BKE_override_library_is_enabled(void);
 
 struct IDOverrideLibrary *BKE_override_library_init(struct ID *local_id, struct ID *reference_id);
 void BKE_override_library_copy(struct ID *dst_id, const struct ID *src_id);
-void BKE_override_library_clear(struct IDOverrideLibrary *override);
-void BKE_override_library_free(struct IDOverrideLibrary **override);
+void BKE_override_library_clear(struct IDOverrideLibrary *override, const bool do_id_user);
+void BKE_override_library_free(struct IDOverrideLibrary **override, const bool do_id_user);
 
 struct ID *BKE_override_library_create_from_id(struct Main *bmain, struct ID *reference_id);
 bool BKE_override_library_create_from_tag(struct Main *bmain);
diff --git a/source/blender/blenkernel/BKE_library_remap.h b/source/blender/blenkernel/BKE_library_remap.h
index d51bc5c2bce..784e1a570c5 100644
--- a/source/blender/blenkernel/BKE_library_remap.h
+++ b/source/blender/blenkernel/BKE_library_remap.h
@@ -25,6 +25,8 @@ extern "C" {
 
 #include "BLI_compiler_attrs.h"
 
+struct wmWindowManager;
+
 /* BKE_libblock_free, delete are declared in BKE_library.h for convenience. */
 
 /* Also IDRemap->flag. */
@@ -83,7 +85,7 @@ void BKE_libblock_relink_ex(struct Main *bmain,
                             void *idv,
                             void *old_idv,
                             void *new_idv,
-                            const bool us_min_never_null) ATTR_NONNULL(1, 2);
+                            const short remap_flags) ATTR_NONNULL(1, 2);
 
 void BKE_libblock_relink_to_newid(struct ID *id) ATTR_NONNULL();
 
diff --git a/source/blender/blenkernel/BKE_light.h b/source/blender/blenkernel/BKE_light.h
index 72bd5857526..e060f587afe 100644
--- a/source/blender/blenkernel/BKE_light.h
+++ b/source/blender/blenkernel/BKE_light.h
@@ -32,7 +32,6 @@ extern "C" {
 
 struct Light;
 struct Main;
-struct Scene;
 
 void BKE_light_init(struct Light *la);
 struct Light *BKE_light_add(struct Main *bmain, const char *name) ATTR_WARN_UNUSED_RESULT;
diff --git a/source/blender/blenkernel/BKE_linestyle.h b/source/blender/blenkernel/BKE_linestyle.h
index 79579b43446..b8950e5df88 100644
--- a/source/blender/blenkernel/BKE_linestyle.h
+++ b/source/blender/blenkernel/BKE_linestyle.h
@@ -38,7 +38,6 @@ extern "C" {
 
 struct ColorBand;
 struct Main;
-struct Object;
 struct ViewLayer;
 struct bContext;
 
diff --git a/source/blender/blenkernel/BKE_main.h b/source/blender/blenkernel/BKE_main.h
index 1c987d5eb8e..1b2e8bcbf42 100644
--- a/source/blender/blenkernel/BKE_main.h
+++ b/source/blender/blenkernel/BKE_main.h
@@ -40,7 +40,6 @@ extern "C" {
 
 struct BLI_mempool;
 struct BlendThumbnail;
-struct Depsgraph;
 struct GHash;
 struct GSet;
 struct ImBuf;
diff --git a/source/blender/blenkernel/BKE_mask.h b/source/blender/blenkernel/BKE_mask.h
index f87c73e35a2..47644ce82e9 100644
--- a/source/blender/blenkernel/BKE_mask.h
+++ b/source/blender/blenkernel/BKE_mask.h
@@ -38,7 +38,6 @@ struct MaskSplinePoint;
 struct MaskSplinePointUW;
 struct MovieClip;
 struct MovieClipUser;
-struct Scene;
 
 /* mask_ops.c */
 typedef enum {
diff --git a/source/blender/blenkernel/BKE_material.h b/source/blender/blenkernel/BKE_material.h
index 5bb69c7166e..7ff9c4e6376 100644
--- a/source/blender/blenkernel/BKE_material.h
+++ b/source/blender/blenkernel/BKE_material.h
@@ -38,6 +38,7 @@ struct bNode;
 /* materials */
 
 void init_def_material(void);
+void BKE_material_gpencil_default_free(void);
 void BKE_material_free(struct Material *ma);
 void test_object_materials(struct Main *bmain, struct Object *ob, struct ID *id);
 void test_all_objects_materials(struct Main *bmain, struct ID *id);
@@ -46,6 +47,7 @@ void BKE_material_resize_object(struct Main *bmain,
                                 const short totcol,
                                 bool do_id_user);
 void BKE_material_init(struct Material *ma);
+void BKE_material_gpencil_init(struct Material *ma);
 void BKE_material_remap_object(struct Object *ob, const unsigned int *remap);
 void BKE_material_remap_object_calc(struct Object *ob_dst,
                                     struct Object *ob_src,
@@ -92,6 +94,7 @@ bool BKE_object_material_slot_add(struct Main *bmain, struct Object *ob);
 bool BKE_object_material_slot_remove(struct Main *bmain, struct Object *ob);
 bool BKE_object_material_slot_used(struct ID *id, short actcol);
 
+struct Material *BKE_material_gpencil_get(struct Object *ob, short act);
 struct MaterialGPencilStyle *BKE_material_gpencil_settings_get(struct Object *ob, short act);
 
 void BKE_texpaint_slot_refresh_cache(struct Scene *scene, struct Material *ma);
@@ -104,9 +107,8 @@ void BKE_material_append_id(struct Main *bmain, struct ID *id, struct Material *
 struct Material *BKE_material_pop_id(struct Main *bmain,
                                      struct ID *id,
                                      /* index is an int because of RNA. */
-                                     int index,
-                                     bool update_data);
-void BKE_material_clear_id(struct Main *bmain, struct ID *id, bool update_data);
+                                     int index);
+void BKE_material_clear_id(struct Main *bmain, struct ID *id);
 /* rendering */
 
 void ramp_blend(int type, float r_col[3], const float fac, const float col[3]);
@@ -124,6 +126,7 @@ struct Depsgraph;
 void BKE_material_eval(struct Depsgraph *depsgraph, struct Material *material);
 
 extern struct Material defmaterial;
+extern struct Material defgpencil_material;
 
 #ifdef __cplusplus
 }
diff --git a/source/blender/blenkernel/BKE_mball.h b/source/blender/blenkernel/BKE_mball.h
index b582e88f9cb..5447fd00866 100644
--- a/source/blender/blenkernel/BKE_mball.h
+++ b/source/blender/blenkernel/BKE_mball.h
@@ -63,7 +63,7 @@ bool BKE_mball_minmax_ex(const struct MetaBall *mb,
 bool BKE_mball_minmax(const struct MetaBall *mb, float min[3], float max[3]);
 bool BKE_mball_center_median(const struct MetaBall *mb, float r_cent[3]);
 bool BKE_mball_center_bounds(const struct MetaBall *mb, float r_cent[3]);
-void BKE_mball_transform(struct MetaBall *mb, float mat[4][4], const bool do_props);
+void BKE_mball_transform(struct MetaBall *mb, const float mat[4][4], const bool do_props);
 void BKE_mball_translate(struct MetaBall *mb, const float offset[3]);
 
 struct MetaElem *BKE_mball_element_add(struct MetaBall *mb, const int type);
diff --git a/source/blender/blenkernel/BKE_mball_tessellate.h b/source/blender/blenkernel/BKE_mball_tessellate.h
index ab8fbc44caf..39740458f2f 100644
--- a/source/blender/blenkernel/BKE_mball_tessellate.h
+++ b/source/blender/blenkernel/BKE_mball_tessellate.h
@@ -20,7 +20,6 @@
  * \ingroup bke
  */
 struct Depsgraph;
-struct Main;
 struct Object;
 struct Scene;
 
diff --git a/source/blender/blenkernel/BKE_mesh.h b/source/blender/blenkernel/BKE_mesh.h
index 94d118bde36..12978e46327 100644
--- a/source/blender/blenkernel/BKE_mesh.h
+++ b/source/blender/blenkernel/BKE_mesh.h
@@ -53,13 +53,11 @@ struct MLoopTri;
 struct MLoopUV;
 struct MPoly;
 struct MVert;
-struct MVertTri;
 struct Main;
 struct MemArena;
 struct Mesh;
 struct ModifierData;
 struct Object;
-struct ReportList;
 struct Scene;
 
 #ifdef __cplusplus
@@ -123,6 +121,13 @@ struct Mesh *BKE_mesh_new_nomain_from_template(const struct Mesh *me_src,
                                                int tessface_len,
                                                int loops_len,
                                                int polys_len);
+struct Mesh *BKE_mesh_new_nomain_from_template_ex(const struct Mesh *me_src,
+                                                  int verts_len,
+                                                  int edges_len,
+                                                  int tessface_len,
+                                                  int loops_len,
+                                                  int polys_len,
+                                                  CustomData_MeshMasks mask);
 
 void BKE_mesh_eval_delete(struct Mesh *me_eval);
 
@@ -185,7 +190,7 @@ void BKE_mesh_material_index_remove(struct Mesh *me, short index);
 bool BKE_mesh_material_index_used(struct Mesh *me, short index);
 void BKE_mesh_material_index_clear(struct Mesh *me);
 void BKE_mesh_material_remap(struct Mesh *me, const unsigned int *remap, unsigned int remap_len);
-void BKE_mesh_smooth_flag_set(struct Object *meshOb, int enableSmooth);
+void BKE_mesh_smooth_flag_set(struct Mesh *me, const bool use_smooth);
 
 const char *BKE_mesh_cmp(struct Mesh *me1, struct Mesh *me2, float thresh);
 
@@ -198,18 +203,6 @@ void BKE_mesh_texspace_get_reference(
     struct Mesh *me, short **r_texflag, float **r_loc, float **r_rot, float **r_size);
 void BKE_mesh_texspace_copy_from_object(struct Mesh *me, struct Object *ob);
 
-bool BKE_mesh_uv_cdlayer_rename_index(struct Mesh *me,
-                                      const int loop_index,
-                                      const int face_index,
-                                      const char *new_name,
-                                      const bool do_tessface);
-bool BKE_mesh_uv_cdlayer_rename(struct Mesh *me,
-                                const char *old_name,
-                                const char *new_name,
-                                bool do_tessface);
-
-float (*BKE_mesh_vertexCos_get(const struct Mesh *me, int *r_numVerts))[3];
-
 void BKE_mesh_split_faces(struct Mesh *mesh, bool free_loop_normals);
 
 /* Create new mesh from the given object at its current state.
@@ -268,8 +261,14 @@ void BKE_mesh_mselect_active_set(struct Mesh *me, int index, int type);
 
 void BKE_mesh_count_selected_items(const struct Mesh *mesh, int r_count[3]);
 
-void BKE_mesh_apply_vert_coords(struct Mesh *mesh, float (*vertCoords)[3]);
-void BKE_mesh_apply_vert_normals(struct Mesh *mesh, short (*vertNormals)[3]);
+float (*BKE_mesh_vert_coords_alloc(const struct Mesh *mesh, int *r_vert_len))[3];
+void BKE_mesh_vert_coords_get(const struct Mesh *mesh, float (*vert_coords)[3]);
+
+void BKE_mesh_vert_coords_apply_with_mat4(struct Mesh *mesh,
+                                          const float (*vert_coords)[3],
+                                          const float mat[4][4]);
+void BKE_mesh_vert_coords_apply(struct Mesh *mesh, const float (*vert_coords)[3]);
+void BKE_mesh_vert_normals_apply(struct Mesh *mesh, const short (*vertNormals)[3]);
 
 /* *** mesh_evaluate.c *** */
 
@@ -309,11 +308,6 @@ void BKE_mesh_calc_normals_poly(struct MVert *mverts,
 void BKE_mesh_calc_normals(struct Mesh *me);
 void BKE_mesh_ensure_normals(struct Mesh *me);
 void BKE_mesh_ensure_normals_for_display(struct Mesh *mesh);
-void BKE_mesh_calc_normals_tessface(struct MVert *mverts,
-                                    int numVerts,
-                                    const struct MFace *mfaces,
-                                    int numFaces,
-                                    float (*r_faceNors)[3]);
 void BKE_mesh_calc_normals_looptri(struct MVert *mverts,
                                    int numVerts,
                                    const struct MLoop *mloop,
@@ -538,14 +532,14 @@ void BKE_mesh_tangent_loops_to_tessdata(struct CustomData *fdata,
                                         unsigned int (*loopindices)[4],
                                         const int num_faces,
                                         const char *layer_name);
-int BKE_mesh_recalc_tessellation(struct CustomData *fdata,
-                                 struct CustomData *ldata,
-                                 struct CustomData *pdata,
-                                 struct MVert *mvert,
-                                 int totface,
-                                 int totloop,
-                                 int totpoly,
-                                 const bool do_face_nor_copy);
+int BKE_mesh_tessface_calc_ex(struct CustomData *fdata,
+                              struct CustomData *ldata,
+                              struct CustomData *pdata,
+                              struct MVert *mvert,
+                              int totface,
+                              int totloop,
+                              int totpoly,
+                              const bool do_face_nor_copy);
 void BKE_mesh_recalc_looptri(const struct MLoop *mloop,
                              const struct MPoly *mpoly,
                              const struct MVert *mvert,
diff --git a/source/blender/blenkernel/BKE_mesh_iterators.h b/source/blender/blenkernel/BKE_mesh_iterators.h
index 28fd4b8bc28..b5b65944079 100644
--- a/source/blender/blenkernel/BKE_mesh_iterators.h
+++ b/source/blender/blenkernel/BKE_mesh_iterators.h
@@ -20,12 +20,6 @@
  * \ingroup bke
  */
 
-struct MEdge;
-struct MLoop;
-struct MLoopTri;
-struct MLoopUV;
-struct MPoly;
-struct MVert;
 struct Mesh;
 
 typedef enum MeshForeachFlag {
diff --git a/source/blender/blenkernel/BKE_mesh_runtime.h b/source/blender/blenkernel/BKE_mesh_runtime.h
index 93d952bcae5..4dc9ac988e7 100644
--- a/source/blender/blenkernel/BKE_mesh_runtime.h
+++ b/source/blender/blenkernel/BKE_mesh_runtime.h
@@ -27,7 +27,6 @@
 
 //#include "BKE_customdata.h"  /* for CustomDataMask */
 
-struct ColorBand;
 struct CustomData;
 struct CustomData_MeshMasks;
 struct Depsgraph;
diff --git a/source/blender/blenkernel/BKE_mesh_tangent.h b/source/blender/blenkernel/BKE_mesh_tangent.h
index bca825ce91c..4a86dfa3832 100644
--- a/source/blender/blenkernel/BKE_mesh_tangent.h
+++ b/source/blender/blenkernel/BKE_mesh_tangent.h
@@ -20,6 +20,8 @@
  * \ingroup bke
  */
 
+struct ReportList;
+
 void BKE_mesh_calc_loop_tangent_single_ex(const struct MVert *mverts,
                                           const int numVerts,
                                           const struct MLoop *mloops,
diff --git a/source/blender/blenkernel/BKE_modifier.h b/source/blender/blenkernel/BKE_modifier.h
index 4aab2a346e7..b2bc30c107c 100644
--- a/source/blender/blenkernel/BKE_modifier.h
+++ b/source/blender/blenkernel/BKE_modifier.h
@@ -35,7 +35,6 @@ struct Mesh;
 struct ModifierData;
 struct Object;
 struct Scene;
-struct ViewLayer;
 struct bArmature;
 
 typedef enum {
diff --git a/source/blender/blenkernel/BKE_node.h b/source/blender/blenkernel/BKE_node.h
index 9b2db5acd3b..cbfc8ffc7d0 100644
--- a/source/blender/blenkernel/BKE_node.h
+++ b/source/blender/blenkernel/BKE_node.h
@@ -59,7 +59,6 @@ struct RenderData;
 struct Scene;
 struct SpaceNode;
 struct Tex;
-struct ViewRender;
 struct bContext;
 struct bNode;
 struct bNodeExecContext;
@@ -385,6 +384,7 @@ void ntreeUserIncrefID(struct bNodeTree *ntree);
 void ntreeUserDecrefID(struct bNodeTree *ntree);
 
 struct bNodeTree *ntreeFromID(const struct ID *id);
+struct ID *BKE_node_tree_find_owner_ID(struct Main *bmain, struct bNodeTree *ntree);
 
 void ntreeMakeLocal(struct Main *bmain,
                     struct bNodeTree *ntree,
@@ -615,6 +615,7 @@ void nodeUpdateInternalLinks(struct bNodeTree *ntree, struct bNode *node);
 
 int nodeSocketIsHidden(struct bNodeSocket *sock);
 void ntreeTagUsedSockets(struct bNodeTree *ntree);
+void nodeSetSocketAvailability(struct bNodeSocket *sock, bool is_available);
 
 /* Node Clipboard */
 void BKE_node_clipboard_init(struct bNodeTree *ntree);
@@ -894,7 +895,7 @@ void BKE_nodetree_remove_layer_n(struct bNodeTree *ntree,
 #define SH_NODE_CURVE_RGB 111
 #define SH_NODE_CAMERA 114
 #define SH_NODE_MATH 115
-#define SH_NODE_VECT_MATH 116
+#define SH_NODE_VECTOR_MATH 116
 #define SH_NODE_SQUEEZE 117
 //#define SH_NODE_MATERIAL_EXT  118
 #define SH_NODE_INVERT 119
@@ -977,6 +978,8 @@ void BKE_nodetree_remove_layer_n(struct bNodeTree *ntree,
 #define SH_NODE_BSDF_HAIR_PRINCIPLED 701
 #define SH_NODE_MAP_RANGE 702
 #define SH_NODE_CLAMP 703
+#define SH_NODE_TEX_WHITE_NOISE 704
+#define SH_NODE_VOLUME_INFO 705
 
 /* custom defines options for Material node */
 #define SH_NODE_MAT_DIFF 1
diff --git a/source/blender/blenkernel/BKE_object.h b/source/blender/blenkernel/BKE_object.h
index f59bf3579be..cf4bce3a209 100644
--- a/source/blender/blenkernel/BKE_object.h
+++ b/source/blender/blenkernel/BKE_object.h
@@ -43,7 +43,6 @@ struct RegionView3D;
 struct RigidBodyWorld;
 struct Scene;
 struct ShaderFxData;
-struct SoftBody;
 struct View3D;
 struct ViewLayer;
 
diff --git a/source/blender/blenkernel/BKE_paint.h b/source/blender/blenkernel/BKE_paint.h
index 37667599488..ed02a34196f 100644
--- a/source/blender/blenkernel/BKE_paint.h
+++ b/source/blender/blenkernel/BKE_paint.h
@@ -32,7 +32,6 @@ struct Depsgraph;
 struct EnumPropertyItem;
 struct GridPaintMask;
 struct ImagePool;
-struct MFace;
 struct MLoop;
 struct MLoopTri;
 struct MVert;
@@ -47,7 +46,6 @@ struct Palette;
 struct PaletteColor;
 struct ReportList;
 struct Scene;
-struct Sculpt;
 struct StrokeCache;
 struct SubdivCCG;
 struct SubdivCCG;
@@ -259,8 +257,15 @@ typedef struct SculptSession {
 
   struct StrokeCache *cache;
 
+  /* Cursor data and active vertex for tools */
   int active_vertex_index;
 
+  float cursor_radius;
+  float cursor_location[3];
+  float cursor_normal[3];
+  float cursor_view_normal[3];
+  struct RegionView3D *rv3d;
+
   union {
     struct {
       struct SculptVertexPaintGeomMap gmap;
diff --git a/source/blender/blenkernel/BKE_particle.h b/source/blender/blenkernel/BKE_particle.h
index 164dbbbf482..e9b4c2b5a66 100644
--- a/source/blender/blenkernel/BKE_particle.h
+++ b/source/blender/blenkernel/BKE_particle.h
@@ -57,7 +57,6 @@ struct ModifierData;
 struct Object;
 struct RNG;
 struct Scene;
-struct ViewLayer;
 
 #define PARTICLE_COLLISION_MAX_COLLISIONS 10
 
diff --git a/source/blender/blenkernel/BKE_pbvh.h b/source/blender/blenkernel/BKE_pbvh.h
index 62544efad2c..f02d41d3c65 100644
--- a/source/blender/blenkernel/BKE_pbvh.h
+++ b/source/blender/blenkernel/BKE_pbvh.h
@@ -28,12 +28,10 @@
 
 struct BMLog;
 struct BMesh;
-struct CCGDerivedMesh;
 struct CCGElem;
 struct CCGKey;
 struct CustomData;
 struct DMFlagMat;
-struct GPUBatch;
 struct GPU_PBVH_Buffers;
 struct IsectRayPrecalc;
 struct MLoop;
@@ -134,8 +132,11 @@ bool BKE_pbvh_node_raycast(PBVH *bvh,
                            float (*origco)[3],
                            bool use_origco,
                            const float ray_start[3],
+                           const float ray_normal[3],
                            struct IsectRayPrecalc *isect_precalc,
-                           float *depth);
+                           float *depth,
+                           int *active_vertex_index,
+                           float *face_normal);
 
 bool BKE_pbvh_bmesh_node_raycast_detail(PBVHNode *node,
                                         const float ray_start[3],
@@ -278,9 +279,9 @@ float *BKE_pbvh_node_layer_disp_get(PBVH *pbvh, PBVHNode *node);
 void BKE_pbvh_node_layer_disp_free(PBVHNode *node);
 
 /* vertex deformer */
-float (*BKE_pbvh_get_vertCos(struct PBVH *pbvh))[3];
-void BKE_pbvh_apply_vertCos(struct PBVH *pbvh, float (*vertCos)[3], const int totvert);
-bool BKE_pbvh_isDeformed(struct PBVH *pbvh);
+float (*BKE_pbvh_vert_coords_alloc(struct PBVH *pbvh))[3];
+void BKE_pbvh_vert_coords_apply(struct PBVH *pbvh, const float (*vertCos)[3], const int totvert);
+bool BKE_pbvh_is_deformed(struct PBVH *pbvh);
 
 /* Vertex Iterator */
 
diff --git a/source/blender/blenkernel/BKE_scene.h b/source/blender/blenkernel/BKE_scene.h
index bcdf1fe48ca..7d564a232a7 100644
--- a/source/blender/blenkernel/BKE_scene.h
+++ b/source/blender/blenkernel/BKE_scene.h
@@ -38,8 +38,6 @@ struct TransformOrientation;
 struct UnitSettings;
 struct View3DCursor;
 struct ViewLayer;
-struct ViewRender;
-struct WorkSpace;
 
 typedef enum eSceneCopyMethod {
   SCE_COPY_NEW = 0,
diff --git a/source/blender/blenkernel/BKE_screen.h b/source/blender/blenkernel/BKE_screen.h
index 5d3e7ad5ec2..a2ce522a895 100644
--- a/source/blender/blenkernel/BKE_screen.h
+++ b/source/blender/blenkernel/BKE_screen.h
@@ -24,7 +24,6 @@
  */
 
 struct ARegion;
-struct GPUFXSettings;
 struct Header;
 struct ID;
 struct ListBase;
@@ -35,7 +34,6 @@ struct ScrArea;
 struct ScrAreaMap;
 struct ScrVert;
 struct SpaceType;
-struct TransformOrientation;
 struct View3D;
 struct View3DShading;
 struct WorkSpace;
diff --git a/source/blender/blenkernel/BKE_sequencer.h b/source/blender/blenkernel/BKE_sequencer.h
index a9e1dfb2392..e1bc16702d5 100644
--- a/source/blender/blenkernel/BKE_sequencer.h
+++ b/source/blender/blenkernel/BKE_sequencer.h
@@ -26,13 +26,11 @@
 
 struct Depsgraph;
 struct Editing;
-struct GPUFX;
 struct GPUOffScreen;
 struct GSet;
 struct ImBuf;
 struct Main;
 struct Mask;
-struct RenderEngineType;
 struct Scene;
 struct Sequence;
 struct SequenceModifierData;
diff --git a/source/blender/blenkernel/BKE_shader_fx.h b/source/blender/blenkernel/BKE_shader_fx.h
index e90dbfcf68d..9fe42a214e9 100644
--- a/source/blender/blenkernel/BKE_shader_fx.h
+++ b/source/blender/blenkernel/BKE_shader_fx.h
@@ -24,22 +24,11 @@
 #include "BLI_compiler_attrs.h"
 #include "BKE_customdata.h"
 
-struct DepsNodeHandle;
-struct Depsgraph;
-struct DerivedMesh;
 struct ID;
 struct ListBase;
-struct Main;
-struct Mesh;
 struct ModifierUpdateDepsgraphContext;
 struct Object;
-struct Scene;
 struct ShaderFxData;
-struct ViewLayer;
-struct bArmature;
-struct bGPDframe;
-struct bGPDlayer;
-struct bGPDstroke;
 
 #define SHADER_FX_ACTIVE(_fx, _is_render) \
   (((_fx->mode & eShaderFxMode_Realtime) && (_is_render == false)) || \
diff --git a/source/blender/blenkernel/BKE_shrinkwrap.h b/source/blender/blenkernel/BKE_shrinkwrap.h
index 8993654254e..9ec75c39fcf 100644
--- a/source/blender/blenkernel/BKE_shrinkwrap.h
+++ b/source/blender/blenkernel/BKE_shrinkwrap.h
@@ -41,7 +41,6 @@
 
 struct BVHTree;
 struct MDeformVert;
-struct MVert;
 struct Mesh;
 struct ModifierEvalContext;
 struct Object;
diff --git a/source/blender/blenkernel/BKE_studiolight.h b/source/blender/blenkernel/BKE_studiolight.h
index d6fff528348..108e93d9caa 100644
--- a/source/blender/blenkernel/BKE_studiolight.h
+++ b/source/blender/blenkernel/BKE_studiolight.h
@@ -87,6 +87,11 @@ enum StudioLightFlag {
   STUDIOLIGHT_RADIANCE_BUFFERS_CALCULATED = (1 << 11),
   STUDIOLIGHT_USER_DEFINED = (1 << 12),
   STUDIOLIGHT_UI_EXPANDED = (1 << 13),
+
+  STUDIOLIGHT_MATCAP_DIFFUSE_GPUTEXTURE = (1 << 14),
+  STUDIOLIGHT_MATCAP_SPECULAR_GPUTEXTURE = (1 << 15),
+  /* Is set for studio lights and matcaps with specular highlight pass. */
+  STUDIOLIGHT_SPECULAR_HIGHLIGHT_PASS = (1 << 16),
 };
 
 #define STUDIOLIGHT_FLAG_ALL (STUDIOLIGHT_INTERNAL | STUDIOLIGHT_EXTERNAL_FILE)
@@ -97,6 +102,11 @@ enum StudioLightFlag {
 
 typedef void StudioLightFreeFunction(struct StudioLight *, void *data);
 
+typedef struct StudioLightImage {
+  ImBuf *ibuf;
+  struct GPUTexture *gputexture;
+} StudioLightImage;
+
 typedef struct StudioLight {
   struct StudioLight *next, *prev;
 
@@ -112,6 +122,8 @@ typedef struct StudioLight {
   int icon_id_matcap_flipped;
   float spherical_harmonics_coefs[STUDIOLIGHT_SH_EFFECTIVE_COEFS_LEN][3];
   float light_direction[3];
+  StudioLightImage matcap_diffuse;
+  StudioLightImage matcap_specular;
   ImBuf *equirect_radiance_buffer;
   ImBuf *equirect_irradiance_buffer;
   ImBuf *radiance_cubemap_buffers[6];
diff --git a/source/blender/blenkernel/BKE_subdiv.h b/source/blender/blenkernel/BKE_subdiv.h
index dd139233ee3..700bf5139e0 100644
--- a/source/blender/blenkernel/BKE_subdiv.h
+++ b/source/blender/blenkernel/BKE_subdiv.h
@@ -29,12 +29,10 @@
 
 struct Mesh;
 struct MultiresModifierData;
-struct Object;
 struct OpenSubdiv_Converter;
 struct OpenSubdiv_Evaluator;
 struct OpenSubdiv_TopologyRefiner;
 struct Subdiv;
-struct SubdivToMeshSettings;
 
 typedef enum eSubdivVtxBoundaryInterpolation {
   /* Do not interpolate boundaries. */
diff --git a/source/blender/blenkernel/BKE_texture.h b/source/blender/blenkernel/BKE_texture.h
index c17db591981..39216e98e82 100644
--- a/source/blender/blenkernel/BKE_texture.h
+++ b/source/blender/blenkernel/BKE_texture.h
@@ -31,17 +31,13 @@ struct Brush;
 struct ColorBand;
 struct FreestyleLineStyle;
 struct ImagePool;
-struct Light;
 struct MTex;
 struct Main;
-struct Material;
 struct ParticleSettings;
 struct PointDensity;
 struct Tex;
 struct TexMapping;
 struct TexResult;
-struct World;
-struct bNode;
 
 /*  in ColorBand struct */
 #define MAXCOLORBAND 32
diff --git a/source/blender/blenkernel/BKE_tracking.h b/source/blender/blenkernel/BKE_tracking.h
index bf417403d43..c64d684de5a 100644
--- a/source/blender/blenkernel/BKE_tracking.h
+++ b/source/blender/blenkernel/BKE_tracking.h
@@ -25,7 +25,6 @@
  */
 
 struct Camera;
-struct Depsgraph;
 struct ImBuf;
 struct ListBase;
 struct MovieClipUser;
diff --git a/source/blender/blenkernel/BKE_workspace.h b/source/blender/blenkernel/BKE_workspace.h
index 133cf2d6cf5..ea8978b7374 100644
--- a/source/blender/blenkernel/BKE_workspace.h
+++ b/source/blender/blenkernel/BKE_workspace.h
@@ -24,9 +24,6 @@
 #include "BLI_compiler_attrs.h"
 
 struct Main;
-struct Scene;
-struct TransformOrientation;
-struct ViewLayer;
 struct bScreen;
 struct bToolRef;
 
diff --git a/source/blender/blenkernel/intern/DerivedMesh.c b/source/blender/blenkernel/intern/DerivedMesh.c
index 6267c095618..010e162f49d 100644
--- a/source/blender/blenkernel/intern/DerivedMesh.c
+++ b/source/blender/blenkernel/intern/DerivedMesh.c
@@ -87,11 +87,8 @@
 #  define ASSERT_IS_VALID_MESH(mesh)
 #endif
 
-static CLG_LogRef LOG = {"bke.derivedmesh"};
 static ThreadRWMutex loops_cache_lock = PTHREAD_RWLOCK_INITIALIZER;
 
-static void shapekey_layers_to_keyblocks(DerivedMesh *dm, Mesh *me, int actshape_uid);
-
 static void mesh_init_origspace(Mesh *mesh);
 
 /* -------------------------------------------------------------------- */
@@ -516,154 +513,6 @@ void DM_ensure_looptri_data(DerivedMesh *dm)
   }
 }
 
-void DM_to_mesh(
-    DerivedMesh *dm, Mesh *me, Object *ob, const CustomData_MeshMasks *mask, bool take_ownership)
-{
-  /* dm might depend on me, so we need to do everything with a local copy */
-  Mesh tmp = *me;
-  int totvert, totedge /*, totface */ /* UNUSED */, totloop, totpoly;
-  int did_shapekeys = 0;
-  eCDAllocType alloctype = CD_DUPLICATE;
-
-  if (take_ownership && dm->type == DM_TYPE_CDDM && dm->needsFree) {
-    bool has_any_referenced_layers = CustomData_has_referenced(&dm->vertData) ||
-                                     CustomData_has_referenced(&dm->edgeData) ||
-                                     CustomData_has_referenced(&dm->loopData) ||
-                                     CustomData_has_referenced(&dm->faceData) ||
-                                     CustomData_has_referenced(&dm->polyData);
-    if (!has_any_referenced_layers) {
-      alloctype = CD_ASSIGN;
-    }
-  }
-
-  CustomData_reset(&tmp.vdata);
-  CustomData_reset(&tmp.edata);
-  CustomData_reset(&tmp.fdata);
-  CustomData_reset(&tmp.ldata);
-  CustomData_reset(&tmp.pdata);
-
-  DM_ensure_normals(dm);
-
-  totvert = tmp.totvert = dm->getNumVerts(dm);
-  totedge = tmp.totedge = dm->getNumEdges(dm);
-  totloop = tmp.totloop = dm->getNumLoops(dm);
-  totpoly = tmp.totpoly = dm->getNumPolys(dm);
-  tmp.totface = 0;
-
-  CustomData_copy(&dm->vertData, &tmp.vdata, mask->vmask, alloctype, totvert);
-  CustomData_copy(&dm->edgeData, &tmp.edata, mask->emask, alloctype, totedge);
-  CustomData_copy(&dm->loopData, &tmp.ldata, mask->lmask, alloctype, totloop);
-  CustomData_copy(&dm->polyData, &tmp.pdata, mask->pmask, alloctype, totpoly);
-  tmp.cd_flag = dm->cd_flag;
-  tmp.runtime.deformed_only = dm->deformedOnly;
-
-  if (CustomData_has_layer(&dm->vertData, CD_SHAPEKEY)) {
-    KeyBlock *kb;
-    int uid;
-
-    if (ob) {
-      kb = BLI_findlink(&me->key->block, ob->shapenr - 1);
-      if (kb) {
-        uid = kb->uid;
-      }
-      else {
-        CLOG_ERROR(&LOG, "could not find active shapekey %d!", ob->shapenr - 1);
-        uid = INT_MAX;
-      }
-    }
-    else {
-      /* if no object, set to INT_MAX so we don't mess up any shapekey layers */
-      uid = INT_MAX;
-    }
-
-    shapekey_layers_to_keyblocks(dm, me, uid);
-    did_shapekeys = 1;
-  }
-
-  /* copy texture space */
-  if (ob) {
-    BKE_mesh_texspace_copy_from_object(&tmp, ob);
-  }
-
-  /* not all DerivedMeshes store their verts/edges/faces in CustomData, so
-   * we set them here in case they are missing */
-  if (!CustomData_has_layer(&tmp.vdata, CD_MVERT)) {
-    CustomData_add_layer(&tmp.vdata,
-                         CD_MVERT,
-                         CD_ASSIGN,
-                         (alloctype == CD_ASSIGN) ? dm->getVertArray(dm) : dm->dupVertArray(dm),
-                         totvert);
-  }
-  if (!CustomData_has_layer(&tmp.edata, CD_MEDGE)) {
-    CustomData_add_layer(&tmp.edata,
-                         CD_MEDGE,
-                         CD_ASSIGN,
-                         (alloctype == CD_ASSIGN) ? dm->getEdgeArray(dm) : dm->dupEdgeArray(dm),
-                         totedge);
-  }
-  if (!CustomData_has_layer(&tmp.pdata, CD_MPOLY)) {
-    tmp.mloop = (alloctype == CD_ASSIGN) ? dm->getLoopArray(dm) : dm->dupLoopArray(dm);
-    tmp.mpoly = (alloctype == CD_ASSIGN) ? dm->getPolyArray(dm) : dm->dupPolyArray(dm);
-
-    CustomData_add_layer(&tmp.ldata, CD_MLOOP, CD_ASSIGN, tmp.mloop, tmp.totloop);
-    CustomData_add_layer(&tmp.pdata, CD_MPOLY, CD_ASSIGN, tmp.mpoly, tmp.totpoly);
-  }
-
-  /* object had got displacement layer, should copy this layer to save sculpted data */
-  /* NOTE: maybe some other layers should be copied? nazgul */
-  if (CustomData_has_layer(&me->ldata, CD_MDISPS)) {
-    if (totloop == me->totloop) {
-      MDisps *mdisps = CustomData_get_layer(&me->ldata, CD_MDISPS);
-      CustomData_add_layer(&tmp.ldata, CD_MDISPS, alloctype, mdisps, totloop);
-    }
-  }
-
-  /* yes, must be before _and_ after tessellate */
-  BKE_mesh_update_customdata_pointers(&tmp, false);
-
-  /* since 2.65 caller must do! */
-  // BKE_mesh_tessface_calc(&tmp);
-
-  CustomData_free(&me->vdata, me->totvert);
-  CustomData_free(&me->edata, me->totedge);
-  CustomData_free(&me->fdata, me->totface);
-  CustomData_free(&me->ldata, me->totloop);
-  CustomData_free(&me->pdata, me->totpoly);
-
-  /* ok, this should now use new CD shapekey data,
-   * which should be fed through the modifier
-   * stack */
-  if (tmp.totvert != me->totvert && !did_shapekeys && me->key) {
-    CLOG_WARN(&LOG, "YEEK! this should be recoded! Shape key loss!: ID '%s'", tmp.id.name);
-    if (tmp.key && !(tmp.id.tag & LIB_TAG_NO_MAIN)) {
-      id_us_min(&tmp.key->id);
-    }
-    tmp.key = NULL;
-  }
-
-  /* Clear selection history */
-  MEM_SAFE_FREE(tmp.mselect);
-  tmp.totselect = 0;
-  BLI_assert(ELEM(tmp.bb, NULL, me->bb));
-  if (me->bb) {
-    MEM_freeN(me->bb);
-    tmp.bb = NULL;
-  }
-
-  /* skip the listbase */
-  MEMCPY_STRUCT_AFTER(me, &tmp, id.prev);
-
-  if (take_ownership) {
-    if (alloctype == CD_ASSIGN) {
-      CustomData_free_typemask(&dm->vertData, dm->numVertData, ~mask->vmask);
-      CustomData_free_typemask(&dm->edgeData, dm->numEdgeData, ~mask->emask);
-      CustomData_free_typemask(&dm->loopData, dm->numLoopData, ~mask->lmask);
-      CustomData_free_typemask(&dm->polyData, dm->numPolyData, ~mask->pmask);
-    }
-    dm->release(dm);
-  }
-}
-
 /** Utility function to convert an (evaluated) Mesh to a shape key block. */
 /* Just a shallow wrapper around BKE_keyblock_convert_from_mesh,
  * that ensures both evaluated mesh and original one has same number of vertices. */
@@ -819,21 +668,6 @@ void DM_interp_vert_data(DerivedMesh *source,
       &source->vertData, &dest->vertData, src_indices, weights, NULL, count, dest_index);
 }
 
-DerivedMesh *mesh_create_derived(Mesh *me, float (*vertCos)[3])
-{
-  DerivedMesh *dm = CDDM_from_mesh(me);
-
-  if (!dm) {
-    return NULL;
-  }
-
-  if (vertCos) {
-    CDDM_apply_vert_coords(dm, vertCos);
-  }
-
-  return dm;
-}
-
 static float (*get_editbmesh_orco_verts(BMEditMesh *em))[3]
 {
   BMIter iter;
@@ -904,7 +738,7 @@ static Mesh *create_orco_mesh(Object *ob, Mesh *me, BMEditMesh *em, int layer)
   orco = get_orco_coords(ob, em, layer, &free);
 
   if (orco) {
-    BKE_mesh_apply_vert_coords(mesh, orco);
+    BKE_mesh_vert_coords_apply(mesh, orco);
     if (free) {
       MEM_freeN(orco);
     }
@@ -924,10 +758,10 @@ static void add_orco_mesh(Object *ob, BMEditMesh *em, Mesh *mesh, Mesh *mesh_orc
     free = 1;
 
     if (mesh_orco->totvert == totvert) {
-      orco = BKE_mesh_vertexCos_get(mesh_orco, NULL);
+      orco = BKE_mesh_vert_coords_alloc(mesh_orco, NULL);
     }
     else {
-      orco = BKE_mesh_vertexCos_get(mesh, NULL);
+      orco = BKE_mesh_vert_coords_alloc(mesh, NULL);
     }
   }
   else {
@@ -963,67 +797,6 @@ static void editmesh_update_statvis_color(const Scene *scene, Object *ob)
   BKE_editmesh_statvis_calc(em, me->runtime.edit_data, &scene->toolsettings->statvis);
 }
 
-static void shapekey_layers_to_keyblocks(DerivedMesh *dm, Mesh *me, int actshape_uid)
-{
-  KeyBlock *kb;
-  int i, j, tot;
-
-  if (!me->key) {
-    return;
-  }
-
-  tot = CustomData_number_of_layers(&dm->vertData, CD_SHAPEKEY);
-  for (i = 0; i < tot; i++) {
-    CustomDataLayer *layer =
-        &dm->vertData.layers[CustomData_get_layer_index_n(&dm->vertData, CD_SHAPEKEY, i)];
-    float(*cos)[3], (*kbcos)[3];
-
-    for (kb = me->key->block.first; kb; kb = kb->next) {
-      if (kb->uid == layer->uid) {
-        break;
-      }
-    }
-
-    if (!kb) {
-      kb = BKE_keyblock_add(me->key, layer->name);
-      kb->uid = layer->uid;
-    }
-
-    if (kb->data) {
-      MEM_freeN(kb->data);
-    }
-
-    cos = CustomData_get_layer_n(&dm->vertData, CD_SHAPEKEY, i);
-    kb->totelem = dm->numVertData;
-
-    kb->data = kbcos = MEM_malloc_arrayN(kb->totelem, 3 * sizeof(float), "kbcos DerivedMesh.c");
-    if (kb->uid == actshape_uid) {
-      MVert *mvert = dm->getVertArray(dm);
-
-      for (j = 0; j < dm->numVertData; j++, kbcos++, mvert++) {
-        copy_v3_v3(*kbcos, mvert->co);
-      }
-    }
-    else {
-      for (j = 0; j < kb->totelem; j++, cos++, kbcos++) {
-        copy_v3_v3(*kbcos, *cos);
-      }
-    }
-  }
-
-  for (kb = me->key->block.first; kb; kb = kb->next) {
-    if (kb->totelem != dm->numVertData) {
-      if (kb->data) {
-        MEM_freeN(kb->data);
-      }
-
-      kb->totelem = dm->numVertData;
-      kb->data = MEM_calloc_arrayN(kb->totelem, 3 * sizeof(float), "kb->data derivedmesh.c");
-      CLOG_ERROR(&LOG, "lost a shapekey layer: '%s'! (bmesh internal error)", kb->name);
-    }
-  }
-}
-
 static void mesh_copy_autosmooth(Mesh *me, Mesh *me_orig)
 {
   if (me_orig->flag & ME_AUTOSMOOTH) {
@@ -1223,14 +996,14 @@ static void mesh_calc_modifiers(struct Depsgraph *depsgraph,
 
       if (mti->type == eModifierTypeType_OnlyDeform && !sculpt_dyntopo) {
         if (!deformed_verts) {
-          deformed_verts = BKE_mesh_vertexCos_get(mesh_input, &num_deformed_verts);
+          deformed_verts = BKE_mesh_vert_coords_alloc(mesh_input, &num_deformed_verts);
         }
         else if (isPrevDeform && mti->dependsOnNormals && mti->dependsOnNormals(md)) {
           if (mesh_final == NULL) {
             mesh_final = BKE_mesh_copy_for_eval(mesh_input, true);
             ASSERT_IS_VALID_MESH(mesh_final);
           }
-          BKE_mesh_apply_vert_coords(mesh_final, deformed_verts);
+          BKE_mesh_vert_coords_apply(mesh_final, deformed_verts);
         }
 
         modwrap_deformVerts(md, &mectx, mesh_final, deformed_verts, num_deformed_verts);
@@ -1254,7 +1027,7 @@ static void mesh_calc_modifiers(struct Depsgraph *depsgraph,
       mesh_deform = BKE_mesh_copy_for_eval(mesh_input, true);
 
       if (deformed_verts) {
-        BKE_mesh_apply_vert_coords(mesh_deform, deformed_verts);
+        BKE_mesh_vert_coords_apply(mesh_deform, deformed_verts);
       }
     }
   }
@@ -1337,10 +1110,10 @@ static void mesh_calc_modifiers(struct Depsgraph *depsgraph,
           /* Deforming a mesh, read the vertex locations
            * out of the mesh and deform them. Once done with this
            * run of deformers verts will be written back. */
-          deformed_verts = BKE_mesh_vertexCos_get(mesh_final, &num_deformed_verts);
+          deformed_verts = BKE_mesh_vert_coords_alloc(mesh_final, &num_deformed_verts);
         }
         else {
-          deformed_verts = BKE_mesh_vertexCos_get(mesh_input, &num_deformed_verts);
+          deformed_verts = BKE_mesh_vert_coords_alloc(mesh_input, &num_deformed_verts);
         }
       }
       /* if this is not the last modifier in the stack then recalculate the normals
@@ -1348,7 +1121,7 @@ static void mesh_calc_modifiers(struct Depsgraph *depsgraph,
       else if (isPrevDeform && mti->dependsOnNormals && mti->dependsOnNormals(md)) {
         /* XXX, this covers bug #23673, but we may need normal calc for other types */
         if (mesh_final) {
-          BKE_mesh_apply_vert_coords(mesh_final, deformed_verts);
+          BKE_mesh_vert_coords_apply(mesh_final, deformed_verts);
         }
       }
 
@@ -1367,7 +1140,7 @@ static void mesh_calc_modifiers(struct Depsgraph *depsgraph,
       /* apply vertex coordinates or build a Mesh as necessary */
       if (mesh_final) {
         if (deformed_verts) {
-          BKE_mesh_apply_vert_coords(mesh_final, deformed_verts);
+          BKE_mesh_vert_coords_apply(mesh_final, deformed_verts);
         }
       }
       else {
@@ -1375,7 +1148,7 @@ static void mesh_calc_modifiers(struct Depsgraph *depsgraph,
         ASSERT_IS_VALID_MESH(mesh_final);
 
         if (deformed_verts) {
-          BKE_mesh_apply_vert_coords(mesh_final, deformed_verts);
+          BKE_mesh_vert_coords_apply(mesh_final, deformed_verts);
         }
 
         /* Initialize original indices the first time we evaluate a
@@ -1554,7 +1327,7 @@ static void mesh_calc_modifiers(struct Depsgraph *depsgraph,
     }
   }
   if (deformed_verts) {
-    BKE_mesh_apply_vert_coords(mesh_final, deformed_verts);
+    BKE_mesh_vert_coords_apply(mesh_final, deformed_verts);
     MEM_freeN(deformed_verts);
     deformed_verts = NULL;
   }
@@ -1614,14 +1387,14 @@ static void mesh_calc_modifiers(struct Depsgraph *depsgraph,
   }
 }
 
-float (*editbmesh_get_vertex_cos(BMEditMesh *em, int *r_numVerts))[3]
+float (*editbmesh_vert_coords_alloc(BMEditMesh *em, int *r_vert_len))[3]
 {
   BMIter iter;
   BMVert *eve;
   float(*cos)[3];
   int i;
 
-  *r_numVerts = em->bm->totvert;
+  *r_vert_len = em->bm->totvert;
 
   cos = MEM_malloc_arrayN(em->bm->totvert, 3 * sizeof(float), "vertexcos");
 
@@ -1793,10 +1566,10 @@ static void editbmesh_calc_modifiers(struct Depsgraph *depsgraph,
           /* Deforming a derived mesh, read the vertex locations
            * out of the mesh and deform them. Once done with this
            * run of deformers verts will be written back. */
-          deformed_verts = BKE_mesh_vertexCos_get(mesh_final, &num_deformed_verts);
+          deformed_verts = BKE_mesh_vert_coords_alloc(mesh_final, &num_deformed_verts);
         }
         else {
-          deformed_verts = editbmesh_get_vertex_cos(em_input, &num_deformed_verts);
+          deformed_verts = editbmesh_vert_coords_alloc(em_input, &num_deformed_verts);
         }
       }
       else if (isPrevDeform && mti->dependsOnNormals && mti->dependsOnNormals(md)) {
@@ -1806,7 +1579,7 @@ static void editbmesh_calc_modifiers(struct Depsgraph *depsgraph,
           mesh_copy_autosmooth(mesh_final, mesh_input);
         }
         BLI_assert(deformed_verts != NULL);
-        BKE_mesh_apply_vert_coords(mesh_final, deformed_verts);
+        BKE_mesh_vert_coords_apply(mesh_final, deformed_verts);
       }
 
       if (mti->deformVertsEM) {
@@ -1826,7 +1599,7 @@ static void editbmesh_calc_modifiers(struct Depsgraph *depsgraph,
             BKE_id_free(NULL, mesh_final);
           }
           mesh_final = mesh_tmp;
-          BKE_mesh_apply_vert_coords(mesh_final, deformed_verts);
+          BKE_mesh_vert_coords_apply(mesh_final, deformed_verts);
         }
         else if (mesh_final == mesh_cage) {
           /* 'me' may be changed by this modifier, so we need to copy it. */
@@ -1840,7 +1613,7 @@ static void editbmesh_calc_modifiers(struct Depsgraph *depsgraph,
         mesh_copy_autosmooth(mesh_final, mesh_input);
 
         if (deformed_verts) {
-          BKE_mesh_apply_vert_coords(mesh_final, deformed_verts);
+          BKE_mesh_vert_coords_apply(mesh_final, deformed_verts);
         }
       }
 
@@ -1910,7 +1683,7 @@ static void editbmesh_calc_modifiers(struct Depsgraph *depsgraph,
     if (r_cage && i == cageIndex) {
       if (mesh_final && deformed_verts) {
         mesh_cage = BKE_mesh_copy_for_eval(mesh_final, false);
-        BKE_mesh_apply_vert_coords(mesh_cage, deformed_verts);
+        BKE_mesh_vert_coords_apply(mesh_cage, deformed_verts);
       }
       else if (mesh_final) {
         mesh_cage = mesh_final;
@@ -1942,7 +1715,7 @@ static void editbmesh_calc_modifiers(struct Depsgraph *depsgraph,
         BKE_id_free(NULL, mesh_final);
       }
       mesh_final = mesh_tmp;
-      BKE_mesh_apply_vert_coords(mesh_final, deformed_verts);
+      BKE_mesh_vert_coords_apply(mesh_final, deformed_verts);
     }
   }
   else if (!deformed_verts && mesh_cage) {
diff --git a/source/blender/blenkernel/intern/anim_sys.c b/source/blender/blenkernel/intern/anim_sys.c
index 1a34c5376f6..ab8ea37377d 100644
--- a/source/blender/blenkernel/intern/anim_sys.c
+++ b/source/blender/blenkernel/intern/anim_sys.c
@@ -648,7 +648,7 @@ char *BKE_animdata_driver_path_hack(bContext *C,
                                     PropertyRNA *prop,
                                     char *base_path)
 {
-  ID *id = (ID *)ptr->id.data;
+  ID *id = ptr->owner_id;
   ScrArea *sa = CTX_wm_area(C);
 
   /* get standard path which may be extended */
@@ -1656,14 +1656,14 @@ static bool animsys_store_rna_setting(PointerRNA *ptr,
   if (path) {
     /* get property to write to */
     if (RNA_path_resolve_property(ptr, path, &r_result->ptr, &r_result->prop)) {
-      if ((ptr->id.data == NULL) || RNA_property_animateable(&r_result->ptr, r_result->prop)) {
+      if ((ptr->owner_id == NULL) || RNA_property_animateable(&r_result->ptr, r_result->prop)) {
         int array_len = RNA_property_array_length(&r_result->ptr, r_result->prop);
 
         if (array_len && array_index >= array_len) {
           if (G.debug & G_DEBUG) {
             CLOG_WARN(&LOG,
                       "Animato: Invalid array index. ID = '%s',  '%s[%d]', array length is %d",
-                      (ptr->id.data) ? (((ID *)ptr->id.data)->name + 2) : "<No ID>",
+                      (ptr->owner_id) ? (ptr->owner_id->name + 2) : "<No ID>",
                       path,
                       array_index,
                       array_len - 1);
@@ -1682,7 +1682,7 @@ static bool animsys_store_rna_setting(PointerRNA *ptr,
       if (G.debug & G_DEBUG) {
         CLOG_WARN(&LOG,
                   "Animato: Invalid path. ID = '%s',  '%s[%d]'",
-                  (ptr->id.data) ? (((ID *)ptr->id.data)->name + 2) : "<No ID>",
+                  (ptr->owner_id) ? (ptr->owner_id->name + 2) : "<No ID>",
                   path,
                   array_index);
       }
@@ -1703,7 +1703,7 @@ static bool animsys_read_rna_setting(PathResolvedRNA *anim_rna, float *r_value)
   float orig_value;
 
   /* caller must ensure this is animatable */
-  BLI_assert(RNA_property_animateable(ptr, prop) || ptr->id.data == NULL);
+  BLI_assert(RNA_property_animateable(ptr, prop) || ptr->owner_id == NULL);
 
   switch (RNA_property_type(prop)) {
     case PROP_BOOLEAN: {
@@ -1765,7 +1765,7 @@ static bool animsys_write_rna_setting(PathResolvedRNA *anim_rna, const float val
   int array_index = anim_rna->prop_index;
 
   /* caller must ensure this is animatable */
-  BLI_assert(RNA_property_animateable(ptr, prop) || ptr->id.data == NULL);
+  BLI_assert(RNA_property_animateable(ptr, prop) || ptr->owner_id == NULL);
 
   /* Check whether value is new. Otherwise we skip all the updates. */
   float old_value;
@@ -1844,7 +1844,7 @@ static bool animsys_construct_orig_pointer_rna(const PointerRNA *ptr, PointerRNA
    * not a valid pointer, but there are exceptions in various places of this file which handles
    * such pointers.
    * We do special trickery here as well, to quickly go from evaluated to original NlaStrip. */
-  if (ptr->id.data == NULL) {
+  if (ptr->owner_id == NULL) {
     if (ptr->type != &RNA_NlaStrip) {
       return false;
     }
@@ -1855,8 +1855,8 @@ static bool animsys_construct_orig_pointer_rna(const PointerRNA *ptr, PointerRNA
     ptr_orig->data = strip->orig_strip;
   }
   else {
-    ptr_orig->id.data = ((ID *)ptr_orig->id.data)->orig_id;
-    ptr_orig->data = ptr_orig->id.data;
+    ptr_orig->owner_id = ptr_orig->owner_id->orig_id;
+    ptr_orig->data = ptr_orig->owner_id;
   }
   return true;
 }
@@ -2003,7 +2003,7 @@ void animsys_evaluate_action_group(PointerRNA *ptr, bAction *act, bActionGroup *
     return;
   }
 
-  action_idcode_patch_check(ptr->id.data, act);
+  action_idcode_patch_check(ptr->owner_id, act);
 
   /* if group is muted, don't evaluated any of the F-Curve */
   if (agrp->flag & AGRP_MUTED) {
@@ -2034,7 +2034,7 @@ static void animsys_evaluate_action_ex(PointerRNA *ptr,
     return;
   }
 
-  action_idcode_patch_check(ptr->id.data, act);
+  action_idcode_patch_check(ptr->owner_id, act);
 
   /* calculate then execute each curve */
   animsys_evaluate_fcurves(ptr, &act->curves, ctime, flush_to_original);
@@ -2643,7 +2643,7 @@ static NlaEvalChannel *nlaevalchan_verify(PointerRNA *ptr, NlaEvalData *nlaeval,
     if (G.debug & G_DEBUG) {
       CLOG_WARN(&LOG,
                 "Animato: Invalid path. ID = '%s',  '%s'",
-                (ptr->id.data) ? (((ID *)ptr->id.data)->name + 2) : "<No ID>",
+                (ptr->owner_id) ? (ptr->owner_id->name + 2) : "<No ID>",
                 path);
     }
 
@@ -2651,7 +2651,7 @@ static NlaEvalChannel *nlaevalchan_verify(PointerRNA *ptr, NlaEvalData *nlaeval,
   }
 
   /* Check that the property can be animated. */
-  if (ptr->id.data != NULL && !RNA_property_animateable(&key.ptr, key.prop)) {
+  if (ptr->owner_id != NULL && !RNA_property_animateable(&key.ptr, key.prop)) {
     return NULL;
   }
 
@@ -2884,7 +2884,7 @@ static bool nlaeval_blend_value(NlaBlendData *blend,
 
   if (index < 0) {
     if (G.debug & G_DEBUG) {
-      ID *id = nec->key.ptr.id.data;
+      ID *id = nec->key.ptr.owner_id;
       CLOG_WARN(&LOG,
                 "Animato: Invalid array index. ID = '%s',  '%s[%d]', array length is %d",
                 id ? (id->name + 2) : "<No ID>",
@@ -3081,7 +3081,7 @@ static void nlastrip_evaluate_actionclip(PointerRNA *ptr,
     return;
   }
 
-  action_idcode_patch_check(ptr->id.data, strip->act);
+  action_idcode_patch_check(ptr->owner_id, strip->act);
 
   /* join this strip's modifiers to the parent's modifiers (own modifiers first) */
   nlaeval_fmodifiers_join_stacks(&tmp_modifiers, &strip->modifiers, modifiers);
diff --git a/source/blender/blenkernel/intern/armature.c b/source/blender/blenkernel/intern/armature.c
index ea927ca9333..168422a4454 100644
--- a/source/blender/blenkernel/intern/armature.c
+++ b/source/blender/blenkernel/intern/armature.c
@@ -268,6 +268,90 @@ void BKE_armature_copy_bone_transforms(bArmature *armature_dst, const bArmature
   }
 }
 
+/** Helper for #ED_armature_transform */
+static void armature_transform_recurse(ListBase *bonebase,
+                                       const float mat[4][4],
+                                       const bool do_props,
+                                       /* Cached from 'mat'. */
+                                       const float mat3[3][3],
+                                       const float scale,
+                                       /* Child bones. */
+                                       const Bone *bone_parent,
+                                       const float arm_mat_parent_inv[4][4])
+{
+  for (Bone *bone = bonebase->first; bone; bone = bone->next) {
+
+    /* Transform the bone's roll. */
+    if (bone_parent == NULL) {
+
+      float roll_mat[3][3];
+      {
+        float delta[3];
+        sub_v3_v3v3(delta, bone->tail, bone->head);
+        vec_roll_to_mat3(delta, bone->roll, roll_mat);
+      }
+
+      /* Transform the roll matrix. */
+      mul_m3_m3m3(roll_mat, mat3, roll_mat);
+
+      /* Apply the transformed roll back. */
+      mat3_to_vec_roll(roll_mat, NULL, &bone->roll);
+    }
+
+    mul_m4_v3(mat, bone->arm_head);
+    mul_m4_v3(mat, bone->arm_tail);
+
+    /* Get the new head and tail */
+    if (bone_parent) {
+      sub_v3_v3v3(bone->head, bone->arm_head, bone_parent->arm_tail);
+      sub_v3_v3v3(bone->tail, bone->arm_tail, bone_parent->arm_tail);
+
+      mul_mat3_m4_v3(arm_mat_parent_inv, bone->head);
+      mul_mat3_m4_v3(arm_mat_parent_inv, bone->tail);
+    }
+    else {
+      copy_v3_v3(bone->head, bone->arm_head);
+      copy_v3_v3(bone->tail, bone->arm_tail);
+    }
+
+    BKE_armature_where_is_bone(bone, bone_parent, false);
+
+    {
+      float arm_mat3[3][3];
+      copy_m3_m4(arm_mat3, bone->arm_mat);
+      mat3_to_vec_roll(arm_mat3, NULL, &bone->arm_roll);
+    }
+
+    if (do_props) {
+      bone->rad_head *= scale;
+      bone->rad_tail *= scale;
+      bone->dist *= scale;
+
+      /* we could be smarter and scale by the matrix along the x & z axis */
+      bone->xwidth *= scale;
+      bone->zwidth *= scale;
+    }
+
+    if (!BLI_listbase_is_empty(&bone->childbase)) {
+      float arm_mat_inv[4][4];
+      invert_m4_m4(arm_mat_inv, bone->arm_mat);
+      armature_transform_recurse(&bone->childbase, mat, do_props, mat3, scale, bone, arm_mat_inv);
+    }
+  }
+}
+
+void BKE_armature_transform(bArmature *arm, const float mat[4][4], const bool do_props)
+{
+  /* Store the scale of the matrix here to use on envelopes. */
+  float scale = mat4_to_scale(mat);
+  float mat3[3][3];
+
+  copy_m3_m4(mat3, mat);
+  normalize_m3(mat3);
+
+  armature_transform_recurse(&arm->bonebase, mat, do_props, mat3, scale, NULL, NULL);
+}
+
 static Bone *get_named_bone_bonechildren(ListBase *lb, const char *name)
 {
   Bone *curBone, *rbone;
@@ -2285,7 +2369,7 @@ void vec_roll_to_mat3(const float vec[3], const float roll, float mat[3][3])
 
 /* recursive part, calculates restposition of entire tree of children */
 /* used by exiting editmode too */
-void BKE_armature_where_is_bone(Bone *bone, Bone *prevbone, const bool use_recursion)
+void BKE_armature_where_is_bone(Bone *bone, const Bone *bone_parent, const bool use_recursion)
 {
   float vec[3];
 
@@ -2301,13 +2385,13 @@ void BKE_armature_where_is_bone(Bone *bone, Bone *prevbone, const bool use_recur
     bone->segments = 1;
   }
 
-  if (prevbone) {
+  if (bone_parent) {
     float offs_bone[4][4];
     /* yoffs(b-1) + root(b) + bonemat(b) */
     BKE_bone_offset_matrix_get(bone, offs_bone);
 
     /* Compose the matrix for this bone  */
-    mul_m4_m4m4(bone->arm_mat, prevbone->arm_mat, offs_bone);
+    mul_m4_m4m4(bone->arm_mat, bone_parent->arm_mat, offs_bone);
   }
   else {
     copy_m4_m3(bone->arm_mat, bone->bone_mat);
@@ -2316,9 +2400,9 @@ void BKE_armature_where_is_bone(Bone *bone, Bone *prevbone, const bool use_recur
 
   /* and the kiddies */
   if (use_recursion) {
-    prevbone = bone;
+    bone_parent = bone;
     for (bone = bone->childbase.first; bone; bone = bone->next) {
-      BKE_armature_where_is_bone(bone, prevbone, use_recursion);
+      BKE_armature_where_is_bone(bone, bone_parent, use_recursion);
     }
   }
 }
diff --git a/source/blender/blenkernel/intern/armature_update.c b/source/blender/blenkernel/intern/armature_update.c
index bf7d81e5d63..00975976130 100644
--- a/source/blender/blenkernel/intern/armature_update.c
+++ b/source/blender/blenkernel/intern/armature_update.c
@@ -887,9 +887,10 @@ void BKE_pose_eval_proxy_copy_bone(struct Depsgraph *depsgraph, Object *object,
   }
   BLI_assert(ID_IS_LINKED(object) && object->proxy_from != NULL);
   bPoseChannel *pchan = pose_pchan_get_indexed(object, pchan_index);
+  BLI_assert(pchan != NULL);
   DEG_debug_print_eval_subdata(
       depsgraph, __func__, object->id.name, object, "pchan", pchan->name, pchan);
-  /* TODO(sergey): Use indexec lookup, once it's guaranteed to be kept
+  /* TODO(sergey): Use indexed lookup, once it's guaranteed to be kept
    * around for the time while proxies are evaluating.
    */
 #if 0
@@ -897,8 +898,13 @@ void BKE_pose_eval_proxy_copy_bone(struct Depsgraph *depsgraph, Object *object,
 #else
   bPoseChannel *pchan_from = BKE_pose_channel_find_name(object->proxy_from->pose, pchan->name);
 #endif
-  BLI_assert(pchan != NULL);
-  BLI_assert(pchan_from != NULL);
+  if (pchan_from == NULL) {
+    printf(
+        "WARNING: Could not find bone %s in linked ID anymore... "
+        "You should delete and re-generate your proxy.\n",
+        pchan->name);
+    return;
+  }
   BKE_pose_copyesult_pchan_result(pchan, pchan_from);
   copy_dq_dq(&pchan->runtime.deform_dual_quat, &pchan_from->runtime.deform_dual_quat);
   BKE_pchan_bbone_segments_cache_copy(pchan, pchan_from);
diff --git a/source/blender/blenkernel/intern/brush.c b/source/blender/blenkernel/intern/brush.c
index b2d3ccfebc3..83e4a582ff1 100644
--- a/source/blender/blenkernel/intern/brush.c
+++ b/source/blender/blenkernel/intern/brush.c
@@ -72,6 +72,7 @@ static void brush_defaults(Brush *brush)
   brush->autosmooth_factor = 0.0f;
   brush->topology_rake_factor = 0.0f;
   brush->crease_pinch_factor = 0.5f;
+  brush->normal_radius_factor = 0.5f;
   brush->sculpt_plane = SCULPT_DISP_DIR_AREA;
   /* How far above or below the plane that is found by averaging the faces. */
   brush->plane_offset = 0.0f;
diff --git a/source/blender/blenkernel/intern/bvhutils.c b/source/blender/blenkernel/intern/bvhutils.c
index 8c125d1609b..85a12027bf2 100644
--- a/source/blender/blenkernel/intern/bvhutils.c
+++ b/source/blender/blenkernel/intern/bvhutils.c
@@ -425,11 +425,11 @@ static BVHTree *bvhtree_from_editmesh_verts_create_tree(float epsilon,
                                                         int tree_type,
                                                         int axis,
                                                         BMEditMesh *em,
-                                                        const int verts_num,
                                                         const BLI_bitmap *verts_mask,
                                                         int verts_num_active)
 {
   BM_mesh_elem_table_ensure(em->bm, BM_VERT);
+  const int verts_num = em->bm->totvert;
   if (verts_mask) {
     BLI_assert(IN_RANGE_INCL(verts_num_active, 0, verts_num));
   }
@@ -516,53 +516,53 @@ BVHTree *bvhtree_from_editmesh_verts_ex(BVHTreeFromEditMesh *data,
                                         int verts_num_active,
                                         float epsilon,
                                         int tree_type,
-                                        int axis)
+                                        int axis,
+                                        const int bvh_cache_type,
+                                        BVHCache **bvh_cache)
 {
-  BVHTree *tree = bvhtree_from_editmesh_verts_create_tree(
-      epsilon, tree_type, axis, em, em->bm->totvert, verts_mask, verts_num_active);
-
-  if (tree) {
-    memset(data, 0, sizeof(*data));
-    data->tree = tree;
-    data->em = em;
-    data->nearest_callback = NULL;
-    data->raycast_callback = editmesh_verts_spherecast;
-  }
-
-  return tree;
-}
+  BVHTree *tree = NULL;
 
-BVHTree *bvhtree_from_editmesh_verts(BVHTreeFromEditMesh *data,
-                                     BMEditMesh *em,
-                                     float epsilon,
-                                     int tree_type,
-                                     int axis,
-                                     BVHCache **bvh_cache)
-{
   if (bvh_cache) {
     BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_READ);
-    data->cached = bvhcache_find(*bvh_cache, BVHTREE_FROM_EM_VERTS, &data->tree);
+    data->cached = bvhcache_find(*bvh_cache, bvh_cache_type, &data->tree);
     BLI_rw_mutex_unlock(&cache_rwlock);
 
     if (data->cached == false) {
       BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
-      data->cached = bvhcache_find(*bvh_cache, BVHTREE_FROM_EM_VERTS, &data->tree);
+      data->cached = bvhcache_find(*bvh_cache, bvh_cache_type, &data->tree);
       if (data->cached == false) {
-        data->tree = bvhtree_from_editmesh_verts_ex(data, em, NULL, -1, epsilon, tree_type, axis);
+        tree = bvhtree_from_editmesh_verts_create_tree(
+            epsilon, tree_type, axis, em, verts_mask, verts_num_active);
 
         /* Save on cache for later use */
         /* printf("BVHTree built and saved on cache\n"); */
-        bvhcache_insert(bvh_cache, data->tree, BVHTREE_FROM_EM_VERTS);
+        bvhcache_insert(bvh_cache, tree, bvh_cache_type);
         data->cached = true;
       }
       BLI_rw_mutex_unlock(&cache_rwlock);
     }
   }
   else {
-    data->tree = bvhtree_from_editmesh_verts_ex(data, em, NULL, -1, epsilon, tree_type, axis);
+    tree = bvhtree_from_editmesh_verts_create_tree(
+        epsilon, tree_type, axis, em, verts_mask, verts_num_active);
   }
 
-  return data->tree;
+  if (tree) {
+    memset(data, 0, sizeof(*data));
+    data->tree = tree;
+    data->em = em;
+    data->nearest_callback = NULL;
+    data->raycast_callback = editmesh_verts_spherecast;
+    data->cached = bvh_cache != NULL;
+  }
+
+  return tree;
+}
+
+BVHTree *bvhtree_from_editmesh_verts(
+    BVHTreeFromEditMesh *data, BMEditMesh *em, float epsilon, int tree_type, int axis)
+{
+  return bvhtree_from_editmesh_verts_ex(data, em, NULL, -1, epsilon, tree_type, axis, 0, NULL);
 }
 
 /**
@@ -580,13 +580,40 @@ BVHTree *bvhtree_from_mesh_verts_ex(BVHTreeFromMesh *data,
                                     int verts_num_active,
                                     float epsilon,
                                     int tree_type,
-                                    int axis)
+                                    int axis,
+                                    const int bvh_cache_type,
+                                    BVHCache **bvh_cache)
 {
-  BVHTree *tree = bvhtree_from_mesh_verts_create_tree(
-      epsilon, tree_type, axis, vert, verts_num, verts_mask, verts_num_active);
+  bool in_cache = false;
+  BVHTree *tree = NULL;
+  if (bvh_cache) {
+    BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_READ);
+    in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
+    BLI_rw_mutex_unlock(&cache_rwlock);
+    if (in_cache == false) {
+      BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
+      in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
+      if (in_cache) {
+        BLI_rw_mutex_unlock(&cache_rwlock);
+      }
+    }
+  }
+
+  if (in_cache == false) {
+    tree = bvhtree_from_mesh_verts_create_tree(
+        epsilon, tree_type, axis, vert, verts_num, verts_mask, verts_num_active);
+
+    if (bvh_cache) {
+      /* Save on cache for later use */
+      /* printf("BVHTree built and saved on cache\n"); */
+      bvhcache_insert(bvh_cache, tree, bvh_cache_type);
+      BLI_rw_mutex_unlock(&cache_rwlock);
+      in_cache = true;
+    }
+  }
 
   /* Setup BVHTreeFromMesh */
-  bvhtree_from_mesh_verts_setup_data(data, tree, false, vert, vert_allocated);
+  bvhtree_from_mesh_verts_setup_data(data, tree, in_cache, vert, vert_allocated);
 
   return tree;
 }
@@ -601,11 +628,12 @@ static BVHTree *bvhtree_from_editmesh_edges_create_tree(float epsilon,
                                                         int tree_type,
                                                         int axis,
                                                         BMEditMesh *em,
-                                                        const int edges_num,
                                                         const BLI_bitmap *edges_mask,
                                                         int edges_num_active)
 {
   BM_mesh_elem_table_ensure(em->bm, BM_EDGE);
+  const int edges_num = em->bm->totedge;
+
   if (edges_mask) {
     BLI_assert(IN_RANGE_INCL(edges_num_active, 0, edges_num));
   }
@@ -705,55 +733,53 @@ BVHTree *bvhtree_from_editmesh_edges_ex(BVHTreeFromEditMesh *data,
                                         int edges_num_active,
                                         float epsilon,
                                         int tree_type,
-                                        int axis)
+                                        int axis,
+                                        const int bvh_cache_type,
+                                        BVHCache **bvh_cache)
 {
-  int edge_num = em->bm->totedge;
-
-  BVHTree *tree = bvhtree_from_editmesh_edges_create_tree(
-      epsilon, tree_type, axis, em, edge_num, edges_mask, edges_num_active);
-
-  if (tree) {
-    memset(data, 0, sizeof(*data));
-    data->tree = tree;
-    data->em = em;
-    data->nearest_callback = NULL; /* TODO */
-    data->raycast_callback = NULL; /* TODO */
-  }
-
-  return tree;
-}
+  BVHTree *tree = NULL;
 
-BVHTree *bvhtree_from_editmesh_edges(BVHTreeFromEditMesh *data,
-                                     BMEditMesh *em,
-                                     float epsilon,
-                                     int tree_type,
-                                     int axis,
-                                     BVHCache **bvh_cache)
-{
   if (bvh_cache) {
     BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_READ);
-    data->cached = bvhcache_find(*bvh_cache, BVHTREE_FROM_EM_EDGES, &data->tree);
+    data->cached = bvhcache_find(*bvh_cache, bvh_cache_type, &data->tree);
     BLI_rw_mutex_unlock(&cache_rwlock);
 
     if (data->cached == false) {
       BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
-      data->cached = bvhcache_find(*bvh_cache, BVHTREE_FROM_EM_EDGES, &data->tree);
+      data->cached = bvhcache_find(*bvh_cache, bvh_cache_type, &data->tree);
       if (data->cached == false) {
-        data->tree = bvhtree_from_editmesh_edges_ex(data, em, NULL, -1, epsilon, tree_type, axis);
+        tree = bvhtree_from_editmesh_edges_create_tree(
+            epsilon, tree_type, axis, em, edges_mask, edges_num_active);
 
         /* Save on cache for later use */
         /* printf("BVHTree built and saved on cache\n"); */
-        bvhcache_insert(bvh_cache, data->tree, BVHTREE_FROM_EM_EDGES);
+        bvhcache_insert(bvh_cache, tree, bvh_cache_type);
         data->cached = true;
       }
       BLI_rw_mutex_unlock(&cache_rwlock);
     }
   }
   else {
-    data->tree = bvhtree_from_editmesh_edges_ex(data, em, NULL, -1, epsilon, tree_type, axis);
+    tree = bvhtree_from_editmesh_edges_create_tree(
+        epsilon, tree_type, axis, em, edges_mask, edges_num_active);
   }
 
-  return data->tree;
+  if (tree) {
+    memset(data, 0, sizeof(*data));
+    data->tree = tree;
+    data->em = em;
+    data->nearest_callback = NULL; /* TODO */
+    data->raycast_callback = NULL; /* TODO */
+    data->cached = bvh_cache != NULL;
+  }
+
+  return tree;
+}
+
+BVHTree *bvhtree_from_editmesh_edges(
+    BVHTreeFromEditMesh *data, BMEditMesh *em, float epsilon, int tree_type, int axis)
+{
+  return bvhtree_from_editmesh_edges_ex(data, em, NULL, -1, epsilon, tree_type, axis, 0, NULL);
 }
 
 /**
@@ -774,14 +800,41 @@ BVHTree *bvhtree_from_mesh_edges_ex(BVHTreeFromMesh *data,
                                     int edges_num_active,
                                     float epsilon,
                                     int tree_type,
-                                    int axis)
+                                    int axis,
+                                    const int bvh_cache_type,
+                                    BVHCache **bvh_cache)
 {
-  BVHTree *tree = bvhtree_from_mesh_edges_create_tree(
-      vert, edge, edges_num, edges_mask, edges_num_active, epsilon, tree_type, axis);
+  bool in_cache = false;
+  BVHTree *tree = NULL;
+  if (bvh_cache) {
+    BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_READ);
+    in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
+    BLI_rw_mutex_unlock(&cache_rwlock);
+    if (in_cache == false) {
+      BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
+      in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
+      if (in_cache) {
+        BLI_rw_mutex_unlock(&cache_rwlock);
+      }
+    }
+  }
+
+  if (in_cache == false) {
+    tree = bvhtree_from_mesh_edges_create_tree(
+        vert, edge, edges_num, edges_mask, edges_num_active, epsilon, tree_type, axis);
+
+    if (bvh_cache) {
+      /* Save on cache for later use */
+      /* printf("BVHTree built and saved on cache\n"); */
+      bvhcache_insert(bvh_cache, tree, bvh_cache_type);
+      BLI_rw_mutex_unlock(&cache_rwlock);
+      in_cache = true;
+    }
+  }
 
   /* Setup BVHTreeFromMesh */
   bvhtree_from_mesh_edges_setup_data(
-      data, tree, false, vert, vert_allocated, edge, edge_allocated);
+      data, tree, in_cache, vert, vert_allocated, edge, edge_allocated);
 
   return tree;
 }
@@ -882,14 +935,41 @@ BVHTree *bvhtree_from_mesh_faces_ex(BVHTreeFromMesh *data,
                                     int faces_num_active,
                                     float epsilon,
                                     int tree_type,
-                                    int axis)
+                                    int axis,
+                                    const int bvh_cache_type,
+                                    BVHCache **bvh_cache)
 {
-  BVHTree *tree = bvhtree_from_mesh_faces_create_tree(
-      epsilon, tree_type, axis, vert, face, numFaces, faces_mask, faces_num_active);
+  bool in_cache = false;
+  BVHTree *tree = NULL;
+  if (bvh_cache) {
+    BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_READ);
+    in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
+    BLI_rw_mutex_unlock(&cache_rwlock);
+    if (in_cache == false) {
+      BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
+      in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
+      if (in_cache) {
+        BLI_rw_mutex_unlock(&cache_rwlock);
+      }
+    }
+  }
+
+  if (in_cache == false) {
+    tree = bvhtree_from_mesh_faces_create_tree(
+        epsilon, tree_type, axis, vert, face, numFaces, faces_mask, faces_num_active);
+
+    if (bvh_cache) {
+      /* Save on cache for later use */
+      /* printf("BVHTree built and saved on cache\n"); */
+      bvhcache_insert(bvh_cache, tree, bvh_cache_type);
+      BLI_rw_mutex_unlock(&cache_rwlock);
+      in_cache = true;
+    }
+  }
 
   /* Setup BVHTreeFromMesh */
   bvhtree_from_mesh_faces_setup_data(
-      data, tree, false, vert, vert_allocated, face, face_allocated);
+      data, tree, in_cache, vert, vert_allocated, face, face_allocated);
 
   return tree;
 }
@@ -904,12 +984,11 @@ static BVHTree *bvhtree_from_editmesh_looptri_create_tree(float epsilon,
                                                           int tree_type,
                                                           int axis,
                                                           BMEditMesh *em,
-                                                          const int looptri_num,
                                                           const BLI_bitmap *looptri_mask,
                                                           int looptri_num_active)
 {
   BVHTree *tree = NULL;
-  int i;
+  const int looptri_num = em->tottri;
 
   if (looptri_num) {
     if (looptri_mask) {
@@ -923,26 +1002,24 @@ static BVHTree *bvhtree_from_editmesh_looptri_create_tree(float epsilon,
     /* printf("%s: building BVH, total=%d\n", __func__, numFaces); */
     tree = BLI_bvhtree_new(looptri_num_active, epsilon, tree_type, axis);
     if (tree) {
-      if (em) {
-        const struct BMLoop *(*looptris)[3] = (void *)em->looptris;
-
-        /* Insert BMesh-tessellation triangles into the bvh tree, unless they are hidden
-         * and/or selected. Even if the faces themselves are not selected for the snapped
-         * transform, having a vertex selected means the face (and thus it's tessellated
-         * triangles) will be moving and will not be a good snap targets. */
-        for (i = 0; i < looptri_num; i++) {
-          const BMLoop **ltri = looptris[i];
-          bool insert = looptri_mask ? BLI_BITMAP_TEST_BOOL(looptri_mask, i) : true;
-
-          if (insert) {
-            /* No reason found to block hit-testing the triangle for snap, so insert it now.*/
-            float co[3][3];
-            copy_v3_v3(co[0], ltri[0]->v->co);
-            copy_v3_v3(co[1], ltri[1]->v->co);
-            copy_v3_v3(co[2], ltri[2]->v->co);
-
-            BLI_bvhtree_insert(tree, i, co[0], 3);
-          }
+      const struct BMLoop *(*looptris)[3] = (void *)em->looptris;
+
+      /* Insert BMesh-tessellation triangles into the bvh tree, unless they are hidden
+       * and/or selected. Even if the faces themselves are not selected for the snapped
+       * transform, having a vertex selected means the face (and thus it's tessellated
+       * triangles) will be moving and will not be a good snap targets. */
+      for (int i = 0; i < looptri_num; i++) {
+        const BMLoop **ltri = looptris[i];
+        bool insert = looptri_mask ? BLI_BITMAP_TEST_BOOL(looptri_mask, i) : true;
+
+        if (insert) {
+          /* No reason found to block hit-testing the triangle for snap, so insert it now.*/
+          float co[3][3];
+          copy_v3_v3(co[0], ltri[0]->v->co);
+          copy_v3_v3(co[1], ltri[1]->v->co);
+          copy_v3_v3(co[2], ltri[2]->v->co);
+
+          BLI_bvhtree_insert(tree, i, co[0], 3);
         }
       }
       BLI_assert(BLI_bvhtree_get_len(tree) == looptri_num_active);
@@ -1035,33 +1112,34 @@ BVHTree *bvhtree_from_editmesh_looptri_ex(BVHTreeFromEditMesh *data,
                                           float epsilon,
                                           int tree_type,
                                           int axis,
-                                          BVHCache **bvhCache)
+                                          const int bvh_cache_type,
+                                          BVHCache **bvh_cache)
 {
   /* BMESH specific check that we have tessfaces,
    * we _could_ tessellate here but rather not - campbell */
 
   BVHTree *tree = NULL;
-  if (bvhCache) {
+  if (bvh_cache) {
     BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_READ);
-    bool in_cache = bvhcache_find(*bvhCache, BVHTREE_FROM_EM_LOOPTRI, &tree);
+    bool in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
     BLI_rw_mutex_unlock(&cache_rwlock);
     if (in_cache == false) {
       BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
-      in_cache = bvhcache_find(*bvhCache, BVHTREE_FROM_EM_LOOPTRI, &tree);
+      in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
       if (in_cache == false) {
         tree = bvhtree_from_editmesh_looptri_create_tree(
-            epsilon, tree_type, axis, em, em->tottri, looptri_mask, looptri_num_active);
+            epsilon, tree_type, axis, em, looptri_mask, looptri_num_active);
 
         /* Save on cache for later use */
         /* printf("BVHTree built and saved on cache\n"); */
-        bvhcache_insert(bvhCache, tree, BVHTREE_FROM_EM_LOOPTRI);
+        bvhcache_insert(bvh_cache, tree, bvh_cache_type);
       }
       BLI_rw_mutex_unlock(&cache_rwlock);
     }
   }
   else {
     tree = bvhtree_from_editmesh_looptri_create_tree(
-        epsilon, tree_type, axis, em, em->tottri, looptri_mask, looptri_num_active);
+        epsilon, tree_type, axis, em, looptri_mask, looptri_num_active);
   }
 
   if (tree) {
@@ -1069,19 +1147,15 @@ BVHTree *bvhtree_from_editmesh_looptri_ex(BVHTreeFromEditMesh *data,
     data->nearest_callback = editmesh_looptri_nearest_point;
     data->raycast_callback = editmesh_looptri_spherecast;
     data->em = em;
-    data->cached = bvhCache != NULL;
+    data->cached = bvh_cache != NULL;
   }
   return tree;
 }
 
-BVHTree *bvhtree_from_editmesh_looptri(BVHTreeFromEditMesh *data,
-                                       BMEditMesh *em,
-                                       float epsilon,
-                                       int tree_type,
-                                       int axis,
-                                       BVHCache **bvhCache)
+BVHTree *bvhtree_from_editmesh_looptri(
+    BVHTreeFromEditMesh *data, BMEditMesh *em, float epsilon, int tree_type, int axis)
 {
-  return bvhtree_from_editmesh_looptri_ex(data, em, NULL, -1, epsilon, tree_type, axis, bvhCache);
+  return bvhtree_from_editmesh_looptri_ex(data, em, NULL, -1, epsilon, tree_type, axis, 0, NULL);
 }
 
 /**
@@ -1101,21 +1175,54 @@ BVHTree *bvhtree_from_mesh_looptri_ex(BVHTreeFromMesh *data,
                                       int looptri_num_active,
                                       float epsilon,
                                       int tree_type,
-                                      int axis)
+                                      int axis,
+                                      const int bvh_cache_type,
+                                      BVHCache **bvh_cache)
 {
-  BVHTree *tree = bvhtree_from_mesh_looptri_create_tree(epsilon,
-                                                        tree_type,
-                                                        axis,
-                                                        vert,
-                                                        mloop,
-                                                        looptri,
-                                                        looptri_num,
-                                                        looptri_mask,
-                                                        looptri_num_active);
+  bool in_cache = false;
+  BVHTree *tree = NULL;
+  if (bvh_cache) {
+    BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_READ);
+    in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
+    BLI_rw_mutex_unlock(&cache_rwlock);
+    if (in_cache == false) {
+      BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
+      in_cache = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
+      if (in_cache) {
+        BLI_rw_mutex_unlock(&cache_rwlock);
+      }
+    }
+  }
+
+  if (in_cache == false) {
+    /* Setup BVHTreeFromMesh */
+    tree = bvhtree_from_mesh_looptri_create_tree(epsilon,
+                                                 tree_type,
+                                                 axis,
+                                                 vert,
+                                                 mloop,
+                                                 looptri,
+                                                 looptri_num,
+                                                 looptri_mask,
+                                                 looptri_num_active);
+
+    if (bvh_cache) {
+      bvhcache_insert(bvh_cache, tree, bvh_cache_type);
+      BLI_rw_mutex_unlock(&cache_rwlock);
+      in_cache = true;
+    }
+  }
 
   /* Setup BVHTreeFromMesh */
-  bvhtree_from_mesh_looptri_setup_data(
-      data, tree, false, vert, vert_allocated, mloop, loop_allocated, looptri, looptri_allocated);
+  bvhtree_from_mesh_looptri_setup_data(data,
+                                       tree,
+                                       in_cache,
+                                       vert,
+                                       vert_allocated,
+                                       mloop,
+                                       loop_allocated,
+                                       looptri,
+                                       looptri_allocated);
 
   return tree;
 }
@@ -1204,164 +1311,154 @@ static BLI_bitmap *looptri_no_hidden_map_get(const MPoly *mpoly,
  */
 BVHTree *BKE_bvhtree_from_mesh_get(struct BVHTreeFromMesh *data,
                                    struct Mesh *mesh,
-                                   const int type,
+                                   const int bvh_cache_type,
                                    const int tree_type)
 {
-  struct BVHTreeFromMesh data_cp = {0};
+  BVHTree *tree = NULL;
+  BVHCache **bvh_cache = &mesh->runtime.bvh_cache;
 
   BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_READ);
-  data_cp.cached = bvhcache_find(mesh->runtime.bvh_cache, type, &data_cp.tree);
+  bool is_cached = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
   BLI_rw_mutex_unlock(&cache_rwlock);
 
-  if (data_cp.cached && data_cp.tree == NULL) {
+  if (is_cached && tree == NULL) {
     memset(data, 0, sizeof(*data));
-    return data_cp.tree;
+    return tree;
   }
 
-  switch (type) {
+  switch (bvh_cache_type) {
     case BVHTREE_FROM_VERTS:
     case BVHTREE_FROM_LOOSEVERTS:
-      data_cp.raycast_callback = mesh_verts_spherecast;
-
-      data_cp.vert = mesh->mvert;
+      if (is_cached == false) {
+        BLI_bitmap *loose_verts_mask = NULL;
+        int loose_vert_len = -1;
+        int verts_len = mesh->totvert;
+
+        if (bvh_cache_type == BVHTREE_FROM_LOOSEVERTS) {
+          loose_verts_mask = loose_verts_map_get(
+              mesh->medge, mesh->totedge, mesh->mvert, verts_len, &loose_vert_len);
+        }
 
-      if (data_cp.cached == false) {
         /* TODO: a global mutex lock held during the expensive operation of
          * building the BVH tree is really bad for performance. */
-        BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
-        data_cp.cached = bvhcache_find(mesh->runtime.bvh_cache, type, &data_cp.tree);
-
-        if (data_cp.cached == false) {
-          BLI_bitmap *loose_verts_mask = NULL;
-          int loose_vert_len = -1;
-          int verts_len = mesh->totvert;
-
-          if (type == BVHTREE_FROM_LOOSEVERTS) {
-            loose_verts_mask = loose_verts_map_get(
-                mesh->medge, mesh->totedge, data_cp.vert, verts_len, &loose_vert_len);
-          }
-
-          data_cp.tree = bvhtree_from_mesh_verts_create_tree(
-              0.0, tree_type, 6, data_cp.vert, verts_len, loose_verts_mask, loose_vert_len);
-
-          if (loose_verts_mask != NULL) {
-            MEM_freeN(loose_verts_mask);
-          }
-
-          /* Save on cache for later use */
-          /* printf("BVHTree built and saved on cache\n"); */
-          bvhcache_insert(&mesh->runtime.bvh_cache, data_cp.tree, type);
+        tree = bvhtree_from_mesh_verts_ex(data,
+                                          mesh->mvert,
+                                          verts_len,
+                                          false,
+                                          loose_verts_mask,
+                                          loose_vert_len,
+                                          0.0f,
+                                          tree_type,
+                                          6,
+                                          bvh_cache_type,
+                                          bvh_cache);
+
+        if (loose_verts_mask != NULL) {
+          MEM_freeN(loose_verts_mask);
         }
-        BLI_rw_mutex_unlock(&cache_rwlock);
+      }
+      else {
+        /* Setup BVHTreeFromMesh */
+        bvhtree_from_mesh_verts_setup_data(data, tree, true, mesh->mvert, false);
       }
       break;
 
     case BVHTREE_FROM_EDGES:
     case BVHTREE_FROM_LOOSEEDGES:
-      data_cp.nearest_callback = mesh_edges_nearest_point;
-      data_cp.raycast_callback = mesh_edges_spherecast;
-
-      data_cp.vert = mesh->mvert;
-      data_cp.edge = mesh->medge;
-
-      if (data_cp.cached == false) {
-        BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
-        data_cp.cached = bvhcache_find(mesh->runtime.bvh_cache, type, &data_cp.tree);
-        if (data_cp.cached == false) {
-          BLI_bitmap *loose_edges_mask = NULL;
-          int loose_edges_len = -1;
-          int edges_len = mesh->totedge;
-
-          if (type == BVHTREE_FROM_LOOSEEDGES) {
-            loose_edges_mask = loose_edges_map_get(data_cp.edge, edges_len, &loose_edges_len);
-          }
+      if (is_cached == false) {
+        BLI_bitmap *loose_edges_mask = NULL;
+        int loose_edges_len = -1;
+        int edges_len = mesh->totedge;
 
-          data_cp.tree = bvhtree_from_mesh_edges_create_tree(data_cp.vert,
-                                                             data_cp.edge,
-                                                             edges_len,
-                                                             loose_edges_mask,
-                                                             loose_edges_len,
-                                                             0.0,
-                                                             tree_type,
-                                                             6);
-
-          if (loose_edges_mask != NULL) {
-            MEM_freeN(loose_edges_mask);
-          }
+        if (bvh_cache_type == BVHTREE_FROM_LOOSEEDGES) {
+          loose_edges_mask = loose_edges_map_get(mesh->medge, edges_len, &loose_edges_len);
+        }
 
-          /* Save on cache for later use */
-          /* printf("BVHTree built and saved on cache\n"); */
-          bvhcache_insert(&mesh->runtime.bvh_cache, data_cp.tree, type);
+        tree = bvhtree_from_mesh_edges_ex(data,
+                                          mesh->mvert,
+                                          false,
+                                          mesh->medge,
+                                          edges_len,
+                                          false,
+                                          loose_edges_mask,
+                                          loose_edges_len,
+                                          0.0,
+                                          tree_type,
+                                          6,
+                                          bvh_cache_type,
+                                          bvh_cache);
+
+        if (loose_edges_mask != NULL) {
+          MEM_freeN(loose_edges_mask);
         }
-        BLI_rw_mutex_unlock(&cache_rwlock);
+      }
+      else {
+        /* Setup BVHTreeFromMesh */
+        bvhtree_from_mesh_edges_setup_data(
+            data, tree, false, mesh->mvert, false, mesh->medge, false);
       }
       break;
 
     case BVHTREE_FROM_FACES:
-      data_cp.nearest_callback = mesh_faces_nearest_point;
-      data_cp.raycast_callback = mesh_faces_spherecast;
-
-      data_cp.vert = mesh->mvert;
-      data_cp.face = mesh->mface;
-
-      if (data_cp.cached == false) {
-        BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
-        data_cp.cached = bvhcache_find(mesh->runtime.bvh_cache, BVHTREE_FROM_FACES, &data_cp.tree);
-        if (data_cp.cached == false) {
-          int num_faces = mesh->totface;
-          BLI_assert(!(num_faces == 0 && mesh->totpoly != 0));
-
-          data_cp.tree = bvhtree_from_mesh_faces_create_tree(
-              0.0, tree_type, 6, data_cp.vert, data_cp.face, num_faces, NULL, -1);
-
-          /* Save on cache for later use */
-          /* printf("BVHTree built and saved on cache\n"); */
-          bvhcache_insert(&mesh->runtime.bvh_cache, data_cp.tree, BVHTREE_FROM_FACES);
-        }
-        BLI_rw_mutex_unlock(&cache_rwlock);
+      if (is_cached == false) {
+        int num_faces = mesh->totface;
+        BLI_assert(!(num_faces == 0 && mesh->totpoly != 0));
+
+        tree = bvhtree_from_mesh_faces_ex(data,
+                                          mesh->mvert,
+                                          false,
+                                          mesh->mface,
+                                          num_faces,
+                                          false,
+                                          NULL,
+                                          -1,
+                                          0.0,
+                                          tree_type,
+                                          6,
+                                          bvh_cache_type,
+                                          bvh_cache);
+      }
+      else {
+        /* Setup BVHTreeFromMesh */
+        bvhtree_from_mesh_faces_setup_data(
+            data, tree, true, mesh->mvert, false, mesh->mface, false);
       }
       break;
 
     case BVHTREE_FROM_LOOPTRI:
     case BVHTREE_FROM_LOOPTRI_NO_HIDDEN:
-      data_cp.nearest_callback = mesh_looptri_nearest_point;
-      data_cp.raycast_callback = mesh_looptri_spherecast;
-
-      data_cp.vert = mesh->mvert;
-      data_cp.loop = mesh->mloop;
-
-      /* TODO: store looptris somewhere? */
-      data_cp.looptri = BKE_mesh_runtime_looptri_ensure(mesh);
-
-      if (data_cp.cached == false) {
-        BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_WRITE);
-        data_cp.cached = bvhcache_find(
-            mesh->runtime.bvh_cache, BVHTREE_FROM_LOOPTRI, &data_cp.tree);
-        if (data_cp.cached == false) {
-          BLI_bitmap *looptri_mask = NULL;
-          int looptri_mask_active_len = -1;
-          int looptri_len = BKE_mesh_runtime_looptri_len(mesh);
-
-          if (type == BVHTREE_FROM_LOOPTRI_NO_HIDDEN) {
-            looptri_mask = looptri_no_hidden_map_get(
-                mesh->mpoly, looptri_len, &looptri_mask_active_len);
-          }
-
-          data_cp.tree = bvhtree_from_mesh_looptri_create_tree(0.0,
-                                                               tree_type,
-                                                               6,
-                                                               data_cp.vert,
-                                                               data_cp.loop,
-                                                               data_cp.looptri,
-                                                               looptri_len,
-                                                               looptri_mask,
-                                                               looptri_mask_active_len);
-
-          /* Save on cache for later use */
-          /* printf("BVHTree built and saved on cache\n"); */
-          bvhcache_insert(&mesh->runtime.bvh_cache, data_cp.tree, BVHTREE_FROM_LOOPTRI);
+      if (is_cached == false) {
+        const MLoopTri *mlooptri = BKE_mesh_runtime_looptri_ensure(mesh);
+        int looptri_len = BKE_mesh_runtime_looptri_len(mesh);
+
+        int looptri_mask_active_len = -1;
+        BLI_bitmap *looptri_mask = NULL;
+        if (bvh_cache_type == BVHTREE_FROM_LOOPTRI_NO_HIDDEN) {
+          looptri_mask = looptri_no_hidden_map_get(
+              mesh->mpoly, looptri_len, &looptri_mask_active_len);
         }
-        BLI_rw_mutex_unlock(&cache_rwlock);
+
+        tree = bvhtree_from_mesh_looptri_ex(data,
+                                            mesh->mvert,
+                                            false,
+                                            mesh->mloop,
+                                            false,
+                                            mlooptri,
+                                            looptri_len,
+                                            false,
+                                            looptri_mask,
+                                            looptri_mask_active_len,
+                                            0.0,
+                                            tree_type,
+                                            6,
+                                            bvh_cache_type,
+                                            bvh_cache);
+      }
+      else {
+        /* Setup BVHTreeFromMesh */
+        const MLoopTri *mlooptri = BKE_mesh_runtime_looptri_ensure(mesh);
+        bvhtree_from_mesh_looptri_setup_data(
+            data, tree, true, mesh->mvert, false, mesh->mloop, false, mlooptri, false);
       }
       break;
     case BVHTREE_FROM_EM_VERTS:
@@ -1371,23 +1468,113 @@ BVHTree *BKE_bvhtree_from_mesh_get(struct BVHTreeFromMesh *data,
       break;
   }
 
-  if (data_cp.tree != NULL) {
+  if (data->tree != NULL) {
 #ifdef DEBUG
-    if (BLI_bvhtree_get_tree_type(data_cp.tree) != tree_type) {
+    if (BLI_bvhtree_get_tree_type(data->tree) != tree_type) {
       printf("tree_type %d obtained instead of %d\n",
-             BLI_bvhtree_get_tree_type(data_cp.tree),
+             BLI_bvhtree_get_tree_type(data->tree),
              tree_type);
     }
 #endif
-    data_cp.cached = true;
-    memcpy(data, &data_cp, sizeof(*data));
+    BLI_assert(data->cached);
   }
   else {
-    free_bvhtree_from_mesh(&data_cp);
+    free_bvhtree_from_mesh(data);
     memset(data, 0, sizeof(*data));
   }
 
-  return data_cp.tree;
+  return tree;
+}
+
+/**
+ * Builds or queries a bvhcache for the cache bvhtree of the request type.
+ */
+BVHTree *BKE_bvhtree_from_editmesh_get(BVHTreeFromEditMesh *data,
+                                       struct BMEditMesh *em,
+                                       const int tree_type,
+                                       const int bvh_cache_type,
+                                       BVHCache **bvh_cache)
+{
+  BVHTree *tree = NULL;
+  bool is_cached = false;
+
+  memset(data, 0, sizeof(*data));
+
+  if (bvh_cache) {
+    BLI_rw_mutex_lock(&cache_rwlock, THREAD_LOCK_READ);
+    is_cached = bvhcache_find(*bvh_cache, bvh_cache_type, &tree);
+    BLI_rw_mutex_unlock(&cache_rwlock);
+
+    if (is_cached && tree == NULL) {
+      return tree;
+    }
+  }
+  data->tree = tree;
+  data->em = em;
+  data->cached = is_cached;
+
+  switch (bvh_cache_type) {
+    case BVHTREE_FROM_EM_VERTS:
+      if (is_cached == false) {
+        tree = bvhtree_from_editmesh_verts_ex(
+            data, em, NULL, -1, 0.0f, tree_type, 6, bvh_cache_type, bvh_cache);
+      }
+      else {
+        data->nearest_callback = NULL;
+        data->raycast_callback = editmesh_verts_spherecast;
+      }
+      break;
+
+    case BVHTREE_FROM_EM_EDGES:
+      if (is_cached == false) {
+        tree = bvhtree_from_editmesh_edges_ex(
+            data, em, NULL, -1, 0.0f, tree_type, 6, bvh_cache_type, bvh_cache);
+      }
+      else {
+        /* Setup BVHTreeFromMesh */
+        data->nearest_callback = NULL; /* TODO */
+        data->raycast_callback = NULL; /* TODO */
+      }
+      break;
+
+    case BVHTREE_FROM_EM_LOOPTRI:
+      if (is_cached == false) {
+        tree = bvhtree_from_editmesh_looptri_ex(
+            data, em, NULL, -1, 0.0f, tree_type, 6, bvh_cache_type, bvh_cache);
+      }
+      else {
+        /* Setup BVHTreeFromMesh */
+        data->nearest_callback = editmesh_looptri_nearest_point;
+        data->raycast_callback = editmesh_looptri_spherecast;
+      }
+      break;
+    case BVHTREE_FROM_VERTS:
+    case BVHTREE_FROM_EDGES:
+    case BVHTREE_FROM_FACES:
+    case BVHTREE_FROM_LOOPTRI:
+    case BVHTREE_FROM_LOOPTRI_NO_HIDDEN:
+    case BVHTREE_FROM_LOOSEVERTS:
+    case BVHTREE_FROM_LOOSEEDGES:
+      BLI_assert(false);
+      break;
+  }
+
+  if (data->tree != NULL) {
+#ifdef DEBUG
+    if (BLI_bvhtree_get_tree_type(data->tree) != tree_type) {
+      printf("tree_type %d obtained instead of %d\n",
+             BLI_bvhtree_get_tree_type(data->tree),
+             tree_type);
+    }
+#endif
+    BLI_assert(data->cached);
+  }
+  else {
+    free_bvhtree_from_editmesh(data);
+    memset(data, 0, sizeof(*data));
+  }
+
+  return tree;
 }
 
 /** \} */
@@ -1473,12 +1660,13 @@ bool bvhcache_has_tree(const BVHCache *cache, const BVHTree *tree)
  * as that will be done when the cache is freed.
  *
  * A call to this assumes that there was no previous cached tree of the given type
+ * \warning The #BVHTree can be NULL.
  */
 void bvhcache_insert(BVHCache **cache_p, BVHTree *tree, int type)
 {
   BVHCacheItem *item = NULL;
 
-  assert(bvhcache_find(*cache_p, type, &(BVHTree *){0}) == false);
+  BLI_assert(bvhcache_find(*cache_p, type, &(BVHTree *){0}) == false);
 
   item = MEM_mallocN(sizeof(BVHCacheItem), "BVHCacheItem");
 
diff --git a/source/blender/blenkernel/intern/cdderivedmesh.c b/source/blender/blenkernel/intern/cdderivedmesh.c
index 28bd9c0cea5..a55b5dc7817 100644
--- a/source/blender/blenkernel/intern/cdderivedmesh.c
+++ b/source/blender/blenkernel/intern/cdderivedmesh.c
@@ -173,16 +173,6 @@ static void cdDM_getVertCo(DerivedMesh *dm, int index, float r_co[3])
   copy_v3_v3(r_co, cddm->mvert[index].co);
 }
 
-static void cdDM_getVertCos(DerivedMesh *dm, float (*r_cos)[3])
-{
-  MVert *mv = CDDM_get_verts(dm);
-  int i;
-
-  for (i = 0; i < dm->numVertData; i++, mv++) {
-    copy_v3_v3(r_cos[i], mv->co);
-  }
-}
-
 static void cdDM_getVertNo(DerivedMesh *dm, int index, float r_no[3])
 {
   CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
@@ -203,278 +193,6 @@ static const MeshElemMap *cdDM_getPolyMap(Object *ob, DerivedMesh *dm)
   return cddm->pmap;
 }
 
-static bool check_sculpt_object_deformed(Object *object, bool for_construction)
-{
-  bool deformed = false;
-
-  /* Active modifiers means extra deformation, which can't be handled correct
-   * on birth of PBVH and sculpt "layer" levels, so use PBVH only for internal brush
-   * stuff and show final DerivedMesh so user would see actual object shape.
-   */
-  deformed |= object->sculpt->modifiers_active;
-
-  if (for_construction) {
-    deformed |= object->sculpt->kb != NULL;
-  }
-  else {
-    /* As in case with modifiers, we can't synchronize deformation made against
-     * PBVH and non-locked keyblock, so also use PBVH only for brushes and
-     * final DM to give final result to user.
-     */
-    deformed |= object->sculpt->kb && (object->shapeflag & OB_SHAPE_LOCK) == 0;
-  }
-
-  return deformed;
-}
-
-static bool can_pbvh_draw(Object *ob, DerivedMesh *dm)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-  Mesh *me = ob->data;
-  bool deformed = check_sculpt_object_deformed(ob, false);
-
-  if (deformed) {
-    return false;
-  }
-
-  return cddm->mvert == me->mvert || ob->sculpt->kb;
-}
-
-static PBVH *cdDM_getPBVH(Object *ob, DerivedMesh *dm)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-
-  if (!ob) {
-    cddm->pbvh = NULL;
-    return NULL;
-  }
-
-  if (!ob->sculpt) {
-    return NULL;
-  }
-
-  if (ob->sculpt->pbvh) {
-    cddm->pbvh = ob->sculpt->pbvh;
-    cddm->pbvh_draw = can_pbvh_draw(ob, dm);
-  }
-
-  /* Sculpting on a BMesh (dynamic-topology) gets a special PBVH */
-  if (!cddm->pbvh && ob->sculpt->bm) {
-    cddm->pbvh = BKE_pbvh_new();
-    cddm->pbvh_draw = true;
-
-    BKE_pbvh_build_bmesh(cddm->pbvh,
-                         ob->sculpt->bm,
-                         ob->sculpt->bm_smooth_shading,
-                         ob->sculpt->bm_log,
-                         ob->sculpt->cd_vert_node_offset,
-                         ob->sculpt->cd_face_node_offset);
-
-    pbvh_show_mask_set(cddm->pbvh, ob->sculpt->show_mask);
-  }
-
-  /* always build pbvh from original mesh, and only use it for drawing if
-   * this derivedmesh is just original mesh. it's the multires subsurf dm
-   * that this is actually for, to support a pbvh on a modified mesh */
-  if (!cddm->pbvh && ob->type == OB_MESH) {
-    Mesh *me = BKE_object_get_original_mesh(ob);
-    const int looptris_num = poly_to_tri_count(me->totpoly, me->totloop);
-    MLoopTri *looptri;
-    bool deformed;
-
-    cddm->pbvh = BKE_pbvh_new();
-    cddm->pbvh_draw = can_pbvh_draw(ob, dm);
-
-    looptri = MEM_malloc_arrayN(looptris_num, sizeof(*looptri), __func__);
-
-    BKE_mesh_recalc_looptri(me->mloop, me->mpoly, me->mvert, me->totloop, me->totpoly, looptri);
-
-    BKE_pbvh_build_mesh(cddm->pbvh,
-                        me->mpoly,
-                        me->mloop,
-                        me->mvert,
-                        me->totvert,
-                        &me->vdata,
-                        &me->ldata,
-                        looptri,
-                        looptris_num);
-
-    pbvh_show_mask_set(cddm->pbvh, ob->sculpt->show_mask);
-
-    deformed = check_sculpt_object_deformed(ob, true);
-
-    if (deformed && ob->derivedDeform) {
-      DerivedMesh *deformdm = ob->derivedDeform;
-      float(*vertCos)[3];
-      int totvert;
-
-      totvert = deformdm->getNumVerts(deformdm);
-      vertCos = MEM_malloc_arrayN(totvert, sizeof(float[3]), "cdDM_getPBVH vertCos");
-      deformdm->getVertCos(deformdm, vertCos);
-      BKE_pbvh_apply_vertCos(cddm->pbvh, vertCos, totvert);
-      MEM_freeN(vertCos);
-    }
-  }
-
-  return cddm->pbvh;
-}
-
-static void cdDM_foreachMappedVert(DerivedMesh *dm,
-                                   void (*func)(void *userData,
-                                                int index,
-                                                const float co[3],
-                                                const float no_f[3],
-                                                const short no_s[3]),
-                                   void *userData,
-                                   DMForeachFlag flag)
-{
-  MVert *mv = CDDM_get_verts(dm);
-  const int *index = DM_get_vert_data_layer(dm, CD_ORIGINDEX);
-  int i;
-
-  if (index) {
-    for (i = 0; i < dm->numVertData; i++, mv++) {
-      const short *no = (flag & DM_FOREACH_USE_NORMAL) ? mv->no : NULL;
-      const int orig = *index++;
-      if (orig == ORIGINDEX_NONE) {
-        continue;
-      }
-      func(userData, orig, mv->co, NULL, no);
-    }
-  }
-  else {
-    for (i = 0; i < dm->numVertData; i++, mv++) {
-      const short *no = (flag & DM_FOREACH_USE_NORMAL) ? mv->no : NULL;
-      func(userData, i, mv->co, NULL, no);
-    }
-  }
-}
-
-static void cdDM_foreachMappedEdge(
-    DerivedMesh *dm,
-    void (*func)(void *userData, int index, const float v0co[3], const float v1co[3]),
-    void *userData)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-  MVert *mv = cddm->mvert;
-  MEdge *med = cddm->medge;
-  int i, orig, *index = DM_get_edge_data_layer(dm, CD_ORIGINDEX);
-
-  for (i = 0; i < dm->numEdgeData; i++, med++) {
-    if (index) {
-      orig = *index++;
-      if (orig == ORIGINDEX_NONE) {
-        continue;
-      }
-      func(userData, orig, mv[med->v1].co, mv[med->v2].co);
-    }
-    else {
-      func(userData, i, mv[med->v1].co, mv[med->v2].co);
-    }
-  }
-}
-
-static void cdDM_foreachMappedLoop(DerivedMesh *dm,
-                                   void (*func)(void *userData,
-                                                int vertex_index,
-                                                int face_index,
-                                                const float co[3],
-                                                const float no[3]),
-                                   void *userData,
-                                   DMForeachFlag flag)
-{
-  /* We can't use dm->getLoopDataLayout(dm) here,
-   * we want to always access dm->loopData, EditDerivedBMesh would
-   * return loop data from bmesh itself. */
-  const float(*lnors)[3] = (flag & DM_FOREACH_USE_NORMAL) ? DM_get_loop_data_layer(dm, CD_NORMAL) :
-                                                            NULL;
-
-  const MVert *mv = CDDM_get_verts(dm);
-  const MLoop *ml = CDDM_get_loops(dm);
-  const MPoly *mp = CDDM_get_polys(dm);
-  const int *v_index = DM_get_vert_data_layer(dm, CD_ORIGINDEX);
-  const int *f_index = DM_get_poly_data_layer(dm, CD_ORIGINDEX);
-  int p_idx, i;
-
-  for (p_idx = 0; p_idx < dm->numPolyData; ++p_idx, ++mp) {
-    for (i = 0; i < mp->totloop; ++i, ++ml) {
-      const int v_idx = v_index ? v_index[ml->v] : ml->v;
-      const int f_idx = f_index ? f_index[p_idx] : p_idx;
-      const float *no = lnors ? *lnors++ : NULL;
-      if (!ELEM(ORIGINDEX_NONE, v_idx, f_idx)) {
-        func(userData, v_idx, f_idx, mv[ml->v].co, no);
-      }
-    }
-  }
-}
-
-static void cdDM_foreachMappedFaceCenter(
-    DerivedMesh *dm,
-    void (*func)(void *userData, int index, const float cent[3], const float no[3]),
-    void *userData,
-    DMForeachFlag flag)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-  MVert *mvert = cddm->mvert;
-  MPoly *mp;
-  MLoop *ml;
-  int i, orig, *index;
-
-  index = CustomData_get_layer(&dm->polyData, CD_ORIGINDEX);
-  mp = cddm->mpoly;
-  for (i = 0; i < dm->numPolyData; i++, mp++) {
-    float cent[3];
-    float *no, _no[3];
-
-    if (index) {
-      orig = *index++;
-      if (orig == ORIGINDEX_NONE) {
-        continue;
-      }
-    }
-    else {
-      orig = i;
-    }
-
-    ml = &cddm->mloop[mp->loopstart];
-    BKE_mesh_calc_poly_center(mp, ml, mvert, cent);
-
-    if (flag & DM_FOREACH_USE_NORMAL) {
-      BKE_mesh_calc_poly_normal(mp, ml, mvert, (no = _no));
-    }
-    else {
-      no = NULL;
-    }
-
-    func(userData, orig, cent, no);
-  }
-}
-
-void CDDM_recalc_tessellation_ex(DerivedMesh *dm, const bool do_face_nor_cpy)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-
-  dm->numTessFaceData = BKE_mesh_recalc_tessellation(&dm->faceData,
-                                                     &dm->loopData,
-                                                     &dm->polyData,
-                                                     cddm->mvert,
-                                                     dm->numTessFaceData,
-                                                     dm->numLoopData,
-                                                     dm->numPolyData,
-                                                     do_face_nor_cpy);
-
-  cddm->mface = CustomData_get_layer(&dm->faceData, CD_MFACE);
-
-  /* Tessellation recreated faceData, and the active layer indices need to get re-propagated
-   * from loops and polys to faces */
-  CustomData_bmesh_update_active_layers(&dm->faceData, &dm->loopData);
-}
-
-void CDDM_recalc_tessellation(DerivedMesh *dm)
-{
-  CDDM_recalc_tessellation_ex(dm, true);
-}
-
 void CDDM_recalc_looptri(DerivedMesh *dm)
 {
   CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
@@ -547,25 +265,13 @@ static CDDerivedMesh *cdDM_create(const char *desc)
   dm->getEdgeDataArray = DM_get_edge_data_layer;
   dm->getTessFaceDataArray = DM_get_tessface_data_layer;
 
-  dm->calcNormals = CDDM_calc_normals;
-  dm->calcLoopNormals = CDDM_calc_loop_normals;
-  dm->calcLoopNormalsSpaceArray = CDDM_calc_loop_normals_spacearr;
-  dm->calcLoopTangents = DM_calc_loop_tangents;
-  dm->recalcTessellation = CDDM_recalc_tessellation;
   dm->recalcLoopTri = CDDM_recalc_looptri;
 
-  dm->getVertCos = cdDM_getVertCos;
   dm->getVertCo = cdDM_getVertCo;
   dm->getVertNo = cdDM_getVertNo;
 
-  dm->getPBVH = cdDM_getPBVH;
   dm->getPolyMap = cdDM_getPolyMap;
 
-  dm->foreachMappedVert = cdDM_foreachMappedVert;
-  dm->foreachMappedEdge = cdDM_foreachMappedEdge;
-  dm->foreachMappedLoop = cdDM_foreachMappedLoop;
-  dm->foreachMappedFaceCenter = cdDM_foreachMappedFaceCenter;
-
   dm->release = cdDM_release;
 
   return cddm;
@@ -664,123 +370,12 @@ DerivedMesh *CDDM_from_mesh_ex(Mesh *mesh,
   return dm;
 }
 
-DerivedMesh *CDDM_from_curve(Object *ob)
-{
-  ListBase disp = {NULL, NULL};
-
-  if (ob->runtime.curve_cache) {
-    disp = ob->runtime.curve_cache->disp;
-  }
-
-  return CDDM_from_curve_displist(ob, &disp);
-}
-
-DerivedMesh *CDDM_from_curve_displist(Object *ob, ListBase *dispbase)
-{
-  Curve *cu = (Curve *)ob->data;
-  DerivedMesh *dm;
-  CDDerivedMesh *cddm;
-  MVert *allvert;
-  MEdge *alledge;
-  MLoop *allloop;
-  MPoly *allpoly;
-  MLoopUV *alluv = NULL;
-  int totvert, totedge, totloop, totpoly;
-  bool use_orco_uv = (cu->flag & CU_UV_ORCO) != 0;
-
-  if (BKE_mesh_nurbs_displist_to_mdata(ob,
-                                       dispbase,
-                                       &allvert,
-                                       &totvert,
-                                       &alledge,
-                                       &totedge,
-                                       &allloop,
-                                       &allpoly,
-                                       (use_orco_uv) ? &alluv : NULL,
-                                       &totloop,
-                                       &totpoly) != 0) {
-    /* Error initializing mdata. This often happens when curve is empty */
-    return CDDM_new(0, 0, 0, 0, 0);
-  }
-
-  dm = CDDM_new(totvert, totedge, 0, totloop, totpoly);
-  dm->deformedOnly = 1;
-  dm->dirty |= DM_DIRTY_NORMALS;
-
-  cddm = (CDDerivedMesh *)dm;
-
-  memcpy(cddm->mvert, allvert, totvert * sizeof(MVert));
-  memcpy(cddm->medge, alledge, totedge * sizeof(MEdge));
-  memcpy(cddm->mloop, allloop, totloop * sizeof(MLoop));
-  memcpy(cddm->mpoly, allpoly, totpoly * sizeof(MPoly));
-
-  if (alluv) {
-    const char *uvname = "Orco";
-    CustomData_add_layer_named(&cddm->dm.loopData, CD_MLOOPUV, CD_ASSIGN, alluv, totloop, uvname);
-  }
-
-  MEM_freeN(allvert);
-  MEM_freeN(alledge);
-  MEM_freeN(allloop);
-  MEM_freeN(allpoly);
-
-  return dm;
-}
-
-static void loops_to_customdata_corners(
-    BMesh *bm, CustomData *facedata, int cdindex, const BMLoop *l3[3], int numCol, int numUV)
-{
-  const BMLoop *l;
-  //  BMFace *f = l3[0]->f;
-  MTFace *texface;
-  MCol *mcol;
-  MLoopCol *mloopcol;
-  MLoopUV *mloopuv;
-  int i, j, hasPCol = CustomData_has_layer(&bm->ldata, CD_PREVIEW_MLOOPCOL);
-
-  for (i = 0; i < numUV; i++) {
-    texface = CustomData_get_n(facedata, CD_MTFACE, cdindex, i);
-
-    for (j = 0; j < 3; j++) {
-      l = l3[j];
-      mloopuv = CustomData_bmesh_get_n(&bm->ldata, l->head.data, CD_MLOOPUV, i);
-      copy_v2_v2(texface->uv[j], mloopuv->uv);
-    }
-  }
-
-  for (i = 0; i < numCol; i++) {
-    mcol = CustomData_get_n(facedata, CD_MCOL, cdindex, i);
-
-    for (j = 0; j < 3; j++) {
-      l = l3[j];
-      mloopcol = CustomData_bmesh_get_n(&bm->ldata, l->head.data, CD_MLOOPCOL, i);
-      MESH_MLOOPCOL_TO_MCOL(mloopcol, &mcol[j]);
-    }
-  }
-
-  if (hasPCol) {
-    mcol = CustomData_get(facedata, cdindex, CD_PREVIEW_MCOL);
-
-    for (j = 0; j < 3; j++) {
-      l = l3[j];
-      mloopcol = CustomData_bmesh_get(&bm->ldata, l->head.data, CD_PREVIEW_MLOOPCOL);
-      MESH_MLOOPCOL_TO_MCOL(mloopcol, &mcol[j]);
-    }
-  }
-}
-
 /* TODO(campbell): remove, use BKE_mesh_from_bmesh_for_eval_nomain instead. */
 
 /* used for both editbmesh and bmesh */
-static DerivedMesh *cddm_from_bmesh_ex(struct BMesh *bm,
-                                       const bool use_mdisps,
-                                       /* EditBMesh vars for use_tessface */
-                                       const bool use_tessface,
-                                       const int em_tottri,
-                                       const BMLoop *(*em_looptris)[3])
+static DerivedMesh *cddm_from_bmesh_ex(struct BMesh *bm, const bool use_mdisps)
 {
-  DerivedMesh *dm = CDDM_new(
-      bm->totvert, bm->totedge, use_tessface ? em_tottri : 0, bm->totloop, bm->totface);
+  DerivedMesh *dm = CDDM_new(bm->totvert, bm->totedge, 0, bm->totloop, bm->totface);
 
   CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
   BMIter iter;
@@ -789,11 +384,8 @@ static DerivedMesh *cddm_from_bmesh_ex(struct BMesh *bm,
   BMFace *efa;
   MVert *mvert = cddm->mvert;
   MEdge *medge = cddm->medge;
-  MFace *mface = cddm->mface;
   MLoop *mloop = cddm->mloop;
   MPoly *mpoly = cddm->mpoly;
-  int numCol = CustomData_number_of_layers(&bm->ldata, CD_MLOOPCOL);
-  int numUV = CustomData_number_of_layers(&bm->ldata, CD_MLOOPUV);
   int *index, add_orig;
   CustomData_MeshMasks mask = {0};
   unsigned int i, j;
@@ -820,11 +412,6 @@ static DerivedMesh *cddm_from_bmesh_ex(struct BMesh *bm,
   CustomData_merge(&bm->ldata, &dm->loopData, mask.lmask, CD_CALLOC, dm->numLoopData);
   CustomData_merge(&bm->pdata, &dm->polyData, mask.pmask, CD_CALLOC, dm->numPolyData);
 
-  /* add tessellation mface layers */
-  if (use_tessface) {
-    CustomData_from_bmeshpoly(&dm->faceData, &dm->loopData, em_tottri);
-  }
-
   index = dm->getVertDataArray(dm, CD_ORIGINDEX);
 
   BM_ITER_MESH_INDEX (eve, &iter, bm, BM_VERTS_OF_MESH, i) {
@@ -883,32 +470,6 @@ static DerivedMesh *cddm_from_bmesh_ex(struct BMesh *bm,
   }
   bm->elem_index_dirty &= ~BM_EDGE;
 
-  /* avoid this where possiblem, takes extra memory */
-  if (use_tessface) {
-
-    BM_mesh_elem_index_ensure(bm, BM_FACE);
-
-    index = dm->getTessFaceDataArray(dm, CD_ORIGINDEX);
-    for (i = 0; i < dm->numTessFaceData; i++) {
-      MFace *mf = &mface[i];
-      const BMLoop **l = em_looptris[i];
-      efa = l[0]->f;
-
-      mf->v1 = BM_elem_index_get(l[0]->v);
-      mf->v2 = BM_elem_index_get(l[1]->v);
-      mf->v3 = BM_elem_index_get(l[2]->v);
-      mf->v4 = 0;
-      mf->mat_nr = efa->mat_nr;
-      mf->flag = BM_face_flag_to_mflag(efa);
-
-      /* map mfaces to polygons in the same cddm intentionally */
-      *index++ = BM_elem_index_get(efa);
-
-      loops_to_customdata_corners(bm, &dm->faceData, i, l, numCol, numUV);
-      test_index_face(mf, &dm->faceData, i, 3);
-    }
-  }
-
   index = CustomData_get_layer(&dm->polyData, CD_ORIGINDEX);
   j = 0;
   BM_ITER_MESH_INDEX (efa, &iter, bm, BM_FACES_OF_MESH, i) {
@@ -948,24 +509,9 @@ static DerivedMesh *cddm_from_bmesh_ex(struct BMesh *bm,
   return dm;
 }
 
-struct DerivedMesh *CDDM_from_bmesh(struct BMesh *bm, const bool use_mdisps)
+DerivedMesh *CDDM_from_editbmesh(BMEditMesh *em, const bool use_mdisps)
 {
-  return cddm_from_bmesh_ex(bm,
-                            use_mdisps,
-                            false,
-                            /* these vars are for editmesh only */
-                            0,
-                            NULL);
-}
-
-DerivedMesh *CDDM_from_editbmesh(BMEditMesh *em, const bool use_mdisps, const bool use_tessface)
-{
-  return cddm_from_bmesh_ex(em->bm,
-                            use_mdisps,
-                            /* editmesh */
-                            use_tessface,
-                            em->tottri,
-                            (const BMLoop *(*)[3])em->looptris);
+  return cddm_from_bmesh_ex(em->bm, use_mdisps);
 }
 
 DerivedMesh *CDDM_copy(DerivedMesh *source)
@@ -1014,366 +560,11 @@ DerivedMesh *CDDM_copy(DerivedMesh *source)
   return dm;
 }
 
-/* note, the CD_ORIGINDEX layers are all 0, so if there is a direct
- * relationship between mesh data this needs to be set by the caller. */
-DerivedMesh *CDDM_from_template_ex(DerivedMesh *source,
-                                   int numVerts,
-                                   int numEdges,
-                                   int numTessFaces,
-                                   int numLoops,
-                                   int numPolys,
-                                   const CustomData_MeshMasks *mask)
-{
-  CDDerivedMesh *cddm = cdDM_create("CDDM_from_template dest");
-  DerivedMesh *dm = &cddm->dm;
-
-  /* ensure these are created if they are made on demand */
-  source->getVertDataArray(source, CD_ORIGINDEX);
-  source->getEdgeDataArray(source, CD_ORIGINDEX);
-  source->getTessFaceDataArray(source, CD_ORIGINDEX);
-  source->getPolyDataArray(source, CD_ORIGINDEX);
-
-  /* this does a copy of all non mvert/medge/mface layers */
-  DM_from_template_ex(
-      dm, source, DM_TYPE_CDDM, numVerts, numEdges, numTessFaces, numLoops, numPolys, mask);
-
-  /* now add mvert/medge/mface layers */
-  CustomData_add_layer(&dm->vertData, CD_MVERT, CD_CALLOC, NULL, numVerts);
-  CustomData_add_layer(&dm->edgeData, CD_MEDGE, CD_CALLOC, NULL, numEdges);
-  CustomData_add_layer(&dm->faceData, CD_MFACE, CD_CALLOC, NULL, numTessFaces);
-  CustomData_add_layer(&dm->loopData, CD_MLOOP, CD_CALLOC, NULL, numLoops);
-  CustomData_add_layer(&dm->polyData, CD_MPOLY, CD_CALLOC, NULL, numPolys);
-
-  if (!CustomData_get_layer(&dm->vertData, CD_ORIGINDEX)) {
-    CustomData_add_layer(&dm->vertData, CD_ORIGINDEX, CD_CALLOC, NULL, numVerts);
-  }
-  if (!CustomData_get_layer(&dm->edgeData, CD_ORIGINDEX)) {
-    CustomData_add_layer(&dm->edgeData, CD_ORIGINDEX, CD_CALLOC, NULL, numEdges);
-  }
-  if (!CustomData_get_layer(&dm->faceData, CD_ORIGINDEX)) {
-    CustomData_add_layer(&dm->faceData, CD_ORIGINDEX, CD_CALLOC, NULL, numTessFaces);
-  }
-
-  cddm->mvert = CustomData_get_layer(&dm->vertData, CD_MVERT);
-  cddm->medge = CustomData_get_layer(&dm->edgeData, CD_MEDGE);
-  cddm->mface = CustomData_get_layer(&dm->faceData, CD_MFACE);
-  cddm->mloop = CustomData_get_layer(&dm->loopData, CD_MLOOP);
-  cddm->mpoly = CustomData_get_layer(&dm->polyData, CD_MPOLY);
-
-  return dm;
-}
-DerivedMesh *CDDM_from_template(
-    DerivedMesh *source, int numVerts, int numEdges, int numTessFaces, int numLoops, int numPolys)
-{
-  return CDDM_from_template_ex(
-      source, numVerts, numEdges, numTessFaces, numLoops, numPolys, &CD_MASK_DERIVEDMESH);
-}
-
-void CDDM_apply_vert_coords(DerivedMesh *dm, float (*vertCoords)[3])
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-  MVert *vert;
-  int i;
-
-  /* this will just return the pointer if it wasn't a referenced layer */
-  vert = CustomData_duplicate_referenced_layer(&dm->vertData, CD_MVERT, dm->numVertData);
-  cddm->mvert = vert;
-
-  for (i = 0; i < dm->numVertData; ++i, ++vert) {
-    copy_v3_v3(vert->co, vertCoords[i]);
-  }
-
-  cddm->dm.dirty |= DM_DIRTY_NORMALS;
-}
-
-void CDDM_apply_vert_normals(DerivedMesh *dm, short (*vertNormals)[3])
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-  MVert *vert;
-  int i;
-
-  /* this will just return the pointer if it wasn't a referenced layer */
-  vert = CustomData_duplicate_referenced_layer(&dm->vertData, CD_MVERT, dm->numVertData);
-  cddm->mvert = vert;
-
-  for (i = 0; i < dm->numVertData; ++i, ++vert) {
-    copy_v3_v3_short(vert->no, vertNormals[i]);
-  }
-
-  cddm->dm.dirty &= ~DM_DIRTY_NORMALS;
-}
-
-void CDDM_calc_normals_mapping_ex(DerivedMesh *dm, const bool only_face_normals)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-  float(*face_nors)[3] = NULL;
-
-  if (dm->numVertData == 0) {
-    cddm->dm.dirty &= ~DM_DIRTY_NORMALS;
-    return;
-  }
-
-  /* now we skip calculating vertex normals for referenced layer,
-   * no need to duplicate verts.
-   * WATCH THIS, bmesh only change!,
-   * need to take care of the side effects here - campbell */
-#if 0
-  /* we don't want to overwrite any referenced layers */
-  cddm->mvert = CustomData_duplicate_referenced_layer(&dm->vertData, CD_MVERT, dm->numVertData);
-#endif
-
-#if 0
-  if (dm->numTessFaceData == 0) {
-    /* No tessellation on this mesh yet, need to calculate one.
-     *
-     * Important not to update face normals from polys since it
-     * interferes with assigning the new normal layer in the following code.
-     */
-    CDDM_recalc_tessellation_ex(dm, false);
-  }
-  else {
-    /* A tessellation already exists, it should always have a CD_ORIGINDEX */
-    BLI_assert(CustomData_has_layer(&dm->faceData, CD_ORIGINDEX));
-    CustomData_free_layers(&dm->faceData, CD_NORMAL, dm->numTessFaceData);
-  }
-#endif
-
-  face_nors = MEM_malloc_arrayN(dm->numPolyData, sizeof(*face_nors), "face_nors");
-
-  /* calculate face normals */
-  BKE_mesh_calc_normals_poly(cddm->mvert,
-                             NULL,
-                             dm->numVertData,
-                             CDDM_get_loops(dm),
-                             CDDM_get_polys(dm),
-                             dm->numLoopData,
-                             dm->numPolyData,
-                             face_nors,
-                             only_face_normals);
-
-  CustomData_add_layer(&dm->polyData, CD_NORMAL, CD_ASSIGN, face_nors, dm->numPolyData);
-
-  cddm->dm.dirty &= ~DM_DIRTY_NORMALS;
-}
-
-void CDDM_calc_normals_mapping(DerivedMesh *dm)
-{
-  /* use this to skip calculating normals on original vert's, this may need to be changed */
-  const bool only_face_normals = CustomData_is_referenced_layer(&dm->vertData, CD_MVERT);
-
-  CDDM_calc_normals_mapping_ex(dm, only_face_normals);
-}
-
-#if 0
-/* bmesh note: this matches what we have in trunk */
-void CDDM_calc_normals(DerivedMesh *dm)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-  float(*poly_nors)[3];
-
-  if (dm->numVertData == 0) {
-    return;
-  }
-
-  /* we don't want to overwrite any referenced layers */
-  cddm->mvert = CustomData_duplicate_referenced_layer(&dm->vertData, CD_MVERT, dm->numVertData);
-
-  /* fill in if it exists */
-  poly_nors = CustomData_get_layer(&dm->polyData, CD_NORMAL);
-  if (!poly_nors) {
-    poly_nors = CustomData_add_layer(&dm->polyData, CD_NORMAL, CD_CALLOC, NULL, dm->numPolyData);
-  }
-
-  BKE_mesh_calc_normals_poly(cddm->mvert,
-                             dm->numVertData,
-                             CDDM_get_loops(dm),
-                             CDDM_get_polys(dm),
-                             dm->numLoopData,
-                             dm->numPolyData,
-                             poly_nors,
-                             false);
-
-  cddm->dm.dirty &= ~DM_DIRTY_NORMALS;
-}
-#else
-
-/* poly normal layer is now only for final display */
-void CDDM_calc_normals(DerivedMesh *dm)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-
-  /* we don't want to overwrite any referenced layers */
-  cddm->mvert = CustomData_duplicate_referenced_layer(&dm->vertData, CD_MVERT, dm->numVertData);
-
-  BKE_mesh_calc_normals_poly(cddm->mvert,
-                             NULL,
-                             dm->numVertData,
-                             CDDM_get_loops(dm),
-                             CDDM_get_polys(dm),
-                             dm->numLoopData,
-                             dm->numPolyData,
-                             NULL,
-                             false);
-
-  cddm->dm.dirty &= ~DM_DIRTY_NORMALS;
-}
-
-#endif
-
-void CDDM_calc_loop_normals(DerivedMesh *dm, const bool use_split_normals, const float split_angle)
-{
-  CDDM_calc_loop_normals_spacearr(dm, use_split_normals, split_angle, NULL);
-}
-
 /* #define DEBUG_CLNORS */
 #ifdef DEBUG_CLNORS
 #  include "BLI_linklist.h"
 #endif
 
-void CDDM_calc_loop_normals_spacearr(DerivedMesh *dm,
-                                     const bool use_split_normals,
-                                     const float split_angle,
-                                     MLoopNorSpaceArray *r_lnors_spacearr)
-{
-  MVert *mverts = dm->getVertArray(dm);
-  MEdge *medges = dm->getEdgeArray(dm);
-  MLoop *mloops = dm->getLoopArray(dm);
-  MPoly *mpolys = dm->getPolyArray(dm);
-
-  CustomData *ldata, *pdata;
-
-  float(*lnors)[3];
-  short(*clnor_data)[2];
-  float(*pnors)[3];
-
-  const int numVerts = dm->getNumVerts(dm);
-  const int numEdges = dm->getNumEdges(dm);
-  const int numLoops = dm->getNumLoops(dm);
-  const int numPolys = dm->getNumPolys(dm);
-
-  ldata = dm->getLoopDataLayout(dm);
-  if (CustomData_has_layer(ldata, CD_NORMAL)) {
-    lnors = CustomData_get_layer(ldata, CD_NORMAL);
-  }
-  else {
-    lnors = CustomData_add_layer(ldata, CD_NORMAL, CD_CALLOC, NULL, numLoops);
-  }
-
-  /* Compute poly (always needed) and vert normals. */
-  /* Note we can't use DM_ensure_normals, since it won't keep computed poly nors... */
-  pdata = dm->getPolyDataLayout(dm);
-  pnors = CustomData_get_layer(pdata, CD_NORMAL);
-  if (!pnors) {
-    pnors = CustomData_add_layer(pdata, CD_NORMAL, CD_CALLOC, NULL, numPolys);
-  }
-  BKE_mesh_calc_normals_poly(mverts,
-                             NULL,
-                             numVerts,
-                             mloops,
-                             mpolys,
-                             numLoops,
-                             numPolys,
-                             pnors,
-                             (dm->dirty & DM_DIRTY_NORMALS) ? false : true);
-
-  dm->dirty &= ~DM_DIRTY_NORMALS;
-
-  clnor_data = CustomData_get_layer(ldata, CD_CUSTOMLOOPNORMAL);
-
-  BKE_mesh_normals_loop_split(mverts,
-                              numVerts,
-                              medges,
-                              numEdges,
-                              mloops,
-                              lnors,
-                              numLoops,
-                              mpolys,
-                              (const float(*)[3])pnors,
-                              numPolys,
-                              use_split_normals,
-                              split_angle,
-                              r_lnors_spacearr,
-                              clnor_data,
-                              NULL);
-#ifdef DEBUG_CLNORS
-  if (r_lnors_spacearr) {
-    int i;
-    for (i = 0; i < numLoops; i++) {
-      if (r_lnors_spacearr->lspacearr[i]->ref_alpha != 0.0f) {
-        LinkNode *loops = r_lnors_spacearr->lspacearr[i]->loops;
-        printf("Loop %d uses lnor space %p:\n", i, r_lnors_spacearr->lspacearr[i]);
-        print_v3("\tfinal lnor", lnors[i]);
-        print_v3("\tauto lnor", r_lnors_spacearr->lspacearr[i]->vec_lnor);
-        print_v3("\tref_vec", r_lnors_spacearr->lspacearr[i]->vec_ref);
-        printf("\talpha: %f\n\tbeta: %f\n\tloops: %p\n",
-               r_lnors_spacearr->lspacearr[i]->ref_alpha,
-               r_lnors_spacearr->lspacearr[i]->ref_beta,
-               r_lnors_spacearr->lspacearr[i]->loops);
-        printf("\t\t(shared with loops");
-        while (loops) {
-          printf(" %d", POINTER_AS_INT(loops->link));
-          loops = loops->next;
-        }
-        printf(")\n");
-      }
-      else {
-        printf("Loop %d has no lnor space\n", i);
-      }
-    }
-  }
-#endif
-}
-
-void CDDM_lower_num_verts(DerivedMesh *dm, int numVerts)
-{
-  BLI_assert(numVerts >= 0);
-  if (numVerts < dm->numVertData) {
-    CustomData_free_elem(&dm->vertData, numVerts, dm->numVertData - numVerts);
-  }
-
-  dm->numVertData = numVerts;
-}
-
-void CDDM_lower_num_edges(DerivedMesh *dm, int numEdges)
-{
-  BLI_assert(numEdges >= 0);
-  if (numEdges < dm->numEdgeData) {
-    CustomData_free_elem(&dm->edgeData, numEdges, dm->numEdgeData - numEdges);
-  }
-
-  dm->numEdgeData = numEdges;
-}
-
-void CDDM_lower_num_tessfaces(DerivedMesh *dm, int numTessFaces)
-{
-  BLI_assert(numTessFaces >= 0);
-  if (numTessFaces < dm->numTessFaceData) {
-    CustomData_free_elem(&dm->faceData, numTessFaces, dm->numTessFaceData - numTessFaces);
-  }
-
-  dm->numTessFaceData = numTessFaces;
-}
-
-void CDDM_lower_num_loops(DerivedMesh *dm, int numLoops)
-{
-  BLI_assert(numLoops >= 0);
-  if (numLoops < dm->numLoopData) {
-    CustomData_free_elem(&dm->loopData, numLoops, dm->numLoopData - numLoops);
-  }
-
-  dm->numLoopData = numLoops;
-}
-
-void CDDM_lower_num_polys(DerivedMesh *dm, int numPolys)
-{
-  BLI_assert(numPolys >= 0);
-  if (numPolys < dm->numPolyData) {
-    CustomData_free_elem(&dm->polyData, numPolys, dm->numPolyData - numPolys);
-  }
-
-  dm->numPolyData = numPolys;
-}
-
 /* mesh element access functions */
 
 MVert *CDDM_get_vert(DerivedMesh *dm, int index)
@@ -1427,58 +618,3 @@ MPoly *CDDM_get_polys(DerivedMesh *dm)
 {
   return ((CDDerivedMesh *)dm)->mpoly;
 }
-
-void CDDM_set_mvert(DerivedMesh *dm, MVert *mvert)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-
-  if (!CustomData_has_layer(&dm->vertData, CD_MVERT)) {
-    CustomData_add_layer(&dm->vertData, CD_MVERT, CD_ASSIGN, mvert, dm->numVertData);
-  }
-
-  cddm->mvert = mvert;
-}
-
-void CDDM_set_medge(DerivedMesh *dm, MEdge *medge)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-
-  if (!CustomData_has_layer(&dm->edgeData, CD_MEDGE)) {
-    CustomData_add_layer(&dm->edgeData, CD_MEDGE, CD_ASSIGN, medge, dm->numEdgeData);
-  }
-
-  cddm->medge = medge;
-}
-
-void CDDM_set_mface(DerivedMesh *dm, MFace *mface)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-
-  if (!CustomData_has_layer(&dm->faceData, CD_MFACE)) {
-    CustomData_add_layer(&dm->faceData, CD_MFACE, CD_ASSIGN, mface, dm->numTessFaceData);
-  }
-
-  cddm->mface = mface;
-}
-
-void CDDM_set_mloop(DerivedMesh *dm, MLoop *mloop)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-
-  if (!CustomData_has_layer(&dm->loopData, CD_MLOOP)) {
-    CustomData_add_layer(&dm->loopData, CD_MLOOP, CD_ASSIGN, mloop, dm->numLoopData);
-  }
-
-  cddm->mloop = mloop;
-}
-
-void CDDM_set_mpoly(DerivedMesh *dm, MPoly *mpoly)
-{
-  CDDerivedMesh *cddm = (CDDerivedMesh *)dm;
-
-  if (!CustomData_has_layer(&dm->polyData, CD_MPOLY)) {
-    CustomData_add_layer(&dm->polyData, CD_MPOLY, CD_ASSIGN, mpoly, dm->numPolyData);
-  }
-
-  cddm->mpoly = mpoly;
-}
diff --git a/source/blender/blenkernel/intern/collection.c b/source/blender/blenkernel/intern/collection.c
index 2031576190e..f2098cc2430 100644
--- a/source/blender/blenkernel/intern/collection.c
+++ b/source/blender/blenkernel/intern/collection.c
@@ -203,6 +203,9 @@ void BKE_collection_copy_data(Main *bmain,
                               const Collection *collection_src,
                               const int flag)
 {
+  BLI_assert(((collection_src->flag & COLLECTION_IS_MASTER) != 0) ==
+             ((collection_src->id.flag & LIB_PRIVATE_DATA) != 0));
+
   /* Do not copy collection's preview (same behavior as for objects). */
   if ((flag & LIB_ID_COPY_NO_PREVIEW) == 0 && false) { /* XXX TODO temp hack */
     BKE_previewimg_id_copy(&collection_dst->id, &collection_src->id);
@@ -338,8 +341,9 @@ Collection *BKE_collection_duplicate(Main *bmain,
                                      const bool do_obdata)
 {
   /* It's not allowed to copy the master collection. */
+  BLI_assert((collection->id.flag & LIB_PRIVATE_DATA) == 0);
+  BLI_assert((collection->flag & COLLECTION_IS_MASTER) == 0);
   if (collection->flag & COLLECTION_IS_MASTER) {
-    BLI_assert("!Master collection can't be duplicated");
     return NULL;
   }
 
@@ -365,15 +369,6 @@ Collection *BKE_collection_duplicate(Main *bmain,
   return collection_new;
 }
 
-Collection *BKE_collection_copy_master(Main *bmain, Collection *collection, const int flag)
-{
-  BLI_assert(collection->flag & COLLECTION_IS_MASTER);
-
-  Collection *collection_dst = MEM_dupallocN(collection);
-  BKE_collection_copy_data(bmain, collection_dst, collection, flag);
-  return collection_dst;
-}
-
 void BKE_collection_make_local(Main *bmain, Collection *collection, const bool lib_local)
 {
   BKE_id_make_local_generic(bmain, &collection->id, true, lib_local);
@@ -499,15 +494,11 @@ Collection *BKE_collection_master_add()
   /* Not an actual datablock, but owned by scene. */
   Collection *master_collection = MEM_callocN(sizeof(Collection), "Master Collection");
   STRNCPY(master_collection->id.name, "GRMaster Collection");
+  master_collection->id.flag |= LIB_PRIVATE_DATA;
   master_collection->flag |= COLLECTION_IS_MASTER;
   return master_collection;
 }
 
-Collection *BKE_collection_master(const Scene *scene)
-{
-  return scene->master_collection;
-}
-
 Scene *BKE_collection_master_scene_search(const Main *bmain, const Collection *master_collection)
 {
   BLI_assert((master_collection->flag & COLLECTION_IS_MASTER) != 0);
@@ -584,7 +575,7 @@ bool BKE_collection_has_object_recursive(Collection *collection, Object *ob)
 
 static Collection *collection_next_find(Main *bmain, Scene *scene, Collection *collection)
 {
-  if (scene && collection == BKE_collection_master(scene)) {
+  if (scene && collection == scene->master_collection) {
     return bmain->collections.first;
   }
   else {
@@ -601,7 +592,7 @@ Collection *BKE_collection_object_find(Main *bmain,
     collection = collection_next_find(bmain, scene, collection);
   }
   else if (scene) {
-    collection = BKE_collection_master(scene);
+    collection = scene->master_collection;
   }
   else {
     collection = bmain->collections.first;
@@ -743,7 +734,7 @@ void BKE_collection_object_add_from(Main *bmain, Scene *scene, Object *ob_src, O
   if (!is_instantiated) {
     /* In case we could not find any non-linked collections in which instantiate our ob_dst,
      * fallback to scene's master collection... */
-    collection_object_add(bmain, BKE_collection_master(scene), ob_dst, 0, true);
+    collection_object_add(bmain, scene->master_collection, ob_dst, 0, true);
   }
 
   BKE_main_collection_sync(bmain);
@@ -885,7 +876,7 @@ void BKE_collections_child_remove_nulls(Main *bmain, Collection *collection)
       collection_null_children_remove(collection);
     }
     for (Scene *scene = bmain->scenes.first; scene != NULL; scene = scene->id.next) {
-      collection_null_children_remove(BKE_collection_master(scene));
+      collection_null_children_remove(scene->master_collection);
     }
 
     for (collection = bmain->collections.first; collection != NULL;
@@ -893,7 +884,7 @@ void BKE_collections_child_remove_nulls(Main *bmain, Collection *collection)
       collection_missing_parents_remove(collection);
     }
     for (Scene *scene = bmain->scenes.first; scene != NULL; scene = scene->id.next) {
-      collection_missing_parents_remove(BKE_collection_master(scene));
+      collection_missing_parents_remove(scene->master_collection);
     }
   }
   else {
@@ -1169,7 +1160,7 @@ static Collection *collection_from_index_recursive(Collection *collection,
 Collection *BKE_collection_from_index(Scene *scene, const int index)
 {
   int index_current = 0;
-  Collection *master_collection = BKE_collection_master(scene);
+  Collection *master_collection = scene->master_collection;
   return collection_from_index_recursive(master_collection, index, &index_current);
 }
 
@@ -1363,7 +1354,7 @@ static void scene_collections_array(Scene *scene, Collection ***collections_arra
     return;
   }
 
-  collection = BKE_collection_master(scene);
+  collection = scene->master_collection;
   BLI_assert(collection != NULL);
   scene_collection_callback(collection, scene_collections_count, tot);
 
diff --git a/source/blender/blenkernel/intern/constraint.c b/source/blender/blenkernel/intern/constraint.c
index e1e4d138fd9..ccacb85651c 100644
--- a/source/blender/blenkernel/intern/constraint.c
+++ b/source/blender/blenkernel/intern/constraint.c
@@ -62,6 +62,7 @@
 #include "BKE_deform.h"
 #include "BKE_displist.h"
 #include "BKE_editmesh.h"
+#include "BKE_fcurve.h"
 #include "BKE_global.h"
 #include "BKE_idprop.h"
 #include "BKE_library.h"
@@ -1808,18 +1809,25 @@ static void rotlike_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *tar
     copy_v3_v3(loc, cob->matrix[3]);
     mat4_to_size(size, cob->matrix);
 
+    /* Select the Euler rotation order, defaulting to the owner. */
+    short rot_order = cob->rotOrder;
+
+    if (data->euler_order != CONSTRAINT_EULER_AUTO) {
+      rot_order = data->euler_order;
+    }
+
     /* To allow compatible rotations, must get both rotations in the order of the owner... */
-    mat4_to_eulO(obeul, cob->rotOrder, cob->matrix);
+    mat4_to_eulO(obeul, rot_order, cob->matrix);
     /* We must get compatible eulers from the beginning because
      * some of them can be modified below (see bug T21875). */
-    mat4_to_compatible_eulO(eul, obeul, cob->rotOrder, ct->matrix);
+    mat4_to_compatible_eulO(eul, obeul, rot_order, ct->matrix);
 
     if ((data->flag & ROTLIKE_X) == 0) {
       eul[0] = obeul[0];
     }
     else {
       if (data->flag & ROTLIKE_OFFSET) {
-        rotate_eulO(eul, cob->rotOrder, 'X', obeul[0]);
+        rotate_eulO(eul, rot_order, 'X', obeul[0]);
       }
 
       if (data->flag & ROTLIKE_X_INVERT) {
@@ -1832,7 +1840,7 @@ static void rotlike_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *tar
     }
     else {
       if (data->flag & ROTLIKE_OFFSET) {
-        rotate_eulO(eul, cob->rotOrder, 'Y', obeul[1]);
+        rotate_eulO(eul, rot_order, 'Y', obeul[1]);
       }
 
       if (data->flag & ROTLIKE_Y_INVERT) {
@@ -1845,7 +1853,7 @@ static void rotlike_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *tar
     }
     else {
       if (data->flag & ROTLIKE_OFFSET) {
-        rotate_eulO(eul, cob->rotOrder, 'Z', obeul[2]);
+        rotate_eulO(eul, rot_order, 'Z', obeul[2]);
       }
 
       if (data->flag & ROTLIKE_Z_INVERT) {
@@ -1856,7 +1864,7 @@ static void rotlike_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *tar
     /* Good to make eulers compatible again,
      * since we don't know how much they were changed above. */
     compatible_eul(eul, obeul);
-    loc_eulO_size_to_mat4(cob->matrix, loc, eul, size, cob->rotOrder);
+    loc_eulO_size_to_mat4(cob->matrix, loc, eul, size, rot_order);
   }
 }
 
@@ -1927,9 +1935,39 @@ static void sizelike_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *ta
   if (VALID_CONS_TARGET(ct)) {
     float obsize[3], size[3];
 
-    mat4_to_size(size, ct->matrix);
     mat4_to_size(obsize, cob->matrix);
 
+    /* Compute one uniform scale factor to apply to all three axes. */
+    if (data->flag & SIZELIKE_UNIFORM) {
+      const int all_axes = SIZELIKE_X | SIZELIKE_Y | SIZELIKE_Z;
+      float total = 1.0f;
+
+      /* If all axes are selected, use the determinant. */
+      if ((data->flag & all_axes) == all_axes) {
+        total = fabsf(mat4_to_volume_scale(ct->matrix));
+      }
+      /* Otherwise multiply individual values. */
+      else {
+        mat4_to_size(size, ct->matrix);
+
+        if (data->flag & SIZELIKE_X) {
+          total *= size[0];
+        }
+        if (data->flag & SIZELIKE_Y) {
+          total *= size[1];
+        }
+        if (data->flag & SIZELIKE_Z) {
+          total *= size[2];
+        }
+      }
+
+      copy_v3_fl(size, cbrt(total));
+    }
+    /* Regular per-axis scaling. */
+    else {
+      mat4_to_size(size, ct->matrix);
+    }
+
     for (int i = 0; i < 3; i++) {
       size[i] = powf(size[i], data->power);
     }
@@ -1948,13 +1986,13 @@ static void sizelike_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *ta
       }
     }
 
-    if ((data->flag & SIZELIKE_X) && (obsize[0] != 0)) {
+    if ((data->flag & (SIZELIKE_X | SIZELIKE_UNIFORM)) && (obsize[0] != 0)) {
       mul_v3_fl(cob->matrix[0], size[0] / obsize[0]);
     }
-    if ((data->flag & SIZELIKE_Y) && (obsize[1] != 0)) {
+    if ((data->flag & (SIZELIKE_Y | SIZELIKE_UNIFORM)) && (obsize[1] != 0)) {
       mul_v3_fl(cob->matrix[1], size[1] / obsize[1]);
     }
-    if ((data->flag & SIZELIKE_Z) && (obsize[2] != 0)) {
+    if ((data->flag & (SIZELIKE_Z | SIZELIKE_UNIFORM)) && (obsize[2] != 0)) {
       mul_v3_fl(cob->matrix[2], size[2] / obsize[2]);
     }
   }
@@ -3689,7 +3727,8 @@ static void transform_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *t
   if (VALID_CONS_TARGET(ct)) {
     float *from_min, *from_max, *to_min, *to_max;
     float loc[3], eul[3], size[3];
-    float dvec[3], sval[3];
+    float dbuf[4], sval[3];
+    float *const dvec = dbuf + 1;
     int i;
 
     /* obtain target effect */
@@ -3708,7 +3747,8 @@ static void transform_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *t
         from_max = data->from_max_scale;
         break;
       case TRANS_ROTATION:
-        mat4_to_eulO(dvec, cob->rotOrder, ct->matrix);
+        BKE_driver_target_matrix_to_rot_channels(
+            ct->matrix, cob->rotOrder, data->from_rotation_mode, -1, true, dbuf);
         from_min = data->from_min_rot;
         from_max = data->from_max_rot;
         break;
@@ -3720,9 +3760,16 @@ static void transform_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *t
         break;
     }
 
+    /* Select the output Euler rotation order, defaulting to the owner. */
+    short rot_order = cob->rotOrder;
+
+    if (data->to == TRANS_ROTATION && data->to_euler_order != CONSTRAINT_EULER_AUTO) {
+      rot_order = data->to_euler_order;
+    }
+
     /* extract components of owner's matrix */
     copy_v3_v3(loc, cob->matrix[3]);
-    mat4_to_eulO(eul, cob->rotOrder, cob->matrix);
+    mat4_to_eulO(eul, rot_order, cob->matrix);
     mat4_to_size(size, cob->matrix);
 
     /* determine where in range current transforms lie */
@@ -3781,7 +3828,7 @@ static void transform_evaluate(bConstraint *con, bConstraintOb *cob, ListBase *t
     }
 
     /* apply to matrix */
-    loc_eulO_size_to_mat4(cob->matrix, loc, eul, size, cob->rotOrder);
+    loc_eulO_size_to_mat4(cob->matrix, loc, eul, size, rot_order);
   }
 }
 
diff --git a/source/blender/blenkernel/intern/context.c b/source/blender/blenkernel/intern/context.c
index f536f21c3e5..bcf6bb338ff 100644
--- a/source/blender/blenkernel/intern/context.c
+++ b/source/blender/blenkernel/intern/context.c
@@ -1053,7 +1053,7 @@ Collection *CTX_data_collection(const bContext *C)
 
   /* fallback */
   Scene *scene = CTX_data_scene(C);
-  return BKE_collection_master(scene);
+  return scene->master_collection;
 }
 
 enum eContextObjectMode CTX_data_mode_enum_ex(const Object *obedit,
diff --git a/source/blender/blenkernel/intern/crazyspace.c b/source/blender/blenkernel/intern/crazyspace.c
index 09bbc6c0d13..fcbebd24b4a 100644
--- a/source/blender/blenkernel/intern/crazyspace.c
+++ b/source/blender/blenkernel/intern/crazyspace.c
@@ -293,7 +293,7 @@ int BKE_crazyspace_get_first_deform_matrices_editbmesh(struct Depsgraph *depsgra
         BLI_linklist_free((LinkNode *)datamasks, NULL);
 
         me = BKE_mesh_from_editmesh_with_coords_thin_wrap(em, &data_mask, NULL);
-        deformedVerts = editbmesh_get_vertex_cos(em, &numVerts);
+        deformedVerts = editbmesh_vert_coords_alloc(em, &numVerts);
         defmats = MEM_mallocN(sizeof(*defmats) * numVerts, "defmats");
 
         for (a = 0; a < numVerts; a++) {
@@ -383,7 +383,7 @@ int BKE_sculpt_get_first_deform_matrices(struct Depsgraph *depsgraph,
          * state. */
         Mesh *me = object_eval.data;
         me_eval = BKE_mesh_copy_for_eval(me, true);
-        deformedVerts = BKE_mesh_vertexCos_get(me, &numVerts);
+        deformedVerts = BKE_mesh_vert_coords_alloc(me, &numVerts);
         defmats = MEM_callocN(sizeof(*defmats) * numVerts, "defmats");
 
         for (a = 0; a < numVerts; a++) {
@@ -485,7 +485,7 @@ void BKE_crazyspace_build_sculpt(struct Depsgraph *depsgraph,
     int a, numVerts;
     Mesh *mesh = (Mesh *)object->data;
 
-    *deformcos = BKE_mesh_vertexCos_get(mesh, &numVerts);
+    *deformcos = BKE_mesh_vert_coords_alloc(mesh, &numVerts);
     *deformmats = MEM_callocN(sizeof(*(*deformmats)) * numVerts, "defmats");
 
     for (a = 0; a < numVerts; a++) {
diff --git a/source/blender/blenkernel/intern/curve.c b/source/blender/blenkernel/intern/curve.c
index 980e8043a42..3112e1b5ed1 100644
--- a/source/blender/blenkernel/intern/curve.c
+++ b/source/blender/blenkernel/intern/curve.c
@@ -4606,18 +4606,13 @@ void BKE_nurb_direction_switch(Nurb *nu)
   }
 }
 
-float (*BKE_curve_nurbs_vertexCos_get(ListBase *lb, int *r_numVerts))[3]
+void BKE_curve_nurbs_vert_coords_get(ListBase *lb, float (*vert_coords)[3], int vert_len)
 {
-  int i, numVerts = *r_numVerts = BKE_nurbList_verts_count(lb);
-  float *co, (*cos)[3] = MEM_malloc_arrayN(numVerts, sizeof(*cos), "cu_vcos");
-  Nurb *nu;
-
-  co = cos[0];
-  for (nu = lb->first; nu; nu = nu->next) {
+  float *co = vert_coords[0];
+  for (Nurb *nu = lb->first; nu; nu = nu->next) {
     if (nu->type == CU_BEZIER) {
       BezTriple *bezt = nu->bezt;
-
-      for (i = 0; i < nu->pntsu; i++, bezt++) {
+      for (int i = 0; i < nu->pntsu; i++, bezt++) {
         copy_v3_v3(co, bezt->vec[0]);
         co += 3;
         copy_v3_v3(co, bezt->vec[1]);
@@ -4628,20 +4623,31 @@ float (*BKE_curve_nurbs_vertexCos_get(ListBase *lb, int *r_numVerts))[3]
     }
     else {
       BPoint *bp = nu->bp;
-
-      for (i = 0; i < nu->pntsu * nu->pntsv; i++, bp++) {
+      for (int i = 0; i < nu->pntsu * nu->pntsv; i++, bp++) {
         copy_v3_v3(co, bp->vec);
         co += 3;
       }
     }
   }
+  BLI_assert(co == vert_coords[vert_len]);
+  UNUSED_VARS_NDEBUG(vert_len);
+}
 
-  return cos;
+float (*BKE_curve_nurbs_vert_coords_alloc(ListBase *lb, int *r_vert_len))[3]
+{
+  const int vert_len = BKE_nurbList_verts_count(lb);
+  float(*vert_coords)[3] = MEM_malloc_arrayN(vert_len, sizeof(*vert_coords), __func__);
+  BKE_curve_nurbs_vert_coords_get(lb, vert_coords, vert_len);
+  *r_vert_len = vert_len;
+  return vert_coords;
 }
 
-void BK_curve_nurbs_vertexCos_apply(ListBase *lb, float (*vertexCos)[3])
+void BKE_curve_nurbs_vert_coords_apply_with_mat4(ListBase *lb,
+                                                 const float (*vert_coords)[3],
+                                                 const float mat[4][4],
+                                                 const bool constrain_2d)
 {
-  const float *co = vertexCos[0];
+  const float *co = vert_coords[0];
   Nurb *nu;
   int i;
 
@@ -4650,6 +4656,44 @@ void BK_curve_nurbs_vertexCos_apply(ListBase *lb, float (*vertexCos)[3])
       BezTriple *bezt = nu->bezt;
 
       for (i = 0; i < nu->pntsu; i++, bezt++) {
+        mul_v3_m4v3(bezt->vec[0], mat, co);
+        co += 3;
+        mul_v3_m4v3(bezt->vec[1], mat, co);
+        co += 3;
+        mul_v3_m4v3(bezt->vec[2], mat, co);
+        co += 3;
+      }
+    }
+    else {
+      BPoint *bp = nu->bp;
+
+      for (i = 0; i < nu->pntsu * nu->pntsv; i++, bp++) {
+        mul_v3_m4v3(bp->vec, mat, co);
+        co += 3;
+      }
+    }
+
+    if (constrain_2d) {
+      if (nu->flag & CU_2D) {
+        BKE_nurb_test_2d(nu);
+      }
+    }
+
+    calchandlesNurb_intern(nu, true);
+  }
+}
+
+void BKE_curve_nurbs_vert_coords_apply(ListBase *lb,
+                                       const float (*vert_coords)[3],
+                                       const bool constrain_2d)
+{
+  const float *co = vert_coords[0];
+
+  for (Nurb *nu = lb->first; nu; nu = nu->next) {
+    if (nu->type == CU_BEZIER) {
+      BezTriple *bezt = nu->bezt;
+
+      for (int i = 0; i < nu->pntsu; i++, bezt++) {
         copy_v3_v3(bezt->vec[0], co);
         co += 3;
         copy_v3_v3(bezt->vec[1], co);
@@ -4661,28 +4705,33 @@ void BK_curve_nurbs_vertexCos_apply(ListBase *lb, float (*vertexCos)[3])
     else {
       BPoint *bp = nu->bp;
 
-      for (i = 0; i < nu->pntsu * nu->pntsv; i++, bp++) {
+      for (int i = 0; i < nu->pntsu * nu->pntsv; i++, bp++) {
         copy_v3_v3(bp->vec, co);
         co += 3;
       }
     }
 
+    if (constrain_2d) {
+      if (nu->flag & CU_2D) {
+        BKE_nurb_test_2d(nu);
+      }
+    }
+
     calchandlesNurb_intern(nu, true);
   }
 }
 
-float (*BKE_curve_nurbs_keyVertexCos_get(ListBase *lb, float *key))[3]
+float (*BKE_curve_nurbs_key_vert_coords_alloc(ListBase *lb, float *key, int *r_vert_len))[3]
 {
-  int i, numVerts = BKE_nurbList_verts_count(lb);
-  float *co, (*cos)[3] = MEM_malloc_arrayN(numVerts, sizeof(*cos), "cu_vcos");
-  Nurb *nu;
+  int vert_len = BKE_nurbList_verts_count(lb);
+  float(*cos)[3] = MEM_malloc_arrayN(vert_len, sizeof(*cos), __func__);
 
-  co = cos[0];
-  for (nu = lb->first; nu; nu = nu->next) {
+  float *co = cos[0];
+  for (Nurb *nu = lb->first; nu; nu = nu->next) {
     if (nu->type == CU_BEZIER) {
       BezTriple *bezt = nu->bezt;
 
-      for (i = 0; i < nu->pntsu; i++, bezt++) {
+      for (int i = 0; i < nu->pntsu; i++, bezt++) {
         copy_v3_v3(co, &key[0]);
         co += 3;
         copy_v3_v3(co, &key[3]);
@@ -4695,18 +4744,18 @@ float (*BKE_curve_nurbs_keyVertexCos_get(ListBase *lb, float *key))[3]
     else {
       BPoint *bp = nu->bp;
 
-      for (i = 0; i < nu->pntsu * nu->pntsv; i++, bp++) {
+      for (int i = 0; i < nu->pntsu * nu->pntsv; i++, bp++) {
         copy_v3_v3(co, key);
         co += 3;
         key += KEYELEM_FLOAT_LEN_BPOINT;
       }
     }
   }
-
+  *r_vert_len = vert_len;
   return cos;
 }
 
-void BKE_curve_nurbs_keyVertexTilts_apply(ListBase *lb, float *key)
+void BKE_curve_nurbs_key_vert_tilts_apply(ListBase *lb, float *key)
 {
   Nurb *nu;
   int i;
diff --git a/source/blender/blenkernel/intern/customdata.c b/source/blender/blenkernel/intern/customdata.c
index 770d8662743..bc982140b29 100644
--- a/source/blender/blenkernel/intern/customdata.c
+++ b/source/blender/blenkernel/intern/customdata.c
@@ -294,7 +294,7 @@ static void layerInterp_mdeformvert(const void **sources,
     }
   }
 
-  /* delay writing to the destination incase dest is in sources */
+  /* Delay writing to the destination in case dest is in sources. */
 
   /* now we know how many unique deform weights there are, so realloc */
   if (dvert->dw && (dvert->totweight == totweight)) {
@@ -441,7 +441,7 @@ static void layerInterp_tface(
     }
   }
 
-  /* delay writing to the destination incase dest is in sources */
+  /* Delay writing to the destination in case dest is in sources. */
   *tf = *(MTFace *)(*sources);
   memcpy(tf->uv, uv, sizeof(tf->uv));
 }
@@ -548,7 +548,7 @@ static void layerInterp_origspace_face(
     }
   }
 
-  /* delay writing to the destination in case dest is in sources */
+  /* Delay writing to the destination in case dest is in sources. */
   memcpy(osf->uv, uv, sizeof(osf->uv));
 }
 
@@ -908,7 +908,7 @@ static void layerInterp_mloopcol(
   /* Subdivide smooth or fractal can cause problems without clamping
    * although weights should also not cause this situation */
 
-  /* also delay writing to the destination incase dest is in sources */
+  /* Also delay writing to the destination in case dest is in sources. */
   mc->r = round_fl_to_uchar_clamp(col.r);
   mc->g = round_fl_to_uchar_clamp(col.g);
   mc->b = round_fl_to_uchar_clamp(col.b);
@@ -1008,7 +1008,7 @@ static void layerInterp_mloopuv(
     }
   }
 
-  /* delay writing to the destination incase dest is in sources */
+  /* Delay writing to the destination in case dest is in sources. */
   copy_v2_v2(((MLoopUV *)dest)->uv, uv);
   ((MLoopUV *)dest)->flag = flag;
 }
@@ -1104,7 +1104,7 @@ static void layerInterp_mloop_origspace(
     }
   }
 
-  /* delay writing to the destination incase dest is in sources */
+  /* Delay writing to the destination in case dest is in sources. */
   copy_v2_v2(((OrigSpaceLoop *)dest)->uv, uv);
 }
 /* --- end copy */
@@ -1152,7 +1152,7 @@ static void layerInterp_mcol(
     }
   }
 
-  /* delay writing to the destination incase dest is in sources */
+  /* Delay writing to the destination in case dest is in sources. */
   for (j = 0; j < 4; ++j) {
 
     /* Subdivide smooth or fractal can cause problems without clamping
@@ -1220,7 +1220,7 @@ static void layerInterp_bweight(const void **sources,
     }
   }
 
-  /* delay writing to the destination incase dest is in sources */
+  /* Delay writing to the destination in case dest is in sources. */
   *((float *)dest) = f;
 }
 
@@ -1251,7 +1251,7 @@ static void layerInterp_shapekey(const void **sources,
     }
   }
 
-  /* delay writing to the destination incase dest is in sources */
+  /* Delay writing to the destination in case dest is in sources. */
   copy_v3_v3((float *)dest, co);
 }
 
@@ -1289,7 +1289,7 @@ static void layerInterp_mvert_skin(const void **sources,
     madd_v3_v3fl(radius, vs_src->radius, w);
   }
 
-  /* delay writing to the destination incase dest is in sources */
+  /* Delay writing to the destination in case dest is in sources. */
   vs_dst = dest;
   copy_v3_v3(vs_dst->radius, radius);
   vs_dst->flag &= ~MVERT_SKIN_ROOT;
@@ -2828,9 +2828,7 @@ void CustomData_interp(const CustomData *source,
   const void *source_buf[SOURCE_BUF_SIZE];
   const void **sources = source_buf;
 
-  /* slow fallback in case we're interpolating a ridiculous number of
-   * elements
-   */
+  /* Slow fallback in case we're interpolating a ridiculous number of elements. */
   if (count > SOURCE_BUF_SIZE) {
     sources = MEM_malloc_arrayN(count, sizeof(*sources), __func__);
   }
@@ -3828,9 +3826,7 @@ void CustomData_bmesh_interp(CustomData *data,
   void *source_buf[SOURCE_BUF_SIZE];
   const void **sources = (const void **)source_buf;
 
-  /* slow fallback in case we're interpolating a ridiculous number of
-   * elements
-   */
+  /* Slow fallback in case we're interpolating a ridiculous number of elements. */
   if (count > SOURCE_BUF_SIZE) {
     sources = MEM_malloc_arrayN(count, sizeof(*sources), __func__);
   }
@@ -4185,7 +4181,7 @@ bool CustomData_verify_versions(struct CustomData *data, int index)
     if (!typeInfo->defaultname && (index > 0) && data->layers[index - 1].type == layer->type) {
       keeplayer = false; /* multiple layers of which we only support one */
     }
-    /* This is a pre-emptive fix for cases that should not happen
+    /* This is a preemptive fix for cases that should not happen
      * (layers that should not be written in .blend files),
      * but can happen due to bugs (see e.g. T62318).
      * Also for forward compatibility, in future,
diff --git a/source/blender/blenkernel/intern/displist.c b/source/blender/blenkernel/intern/displist.c
index 83a10344b96..cbd3c91acc7 100644
--- a/source/blender/blenkernel/intern/displist.c
+++ b/source/blender/blenkernel/intern/displist.c
@@ -892,8 +892,7 @@ static void curve_calc_modifiers_pre(
        * tilts, which is passed through in the modifier stack.
        * this is also the reason curves do not use a virtual
        * shape key modifier yet. */
-      deformedVerts = BKE_curve_nurbs_keyVertexCos_get(nurb, keyVerts);
-      numVerts = BKE_nurbList_verts_count(nurb);
+      deformedVerts = BKE_curve_nurbs_key_vert_coords_alloc(nurb, keyVerts, &numVerts);
     }
   }
 
@@ -909,7 +908,7 @@ static void curve_calc_modifiers_pre(
       }
 
       if (!deformedVerts) {
-        deformedVerts = BKE_curve_nurbs_vertexCos_get(nurb, &numVerts);
+        deformedVerts = BKE_curve_nurbs_vert_coords_alloc(nurb, &numVerts);
       }
 
       mti->deformVerts(md, &mectx, NULL, deformedVerts, numVerts);
@@ -921,11 +920,11 @@ static void curve_calc_modifiers_pre(
   }
 
   if (deformedVerts) {
-    BK_curve_nurbs_vertexCos_apply(nurb, deformedVerts);
+    BKE_curve_nurbs_vert_coords_apply(nurb, deformedVerts, false);
     MEM_freeN(deformedVerts);
   }
   if (keyVerts) { /* these are not passed through modifier stack */
-    BKE_curve_nurbs_keyVertexTilts_apply(nurb, keyVerts);
+    BKE_curve_nurbs_key_vert_tilts_apply(nurb, keyVerts);
   }
 
   if (keyVerts) {
@@ -933,18 +932,18 @@ static void curve_calc_modifiers_pre(
   }
 }
 
-static float (*displist_get_allverts(ListBase *dispbase, int *totvert))[3]
+static float (*displist_vert_coords_alloc(ListBase *dispbase, int *r_vert_len))[3]
 {
   DispList *dl;
   float(*allverts)[3], *fp;
 
-  *totvert = 0;
+  *r_vert_len = 0;
 
   for (dl = dispbase->first; dl; dl = dl->next) {
-    *totvert += (dl->type == DL_INDEX3) ? dl->nr : dl->parts * dl->nr;
+    *r_vert_len += (dl->type == DL_INDEX3) ? dl->nr : dl->parts * dl->nr;
   }
 
-  allverts = MEM_mallocN((*totvert) * sizeof(float) * 3, "displist_get_allverts allverts");
+  allverts = MEM_mallocN((*r_vert_len) * sizeof(float) * 3, "displist_vert_coords_alloc allverts");
   fp = (float *)allverts;
   for (dl = dispbase->first; dl; dl = dl->next) {
     int offs = 3 * ((dl->type == DL_INDEX3) ? dl->nr : dl->parts * dl->nr);
@@ -955,7 +954,7 @@ static float (*displist_get_allverts(ListBase *dispbase, int *totvert))[3]
   return allverts;
 }
 
-static void displist_apply_allverts(ListBase *dispbase, float (*allverts)[3])
+static void displist_vert_coords_apply(ListBase *dispbase, float (*allverts)[3])
 {
   DispList *dl;
   const float *fp;
@@ -1027,7 +1026,7 @@ static void curve_calc_modifiers_post(Depsgraph *depsgraph,
      * we need to create a Mesh for each curve that uses modifiers. */
     if (modified == NULL /* && need_normal */) {
       if (vertCos != NULL) {
-        displist_apply_allverts(dispbase, vertCos);
+        displist_vert_coords_apply(dispbase, vertCos);
       }
 
       if (ELEM(ob->type, OB_CURVE, OB_FONT) && (cu->flag & CU_DEFORM_FILL)) {
@@ -1041,7 +1040,7 @@ static void curve_calc_modifiers_post(Depsgraph *depsgraph,
         (mti->type == eModifierTypeType_DeformOrConstruct && !modified)) {
       if (modified) {
         if (!vertCos) {
-          vertCos = BKE_mesh_vertexCos_get(modified, &totvert);
+          vertCos = BKE_mesh_vert_coords_alloc(modified, &totvert);
         }
         if (need_normal) {
           BKE_mesh_ensure_normals(modified);
@@ -1050,7 +1049,7 @@ static void curve_calc_modifiers_post(Depsgraph *depsgraph,
       }
       else {
         if (!vertCos) {
-          vertCos = displist_get_allverts(dispbase, &totvert);
+          vertCos = displist_vert_coords_alloc(dispbase, &totvert);
         }
         mti->deformVerts(md, &mectx_deform, NULL, vertCos, totvert);
       }
@@ -1071,12 +1070,12 @@ static void curve_calc_modifiers_post(Depsgraph *depsgraph,
           BKE_id_free(NULL, modified);
           modified = temp_mesh;
 
-          BKE_mesh_apply_vert_coords(modified, vertCos);
+          BKE_mesh_vert_coords_apply(modified, vertCos);
         }
       }
       else {
         if (vertCos) {
-          displist_apply_allverts(dispbase, vertCos);
+          displist_vert_coords_apply(dispbase, vertCos);
         }
 
         if (ELEM(ob->type, OB_CURVE, OB_FONT) && (cu->flag & CU_DEFORM_FILL)) {
@@ -1115,13 +1114,13 @@ static void curve_calc_modifiers_post(Depsgraph *depsgraph,
       BKE_id_free(NULL, modified);
       modified = temp_mesh;
 
-      BKE_mesh_apply_vert_coords(modified, vertCos);
+      BKE_mesh_vert_coords_apply(modified, vertCos);
       BKE_mesh_calc_normals_mapping_simple(modified);
 
       MEM_freeN(vertCos);
     }
     else {
-      displist_apply_allverts(dispbase, vertCos);
+      displist_vert_coords_apply(dispbase, vertCos);
       MEM_freeN(vertCos);
       vertCos = NULL;
     }
@@ -1129,15 +1128,6 @@ static void curve_calc_modifiers_post(Depsgraph *depsgraph,
 
   if (r_final) {
     if (modified) {
-      /* see: mesh_calc_modifiers */
-      if (modified->totface == 0) {
-        BKE_mesh_tessface_calc(modified);
-      }
-      /* Even if tessellation is not needed, some modifiers might have modified CD layers
-       * (like mloopcol or mloopuv), hence we have to update those. */
-      else if (modified->runtime.cd_dirty_vert & CD_MASK_TESSLOOPNORMAL) {
-        BKE_mesh_tessface_calc(modified);
-      }
 
       /* XXX2.8(Sybren): make sure the face normals are recalculated as well */
       BKE_mesh_ensure_normals(modified);
diff --git a/source/blender/blenkernel/intern/dynamicpaint.c b/source/blender/blenkernel/intern/dynamicpaint.c
index 04cbdbb6bcd..91b169590ac 100644
--- a/source/blender/blenkernel/intern/dynamicpaint.c
+++ b/source/blender/blenkernel/intern/dynamicpaint.c
@@ -2755,7 +2755,7 @@ static bool dynamicPaint_symmetrizeAdjData(PaintAdjData *ed, int active_points)
       for (int i = 0, idx = ed->n_index[index]; i < ed->n_num[index]; i++) {
         const int target = ed->n_target[idx + i];
 
-        assert(!(ed->flags[target] & ADJ_BORDER_PIXEL));
+        BLI_assert(!(ed->flags[target] & ADJ_BORDER_PIXEL));
 
         if (!dynamicPaint_pointHasNeighbor(ed, target, index)) {
           new_n_num[target]++;
@@ -2782,7 +2782,7 @@ static bool dynamicPaint_symmetrizeAdjData(PaintAdjData *ed, int active_points)
         new_n_num[index] = ed->n_num[index];
       }
 
-      assert(n_pos == total_targets);
+      BLI_assert(n_pos == total_targets);
 
       /* Add symmetrized - this loop behavior must exactly match the count pass above */
       for (int index = 0; index < active_points; index++) {
@@ -5572,7 +5572,7 @@ static void dynamic_paint_border_cb(void *__restrict userdata,
 
     PaintPoint *pPoint2 = &((PaintPoint *)sData->type_data)[target];
 
-    assert(!(sData->adj_data->flags[target] & ADJ_BORDER_PIXEL));
+    BLI_assert(!(sData->adj_data->flags[target] & ADJ_BORDER_PIXEL));
 
     madd_v3_v3fl(mix_color, pPoint2->color, pPoint2->color[3]);
     mix_color[3] += pPoint2->color[3];
diff --git a/source/blender/blenkernel/intern/editderivedmesh.c b/source/blender/blenkernel/intern/editderivedmesh.c
index 06f297b23e0..2df3d2f0fe9 100644
--- a/source/blender/blenkernel/intern/editderivedmesh.c
+++ b/source/blender/blenkernel/intern/editderivedmesh.c
@@ -517,8 +517,8 @@ static void cage_mapped_verts_callback(void *userData,
   }
 }
 
-float (*BKE_editmesh_vertexCos_get(
-    struct Depsgraph *depsgraph, BMEditMesh *em, Scene *scene, int *r_numVerts))[3]
+float (*BKE_editmesh_vert_coords_alloc(
+    struct Depsgraph *depsgraph, BMEditMesh *em, Scene *scene, int *r_vert_len))[3]
 {
   Mesh *cage;
   BLI_bitmap *visit_bitmap;
@@ -540,8 +540,8 @@ float (*BKE_editmesh_vertexCos_get(
 
   MEM_freeN(visit_bitmap);
 
-  if (r_numVerts) {
-    *r_numVerts = em->bm->totvert;
+  if (r_vert_len) {
+    *r_vert_len = em->bm->totvert;
   }
 
   return cos_cage;
diff --git a/source/blender/blenkernel/intern/editmesh.c b/source/blender/blenkernel/intern/editmesh.c
index b8234ccc5bb..866c494d354 100644
--- a/source/blender/blenkernel/intern/editmesh.c
+++ b/source/blender/blenkernel/intern/editmesh.c
@@ -39,7 +39,7 @@ BMEditMesh *BKE_editmesh_create(BMesh *bm, const bool do_tessellate)
 
   em->bm = bm;
   if (do_tessellate) {
-    BKE_editmesh_tessface_calc(em);
+    BKE_editmesh_looptri_calc(em);
   }
 
   return em;
@@ -61,11 +61,10 @@ BMEditMesh *BKE_editmesh_copy(BMEditMesh *em)
 
   /* The tessellation is NOT calculated on the copy here,
    * because currently all the callers of this function use
-   * it to make a backup copy of the BMEditMesh to restore
-   * it in the case of errors in an operation. For perf
-   * reasons, in that case it makes more sense to do the
-   * tessellation only when/if that copy ends up getting
-   * used.*/
+   * it to make a backup copy of the #BMEditMesh to restore
+   * it in the case of errors in an operation. For performance reasons,
+   * in that case it makes more sense to do the
+   * tessellation only when/if that copy ends up getting used. */
   em_copy->looptris = NULL;
 
   return em_copy;
@@ -97,8 +96,8 @@ static void editmesh_tessface_calc_intern(BMEditMesh *em)
 
   BMesh *bm = em->bm;
 
-  /* this assumes all faces can be scan-filled, which isn't always true,
-   * worst case we over alloc a little which is acceptable */
+  /* This assumes all faces can be scan-filled, which isn't always true,
+   * worst case we over allocate a little which is acceptable. */
   const int looptris_tot = poly_to_tri_count(bm->totface, bm->totloop);
   const int looptris_tot_prev_alloc = em->looptris ?
                                           (MEM_allocN_len(em->looptris) / sizeof(*em->looptris)) :
@@ -109,7 +108,7 @@ static void editmesh_tessface_calc_intern(BMEditMesh *em)
   /* this means no reallocs for quad dominant models, for */
   if ((em->looptris != NULL) &&
       /* (*em->tottri >= looptris_tot)) */
-      /* check against alloc'd size incase we over alloc'd a little */
+      /* Check against allocated size in case we over allocated a little. */
       ((looptris_tot_prev_alloc >= looptris_tot) &&
        (looptris_tot_prev_alloc <= looptris_tot * 2))) {
     looptris = em->looptris;
@@ -127,7 +126,7 @@ static void editmesh_tessface_calc_intern(BMEditMesh *em)
   BM_mesh_calc_tessellation(em->bm, em->looptris, &em->tottri);
 }
 
-void BKE_editmesh_tessface_calc(BMEditMesh *em)
+void BKE_editmesh_looptri_calc(BMEditMesh *em)
 {
   editmesh_tessface_calc_intern(em);
 
@@ -210,7 +209,7 @@ void BKE_editmesh_color_ensure(BMEditMesh *em, const char htype)
   }
 }
 
-float (*BKE_editmesh_vertexCos_get_orco(BMEditMesh *em, int *r_numVerts))[3]
+float (*BKE_editmesh_vert_coords_alloc_orco(BMEditMesh *em, int *r_vert_len))[3]
 {
   BMIter iter;
   BMVert *eve;
@@ -223,7 +222,7 @@ float (*BKE_editmesh_vertexCos_get_orco(BMEditMesh *em, int *r_numVerts))[3]
     copy_v3_v3(orco[i], eve->co);
   }
 
-  *r_numVerts = em->bm->totvert;
+  *r_vert_len = em->bm->totvert;
 
   return orco;
 }
diff --git a/source/blender/blenkernel/intern/editmesh_bvh.c b/source/blender/blenkernel/intern/editmesh_bvh.c
index f23276d7be0..5af4e4c0023 100644
--- a/source/blender/blenkernel/intern/editmesh_bvh.c
+++ b/source/blender/blenkernel/intern/editmesh_bvh.c
@@ -73,7 +73,7 @@ BMBVHTree *BKE_bmbvh_new_ex(BMesh *bm,
   BMFace *f_test, *f_test_prev;
   bool test_fn_ret;
 
-  /* BKE_editmesh_tessface_calc() must be called already */
+  /* BKE_editmesh_looptri_calc() must be called already */
   BLI_assert(looptris_tot != 0 || bm->totface == 0);
 
   if (cos_cage) {
diff --git a/source/blender/blenkernel/intern/fcurve.c b/source/blender/blenkernel/intern/fcurve.c
index 9580ea763fb..b596eeb9e35 100644
--- a/source/blender/blenkernel/intern/fcurve.c
+++ b/source/blender/blenkernel/intern/fcurve.c
@@ -382,8 +382,8 @@ FCurve *rna_get_fcurve_context_ui(bContext *C,
   }
 
   /* there must be some RNA-pointer + property combon */
-  if (prop && tptr.id.data && RNA_property_animateable(&tptr, prop)) {
-    AnimData *adt = BKE_animdata_from_id(tptr.id.data);
+  if (prop && tptr.owner_id && RNA_property_animateable(&tptr, prop)) {
+    AnimData *adt = BKE_animdata_from_id(tptr.owner_id);
     int step = (
         /* Always 1 in case we have no context (can't check in 'ancestors' of given RNA ptr). */
         C ? 2 : 1);
@@ -391,7 +391,7 @@ FCurve *rna_get_fcurve_context_ui(bContext *C,
 
     if (!adt && C) {
       path = BKE_animdata_driver_path_hack(C, &tptr, prop, NULL);
-      adt = BKE_animdata_from_id(tptr.id.data);
+      adt = BKE_animdata_from_id(tptr.owner_id);
       step--;
     }
 
@@ -438,7 +438,7 @@ FCurve *rna_get_fcurve_context_ui(bContext *C,
             if (tpath && tpath != path) {
               MEM_freeN(path);
               path = tpath;
-              adt = BKE_animdata_from_id(tptr.id.data);
+              adt = BKE_animdata_from_id(tptr.owner_id);
             }
             else {
               adt = NULL;
@@ -1714,7 +1714,10 @@ static float dvar_eval_transChan(ChannelDriver *driver, DriverVar *dvar)
      * of scale over all three axes unless the matrix includes shear. */
     return cbrtf(mat4_to_volume_scale(mat));
   }
-  else if (dtar->transChan >= DTAR_TRANSCHAN_SCALEX) {
+  else if (ELEM(dtar->transChan,
+                DTAR_TRANSCHAN_SCALEX,
+                DTAR_TRANSCHAN_SCALEY,
+                DTAR_TRANSCHAN_SCALEZ)) {
     /* Extract scale, and choose the right axis,
      * inline 'mat4_to_size'. */
     return len_v3(mat[dtar->transChan - DTAR_TRANSCHAN_SCALEX]);
@@ -1728,15 +1731,25 @@ static float dvar_eval_transChan(ChannelDriver *driver, DriverVar *dvar)
      *     b) [NOT USED] directly use the original values (no decomposition)
      *         - only an option for "transform space", if quality is really bad with a)
      */
-    float eul[3];
+    float quat[4];
+    int channel;
 
-    mat4_to_eulO(eul, rot_order, mat);
+    if (dtar->transChan == DTAR_TRANSCHAN_ROTW) {
+      channel = 0;
+    }
+    else {
+      channel = 1 + dtar->transChan - DTAR_TRANSCHAN_ROTX;
+      BLI_assert(channel < 4);
+    }
 
-    if (use_eulers) {
-      compatible_eul(eul, oldEul);
+    BKE_driver_target_matrix_to_rot_channels(
+        mat, rot_order, dtar->rotation_mode, channel, false, quat);
+
+    if (use_eulers && dtar->rotation_mode == DTAR_ROTMODE_AUTO) {
+      compatible_eul(quat + 1, oldEul);
     }
 
-    return eul[dtar->transChan - DTAR_TRANSCHAN_ROTX];
+    return quat[channel];
   }
   else {
     /* extract location and choose right axis */
@@ -1744,6 +1757,71 @@ static float dvar_eval_transChan(ChannelDriver *driver, DriverVar *dvar)
   }
 }
 
+/* Convert a quaternion to pseudo-angles representing the weighted amount of rotation. */
+static void quaternion_to_angles(float quat[4], int channel)
+{
+  if (channel < 0) {
+    quat[0] = 2.0f * saacosf(quat[0]);
+
+    for (int i = 1; i < 4; i++) {
+      quat[i] = 2.0f * saasinf(quat[i]);
+    }
+  }
+  else if (channel == 0) {
+    quat[0] = 2.0f * saacosf(quat[0]);
+  }
+  else {
+    quat[channel] = 2.0f * saasinf(quat[channel]);
+  }
+}
+
+/* Compute channel values for a rotational Transform Channel driver variable. */
+void BKE_driver_target_matrix_to_rot_channels(
+    float mat[4][4], int auto_order, int rotation_mode, int channel, bool angles, float r_buf[4])
+{
+  float *const quat = r_buf;
+  float *const eul = r_buf + 1;
+
+  zero_v4(r_buf);
+
+  if (rotation_mode == DTAR_ROTMODE_AUTO) {
+    mat4_to_eulO(eul, auto_order, mat);
+  }
+  else if (rotation_mode >= DTAR_ROTMODE_EULER_MIN && rotation_mode <= DTAR_ROTMODE_EULER_MAX) {
+    mat4_to_eulO(eul, rotation_mode, mat);
+  }
+  else if (rotation_mode == DTAR_ROTMODE_QUATERNION) {
+    mat4_to_quat(quat, mat);
+
+    /* For Transformation constraint convenience, convert to pseudo-angles. */
+    if (angles) {
+      quaternion_to_angles(quat, channel);
+    }
+  }
+  else if (rotation_mode >= DTAR_ROTMODE_SWING_TWIST_X &&
+           rotation_mode <= DTAR_ROTMODE_SWING_TWIST_Z) {
+    int axis = rotation_mode - DTAR_ROTMODE_SWING_TWIST_X;
+    float raw_quat[4], twist;
+
+    mat4_to_quat(raw_quat, mat);
+
+    if (channel == axis + 1) {
+      /* If only the twist angle is needed, skip computing swing. */
+      twist = quat_split_swing_and_twist(raw_quat, axis, NULL, NULL);
+    }
+    else {
+      twist = quat_split_swing_and_twist(raw_quat, axis, quat, NULL);
+
+      quaternion_to_angles(quat, channel);
+    }
+
+    quat[axis + 1] = twist;
+  }
+  else {
+    BLI_assert(false);
+  }
+}
+
 /* ......... */
 
 /* Table of Driver Variable Type Info Data */
diff --git a/source/blender/blenkernel/intern/gpencil.c b/source/blender/blenkernel/intern/gpencil.c
index 220584bac6a..a00be90297b 100644
--- a/source/blender/blenkernel/intern/gpencil.c
+++ b/source/blender/blenkernel/intern/gpencil.c
@@ -152,19 +152,19 @@ bool BKE_gpencil_free_strokes(bGPDframe *gpf)
 }
 
 /* Free strokes and colors belonging to a gp-frame */
-bool BKE_gpencil_free_frame_runtime_data(bGPDframe *derived_gpf)
+bool BKE_gpencil_free_frame_runtime_data(bGPDframe *gpf_eval)
 {
   bGPDstroke *gps_next;
-  if (!derived_gpf) {
+  if (!gpf_eval) {
     return false;
   }
 
   /* free strokes */
-  for (bGPDstroke *gps = derived_gpf->strokes.first; gps; gps = gps_next) {
+  for (bGPDstroke *gps = gpf_eval->strokes.first; gps; gps = gps_next) {
     gps_next = gps->next;
     BKE_gpencil_free_stroke(gps);
   }
-  BLI_listbase_clear(&derived_gpf->strokes);
+  BLI_listbase_clear(&gpf_eval->strokes);
 
   return true;
 }
@@ -267,7 +267,8 @@ bGPDframe *BKE_gpencil_frame_addnew(bGPDlayer *gpl, int cframe)
 
   /* check whether frame was added successfully */
   if (state == -1) {
-    CLOG_ERROR(&LOG, "Frame (%d) existed already for this layer. Using existing frame", cframe);
+    CLOG_ERROR(
+        &LOG, "Frame (%d) existed already for this layer_active. Using existing frame", cframe);
 
     /* free the newly created one, and use the old one instead */
     MEM_freeN(gpf);
@@ -1040,6 +1041,37 @@ void BKE_gpencil_layer_setactive(bGPdata *gpd, bGPDlayer *active)
   }
 }
 
+/* Set locked layers for autolock mode. */
+void BKE_gpencil_layer_autolock_set(bGPdata *gpd)
+{
+  BLI_assert(gpd != NULL);
+
+  bGPDlayer *gpl;
+
+  if (gpd->flag & GP_DATA_AUTOLOCK_LAYERS) {
+    bGPDlayer *layer_active = BKE_gpencil_layer_getactive(gpd);
+
+    /* Lock all other layers */
+    for (gpl = gpd->layers.first; gpl; gpl = gpl->next) {
+      /* unlock active layer */
+      if (gpl == layer_active) {
+        gpl->flag &= ~GP_LAYER_LOCKED;
+      }
+      else {
+        gpl->flag |= GP_LAYER_LOCKED;
+      }
+    }
+  }
+  else {
+    /* If disable is better unlock all layers by default or it looks there is
+     * a problem in the UI because the user expects all layers will be unlocked
+     */
+    for (gpl = gpd->layers.first; gpl; gpl = gpl->next) {
+      gpl->flag &= ~GP_LAYER_LOCKED;
+    }
+  }
+}
+
 /* delete the active gp-layer */
 void BKE_gpencil_layer_delete(bGPdata *gpd, bGPDlayer *gpl)
 {
@@ -1190,38 +1222,34 @@ Material *BKE_gpencil_object_material_ensure_from_active_input_brush(Main *bmain
       brush->gpencil_settings->flag &= ~GP_BRUSH_MATERIAL_PINNED;
     }
   }
-  return BKE_gpencil_object_material_ensure_from_active_input_material(bmain, ob);
+  return BKE_gpencil_object_material_ensure_from_active_input_material(ob);
 }
 
 /**
  * Guaranteed to return a material assigned to object. Returns never NULL.
  * Only use this for materials unrelated to user input.
  */
-Material *BKE_gpencil_object_material_ensure_from_active_input_material(Main *bmain, Object *ob)
+Material *BKE_gpencil_object_material_ensure_from_active_input_material(Object *ob)
 {
   Material *ma = give_current_material(ob, ob->actcol);
   if (ma) {
     return ma;
   }
-  /* If the slot is empty, remove because will be added again,
-   * if not, we will get an empty slot. */
-  if ((ob->totcol > 0) && (ob->actcol == ob->totcol)) {
-    BKE_object_material_slot_remove(bmain, ob);
-  }
-  return BKE_gpencil_object_material_new(bmain, ob, "Material", NULL);
+
+  return &defgpencil_material;
 }
 
 /* Get active color, and add all default settings if we don't find anything */
-Material *BKE_gpencil_object_material_ensure_active(Main *bmain, Object *ob)
+Material *BKE_gpencil_object_material_ensure_active(Object *ob)
 {
   Material *ma = NULL;
 
   /* sanity checks */
-  if (ELEM(NULL, bmain, ob)) {
+  if (ob == NULL) {
     return NULL;
   }
 
-  ma = BKE_gpencil_object_material_ensure_from_active_input_material(bmain, ob);
+  ma = BKE_gpencil_object_material_ensure_from_active_input_material(ob);
   if (ma->gp_style == NULL) {
     BKE_material_init_gpencil_settings(ma);
   }
@@ -2048,33 +2076,55 @@ bool BKE_gpencil_smooth_stroke_strength(bGPDstroke *gps, int point_index, float
   bGPDspoint *ptb = &gps->points[point_index];
 
   /* Do nothing if not enough points */
-  if (gps->totpoints <= 2) {
+  if ((gps->totpoints <= 2) || (point_index < 1)) {
     return false;
   }
+  /* Only affect endpoints by a fraction of the normal influence */
+  float inf = influence;
+  if ((point_index == 0) || (point_index == gps->totpoints - 1)) {
+    inf *= 0.01f;
+  }
+  /* Limit max influence to reduce pop effect. */
+  CLAMP_MAX(inf, 0.98f);
 
-  /* Compute theoretical optimal value using distances */
-  bGPDspoint *pta, *ptc;
-  int before = point_index - 1;
-  int after = point_index + 1;
+  float total = 0.0f;
+  float max_strength = 0.0f;
+  const int steps = 4;
+  const float average_fac = 1.0f / (float)(steps * 2 + 1);
+  int step;
 
-  CLAMP_MIN(before, 0);
-  CLAMP_MAX(after, gps->totpoints - 1);
+  /* add the point itself */
+  total += ptb->strength * average_fac;
+  max_strength = ptb->strength;
 
-  pta = &gps->points[before];
-  ptc = &gps->points[after];
+  /* n-steps before/after current point */
+  for (step = 1; step <= steps; step++) {
+    bGPDspoint *pt1, *pt2;
+    int before = point_index - step;
+    int after = point_index + step;
 
-  /* the optimal value is the corresponding to the interpolation of the strength
-   * at the distance of point b
-   */
-  float fac = line_point_factor_v3(&ptb->x, &pta->x, &ptc->x);
-  /* sometimes the factor can be wrong due stroke geometry, so use middle point */
-  if ((fac < 0.0f) || (fac > 1.0f)) {
-    fac = 0.5f;
+    CLAMP_MIN(before, 0);
+    CLAMP_MAX(after, gps->totpoints - 1);
+
+    pt1 = &gps->points[before];
+    pt2 = &gps->points[after];
+
+    /* add both these points to the average-sum (s += p[i]/n) */
+    total += pt1->strength * average_fac;
+    total += pt2->strength * average_fac;
+    /* Save max value. */
+    if (max_strength < pt1->strength) {
+      max_strength = pt1->strength;
+    }
+    if (max_strength < pt2->strength) {
+      max_strength = pt2->strength;
+    }
   }
-  const float optimal = (1.0f - fac) * pta->strength + fac * ptc->strength;
 
-  /* Based on influence factor, blend between original and optimal */
-  ptb->strength = (1.0f - influence) * ptb->strength + influence * optimal;
+  /* Based on influence factor, blend between original and optimal smoothed value. */
+  ptb->strength = interpf(ptb->strength, total, inf);
+  /* Clamp to maximum stroke strength to avoid weird results. */
+  CLAMP_MAX(ptb->strength, max_strength);
 
   return true;
 }
@@ -2089,31 +2139,52 @@ bool BKE_gpencil_smooth_stroke_thickness(bGPDstroke *gps, int point_index, float
   if ((gps->totpoints <= 2) || (point_index < 1)) {
     return false;
   }
+  /* Only affect endpoints by a fraction of the normal influence */
+  float inf = influence;
+  if ((point_index == 0) || (point_index == gps->totpoints - 1)) {
+    inf *= 0.01f;
+  }
+  /* Limit max influence to reduce pop effect. */
+  CLAMP_MAX(inf, 0.98f);
 
-  /* Compute theoretical optimal value using distances */
-  bGPDspoint *pta, *ptc;
-  int before = point_index - 1;
-  int after = point_index + 1;
+  float total = 0.0f;
+  float max_pressure = 0.0f;
+  const int steps = 4;
+  const float average_fac = 1.0f / (float)(steps * 2 + 1);
+  int step;
 
-  CLAMP_MIN(before, 0);
-  CLAMP_MAX(after, gps->totpoints - 1);
+  /* add the point itself */
+  total += ptb->pressure * average_fac;
+  max_pressure = ptb->pressure;
 
-  pta = &gps->points[before];
-  ptc = &gps->points[after];
+  /* n-steps before/after current point */
+  for (step = 1; step <= steps; step++) {
+    bGPDspoint *pt1, *pt2;
+    int before = point_index - step;
+    int after = point_index + step;
 
-  /* the optimal value is the corresponding to the interpolation of the pressure
-   * at the distance of point b
-   */
-  float fac = line_point_factor_v3(&ptb->x, &pta->x, &ptc->x);
-  /* sometimes the factor can be wrong due stroke geometry, so use middle point */
-  if ((fac < 0.0f) || (fac > 1.0f)) {
-    fac = 0.5f;
-  }
-  float optimal = interpf(ptc->pressure, pta->pressure, fac);
+    CLAMP_MIN(before, 0);
+    CLAMP_MAX(after, gps->totpoints - 1);
 
-  /* Based on influence factor, blend between original and optimal */
-  ptb->pressure = interpf(optimal, ptb->pressure, influence);
+    pt1 = &gps->points[before];
+    pt2 = &gps->points[after];
+
+    /* add both these points to the average-sum (s += p[i]/n) */
+    total += pt1->pressure * average_fac;
+    total += pt2->pressure * average_fac;
+    /* Save max value. */
+    if (max_pressure < pt1->pressure) {
+      max_pressure = pt1->pressure;
+    }
+    if (max_pressure < pt2->pressure) {
+      max_pressure = pt2->pressure;
+    }
+  }
 
+  /* Based on influence factor, blend between original and optimal smoothed value. */
+  ptb->pressure = interpf(ptb->pressure, total, inf);
+  /* Clamp to maximum stroke thickness to avoid weird results. */
+  CLAMP_MAX(ptb->pressure, max_pressure);
   return true;
 }
 
@@ -2229,6 +2300,7 @@ void BKE_gpencil_material_index_reassign(bGPdata *gpd, int totcol, int index)
         /* reassign strokes */
         if ((gps->mat_nr > index) || (gps->mat_nr > totcol - 1)) {
           gps->mat_nr--;
+          CLAMP_MIN(gps->mat_nr, 0);
         }
       }
     }
@@ -3134,3 +3206,397 @@ void BKE_gpencil_merge_distance_stroke(bGPDframe *gpf,
     BKE_gpencil_dissolve_points(gpf, gps, GP_SPOINT_TAG);
   }
 }
+
+/* Helper: Check materials with same color. */
+static int gpencil_check_same_material_color(Object *ob_gp, float color[4], Material **r_mat)
+{
+  Material *ma = NULL;
+  float color_cu[4];
+  linearrgb_to_srgb_v3_v3(color_cu, color);
+  float hsv1[4];
+  rgb_to_hsv_v(color_cu, hsv1);
+  hsv1[3] = color[3];
+
+  for (int i = 1; i <= ob_gp->totcol; i++) {
+    ma = give_current_material(ob_gp, i);
+    MaterialGPencilStyle *gp_style = ma->gp_style;
+    /* Check color with small tolerance (better in HSV). */
+    float hsv2[4];
+    rgb_to_hsv_v(gp_style->fill_rgba, hsv2);
+    hsv2[3] = gp_style->fill_rgba[3];
+    if ((gp_style->fill_style == GP_STYLE_FILL_STYLE_SOLID) && (compare_v4v4(hsv1, hsv2, 0.01f))) {
+      *r_mat = ma;
+      return i - 1;
+    }
+  }
+
+  *r_mat = NULL;
+  return -1;
+}
+
+/* Helper: Add gpencil material using curve material as base. */
+static Material *gpencil_add_from_curve_material(Main *bmain,
+                                                 Object *ob_gp,
+                                                 float cu_color[4],
+                                                 const bool gpencil_lines,
+                                                 const bool fill,
+                                                 int *r_idx)
+{
+  Material *mat_gp = BKE_gpencil_object_material_new(
+      bmain, ob_gp, (fill) ? "Material" : "Unassigned", r_idx);
+  MaterialGPencilStyle *gp_style = mat_gp->gp_style;
+
+  /* Stroke color. */
+  if (gpencil_lines) {
+    ARRAY_SET_ITEMS(gp_style->stroke_rgba, 0.0f, 0.0f, 0.0f, 1.0f);
+    gp_style->flag |= GP_STYLE_STROKE_SHOW;
+  }
+  else {
+    linearrgb_to_srgb_v4(gp_style->stroke_rgba, cu_color);
+    gp_style->flag &= ~GP_STYLE_STROKE_SHOW;
+  }
+
+  /* Fill color. */
+  linearrgb_to_srgb_v4(gp_style->fill_rgba, cu_color);
+  /* Fill is false if the original curve hasn't material assigned, so enable it. */
+  if (fill) {
+    gp_style->flag |= GP_STYLE_FILL_SHOW;
+  }
+
+  /* Check at least one is enabled. */
+  if (((gp_style->flag & GP_STYLE_STROKE_SHOW) == 0) &&
+      ((gp_style->flag & GP_STYLE_FILL_SHOW) == 0)) {
+    gp_style->flag |= GP_STYLE_STROKE_SHOW;
+  }
+
+  return mat_gp;
+}
+
+/* Helper: Create new stroke section. */
+static void gpencil_add_new_points(bGPDstroke *gps,
+                                   float *coord_array,
+                                   float pressure,
+                                   int init,
+                                   int totpoints,
+                                   float init_co[3],
+                                   bool last)
+{
+  for (int i = 0; i < totpoints; i++) {
+    bGPDspoint *pt = &gps->points[i + init];
+    copy_v3_v3(&pt->x, &coord_array[3 * i]);
+    /* Be sure the last point is not on top of the first point of the curve or
+     * the close of the stroke will produce glitches. */
+    if ((last) && (i > 0) && (i == totpoints - 1)) {
+      float dist = len_v3v3(init_co, &pt->x);
+      if (dist < 0.1f) {
+        /* Interpolate between previous point and current to back slightly. */
+        bGPDspoint *pt_prev = &gps->points[i + init - 1];
+        interp_v3_v3v3(&pt->x, &pt_prev->x, &pt->x, 0.95f);
+      }
+    }
+
+    pt->pressure = pressure;
+    pt->strength = 1.0f;
+  }
+}
+
+/* Helper: Get the first collection that includes the object. */
+static Collection *gpencil_get_parent_collection(Scene *scene, Object *ob)
+{
+  Collection *mycol = NULL;
+  FOREACH_SCENE_COLLECTION_BEGIN (scene, collection) {
+    for (CollectionObject *cob = collection->gobject.first; cob; cob = cob->next) {
+      if ((mycol == NULL) && (cob->ob == ob)) {
+        mycol = collection;
+      }
+    }
+  }
+  FOREACH_SCENE_COLLECTION_END;
+
+  return mycol;
+}
+
+/* Helper: Convert one spline to grease pencil stroke. */
+static void gpencil_convert_spline(Main *bmain,
+                                   Object *ob_gp,
+                                   Object *ob_cu,
+                                   const bool gpencil_lines,
+                                   const bool only_stroke,
+                                   bGPDframe *gpf,
+                                   Nurb *nu)
+{
+  Curve *cu = (Curve *)ob_cu->data;
+  bool cyclic = true;
+
+  /* Create Stroke. */
+  bGPDstroke *gps = MEM_callocN(sizeof(bGPDstroke), "bGPDstroke");
+  gps->thickness = 1.0f;
+  gps->gradient_f = 1.0f;
+  ARRAY_SET_ITEMS(gps->gradient_s, 1.0f, 1.0f);
+  ARRAY_SET_ITEMS(gps->caps, GP_STROKE_CAP_ROUND, GP_STROKE_CAP_ROUND);
+  gps->inittime = 0.0f;
+
+  /* Enable recalculation flag by default. */
+  gps->flag |= GP_STROKE_RECALC_GEOMETRY;
+  gps->flag &= ~GP_STROKE_SELECT;
+  gps->flag |= GP_STROKE_3DSPACE;
+
+  gps->mat_nr = 0;
+  /* Count total points
+   * The total of points must consider that last point of each segment is equal to the first
+   * point of next segment.
+   */
+  int totpoints = 0;
+  int segments = 0;
+  int resolu = nu->resolu + 1;
+  segments = nu->pntsu;
+  if (((nu->flagu & CU_NURB_CYCLIC) == 0) || (nu->pntsu == 2)) {
+    segments--;
+    cyclic = false;
+  }
+  totpoints = (resolu * segments) - (segments - 1);
+
+  /* Initialize triangle memory to dummy data. */
+  gps->tot_triangles = 0;
+  gps->triangles = NULL;
+
+  /* Materials
+   * Notice: The color of the material is the color of viewport and not the final shader color.
+   */
+  Material *mat_gp = NULL;
+  bool fill = true;
+  /* Check if grease pencil has a material with same color.*/
+  float color[4];
+  if ((cu->mat) && (*cu->mat)) {
+    Material *mat_cu = *cu->mat;
+    copy_v4_v4(color, &mat_cu->r);
+  }
+  else {
+    /* Gray (unassigned from SVG add-on) */
+    zero_v4(color);
+    add_v3_fl(color, 0.6f);
+    color[3] = 1.0f;
+    fill = false;
+  }
+
+  /* Special case: If the color was created by the SVG add-on and the name contains '_stroke' and
+   * there is only one color, the stroke must not be closed, fill to false and use for
+   * stroke the fill color.
+   */
+  bool do_stroke = false;
+  if (ob_cu->totcol == 1) {
+    Material *ma_stroke = give_current_material(ob_cu, 1);
+    if ((ma_stroke) && (strstr(ma_stroke->id.name, "_stroke") != NULL)) {
+      do_stroke = true;
+    }
+  }
+
+  int r_idx = gpencil_check_same_material_color(ob_gp, color, &mat_gp);
+  if ((ob_cu->totcol > 0) && (r_idx < 0)) {
+    Material *mat_curve = give_current_material(ob_cu, 1);
+    mat_gp = gpencil_add_from_curve_material(bmain, ob_gp, color, gpencil_lines, fill, &r_idx);
+
+    if ((mat_curve) && (mat_curve->gp_style != NULL)) {
+      MaterialGPencilStyle *gp_style_cur = mat_curve->gp_style;
+      MaterialGPencilStyle *gp_style_gp = mat_gp->gp_style;
+
+      copy_v4_v4(gp_style_gp->mix_rgba, gp_style_cur->mix_rgba);
+      gp_style_gp->fill_style = gp_style_cur->fill_style;
+      gp_style_gp->mix_factor = gp_style_cur->mix_factor;
+      gp_style_gp->gradient_angle = gp_style_cur->gradient_angle;
+    }
+
+    /* If object has more than 1 material, use second material for stroke color. */
+    if ((!only_stroke) && (ob_cu->totcol > 1) && (give_current_material(ob_cu, 2))) {
+      mat_curve = give_current_material(ob_cu, 2);
+      linearrgb_to_srgb_v3_v3(mat_gp->gp_style->stroke_rgba, &mat_curve->r);
+      mat_gp->gp_style->stroke_rgba[3] = mat_curve->a;
+    }
+    else if ((only_stroke) || (do_stroke)) {
+      /* Also use the first color if the fill is none for stroke color. */
+      if (ob_cu->totcol > 0) {
+        mat_curve = give_current_material(ob_cu, 1);
+        linearrgb_to_srgb_v3_v3(mat_gp->gp_style->stroke_rgba, &mat_curve->r);
+        mat_gp->gp_style->stroke_rgba[3] = mat_curve->a;
+        /* Set fill and stroke depending of curve type (3D or 2D). */
+        if ((cu->flag & CU_3D) || ((cu->flag & (CU_FRONT | CU_BACK)) == 0)) {
+          mat_gp->gp_style->flag |= GP_STYLE_STROKE_SHOW;
+          mat_gp->gp_style->flag &= ~GP_STYLE_FILL_SHOW;
+        }
+        else {
+          mat_gp->gp_style->flag &= ~GP_STYLE_STROKE_SHOW;
+          mat_gp->gp_style->flag |= GP_STYLE_FILL_SHOW;
+        }
+      }
+    }
+  }
+  CLAMP_MIN(r_idx, 0);
+
+  /* Assign material index to stroke. */
+  gps->mat_nr = r_idx;
+
+  /* Add stroke to frame.*/
+  BLI_addtail(&gpf->strokes, gps);
+
+  float *coord_array = NULL;
+  float init_co[3];
+
+  switch (nu->type) {
+    case CU_POLY: {
+      /* Allocate memory for storage points. */
+      gps->totpoints = nu->pntsu;
+      gps->points = MEM_callocN(sizeof(bGPDspoint) * gps->totpoints, "gp_stroke_points");
+      /* Increase thickness for this type. */
+      gps->thickness = 10.0f;
+
+      /* Get all curve points */
+      for (int s = 0; s < gps->totpoints; s++) {
+        BPoint *bp = &nu->bp[s];
+        bGPDspoint *pt = &gps->points[s];
+        copy_v3_v3(&pt->x, bp->vec);
+        pt->pressure = bp->radius;
+        pt->strength = 1.0f;
+      }
+      break;
+    }
+    case CU_BEZIER: {
+      /* Allocate memory for storage points. */
+      gps->totpoints = totpoints;
+      gps->points = MEM_callocN(sizeof(bGPDspoint) * gps->totpoints, "gp_stroke_points");
+
+      int init = 0;
+      resolu = nu->resolu + 1;
+      segments = nu->pntsu;
+      if (((nu->flagu & CU_NURB_CYCLIC) == 0) || (nu->pntsu == 2)) {
+        segments--;
+      }
+      /* Get all interpolated curve points of Beziert */
+      for (int s = 0; s < segments; s++) {
+        int inext = (s + 1) % nu->pntsu;
+        BezTriple *prevbezt = &nu->bezt[s];
+        BezTriple *bezt = &nu->bezt[inext];
+        bool last = (bool)(s == segments - 1);
+
+        coord_array = MEM_callocN((size_t)3 * resolu * sizeof(float), __func__);
+
+        for (int j = 0; j < 3; j++) {
+          BKE_curve_forward_diff_bezier(prevbezt->vec[1][j],
+                                        prevbezt->vec[2][j],
+                                        bezt->vec[0][j],
+                                        bezt->vec[1][j],
+                                        coord_array + j,
+                                        resolu - 1,
+                                        3 * sizeof(float));
+        }
+        /* Save first point coordinates. */
+        if (s == 0) {
+          copy_v3_v3(init_co, &coord_array[0]);
+        }
+        /* Add points to the stroke */
+        gpencil_add_new_points(gps, coord_array, bezt->radius, init, resolu, init_co, last);
+        /* Free memory. */
+        MEM_SAFE_FREE(coord_array);
+
+        /* As the last point of segment is the first point of next segment, back one array
+         * element to avoid duplicated points on the same location.
+         */
+        init += resolu - 1;
+      }
+      break;
+    }
+    case CU_NURBS: {
+      if (nu->pntsv == 1) {
+
+        int nurb_points;
+        if (nu->flagu & CU_NURB_CYCLIC) {
+          resolu++;
+          nurb_points = nu->pntsu * resolu;
+        }
+        else {
+          nurb_points = (nu->pntsu - 1) * resolu;
+        }
+        /* Get all curve points. */
+        coord_array = MEM_callocN(sizeof(float[3]) * nurb_points, __func__);
+        BKE_nurb_makeCurve(nu, coord_array, NULL, NULL, NULL, resolu, sizeof(float[3]));
+
+        /* Allocate memory for storage points. */
+        gps->totpoints = nurb_points - 1;
+        gps->points = MEM_callocN(sizeof(bGPDspoint) * gps->totpoints, "gp_stroke_points");
+
+        /* Add points. */
+        gpencil_add_new_points(gps, coord_array, 1.0f, 0, gps->totpoints, init_co, false);
+
+        MEM_SAFE_FREE(coord_array);
+      }
+      break;
+    }
+    default: {
+      break;
+    }
+  }
+  /* Cyclic curve, close stroke. */
+  if ((cyclic) && (!do_stroke)) {
+    BKE_gpencil_close_stroke(gps);
+  }
+}
+
+/* Convert a curve object to grease pencil stroke.
+ *
+ * \param bmain: Main thread pointer
+ * \param scene: Original scene.
+ * \param ob_gp: Grease pencil object to add strokes.
+ * \param ob_cu: Curve to convert.
+ * \param gpencil_lines: Use lines for strokes.
+ * \param use_collections: Create layers using collection names.
+ * \param only_stroke: The material must be only stroke without fill.
+ */
+void BKE_gpencil_convert_curve(Main *bmain,
+                               Scene *scene,
+                               Object *ob_gp,
+                               Object *ob_cu,
+                               const bool gpencil_lines,
+                               const bool use_collections,
+                               const bool only_stroke)
+{
+  if (ELEM(NULL, ob_gp, ob_cu) || (ob_gp->type != OB_GPENCIL) || (ob_gp->data == NULL)) {
+    return;
+  }
+
+  Curve *cu = (Curve *)ob_cu->data;
+  bGPdata *gpd = (bGPdata *)ob_gp->data;
+  bGPDlayer *gpl = NULL;
+
+  /* If the curve is empty, cancel. */
+  if (cu->nurb.first == NULL) {
+    return;
+  }
+
+  /* Check if there is an active layer. */
+  if (use_collections) {
+    Collection *collection = gpencil_get_parent_collection(scene, ob_cu);
+    if (collection != NULL) {
+      gpl = BLI_findstring(&gpd->layers, collection->id.name + 2, offsetof(bGPDlayer, info));
+      if (gpl == NULL) {
+        gpl = BKE_gpencil_layer_addnew(gpd, collection->id.name + 2, true);
+      }
+    }
+  }
+
+  if (gpl == NULL) {
+    gpl = BKE_gpencil_layer_getactive(gpd);
+    if (gpl == NULL) {
+      gpl = BKE_gpencil_layer_addnew(gpd, DATA_("GP_Layer"), true);
+    }
+  }
+
+  /* Check if there is an active frame and add if needed. */
+  bGPDframe *gpf = BKE_gpencil_layer_getframe(gpl, CFRA, GP_GETFRAME_ADD_COPY);
+
+  /* Read all splines of the curve and create a stroke for each. */
+  for (Nurb *nu = cu->nurb.first; nu; nu = nu->next) {
+    gpencil_convert_spline(bmain, ob_gp, ob_cu, gpencil_lines, only_stroke, gpf, nu);
+  }
+
+  /* Tag for recalculation */
+  DEG_id_tag_update(&gpd->id, ID_RECALC_GEOMETRY | ID_RECALC_COPY_ON_WRITE);
+}
diff --git a/source/blender/blenkernel/intern/gpencil_modifier.c b/source/blender/blenkernel/intern/gpencil_modifier.c
index 36d2d6316f7..1750a389788 100644
--- a/source/blender/blenkernel/intern/gpencil_modifier.c
+++ b/source/blender/blenkernel/intern/gpencil_modifier.c
@@ -779,26 +779,28 @@ void BKE_gpencil_subdivide(bGPDstroke *gps, int level, int flag)
 }
 
 /* Copy frame but do not assign new memory */
-static void gpencil_copy_frame(Object *ob, bGPDframe *gpf, bGPDframe *derived_gpf)
+static void gpencil_frame_copy_noalloc(Object *ob, bGPDframe *gpf, bGPDframe *gpf_eval)
 {
-  derived_gpf->prev = gpf->prev;
-  derived_gpf->next = gpf->next;
-  derived_gpf->framenum = gpf->framenum;
-  derived_gpf->flag = gpf->flag;
-  derived_gpf->key_type = gpf->key_type;
-  derived_gpf->runtime = gpf->runtime;
-  copy_m4_m4(derived_gpf->runtime.parent_obmat, gpf->runtime.parent_obmat);
+  gpf_eval->prev = gpf->prev;
+  gpf_eval->next = gpf->next;
+  gpf_eval->framenum = gpf->framenum;
+  gpf_eval->flag = gpf->flag;
+  gpf_eval->key_type = gpf->key_type;
+  gpf_eval->runtime = gpf->runtime;
+  copy_m4_m4(gpf_eval->runtime.parent_obmat, gpf->runtime.parent_obmat);
 
   /* copy strokes */
-  BLI_listbase_clear(&derived_gpf->strokes);
+  BLI_listbase_clear(&gpf_eval->strokes);
   for (bGPDstroke *gps_src = gpf->strokes.first; gps_src; gps_src = gps_src->next) {
     /* make copy of source stroke */
     bGPDstroke *gps_dst = BKE_gpencil_stroke_duplicate(gps_src);
 
     /* copy color to temp fields to apply temporal changes in the stroke */
     MaterialGPencilStyle *gp_style = BKE_material_gpencil_settings_get(ob, gps_src->mat_nr + 1);
-    copy_v4_v4(gps_dst->runtime.tmp_stroke_rgba, gp_style->stroke_rgba);
-    copy_v4_v4(gps_dst->runtime.tmp_fill_rgba, gp_style->fill_rgba);
+    if (gp_style) {
+      copy_v4_v4(gps_dst->runtime.tmp_stroke_rgba, gp_style->stroke_rgba);
+      copy_v4_v4(gps_dst->runtime.tmp_fill_rgba, gp_style->fill_rgba);
+    }
 
     /* Save original pointers for using in edit and select operators. */
     gps_dst->runtime.gps_orig = gps_src;
@@ -808,25 +810,27 @@ static void gpencil_copy_frame(Object *ob, bGPDframe *gpf, bGPDframe *derived_gp
       pt_dst->runtime.idx_orig = i;
     }
 
-    BLI_addtail(&derived_gpf->strokes, gps_dst);
+    BLI_addtail(&gpf_eval->strokes, gps_dst);
   }
 }
 
-/* Ensure there is a derived frame */
-static void gpencil_ensure_derived_frame(
-    int idx, Object *ob, bGPDlayer *gpl, bGPDframe *gpf, bGPDframe **derived_gpf)
+/* Ensure there is a evaluated frame */
+static void gpencil_evaluated_frame_ensure(int idx,
+                                           Object *ob,
+                                           bGPDframe *gpf,
+                                           bGPDframe **gpf_eval)
 {
-  /* create derived frames array data or expand */
-  bGPDframe *derived_frames = ob->runtime.derived_frames;
-  *derived_gpf = &derived_frames[idx];
+  /* Create evaluated frames array data or expand. */
+  bGPDframe *evaluated_frames = ob->runtime.gpencil_evaluated_frames;
+  *gpf_eval = &evaluated_frames[idx];
 
-  /* if derived frame create a new one */
-  if (*derived_gpf != NULL) {
+  /* If already exist a evaluated frame create a new one. */
+  if (*gpf_eval != NULL) {
     /* first clear temp data */
-    BKE_gpencil_free_frame_runtime_data(*derived_gpf);
+    BKE_gpencil_free_frame_runtime_data(*gpf_eval);
   }
-  /* copy data (do not assign new memory)*/
-  gpencil_copy_frame(ob, gpf, *derived_gpf);
+  /* Copy data (do not assign new memory). */
+  gpencil_frame_copy_noalloc(ob, gpf, *gpf_eval);
 }
 
 /* Calculate gpencil modifiers */
@@ -836,21 +840,22 @@ void BKE_gpencil_modifiers_calc(Depsgraph *depsgraph, Scene *scene, Object *ob)
   Object *ob_orig = DEG_get_original_object(ob);
   bGPdata *gpd = (bGPdata *)ob_orig->data;
   const bool is_multiedit = (bool)GPENCIL_MULTIEDIT_SESSIONS_ON(gpd);
-  const bool simplify_modif = GP_SIMPLIFY_MODIF(scene, false);
+  const bool simplify_modif = GPENCIL_SIMPLIFY_MODIF(scene, false);
   const bool is_render = (bool)(DEG_get_mode(depsgraph) == DAG_EVAL_RENDER);
   const bool time_remap = BKE_gpencil_has_time_modifiers(ob);
   int cfra_eval = (int)DEG_get_ctime(depsgraph);
 
-  /* Create array of derived frames equal to number of layers. */
-  ob->runtime.tot_layers = BLI_listbase_count(&gpd->layers);
-  CLAMP_MIN(ob->runtime.tot_layers, 1);
-  if (ob->runtime.derived_frames == NULL) {
-    ob->runtime.derived_frames = MEM_callocN(sizeof(struct bGPDframe) * ob->runtime.tot_layers,
-                                             __func__);
+  /* Create array of evaluated frames equal to number of layers. */
+  ob->runtime.gpencil_tot_layers = BLI_listbase_count(&gpd->layers);
+  CLAMP_MIN(ob->runtime.gpencil_tot_layers, 1);
+  if (ob->runtime.gpencil_evaluated_frames == NULL) {
+    ob->runtime.gpencil_evaluated_frames = MEM_callocN(
+        sizeof(struct bGPDframe) * ob->runtime.gpencil_tot_layers, __func__);
   }
   else {
-    ob->runtime.derived_frames = MEM_recallocN(ob->runtime.derived_frames,
-                                               sizeof(struct bGPDframe) * ob->runtime.tot_layers);
+    ob->runtime.gpencil_evaluated_frames = MEM_recallocN(ob->runtime.gpencil_evaluated_frames,
+                                                         sizeof(struct bGPDframe) *
+                                                             ob->runtime.gpencil_tot_layers);
   }
 
   /* Init general modifiers data. */
@@ -877,8 +882,8 @@ void BKE_gpencil_modifiers_calc(Depsgraph *depsgraph, Scene *scene, Object *ob)
     }
 
     /* Create a duplicate data set of stroke to modify. */
-    bGPDframe *derived_gpf = NULL;
-    gpencil_ensure_derived_frame(idx, ob, gpl, gpf, &derived_gpf);
+    bGPDframe *gpf_eval = NULL;
+    gpencil_evaluated_frame_ensure(idx, ob, gpf, &gpf_eval);
 
     /* Skip all if some disable flag is enabled. */
     if ((ob->greasepencil_modifiers.first == NULL) || (is_multiedit) || (simplify_modif)) {
@@ -888,13 +893,13 @@ void BKE_gpencil_modifiers_calc(Depsgraph *depsgraph, Scene *scene, Object *ob)
 
     /* Apply geometry modifiers (create new geometry). */
     if (BKE_gpencil_has_geometry_modifiers(ob)) {
-      BKE_gpencil_geometry_modifiers(depsgraph, ob, gpl, derived_gpf, is_render);
+      BKE_gpencil_geometry_modifiers(depsgraph, ob, gpl, gpf_eval, is_render);
     }
 
     /* Loop all strokes and deform them. */
-    for (bGPDstroke *gps = derived_gpf->strokes.first; gps; gps = gps->next) {
+    for (bGPDstroke *gps = gpf_eval->strokes.first; gps; gps = gps->next) {
       /* Apply modifiers that only deform geometry */
-      BKE_gpencil_stroke_modifiers(depsgraph, ob, gpl, derived_gpf, gps, is_render);
+      BKE_gpencil_stroke_modifiers(depsgraph, ob, gpl, gpf_eval, gps, is_render);
     }
 
     idx++;
diff --git a/source/blender/blenkernel/intern/key.c b/source/blender/blenkernel/intern/key.c
index 61de6a8c06a..9a38f348361 100644
--- a/source/blender/blenkernel/intern/key.c
+++ b/source/blender/blenkernel/intern/key.c
@@ -2066,9 +2066,9 @@ void BKE_keyblock_mesh_calc_normals(struct KeyBlock *kb,
 }
 
 /************************* raw coords ************************/
-void BKE_keyblock_update_from_vertcos(Object *ob, KeyBlock *kb, float (*vertCos)[3])
+void BKE_keyblock_update_from_vertcos(Object *ob, KeyBlock *kb, const float (*vertCos)[3])
 {
-  float(*co)[3] = vertCos;
+  const float(*co)[3] = vertCos;
   float *fp = kb->data;
   int tot, a;
 
@@ -2126,7 +2126,7 @@ void BKE_keyblock_update_from_vertcos(Object *ob, KeyBlock *kb, float (*vertCos)
   }
 }
 
-void BKE_keyblock_convert_from_vertcos(Object *ob, KeyBlock *kb, float (*vertCos)[3])
+void BKE_keyblock_convert_from_vertcos(Object *ob, KeyBlock *kb, const float (*vertCos)[3])
 {
   int tot = 0, elemsize;
 
@@ -2219,7 +2219,7 @@ float (*BKE_keyblock_convert_to_vertcos(Object *ob, KeyBlock *kb))[3]
 }
 
 /************************* raw coord offsets ************************/
-void BKE_keyblock_update_from_offset(Object *ob, KeyBlock *kb, float (*ofs)[3])
+void BKE_keyblock_update_from_offset(Object *ob, KeyBlock *kb, const float (*ofs)[3])
 {
   int a;
   float *fp = kb->data;
diff --git a/source/blender/blenkernel/intern/lattice.c b/source/blender/blenkernel/intern/lattice.c
index 83b354ffb4a..e46b7ca5130 100644
--- a/source/blender/blenkernel/intern/lattice.c
+++ b/source/blender/blenkernel/intern/lattice.c
@@ -139,7 +139,7 @@ void BKE_lattice_resize(Lattice *lt, int uNew, int vNew, int wNew, Object *ltOb)
   BPoint *bp;
   int i, u, v, w;
   float fu, fv, fw, uc, vc, wc, du = 0.0, dv = 0.0, dw = 0.0;
-  float *co, (*vertexCos)[3] = NULL;
+  float *co, (*vert_coords)[3] = NULL;
 
   /* vertex weight groups are just freed all for now */
   if (lt->dvert) {
@@ -159,7 +159,7 @@ void BKE_lattice_resize(Lattice *lt, int uNew, int vNew, int wNew, Object *ltOb)
     }
   }
 
-  vertexCos = MEM_mallocN(sizeof(*vertexCos) * uNew * vNew * wNew, "tmp_vcos");
+  vert_coords = MEM_mallocN(sizeof(*vert_coords) * uNew * vNew * wNew, "tmp_vcos");
 
   calc_lat_fudu(lt->flag, uNew, &fu, &du);
   calc_lat_fudu(lt->flag, vNew, &fv, &dv);
@@ -187,7 +187,7 @@ void BKE_lattice_resize(Lattice *lt, int uNew, int vNew, int wNew, Object *ltOb)
     }
   }
 
-  co = vertexCos[0];
+  co = vert_coords[0];
   for (w = 0, wc = fw; w < wNew; w++, wc += dw) {
     for (v = 0, vc = fv; v < vNew; v++, vc += dv) {
       for (u = 0, uc = fu; u < uNew; u++, co += 3, uc += du) {
@@ -212,7 +212,7 @@ void BKE_lattice_resize(Lattice *lt, int uNew, int vNew, int wNew, Object *ltOb)
 
     copy_m4_m4(mat, ltOb->obmat);
     unit_m4(ltOb->obmat);
-    lattice_deform_verts(ltOb, NULL, NULL, vertexCos, uNew * vNew * wNew, NULL, 1.0f);
+    lattice_deform_verts(ltOb, NULL, NULL, vert_coords, uNew * vNew * wNew, NULL, 1.0f);
     copy_m4_m4(ltOb->obmat, mat);
 
     lt->typeu = typeu;
@@ -238,10 +238,10 @@ void BKE_lattice_resize(Lattice *lt, int uNew, int vNew, int wNew, Object *ltOb)
   bp = lt->def;
 
   for (i = 0; i < lt->pntsu * lt->pntsv * lt->pntsw; i++, bp++) {
-    copy_v3_v3(bp->vec, vertexCos[i]);
+    copy_v3_v3(bp->vec, vert_coords[i]);
   }
 
-  MEM_freeN(vertexCos);
+  MEM_freeN(vert_coords);
 }
 
 void BKE_lattice_init(Lattice *lt)
@@ -747,7 +747,7 @@ static bool calc_curve_deform(
 
 void curve_deform_verts(Object *cuOb,
                         Object *target,
-                        float (*vertexCos)[3],
+                        float (*vert_coords)[3],
                         int numVerts,
                         MDeformVert *dvert,
                         const int defgrp_index,
@@ -786,11 +786,11 @@ void curve_deform_verts(Object *cuOb,
         const float weight = defvert_find_weight(dvert_iter, defgrp_index);
 
         if (weight > 0.0f) {
-          mul_m4_v3(cd.curvespace, vertexCos[a]);
-          copy_v3_v3(vec, vertexCos[a]);
+          mul_m4_v3(cd.curvespace, vert_coords[a]);
+          copy_v3_v3(vec, vert_coords[a]);
           calc_curve_deform(cuOb, vec, defaxis, &cd, NULL);
-          interp_v3_v3v3(vertexCos[a], vertexCos[a], vec, weight);
-          mul_m4_v3(cd.objectspace, vertexCos[a]);
+          interp_v3_v3v3(vert_coords[a], vert_coords[a], vec, weight);
+          mul_m4_v3(cd.objectspace, vert_coords[a]);
         }
       }
     }
@@ -800,8 +800,8 @@ void curve_deform_verts(Object *cuOb,
 
       for (a = 0, dvert_iter = dvert; a < numVerts; a++, dvert_iter++) {
         if (defvert_find_weight(dvert_iter, defgrp_index) > 0.0f) {
-          mul_m4_v3(cd.curvespace, vertexCos[a]);
-          minmax_v3v3_v3(cd.dmin, cd.dmax, vertexCos[a]);
+          mul_m4_v3(cd.curvespace, vert_coords[a]);
+          minmax_v3v3_v3(cd.dmin, cd.dmax, vert_coords[a]);
         }
       }
 
@@ -810,10 +810,10 @@ void curve_deform_verts(Object *cuOb,
 
         if (weight > 0.0f) {
           /* already in 'cd.curvespace', prev for loop */
-          copy_v3_v3(vec, vertexCos[a]);
+          copy_v3_v3(vec, vert_coords[a]);
           calc_curve_deform(cuOb, vec, defaxis, &cd, NULL);
-          interp_v3_v3v3(vertexCos[a], vertexCos[a], vec, weight);
-          mul_m4_v3(cd.objectspace, vertexCos[a]);
+          interp_v3_v3v3(vert_coords[a], vert_coords[a], vec, weight);
+          mul_m4_v3(cd.objectspace, vert_coords[a]);
         }
       }
     }
@@ -821,9 +821,9 @@ void curve_deform_verts(Object *cuOb,
   else {
     if (cu->flag & CU_DEFORM_BOUNDS_OFF) {
       for (a = 0; a < numVerts; a++) {
-        mul_m4_v3(cd.curvespace, vertexCos[a]);
-        calc_curve_deform(cuOb, vertexCos[a], defaxis, &cd, NULL);
-        mul_m4_v3(cd.objectspace, vertexCos[a]);
+        mul_m4_v3(cd.curvespace, vert_coords[a]);
+        calc_curve_deform(cuOb, vert_coords[a], defaxis, &cd, NULL);
+        mul_m4_v3(cd.objectspace, vert_coords[a]);
       }
     }
     else {
@@ -831,14 +831,14 @@ void curve_deform_verts(Object *cuOb,
       INIT_MINMAX(cd.dmin, cd.dmax);
 
       for (a = 0; a < numVerts; a++) {
-        mul_m4_v3(cd.curvespace, vertexCos[a]);
-        minmax_v3v3_v3(cd.dmin, cd.dmax, vertexCos[a]);
+        mul_m4_v3(cd.curvespace, vert_coords[a]);
+        minmax_v3v3_v3(cd.dmin, cd.dmax, vert_coords[a]);
       }
 
       for (a = 0; a < numVerts; a++) {
         /* already in 'cd.curvespace', prev for loop */
-        calc_curve_deform(cuOb, vertexCos[a], defaxis, &cd, NULL);
-        mul_m4_v3(cd.objectspace, vertexCos[a]);
+        calc_curve_deform(cuOb, vert_coords[a], defaxis, &cd, NULL);
+        mul_m4_v3(cd.objectspace, vert_coords[a]);
       }
     }
   }
@@ -881,7 +881,7 @@ void curve_deform_vector(
 
 typedef struct LatticeDeformUserdata {
   LatticeDeformData *lattice_deform_data;
-  float (*vertexCos)[3];
+  float (*vert_coords)[3];
   MDeformVert *dvert;
   int defgrp_index;
   float fac;
@@ -896,18 +896,18 @@ static void lattice_deform_vert_task(void *__restrict userdata,
   if (data->dvert != NULL) {
     const float weight = defvert_find_weight(data->dvert + index, data->defgrp_index);
     if (weight > 0.0f) {
-      calc_latt_deform(data->lattice_deform_data, data->vertexCos[index], weight * data->fac);
+      calc_latt_deform(data->lattice_deform_data, data->vert_coords[index], weight * data->fac);
     }
   }
   else {
-    calc_latt_deform(data->lattice_deform_data, data->vertexCos[index], data->fac);
+    calc_latt_deform(data->lattice_deform_data, data->vert_coords[index], data->fac);
   }
 }
 
 void lattice_deform_verts(Object *laOb,
                           Object *target,
                           Mesh *mesh,
-                          float (*vertexCos)[3],
+                          float (*vert_coords)[3],
                           int numVerts,
                           const char *vgroup,
                           float fac)
@@ -942,11 +942,13 @@ void lattice_deform_verts(Object *laOb,
     }
   }
 
-  LatticeDeformUserdata data = {.lattice_deform_data = lattice_deform_data,
-                                .vertexCos = vertexCos,
-                                .dvert = dvert,
-                                .defgrp_index = defgrp_index,
-                                .fac = fac};
+  LatticeDeformUserdata data = {
+      .lattice_deform_data = lattice_deform_data,
+      .vert_coords = vert_coords,
+      .dvert = dvert,
+      .defgrp_index = defgrp_index,
+      .fac = fac,
+  };
 
   TaskParallelSettings settings;
   BLI_parallel_range_settings_defaults(&settings);
@@ -1055,33 +1057,37 @@ void outside_lattice(Lattice *lt)
   }
 }
 
-float (*BKE_lattice_vertexcos_get(struct Object *ob, int *r_numVerts))[3]
+void BKE_lattice_vert_coords_get(const Lattice *lt, float (*vert_coords)[3])
 {
-  Lattice *lt = ob->data;
-  int i, numVerts;
-  float(*vertexCos)[3];
-
-  if (lt->editlatt) {
-    lt = lt->editlatt->latt;
+  const int vert_len = lt->pntsu * lt->pntsv * lt->pntsw;
+  for (int i = 0; i < vert_len; i++) {
+    copy_v3_v3(vert_coords[i], lt->def[i].vec);
   }
-  numVerts = *r_numVerts = lt->pntsu * lt->pntsv * lt->pntsw;
+}
 
-  vertexCos = MEM_mallocN(sizeof(*vertexCos) * numVerts, "lt_vcos");
+float (*BKE_lattice_vert_coords_alloc(const Lattice *lt, int *r_vert_len))[3]
+{
+  const int vert_len = *r_vert_len = lt->pntsu * lt->pntsv * lt->pntsw;
+  float(*vert_coords)[3] = MEM_mallocN(sizeof(*vert_coords) * vert_len, __func__);
+  BKE_lattice_vert_coords_get(lt, vert_coords);
+  return vert_coords;
+}
 
+void BKE_lattice_vert_coords_apply_with_mat4(struct Lattice *lt,
+                                             const float (*vertexCos)[3],
+                                             const float mat[4][4])
+{
+  int i, numVerts = lt->pntsu * lt->pntsv * lt->pntsw;
   for (i = 0; i < numVerts; i++) {
-    copy_v3_v3(vertexCos[i], lt->def[i].vec);
+    mul_v3_m4v3(lt->def[i].vec, mat, vertexCos[i]);
   }
-
-  return vertexCos;
 }
 
-void BKE_lattice_vertexcos_apply(struct Object *ob, float (*vertexCos)[3])
+void BKE_lattice_vert_coords_apply(Lattice *lt, const float (*vert_coords)[3])
 {
-  Lattice *lt = ob->data;
-  int i, numVerts = lt->pntsu * lt->pntsv * lt->pntsw;
-
-  for (i = 0; i < numVerts; i++) {
-    copy_v3_v3(lt->def[i].vec, vertexCos[i]);
+  const int vert_len = lt->pntsu * lt->pntsv * lt->pntsw;
+  for (int i = 0; i < vert_len; i++) {
+    copy_v3_v3(lt->def[i].vec, vert_coords[i]);
   }
 }
 
@@ -1093,7 +1099,7 @@ void BKE_lattice_modifiers_calc(struct Depsgraph *depsgraph, Scene *scene, Objec
   Object *ob_orig = DEG_get_original_object(ob);
   VirtualModifierData virtualModifierData;
   ModifierData *md = modifiers_getVirtualModifierList(ob, &virtualModifierData);
-  float(*vertexCos)[3] = NULL;
+  float(*vert_coords)[3] = NULL;
   int numVerts, editmode = (lt->editlatt != NULL);
   const ModifierEvalContext mectx = {depsgraph, ob, 0};
 
@@ -1123,29 +1129,37 @@ void BKE_lattice_modifiers_calc(struct Depsgraph *depsgraph, Scene *scene, Objec
       continue;
     }
 
-    if (!vertexCos) {
-      vertexCos = BKE_lattice_vertexcos_get(ob_orig, &numVerts);
+    if (!vert_coords) {
+      Lattice *lt_orig = ob_orig->data;
+      if (lt_orig->editlatt) {
+        lt_orig = lt_orig->editlatt->latt;
+      }
+      vert_coords = BKE_lattice_vert_coords_alloc(lt_orig, &numVerts);
     }
-    mti->deformVerts(md, &mectx, NULL, vertexCos, numVerts);
+    mti->deformVerts(md, &mectx, NULL, vert_coords, numVerts);
   }
 
   if (ob->id.tag & LIB_TAG_COPIED_ON_WRITE) {
-    if (vertexCos) {
-      BKE_lattice_vertexcos_apply(ob, vertexCos);
-      MEM_freeN(vertexCos);
+    if (vert_coords) {
+      BKE_lattice_vert_coords_apply(ob->data, vert_coords);
+      MEM_freeN(vert_coords);
     }
   }
   else {
     /* Displist won't do anything; this is just for posterity's sake until we remove it. */
-    if (!vertexCos) {
-      vertexCos = BKE_lattice_vertexcos_get(ob_orig, &numVerts);
+    if (!vert_coords) {
+      Lattice *lt_orig = ob_orig->data;
+      if (lt_orig->editlatt) {
+        lt_orig = lt_orig->editlatt->latt;
+      }
+      vert_coords = BKE_lattice_vert_coords_alloc(lt_orig, &numVerts);
     }
 
     DispList *dl = MEM_callocN(sizeof(*dl), "lt_dl");
     dl->type = DL_VERTS;
     dl->parts = 1;
     dl->nr = numVerts;
-    dl->verts = (float *)vertexCos;
+    dl->verts = (float *)vert_coords;
 
     BLI_addtail(&ob->runtime.curve_cache->disp, dl);
   }
diff --git a/source/blender/blenkernel/intern/layer.c b/source/blender/blenkernel/intern/layer.c
index 40608285785..de105b9b62a 100644
--- a/source/blender/blenkernel/intern/layer.c
+++ b/source/blender/blenkernel/intern/layer.c
@@ -494,6 +494,36 @@ static LayerCollection *collection_from_index(ListBase *lb, const int number, in
 }
 
 /**
+ * Determine if a collection is hidden, viewport visibility restricted, or excluded
+ */
+static bool layer_collection_hidden(ViewLayer *view_layer, LayerCollection *lc)
+{
+  if (lc->flag & LAYER_COLLECTION_EXCLUDE) {
+    return true;
+  }
+
+  /* Check visiblilty restriction flags */
+  if (lc->flag & LAYER_COLLECTION_HIDE || lc->collection->flag & COLLECTION_RESTRICT_VIEWPORT) {
+    return true;
+  }
+  else {
+    /* Restriction flags stay set, so we need to check parents */
+    CollectionParent *parent = lc->collection->parents.first;
+
+    if (parent) {
+      lc = BKE_layer_collection_first_from_scene_collection(view_layer, parent->collection);
+
+      return lc && layer_collection_hidden(view_layer, lc);
+    }
+    else {
+      return false;
+    }
+  }
+
+  return false;
+}
+
+/**
  * Get the collection for a given index
  */
 LayerCollection *BKE_layer_collection_from_index(ViewLayer *view_layer, const int index)
@@ -537,8 +567,9 @@ LayerCollection *BKE_layer_collection_activate_parent(ViewLayer *view_layer, Lay
     lc = NULL;
   }
 
-  if (lc && (lc->flag & LAYER_COLLECTION_EXCLUDE)) {
-    /* Don't activate excluded collections. */
+  /* Don't activate excluded or hidden collections to prevent creating objects in a hidden
+   * collection from the UI */
+  if (lc && layer_collection_hidden(view_layer, lc)) {
     return BKE_layer_collection_activate_parent(view_layer, lc);
   }
 
@@ -817,8 +848,7 @@ void BKE_layer_collection_sync(const Scene *scene, ViewLayer *view_layer)
 
   /* Always set a valid active collection. */
   LayerCollection *active = view_layer->active_collection;
-
-  if (active && (active->flag & LAYER_COLLECTION_EXCLUDE)) {
+  if (active && layer_collection_hidden(view_layer, active)) {
     BKE_layer_collection_activate_parent(view_layer, active);
   }
   else if (active == NULL) {
diff --git a/source/blender/blenkernel/intern/library.c b/source/blender/blenkernel/intern/library.c
index de6f5142a19..f45eba2b53f 100644
--- a/source/blender/blenkernel/intern/library.c
+++ b/source/blender/blenkernel/intern/library.c
@@ -851,8 +851,8 @@ void BKE_id_swap(Main *bmain, ID *id_a, ID *id_b)
   id_b->properties = id_a_back.properties;
 
   /* Swap will have broken internal references to itself, restore them. */
-  BKE_libblock_relink_ex(bmain, id_a, id_b, id_a, false);
-  BKE_libblock_relink_ex(bmain, id_b, id_a, id_b, false);
+  BKE_libblock_relink_ex(bmain, id_a, id_b, id_a, ID_REMAP_SKIP_NEVER_NULL_USAGE);
+  BKE_libblock_relink_ex(bmain, id_b, id_a, id_b, ID_REMAP_SKIP_NEVER_NULL_USAGE);
 }
 
 /** Does *not* set ID->newid pointer. */
@@ -1408,21 +1408,24 @@ void *BKE_id_new_nomain(const short type, const char *name)
   return id;
 }
 
-void BKE_libblock_copy_ex(Main *bmain, const ID *id, ID **r_newid, const int flag)
+void BKE_libblock_copy_ex(Main *bmain, const ID *id, ID **r_newid, int flag)
 {
   ID *new_id = *r_newid;
 
-  /* Grrrrrrrrr... Not adding 'root' nodetrees to bmain.... grrrrrrrrrrrrrrrrrrrr! */
-  /* This is taken from original ntree copy code, might be weak actually? */
-  const bool use_nodetree_alloc_exception = ((GS(id->name) == ID_NT) && (bmain != NULL) &&
-                                             (BLI_findindex(&bmain->nodetrees, id) < 0));
-
   BLI_assert((flag & LIB_ID_CREATE_NO_MAIN) != 0 || bmain != NULL);
   BLI_assert((flag & LIB_ID_CREATE_NO_MAIN) != 0 || (flag & LIB_ID_CREATE_NO_ALLOCATE) == 0);
   BLI_assert((flag & LIB_ID_CREATE_NO_MAIN) == 0 || (flag & LIB_ID_CREATE_NO_USER_REFCOUNT) != 0);
   /* Never implicitly copy shapekeys when generating temp data outside of Main database. */
   BLI_assert((flag & LIB_ID_CREATE_NO_MAIN) == 0 || (flag & LIB_ID_COPY_SHAPEKEY) == 0);
 
+  /* 'Private ID' data handling. */
+  if ((bmain != NULL) && (id->flag & LIB_PRIVATE_DATA) != 0) {
+    flag |= LIB_ID_CREATE_NO_MAIN;
+  }
+
+  /* The id->flag bits to copy over. */
+  const int copy_flag_mask = LIB_PRIVATE_DATA;
+
   if ((flag & LIB_ID_CREATE_NO_ALLOCATE) != 0) {
     /* r_newid already contains pointer to allocated memory. */
     /* TODO do we want to memset(0) whole mem before filling it? */
@@ -1432,10 +1435,7 @@ void BKE_libblock_copy_ex(Main *bmain, const ID *id, ID **r_newid, const int fla
     /* TODO Do we want/need to copy more from ID struct itself? */
   }
   else {
-    new_id = BKE_libblock_alloc(bmain,
-                                GS(id->name),
-                                id->name + 2,
-                                flag | (use_nodetree_alloc_exception ? LIB_ID_CREATE_NO_MAIN : 0));
+    new_id = BKE_libblock_alloc(bmain, GS(id->name), id->name + 2, flag);
   }
   BLI_assert(new_id != NULL);
 
@@ -1448,6 +1448,8 @@ void BKE_libblock_copy_ex(Main *bmain, const ID *id, ID **r_newid, const int fla
     memcpy(cpn + id_offset, cp + id_offset, id_len - id_offset);
   }
 
+  new_id->flag = (new_id->flag & ~copy_flag_mask) | (id->flag & copy_flag_mask);
+
   if (id->properties) {
     new_id->properties = IDP_CopyProperty_ex(id->properties, flag);
   }
@@ -1727,8 +1729,8 @@ bool BKE_id_new_name_validate(ListBase *lb, ID *id, const char *tname)
 
   /* This was in 2.43 and previous releases
    * however all data in blender should be sorted, not just duplicate names
-   * sorting should not hurt, but noting just incase it alters the way other
-   * functions work, so sort every time */
+   * sorting should not hurt, but noting just in case it alters the way other
+   * functions work, so sort every time. */
 #if 0
   if (result) {
     id_sort_by_name(lb, id);
diff --git a/source/blender/blenkernel/intern/library_override.c b/source/blender/blenkernel/intern/library_override.c
index e435f07e38d..6532fce5778 100644
--- a/source/blender/blenkernel/intern/library_override.c
+++ b/source/blender/blenkernel/intern/library_override.c
@@ -111,11 +111,11 @@ void BKE_override_library_copy(ID *dst_id, const ID *src_id)
 
   if (dst_id->override_library != NULL) {
     if (src_id->override_library == NULL) {
-      BKE_override_library_free(&dst_id->override_library);
+      BKE_override_library_free(&dst_id->override_library, true);
       return;
     }
     else {
-      BKE_override_library_clear(dst_id->override_library);
+      BKE_override_library_clear(dst_id->override_library, true);
     }
   }
   else if (src_id->override_library == NULL) {
@@ -144,7 +144,7 @@ void BKE_override_library_copy(ID *dst_id, const ID *src_id)
 }
 
 /** Clear any overriding data from given \a override. */
-void BKE_override_library_clear(IDOverrideLibrary *override)
+void BKE_override_library_clear(IDOverrideLibrary *override, const bool do_id_user)
 {
   BLI_assert(override != NULL);
 
@@ -153,16 +153,18 @@ void BKE_override_library_clear(IDOverrideLibrary *override)
   }
   BLI_freelistN(&override->properties);
 
-  id_us_min(override->reference);
-  /* override->storage should never be refcounted... */
+  if (do_id_user) {
+    id_us_min(override->reference);
+    /* override->storage should never be refcounted... */
+  }
 }
 
 /** Free given \a override. */
-void BKE_override_library_free(struct IDOverrideLibrary **override)
+void BKE_override_library_free(struct IDOverrideLibrary **override, const bool do_id_user)
 {
   BLI_assert(*override != NULL);
 
-  BKE_override_library_clear(*override);
+  BKE_override_library_clear(*override, do_id_user);
   MEM_freeN(*override);
   *override = NULL;
 }
@@ -205,6 +207,11 @@ ID *BKE_override_library_create_from_id(Main *bmain, ID *reference_id)
  * \note Set id->newid of overridden libs with newly created overrides,
  * caller is responsible to clean those pointers before/after usage as needed.
  *
+ * \note By default, it will only remap newly created local overriding data-blocks between
+ * themselves, to avoid 'enforcing' those overrides into all other usages of the linked data in
+ * main. You can add more local IDs to be remapped to use new overriding ones by setting their
+ * LIB_TAG_DOIT tag.
+ *
  * \return \a true on success, \a false otherwise.
  */
 bool BKE_override_library_create_from_tag(Main *bmain)
@@ -212,26 +219,59 @@ bool BKE_override_library_create_from_tag(Main *bmain)
   ID *reference_id;
   bool ret = true;
 
+  ListBase todo_ids = {NULL};
+  LinkData *todo_id_iter;
+
+  /* Get all IDs we want to override. */
   FOREACH_MAIN_ID_BEGIN (bmain, reference_id) {
     if ((reference_id->tag & LIB_TAG_DOIT) != 0 && reference_id->lib != NULL) {
-      if ((reference_id->newid = override_library_create_from(bmain, reference_id)) == NULL) {
-        ret = false;
-      }
+      todo_id_iter = MEM_callocN(sizeof(*todo_id_iter), __func__);
+      todo_id_iter->data = reference_id;
+      BLI_addtail(&todo_ids, todo_id_iter);
     }
   }
   FOREACH_MAIN_ID_END;
 
-  FOREACH_MAIN_ID_BEGIN (bmain, reference_id) {
-    if ((reference_id->tag & LIB_TAG_DOIT) != 0 && reference_id->lib != NULL &&
-        reference_id->newid != NULL) {
-      ID *local_id = reference_id->newid;
-      BKE_libblock_remap(bmain,
-                         reference_id,
-                         local_id,
-                         ID_REMAP_SKIP_INDIRECT_USAGE | ID_REMAP_SKIP_OVERRIDE_LIBRARY);
+  /* Override the IDs. */
+  for (todo_id_iter = todo_ids.first; todo_id_iter != NULL; todo_id_iter = todo_id_iter->next) {
+    reference_id = todo_id_iter->data;
+    if ((reference_id->newid = override_library_create_from(bmain, reference_id)) == NULL) {
+      ret = false;
+    }
+    else {
+      /* We also tag the new IDs so that in next step we can remap their pointers too. */
+      reference_id->newid->tag |= LIB_TAG_DOIT;
     }
   }
-  FOREACH_MAIN_ID_END;
+
+  /* Only remap new local ID's pointers, we don't want to force our new overrides onto our whole
+   * existing linked IDs usages. */
+  for (todo_id_iter = todo_ids.first; todo_id_iter != NULL; todo_id_iter = todo_id_iter->next) {
+    ID *other_id;
+    reference_id = todo_id_iter->data;
+
+    if (reference_id->newid == NULL) {
+      continue;
+    }
+
+    /* Still checking the whole Main, that way we can tag other local IDs as needing to be remapped
+     * to use newly created overriding IDs, if needed. */
+    FOREACH_MAIN_ID_BEGIN (bmain, other_id) {
+      if ((other_id->tag & LIB_TAG_DOIT) != 0 && other_id->lib == NULL) {
+        ID *local_id = reference_id->newid;
+        /* Note that using ID_REMAP_SKIP_INDIRECT_USAGE below is superfluous, as we only remap
+         * local IDs usages anyway... */
+        BKE_libblock_relink_ex(bmain,
+                               other_id,
+                               reference_id,
+                               local_id,
+                               ID_REMAP_SKIP_INDIRECT_USAGE | ID_REMAP_SKIP_OVERRIDE_LIBRARY);
+      }
+    }
+    FOREACH_MAIN_ID_END;
+  }
+
+  BLI_freelistN(&todo_ids);
 
   return ret;
 }
@@ -628,7 +668,7 @@ void BKE_main_override_library_operations_create(Main *bmain, const bool force_a
   ID *id;
 
   FOREACH_MAIN_ID_BEGIN (bmain, id) {
-    if (force_auto ||
+    if ((ID_IS_OVERRIDE_LIBRARY(id) && force_auto) ||
         (ID_IS_OVERRIDE_LIBRARY_AUTO(id) && (id->tag & LIB_TAG_OVERRIDE_LIBRARY_AUTOREFRESH))) {
       BKE_override_library_operations_create(bmain, id, force_auto);
       id->tag &= ~LIB_TAG_OVERRIDE_LIBRARY_AUTOREFRESH;
diff --git a/source/blender/blenkernel/intern/library_remap.c b/source/blender/blenkernel/intern/library_remap.c
index 8fe2552c03f..61f506a8a6c 100644
--- a/source/blender/blenkernel/intern/library_remap.c
+++ b/source/blender/blenkernel/intern/library_remap.c
@@ -652,12 +652,11 @@ void BKE_libblock_unlink(Main *bmain,
  *     ... sigh
  */
 void BKE_libblock_relink_ex(
-    Main *bmain, void *idv, void *old_idv, void *new_idv, const bool us_min_never_null)
+    Main *bmain, void *idv, void *old_idv, void *new_idv, const short remap_flags)
 {
   ID *id = idv;
   ID *old_id = old_idv;
   ID *new_id = new_idv;
-  int remap_flags = us_min_never_null ? 0 : ID_REMAP_SKIP_NEVER_NULL_USAGE;
 
   /* No need to lock here, we are only affecting given ID, not bmain database. */
 
@@ -757,7 +756,7 @@ void BKE_libblock_free_data(ID *id, const bool do_id_user)
   }
 
   if (id->override_library) {
-    BKE_override_library_free(&id->override_library);
+    BKE_override_library_free(&id->override_library, do_id_user);
   }
 
   /* XXX TODO remove animdata handling from each type's freeing func,
@@ -945,7 +944,7 @@ void BKE_id_free_ex(Main *bmain, void *idv, int flag, const bool use_flag_from_i
 #endif
 
   if ((flag & LIB_ID_FREE_NO_USER_REFCOUNT) == 0) {
-    BKE_libblock_relink_ex(bmain, id, NULL, NULL, true);
+    BKE_libblock_relink_ex(bmain, id, NULL, NULL, 0);
   }
 
   BKE_libblock_free_datablock(id, flag);
@@ -1091,7 +1090,7 @@ static void id_delete(Main *bmain, const bool do_tagged_deletion)
             bmain, id, NULL, ID_REMAP_FLAG_NEVER_NULL_USAGE | ID_REMAP_FORCE_NEVER_NULL_USAGE);
         /* Since we removed ID from Main,
          * we also need to unlink its own other IDs usages ourself. */
-        BKE_libblock_relink_ex(bmain, id, NULL, NULL, true);
+        BKE_libblock_relink_ex(bmain, id, NULL, NULL, 0);
         /* Now we can safely mark that ID as not being in Main database anymore. */
         id->tag |= LIB_TAG_NO_MAIN;
         /* This is needed because we may not have remapped usages
diff --git a/source/blender/blenkernel/intern/light.c b/source/blender/blenkernel/intern/light.c
index 75c9e0e42a5..07ec8d70af1 100644
--- a/source/blender/blenkernel/intern/light.c
+++ b/source/blender/blenkernel/intern/light.c
@@ -108,12 +108,15 @@ Light *BKE_light_add(Main *bmain, const char *name)
  */
 void BKE_light_copy_data(Main *bmain, Light *la_dst, const Light *la_src, const int flag)
 {
+  /* We never handle usercount here for own data. */
+  const int flag_subdata = flag | LIB_ID_CREATE_NO_USER_REFCOUNT;
+  /* We always need allocation of our private ID data. */
+  const int flag_private_id_data = flag_subdata & ~LIB_ID_CREATE_NO_ALLOCATE;
+
   la_dst->curfalloff = BKE_curvemapping_copy(la_src->curfalloff);
 
   if (la_src->nodetree) {
-    /* Note: nodetree is *not* in bmain, however this specific case is handled at lower level
-     *       (see BKE_libblock_copy_ex()). */
-    BKE_id_copy_ex(bmain, (ID *)la_src->nodetree, (ID **)&la_dst->nodetree, flag);
+    BKE_id_copy_ex(bmain, (ID *)la_src->nodetree, (ID **)&la_dst->nodetree, flag_private_id_data);
   }
 
   if ((flag & LIB_ID_COPY_NO_PREVIEW) == 0) {
diff --git a/source/blender/blenkernel/intern/linestyle.c b/source/blender/blenkernel/intern/linestyle.c
index 7bfe5a7c8ff..e5fa5ff08b0 100644
--- a/source/blender/blenkernel/intern/linestyle.c
+++ b/source/blender/blenkernel/intern/linestyle.c
@@ -167,6 +167,8 @@ void BKE_linestyle_copy_data(struct Main *bmain,
 {
   /* We never handle usercount here for own data. */
   const int flag_subdata = flag | LIB_ID_CREATE_NO_USER_REFCOUNT;
+  /* We always need allocation of our private ID data. */
+  const int flag_private_id_data = flag_subdata & ~LIB_ID_CREATE_NO_ALLOCATE;
 
   for (int a = 0; a < MAX_MTEX; a++) {
     if (linestyle_src->mtex[a]) {
@@ -176,9 +178,10 @@ void BKE_linestyle_copy_data(struct Main *bmain,
   }
 
   if (linestyle_src->nodetree) {
-    /* Note: nodetree is *not* in bmain, however this specific case is handled at lower level
-     *       (see BKE_libblock_copy_ex()). */
-    BKE_id_copy_ex(bmain, (ID *)linestyle_src->nodetree, (ID **)&linestyle_dst->nodetree, flag);
+    BKE_id_copy_ex(bmain,
+                   (ID *)linestyle_src->nodetree,
+                   (ID **)&linestyle_dst->nodetree,
+                   flag_private_id_data);
   }
 
   LineStyleModifier *m;
diff --git a/source/blender/blenkernel/intern/mask.c b/source/blender/blenkernel/intern/mask.c
index b6d61563fe8..fc087ff91b2 100644
--- a/source/blender/blenkernel/intern/mask.c
+++ b/source/blender/blenkernel/intern/mask.c
@@ -1628,7 +1628,9 @@ MaskLayerShape *BKE_mask_layer_shape_find_frame(MaskLayer *masklay, const int fr
   return NULL;
 }
 
-/* when returning 2 - the frame isnt found but before/after frames are */
+/**
+ * When returning 2 - the frame isn't found but before/after frames are.
+ */
 int BKE_mask_layer_shape_find_frame_range(MaskLayer *masklay,
                                           const float frame,
                                           MaskLayerShape **r_masklay_shape_a,
diff --git a/source/blender/blenkernel/intern/material.c b/source/blender/blenkernel/intern/material.c
index b01c1189fd1..c12541b3b22 100644
--- a/source/blender/blenkernel/intern/material.c
+++ b/source/blender/blenkernel/intern/material.c
@@ -71,6 +71,7 @@
 
 /* used in UI and render */
 Material defmaterial;
+Material defgpencil_material;
 
 static CLG_LogRef LOG = {"bke.material"};
 
@@ -78,6 +79,13 @@ static CLG_LogRef LOG = {"bke.material"};
 void init_def_material(void)
 {
   BKE_material_init(&defmaterial);
+  BKE_material_gpencil_init(&defgpencil_material);
+}
+
+/* Free the GPencil data of the default material, creator.c */
+void BKE_material_gpencil_default_free(void)
+{
+  MEM_SAFE_FREE(defgpencil_material.gp_style);
 }
 
 /** Free (or release) any data used by this material (does not free the material itself). */
@@ -144,6 +152,16 @@ void BKE_material_init(Material *ma)
   ma->blend_shadow = MA_BS_SOLID;
 }
 
+void BKE_material_gpencil_init(Material *ma)
+{
+  BKE_material_init(ma);
+
+  /* grease pencil settings */
+  strcpy(ma->id.name, "MADefault GPencil");
+  BKE_material_init_gpencil_settings(ma);
+  add_v3_fl(&ma->gp_style->stroke_rgba[0], 0.6f);
+}
+
 Material *BKE_material_add(Main *bmain, const char *name)
 {
   Material *ma;
@@ -180,10 +198,13 @@ Material *BKE_material_add_gpencil(Main *bmain, const char *name)
  */
 void BKE_material_copy_data(Main *bmain, Material *ma_dst, const Material *ma_src, const int flag)
 {
+  /* We never handle usercount here for own data. */
+  const int flag_subdata = flag | LIB_ID_CREATE_NO_USER_REFCOUNT;
+  /* We always need allocation of our private ID data. */
+  const int flag_private_id_data = flag_subdata & ~LIB_ID_CREATE_NO_ALLOCATE;
+
   if (ma_src->nodetree) {
-    /* Note: nodetree is *not* in bmain, however this specific case is handled at lower level
-     *       (see BKE_libblock_copy_ex()). */
-    BKE_id_copy_ex(bmain, (ID *)ma_src->nodetree, (ID **)&ma_dst->nodetree, flag);
+    BKE_id_copy_ex(bmain, (ID *)ma_src->nodetree, (ID **)&ma_dst->nodetree, flag_private_id_data);
   }
 
   if ((flag & LIB_ID_COPY_NO_PREVIEW) == 0) {
@@ -461,7 +482,7 @@ void BKE_material_append_id(Main *bmain, ID *id, Material *ma)
   }
 }
 
-Material *BKE_material_pop_id(Main *bmain, ID *id, int index_i, bool update_data)
+Material *BKE_material_pop_id(Main *bmain, ID *id, int index_i)
 {
   short index = (short)index_i;
   Material *ret = NULL;
@@ -489,10 +510,7 @@ Material *BKE_material_pop_id(Main *bmain, ID *id, int index_i, bool update_data
         test_all_objects_materials(bmain, id);
       }
 
-      if (update_data) {
-        /* decrease mat_nr index */
-        material_data_index_remove_id(id, index);
-      }
+      material_data_index_remove_id(id, index);
 
       DEG_id_tag_update(id, ID_RECALC_COPY_ON_WRITE);
       DEG_relations_tag_update(bmain);
@@ -502,7 +520,7 @@ Material *BKE_material_pop_id(Main *bmain, ID *id, int index_i, bool update_data
   return ret;
 }
 
-void BKE_material_clear_id(Main *bmain, ID *id, bool update_data)
+void BKE_material_clear_id(Main *bmain, ID *id)
 {
   Material ***matar;
   if ((matar = give_matarar_id(id))) {
@@ -516,12 +534,9 @@ void BKE_material_clear_id(Main *bmain, ID *id, bool update_data)
       MEM_freeN(*matar);
       *matar = NULL;
     }
-    test_all_objects_materials(bmain, id);
 
-    if (update_data) {
-      /* decrease mat_nr index */
-      material_data_index_clear_id(id);
-    }
+    test_all_objects_materials(bmain, id);
+    material_data_index_clear_id(id);
 
     DEG_id_tag_update(id, ID_RECALC_COPY_ON_WRITE);
     DEG_relations_tag_update(bmain);
@@ -586,6 +601,17 @@ Material *give_current_material(Object *ob, short act)
   return ma_p ? *ma_p : NULL;
 }
 
+Material *BKE_material_gpencil_get(Object *ob, short act)
+{
+  Material *ma = give_current_material(ob, act);
+  if (ma != NULL) {
+    return ma;
+  }
+  else {
+    return &defgpencil_material;
+  }
+}
+
 MaterialGPencilStyle *BKE_material_gpencil_settings_get(Object *ob, short act)
 {
   Material *ma = give_current_material(ob, act);
@@ -597,7 +623,7 @@ MaterialGPencilStyle *BKE_material_gpencil_settings_get(Object *ob, short act)
     return ma->gp_style;
   }
   else {
-    return NULL;
+    return defgpencil_material.gp_style;
   }
 }
 
@@ -1076,10 +1102,6 @@ bool BKE_object_material_slot_remove(Main *bmain, Object *ob)
   }
   /* check indices from gpencil */
   else if (ob->type == OB_GPENCIL) {
-    /* need one color */
-    if (ob->totcol == 0) {
-      BKE_gpencil_object_material_ensure_from_active_input_material(bmain, ob);
-    }
     BKE_gpencil_material_index_reassign((bGPdata *)ob->data, ob->totcol, actcol - 1);
   }
 
diff --git a/source/blender/blenkernel/intern/mball.c b/source/blender/blenkernel/intern/mball.c
index d6fa071009e..19009322975 100644
--- a/source/blender/blenkernel/intern/mball.c
+++ b/source/blender/blenkernel/intern/mball.c
@@ -364,12 +364,13 @@ bool BKE_mball_is_any_unselected(const MetaBall *mb)
   return false;
 }
 
-/* \brief copy some properties from object to other metaball object with same base name
+/**
+ * \brief copy some properties from object to other metaball object with same base name
  *
  * When some properties (wiresize, threshold, update flags) of metaball are changed, then this
  * properties are copied to all metaballs in same "group" (metaballs with same base name: MBall,
  * MBall.001, MBall.002, etc). The most important is to copy properties to the base metaball,
- * because this metaball influence polygonisation of metaballs. */
+ * because this metaball influence polygonization of metaballs. */
 void BKE_mball_properties_copy(Scene *scene, Object *active_object)
 {
   Scene *sce_iter = scene;
@@ -525,7 +526,7 @@ bool BKE_mball_center_bounds(const MetaBall *mb, float r_cent[3])
   return false;
 }
 
-void BKE_mball_transform(MetaBall *mb, float mat[4][4], const bool do_props)
+void BKE_mball_transform(MetaBall *mb, const float mat[4][4], const bool do_props)
 {
   float quat[4];
   const float scale = mat4_to_scale(mat);
diff --git a/source/blender/blenkernel/intern/mesh.c b/source/blender/blenkernel/intern/mesh.c
index 4f39d34574e..61b841591a4 100644
--- a/source/blender/blenkernel/intern/mesh.c
+++ b/source/blender/blenkernel/intern/mesh.c
@@ -661,13 +661,13 @@ Mesh *BKE_mesh_new_nomain(
   return mesh;
 }
 
-static Mesh *mesh_new_nomain_from_template_ex(const Mesh *me_src,
-                                              int verts_len,
-                                              int edges_len,
-                                              int tessface_len,
-                                              int loops_len,
-                                              int polys_len,
-                                              CustomData_MeshMasks mask)
+Mesh *BKE_mesh_new_nomain_from_template_ex(const Mesh *me_src,
+                                           int verts_len,
+                                           int edges_len,
+                                           int tessface_len,
+                                           int loops_len,
+                                           int polys_len,
+                                           CustomData_MeshMasks mask)
 {
   /* Only do tessface if we are creating tessfaces or copying from mesh with only tessfaces. */
   const bool do_tessface = (tessface_len || ((me_src->totface != 0) && (me_src->totpoly == 0)));
@@ -713,7 +713,7 @@ Mesh *BKE_mesh_new_nomain_from_template(const Mesh *me_src,
                                         int loops_len,
                                         int polys_len)
 {
-  return mesh_new_nomain_from_template_ex(
+  return BKE_mesh_new_nomain_from_template_ex(
       me_src, verts_len, edges_len, tessface_len, loops_len, polys_len, CD_MASK_EVERYTHING);
 }
 
@@ -801,7 +801,7 @@ Mesh *BKE_mesh_from_editmesh_with_coords_thin_wrap(BMEditMesh *em,
   me->runtime.is_original = true;
   if (vertexCos) {
     /* We will own this array in the future. */
-    BKE_mesh_apply_vert_coords(me, vertexCos);
+    BKE_mesh_vert_coords_apply(me, vertexCos);
     MEM_freeN(vertexCos);
     me->runtime.is_original = false;
   }
@@ -813,92 +813,6 @@ void BKE_mesh_make_local(Main *bmain, Mesh *me, const bool lib_local)
   BKE_id_make_local_generic(bmain, &me->id, true, lib_local);
 }
 
-bool BKE_mesh_uv_cdlayer_rename_index(Mesh *me,
-                                      const int loop_index,
-                                      const int face_index,
-                                      const char *new_name,
-                                      const bool do_tessface)
-{
-  CustomData *ldata, *fdata;
-  CustomDataLayer *cdlu, *cdlf;
-
-  if (me->edit_mesh) {
-    ldata = &me->edit_mesh->bm->ldata;
-    fdata = NULL; /* No tessellated data in BMesh! */
-  }
-  else {
-    ldata = &me->ldata;
-    fdata = &me->fdata;
-  }
-
-  cdlu = &ldata->layers[loop_index];
-  cdlf = (face_index != -1) && fdata && do_tessface ? &fdata->layers[face_index] : NULL;
-
-  if (cdlu->name != new_name) {
-    /* Mesh validate passes a name from the CD layer as the new name,
-     * Avoid memcpy from self to self in this case.
-     */
-    BLI_strncpy(cdlu->name, new_name, sizeof(cdlu->name));
-    CustomData_set_layer_unique_name(ldata, loop_index);
-  }
-
-  if (cdlf == NULL) {
-    return false;
-  }
-
-  BLI_strncpy(cdlf->name, cdlu->name, sizeof(cdlf->name));
-  CustomData_set_layer_unique_name(fdata, face_index);
-
-  return true;
-}
-
-bool BKE_mesh_uv_cdlayer_rename(Mesh *me,
-                                const char *old_name,
-                                const char *new_name,
-                                bool do_tessface)
-{
-  CustomData *ldata, *fdata;
-  if (me->edit_mesh) {
-    ldata = &me->edit_mesh->bm->ldata;
-    /* No tessellated data in BMesh! */
-    fdata = NULL;
-    do_tessface = false;
-  }
-  else {
-    ldata = &me->ldata;
-    fdata = &me->fdata;
-    do_tessface = (do_tessface && fdata->totlayer);
-  }
-
-  {
-    const int lidx_start = CustomData_get_layer_index(ldata, CD_MLOOPUV);
-    const int fidx_start = do_tessface ? CustomData_get_layer_index(fdata, CD_MTFACE) : -1;
-    int lidx = CustomData_get_named_layer(ldata, CD_MLOOPUV, old_name);
-    int fidx = do_tessface ? CustomData_get_named_layer(fdata, CD_MTFACE, old_name) : -1;
-
-    /* None of those cases should happen, in theory!
-     * Note this assume we have the same number of mtexpoly, mloopuv and mtface layers!
-     */
-    if (lidx == -1) {
-      if (fidx == -1) {
-        /* No layer found with this name! */
-        return false;
-      }
-      else {
-        lidx = fidx;
-      }
-    }
-
-    /* Go back to absolute indices! */
-    lidx += lidx_start;
-    if (fidx != -1) {
-      fidx += fidx_start;
-    }
-
-    return BKE_mesh_uv_cdlayer_rename_index(me, lidx, fidx, new_name, do_tessface);
-  }
-}
-
 void BKE_mesh_boundbox_calc(Mesh *me, float r_loc[3], float r_size[3])
 {
   BoundBox *bb;
@@ -1273,54 +1187,21 @@ void BKE_mesh_material_remap(Mesh *me, const unsigned int *remap, unsigned int r
 #undef MAT_NR_REMAP
 }
 
-void BKE_mesh_smooth_flag_set(Object *meshOb, int enableSmooth)
+void BKE_mesh_smooth_flag_set(Mesh *me, const bool use_smooth)
 {
-  Mesh *me = meshOb->data;
-  int i;
-
-  for (i = 0; i < me->totpoly; i++) {
-    MPoly *mp = &me->mpoly[i];
-
-    if (enableSmooth) {
-      mp->flag |= ME_SMOOTH;
-    }
-    else {
-      mp->flag &= ~ME_SMOOTH;
+  if (use_smooth) {
+    for (int i = 0; i < me->totpoly; i++) {
+      me->mpoly[i].flag |= ME_SMOOTH;
     }
   }
-
-  for (i = 0; i < me->totface; i++) {
-    MFace *mf = &me->mface[i];
-
-    if (enableSmooth) {
-      mf->flag |= ME_SMOOTH;
-    }
-    else {
-      mf->flag &= ~ME_SMOOTH;
+  else {
+    for (int i = 0; i < me->totpoly; i++) {
+      me->mpoly[i].flag &= ~ME_SMOOTH;
     }
   }
 }
 
 /**
- * Return a newly MEM_malloc'd array of all the mesh vertex locations
- * \note \a r_verts_len may be NULL
- */
-float (*BKE_mesh_vertexCos_get(const Mesh *me, int *r_verts_len))[3]
-{
-  int i, verts_len = me->totvert;
-  float(*cos)[3] = MEM_malloc_arrayN(verts_len, sizeof(*cos), "vertexcos1");
-
-  if (r_verts_len) {
-    *r_verts_len = verts_len;
-  }
-  for (i = 0; i < verts_len; i++) {
-    copy_v3_v3(cos[i], me->mvert[i].co);
-  }
-
-  return cos;
-}
-
-/**
  * Find the index of the loop in 'poly' which references vertex,
  * returns -1 if not found
  */
@@ -1471,7 +1352,7 @@ void BKE_mesh_ensure_navmesh(Mesh *me)
 
 void BKE_mesh_tessface_calc(Mesh *mesh)
 {
-  mesh->totface = BKE_mesh_recalc_tessellation(
+  mesh->totface = BKE_mesh_tessface_calc_ex(
       &mesh->fdata,
       &mesh->ldata,
       &mesh->pdata,
@@ -1665,35 +1546,56 @@ void BKE_mesh_count_selected_items(const Mesh *mesh, int r_count[3])
   /* We could support faces in paint modes. */
 }
 
-void BKE_mesh_apply_vert_coords(Mesh *mesh, float (*vertCoords)[3])
+void BKE_mesh_vert_coords_get(const Mesh *mesh, float (*vert_coords)[3])
 {
-  MVert *vert;
-  int i;
-
-  /* this will just return the pointer if it wasn't a referenced layer */
-  vert = CustomData_duplicate_referenced_layer(&mesh->vdata, CD_MVERT, mesh->totvert);
-  mesh->mvert = vert;
+  const MVert *mv = mesh->mvert;
+  for (int i = 0; i < mesh->totvert; i++, mv++) {
+    copy_v3_v3(vert_coords[i], mv->co);
+  }
+}
 
-  for (i = 0; i < mesh->totvert; ++i, ++vert) {
-    copy_v3_v3(vert->co, vertCoords[i]);
+float (*BKE_mesh_vert_coords_alloc(const Mesh *mesh, int *r_vert_len))[3]
+{
+  float(*vert_coords)[3] = MEM_mallocN(sizeof(float[3]) * mesh->totvert, __func__);
+  BKE_mesh_vert_coords_get(mesh, vert_coords);
+  if (r_vert_len) {
+    *r_vert_len = mesh->totvert;
   }
+  return vert_coords;
+}
 
+void BKE_mesh_vert_coords_apply(Mesh *mesh, const float (*vert_coords)[3])
+{
+  /* This will just return the pointer if it wasn't a referenced layer. */
+  MVert *mv = CustomData_duplicate_referenced_layer(&mesh->vdata, CD_MVERT, mesh->totvert);
+  mesh->mvert = mv;
+  for (int i = 0; i < mesh->totvert; i++, mv++) {
+    copy_v3_v3(mv->co, vert_coords[i]);
+  }
   mesh->runtime.cd_dirty_vert |= CD_MASK_NORMAL;
 }
 
-void BKE_mesh_apply_vert_normals(Mesh *mesh, short (*vertNormals)[3])
+void BKE_mesh_vert_coords_apply_with_mat4(Mesh *mesh,
+                                          const float (*vert_coords)[3],
+                                          const float mat[4][4])
 {
-  MVert *vert;
-  int i;
-
-  /* this will just return the pointer if it wasn't a referenced layer */
-  vert = CustomData_duplicate_referenced_layer(&mesh->vdata, CD_MVERT, mesh->totvert);
-  mesh->mvert = vert;
-
-  for (i = 0; i < mesh->totvert; ++i, ++vert) {
-    copy_v3_v3_short(vert->no, vertNormals[i]);
+  /* This will just return the pointer if it wasn't a referenced layer. */
+  MVert *mv = CustomData_duplicate_referenced_layer(&mesh->vdata, CD_MVERT, mesh->totvert);
+  mesh->mvert = mv;
+  for (int i = 0; i < mesh->totvert; i++, mv++) {
+    mul_v3_m4v3(mv->co, mat, vert_coords[i]);
   }
+  mesh->runtime.cd_dirty_vert |= CD_MASK_NORMAL;
+}
 
+void BKE_mesh_vert_normals_apply(Mesh *mesh, const short (*vert_normals)[3])
+{
+  /* This will just return the pointer if it wasn't a referenced layer. */
+  MVert *mv = CustomData_duplicate_referenced_layer(&mesh->vdata, CD_MVERT, mesh->totvert);
+  mesh->mvert = mv;
+  for (int i = 0; i < mesh->totvert; i++, mv++) {
+    copy_v3_v3_short(mv->no, vert_normals[i]);
+  }
   mesh->runtime.cd_dirty_vert &= ~CD_MASK_NORMAL;
 }
 
diff --git a/source/blender/blenkernel/intern/mesh_convert.c b/source/blender/blenkernel/intern/mesh_convert.c
index 3a2ba078dce..e072a37acee 100644
--- a/source/blender/blenkernel/intern/mesh_convert.c
+++ b/source/blender/blenkernel/intern/mesh_convert.c
@@ -1341,11 +1341,11 @@ Mesh *BKE_mesh_create_derived_for_modifier(struct Depsgraph *depsgraph,
 
   if (mti->type == eModifierTypeType_OnlyDeform) {
     int numVerts;
-    float(*deformedVerts)[3] = BKE_mesh_vertexCos_get(me, &numVerts);
+    float(*deformedVerts)[3] = BKE_mesh_vert_coords_alloc(me, &numVerts);
 
     BKE_id_copy_ex(NULL, &me->id, (ID **)&result, LIB_ID_COPY_LOCALIZE);
     mti->deformVerts(md_eval, &mectx, result, deformedVerts, numVerts);
-    BKE_mesh_apply_vert_coords(result, deformedVerts);
+    BKE_mesh_vert_coords_apply(result, deformedVerts);
 
     if (build_shapekey_layers) {
       add_shapekey_layers(result, me);
@@ -1434,7 +1434,6 @@ static void shapekey_layers_to_keyblocks(Mesh *mesh_src, Mesh *mesh_dst, int act
   }
 }
 
-/* This is a Mesh-based copy of DM_to_mesh() */
 void BKE_mesh_nomain_to_mesh(Mesh *mesh_src,
                              Mesh *mesh_dst,
                              Object *ob,
@@ -1442,7 +1441,7 @@ void BKE_mesh_nomain_to_mesh(Mesh *mesh_src,
                              bool take_ownership)
 {
   /* mesh_src might depend on mesh_dst, so we need to do everything with a local copy */
-  /* TODO(Sybren): the above claim came from DM_to_mesh();
+  /* TODO(Sybren): the above claim came from 2.7x derived-mesh code (DM_to_mesh);
    * check whether it is still true with Mesh */
   Mesh tmp = *mesh_dst;
   int totvert, totedge /*, totface */ /* UNUSED */, totloop, totpoly;
@@ -1594,7 +1593,6 @@ void BKE_mesh_nomain_to_mesh(Mesh *mesh_src,
   }
 }
 
-/* This is a Mesh-based copy of DM_to_meshkey() */
 void BKE_mesh_nomain_to_meshkey(Mesh *mesh_src, Mesh *mesh_dst, KeyBlock *kb)
 {
   int a, totvert = mesh_src->totvert;
diff --git a/source/blender/blenkernel/intern/mesh_evaluate.c b/source/blender/blenkernel/intern/mesh_evaluate.c
index 2ea275cdfb0..38762109167 100644
--- a/source/blender/blenkernel/intern/mesh_evaluate.c
+++ b/source/blender/blenkernel/intern/mesh_evaluate.c
@@ -448,64 +448,6 @@ void BKE_mesh_calc_normals(Mesh *mesh)
   mesh->runtime.cd_dirty_vert &= ~CD_MASK_NORMAL;
 }
 
-void BKE_mesh_calc_normals_tessface(
-    MVert *mverts, int numVerts, const MFace *mfaces, int numFaces, float (*r_faceNors)[3])
-{
-  float(*tnorms)[3] = MEM_calloc_arrayN((size_t)numVerts, sizeof(*tnorms), "tnorms");
-  float(*fnors)[3] = (r_faceNors) ?
-                         r_faceNors :
-                         MEM_calloc_arrayN((size_t)numFaces, sizeof(*fnors), "meshnormals");
-  int i;
-
-  if (!tnorms || !fnors) {
-    goto cleanup;
-  }
-
-  for (i = 0; i < numFaces; i++) {
-    const MFace *mf = &mfaces[i];
-    float *f_no = fnors[i];
-    float *n4 = (mf->v4) ? tnorms[mf->v4] : NULL;
-    const float *c4 = (mf->v4) ? mverts[mf->v4].co : NULL;
-
-    if (mf->v4) {
-      normal_quad_v3(
-          f_no, mverts[mf->v1].co, mverts[mf->v2].co, mverts[mf->v3].co, mverts[mf->v4].co);
-    }
-    else {
-      normal_tri_v3(f_no, mverts[mf->v1].co, mverts[mf->v2].co, mverts[mf->v3].co);
-    }
-
-    accumulate_vertex_normals_v3(tnorms[mf->v1],
-                                 tnorms[mf->v2],
-                                 tnorms[mf->v3],
-                                 n4,
-                                 f_no,
-                                 mverts[mf->v1].co,
-                                 mverts[mf->v2].co,
-                                 mverts[mf->v3].co,
-                                 c4);
-  }
-
-  /* following Mesh convention; we use vertex coordinate itself for normal in this case */
-  for (i = 0; i < numVerts; i++) {
-    MVert *mv = &mverts[i];
-    float *no = tnorms[i];
-
-    if (UNLIKELY(normalize_v3(no) == 0.0f)) {
-      normalize_v3_v3(no, mv->co);
-    }
-
-    normal_float_to_short_v3(mv->no, no);
-  }
-
-cleanup:
-  MEM_freeN(tnorms);
-
-  if (fnors != r_faceNors) {
-    MEM_freeN(fnors);
-  }
-}
-
 void BKE_mesh_calc_normals_looptri(MVert *mverts,
                                    int numVerts,
                                    const MLoop *mloop,
@@ -1320,7 +1262,7 @@ static void split_loop_nor_fan_do(LoopSplitTaskDataCommon *common_data, LoopSpli
           }
           // print_v2("new clnors", clnors_avg);
         }
-        /* Extra bonus: since smallstack is local to this func,
+        /* Extra bonus: since small-stack is local to this function,
          * no more need to empty it at all cost! */
 
         BKE_lnor_space_custom_data_to_normal(lnor_space, *clnor_ref, lnor);
@@ -1336,7 +1278,7 @@ static void split_loop_nor_fan_do(LoopSplitTaskDataCommon *common_data, LoopSpli
         copy_v3_v3(nor, lnor);
       }
     }
-    /* Extra bonus: since smallstack is local to this func,
+    /* Extra bonus: since small-stack is local to this funcion,
      * no more need to empty it at all cost! */
   }
 }
@@ -3003,14 +2945,14 @@ void BKE_mesh_tangent_loops_to_tessdata(CustomData *fdata,
  *
  * \return number of tessellation faces.
  */
-int BKE_mesh_recalc_tessellation(CustomData *fdata,
-                                 CustomData *ldata,
-                                 CustomData *pdata,
-                                 MVert *mvert,
-                                 int totface,
-                                 int totloop,
-                                 int totpoly,
-                                 const bool do_face_nor_copy)
+int BKE_mesh_tessface_calc_ex(CustomData *fdata,
+                              CustomData *ldata,
+                              CustomData *pdata,
+                              MVert *mvert,
+                              int totface,
+                              int totloop,
+                              int totpoly,
+                              const bool do_face_nor_copy)
 {
   /* use this to avoid locking pthread for _every_ polygon
    * and calling the fill function */
diff --git a/source/blender/blenkernel/intern/mesh_mapping.c b/source/blender/blenkernel/intern/mesh_mapping.c
index 78c0fa184f4..d908a248a28 100644
--- a/source/blender/blenkernel/intern/mesh_mapping.c
+++ b/source/blender/blenkernel/intern/mesh_mapping.c
@@ -845,7 +845,7 @@ static bool poly_is_island_boundary_smooth_cb(const MPoly *mp,
     const MPoly *mp_other = (mp == &mpoly_array[edge_poly_map->indices[0]]) ?
                                 &mpoly_array[edge_poly_map->indices[1]] :
                                 &mpoly_array[edge_poly_map->indices[0]];
-    return (mp_other->flag & ME_SMOOTH) != 0;
+    return (mp_other->flag & ME_SMOOTH) == 0;
   }
   return true;
 }
diff --git a/source/blender/blenkernel/intern/mesh_merge.c b/source/blender/blenkernel/intern/mesh_merge.c
index 71cc20c78b7..ae808b85323 100644
--- a/source/blender/blenkernel/intern/mesh_merge.c
+++ b/source/blender/blenkernel/intern/mesh_merge.c
@@ -235,7 +235,7 @@ static bool poly_gset_compare_fn(const void *k1, const void *k2)
  * but the additional checks are costly and not necessary in the case of mirror,
  * because each vertex is only merged to its own mirror.
  *
- * \note #BKE_mesh_recalc_tessellation has to run on the returned DM
+ * \note #BKE_mesh_tessface_calc_ex has to run on the returned DM
  * if you want to access tessfaces.
  */
 Mesh *BKE_mesh_merge_verts(Mesh *mesh,
diff --git a/source/blender/blenkernel/intern/mesh_remap.c b/source/blender/blenkernel/intern/mesh_remap.c
index db158ca8fb2..834008ca1ac 100644
--- a/source/blender/blenkernel/intern/mesh_remap.c
+++ b/source/blender/blenkernel/intern/mesh_remap.c
@@ -291,7 +291,7 @@ void BKE_mesh_remap_find_best_match_from_mesh(const MVert *verts_dst,
   float best_match = FLT_MAX, match;
 
   const int numverts_src = me_src->totvert;
-  float(*vcos_src)[3] = BKE_mesh_vertexCos_get(me_src, NULL);
+  float(*vcos_src)[3] = BKE_mesh_vert_coords_alloc(me_src, NULL);
 
   mesh_calc_eigen_matrix(NULL, (const float(*)[3])vcos_src, numverts_src, mat_src);
   mesh_calc_eigen_matrix(verts_dst, NULL, numverts_dst, mat_dst);
@@ -548,7 +548,7 @@ void BKE_mesh_remap_calc_verts_from_mesh(const int mode,
     }
     else if (ELEM(mode, MREMAP_MODE_VERT_EDGE_NEAREST, MREMAP_MODE_VERT_EDGEINTERP_NEAREST)) {
       MEdge *edges_src = me_src->medge;
-      float(*vcos_src)[3] = BKE_mesh_vertexCos_get(me_src, NULL);
+      float(*vcos_src)[3] = BKE_mesh_vert_coords_alloc(me_src, NULL);
 
       BKE_bvhtree_from_mesh_get(&treedata, me_src, BVHTREE_FROM_EDGES, 2);
       nearest.index = -1;
@@ -602,7 +602,7 @@ void BKE_mesh_remap_calc_verts_from_mesh(const int mode,
                   MREMAP_MODE_VERT_POLYINTERP_VNORPROJ)) {
       MPoly *polys_src = me_src->mpoly;
       MLoop *loops_src = me_src->mloop;
-      float(*vcos_src)[3] = BKE_mesh_vertexCos_get(me_src, NULL);
+      float(*vcos_src)[3] = BKE_mesh_vert_coords_alloc(me_src, NULL);
 
       size_t tmp_buff_size = MREMAP_DEFAULT_BUFSIZE;
       float(*vcos)[3] = MEM_mallocN(sizeof(*vcos) * tmp_buff_size, __func__);
@@ -752,7 +752,7 @@ void BKE_mesh_remap_calc_edges_from_mesh(const int mode,
       const int num_verts_src = me_src->totvert;
       const int num_edges_src = me_src->totedge;
       MEdge *edges_src = me_src->medge;
-      float(*vcos_src)[3] = BKE_mesh_vertexCos_get(me_src, NULL);
+      float(*vcos_src)[3] = BKE_mesh_vert_coords_alloc(me_src, NULL);
 
       MeshElemMap *vert_to_edge_src_map;
       int *vert_to_edge_src_map_mem;
@@ -901,7 +901,7 @@ void BKE_mesh_remap_calc_edges_from_mesh(const int mode,
       MEdge *edges_src = me_src->medge;
       MPoly *polys_src = me_src->mpoly;
       MLoop *loops_src = me_src->mloop;
-      float(*vcos_src)[3] = BKE_mesh_vertexCos_get(me_src, NULL);
+      float(*vcos_src)[3] = BKE_mesh_vert_coords_alloc(me_src, NULL);
 
       BKE_bvhtree_from_mesh_get(&treedata, me_src, BVHTREE_FROM_LOOPTRI, 2);
 
@@ -1350,7 +1350,7 @@ void BKE_mesh_remap_calc_loops_from_mesh(const int mode,
     size_t islands_res_buff_size = MREMAP_DEFAULT_BUFSIZE;
 
     if (!use_from_vert) {
-      vcos_src = BKE_mesh_vertexCos_get(me_src, NULL);
+      vcos_src = BKE_mesh_vert_coords_alloc(me_src, NULL);
 
       vcos_interp = MEM_mallocN(sizeof(*vcos_interp) * buff_size_interp, __func__);
       indices_interp = MEM_mallocN(sizeof(*indices_interp) * buff_size_interp, __func__);
@@ -1553,7 +1553,9 @@ void BKE_mesh_remap_calc_loops_from_mesh(const int mode,
                                      num_verts_active,
                                      0.0,
                                      2,
-                                     6);
+                                     6,
+                                     0,
+                                     NULL);
         }
 
         MEM_freeN(verts_active);
@@ -1594,7 +1596,9 @@ void BKE_mesh_remap_calc_loops_from_mesh(const int mode,
                                        num_looptri_active,
                                        0.0,
                                        2,
-                                       6);
+                                       6,
+                                       0,
+                                       NULL);
         }
 
         MEM_freeN(looptri_active);
diff --git a/source/blender/blenkernel/intern/multires_reshape.c b/source/blender/blenkernel/intern/multires_reshape.c
index d939267ac5a..9947b96105d 100644
--- a/source/blender/blenkernel/intern/multires_reshape.c
+++ b/source/blender/blenkernel/intern/multires_reshape.c
@@ -891,7 +891,7 @@ bool multiresModifier_reshapeFromObject(struct Depsgraph *depsgraph,
   Object *src_eval = DEG_get_evaluated_object(depsgraph, src);
   Mesh *src_mesh_eval = mesh_get_eval_final(depsgraph, scene_eval, src_eval, &CD_MASK_BAREMESH);
   int num_deformed_verts;
-  float(*deformed_verts)[3] = BKE_mesh_vertexCos_get(src_mesh_eval, &num_deformed_verts);
+  float(*deformed_verts)[3] = BKE_mesh_vert_coords_alloc(src_mesh_eval, &num_deformed_verts);
   bool result = multires_reshape_from_vertcos(
       depsgraph, dst, &reshape_mmd, deformed_verts, num_deformed_verts, false);
   MEM_freeN(deformed_verts);
@@ -926,7 +926,7 @@ bool multiresModifier_reshapeFromDeformModifier(struct Depsgraph *depsgraph,
    * deformation modifiers will be applied though). */
   Mesh *multires_mesh = BKE_multires_create_mesh(depsgraph, scene_eval, &highest_mmd, object);
   int num_deformed_verts;
-  float(*deformed_verts)[3] = BKE_mesh_vertexCos_get(multires_mesh, &num_deformed_verts);
+  float(*deformed_verts)[3] = BKE_mesh_vert_coords_alloc(multires_mesh, &num_deformed_verts);
   /* Apply deformation modifier on the multires, */
   const ModifierEvalContext modifier_ctx = {
       .depsgraph = depsgraph,
diff --git a/source/blender/blenkernel/intern/node.c b/source/blender/blenkernel/intern/node.c
index 986571e34bd..2eba71fa6bd 100644
--- a/source/blender/blenkernel/intern/node.c
+++ b/source/blender/blenkernel/intern/node.c
@@ -1406,6 +1406,7 @@ bNodeTree *ntreeAddTree(Main *bmain, const char *name, const char *idname)
   }
   else {
     ntree = MEM_callocN(sizeof(bNodeTree), "new node tree");
+    ntree->id.flag |= LIB_PRIVATE_DATA;
     *((short *)ntree->id.name) = ID_NT;
     BLI_strncpy(ntree->id.name + 2, name, sizeof(ntree->id.name));
   }
@@ -2172,6 +2173,7 @@ void ntreeSetOutput(bNodeTree *ntree)
    * might be different for editor or for "real" use... */
 }
 
+/* Returns the private NodeTree object of the datablock, if it has one. */
 bNodeTree *ntreeFromID(const ID *id)
 {
   switch (GS(id->name)) {
@@ -2192,6 +2194,28 @@ bNodeTree *ntreeFromID(const ID *id)
   }
 }
 
+/* Finds and returns the datablock that privately owns the given tree, or NULL. */
+ID *BKE_node_tree_find_owner_ID(Main *bmain, struct bNodeTree *ntree)
+{
+  ListBase *lists[] = {&bmain->materials,
+                       &bmain->lights,
+                       &bmain->worlds,
+                       &bmain->textures,
+                       &bmain->scenes,
+                       &bmain->linestyles,
+                       NULL};
+
+  for (int i = 0; lists[i] != NULL; i++) {
+    LISTBASE_FOREACH (ID *, id, lists[i]) {
+      if (ntreeFromID(id) == ntree) {
+        return id;
+      }
+    }
+  }
+
+  return NULL;
+}
+
 void ntreeMakeLocal(Main *bmain, bNodeTree *ntree, bool id_in_mainlist, const bool lib_local)
 {
   BKE_id_make_local_generic(bmain, &ntree->id, id_in_mainlist, lib_local);
@@ -2799,6 +2823,16 @@ int nodeSocketIsHidden(bNodeSocket *sock)
   return ((sock->flag & (SOCK_HIDDEN | SOCK_UNAVAIL)) != 0);
 }
 
+void nodeSetSocketAvailability(bNodeSocket *sock, bool is_available)
+{
+  if (is_available) {
+    sock->flag &= ~SOCK_UNAVAIL;
+  }
+  else {
+    sock->flag |= SOCK_UNAVAIL;
+  }
+}
+
 /* ************** Node Clipboard *********** */
 
 #define USE_NODE_CB_VALIDATE
@@ -3222,11 +3256,6 @@ void ntreeTagUsedSockets(bNodeTree *ntree)
   }
 
   for (link = ntree->links.first; link; link = link->next) {
-    /* link is unused if either side is disabled */
-    if ((link->fromsock->flag & SOCK_UNAVAIL) || (link->tosock->flag & SOCK_UNAVAIL)) {
-      continue;
-    }
-
     link->fromsock->flag |= SOCK_IN_USE;
     link->tosock->flag |= SOCK_IN_USE;
   }
@@ -3928,6 +3957,7 @@ static void registerShaderNodes(void)
   register_node_type_sh_tex_brick();
   register_node_type_sh_tex_pointdensity();
   register_node_type_sh_tex_ies();
+  register_node_type_sh_tex_white_noise();
 }
 
 static void registerTextureNodes(void)
@@ -3954,6 +3984,7 @@ static void registerTextureNodes(void)
   register_node_type_sh_tangent();
   register_node_type_sh_normal_map();
   register_node_type_sh_hair_info();
+  register_node_type_sh_volume_info();
 
   register_node_type_tex_checker();
   register_node_type_tex_texture();
diff --git a/source/blender/blenkernel/intern/object.c b/source/blender/blenkernel/intern/object.c
index d95c02cdf7f..ed517bfc513 100644
--- a/source/blender/blenkernel/intern/object.c
+++ b/source/blender/blenkernel/intern/object.c
@@ -2985,6 +2985,15 @@ bool BKE_object_empty_image_data_is_visible_in_view3d(const Object *ob, const Re
     }
   }
 
+  if (visibility_flag & OB_EMPTY_IMAGE_HIDE_NON_AXIS_ALIGNED) {
+    float proj[3];
+    project_plane_v3_v3v3(proj, ob->obmat[2], rv3d->viewinv[2]);
+    const float proj_length_sq = len_squared_v3(proj);
+    if (proj_length_sq > 1e-5f) {
+      return false;
+    }
+  }
+
   return true;
 }
 
@@ -3587,7 +3596,7 @@ bool BKE_object_shapekey_free(Main *bmain, Object *ob)
 
   BKE_id_free_us(bmain, key);
 
-  return false;
+  return true;
 }
 
 bool BKE_object_shapekey_remove(Main *bmain, Object *ob, KeyBlock *kb)
@@ -3871,7 +3880,7 @@ int BKE_object_scenes_users_get(Main *bmain, Object *ob)
 {
   int num_scenes = 0;
   for (Scene *scene = bmain->scenes.first; scene != NULL; scene = scene->id.next) {
-    if (BKE_collection_has_object_recursive(BKE_collection_master(scene), ob)) {
+    if (BKE_collection_has_object_recursive(scene->master_collection, ob)) {
       num_scenes++;
     }
   }
diff --git a/source/blender/blenkernel/intern/object_update.c b/source/blender/blenkernel/intern/object_update.c
index 9b717650e8e..3a330ea0d5a 100644
--- a/source/blender/blenkernel/intern/object_update.c
+++ b/source/blender/blenkernel/intern/object_update.c
@@ -436,7 +436,7 @@ void BKE_object_eval_eval_base_flags(Depsgraph *depsgraph,
   BKE_base_eval_flags(base);
 
   /* For render, compute base visibility again since BKE_base_eval_flags
-   * assumed viewport visibility. Selectability does not matter here. */
+   * assumed viewport visibility. Select-ability does not matter here. */
   if (DEG_get_mode(depsgraph) == DAG_EVAL_RENDER) {
     if (base->flag & BASE_ENABLED_RENDER) {
       base->flag |= BASE_VISIBLE;
diff --git a/source/blender/blenkernel/intern/ocean.c b/source/blender/blenkernel/intern/ocean.c
index 39fb668c873..fcceebc3913 100644
--- a/source/blender/blenkernel/intern/ocean.c
+++ b/source/blender/blenkernel/intern/ocean.c
@@ -519,7 +519,7 @@ static void ocean_compute_htilda(void *__restrict userdata,
 
   int j;
 
-  /* note the <= _N/2 here, see the fftw doco
+  /* Note the <= _N/2 here, see the FFTW documentation
    * about the mechanics of the complex->real fft storage. */
   for (j = 0; j <= o->_N / 2; ++j) {
     fftw_complex exp_param1;
diff --git a/source/blender/blenkernel/intern/paint.c b/source/blender/blenkernel/intern/paint.c
index 4a41ffbfa8c..a46bb36c883 100644
--- a/source/blender/blenkernel/intern/paint.c
+++ b/source/blender/blenkernel/intern/paint.c
@@ -1222,10 +1222,10 @@ static void sculpt_update_object(
       BKE_sculptsession_free_deformMats(ss);
 
       ss->orig_cos = (ss->kb) ? BKE_keyblock_convert_to_vertcos(ob, ss->kb) :
-                                BKE_mesh_vertexCos_get(me, NULL);
+                                BKE_mesh_vert_coords_alloc(me, NULL);
 
       BKE_crazyspace_build_sculpt(depsgraph, scene, ob, &ss->deform_imats, &ss->deform_cos);
-      BKE_pbvh_apply_vertCos(ss->pbvh, ss->deform_cos, me->totvert);
+      BKE_pbvh_vert_coords_apply(ss->pbvh, ss->deform_cos, me->totvert);
 
       for (a = 0; a < me->totvert; ++a) {
         invert_m3(ss->deform_imats[a]);
@@ -1242,14 +1242,14 @@ static void sculpt_update_object(
 
   /* if pbvh is deformed, key block is already applied to it */
   if (ss->kb) {
-    bool pbvh_deformed = BKE_pbvh_isDeformed(ss->pbvh);
+    bool pbvh_deformed = BKE_pbvh_is_deformed(ss->pbvh);
     if (!pbvh_deformed || ss->deform_cos == NULL) {
       float(*vertCos)[3] = BKE_keyblock_convert_to_vertcos(ob, ss->kb);
 
       if (vertCos) {
         if (!pbvh_deformed) {
           /* apply shape keys coordinates to PBVH */
-          BKE_pbvh_apply_vertCos(ss->pbvh, vertCos, me->totvert);
+          BKE_pbvh_vert_coords_apply(ss->pbvh, vertCos, me->totvert);
         }
         if (ss->deform_cos == NULL) {
           ss->deform_cos = vertCos;
@@ -1477,8 +1477,8 @@ static PBVH *build_pbvh_from_regular_mesh(Object *ob, Mesh *me_eval_deform)
   const bool is_deformed = check_sculpt_object_deformed(ob, true);
   if (is_deformed && me_eval_deform != NULL) {
     int totvert;
-    float(*v_cos)[3] = BKE_mesh_vertexCos_get(me_eval_deform, &totvert);
-    BKE_pbvh_apply_vertCos(pbvh, v_cos, totvert);
+    float(*v_cos)[3] = BKE_mesh_vert_coords_alloc(me_eval_deform, &totvert);
+    BKE_pbvh_vert_coords_apply(pbvh, v_cos, totvert);
     MEM_freeN(v_cos);
   }
 
diff --git a/source/blender/blenkernel/intern/particle.c b/source/blender/blenkernel/intern/particle.c
index ffeba0148a2..2de8d4a643a 100644
--- a/source/blender/blenkernel/intern/particle.c
+++ b/source/blender/blenkernel/intern/particle.c
@@ -2691,7 +2691,7 @@ static void psys_thread_create_path(ParticleTask *task,
       pa = &psys->particles[cpa->parent];
       par = pcache[cpa->parent];
 
-      /* If particle is unexisting, try to pick a viable parent from particles
+      /* If particle is non-existing, try to pick a viable parent from particles
        * used for interpolation. */
       for (k = 0; k < 4 && pa && (pa->flag & PARS_UNEXIST); k++) {
         if (cpa->pa[k] >= 0) {
diff --git a/source/blender/blenkernel/intern/particle_system.c b/source/blender/blenkernel/intern/particle_system.c
index 31484b59127..e328ae8952e 100644
--- a/source/blender/blenkernel/intern/particle_system.c
+++ b/source/blender/blenkernel/intern/particle_system.c
@@ -3490,10 +3490,10 @@ static void do_hair_dynamics(ParticleSimulationData *sim)
   psys->clmd->sim_parms->effector_weights = psys->part->effector_weights;
 
   BKE_id_copy_ex(NULL, &psys->hair_in_mesh->id, (ID **)&psys->hair_out_mesh, LIB_ID_COPY_LOCALIZE);
-  deformedVerts = BKE_mesh_vertexCos_get(psys->hair_out_mesh, NULL);
+  deformedVerts = BKE_mesh_vert_coords_alloc(psys->hair_out_mesh, NULL);
   clothModifier_do(
       psys->clmd, sim->depsgraph, sim->scene, sim->ob, psys->hair_in_mesh, deformedVerts);
-  BKE_mesh_apply_vert_coords(psys->hair_out_mesh, deformedVerts);
+  BKE_mesh_vert_coords_apply(psys->hair_out_mesh, deformedVerts);
 
   MEM_freeN(deformedVerts);
 
diff --git a/source/blender/blenkernel/intern/pbvh.c b/source/blender/blenkernel/intern/pbvh.c
index 64a16a0e3be..7a8c082842e 100644
--- a/source/blender/blenkernel/intern/pbvh.c
+++ b/source/blender/blenkernel/intern/pbvh.c
@@ -533,7 +533,7 @@ static void pbvh_build(PBVH *bvh, BB *cb, BBC *prim_bbc, int totprim)
  * Do a full rebuild with on Mesh data structure.
  *
  * \note Unlike mpoly/mloop/verts, looptri is **totally owned** by PBVH
- * (which means it may rewrite it if needed, see #BKE_pbvh_apply_vertCos().
+ * (which means it may rewrite it if needed, see #BKE_pbvh_vert_coords_apply().
  */
 void BKE_pbvh_build_mesh(PBVH *bvh,
                          const MPoly *mpoly,
@@ -1753,14 +1753,21 @@ static bool pbvh_faces_node_raycast(PBVH *bvh,
                                     const PBVHNode *node,
                                     float (*origco)[3],
                                     const float ray_start[3],
+                                    const float ray_normal[3],
                                     struct IsectRayPrecalc *isect_precalc,
-                                    float *depth)
+                                    float *depth,
+                                    int *r_active_vertex_index,
+                                    float *r_face_normal)
 {
   const MVert *vert = bvh->verts;
   const MLoop *mloop = bvh->mloop;
   const int *faces = node->prim_indices;
   int i, totface = node->totprim;
   bool hit = false;
+  float min_depth = FLT_MAX;
+  float location[3] = {0.0f};
+  float nearest_vertex_co[3];
+  copy_v3_fl(nearest_vertex_co, 0.0f);
 
   for (i = 0; i < totface; ++i) {
     const MLoopTri *lt = &bvh->looptri[faces[i]];
@@ -1787,6 +1794,22 @@ static bool pbvh_faces_node_raycast(PBVH *bvh,
                                        vert[mloop[lt->tri[1]].v].co,
                                        vert[mloop[lt->tri[2]].v].co,
                                        depth);
+
+      if (hit && *depth < min_depth) {
+        min_depth = *depth;
+        normal_tri_v3(r_face_normal,
+                      vert[mloop[lt->tri[0]].v].co,
+                      vert[mloop[lt->tri[1]].v].co,
+                      vert[mloop[lt->tri[2]].v].co);
+        madd_v3_v3v3fl(location, ray_start, ray_normal, *depth);
+        for (int j = 0; j < 3; j++) {
+          if (len_squared_v3v3(location, vert[mloop[lt->tri[j]].v].co) <
+              len_squared_v3v3(location, nearest_vertex_co)) {
+            copy_v3_v3(nearest_vertex_co, vert[mloop[lt->tri[j]].v].co);
+            *r_active_vertex_index = mloop[lt->tri[j]].v;
+          }
+        }
+      }
     }
   }
 
@@ -1857,8 +1880,11 @@ bool BKE_pbvh_node_raycast(PBVH *bvh,
                            float (*origco)[3],
                            bool use_origco,
                            const float ray_start[3],
+                           const float ray_normal[3],
                            struct IsectRayPrecalc *isect_precalc,
-                           float *depth)
+                           float *depth,
+                           int *active_vertex_index,
+                           float *face_normal)
 {
   bool hit = false;
 
@@ -1868,13 +1894,29 @@ bool BKE_pbvh_node_raycast(PBVH *bvh,
 
   switch (bvh->type) {
     case PBVH_FACES:
-      hit |= pbvh_faces_node_raycast(bvh, node, origco, ray_start, isect_precalc, depth);
+      hit |= pbvh_faces_node_raycast(bvh,
+                                     node,
+                                     origco,
+                                     ray_start,
+                                     ray_normal,
+                                     isect_precalc,
+                                     depth,
+                                     active_vertex_index,
+                                     face_normal);
       break;
     case PBVH_GRIDS:
       hit |= pbvh_grids_node_raycast(bvh, node, origco, ray_start, isect_precalc, depth);
       break;
     case PBVH_BMESH:
-      hit = pbvh_bmesh_node_raycast(node, ray_start, isect_precalc, depth, use_origco);
+      BM_mesh_elem_index_ensure(bvh->bm, BM_VERT);
+      hit = pbvh_bmesh_node_raycast(node,
+                                    ray_start,
+                                    ray_normal,
+                                    isect_precalc,
+                                    depth,
+                                    use_origco,
+                                    active_vertex_index,
+                                    face_normal);
       break;
   }
 
@@ -2306,7 +2348,7 @@ void BKE_pbvh_node_layer_disp_free(PBVHNode *node)
   }
 }
 
-float (*BKE_pbvh_get_vertCos(PBVH *pbvh))[3]
+float (*BKE_pbvh_vert_coords_alloc(PBVH *pbvh))[3]
 {
   float(*vertCos)[3] = NULL;
 
@@ -2324,7 +2366,7 @@ float (*BKE_pbvh_get_vertCos(PBVH *pbvh))[3]
   return vertCos;
 }
 
-void BKE_pbvh_apply_vertCos(PBVH *pbvh, float (*vertCos)[3], const int totvert)
+void BKE_pbvh_vert_coords_apply(PBVH *pbvh, const float (*vertCos)[3], const int totvert)
 {
   if (totvert != pbvh->totvert) {
     BLI_assert(!"PBVH: Given deforming vcos number does not natch PBVH vertex number!");
@@ -2368,7 +2410,7 @@ void BKE_pbvh_apply_vertCos(PBVH *pbvh, float (*vertCos)[3], const int totvert)
   }
 }
 
-bool BKE_pbvh_isDeformed(PBVH *pbvh)
+bool BKE_pbvh_is_deformed(PBVH *pbvh)
 {
   return pbvh->deformed;
 }
diff --git a/source/blender/blenkernel/intern/pbvh_bmesh.c b/source/blender/blenkernel/intern/pbvh_bmesh.c
index 1d8088c6605..c04e172f116 100644
--- a/source/blender/blenkernel/intern/pbvh_bmesh.c
+++ b/source/blender/blenkernel/intern/pbvh_bmesh.c
@@ -1508,12 +1508,18 @@ static bool pbvh_bmesh_collapse_short_edges(EdgeQueueContext *eq_ctx,
 
 bool pbvh_bmesh_node_raycast(PBVHNode *node,
                              const float ray_start[3],
+                             const float ray_normal[3],
                              struct IsectRayPrecalc *isect_precalc,
                              float *depth,
-                             bool use_original)
+                             bool use_original,
+                             int *r_active_vertex_index,
+                             float *r_face_normal)
 {
   bool hit = false;
 
+  float min_depth = FLT_MAX;
+  float nearest_vertex_co[3] = {0.0f};
+  float location[3] = {0.0f};
   if (use_original && node->bm_tot_ortri) {
     for (int i = 0; i < node->bm_tot_ortri; i++) {
       const int *t = node->bm_ortri[i];
@@ -1538,6 +1544,19 @@ bool pbvh_bmesh_node_raycast(PBVHNode *node,
         BM_face_as_array_vert_tri(f, v_tri);
         hit |= ray_face_intersection_tri(
             ray_start, isect_precalc, v_tri[0]->co, v_tri[1]->co, v_tri[2]->co, depth);
+
+        if (hit && *depth < min_depth) {
+          min_depth = *depth;
+          normal_tri_v3(r_face_normal, v_tri[0]->co, v_tri[1]->co, v_tri[2]->co);
+          madd_v3_v3v3fl(location, ray_start, ray_normal, *depth);
+          for (int j = 0; j < 3; j++) {
+            if (len_squared_v3v3(location, v_tri[j]->co) <
+                len_squared_v3v3(location, nearest_vertex_co)) {
+              copy_v3_v3(nearest_vertex_co, v_tri[j]->co);
+              *r_active_vertex_index = BM_elem_index_get(v_tri[j]);
+            }
+          }
+        }
       }
     }
   }
diff --git a/source/blender/blenkernel/intern/pbvh_intern.h b/source/blender/blenkernel/intern/pbvh_intern.h
index e74a8d43c68..bad103743eb 100644
--- a/source/blender/blenkernel/intern/pbvh_intern.h
+++ b/source/blender/blenkernel/intern/pbvh_intern.h
@@ -206,9 +206,12 @@ void pbvh_update_BB_redraw(PBVH *bvh, PBVHNode **nodes, int totnode, int flag);
 /* pbvh_bmesh.c */
 bool pbvh_bmesh_node_raycast(PBVHNode *node,
                              const float ray_start[3],
+                             const float ray_normal[3],
                              struct IsectRayPrecalc *isect_precalc,
                              float *dist,
-                             bool use_original);
+                             bool use_original,
+                             int *r_active_vertex_index,
+                             float *r_face_normal);
 bool pbvh_bmesh_node_nearest_to_ray(PBVHNode *node,
                                     const float ray_start[3],
                                     const float ray_normal[3],
diff --git a/source/blender/blenkernel/intern/pointcache.c b/source/blender/blenkernel/intern/pointcache.c
index 13d0f1adb84..a31cb96c6c9 100644
--- a/source/blender/blenkernel/intern/pointcache.c
+++ b/source/blender/blenkernel/intern/pointcache.c
@@ -1831,6 +1831,10 @@ void BKE_ptcache_id_from_rigidbody(PTCacheID *pid, Object *ob, RigidBodyWorld *r
   pid->file_type = PTCACHE_FILE_PTCACHE;
 }
 
+/**
+ * \param ob: Optional, may be NULL.
+ * \param scene: Optional may be NULL.
+ */
 PTCacheID BKE_ptcache_id_find(Object *ob, Scene *scene, PointCache *cache)
 {
   PTCacheID result = {0};
@@ -1934,40 +1938,45 @@ static bool foreach_object_ptcache(
     Scene *scene, Object *object, int duplis, ForeachPtcacheCb callback, void *callback_user_data)
 {
   PTCacheID pid;
-  /* Soft body. */
-  if (object->soft != NULL) {
-    BKE_ptcache_id_from_softbody(&pid, object, object->soft);
-    if (!callback(&pid, callback_user_data)) {
+
+  if (object != NULL) {
+    /* Soft body. */
+    if (object->soft != NULL) {
+      BKE_ptcache_id_from_softbody(&pid, object, object->soft);
+      if (!callback(&pid, callback_user_data)) {
+        return false;
+      }
+    }
+    /* Particle systems. */
+    if (!foreach_object_particle_ptcache(object, callback, callback_user_data)) {
       return false;
     }
+    /* Modifiers. */
+    if (!foreach_object_modifier_ptcache(object, callback, callback_user_data)) {
+      return false;
+    }
+    /* Consider all object in dupli groups to be part of the same object,
+     * for baking with linking dupligroups. Once we have better overrides
+     * this can be revisited so users select the local objects directly. */
+    if (scene != NULL && (duplis-- > 0) && (object->instance_collection != NULL)) {
+      FOREACH_COLLECTION_OBJECT_RECURSIVE_BEGIN (object->instance_collection, current_object) {
+        if (current_object == object) {
+          continue;
+        }
+        foreach_object_ptcache(scene, current_object, duplis, callback, callback_user_data);
+      }
+      FOREACH_COLLECTION_OBJECT_RECURSIVE_END;
+    }
   }
-  /* Particle systems. */
-  if (!foreach_object_particle_ptcache(object, callback, callback_user_data)) {
-    return false;
-  }
-  /* Modifiers. */
-  if (!foreach_object_modifier_ptcache(object, callback, callback_user_data)) {
-    return false;
-  }
+
   /* Rigid body. */
-  if (scene != NULL && object->rigidbody_object != NULL && scene->rigidbody_world != NULL) {
+  if (scene != NULL && (object == NULL || object->rigidbody_object != NULL) &&
+      scene->rigidbody_world != NULL) {
     BKE_ptcache_id_from_rigidbody(&pid, object, scene->rigidbody_world);
     if (!callback(&pid, callback_user_data)) {
       return false;
     }
   }
-  /* Consider all object in dupli groups to be part of the same object,
-   * for baking with linking dupligroups. Once we have better overrides
-   * this can be revisited so users select the local objects directly. */
-  if (scene != NULL && (duplis-- > 0) && (object->instance_collection != NULL)) {
-    FOREACH_COLLECTION_OBJECT_RECURSIVE_BEGIN (object->instance_collection, current_object) {
-      if (current_object == object) {
-        continue;
-      }
-      foreach_object_ptcache(scene, current_object, duplis, callback, callback_user_data);
-    }
-    FOREACH_COLLECTION_OBJECT_RECURSIVE_END;
-  }
   return true;
 }
 
diff --git a/source/blender/blenkernel/intern/rigidbody.c b/source/blender/blenkernel/intern/rigidbody.c
index ec73406c14c..514f000d73d 100644
--- a/source/blender/blenkernel/intern/rigidbody.c
+++ b/source/blender/blenkernel/intern/rigidbody.c
@@ -1452,7 +1452,7 @@ void BKE_rigidbody_remove_object(Main *bmain, Scene *scene, Object *ob)
       /* Some users seems to find it funny to use a view-layer instancing collection
        * as RBW collection... Despite this being a bad (ab)use of the system, avoid losing objects
        * when we remove them from RB simulation. */
-      BKE_collection_object_add(bmain, BKE_collection_master(scene), ob);
+      BKE_collection_object_add(bmain, scene->master_collection, ob);
     }
     BKE_collection_object_remove(bmain, rbw->group, ob, false);
   }
diff --git a/source/blender/blenkernel/intern/scene.c b/source/blender/blenkernel/intern/scene.c
index 3440c8dfb69..c4ce0cf6ed7 100644
--- a/source/blender/blenkernel/intern/scene.c
+++ b/source/blender/blenkernel/intern/scene.c
@@ -266,6 +266,8 @@ void BKE_scene_copy_data(Main *bmain, Scene *sce_dst, const Scene *sce_src, cons
 {
   /* We never handle usercount here for own data. */
   const int flag_subdata = flag | LIB_ID_CREATE_NO_USER_REFCOUNT;
+  /* We always need allocation of our private ID data. */
+  const int flag_private_id_data = flag_subdata & ~LIB_ID_CREATE_NO_ALLOCATE;
 
   sce_dst->ed = NULL;
   sce_dst->depsgraph_hash = NULL;
@@ -273,8 +275,10 @@ void BKE_scene_copy_data(Main *bmain, Scene *sce_dst, const Scene *sce_src, cons
 
   /* Master Collection */
   if (sce_src->master_collection) {
-    sce_dst->master_collection = BKE_collection_copy_master(
-        bmain, sce_src->master_collection, flag);
+    BKE_id_copy_ex(bmain,
+                   (ID *)sce_src->master_collection,
+                   (ID **)&sce_dst->master_collection,
+                   flag_private_id_data);
   }
 
   /* View Layers */
@@ -292,10 +296,13 @@ void BKE_scene_copy_data(Main *bmain, Scene *sce_dst, const Scene *sce_src, cons
   BKE_keyingsets_copy(&(sce_dst->keyingsets), &(sce_src->keyingsets));
 
   if (sce_src->nodetree) {
-    /* Note: nodetree is *not* in bmain, however this specific case is handled at lower level
-     *       (see BKE_libblock_copy_ex()). */
-    BKE_id_copy_ex(bmain, (ID *)sce_src->nodetree, (ID **)&sce_dst->nodetree, flag);
-    BKE_libblock_relink_ex(bmain, sce_dst->nodetree, (void *)(&sce_src->id), &sce_dst->id, false);
+    BKE_id_copy_ex(
+        bmain, (ID *)sce_src->nodetree, (ID **)&sce_dst->nodetree, flag_private_id_data);
+    BKE_libblock_relink_ex(bmain,
+                           sce_dst->nodetree,
+                           (void *)(&sce_src->id),
+                           &sce_dst->id,
+                           ID_REMAP_SKIP_NEVER_NULL_USAGE);
   }
 
   if (sce_src->rigidbody_world) {
diff --git a/source/blender/blenkernel/intern/seqeffects.c b/source/blender/blenkernel/intern/seqeffects.c
index b151bc7f092..89f3f9ef9fd 100644
--- a/source/blender/blenkernel/intern/seqeffects.c
+++ b/source/blender/blenkernel/intern/seqeffects.c
@@ -1646,7 +1646,7 @@ static void do_blend_effect_float(
     case SEQ_TYPE_DARKEN:
       apply_blend_function_float(facf0, facf1, x, y, rect1, rect2, out, blend_color_darken_float);
       break;
-    case SEQ_TYPE_BURN:
+    case SEQ_TYPE_COLOR_BURN:
       apply_blend_function_float(facf0, facf1, x, y, rect1, rect2, out, blend_color_burn_float);
       break;
     case SEQ_TYPE_LINEAR_BURN:
@@ -1734,7 +1734,7 @@ static void do_blend_effect_byte(float facf0,
     case SEQ_TYPE_DARKEN:
       apply_blend_function_byte(facf0, facf1, x, y, rect1, rect2, out, blend_color_darken_byte);
       break;
-    case SEQ_TYPE_BURN:
+    case SEQ_TYPE_COLOR_BURN:
       apply_blend_function_byte(facf0, facf1, x, y, rect1, rect2, out, blend_color_burn_byte);
       break;
     case SEQ_TYPE_LINEAR_BURN:
@@ -4130,7 +4130,7 @@ static struct SeqEffectHandle get_sequence_effect_impl(int seq_type)
       break;
     case SEQ_TYPE_SCREEN:
     case SEQ_TYPE_OVERLAY:
-    case SEQ_TYPE_BURN:
+    case SEQ_TYPE_COLOR_BURN:
     case SEQ_TYPE_LINEAR_BURN:
     case SEQ_TYPE_DARKEN:
     case SEQ_TYPE_LIGHTEN:
diff --git a/source/blender/blenkernel/intern/softbody.c b/source/blender/blenkernel/intern/softbody.c
index 20736e28a53..d3b72fb295d 100644
--- a/source/blender/blenkernel/intern/softbody.c
+++ b/source/blender/blenkernel/intern/softbody.c
@@ -2705,7 +2705,7 @@ static void mesh_to_softbody(Scene *scene, Object *ob)
 
   for (a = 0; a < me->totvert; a++, bp++) {
     /* get scalar values needed  *per vertex* from vertex group functions,
-     * so we can *paint* them nicly ..
+     * so we can *paint* them nicely ..
      * they are normalized [0.0..1.0] so may be we need amplitude for scale
      * which can be done by caller but still .. i'd like it to go this way
      */
diff --git a/source/blender/blenkernel/intern/studiolight.c b/source/blender/blenkernel/intern/studiolight.c
index 6e83f5d75e2..b3f2a624ebe 100644
--- a/source/blender/blenkernel/intern/studiolight.c
+++ b/source/blender/blenkernel/intern/studiolight.c
@@ -46,13 +46,16 @@
 
 #include "MEM_guardedalloc.h"
 
+#include "intern/openexr/openexr_multi.h"
+
 /* Statics */
 static ListBase studiolights;
 static int last_studiolight_id = 0;
 #define STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE 96
 #define STUDIOLIGHT_IRRADIANCE_EQUIRECT_HEIGHT 32
 #define STUDIOLIGHT_IRRADIANCE_EQUIRECT_WIDTH (STUDIOLIGHT_IRRADIANCE_EQUIRECT_HEIGHT * 2)
-
+#define STUDIOLIGHT_PASSNAME_DIFFUSE "diffuse"
+#define STUDIOLIGHT_PASSNAME_SPECULAR "specular"
 /*
  * The method to calculate the irradiance buffers
  * The irradiance buffer is only shown in the background when in LookDev.
@@ -152,6 +155,10 @@ static void studiolight_free(struct StudioLight *sl)
   GPU_TEXTURE_SAFE_FREE(sl->equirect_irradiance_gputexture);
   IMB_SAFE_FREE(sl->equirect_radiance_buffer);
   IMB_SAFE_FREE(sl->equirect_irradiance_buffer);
+  GPU_TEXTURE_SAFE_FREE(sl->matcap_diffuse.gputexture);
+  GPU_TEXTURE_SAFE_FREE(sl->matcap_specular.gputexture);
+  IMB_SAFE_FREE(sl->matcap_diffuse.ibuf);
+  IMB_SAFE_FREE(sl->matcap_specular.ibuf);
   MEM_SAFE_FREE(sl->path_irr_cache);
   MEM_SAFE_FREE(sl->path_sh_cache);
   MEM_SAFE_FREE(sl);
@@ -342,72 +349,232 @@ static void cube_face_uv_to_direction(float r_dir[3], float x, float y, int face
   normalize_v3(r_dir);
 }
 
+typedef struct MultilayerConvertContext {
+  int num_diffuse_channels;
+  float *diffuse_pass;
+  int num_specular_channels;
+  float *specular_pass;
+} MultilayerConvertContext;
+
+static void *studiolight_multilayer_addview(void *UNUSED(base), const char *UNUSED(view_name))
+{
+  return NULL;
+}
+static void *studiolight_multilayer_addlayer(void *base, const char *UNUSED(layer_name))
+{
+  return base;
+}
+
+/* Convert a multilayer pass to ImBuf channel 4 float buffer.
+ * NOTE: Parameter rect will become invalid. Do not use rect after calling this
+ * function */
+static float *studiolight_multilayer_convert_pass(ImBuf *ibuf,
+                                                  float *rect,
+                                                  const unsigned int channels)
+{
+  if (channels == 4) {
+    return rect;
+  }
+  else {
+    float *new_rect = MEM_callocN(sizeof(float[4]) * ibuf->x * ibuf->y, __func__);
+
+    IMB_buffer_float_from_float(new_rect,
+                                rect,
+                                channels,
+                                IB_PROFILE_LINEAR_RGB,
+                                IB_PROFILE_LINEAR_RGB,
+                                false,
+                                ibuf->x,
+                                ibuf->y,
+                                ibuf->x,
+                                ibuf->x);
+
+    MEM_freeN(rect);
+    return new_rect;
+  }
+}
+
+static void studiolight_multilayer_addpass(void *base,
+                                           void *UNUSED(lay),
+                                           const char *pass_name,
+                                           float *rect,
+                                           int num_channels,
+                                           const char *UNUSED(chan_id),
+                                           const char *UNUSED(view_name))
+{
+  MultilayerConvertContext *ctx = base;
+  /* NOTE: This function must free pass pixels data if it is not used, this
+   * is how IMB_exr_multilayer_convert() is working. */
+  /* If we've found a first combined pass, skip all the rest ones. */
+  if (STREQ(pass_name, STUDIOLIGHT_PASSNAME_DIFFUSE)) {
+    ctx->diffuse_pass = rect;
+    ctx->num_diffuse_channels = num_channels;
+  }
+  else if (STREQ(pass_name, STUDIOLIGHT_PASSNAME_SPECULAR)) {
+    ctx->specular_pass = rect;
+    ctx->num_specular_channels = num_channels;
+  }
+  else {
+    MEM_freeN(rect);
+  }
+}
+
 static void studiolight_load_equirect_image(StudioLight *sl)
 {
   if (sl->flag & STUDIOLIGHT_EXTERNAL_FILE) {
-    ImBuf *ibuf = NULL;
-    ibuf = IMB_loadiffname(sl->path, 0, NULL);
-    if (ibuf == NULL) {
-      float *colbuf = MEM_mallocN(sizeof(float[4]), __func__);
-      copy_v4_fl4(colbuf, 1.0f, 0.0f, 1.0f, 1.0f);
-      ibuf = IMB_allocFromBuffer(NULL, colbuf, 1, 1);
+    ImBuf *ibuf = IMB_loadiffname(sl->path, IB_multilayer, NULL);
+    ImBuf *specular_ibuf = NULL;
+    ImBuf *diffuse_ibuf = NULL;
+    const bool failed = (ibuf == NULL);
+
+    if (ibuf) {
+      if (ibuf->ftype == IMB_FTYPE_OPENEXR && ibuf->userdata) {
+        /* the read file is a multilayered openexr file (userdata != NULL)
+         * This file is currently only supported for MATCAPS where
+         * the first found 'diffuse' pass will be used for diffuse lighting
+         * and the first found 'specular' pass will be used for specular lighting */
+        MultilayerConvertContext ctx = {0};
+        IMB_exr_multilayer_convert(ibuf->userdata,
+                                   &ctx,
+                                   &studiolight_multilayer_addview,
+                                   &studiolight_multilayer_addlayer,
+                                   &studiolight_multilayer_addpass);
+
+        /* `ctx.diffuse_pass` and `ctx.specular_pass` can be freed inside
+         * `studiolight_multilayer_convert_pass` when conversion happens.
+         * When not converted we move the ownership of the buffer to the
+         * `converted_pass`. We only need to free `converted_pass` as it holds
+         * the unmodified allocation from the `ctx.*_pass` or the converted data.
+         */
+        if (ctx.diffuse_pass != NULL) {
+          float *converted_pass = studiolight_multilayer_convert_pass(
+              ibuf, ctx.diffuse_pass, ctx.num_diffuse_channels);
+          diffuse_ibuf = IMB_allocFromBuffer(
+              NULL, converted_pass, ibuf->x, ibuf->y, ctx.num_diffuse_channels);
+          MEM_freeN(converted_pass);
+        }
+
+        if (ctx.specular_pass != NULL) {
+          float *converted_pass = studiolight_multilayer_convert_pass(
+              ibuf, ctx.specular_pass, ctx.num_specular_channels);
+          specular_ibuf = IMB_allocFromBuffer(
+              NULL, converted_pass, ibuf->x, ibuf->y, ctx.num_specular_channels);
+          MEM_freeN(converted_pass);
+        }
+
+        IMB_exr_close(ibuf->userdata);
+        ibuf->userdata = NULL;
+        IMB_freeImBuf(ibuf);
+        ibuf = NULL;
+      }
+      else {
+        /* read file is an single layer openexr file or the read file isn't
+         * an openexr file */
+        IMB_float_from_rect(ibuf);
+        diffuse_ibuf = ibuf;
+        ibuf = NULL;
+      }
+    }
+
+    if (diffuse_ibuf == NULL) {
+      /* Create 1x1 diffuse buffer, in case image failed to load or if there was
+       * only a specular pass in the multilayer file or no passes were found. */
+      const float black[4] = {0.0f, 0.0f, 0.0f, 1.0f};
+      const float magenta[4] = {1.0f, 0.0f, 1.0f, 1.0f};
+      diffuse_ibuf = IMB_allocFromBuffer(
+          NULL, (failed || (specular_ibuf == NULL)) ? magenta : black, 1, 1, 4);
+    }
+
+    if ((sl->flag & STUDIOLIGHT_TYPE_MATCAP)) {
+      sl->matcap_diffuse.ibuf = diffuse_ibuf;
+      sl->matcap_specular.ibuf = specular_ibuf;
+      if (specular_ibuf != NULL) {
+        sl->flag |= STUDIOLIGHT_SPECULAR_HIGHLIGHT_PASS;
+      }
+    }
+    else {
+      sl->equirect_radiance_buffer = diffuse_ibuf;
+      if (specular_ibuf != NULL) {
+        IMB_freeImBuf(specular_ibuf);
+      }
     }
-    IMB_float_from_rect(ibuf);
-    sl->equirect_radiance_buffer = ibuf;
   }
+
   sl->flag |= STUDIOLIGHT_EXTERNAL_IMAGE_LOADED;
 }
 
 static void studiolight_create_equirect_radiance_gputexture(StudioLight *sl)
 {
   if (sl->flag & STUDIOLIGHT_EXTERNAL_FILE) {
-    char error[256];
     BKE_studiolight_ensure_flag(sl, STUDIOLIGHT_EXTERNAL_IMAGE_LOADED);
     ImBuf *ibuf = sl->equirect_radiance_buffer;
 
-    if (sl->flag & STUDIOLIGHT_TYPE_MATCAP) {
-      float *gpu_matcap_3components = MEM_callocN(sizeof(float[3]) * ibuf->x * ibuf->y, __func__);
+    sl->equirect_radiance_gputexture = GPU_texture_create_2d(
+        ibuf->x, ibuf->y, GPU_RGBA16F, ibuf->rect_float, NULL);
+    GPUTexture *tex = sl->equirect_radiance_gputexture;
+    GPU_texture_bind(tex, 0);
+    GPU_texture_filter_mode(tex, true);
+    GPU_texture_wrap_mode(tex, true);
+    GPU_texture_unbind(tex);
+  }
+  sl->flag |= STUDIOLIGHT_EQUIRECT_RADIANCE_GPUTEXTURE;
+}
 
-      float(*offset4)[4] = (float(*)[4])ibuf->rect_float;
-      float(*offset3)[3] = (float(*)[3])gpu_matcap_3components;
-      for (int i = 0; i < ibuf->x * ibuf->y; i++, offset4++, offset3++) {
-        copy_v3_v3(*offset3, *offset4);
-      }
+static void studiolight_create_matcap_gputexture(StudioLightImage *sli)
+{
+  BLI_assert(sli->ibuf);
+  ImBuf *ibuf = sli->ibuf;
+  float *gpu_matcap_3components = MEM_callocN(sizeof(float[3]) * ibuf->x * ibuf->y, __func__);
+
+  float(*offset4)[4] = (float(*)[4])ibuf->rect_float;
+  float(*offset3)[3] = (float(*)[3])gpu_matcap_3components;
+  for (int i = 0; i < ibuf->x * ibuf->y; i++, offset4++, offset3++) {
+    copy_v3_v3(*offset3, *offset4);
+  }
+
+  sli->gputexture = GPU_texture_create_nD(ibuf->x,
+                                          ibuf->y,
+                                          0,
+                                          2,
+                                          gpu_matcap_3components,
+                                          GPU_R11F_G11F_B10F,
+                                          GPU_DATA_FLOAT,
+                                          0,
+                                          false,
+                                          NULL);
+  MEM_SAFE_FREE(gpu_matcap_3components);
+}
 
-      sl->equirect_radiance_gputexture = GPU_texture_create_nD(ibuf->x,
-                                                               ibuf->y,
-                                                               0,
-                                                               2,
-                                                               gpu_matcap_3components,
-                                                               GPU_R11F_G11F_B10F,
-                                                               GPU_DATA_FLOAT,
-                                                               0,
-                                                               false,
-                                                               error);
-
-      MEM_SAFE_FREE(gpu_matcap_3components);
+static void studiolight_create_matcap_diffuse_gputexture(StudioLight *sl)
+{
+  if (sl->flag & STUDIOLIGHT_EXTERNAL_FILE) {
+    if (sl->flag & STUDIOLIGHT_TYPE_MATCAP) {
+      BKE_studiolight_ensure_flag(sl, STUDIOLIGHT_EXTERNAL_IMAGE_LOADED);
+      studiolight_create_matcap_gputexture(&sl->matcap_diffuse);
     }
-    else {
-      sl->equirect_radiance_gputexture = GPU_texture_create_2d(
-          ibuf->x, ibuf->y, GPU_RGBA16F, ibuf->rect_float, error);
-      GPUTexture *tex = sl->equirect_radiance_gputexture;
-      GPU_texture_bind(tex, 0);
-      GPU_texture_filter_mode(tex, true);
-      GPU_texture_wrap_mode(tex, true);
-      GPU_texture_unbind(tex);
+  }
+  sl->flag |= STUDIOLIGHT_MATCAP_DIFFUSE_GPUTEXTURE;
+}
+static void studiolight_create_matcap_specular_gputexture(StudioLight *sl)
+{
+  if (sl->flag & STUDIOLIGHT_EXTERNAL_FILE) {
+    if (sl->flag & STUDIOLIGHT_TYPE_MATCAP) {
+      BKE_studiolight_ensure_flag(sl, STUDIOLIGHT_EXTERNAL_IMAGE_LOADED);
+      if (sl->matcap_specular.ibuf) {
+        studiolight_create_matcap_gputexture(&sl->matcap_specular);
+      }
     }
   }
-  sl->flag |= STUDIOLIGHT_EQUIRECT_RADIANCE_GPUTEXTURE;
+  sl->flag |= STUDIOLIGHT_MATCAP_SPECULAR_GPUTEXTURE;
 }
 
 static void studiolight_create_equirect_irradiance_gputexture(StudioLight *sl)
 {
   if (sl->flag & STUDIOLIGHT_EXTERNAL_FILE) {
-    char error[256];
     BKE_studiolight_ensure_flag(sl, STUDIOLIGHT_EQUIRECT_IRRADIANCE_IMAGE_CALCULATED);
     ImBuf *ibuf = sl->equirect_irradiance_buffer;
     sl->equirect_irradiance_gputexture = GPU_texture_create_2d(
-        ibuf->x, ibuf->y, GPU_RGBA16F, ibuf->rect_float, error);
+        ibuf->x, ibuf->y, GPU_RGBA16F, ibuf->rect_float, NULL);
     GPUTexture *tex = sl->equirect_irradiance_gputexture;
     GPU_texture_bind(tex, 0);
     GPU_texture_filter_mode(tex, true);
@@ -457,32 +624,32 @@ static void studiolight_calculate_radiance_cubemap_buffers(StudioLight *sl)
       /* front */
       studiolight_calculate_radiance_buffer(ibuf, colbuf, 0, 2, 1, 1, -1, 1);
       sl->radiance_cubemap_buffers[STUDIOLIGHT_Y_POS] = IMB_allocFromBuffer(
-          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE);
+          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, 4);
 
       /* back */
       studiolight_calculate_radiance_buffer(ibuf, colbuf, 0, 2, 1, 1, 1, -1);
       sl->radiance_cubemap_buffers[STUDIOLIGHT_Y_NEG] = IMB_allocFromBuffer(
-          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE);
+          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, 4);
 
       /* left */
       studiolight_calculate_radiance_buffer(ibuf, colbuf, 2, 1, 0, 1, -1, 1);
       sl->radiance_cubemap_buffers[STUDIOLIGHT_X_POS] = IMB_allocFromBuffer(
-          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE);
+          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, 4);
 
       /* right */
       studiolight_calculate_radiance_buffer(ibuf, colbuf, 2, 1, 0, -1, -1, -1);
       sl->radiance_cubemap_buffers[STUDIOLIGHT_X_NEG] = IMB_allocFromBuffer(
-          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE);
+          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, 4);
 
       /* top */
       studiolight_calculate_radiance_buffer(ibuf, colbuf, 0, 1, 2, -1, -1, 1);
       sl->radiance_cubemap_buffers[STUDIOLIGHT_Z_NEG] = IMB_allocFromBuffer(
-          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE);
+          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, 4);
 
       /* bottom */
       studiolight_calculate_radiance_buffer(ibuf, colbuf, 0, 1, 2, 1, -1, -1);
       sl->radiance_cubemap_buffers[STUDIOLIGHT_Z_POS] = IMB_allocFromBuffer(
-          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE);
+          NULL, colbuf, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, STUDIOLIGHT_RADIANCE_CUBEMAP_SIZE, 4);
 
 #if 0
       IMB_saveiff(sl->radiance_cubemap_buffers[STUDIOLIGHT_X_POS],
@@ -603,7 +770,8 @@ static void studiolight_spherical_harmonics_calculate_coefficients(StudioLight *
 /* Take monochrome SH as input */
 static float studiolight_spherical_harmonics_lambda_get(float *sh, float max_laplacian)
 {
-  /* From Peter-Pike Sloan's Stupid SH Tricks http://www.ppsloan.org/publications/StupidSH36.pdf */
+  /* From Peter-Pike Sloan's Stupid SH Tricks http://www.ppsloan.org/publications/StupidSH36.pdf
+   */
   float table_l[STUDIOLIGHT_SH_BANDS];
   float table_b[STUDIOLIGHT_SH_BANDS];
 
@@ -720,7 +888,6 @@ BLI_INLINE void studiolight_spherical_harmonics_eval(StudioLight *sl,
   }
   return;
 #else
-
   /* L0 */
   mul_v3_v3fl(color, sl->spherical_harmonics_coefs[0], 0.282095f);
 #  if STUDIOLIGHT_SH_BANDS > 1 /* L1 */
@@ -1045,7 +1212,8 @@ static void studiolight_calculate_irradiance_equirect_image(StudioLight *sl)
     sl->equirect_irradiance_buffer = IMB_allocFromBuffer(NULL,
                                                          colbuf,
                                                          STUDIOLIGHT_IRRADIANCE_EQUIRECT_WIDTH,
-                                                         STUDIOLIGHT_IRRADIANCE_EQUIRECT_HEIGHT);
+                                                         STUDIOLIGHT_IRRADIANCE_EQUIRECT_HEIGHT,
+                                                         4);
     MEM_freeN(colbuf);
 
 #ifdef STUDIOLIGHT_IRRADIANCE_METHOD_RADIANCE
@@ -1196,7 +1364,8 @@ static void studiolight_matcap_preview(uint *icon_buffer, StudioLight *sl, bool
 {
   BKE_studiolight_ensure_flag(sl, STUDIOLIGHT_EXTERNAL_IMAGE_LOADED);
 
-  ImBuf *ibuf = sl->equirect_radiance_buffer;
+  ImBuf *diffuse_buffer = sl->matcap_diffuse.ibuf;
+  ImBuf *specular_buffer = sl->matcap_specular.ibuf;
 
   ITER_PIXELS (uint, icon_buffer, 1, STUDIOLIGHT_ICON_SIZE, STUDIOLIGHT_ICON_SIZE) {
     float dy = RESCALE_COORD(y);
@@ -1206,7 +1375,15 @@ static void studiolight_matcap_preview(uint *icon_buffer, StudioLight *sl, bool
     }
 
     float color[4];
-    nearest_interpolation_color(ibuf, NULL, color, dx * ibuf->x - 1.0f, dy * ibuf->y - 1.0f);
+    float u = dx * diffuse_buffer->x - 1.0f;
+    float v = dy * diffuse_buffer->y - 1.0f;
+    nearest_interpolation_color(diffuse_buffer, NULL, color, u, v);
+
+    if (specular_buffer) {
+      float specular[4];
+      nearest_interpolation_color(specular_buffer, NULL, specular, u, v);
+      add_v3_v3(color, specular);
+    }
 
     uint alphamask = alpha_circle_mask(dx, dy, 0.5f - texel_size[0], 0.5f);
 
@@ -1303,9 +1480,9 @@ void BKE_studiolight_default(SolidLight lights[4], float light_ambient[4])
 void BKE_studiolight_init(void)
 {
   /* Add default studio light */
-  StudioLight *sl = studiolight_create(STUDIOLIGHT_INTERNAL |
-                                       STUDIOLIGHT_SPHERICAL_HARMONICS_COEFFICIENTS_CALCULATED |
-                                       STUDIOLIGHT_TYPE_STUDIO);
+  StudioLight *sl = studiolight_create(
+      STUDIOLIGHT_INTERNAL | STUDIOLIGHT_SPHERICAL_HARMONICS_COEFFICIENTS_CALCULATED |
+      STUDIOLIGHT_TYPE_STUDIO | STUDIOLIGHT_SPECULAR_HIGHLIGHT_PASS);
   BLI_strncpy(sl->name, "Default", FILE_MAXFILE);
 
   BLI_addtail(&studiolights, sl);
@@ -1317,7 +1494,8 @@ void BKE_studiolight_init(void)
   if (!BKE_appdir_app_is_portable_install()) {
     studiolight_add_files_from_datafolder(BLENDER_USER_DATAFILES,
                                           STUDIOLIGHT_LIGHTS_FOLDER,
-                                          STUDIOLIGHT_TYPE_STUDIO | STUDIOLIGHT_USER_DEFINED);
+                                          STUDIOLIGHT_TYPE_STUDIO | STUDIOLIGHT_USER_DEFINED |
+                                              STUDIOLIGHT_SPECULAR_HIGHLIGHT_PASS);
     studiolight_add_files_from_datafolder(BLENDER_USER_DATAFILES,
                                           STUDIOLIGHT_WORLD_FOLDER,
                                           STUDIOLIGHT_TYPE_WORLD | STUDIOLIGHT_USER_DEFINED);
@@ -1325,8 +1503,10 @@ void BKE_studiolight_init(void)
                                           STUDIOLIGHT_MATCAP_FOLDER,
                                           STUDIOLIGHT_TYPE_MATCAP | STUDIOLIGHT_USER_DEFINED);
   }
-  studiolight_add_files_from_datafolder(
-      BLENDER_SYSTEM_DATAFILES, STUDIOLIGHT_LIGHTS_FOLDER, STUDIOLIGHT_TYPE_STUDIO);
+  studiolight_add_files_from_datafolder(BLENDER_SYSTEM_DATAFILES,
+                                        STUDIOLIGHT_LIGHTS_FOLDER,
+                                        STUDIOLIGHT_TYPE_STUDIO |
+                                            STUDIOLIGHT_SPECULAR_HIGHLIGHT_PASS);
   studiolight_add_files_from_datafolder(
       BLENDER_SYSTEM_DATAFILES, STUDIOLIGHT_WORLD_FOLDER, STUDIOLIGHT_TYPE_WORLD);
   studiolight_add_files_from_datafolder(
@@ -1456,6 +1636,12 @@ void BKE_studiolight_ensure_flag(StudioLight *sl, int flag)
       studiolight_calculate_irradiance_equirect_image(sl);
     }
   }
+  if ((flag & STUDIOLIGHT_MATCAP_DIFFUSE_GPUTEXTURE)) {
+    studiolight_create_matcap_diffuse_gputexture(sl);
+  }
+  if ((flag & STUDIOLIGHT_MATCAP_SPECULAR_GPUTEXTURE)) {
+    studiolight_create_matcap_specular_gputexture(sl);
+  }
 }
 
 /*
diff --git a/source/blender/blenkernel/intern/subdiv_mesh.c b/source/blender/blenkernel/intern/subdiv_mesh.c
index 1ff9140681f..9d713d1937d 100644
--- a/source/blender/blenkernel/intern/subdiv_mesh.c
+++ b/source/blender/blenkernel/intern/subdiv_mesh.c
@@ -498,9 +498,14 @@ static bool subdiv_mesh_topology_info(const SubdivForeachContext *foreach_contex
                                       const int num_loops,
                                       const int num_polygons)
 {
+  /* Multires grid data will be applied or become invalid after subdivision,
+   * so don't try to preserve it and use memory. */
+  CustomData_MeshMasks mask = CD_MASK_EVERYTHING;
+  mask.lmask &= ~CD_MASK_MULTIRES_GRIDS;
+
   SubdivMeshContext *subdiv_context = foreach_context->user_data;
-  subdiv_context->subdiv_mesh = BKE_mesh_new_nomain_from_template(
-      subdiv_context->coarse_mesh, num_vertices, num_edges, 0, num_loops, num_polygons);
+  subdiv_context->subdiv_mesh = BKE_mesh_new_nomain_from_template_ex(
+      subdiv_context->coarse_mesh, num_vertices, num_edges, 0, num_loops, num_polygons, mask);
   subdiv_mesh_ctx_cache_custom_data_layers(subdiv_context);
   subdiv_mesh_prepare_accumulator(subdiv_context, num_vertices);
   return true;
diff --git a/source/blender/blenkernel/intern/subsurf_ccg.c b/source/blender/blenkernel/intern/subsurf_ccg.c
index 3f57ae169ff..77dc438cd04 100644
--- a/source/blender/blenkernel/intern/subsurf_ccg.c
+++ b/source/blender/blenkernel/intern/subsurf_ccg.c
@@ -83,8 +83,6 @@ static CCGDerivedMesh *getCCGDerivedMesh(CCGSubSurf *ss,
                                          int useSubsurfUv,
                                          DerivedMesh *dm,
                                          bool use_gpu_backend);
-static int ccgDM_use_grid_pbvh(CCGDerivedMesh *ccgdm);
-
 ///
 
 static void *arena_alloc(CCGAllocatorHDL a, int numBytes)
@@ -1580,221 +1578,6 @@ static void ccgDM_copyFinalPolyArray(DerivedMesh *dm, MPoly *mpoly)
   }
 }
 
-static void ccgdm_getVertCos(DerivedMesh *dm, float (*cos)[3])
-{
-  CCGDerivedMesh *ccgdm = (CCGDerivedMesh *)dm;
-  CCGSubSurf *ss = ccgdm->ss;
-  int edgeSize = ccgSubSurf_getEdgeSize(ss);
-  int gridSize = ccgSubSurf_getGridSize(ss);
-  int i;
-  CCGVertIterator vi;
-  CCGEdgeIterator ei;
-  CCGFaceIterator fi;
-  CCGFace **faceMap2;
-  CCGEdge **edgeMap2;
-  CCGVert **vertMap2;
-  int index, totvert, totedge, totface;
-
-  totvert = ccgSubSurf_getNumVerts(ss);
-  vertMap2 = MEM_mallocN(totvert * sizeof(*vertMap2), "vertmap");
-  for (ccgSubSurf_initVertIterator(ss, &vi); !ccgVertIterator_isStopped(&vi);
-       ccgVertIterator_next(&vi)) {
-    CCGVert *v = ccgVertIterator_getCurrent(&vi);
-
-    vertMap2[POINTER_AS_INT(ccgSubSurf_getVertVertHandle(v))] = v;
-  }
-
-  totedge = ccgSubSurf_getNumEdges(ss);
-  edgeMap2 = MEM_mallocN(totedge * sizeof(*edgeMap2), "edgemap");
-  for (ccgSubSurf_initEdgeIterator(ss, &ei), i = 0; !ccgEdgeIterator_isStopped(&ei);
-       i++, ccgEdgeIterator_next(&ei)) {
-    CCGEdge *e = ccgEdgeIterator_getCurrent(&ei);
-
-    edgeMap2[POINTER_AS_INT(ccgSubSurf_getEdgeEdgeHandle(e))] = e;
-  }
-
-  totface = ccgSubSurf_getNumFaces(ss);
-  faceMap2 = MEM_mallocN(totface * sizeof(*faceMap2), "facemap");
-  for (ccgSubSurf_initFaceIterator(ss, &fi); !ccgFaceIterator_isStopped(&fi);
-       ccgFaceIterator_next(&fi)) {
-    CCGFace *f = ccgFaceIterator_getCurrent(&fi);
-
-    faceMap2[POINTER_AS_INT(ccgSubSurf_getFaceFaceHandle(f))] = f;
-  }
-
-  i = 0;
-  for (index = 0; index < totface; index++) {
-    CCGFace *f = faceMap2[index];
-    int x, y, S, numVerts = ccgSubSurf_getFaceNumVerts(f);
-
-    copy_v3_v3(cos[i++], ccgSubSurf_getFaceCenterData(f));
-
-    for (S = 0; S < numVerts; S++) {
-      for (x = 1; x < gridSize - 1; x++) {
-        copy_v3_v3(cos[i++], ccgSubSurf_getFaceGridEdgeData(ss, f, S, x));
-      }
-    }
-
-    for (S = 0; S < numVerts; S++) {
-      for (y = 1; y < gridSize - 1; y++) {
-        for (x = 1; x < gridSize - 1; x++) {
-          copy_v3_v3(cos[i++], ccgSubSurf_getFaceGridData(ss, f, S, x, y));
-        }
-      }
-    }
-  }
-
-  for (index = 0; index < totedge; index++) {
-    CCGEdge *e = edgeMap2[index];
-    int x;
-
-    for (x = 1; x < edgeSize - 1; x++) {
-      copy_v3_v3(cos[i++], ccgSubSurf_getEdgeData(ss, e, x));
-    }
-  }
-
-  for (index = 0; index < totvert; index++) {
-    CCGVert *v = vertMap2[index];
-    copy_v3_v3(cos[i++], ccgSubSurf_getVertData(ss, v));
-  }
-
-  MEM_freeN(vertMap2);
-  MEM_freeN(edgeMap2);
-  MEM_freeN(faceMap2);
-}
-
-static void ccgDM_foreachMappedVert(DerivedMesh *dm,
-                                    void (*func)(void *userData,
-                                                 int index,
-                                                 const float co[3],
-                                                 const float no_f[3],
-                                                 const short no_s[3]),
-                                    void *userData,
-                                    DMForeachFlag flag)
-{
-  CCGDerivedMesh *ccgdm = (CCGDerivedMesh *)dm;
-  CCGVertIterator vi;
-  CCGKey key;
-  CCG_key_top_level(&key, ccgdm->ss);
-
-  for (ccgSubSurf_initVertIterator(ccgdm->ss, &vi); !ccgVertIterator_isStopped(&vi);
-       ccgVertIterator_next(&vi)) {
-    CCGVert *v = ccgVertIterator_getCurrent(&vi);
-    const int index = ccgDM_getVertMapIndex(ccgdm->ss, v);
-
-    if (index != -1) {
-      CCGElem *vd = ccgSubSurf_getVertData(ccgdm->ss, v);
-      const float *no = (flag & DM_FOREACH_USE_NORMAL) ? CCG_elem_no(&key, vd) : NULL;
-      func(userData, index, CCG_elem_co(&key, vd), no, NULL);
-    }
-  }
-}
-
-static void ccgDM_foreachMappedEdge(
-    DerivedMesh *dm,
-    void (*func)(void *userData, int index, const float v0co[3], const float v1co[3]),
-    void *userData)
-{
-  CCGDerivedMesh *ccgdm = (CCGDerivedMesh *)dm;
-  CCGSubSurf *ss = ccgdm->ss;
-  CCGEdgeIterator ei;
-  CCGKey key;
-  int i, edgeSize = ccgSubSurf_getEdgeSize(ss);
-
-  CCG_key_top_level(&key, ss);
-
-  for (ccgSubSurf_initEdgeIterator(ss, &ei); !ccgEdgeIterator_isStopped(&ei);
-       ccgEdgeIterator_next(&ei)) {
-    CCGEdge *e = ccgEdgeIterator_getCurrent(&ei);
-    const int index = ccgDM_getEdgeMapIndex(ss, e);
-
-    if (index != -1) {
-      CCGElem *edgeData = ccgSubSurf_getEdgeDataArray(ss, e);
-      for (i = 0; i < edgeSize - 1; i++) {
-        func(userData,
-             index,
-             CCG_elem_offset_co(&key, edgeData, i),
-             CCG_elem_offset_co(&key, edgeData, i + 1));
-      }
-    }
-  }
-}
-
-static void ccgDM_foreachMappedLoop(DerivedMesh *dm,
-                                    void (*func)(void *userData,
-                                                 int vertex_index,
-                                                 int face_index,
-                                                 const float co[3],
-                                                 const float no[3]),
-                                    void *userData,
-                                    DMForeachFlag flag)
-{
-  /* We can't use dm->getLoopDataLayout(dm) here, we want to always access dm->loopData,
-   * EditDerivedBMesh would return loop data from bmesh itself. */
-  const float(*lnors)[3] = (flag & DM_FOREACH_USE_NORMAL) ? DM_get_loop_data_layer(dm, CD_NORMAL) :
-                                                            NULL;
-
-  MVert *mv = dm->getVertArray(dm);
-  MLoop *ml = dm->getLoopArray(dm);
-  MPoly *mp = dm->getPolyArray(dm);
-  const int *v_index = dm->getVertDataArray(dm, CD_ORIGINDEX);
-  const int *f_index = dm->getPolyDataArray(dm, CD_ORIGINDEX);
-  int p_idx, i;
-
-  for (p_idx = 0; p_idx < dm->numPolyData; ++p_idx, ++mp) {
-    for (i = 0; i < mp->totloop; ++i, ++ml) {
-      const int v_idx = v_index ? v_index[ml->v] : ml->v;
-      const int f_idx = f_index ? f_index[p_idx] : p_idx;
-      const float *no = lnors ? *lnors++ : NULL;
-      if (!ELEM(ORIGINDEX_NONE, v_idx, f_idx)) {
-        func(userData, v_idx, f_idx, mv[ml->v].co, no);
-      }
-    }
-  }
-}
-
-static void UNUSED_FUNCTION(ccgdm_pbvh_update)(CCGDerivedMesh *ccgdm)
-{
-  if (ccgdm->pbvh && ccgDM_use_grid_pbvh(ccgdm)) {
-    CCGFace **faces;
-    int totface;
-
-    BKE_pbvh_get_grid_updates(ccgdm->pbvh, 1, (void ***)&faces, &totface);
-    if (totface) {
-      ccgSubSurf_updateFromFaces(ccgdm->ss, 0, faces, totface);
-      ccgSubSurf_updateNormals(ccgdm->ss, faces, totface);
-      MEM_freeN(faces);
-    }
-  }
-}
-
-static void ccgDM_foreachMappedFaceCenter(
-    DerivedMesh *dm,
-    void (*func)(void *userData, int index, const float co[3], const float no[3]),
-    void *userData,
-    DMForeachFlag flag)
-{
-  CCGDerivedMesh *ccgdm = (CCGDerivedMesh *)dm;
-  CCGSubSurf *ss = ccgdm->ss;
-  CCGKey key;
-  CCGFaceIterator fi;
-
-  CCG_key_top_level(&key, ss);
-
-  for (ccgSubSurf_initFaceIterator(ss, &fi); !ccgFaceIterator_isStopped(&fi);
-       ccgFaceIterator_next(&fi)) {
-    CCGFace *f = ccgFaceIterator_getCurrent(&fi);
-    const int index = ccgDM_getFaceMapIndex(ss, f);
-
-    if (index != -1) {
-      /* Face center data normal isn't updated atm. */
-      CCGElem *vd = ccgSubSurf_getFaceGridData(ss, f, 0, 0, 0);
-      const float *no = (flag & DM_FOREACH_USE_NORMAL) ? CCG_elem_no(&key, vd) : NULL;
-      func(userData, index, CCG_elem_co(&key, vd), no);
-    }
-  }
-}
-
 static void ccgDM_release(DerivedMesh *dm)
 {
   CCGDerivedMesh *ccgdm = (CCGDerivedMesh *)dm;
@@ -2222,174 +2005,6 @@ static const MeshElemMap *ccgDM_getPolyMap(Object *ob, DerivedMesh *dm)
   return ccgdm->pmap;
 }
 
-static int ccgDM_use_grid_pbvh(CCGDerivedMesh *ccgdm)
-{
-  MultiresModifierData *mmd = ccgdm->multires.mmd;
-
-  /* both of multires and subsurf modifiers are CCG, but
-   * grids should only be used when sculpting on multires */
-  if (!mmd) {
-    return 0;
-  }
-
-  return 1;
-}
-
-static struct PBVH *ccgDM_getPBVH(Object *ob, DerivedMesh *dm)
-{
-  CCGDerivedMesh *ccgdm = (CCGDerivedMesh *)dm;
-  CCGKey key;
-  int numGrids;
-
-  CCG_key_top_level(&key, ccgdm->ss);
-
-  if (!ob) {
-    ccgdm->pbvh = NULL;
-    return NULL;
-  }
-
-  if (!ob->sculpt) {
-    return NULL;
-  }
-
-  bool grid_pbvh = ccgDM_use_grid_pbvh(ccgdm);
-  if ((ob->mode & OB_MODE_SCULPT) == 0) {
-    /* In vwpaint, we may use a grid_pbvh for multires/subsurf, under certain conditions.
-     * More complex cases break 'history' trail back to original vertices,
-     * in that case we fall back to deformed cage only (i.e. original deformed mesh). */
-    VirtualModifierData virtualModifierData;
-    ModifierData *md = modifiers_getVirtualModifierList(ob, &virtualModifierData);
-
-    grid_pbvh = true;
-    bool has_one_ccg_modifier = false;
-    for (; md; md = md->next) {
-      /* We can only accept to use this ccgdm if:
-       *   - it's the only active ccgdm in the stack.
-       *   - there is no topology-modifying modifier in the stack.
-       * Otherwise, there is no way to map back to original geometry from grid-generated PBVH.
-       */
-      const ModifierTypeInfo *mti = modifierType_getInfo(md->type);
-
-      if (!modifier_isEnabled(NULL, md, eModifierMode_Realtime)) {
-        continue;
-      }
-      if (ELEM(mti->type, eModifierTypeType_OnlyDeform, eModifierTypeType_NonGeometrical)) {
-        continue;
-      }
-
-      if (ELEM(md->type, eModifierType_Subsurf, eModifierType_Multires)) {
-        if (has_one_ccg_modifier) {
-          /* We only allow a single active ccg modifier in the stack. */
-          grid_pbvh = false;
-          break;
-        }
-        has_one_ccg_modifier = true;
-        continue;
-      }
-
-      /* Any other non-deforming modifier makes it impossible to use grid pbvh. */
-      grid_pbvh = false;
-      break;
-    }
-  }
-
-  if (ob->sculpt->pbvh) {
-    /* Note that we have to clean up existing pbvh instead of updating it in case it does not
-     * match current grid_pbvh status. */
-    const PBVHType pbvh_type = BKE_pbvh_type(ob->sculpt->pbvh);
-    if (grid_pbvh) {
-      if (pbvh_type == PBVH_GRIDS) {
-        /* pbvh's grids, gridadj and gridfaces points to data inside ccgdm
-         * but this can be freed on ccgdm release, this updates the pointers
-         * when the ccgdm gets remade, the assumption is that the topology
-         * does not change. */
-        ccgdm_create_grids(dm);
-        BKE_pbvh_grids_update(ob->sculpt->pbvh,
-                              ccgdm->gridData,
-                              (void **)ccgdm->gridFaces,
-                              ccgdm->gridFlagMats,
-                              ccgdm->gridHidden);
-      }
-      else {
-        BKE_pbvh_free(ob->sculpt->pbvh);
-        ob->sculpt->pbvh = NULL;
-      }
-    }
-    else if (pbvh_type == PBVH_GRIDS) {
-      BKE_pbvh_free(ob->sculpt->pbvh);
-      ob->sculpt->pbvh = NULL;
-    }
-
-    ccgdm->pbvh = ob->sculpt->pbvh;
-  }
-
-  if (ccgdm->pbvh) {
-    return ccgdm->pbvh;
-  }
-
-  /* No pbvh exists yet, we need to create one. only in case of multires
-   * we build a pbvh over the modified mesh, in other cases the base mesh
-   * is being sculpted, so we build a pbvh from that. */
-  /* Note: vwpaint tries to always build a pbvh over the modified mesh. */
-  if (grid_pbvh) {
-    ccgdm_create_grids(dm);
-
-    numGrids = ccgDM_getNumGrids(dm);
-
-    ob->sculpt->pbvh = ccgdm->pbvh = BKE_pbvh_new();
-    BKE_pbvh_build_grids(ccgdm->pbvh,
-                         ccgdm->gridData,
-                         numGrids,
-                         &key,
-                         (void **)ccgdm->gridFaces,
-                         ccgdm->gridFlagMats,
-                         ccgdm->gridHidden);
-  }
-  else if (ob->type == OB_MESH) {
-    Mesh *me = BKE_object_get_original_mesh(ob);
-    const int looptris_num = poly_to_tri_count(me->totpoly, me->totloop);
-    MLoopTri *looptri;
-
-    looptri = MEM_mallocN(sizeof(*looptri) * looptris_num, __func__);
-
-    BKE_mesh_recalc_looptri(me->mloop, me->mpoly, me->mvert, me->totloop, me->totpoly, looptri);
-
-    ob->sculpt->pbvh = ccgdm->pbvh = BKE_pbvh_new();
-    BKE_pbvh_build_mesh(ccgdm->pbvh,
-                        me->mpoly,
-                        me->mloop,
-                        me->mvert,
-                        me->totvert,
-                        &me->vdata,
-                        &me->ldata,
-                        looptri,
-                        looptris_num);
-
-    if (ob->sculpt->modifiers_active && ob->derivedDeform != NULL) {
-      DerivedMesh *deformdm = ob->derivedDeform;
-      float(*vertCos)[3];
-      int totvert;
-
-      totvert = deformdm->getNumVerts(deformdm);
-      vertCos = MEM_malloc_arrayN(totvert, sizeof(float[3]), "ccgDM_getPBVH vertCos");
-      deformdm->getVertCos(deformdm, vertCos);
-      BKE_pbvh_apply_vertCos(ccgdm->pbvh, vertCos, totvert);
-      MEM_freeN(vertCos);
-    }
-  }
-
-  if (ccgdm->pbvh != NULL) {
-    pbvh_show_mask_set(ccgdm->pbvh, ob->sculpt->show_mask);
-  }
-
-  return ccgdm->pbvh;
-}
-
-static void ccgDM_recalcTessellation(DerivedMesh *UNUSED(dm))
-{
-  /* Nothing to do: CCG handles creating its own tessfaces */
-}
-
 /* WARNING! *MUST* be called in an 'loops_cache_rwlock' protected thread context! */
 static void ccgDM_recalcLoopTri(DerivedMesh *dm)
 {
@@ -2425,12 +2040,6 @@ static void ccgDM_recalcLoopTri(DerivedMesh *dm)
   dm->looptris.array_wip = NULL;
 }
 
-static void ccgDM_calcNormals(DerivedMesh *dm)
-{
-  /* Nothing to do: CCG calculates normals during drawing */
-  dm->dirty &= ~DM_DIRTY_NORMALS;
-}
-
 static void set_default_ccgdm_callbacks(CCGDerivedMesh *ccgdm)
 {
   ccgdm->dm.getMinMax = ccgDM_getMinMax;
@@ -2471,21 +2080,9 @@ static void set_default_ccgdm_callbacks(CCGDerivedMesh *ccgdm)
   ccgdm->dm.getGridFlagMats = ccgDM_getGridFlagMats;
   ccgdm->dm.getGridHidden = ccgDM_getGridHidden;
   ccgdm->dm.getPolyMap = ccgDM_getPolyMap;
-  ccgdm->dm.getPBVH = ccgDM_getPBVH;
 
-  ccgdm->dm.calcNormals = ccgDM_calcNormals;
-  ccgdm->dm.calcLoopNormals = CDDM_calc_loop_normals;
-  ccgdm->dm.calcLoopNormalsSpaceArray = CDDM_calc_loop_normals_spacearr;
-  ccgdm->dm.calcLoopTangents = DM_calc_loop_tangents;
-  ccgdm->dm.recalcTessellation = ccgDM_recalcTessellation;
   ccgdm->dm.recalcLoopTri = ccgDM_recalcLoopTri;
 
-  ccgdm->dm.getVertCos = ccgdm_getVertCos;
-  ccgdm->dm.foreachMappedVert = ccgDM_foreachMappedVert;
-  ccgdm->dm.foreachMappedEdge = ccgDM_foreachMappedEdge;
-  ccgdm->dm.foreachMappedLoop = ccgDM_foreachMappedLoop;
-  ccgdm->dm.foreachMappedFaceCenter = ccgDM_foreachMappedFaceCenter;
-
   ccgdm->dm.release = ccgDM_release;
 }
 
diff --git a/source/blender/blenkernel/intern/texture.c b/source/blender/blenkernel/intern/texture.c
index ad7c5e3f660..5c03b7730aa 100644
--- a/source/blender/blenkernel/intern/texture.c
+++ b/source/blender/blenkernel/intern/texture.c
@@ -423,6 +423,11 @@ MTex *BKE_texture_mtex_add_id(ID *id, int slot)
  */
 void BKE_texture_copy_data(Main *bmain, Tex *tex_dst, const Tex *tex_src, const int flag)
 {
+  /* We never handle usercount here for own data. */
+  const int flag_subdata = flag | LIB_ID_CREATE_NO_USER_REFCOUNT;
+  /* We always need allocation of our private ID data. */
+  const int flag_private_id_data = flag_subdata & ~LIB_ID_CREATE_NO_ALLOCATE;
+
   if (!BKE_texture_is_image_user(tex_src)) {
     tex_dst->ima = NULL;
   }
@@ -434,9 +439,8 @@ void BKE_texture_copy_data(Main *bmain, Tex *tex_dst, const Tex *tex_src, const
     if (tex_src->nodetree->execdata) {
       ntreeTexEndExecTree(tex_src->nodetree->execdata);
     }
-    /* Note: nodetree is *not* in bmain, however this specific case is handled at lower level
-     *       (see BKE_libblock_copy_ex()). */
-    BKE_id_copy_ex(bmain, (ID *)tex_src->nodetree, (ID **)&tex_dst->nodetree, flag);
+    BKE_id_copy_ex(
+        bmain, (ID *)tex_src->nodetree, (ID **)&tex_dst->nodetree, flag_private_id_data);
   }
 
   if ((flag & LIB_ID_COPY_NO_PREVIEW) == 0) {
diff --git a/source/blender/blenkernel/intern/tracking_solver.c b/source/blender/blenkernel/intern/tracking_solver.c
index 7e7a839b645..ab741eed410 100644
--- a/source/blender/blenkernel/intern/tracking_solver.c
+++ b/source/blender/blenkernel/intern/tracking_solver.c
@@ -221,7 +221,7 @@ static bool reconstruct_retrieve_libmv_tracks(MovieReconstructContext *context,
        *
        * There's one weak part tho, which is requirement object
        * motion starts at the same frame as camera motion does,
-       * otherwise that;' be a russian roulette whether object is
+       * otherwise that;' be a Russian roulette whether object is
        * aligned correct or not.
        */
       if (!origin_set) {
diff --git a/source/blender/blenkernel/intern/tracking_stabilize.c b/source/blender/blenkernel/intern/tracking_stabilize.c
index 03229c654fb..7da36ba9af9 100644
--- a/source/blender/blenkernel/intern/tracking_stabilize.c
+++ b/source/blender/blenkernel/intern/tracking_stabilize.c
@@ -1129,8 +1129,8 @@ static void stabilization_data_to_mat4(float pixel_aspect,
 }
 
 /* Calculate scale factor necessary to eliminate black image areas
- * caused by the compensating movements of the stabilizator.
- * This function visits every frame where stabilisation data is
+ * caused by the compensating movements of the stabilizer.
+ * This function visits every frame where stabilization data is
  * available and determines the factor for this frame. The overall
  * largest factor found is returned as result.
  *
diff --git a/source/blender/blenkernel/intern/undo_system.c b/source/blender/blenkernel/intern/undo_system.c
index 07cf5205cab..36b950fb8f9 100644
--- a/source/blender/blenkernel/intern/undo_system.c
+++ b/source/blender/blenkernel/intern/undo_system.c
@@ -503,7 +503,7 @@ bool BKE_undosys_step_push_with_type(UndoStack *ustack,
   /* Might not be final place for this to be called - probably only want to call it from some
    * undo handlers, not all of them? */
   if (BKE_override_library_is_enabled()) {
-    BKE_main_override_library_operations_create(G.main, false);
+    BKE_main_override_library_operations_create(CTX_data_main(C), false);
   }
 
   /* Remove all undos after (also when 'ustack->step_active == NULL'). */
diff --git a/source/blender/blenkernel/intern/unit.c b/source/blender/blenkernel/intern/unit.c
index e30ea687b13..ac1b5e4ab0b 100644
--- a/source/blender/blenkernel/intern/unit.c
+++ b/source/blender/blenkernel/intern/unit.c
@@ -501,12 +501,12 @@ static PreferredUnits preferred_units_from_UnitSettings(const UnitSettings *sett
   return units;
 }
 
-static size_t unit_as_string_splitted(char *str,
-                                      int len_max,
-                                      double value,
-                                      int prec,
-                                      const bUnitCollection *usys,
-                                      const bUnitDef *main_unit)
+static size_t unit_as_string_split_pair(char *str,
+                                        int len_max,
+                                        double value,
+                                        int prec,
+                                        const bUnitCollection *usys,
+                                        const bUnitDef *main_unit)
 {
   const bUnitDef *unit_a, *unit_b;
   double value_a, value_b;
@@ -602,7 +602,7 @@ static size_t unit_as_string_main(char *str,
   }
 
   if (split && unit_should_be_split(type)) {
-    int length = unit_as_string_splitted(str, len_max, value, prec, usys, main_unit);
+    int length = unit_as_string_split_pair(str, len_max, value, prec, usys, main_unit);
     /* failed when length is negative, fallback to no split */
     if (length >= 0) {
       return length;
diff --git a/source/blender/blenkernel/intern/world.c b/source/blender/blenkernel/intern/world.c
index 109d615ae83..31fc87d7021 100644
--- a/source/blender/blenkernel/intern/world.c
+++ b/source/blender/blenkernel/intern/world.c
@@ -106,10 +106,14 @@ World *BKE_world_add(Main *bmain, const char *name)
  */
 void BKE_world_copy_data(Main *bmain, World *wrld_dst, const World *wrld_src, const int flag)
 {
+  /* We never handle usercount here for own data. */
+  const int flag_subdata = flag | LIB_ID_CREATE_NO_USER_REFCOUNT;
+  /* We always need allocation of our private ID data. */
+  const int flag_private_id_data = flag_subdata & ~LIB_ID_CREATE_NO_ALLOCATE;
+
   if (wrld_src->nodetree) {
-    /* Note: nodetree is *not* in bmain, however this specific case is handled at lower level
-     *       (see BKE_libblock_copy_ex()). */
-    BKE_id_copy_ex(bmain, (ID *)wrld_src->nodetree, (ID **)&wrld_dst->nodetree, flag);
+    BKE_id_copy_ex(
+        bmain, (ID *)wrld_src->nodetree, (ID **)&wrld_dst->nodetree, flag_private_id_data);
   }
 
   BLI_listbase_clear(&wrld_dst->gpumaterial);
diff --git a/source/blender/blenkernel/nla_private.h b/source/blender/blenkernel/nla_private.h
index 1705cd4effc..ed653182a3e 100644
--- a/source/blender/blenkernel/nla_private.h
+++ b/source/blender/blenkernel/nla_private.h
@@ -24,8 +24,6 @@
 #ifndef __NLA_PRIVATE_H__
 #define __NLA_PRIVATE_H__
 
-struct Depsgraph;
-
 #include "RNA_types.h"
 #include "BLI_bitmap.h"
 #include "BLI_ghash.h"
diff --git a/source/blender/blenlib/BLI_array.h b/source/blender/blenlib/BLI_array.h
index a61eff7329e..d687838a896 100644
--- a/source/blender/blenlib/BLI_array.h
+++ b/source/blender/blenlib/BLI_array.h
@@ -83,9 +83,9 @@ void _bli_array_grow_func(void **arr_p,
           ((void *)(_##arr##_static) != \
            NULL) && /* don't add _##arr##_len below because it must be zero */ \
           (_bli_array_totalsize_static(arr) >= \
-           _##arr##_len + (num))) ? /* we have an empty array and a static var big enough */ \
+           (size_t)(_##arr##_len + (num)))) ? /* we have an empty array and a static var big enough */ \
              (void)(arr = (void *)_##arr##_static) : /* use existing static array or allocate */ \
-             (LIKELY(_bli_array_totalsize(arr) >= _##arr##_len + (num)) ? \
+             (LIKELY(_bli_array_totalsize(arr) >= (size_t)(_##arr##_len + (num))) ? \
                   (void)0 /* do nothing */ : \
                   _bli_array_grow_func((void **)&(arr), \
                                        _##arr##_static, \
diff --git a/source/blender/blenlib/BLI_dynstr.h b/source/blender/blenlib/BLI_dynstr.h
index 72a6a1b0150..87e58d90d3f 100644
--- a/source/blender/blenlib/BLI_dynstr.h
+++ b/source/blender/blenlib/BLI_dynstr.h
@@ -35,7 +35,6 @@
 #include "BLI_compiler_attrs.h"
 
 struct DynStr;
-struct MemArena;
 
 /** The abstract DynStr type */
 typedef struct DynStr DynStr;
diff --git a/source/blender/blenlib/BLI_fileops.h b/source/blender/blenlib/BLI_fileops.h
index d8daa81b58d..d78f167a8fd 100644
--- a/source/blender/blenlib/BLI_fileops.h
+++ b/source/blender/blenlib/BLI_fileops.h
@@ -99,8 +99,13 @@ void BLI_filelist_entry_size_to_string(const struct stat *st,
 void BLI_filelist_entry_mode_to_string(
     const struct stat *st, const bool compact, char r_mode1[], char r_mode2[], char r_mode3[]);
 void BLI_filelist_entry_owner_to_string(const struct stat *st, const bool compact, char r_owner[]);
-void BLI_filelist_entry_datetime_to_string(
-    const struct stat *st, const int64_t ts, const bool compact, char r_time[], char r_date[]);
+void BLI_filelist_entry_datetime_to_string(const struct stat *st,
+                                           const int64_t ts,
+                                           const bool compact,
+                                           char r_time[],
+                                           char r_date[],
+                                           bool *r_is_today,
+                                           bool *r_is_yesterday);
 
 /* Files */
 
diff --git a/source/blender/blenlib/BLI_kdopbvh.h b/source/blender/blenlib/BLI_kdopbvh.h
index a348694c4da..abfd561200c 100644
--- a/source/blender/blenlib/BLI_kdopbvh.h
+++ b/source/blender/blenlib/BLI_kdopbvh.h
@@ -177,6 +177,12 @@ int BLI_bvhtree_find_nearest(BVHTree *tree,
                              BVHTree_NearestPointCallback callback,
                              void *userdata);
 
+int BLI_bvhtree_find_nearest_first(BVHTree *tree,
+                                   const float co[3],
+                                   const float dist_sq,
+                                   BVHTree_NearestPointCallback callback,
+                                   void *userdata);
+
 int BLI_bvhtree_ray_cast_ex(BVHTree *tree,
                             const float co[3],
                             const float dir[3],
diff --git a/source/blender/blenlib/BLI_listbase.h b/source/blender/blenlib/BLI_listbase.h
index c235a72e7e4..b8b62dd3451 100644
--- a/source/blender/blenlib/BLI_listbase.h
+++ b/source/blender/blenlib/BLI_listbase.h
@@ -171,12 +171,21 @@ struct LinkData *BLI_genericNodeN(void *data);
 #define LISTBASE_FOREACH(type, var, list) \
   for (type var = (type)((list)->first); var != NULL; var = (type)(((Link *)(var))->next))
 
+#define LISTBASE_FOREACH_BACKWARD(type, var, list) \
+  for (type var = (type)((list)->last); var != NULL; var = (type)(((Link *)(var))->prev))
+
 /** A version of #LISTBASE_FOREACH that supports removing the item we're looping over. */
 #define LISTBASE_FOREACH_MUTABLE(type, var, list) \
   for (type var = (type)((list)->first), *var##_iter_next; \
        ((var != NULL) ? ((void)(var##_iter_next = (type)(((Link *)(var))->next)), 1) : 0); \
        var = var##_iter_next)
 
+/** A version of #LISTBASE_FOREACH_BACKWARD that supports removing the item we're looping over. */
+#define LISTBASE_FOREACH_BACKWARD_MUTABLE(type, var, list) \
+  for (type var = (type)((list)->last), *var##_iter_prev; \
+       ((var != NULL) ? ((void)(var##_iter_prev = (type)(((Link *)(var))->prev)), 1) : 0); \
+       var = var##_iter_prev)
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/source/blender/blenlib/BLI_math_base.h b/source/blender/blenlib/BLI_math_base.h
index deb29605552..7b770f2dd3e 100644
--- a/source/blender/blenlib/BLI_math_base.h
+++ b/source/blender/blenlib/BLI_math_base.h
@@ -121,6 +121,9 @@ MINLINE float max_fff(float a, float b, float c);
 MINLINE float min_ffff(float a, float b, float c, float d);
 MINLINE float max_ffff(float a, float b, float c, float d);
 
+MINLINE double min_dd(double a, double b);
+MINLINE double max_dd(double a, double b);
+
 MINLINE int min_ii(int a, int b);
 MINLINE int max_ii(int a, int b);
 MINLINE int min_iii(int a, int b, int c);
diff --git a/source/blender/blenlib/BLI_math_geom.h b/source/blender/blenlib/BLI_math_geom.h
index 39b1b96d009..caecc0aebc2 100644
--- a/source/blender/blenlib/BLI_math_geom.h
+++ b/source/blender/blenlib/BLI_math_geom.h
@@ -188,6 +188,10 @@ float dist_squared_to_projected_aabb_simple(const float projmat[4][4],
                                             const float bbmax[3]);
 
 float closest_to_line_v2(float r_close[2], const float p[2], const float l1[2], const float l2[2]);
+double closest_to_line_v2_db(double r_close[2],
+                             const double p[2],
+                             const double l1[2],
+                             const double l2[2]);
 float closest_to_line_v3(float r_close[3], const float p[3], const float l1[3], const float l2[3]);
 void closest_to_line_segment_v2(float r_close[2],
                                 const float p[2],
@@ -267,7 +271,12 @@ bool isect_seg_seg_v2_simple(const float v1[2],
                              const float v2[2],
                              const float v3[2],
                              const float v4[2]);
-
+int isect_seg_seg_v2_lambda_mu_db(const double v1[2],
+                                  const double v2[2],
+                                  const double v3[2],
+                                  const double v4[2],
+                                  double *r_lambda,
+                                  double *r_mu);
 int isect_line_sphere_v3(const float l1[3],
                          const float l2[3],
                          const float sp[3],
diff --git a/source/blender/blenlib/BLI_math_matrix.h b/source/blender/blenlib/BLI_math_matrix.h
index 52d976daa2d..f5d87667b73 100644
--- a/source/blender/blenlib/BLI_math_matrix.h
+++ b/source/blender/blenlib/BLI_math_matrix.h
@@ -41,6 +41,7 @@ void zero_m4(float R[4][4]);
 void unit_m2(float R[2][2]);
 void unit_m3(float R[3][3]);
 void unit_m4(float R[4][4]);
+void unit_m4_db(double m[4][4]);
 
 void copy_m2_m2(float R[2][2], const float A[2][2]);
 void copy_m3_m3(float R[3][3], const float A[3][3]);
@@ -48,9 +49,14 @@ void copy_m4_m4(float R[4][4], const float A[4][4]);
 void copy_m3_m4(float R[3][3], const float A[4][4]);
 void copy_m4_m3(float R[4][4], const float A[3][3]);
 
+void copy_m4_m4_db(double m1[4][4], const double m2[4][4]);
+
 /* double->float */
 void copy_m3_m3d(float R[3][3], const double A[3][3]);
 
+/* float->double */
+void copy_m4d_m4(double R[4][4], const float A[4][4]);
+
 void swap_m3m3(float A[3][3], float B[3][3]);
 void swap_m4m4(float A[4][4], float B[4][4]);
 
@@ -82,6 +88,8 @@ void mul_m3_m3m3_uniq(float R[3][3], const float A[3][3], const float B[3][3]);
 void mul_m3_m3_pre(float R[3][3], const float A[3][3]);
 void mul_m3_m3_post(float R[3][3], const float B[3][3]);
 void mul_m4_m4m4_uniq(float R[4][4], const float A[4][4], const float B[4][4]);
+void mul_m4_m4m4_db_uniq(double R[4][4], const double A[4][4], const double B[4][4]);
+void mul_m4db_m4db_m4fl_uniq(double R[4][4], const double A[4][4], const float B[4][4]);
 void mul_m4_m4_pre(float R[4][4], const float A[4][4]);
 void mul_m4_m4_post(float R[4][4], const float B[4][4]);
 
@@ -173,11 +181,14 @@ void _va_mul_m4_series_9(float R[4][4],
 
 void mul_m4_v3(const float M[4][4], float r[3]);
 void mul_v3_m4v3(float r[3], const float M[4][4], const float v[3]);
+void mul_v3_m4v3_db(double r[3], const double mat[4][4], const double vec[3]);
+void mul_v4_m4v3_db(double r[4], const double mat[4][4], const double vec[3]);
 void mul_v2_m4v3(float r[2], const float M[4][4], const float v[3]);
 void mul_v2_m2v2(float r[2], const float M[2][2], const float v[2]);
 void mul_m2v2(const float M[2][2], float v[2]);
 void mul_mat3_m4_v3(const float M[4][4], float r[3]);
 void mul_v3_mat3_m4v3(float r[3], const float M[4][4], const float v[3]);
+void mul_v3_mat3_m4v3_db(double r[3], const double M[4][4], const double v[3]);
 void mul_m4_v4(const float M[4][4], float r[4]);
 void mul_v4_m4v4(float r[4], const float M[4][4], const float v[4]);
 void mul_v4_m4v3(float r[4], const float M[4][4], const float v[3]); /* v has implicit w = 1.0f */
diff --git a/source/blender/blenlib/BLI_math_rotation.h b/source/blender/blenlib/BLI_math_rotation.h
index 7a4ac14970f..1e56b80bcf2 100644
--- a/source/blender/blenlib/BLI_math_rotation.h
+++ b/source/blender/blenlib/BLI_math_rotation.h
@@ -96,6 +96,8 @@ void rotation_between_vecs_to_mat3(float m[3][3], const float v1[3], const float
 void rotation_between_vecs_to_quat(float q[4], const float v1[3], const float v2[3]);
 void rotation_between_quats_to_quat(float q[4], const float q1[4], const float q2[4]);
 
+float quat_split_swing_and_twist(const float q[4], int axis, float r_swing[4], float r_twist[4]);
+
 float angle_normalized_qt(const float q[4]);
 float angle_normalized_qtqt(const float q1[4], const float q2[4]);
 float angle_qt(const float q[4]);
diff --git a/source/blender/blenlib/BLI_math_vector.h b/source/blender/blenlib/BLI_math_vector.h
index 53e686c87a6..31d725af5ad 100644
--- a/source/blender/blenlib/BLI_math_vector.h
+++ b/source/blender/blenlib/BLI_math_vector.h
@@ -103,9 +103,11 @@ MINLINE void add_v2_fl(float r[2], float f);
 MINLINE void add_v3_fl(float r[3], float f);
 MINLINE void add_v4_fl(float r[4], float f);
 MINLINE void add_v2_v2(float r[2], const float a[2]);
+MINLINE void add_v2_v2_db(double r[2], const double a[2]);
 MINLINE void add_v2_v2v2(float r[2], const float a[2], const float b[2]);
 MINLINE void add_v2_v2v2_int(int r[2], const int a[2], const int b[2]);
 MINLINE void add_v3_v3(float r[3], const float a[3]);
+MINLINE void add_v3_v3_db(double r[3], const double a[3]);
 MINLINE void add_v3_v3v3(float r[3], const float a[3], const float b[3]);
 MINLINE void add_v4_v4(float r[4], const float a[4]);
 MINLINE void add_v4_v4v4(float r[4], const float a[4], const float b[4]);
@@ -115,10 +117,12 @@ MINLINE void add_v3fl_v3fl_v3s(float r[3], const float a[3], const short b[3]);
 
 MINLINE void sub_v2_v2(float r[2], const float a[2]);
 MINLINE void sub_v2_v2v2(float r[2], const float a[2], const float b[2]);
+MINLINE void sub_v2_v2v2_db(double r[2], const double a[2], const double b[2]);
 MINLINE void sub_v2_v2v2_int(int r[2], const int a[2], const int b[2]);
 MINLINE void sub_v3_v3(float r[3], const float a[3]);
 MINLINE void sub_v3_v3v3(float r[3], const float a[3], const float b[3]);
 MINLINE void sub_v3_v3v3_int(int r[3], const int a[3], const int b[3]);
+MINLINE void sub_v3_v3v3_db(double r[3], const double a[3], const double b[3]);
 MINLINE void sub_v4_v4(float r[4], const float a[4]);
 MINLINE void sub_v4_v4v4(float r[4], const float a[4], const float b[4]);
 
@@ -127,6 +131,7 @@ MINLINE void sub_v3db_v3fl_v3fl(double r[3], const float a[3], const float b[3])
 MINLINE void mul_v2_fl(float r[2], float f);
 MINLINE void mul_v2_v2fl(float r[2], const float a[2], float f);
 MINLINE void mul_v3_fl(float r[3], float f);
+MINLINE void mul_v3db_db(double r[3], double f);
 MINLINE void mul_v3_v3fl(float r[3], const float a[3], float f);
 MINLINE void mul_v2_v2(float r[2], const float a[2]);
 MINLINE void mul_v2_v2v2(float r[2], const float a[2], const float b[2]);
@@ -181,6 +186,7 @@ MINLINE void abs_v4(float r[4]);
 MINLINE void abs_v4_v4(float r[4], const float a[4]);
 
 MINLINE float dot_v2v2(const float a[2], const float b[2]) ATTR_WARN_UNUSED_RESULT;
+MINLINE double dot_v2v2_db(const double a[2], const double b[2]) ATTR_WARN_UNUSED_RESULT;
 MINLINE float dot_v3v3(const float a[3], const float b[3]) ATTR_WARN_UNUSED_RESULT;
 MINLINE float dot_v3v3v3(const float p[3],
                          const float a[3],
@@ -189,9 +195,12 @@ MINLINE float dot_v4v4(const float a[4], const float b[4]) ATTR_WARN_UNUSED_RESU
 
 MINLINE double dot_v3db_v3fl(const double a[3], const float b[3]) ATTR_WARN_UNUSED_RESULT;
 
+MINLINE double dot_v3v3_db(const double a[3], const double b[3]) ATTR_WARN_UNUSED_RESULT;
+
 MINLINE float cross_v2v2(const float a[2], const float b[2]) ATTR_WARN_UNUSED_RESULT;
 MINLINE void cross_v3_v3v3(float r[3], const float a[3], const float b[3]);
 MINLINE void cross_v3_v3v3_hi_prec(float r[3], const float a[3], const float b[3]);
+MINLINE void cross_v3_v3v3_db(double r[3], const double a[3], const double b[3]);
 
 MINLINE void add_newell_cross_v3_v3v3(float n[3], const float v_prev[3], const float v_curr[3]);
 
@@ -206,8 +215,10 @@ MINLINE int len_manhattan_v2_int(const int v[2]) ATTR_WARN_UNUSED_RESULT;
 MINLINE float len_manhattan_v3(const float v[3]) ATTR_WARN_UNUSED_RESULT;
 MINLINE float len_v2(const float a[2]) ATTR_WARN_UNUSED_RESULT;
 MINLINE float len_v2v2(const float a[2], const float b[2]) ATTR_WARN_UNUSED_RESULT;
+MINLINE double len_v2v2_db(const double a[2], const double b[2]) ATTR_WARN_UNUSED_RESULT;
 MINLINE float len_v2v2_int(const int v1[2], const int v2[2]);
 MINLINE float len_squared_v2v2(const float a[2], const float b[2]) ATTR_WARN_UNUSED_RESULT;
+MINLINE double len_squared_v2v2_db(const double a[2], const double b[2]) ATTR_WARN_UNUSED_RESULT;
 MINLINE float len_squared_v3v3(const float a[3], const float b[3]) ATTR_WARN_UNUSED_RESULT;
 MINLINE float len_squared_v4v4(const float a[4], const float b[4]) ATTR_WARN_UNUSED_RESULT;
 MINLINE float len_manhattan_v2v2(const float a[2], const float b[2]) ATTR_WARN_UNUSED_RESULT;
diff --git a/source/blender/blenlib/BLI_memiter.h b/source/blender/blenlib/BLI_memiter.h
index 6f6316e15e4..9f5be292dd8 100644
--- a/source/blender/blenlib/BLI_memiter.h
+++ b/source/blender/blenlib/BLI_memiter.h
@@ -33,7 +33,6 @@ extern "C" {
 #define BLI_MEMITER_DEFAULT_SIZE (1 << 19)
 
 struct BLI_memiter;
-struct BLI_memiter_chunk;
 
 typedef struct BLI_memiter BLI_memiter;
 
diff --git a/source/blender/blenlib/BLI_path_util.h b/source/blender/blenlib/BLI_path_util.h
index 99e86615e50..75d5cb286ac 100644
--- a/source/blender/blenlib/BLI_path_util.h
+++ b/source/blender/blenlib/BLI_path_util.h
@@ -29,8 +29,6 @@ extern "C" {
 
 #include "BLI_compiler_attrs.h"
 
-struct ListBase;
-
 void BLI_setenv(const char *env, const char *val) ATTR_NONNULL(1);
 void BLI_setenv_if_new(const char *env, const char *val) ATTR_NONNULL(1);
 const char *BLI_getenv(const char *env) ATTR_NONNULL(1);
diff --git a/source/blender/blenlib/BLI_polyfill_2d_beautify.h b/source/blender/blenlib/BLI_polyfill_2d_beautify.h
index f815459fdf5..042cb7e0ea9 100644
--- a/source/blender/blenlib/BLI_polyfill_2d_beautify.h
+++ b/source/blender/blenlib/BLI_polyfill_2d_beautify.h
@@ -36,9 +36,10 @@ float BLI_polyfill_beautify_quad_rotate_calc_ex(const float v1[2],
                                                 const float v2[2],
                                                 const float v3[2],
                                                 const float v4[2],
-                                                const bool lock_degenerate);
+                                                const bool lock_degenerate,
+                                                float *r_area);
 #define BLI_polyfill_beautify_quad_rotate_calc(v1, v2, v3, v4) \
-  BLI_polyfill_beautify_quad_rotate_calc_ex(v1, v2, v3, v4, false)
+  BLI_polyfill_beautify_quad_rotate_calc_ex(v1, v2, v3, v4, false, NULL)
 
 /* avoid realloc's when creating new structures for polyfill ngons */
 #define BLI_POLYFILL_ALLOC_NGON_RESERVE 64
diff --git a/source/blender/blenlib/BLI_string.h b/source/blender/blenlib/BLI_string.h
index 70ffb46e952..cab2e68ca2b 100644
--- a/source/blender/blenlib/BLI_string.h
+++ b/source/blender/blenlib/BLI_string.h
@@ -100,7 +100,7 @@ char *BLI_strncasestr(const char *s, const char *find, size_t len) ATTR_WARN_UNU
 int BLI_strcasecmp(const char *s1, const char *s2) ATTR_WARN_UNUSED_RESULT ATTR_NONNULL();
 int BLI_strncasecmp(const char *s1, const char *s2, size_t len) ATTR_WARN_UNUSED_RESULT
     ATTR_NONNULL();
-int BLI_natstrcmp(const char *s1, const char *s2) ATTR_WARN_UNUSED_RESULT ATTR_NONNULL();
+int BLI_strcasecmp_natural(const char *s1, const char *s2) ATTR_WARN_UNUSED_RESULT ATTR_NONNULL();
 int BLI_strcmp_ignore_pad(const char *str1,
                           const char *str2,
                           const char pad) ATTR_WARN_UNUSED_RESULT ATTR_NONNULL();
diff --git a/source/blender/blenlib/BLI_system.h b/source/blender/blenlib/BLI_system.h
index f4c0399e959..8c0c9ad99bf 100644
--- a/source/blender/blenlib/BLI_system.h
+++ b/source/blender/blenlib/BLI_system.h
@@ -19,11 +19,16 @@
 
 #include <stdio.h>
 
+#ifdef __cplusplus
+extern "C" {
+#endif
+
 /** \file
  * \ingroup bli
  */
 
 int BLI_cpu_support_sse2(void);
+int BLI_cpu_support_sse41(void);
 void BLI_system_backtrace(FILE *fp);
 
 /* Get CPU brand, result is to be MEM_freeN()-ed. */
@@ -52,4 +57,8 @@ int BLI_system_memory_max_in_megabytes_int(void);
 #  define BLI_SYSTEM_PID_H <unistd.h>
 #endif
 
+#ifdef __cplusplus
+}
+#endif
+
 #endif /* __BLI_SYSTEM_H__ */
diff --git a/source/blender/blenlib/intern/BLI_filelist.c b/source/blender/blenlib/intern/BLI_filelist.c
index 91f16ca9b7b..c92505acb54 100644
--- a/source/blender/blenlib/intern/BLI_filelist.c
+++ b/source/blender/blenlib/intern/BLI_filelist.c
@@ -107,7 +107,7 @@ static int bli_compare(struct direntry *entry1, struct direntry *entry2)
     return 1;
   }
 
-  return (BLI_natstrcmp(entry1->relname, entry2->relname));
+  return (BLI_strcasecmp_natural(entry1->relname, entry2->relname));
 }
 
 struct BuildDirCtx {
@@ -261,36 +261,17 @@ unsigned int BLI_filelist_dir_contents(const char *dirname, struct direntry **r_
  */
 void BLI_filelist_entry_size_to_string(const struct stat *st,
                                        const uint64_t sz,
-                                       const bool compact,
+                                       /* Used to change MB -> M, etc. - is that really useful? */
+                                       const bool UNUSED(compact),
                                        char r_size[FILELIST_DIRENTRY_SIZE_LEN])
 {
-  double size;
-  const char *fmt;
-  const char *units[] = {"KiB", "MiB", "GiB", "TiB", NULL};
-  const char *units_compact[] = {"K", "M", "G", "T", NULL};
-  const char *unit = "B";
-
   /*
    * Seems st_size is signed 32-bit value in *nix and Windows.  This
    * will buy us some time until files get bigger than 4GB or until
    * everyone starts using __USE_FILE_OFFSET64 or equivalent.
    */
-  size = (double)(st ? st->st_size : sz);
-
-  if (size > 1024.0) {
-    const char **u;
-    for (u = compact ? units_compact : units, size /= 1024.0; size > 1024.0 && *(u + 1);
-         u++, size /= 1024.0) {
-      /* pass */
-    }
-    fmt = size > 100.0 ? "%.0f %s" : (size > 10.0 ? "%.1f %s" : "%.2f %s");
-    unit = *u;
-  }
-  else {
-    fmt = "%.0f %s";
-  }
-
-  BLI_snprintf(r_size, sizeof(*r_size) * FILELIST_DIRENTRY_SIZE_LEN, fmt, size, unit);
+  double size = (double)(st ? st->st_size : sz);
+  BLI_str_format_byte_unit(r_size, size, true);
 }
 
 /**
@@ -366,14 +347,45 @@ void BLI_filelist_entry_owner_to_string(const struct stat *st,
 
 /**
  * Convert given entry's time into human-readable strings.
+ *
+ * \param r_is_today: optional, returns true if the date matches today's.
+ * \param r_is_yesterday: optional, returns true if the date matches yesterday's.
  */
 void BLI_filelist_entry_datetime_to_string(const struct stat *st,
                                            const int64_t ts,
                                            const bool compact,
                                            char r_time[FILELIST_DIRENTRY_TIME_LEN],
-                                           char r_date[FILELIST_DIRENTRY_DATE_LEN])
+                                           char r_date[FILELIST_DIRENTRY_DATE_LEN],
+                                           bool *r_is_today,
+                                           bool *r_is_yesterday)
 {
-  time_t ts_mtime = ts;
+  int today_year = 0;
+  int today_yday = 0;
+  int yesterday_year = 0;
+  int yesterday_yday = 0;
+
+  if (r_is_today || r_is_yesterday) {
+    /* Localtime() has only one buffer so need to get data out before called again. */
+    const time_t ts_now = time(NULL);
+    struct tm *today = localtime(&ts_now);
+
+    today_year = today->tm_year;
+    today_yday = today->tm_yday;
+    /* Handle a yesterday that spans a year */
+    today->tm_mday--;
+    mktime(today);
+    yesterday_year = today->tm_year;
+    yesterday_yday = today->tm_yday;
+
+    if (r_is_today) {
+      *r_is_today = false;
+    }
+    if (r_is_yesterday) {
+      *r_is_yesterday = false;
+    }
+  }
+
+  const time_t ts_mtime = ts;
   const struct tm *tm = localtime(st ? &st->st_mtime : &ts_mtime);
   const time_t zero = 0;
 
@@ -385,12 +397,20 @@ void BLI_filelist_entry_datetime_to_string(const struct stat *st,
   if (r_time) {
     strftime(r_time, sizeof(*r_time) * FILELIST_DIRENTRY_TIME_LEN, "%H:%M", tm);
   }
+
   if (r_date) {
     strftime(r_date,
              sizeof(*r_date) * FILELIST_DIRENTRY_DATE_LEN,
-             compact ? "%d/%m/%y" : "%d-%b-%y",
+             compact ? "%d/%m/%y" : "%d %b %Y",
              tm);
   }
+
+  if (r_is_today && (tm->tm_year == today_year) && (tm->tm_yday == today_yday)) {
+    *r_is_today = true;
+  }
+  else if (r_is_yesterday && (tm->tm_year == yesterday_year) && (tm->tm_yday == yesterday_yday)) {
+    *r_is_yesterday = true;
+  }
 }
 
 /**
diff --git a/source/blender/blenlib/intern/BLI_ghash.c b/source/blender/blenlib/intern/BLI_ghash.c
index 769bb02e2b9..05ffb02597d 100644
--- a/source/blender/blenlib/intern/BLI_ghash.c
+++ b/source/blender/blenlib/intern/BLI_ghash.c
@@ -886,7 +886,7 @@ bool BLI_ghash_ensure_p_ex(GHash *gh, const void *key, void ***r_key, void ***r_
   const bool haskey = (e != NULL);
 
   if (!haskey) {
-    /* pass 'key' incase we resize */
+    /* Pass 'key' in case we resize. */
     e = BLI_mempool_alloc(gh->entrypool);
     ghash_insert_ex_keyonly_entry(gh, (void *)key, bucket_index, (Entry *)e);
     e->e.key = NULL; /* caller must re-assign */
@@ -1189,7 +1189,7 @@ bool BLI_gset_ensure_p_ex(GSet *gs, const void *key, void ***r_key)
   const bool haskey = (e != NULL);
 
   if (!haskey) {
-    /* pass 'key' incase we resize */
+    /* Pass 'key' in case we resize */
     e = BLI_mempool_alloc(((GHash *)gs)->entrypool);
     ghash_insert_ex_keyonly_entry((GHash *)gs, (void *)key, bucket_index, (Entry *)e);
     e->key = NULL; /* caller must re-assign */
diff --git a/source/blender/blenlib/intern/BLI_kdopbvh.c b/source/blender/blenlib/intern/BLI_kdopbvh.c
index 0e93fd8e13b..ed3c7096865 100644
--- a/source/blender/blenlib/intern/BLI_kdopbvh.c
+++ b/source/blender/blenlib/intern/BLI_kdopbvh.c
@@ -1467,6 +1467,95 @@ int BLI_bvhtree_find_nearest(BVHTree *tree,
 /** \} */
 
 /* -------------------------------------------------------------------- */
+/** \name BLI_bvhtree_find_nearest_first
+ * \{ */
+
+static bool isect_aabb_v3(BVHNode *node, const float co[3])
+{
+  const BVHTreeAxisRange *bv = (const BVHTreeAxisRange *)node->bv;
+
+  if (co[0] > bv[0].min && co[0] < bv[0].max && co[1] > bv[1].min && co[1] < bv[1].max &&
+      co[2] > bv[2].min && co[2] < bv[2].max) {
+    return true;
+  }
+
+  return false;
+}
+
+static bool dfs_find_duplicate_fast_dfs(BVHNearestData *data, BVHNode *node)
+{
+  if (node->totnode == 0) {
+    if (isect_aabb_v3(node, data->co)) {
+      if (data->callback) {
+        const float dist_sq = data->nearest.dist_sq;
+        data->callback(data->userdata, node->index, data->co, &data->nearest);
+        return (data->nearest.dist_sq < dist_sq);
+      }
+      else {
+        data->nearest.index = node->index;
+        return true;
+      }
+    }
+  }
+  else {
+    /* Better heuristic to pick the closest node to dive on */
+    int i;
+
+    if (data->proj[node->main_axis] <= node->children[0]->bv[node->main_axis * 2 + 1]) {
+      for (i = 0; i != node->totnode; i++) {
+        if (isect_aabb_v3(node->children[i], data->co)) {
+          if (dfs_find_duplicate_fast_dfs(data, node->children[i])) {
+            return true;
+          }
+        }
+      }
+    }
+    else {
+      for (i = node->totnode; i--;) {
+        if (isect_aabb_v3(node->children[i], data->co)) {
+          if (dfs_find_duplicate_fast_dfs(data, node->children[i])) {
+            return true;
+          }
+        }
+      }
+    }
+  }
+  return false;
+}
+
+/**
+ * Find the first node nearby.
+ * Favors speed over quality since it doesn't find the best target node.
+ */
+int BLI_bvhtree_find_nearest_first(BVHTree *tree,
+                                   const float co[3],
+                                   const float dist_sq,
+                                   BVHTree_NearestPointCallback callback,
+                                   void *userdata)
+{
+  BVHNearestData data;
+  BVHNode *root = tree->nodes[tree->totleaf];
+
+  /* init data to search */
+  data.tree = tree;
+  data.co = co;
+
+  data.callback = callback;
+  data.userdata = userdata;
+  data.nearest.index = -1;
+  data.nearest.dist_sq = dist_sq;
+
+  /* dfs search */
+  if (root) {
+    dfs_find_duplicate_fast_dfs(&data, root);
+  }
+
+  return data.nearest.index;
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name BLI_bvhtree_ray_cast
  *
  * raycast is done by performing a DFS on the BVHTree and saving the closest hit.
diff --git a/source/blender/blenlib/intern/boxpack_2d.c b/source/blender/blenlib/intern/boxpack_2d.c
index 196d45967be..ddc7f9ee4c7 100644
--- a/source/blender/blenlib/intern/boxpack_2d.c
+++ b/source/blender/blenlib/intern/boxpack_2d.c
@@ -273,12 +273,12 @@ static int vertex_sort(const void *p1, const void *p2, void *vs_ctx_p)
 /** \} */
 
 /**
- * Main boxpacking function accessed from other functions
+ * Main box-packing function accessed from other functions
  * This sets boxes x,y to positive values, sorting from 0,0 outwards.
  * There is no limit to the space boxes may take, only that they will be packed
  * tightly into the lower left hand corner (0,0)
  *
- * \param boxarray: a pre allocated array of boxes.
+ * \param boxarray: a pre-allocated array of boxes.
  *      only the 'box->x' and 'box->y' are set, 'box->w' and 'box->h' are used,
  *      'box->index' is not used at all, the only reason its there
  *          is that the box array is sorted by area and programs need to be able
diff --git a/source/blender/blenlib/intern/delaunay_2d.c b/source/blender/blenlib/intern/delaunay_2d.c
index 23f560c5463..102ea8eb234 100644
--- a/source/blender/blenlib/intern/delaunay_2d.c
+++ b/source/blender/blenlib/intern/delaunay_2d.c
@@ -115,99 +115,6 @@ static void validate_face_centroid(SymEdge *se);
 static void validate_cdt(CDT_state *cdt, bool check_all_tris);
 #endif
 
-/* TODO: move these to BLI_vector... and BLI_math... */
-static double max_dd(const double a, const double b)
-{
-  return (a > b) ? a : b;
-}
-
-static double len_v2v2_db(const double a[2], const double b[2])
-{
-  return sqrt((b[0] - a[0]) * (b[0] - a[0]) + (b[1] - a[1]) * (b[1] - a[1]));
-}
-
-static double len_squared_v2v2_db(const double a[2], const double b[2])
-{
-  return (b[0] - a[0]) * (b[0] - a[0]) + (b[1] - a[1]) * (b[1] - a[1]);
-}
-
-static void add_v2_v2_db(double a[2], const double b[2])
-{
-  a[0] += b[0];
-  a[1] += b[1];
-}
-
-static void sub_v2_v2v2_db(double *a, const double *b, const double *c)
-{
-  a[0] = b[0] - c[0];
-  a[1] = b[1] - c[1];
-}
-
-static double dot_v2v2_db(const double *a, const double *b)
-{
-  return a[0] * b[0] + a[1] * b[1];
-}
-
-static double closest_to_line_v2_db(double r_close[2],
-                                    const double p[2],
-                                    const double l1[2],
-                                    const double l2[2])
-{
-  double h[2], u[2], lambda, denom;
-  sub_v2_v2v2_db(u, l2, l1);
-  sub_v2_v2v2_db(h, p, l1);
-  denom = dot_v2v2_db(u, u);
-  if (denom < DBL_EPSILON) {
-    r_close[0] = l1[0];
-    r_close[1] = l1[1];
-    return 0.0;
-  }
-  lambda = dot_v2v2_db(u, h) / dot_v2v2_db(u, u);
-  r_close[0] = l1[0] + u[0] * lambda;
-  r_close[1] = l1[1] + u[1] * lambda;
-  return lambda;
-}
-
-/**
- * If intersection == ISECT_LINE_LINE_CROSS or ISECT_LINE_LINE_NONE:
- * <pre>
- * pt = v1 + lamba * (v2 - v1) = v3 + mu * (v4 - v3)
- * </pre>
- */
-static int isect_seg_seg_v2_lambda_mu_db(const double v1[2],
-                                         const double v2[2],
-                                         const double v3[2],
-                                         const double v4[2],
-                                         double *r_lambda,
-                                         double *r_mu)
-{
-  double div, lambda, mu;
-
-  div = (v2[0] - v1[0]) * (v4[1] - v3[1]) - (v2[1] - v1[1]) * (v4[0] - v3[0]);
-  if (fabs(div) < DBL_EPSILON) {
-    return ISECT_LINE_LINE_COLINEAR;
-  }
-
-  lambda = ((v1[1] - v3[1]) * (v4[0] - v3[0]) - (v1[0] - v3[0]) * (v4[1] - v3[1])) / div;
-
-  mu = ((v1[1] - v3[1]) * (v2[0] - v1[0]) - (v1[0] - v3[0]) * (v2[1] - v1[1])) / div;
-
-  if (r_lambda) {
-    *r_lambda = lambda;
-  }
-  if (r_mu) {
-    *r_mu = mu;
-  }
-
-  if (lambda >= 0.0 && lambda <= 1.0 && mu >= 0.0 && mu <= 1.0) {
-    if (lambda == 0.0 || lambda == 1.0 || mu == 0.0 || mu == 1.0) {
-      return ISECT_LINE_LINE_EXACT;
-    }
-    return ISECT_LINE_LINE_CROSS;
-  }
-  return ISECT_LINE_LINE_NONE;
-}
-
 /** return 1 if a,b,c forms CCW angle, -1 if a CW angle, 0 if straight  */
 static int CCW_test(const double a[2], const double b[2], const double c[2])
 {
diff --git a/source/blender/blenlib/intern/freetypefont.c b/source/blender/blenlib/intern/freetypefont.c
index 7be7674069c..b5a950d6851 100644
--- a/source/blender/blenlib/intern/freetypefont.c
+++ b/source/blender/blenlib/intern/freetypefont.c
@@ -565,7 +565,7 @@ VChar *BLI_vfontchar_copy(const VChar *vchar_src, const int UNUSED(flag))
  *       between them
  * </pre>
  *
- * Each glyph's original outline points are located on a grid of indivisible units.
+ * Each glyphs original outline points are located on a grid of indivisible units.
  * The points are stored in the font file as 16-bit integer grid coordinates,
  * with the grid origin's being at (0, 0); they thus range from -16384 to 16383.
  *
diff --git a/source/blender/blenlib/intern/math_base_inline.c b/source/blender/blenlib/intern/math_base_inline.c
index 320f8a0f1ab..47327a878d4 100644
--- a/source/blender/blenlib/intern/math_base_inline.c
+++ b/source/blender/blenlib/intern/math_base_inline.c
@@ -353,6 +353,15 @@ MINLINE float max_ff(float a, float b)
   return (a > b) ? a : b;
 }
 
+MINLINE double min_dd(double a, double b)
+{
+  return (a < b) ? a : b;
+}
+MINLINE double max_dd(double a, double b)
+{
+  return (a > b) ? a : b;
+}
+
 MINLINE int min_ii(int a, int b)
 {
   return (a < b) ? a : b;
diff --git a/source/blender/blenlib/intern/math_color.c b/source/blender/blenlib/intern/math_color.c
index f07c2ec159e..cc29ebe4f20 100644
--- a/source/blender/blenlib/intern/math_color.c
+++ b/source/blender/blenlib/intern/math_color.c
@@ -153,7 +153,7 @@ void rgb_to_ycc(float r, float g, float b, float *ly, float *lcb, float *lcr, in
       cr = (0.5f * sr) - (0.41869f * sg) - (0.08131f * sb) + 128.0f;
       break;
     default:
-      assert(!"invalid colorspace");
+      BLI_assert(!"invalid colorspace");
       break;
   }
 
diff --git a/source/blender/blenlib/intern/math_geom.c b/source/blender/blenlib/intern/math_geom.c
index 06fd5c70772..f1575ed2b2d 100644
--- a/source/blender/blenlib/intern/math_geom.c
+++ b/source/blender/blenlib/intern/math_geom.c
@@ -1356,6 +1356,52 @@ bool isect_seg_seg_v2_simple(const float v1[2],
 }
 
 /**
+ * If intersection == ISECT_LINE_LINE_CROSS or ISECT_LINE_LINE_NONE:
+ * <pre>
+ * pt = v1 + lambda * (v2 - v1) = v3 + mu * (v4 - v3)
+ * </pre>
+ * \returns intersection type:
+ * - ISECT_LINE_LINE_COLINEAR: collinear.
+ * - ISECT_LINE_LINE_EXACT: intersection at an endpoint of either.
+ * - ISECT_LINE_LINE_CROSS: interaction, not at an endpoint.
+ * - ISECT_LINE_LINE_NONE: no intersection.
+ * Also returns lambda and mu in r_lambda and r_mu.
+ */
+int isect_seg_seg_v2_lambda_mu_db(const double v1[2],
+                                  const double v2[2],
+                                  const double v3[2],
+                                  const double v4[2],
+                                  double *r_lambda,
+                                  double *r_mu)
+{
+  double div, lambda, mu;
+
+  div = (v2[0] - v1[0]) * (v4[1] - v3[1]) - (v2[1] - v1[1]) * (v4[0] - v3[0]);
+  if (fabs(div) < DBL_EPSILON) {
+    return ISECT_LINE_LINE_COLINEAR;
+  }
+
+  lambda = ((v1[1] - v3[1]) * (v4[0] - v3[0]) - (v1[0] - v3[0]) * (v4[1] - v3[1])) / div;
+
+  mu = ((v1[1] - v3[1]) * (v2[0] - v1[0]) - (v1[0] - v3[0]) * (v2[1] - v1[1])) / div;
+
+  if (r_lambda) {
+    *r_lambda = lambda;
+  }
+  if (r_mu) {
+    *r_mu = mu;
+  }
+
+  if (lambda >= 0.0 && lambda <= 1.0 && mu >= 0.0 && mu <= 1.0) {
+    if (lambda == 0.0 || lambda == 1.0 || mu == 0.0 || mu == 1.0) {
+      return ISECT_LINE_LINE_EXACT;
+    }
+    return ISECT_LINE_LINE_CROSS;
+  }
+  return ISECT_LINE_LINE_NONE;
+}
+
+/**
  * \param l1, l2: Coordinates (point of line).
  * \param sp, r:  Coordinate and radius (sphere).
  * \return r_p1, r_p2: Intersection coordinates.
@@ -3187,6 +3233,26 @@ float closest_to_line_v2(float r_close[2], const float p[2], const float l1[2],
   return lambda;
 }
 
+double closest_to_line_v2_db(double r_close[2],
+                             const double p[2],
+                             const double l1[2],
+                             const double l2[2])
+{
+  double h[2], u[2], lambda, denom;
+  sub_v2_v2v2_db(u, l2, l1);
+  sub_v2_v2v2_db(h, p, l1);
+  denom = dot_v2v2_db(u, u);
+  if (denom < DBL_EPSILON) {
+    r_close[0] = l1[0];
+    r_close[1] = l1[1];
+    return 0.0;
+  }
+  lambda = dot_v2v2_db(u, h) / dot_v2v2_db(u, u);
+  r_close[0] = l1[0] + u[0] * lambda;
+  r_close[1] = l1[1] + u[1] * lambda;
+  return lambda;
+}
+
 float ray_point_factor_v3_ex(const float p[3],
                              const float ray_origin[3],
                              const float ray_direction[3],
@@ -3452,7 +3518,7 @@ bool clip_segment_v3_plane(
     }
   }
 
-  /* incase input/output values match (above also) */
+  /* In case input/output values match (above also). */
   const float p1_copy[3] = {UNPACK3(p1)};
   copy_v3_v3(r_p2, p2);
   copy_v3_v3(r_p1, p1_copy);
@@ -3511,7 +3577,7 @@ bool clip_segment_v3_plane_n(const float p1[3],
     }
   }
 
-  /* incase input/output values match */
+  /* In case input/output values match. */
   const float p1_copy[3] = {UNPACK3(p1)};
 
   madd_v3_v3v3fl(r_p1, p1_copy, dp, p1_fac);
diff --git a/source/blender/blenlib/intern/math_matrix.c b/source/blender/blenlib/intern/math_matrix.c
index 7c64206134b..2568d42566c 100644
--- a/source/blender/blenlib/intern/math_matrix.c
+++ b/source/blender/blenlib/intern/math_matrix.c
@@ -71,6 +71,15 @@ void unit_m4(float m[4][4])
   m[3][0] = m[3][1] = m[3][2] = 0.0f;
 }
 
+void unit_m4_db(double m[4][4])
+{
+  m[0][0] = m[1][1] = m[2][2] = m[3][3] = 1.0f;
+  m[0][1] = m[0][2] = m[0][3] = 0.0f;
+  m[1][0] = m[1][2] = m[1][3] = 0.0f;
+  m[2][0] = m[2][1] = m[2][3] = 0.0f;
+  m[3][0] = m[3][1] = m[3][2] = 0.0f;
+}
+
 void copy_m2_m2(float m1[2][2], const float m2[2][2])
 {
   memcpy(m1, m2, sizeof(float[2][2]));
@@ -87,6 +96,11 @@ void copy_m4_m4(float m1[4][4], const float m2[4][4])
   memcpy(m1, m2, sizeof(float[4][4]));
 }
 
+void copy_m4_m4_db(double m1[4][4], const double m2[4][4])
+{
+  memcpy(m1, m2, sizeof(double[4][4]));
+}
+
 void copy_m3_m4(float m1[3][3], const float m2[4][4])
 {
   m1[0][0] = m2[0][0];
@@ -127,6 +141,29 @@ void copy_m4_m3(float m1[4][4], const float m2[3][3]) /* no clear */
   m1[3][3] = 1.0f;
 }
 
+void copy_m4d_m4(double m1[4][4], const float m2[4][4])
+{
+  m1[0][0] = m2[0][0];
+  m1[0][1] = m2[0][1];
+  m1[0][2] = m2[0][2];
+  m1[0][3] = m2[0][3];
+
+  m1[1][0] = m2[1][0];
+  m1[1][1] = m2[1][1];
+  m1[1][2] = m2[1][2];
+  m1[1][3] = m2[1][3];
+
+  m1[2][0] = m2[2][0];
+  m1[2][1] = m2[2][1];
+  m1[2][2] = m2[2][2];
+  m1[2][3] = m2[2][3];
+
+  m1[3][0] = m2[3][0];
+  m1[3][1] = m2[3][1];
+  m1[3][2] = m2[3][2];
+  m1[3][3] = m2[3][3];
+}
+
 void copy_m3_m3d(float R[3][3], const double A[3][3])
 {
   /* Keep it stupid simple for better data flow in CPU. */
@@ -231,6 +268,61 @@ void mul_m4_m4m4_uniq(float R[4][4], const float A[4][4], const float B[4][4])
 #endif
 }
 
+void mul_m4_m4m4_db_uniq(double R[4][4], const double A[4][4], const double B[4][4])
+{
+  BLI_assert(R != A && R != B);
+
+  /* matrix product: R[j][k] = A[j][i] . B[i][k] */
+
+  R[0][0] = B[0][0] * A[0][0] + B[0][1] * A[1][0] + B[0][2] * A[2][0] + B[0][3] * A[3][0];
+  R[0][1] = B[0][0] * A[0][1] + B[0][1] * A[1][1] + B[0][2] * A[2][1] + B[0][3] * A[3][1];
+  R[0][2] = B[0][0] * A[0][2] + B[0][1] * A[1][2] + B[0][2] * A[2][2] + B[0][3] * A[3][2];
+  R[0][3] = B[0][0] * A[0][3] + B[0][1] * A[1][3] + B[0][2] * A[2][3] + B[0][3] * A[3][3];
+
+  R[1][0] = B[1][0] * A[0][0] + B[1][1] * A[1][0] + B[1][2] * A[2][0] + B[1][3] * A[3][0];
+  R[1][1] = B[1][0] * A[0][1] + B[1][1] * A[1][1] + B[1][2] * A[2][1] + B[1][3] * A[3][1];
+  R[1][2] = B[1][0] * A[0][2] + B[1][1] * A[1][2] + B[1][2] * A[2][2] + B[1][3] * A[3][2];
+  R[1][3] = B[1][0] * A[0][3] + B[1][1] * A[1][3] + B[1][2] * A[2][3] + B[1][3] * A[3][3];
+
+  R[2][0] = B[2][0] * A[0][0] + B[2][1] * A[1][0] + B[2][2] * A[2][0] + B[2][3] * A[3][0];
+  R[2][1] = B[2][0] * A[0][1] + B[2][1] * A[1][1] + B[2][2] * A[2][1] + B[2][3] * A[3][1];
+  R[2][2] = B[2][0] * A[0][2] + B[2][1] * A[1][2] + B[2][2] * A[2][2] + B[2][3] * A[3][2];
+  R[2][3] = B[2][0] * A[0][3] + B[2][1] * A[1][3] + B[2][2] * A[2][3] + B[2][3] * A[3][3];
+
+  R[3][0] = B[3][0] * A[0][0] + B[3][1] * A[1][0] + B[3][2] * A[2][0] + B[3][3] * A[3][0];
+  R[3][1] = B[3][0] * A[0][1] + B[3][1] * A[1][1] + B[3][2] * A[2][1] + B[3][3] * A[3][1];
+  R[3][2] = B[3][0] * A[0][2] + B[3][1] * A[1][2] + B[3][2] * A[2][2] + B[3][3] * A[3][2];
+  R[3][3] = B[3][0] * A[0][3] + B[3][1] * A[1][3] + B[3][2] * A[2][3] + B[3][3] * A[3][3];
+}
+
+void mul_m4db_m4db_m4fl_uniq(double R[4][4], const double A[4][4], const float B[4][4])
+{
+  /* Remove second check since types don't match. */
+  BLI_assert(R != A /* && R != B */);
+
+  /* matrix product: R[j][k] = A[j][i] . B[i][k] */
+
+  R[0][0] = B[0][0] * A[0][0] + B[0][1] * A[1][0] + B[0][2] * A[2][0] + B[0][3] * A[3][0];
+  R[0][1] = B[0][0] * A[0][1] + B[0][1] * A[1][1] + B[0][2] * A[2][1] + B[0][3] * A[3][1];
+  R[0][2] = B[0][0] * A[0][2] + B[0][1] * A[1][2] + B[0][2] * A[2][2] + B[0][3] * A[3][2];
+  R[0][3] = B[0][0] * A[0][3] + B[0][1] * A[1][3] + B[0][2] * A[2][3] + B[0][3] * A[3][3];
+
+  R[1][0] = B[1][0] * A[0][0] + B[1][1] * A[1][0] + B[1][2] * A[2][0] + B[1][3] * A[3][0];
+  R[1][1] = B[1][0] * A[0][1] + B[1][1] * A[1][1] + B[1][2] * A[2][1] + B[1][3] * A[3][1];
+  R[1][2] = B[1][0] * A[0][2] + B[1][1] * A[1][2] + B[1][2] * A[2][2] + B[1][3] * A[3][2];
+  R[1][3] = B[1][0] * A[0][3] + B[1][1] * A[1][3] + B[1][2] * A[2][3] + B[1][3] * A[3][3];
+
+  R[2][0] = B[2][0] * A[0][0] + B[2][1] * A[1][0] + B[2][2] * A[2][0] + B[2][3] * A[3][0];
+  R[2][1] = B[2][0] * A[0][1] + B[2][1] * A[1][1] + B[2][2] * A[2][1] + B[2][3] * A[3][1];
+  R[2][2] = B[2][0] * A[0][2] + B[2][1] * A[1][2] + B[2][2] * A[2][2] + B[2][3] * A[3][2];
+  R[2][3] = B[2][0] * A[0][3] + B[2][1] * A[1][3] + B[2][2] * A[2][3] + B[2][3] * A[3][3];
+
+  R[3][0] = B[3][0] * A[0][0] + B[3][1] * A[1][0] + B[3][2] * A[2][0] + B[3][3] * A[3][0];
+  R[3][1] = B[3][0] * A[0][1] + B[3][1] * A[1][1] + B[3][2] * A[2][1] + B[3][3] * A[3][1];
+  R[3][2] = B[3][0] * A[0][2] + B[3][1] * A[1][2] + B[3][2] * A[2][2] + B[3][3] * A[3][2];
+  R[3][3] = B[3][0] * A[0][3] + B[3][1] * A[1][3] + B[3][2] * A[2][3] + B[3][3] * A[3][3];
+}
+
 void mul_m4_m4_pre(float R[4][4], const float A[4][4])
 {
   BLI_assert(A != R);
@@ -604,6 +696,26 @@ void mul_v3_m4v3(float r[3], const float mat[4][4], const float vec[3])
   r[2] = x * mat[0][2] + y * mat[1][2] + mat[2][2] * vec[2] + mat[3][2];
 }
 
+void mul_v3_m4v3_db(double r[3], const double mat[4][4], const double vec[3])
+{
+  const double x = vec[0];
+  const double y = vec[1];
+
+  r[0] = x * mat[0][0] + y * mat[1][0] + mat[2][0] * vec[2] + mat[3][0];
+  r[1] = x * mat[0][1] + y * mat[1][1] + mat[2][1] * vec[2] + mat[3][1];
+  r[2] = x * mat[0][2] + y * mat[1][2] + mat[2][2] * vec[2] + mat[3][2];
+}
+void mul_v4_m4v3_db(double r[4], const double mat[4][4], const double vec[3])
+{
+  const double x = vec[0];
+  const double y = vec[1];
+
+  r[0] = x * mat[0][0] + y * mat[1][0] + mat[2][0] * vec[2] + mat[3][0];
+  r[1] = x * mat[0][1] + y * mat[1][1] + mat[2][1] * vec[2] + mat[3][1];
+  r[2] = x * mat[0][2] + y * mat[1][2] + mat[2][2] * vec[2] + mat[3][2];
+  r[3] = x * mat[0][3] + y * mat[1][3] + mat[2][3] * vec[2] + mat[3][3];
+}
+
 void mul_v2_m4v3(float r[2], const float mat[4][4], const float vec[3])
 {
   const float x = vec[0];
@@ -646,6 +758,16 @@ void mul_v3_mat3_m4v3(float r[3], const float mat[4][4], const float vec[3])
   r[2] = x * mat[0][2] + y * mat[1][2] + mat[2][2] * vec[2];
 }
 
+void mul_v3_mat3_m4v3_db(double r[3], const double mat[4][4], const double vec[3])
+{
+  const double x = vec[0];
+  const double y = vec[1];
+
+  r[0] = x * mat[0][0] + y * mat[1][0] + mat[2][0] * vec[2];
+  r[1] = x * mat[0][1] + y * mat[1][1] + mat[2][1] * vec[2];
+  r[2] = x * mat[0][2] + y * mat[1][2] + mat[2][2] * vec[2];
+}
+
 void mul_project_m4_v3(const float mat[4][4], float vec[3])
 {
   /* absolute value to not flip the frustum upside down behind the camera */
@@ -1891,7 +2013,7 @@ void rotate_m4(float mat[4][4], const char axis, const float angle)
   const float angle_cos = cosf(angle);
   const float angle_sin = sinf(angle);
 
-  assert(axis >= 'X' && axis <= 'Z');
+  BLI_assert(axis >= 'X' && axis <= 'Z');
 
   switch (axis) {
     case 'X':
diff --git a/source/blender/blenlib/intern/math_rotation.c b/source/blender/blenlib/intern/math_rotation.c
index 47814b1080c..5762d164914 100644
--- a/source/blender/blenlib/intern/math_rotation.c
+++ b/source/blender/blenlib/intern/math_rotation.c
@@ -535,6 +535,49 @@ void rotation_between_quats_to_quat(float q[4], const float q1[4], const float q
   mul_qt_qtqt(q, tquat, q2);
 }
 
+/** Decompose a quaternion into a swing rotation (quaternion with the selected
+ *  axis component locked at zero), followed by a twist rotation around the axis.
+ *
+ *  \param q: input quaternion.
+ *  \param axis: twist axis in [0,1,2]
+ *  \param r_swing[out]: if not NULL, receives the swing quaternion.
+ *  \param r_twist[out]: if not NULL, receives the twist quaternion.
+ *  \returns twist angle.
+ */
+float quat_split_swing_and_twist(const float q[4], int axis, float r_swing[4], float r_twist[4])
+{
+  BLI_assert(axis >= 0 && axis <= 2);
+
+  /* Half-twist angle can be computed directly. */
+  float t = atan2f(q[axis + 1], q[0]);
+
+  if (r_swing || r_twist) {
+    float sin_t = sinf(t), cos_t = cosf(t);
+
+    /* Compute swing by multiplying the original quaternion by inverted twist. */
+    if (r_swing) {
+      float twist_inv[4];
+
+      twist_inv[0] = cos_t;
+      zero_v3(twist_inv + 1);
+      twist_inv[axis + 1] = -sin_t;
+
+      mul_qt_qtqt(r_swing, q, twist_inv);
+
+      BLI_assert(fabsf(r_swing[axis + 1]) < BLI_ASSERT_UNIT_EPSILON);
+    }
+
+    /* Output twist last just in case q ovelaps r_twist. */
+    if (r_twist) {
+      r_twist[0] = cos_t;
+      zero_v3(r_twist + 1);
+      r_twist[axis + 1] = sin_t;
+    }
+  }
+
+  return 2.0f * t;
+}
+
 /* -------------------------------------------------------------------- */
 /** \name Quaternion Angle
  *
@@ -644,8 +687,8 @@ void vec_to_quat(float q[4], const float vec[3], short axis, const short upflag)
   float nor[3], tvec[3];
   float angle, si, co, len;
 
-  assert(axis >= 0 && axis <= 5);
-  assert(upflag >= 0 && upflag <= 2);
+  BLI_assert(axis >= 0 && axis <= 5);
+  BLI_assert(upflag >= 0 && upflag <= 2);
 
   /* first set the quat to unit */
   unit_qt(q);
@@ -1194,7 +1237,7 @@ void axis_angle_to_quat_single(float q[4], const char axis, const float angle)
   const float angle_sin = sinf(angle_half);
   const int axis_index = (axis - 'X');
 
-  assert(axis >= 'X' && axis <= 'Z');
+  BLI_assert(axis >= 'X' && axis <= 'Z');
 
   q[0] = angle_cos;
   zero_v3(q + 1);
@@ -1398,7 +1441,7 @@ void rotate_eul(float beul[3], const char axis, const float ang)
 {
   float eul[3], mat1[3][3], mat2[3][3], totmat[3][3];
 
-  assert(axis >= 'X' && axis <= 'Z');
+  BLI_assert(axis >= 'X' && axis <= 'Z');
 
   eul[0] = eul[1] = eul[2] = 0.0f;
   if (axis == 'X') {
@@ -1544,7 +1587,7 @@ static const RotOrderInfo rotOrders[] = {
  */
 static const RotOrderInfo *get_rotation_order_info(const short order)
 {
-  assert(order >= 0 && order <= 6);
+  BLI_assert(order >= 0 && order <= 6);
   if (order < 1) {
     return &rotOrders[0];
   }
@@ -1813,7 +1856,7 @@ void rotate_eulO(float beul[3], const short order, char axis, float ang)
 {
   float eul[3], mat1[3][3], mat2[3][3], totmat[3][3];
 
-  assert(axis >= 'X' && axis <= 'Z');
+  BLI_assert(axis >= 'X' && axis <= 'Z');
 
   zero_v3(eul);
 
@@ -2119,8 +2162,8 @@ void quat_apply_track(float quat[4], short axis, short upflag)
       {0.0, sqrt_1_2, sqrt_1_2, 0.0},
   };
 
-  assert(axis >= 0 && axis <= 5);
-  assert(upflag >= 0 && upflag <= 2);
+  BLI_assert(axis >= 0 && axis <= 5);
+  BLI_assert(upflag >= 0 && upflag <= 2);
 
   mul_qt_qtqt(quat, quat, quat_track[axis]);
 
@@ -2142,7 +2185,7 @@ void vec_apply_track(float vec[3], short axis)
 {
   float tvec[3];
 
-  assert(axis >= 0 && axis <= 5);
+  BLI_assert(axis >= 0 && axis <= 5);
 
   copy_v3_v3(tvec, vec);
 
diff --git a/source/blender/blenlib/intern/math_vector_inline.c b/source/blender/blenlib/intern/math_vector_inline.c
index 865c2f5dc25..b4b53a1dd58 100644
--- a/source/blender/blenlib/intern/math_vector_inline.c
+++ b/source/blender/blenlib/intern/math_vector_inline.c
@@ -355,6 +355,12 @@ MINLINE void add_v2_v2(float r[2], const float a[2])
   r[1] += a[1];
 }
 
+MINLINE void add_v2_v2_db(double r[2], const double a[2])
+{
+  r[0] += a[0];
+  r[1] += a[1];
+}
+
 MINLINE void add_v2_v2v2(float r[2], const float a[2], const float b[2])
 {
   r[0] = a[0] + b[0];
@@ -374,6 +380,13 @@ MINLINE void add_v3_v3(float r[3], const float a[3])
   r[2] += a[2];
 }
 
+MINLINE void add_v3_v3_db(double r[3], const double a[3])
+{
+  r[0] += a[0];
+  r[1] += a[1];
+  r[2] += a[2];
+}
+
 MINLINE void add_v3_v3v3(float r[3], const float a[3], const float b[3])
 {
   r[0] = a[0] + b[0];
@@ -423,6 +436,12 @@ MINLINE void sub_v2_v2v2(float r[2], const float a[2], const float b[2])
   r[1] = a[1] - b[1];
 }
 
+MINLINE void sub_v2_v2v2_db(double r[2], const double a[2], const double b[2])
+{
+  r[0] = a[0] - b[0];
+  r[1] = a[1] - b[1];
+}
+
 MINLINE void sub_v2_v2v2_int(int r[2], const int a[2], const int b[2])
 {
   r[0] = a[0] - b[0];
@@ -450,6 +469,13 @@ MINLINE void sub_v3_v3v3_int(int r[3], const int a[3], const int b[3])
   r[2] = a[2] - b[2];
 }
 
+MINLINE void sub_v3_v3v3_db(double r[3], const double a[3], const double b[3])
+{
+  r[0] = a[0] - b[0];
+  r[1] = a[1] - b[1];
+  r[2] = a[2] - b[2];
+}
+
 MINLINE void sub_v3db_v3fl_v3fl(double r[3], const float a[3], const float b[3])
 {
   r[0] = (double)a[0] - (double)b[0];
@@ -492,6 +518,13 @@ MINLINE void mul_v3_fl(float r[3], float f)
   r[2] *= f;
 }
 
+MINLINE void mul_v3db_db(double r[3], double f)
+{
+  r[0] *= f;
+  r[1] *= f;
+  r[2] *= f;
+}
+
 MINLINE void mul_v3_v3fl(float r[3], const float a[3], float f)
 {
   r[0] = a[0] * f;
@@ -809,6 +842,11 @@ MINLINE float dot_v2v2(const float a[2], const float b[2])
   return a[0] * b[0] + a[1] * b[1];
 }
 
+MINLINE double dot_v2v2_db(const double a[2], const double b[2])
+{
+  return a[0] * b[0] + a[1] * b[1];
+}
+
 MINLINE float dot_v3v3(const float a[3], const float b[3])
 {
   return a[0] * b[0] + a[1] * b[1] + a[2] * b[2];
@@ -836,6 +874,11 @@ MINLINE double dot_v3db_v3fl(const double a[3], const float b[3])
   return a[0] * (double)b[0] + a[1] * (double)b[1] + a[2] * (double)b[2];
 }
 
+MINLINE double dot_v3v3_db(const double a[3], const double b[3])
+{
+  return a[0] * b[0] + a[1] * b[1] + a[2] * b[2];
+}
+
 MINLINE float cross_v2v2(const float a[2], const float b[2])
 {
   return a[0] * b[1] - a[1] * b[0];
@@ -859,6 +902,14 @@ MINLINE void cross_v3_v3v3_hi_prec(float r[3], const float a[3], const float b[3
   r[2] = (float)((double)a[0] * (double)b[1] - (double)a[1] * (double)b[0]);
 }
 
+MINLINE void cross_v3_v3v3_db(double r[3], const double a[3], const double b[3])
+{
+  BLI_assert(r != a && r != b);
+  r[0] = a[1] * b[2] - a[2] * b[1];
+  r[1] = a[2] * b[0] - a[0] * b[2];
+  r[2] = a[0] * b[1] - a[1] * b[0];
+}
+
 /* Newell's Method */
 /* excuse this fairly specific function,
  * its used for polygon normals all over the place
@@ -922,6 +973,15 @@ MINLINE float len_v2v2(const float v1[2], const float v2[2])
   return sqrtf(x * x + y * y);
 }
 
+MINLINE double len_v2v2_db(const double v1[2], const double v2[2])
+{
+  double x, y;
+
+  x = v1[0] - v2[0];
+  y = v1[1] - v2[1];
+  return sqrt(x * x + y * y);
+}
+
 MINLINE float len_v2v2_int(const int v1[2], const int v2[2])
 {
   float x, y;
@@ -944,6 +1004,14 @@ MINLINE float len_squared_v2v2(const float a[2], const float b[2])
   return dot_v2v2(d, d);
 }
 
+MINLINE double len_squared_v2v2_db(const double a[2], const double b[2])
+{
+  double d[2];
+
+  sub_v2_v2v2_db(d, b, a);
+  return dot_v2v2_db(d, d);
+}
+
 MINLINE float len_squared_v3v3(const float a[3], const float b[3])
 {
   float d[3];
diff --git a/source/blender/blenlib/intern/polyfill_2d_beautify.c b/source/blender/blenlib/intern/polyfill_2d_beautify.c
index 3e94ae8de1f..ab397b86b1a 100644
--- a/source/blender/blenlib/intern/polyfill_2d_beautify.c
+++ b/source/blender/blenlib/intern/polyfill_2d_beautify.c
@@ -75,7 +75,7 @@ static int oedge_cmp(const void *a1, const void *a2)
     return -1;
   }
 
-  /* only for pradictability */
+  /* Only for predictability. */
   if (x1->e_half > x2->e_half) {
     return 1;
   }
@@ -100,6 +100,10 @@ BLI_INLINE bool is_boundary_edge(uint i_a, uint i_b, const uint coord_last)
  * - When true, an existing zero area face on either side of the (2 - 4
  *   split will return a positive value.
  * - When false, the check must be non-biased towards either split direction.
+ * \param r_area: Return the area of the quad,
+ * This can be useful when comparing the return value with near zero epsilons.
+ * In this case the epsilon can be scaled by the area to avoid the return value
+ * of very large faces not having a reliable way to detect near-zero output.
  *
  * \return (negative number means the edge can be rotated, lager == better).
  */
@@ -107,7 +111,8 @@ float BLI_polyfill_beautify_quad_rotate_calc_ex(const float v1[2],
                                                 const float v2[2],
                                                 const float v3[2],
                                                 const float v4[2],
-                                                const bool lock_degenerate)
+                                                const bool lock_degenerate,
+                                                float *r_area)
 {
   /* not a loop (only to be able to break out) */
   do {
@@ -121,6 +126,13 @@ float BLI_polyfill_beautify_quad_rotate_calc_ex(const float v1[2],
 
     BLI_assert((ELEM(v1, v2, v3, v4) == false) && (ELEM(v2, v1, v3, v4) == false) &&
                (ELEM(v3, v1, v2, v4) == false) && (ELEM(v4, v1, v2, v3) == false));
+
+    if (r_area) {
+      *r_area = fabsf(area_2x_234) + fabsf(area_2x_241) +
+                /* Include both pairs for predictable results. */
+                fabsf(area_2x_123) + fabsf(area_2x_134) / 8.0f;
+    }
+
     /*
      * Test for unusable (1-3) state.
      * - Area sign flipping to check faces aren't going to point in opposite directions.
@@ -191,7 +203,8 @@ float BLI_polyfill_beautify_quad_rotate_calc_ex(const float v1[2],
 
 static float polyedge_rotate_beauty_calc(const float (*coords)[2],
                                          const struct HalfEdge *edges,
-                                         const struct HalfEdge *e_a)
+                                         const struct HalfEdge *e_a,
+                                         float *r_area)
 {
   const struct HalfEdge *e_b = &edges[e_a->e_radial];
 
@@ -205,7 +218,7 @@ static float polyedge_rotate_beauty_calc(const float (*coords)[2],
   v3 = coords[e_b_other->v];
   v4 = coords[e_b->v];
 
-  return BLI_polyfill_beautify_quad_rotate_calc(v1, v2, v3, v4);
+  return BLI_polyfill_beautify_quad_rotate_calc_ex(v1, v2, v3, v4, false, r_area);
 }
 
 static void polyedge_beauty_cost_update_single(const float (*coords)[2],
@@ -216,13 +229,18 @@ static void polyedge_beauty_cost_update_single(const float (*coords)[2],
 {
   const uint i = e->base_index;
   /* recalculate edge */
-  const float cost = polyedge_rotate_beauty_calc(coords, edges, e);
+  float area;
+  const float cost = polyedge_rotate_beauty_calc(coords, edges, e, &area);
   /* We can get cases where both choices generate very small negative costs,
    * which leads to infinite loop. Anyway, costs above that are not worth recomputing,
    * maybe we could even optimize it to a smaller limit?
    * Actually, FLT_EPSILON is too small in some cases, 1e-6f seems to work OK hopefully?
-   * See T43578, T49478. */
-  if (cost < -1e-6f) {
+   * See T43578, T49478.
+   *
+   * In fact a larger epsilon can still fail when the area of the face is very large,
+   * now the epsilon is scaled by the face area.
+   * See T56532. */
+  if (cost < -1e-6f * max_ff(area, 1.0f)) {
     BLI_heap_insert_or_update(eheap, &eheap_table[i], cost, e);
   }
   else {
@@ -381,7 +399,7 @@ void BLI_polyfill_beautify(const float (*coords)[2],
     for (uint i = 0; i < half_edges_len; i++, e++) {
       /* Accounts for boundary edged too (UINT_MAX). */
       if (e->e_radial < i) {
-        const float cost = polyedge_rotate_beauty_calc(coords, half_edges, e);
+        const float cost = polyedge_rotate_beauty_calc(coords, half_edges, e, NULL);
         if (cost < 0.0f) {
           eheap_table[e->base_index] = BLI_heap_insert(eheap, cost, e);
         }
diff --git a/source/blender/blenlib/intern/string.c b/source/blender/blenlib/intern/string.c
index 3b69e257f45..4059e0401c0 100644
--- a/source/blender/blenlib/intern/string.c
+++ b/source/blender/blenlib/intern/string.c
@@ -663,8 +663,11 @@ static int left_number_strcmp(const char *s1, const char *s2, int *tiebreaker)
   return 0;
 }
 
-/* natural string compare, keeping numbers in order */
-int BLI_natstrcmp(const char *s1, const char *s2)
+/**
+ * Case insensitive, *natural* string comparison,
+ * keeping numbers in order.
+ */
+int BLI_strcasecmp_natural(const char *s1, const char *s2)
 {
   register int d1 = 0, d2 = 0;
   register char c1, c2;
@@ -675,10 +678,7 @@ int BLI_natstrcmp(const char *s1, const char *s2)
    * numeric, else do a tolower and char compare */
 
   while (1) {
-    c1 = tolower(s1[d1]);
-    c2 = tolower(s2[d2]);
-
-    if (isdigit(c1) && isdigit(c2)) {
+    if (isdigit(s1[d1]) && isdigit(s2[d2])) {
       int numcompare = left_number_strcmp(s1 + d1, s2 + d2, &tiebreaker);
 
       if (numcompare != 0) {
@@ -693,11 +693,11 @@ int BLI_natstrcmp(const char *s1, const char *s2)
       while (isdigit(s2[d2])) {
         d2++;
       }
-
-      c1 = tolower(s1[d1]);
-      c2 = tolower(s2[d2]);
     }
 
+    c1 = tolower(s1[d1]);
+    c2 = tolower(s2[d2]);
+
     /* first check for '.' so "foo.bar" comes before "foo 1.bar" */
     if (c1 == '.' && c2 != '.') {
       return -1;
diff --git a/source/blender/blenlib/intern/system.c b/source/blender/blenlib/intern/system.c
index 88f2e2625e8..941c2b608e6 100644
--- a/source/blender/blenlib/intern/system.c
+++ b/source/blender/blenlib/intern/system.c
@@ -179,6 +179,19 @@ char *BLI_cpu_brand_string(void)
   return NULL;
 }
 
+int BLI_cpu_support_sse41(void)
+{
+  int result[4], num;
+  __cpuid(result, 0);
+  num = result[0];
+
+  if (num >= 1) {
+    __cpuid(result, 0x00000001);
+    return (result[2] & ((int)1 << 19)) != 0;
+  }
+  return 0;
+}
+
 void BLI_hostname_get(char *buffer, size_t bufsize)
 {
 #ifndef WIN32
diff --git a/source/blender/blenlib/intern/winstuff.c b/source/blender/blenlib/intern/winstuff.c
index 9e6ea0bb6ee..724f27f4667 100644
--- a/source/blender/blenlib/intern/winstuff.c
+++ b/source/blender/blenlib/intern/winstuff.c
@@ -172,17 +172,7 @@ void RegisterBlendExtension(void)
 
   BLI_getInstallationDir(InstallDir);
   GetSystemDirectory(SysDir, FILE_MAXDIR);
-#  ifdef _WIN64
-  ThumbHandlerDLL = "BlendThumb64.dll";
-#  else
-  IsWow64Process(GetCurrentProcess(), &IsWOW64);
-  if (IsWOW64 == true) {
-    ThumbHandlerDLL = "BlendThumb64.dll";
-  }
-  else {
-    ThumbHandlerDLL = "BlendThumb.dll";
-  }
-#  endif
+  ThumbHandlerDLL = "BlendThumb.dll";
   snprintf(
       RegCmd, MAX_PATH * 2, "%s\\regsvr32 /s \"%s\\%s\"", SysDir, InstallDir, ThumbHandlerDLL);
   system(RegCmd);
diff --git a/source/blender/blenloader/BLO_blend_defs.h b/source/blender/blenloader/BLO_blend_defs.h
index 0787d054141..fec61605dca 100644
--- a/source/blender/blenloader/BLO_blend_defs.h
+++ b/source/blender/blenloader/BLO_blend_defs.h
@@ -18,7 +18,7 @@
 
 /** \file
  * \ingroup blenloader
- * \brief defines for blendfile codes
+ * \brief defines for blend-file codes.
  */
 
 /* INTEGER CODES */
diff --git a/source/blender/blenloader/BLO_readfile.h b/source/blender/blenloader/BLO_readfile.h
index 8944dfdb711..adf3bf00d48 100644
--- a/source/blender/blenloader/BLO_readfile.h
+++ b/source/blender/blenloader/BLO_readfile.h
@@ -41,7 +41,6 @@ struct UserDef;
 struct View3D;
 struct ViewLayer;
 struct WorkSpace;
-struct bContext;
 struct bScreen;
 struct wmWindowManager;
 
diff --git a/source/blender/blenloader/intern/readfile.c b/source/blender/blenloader/intern/readfile.c
index 91221ed3740..52d80c0cba1 100644
--- a/source/blender/blenloader/intern/readfile.c
+++ b/source/blender/blenloader/intern/readfile.c
@@ -223,7 +223,7 @@
  * which keeps large arrays in memory from data-blocks we may not even use.
  *
  * \note This is disabled when using compression,
- * while zlib supports seek ist's unusably slow, see: T61880.
+ * while zlib supports seek it's unusably slow, see: T61880.
  */
 #define USE_BHEAD_READ_ON_DEMAND
 
@@ -234,7 +234,7 @@
 #define USE_GHASH_RESTORE_POINTER
 
 /* Define this to have verbose debug prints. */
-#define USE_DEBUG_PRINT
+//#define USE_DEBUG_PRINT
 
 #ifdef USE_DEBUG_PRINT
 #  define DEBUG_PRINTF(...) printf(__VA_ARGS__)
@@ -272,13 +272,13 @@ typedef struct BHeadN {
  * because ID names are used in lookup tables. */
 #define BHEAD_USE_READ_ON_DEMAND(bhead) ((bhead)->code == DATA)
 
-/* this function ensures that reports are printed,
- * in the case of libraray linking errors this is important!
+/**
+ * This function ensures that reports are printed,
+ * in the case of library linking errors this is important!
  *
  * bit kludge but better then doubling up on prints,
  * we could alternatively have a versions of a report function which forces printing - campbell
  */
-
 void blo_reportf_wrap(ReportList *reports, ReportType type, const char *format, ...)
 {
   char fixed_buf[1024]; /* should be long enough */
@@ -534,7 +534,7 @@ static void split_libdata(ListBase *lb_src, Main **lib_main_array, const uint li
 
     if (id->lib) {
       if (((uint)id->lib->temp_index < lib_main_array_len) &&
-          /* this check should never fail, just incase 'id->lib' is a dangling pointer. */
+          /* this check should never fail, just in case 'id->lib' is a dangling pointer. */
           (lib_main_array[id->lib->temp_index]->curlib == id->lib)) {
         Main *mainvar = lib_main_array[id->lib->temp_index];
         ListBase *lb_dst = which_libbase(mainvar, GS(id->name));
@@ -2499,16 +2499,13 @@ static void IDP_DirectLinkProperty(IDProperty *prop, int switch_endian, FileData
       IDP_DirectLinkIDPArray(prop, switch_endian, fd);
       break;
     case IDP_DOUBLE:
-      /* erg, stupid doubles.  since I'm storing them
-       * in the same field as int val; val2 in the
-       * IDPropertyData struct, they have to deal with
-       * endianness specifically
+      /* Workaround for doubles.
+       * They are stored in the same field as `int val, val2` in the IDPropertyData struct,
+       * they have to deal with endianness specifically.
        *
-       * in theory, val and val2 would've already been swapped
+       * In theory, val and val2 would've already been swapped
        * if switch_endian is true, so we have to first unswap
-       * them then reswap them as a single 64-bit entity.
-       */
-
+       * them then re-swap them as a single 64-bit entity. */
       if (switch_endian) {
         BLI_endian_switch_int32(&prop->data.val);
         BLI_endian_switch_int32(&prop->data.val2);
@@ -4899,21 +4896,13 @@ static void lib_link_mesh(FileData *fd, Main *main)
         G_MAIN = gmain;
       }
 
-      /*
-       * Re-tessellate, even if the polys were just created from tessfaces, this
-       * is important because it:
-       * - fill the CD_ORIGINDEX layer
-       * - gives consistency of tessface between loading from a file and
-       *   converting an edited BMesh back into a mesh (i.e. it replaces
-       *   quad tessfaces in a loaded mesh immediately, instead of lazily
-       *   waiting until edit mode has been entered/exited, making it easier
-       *   to recognize problems that would otherwise only show up after edits).
-       */
-#ifdef USE_TESSFACE_DEFAULT
-      BKE_mesh_tessface_calc(me);
-#else
+      /* Deprecated, only kept for conversion. */
       BKE_mesh_tessface_clear(me);
-#endif
+
+      /* Moved from do_versions because we need updated polygons for calculating normals. */
+      if (MAIN_VERSION_OLDER(main, 256, 6)) {
+        BKE_mesh_calc_normals(me);
+      }
 
       me->id.tag &= ~LIB_TAG_NEED_LINK;
     }
@@ -7719,8 +7708,8 @@ static void direct_link_windowmanager(FileData *fd, wmWindowManager *wm)
     win->addmousemove = true;
     win->stereo3d_format = newdataadr(fd, win->stereo3d_format);
 
-    /* multiview always fallback to anaglyph at file opening
-     * otherwise quadbuffer saved files can break Blender */
+    /* Multi-view always fallback to anaglyph at file opening
+     * otherwise quad-buffer saved files can break Blender. */
     if (win->stereo3d_format) {
       win->stereo3d_format->display_mode = S3D_DISPLAY_ANAGLYPH;
     }
@@ -9225,7 +9214,7 @@ static BHead *read_libblock(FileData *fd, Main *main, BHead *bhead, const int ta
    * flags dependency graph does not do animation update to avoid loss of unkeyed changes.,
    * which conflicts with undo/redo of changes to animation data itself.
    *
-   * But for regular file load we clear the flag, since the flags might have been changed sinde
+   * But for regular file load we clear the flag, since the flags might have been changed since
    * the version the file has been saved with. */
   if (!fd->memfile) {
     id->recalc = 0;
@@ -11314,7 +11303,7 @@ static void add_collections_to_scene(Main *mainvar,
     }
     /* We do not want to force instantiation of indirectly linked collections,
      * not even when appending. Users can now easily instantiate collections (and their objects)
-     * as needed by themsleves. See T67032. */
+     * as needed by themselves. See T67032. */
     else if ((collection->id.tag & LIB_TAG_INDIRECT) == 0) {
       bool do_add_collection = (collection->id.tag & LIB_TAG_DOIT) != 0;
       if (!do_add_collection) {
diff --git a/source/blender/blenloader/intern/versioning_250.c b/source/blender/blenloader/intern/versioning_250.c
index 6a5cfc45628..ba4dc2d33c7 100644
--- a/source/blender/blenloader/intern/versioning_250.c
+++ b/source/blender/blenloader/intern/versioning_250.c
@@ -2080,11 +2080,12 @@ void blo_do_versions_250(FileData *fd, Library *lib, Main *bmain)
     }
   }
 
-  if (bmain->versionfile < 256 || (bmain->versionfile == 256 && bmain->subversionfile < 6)) {
-    Mesh *me;
-
-    for (me = bmain->meshes.first; me; me = me->id.next) {
-      BKE_mesh_calc_normals_tessface(me->mvert, me->totvert, me->mface, me->totface, NULL);
+  if (0) {
+    if (bmain->versionfile < 256 || (bmain->versionfile == 256 && bmain->subversionfile < 6)) {
+      for (Mesh *me = bmain->meshes.first; me; me = me->id.next) {
+        /* Vertex normal calculation from legacy 'MFace' has been removed.
+         * update after calculating polygons in file reading code instead. */
+      }
     }
   }
 
diff --git a/source/blender/blenloader/intern/versioning_270.c b/source/blender/blenloader/intern/versioning_270.c
index b61692799ed..5010cc8071d 100644
--- a/source/blender/blenloader/intern/versioning_270.c
+++ b/source/blender/blenloader/intern/versioning_270.c
@@ -1748,7 +1748,7 @@ void blo_do_versions_270(FileData *fd, Library *UNUSED(lib), Main *bmain)
       for (Brush *br = bmain->brushes.first; br; br = br->id.next) {
         br->falloff_angle = DEG2RADF(80);
         br->flag &= ~(BRUSH_FLAG_UNUSED_1 | BRUSH_FLAG_UNUSED_6 | BRUSH_FLAG_UNUSED_7 |
-                      BRUSH_FLAG_UNUSED_17 | BRUSH_FRONTFACE_FALLOFF);
+                      BRUSH_SCENE_SPACING | BRUSH_FRONTFACE_FALLOFF);
       }
 
       for (Scene *scene = bmain->scenes.first; scene; scene = scene->id.next) {
diff --git a/source/blender/blenloader/intern/versioning_280.c b/source/blender/blenloader/intern/versioning_280.c
index 1d2490229da..51c8a0d2a72 100644
--- a/source/blender/blenloader/intern/versioning_280.c
+++ b/source/blender/blenloader/intern/versioning_280.c
@@ -125,8 +125,8 @@ static void do_version_workspaces_create_from_screens(Main *bmain)
     }
 
     if (screen_parent) {
-      /* fullscreen with "Back to Previous" option, don't create
-       * a new workspace, add layout workspace containing parent */
+      /* Full-screen with "Back to Previous" option, don't create
+       * a new workspace, add layout workspace containing parent. */
       workspace = BLI_findstring(
           &bmain->workspaces, screen_parent->id.name + 2, offsetof(ID, name) + 2);
     }
@@ -406,7 +406,7 @@ static void do_version_layers_to_collections(Main *bmain, Scene *scene)
   }
 
   /* Create collections from layers. */
-  Collection *collection_master = BKE_collection_master(scene);
+  Collection *collection_master = scene->master_collection;
   Collection *collections[20] = {NULL};
 
   for (int layer = 0; layer < 20; layer++) {
@@ -735,13 +735,9 @@ static void do_versions_seq_alloc_transform_and_crop(ListBase *seqbase)
 }
 
 /* Return true if there is something to convert. */
-static bool do_versions_material_convert_legacy_blend_mode(bNodeTree *ntree,
-                                                           char blend_method,
-                                                           GSet *nodegrp_tree_set,
-                                                           GSet *nooutput_tree_set)
+static void do_versions_material_convert_legacy_blend_mode(bNodeTree *ntree, char blend_method)
 {
   bool need_update = false;
-  bool do_conversion = false;
 
   /* Iterate backwards from end so we don't encounter newly added links. */
   bNodeLink *prevlink;
@@ -754,31 +750,6 @@ static bool do_versions_material_convert_legacy_blend_mode(bNodeTree *ntree,
     bNode *tonode = link->tonode;
     bNodeSocket *tosock = link->tosock;
 
-    if (nodegrp_tree_set) {
-      if (fromnode->type == NODE_GROUP && fromnode->id != NULL) {
-        bNodeTree *group_ntree = (bNodeTree *)fromnode->id;
-        if (BLI_gset_add(nodegrp_tree_set, group_ntree)) {
-          /* Recursive but not convert (blend_method = -1). Conversion happens after. */
-          if (!do_versions_material_convert_legacy_blend_mode(
-                  group_ntree, -1, nodegrp_tree_set, nooutput_tree_set)) {
-            /* There is no output to convert in the tree. */
-            BLI_gset_add(nooutput_tree_set, group_ntree);
-          }
-        }
-      }
-      if (tonode->type == NODE_GROUP && tonode->id != NULL) {
-        bNodeTree *group_ntree = (bNodeTree *)tonode->id;
-        if (BLI_gset_add(nodegrp_tree_set, group_ntree)) {
-          /* Recursive but not convert (blend_method = -1). Conversion happens after. */
-          if (!do_versions_material_convert_legacy_blend_mode(
-                  group_ntree, -1, nodegrp_tree_set, nooutput_tree_set)) {
-            /* There is no output to convert in the tree. */
-            BLI_gset_add(nooutput_tree_set, group_ntree);
-          }
-        }
-      }
-    }
-
     if (!(tonode->type == SH_NODE_OUTPUT_MATERIAL && STREQ(tosock->identifier, "Surface"))) {
       continue;
     }
@@ -788,8 +759,6 @@ static bool do_versions_material_convert_legacy_blend_mode(bNodeTree *ntree,
       continue;
     }
 
-    do_conversion = true;
-
     if (blend_method == 1 /* MA_BM_ADD */) {
       nodeRemLink(ntree, link);
 
@@ -854,11 +823,9 @@ static bool do_versions_material_convert_legacy_blend_mode(bNodeTree *ntree,
   if (need_update) {
     ntreeUpdateTree(NULL, ntree);
   }
-
-  return do_conversion;
 }
 
-void do_versions_after_linking_280(Main *bmain, ReportList *reports)
+void do_versions_after_linking_280(Main *bmain, ReportList *UNUSED(reports))
 {
   bool use_collection_compat_28 = true;
 
@@ -1259,74 +1226,20 @@ void do_versions_after_linking_280(Main *bmain, ReportList *reports)
      * now that we use dualsource blending. */
     /* We take care of doing only nodetrees that are always part of materials
      * with old blending modes. */
-    GSet *ntrees_additive = BLI_gset_new(BLI_ghashutil_ptrhash, BLI_ghashutil_ptrcmp, __func__);
-    GSet *ntrees_multiply = BLI_gset_new(BLI_ghashutil_ptrhash, BLI_ghashutil_ptrcmp, __func__);
-    GSet *ntrees_nolegacy = BLI_gset_new(BLI_ghashutil_ptrhash, BLI_ghashutil_ptrcmp, __func__);
-    GSet *ntrees_nooutput = BLI_gset_new(BLI_ghashutil_ptrhash, BLI_ghashutil_ptrcmp, __func__);
     for (Material *ma = bmain->materials.first; ma; ma = ma->id.next) {
       bNodeTree *ntree = ma->nodetree;
       if (ma->blend_method == 1 /* MA_BM_ADD */) {
         if (ma->use_nodes) {
-          do_versions_material_convert_legacy_blend_mode(
-              ntree, ma->blend_method, ntrees_additive, ntrees_nooutput);
+          do_versions_material_convert_legacy_blend_mode(ntree, 1 /* MA_BM_ADD */);
         }
         ma->blend_method = MA_BM_BLEND;
       }
       else if (ma->blend_method == 2 /* MA_BM_MULTIPLY */) {
         if (ma->use_nodes) {
-          do_versions_material_convert_legacy_blend_mode(
-              ntree, ma->blend_method, ntrees_multiply, ntrees_nooutput);
+          do_versions_material_convert_legacy_blend_mode(ntree, 2 /* MA_BM_MULTIPLY */);
         }
         ma->blend_method = MA_BM_BLEND;
       }
-      else {
-        /* Still tag the group nodes as not using legacy blend modes. */
-        if (ma->use_nodes) {
-          do_versions_material_convert_legacy_blend_mode(
-              ntree, -1, ntrees_nolegacy, ntrees_nooutput);
-        }
-      }
-    }
-    GHashIterState iter = {0};
-    bNodeTree *ntree;
-    /* Remove trees that have no output nodes.
-     * This is done separately to avoid infinite recursion. */
-    while (BLI_gset_pop(ntrees_nooutput, (GSetIterState *)&iter, (void **)&ntree)) {
-      BLI_gset_remove(ntrees_additive, ntree, NULL);
-      BLI_gset_remove(ntrees_multiply, ntree, NULL);
-      BLI_gset_remove(ntrees_nolegacy, ntree, NULL);
-    }
-    BLI_gset_free(ntrees_nooutput, NULL);
-    /* Remove group nodetree that are used by material using non-legacy blend mode. */
-    GHashIterState iter_rm = {0};
-    bool error = false;
-    while (BLI_gset_pop(ntrees_nolegacy, (GSetIterState *)&iter_rm, (void **)&ntree)) {
-      if (BLI_gset_remove(ntrees_additive, ntree, NULL)) {
-        error = true;
-      }
-      if (BLI_gset_remove(ntrees_multiply, ntree, NULL)) {
-        error = true;
-      }
-    }
-    BLI_gset_free(ntrees_nolegacy, NULL);
-    /* Convert remaining group nodetree. */
-    GHashIterState iter_add = {0};
-    GHashIterState iter_mul = {0};
-    while (BLI_gset_pop(ntrees_additive, (GSetIterState *)&iter_add, (void **)&ntree)) {
-      do_versions_material_convert_legacy_blend_mode(ntree, 1 /* MA_BM_ADD */, NULL, NULL);
-    }
-    while (BLI_gset_pop(ntrees_multiply, (GSetIterState *)&iter_mul, (void **)&ntree)) {
-      do_versions_material_convert_legacy_blend_mode(ntree, 2 /* MA_BM_MULTIPLY */, NULL, NULL);
-    }
-    BLI_gset_free(ntrees_additive, NULL);
-    BLI_gset_free(ntrees_multiply, NULL);
-
-    if (error) {
-      BKE_report(reports, RPT_ERROR, "Eevee material conversion problem. Error in console");
-      printf(
-          "One or more group nodetrees containing a material output were found"
-          " in both a material using deprecated blend mode and a normal one.\n"
-          "Nothing in these nodetrees was changed and manual update is required.\n");
     }
   }
 }
@@ -3792,8 +3705,7 @@ void blo_do_versions_280(FileData *fd, Library *UNUSED(lib), Main *bmain)
     }
   }
 
-  {
-    /* Versioning code until next subversion bump goes here. */
+  if (!MAIN_VERSION_ATLEAST(bmain, 281, 3)) {
     if (U.view_rotate_sensitivity_turntable == 0) {
       U.view_rotate_sensitivity_turntable = DEG2RADF(0.4f);
       U.view_rotate_sensitivity_trackball = 1.0f;
@@ -3808,6 +3720,12 @@ void blo_do_versions_280(FileData *fd, Library *UNUSED(lib), Main *bmain)
               ar->alignment = RGN_ALIGN_RIGHT;
             }
           }
+          /* Mark outliners as dirty for syncing and enable synced selection */
+          if (sl->spacetype == SPACE_OUTLINER) {
+            SpaceOutliner *soutliner = (SpaceOutliner *)sl;
+            soutliner->sync_select_dirty |= WM_OUTLINER_SYNC_SELECT_FROM_ALL;
+            soutliner->flag |= SO_SYNC_SELECT;
+          }
         }
       }
     }
@@ -3817,4 +3735,69 @@ void blo_do_versions_280(FileData *fd, Library *UNUSED(lib), Main *bmain)
       }
     }
   }
+
+  if (!MAIN_VERSION_ATLEAST(bmain, 281, 4)) {
+    ID *id;
+    FOREACH_MAIN_ID_BEGIN (bmain, id) {
+      bNodeTree *ntree = ntreeFromID(id);
+      if (ntree) {
+        ntree->id.flag |= LIB_PRIVATE_DATA;
+      }
+    }
+    FOREACH_MAIN_ID_END;
+  }
+
+  if (!MAIN_VERSION_ATLEAST(bmain, 281, 5)) {
+    for (Brush *br = bmain->brushes.first; br; br = br->id.next) {
+      if (br->ob_mode & OB_MODE_SCULPT && br->normal_radius_factor == 0.0f) {
+        br->normal_radius_factor = 0.5f;
+      }
+    }
+
+    LISTBASE_FOREACH (Scene *, scene, &bmain->scenes) {
+      /* Older files do not have a master collection, which is then added through
+       * `BKE_collection_master_add()`, so everything is fine. */
+      if (scene->master_collection != NULL) {
+        scene->master_collection->id.flag |= LIB_PRIVATE_DATA;
+      }
+    }
+  }
+
+  {
+    /* Versioning code until next subversion bump goes here. */
+
+    for (bScreen *screen = bmain->screens.first; screen; screen = screen->id.next) {
+      for (ScrArea *sa = screen->areabase.first; sa; sa = sa->next) {
+        for (SpaceLink *sl = sa->spacedata.first; sl; sl = sl->next) {
+          if (sl->spacetype == SPACE_FILE) {
+            SpaceFile *sfile = (SpaceFile *)sl;
+            ListBase *regionbase = (sl == sa->spacedata.first) ? &sa->regionbase : &sl->regionbase;
+            ARegion *ar_ui = do_versions_find_region(regionbase, RGN_TYPE_UI);
+            ARegion *ar_header = do_versions_find_region(regionbase, RGN_TYPE_HEADER);
+            ARegion *ar_toolprops = do_versions_find_region_or_null(regionbase,
+                                                                    RGN_TYPE_TOOL_PROPS);
+
+            /* Reinsert UI region so that it spawns entire area width */
+            BLI_remlink(regionbase, ar_ui);
+            BLI_insertlinkafter(regionbase, ar_header, ar_ui);
+
+            ar_ui->flag |= RGN_FLAG_DYNAMIC_SIZE;
+
+            if (ar_toolprops && (ar_toolprops->alignment == (RGN_ALIGN_BOTTOM | RGN_SPLIT_PREV))) {
+              SpaceType *stype = BKE_spacetype_from_id(sl->spacetype);
+
+              /* Remove empty region at old location. */
+              BLI_assert(sfile->op == NULL);
+              BKE_area_region_free(stype, ar_toolprops);
+              BLI_freelinkN(regionbase, ar_toolprops);
+            }
+
+            if (sfile->params) {
+              sfile->params->details_flags |= FILE_DETAILS_SIZE | FILE_DETAILS_DATETIME;
+            }
+          }
+        }
+      }
+    }
+  }
 }
diff --git a/source/blender/blenloader/intern/versioning_cycles.c b/source/blender/blenloader/intern/versioning_cycles.c
index cebe15e2719..a3e9b8dc206 100644
--- a/source/blender/blenloader/intern/versioning_cycles.c
+++ b/source/blender/blenloader/intern/versioning_cycles.c
@@ -26,6 +26,7 @@
 
 #include "BLI_math.h"
 #include "BLI_string.h"
+#include "BLI_listbase.h"
 #include "BLI_utildefines.h"
 
 #include "DNA_color_types.h"
@@ -44,6 +45,11 @@
 #include "BLO_readfile.h"
 #include "readfile.h"
 
+static bool socket_is_used(bNodeSocket *sock)
+{
+  return sock->flag & SOCK_IN_USE;
+}
+
 static float *cycles_node_socket_float_value(bNodeSocket *socket)
 {
   bNodeSocketValueFloat *socket_data = socket->default_value;
@@ -56,6 +62,12 @@ static float *cycles_node_socket_rgba_value(bNodeSocket *socket)
   return socket_data->value;
 }
 
+static float *cycles_node_socket_vector_value(bNodeSocket *socket)
+{
+  bNodeSocketValueVector *socket_data = socket->default_value;
+  return socket_data->value;
+}
+
 static IDProperty *cycles_properties_from_ID(ID *id)
 {
   IDProperty *idprop = IDP_GetProperties(id, false);
@@ -185,7 +197,7 @@ static void square_roughness_node_insert(bNodeTree *ntree)
 
     /* Add sqrt node. */
     bNode *node = nodeAddStaticNode(NULL, ntree, SH_NODE_MATH);
-    node->custom1 = NODE_MATH_POW;
+    node->custom1 = NODE_MATH_POWER;
     node->locx = 0.5f * (fromnode->locx + tonode->locx);
     node->locy = 0.5f * (fromnode->locy + tonode->locy);
 
@@ -385,6 +397,365 @@ static void light_emission_unify(Light *light, const char *engine)
   }
 }
 
+/* The B input of the Math node is no longer used for single-operand operators.
+ * Previously, if the B input was linked and the A input was not, the B input
+ * was used as the input of the operator. To correct this, we move the link
+ * from B to A if B is linked and A is not.
+ */
+static void update_math_node_single_operand_operators(bNodeTree *ntree)
+{
+  bool need_update = false;
+
+  for (bNode *node = ntree->nodes.first; node; node = node->next) {
+    if (node->type == SH_NODE_MATH) {
+      if (ELEM(node->custom1,
+               NODE_MATH_SQRT,
+               NODE_MATH_CEIL,
+               NODE_MATH_SINE,
+               NODE_MATH_ROUND,
+               NODE_MATH_FLOOR,
+               NODE_MATH_COSINE,
+               NODE_MATH_ARCSINE,
+               NODE_MATH_TANGENT,
+               NODE_MATH_ABSOLUTE,
+               NODE_MATH_FRACTION,
+               NODE_MATH_ARCCOSINE,
+               NODE_MATH_ARCTANGENT)) {
+        bNodeSocket *sockA = BLI_findlink(&node->inputs, 0);
+        bNodeSocket *sockB = BLI_findlink(&node->inputs, 1);
+        if (!sockA->link && sockB->link) {
+          nodeAddLink(ntree, sockB->link->fromnode, sockB->link->fromsock, node, sockA);
+          nodeRemLink(ntree, sockB->link);
+          need_update = true;
+        }
+      }
+    }
+  }
+
+  if (need_update) {
+    ntreeUpdateTree(NULL, ntree);
+  }
+}
+
+/* The Value output of the Vector Math node is no longer available in the Add
+ * and Subtract operators. Previously, this Value output was computed from the
+ * Vector output V as follows:
+ *
+ *   Value = (abs(V.x) + abs(V.y) + abs(V.z)) / 3
+ *
+ * Or more compactly using vector operators:
+ *
+ *   Value = dot(abs(V), (1 / 3, 1 / 3, 1 / 3))
+ *
+ * To correct this, if the Value output was used, we are going to compute
+ * it using the second equation by adding an absolute and a dot node, and
+ * then connect them appropriately.
+ */
+static void update_vector_math_node_add_and_subtract_operators(bNodeTree *ntree)
+{
+  bool need_update = false;
+
+  for (bNode *node = ntree->nodes.first; node; node = node->next) {
+    if (node->type == SH_NODE_VECTOR_MATH) {
+      bNodeSocket *sockOutValue = nodeFindSocket(node, SOCK_OUT, "Value");
+      if (socket_is_used(sockOutValue) &&
+          ELEM(node->custom1, NODE_VECTOR_MATH_ADD, NODE_VECTOR_MATH_SUBTRACT)) {
+
+        bNode *absNode = nodeAddStaticNode(NULL, ntree, SH_NODE_VECTOR_MATH);
+        absNode->custom1 = NODE_VECTOR_MATH_ABSOLUTE;
+        absNode->locx = node->locx + node->width + 20.0f;
+        absNode->locy = node->locy;
+
+        bNode *dotNode = nodeAddStaticNode(NULL, ntree, SH_NODE_VECTOR_MATH);
+        dotNode->custom1 = NODE_VECTOR_MATH_DOT_PRODUCT;
+        dotNode->locx = absNode->locx + absNode->width + 20.0f;
+        dotNode->locy = absNode->locy;
+        bNodeSocket *sockDotB = BLI_findlink(&dotNode->inputs, 1);
+        bNodeSocket *sockDotOutValue = nodeFindSocket(dotNode, SOCK_OUT, "Value");
+        copy_v3_fl(cycles_node_socket_vector_value(sockDotB), 1 / 3.0f);
+
+        LISTBASE_FOREACH_BACKWARD_MUTABLE (bNodeLink *, link, &ntree->links) {
+          if (link->fromsock == sockOutValue) {
+            nodeAddLink(ntree, dotNode, sockDotOutValue, link->tonode, link->tosock);
+            nodeRemLink(ntree, link);
+          }
+        }
+
+        bNodeSocket *sockAbsA = BLI_findlink(&absNode->inputs, 0);
+        bNodeSocket *sockDotA = BLI_findlink(&dotNode->inputs, 0);
+        bNodeSocket *sockOutVector = nodeFindSocket(node, SOCK_OUT, "Vector");
+        bNodeSocket *sockAbsOutVector = nodeFindSocket(absNode, SOCK_OUT, "Vector");
+
+        nodeAddLink(ntree, node, sockOutVector, absNode, sockAbsA);
+        nodeAddLink(ntree, absNode, sockAbsOutVector, dotNode, sockDotA);
+
+        need_update = true;
+      }
+    }
+  }
+
+  if (need_update) {
+    ntreeUpdateTree(NULL, ntree);
+  }
+}
+
+/* The Vector output of the Vector Math node is no longer available in the Dot
+ * Product operator. Previously, this Vector was always zero initialized. To
+ * correct this, we zero out any socket the Vector Output was connected to.
+ */
+static void update_vector_math_node_dot_product_operator(bNodeTree *ntree)
+{
+  bool need_update = false;
+
+  for (bNode *node = ntree->nodes.first; node; node = node->next) {
+    if (node->type == SH_NODE_VECTOR_MATH) {
+      bNodeSocket *sockOutVector = nodeFindSocket(node, SOCK_OUT, "Vector");
+      if (socket_is_used(sockOutVector) && node->custom1 == NODE_VECTOR_MATH_DOT_PRODUCT) {
+        LISTBASE_FOREACH_MUTABLE (bNodeLink *, link, &ntree->links) {
+          if (link->fromsock == sockOutVector) {
+            switch (link->tosock->type) {
+              case SOCK_FLOAT:
+                *cycles_node_socket_float_value(link->tosock) = 0.0f;
+                break;
+              case SOCK_VECTOR:
+                copy_v3_fl(cycles_node_socket_vector_value(link->tosock), 0.0f);
+                break;
+              case SOCK_RGBA:
+                copy_v4_fl(cycles_node_socket_rgba_value(link->tosock), 0.0f);
+                break;
+            }
+            nodeRemLink(ntree, link);
+          }
+        }
+        need_update = true;
+      }
+    }
+  }
+
+  if (need_update) {
+    ntreeUpdateTree(NULL, ntree);
+  }
+}
+
+/* Previously, the Vector output of the cross product operator was normalized.
+ * To correct this, a Normalize node is added to normalize the output if used.
+ * Moreover, the Value output was removed. This Value was equal to the length
+ * of the cross product. To correct this, a Length node is added if needed.
+ */
+static void update_vector_math_node_cross_product_operator(bNodeTree *ntree)
+{
+  bool need_update = false;
+
+  for (bNode *node = ntree->nodes.first; node; node = node->next) {
+    if (node->type == SH_NODE_VECTOR_MATH) {
+      if (node->custom1 == NODE_VECTOR_MATH_CROSS_PRODUCT) {
+        bNodeSocket *sockOutVector = nodeFindSocket(node, SOCK_OUT, "Vector");
+        if (socket_is_used(sockOutVector)) {
+          bNode *normalizeNode = nodeAddStaticNode(NULL, ntree, SH_NODE_VECTOR_MATH);
+          normalizeNode->custom1 = NODE_VECTOR_MATH_NORMALIZE;
+          normalizeNode->locx = node->locx + node->width + 20.0f;
+          normalizeNode->locy = node->locy;
+          bNodeSocket *sockNormalizeOut = nodeFindSocket(normalizeNode, SOCK_OUT, "Vector");
+
+          LISTBASE_FOREACH_BACKWARD_MUTABLE (bNodeLink *, link, &ntree->links) {
+            if (link->fromsock == sockOutVector) {
+              nodeAddLink(ntree, normalizeNode, sockNormalizeOut, link->tonode, link->tosock);
+              nodeRemLink(ntree, link);
+            }
+          }
+          bNodeSocket *sockNormalizeA = BLI_findlink(&normalizeNode->inputs, 0);
+          nodeAddLink(ntree, node, sockOutVector, normalizeNode, sockNormalizeA);
+
+          need_update = true;
+        }
+
+        bNodeSocket *sockOutValue = nodeFindSocket(node, SOCK_OUT, "Value");
+        if (socket_is_used(sockOutValue)) {
+          bNode *lengthNode = nodeAddStaticNode(NULL, ntree, SH_NODE_VECTOR_MATH);
+          lengthNode->custom1 = NODE_VECTOR_MATH_LENGTH;
+          lengthNode->locx = node->locx + node->width + 20.0f;
+          if (socket_is_used(sockOutVector)) {
+            lengthNode->locy = node->locy - lengthNode->height - 20.0f;
+          }
+          else {
+            lengthNode->locy = node->locy;
+          }
+          bNodeSocket *sockLengthOut = nodeFindSocket(lengthNode, SOCK_OUT, "Value");
+
+          LISTBASE_FOREACH_BACKWARD_MUTABLE (bNodeLink *, link, &ntree->links) {
+            if (link->fromsock == sockOutValue) {
+              nodeAddLink(ntree, lengthNode, sockLengthOut, link->tonode, link->tosock);
+              nodeRemLink(ntree, link);
+            }
+          }
+          bNodeSocket *sockLengthA = BLI_findlink(&lengthNode->inputs, 0);
+          nodeAddLink(ntree, node, sockOutVector, lengthNode, sockLengthA);
+
+          need_update = true;
+        }
+      }
+    }
+  }
+
+  if (need_update) {
+    ntreeUpdateTree(NULL, ntree);
+  }
+}
+
+/* The Value output of the Vector Math node is no longer available in the
+ * Normalize operator. This Value output was equal to the length of the
+ * the input vector A. To correct this, we either add a Length node or
+ * convert the Normalize node into a Length node, depending on if the
+ * Vector output is needed.
+ */
+static void update_vector_math_node_normalize_operator(bNodeTree *ntree)
+{
+  bool need_update = false;
+
+  for (bNode *node = ntree->nodes.first; node; node = node->next) {
+    if (node->type == SH_NODE_VECTOR_MATH) {
+      bNodeSocket *sockOutValue = nodeFindSocket(node, SOCK_OUT, "Value");
+      if (node->custom1 == NODE_VECTOR_MATH_NORMALIZE && socket_is_used(sockOutValue)) {
+        bNodeSocket *sockOutVector = nodeFindSocket(node, SOCK_OUT, "Vector");
+        if (socket_is_used(sockOutVector)) {
+          bNode *lengthNode = nodeAddStaticNode(NULL, ntree, SH_NODE_VECTOR_MATH);
+          lengthNode->custom1 = NODE_VECTOR_MATH_LENGTH;
+          lengthNode->locx = node->locx + node->width + 20.0f;
+          lengthNode->locy = node->locy;
+          bNodeSocket *sockLengthValue = nodeFindSocket(lengthNode, SOCK_OUT, "Value");
+
+          LISTBASE_FOREACH_BACKWARD_MUTABLE (bNodeLink *, link, &ntree->links) {
+            if (link->fromsock == sockOutValue) {
+              nodeAddLink(ntree, lengthNode, sockLengthValue, link->tonode, link->tosock);
+              nodeRemLink(ntree, link);
+            }
+          }
+          bNodeSocket *sockA = BLI_findlink(&node->inputs, 0);
+          bNodeSocket *sockLengthA = BLI_findlink(&lengthNode->inputs, 0);
+          if (sockA->link) {
+            bNodeLink *link = sockA->link;
+            nodeAddLink(ntree, link->fromnode, link->fromsock, lengthNode, sockLengthA);
+          }
+          else {
+            copy_v3_v3(cycles_node_socket_vector_value(sockLengthA),
+                       cycles_node_socket_vector_value(sockA));
+          }
+
+          need_update = true;
+        }
+        else {
+          node->custom1 = NODE_VECTOR_MATH_LENGTH;
+        }
+      }
+    }
+  }
+  if (need_update) {
+    ntreeUpdateTree(NULL, ntree);
+  }
+}
+
+/* The Vector Math operator types didn't have an enum, but rather, their
+ * values were hard coded into the code. After the enum was created and
+ * after more vector operators were added, the hard coded values needs
+ * to be remapped to their correct enum values. To fix this, we remap
+ * the values according to the following rules:
+ *
+ * Dot Product Operator : 3 -> 7
+ * Normalize Operator   : 5 -> 11
+ *
+ * Additionally, since the Average operator was removed, it is assigned
+ * a value of -1 just to be identified later in the versioning code:
+ *
+ * Average Operator : 2 -> -1
+ *
+ */
+static void update_vector_math_node_operators_enum_mapping(bNodeTree *ntree)
+{
+  for (bNode *node = ntree->nodes.first; node; node = node->next) {
+    if (node->type == SH_NODE_VECTOR_MATH) {
+      switch (node->custom1) {
+        case 2:
+          node->custom1 = -1;
+          break;
+        case 3:
+          node->custom1 = 7;
+          break;
+        case 5:
+          node->custom1 = 11;
+          break;
+      }
+    }
+  }
+}
+
+/* The Average operator is no longer available in the Vector Math node.
+ * The Vector output was equal to the normalized sum of input vectors while
+ * the Value output was equal to the length of the sum of input vectors.
+ * To correct this, we convert the node into an Add node and add a length
+ * node or a normalize node if needed.
+ */
+static void update_vector_math_node_average_operator(bNodeTree *ntree)
+{
+  bool need_update = false;
+
+  for (bNode *node = ntree->nodes.first; node; node = node->next) {
+    if (node->type == SH_NODE_VECTOR_MATH) {
+      /* See update_vector_math_node_operators_enum_mapping. */
+      if (node->custom1 == -1) {
+        node->custom1 = NODE_VECTOR_MATH_ADD;
+        bNodeSocket *sockOutVector = nodeFindSocket(node, SOCK_OUT, "Vector");
+        if (socket_is_used(sockOutVector)) {
+          bNode *normalizeNode = nodeAddStaticNode(NULL, ntree, SH_NODE_VECTOR_MATH);
+          normalizeNode->custom1 = NODE_VECTOR_MATH_NORMALIZE;
+          normalizeNode->locx = node->locx + node->width + 20.0f;
+          normalizeNode->locy = node->locy;
+          bNodeSocket *sockNormalizeOut = nodeFindSocket(normalizeNode, SOCK_OUT, "Vector");
+
+          LISTBASE_FOREACH_BACKWARD_MUTABLE (bNodeLink *, link, &ntree->links) {
+            if (link->fromsock == sockOutVector) {
+              nodeAddLink(ntree, normalizeNode, sockNormalizeOut, link->tonode, link->tosock);
+              nodeRemLink(ntree, link);
+            }
+          }
+          bNodeSocket *sockNormalizeA = BLI_findlink(&normalizeNode->inputs, 0);
+          nodeAddLink(ntree, node, sockOutVector, normalizeNode, sockNormalizeA);
+
+          need_update = true;
+        }
+
+        bNodeSocket *sockOutValue = nodeFindSocket(node, SOCK_OUT, "Value");
+        if (socket_is_used(sockOutValue)) {
+          bNode *lengthNode = nodeAddStaticNode(NULL, ntree, SH_NODE_VECTOR_MATH);
+          lengthNode->custom1 = NODE_VECTOR_MATH_LENGTH;
+          lengthNode->locx = node->locx + node->width + 20.0f;
+          if (socket_is_used(sockOutVector)) {
+            lengthNode->locy = node->locy - lengthNode->height - 20.0f;
+          }
+          else {
+            lengthNode->locy = node->locy;
+          }
+          bNodeSocket *sockLengthOut = nodeFindSocket(lengthNode, SOCK_OUT, "Value");
+
+          LISTBASE_FOREACH_BACKWARD_MUTABLE (bNodeLink *, link, &ntree->links) {
+            if (link->fromsock == sockOutValue) {
+              nodeAddLink(ntree, lengthNode, sockLengthOut, link->tonode, link->tosock);
+              nodeRemLink(ntree, link);
+            }
+          }
+          bNodeSocket *sockLengthA = BLI_findlink(&lengthNode->inputs, 0);
+          nodeAddLink(ntree, node, sockOutVector, lengthNode, sockLengthA);
+
+          need_update = true;
+        }
+      }
+    }
+  }
+
+  if (need_update) {
+    ntreeUpdateTree(NULL, ntree);
+  }
+}
+
 void blo_do_versions_cycles(FileData *UNUSED(fd), Library *UNUSED(lib), Main *bmain)
 {
   /* Particle shape shared with Eevee. */
@@ -417,6 +788,15 @@ void blo_do_versions_cycles(FileData *UNUSED(fd), Library *UNUSED(lib), Main *bm
       }
     }
   }
+
+  if (!MAIN_VERSION_ATLEAST(bmain, 281, 3)) {
+    FOREACH_NODETREE_BEGIN (bmain, ntree, id) {
+      if (ntree->type == NTREE_SHADER) {
+        update_vector_math_node_operators_enum_mapping(ntree);
+      }
+    }
+    FOREACH_NODETREE_END;
+  }
 }
 
 void do_versions_after_linking_cycles(Main *bmain)
@@ -526,4 +906,26 @@ void do_versions_after_linking_cycles(Main *bmain)
       }
     }
   }
+
+  if (!MAIN_VERSION_ATLEAST(bmain, 281, 2)) {
+    FOREACH_NODETREE_BEGIN (bmain, ntree, id) {
+      if (ntree->type == NTREE_SHADER) {
+        update_math_node_single_operand_operators(ntree);
+      }
+    }
+    FOREACH_NODETREE_END;
+  }
+
+  if (!MAIN_VERSION_ATLEAST(bmain, 281, 3)) {
+    FOREACH_NODETREE_BEGIN (bmain, ntree, id) {
+      if (ntree->type == NTREE_SHADER) {
+        update_vector_math_node_add_and_subtract_operators(ntree);
+        update_vector_math_node_dot_product_operator(ntree);
+        update_vector_math_node_cross_product_operator(ntree);
+        update_vector_math_node_normalize_operator(ntree);
+        update_vector_math_node_average_operator(ntree);
+      }
+    }
+    FOREACH_NODETREE_END;
+  }
 }
diff --git a/source/blender/blenloader/intern/versioning_defaults.c b/source/blender/blenloader/intern/versioning_defaults.c
index fa69892584a..fa22fd46068 100644
--- a/source/blender/blenloader/intern/versioning_defaults.c
+++ b/source/blender/blenloader/intern/versioning_defaults.c
@@ -329,6 +329,14 @@ void BLO_update_defaults_startup_blend(Main *bmain, const char *app_template)
       BLI_strncpy(screen->id.name + 2, workspace->id.name + 2, sizeof(screen->id.name) - 2);
       BLI_libblock_ensure_unique_name(bmain, screen->id.name);
     }
+
+    /* For some reason we have unused screens, needed until re-saving.
+     * Clear unused layouts because they're visible in the outliner & Python API. */
+    LISTBASE_FOREACH_MUTABLE (WorkSpaceLayout *, layout_iter, &workspace->layouts) {
+      if (layout != layout_iter) {
+        BKE_workspace_layout_remove(bmain, workspace, layout_iter);
+      }
+    }
   }
 
   /* Scenes */
diff --git a/source/blender/blenloader/intern/versioning_userdef.c b/source/blender/blenloader/intern/versioning_userdef.c
index e987a623d0b..67b8ce1d0de 100644
--- a/source/blender/blenloader/intern/versioning_userdef.c
+++ b/source/blender/blenloader/intern/versioning_userdef.c
@@ -141,10 +141,16 @@ static void do_versions_theme(const UserDef *userdef, bTheme *btheme)
     FROM_DEFAULT_V4_UCHAR(space_outliner.row_alternate);
   }
 
+  if (!USER_VERSION_ATLEAST(281, 3)) {
+    FROM_DEFAULT_V4_UCHAR(space_outliner.selected_highlight);
+    FROM_DEFAULT_V4_UCHAR(space_outliner.active);
+  }
+
   /**
    * Include next version bump.
    */
   {
+    FROM_DEFAULT_V4_UCHAR(space_file.execution_buts);
   }
 
 #undef FROM_DEFAULT_V4_UCHAR
diff --git a/source/blender/blenloader/intern/writefile.c b/source/blender/blenloader/intern/writefile.c
index d8889db7f2d..e81fa9fd2ff 100644
--- a/source/blender/blenloader/intern/writefile.c
+++ b/source/blender/blenloader/intern/writefile.c
@@ -1143,9 +1143,12 @@ typedef struct RenderInfo {
   char scene_name[MAX_ID_NAME - 2];
 } RenderInfo;
 
-/* was for historic render-deamon feature,
- * now write because it can be easily extracted without
- * reading the whole blend file */
+/**
+ * This was originally added for the historic render-daemon feature,
+ * now write because it can be easily extracted without reading the whole blend file.
+ *
+ * See: `release/scripts/modules/blend_render_info.py`
+ */
 static void write_renderinfo(WriteData *wd, Main *mainvar)
 {
   bScreen *curscreen;
diff --git a/source/blender/blentranslation/msgfmt/msgfmt.c b/source/blender/blentranslation/msgfmt/msgfmt.c
index 215c92f87de..4691d791301 100644
--- a/source/blender/blentranslation/msgfmt/msgfmt.c
+++ b/source/blender/blentranslation/msgfmt/msgfmt.c
@@ -24,7 +24,7 @@
  *
  * Generate binary message catalog from textual translation description.
  *
- * This program converts a textual Uniforum-style message catalog (.po file)
+ * This program converts a textual Uniform-style message catalog (.po file)
  * into a binary GNU catalog (.mo file).
  * This is essentially the same function as the GNU msgfmt program,
  * however, it is a simpler implementation.
diff --git a/source/blender/bmesh/intern/bmesh_marking.c b/source/blender/bmesh/intern/bmesh_marking.c
index 5aec59ccd5d..788edc348d9 100644
--- a/source/blender/bmesh/intern/bmesh_marking.c
+++ b/source/blender/bmesh/intern/bmesh_marking.c
@@ -870,9 +870,11 @@ void BM_editselection_normal(BMEditSelection *ese, float r_normal[3])
   }
 }
 
-/* Calculate a plane that is rightangles to the edge/vert/faces normal
+/**
+ * Calculate a plane that is right angles to the edge/vert/faces normal
  * also make the plane run along an axis that is related to the geometry,
- * because this is used for the gizmos Y axis. */
+ * because this is used for the gizmos Y axis.
+ */
 void BM_editselection_plane(BMEditSelection *ese, float r_plane[3])
 {
   if (ese->htype == BM_VERT) {
diff --git a/source/blender/bmesh/intern/bmesh_mesh_conv.c b/source/blender/bmesh/intern/bmesh_mesh_conv.c
index 2000689b496..9bab959f0a2 100644
--- a/source/blender/bmesh/intern/bmesh_mesh_conv.c
+++ b/source/blender/bmesh/intern/bmesh_mesh_conv.c
@@ -422,7 +422,7 @@ void BM_mesh_bm_from_me(BMesh *bm, const Mesh *me, const struct BMeshFromMeshPar
       /* don't use 'j' since we may have skipped some faces, hence some loops. */
       BM_elem_index_set(l_iter, totloops++); /* set_ok */
 
-      /* Save index of correspsonding MLoop */
+      /* Save index of corresponding #MLoop. */
       CustomData_to_bmesh_block(&me->ldata, &bm->ldata, j++, &l_iter->head.data, true);
     } while ((l_iter = l_iter->next) != l_first);
 
diff --git a/source/blender/bmesh/intern/bmesh_opdefines.c b/source/blender/bmesh/intern/bmesh_opdefines.c
index b9f0bcc05f0..7086cea1ace 100644
--- a/source/blender/bmesh/intern/bmesh_opdefines.c
+++ b/source/blender/bmesh/intern/bmesh_opdefines.c
@@ -369,28 +369,6 @@ static BMOpDefine bmo_remove_doubles_def = {
 };
 
 /*
- * Auto Merge.
- *
- * Finds groups of vertices closer then **dist** and merges them together,
- * using the weld verts bmop.  The merges must go from a vert not in
- * **verts** to one in **verts**.
- */
-static BMOpDefine bmo_automerge_def = {
-  "automerge",
-  /* slots_in */
-  {{"verts", BMO_OP_SLOT_ELEMENT_BUF, {BM_VERT}}, /* input verts */
-   {"dist",         BMO_OP_SLOT_FLT}, /* maximum distance */
-   {{'\0'}},
-  },
-  {{{'\0'}}},  /* no output */
-  bmo_automerge_exec,
-  (BMO_OPTYPE_FLAG_UNTAN_MULTIRES |
-   BMO_OPTYPE_FLAG_NORMALS_CALC |
-   BMO_OPTYPE_FLAG_SELECT_FLUSH |
-   BMO_OPTYPE_FLAG_SELECT_VALIDATE),
-};
-
-/*
  * Collapse Connected.
  *
  * Collapses connected vertices
@@ -2073,7 +2051,6 @@ static BMOpDefine bmo_symmetrize_def = {
 /* clang-format on */
 
 const BMOpDefine *bmo_opdefines[] = {
-    &bmo_automerge_def,
     &bmo_average_vert_facedata_def,
     &bmo_beautify_fill_def,
     &bmo_bevel_def,
diff --git a/source/blender/bmesh/intern/bmesh_operators_private.h b/source/blender/bmesh/intern/bmesh_operators_private.h
index 8658f6c233c..137c5aa338e 100644
--- a/source/blender/bmesh/intern/bmesh_operators_private.h
+++ b/source/blender/bmesh/intern/bmesh_operators_private.h
@@ -24,7 +24,6 @@
 struct BMOperator;
 struct BMesh;
 
-void bmo_automerge_exec(BMesh *bm, BMOperator *op);
 void bmo_average_vert_facedata_exec(BMesh *bm, BMOperator *op);
 void bmo_beautify_fill_exec(BMesh *bm, BMOperator *op);
 void bmo_bevel_exec(BMesh *bm, BMOperator *op);
diff --git a/source/blender/bmesh/intern/bmesh_polygon.c b/source/blender/bmesh/intern/bmesh_polygon.c
index dc839054987..915ad6bf5c4 100644
--- a/source/blender/bmesh/intern/bmesh_polygon.c
+++ b/source/blender/bmesh/intern/bmesh_polygon.c
@@ -483,7 +483,7 @@ void BM_face_calc_tangent_edge_diagonal(const BMFace *f, float r_tangent[3])
 
   l_iter = l_first = BM_FACE_FIRST_LOOP(f);
 
-  /* incase of degenerate faces */
+  /* In case of degenerate faces. */
   zero_v3(r_tangent);
 
   /* warning: O(n^2) loop here, take care! */
@@ -520,7 +520,7 @@ void BM_face_calc_tangent_vert_diagonal(const BMFace *f, float r_tangent[3])
 
   l_iter = l_first = BM_FACE_FIRST_LOOP(f);
 
-  /* incase of degenerate faces */
+  /* In case of degenerate faces. */
   zero_v3(r_tangent);
 
   /* warning: O(n^2) loop here, take care! */
diff --git a/source/blender/bmesh/intern/bmesh_query.c b/source/blender/bmesh/intern/bmesh_query.c
index 4a47bcccb25..219bec15e5b 100644
--- a/source/blender/bmesh/intern/bmesh_query.c
+++ b/source/blender/bmesh/intern/bmesh_query.c
@@ -207,6 +207,34 @@ bool BM_vert_pair_share_face_check_cb(BMVert *v_a,
   return false;
 }
 
+BMFace *BM_vert_pair_shared_face_cb(BMVert *v_a,
+                                    BMVert *v_b,
+                                    const bool allow_adjacent,
+                                    bool (*callback)(BMFace *, BMLoop *, BMLoop *, void *userdata),
+                                    void *user_data,
+                                    BMLoop **r_l_a,
+                                    BMLoop **r_l_b)
+{
+  if (v_a->e && v_b->e) {
+    BMIter iter;
+    BMLoop *l_a, *l_b;
+
+    BM_ITER_ELEM (l_a, &iter, v_a, BM_LOOPS_OF_VERT) {
+      BMFace *f = l_a->f;
+      l_b = BM_face_vert_share_loop(f, v_b);
+      if (l_b && (allow_adjacent || !BM_loop_is_adjacent(l_a, l_b)) &&
+          callback(f, l_a, l_b, user_data)) {
+        *r_l_a = l_a;
+        *r_l_b = l_b;
+
+        return f;
+      }
+    }
+  }
+
+  return NULL;
+}
+
 /**
  * Given 2 verts, find the smallest face they share and give back both loops.
  */
@@ -1542,7 +1570,7 @@ float BM_loop_calc_face_normal_safe_ex(const BMLoop *l, const float epsilon_sq,
 /**
  * #BM_loop_calc_face_normal_safe_ex with pre-defined sane epsilon.
  *
- * Since this doesn't scale baed on triangle size, fixed value works well.
+ * Since this doesn't scale based on triangle size, fixed value works well.
  */
 float BM_loop_calc_face_normal_safe(const BMLoop *l, float r_normal[3])
 {
diff --git a/source/blender/bmesh/intern/bmesh_query.h b/source/blender/bmesh/intern/bmesh_query.h
index e96984490c0..3a864fbb5dd 100644
--- a/source/blender/bmesh/intern/bmesh_query.h
+++ b/source/blender/bmesh/intern/bmesh_query.h
@@ -62,6 +62,13 @@ bool BM_vert_pair_share_face_check_cb(BMVert *v_a,
                                       bool (*test_fn)(BMFace *f, void *user_data),
                                       void *user_data) ATTR_WARN_UNUSED_RESULT
     ATTR_NONNULL(1, 2, 3);
+BMFace *BM_vert_pair_shared_face_cb(BMVert *v_a,
+                                    BMVert *v_b,
+                                    const bool allow_adjacent,
+                                    bool (*callback)(BMFace *, BMLoop *, BMLoop *, void *userdata),
+                                    void *user_data,
+                                    BMLoop **r_l_a,
+                                    BMLoop **r_l_b) ATTR_NONNULL(1, 2, 4, 6, 7);
 BMFace *BM_vert_pair_share_face_by_len(BMVert *v_a,
                                        BMVert *v_b,
                                        BMLoop **r_l_a,
diff --git a/source/blender/bmesh/intern/bmesh_walkers_impl.c b/source/blender/bmesh/intern/bmesh_walkers_impl.c
index ade6fdfcbed..f317c59b8d1 100644
--- a/source/blender/bmesh/intern/bmesh_walkers_impl.c
+++ b/source/blender/bmesh/intern/bmesh_walkers_impl.c
@@ -648,7 +648,7 @@ static void *bmw_ConnectedVertexWalker_step(BMWalker *walker)
  * \note that this doesn't work on non-manifold geometry.
  * it might be better to rewrite this to extract
  * boundary info from the island walker, rather then directly walking
- * over the boundary.  raises an error if it encounters nonmanifold geometry.
+ * over the boundary.  raises an error if it encounters non-manifold geometry.
  *
  * \todo Add restriction flag/callback for wire edges.
  */
diff --git a/source/blender/bmesh/operators/bmo_inset.c b/source/blender/bmesh/operators/bmo_inset.c
index f23c420295b..1570c802bee 100644
--- a/source/blender/bmesh/operators/bmo_inset.c
+++ b/source/blender/bmesh/operators/bmo_inset.c
@@ -588,7 +588,7 @@ void bmo_inset_region_exec(BMesh *bm, BMOperator *op)
         /* tag if boundary is enabled */
         (use_boundary && BM_edge_is_boundary(e) && BM_elem_flag_test(e->l->f, BM_ELEM_TAG)) ||
 
-        /* tag if edge is an interior edge inbetween a tagged and untagged face */
+        /* tag if edge is an interior edge in between a tagged and untagged face */
         (bm_edge_is_mixed_face_tag(e->l))) {
       /* tag */
       BM_elem_flag_enable(e->v1, BM_ELEM_TAG);
diff --git a/source/blender/bmesh/operators/bmo_removedoubles.c b/source/blender/bmesh/operators/bmo_removedoubles.c
index d783842c017..1d28d8223cd 100644
--- a/source/blender/bmesh/operators/bmo_removedoubles.c
+++ b/source/blender/bmesh/operators/bmo_removedoubles.c
@@ -706,34 +706,3 @@ void bmo_find_doubles_exec(BMesh *bm, BMOperator *op)
   slot_targetmap_out = BMO_slot_get(op->slots_out, "targetmap.out");
   bmesh_find_doubles_common(bm, op, op, slot_targetmap_out);
 }
-
-void bmo_automerge_exec(BMesh *bm, BMOperator *op)
-{
-  BMOperator findop, weldop;
-  BMIter viter;
-  BMVert *v;
-
-  /* The "verts" input sent to this op is the set of verts that
-   * can be merged away into any other verts. Mark all other verts
-   * as VERT_KEEP. */
-  BMO_slot_buffer_flag_enable(bm, op->slots_in, "verts", BM_VERT, VERT_IN);
-  BM_ITER_MESH (v, &viter, bm, BM_VERTS_OF_MESH) {
-    if (!BMO_vert_flag_test(bm, v, VERT_IN)) {
-      BMO_vert_flag_enable(bm, v, VERT_KEEP);
-    }
-  }
-
-  /* Search for doubles among all vertices, but only merge non-VERT_KEEP
-   * vertices into VERT_KEEP vertices. */
-  BMO_op_initf(bm, &findop, op->flag, "find_doubles verts=%av keep_verts=%fv", VERT_KEEP);
-  BMO_slot_copy(op, slots_in, "dist", &findop, slots_in, "dist");
-  BMO_op_exec(bm, &findop);
-
-  /* weld the vertices */
-  BMO_op_init(bm, &weldop, op->flag, "weld_verts");
-  BMO_slot_copy(&findop, slots_out, "targetmap.out", &weldop, slots_in, "targetmap");
-  BMO_op_exec(bm, &weldop);
-
-  BMO_op_finish(bm, &findop);
-  BMO_op_finish(bm, &weldop);
-}
diff --git a/source/blender/bmesh/tools/bmesh_beautify.c b/source/blender/bmesh/tools/bmesh_beautify.c
index 5d511374989..dabdbc3c97a 100644
--- a/source/blender/bmesh/tools/bmesh_beautify.c
+++ b/source/blender/bmesh/tools/bmesh_beautify.c
@@ -199,7 +199,7 @@ static float bm_edge_calc_rotate_beauty__area(const float v1[3],
      * Allowing to rotate out of a degenerate state can flip the faces
      * (when performed iteratively).
      */
-    return BLI_polyfill_beautify_quad_rotate_calc_ex(v1_xy, v2_xy, v3_xy, v4_xy, true);
+    return BLI_polyfill_beautify_quad_rotate_calc_ex(v1_xy, v2_xy, v3_xy, v4_xy, true, NULL);
   } while (false);
 
   return FLT_MAX;
diff --git a/source/blender/bmesh/tools/bmesh_bevel.c b/source/blender/bmesh/tools/bmesh_bevel.c
index 94935f2090b..797e2ca864e 100644
--- a/source/blender/bmesh/tools/bmesh_bevel.c
+++ b/source/blender/bmesh/tools/bmesh_bevel.c
@@ -3122,6 +3122,7 @@ static void adjust_offsets(BevelParams *bp, BMesh *bm)
       }
       if (!iscycle) {
         /* right->left direction, changing vchainstart at each step */
+        v->adjchain = NULL;
         v = vchainstart;
         bvcur = bv;
         do {
diff --git a/source/blender/bmesh/tools/bmesh_decimate_collapse.c b/source/blender/bmesh/tools/bmesh_decimate_collapse.c
index e32a9334343..198b4c8e76b 100644
--- a/source/blender/bmesh/tools/bmesh_decimate_collapse.c
+++ b/source/blender/bmesh/tools/bmesh_decimate_collapse.c
@@ -363,7 +363,7 @@ static void bm_decim_build_edge_cost(BMesh *bm,
 struct KD_Symmetry_Data {
   /* pre-flipped coords */
   float e_v1_co[3], e_v2_co[3];
-  /* Use to compare the correct endpoints incase v1/v2 are swapped */
+  /* Use to compare the correct endpoints in case v1/v2 are swapped. */
   float e_dir[3];
 
   int e_found_index;
@@ -1371,8 +1371,8 @@ void BM_mesh_decimate_collapse(BMesh *bm,
       /* handy to detect corruptions elsewhere */
       BLI_assert(BM_elem_index_get(e) < tot_edge_orig);
 
-      /* under normal conditions wont be accessed again,
-       * but NULL just incase so we don't use freed node */
+      /* Under normal conditions wont be accessed again,
+       * but NULL just in case so we don't use freed node. */
       eheap_table[BM_elem_index_get(e)] = NULL;
 
       bm_decim_edge_collapse(bm,
diff --git a/source/blender/bmesh/tools/bmesh_decimate_unsubdivide.c b/source/blender/bmesh/tools/bmesh_decimate_unsubdivide.c
index 27b4fa15f26..2cc86a7c93f 100644
--- a/source/blender/bmesh/tools/bmesh_decimate_unsubdivide.c
+++ b/source/blender/bmesh/tools/bmesh_decimate_unsubdivide.c
@@ -231,7 +231,7 @@ void BM_mesh_decimate_unsubdivide_ex(BMesh *bm, const int iterations, const bool
           if (BMO_vert_flag_test(bm, v, ELE_VERT_TAG))
 #endif
           {
-            /* check again incase the topology changed */
+            /* Check again in case the topology changed. */
             if (bm_vert_dissolve_fan_test(v)) {
               v_first = v;
             }
diff --git a/source/blender/bmesh/tools/bmesh_region_match.c b/source/blender/bmesh/tools/bmesh_region_match.c
index 943f7532960..c30992fa296 100644
--- a/source/blender/bmesh/tools/bmesh_region_match.c
+++ b/source/blender/bmesh/tools/bmesh_region_match.c
@@ -1099,7 +1099,7 @@ static BMEdge *bm_face_region_pivot_edge_find(BMFace **faces_region,
         }
       }
       else {
-        /* use incase (depth == 0), no interior verts */
+        /* Use in case (depth == 0), no interior verts. */
         e_pivot_fallback = e;
       }
     } while ((l_iter = l_iter->next) != l_first);
diff --git a/source/blender/collada/AnimationExporter.h b/source/blender/collada/AnimationExporter.h
index 27474cc2b58..2be2dadeeeb 100644
--- a/source/blender/collada/AnimationExporter.h
+++ b/source/blender/collada/AnimationExporter.h
@@ -78,8 +78,6 @@ extern "C" {
 #include <map>
 #include <algorithm>  // std::find
 
-struct Depsgraph;
-
 typedef enum BC_animation_source_type {
   BC_SOURCE_TYPE_VALUE,
   BC_SOURCE_TYPE_ANGLE,
diff --git a/source/blender/collada/AnimationImporter.cpp b/source/blender/collada/AnimationImporter.cpp
index 54177560eb5..79593f07383 100644
--- a/source/blender/collada/AnimationImporter.cpp
+++ b/source/blender/collada/AnimationImporter.cpp
@@ -733,7 +733,7 @@ void AnimationImporter::Assign_float_animations(const COLLADAFW::UniqueId &listi
            * Reason: old blender versions stored spot_size in radians (was a bug)
            */
           if (this->import_from_version == "" ||
-              BLI_natstrcmp(this->import_from_version.c_str(), "2.69.10") != -1) {
+              BLI_strcasecmp_natural(this->import_from_version.c_str(), "2.69.10") != -1) {
             fcurve_deg_to_rad(fcu);
           }
         }
diff --git a/source/blender/collada/CMakeLists.txt b/source/blender/collada/CMakeLists.txt
index 5f28ca4ffbb..40762db759e 100644
--- a/source/blender/collada/CMakeLists.txt
+++ b/source/blender/collada/CMakeLists.txt
@@ -28,10 +28,7 @@ FIND_FILE(OPENCOLLADA_ANIMATION_CLIP
     )
 
 IF(OPENCOLLADA_ANIMATION_CLIP)
-   message(STATUS "Found opencollada: ${OPENCOLLADA_ANIMATION_CLIP} ")
    add_definitions(-DWITH_OPENCOLLADA_ANIMATION_CLIP)
-ELSE()
-   message(STATUS "opencollada: Build without animation clip support")
 ENDIF()
 
 set(INC
diff --git a/source/blender/collada/ControllerExporter.h b/source/blender/collada/ControllerExporter.h
index 200f8431f62..ce2ed9fe453 100644
--- a/source/blender/collada/ControllerExporter.h
+++ b/source/blender/collada/ControllerExporter.h
@@ -47,7 +47,6 @@
 
 #include "BKE_key.h"
 
-struct Depsgraph;
 class SceneExporter;
 
 class ControllerExporter : public COLLADASW::LibraryControllers,
diff --git a/source/blender/collada/GeometryExporter.h b/source/blender/collada/GeometryExporter.h
index 7262784db82..8c7a38fc407 100644
--- a/source/blender/collada/GeometryExporter.h
+++ b/source/blender/collada/GeometryExporter.h
@@ -39,8 +39,6 @@
 #include "BlenderContext.h"
 #include "BKE_key.h"
 
-struct Depsgraph;
-
 class Normal {
  public:
   float x;
@@ -58,8 +56,6 @@ class GeometryExporter : COLLADASW::LibraryGeometries {
     unsigned int v1, v2, v3, v4;
   };
 
-  Normal n;
-
  public:
   /* TODO: optimize UV sets by making indexed list with duplicates removed */
   GeometryExporter(BlenderContext &blender_context,
diff --git a/source/blender/collada/collada.h b/source/blender/collada/collada.h
index df6e72cbf3a..72753e170a3 100644
--- a/source/blender/collada/collada.h
+++ b/source/blender/collada/collada.h
@@ -34,9 +34,6 @@ extern "C" {
 #include "BLI_path_util.h"
 #include "RNA_types.h"
 
-struct Depsgraph;
-struct Scene;
-struct ViewLayer;
 struct bContext;
 
 /*
diff --git a/source/blender/collada/collada_utils.h b/source/blender/collada/collada_utils.h
index c23e33df76d..c0425e59d1a 100644
--- a/source/blender/collada/collada_utils.h
+++ b/source/blender/collada/collada_utils.h
@@ -69,8 +69,6 @@ extern "C" {
 
 constexpr int LIMITTED_PRECISION = 6;
 
-struct Depsgraph;
-
 typedef std::map<COLLADAFW::UniqueId, Image *> UidImageMap;
 typedef std::map<std::string, Image *> KeyImageMap;
 typedef std::map<COLLADAFW::TextureMapId, std::vector<MTex *>> TexIndexTextureArrayMap;
diff --git a/source/blender/compositor/CMakeLists.txt b/source/blender/compositor/CMakeLists.txt
index 50b5951f99f..2a8914c8cd9 100644
--- a/source/blender/compositor/CMakeLists.txt
+++ b/source/blender/compositor/CMakeLists.txt
@@ -280,8 +280,8 @@ set(SRC
   nodes/COM_BlurNode.h
   nodes/COM_BokehBlurNode.cpp
   nodes/COM_BokehBlurNode.h
-  nodes/COM_DenoiseNode.h
   nodes/COM_DenoiseNode.cpp
+  nodes/COM_DenoiseNode.h
   nodes/COM_DespeckleNode.cpp
   nodes/COM_DespeckleNode.h
   nodes/COM_DilateErodeNode.cpp
@@ -491,8 +491,8 @@ set(SRC
   operations/COM_ConvolutionEdgeFilterOperation.h
   operations/COM_ConvolutionFilterOperation.cpp
   operations/COM_ConvolutionFilterOperation.h
-  operations/COM_DenoiseOperation.h
   operations/COM_DenoiseOperation.cpp
+  operations/COM_DenoiseOperation.h
   operations/COM_DespeckleOperation.cpp
   operations/COM_DespeckleOperation.h
   operations/COM_DilateErodeOperation.cpp
diff --git a/source/blender/compositor/intern/COM_CompositorContext.h b/source/blender/compositor/intern/COM_CompositorContext.h
index f4cd60e3ee0..b28d5ff0cdf 100644
--- a/source/blender/compositor/intern/COM_CompositorContext.h
+++ b/source/blender/compositor/intern/COM_CompositorContext.h
@@ -215,7 +215,7 @@ class CompositorContext {
   }
 
   /**
-   * \brief get the current framenumber of the scene in this context
+   * \brief get the current frame-number of the scene in this context
    */
   int getFramenumber() const;
 
diff --git a/source/blender/compositor/intern/COM_MemoryBuffer.h b/source/blender/compositor/intern/COM_MemoryBuffer.h
index df936818f33..7e5b0264aa3 100644
--- a/source/blender/compositor/intern/COM_MemoryBuffer.h
+++ b/source/blender/compositor/intern/COM_MemoryBuffer.h
@@ -73,7 +73,7 @@ class MemoryBuffer {
   rcti m_rect;
 
   /**
-   * brief refers to the chunknumber within the executiongroup where related to the MemoryProxy
+   * brief refers to the chunk-number within the execution-group where related to the MemoryProxy
    * \see memoryProxy
    */
   unsigned int m_chunkNumber;
diff --git a/source/blender/compositor/intern/COM_NodeOperation.h b/source/blender/compositor/intern/COM_NodeOperation.h
index af9ed2648c9..6b073452771 100644
--- a/source/blender/compositor/intern/COM_NodeOperation.h
+++ b/source/blender/compositor/intern/COM_NodeOperation.h
@@ -368,7 +368,7 @@ class NodeOperation : public SocketReader {
     return true;
   }
 
-  inline bool isBreaked() const
+  inline bool isBraked() const
   {
     return this->m_btree->test_break(this->m_btree->tbh);
   }
diff --git a/source/blender/compositor/intern/COM_OpenCLDevice.cpp b/source/blender/compositor/intern/COM_OpenCLDevice.cpp
index 0aa054b8c77..2529637801d 100644
--- a/source/blender/compositor/intern/COM_OpenCLDevice.cpp
+++ b/source/blender/compositor/intern/COM_OpenCLDevice.cpp
@@ -248,7 +248,7 @@ void OpenCLDevice::COM_clEnqueueRange(cl_kernel kernel,
         printf("CLERROR[%d]: %s\n", error, clewErrorString(error));
       }
       clFlush(this->m_queue);
-      if (operation->isBreaked()) {
+      if (operation->isBraked()) {
         breaked = false;
       }
     }
diff --git a/source/blender/compositor/intern/COM_OpenCLDevice.h b/source/blender/compositor/intern/COM_OpenCLDevice.h
index 7a83bda162c..45ce77acac7 100644
--- a/source/blender/compositor/intern/COM_OpenCLDevice.h
+++ b/source/blender/compositor/intern/COM_OpenCLDevice.h
@@ -78,8 +78,8 @@ class OpenCLDevice : public Device {
   bool initialize();
 
   /**
-   * \brief deinitialize the device
-   * During deintiialization the command queue is cleared
+   * \brief de-initialize the device
+   * During de-initialization the command queue is cleared
    */
   void deinitialize();
 
diff --git a/source/blender/compositor/nodes/COM_ImageNode.cpp b/source/blender/compositor/nodes/COM_ImageNode.cpp
index dc3f65deb1f..6bce56ffd52 100644
--- a/source/blender/compositor/nodes/COM_ImageNode.cpp
+++ b/source/blender/compositor/nodes/COM_ImageNode.cpp
@@ -195,7 +195,7 @@ void ImageNode::convertToOperations(NodeConverter &converter,
             }
           }
 
-          /* incase we can't load the layer */
+          /* In case we can't load the layer. */
           if (operation == NULL) {
             converter.setInvalidOutput(getOutputSocket(index));
           }
diff --git a/source/blender/compositor/nodes/COM_MathNode.cpp b/source/blender/compositor/nodes/COM_MathNode.cpp
index 5a2f934f37f..d13b34bb6b5 100644
--- a/source/blender/compositor/nodes/COM_MathNode.cpp
+++ b/source/blender/compositor/nodes/COM_MathNode.cpp
@@ -29,61 +29,61 @@ void MathNode::convertToOperations(NodeConverter &converter,
     case NODE_MATH_ADD:
       operation = new MathAddOperation();
       break;
-    case NODE_MATH_SUB:
+    case NODE_MATH_SUBTRACT:
       operation = new MathSubtractOperation();
       break;
-    case NODE_MATH_MUL:
+    case NODE_MATH_MULTIPLY:
       operation = new MathMultiplyOperation();
       break;
     case NODE_MATH_DIVIDE:
       operation = new MathDivideOperation();
       break;
-    case NODE_MATH_SIN:
+    case NODE_MATH_SINE:
       operation = new MathSineOperation();
       break;
-    case NODE_MATH_COS:
+    case NODE_MATH_COSINE:
       operation = new MathCosineOperation();
       break;
-    case NODE_MATH_TAN:
+    case NODE_MATH_TANGENT:
       operation = new MathTangentOperation();
       break;
-    case NODE_MATH_ASIN:
+    case NODE_MATH_ARCSINE:
       operation = new MathArcSineOperation();
       break;
-    case NODE_MATH_ACOS:
+    case NODE_MATH_ARCCOSINE:
       operation = new MathArcCosineOperation();
       break;
-    case NODE_MATH_ATAN:
+    case NODE_MATH_ARCTANGENT:
       operation = new MathArcTangentOperation();
       break;
-    case NODE_MATH_POW:
+    case NODE_MATH_POWER:
       operation = new MathPowerOperation();
       break;
-    case NODE_MATH_LOG:
+    case NODE_MATH_LOGARITHM:
       operation = new MathLogarithmOperation();
       break;
-    case NODE_MATH_MIN:
+    case NODE_MATH_MINIMUM:
       operation = new MathMinimumOperation();
       break;
-    case NODE_MATH_MAX:
+    case NODE_MATH_MAXIMUM:
       operation = new MathMaximumOperation();
       break;
     case NODE_MATH_ROUND:
       operation = new MathRoundOperation();
       break;
-    case NODE_MATH_LESS:
+    case NODE_MATH_LESS_THAN:
       operation = new MathLessThanOperation();
       break;
-    case NODE_MATH_GREATER:
+    case NODE_MATH_GREATER_THAN:
       operation = new MathGreaterThanOperation();
       break;
-    case NODE_MATH_MOD:
+    case NODE_MATH_MODULO:
       operation = new MathModuloOperation();
       break;
-    case NODE_MATH_ABS:
+    case NODE_MATH_ABSOLUTE:
       operation = new MathAbsoluteOperation();
       break;
-    case NODE_MATH_ATAN2:
+    case NODE_MATH_ARCTAN2:
       operation = new MathArcTan2Operation();
       break;
     case NODE_MATH_FLOOR:
@@ -92,7 +92,7 @@ void MathNode::convertToOperations(NodeConverter &converter,
     case NODE_MATH_CEIL:
       operation = new MathCeilOperation();
       break;
-    case NODE_MATH_FRACT:
+    case NODE_MATH_FRACTION:
       operation = new MathFractOperation();
       break;
     case NODE_MATH_SQRT:
diff --git a/source/blender/compositor/nodes/COM_MixNode.cpp b/source/blender/compositor/nodes/COM_MixNode.cpp
index 27a605e28d4..b3bcefaccb5 100644
--- a/source/blender/compositor/nodes/COM_MixNode.cpp
+++ b/source/blender/compositor/nodes/COM_MixNode.cpp
@@ -52,7 +52,7 @@ void MixNode::convertToOperations(NodeConverter &converter,
       convertProg = new MixLightenOperation();
       break;
     case MA_RAMP_BURN:
-      convertProg = new MixBurnOperation();
+      convertProg = new MixColorBurnOperation();
       break;
     case MA_RAMP_HUE:
       convertProg = new MixHueOperation();
diff --git a/source/blender/compositor/operations/COM_ColorBalanceLGGOperation.cpp b/source/blender/compositor/operations/COM_ColorBalanceLGGOperation.cpp
index 68b5af3089a..1578a805d1e 100644
--- a/source/blender/compositor/operations/COM_ColorBalanceLGGOperation.cpp
+++ b/source/blender/compositor/operations/COM_ColorBalanceLGGOperation.cpp
@@ -22,7 +22,7 @@
 inline float colorbalance_lgg(float in, float lift_lgg, float gamma_inv, float gain)
 {
   /* 1:1 match with the sequencer with linear/srgb conversions, the conversion isnt pretty
-   * but best keep it this way, sice testing for durian shows a similar calculation
+   * but best keep it this way, since testing for durian shows a similar calculation
    * without lin/srgb conversions gives bad results (over-saturated shadows) with colors
    * slightly below 1.0. some correction can be done but it ends up looking bad for shadows or
    * lighter tones - campbell */
diff --git a/source/blender/compositor/operations/COM_CompositorOperation.cpp b/source/blender/compositor/operations/COM_CompositorOperation.cpp
index 40315d217ce..5bd466658c0 100644
--- a/source/blender/compositor/operations/COM_CompositorOperation.cpp
+++ b/source/blender/compositor/operations/COM_CompositorOperation.cpp
@@ -78,7 +78,7 @@ void CompositorOperation::deinitExecution()
     return;
   }
 
-  if (!isBreaked()) {
+  if (!isBraked()) {
     Render *re = RE_GetSceneRender(this->m_scene);
     RenderResult *rr = RE_AcquireResultWrite(re);
 
@@ -207,7 +207,7 @@ void CompositorOperation::executeRegion(rcti *rect, unsigned int /*tileNumber*/)
       zbuffer[offset] = color[0];
       offset4 += COM_NUM_CHANNELS_COLOR;
       offset++;
-      if (isBreaked()) {
+      if (isBraked()) {
         breaked = true;
       }
     }
diff --git a/source/blender/compositor/operations/COM_DenoiseOperation.cpp b/source/blender/compositor/operations/COM_DenoiseOperation.cpp
index af568490c72..3aeaba539f9 100644
--- a/source/blender/compositor/operations/COM_DenoiseOperation.cpp
+++ b/source/blender/compositor/operations/COM_DenoiseOperation.cpp
@@ -21,16 +21,19 @@
 
 #include "COM_DenoiseOperation.h"
 #include "BLI_math.h"
+#include "BLI_system.h"
 #ifdef WITH_OPENIMAGEDENOISE
+#  include "BLI_threads.h"
 #  include <OpenImageDenoise/oidn.hpp>
+static pthread_mutex_t oidn_lock = BLI_MUTEX_INITIALIZER;
 #endif
 #include <iostream>
 
 DenoiseOperation::DenoiseOperation() : SingleThreadedOperation()
 {
   this->addInputSocket(COM_DT_COLOR);
-  this->addInputSocket(COM_DT_COLOR);
   this->addInputSocket(COM_DT_VECTOR);
+  this->addInputSocket(COM_DT_COLOR);
   this->addOutputSocket(COM_DT_COLOR);
   this->m_settings = NULL;
 }
@@ -38,23 +41,23 @@ void DenoiseOperation::initExecution()
 {
   SingleThreadedOperation::initExecution();
   this->m_inputProgramColor = getInputSocketReader(0);
-  this->m_inputProgramAlbedo = getInputSocketReader(1);
-  this->m_inputProgramNormal = getInputSocketReader(2);
+  this->m_inputProgramNormal = getInputSocketReader(1);
+  this->m_inputProgramAlbedo = getInputSocketReader(2);
 }
 
 void DenoiseOperation::deinitExecution()
 {
   this->m_inputProgramColor = NULL;
-  this->m_inputProgramAlbedo = NULL;
   this->m_inputProgramNormal = NULL;
+  this->m_inputProgramAlbedo = NULL;
   SingleThreadedOperation::deinitExecution();
 }
 
 MemoryBuffer *DenoiseOperation::createMemoryBuffer(rcti *rect2)
 {
   MemoryBuffer *tileColor = (MemoryBuffer *)this->m_inputProgramColor->initializeTileData(rect2);
-  MemoryBuffer *tileAlbedo = (MemoryBuffer *)this->m_inputProgramAlbedo->initializeTileData(rect2);
   MemoryBuffer *tileNormal = (MemoryBuffer *)this->m_inputProgramNormal->initializeTileData(rect2);
+  MemoryBuffer *tileAlbedo = (MemoryBuffer *)this->m_inputProgramAlbedo->initializeTileData(rect2);
   rcti rect;
   rect.xmin = 0;
   rect.ymin = 0;
@@ -62,7 +65,7 @@ MemoryBuffer *DenoiseOperation::createMemoryBuffer(rcti *rect2)
   rect.ymax = getHeight();
   MemoryBuffer *result = new MemoryBuffer(COM_DT_COLOR, &rect);
   float *data = result->getBuffer();
-  this->generateDenoise(data, tileColor, tileAlbedo, tileNormal, this->m_settings);
+  this->generateDenoise(data, tileColor, tileNormal, tileAlbedo, this->m_settings);
   return result;
 }
 
@@ -85,8 +88,8 @@ bool DenoiseOperation::determineDependingAreaOfInterest(rcti * /*input*/,
 
 void DenoiseOperation::generateDenoise(float *data,
                                        MemoryBuffer *inputTileColor,
-                                       MemoryBuffer *inputTileAlbedo,
                                        MemoryBuffer *inputTileNormal,
+                                       MemoryBuffer *inputTileAlbedo,
                                        NodeDenoise *settings)
 {
   float *inputBufferColor = inputTileColor->getBuffer();
@@ -95,60 +98,69 @@ void DenoiseOperation::generateDenoise(float *data,
     return;
   }
 #ifdef WITH_OPENIMAGEDENOISE
-  oidn::DeviceRef device = oidn::newDevice();
-  device.commit();
+  if (BLI_cpu_support_sse41()) {
+    oidn::DeviceRef device = oidn::newDevice();
+    device.commit();
 
-  oidn::FilterRef filter = device.newFilter("RT");
-  filter.setImage("color",
-                  inputBufferColor,
-                  oidn::Format::Float3,
-                  inputTileColor->getWidth(),
-                  inputTileColor->getHeight(),
-                  0,
-                  4 * sizeof(float));
-  if (inputTileAlbedo && inputTileAlbedo->getBuffer()) {
-    filter.setImage("albedo",
-                    inputTileAlbedo->getBuffer(),
+    oidn::FilterRef filter = device.newFilter("RT");
+    filter.setImage("color",
+                    inputBufferColor,
                     oidn::Format::Float3,
-                    inputTileAlbedo->getWidth(),
-                    inputTileAlbedo->getHeight(),
+                    inputTileColor->getWidth(),
+                    inputTileColor->getHeight(),
                     0,
                     4 * sizeof(float));
-  }
-  if (inputTileNormal && inputTileNormal->getBuffer()) {
-    filter.setImage("normal",
-                    inputTileNormal->getBuffer(),
+    if (inputTileNormal && inputTileNormal->getBuffer()) {
+      filter.setImage("normal",
+                      inputTileNormal->getBuffer(),
+                      oidn::Format::Float3,
+                      inputTileNormal->getWidth(),
+                      inputTileNormal->getHeight(),
+                      0,
+                      3 * sizeof(float));
+    }
+    if (inputTileAlbedo && inputTileAlbedo->getBuffer()) {
+      filter.setImage("albedo",
+                      inputTileAlbedo->getBuffer(),
+                      oidn::Format::Float3,
+                      inputTileAlbedo->getWidth(),
+                      inputTileAlbedo->getHeight(),
+                      0,
+                      4 * sizeof(float));
+    }
+    filter.setImage("output",
+                    data,
                     oidn::Format::Float3,
-                    inputTileNormal->getWidth(),
-                    inputTileNormal->getHeight(),
+                    inputTileColor->getWidth(),
+                    inputTileColor->getHeight(),
                     0,
-                    3 * sizeof(float));
-  }
-  filter.setImage("output",
-                  data,
-                  oidn::Format::Float3,
-                  inputTileColor->getWidth(),
-                  inputTileColor->getHeight(),
-                  0,
-                  4 * sizeof(float));
+                    4 * sizeof(float));
 
-  BLI_assert(settings);
-  if (settings) {
-    filter.set("hdr", settings->hdr);
-    filter.set("srgb", false);
-  }
+    BLI_assert(settings);
+    if (settings) {
+      filter.set("hdr", settings->hdr);
+      filter.set("srgb", false);
+    }
 
-  filter.commit();
-  filter.execute();
+    filter.commit();
+    /* Since it's memory intensive, it's better to run only one instance of OIDN at a time.
+     * OpenImageDenoise is multithreaded internally and should use all available cores nonetheless.
+     */
+    BLI_mutex_lock(&oidn_lock);
+    filter.execute();
+    BLI_mutex_unlock(&oidn_lock);
 
-  /* copy the alpha channel, OpenImageDenoise currently only supports RGB */
-  size_t numPixels = inputTileColor->getWidth() * inputTileColor->getHeight();
-  for (size_t i = 0; i < numPixels; ++i) {
-    data[i * 4 + 3] = inputBufferColor[i * 4 + 3];
+    /* copy the alpha channel, OpenImageDenoise currently only supports RGB */
+    size_t numPixels = inputTileColor->getWidth() * inputTileColor->getHeight();
+    for (size_t i = 0; i < numPixels; ++i) {
+      data[i * 4 + 3] = inputBufferColor[i * 4 + 3];
+    }
+    return;
   }
-#else
+#endif
+  /* If built without OIDN or running on an unsupported CPU, just pass through. */
+  UNUSED_VARS(inputTileAlbedo, inputTileNormal, settings);
   ::memcpy(data,
            inputBufferColor,
            inputTileColor->getWidth() * inputTileColor->getHeight() * sizeof(float) * 4);
-#endif
 }
diff --git a/source/blender/compositor/operations/COM_DenoiseOperation.h b/source/blender/compositor/operations/COM_DenoiseOperation.h
index 6e19bd6034a..73165174932 100644
--- a/source/blender/compositor/operations/COM_DenoiseOperation.h
+++ b/source/blender/compositor/operations/COM_DenoiseOperation.h
@@ -62,8 +62,8 @@ class DenoiseOperation : public SingleThreadedOperation {
  protected:
   void generateDenoise(float *data,
                        MemoryBuffer *inputTileColor,
-                       MemoryBuffer *inputTileAlbedo,
                        MemoryBuffer *inputTileNormal,
+                       MemoryBuffer *inputTileAlbedo,
                        NodeDenoise *settings);
 
   MemoryBuffer *createMemoryBuffer(rcti *rect);
diff --git a/source/blender/compositor/operations/COM_GlareGhostOperation.cpp b/source/blender/compositor/operations/COM_GlareGhostOperation.cpp
index 944a1d9c5dc..9f01cf5d63a 100644
--- a/source/blender/compositor/operations/COM_GlareGhostOperation.cpp
+++ b/source/blender/compositor/operations/COM_GlareGhostOperation.cpp
@@ -51,7 +51,7 @@ void GlareGhostOperation::generateGlare(float *data, MemoryBuffer *inputTile, No
   if (!breaked) {
     FastGaussianBlurOperation::IIR_gauss(tbuf1, s1, 1, 3);
   }
-  if (isBreaked()) {
+  if (isBraked()) {
     breaked = true;
   }
   if (!breaked) {
@@ -60,19 +60,19 @@ void GlareGhostOperation::generateGlare(float *data, MemoryBuffer *inputTile, No
 
   MemoryBuffer *tbuf2 = tbuf1->duplicate();
 
-  if (isBreaked()) {
+  if (isBraked()) {
     breaked = true;
   }
   if (!breaked) {
     FastGaussianBlurOperation::IIR_gauss(tbuf2, s2, 0, 3);
   }
-  if (isBreaked()) {
+  if (isBraked()) {
     breaked = true;
   }
   if (!breaked) {
     FastGaussianBlurOperation::IIR_gauss(tbuf2, s2, 1, 3);
   }
-  if (isBreaked()) {
+  if (isBraked()) {
     breaked = true;
   }
   if (!breaked) {
@@ -117,7 +117,7 @@ void GlareGhostOperation::generateGlare(float *data, MemoryBuffer *inputTile, No
 
       gbuf->writePixel(x, y, c);
     }
-    if (isBreaked()) {
+    if (isBraked()) {
       breaked = true;
     }
   }
@@ -142,7 +142,7 @@ void GlareGhostOperation::generateGlare(float *data, MemoryBuffer *inputTile, No
         }
         tbuf1->addPixel(x, y, tc);
       }
-      if (isBreaked()) {
+      if (isBraked()) {
         breaked = true;
       }
     }
diff --git a/source/blender/compositor/operations/COM_GlareSimpleStarOperation.cpp b/source/blender/compositor/operations/COM_GlareSimpleStarOperation.cpp
index 1ceeba18960..75c2ae51bde 100644
--- a/source/blender/compositor/operations/COM_GlareSimpleStarOperation.cpp
+++ b/source/blender/compositor/operations/COM_GlareSimpleStarOperation.cpp
@@ -58,7 +58,7 @@ void GlareSimpleStarOperation::generateGlare(float *data,
         c[3] = 1.0f;
         tbuf2->writePixel(x, y, c);
       }
-      if (isBreaked()) {
+      if (isBraked()) {
         breaked = true;
       }
     }
@@ -87,7 +87,7 @@ void GlareSimpleStarOperation::generateGlare(float *data,
         c[3] = 1.0f;
         tbuf2->writePixel(x, y, c);
       }
-      if (isBreaked()) {
+      if (isBraked()) {
         breaked = true;
       }
     }
diff --git a/source/blender/compositor/operations/COM_GlareStreaksOperation.cpp b/source/blender/compositor/operations/COM_GlareStreaksOperation.cpp
index 78ca373faaf..951dec9281e 100644
--- a/source/blender/compositor/operations/COM_GlareStreaksOperation.cpp
+++ b/source/blender/compositor/operations/COM_GlareStreaksOperation.cpp
@@ -78,7 +78,7 @@ void GlareStreaksOperation::generateGlare(float *data,
           tdstcol[2] = 0.5f * (tdstcol[2] + c1[2] + wt * (c2[2] + wt * (c3[2] + wt * c4[2])));
           tdstcol[3] = 1.0f;
         }
-        if (isBreaked()) {
+        if (isBraked()) {
           breaked = true;
         }
       }
diff --git a/source/blender/compositor/operations/COM_MaskOperation.cpp b/source/blender/compositor/operations/COM_MaskOperation.cpp
index 0c9208fb6bb..88a3a5c535c 100644
--- a/source/blender/compositor/operations/COM_MaskOperation.cpp
+++ b/source/blender/compositor/operations/COM_MaskOperation.cpp
@@ -148,7 +148,7 @@ void MaskOperation::executePixelSampled(float output[4],
     }
   }
   else {
-    /* incase loop below fails */
+    /* In case loop below fails. */
     output[0] = 0.0f;
 
     for (unsigned int i = 0; i < this->m_rasterMaskHandleTot; i++) {
diff --git a/source/blender/compositor/operations/COM_MixOperation.cpp b/source/blender/compositor/operations/COM_MixOperation.cpp
index a610cb0d1e8..6b8a0caa13b 100644
--- a/source/blender/compositor/operations/COM_MixOperation.cpp
+++ b/source/blender/compositor/operations/COM_MixOperation.cpp
@@ -162,12 +162,15 @@ void MixBlendOperation::executePixelSampled(float output[4],
 
 /* ******** Mix Burn Operation ******** */
 
-MixBurnOperation::MixBurnOperation() : MixBaseOperation()
+MixColorBurnOperation::MixColorBurnOperation() : MixBaseOperation()
 {
   /* pass */
 }
 
-void MixBurnOperation::executePixelSampled(float output[4], float x, float y, PixelSampler sampler)
+void MixColorBurnOperation::executePixelSampled(float output[4],
+                                                float x,
+                                                float y,
+                                                PixelSampler sampler)
 {
   float inputColor1[4];
   float inputColor2[4];
diff --git a/source/blender/compositor/operations/COM_MixOperation.h b/source/blender/compositor/operations/COM_MixOperation.h
index 6ea55a3edfd..fbf60fb9d37 100644
--- a/source/blender/compositor/operations/COM_MixOperation.h
+++ b/source/blender/compositor/operations/COM_MixOperation.h
@@ -95,9 +95,9 @@ class MixBlendOperation : public MixBaseOperation {
   void executePixelSampled(float output[4], float x, float y, PixelSampler sampler);
 };
 
-class MixBurnOperation : public MixBaseOperation {
+class MixColorBurnOperation : public MixBaseOperation {
  public:
-  MixBurnOperation();
+  MixColorBurnOperation();
   void executePixelSampled(float output[4], float x, float y, PixelSampler sampler);
 };
 
diff --git a/source/blender/compositor/operations/COM_ViewerOperation.cpp b/source/blender/compositor/operations/COM_ViewerOperation.cpp
index d5b9edae719..3f7619523e3 100644
--- a/source/blender/compositor/operations/COM_ViewerOperation.cpp
+++ b/source/blender/compositor/operations/COM_ViewerOperation.cpp
@@ -111,7 +111,7 @@ void ViewerOperation::executeRegion(rcti *rect, unsigned int /*tileNumber*/)
       offset++;
       offset4 += 4;
     }
-    if (isBreaked()) {
+    if (isBraked()) {
       breaked = true;
     }
     offset += offsetadd;
diff --git a/source/blender/compositor/operations/COM_WriteBufferOperation.cpp b/source/blender/compositor/operations/COM_WriteBufferOperation.cpp
index beba29e6755..6c5e45472a8 100644
--- a/source/blender/compositor/operations/COM_WriteBufferOperation.cpp
+++ b/source/blender/compositor/operations/COM_WriteBufferOperation.cpp
@@ -76,7 +76,7 @@ void WriteBufferOperation::executeRegion(rcti *rect, unsigned int /*tileNumber*/
         this->m_input->read(&(buffer[offset4]), x, y, data);
         offset4 += num_channels;
       }
-      if (isBreaked()) {
+      if (isBraked()) {
         breaked = true;
       }
     }
@@ -100,7 +100,7 @@ void WriteBufferOperation::executeRegion(rcti *rect, unsigned int /*tileNumber*/
         this->m_input->readSampled(&(buffer[offset4]), x, y, COM_PS_NEAREST);
         offset4 += num_channels;
       }
-      if (isBreaked()) {
+      if (isBraked()) {
         breaked = true;
       }
     }
@@ -119,7 +119,7 @@ void WriteBufferOperation::executeOpenCLRegion(OpenCLDevice *device,
   /*
    * 1. create cl_mem from outputbuffer
    * 2. call NodeOperation (input) executeOpenCLChunk(.....)
-   * 3. schedule readback from opencl to main device (outputbuffer)
+   * 3. schedule read back from opencl to main device (outputbuffer)
    * 4. schedule native callback
    *
    * note: list of cl_mem will be filled by 2, and needs to be cleaned up by 4
diff --git a/source/blender/depsgraph/DEG_depsgraph.h b/source/blender/depsgraph/DEG_depsgraph.h
index bdcb4061096..7553eca9b64 100644
--- a/source/blender/depsgraph/DEG_depsgraph.h
+++ b/source/blender/depsgraph/DEG_depsgraph.h
@@ -55,9 +55,6 @@ typedef struct Depsgraph Depsgraph;
 
 struct Main;
 
-struct PointerRNA;
-struct PropertyRNA;
-struct RenderEngineType;
 struct Scene;
 struct ViewLayer;
 
@@ -111,6 +108,8 @@ void DEG_graph_on_visible_update(struct Main *bmain, Depsgraph *depsgraph, const
 /* Update all dependency graphs when visible scenes/layers changes. */
 void DEG_on_visible_update(struct Main *bmain, const bool do_time);
 
+/* NOTE: Will return NULL if the flag is not known, allowing to gracefully handle situations
+ * when recalc flag has been removed. */
 const char *DEG_update_tag_as_string(IDRecalcFlag flag);
 
 void DEG_id_tag_update(struct ID *id, int flag);
diff --git a/source/blender/depsgraph/DEG_depsgraph_build.h b/source/blender/depsgraph/DEG_depsgraph_build.h
index 18e06410adf..37dfaf2c3e2 100644
--- a/source/blender/depsgraph/DEG_depsgraph_build.h
+++ b/source/blender/depsgraph/DEG_depsgraph_build.h
@@ -34,12 +34,9 @@ struct Depsgraph;
 /* ------------------------------------------------ */
 
 struct CacheFile;
-struct Collection;
 struct CustomData_MeshMasks;
-struct EffectorWeights;
 struct ID;
 struct Main;
-struct ModifierData;
 struct Object;
 struct Scene;
 struct ViewLayer;
diff --git a/source/blender/depsgraph/DEG_depsgraph_query.h b/source/blender/depsgraph/DEG_depsgraph_query.h
index e3381e79e9c..82f7f33411a 100644
--- a/source/blender/depsgraph/DEG_depsgraph_query.h
+++ b/source/blender/depsgraph/DEG_depsgraph_query.h
@@ -32,14 +32,13 @@
 /* Needed for the instance iterator. */
 #include "DNA_object_types.h"
 
-struct ID;
-
 struct BLI_Iterator;
-struct Base;
 struct CustomData_MeshMasks;
 struct Depsgraph;
 struct DupliObject;
+struct ID;
 struct ListBase;
+struct PointerRNA;
 struct Scene;
 struct ViewLayer;
 
diff --git a/source/blender/depsgraph/intern/builder/deg_builder_cache.cc b/source/blender/depsgraph/intern/builder/deg_builder_cache.cc
index 3df707e92c1..3cfb4f95e5e 100644
--- a/source/blender/depsgraph/intern/builder/deg_builder_cache.cc
+++ b/source/blender/depsgraph/intern/builder/deg_builder_cache.cc
@@ -103,9 +103,9 @@ void animated_property_cb(ID * /*id*/, FCurve *fcurve, void *data_v)
   /* Get storage for the ID.
    * This is needed to deal with cases when nested datablock is animated by its parent. */
   AnimatedPropertyStorage *animated_property_storage = data->animated_property_storage;
-  if (pointer_rna.id.data != data->pointer_rna.id.data) {
+  if (pointer_rna.owner_id != data->pointer_rna.owner_id) {
     animated_property_storage = data->builder_cache->ensureAnimatedPropertyStorage(
-        reinterpret_cast<ID *>(pointer_rna.id.data));
+        pointer_rna.owner_id);
   }
   /* Set the property as animated. */
   animated_property_storage->tagPropertyAsAnimated(&pointer_rna, property_rna);
diff --git a/source/blender/depsgraph/intern/builder/deg_builder_cycle.cc b/source/blender/depsgraph/intern/builder/deg_builder_cycle.cc
index d11a60b77dd..bfb1a981c4d 100644
--- a/source/blender/depsgraph/intern/builder/deg_builder_cycle.cc
+++ b/source/blender/depsgraph/intern/builder/deg_builder_cycle.cc
@@ -149,7 +149,7 @@ bool check_relation_can_murder(Relation *relation)
 
 Relation *select_relation_to_murder(Relation *relation, StackEntry *cycle_start_entry)
 {
-  /* More or less russian roulette solver, which will make sure only
+  /* More or less Russian roulette solver, which will make sure only
    * specially marked relations are kept alive.
    *
    * TODO(sergey): There might be better strategies here. */
diff --git a/source/blender/depsgraph/intern/builder/deg_builder_map.h b/source/blender/depsgraph/intern/builder/deg_builder_map.h
index 0767837a587..65b493e2467 100644
--- a/source/blender/depsgraph/intern/builder/deg_builder_map.h
+++ b/source/blender/depsgraph/intern/builder/deg_builder_map.h
@@ -25,7 +25,6 @@
 
 #include "intern/depsgraph_type.h"
 
-struct GSet;
 struct ID;
 
 namespace DEG {
diff --git a/source/blender/depsgraph/intern/builder/deg_builder_nodes.h b/source/blender/depsgraph/intern/builder/deg_builder_nodes.h
index de9f0e4d6cd..65f3521b556 100644
--- a/source/blender/depsgraph/intern/builder/deg_builder_nodes.h
+++ b/source/blender/depsgraph/intern/builder/deg_builder_nodes.h
@@ -31,7 +31,6 @@
 
 #include "DEG_depsgraph.h"
 
-struct Base;
 struct CacheFile;
 struct Camera;
 struct Collection;
@@ -44,14 +43,12 @@ struct LayerCollection;
 struct Light;
 struct LightProbe;
 struct ListBase;
-struct MTex;
 struct Main;
 struct Mask;
 struct Material;
 struct MovieClip;
 struct Object;
 struct ParticleSettings;
-struct Probe;
 struct Scene;
 struct Speaker;
 struct Tex;
@@ -64,15 +61,12 @@ struct bNodeTree;
 struct bPoseChannel;
 struct bSound;
 
-struct PropertyRNA;
-
 namespace DEG {
 
 struct ComponentNode;
 struct Depsgraph;
 class DepsgraphBuilderCache;
 struct IDNode;
-struct Node;
 struct OperationNode;
 struct TimeSourceNode;
 
diff --git a/source/blender/depsgraph/intern/builder/deg_builder_relations.cc b/source/blender/depsgraph/intern/builder/deg_builder_relations.cc
index 3c226338bfd..c7b545e9feb 100644
--- a/source/blender/depsgraph/intern/builder/deg_builder_relations.cc
+++ b/source/blender/depsgraph/intern/builder/deg_builder_relations.cc
@@ -807,13 +807,25 @@ void DepsgraphRelationBuilder::build_object_parent(Object *object)
 {
   Object *parent = object->parent;
   ID *parent_id = &object->parent->id;
-  ComponentKey ob_key(&object->id, NodeType::TRANSFORM);
-  /* Type-specific links/ */
+  ComponentKey object_transform_key(&object->id, NodeType::TRANSFORM);
+  /* Type-specific links. */
   switch (object->partype) {
     /* Armature Deform (Virtual Modifier) */
     case PARSKEL: {
-      ComponentKey parent_key(parent_id, NodeType::TRANSFORM);
-      add_relation(parent_key, ob_key, "Armature Deform Parent");
+      ComponentKey parent_transform_key(parent_id, NodeType::TRANSFORM);
+      add_relation(parent_transform_key, object_transform_key, "Parent Armature Transform");
+
+      if (parent->type == OB_ARMATURE) {
+        ComponentKey object_geometry_key(&object->id, NodeType::GEOMETRY);
+        ComponentKey parent_pose_key(parent_id, NodeType::EVAL_POSE);
+        add_relation(
+            parent_transform_key, object_geometry_key, "Parent Armature Transform -> Geometry");
+        add_relation(parent_pose_key, object_geometry_key, "Parent Armature Pose -> Geometry");
+
+        add_depends_on_transform_relation(
+            &object->id, object_geometry_key, "Virtual Armature Modifier");
+      }
+
       break;
     }
 
@@ -821,7 +833,7 @@ void DepsgraphRelationBuilder::build_object_parent(Object *object)
     case PARVERT1:
     case PARVERT3: {
       ComponentKey parent_key(parent_id, NodeType::GEOMETRY);
-      add_relation(parent_key, ob_key, "Vertex Parent");
+      add_relation(parent_key, object_transform_key, "Vertex Parent");
       /* Original index is used for optimizations of lookups for subdiv
        * only meshes.
        * TODO(sergey): This optimization got lost at 2.8, so either verify
@@ -833,7 +845,7 @@ void DepsgraphRelationBuilder::build_object_parent(Object *object)
                               DEGCustomDataMeshMasks::MaskFace(CD_MASK_ORIGINDEX) |
                               DEGCustomDataMeshMasks::MaskPoly(CD_MASK_ORIGINDEX));
       ComponentKey transform_key(parent_id, NodeType::TRANSFORM);
-      add_relation(transform_key, ob_key, "Vertex Parent TFM");
+      add_relation(transform_key, object_transform_key, "Vertex Parent TFM");
       break;
     }
 
@@ -842,8 +854,8 @@ void DepsgraphRelationBuilder::build_object_parent(Object *object)
       ComponentKey parent_bone_key(parent_id, NodeType::BONE, object->parsubstr);
       OperationKey parent_transform_key(
           parent_id, NodeType::TRANSFORM, OperationCode::TRANSFORM_FINAL);
-      add_relation(parent_bone_key, ob_key, "Bone Parent");
-      add_relation(parent_transform_key, ob_key, "Armature Parent");
+      add_relation(parent_bone_key, object_transform_key, "Bone Parent");
+      add_relation(parent_transform_key, object_transform_key, "Armature Parent");
       break;
     }
 
@@ -852,8 +864,8 @@ void DepsgraphRelationBuilder::build_object_parent(Object *object)
         /* Lattice Deform Parent - Virtual Modifier. */
         ComponentKey parent_key(parent_id, NodeType::TRANSFORM);
         ComponentKey geom_key(parent_id, NodeType::GEOMETRY);
-        add_relation(parent_key, ob_key, "Lattice Deform Parent");
-        add_relation(geom_key, ob_key, "Lattice Deform Parent Geom");
+        add_relation(parent_key, object_transform_key, "Lattice Deform Parent");
+        add_relation(geom_key, object_transform_key, "Lattice Deform Parent Geom");
       }
       else if (object->parent->type == OB_CURVE) {
         Curve *cu = (Curve *)object->parent->data;
@@ -861,20 +873,20 @@ void DepsgraphRelationBuilder::build_object_parent(Object *object)
         if (cu->flag & CU_PATH) {
           /* Follow Path. */
           ComponentKey parent_key(parent_id, NodeType::GEOMETRY);
-          add_relation(parent_key, ob_key, "Curve Follow Parent");
+          add_relation(parent_key, object_transform_key, "Curve Follow Parent");
           ComponentKey transform_key(parent_id, NodeType::TRANSFORM);
-          add_relation(transform_key, ob_key, "Curve Follow TFM");
+          add_relation(transform_key, object_transform_key, "Curve Follow TFM");
         }
         else {
           /* Standard Parent. */
           ComponentKey parent_key(parent_id, NodeType::TRANSFORM);
-          add_relation(parent_key, ob_key, "Curve Parent");
+          add_relation(parent_key, object_transform_key, "Curve Parent");
         }
       }
       else {
         /* Standard Parent. */
         ComponentKey parent_key(parent_id, NodeType::TRANSFORM);
-        add_relation(parent_key, ob_key, "Parent");
+        add_relation(parent_key, object_transform_key, "Parent");
       }
       break;
     }
@@ -887,7 +899,7 @@ void DepsgraphRelationBuilder::build_object_parent(Object *object)
     ComponentKey parent_geometry_key(parent_id, NodeType::GEOMETRY);
     /* NOTE: Metaballs are evaluating geometry only after their transform,
      * so we only hook up to transform channel here. */
-    add_relation(parent_geometry_key, ob_key, "Parent");
+    add_relation(parent_geometry_key, object_transform_key, "Parent");
   }
 
   /* Dupliverts uses original vertex index. */
@@ -1384,7 +1396,7 @@ void DepsgraphRelationBuilder::build_driver_data(ID *id, FCurve *fcu)
    * it. This is necessary to provide more granular dependencies specifically for
    * Bone objects, because the armature data doesn't have per-bone components,
    * and generic add_relation can only add one link. */
-  ID *id_ptr = (ID *)property_entry_key.ptr.id.data;
+  ID *id_ptr = property_entry_key.ptr.owner_id;
   bool is_bone = id_ptr && property_entry_key.ptr.type == &RNA_Bone;
   /* If the Bone property is referenced via obj.pose.bones[].bone,
    * the RNA pointer refers to the Object ID, so skip to data. */
@@ -1433,8 +1445,8 @@ void DepsgraphRelationBuilder::build_driver_data(ID *id, FCurve *fcu)
       PointerRNA ptr;
       RNA_id_pointer_create(id, &id_ptr);
       if (RNA_path_resolve_full(&id_ptr, fcu->rna_path, &ptr, NULL, NULL)) {
-        if (id_ptr.id.data != ptr.id.data) {
-          ComponentKey cow_key((ID *)ptr.id.data, NodeType::COPY_ON_WRITE);
+        if (id_ptr.owner_id != ptr.owner_id) {
+          ComponentKey cow_key(ptr.owner_id, NodeType::COPY_ON_WRITE);
           add_relation(cow_key, driver_key, "Driven CoW -> Driver", RELATION_CHECK_BEFORE_ADD);
         }
       }
diff --git a/source/blender/depsgraph/intern/builder/deg_builder_relations.h b/source/blender/depsgraph/intern/builder/deg_builder_relations.h
index 0e15818622f..e58ef989ac9 100644
--- a/source/blender/depsgraph/intern/builder/deg_builder_relations.h
+++ b/source/blender/depsgraph/intern/builder/deg_builder_relations.h
@@ -51,7 +51,6 @@ struct Camera;
 struct Collection;
 struct EffectorWeights;
 struct FCurve;
-struct GHash;
 struct ID;
 struct Image;
 struct Key;
@@ -59,11 +58,9 @@ struct LayerCollection;
 struct Light;
 struct LightProbe;
 struct ListBase;
-struct MTex;
 struct Main;
 struct Mask;
 struct Material;
-struct ModifierData;
 struct MovieClip;
 struct Object;
 struct ParticleSettings;
diff --git a/source/blender/depsgraph/intern/builder/deg_builder_rna.cc b/source/blender/depsgraph/intern/builder/deg_builder_rna.cc
index be494104522..d3ae3da9b56 100644
--- a/source/blender/depsgraph/intern/builder/deg_builder_rna.cc
+++ b/source/blender/depsgraph/intern/builder/deg_builder_rna.cc
@@ -177,7 +177,7 @@ RNANodeIdentifier RNANodeQuery::construct_node_identifier(const PointerRNA *ptr,
     return node_identifier;
   }
   /* Set default values for returns. */
-  node_identifier.id = static_cast<ID *>(ptr->id.data);
+  node_identifier.id = ptr->owner_id;
   node_identifier.component_name = "";
   node_identifier.operation_code = OperationCode::OPERATION;
   node_identifier.operation_name = "";
@@ -239,7 +239,7 @@ RNANodeIdentifier RNANodeQuery::construct_node_identifier(const PointerRNA *ptr,
     return node_identifier;
   }
   else if (RNA_struct_is_a(ptr->type, &RNA_Constraint)) {
-    const Object *object = static_cast<const Object *>(ptr->id.data);
+    const Object *object = reinterpret_cast<const Object *>(ptr->owner_id);
     const bConstraint *constraint = static_cast<const bConstraint *>(ptr->data);
     RNANodeQueryIDData *id_data = ensure_id_data(&object->id);
     /* Check whether is object or bone constraint. */
@@ -259,7 +259,7 @@ RNANodeIdentifier RNANodeQuery::construct_node_identifier(const PointerRNA *ptr,
     return node_identifier;
   }
   else if (ELEM(ptr->type, &RNA_ConstraintTarget, &RNA_ConstraintTargetBone)) {
-    Object *object = (Object *)ptr->id.data;
+    Object *object = reinterpret_cast<Object *>(ptr->owner_id);
     bConstraintTarget *tgt = (bConstraintTarget *)ptr->data;
     /* Check whether is object or bone constraint. */
     bPoseChannel *pchan = NULL;
@@ -323,14 +323,14 @@ RNANodeIdentifier RNANodeQuery::construct_node_identifier(const PointerRNA *ptr,
   }
   else if (ptr->type == &RNA_ShapeKey) {
     KeyBlock *key_block = static_cast<KeyBlock *>(ptr->data);
-    node_identifier.id = static_cast<ID *>(ptr->id.data);
+    node_identifier.id = ptr->owner_id;
     node_identifier.type = NodeType::PARAMETERS;
     node_identifier.operation_code = OperationCode::PARAMETERS_EVAL;
     node_identifier.operation_name = key_block->name;
     return node_identifier;
   }
   else if (ptr->type == &RNA_Key) {
-    node_identifier.id = static_cast<ID *>(ptr->id.data);
+    node_identifier.id = ptr->owner_id;
     node_identifier.type = NodeType::GEOMETRY;
     return node_identifier;
   }
@@ -348,12 +348,12 @@ RNANodeIdentifier RNANodeQuery::construct_node_identifier(const PointerRNA *ptr,
     return node_identifier;
   }
   else if (ELEM(ptr->type, &RNA_Curve, &RNA_TextCurve)) {
-    node_identifier.id = (ID *)ptr->id.data;
+    node_identifier.id = ptr->owner_id;
     node_identifier.type = NodeType::GEOMETRY;
     return node_identifier;
   }
   else if (ELEM(ptr->type, &RNA_BezierSplinePoint, &RNA_SplinePoint)) {
-    node_identifier.id = (ID *)ptr->id.data;
+    node_identifier.id = ptr->owner_id;
     node_identifier.type = NodeType::GEOMETRY;
     return node_identifier;
   }
diff --git a/source/blender/depsgraph/intern/depsgraph.h b/source/blender/depsgraph/intern/depsgraph.h
index 15ce57cb169..96b1a2a1f8a 100644
--- a/source/blender/depsgraph/intern/depsgraph.h
+++ b/source/blender/depsgraph/intern/depsgraph.h
@@ -45,13 +45,11 @@
 struct GHash;
 struct GSet;
 struct ID;
-struct Main;
 struct Scene;
 struct ViewLayer;
 
 namespace DEG {
 
-struct ComponentNode;
 struct IDNode;
 struct Node;
 struct OperationNode;
diff --git a/source/blender/depsgraph/intern/depsgraph_query.cc b/source/blender/depsgraph/intern/depsgraph_query.cc
index 23f2bf4194f..8556a351e2b 100644
--- a/source/blender/depsgraph/intern/depsgraph_query.cc
+++ b/source/blender/depsgraph/intern/depsgraph_query.cc
@@ -205,11 +205,11 @@ void DEG_get_evaluated_rna_pointer(const Depsgraph *depsgraph,
   if ((ptr == NULL) || (r_ptr_eval == NULL)) {
     return;
   }
-  ID *orig_id = (ID *)ptr->id.data;
+  ID *orig_id = ptr->owner_id;
   ID *cow_id = DEG_get_evaluated_id(depsgraph, orig_id);
-  if (ptr->id.data == ptr->data) {
+  if (ptr->owner_id == ptr->data) {
     /* For ID pointers, it's easy... */
-    r_ptr_eval->id.data = (void *)cow_id;
+    r_ptr_eval->owner_id = cow_id;
     r_ptr_eval->data = (void *)cow_id;
     r_ptr_eval->type = ptr->type;
   }
@@ -220,7 +220,7 @@ void DEG_get_evaluated_rna_pointer(const Depsgraph *depsgraph,
     const Object *ob_eval = (Object *)cow_id;
     bPoseChannel *pchan = (bPoseChannel *)ptr->data;
     const bPoseChannel *pchan_eval = BKE_pose_channel_find_name(ob_eval->pose, pchan->name);
-    r_ptr_eval->id.data = (void *)cow_id;
+    r_ptr_eval->owner_id = cow_id;
     r_ptr_eval->data = (void *)pchan_eval;
     r_ptr_eval->type = ptr->type;
   }
diff --git a/source/blender/depsgraph/intern/depsgraph_tag.cc b/source/blender/depsgraph/intern/depsgraph_tag.cc
index 392514990e3..72ef495f1d5 100644
--- a/source/blender/depsgraph/intern/depsgraph_tag.cc
+++ b/source/blender/depsgraph/intern/depsgraph_tag.cc
@@ -412,11 +412,15 @@ static void graph_id_tag_update_single_flag(Main *bmain,
 
 string stringify_append_bit(const string &str, IDRecalcFlag tag)
 {
+  const char *tag_name = DEG_update_tag_as_string(tag);
+  if (tag_name == NULL) {
+    return str;
+  }
   string result = str;
   if (!result.empty()) {
     result += ", ";
   }
-  result += DEG_update_tag_as_string(tag);
+  result += tag_name;
   return result;
 }
 
@@ -638,7 +642,7 @@ void graph_id_tag_update(
    * This way IDs in the undo steps will have this flag preserved, making it possible to restore
    * all needed tags when new dependency graph is created on redo.
    * This is the only way to ensure modifications to animation data (such as keyframes i.e.)
-   * properly triggers animation update for the newely constructed dependency graph on redo (while
+   * properly triggers animation update for the newly constructed dependency graph on redo (while
    * usually newly created dependency graph skips animation update to avoid loss of unkeyed
    * changes). */
   if (update_source == DEG_UPDATE_SOURCE_USER_EDIT) {
@@ -717,8 +721,7 @@ const char *DEG_update_tag_as_string(IDRecalcFlag flag)
     case ID_RECALC_ALL:
       return "ALL";
   }
-  BLI_assert(!"Unhandled update flag, should never happen!");
-  return "UNKNOWN";
+  return NULL;
 }
 
 /* Data-Based Tagging  */
diff --git a/source/blender/depsgraph/intern/node/deg_node.h b/source/blender/depsgraph/intern/node/deg_node.h
index eea69502baa..acfc8d19bc7 100644
--- a/source/blender/depsgraph/intern/node/deg_node.h
+++ b/source/blender/depsgraph/intern/node/deg_node.h
@@ -29,7 +29,6 @@
 
 #include "DEG_depsgraph_build.h"
 
-struct GHash;
 struct ID;
 struct Scene;
 
diff --git a/source/blender/depsgraph/intern/node/deg_node_id.h b/source/blender/depsgraph/intern/node/deg_node_id.h
index e6eb91a6b2d..35184253f5c 100644
--- a/source/blender/depsgraph/intern/node/deg_node_id.h
+++ b/source/blender/depsgraph/intern/node/deg_node_id.h
@@ -26,6 +26,8 @@
 #include "intern/node/deg_node.h"
 #include "BLI_sys_types.h"
 
+struct GHash;
+
 namespace DEG {
 
 struct ComponentNode;
diff --git a/source/blender/depsgraph/intern/node/deg_node_operation.h b/source/blender/depsgraph/intern/node/deg_node_operation.h
index b0776d186bb..87c1a7f10a1 100644
--- a/source/blender/depsgraph/intern/node/deg_node_operation.h
+++ b/source/blender/depsgraph/intern/node/deg_node_operation.h
@@ -28,7 +28,6 @@
 #include "intern/depsgraph_type.h"
 
 struct Depsgraph;
-struct ID;
 
 namespace DEG {
 
diff --git a/source/blender/draw/CMakeLists.txt b/source/blender/draw/CMakeLists.txt
index fc91ac371cb..9b0f5f743f6 100644
--- a/source/blender/draw/CMakeLists.txt
+++ b/source/blender/draw/CMakeLists.txt
@@ -328,6 +328,7 @@ data_to_c_simple(modes/shaders/overlay_face_wireframe_geom.glsl SRC)
 data_to_c_simple(modes/shaders/overlay_face_wireframe_frag.glsl SRC)
 data_to_c_simple(modes/shaders/object_camera_image_frag.glsl SRC)
 data_to_c_simple(modes/shaders/object_camera_image_vert.glsl SRC)
+data_to_c_simple(modes/shaders/object_color_axes_vert.glsl SRC)
 data_to_c_simple(modes/shaders/object_empty_axes_vert.glsl SRC)
 data_to_c_simple(modes/shaders/object_empty_image_frag.glsl SRC)
 data_to_c_simple(modes/shaders/object_empty_image_vert.glsl SRC)
diff --git a/source/blender/draw/DRW_engine.h b/source/blender/draw/DRW_engine.h
index 6cae9ceb7d6..34d3ae1d207 100644
--- a/source/blender/draw/DRW_engine.h
+++ b/source/blender/draw/DRW_engine.h
@@ -26,28 +26,20 @@
 #include "BLI_sys_types.h" /* for bool */
 
 struct ARegion;
-struct Base;
 struct DRWInstanceDataList;
-struct DRWPass;
 struct Depsgraph;
 struct DrawEngineType;
 struct GPUMaterial;
 struct GPUOffScreen;
 struct GPUViewport;
 struct ID;
-struct IDProperty;
 struct Main;
-struct Material;
 struct Object;
-struct RegionView3D;
 struct RenderEngine;
 struct RenderEngineType;
 struct Scene;
 struct View3D;
-struct ViewContext;
 struct ViewLayer;
-struct ViewportEngineData;
-struct WorkSpace;
 struct bContext;
 struct rcti;
 
@@ -166,8 +158,8 @@ void DRW_opengl_context_disable_ex(bool restore);
 
 void DRW_opengl_render_context_enable(void *re_gl_context);
 void DRW_opengl_render_context_disable(void *re_gl_context);
-void DRW_gawain_render_context_enable(void *re_gpu_context);
-void DRW_gawain_render_context_disable(void *re_gpu_context);
+void DRW_gpu_render_context_enable(void *re_gpu_context);
+void DRW_gpu_render_context_disable(void *re_gpu_context);
 
 void DRW_deferred_shader_remove(struct GPUMaterial *mat);
 
diff --git a/source/blender/draw/DRW_select_buffer.h b/source/blender/draw/DRW_select_buffer.h
index 4aa1c403710..0ad13af81ba 100644
--- a/source/blender/draw/DRW_select_buffer.h
+++ b/source/blender/draw/DRW_select_buffer.h
@@ -30,7 +30,6 @@ struct Base;
 struct Depsgraph;
 struct Object;
 struct View3D;
-struct ViewLayer;
 struct rcti;
 
 typedef struct SELECTID_ObjectData {
@@ -58,9 +57,6 @@ struct ObjectOffsets {
 };
 
 struct SELECTID_Context {
-  struct GPUFrameBuffer *framebuffer_select_id;
-  struct GPUTexture *texture_u32;
-
   /* All context objects */
   struct Object **objects;
   uint objects_len;
@@ -131,7 +127,4 @@ void DRW_select_buffer_context_create(struct Base **bases,
                                       const uint bases_len,
                                       short select_mode);
 
-/* select_engine.c */
-struct SELECTID_Context *DRW_select_engine_context_get(void);
-
 #endif /* __DRW_SELECT_BUFFER_H__ */
diff --git a/source/blender/draw/engines/eevee/eevee_effects.c b/source/blender/draw/engines/eevee/eevee_effects.c
index 93521c71127..0ca1e0b2858 100644
--- a/source/blender/draw/engines/eevee/eevee_effects.c
+++ b/source/blender/draw/engines/eevee/eevee_effects.c
@@ -509,7 +509,7 @@ void EEVEE_downsample_buffer(EEVEE_Data *vedata, GPUTexture *texture_src, int le
 }
 
 /**
- * Simple downsampling algorithm for cubemap. Reconstruct mip chain up to mip level.
+ * Simple down-sampling algorithm for cubemap. Reconstruct mip chain up to mip level.
  */
 void EEVEE_downsample_cube_buffer(EEVEE_Data *vedata, GPUTexture *texture_src, int level)
 {
@@ -580,7 +580,7 @@ void EEVEE_draw_effects(EEVEE_ViewLayerData *UNUSED(sldata), EEVEE_Data *vedata)
   /* NOTE: Lookdev drawing happens before TAA but after
    * motion blur and dof to avoid distortions.
    * Velocity resolve use a hack to exclude lookdev
-   * spheres from creating shimering reprojection vectors. */
+   * spheres from creating shimmering re-projection vectors. */
   EEVEE_lookdev_draw(vedata);
   EEVEE_velocity_resolve(vedata);
 
diff --git a/source/blender/draw/engines/eevee/eevee_lightcache.c b/source/blender/draw/engines/eevee/eevee_lightcache.c
index c82a112b343..05aea652591 100644
--- a/source/blender/draw/engines/eevee/eevee_lightcache.c
+++ b/source/blender/draw/engines/eevee/eevee_lightcache.c
@@ -80,8 +80,8 @@ extern void DRW_opengl_context_disable(void);
 
 extern void DRW_opengl_render_context_enable(void *re_gl_context);
 extern void DRW_opengl_render_context_disable(void *re_gl_context);
-extern void DRW_gawain_render_context_enable(void *re_gpu_context);
-extern void DRW_gawain_render_context_disable(void *re_gpu_context);
+extern void DRW_gpu_render_context_enable(void *re_gpu_context);
+extern void DRW_gpu_render_context_disable(void *re_gpu_context);
 
 typedef struct EEVEE_LightBake {
   Depsgraph *depsgraph;
@@ -138,7 +138,7 @@ typedef struct EEVEE_LightBake {
   float vis_res;
   /** Result of previous light bounce. */
   GPUTexture *grid_prev;
-  /** Pointer to the id.data of the probe object. */
+  /** Pointer to the owner_id of the probe object. */
   LightProbe **grid_prb;
 
   /* Reflection probe */
@@ -148,7 +148,7 @@ typedef struct EEVEE_LightBake {
   int ref_cube_res;
   /** Index of the current cube. */
   int cube_offset;
-  /** Pointer to the id.data of the probe object. */
+  /** Pointer to the owner_id of the probe object. */
   LightProbe **cube_prb;
 
   /* Dummy Textures */
@@ -412,7 +412,7 @@ static void eevee_lightbake_context_enable(EEVEE_LightBake *lbake)
     if (lbake->gpu_context == NULL) {
       lbake->gpu_context = GPU_context_create(0);
     }
-    DRW_gawain_render_context_enable(lbake->gpu_context);
+    DRW_gpu_render_context_enable(lbake->gpu_context);
   }
   else {
     DRW_opengl_context_enable();
@@ -422,7 +422,7 @@ static void eevee_lightbake_context_enable(EEVEE_LightBake *lbake)
 static void eevee_lightbake_context_disable(EEVEE_LightBake *lbake)
 {
   if (lbake->gl_context) {
-    DRW_gawain_render_context_disable(lbake->gpu_context);
+    DRW_gpu_render_context_disable(lbake->gpu_context);
     DRW_opengl_render_context_disable(lbake->gl_context);
   }
   else {
@@ -654,7 +654,7 @@ static void eevee_lightbake_delete_resources(EEVEE_LightBake *lbake)
 
   if (lbake->gl_context) {
     DRW_opengl_render_context_enable(lbake->gl_context);
-    DRW_gawain_render_context_enable(lbake->gpu_context);
+    DRW_gpu_render_context_enable(lbake->gpu_context);
   }
   else if (!lbake->resource_only) {
     DRW_opengl_context_enable();
@@ -675,8 +675,8 @@ static void eevee_lightbake_delete_resources(EEVEE_LightBake *lbake)
   }
 
   if (lbake->gpu_context) {
-    DRW_gawain_render_context_disable(lbake->gpu_context);
-    DRW_gawain_render_context_enable(lbake->gpu_context);
+    DRW_gpu_render_context_disable(lbake->gpu_context);
+    DRW_gpu_render_context_enable(lbake->gpu_context);
     GPU_context_discard(lbake->gpu_context);
   }
 
diff --git a/source/blender/draw/engines/eevee/eevee_lightcache.h b/source/blender/draw/engines/eevee/eevee_lightcache.h
index 5aea712a910..ede2de13dce 100644
--- a/source/blender/draw/engines/eevee/eevee_lightcache.h
+++ b/source/blender/draw/engines/eevee/eevee_lightcache.h
@@ -26,7 +26,6 @@
 #include "BLI_sys_types.h" /* for bool */
 
 struct EEVEE_Data;
-struct EEVEE_LightBake;
 struct EEVEE_ViewLayerData;
 struct LightCache;
 struct Scene;
diff --git a/source/blender/draw/engines/eevee/eevee_lightprobes.c b/source/blender/draw/engines/eevee/eevee_lightprobes.c
index 2026b44fe87..8b1309e8537 100644
--- a/source/blender/draw/engines/eevee/eevee_lightprobes.c
+++ b/source/blender/draw/engines/eevee/eevee_lightprobes.c
@@ -1213,7 +1213,7 @@ void EEVEE_lightbake_filter_visibility(EEVEE_ViewLayerData *sldata,
   DRW_draw_pass(psl->probe_visibility_compute);
 }
 
-/* Actually a simple downsampling */
+/* Actually a simple down-sampling. */
 static void downsample_planar(void *vedata, int level)
 {
   EEVEE_PassList *psl = ((EEVEE_Data *)vedata)->psl;
diff --git a/source/blender/draw/engines/eevee/eevee_lights.c b/source/blender/draw/engines/eevee/eevee_lights.c
index c15872b31fa..0bfc23b8354 100644
--- a/source/blender/draw/engines/eevee/eevee_lights.c
+++ b/source/blender/draw/engines/eevee/eevee_lights.c
@@ -779,7 +779,7 @@ static void eevee_light_setup(Object *ob, EEVEE_Light *evli)
 
 /**
  * Special ball distribution:
- * Point are distributed in a way that when they are orthogonaly
+ * Point are distributed in a way that when they are orthogonally
  * projected into any plane, the resulting distribution is (close to)
  * a uniform disc distribution.
  */
diff --git a/source/blender/draw/engines/eevee/eevee_lut.c b/source/blender/draw/engines/eevee/eevee_lut.c
index 8db45e4cee6..b380269db99 100644
--- a/source/blender/draw/engines/eevee/eevee_lut.c
+++ b/source/blender/draw/engines/eevee/eevee_lut.c
@@ -21,6 +21,8 @@
  * \ingroup gpu
  */
 
+#include "eevee_lut.h" /* own include */
+
 const float ltc_mat_ggx[64 * 64 * 4] = {
     1.000000, 0.000000,  0.000000, 0.000020, 1.000000, 0.000000,  0.000000,  0.000504,
     1.000000, 0.000000,  0.000000, 0.002016, 1.000000, 0.000000,  0.000000,  0.004535,
diff --git a/source/blender/draw/engines/eevee/eevee_private.h b/source/blender/draw/engines/eevee/eevee_private.h
index 27edd44075b..3cdafee95a2 100644
--- a/source/blender/draw/engines/eevee/eevee_private.h
+++ b/source/blender/draw/engines/eevee/eevee_private.h
@@ -30,7 +30,6 @@ struct EEVEE_ShadowCasterBuffer;
 struct GPUFrameBuffer;
 struct Object;
 struct RenderLayer;
-struct RenderResult;
 
 extern struct DrawEngineType draw_engine_eevee_type;
 
diff --git a/source/blender/draw/engines/gpencil/gpencil_cache_utils.c b/source/blender/draw/engines/gpencil/gpencil_cache_utils.c
index b0608343ab3..33e6d73eeeb 100644
--- a/source/blender/draw/engines/gpencil/gpencil_cache_utils.c
+++ b/source/blender/draw/engines/gpencil/gpencil_cache_utils.c
@@ -349,13 +349,13 @@ void DRW_gpencil_freecache(struct Object *ob)
     }
   }
 
-  /* clear all frames derived data */
-  for (int i = 0; i < ob->runtime.tot_layers; i++) {
-    bGPDframe *derived_gpf = &ob->runtime.derived_frames[i];
-    BKE_gpencil_free_frame_runtime_data(derived_gpf);
-    derived_gpf = NULL;
+  /* clear all frames evaluated data */
+  for (int i = 0; i < ob->runtime.gpencil_tot_layers; i++) {
+    bGPDframe *gpf_eval = &ob->runtime.gpencil_evaluated_frames[i];
+    BKE_gpencil_free_frame_runtime_data(gpf_eval);
+    gpf_eval = NULL;
   }
 
-  ob->runtime.tot_layers = 0;
-  MEM_SAFE_FREE(ob->runtime.derived_frames);
+  ob->runtime.gpencil_tot_layers = 0;
+  MEM_SAFE_FREE(ob->runtime.gpencil_evaluated_frames);
 }
diff --git a/source/blender/draw/engines/gpencil/gpencil_draw_cache_impl.c b/source/blender/draw/engines/gpencil/gpencil_draw_cache_impl.c
index 1ec02cac109..541a9e31586 100644
--- a/source/blender/draw/engines/gpencil/gpencil_draw_cache_impl.c
+++ b/source/blender/draw/engines/gpencil/gpencil_draw_cache_impl.c
@@ -821,7 +821,7 @@ void gpencil_get_edit_geom(struct GpencilBatchCacheElem *be,
 void gpencil_get_edlin_geom(struct GpencilBatchCacheElem *be,
                             bGPDstroke *gps,
                             float alpha,
-                            short UNUSED(dflag))
+                            const bool hide_select)
 {
   const DRWContextState *draw_ctx = DRW_context_state_get();
   Object *ob = draw_ctx->obact;
@@ -866,7 +866,7 @@ void gpencil_get_edlin_geom(struct GpencilBatchCacheElem *be,
       copy_v4_v4(fcolor, selectColor);
     }
     else {
-      if (pt->flag & GP_SPOINT_SELECT) {
+      if ((pt->flag & GP_SPOINT_SELECT) && (!hide_select)) {
         copy_v4_v4(fcolor, selectColor);
       }
       else {
diff --git a/source/blender/draw/engines/gpencil/gpencil_draw_utils.c b/source/blender/draw/engines/gpencil/gpencil_draw_utils.c
index 85ad1689c23..bef72b9a054 100644
--- a/source/blender/draw/engines/gpencil/gpencil_draw_utils.c
+++ b/source/blender/draw/engines/gpencil/gpencil_draw_utils.c
@@ -65,8 +65,7 @@
 static void gpencil_calc_vertex(GPENCIL_StorageList *stl,
                                 tGPencilObjectCache *cache_ob,
                                 GpencilBatchCache *cache,
-                                bGPdata *gpd,
-                                int cfra_eval)
+                                bGPdata *gpd)
 {
   if (!cache->is_dirty) {
     return;
@@ -88,13 +87,13 @@ static void gpencil_calc_vertex(GPENCIL_StorageList *stl,
 
   cache_ob->tot_vertex = 0;
   cache_ob->tot_triangles = 0;
-  int derived_idx = 0;
+  int idx_eval = 0;
 
   for (bGPDlayer *gpl = gpd->layers.first; gpl; gpl = gpl->next) {
     bGPDframe *init_gpf = NULL;
     const bool is_onion = ((do_onion) && (gpl->onion_flag & GP_LAYER_ONIONSKIN));
     if (gpl->flag & GP_LAYER_HIDE) {
-      derived_idx++;
+      idx_eval++;
       continue;
     }
 
@@ -103,7 +102,7 @@ static void gpencil_calc_vertex(GPENCIL_StorageList *stl,
       init_gpf = gpl->frames.first;
     }
     else {
-      init_gpf = &ob->runtime.derived_frames[derived_idx];
+      init_gpf = &ob->runtime.gpencil_evaluated_frames[idx_eval];
     }
 
     if (init_gpf == NULL) {
@@ -119,7 +118,7 @@ static void gpencil_calc_vertex(GPENCIL_StorageList *stl,
         break;
       }
     }
-    derived_idx++;
+    idx_eval++;
   }
 
   cache->b_fill.tot_vertex = cache_ob->tot_triangles * 3;
@@ -939,6 +938,17 @@ static void gpencil_add_editpoints_vertexdata(GpencilBatchCache *cache,
 {
   const DRWContextState *draw_ctx = DRW_context_state_get();
   View3D *v3d = draw_ctx->v3d;
+  ToolSettings *ts = draw_ctx->scene->toolsettings;
+  const bool use_sculpt_mask = (GPENCIL_SCULPT_MODE(gpd) && (ts->gpencil_selectmode_sculpt &
+                                                             (GP_SCULPT_MASK_SELECTMODE_POINT |
+                                                              GP_SCULPT_MASK_SELECTMODE_STROKE |
+                                                              GP_SCULPT_MASK_SELECTMODE_SEGMENT)));
+
+  const bool show_sculpt_points = (GPENCIL_SCULPT_MODE(gpd) &&
+                                   (ts->gpencil_selectmode_sculpt &
+                                    (GP_SCULPT_MASK_SELECTMODE_POINT |
+                                     GP_SCULPT_MASK_SELECTMODE_SEGMENT)));
+
   MaterialGPencilStyle *gp_style = BKE_material_gpencil_settings_get(ob, gps->mat_nr + 1);
 
   /* alpha factor for edit points/line to make them more subtle */
@@ -951,11 +961,24 @@ static void gpencil_add_editpoints_vertexdata(GpencilBatchCache *cache,
     }
     const bool is_weight_paint = (gpd) && (gpd->flag & GP_DATA_STROKE_WEIGHTMODE);
 
+    /* If Sculpt mode and the mask is disabled, the select must be hidden. */
+    const bool hide_select = GPENCIL_SCULPT_MODE(gpd) && !use_sculpt_mask;
+
+    /* Show Edit points if:
+     *  Edit mode: Not in Stroke selection mode
+     *  Sculpt mode: Not in Stroke mask mode and any other mask mode enabled
+     *  Weight mode: Always
+     */
+    const bool show_points = (show_sculpt_points) || (is_weight_paint) ||
+                             (GPENCIL_EDIT_MODE(gpd) &&
+                              ((ts->gpencil_selectmode_edit & GP_SELECTMODE_STROKE) == 0));
+
     if (cache->is_dirty) {
       if ((obact == ob) && ((v3d->flag2 & V3D_HIDE_OVERLAYS) == 0) &&
           (v3d->gp_flag & V3D_GP_SHOW_EDIT_LINES)) {
+
         /* line of the original stroke */
-        gpencil_get_edlin_geom(&cache->b_edlin, gps, edit_alpha, gpd->flag);
+        gpencil_get_edlin_geom(&cache->b_edlin, gps, edit_alpha, hide_select);
 
         /* add to list of groups */
         cache->grp_cache = gpencil_group_cache_add(cache->grp_cache,
@@ -968,6 +991,12 @@ static void gpencil_add_editpoints_vertexdata(GpencilBatchCache *cache,
                                                    &cache->grp_size,
                                                    &cache->grp_used);
       }
+
+      /* If the points are hidden return. */
+      if ((!show_points) || (hide_select)) {
+        return;
+      }
+
       /* edit points */
       if ((gps->flag & GP_STROKE_SELECT) || (is_weight_paint)) {
         if ((gpl->flag & GP_LAYER_UNLOCK_COLOR) ||
@@ -1797,7 +1826,7 @@ void gpencil_populate_multiedit(GPENCIL_e_data *e_data,
   const bool playing = stl->storage->is_playing;
 
   /* calc max size of VBOs */
-  gpencil_calc_vertex(stl, cache_ob, cache, gpd, cfra_eval);
+  gpencil_calc_vertex(stl, cache_ob, cache, gpd);
 
   /* draw strokes */
   for (bGPDlayer *gpl = gpd->layers.first; gpl; gpl = gpl->next) {
@@ -1860,6 +1889,14 @@ void gpencil_populate_datablock(GPENCIL_e_data *e_data,
   const ViewLayer *view_layer = DEG_get_evaluated_view_layer(draw_ctx->depsgraph);
   Scene *scene = draw_ctx->scene;
 
+  /* TODO: Review why is needed this recalc when render cycles + GP object in background.
+   * We need these lines to keep running the background render, but asap we get an alternative
+   * solution, we must remove it and keep all logic inside gpencil_modifier module. (antoniov)
+   */
+  if (ob->runtime.gpencil_tot_layers == 0) {
+    BKE_gpencil_modifiers_calc(draw_ctx->depsgraph, draw_ctx->scene, ob);
+  }
+
   /* Use original data to shared in edit/transform operators */
   bGPdata *gpd_eval = (bGPdata *)ob->data;
   bGPdata *gpd = (bGPdata *)DEG_get_original_id(&gpd_eval->id);
@@ -1867,7 +1904,7 @@ void gpencil_populate_datablock(GPENCIL_e_data *e_data,
   View3D *v3d = draw_ctx->v3d;
   int cfra_eval = (int)DEG_get_ctime(draw_ctx->depsgraph);
 
-  bGPDframe *derived_gpf = NULL;
+  bGPDframe *gpf_eval = NULL;
   const bool overlay = v3d != NULL ? (bool)((v3d->flag2 & V3D_HIDE_OVERLAYS) == 0) : true;
   const bool time_remap = BKE_gpencil_has_time_modifiers(ob);
 
@@ -1887,7 +1924,7 @@ void gpencil_populate_datablock(GPENCIL_e_data *e_data,
   }
 
   /* calc max size of VBOs */
-  gpencil_calc_vertex(stl, cache_ob, cache, gpd, cfra_eval);
+  gpencil_calc_vertex(stl, cache_ob, cache, gpd);
 
   /* draw normal strokes */
   for (bGPDlayer *gpl = gpd->layers.first; gpl; gpl = gpl->next) {
@@ -1936,9 +1973,9 @@ void gpencil_populate_datablock(GPENCIL_e_data *e_data,
       opacity = opacity * v3d->overlay.gpencil_fade_layer;
     }
 
-    /* Get derived frames array data */
-    int derived_idx = BLI_findindex(&gpd->layers, gpl);
-    derived_gpf = &ob->runtime.derived_frames[derived_idx];
+    /* Get evaluated frames array data */
+    int idx_eval = BLI_findindex(&gpd->layers, gpl);
+    gpf_eval = &ob->runtime.gpencil_evaluated_frames[idx_eval];
 
     /* draw onion skins */
     if (!ID_IS_LINKED(&gpd->id)) {
@@ -1952,17 +1989,8 @@ void gpencil_populate_datablock(GPENCIL_e_data *e_data,
       }
     }
     /* draw normal strokes */
-    gpencil_draw_strokes(cache,
-                         e_data,
-                         vedata,
-                         ob,
-                         gpd,
-                         gpl,
-                         derived_gpf,
-                         opacity,
-                         gpl->tintcolor,
-                         false,
-                         cache_ob);
+    gpencil_draw_strokes(
+        cache, e_data, vedata, ob, gpd, gpl, gpf_eval, opacity, gpl->tintcolor, false, cache_ob);
   }
 
   /* create batchs and shading groups */
diff --git a/source/blender/draw/engines/gpencil/gpencil_engine.c b/source/blender/draw/engines/gpencil/gpencil_engine.c
index b0898cb33f3..d03f975ca37 100644
--- a/source/blender/draw/engines/gpencil/gpencil_engine.c
+++ b/source/blender/draw/engines/gpencil/gpencil_engine.c
@@ -405,10 +405,10 @@ void GPENCIL_cache_init(void *vedata)
     }
 
     /* save simplify flags (can change while drawing, so it's better to save) */
-    stl->storage->simplify_fill = GP_SIMPLIFY_FILL(scene, stl->storage->is_playing);
-    stl->storage->simplify_modif = GP_SIMPLIFY_MODIF(scene, stl->storage->is_playing);
-    stl->storage->simplify_fx = GP_SIMPLIFY_FX(scene, stl->storage->is_playing);
-    stl->storage->simplify_blend = GP_SIMPLIFY_BLEND(scene, stl->storage->is_playing);
+    stl->storage->simplify_fill = GPENCIL_SIMPLIFY_FILL(scene, stl->storage->is_playing);
+    stl->storage->simplify_modif = GPENCIL_SIMPLIFY_MODIF(scene, stl->storage->is_playing);
+    stl->storage->simplify_fx = GPENCIL_SIMPLIFY_FX(scene, stl->storage->is_playing);
+    stl->storage->simplify_blend = GPENCIL_SIMPLIFY_BLEND(scene, stl->storage->is_playing);
 
     /* xray mode */
     if (v3d) {
diff --git a/source/blender/draw/engines/gpencil/gpencil_engine.h b/source/blender/draw/engines/gpencil/gpencil_engine.h
index c3167e21f27..f29e83f64cf 100644
--- a/source/blender/draw/engines/gpencil/gpencil_engine.h
+++ b/source/blender/draw/engines/gpencil/gpencil_engine.h
@@ -28,15 +28,14 @@
 struct GPENCIL_Data;
 struct GPENCIL_StorageList;
 struct MaterialGPencilStyle;
-struct ModifierData;
 struct Object;
 struct RenderEngine;
 struct RenderLayer;
 struct bGPDstroke;
-struct tGPspoint;
-struct GpencilBatchCache;
-struct GpencilBatchCacheElem;
-struct GpencilBatchGroup;
+
+struct GPUBatch;
+struct GPUVertBuf;
+struct GPUVertFormat;
 
 #define GPENCIL_CACHE_BLOCK_SIZE 8
 #define GPENCIL_MAX_SHGROUPS 65536
@@ -327,6 +326,64 @@ typedef struct GPENCIL_e_data {
 
 } GPENCIL_e_data; /* Engine data */
 
+/* GPUBatch Cache Element */
+typedef struct GpencilBatchCacheElem {
+  GPUBatch *batch;
+  GPUVertBuf *vbo;
+  int vbo_len;
+  /* attr ids */
+  GPUVertFormat *format;
+  uint pos_id;
+  uint color_id;
+  uint thickness_id;
+  uint uvdata_id;
+  uint prev_pos_id;
+
+  /* size for VBO alloc */
+  int tot_vertex;
+} GpencilBatchCacheElem;
+
+/* Defines each batch group to define later the shgroup */
+typedef struct GpencilBatchGroup {
+  struct bGPDlayer *gpl;  /* reference to original layer */
+  struct bGPDframe *gpf;  /* reference to original frame */
+  struct bGPDstroke *gps; /* reference to original stroke */
+  short type;             /* type of element */
+  bool onion;             /* the group is part of onion skin */
+  int vertex_idx;         /* index of vertex data */
+} GpencilBatchGroup;
+
+typedef enum GpencilBatchGroup_Type {
+  eGpencilBatchGroupType_Stroke = 1,
+  eGpencilBatchGroupType_Point = 2,
+  eGpencilBatchGroupType_Fill = 3,
+  eGpencilBatchGroupType_Edit = 4,
+  eGpencilBatchGroupType_Edlin = 5,
+} GpencilBatchGroup_Type;
+
+/* Runtime data for GPU and evaluated frames after applying modifiers */
+typedef struct GpencilBatchCache {
+  GpencilBatchCacheElem b_stroke;
+  GpencilBatchCacheElem b_point;
+  GpencilBatchCacheElem b_fill;
+  GpencilBatchCacheElem b_edit;
+  GpencilBatchCacheElem b_edlin;
+
+  /** Cache is dirty */
+  bool is_dirty;
+  /** Edit mode flag */
+  bool is_editmode;
+  /** Last cache frame */
+  int cache_frame;
+
+  /** Total groups in arrays */
+  int grp_used;
+  /** Max size of the array */
+  int grp_size;
+  /** Array of cache elements */
+  struct GpencilBatchGroup *grp_cache;
+} GpencilBatchCache;
+
 /* general drawing functions */
 struct DRWShadingGroup *gpencil_shgroup_stroke_create(struct GPENCIL_Data *vedata,
                                                       struct DRWPass *pass,
@@ -380,7 +437,7 @@ void gpencil_get_edit_geom(struct GpencilBatchCacheElem *be,
 void gpencil_get_edlin_geom(struct GpencilBatchCacheElem *be,
                             struct bGPDstroke *gps,
                             float alpha,
-                            short dflag);
+                            const bool hide_select);
 
 struct GPUBatch *gpencil_get_buffer_stroke_geom(struct bGPdata *gpd, short thickness);
 struct GPUBatch *gpencil_get_buffer_fill_geom(struct bGPdata *gpd);
diff --git a/source/blender/draw/engines/select/select_draw_utils.c b/source/blender/draw/engines/select/select_draw_utils.c
index 268cd60a8aa..6a51c28c242 100644
--- a/source/blender/draw/engines/select/select_draw_utils.c
+++ b/source/blender/draw/engines/select/select_draw_utils.c
@@ -44,48 +44,6 @@
 /** \name Draw Utilities
  * \{ */
 
-static void select_id_framebuffer_setup(struct SELECTID_Context *select_ctx)
-{
-  DefaultTextureList *dtxl = DRW_viewport_texture_list_get();
-  int size[2];
-  size[0] = GPU_texture_width(dtxl->depth);
-  size[1] = GPU_texture_height(dtxl->depth);
-
-  if (select_ctx->framebuffer_select_id == NULL) {
-    select_ctx->framebuffer_select_id = GPU_framebuffer_create();
-  }
-
-  if ((select_ctx->texture_u32 != NULL) &&
-      ((GPU_texture_width(select_ctx->texture_u32) != size[0]) ||
-       (GPU_texture_height(select_ctx->texture_u32) != size[1]))) {
-    GPU_texture_free(select_ctx->texture_u32);
-    select_ctx->texture_u32 = NULL;
-  }
-
-  /* Make sure the depth texture is attached.
-   * It may disappear when loading another Blender session. */
-  GPU_framebuffer_texture_attach(select_ctx->framebuffer_select_id, dtxl->depth, 0, 0);
-
-  if (select_ctx->texture_u32 == NULL) {
-    select_ctx->texture_u32 = GPU_texture_create_2d(size[0], size[1], GPU_R32UI, NULL, NULL);
-    GPU_framebuffer_texture_attach(
-        select_ctx->framebuffer_select_id, select_ctx->texture_u32, 0, 0);
-
-    GPU_framebuffer_check_valid(select_ctx->framebuffer_select_id, NULL);
-  }
-}
-
-/* Remove all tags from drawn or culled objects. */
-void select_id_context_clear(struct SELECTID_Context *select_ctx)
-{
-  select_ctx->objects_drawn_len = 0;
-  select_ctx->index_drawn_len = 1;
-  select_id_framebuffer_setup(select_ctx);
-  GPU_framebuffer_bind(select_ctx->framebuffer_select_id);
-  GPU_framebuffer_clear_color_depth(
-      select_ctx->framebuffer_select_id, (const float[4]){0.0f}, 1.0f);
-}
-
 void select_id_object_min_max(Object *obj, float r_min[3], float r_max[3])
 {
   BoundBox *bb;
@@ -104,13 +62,19 @@ void select_id_object_min_max(Object *obj, float r_min[3], float r_max[3])
 short select_id_get_object_select_mode(Scene *scene, Object *ob)
 {
   short r_select_mode = 0;
-  if (ob->mode & (OB_MODE_WEIGHT_PAINT | OB_MODE_VERTEX_PAINT | OB_MODE_TEXTURE_PAINT)) {
+  if (ob->mode & (OB_MODE_WEIGHT_PAINT | OB_MODE_VERTEX_PAINT)) {
     Mesh *me_orig = DEG_get_original_object(ob)->data;
     if (me_orig->editflag & ME_EDIT_PAINT_FACE_SEL) {
       r_select_mode = SCE_SELECT_FACE;
     }
-    if (me_orig->editflag & ME_EDIT_PAINT_VERT_SEL) {
-      r_select_mode |= SCE_SELECT_VERTEX;
+    else if (me_orig->editflag & ME_EDIT_PAINT_VERT_SEL) {
+      r_select_mode = SCE_SELECT_VERTEX;
+    }
+  }
+  else if (ob->mode & OB_MODE_TEXTURE_PAINT) {
+    Mesh *me_orig = DEG_get_original_object(ob)->data;
+    if (me_orig->editflag & ME_EDIT_PAINT_FACE_SEL) {
+      r_select_mode = SCE_SELECT_FACE;
     }
   }
   else {
@@ -151,12 +115,11 @@ static void draw_select_id_edit_mesh(SELECTID_StorageList *stl,
 
   BM_mesh_elem_table_ensure(em->bm, BM_VERT | BM_EDGE | BM_FACE);
 
-  struct GPUBatch *geom_faces;
-  DRWShadingGroup *face_shgrp;
   if (select_mode & SCE_SELECT_FACE) {
-    geom_faces = DRW_mesh_batch_cache_get_triangles_with_select_id(me);
-    face_shgrp = DRW_shgroup_create_sub(stl->g_data->shgrp_face_flat);
+    struct GPUBatch *geom_faces = DRW_mesh_batch_cache_get_triangles_with_select_id(me);
+    DRWShadingGroup *face_shgrp = DRW_shgroup_create_sub(stl->g_data->shgrp_face_flat);
     DRW_shgroup_uniform_int_copy(face_shgrp, "offset", *(int *)&initial_offset);
+    DRW_shgroup_call_no_cull(face_shgrp, geom_faces, ob);
 
     if (draw_facedot) {
       struct GPUBatch *geom_facedots = DRW_mesh_batch_cache_get_facedots_with_select_id(me);
@@ -165,11 +128,13 @@ static void draw_select_id_edit_mesh(SELECTID_StorageList *stl,
     *r_face_offset = initial_offset + em->bm->totface;
   }
   else {
-    geom_faces = DRW_mesh_batch_cache_get_surface(me);
-    face_shgrp = stl->g_data->shgrp_face_unif;
+    if (ob->dt >= OB_SOLID) {
+      struct GPUBatch *geom_faces = DRW_mesh_batch_cache_get_surface(me);
+      DRWShadingGroup *face_shgrp = stl->g_data->shgrp_face_unif;
+      DRW_shgroup_call_no_cull(face_shgrp, geom_faces, ob);
+    }
     *r_face_offset = initial_offset;
   }
-  DRW_shgroup_call_no_cull(face_shgrp, geom_faces, ob);
 
   /* Unlike faces, only draw edges if edge select mode. */
   if (select_mode & SCE_SELECT_EDGE) {
diff --git a/source/blender/draw/engines/select/select_engine.c b/source/blender/draw/engines/select/select_engine.c
index 4c158312e0f..c294bba2083 100644
--- a/source/blender/draw/engines/select/select_engine.c
+++ b/source/blender/draw/engines/select/select_engine.c
@@ -40,10 +40,13 @@
 /* *********** STATIC *********** */
 
 static struct {
+  struct GPUFrameBuffer *framebuffer_select_id;
+  struct GPUTexture *texture_u32;
+
   SELECTID_Shaders sh_data[GPU_SHADER_CFG_LEN];
   struct SELECTID_Context context;
   uint runtime_new_objects;
-} e_data = {{{NULL}}}; /* Engine data */
+} e_data = {NULL}; /* Engine data */
 
 /* Shaders */
 extern char datatoc_common_view_lib_glsl[];
@@ -51,6 +54,41 @@ extern char datatoc_selection_id_3D_vert_glsl[];
 extern char datatoc_selection_id_frag_glsl[];
 
 /* -------------------------------------------------------------------- */
+/** \name Utils
+ * \{ */
+
+static void select_engine_framebuffer_setup(void)
+{
+  DefaultTextureList *dtxl = DRW_viewport_texture_list_get();
+  int size[2];
+  size[0] = GPU_texture_width(dtxl->depth);
+  size[1] = GPU_texture_height(dtxl->depth);
+
+  if (e_data.framebuffer_select_id == NULL) {
+    e_data.framebuffer_select_id = GPU_framebuffer_create();
+  }
+
+  if ((e_data.texture_u32 != NULL) && ((GPU_texture_width(e_data.texture_u32) != size[0]) ||
+                                       (GPU_texture_height(e_data.texture_u32) != size[1]))) {
+    GPU_texture_free(e_data.texture_u32);
+    e_data.texture_u32 = NULL;
+  }
+
+  /* Make sure the depth texture is attached.
+   * It may disappear when loading another Blender session. */
+  GPU_framebuffer_texture_attach(e_data.framebuffer_select_id, dtxl->depth, 0, 0);
+
+  if (e_data.texture_u32 == NULL) {
+    e_data.texture_u32 = GPU_texture_create_2d(size[0], size[1], GPU_R32UI, NULL, NULL);
+    GPU_framebuffer_texture_attach(e_data.framebuffer_select_id, e_data.texture_u32, 0, 0);
+
+    GPU_framebuffer_check_valid(e_data.framebuffer_select_id, NULL);
+  }
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Engine Functions
  * \{ */
 
@@ -125,6 +163,13 @@ static void select_cache_init(void *vedata)
   if (e_data.context.select_mode == -1) {
     e_data.context.select_mode = select_id_get_object_select_mode(draw_ctx->scene,
                                                                   draw_ctx->obact);
+    if (e_data.context.select_mode == 0) {
+      /* Need for sampling weights. */
+      if (draw_ctx->object_mode & OB_MODE_WEIGHT_PAINT) {
+        e_data.context.select_mode = SCE_SELECT_FACE;
+      }
+    }
+
     BLI_assert(e_data.context.select_mode != 0);
   }
 
@@ -186,8 +231,13 @@ static void select_cache_init(void *vedata)
   float(*persmat)[4] = draw_ctx->rv3d->persmat;
   e_data.context.is_dirty = !compare_m4m4(e_data.context.persmat, persmat, FLT_EPSILON);
   if (e_data.context.is_dirty) {
+    /* Remove all tags from drawn or culled objects. */
     copy_m4_m4(e_data.context.persmat, persmat);
-    select_id_context_clear(&e_data.context);
+    e_data.context.objects_drawn_len = 0;
+    e_data.context.index_drawn_len = 1;
+    select_engine_framebuffer_setup();
+    GPU_framebuffer_bind(e_data.framebuffer_select_id);
+    GPU_framebuffer_clear_color_depth(e_data.framebuffer_select_id, (const float[4]){0.0f}, 1.0f);
   }
   e_data.runtime_new_objects = 0;
 }
@@ -204,14 +254,14 @@ static void select_cache_populate(void *vedata, Object *ob)
     /* The object indices have already been drawn. Fill depth pass.
      * Opti: Most of the time this depth pass is not used. */
     struct Mesh *me = ob->data;
-    struct GPUBatch *geom_faces;
     if (e_data.context.select_mode & SCE_SELECT_FACE) {
-      geom_faces = DRW_mesh_batch_cache_get_triangles_with_select_id(me);
+      struct GPUBatch *geom_faces = DRW_mesh_batch_cache_get_triangles_with_select_id(me);
+      DRW_shgroup_call_obmat(stl->g_data->shgrp_depth_only, geom_faces, ob->obmat);
     }
-    else {
-      geom_faces = DRW_mesh_batch_cache_get_surface(me);
+    else if (ob->dt >= OB_SOLID) {
+      struct GPUBatch *geom_faces = DRW_mesh_batch_cache_get_surface(me);
+      DRW_shgroup_call_obmat(stl->g_data->shgrp_depth_only, geom_faces, ob->obmat);
     }
-    DRW_shgroup_call_obmat(stl->g_data->shgrp_depth_only, geom_faces, ob->obmat);
 
     if (e_data.context.select_mode & SCE_SELECT_EDGE) {
       struct GPUBatch *geom_edges = DRW_mesh_batch_cache_get_edges_with_select_id(me);
@@ -283,7 +333,7 @@ static void select_draw_scene(void *vedata)
   }
 
   /* Setup framebuffer */
-  GPU_framebuffer_bind(e_data.context.framebuffer_select_id);
+  GPU_framebuffer_bind(e_data.framebuffer_select_id);
 
   DRW_draw_pass(psl->select_id_face_pass);
 
@@ -306,8 +356,8 @@ static void select_engine_free(void)
     DRW_SHADER_FREE_SAFE(sh_data->select_id_uniform);
   }
 
-  DRW_TEXTURE_FREE_SAFE(e_data.context.texture_u32);
-  GPU_FRAMEBUFFER_FREE_SAFE(e_data.context.framebuffer_select_id);
+  DRW_TEXTURE_FREE_SAFE(e_data.texture_u32);
+  GPU_FRAMEBUFFER_FREE_SAFE(e_data.framebuffer_select_id);
   MEM_SAFE_FREE(e_data.context.objects);
   MEM_SAFE_FREE(e_data.context.index_offsets);
   MEM_SAFE_FREE(e_data.context.objects_drawn);
@@ -368,6 +418,16 @@ struct SELECTID_Context *DRW_select_engine_context_get(void)
   return &e_data.context;
 }
 
+GPUFrameBuffer *DRW_engine_select_framebuffer_get(void)
+{
+  return e_data.framebuffer_select_id;
+}
+
+GPUTexture *DRW_engine_select_texture_get(void)
+{
+  return e_data.texture_u32;
+}
+
 /** \} */
 
 #undef SELECT_ENGINE
diff --git a/source/blender/draw/engines/select/select_engine.h b/source/blender/draw/engines/select/select_engine.h
index 5b900ccaf27..79139d9deaf 100644
--- a/source/blender/draw/engines/select/select_engine.h
+++ b/source/blender/draw/engines/select/select_engine.h
@@ -26,4 +26,9 @@
 extern DrawEngineType draw_engine_select_type;
 extern RenderEngineType DRW_engine_viewport_select_type;
 
-#endif /* __SELECT_ID_ENGINE_H__ */
+struct SELECTID_Context *DRW_select_engine_context_get(void);
+
+struct GPUFrameBuffer *DRW_engine_select_framebuffer_get(void);
+struct GPUTexture *DRW_engine_select_texture_get(void);
+
+#endif /* __SELECT_ENGINE_H__ */
diff --git a/source/blender/draw/engines/select/select_private.h b/source/blender/draw/engines/select/select_private.h
index e48ce4314ae..642cd6ffc56 100644
--- a/source/blender/draw/engines/select/select_private.h
+++ b/source/blender/draw/engines/select/select_private.h
@@ -66,7 +66,6 @@ typedef struct SELECTID_PrivateData {
 } SELECTID_PrivateData; /* Transient data */
 
 /* select_draw_utils.c */
-void select_id_context_clear(struct SELECTID_Context *select_ctx);
 void select_id_object_min_max(struct Object *obj, float r_min[3], float r_max[3]);
 short select_id_get_object_select_mode(Scene *scene, Object *ob);
 void select_id_draw_object(void *vedata,
diff --git a/source/blender/draw/engines/select/shaders/selection_id_frag.glsl b/source/blender/draw/engines/select/shaders/selection_id_frag.glsl
index ea86ddc7301..a84bbbb2cac 100644
--- a/source/blender/draw/engines/select/shaders/selection_id_frag.glsl
+++ b/source/blender/draw/engines/select/shaders/selection_id_frag.glsl
@@ -1,14 +1,15 @@
 
 #ifdef UNIFORM_ID
 uniform int id;
-#  define id floatBitsToUint(intBitsToFloat(id))
+#  define _id floatBitsToUint(intBitsToFloat(id))
 #else
 flat in uint id;
+#  define _id id
 #endif
 
 out uint fragColor;
 
 void main()
 {
-  fragColor = id;
+  fragColor = _id;
 }
diff --git a/source/blender/draw/engines/workbench/shaders/workbench_deferred_composite_frag.glsl b/source/blender/draw/engines/workbench/shaders/workbench_deferred_composite_frag.glsl
index a1f80440404..fd4cea4279a 100644
--- a/source/blender/draw/engines/workbench/shaders/workbench_deferred_composite_frag.glsl
+++ b/source/blender/draw/engines/workbench/shaders/workbench_deferred_composite_frag.glsl
@@ -5,7 +5,8 @@ uniform sampler2D materialBuffer;
 uniform sampler2D normalBuffer;
 /* normalBuffer contains viewport normals */
 uniform sampler2D cavityBuffer;
-uniform sampler2D matcapImage;
+uniform sampler2D matcapDiffuseImage;
+uniform sampler2D matcapSpecularImage;
 
 uniform vec2 invertedViewportSize;
 uniform vec4 viewvecs[3];
@@ -55,8 +56,15 @@ void main()
   normal_viewport = (metallic > 0.0) ? normal_viewport : -normal_viewport;
   bool flipped = world_data.matcap_orientation != 0;
   vec2 matcap_uv = matcap_uv_compute(I_vs, normal_viewport, flipped);
-  vec3 matcap = textureLod(matcapImage, matcap_uv, 0.0).rgb;
-  vec3 shaded_color = matcap * base_color;
+  vec3 matcap_diffuse = textureLod(matcapDiffuseImage, matcap_uv, 0.0).rgb;
+
+#  ifdef V3D_SHADING_SPECULAR_HIGHLIGHT
+  vec3 matcap_specular = textureLod(matcapSpecularImage, matcap_uv, 0.0).rgb;
+#  else
+  vec3 matcap_specular = vec3(0.0);
+#  endif
+
+  vec3 shaded_color = matcap_diffuse * base_color + matcap_specular;
 
 #elif defined(V3D_LIGHTING_STUDIO)
 
diff --git a/source/blender/draw/engines/workbench/shaders/workbench_forward_transparent_accum_frag.glsl b/source/blender/draw/engines/workbench/shaders/workbench_forward_transparent_accum_frag.glsl
index c78b2182d04..3333dfeff38 100644
--- a/source/blender/draw/engines/workbench/shaders/workbench_forward_transparent_accum_frag.glsl
+++ b/source/blender/draw/engines/workbench/shaders/workbench_forward_transparent_accum_frag.glsl
@@ -27,7 +27,8 @@ in vec2 uv_interp;
 in vec3 vertexColor;
 #endif
 #ifdef V3D_LIGHTING_MATCAP
-uniform sampler2D matcapImage;
+uniform sampler2D matcapDiffuseImage;
+uniform sampler2D matcapSpecularImage;
 #endif
 
 layout(std140) uniform world_block
@@ -68,8 +69,13 @@ void main()
 #elif defined(V3D_LIGHTING_MATCAP)
   bool flipped = world_data.matcap_orientation != 0;
   vec2 matcap_uv = matcap_uv_compute(I_vs, nor, flipped);
-  vec3 matcap = textureLod(matcapImage, matcap_uv, 0.0).rgb;
-  vec3 shaded_color = matcap * diffuse_color.rgb;
+  vec3 matcap_diffuse = textureLod(matcapDiffuseImage, matcap_uv, 0.0).rgb;
+#  ifdef V3D_SHADING_SPECULAR_HIGHLIGHT
+  vec3 matcap_specular = textureLod(matcapSpecularImage, matcap_uv, 0.0).rgb;
+#  else
+  vec3 matcap_specular = vec3(0.0);
+#  endif
+  vec3 shaded_color = matcap_diffuse * diffuse_color.rgb + matcap_specular;
 
 #elif defined(V3D_LIGHTING_STUDIO)
   vec3 shaded_color = get_world_lighting(
diff --git a/source/blender/draw/engines/workbench/workbench_deferred.c b/source/blender/draw/engines/workbench/workbench_deferred.c
index add49462de1..c1c9d8b5a96 100644
--- a/source/blender/draw/engines/workbench/workbench_deferred.c
+++ b/source/blender/draw/engines/workbench/workbench_deferred.c
@@ -694,16 +694,22 @@ static void workbench_composite_uniforms(WORKBENCH_PrivateData *wpd, DRWShadingG
   if (CAVITY_ENABLED(wpd)) {
     DRW_shgroup_uniform_texture_ref(grp, "cavityBuffer", &e_data.cavity_buffer_tx);
   }
-  if (SPECULAR_HIGHLIGHT_ENABLED(wpd) || STUDIOLIGHT_TYPE_MATCAP_ENABLED(wpd)) {
+  if (workbench_is_specular_highlight_enabled(wpd) || STUDIOLIGHT_TYPE_MATCAP_ENABLED(wpd)) {
     DRW_shgroup_uniform_vec4(grp, "viewvecs[0]", (float *)wpd->viewvecs, 3);
   }
-  if (SPECULAR_HIGHLIGHT_ENABLED(wpd) || STUDIOLIGHT_TYPE_MATCAP_ENABLED(wpd)) {
+  if (workbench_is_specular_highlight_enabled(wpd) || STUDIOLIGHT_TYPE_MATCAP_ENABLED(wpd)) {
     DRW_shgroup_uniform_vec2(grp, "invertedViewportSize", DRW_viewport_invert_size_get(), 1);
   }
   if (STUDIOLIGHT_TYPE_MATCAP_ENABLED(wpd)) {
-    BKE_studiolight_ensure_flag(wpd->studio_light, STUDIOLIGHT_EQUIRECT_RADIANCE_GPUTEXTURE);
+    BKE_studiolight_ensure_flag(wpd->studio_light,
+                                STUDIOLIGHT_MATCAP_DIFFUSE_GPUTEXTURE |
+                                    STUDIOLIGHT_MATCAP_SPECULAR_GPUTEXTURE);
     DRW_shgroup_uniform_texture(
-        grp, "matcapImage", wpd->studio_light->equirect_radiance_gputexture);
+        grp, "matcapDiffuseImage", wpd->studio_light->matcap_diffuse.gputexture);
+    if (workbench_is_specular_highlight_enabled(wpd)) {
+      DRW_shgroup_uniform_texture(
+          grp, "matcapSpecularImage", wpd->studio_light->matcap_specular.gputexture);
+    }
   }
 }
 
diff --git a/source/blender/draw/engines/workbench/workbench_forward.c b/source/blender/draw/engines/workbench/workbench_forward.c
index 4c1fce550e8..824c6e7bd76 100644
--- a/source/blender/draw/engines/workbench/workbench_forward.c
+++ b/source/blender/draw/engines/workbench/workbench_forward.c
@@ -186,11 +186,17 @@ WORKBENCH_MaterialData *workbench_forward_get_or_create_material_data(WORKBENCH_
     DRW_shgroup_uniform_vec4(grp, "viewvecs[0]", (float *)wpd->viewvecs, 3);
     workbench_material_copy(material, &material_template);
     if (STUDIOLIGHT_TYPE_MATCAP_ENABLED(wpd)) {
-      BKE_studiolight_ensure_flag(wpd->studio_light, STUDIOLIGHT_EQUIRECT_RADIANCE_GPUTEXTURE);
+      BKE_studiolight_ensure_flag(wpd->studio_light,
+                                  STUDIOLIGHT_MATCAP_DIFFUSE_GPUTEXTURE |
+                                      STUDIOLIGHT_MATCAP_SPECULAR_GPUTEXTURE);
       DRW_shgroup_uniform_texture(
-          grp, "matcapImage", wpd->studio_light->equirect_radiance_gputexture);
+          grp, "matcapDiffuseImage", wpd->studio_light->matcap_diffuse.gputexture);
+      if (workbench_is_specular_highlight_enabled(wpd)) {
+        DRW_shgroup_uniform_texture(
+            grp, "matcapSpecularImage", wpd->studio_light->matcap_specular.gputexture);
+      }
     }
-    if (SPECULAR_HIGHLIGHT_ENABLED(wpd) || MATCAP_ENABLED(wpd)) {
+    if (workbench_is_specular_highlight_enabled(wpd) || MATCAP_ENABLED(wpd)) {
       DRW_shgroup_uniform_vec2(grp, "invertedViewportSize", DRW_viewport_invert_size_get(), 1);
     }
     if (SHADOW_ENABLED(wpd)) {
@@ -528,11 +534,17 @@ static void workbench_forward_cache_populate_particles(WORKBENCH_Data *vedata, O
       float hair_alpha = XRAY_ALPHA(wpd) * 0.33f;
       DRW_shgroup_uniform_float_copy(shgrp, "alpha", hair_alpha);
       if (STUDIOLIGHT_TYPE_MATCAP_ENABLED(wpd)) {
-        BKE_studiolight_ensure_flag(wpd->studio_light, STUDIOLIGHT_EQUIRECT_RADIANCE_GPUTEXTURE);
+        BKE_studiolight_ensure_flag(wpd->studio_light,
+                                    STUDIOLIGHT_MATCAP_DIFFUSE_GPUTEXTURE |
+                                        STUDIOLIGHT_MATCAP_SPECULAR_GPUTEXTURE);
         DRW_shgroup_uniform_texture(
-            shgrp, "matcapImage", wpd->studio_light->equirect_radiance_gputexture);
+            shgrp, "matcapDiffuseImage", wpd->studio_light->matcap_diffuse.gputexture);
+        if (workbench_is_specular_highlight_enabled(wpd)) {
+          DRW_shgroup_uniform_texture(
+              shgrp, "matcapSpecularImage", wpd->studio_light->matcap_specular.gputexture);
+        }
       }
-      if (SPECULAR_HIGHLIGHT_ENABLED(wpd) || MATCAP_ENABLED(wpd)) {
+      if (workbench_is_specular_highlight_enabled(wpd) || MATCAP_ENABLED(wpd)) {
         DRW_shgroup_uniform_vec2(shgrp, "invertedViewportSize", DRW_viewport_invert_size_get(), 1);
       }
 
diff --git a/source/blender/draw/engines/workbench/workbench_materials.c b/source/blender/draw/engines/workbench/workbench_materials.c
index 6699a1954ba..e050877187e 100644
--- a/source/blender/draw/engines/workbench/workbench_materials.c
+++ b/source/blender/draw/engines/workbench/workbench_materials.c
@@ -79,7 +79,7 @@ void workbench_material_update_data(WORKBENCH_PrivateData *wpd,
     /* V3D_SHADING_MATERIAL_COLOR or V3D_SHADING_TEXTURE_COLOR */
     if (mat) {
       data->alpha *= mat->a;
-      if (SPECULAR_HIGHLIGHT_ENABLED(wpd)) {
+      if (workbench_is_specular_highlight_enabled(wpd)) {
         copy_v3_v3(data->base_color, &mat->r);
         mul_v3_v3fl(data->diffuse_color, &mat->r, 1.0f - mat->metallic);
         mul_v3_v3fl(data->specular_color, &mat->r, mat->metallic);
@@ -121,7 +121,7 @@ char *workbench_material_build_defines(WORKBENCH_PrivateData *wpd,
   if (SSAO_ENABLED(wpd) || CURVATURE_ENABLED(wpd)) {
     BLI_dynstr_append(ds, "#define WB_CAVITY\n");
   }
-  if (SPECULAR_HIGHLIGHT_ENABLED(wpd)) {
+  if (workbench_is_specular_highlight_enabled(wpd)) {
     BLI_dynstr_append(ds, "#define V3D_SHADING_SPECULAR_HIGHLIGHT\n");
   }
   if (STUDIOLIGHT_ENABLED(wpd)) {
@@ -195,11 +195,12 @@ int workbench_material_get_composite_shader_index(WORKBENCH_PrivateData *wpd)
   /* NOTE: change MAX_COMPOSITE_SHADERS accordingly when modifying this function. */
   int index = 0;
   /* 2 bits FLAT/STUDIO/MATCAP + Specular highlight */
-  index = SPECULAR_HIGHLIGHT_ENABLED(wpd) ? 3 : wpd->shading.light;
+  index = wpd->shading.light;
   SET_FLAG_FROM_TEST(index, wpd->shading.flag & V3D_SHADING_SHADOW, 1 << 2);
   SET_FLAG_FROM_TEST(index, wpd->shading.flag & V3D_SHADING_CAVITY, 1 << 3);
   SET_FLAG_FROM_TEST(index, wpd->shading.flag & V3D_SHADING_OBJECT_OUTLINE, 1 << 4);
   SET_FLAG_FROM_TEST(index, workbench_is_matdata_pass_enabled(wpd), 1 << 5);
+  SET_FLAG_FROM_TEST(index, workbench_is_specular_highlight_enabled(wpd), 1 << 6);
   BLI_assert(index < MAX_COMPOSITE_SHADERS);
   return index;
 }
@@ -246,12 +247,13 @@ int workbench_material_get_accum_shader_index(WORKBENCH_PrivateData *wpd,
   /* NOTE: change MAX_ACCUM_SHADERS accordingly when modifying this function. */
   int index = 0;
   /* 2 bits FLAT/STUDIO/MATCAP + Specular highlight */
-  index = SPECULAR_HIGHLIGHT_ENABLED(wpd) ? 3 : wpd->shading.light;
+  index = wpd->shading.light;
   SET_FLAG_FROM_TEST(index, use_textures, 1 << 2);
   SET_FLAG_FROM_TEST(index, use_vertex_colors, 1 << 3);
   SET_FLAG_FROM_TEST(index, is_hair, 1 << 4);
   /* 1 bits SHADOWS (only facing factor) */
   SET_FLAG_FROM_TEST(index, SHADOW_ENABLED(wpd), 1 << 5);
+  SET_FLAG_FROM_TEST(index, workbench_is_specular_highlight_enabled(wpd), 1 << 6);
   BLI_assert(index < MAX_ACCUM_SHADERS);
   return index;
 }
@@ -333,7 +335,7 @@ void workbench_material_shgroup_uniform(WORKBENCH_PrivateData *wpd,
                                1);
     }
 
-    if (SPECULAR_HIGHLIGHT_ENABLED(wpd)) {
+    if (workbench_is_specular_highlight_enabled(wpd)) {
       if (use_metallic) {
         DRW_shgroup_uniform_float(grp, "materialMetallic", &material->metallic, 1);
       }
diff --git a/source/blender/draw/engines/workbench/workbench_private.h b/source/blender/draw/engines/workbench/workbench_private.h
index 255b036eebb..d880d5d58b5 100644
--- a/source/blender/draw/engines/workbench/workbench_private.h
+++ b/source/blender/draw/engines/workbench/workbench_private.h
@@ -36,9 +36,9 @@
 
 #define WORKBENCH_ENGINE "BLENDER_WORKBENCH"
 #define M_GOLDEN_RATION_CONJUGATE 0.618033988749895
-#define MAX_COMPOSITE_SHADERS (1 << 6)
+#define MAX_COMPOSITE_SHADERS (1 << 7)
 #define MAX_PREPASS_SHADERS (1 << 7)
-#define MAX_ACCUM_SHADERS (1 << 6)
+#define MAX_ACCUM_SHADERS (1 << 7)
 #define MAX_CAVITY_SHADERS (1 << 3)
 
 #define TEXTURE_DRAWING_ENABLED(wpd) (wpd->shading.color_type == V3D_SHADING_TEXTURE_COLOR)
@@ -77,9 +77,6 @@
 #define IS_NAVIGATING(wpd) \
   ((DRW_context_state_get()->rv3d) && (DRW_context_state_get()->rv3d->rflag & RV3D_NAVIGATING))
 
-#define SPECULAR_HIGHLIGHT_ENABLED(wpd) \
-  (STUDIOLIGHT_ENABLED(wpd) && (wpd->shading.flag & V3D_SHADING_SPECULAR_HIGHLIGHT) && \
-   (!STUDIOLIGHT_TYPE_MATCAP_ENABLED(wpd)))
 #define OBJECT_OUTLINE_ENABLED(wpd) (wpd->shading.flag & V3D_SHADING_OBJECT_OUTLINE)
 #define OBJECT_ID_PASS_ENABLED(wpd) (OBJECT_OUTLINE_ENABLED(wpd) || CURVATURE_ENABLED(wpd))
 #define NORMAL_VIEWPORT_COMP_PASS_ENABLED(wpd) \
@@ -316,6 +313,16 @@ typedef struct WORKBENCH_ObjectData {
 } WORKBENCH_ObjectData;
 
 /* inline helper functions */
+BLI_INLINE bool workbench_is_specular_highlight_enabled(WORKBENCH_PrivateData *wpd)
+{
+  if ((wpd->shading.flag & V3D_SHADING_SPECULAR_HIGHLIGHT)) {
+    if (STUDIOLIGHT_ENABLED(wpd) || MATCAP_ENABLED(wpd)) {
+      return (wpd->studio_light->flag & STUDIOLIGHT_SPECULAR_HIGHLIGHT_PASS) != 0;
+    }
+  }
+  return false;
+}
+
 BLI_INLINE bool workbench_is_taa_enabled(WORKBENCH_PrivateData *wpd)
 {
   if (DRW_state_is_image_render()) {
diff --git a/source/blender/draw/intern/DRW_render.h b/source/blender/draw/intern/DRW_render.h
index 8e4ca1cefab..a289152807c 100644
--- a/source/blender/draw/intern/DRW_render.h
+++ b/source/blender/draw/intern/DRW_render.h
@@ -61,7 +61,6 @@
 
 #include "DEG_depsgraph.h"
 
-struct DRWTextStore;
 struct DefaultFramebufferList;
 struct DefaultTextureList;
 struct GPUBatch;
@@ -70,12 +69,9 @@ struct GPUMaterial;
 struct GPUShader;
 struct GPUTexture;
 struct GPUUniformBuffer;
-struct LightEngineData;
 struct Object;
 struct ParticleSystem;
 struct RenderEngineType;
-struct ViewportEngineData;
-struct ViewportEngineData_Info;
 struct bContext;
 struct rcti;
 
diff --git a/source/blender/draw/intern/draw_cache.c b/source/blender/draw/intern/draw_cache.c
index bad4b55eb1a..520932bc429 100644
--- a/source/blender/draw/intern/draw_cache.c
+++ b/source/blender/draw/intern/draw_cache.c
@@ -4001,7 +4001,7 @@ GPUBatch *DRW_cache_cursor_get(bool crosshair_lines)
 /** \} */
 
 /* -------------------------------------------------------------------- */
-/** \name Batch Cache Impl. common
+/** \name Batch Cache Implementation (common)
  * \{ */
 
 void drw_batch_cache_validate(Object *ob)
diff --git a/source/blender/draw/intern/draw_cache.h b/source/blender/draw/intern/draw_cache.h
index 5dadcdc1457..273e97c6e49 100644
--- a/source/blender/draw/intern/draw_cache.h
+++ b/source/blender/draw/intern/draw_cache.h
@@ -27,8 +27,8 @@ struct GPUBatch;
 struct GPUMaterial;
 struct ModifierData;
 struct Object;
-struct ParticleSystem;
 struct PTCacheEdit;
+struct ParticleSystem;
 
 void DRW_shape_cache_free(void);
 void DRW_shape_cache_reset(void);
diff --git a/source/blender/draw/intern/draw_cache_extract_mesh.c b/source/blender/draw/intern/draw_cache_extract_mesh.c
index 95b984747bd..d952965de35 100644
--- a/source/blender/draw/intern/draw_cache_extract_mesh.c
+++ b/source/blender/draw/intern/draw_cache_extract_mesh.c
@@ -454,7 +454,8 @@ static void *extract_tris_init(const MeshRenderData *mr, void *UNUSED(ibo))
     BMFace *efa;
     BM_ITER_MESH (efa, &iter, mr->bm, BM_FACES_OF_MESH) {
       if (!BM_elem_flag_test(efa, BM_ELEM_HIDDEN)) {
-        mat_tri_len[efa->mat_nr] += efa->len - 2;
+        int mat = min_ii(efa->mat_nr, mr->mat_len - 1);
+        mat_tri_len[mat] += efa->len - 2;
       }
     }
   }
@@ -462,7 +463,8 @@ static void *extract_tris_init(const MeshRenderData *mr, void *UNUSED(ibo))
     const MPoly *mpoly = mr->mpoly;
     for (int p = 0; p < mr->poly_len; p++, mpoly++) {
       if (!(mr->use_hide && (mpoly->flag & ME_HIDE))) {
-        mat_tri_len[mpoly->mat_nr] += mpoly->totloop - 2;
+        int mat = min_ii(mpoly->mat_nr, mr->mat_len - 1);
+        mat_tri_len[mat] += mpoly->totloop - 2;
       }
     }
   }
@@ -505,11 +507,12 @@ static void extract_tris_looptri_mesh(const MeshRenderData *mr,
                                       void *_data)
 {
   const MPoly *mpoly = &mr->mpoly[mlt->poly];
-  if (!(mpoly->flag & ME_HIDE)) {
+  if (!(mr->use_hide && (mpoly->flag & ME_HIDE))) {
     MeshExtract_Tri_Data *data = _data;
     int *mat_tri_ofs = data->tri_mat_end;
+    int mat = min_ii(mpoly->mat_nr, mr->mat_len - 1);
     GPU_indexbuf_set_tri_verts(
-        &data->elb, mat_tri_ofs[mpoly->mat_nr]++, mlt->tri[0], mlt->tri[1], mlt->tri[2]);
+        &data->elb, mat_tri_ofs[mat]++, mlt->tri[0], mlt->tri[1], mlt->tri[2]);
   }
 }
 
@@ -534,18 +537,20 @@ static void extract_tris_finish(const MeshRenderData *mr, void *ibo, void *_data
   MEM_freeN(data);
 }
 
-const MeshExtract extract_tris = {extract_tris_init,
-                                  extract_tris_looptri_bmesh,
-                                  extract_tris_looptri_mesh,
-                                  NULL,
-                                  NULL,
-                                  NULL,
-                                  NULL,
-                                  NULL,
-                                  NULL,
-                                  extract_tris_finish,
-                                  0,
-                                  false};
+static const MeshExtract extract_tris = {
+    extract_tris_init,
+    extract_tris_looptri_bmesh,
+    extract_tris_looptri_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_tris_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -645,18 +650,20 @@ static void extract_lines_finish(const MeshRenderData *mr, void *ibo, void *elb)
   }
 }
 
-const MeshExtract extract_lines = {extract_lines_init,
-                                   NULL,
-                                   NULL,
-                                   extract_lines_loop_bmesh,
-                                   extract_lines_loop_mesh,
-                                   extract_lines_ledge_bmesh,
-                                   extract_lines_ledge_mesh,
-                                   NULL,
-                                   NULL,
-                                   extract_lines_finish,
-                                   0,
-                                   false};
+static const MeshExtract extract_lines = {
+    extract_lines_init,
+    NULL,
+    NULL,
+    extract_lines_loop_bmesh,
+    extract_lines_loop_mesh,
+    extract_lines_ledge_bmesh,
+    extract_lines_ledge_mesh,
+    NULL,
+    NULL,
+    extract_lines_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -750,18 +757,20 @@ static void extract_points_finish(const MeshRenderData *UNUSED(mr), void *ibo, v
   MEM_freeN(elb);
 }
 
-const MeshExtract extract_points = {extract_points_init,
-                                    NULL,
-                                    NULL,
-                                    extract_points_loop_bmesh,
-                                    extract_points_loop_mesh,
-                                    extract_points_ledge_bmesh,
-                                    extract_points_ledge_mesh,
-                                    extract_points_lvert_bmesh,
-                                    extract_points_lvert_mesh,
-                                    extract_points_finish,
-                                    0,
-                                    false};
+static const MeshExtract extract_points = {
+    extract_points_init,
+    NULL,
+    NULL,
+    extract_points_loop_bmesh,
+    extract_points_loop_mesh,
+    extract_points_ledge_bmesh,
+    extract_points_ledge_mesh,
+    extract_points_lvert_bmesh,
+    extract_points_lvert_mesh,
+    extract_points_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -813,18 +822,20 @@ static void extract_fdots_finish(const MeshRenderData *UNUSED(mr), void *ibo, vo
   MEM_freeN(elb);
 }
 
-const MeshExtract extract_fdots = {extract_fdots_init,
-                                   NULL,
-                                   NULL,
-                                   extract_fdots_loop_bmesh,
-                                   extract_fdots_loop_mesh,
-                                   NULL,
-                                   NULL,
-                                   NULL,
-                                   NULL,
-                                   extract_fdots_finish,
-                                   0,
-                                   false};
+static const MeshExtract extract_fdots = {
+    extract_fdots_init,
+    NULL,
+    NULL,
+    extract_fdots_loop_bmesh,
+    extract_fdots_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_fdots_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -886,18 +897,20 @@ static void extract_lines_paint_mask_finish(const MeshRenderData *UNUSED(mr),
   MEM_freeN(data);
 }
 
-const MeshExtract extract_lines_paint_mask = {extract_lines_paint_mask_init,
-                                              NULL,
-                                              NULL,
-                                              NULL,
-                                              extract_lines_paint_mask_loop_mesh,
-                                              NULL,
-                                              NULL,
-                                              NULL,
-                                              NULL,
-                                              extract_lines_paint_mask_finish,
-                                              0,
-                                              false};
+static const MeshExtract extract_lines_paint_mask = {
+    extract_lines_paint_mask_init,
+    NULL,
+    NULL,
+    NULL,
+    extract_lines_paint_mask_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_lines_paint_mask_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -919,7 +932,7 @@ static void *extract_lines_adjacency_init(const MeshRenderData *mr, void *UNUSED
 {
   /* Similar to poly_to_tri_count().
    * There is always loop + tri - 1 edges inside a polygon.
-   * Cummulate for all polys and you get : */
+   * Accumulate for all polys and you get : */
   uint tess_edge_len = mr->loop_len + mr->tri_len - mr->poly_len;
 
   size_t vert_to_loop_size = sizeof(uint) * mr->vert_len;
@@ -1038,18 +1051,20 @@ static void extract_lines_adjacency_finish(const MeshRenderData *mr, void *ibo,
 
 #undef NO_EDGE
 
-const MeshExtract extract_lines_adjacency = {extract_lines_adjacency_init,
-                                             extract_lines_adjacency_looptri_bmesh,
-                                             extract_lines_adjacency_looptri_mesh,
-                                             NULL,
-                                             NULL,
-                                             NULL,
-                                             NULL,
-                                             NULL,
-                                             NULL,
-                                             extract_lines_adjacency_finish,
-                                             0,
-                                             false};
+static const MeshExtract extract_lines_adjacency = {
+    extract_lines_adjacency_init,
+    extract_lines_adjacency_looptri_bmesh,
+    extract_lines_adjacency_looptri_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_lines_adjacency_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -1112,18 +1127,20 @@ static void extract_edituv_tris_finish(const MeshRenderData *UNUSED(mr), void *i
   MEM_freeN(extract_data);
 }
 
-const MeshExtract extract_edituv_tris = {extract_edituv_tris_init,
-                                         extract_edituv_tris_looptri_bmesh,
-                                         extract_edituv_tris_looptri_mesh,
-                                         NULL,
-                                         NULL,
-                                         NULL,
-                                         NULL,
-                                         NULL,
-                                         NULL,
-                                         extract_edituv_tris_finish,
-                                         0,
-                                         false};
+static const MeshExtract extract_edituv_tris = {
+    extract_edituv_tris_init,
+    extract_edituv_tris_looptri_bmesh,
+    extract_edituv_tris_looptri_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_edituv_tris_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -1185,18 +1202,20 @@ static void extract_edituv_lines_finish(const MeshRenderData *UNUSED(mr), void *
   MEM_freeN(extract_data);
 }
 
-const MeshExtract extract_edituv_lines = {extract_edituv_lines_init,
-                                          NULL,
-                                          NULL,
-                                          extract_edituv_lines_loop_bmesh,
-                                          extract_edituv_lines_loop_mesh,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          extract_edituv_lines_finish,
-                                          0,
-                                          false};
+static const MeshExtract extract_edituv_lines = {
+    extract_edituv_lines_init,
+    NULL,
+    NULL,
+    extract_edituv_lines_loop_bmesh,
+    extract_edituv_lines_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_edituv_lines_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -1254,18 +1273,20 @@ static void extract_edituv_points_finish(const MeshRenderData *UNUSED(mr), void
   MEM_freeN(extract_data);
 }
 
-const MeshExtract extract_edituv_points = {extract_edituv_points_init,
-                                           NULL,
-                                           NULL,
-                                           extract_edituv_points_loop_bmesh,
-                                           extract_edituv_points_loop_mesh,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           extract_edituv_points_finish,
-                                           0,
-                                           false};
+static const MeshExtract extract_edituv_points = {
+    extract_edituv_points_init,
+    NULL,
+    NULL,
+    extract_edituv_points_loop_bmesh,
+    extract_edituv_points_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_edituv_points_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -1330,18 +1351,20 @@ static void extract_edituv_fdots_finish(const MeshRenderData *UNUSED(mr), void *
   MEM_freeN(data);
 }
 
-const MeshExtract extract_edituv_fdots = {extract_edituv_fdots_init,
-                                          NULL,
-                                          NULL,
-                                          extract_edituv_fdots_loop_bmesh,
-                                          extract_edituv_fdots_loop_mesh,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          extract_edituv_fdots_finish,
-                                          0,
-                                          false};
+static const MeshExtract extract_edituv_fdots = {
+    extract_edituv_fdots_init,
+    NULL,
+    NULL,
+    extract_edituv_fdots_loop_bmesh,
+    extract_edituv_fdots_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_edituv_fdots_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -1410,7 +1433,7 @@ static void extract_pos_nor_loop_mesh(const MeshRenderData *mr,
                                       int l,
                                       const MLoop *mloop,
                                       int UNUSED(p),
-                                      const MPoly *mpoly,
+                                      const MPoly *UNUSED(mpoly),
                                       void *_data)
 {
   MeshExtract_PosNor_Data *data = _data;
@@ -1419,12 +1442,15 @@ static void extract_pos_nor_loop_mesh(const MeshRenderData *mr,
   copy_v3_v3(vert->pos, mvert->co);
   vert->nor = data->packed_nor[mloop->v];
   /* Flag for paint mode overlay. */
-  if (mpoly->flag & ME_HIDE)
+  if (mvert->flag & ME_HIDE) {
     vert->nor.w = -1;
-  else if (mpoly->flag & ME_FACE_SEL)
+  }
+  else if (mvert->flag & SELECT) {
     vert->nor.w = 1;
-  else
+  }
+  else {
     vert->nor.w = 0;
+  }
 }
 
 static void extract_pos_nor_ledge_bmesh(const MeshRenderData *mr, int e, BMEdge *eed, void *_data)
@@ -1479,18 +1505,20 @@ static void extract_pos_nor_finish(const MeshRenderData *UNUSED(mr), void *UNUSE
   MEM_freeN(data);
 }
 
-const MeshExtract extract_pos_nor = {extract_pos_nor_init,
-                                     NULL,
-                                     NULL,
-                                     extract_pos_nor_loop_bmesh,
-                                     extract_pos_nor_loop_mesh,
-                                     extract_pos_nor_ledge_bmesh,
-                                     extract_pos_nor_ledge_mesh,
-                                     extract_pos_nor_lvert_bmesh,
-                                     extract_pos_nor_lvert_mesh,
-                                     extract_pos_nor_finish,
-                                     0,
-                                     true};
+static const MeshExtract extract_pos_nor = {
+    extract_pos_nor_init,
+    NULL,
+    NULL,
+    extract_pos_nor_loop_bmesh,
+    extract_pos_nor_loop_mesh,
+    extract_pos_nor_ledge_bmesh,
+    extract_pos_nor_ledge_mesh,
+    extract_pos_nor_lvert_bmesh,
+    extract_pos_nor_lvert_mesh,
+    extract_pos_nor_finish,
+    0,
+    true,
+};
 /** \} */
 
 /* ---------------------------------------------------------------------- */
@@ -1501,7 +1529,7 @@ static void *extract_lnor_init(const MeshRenderData *mr, void *buf)
 {
   static GPUVertFormat format = {0};
   if (format.attr_len == 0) {
-    GPU_vertformat_attr_add(&format, "nor", GPU_COMP_I10, 3, GPU_FETCH_INT_TO_FLOAT_UNIT);
+    GPU_vertformat_attr_add(&format, "nor", GPU_COMP_I10, 4, GPU_FETCH_INT_TO_FLOAT_UNIT);
     GPU_vertformat_alias_add(&format, "lnor");
   }
   GPUVertBuf *vbo = buf;
@@ -1536,20 +1564,32 @@ static void extract_lnor_loop_mesh(
   else {
     ((GPUPackedNormal *)data)[l] = GPU_normal_convert_i10_v3(mr->poly_normals[p]);
   }
-}
-
-const MeshExtract extract_lnor = {extract_lnor_init,
-                                  NULL,
-                                  NULL,
-                                  extract_lnor_loop_bmesh,
-                                  extract_lnor_loop_mesh,
-                                  NULL,
-                                  NULL,
-                                  NULL,
-                                  NULL,
-                                  NULL,
-                                  MR_DATA_LOOP_NOR,
-                                  true};
+  /* Flag for paint mode overlay. */
+  if (mpoly->flag & ME_HIDE) {
+    ((GPUPackedNormal *)data)[l].w = -1;
+  }
+  else if (mpoly->flag & ME_FACE_SEL) {
+    ((GPUPackedNormal *)data)[l].w = 1;
+  }
+  else {
+    ((GPUPackedNormal *)data)[l].w = 0;
+  }
+}
+
+static const MeshExtract extract_lnor = {
+    extract_lnor_init,
+    NULL,
+    NULL,
+    extract_lnor_loop_bmesh,
+    extract_lnor_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    MR_DATA_LOOP_NOR,
+    true,
+};
 
 /** \} */
 
@@ -1633,8 +1673,21 @@ static void *extract_uv_init(const MeshRenderData *mr, void *buf)
   return NULL;
 }
 
-const MeshExtract extract_uv = {
-    extract_uv_init, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, 0, false};
+static const MeshExtract extract_uv = {
+    extract_uv_init,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    0,
+    false,
+};
+
 /** \} */
 
 /* ---------------------------------------------------------------------- */
@@ -1775,18 +1828,20 @@ static void *extract_tan_init(const MeshRenderData *mr, void *buf)
   return NULL;
 }
 
-const MeshExtract extract_tan = {extract_tan_init,
-                                 NULL,
-                                 NULL,
-                                 NULL,
-                                 NULL,
-                                 NULL,
-                                 NULL,
-                                 NULL,
-                                 NULL,
-                                 NULL,
-                                 MR_DATA_POLY_NOR | MR_DATA_TAN_LOOP_NOR | MR_DATA_LOOPTRI,
-                                 false};
+static const MeshExtract extract_tan = {
+    extract_tan_init,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    MR_DATA_POLY_NOR | MR_DATA_TAN_LOOP_NOR | MR_DATA_LOOPTRI,
+    false,
+};
 
 /** \} */
 
@@ -1840,10 +1895,21 @@ static void *extract_vcol_init(const MeshRenderData *mr, void *buf)
   return NULL;
 }
 
-const MeshExtract extract_vcol = {
-    extract_vcol_init, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, 0, false};
+static const MeshExtract extract_vcol = {
+    extract_vcol_init,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    0,
+    false,
+};
 
-/** \} */ /** \} */
+/** \} */
 
 /* ---------------------------------------------------------------------- */
 /** \name Extract Orco
@@ -1859,8 +1925,8 @@ static void *extract_orco_init(const MeshRenderData *mr, void *buf)
   static GPUVertFormat format = {0};
   if (format.attr_len == 0) {
     /* FIXME(fclem): We use the last component as a way to differentiate from generic vertex
-     * attribs. This is a substential waste of Vram and should be done another way.
-     * Unfortunately, at the time of writting, I did not found any other "non disruptive"
+     * attribs. This is a substantial waste of Vram and should be done another way.
+     * Unfortunately, at the time of writing, I did not found any other "non disruptive"
      * alternative. */
     GPU_vertformat_attr_add(&format, "orco", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
   }
@@ -1908,18 +1974,20 @@ static void extract_orco_finish(const MeshRenderData *UNUSED(mr), void *UNUSED(b
   MEM_freeN(data);
 }
 
-const MeshExtract extract_orco = {extract_orco_init,
-                                  NULL,
-                                  NULL,
-                                  extract_orco_loop_bmesh,
-                                  extract_orco_loop_mesh,
-                                  NULL,
-                                  NULL,
-                                  NULL,
-                                  NULL,
-                                  extract_orco_finish,
-                                  0,
-                                  true};
+static const MeshExtract extract_orco = {
+    extract_orco_init,
+    NULL,
+    NULL,
+    extract_orco_loop_bmesh,
+    extract_orco_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_orco_finish,
+    0,
+    true,
+};
 
 /** \} */
 
@@ -2085,18 +2153,20 @@ static void extract_edge_fac_finish(const MeshRenderData *mr, void *buf, void *_
   MEM_freeN(data);
 }
 
-const MeshExtract extract_edge_fac = {extract_edge_fac_init,
-                                      NULL,
-                                      NULL,
-                                      extract_edge_fac_loop_bmesh,
-                                      extract_edge_fac_loop_mesh,
-                                      extract_edge_fac_ledge_bmesh,
-                                      extract_edge_fac_ledge_mesh,
-                                      NULL,
-                                      NULL,
-                                      extract_edge_fac_finish,
-                                      MR_DATA_POLY_NOR,
-                                      false};
+static const MeshExtract extract_edge_fac = {
+    extract_edge_fac_init,
+    NULL,
+    NULL,
+    extract_edge_fac_loop_bmesh,
+    extract_edge_fac_loop_mesh,
+    extract_edge_fac_ledge_bmesh,
+    extract_edge_fac_ledge_mesh,
+    NULL,
+    NULL,
+    extract_edge_fac_finish,
+    MR_DATA_POLY_NOR,
+    false,
+};
 
 /** \} */
 /* ---------------------------------------------------------------------- */
@@ -2213,18 +2283,20 @@ static void extract_weights_finish(const MeshRenderData *UNUSED(mr), void *UNUSE
   MEM_freeN(data);
 }
 
-const MeshExtract extract_weights = {extract_weights_init,
-                                     NULL,
-                                     NULL,
-                                     extract_weights_loop_bmesh,
-                                     extract_weights_loop_mesh,
-                                     NULL,
-                                     NULL,
-                                     NULL,
-                                     NULL,
-                                     extract_weights_finish,
-                                     0,
-                                     true};
+static const MeshExtract extract_weights = {
+    extract_weights_init,
+    NULL,
+    NULL,
+    extract_weights_loop_bmesh,
+    extract_weights_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_weights_finish,
+    0,
+    true,
+};
 
 /** \} */
 
@@ -2479,18 +2551,20 @@ static void extract_edit_data_lvert_mesh(const MeshRenderData *mr,
   }
 }
 
-const MeshExtract extract_edit_data = {extract_edit_data_init,
-                                       NULL,
-                                       NULL,
-                                       extract_edit_data_loop_bmesh,
-                                       extract_edit_data_loop_mesh,
-                                       extract_edit_data_ledge_bmesh,
-                                       extract_edit_data_ledge_mesh,
-                                       extract_edit_data_lvert_bmesh,
-                                       extract_edit_data_lvert_mesh,
-                                       NULL,
-                                       0,
-                                       true};
+static const MeshExtract extract_edit_data = {
+    extract_edit_data_init,
+    NULL,
+    NULL,
+    extract_edit_data_loop_bmesh,
+    extract_edit_data_loop_mesh,
+    extract_edit_data_ledge_bmesh,
+    extract_edit_data_ledge_mesh,
+    extract_edit_data_lvert_bmesh,
+    extract_edit_data_lvert_mesh,
+    NULL,
+    0,
+    true,
+};
 
 /** \} */
 
@@ -2516,7 +2590,7 @@ static void *extract_edituv_data_init(const MeshRenderData *mr, void *buf)
   GPU_vertbuf_init_with_format(vbo, &format);
   GPU_vertbuf_data_alloc(vbo, mr->loop_len);
 
-  CustomData *cd_ldata = &mr->me->ldata;
+  CustomData *cd_ldata = (mr->extract_type == MR_EXTRACT_BMESH) ? &mr->bm->ldata : &mr->me->ldata;
 
   MeshExtract_EditUVData_Data *data = MEM_callocN(sizeof(*data), __func__);
   data->vbo_data = (EditLoopData *)vbo->data;
@@ -2577,18 +2651,20 @@ static void extract_edituv_data_finish(const MeshRenderData *UNUSED(mr),
   MEM_freeN(data);
 }
 
-const MeshExtract extract_edituv_data = {extract_edituv_data_init,
-                                         NULL,
-                                         NULL,
-                                         extract_edituv_data_loop_bmesh,
-                                         extract_edituv_data_loop_mesh,
-                                         NULL,
-                                         NULL,
-                                         NULL,
-                                         NULL,
-                                         extract_edituv_data_finish,
-                                         0,
-                                         true};
+static const MeshExtract extract_edituv_data = {
+    extract_edituv_data_init,
+    NULL,
+    NULL,
+    extract_edituv_data_loop_bmesh,
+    extract_edituv_data_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_edituv_data_finish,
+    0,
+    true,
+};
 
 /** \} */
 
@@ -2708,18 +2784,20 @@ static void mesh_stretch_area_finish(const MeshRenderData *mr, void *buf, void *
   MEM_freeN(area_ratio);
 }
 
-const MeshExtract extract_stretch_area = {extract_stretch_area_init,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          NULL,
-                                          mesh_stretch_area_finish,
-                                          0,
-                                          false};
+static const MeshExtract extract_stretch_area = {
+    extract_stretch_area_init,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    mesh_stretch_area_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -2744,8 +2822,8 @@ static void compute_normalize_edge_vectors(float auv[2][2],
                                            float av[2][3],
                                            const float uv[2],
                                            const float uv_prev[2],
-                                           const float co[2],
-                                           const float co_prev[2])
+                                           const float co[3],
+                                           const float co_prev[3])
 {
   /* Move previous edge. */
   copy_v2_v2(auv[0], auv[1]);
@@ -2896,18 +2974,20 @@ static void extract_stretch_angle_finish(const MeshRenderData *UNUSED(mr),
   MEM_freeN(data);
 }
 
-const MeshExtract extract_stretch_angle = {extract_stretch_angle_init,
-                                           NULL,
-                                           NULL,
-                                           extract_stretch_angle_loop_bmesh,
-                                           extract_stretch_angle_loop_mesh,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           extract_stretch_angle_finish,
-                                           0,
-                                           true};
+static const MeshExtract extract_stretch_angle = {
+    extract_stretch_angle_init,
+    NULL,
+    NULL,
+    extract_stretch_angle_loop_bmesh,
+    extract_stretch_angle_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_stretch_angle_finish,
+    0,
+    false,
+};
 
 /** \} */
 
@@ -3479,20 +3559,22 @@ static void extract_mesh_analysis_finish(const MeshRenderData *mr, void *buf, vo
   }
 }
 
-const MeshExtract extract_mesh_analysis = {extract_mesh_analysis_init,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           NULL,
-                                           extract_mesh_analysis_finish,
-                                           /* This is not needed for all vis type.
-                                            * Maybe split into different extract. */
-                                           MR_DATA_POLY_NOR | MR_DATA_LOOPTRI,
-                                           false};
+static const MeshExtract extract_mesh_analysis = {
+    extract_mesh_analysis_init,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_mesh_analysis_finish,
+    /* This is not needed for all vis type.
+     * Maybe split into different extract. */
+    MR_DATA_POLY_NOR | MR_DATA_LOOPTRI,
+    false,
+};
 
 /** \} */
 
@@ -3546,18 +3628,20 @@ static void extract_fdots_pos_loop_mesh(const MeshRenderData *mr,
   }
 }
 
-const MeshExtract extract_fdots_pos = {extract_fdots_pos_init,
-                                       NULL,
-                                       NULL,
-                                       extract_fdots_pos_loop_bmesh,
-                                       extract_fdots_pos_loop_mesh,
-                                       NULL,
-                                       NULL,
-                                       NULL,
-                                       NULL,
-                                       NULL,
-                                       0,
-                                       true};
+static const MeshExtract extract_fdots_pos = {
+    extract_fdots_pos_init,
+    NULL,
+    NULL,
+    extract_fdots_pos_loop_bmesh,
+    extract_fdots_pos_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    0,
+    true,
+};
 
 /** \} */
 
@@ -3604,18 +3688,20 @@ static void extract_fdots_nor_finish(const MeshRenderData *mr, void *buf, void *
   }
 }
 
-const MeshExtract extract_fdots_nor = {extract_fdots_nor_init,
-                                       NULL,
-                                       NULL,
-                                       NULL,
-                                       NULL,
-                                       NULL,
-                                       NULL,
-                                       NULL,
-                                       NULL,
-                                       extract_fdots_nor_finish,
-                                       MR_DATA_POLY_NOR,
-                                       false};
+static const MeshExtract extract_fdots_nor = {
+    extract_fdots_nor_init,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_fdots_nor_finish,
+    MR_DATA_POLY_NOR,
+    false,
+};
 
 /** \} */
 
@@ -3686,18 +3772,20 @@ static void extract_fdots_uv_finish(const MeshRenderData *UNUSED(mr),
   MEM_freeN(data);
 }
 
-const MeshExtract extract_fdots_uv = {extract_fdots_uv_init,
-                                      NULL,
-                                      NULL,
-                                      extract_fdots_uv_loop_bmesh,
-                                      extract_fdots_uv_loop_mesh,
-                                      NULL,
-                                      NULL,
-                                      NULL,
-                                      NULL,
-                                      extract_fdots_uv_finish,
-                                      0,
-                                      true};
+static const MeshExtract extract_fdots_uv = {
+    extract_fdots_uv_init,
+    NULL,
+    NULL,
+    extract_fdots_uv_loop_bmesh,
+    extract_fdots_uv_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_fdots_uv_finish,
+    0,
+    true,
+};
 /** \} */
 
 /* ---------------------------------------------------------------------- */
@@ -3759,18 +3847,20 @@ static void extract_fdots_edituv_data_finish(const MeshRenderData *UNUSED(mr),
   MEM_freeN(data);
 }
 
-const MeshExtract extract_fdots_edituv_data = {extract_fdots_edituv_data_init,
-                                               NULL,
-                                               NULL,
-                                               extract_fdots_edituv_data_loop_bmesh,
-                                               extract_fdots_edituv_data_loop_mesh,
-                                               NULL,
-                                               NULL,
-                                               NULL,
-                                               NULL,
-                                               extract_fdots_edituv_data_finish,
-                                               0,
-                                               true};
+static const MeshExtract extract_fdots_edituv_data = {
+    extract_fdots_edituv_data_init,
+    NULL,
+    NULL,
+    extract_fdots_edituv_data_loop_bmesh,
+    extract_fdots_edituv_data_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    extract_fdots_edituv_data_finish,
+    0,
+    true,
+};
 /** \} */
 
 /* ---------------------------------------------------------------------- */
@@ -3790,7 +3880,7 @@ static void *extract_select_idx_init(const MeshRenderData *mr, void *buf)
   return vbo->data;
 }
 
-/* TODO Use glVertexID to get loop index and use the data structure on the CPU to retreive the
+/* TODO Use glVertexID to get loop index and use the data structure on the CPU to retrieve the
  * select element associated with this loop ID. This would remove the need for this separate index
  * VBOs. We could upload the p/e/v_origindex as a buffer texture and sample it inside the shader to
  * output original index. */
@@ -3898,44 +3988,50 @@ static void extract_vert_idx_lvert_mesh(const MeshRenderData *mr,
   ((uint32_t *)data)[mr->loop_len + mr->edge_loose_len * 2 + v] = v_orig;
 }
 
-const MeshExtract extract_poly_idx = {extract_select_idx_init,
-                                      NULL,
-                                      NULL,
-                                      extract_poly_idx_loop_bmesh,
-                                      extract_poly_idx_loop_mesh,
-                                      NULL,
-                                      NULL,
-                                      NULL,
-                                      NULL,
-                                      NULL,
-                                      0,
-                                      true};
-
-const MeshExtract extract_edge_idx = {extract_select_idx_init,
-                                      NULL,
-                                      NULL,
-                                      extract_edge_idx_loop_bmesh,
-                                      extract_edge_idx_loop_mesh,
-                                      extract_edge_idx_ledge_bmesh,
-                                      extract_edge_idx_ledge_mesh,
-                                      NULL,
-                                      NULL,
-                                      NULL,
-                                      0,
-                                      true};
-
-const MeshExtract extract_vert_idx = {extract_select_idx_init,
-                                      NULL,
-                                      NULL,
-                                      extract_vert_idx_loop_bmesh,
-                                      extract_vert_idx_loop_mesh,
-                                      extract_vert_idx_ledge_bmesh,
-                                      extract_vert_idx_ledge_mesh,
-                                      extract_vert_idx_lvert_bmesh,
-                                      extract_vert_idx_lvert_mesh,
-                                      NULL,
-                                      0,
-                                      true};
+static const MeshExtract extract_poly_idx = {
+    extract_select_idx_init,
+    NULL,
+    NULL,
+    extract_poly_idx_loop_bmesh,
+    extract_poly_idx_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    0,
+    true,
+};
+
+static const MeshExtract extract_edge_idx = {
+    extract_select_idx_init,
+    NULL,
+    NULL,
+    extract_edge_idx_loop_bmesh,
+    extract_edge_idx_loop_mesh,
+    extract_edge_idx_ledge_bmesh,
+    extract_edge_idx_ledge_mesh,
+    NULL,
+    NULL,
+    NULL,
+    0,
+    true,
+};
+
+static const MeshExtract extract_vert_idx = {
+    extract_select_idx_init,
+    NULL,
+    NULL,
+    extract_vert_idx_loop_bmesh,
+    extract_vert_idx_loop_mesh,
+    extract_vert_idx_ledge_bmesh,
+    extract_vert_idx_ledge_mesh,
+    extract_vert_idx_lvert_bmesh,
+    extract_vert_idx_lvert_mesh,
+    NULL,
+    0,
+    true,
+};
 
 static void *extract_select_fdot_idx_init(const MeshRenderData *mr, void *buf)
 {
@@ -3968,18 +4064,20 @@ static void extract_fdot_idx_loop_mesh(const MeshRenderData *mr,
   ((uint32_t *)data)[p] = (mr->p_origindex) ? mr->p_origindex[p] : p;
 }
 
-const MeshExtract extract_fdot_idx = {extract_select_fdot_idx_init,
-                                      NULL,
-                                      NULL,
-                                      extract_fdot_idx_loop_bmesh,
-                                      extract_fdot_idx_loop_mesh,
-                                      NULL,
-                                      NULL,
-                                      NULL,
-                                      NULL,
-                                      NULL,
-                                      0,
-                                      true};
+static const MeshExtract extract_fdot_idx = {
+    extract_select_fdot_idx_init,
+    NULL,
+    NULL,
+    extract_fdot_idx_loop_bmesh,
+    extract_fdot_idx_loop_mesh,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    NULL,
+    0,
+    true,
+};
 
 /** \} */
 
diff --git a/source/blender/draw/intern/draw_cache_impl.h b/source/blender/draw/intern/draw_cache_impl.h
index d392db63938..3e33346c7d8 100644
--- a/source/blender/draw/intern/draw_cache_impl.h
+++ b/source/blender/draw/intern/draw_cache_impl.h
@@ -23,18 +23,14 @@
 #ifndef __DRAW_CACHE_IMPL_H__
 #define __DRAW_CACHE_IMPL_H__
 
-struct CurveCache;
 struct GPUBatch;
 struct GPUIndexBuf;
 struct GPUMaterial;
-struct GPUTexture;
 struct GPUVertBuf;
 struct ListBase;
 struct ModifierData;
 struct PTCacheEdit;
 struct ParticleSystem;
-struct SpaceImage;
-struct ToolSettings;
 
 struct Curve;
 struct Lattice;
diff --git a/source/blender/draw/intern/draw_cache_impl_displist.c b/source/blender/draw/intern/draw_cache_impl_displist.c
index 90bbf996851..eef108b1f2f 100644
--- a/source/blender/draw/intern/draw_cache_impl_displist.c
+++ b/source/blender/draw/intern/draw_cache_impl_displist.c
@@ -66,7 +66,7 @@ static int dl_tri_len(const DispList *dl)
   return 0;
 }
 
-/* see: displist_get_allverts */
+/* see: displist_vert_coords_alloc */
 static int curve_render_surface_vert_len_get(const ListBase *lb)
 {
   int vert_len = 0;
diff --git a/source/blender/draw/intern/draw_cache_impl_mesh.c b/source/blender/draw/intern/draw_cache_impl_mesh.c
index 12c6a715685..456e21db6ed 100644
--- a/source/blender/draw/intern/draw_cache_impl_mesh.c
+++ b/source/blender/draw/intern/draw_cache_impl_mesh.c
@@ -549,6 +549,7 @@ void DRW_mesh_batch_cache_dirty_tag(Mesh *me, int mode)
       {
         GPU_INDEXBUF_DISCARD_SAFE(mbufcache->ibo.lines_paint_mask);
         GPU_VERTBUF_DISCARD_SAFE(mbufcache->vbo.pos_nor);
+        GPU_VERTBUF_DISCARD_SAFE(mbufcache->vbo.lnor);
       }
       GPU_BATCH_DISCARD_SAFE(cache->batch.surface);
       GPU_BATCH_DISCARD_SAFE(cache->batch.wire_loops);
@@ -969,6 +970,7 @@ void DRW_mesh_batch_cache_free_old(Mesh *me, int ctime)
 void DRW_mesh_batch_cache_create_requested(
     Object *ob, Mesh *me, const Scene *scene, const bool is_paint_mode, const bool use_hide)
 {
+  GPUIndexBuf **saved_elem_ranges = NULL;
   const ToolSettings *ts = NULL;
   if (scene) {
     ts = scene->toolsettings;
@@ -1031,6 +1033,17 @@ void DRW_mesh_batch_cache_create_requested(
           GPU_VERTBUF_DISCARD_SAFE(mbuffercache->vbo.vcol);
         }
       }
+      /* XXX save element buffer to avoid recreating them.
+       * This is only if the cd_needed changes so it is ok to keep them.*/
+      if (cache->surface_per_mat[0] && cache->surface_per_mat[0]->elem) {
+        saved_elem_ranges = MEM_callocN(sizeof(saved_elem_ranges) * cache->mat_len, __func__);
+        for (int i = 0; i < cache->mat_len; ++i) {
+          saved_elem_ranges[i] = cache->surface_per_mat[i]->elem;
+          /* Avoid deletion as the batch is owner. */
+          cache->surface_per_mat[i]->elem = NULL;
+          cache->surface_per_mat[i]->owns_flag &= ~GPU_BATCH_OWNS_INDEX;
+        }
+      }
       /* We can't discard batches at this point as they have been
        * referenced for drawing. Just clear them in place. */
       for (int i = 0; i < cache->mat_len; ++i) {
@@ -1128,6 +1141,8 @@ void DRW_mesh_batch_cache_create_requested(
   }
   if (DRW_batch_requested(cache->batch.wire_loops, GPU_PRIM_LINES)) {
     DRW_ibo_request(cache->batch.wire_loops, &mbufcache->ibo.lines_paint_mask);
+    /* Order matters. First ones override latest vbos' attribs. */
+    DRW_vbo_request(cache->batch.wire_loops, &mbufcache->vbo.lnor);
     DRW_vbo_request(cache->batch.wire_loops, &mbufcache->vbo.pos_nor);
   }
   if (DRW_batch_requested(cache->batch.wire_edges, GPU_PRIM_LINES)) {
@@ -1151,7 +1166,13 @@ void DRW_mesh_batch_cache_create_requested(
   /* Per Material */
   for (int i = 0; i < cache->mat_len; ++i) {
     if (DRW_batch_requested(cache->surface_per_mat[i], GPU_PRIM_TRIS)) {
-      DRW_ibo_request(cache->surface_per_mat[i], &mbufcache->ibo.tris);
+      if (saved_elem_ranges && saved_elem_ranges[i]) {
+        /* XXX assign old element buffer range (it did not change).*/
+        GPU_batch_elembuf_set(cache->surface_per_mat[i], saved_elem_ranges[i], true);
+      }
+      else {
+        DRW_ibo_request(cache->surface_per_mat[i], &mbufcache->ibo.tris);
+      }
       /* Order matters. First ones override latest vbos' attribs. */
       DRW_vbo_request(cache->surface_per_mat[i], &mbufcache->vbo.lnor);
       DRW_vbo_request(cache->surface_per_mat[i], &mbufcache->vbo.pos_nor);
@@ -1170,6 +1191,8 @@ void DRW_mesh_batch_cache_create_requested(
     }
   }
 
+  MEM_SAFE_FREE(saved_elem_ranges);
+
   mbufcache = (do_cage) ? &cache->cage : &cache->final;
 
   /* Edit Mesh */
diff --git a/source/blender/draw/intern/draw_common.c b/source/blender/draw/intern/draw_common.c
index e3867eef8d8..2f4a08d4525 100644
--- a/source/blender/draw/intern/draw_common.c
+++ b/source/blender/draw/intern/draw_common.c
@@ -243,6 +243,7 @@ extern char datatoc_gpu_shader_point_varying_color_frag_glsl[];
 
 extern char datatoc_object_mball_handles_vert_glsl[];
 extern char datatoc_object_empty_axes_vert_glsl[];
+extern char datatoc_object_color_axes_vert_glsl[];
 
 typedef struct COMMON_Shaders {
   struct GPUShader *shape_outline;
@@ -262,6 +263,7 @@ typedef struct COMMON_Shaders {
   struct GPUShader *volume_velocity_needle_sh;
   struct GPUShader *volume_velocity_sh;
   struct GPUShader *empty_axes_sh;
+  struct GPUShader *color_axes_sh;
 
   struct GPUShader *mball_handles;
 } COMMON_Shaders;
@@ -549,6 +551,37 @@ struct DRWCallBuffer *buffer_instance_empty_axes(DRWPass *pass,
   return DRW_shgroup_call_buffer_instance(grp, g_formats.instance_sized, geom);
 }
 
+struct DRWCallBuffer *buffer_instance_color_axes(DRWPass *pass,
+                                                 struct GPUBatch *geom,
+                                                 DRWShadingGroup **r_grp,
+                                                 eGPUShaderConfig sh_cfg)
+{
+  COMMON_Shaders *sh_data = &g_shaders[sh_cfg];
+  if (sh_data->color_axes_sh == NULL) {
+    const GPUShaderConfigData *sh_cfg_data = &GPU_shader_cfg_data[sh_cfg];
+    sh_data->color_axes_sh = GPU_shader_create_from_arrays({
+        .vert = (const char *[]){sh_cfg_data->lib, datatoc_object_color_axes_vert_glsl, NULL},
+        .frag = (const char *[]){datatoc_gpu_shader_flat_color_frag_glsl, NULL},
+        .defs = (const char *[]){sh_cfg_data->def, NULL},
+    });
+  }
+
+  DRW_shgroup_instance_format(g_formats.instance_sized,
+                              {
+                                  {"color", DRW_ATTR_FLOAT, 3},
+                                  {"size", DRW_ATTR_FLOAT, 1},
+                                  {"InstanceModelMatrix", DRW_ATTR_FLOAT, 16},
+                              });
+
+  DRWShadingGroup *grp = DRW_shgroup_create(sh_data->color_axes_sh, pass);
+  DRW_shgroup_uniform_vec3(grp, "screenVecs[0]", DRW_viewport_screenvecs_get(), 2);
+  if (sh_cfg == GPU_SHADER_CFG_CLIPPED) {
+    DRW_shgroup_state_enable(grp, DRW_STATE_CLIP_PLANES);
+  }
+  *r_grp = grp;
+  return DRW_shgroup_call_buffer_instance(grp, g_formats.instance_sized, geom);
+}
+
 struct DRWCallBuffer *buffer_instance_outline(DRWPass *pass,
                                               struct GPUBatch *geom,
                                               const int *baseid,
diff --git a/source/blender/draw/intern/draw_common.h b/source/blender/draw/intern/draw_common.h
index 05d7bafa00d..9899b6c0194 100644
--- a/source/blender/draw/intern/draw_common.h
+++ b/source/blender/draw/intern/draw_common.h
@@ -30,7 +30,6 @@ struct GPUBatch;
 struct GPUMaterial;
 struct ModifierData;
 struct Object;
-struct PTCacheEdit;
 struct ParticleSystem;
 struct ViewLayer;
 
@@ -113,7 +112,6 @@ typedef struct GlobalsUboStorage {
   /* Pack individual float at the end of the buffer to avoid alignment errors */
   float sizeLightCenter, sizeLightCircle, sizeLightCircleShadow;
   float sizeVertex, sizeEdge, sizeEdgeFix, sizeFaceDot;
-  float gridDistance, gridResolution, gridSubdivisions, gridScale;
 
   float pad_globalsBlock;
 } GlobalsUboStorage;
@@ -162,6 +160,10 @@ struct DRWCallBuffer *buffer_instance_screen_aligned(struct DRWPass *pass,
 struct DRWCallBuffer *buffer_instance_empty_axes(struct DRWPass *pass,
                                                  struct GPUBatch *geom,
                                                  eGPUShaderConfig sh_cfg);
+struct DRWCallBuffer *buffer_instance_color_axes(struct DRWPass *pass,
+                                                 struct GPUBatch *geom,
+                                                 struct DRWShadingGroup **r_grp,
+                                                 eGPUShaderConfig sh_cfg);
 struct DRWCallBuffer *buffer_instance_scaled(struct DRWPass *pass,
                                              struct GPUBatch *geom,
                                              eGPUShaderConfig sh_cfg);
diff --git a/source/blender/draw/intern/draw_hair.c b/source/blender/draw/intern/draw_hair.c
index fa1f1f2aab4..44ed01c47aa 100644
--- a/source/blender/draw/intern/draw_hair.c
+++ b/source/blender/draw/intern/draw_hair.c
@@ -253,7 +253,7 @@ void DRW_hair_update(void)
    * On some system it crashes (see T58489) and on some other it renders garbage (see T60171).
    *
    * So instead of using transform feedback we render to a texture,
-   * readback the result to system memory and reupload as VBO data.
+   * read back the result to system memory and re-upload as VBO data.
    * It is really not ideal performance wise, but it is the simplest
    * and the most local workaround that still uses the power of the GPU.
    */
diff --git a/source/blender/draw/intern/draw_instance_data.c b/source/blender/draw/intern/draw_instance_data.c
index 802f49d6549..8f26cc72a02 100644
--- a/source/blender/draw/intern/draw_instance_data.c
+++ b/source/blender/draw/intern/draw_instance_data.c
@@ -23,7 +23,7 @@
 /**
  * DRW Instance Data Manager
  * This is a special memory manager that keeps memory blocks ready to send as vbo data in one
- * continuous allocation. This way we avoid feeding gawain each instance data one by one and
+ * continuous allocation. This way we avoid feeding #GPUBatch each instance data one by one and
  * unnecessary memcpy. Since we loose which memory block was used each #DRWShadingGroup we need to
  * redistribute them in the same order/size to avoid to realloc each frame. This is why
  * #DRWInstanceDatas are sorted in a list for each different data size.
diff --git a/source/blender/draw/intern/draw_instance_data.h b/source/blender/draw/intern/draw_instance_data.h
index d88de1a58e2..2ede68e16d8 100644
--- a/source/blender/draw/intern/draw_instance_data.h
+++ b/source/blender/draw/intern/draw_instance_data.h
@@ -35,8 +35,6 @@
 typedef struct DRWInstanceData DRWInstanceData;
 typedef struct DRWInstanceDataList DRWInstanceDataList;
 
-struct DRWShadingGroup;
-
 void *DRW_instance_data_next(DRWInstanceData *idata);
 DRWInstanceData *DRW_instance_data_request(DRWInstanceDataList *idatalist, uint attr_size);
 
diff --git a/source/blender/draw/intern/draw_manager.c b/source/blender/draw/intern/draw_manager.c
index ebbe3694df8..d88629e9bb7 100644
--- a/source/blender/draw/intern/draw_manager.c
+++ b/source/blender/draw/intern/draw_manager.c
@@ -33,6 +33,7 @@
 
 #include "BKE_anim.h"
 #include "BKE_colortools.h"
+#include "BKE_context.h"
 #include "BKE_curve.h"
 #include "BKE_editmesh.h"
 #include "BKE_global.h"
@@ -41,6 +42,7 @@
 #include "BKE_main.h"
 #include "BKE_mball.h"
 #include "BKE_mesh.h"
+#include "BKE_modifier.h"
 #include "BKE_object.h"
 #include "BKE_particle.h"
 #include "BKE_paint.h"
@@ -604,6 +606,7 @@ static DRWCallState *draw_unit_state_create(void)
 
   state->ob_index = 0;
   state->ob_random = 0.0f;
+  copy_v3_fl(state->ob_color, 1.0f);
 
   /* TODO(fclem) get rid of this. */
   state->culling = BLI_memblock_alloc(DST.vmempool->cullstates);
@@ -1865,7 +1868,7 @@ void DRW_render_gpencil(struct RenderEngine *engine, struct Depsgraph *depsgraph
     DRW_opengl_render_context_enable(re_gl_context);
     /* We need to query gpu context after a gl context has been bound. */
     re_gpu_context = RE_gpu_context_get(render);
-    DRW_gawain_render_context_enable(re_gpu_context);
+    DRW_gpu_render_context_enable(re_gpu_context);
   }
   else {
     DRW_opengl_context_enable();
@@ -1950,13 +1953,13 @@ void DRW_render_to_image(RenderEngine *engine, struct Depsgraph *depsgraph)
     DRW_opengl_render_context_enable(re_gl_context);
     /* We need to query gpu context after a gl context has been bound. */
     re_gpu_context = RE_gpu_context_get(render);
-    DRW_gawain_render_context_enable(re_gpu_context);
+    DRW_gpu_render_context_enable(re_gpu_context);
   }
   else {
     DRW_opengl_context_enable();
   }
 
-  /* IMPORTANT: We dont support immediate mode in render mode!
+  /* IMPORTANT: We don't support immediate mode in render mode!
    * This shall remain in effect until immediate mode supports
    * multiple threads. */
 
@@ -2041,7 +2044,7 @@ void DRW_render_to_image(RenderEngine *engine, struct Depsgraph *depsgraph)
 
   /* Changing Context */
   if (re_gl_context != NULL) {
-    DRW_gawain_render_context_disable(re_gpu_context);
+    DRW_gpu_render_context_disable(re_gpu_context);
     DRW_opengl_render_context_disable(re_gl_context);
   }
   else {
@@ -2204,24 +2207,43 @@ void DRW_draw_select_loop(struct Depsgraph *depsgraph,
   drw_state_prepare_clean_for_draw(&DST);
 
   bool use_obedit = false;
-  int obedit_mode = 0;
+  /* obedit_ctx_mode is used for selecting the right draw engines */
+  eContextObjectMode obedit_ctx_mode;
+  /* object_mode is used for filtering objects in the depsgraph */
+  eObjectMode object_mode;
+  int object_type = 0;
   if (obedit != NULL) {
+    object_type = obedit->type;
+    object_mode = obedit->mode;
     if (obedit->type == OB_MBALL) {
       use_obedit = true;
-      obedit_mode = CTX_MODE_EDIT_METABALL;
+      obedit_ctx_mode = CTX_MODE_EDIT_METABALL;
     }
     else if (obedit->type == OB_ARMATURE) {
       use_obedit = true;
-      obedit_mode = CTX_MODE_EDIT_ARMATURE;
+      obedit_ctx_mode = CTX_MODE_EDIT_ARMATURE;
     }
   }
   if (v3d->overlay.flag & V3D_OVERLAY_BONE_SELECT) {
     if (!(v3d->flag2 & V3D_HIDE_OVERLAYS)) {
       /* Note: don't use "BKE_object_pose_armature_get" here, it breaks selection. */
       Object *obpose = OBPOSE_FROM_OBACT(obact);
+      if (obpose == NULL) {
+        Object *obweight = OBWEIGHTPAINT_FROM_OBACT(obact);
+        if (obweight) {
+          /* Only use Armature pose selection, when connected armature is in pose mode. */
+          Object *ob_armature = modifiers_isDeformedByArmature(obweight);
+          if (ob_armature && ob_armature->mode == OB_MODE_POSE) {
+            obpose = ob_armature;
+          }
+        }
+      }
+
       if (obpose) {
         use_obedit = true;
-        obedit_mode = CTX_MODE_POSE;
+        object_type = obpose->type;
+        object_mode = obpose->mode;
+        obedit_ctx_mode = CTX_MODE_POSE;
       }
     }
   }
@@ -2235,8 +2257,8 @@ void DRW_draw_select_loop(struct Depsgraph *depsgraph,
 
   /* Get list of enabled engines */
   if (use_obedit) {
-    drw_engines_enable_from_paint_mode(obedit_mode);
-    drw_engines_enable_from_mode(obedit_mode);
+    drw_engines_enable_from_paint_mode(obedit_ctx_mode);
+    drw_engines_enable_from_mode(obedit_ctx_mode);
   }
   else if (!draw_surface) {
     /* grease pencil selection */
@@ -2283,7 +2305,7 @@ void DRW_draw_select_loop(struct Depsgraph *depsgraph,
     drw_engines_world_update(scene);
 
     if (use_obedit) {
-      FOREACH_OBJECT_IN_MODE_BEGIN (view_layer, v3d, obact->type, obact->mode, ob_iter) {
+      FOREACH_OBJECT_IN_MODE_BEGIN (view_layer, v3d, object_type, object_mode, ob_iter) {
         drw_engines_cache_populate(ob_iter);
       }
       FOREACH_OBJECT_IN_MODE_END;
@@ -2563,7 +2585,6 @@ void DRW_draw_select_id(Depsgraph *depsgraph, ARegion *ar, View3D *v3d, const rc
 
   /* Reset before using it. */
   drw_state_prepare_clean_for_draw(&DST);
-  DST.buffer_finish_called = true;
 
   /* Instead of 'DRW_context_state_init(C, &DST.draw_ctx)', assign from args */
   DST.draw_ctx = (DRWContextState){
@@ -2601,6 +2622,8 @@ void DRW_draw_select_id(Depsgraph *depsgraph, ARegion *ar, View3D *v3d, const rc
     }
 
     drw_engines_cache_finish();
+
+    DRW_render_instance_buffer_finish();
   }
 
   /* Start Drawing */
@@ -2610,6 +2633,8 @@ void DRW_draw_select_id(Depsgraph *depsgraph, ARegion *ar, View3D *v3d, const rc
 
   drw_engines_disable();
 
+  drw_viewport_cache_resize();
+
 #ifdef DEBUG
   /* Avoid accidental reuse. */
   drw_state_ensure_not_reused(&DST);
@@ -2923,7 +2948,7 @@ void DRW_opengl_context_create(void)
   /* This changes the active context. */
   DST.gl_context = WM_opengl_context_create();
   WM_opengl_context_activate(DST.gl_context);
-  /* Be sure to create gawain.context too. */
+  /* Be sure to create gpu_context too. */
   DST.gpu_context = GPU_context_create(0);
   if (!G.background) {
     immActivate();
@@ -3022,7 +3047,7 @@ void DRW_opengl_render_context_disable(void *re_gl_context)
 }
 
 /* Needs to be called AFTER DRW_opengl_render_context_enable() */
-void DRW_gawain_render_context_enable(void *re_gpu_context)
+void DRW_gpu_render_context_enable(void *re_gpu_context)
 {
   /* If thread is main you should use DRW_opengl_context_enable(). */
   BLI_assert(!BLI_thread_is_main());
@@ -3032,7 +3057,7 @@ void DRW_gawain_render_context_enable(void *re_gpu_context)
 }
 
 /* Needs to be called BEFORE DRW_opengl_render_context_disable() */
-void DRW_gawain_render_context_disable(void *UNUSED(re_gpu_context))
+void DRW_gpu_render_context_disable(void *UNUSED(re_gpu_context))
 {
   DRW_shape_cache_reset(); /* XXX fix that too. */
   GPU_context_active_set(NULL);
diff --git a/source/blender/draw/intern/draw_manager.h b/source/blender/draw/intern/draw_manager.h
index f37e713e374..85f6cf05e83 100644
--- a/source/blender/draw/intern/draw_manager.h
+++ b/source/blender/draw/intern/draw_manager.h
@@ -101,6 +101,7 @@ enum {
   DRW_CALL_MODELVIEWPROJECTION = (1 << 1),
   DRW_CALL_ORCOTEXFAC = (1 << 2),
   DRW_CALL_OBJECTINFO = (1 << 3),
+  DRW_CALL_OBJECTCOLOR = (1 << 4),
 };
 
 typedef struct DRWCullingState {
@@ -122,6 +123,7 @@ typedef struct DRWCallState {
   float modelinverse[4][4];
   float orcotexfac[2][3];
   float ob_random;
+  float ob_color[4];
 } DRWCallState;
 
 typedef struct DRWCall {
@@ -196,6 +198,7 @@ struct DRWShadingGroup {
   int orcotexfac;
   int callid;
   int objectinfo;
+  int objectcolor;
   uchar matflag; /* Matrices needed, same as DRWCall.flag */
 
   DRWPass *pass_parent; /* backlink to pass we're in */
diff --git a/source/blender/draw/intern/draw_manager_data.c b/source/blender/draw/intern/draw_manager_data.c
index 99ab25645d2..8b7cb9c1dad 100644
--- a/source/blender/draw/intern/draw_manager_data.c
+++ b/source/blender/draw/intern/draw_manager_data.c
@@ -387,6 +387,10 @@ static void drw_call_state_update_matflag(DRWCallState *state,
     }
     state->ob_random = random * (1.0f / (float)0xFFFFFFFF);
   }
+
+  if (new_flags & DRW_CALL_OBJECTCOLOR) {
+    copy_v4_v4(state->ob_color, ob->color);
+  }
 }
 
 static DRWCallState *drw_call_state_create(DRWShadingGroup *shgroup, float (*obmat)[4], Object *ob)
@@ -836,6 +840,7 @@ static void drw_shgroup_init(DRWShadingGroup *shgroup, GPUShader *shader)
   shgroup->modelviewprojection = GPU_shader_get_builtin_uniform(shader, GPU_UNIFORM_MVP);
   shgroup->orcotexfac = GPU_shader_get_builtin_uniform(shader, GPU_UNIFORM_ORCO);
   shgroup->objectinfo = GPU_shader_get_builtin_uniform(shader, GPU_UNIFORM_OBJECT_INFO);
+  shgroup->objectcolor = GPU_shader_get_builtin_uniform(shader, GPU_UNIFORM_OBJECT_COLOR);
   shgroup->callid = GPU_shader_get_builtin_uniform(shader, GPU_UNIFORM_CALLID);
 
   shgroup->matflag = 0;
@@ -851,6 +856,9 @@ static void drw_shgroup_init(DRWShadingGroup *shgroup, GPUShader *shader)
   if (shgroup->objectinfo > -1) {
     shgroup->matflag |= DRW_CALL_OBJECTINFO;
   }
+  if (shgroup->objectcolor > -1) {
+    shgroup->matflag |= DRW_CALL_OBJECTCOLOR;
+  }
 }
 
 static DRWShadingGroup *drw_shgroup_create_ex(struct GPUShader *shader, DRWPass *pass)
diff --git a/source/blender/draw/intern/draw_manager_exec.c b/source/blender/draw/intern/draw_manager_exec.c
index 3bf442b4f39..50408015fbc 100644
--- a/source/blender/draw/intern/draw_manager_exec.c
+++ b/source/blender/draw/intern/draw_manager_exec.c
@@ -592,6 +592,10 @@ static void draw_geometry_prepare(DRWShadingGroup *shgroup, DRWCall *call)
     infos[3] = (state->flag & DRW_CALL_NEGSCALE) ? -1.0f : 1.0f;
     GPU_shader_uniform_vector(shgroup->shader, shgroup->objectinfo, 4, 1, (float *)infos);
   }
+  if (shgroup->objectcolor != -1) {
+    GPU_shader_uniform_vector(
+        shgroup->shader, shgroup->objectcolor, 4, 1, (float *)state->ob_color);
+  }
   if (shgroup->orcotexfac != -1) {
     GPU_shader_uniform_vector(
         shgroup->shader, shgroup->orcotexfac, 3, 2, (float *)state->orcotexfac);
@@ -622,12 +626,12 @@ BLI_INLINE void draw_geometry_execute(DRWShadingGroup *shgroup,
     GPU_batch_bind(geom);
   }
 
-  /* XXX hacking gawain. we don't want to call glUseProgram! (huge performance loss) */
+  /* XXX hacking #GPUBatch. we don't want to call glUseProgram! (huge performance loss) */
   geom->program_in_use = true;
 
   GPU_batch_draw_advanced(geom, vert_first, vert_count, inst_first, inst_count);
 
-  geom->program_in_use = false; /* XXX hacking gawain */
+  geom->program_in_use = false; /* XXX hacking #GPUBatch */
 }
 
 enum {
diff --git a/source/blender/draw/intern/draw_manager_shader.c b/source/blender/draw/intern/draw_manager_shader.c
index d0aa6d55c03..9c34cdbac3b 100644
--- a/source/blender/draw/intern/draw_manager_shader.c
+++ b/source/blender/draw/intern/draw_manager_shader.c
@@ -178,7 +178,7 @@ static void drw_deferred_shader_compilation_free(void *custom_data)
 
 static void drw_deferred_shader_add(GPUMaterial *mat, bool deferred)
 {
-  /* Do not deferre the compilation if we are rendering for image.
+  /* Do not defer the compilation if we are rendering for image.
    * deferred rendering is only possible when `evil_C` is available */
   if (DST.draw_ctx.evil_C == NULL || DRW_state_is_image_render() || !USE_DEFERRED_COMPILATION ||
       !deferred) {
diff --git a/source/blender/draw/intern/draw_select_buffer.c b/source/blender/draw/intern/draw_select_buffer.c
index 974ea22ccea..6454fc71e56 100644
--- a/source/blender/draw/intern/draw_select_buffer.c
+++ b/source/blender/draw/intern/draw_select_buffer.c
@@ -75,14 +75,15 @@ uint *DRW_select_buffer_read(struct Depsgraph *depsgraph,
     DRW_draw_select_id(depsgraph, ar, v3d, rect);
 
     if (select_ctx->index_drawn_len > 1) {
-      BLI_assert(ar->winx == GPU_texture_width(select_ctx->texture_u32) &&
-                 ar->winy == GPU_texture_height(select_ctx->texture_u32));
+      BLI_assert(ar->winx == GPU_texture_width(DRW_engine_select_texture_get()) &&
+                 ar->winy == GPU_texture_height(DRW_engine_select_texture_get()));
 
       /* Read the UI32 pixels. */
       buf_len = BLI_rcti_size_x(rect) * BLI_rcti_size_y(rect);
       r_buf = MEM_mallocN(buf_len * sizeof(*r_buf), __func__);
 
-      GPU_framebuffer_bind(select_ctx->framebuffer_select_id);
+      GPUFrameBuffer *select_id_fb = DRW_engine_select_framebuffer_get();
+      GPU_framebuffer_bind(select_id_fb);
       glReadBuffer(GL_COLOR_ATTACHMENT0);
       glReadPixels(rect_clamp.xmin,
                    rect_clamp.ymin,
diff --git a/source/blender/draw/modes/object_mode.c b/source/blender/draw/modes/object_mode.c
index 87fc74f1f72..716372560db 100644
--- a/source/blender/draw/modes/object_mode.c
+++ b/source/blender/draw/modes/object_mode.c
@@ -234,6 +234,7 @@ typedef struct OBJECT_ShadingGroupList {
   /* Helpers */
   DRWCallBuffer *relationship_lines;
   DRWCallBuffer *constraint_lines;
+  DRWCallBuffer *origin_xform;
 
   /* Camera */
   DRWCallBuffer *camera;
@@ -332,7 +333,7 @@ static struct {
 
   OBJECT_Shaders sh_data[GPU_SHADER_CFG_LEN];
 
-  float grid_settings[5];
+  float grid_distance;
   float grid_mesh_size;
   int grid_flag;
   float grid_axes[3];
@@ -340,6 +341,7 @@ static struct {
   int zneg_flag;
   float zplane_axes[3];
   float inv_viewport_size[2];
+  float grid_steps[8];
   bool draw_grid;
   /* Temp buffer textures */
   struct GPUTexture *outlines_depth_tx;
@@ -559,8 +561,6 @@ static void OBJECT_engine_init(void *vedata)
     View3D *v3d = draw_ctx->v3d;
     Scene *scene = draw_ctx->scene;
     RegionView3D *rv3d = draw_ctx->rv3d;
-    float grid_scale = ED_view3d_grid_scale(scene, v3d, NULL);
-    float grid_res;
 
     const bool show_axis_x = (v3d->gridflag & V3D_SHOW_X) != 0;
     const bool show_axis_y = (v3d->gridflag & V3D_SHOW_Y) != 0;
@@ -576,21 +576,6 @@ static void OBJECT_engine_init(void *vedata)
 
     /* if perps */
     if (winmat[3][3] == 0.0f) {
-      float fov;
-      float viewvecs[2][4] = {
-          {1.0f, -1.0f, -1.0f, 1.0f},
-          {-1.0f, 1.0f, -1.0f, 1.0f},
-      };
-
-      /* convert the view vectors to view space */
-      for (int i = 0; i < 2; i++) {
-        mul_m4_v4(wininv, viewvecs[i]);
-        mul_v3_fl(viewvecs[i], 1.0f / viewvecs[i][2]); /* perspective divide */
-      }
-
-      fov = angle_v3v3(viewvecs[0], viewvecs[1]) / 2.0f;
-      grid_res = fabsf(tanf(fov)) / grid_scale;
-
       e_data.grid_flag = (1 << 4); /* XY plane */
       if (show_axis_x) {
         e_data.grid_flag |= SHOW_AXIS_X;
@@ -603,14 +588,6 @@ static void OBJECT_engine_init(void *vedata)
       }
     }
     else {
-      if (rv3d->view != RV3D_VIEW_USER) {
-        /* Allow 3 more subdivisions. */
-        grid_scale /= powf(v3d->gridsubdiv, 3);
-      }
-
-      float viewdist = 1.0f / max_ff(fabsf(winmat[0][0]), fabsf(winmat[1][1]));
-      grid_res = viewdist / grid_scale;
-
       if (ELEM(rv3d->view, RV3D_VIEW_RIGHT, RV3D_VIEW_LEFT)) {
         e_data.draw_grid = show_ortho_grid;
         e_data.grid_flag = PLANE_YZ | SHOW_AXIS_Y | SHOW_AXIS_Z | SHOW_GRID | GRID_BACK;
@@ -688,12 +665,7 @@ static void OBJECT_engine_init(void *vedata)
       dist = v3d->clip_end;
     }
 
-    e_data.grid_settings[0] = dist / 2.0f;     /* gridDistance */
-    e_data.grid_settings[1] = grid_res;        /* gridResolution */
-    e_data.grid_settings[2] = grid_scale;      /* gridScale */
-    e_data.grid_settings[3] = v3d->gridsubdiv; /* gridSubdiv */
-    e_data.grid_settings[4] = (v3d->gridsubdiv > 1) ? 1.0f / logf(v3d->gridsubdiv) :
-                                                      0.0f; /* 1/log(gridSubdiv) */
+    e_data.grid_distance = dist / 2.0f; /* gridDistance */
 
     if (winmat[3][3] == 0.0f) {
       e_data.grid_mesh_size = dist;
@@ -702,6 +674,8 @@ static void OBJECT_engine_init(void *vedata)
       float viewdist = 1.0f / min_ff(fabsf(winmat[0][0]), fabsf(winmat[1][1]));
       e_data.grid_mesh_size = viewdist * dist;
     }
+
+    ED_view3d_grid_steps(scene, v3d, rv3d, e_data.grid_steps);
   }
 
   copy_v2_v2(e_data.inv_viewport_size, DRW_viewport_size_get());
@@ -1512,10 +1486,10 @@ static void OBJECT_cache_init(void *vedata)
     DRWShadingGroup *grp = DRW_shgroup_create(sh_data->grid, psl->grid);
     DRW_shgroup_uniform_int(grp, "gridFlag", &e_data.zneg_flag, 1);
     DRW_shgroup_uniform_vec3(grp, "planeAxes", e_data.zplane_axes, 1);
-    DRW_shgroup_uniform_vec4(grp, "gridSettings", e_data.grid_settings, 1);
+    DRW_shgroup_uniform_vec3(grp, "screenVecs[0]", DRW_viewport_screenvecs_get(), 2);
+    DRW_shgroup_uniform_float(grp, "gridDistance", &e_data.grid_distance, 1);
     DRW_shgroup_uniform_float_copy(grp, "lineKernel", grid_line_size);
     DRW_shgroup_uniform_float_copy(grp, "meshSize", e_data.grid_mesh_size);
-    DRW_shgroup_uniform_float(grp, "gridOneOverLogSubdiv", &e_data.grid_settings[4], 1);
     DRW_shgroup_uniform_block(grp, "globalsBlock", G_draw.block_ubo);
     DRW_shgroup_uniform_texture_ref(grp, "depthBuffer", &dtxl->depth);
     DRW_shgroup_call(grp, geom, NULL);
@@ -1525,6 +1499,7 @@ static void OBJECT_cache_init(void *vedata)
     DRW_shgroup_uniform_vec3(grp, "planeAxes", e_data.grid_axes, 1);
     DRW_shgroup_uniform_block(grp, "globalsBlock", G_draw.block_ubo);
     DRW_shgroup_uniform_texture_ref(grp, "depthBuffer", &dtxl->depth);
+    DRW_shgroup_uniform_float(grp, "gridSteps", e_data.grid_steps, ARRAY_SIZE(e_data.grid_steps));
     DRW_shgroup_call(grp, geom, NULL);
 
     grp = DRW_shgroup_create(sh_data->grid, psl->grid);
@@ -1761,6 +1736,16 @@ static void OBJECT_cache_init(void *vedata)
     sgl->constraint_lines = buffer_dynlines_dashed_uniform_color(
         sgl->non_meshes, gb->colorGridAxisZ, draw_ctx->sh_cfg);
 
+    {
+      DRWShadingGroup *grp_axes;
+      sgl->origin_xform = buffer_instance_color_axes(
+          sgl->non_meshes, DRW_cache_bone_arrows_get(), &grp_axes, draw_ctx->sh_cfg);
+
+      DRW_shgroup_state_disable(grp_axes, DRW_STATE_DEPTH_LESS_EQUAL);
+      DRW_shgroup_state_enable(grp_axes, DRW_STATE_DEPTH_ALWAYS);
+      DRW_shgroup_state_enable(grp_axes, DRW_STATE_WIRE_SMOOTH);
+    }
+
     /* Force Field Curve Guide End (here because of stipple) */
     /* TODO port to shader stipple */
     geom = DRW_cache_screenspace_circle_get();
@@ -3701,6 +3686,15 @@ static void OBJECT_cache_populate(void *vedata, Object *ob)
     DRW_shgroup_bounds(sgl, ob, theme_id);
   }
 
+  /* Helpers for when we're transforming origins. */
+  if (scene->toolsettings->transform_flag & SCE_XFORM_DATA_ORIGIN) {
+    if (ob->base_flag & BASE_SELECTED) {
+      const float color[4] = {0.75, 0.75, 0.75, 0.5};
+      float axes_size = 1.0f;
+      DRW_buffer_add_entry(sgl->origin_xform, color, &axes_size, ob->obmat);
+    }
+  }
+
   /* don't show object extras in set's */
   if ((ob->base_flag & (BASE_FROM_SET | BASE_FROM_DUPLI)) == 0) {
     if ((draw_ctx->object_mode & (OB_MODE_ALL_PAINT | OB_MODE_ALL_PAINT_GPENCIL)) == 0) {
@@ -3711,7 +3705,7 @@ static void OBJECT_cache_populate(void *vedata, Object *ob)
       DRW_shgroup_relationship_lines(sgl, draw_ctx->depsgraph, scene, ob);
     }
 
-    const bool draw_extra = (ob->dtx != 0);
+    const bool draw_extra = ob->dtx & (OB_DRAWNAME | OB_TEXSPACE | OB_DRAWBOUNDOX);
     if (draw_extra && (theme_id == TH_UNDEFINED)) {
       theme_id = DRW_object_wire_theme_get(ob, view_layer, NULL);
     }
diff --git a/source/blender/draw/modes/shaders/common_globals_lib.glsl b/source/blender/draw/modes/shaders/common_globals_lib.glsl
index f75ef06b6d9..e559224eb9e 100644
--- a/source/blender/draw/modes/shaders/common_globals_lib.glsl
+++ b/source/blender/draw/modes/shaders/common_globals_lib.glsl
@@ -75,11 +75,6 @@ layout(std140) uniform globalsBlock
   float sizeEdgeFix;
   float sizeFaceDot;
 
-  float gridDistance;
-  float gridResolution;
-  float gridSubdivisions;
-  float gridScale;
-
   float pad_globalsBlock;
 };
 
diff --git a/source/blender/draw/modes/shaders/object_color_axes_vert.glsl b/source/blender/draw/modes/shaders/object_color_axes_vert.glsl
new file mode 100644
index 00000000000..239dec30c42
--- /dev/null
+++ b/source/blender/draw/modes/shaders/object_color_axes_vert.glsl
@@ -0,0 +1,35 @@
+
+uniform mat4 ViewProjectionMatrix;
+uniform vec3 screenVecs[3];
+
+/* ---- Instantiated Attrs ---- */
+in float axis; /* position on the axis. [0.0-1.0] is X axis, [1.0-2.0] is Y, etc... */
+in vec2 screenPos;
+in vec3 colorAxis;
+
+/* ---- Per instance Attrs ---- */
+in mat4 InstanceModelMatrix;
+in vec4 color;
+
+flat out vec4 finalColor;
+
+void main()
+{
+  float draw_size = 4.0;
+  vec3 chosen_axis = InstanceModelMatrix[int(axis)].xyz;
+  vec3 loc = InstanceModelMatrix[3].xyz;
+  vec3 wpos = loc + chosen_axis * fract(axis) * draw_size;
+  vec3 spos = screenVecs[0].xyz * screenPos.x + screenVecs[1].xyz * screenPos.y;
+  /* Scale uniformly by axis length */
+  spos *= length(chosen_axis) * draw_size;
+
+  vec4 pos_4d = vec4(wpos + spos, 1.0);
+  gl_Position = ViewProjectionMatrix * pos_4d;
+
+  finalColor.rgb = mix(colorAxis, color.rgb, color.a);
+  finalColor.a = 1.0;
+
+#ifdef USE_WORLD_CLIP_PLANES
+  world_clip_planes_calc_clip_distance(pos_4d.xyz);
+#endif
+}
diff --git a/source/blender/draw/modes/shaders/object_empty_image_frag.glsl b/source/blender/draw/modes/shaders/object_empty_image_frag.glsl
index 7dfbf469adc..e33aa6cdcc1 100644
--- a/source/blender/draw/modes/shaders/object_empty_image_frag.glsl
+++ b/source/blender/draw/modes/shaders/object_empty_image_frag.glsl
@@ -37,9 +37,15 @@ void main()
 
   if (depthMode == DEPTH_BACK) {
     gl_FragDepth = 0.999999;
+#ifdef USE_WIRE
+    gl_FragDepth -= 1e-5;
+#endif
   }
   else if (depthMode == DEPTH_FRONT) {
     gl_FragDepth = 0.000001;
+#ifdef USE_WIRE
+    gl_FragDepth -= 1e-5;
+#endif
   }
   else if (depthMode == DEPTH_UNCHANGED) {
     gl_FragDepth = gl_FragCoord.z;
diff --git a/source/blender/draw/modes/shaders/object_grid_frag.glsl b/source/blender/draw/modes/shaders/object_grid_frag.glsl
index a20f12efd93..751d839eb79 100644
--- a/source/blender/draw/modes/shaders/object_grid_frag.glsl
+++ b/source/blender/draw/modes/shaders/object_grid_frag.glsl
@@ -9,21 +9,19 @@ in vec3 local_pos;
 out vec4 FragColor;
 
 uniform vec3 planeAxes;
-uniform vec4 gridSettings;
+uniform vec3 screenVecs[2];
+uniform float gridDistance;
 uniform float meshSize;
 uniform float lineKernel = 0.0;
-uniform float gridOneOverLogSubdiv;
 uniform sampler2D depthBuffer;
 
-#define gridDistance gridSettings.x
-#define gridResolution gridSettings.y
-#define gridScale gridSettings.z
-#define gridSubdiv gridSettings.w
-
 #define cameraPos (ViewMatrixInverse[3].xyz)
 
 uniform int gridFlag;
 
+#define STEPS_LEN 8
+uniform float gridSteps[STEPS_LEN] = float[](0.001, 0.01, 0.1, 1.0, 10.0, 100.0, 1000.0, 10000.0);
+
 #define AXIS_X (1 << 0)
 #define AXIS_Y (1 << 1)
 #define AXIS_Z (1 << 2)
@@ -73,10 +71,14 @@ vec3 get_axes(vec3 co, vec3 fwidthCos, float line_size)
                           axes_domain - (line_size + lineKernel));
 }
 
+#define linearstep(p0, p1, v) (clamp(((v) - (p0)) / abs((p1) - (p0)), 0.0, 1.0))
+
 void main()
 {
   vec3 wPos = local_pos * meshSize;
-  vec3 fwidthPos = fwidth(wPos);
+  vec3 dFdxPos = dFdx(wPos);
+  vec3 dFdyPos = dFdy(wPos);
+  vec3 fwidthPos = abs(dFdxPos) + abs(dFdyPos);
   wPos += cameraPos * planeAxes;
 
   float dist, fade;
@@ -116,15 +118,54 @@ void main()
   }
 
   if ((gridFlag & GRID) != 0) {
-    float grid_res = log(dist * gridResolution) * gridOneOverLogSubdiv;
+    /* Using `max(dot(dFdxPos, screenVecs[0]), dot(dFdyPos, screenVecs[1]))`
+     * would be more accurate, but not really necessary. */
+    float grid_res = dot(dFdxPos, screenVecs[0]);
 
-    float blend = fract(-max(grid_res, 0.0));
-    float lvl = floor(grid_res);
+    /* The gride begins to appear when it comprises 4 pixels */
+    grid_res *= 4;
 
     /* from biggest to smallest */
-    float scaleA = gridScale * pow(gridSubdiv, max(lvl - 1.0, 0.0));
-    float scaleB = gridScale * pow(gridSubdiv, max(lvl + 0.0, 0.0));
-    float scaleC = gridScale * pow(gridSubdiv, max(lvl + 1.0, 1.0));
+    vec4 scale;
+#if 0
+    int step_id = 0;
+    scale[0] = 0.0;
+    scale[1] = gridSteps[0];
+    while (scale[1] < grid_res && step_id != STEPS_LEN - 1) {
+      scale[0] = scale[1];
+      scale[1] = gridSteps[++step_id];
+    }
+    scale[2] = gridSteps[min(step_id + 1, STEPS_LEN - 1)];
+    scale[3] = gridSteps[min(step_id + 2, STEPS_LEN - 1)];
+#else
+    /* For more efficiency, unroll the loop above. */
+    if (gridSteps[0] > grid_res) {
+      scale = vec4(0.0, gridSteps[0], gridSteps[1], gridSteps[2]);
+    }
+    else if (gridSteps[1] > grid_res) {
+      scale = vec4(gridSteps[0], gridSteps[1], gridSteps[2], gridSteps[3]);
+    }
+    else if (gridSteps[2] > grid_res) {
+      scale = vec4(gridSteps[1], gridSteps[2], gridSteps[3], gridSteps[4]);
+    }
+    else if (gridSteps[3] > grid_res) {
+      scale = vec4(gridSteps[2], gridSteps[3], gridSteps[4], gridSteps[5]);
+    }
+    else if (gridSteps[4] > grid_res) {
+      scale = vec4(gridSteps[3], gridSteps[4], gridSteps[5], gridSteps[6]);
+    }
+    else if (gridSteps[5] > grid_res) {
+      scale = vec4(gridSteps[4], gridSteps[5], gridSteps[6], gridSteps[7]);
+    }
+    else if (gridSteps[6] > grid_res) {
+      scale = vec4(gridSteps[5], gridSteps[6], gridSteps[7], gridSteps[7]);
+    }
+    else {
+      scale = vec4(gridSteps[6], gridSteps[7], gridSteps[7], gridSteps[7]);
+    }
+#endif
+    float blend = 1.0 - linearstep(scale[0], scale[1], grid_res);
+    blend = blend * blend * blend;
 
     vec2 grid_pos, grid_fwidth;
     if ((gridFlag & PLANE_XZ) != 0) {
@@ -140,9 +181,9 @@ void main()
       grid_fwidth = fwidthPos.xy;
     }
 
-    float gridA = get_grid(grid_pos, grid_fwidth, scaleA);
-    float gridB = get_grid(grid_pos, grid_fwidth, scaleB);
-    float gridC = get_grid(grid_pos, grid_fwidth, scaleC);
+    float gridA = get_grid(grid_pos, grid_fwidth, scale[1]);
+    float gridB = get_grid(grid_pos, grid_fwidth, scale[2]);
+    float gridC = get_grid(grid_pos, grid_fwidth, scale[3]);
 
     FragColor = colorGrid;
     FragColor.a *= gridA * blend;
diff --git a/source/blender/draw/modes/shaders/object_grid_vert.glsl b/source/blender/draw/modes/shaders/object_grid_vert.glsl
index d247967b03a..496bb011c74 100644
--- a/source/blender/draw/modes/shaders/object_grid_vert.glsl
+++ b/source/blender/draw/modes/shaders/object_grid_vert.glsl
@@ -3,14 +3,8 @@
  * Clément Foucault */
 
 uniform vec3 planeAxes;
-uniform vec4 gridSettings;
 uniform float meshSize;
 
-#define gridDistance gridSettings.x
-#define gridResolution gridSettings.y
-#define gridScale gridSettings.z
-#define gridSubdiv gridSettings.w
-
 uniform int gridFlag;
 
 #define cameraPos (ViewMatrixInverse[3].xyz)
diff --git a/source/blender/editors/animation/anim_channels_defines.c b/source/blender/editors/animation/anim_channels_defines.c
index 1649744ba8d..d80b96f0d74 100644
--- a/source/blender/editors/animation/anim_channels_defines.c
+++ b/source/blender/editors/animation/anim_channels_defines.c
@@ -4847,7 +4847,7 @@ void ANIM_channel_draw_widgets(const bContext *C,
 
   /* step 4) draw text - check if renaming widget is in use... */
   if (is_being_renamed) {
-    PointerRNA ptr = {{NULL}};
+    PointerRNA ptr = {NULL};
     PropertyRNA *prop = NULL;
 
     /* draw renaming widget if we can get RNA pointer for it
diff --git a/source/blender/editors/animation/anim_channels_edit.c b/source/blender/editors/animation/anim_channels_edit.c
index 7e913014a87..61b8e4a2341 100644
--- a/source/blender/editors/animation/anim_channels_edit.c
+++ b/source/blender/editors/animation/anim_channels_edit.c
@@ -3114,6 +3114,7 @@ static int mouse_anim_channels(bContext *C, bAnimContext *ac, int channel_index,
         ANIM_set_active_channel(ac, ac->data, ac->datatype, filter, gpl, ANIMTYPE_GPLAYER);
         /* update other layer status */
         BKE_gpencil_layer_setactive(gpd, gpl);
+        BKE_gpencil_layer_autolock_set(gpd);
       }
 
       /* Grease Pencil updates */
diff --git a/source/blender/editors/animation/anim_filter.c b/source/blender/editors/animation/anim_filter.c
index a78a63f1347..48493c9e961 100644
--- a/source/blender/editors/animation/anim_filter.c
+++ b/source/blender/editors/animation/anim_filter.c
@@ -630,9 +630,8 @@ static bAnimListElem *make_new_animlistelem(void *data,
     /* do specifics */
     switch (datatype) {
       case ANIMTYPE_SUMMARY: {
-        /* nothing to include for now... this is just a dummy wrappy around all the other channels
-         * in the DopeSheet, and gets included at the start of the list
-         */
+        /* Nothing to include for now... this is just a dummy wrapper around
+         * all the other channels in the DopeSheet, and gets included at the start of the list. */
         ale->key_data = NULL;
         ale->datatype = ALE_ALL;
         break;
diff --git a/source/blender/editors/animation/anim_ipo_utils.c b/source/blender/editors/animation/anim_ipo_utils.c
index fad9a1a8e49..5b729c856c0 100644
--- a/source/blender/editors/animation/anim_ipo_utils.c
+++ b/source/blender/editors/animation/anim_ipo_utils.c
@@ -122,7 +122,7 @@ int getname_anim_fcurve(char *name, ID *id, FCurve *fcu)
           MEM_freeN(constName);
         }
       }
-      else if (ptr.data != ptr.id.data) {
+      else if (ptr.data != ptr.owner_id) {
         PropertyRNA *nameprop = RNA_struct_name_property(ptr.type);
         if (nameprop) {
           /* this gets a string which will need to be freed */
diff --git a/source/blender/editors/animation/anim_markers.c b/source/blender/editors/animation/anim_markers.c
index ded59466370..8c873eb6b45 100644
--- a/source/blender/editors/animation/anim_markers.c
+++ b/source/blender/editors/animation/anim_markers.c
@@ -486,7 +486,7 @@ static void draw_marker(
   float name_y = UI_DPI_FAC * 18;
   /* Give an offset to the marker name when selected,
    * or when near the current frame (5 frames range, starting from the current one). */
-  if ((marker->flag & SELECT) || (IN_RANGE_INCL(marker->frame, cfra, cfra - 4))) {
+  if ((marker->flag & SELECT) || (IN_RANGE_INCL(marker->frame, cfra - 4, cfra))) {
     name_y += UI_DPI_FAC * 10;
   }
   draw_marker_name(fstyle, marker, xpos, name_y);
diff --git a/source/blender/editors/animation/anim_motion_paths.c b/source/blender/editors/animation/anim_motion_paths.c
index 7a5b57b1ce6..bd4886817cd 100644
--- a/source/blender/editors/animation/anim_motion_paths.c
+++ b/source/blender/editors/animation/anim_motion_paths.c
@@ -60,9 +60,8 @@ typedef struct MPathTarget {
   Object *ob;          /* source object */
   bPoseChannel *pchan; /* source posechannel (if applicable) */
 
-  /* "Evaluated" Copies (these come from the background COW copie
-   * that provide all the coordinates we want to save off)
-   */
+  /* "Evaluated" Copies (these come from the background COW copy
+   * that provide all the coordinates we want to save off). */
   Object *ob_eval; /* evaluated object */
 } MPathTarget;
 
diff --git a/source/blender/editors/animation/drivers.c b/source/blender/editors/animation/drivers.c
index 935d11a388f..7b9e6a10f44 100644
--- a/source/blender/editors/animation/drivers.c
+++ b/source/blender/editors/animation/drivers.c
@@ -218,7 +218,7 @@ static int add_driver_with_target(ReportList *UNUSED(reports),
 
     /* Create a driver variable for the target
      *   - For transform properties, we want to automatically use "transform channel" instead
-     *     (The only issue is with quat rotations vs euler channels...)
+     *     (The only issue is with quaternion rotations vs euler channels...)
      *   - To avoid problems with transform properties depending on the final transform that they
      *     control (thus creating pseudo-cycles - see T48734), we don't use transform channels
      *     when both the source and destinations are in same places.
@@ -934,7 +934,7 @@ static const EnumPropertyItem *driver_mapping_type_itemsf(bContext *C,
   EnumPropertyItem *input = prop_driver_create_mapping_types;
   EnumPropertyItem *item = NULL;
 
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   int index;
 
@@ -946,7 +946,7 @@ static const EnumPropertyItem *driver_mapping_type_itemsf(bContext *C,
 
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (ptr.id.data && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
+  if (ptr.owner_id && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
     const bool is_array = RNA_property_array_check(prop);
 
     while (input->identifier) {
@@ -971,7 +971,7 @@ static const EnumPropertyItem *driver_mapping_type_itemsf(bContext *C,
 
 static bool add_driver_button_poll(bContext *C)
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   int index;
   bool driven, special;
@@ -979,7 +979,7 @@ static bool add_driver_button_poll(bContext *C)
   /* this operator can only run if there's a property button active, and it can be animated */
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (!(ptr.id.data && ptr.data && prop)) {
+  if (!(ptr.owner_id && ptr.data && prop)) {
     return false;
   }
   if (!RNA_property_animateable(&ptr, prop)) {
@@ -995,7 +995,7 @@ static bool add_driver_button_poll(bContext *C)
  * (i.e. "manual/add later"). */
 static int add_driver_button_none(bContext *C, wmOperator *op, short mapping_type)
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   int index;
   int success = 0;
@@ -1006,12 +1006,13 @@ static int add_driver_button_none(bContext *C, wmOperator *op, short mapping_typ
     index = -1;
   }
 
-  if (ptr.id.data && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
+  if (ptr.owner_id && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
     char *path = BKE_animdata_driver_path_hack(C, &ptr, prop, NULL);
     short flags = CREATEDRIVER_WITH_DEFAULT_DVAR;
 
     if (path) {
-      success += ANIM_add_driver(op->reports, ptr.id.data, path, index, flags, DRIVER_TYPE_PYTHON);
+      success += ANIM_add_driver(
+          op->reports, ptr.owner_id, path, index, flags, DRIVER_TYPE_PYTHON);
       MEM_freeN(path);
     }
   }
@@ -1095,28 +1096,29 @@ static void UNUSED_FUNCTION(ANIM_OT_driver_button_add_menu)(wmOperatorType *ot)
 
 static int add_driver_button_invoke(bContext *C, wmOperator *op, const wmEvent *UNUSED(event))
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   int index;
 
   /* try to find driver using property retrieved from UI */
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (ptr.id.data && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
+  if (ptr.owner_id && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
     /* 1) Create a new "empty" driver for this property */
     char *path = BKE_animdata_driver_path_hack(C, &ptr, prop, NULL);
     short flags = CREATEDRIVER_WITH_DEFAULT_DVAR;
     short success = 0;
 
     if (path) {
-      success += ANIM_add_driver(op->reports, ptr.id.data, path, index, flags, DRIVER_TYPE_PYTHON);
+      success += ANIM_add_driver(
+          op->reports, ptr.owner_id, path, index, flags, DRIVER_TYPE_PYTHON);
       MEM_freeN(path);
     }
 
     if (success) {
       /* send updates */
       UI_context_update_anim_flag(C);
-      DEG_id_tag_update(ptr.id.data, ID_RECALC_COPY_ON_WRITE);
+      DEG_id_tag_update(ptr.owner_id, ID_RECALC_COPY_ON_WRITE);
       DEG_relations_tag_update(CTX_data_main(C));
       WM_event_add_notifier(C, NC_ANIMATION | ND_FCURVES_ORDER, NULL);
     }
@@ -1149,7 +1151,7 @@ void ANIM_OT_driver_button_add(wmOperatorType *ot)
 
 static int remove_driver_button_exec(bContext *C, wmOperator *op)
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   short success = 0;
   int index;
@@ -1162,11 +1164,11 @@ static int remove_driver_button_exec(bContext *C, wmOperator *op)
     index = -1;
   }
 
-  if (ptr.id.data && ptr.data && prop) {
+  if (ptr.owner_id && ptr.data && prop) {
     char *path = BKE_animdata_driver_path_hack(C, &ptr, prop, NULL);
 
     if (path) {
-      success = ANIM_remove_driver(op->reports, ptr.id.data, path, index, 0);
+      success = ANIM_remove_driver(op->reports, ptr.owner_id, path, index, 0);
 
       MEM_freeN(path);
     }
@@ -1205,14 +1207,14 @@ void ANIM_OT_driver_button_remove(wmOperatorType *ot)
 
 static int edit_driver_button_exec(bContext *C, wmOperator *op)
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   int index;
 
   /* try to find driver using property retrieved from UI */
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (ptr.id.data && ptr.data && prop) {
+  if (ptr.owner_id && ptr.data && prop) {
     UI_popover_panel_invoke(C, "GRAPH_PT_drivers_popover", true, op->reports);
   }
 
@@ -1239,7 +1241,7 @@ void ANIM_OT_driver_button_edit(wmOperatorType *ot)
 
 static int copy_driver_button_exec(bContext *C, wmOperator *op)
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   short success = 0;
   int index;
@@ -1247,12 +1249,12 @@ static int copy_driver_button_exec(bContext *C, wmOperator *op)
   /* try to create driver using property retrieved from UI */
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (ptr.id.data && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
+  if (ptr.owner_id && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
     char *path = BKE_animdata_driver_path_hack(C, &ptr, prop, NULL);
 
     if (path) {
       /* only copy the driver for the button that this was involved for */
-      success = ANIM_copy_driver(op->reports, ptr.id.data, path, index, 0);
+      success = ANIM_copy_driver(op->reports, ptr.owner_id, path, index, 0);
 
       UI_context_update_anim_flag(C);
 
@@ -1283,7 +1285,7 @@ void ANIM_OT_copy_driver_button(wmOperatorType *ot)
 
 static int paste_driver_button_exec(bContext *C, wmOperator *op)
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   short success = 0;
   int index;
@@ -1291,18 +1293,18 @@ static int paste_driver_button_exec(bContext *C, wmOperator *op)
   /* try to create driver using property retrieved from UI */
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (ptr.id.data && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
+  if (ptr.owner_id && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
     char *path = BKE_animdata_driver_path_hack(C, &ptr, prop, NULL);
 
     if (path) {
       /* only copy the driver for the button that this was involved for */
-      success = ANIM_paste_driver(op->reports, ptr.id.data, path, index, 0);
+      success = ANIM_paste_driver(op->reports, ptr.owner_id, path, index, 0);
 
       UI_context_update_anim_flag(C);
 
       DEG_relations_tag_update(CTX_data_main(C));
 
-      DEG_id_tag_update(ptr.id.data, ID_RECALC_ANIMATION);
+      DEG_id_tag_update(ptr.owner_id, ID_RECALC_ANIMATION);
 
       WM_event_add_notifier(C, NC_ANIMATION | ND_KEYFRAME_PROP, NULL);  // XXX
 
diff --git a/source/blender/editors/animation/fmodifier_ui.c b/source/blender/editors/animation/fmodifier_ui.c
index b42e8102c5b..705351522f8 100644
--- a/source/blender/editors/animation/fmodifier_ui.c
+++ b/source/blender/editors/animation/fmodifier_ui.c
@@ -197,7 +197,7 @@ static void draw_modifier__generator(uiLayout *layout,
           &data->poly_order,
           1,
           100,
-          0,
+          1,
           0,
           TIP_("'Order' of the Polynomial (for a polynomial with n terms, 'order' is n-1)"));
       UI_but_func_set(but, validate_fmodifier_cb, fcm, fcurve_owner_id);
@@ -335,7 +335,7 @@ static void draw_modifier__generator(uiLayout *layout,
           &data->poly_order,
           1,
           100,
-          0,
+          1,
           0,
           TIP_("'Order' of the Polynomial (for a polynomial with n terms, 'order' is n-1)"));
       UI_but_func_set(but, validate_fmodifier_cb, fcm, fcurve_owner_id);
diff --git a/source/blender/editors/animation/keyframes_draw.c b/source/blender/editors/animation/keyframes_draw.c
index 889d27480df..ca7e0eae136 100644
--- a/source/blender/editors/animation/keyframes_draw.c
+++ b/source/blender/editors/animation/keyframes_draw.c
@@ -814,8 +814,10 @@ static void draw_keylist(View2D *v2d,
       uint outline_color_id = GPU_vertformat_attr_add(
           format, "outlineColor", GPU_COMP_U8, 4, GPU_FETCH_INT_TO_FLOAT_UNIT);
       uint flags_id = GPU_vertformat_attr_add(format, "flags", GPU_COMP_U32, 1, GPU_FETCH_INT);
-      immBindBuiltinProgram(GPU_SHADER_KEYFRAME_DIAMOND);
+
       GPU_program_point_size(true);
+      immBindBuiltinProgram(GPU_SHADER_KEYFRAME_DIAMOND);
+      immUniform1f("outline_scale", 1.0f);
       immUniform2f(
           "ViewportSize", BLI_rcti_size_x(&v2d->mask) + 1, BLI_rcti_size_y(&v2d->mask) + 1);
       immBegin(GPU_PRIM_POINTS, key_len);
diff --git a/source/blender/editors/animation/keyframing.c b/source/blender/editors/animation/keyframing.c
index dcc596e67e1..0f8b8742659 100644
--- a/source/blender/editors/animation/keyframing.c
+++ b/source/blender/editors/animation/keyframing.c
@@ -222,7 +222,7 @@ FCurve *verify_fcurve(Main *bmain,
 
         /* sync bone group colors if applicable */
         if (ptr && (ptr->type == &RNA_PoseBone)) {
-          Object *ob = (Object *)ptr->id.data;
+          Object *ob = (Object *)ptr->owner_id;
           bPoseChannel *pchan = (bPoseChannel *)ptr->data;
           bPose *pose = ob->pose;
           bActionGroup *grp;
@@ -286,7 +286,7 @@ void update_autoflags_fcurve(FCurve *fcu, bContext *C, ReportList *reports, Poin
   PropertyRNA *prop;
   int old_flag = fcu->flag;
 
-  if ((ptr->id.data == NULL) && (ptr->data == NULL)) {
+  if ((ptr->owner_id == NULL) && (ptr->data == NULL)) {
     BKE_report(reports, RPT_ERROR, "No RNA pointer available to retrieve values for this fcurve");
     return;
   }
@@ -294,7 +294,7 @@ void update_autoflags_fcurve(FCurve *fcu, bContext *C, ReportList *reports, Poin
   /* try to get property we should be affecting */
   if (RNA_path_resolve_property(ptr, fcu->rna_path, &tmp_ptr, &prop) == false) {
     /* property not found... */
-    const char *idname = (ptr->id.data) ? ((ID *)ptr->id.data)->name : TIP_("<No ID pointer>");
+    const char *idname = (ptr->owner_id) ? ptr->owner_id->name : TIP_("<No ID pointer>");
 
     BKE_reportf(reports,
                 RPT_ERROR,
@@ -1203,7 +1203,7 @@ bool insert_keyframe_direct(ReportList *reports,
   }
 
   /* if no property given yet, try to validate from F-Curve info */
-  if ((ptr.id.data == NULL) && (ptr.data == NULL)) {
+  if ((ptr.owner_id == NULL) && (ptr.data == NULL)) {
     BKE_report(
         reports, RPT_ERROR, "No RNA pointer available to retrieve values for keyframing from");
     return false;
@@ -1214,7 +1214,7 @@ bool insert_keyframe_direct(ReportList *reports,
     /* try to get property we should be affecting */
     if (RNA_path_resolve_property(&ptr, fcu->rna_path, &tmp_ptr, &prop) == false) {
       /* property not found... */
-      const char *idname = (ptr.id.data) ? ((ID *)ptr.id.data)->name : TIP_("<No ID pointer>");
+      const char *idname = (ptr.owner_id) ? ptr.owner_id->name : TIP_("<No ID pointer>");
 
       BKE_reportf(reports,
                   RPT_ERROR,
@@ -1544,9 +1544,9 @@ static bool delete_keyframe_fcurve(AnimData *adt, FCurve *fcu, float cfra)
 static void deg_tag_after_keyframe_delete(Main *bmain, ID *id, AnimData *adt)
 {
   if (adt->action == NULL) {
-    /* In the case last f-curve wes removed need to inform dependency graph
+    /* In the case last f-curve was removed need to inform dependency graph
      * about relations update, since it needs to get rid of animation operation
-     * for this datablock. */
+     * for this data-block. */
     DEG_id_tag_update_ex(bmain, id, ID_RECALC_ANIMATION_NO_FLUSH);
     DEG_relations_tag_update(bmain);
   }
@@ -2350,7 +2350,7 @@ static int insert_key_button_exec(bContext *C, wmOperator *op)
   Main *bmain = CTX_data_main(C);
   Scene *scene = CTX_data_scene(C);
   ToolSettings *ts = scene->toolsettings;
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   char *path;
   uiBut *but;
@@ -2369,7 +2369,7 @@ static int insert_key_button_exec(bContext *C, wmOperator *op)
     return (OPERATOR_CANCELLED | OPERATOR_PASS_THROUGH);
   }
 
-  if ((ptr.id.data && ptr.data && prop) && RNA_property_animateable(&ptr, prop)) {
+  if ((ptr.owner_id && ptr.data && prop) && RNA_property_animateable(&ptr, prop)) {
     if (ptr.type == &RNA_NlaStrip) {
       /* Handle special properties for NLA Strips, whose F-Curves are stored on the
        * strips themselves. These are stored separately or else the properties will
@@ -2435,7 +2435,7 @@ static int insert_key_button_exec(bContext *C, wmOperator *op)
 
         success = insert_keyframe(bmain,
                                   op->reports,
-                                  ptr.id.data,
+                                  ptr.owner_id,
                                   NULL,
                                   group,
                                   path,
@@ -2473,7 +2473,7 @@ static int insert_key_button_exec(bContext *C, wmOperator *op)
   }
 
   if (success) {
-    ID *id = ptr.id.data;
+    ID *id = ptr.owner_id;
     AnimData *adt = BKE_animdata_from_id(id);
     if (adt->action != NULL) {
       DEG_id_tag_update(&adt->action->id, ID_RECALC_ANIMATION_NO_FLUSH);
@@ -2513,7 +2513,7 @@ void ANIM_OT_keyframe_insert_button(wmOperatorType *ot)
 static int delete_key_button_exec(bContext *C, wmOperator *op)
 {
   Scene *scene = CTX_data_scene(C);
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   Main *bmain = CTX_data_main(C);
   char *path;
@@ -2528,13 +2528,13 @@ static int delete_key_button_exec(bContext *C, wmOperator *op)
     return (OPERATOR_CANCELLED | OPERATOR_PASS_THROUGH);
   }
 
-  if (ptr.id.data && ptr.data && prop) {
+  if (ptr.owner_id && ptr.data && prop) {
     if (BKE_nlastrip_has_curves_for_property(&ptr, prop)) {
       /* Handle special properties for NLA Strips, whose F-Curves are stored on the
        * strips themselves. These are stored separately or else the properties will
        * not have any effect.
        */
-      ID *id = ptr.id.data;
+      ID *id = ptr.owner_id;
       NlaStrip *strip = (NlaStrip *)ptr.data;
       FCurve *fcu = list_find_fcurve(&strip->fcurves, RNA_property_identifier(prop), 0);
 
@@ -2577,7 +2577,7 @@ static int delete_key_button_exec(bContext *C, wmOperator *op)
         }
 
         success = delete_keyframe(
-            bmain, op->reports, ptr.id.data, NULL, NULL, path, index, cfra, 0);
+            bmain, op->reports, ptr.owner_id, NULL, NULL, path, index, cfra, 0);
         MEM_freeN(path);
       }
       else if (G.debug & G_DEBUG) {
@@ -2622,7 +2622,7 @@ void ANIM_OT_keyframe_delete_button(wmOperatorType *ot)
 
 static int clear_key_button_exec(bContext *C, wmOperator *op)
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   Main *bmain = CTX_data_main(C);
   char *path;
@@ -2636,7 +2636,7 @@ static int clear_key_button_exec(bContext *C, wmOperator *op)
     return (OPERATOR_CANCELLED | OPERATOR_PASS_THROUGH);
   }
 
-  if (ptr.id.data && ptr.data && prop) {
+  if (ptr.owner_id && ptr.data && prop) {
     path = RNA_path_from_ID_to_property(&ptr, prop);
 
     if (path) {
@@ -2645,7 +2645,7 @@ static int clear_key_button_exec(bContext *C, wmOperator *op)
         index = -1;
       }
 
-      success += clear_keyframe(bmain, op->reports, ptr.id.data, NULL, NULL, path, index, 0);
+      success += clear_keyframe(bmain, op->reports, ptr.owner_id, NULL, NULL, path, index, 0);
       MEM_freeN(path);
     }
     else if (G.debug & G_DEBUG) {
diff --git a/source/blender/editors/animation/keyingsets.c b/source/blender/editors/animation/keyingsets.c
index ccd0fc54611..258c0e4f1f6 100644
--- a/source/blender/editors/animation/keyingsets.c
+++ b/source/blender/editors/animation/keyingsets.c
@@ -287,7 +287,7 @@ static int add_keyingset_button_exec(bContext *C, wmOperator *op)
   Scene *scene = CTX_data_scene(C);
   KeyingSet *ks = NULL;
   PropertyRNA *prop = NULL;
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   char *path = NULL;
   short success = 0;
   int index = 0, pflag = 0;
@@ -332,7 +332,7 @@ static int add_keyingset_button_exec(bContext *C, wmOperator *op)
   }
 
   /* check if property is able to be added */
-  if (ptr.id.data && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
+  if (ptr.owner_id && ptr.data && prop && RNA_property_animateable(&ptr, prop)) {
     path = RNA_path_from_ID_to_property(&ptr, prop);
 
     if (path) {
@@ -348,7 +348,7 @@ static int add_keyingset_button_exec(bContext *C, wmOperator *op)
       }
 
       /* add path to this setting */
-      BKE_keyingset_add_path(ks, ptr.id.data, NULL, path, index, pflag, KSP_GROUP_KSNAME);
+      BKE_keyingset_add_path(ks, ptr.owner_id, NULL, path, index, pflag, KSP_GROUP_KSNAME);
       ks->active_path = BLI_listbase_count(&ks->paths);
       success = 1;
 
@@ -393,7 +393,7 @@ static int remove_keyingset_button_exec(bContext *C, wmOperator *op)
   Scene *scene = CTX_data_scene(C);
   KeyingSet *ks = NULL;
   PropertyRNA *prop = NULL;
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   char *path = NULL;
   short success = 0;
   int index = 0;
@@ -420,14 +420,14 @@ static int remove_keyingset_button_exec(bContext *C, wmOperator *op)
     ks = BLI_findlink(&scene->keyingsets, scene->active_keyingset - 1);
   }
 
-  if (ptr.id.data && ptr.data && prop) {
+  if (ptr.owner_id && ptr.data && prop) {
     path = RNA_path_from_ID_to_property(&ptr, prop);
 
     if (path) {
       KS_Path *ksp;
 
       /* try to find a path matching this description */
-      ksp = BKE_keyingset_find_path(ks, ptr.id.data, ks->name, path, index, KSP_GROUP_KSNAME);
+      ksp = BKE_keyingset_find_path(ks, ptr.owner_id, ks->name, path, index, KSP_GROUP_KSNAME);
 
       if (ksp) {
         BKE_keyingset_free_path(ks, ksp);
@@ -712,9 +712,9 @@ int ANIM_scene_get_keyingset_index(Scene *scene, KeyingSet *ks)
     }
   }
 
-  /* still here, so try builtins list too
-   * - builtins are from (<= -1)
-   * - none/invalid is (= 0)
+  /* Still here, so try built-ins list too:
+   * - Built-ins are from (<= -1).
+   * - None/Invalid is (= 0).
    */
   index = BLI_findindex(&builtin_keyingsets, ks);
   if (index != -1) {
diff --git a/source/blender/editors/armature/armature_add.c b/source/blender/editors/armature/armature_add.c
index d2fa77f90be..1073034383d 100644
--- a/source/blender/editors/armature/armature_add.c
+++ b/source/blender/editors/armature/armature_add.c
@@ -769,7 +769,7 @@ static int armature_symmetrize_exec(bContext *C, wmOperator *op)
 
         EditBone *ebone = ebone_iter->temp.ebone;
 
-        /* copy flags incase bone is pre-existing data */
+        /* Copy flags in case bone is pre-existing data. */
         ebone->flag = (ebone->flag & ~flag_copy) | (ebone_iter->flag & flag_copy);
 
         if (ebone_iter->parent == NULL) {
diff --git a/source/blender/editors/armature/armature_edit.c b/source/blender/editors/armature/armature_edit.c
index 4e6661b1d15..c4c10549da3 100644
--- a/source/blender/editors/armature/armature_edit.c
+++ b/source/blender/editors/armature/armature_edit.c
@@ -62,22 +62,10 @@
 /* ************************** Object Tools Exports ******************************* */
 /* NOTE: these functions are exported to the Object module to be called from the tools there */
 
-void ED_armature_transform_apply(Main *bmain, Object *ob, float mat[4][4], const bool do_props)
-{
-  bArmature *arm = ob->data;
-
-  /* Put the armature into editmode */
-  ED_armature_to_edit(arm);
-
-  /* Transform the bones */
-  ED_armature_transform_bones(arm, mat, do_props);
-
-  /* Turn the list into an armature */
-  ED_armature_from_edit(bmain, arm);
-  ED_armature_edit_free(arm);
-}
-
-void ED_armature_transform_bones(struct bArmature *arm, float mat[4][4], const bool do_props)
+/**
+ * See #BKE_armature_transform for object-mode transform.
+ */
+void ED_armature_edit_transform(bArmature *arm, const float mat[4][4], const bool do_props)
 {
   EditBone *ebone;
   float scale = mat4_to_scale(mat); /* store the scale of the matrix here to use on envelopes */
@@ -114,21 +102,13 @@ void ED_armature_transform_bones(struct bArmature *arm, float mat[4][4], const b
   }
 }
 
-void ED_armature_transform(Main *bmain, bArmature *arm, float mat[4][4], const bool do_props)
+void ED_armature_transform(bArmature *arm, const float mat[4][4], const bool do_props)
 {
   if (arm->edbo) {
-    ED_armature_transform_bones(arm, mat, do_props);
+    ED_armature_edit_transform(arm, mat, do_props);
   }
   else {
-    /* Put the armature into editmode */
-    ED_armature_to_edit(arm);
-
-    /* Transform the bones */
-    ED_armature_transform_bones(arm, mat, do_props);
-
-    /* Go back to object mode*/
-    ED_armature_from_edit(bmain, arm);
-    ED_armature_edit_free(arm);
+    BKE_armature_transform(arm, mat, do_props);
   }
 }
 
diff --git a/source/blender/editors/armature/armature_intern.h b/source/blender/editors/armature/armature_intern.h
index 569eb7e2e04..fa562ab0f44 100644
--- a/source/blender/editors/armature/armature_intern.h
+++ b/source/blender/editors/armature/armature_intern.h
@@ -30,7 +30,6 @@ struct wmOperatorType;
 struct Base;
 struct Object;
 struct Scene;
-struct bAction;
 struct bContext;
 struct bPoseChannel;
 
diff --git a/source/blender/editors/armature/armature_select.c b/source/blender/editors/armature/armature_select.c
index 23ddf77e63d..eff621d7b71 100644
--- a/source/blender/editors/armature/armature_select.c
+++ b/source/blender/editors/armature/armature_select.c
@@ -46,6 +46,7 @@
 
 #include "ED_armature.h"
 #include "ED_object.h"
+#include "ED_outliner.h"
 #include "ED_screen.h"
 #include "ED_select_utils.h"
 #include "ED_view3d.h"
@@ -356,6 +357,8 @@ static int armature_select_linked_invoke(bContext *C, wmOperator *op, const wmEv
     }
   }
 
+  ED_outliner_select_sync_from_edit_bone_tag(C);
+
   ED_armature_edit_sync_selection(arm->edbo);
 
   WM_event_add_notifier(C, NC_OBJECT | ND_BONE_SELECT, base->object);
@@ -1027,6 +1030,8 @@ static int armature_de_select_all_exec(bContext *C, wmOperator *op)
   }
   CTX_DATA_END;
 
+  ED_outliner_select_sync_from_edit_bone_tag(C);
+
   WM_event_add_notifier(C, NC_OBJECT | ND_BONE_SELECT, NULL);
 
   return OPERATOR_FINISHED;
@@ -1148,6 +1153,8 @@ static int armature_de_select_more_exec(bContext *C, wmOperator *UNUSED(op))
     WM_event_add_notifier(C, NC_OBJECT | ND_BONE_SELECT, ob);
   }
   MEM_freeN(objects);
+
+  ED_outliner_select_sync_from_edit_bone_tag(C);
   return OPERATOR_FINISHED;
 }
 
@@ -1178,6 +1185,8 @@ static int armature_de_select_less_exec(bContext *C, wmOperator *UNUSED(op))
     WM_event_add_notifier(C, NC_OBJECT | ND_BONE_SELECT, ob);
   }
   MEM_freeN(objects);
+
+  ED_outliner_select_sync_from_edit_bone_tag(C);
   return OPERATOR_FINISHED;
 }
 
@@ -1569,6 +1578,8 @@ static int armature_select_similar_exec(bContext *C, wmOperator *op)
 
 #undef STRUCT_SIZE_AND_OFFSET
 
+  ED_outliner_select_sync_from_edit_bone_tag(C);
+
   return OPERATOR_FINISHED;
 }
 
@@ -1663,6 +1674,8 @@ static int armature_select_hierarchy_exec(bContext *C, wmOperator *op)
     return OPERATOR_CANCELLED;
   }
 
+  ED_outliner_select_sync_from_edit_bone_tag(C);
+
   ED_armature_edit_sync_selection(arm->edbo);
 
   WM_event_add_notifier(C, NC_OBJECT | ND_BONE_SELECT, ob);
@@ -1748,6 +1761,8 @@ static int armature_select_mirror_exec(bContext *C, wmOperator *op)
       arm->act_edbone = ebone_mirror_act;
     }
 
+    ED_outliner_select_sync_from_edit_bone_tag(C);
+
     ED_armature_edit_sync_selection(arm->edbo);
 
     WM_event_add_notifier(C, NC_OBJECT | ND_BONE_SELECT, ob);
@@ -1876,6 +1891,7 @@ static int armature_shortest_path_pick_invoke(bContext *C, wmOperator *op, const
 
   if (changed) {
     arm->act_edbone = ebone_dst;
+    ED_outliner_select_sync_from_edit_bone_tag(C);
     ED_armature_edit_sync_selection(arm->edbo);
     WM_event_add_notifier(C, NC_OBJECT | ND_BONE_SELECT, obedit);
 
diff --git a/source/blender/editors/armature/armature_utils.c b/source/blender/editors/armature/armature_utils.c
index d8777b7e0b7..cd299906b4c 100644
--- a/source/blender/editors/armature/armature_utils.c
+++ b/source/blender/editors/armature/armature_utils.c
@@ -381,23 +381,37 @@ void armature_tag_unselect(bArmature *arm)
 
 void ED_armature_ebone_transform_mirror_update(bArmature *arm, EditBone *ebo, bool check_select)
 {
+  /* TODO When this function is called by property updates,
+   * cancelling the value change will not restore mirrored bone correctly. */
+
+  /* Currently check_select==true when this function is called from a transform operator,
+   * eg. from 3d viewport. */
+
   /* no layer check, correct mirror is more important */
   if (!check_select || ebo->flag & (BONE_TIPSEL | BONE_ROOTSEL)) {
     EditBone *eboflip = ED_armature_ebone_get_mirrored(arm->edbo, ebo);
     if (eboflip) {
-      /* we assume X-axis flipping for now */
-      if (check_select && ebo->flag & BONE_TIPSEL) {
-        EditBone *children;
+      /* We assume X-axis flipping for now. */
+
+      /* Always mirror roll, since it can be changed by moving either head or tail. */
+      eboflip->roll = -ebo->roll;
+
+      if (!check_select || ebo->flag & BONE_TIPSEL) {
+        /* Mirror tail properties. */
 
         eboflip->tail[0] = -ebo->tail[0];
         eboflip->tail[1] = ebo->tail[1];
         eboflip->tail[2] = ebo->tail[2];
         eboflip->rad_tail = ebo->rad_tail;
-        eboflip->roll = -ebo->roll;
         eboflip->curve_out_x = -ebo->curve_out_x;
+        eboflip->curve_out_y = ebo->curve_out_y;
+        eboflip->scale_out_x = ebo->scale_out_x;
+        eboflip->scale_out_y = ebo->scale_out_y;
+        eboflip->ease2 = ebo->ease2;
         eboflip->roll2 = -ebo->roll2;
 
-        /* Also move connected children, in case children's name aren't mirrored properly */
+        /* Also move connected children, in case children's name aren't mirrored properly. */
+        EditBone *children;
         for (children = arm->edbo->first; children; children = children->next) {
           if (children->parent == eboflip && children->flag & BONE_CONNECTED) {
             copy_v3_v3(children->head, eboflip->tail);
@@ -405,32 +419,39 @@ void ED_armature_ebone_transform_mirror_update(bArmature *arm, EditBone *ebo, bo
           }
         }
       }
+
       if (!check_select || ebo->flag & BONE_ROOTSEL) {
+        /* Mirror head properties. */
         eboflip->head[0] = -ebo->head[0];
         eboflip->head[1] = ebo->head[1];
         eboflip->head[2] = ebo->head[2];
         eboflip->rad_head = ebo->rad_head;
-        eboflip->roll = -ebo->roll;
+
         eboflip->curve_in_x = -ebo->curve_in_x;
+        eboflip->curve_in_y = ebo->curve_in_y;
+        eboflip->scale_in_x = ebo->scale_in_x;
+        eboflip->scale_in_y = ebo->scale_in_y;
+        eboflip->ease1 = ebo->ease1;
         eboflip->roll1 = -ebo->roll1;
 
-        /* Also move connected parent, in case parent's name isn't mirrored properly */
+        /* Also move connected parent, in case parent's name isn't mirrored properly. */
         if (eboflip->parent && eboflip->flag & BONE_CONNECTED) {
           EditBone *parent = eboflip->parent;
           copy_v3_v3(parent->tail, eboflip->head);
           parent->rad_tail = ebo->rad_head;
         }
       }
+
       if (!check_select || ebo->flag & BONE_SELECTED) {
+        /* Mirror bone body properties (both head and tail are selected). */
+        /* TODO: These values can also be changed from pose mode,
+         * so only mirroring them in edit mode is not ideal. */
         eboflip->dist = ebo->dist;
-        eboflip->roll = -ebo->roll;
+        eboflip->weight = ebo->weight;
+
+        eboflip->segments = ebo->segments;
         eboflip->xwidth = ebo->xwidth;
         eboflip->zwidth = ebo->zwidth;
-
-        eboflip->curve_in_x = -ebo->curve_in_x;
-        eboflip->curve_out_x = -ebo->curve_out_x;
-        eboflip->roll1 = -ebo->roll1;
-        eboflip->roll2 = -ebo->roll2;
       }
     }
   }
diff --git a/source/blender/editors/armature/pose_select.c b/source/blender/editors/armature/pose_select.c
index 8434fee6e78..a59067e60c1 100644
--- a/source/blender/editors/armature/pose_select.c
+++ b/source/blender/editors/armature/pose_select.c
@@ -54,6 +54,7 @@
 #include "ED_keyframing.h"
 #include "ED_mesh.h"
 #include "ED_object.h"
+#include "ED_outliner.h"
 #include "ED_screen.h"
 #include "ED_select_utils.h"
 #include "ED_view3d.h"
@@ -449,6 +450,8 @@ static int pose_select_connected_invoke(bContext *C, wmOperator *op, const wmEve
     selectconnected_posebonechildren(base->object, curBone, extend);
   }
 
+  ED_outliner_select_sync_from_pose_bone_tag(C);
+
   ED_pose_bone_select_tag_update(base->object);
 
   return OPERATOR_FINISHED;
@@ -514,6 +517,8 @@ static int pose_de_select_all_exec(bContext *C, wmOperator *op)
   }
   CTX_DATA_END;
 
+  ED_outliner_select_sync_from_pose_bone_tag(C);
+
   WM_event_add_notifier(C, NC_OBJECT | ND_BONE_SELECT, NULL);
 
   return OPERATOR_FINISHED;
@@ -560,6 +565,8 @@ static int pose_select_parent_exec(bContext *C, wmOperator *UNUSED(op))
     return OPERATOR_CANCELLED;
   }
 
+  ED_outliner_select_sync_from_pose_bone_tag(C);
+
   ED_pose_bone_select_tag_update(ob);
   return OPERATOR_FINISHED;
 }
@@ -624,6 +631,8 @@ static int pose_select_constraint_target_exec(bContext *C, wmOperator *UNUSED(op
     return OPERATOR_CANCELLED;
   }
 
+  ED_outliner_select_sync_from_pose_bone_tag(C);
+
   return OPERATOR_FINISHED;
 }
 
@@ -712,6 +721,8 @@ static int pose_select_hierarchy_exec(bContext *C, wmOperator *op)
     return OPERATOR_CANCELLED;
   }
 
+  ED_outliner_select_sync_from_pose_bone_tag(C);
+
   ED_pose_bone_select_tag_update(ob);
 
   return OPERATOR_FINISHED;
@@ -796,7 +807,7 @@ static bool pose_select_same_group(bContext *C, bool extend)
   group_flags = NULL;
   ob_index = -1;
   ob_prev = NULL;
-  CTX_DATA_BEGIN_WITH_ID (C, bPoseChannel *, pchan, visible_pose_bones, Object, *ob) {
+  CTX_DATA_BEGIN_WITH_ID (C, bPoseChannel *, pchan, visible_pose_bones, Object *, ob) {
     if (ob != ob_prev) {
       ob_index++;
       group_flags = group_flags_array + (ob_index * groups_len);
@@ -1061,6 +1072,8 @@ static int pose_select_grouped_exec(bContext *C, wmOperator *op)
 
   /* report done status */
   if (changed) {
+    ED_outliner_select_sync_from_pose_bone_tag(C);
+
     return OPERATOR_FINISHED;
   }
   else {
@@ -1172,6 +1185,8 @@ static int pose_select_mirror_exec(bContext *C, wmOperator *op)
   }
   MEM_freeN(objects);
 
+  ED_outliner_select_sync_from_pose_bone_tag(C);
+
   return OPERATOR_FINISHED;
 }
 
diff --git a/source/blender/editors/armature/pose_slide.c b/source/blender/editors/armature/pose_slide.c
index c8e79b879a4..6274eb549da 100644
--- a/source/blender/editors/armature/pose_slide.c
+++ b/source/blender/editors/armature/pose_slide.c
@@ -442,7 +442,7 @@ static void pose_slide_apply_props(tPoseSlideOp *pso,
                                    tPChanFCurveLink *pfl,
                                    const char prop_prefix[])
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   LinkData *ld;
   int len = strlen(pfl->pchan_path);
 
diff --git a/source/blender/editors/curve/editcurve.c b/source/blender/editors/curve/editcurve.c
index c93531bb6cc..d7650db546d 100644
--- a/source/blender/editors/curve/editcurve.c
+++ b/source/blender/editors/curve/editcurve.c
@@ -51,6 +51,7 @@
 
 #include "DEG_depsgraph.h"
 #include "DEG_depsgraph_build.h"
+#include "DEG_depsgraph_query.h"
 
 #include "WM_api.h"
 #include "WM_types.h"
@@ -5679,6 +5680,7 @@ static int add_vertex_invoke(bContext *C, wmOperator *op, const wmEvent *event)
           },
           mval,
           NULL,
+          NULL,
           location,
           NULL);
 
@@ -7104,14 +7106,15 @@ static int match_texture_space_exec(bContext *C, wmOperator *UNUSED(op))
   (void)depsgraph;
 
   Object *object = CTX_data_active_object(C);
+  Object *object_eval = DEG_get_evaluated_object(depsgraph, object);
   Curve *curve = (Curve *)object->data;
   float min[3], max[3], size[3], loc[3];
   int a;
 
-  BLI_assert(object->runtime.curve_cache != NULL);
+  BLI_assert(object_eval->runtime.curve_cache != NULL);
 
   INIT_MINMAX(min, max);
-  BKE_displist_minmax(&object->runtime.curve_cache->disp, min, max);
+  BKE_displist_minmax(&object_eval->runtime.curve_cache->disp, min, max);
 
   mid_v3_v3v3(loc, min, max);
 
@@ -7138,6 +7141,7 @@ static int match_texture_space_exec(bContext *C, wmOperator *UNUSED(op))
   curve->texflag &= ~CU_AUTOSPACE;
 
   WM_event_add_notifier(C, NC_GEOM | ND_DATA, curve);
+  DEG_id_tag_update(&curve->id, ID_RECALC_GEOMETRY);
 
   return OPERATOR_FINISHED;
 }
diff --git a/source/blender/editors/curve/editcurve_paint.c b/source/blender/editors/curve/editcurve_paint.c
index 5e0053782d4..c7c19aa2d02 100644
--- a/source/blender/editors/curve/editcurve_paint.c
+++ b/source/blender/editors/curve/editcurve_paint.c
@@ -120,7 +120,7 @@ struct CurveDrawData {
 
   struct {
     float mouse[2];
-    /* used incase we can't calculate the depth */
+    /* Used in case we can't calculate the depth. */
     float location_world[3];
 
     float location_world_valid[3];
@@ -1053,7 +1053,7 @@ static int curve_draw_invoke(bContext *C, wmOperator *op, const wmEvent *event)
 
   const bool is_modal = RNA_boolean_get(op->ptr, "wait_for_input");
 
-  /* fallback (incase we can't find the depth on first test) */
+  /* Fallback (in case we can't find the depth on first test). */
   {
     const float mval_fl[2] = {UNPACK2(event->mval)};
     float center[3];
diff --git a/source/blender/editors/curve/editcurve_select.c b/source/blender/editors/curve/editcurve_select.c
index d3f0ebfda3c..f67ccf1e4bd 100644
--- a/source/blender/editors/curve/editcurve_select.c
+++ b/source/blender/editors/curve/editcurve_select.c
@@ -1417,7 +1417,7 @@ void CURVE_OT_select_nth(wmOperatorType *ot)
 {
   /* identifiers */
   ot->name = "Checker Deselect";
-  ot->description = "Deselect every other vertex";
+  ot->description = "Deselect every Nth point starting from the active one";
   ot->idname = "CURVE_OT_select_nth";
 
   /* api callbacks */
diff --git a/source/blender/editors/curve/editfont.c b/source/blender/editors/curve/editfont.c
index f0c1abff201..781eb2634fb 100644
--- a/source/blender/editors/curve/editfont.c
+++ b/source/blender/editors/curve/editfont.c
@@ -71,7 +71,11 @@
 
 static int kill_selection(Object *obedit, int ins);
 
-/************************* utilities ******************************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Internal Utilities
+ * \{ */
 
 static wchar_t findaccent(wchar_t char1, unsigned int code)
 {
@@ -440,8 +444,48 @@ static void text_update_edited(bContext *C, Object *obedit, int mode)
   WM_event_add_notifier(C, NC_GEOM | ND_DATA, obedit->data);
 }
 
+static int kill_selection(Object *obedit, int ins) /* 1 == new character */
+{
+  Curve *cu = obedit->data;
+  EditFont *ef = cu->editfont;
+  int selend, selstart, direction;
+  int offset = 0;
+  int getfrom;
+
+  direction = BKE_vfont_select_get(obedit, &selstart, &selend);
+  if (direction) {
+    int size;
+    if (ins) {
+      offset = 1;
+    }
+    if (ef->pos >= selstart) {
+      ef->pos = selstart + offset;
+    }
+    if ((direction == -1) && ins) {
+      selstart++;
+      selend++;
+    }
+    getfrom = selend + offset;
+    if (ins == 0) {
+      getfrom++;
+    }
+    size = (ef->len * sizeof(wchar_t)) - (selstart * sizeof(wchar_t)) + (offset * sizeof(wchar_t));
+    memmove(ef->textbuf + selstart, ef->textbuf + getfrom, size);
+    memmove(ef->textbufinfo + selstart,
+            ef->textbufinfo + getfrom,
+            ((ef->len - selstart) + offset) * sizeof(CharInfo));
+    ef->len -= ((selend - selstart) + 1);
+    ef->selstart = ef->selend = 0;
+  }
+
+  return (direction);
+}
+
+/** \} */
+
 /* -------------------------------------------------------------------- */
-/* Generic Paste Functions */
+/** \name Generic Paste Functions
+ * \{ */
 
 /* text_update_edited(C, scene, obedit, 1, FO_EDIT); */
 static bool font_paste_wchar(Object *obedit,
@@ -506,8 +550,11 @@ static bool font_paste_utf8(bContext *C, const char *str, const size_t str_len)
   return retval;
 }
 
+/** \} */
+
 /* -------------------------------------------------------------------- */
-/* Paste From File*/
+/** \name Paste From File Operator
+ * \{ */
 
 static int paste_from_file(bContext *C, ReportList *reports, const char *filename)
 {
@@ -585,7 +632,11 @@ void FONT_OT_text_paste_from_file(wmOperatorType *ot)
                                  FILE_SORT_ALPHA);
 }
 
-/******************* text to object operator ********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Text To Object
+ * \{ */
 
 static void txt_add_object(bContext *C, TextLine *firstline, int totline, const float offset[3])
 {
@@ -703,46 +754,11 @@ void ED_text_to_object(bContext *C, Text *text, const bool split_lines)
   }
 }
 
-/********************** utilities ***************************/
-
-static int kill_selection(Object *obedit, int ins) /* 1 == new character */
-{
-  Curve *cu = obedit->data;
-  EditFont *ef = cu->editfont;
-  int selend, selstart, direction;
-  int offset = 0;
-  int getfrom;
-
-  direction = BKE_vfont_select_get(obedit, &selstart, &selend);
-  if (direction) {
-    int size;
-    if (ins) {
-      offset = 1;
-    }
-    if (ef->pos >= selstart) {
-      ef->pos = selstart + offset;
-    }
-    if ((direction == -1) && ins) {
-      selstart++;
-      selend++;
-    }
-    getfrom = selend + offset;
-    if (ins == 0) {
-      getfrom++;
-    }
-    size = (ef->len * sizeof(wchar_t)) - (selstart * sizeof(wchar_t)) + (offset * sizeof(wchar_t));
-    memmove(ef->textbuf + selstart, ef->textbuf + getfrom, size);
-    memmove(ef->textbufinfo + selstart,
-            ef->textbufinfo + getfrom,
-            ((ef->len - selstart) + offset) * sizeof(CharInfo));
-    ef->len -= ((selend - selstart) + 1);
-    ef->selstart = ef->selend = 0;
-  }
-
-  return (direction);
-}
+/** \} */
 
-/******************* set style operator ********************/
+/* -------------------------------------------------------------------- */
+/** \name Set Style Operator
+ * \{ */
 
 static const EnumPropertyItem style_items[] = {
     {CU_CHINFO_BOLD, "BOLD", 0, "Bold", ""},
@@ -806,7 +822,11 @@ void FONT_OT_style_set(wmOperatorType *ot)
   RNA_def_boolean(ot->srna, "clear", 0, "Clear", "Clear style rather than setting it");
 }
 
-/******************* toggle style operator ********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Toggle Style Operator
+ * \{ */
 
 static int toggle_style_exec(bContext *C, wmOperator *op)
 {
@@ -845,8 +865,11 @@ void FONT_OT_style_toggle(wmOperatorType *ot)
       ot->srna, "style", style_items, CU_CHINFO_BOLD, "Style", "Style to set selection to");
 }
 
+/** \} */
+
 /* -------------------------------------------------------------------- */
-/* Select All */
+/** \name Select All Operator
+ * \{ */
 
 static int font_select_all_exec(bContext *C, wmOperator *UNUSED(op))
 {
@@ -883,7 +906,11 @@ void FONT_OT_select_all(wmOperatorType *ot)
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 }
 
-/******************* copy text operator ********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Copy Text Operator
+ * \{ */
 
 static void copy_selection(Object *obedit)
 {
@@ -932,7 +959,11 @@ void FONT_OT_text_copy(wmOperatorType *ot)
   ot->poll = ED_operator_editfont;
 }
 
-/******************* cut text operator ********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Cut Text Operator
+ * \{ */
 
 static int cut_text_exec(bContext *C, wmOperator *UNUSED(op))
 {
@@ -966,7 +997,11 @@ void FONT_OT_text_cut(wmOperatorType *ot)
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 }
 
-/******************* paste text operator ********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Paste Text Operator
+ * \{ */
 
 static bool paste_selection(Object *obedit, ReportList *reports)
 {
@@ -1066,7 +1101,11 @@ void FONT_OT_text_paste(wmOperatorType *ot)
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 }
 
-/************************ move operator ************************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Move Operator
+ * \{ */
 
 static const EnumPropertyItem move_type_items[] = {
     {LINE_BEGIN, "LINE_BEGIN", 0, "Line Begin", ""},
@@ -1232,7 +1271,11 @@ void FONT_OT_move(wmOperatorType *ot)
   RNA_def_enum(ot->srna, "type", move_type_items, LINE_BEGIN, "Type", "Where to move cursor to");
 }
 
-/******************* move select operator ********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Move Select Operator
+ * \{ */
 
 static int move_select_exec(bContext *C, wmOperator *op)
 {
@@ -1264,7 +1307,11 @@ void FONT_OT_move_select(wmOperatorType *ot)
                "Where to move cursor to, to make a selection");
 }
 
-/************************* change spacing **********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Change Spacing
+ * \{ */
 
 static int change_spacing_exec(bContext *C, wmOperator *op)
 {
@@ -1314,7 +1361,11 @@ void FONT_OT_change_spacing(wmOperatorType *ot)
               20);
 }
 
-/************************* change character **********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Change Character
+ * \{ */
 
 static int change_character_exec(bContext *C, wmOperator *op)
 {
@@ -1368,7 +1419,11 @@ void FONT_OT_change_character(wmOperatorType *ot)
               255);
 }
 
-/******************* line break operator ********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Line Break Operator
+ * \{ */
 
 static int line_break_exec(bContext *C, wmOperator *UNUSED(op))
 {
@@ -1400,7 +1455,11 @@ void FONT_OT_line_break(wmOperatorType *ot)
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 }
 
-/******************* delete operator **********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Delete Operator
+ * \{ */
 
 static const EnumPropertyItem delete_type_items[] = {
     {DEL_NEXT_CHAR, "NEXT_CHARACTER", 0, "Next Character", ""},
@@ -1549,7 +1608,11 @@ void FONT_OT_delete(wmOperatorType *ot)
                "Which part of the text to delete");
 }
 
-/*********************** insert text operator *************************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Insert Text Operator
+ * \{ */
 
 static int insert_text_exec(bContext *C, wmOperator *op)
 {
@@ -1700,7 +1763,12 @@ void FONT_OT_text_insert(wmOperatorType *ot)
       "Next typed character will strike through previous, for special character input");
 }
 
-/*********************** textbox add operator *************************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Textbox Add Operator
+ * \{ */
+
 static int textbox_add_exec(bContext *C, wmOperator *UNUSED(op))
 {
   Object *obedit = CTX_data_active_object(C);
@@ -1736,7 +1804,11 @@ void FONT_OT_textbox_add(wmOperatorType *ot)
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 }
 
-/*********************** textbox remove operator *************************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Textbox Remove Operator
+ * \{ */
 
 static int textbox_remove_exec(bContext *C, wmOperator *op)
 {
@@ -1778,7 +1850,11 @@ void FONT_OT_textbox_remove(wmOperatorType *ot)
   RNA_def_int(ot->srna, "index", 0, 0, INT_MAX, "Index", "The current text box", 0, INT_MAX);
 }
 
-/***************** editmode enter/exit ********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Editmode Enter/Exit
+ * \{ */
 
 void ED_curve_editfont_make(Object *obedit)
 {
@@ -1851,7 +1927,11 @@ void ED_curve_editfont_free(Object *obedit)
   BKE_curve_editfont_free((Curve *)obedit->data);
 }
 
-/********************** set case operator *********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Set Case Operator
+ * \{ */
 
 static const EnumPropertyItem case_items[] = {
     {CASE_LOWER, "LOWER", 0, "Lower", ""},
@@ -1920,7 +2000,11 @@ void FONT_OT_case_set(wmOperatorType *ot)
   RNA_def_enum(ot->srna, "case", case_items, CASE_LOWER, "Case", "Lower or upper case");
 }
 
-/********************** toggle case operator *********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Toggle Case Operator
+ * \{ */
 
 static int toggle_case_exec(bContext *C, wmOperator *UNUSED(op))
 {
@@ -2031,7 +2115,7 @@ static int open_invoke(bContext *C, wmOperator *op, const wmEvent *UNUSED(event)
 
   if (pprop->prop) {
     idptr = RNA_property_pointer_get((PointerRNA *)pprop, pprop->prop);
-    vfont = idptr.id.data;
+    vfont = (VFont *)idptr.owner_id;
   }
 
   path = (vfont && !BKE_vfont_is_builtin(vfont)) ? vfont->name : U.fontdir;
@@ -2071,7 +2155,11 @@ void FONT_OT_open(wmOperatorType *ot)
                                  FILE_SORT_ALPHA);
 }
 
-/******************* delete operator *********************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Delete Operator
+ * \{ */
 
 static int font_unlink_exec(bContext *C, wmOperator *op)
 {
@@ -2190,3 +2278,5 @@ bool ED_curve_editfont_select_pick(
     return false;
   }
 }
+
+/** \} */
diff --git a/source/blender/editors/gizmo_library/gizmo_types/arrow3d_gizmo.c b/source/blender/editors/gizmo_library/gizmo_types/arrow3d_gizmo.c
index d23965269ab..fa9c0f1fbb2 100644
--- a/source/blender/editors/gizmo_library/gizmo_types/arrow3d_gizmo.c
+++ b/source/blender/editors/gizmo_library/gizmo_types/arrow3d_gizmo.c
@@ -363,8 +363,8 @@ static void gizmo_arrow_exit(bContext *C, wmGizmo *gz, const bool cancel)
   const bool is_prop_valid = WM_gizmo_target_property_is_valid(gz_prop);
 
   if (!cancel) {
-    /* Assign incase applying the operation needs an updated offset
-     * editmesh bisect needs this. */
+    /* Assign in case applying the operation needs an updated offset
+     * edit-mesh bisect needs this. */
     if (is_prop_valid) {
       const int transform_flag = RNA_enum_get(arrow->gizmo.ptr, "transform");
       const bool constrained = (transform_flag & ED_GIZMO_ARROW_XFORM_FLAG_CONSTRAINED) != 0;
diff --git a/source/blender/editors/gizmo_library/gizmo_types/cage3d_gizmo.c b/source/blender/editors/gizmo_library/gizmo_types/cage3d_gizmo.c
index e2a86469da1..406f76bc65e 100644
--- a/source/blender/editors/gizmo_library/gizmo_types/cage3d_gizmo.c
+++ b/source/blender/editors/gizmo_library/gizmo_types/cage3d_gizmo.c
@@ -358,7 +358,7 @@ static void gizmo_cage3d_draw_intern(
       bool show = false;
       if (gz->highlight_part == ED_GIZMO_CAGE3D_PART_TRANSLATE) {
         /* Only show if we're drawing the center handle
-         * otherwise the entire rectangle is the hotspot. */
+         * otherwise the entire rectangle is the hot-spot. */
         if (draw_options & ED_GIZMO_CAGE2D_DRAW_FLAG_XFORM_CENTER_HANDLE) {
           show = true;
         }
diff --git a/source/blender/editors/gizmo_library/gizmo_types/move3d_gizmo.c b/source/blender/editors/gizmo_library/gizmo_types/move3d_gizmo.c
index 21f779b72b1..37ee95d5058 100644
--- a/source/blender/editors/gizmo_library/gizmo_types/move3d_gizmo.c
+++ b/source/blender/editors/gizmo_library/gizmo_types/move3d_gizmo.c
@@ -283,6 +283,7 @@ static int gizmo_move_modal(bContext *C,
                   .use_occlusion_test = true,
               },
               mval_fl,
+              NULL,
               &dist_px,
               co,
               NULL)) {
diff --git a/source/blender/editors/gpencil/annotate_paint.c b/source/blender/editors/gpencil/annotate_paint.c
index 07b61751b22..22f1753a810 100644
--- a/source/blender/editors/gpencil/annotate_paint.c
+++ b/source/blender/editors/gpencil/annotate_paint.c
@@ -1156,7 +1156,7 @@ static tGPsdata *gp_session_initpaint(bContext *C)
   /* create new context data */
   p = MEM_callocN(sizeof(tGPsdata), "Annotation Drawing Data");
 
-  /* Try to initialise context data
+  /* Try to initialize context data
    * WARNING: This may not always succeed (e.g. using GP in an annotation-only context)
    */
   if (gp_session_initdata(C, p) == 0) {
@@ -1252,15 +1252,6 @@ static void gp_paint_initstroke(tGPsdata *p, eGPencil_PaintModes paintmode, Deps
     /* Ensure active frame is set correctly... */
     p->gpf = p->gpl->actframe;
 
-    /* Restrict eraser to only affecting selected strokes, if the "selection mask" is on
-     * (though this is only available in editmode)
-     */
-    if (p->gpd->flag & GP_DATA_STROKE_EDITMODE) {
-      if (ts->gp_sculpt.flag & GP_SCULPT_SETT_FLAG_SELECT_MASK) {
-        p->flags |= GP_PAINTFLAG_SELECTMASK;
-      }
-    }
-
     if (has_layer_to_erase == false) {
       p->status = GP_STATUS_CAPTURE;
       // if (G.debug & G_DEBUG)
diff --git a/source/blender/editors/gpencil/gpencil_add_monkey.c b/source/blender/editors/gpencil/gpencil_add_monkey.c
index c4528518009..139697ad0e3 100644
--- a/source/blender/editors/gpencil/gpencil_add_monkey.c
+++ b/source/blender/editors/gpencil/gpencil_add_monkey.c
@@ -54,7 +54,7 @@ static int gpencil_monkey_color(
   short *totcol = give_totcolp(ob);
   Material *ma = NULL;
   for (short i = 0; i < *totcol; i++) {
-    ma = give_current_material(ob, i + 1);
+    ma = BKE_material_gpencil_get(ob, i + 1);
     if (STREQ(ma->id.name, pct->name)) {
       return i;
     }
diff --git a/source/blender/editors/gpencil/gpencil_add_stroke.c b/source/blender/editors/gpencil/gpencil_add_stroke.c
index 80e239c9ae5..74617599eaa 100644
--- a/source/blender/editors/gpencil/gpencil_add_stroke.c
+++ b/source/blender/editors/gpencil/gpencil_add_stroke.c
@@ -53,7 +53,7 @@ static int gp_stroke_material(Main *bmain, Object *ob, const ColorTemplate *pct,
   short *totcol = give_totcolp(ob);
   Material *ma = NULL;
   for (short i = 0; i < *totcol; i++) {
-    ma = give_current_material(ob, i + 1);
+    ma = BKE_material_gpencil_get(ob, i + 1);
     if (STREQ(ma->id.name, pct->name)) {
       return i;
     }
diff --git a/source/blender/editors/gpencil/gpencil_brush.c b/source/blender/editors/gpencil/gpencil_brush.c
index 9124c0f5d51..d1c4f271321 100644
--- a/source/blender/editors/gpencil/gpencil_brush.c
+++ b/source/blender/editors/gpencil/gpencil_brush.c
@@ -105,6 +105,7 @@ typedef struct tGP_BrushEditData {
   eGP_Sculpt_Types brush_type;
   eGP_Sculpt_Types brush_type_old;
   eGP_Sculpt_Flag flag;
+  eGP_Sculpt_SelectMaskFlag mask;
 
   /* Space Conversion Data */
   GP_SpaceConversion gsc;
@@ -133,8 +134,8 @@ typedef struct tGP_BrushEditData {
   /* - effect vector (e.g. 2D/3D translation for grab brush) */
   float dvec[3];
 
-  /* rotation for derived data */
-  float rot;
+  /* rotation for evaluated data */
+  float rot_eval;
 
   /* - multiframe falloff factor */
   float mf_falloff;
@@ -300,6 +301,20 @@ static float gp_brush_influence_calc(tGP_BrushEditData *gso, const int radius, c
   return influence;
 }
 
+/* Force recal filling data */
+static void gp_recalc_geometry(bGPDstroke *gps)
+{
+  bGPDstroke *gps_orig = gps->runtime.gps_orig;
+  if (gps_orig) {
+    gps_orig->flag |= GP_STROKE_RECALC_GEOMETRY;
+    gps_orig->tot_triangles = 0;
+  }
+  else {
+    gps->flag |= GP_STROKE_RECALC_GEOMETRY;
+    gps->tot_triangles = 0;
+  }
+}
+
 /* ************************************************ */
 /* Brush Callbacks */
 /* This section defines the callbacks used by each brush to perform their magic.
@@ -313,7 +328,7 @@ static float gp_brush_influence_calc(tGP_BrushEditData *gso, const int radius, c
  * smooth-brush implementation to test the algorithm for stroke smoothing. */
 static bool gp_brush_smooth_apply(tGP_BrushEditData *gso,
                                   bGPDstroke *gps,
-                                  float UNUSED(rot),
+                                  float UNUSED(rot_eval),
                                   int pt_index,
                                   const int radius,
                                   const int co[2])
@@ -341,7 +356,7 @@ static bool gp_brush_smooth_apply(tGP_BrushEditData *gso,
     BKE_gpencil_smooth_stroke_uv(gps, pt_index, inf);
   }
 
-  gps->flag |= GP_STROKE_RECALC_GEOMETRY;
+  gp_recalc_geometry(gps);
 
   return true;
 }
@@ -352,7 +367,7 @@ static bool gp_brush_smooth_apply(tGP_BrushEditData *gso,
 /* Make lines thicker or thinner by the specified amounts */
 static bool gp_brush_thickness_apply(tGP_BrushEditData *gso,
                                      bGPDstroke *gps,
-                                     float UNUSED(rot),
+                                     float UNUSED(rot_eval),
                                      int pt_index,
                                      const int radius,
                                      const int co[2])
@@ -396,7 +411,7 @@ static bool gp_brush_thickness_apply(tGP_BrushEditData *gso,
 /* Make color more or less transparent by the specified amounts */
 static bool gp_brush_strength_apply(tGP_BrushEditData *gso,
                                     bGPDstroke *gps,
-                                    float UNUSED(rot),
+                                    float UNUSED(rot_eval),
                                     int pt_index,
                                     const int radius,
                                     const int co[2])
@@ -443,7 +458,7 @@ typedef struct tGPSB_Grab_StrokeData {
   /* array of influence weights for each of the included points */
   float *weights;
   /* angles to calc transformation */
-  float *rot;
+  float *rot_eval;
 
   /* capacity of the arrays */
   int capacity;
@@ -471,7 +486,7 @@ static void gp_brush_grab_stroke_init(tGP_BrushEditData *gso, bGPDstroke *gps)
 
     memset(data->points, 0, sizeof(int) * data->capacity);
     memset(data->weights, 0, sizeof(float) * data->capacity);
-    memset(data->rot, 0, sizeof(float) * data->capacity);
+    memset(data->rot_eval, 0, sizeof(float) * data->capacity);
   }
   else {
     /* Create new instance */
@@ -482,7 +497,7 @@ static void gp_brush_grab_stroke_init(tGP_BrushEditData *gso, bGPDstroke *gps)
 
     data->points = MEM_callocN(sizeof(int) * data->capacity, "GP Stroke Grab Indices");
     data->weights = MEM_callocN(sizeof(float) * data->capacity, "GP Stroke Grab Weights");
-    data->rot = MEM_callocN(sizeof(float) * data->capacity, "GP Stroke Grab X");
+    data->rot_eval = MEM_callocN(sizeof(float) * data->capacity, "GP Stroke Grab Rotations");
 
     /* hook up to the cache */
     BLI_ghash_insert(gso->stroke_customdata, gps, data);
@@ -492,7 +507,7 @@ static void gp_brush_grab_stroke_init(tGP_BrushEditData *gso, bGPDstroke *gps)
 /* store references to stroke points in the initial stage */
 static bool gp_brush_grab_store_points(tGP_BrushEditData *gso,
                                        bGPDstroke *gps,
-                                       float rot,
+                                       float rot_eval,
                                        int pt_index,
                                        const int radius,
                                        const int co[2])
@@ -506,7 +521,7 @@ static bool gp_brush_grab_store_points(tGP_BrushEditData *gso,
   /* insert this point into the set of affected points */
   data->points[data->size] = pt_index;
   data->weights[data->size] = inf;
-  data->rot[data->size] = rot;
+  data->rot_eval[data->size] = rot_eval;
   data->size++;
 
   /* done */
@@ -530,10 +545,10 @@ static void gp_brush_grab_calc_dvec(tGP_BrushEditData *gso)
     mval_f[0] = (float)(gso->mval[0] - gso->mval_prev[0]);
     mval_f[1] = (float)(gso->mval[1] - gso->mval_prev[1]);
 
-    /* apply derived data transformation */
-    if (gso->rot != 0.0f) {
-      const float cval = cos(gso->rot);
-      const float sval = sin(gso->rot);
+    /* apply evaluated data transformation */
+    if (gso->rot_eval != 0.0f) {
+      const float cval = cos(gso->rot_eval);
+      const float sval = sin(gso->rot_eval);
       float r[2];
       r[0] = (mval_f[0] * cval) - (mval_f[1] * sval);
       r[1] = (mval_f[0] * sval) + (mval_f[1] * cval);
@@ -564,8 +579,8 @@ static void gp_brush_grab_apply_cached(tGP_BrushEditData *gso,
     bGPDspoint *pt = &gps->points[data->points[i]];
     float delta[3] = {0.0f};
 
-    /* get derived transformation */
-    gso->rot = data->rot[i];
+    /* get evaluated transformation */
+    gso->rot_eval = data->rot_eval[i];
     gp_brush_grab_calc_dvec(gso);
 
     /* adjust the amount of displacement to apply */
@@ -586,7 +601,7 @@ static void gp_brush_grab_apply_cached(tGP_BrushEditData *gso,
     /* compute lock axis */
     gpsculpt_compute_lock_axis(gso, pt, save_pt);
   }
-  gps->flag |= GP_STROKE_RECALC_GEOMETRY;
+  gp_recalc_geometry(gps);
 }
 
 /* free customdata used for handling this stroke */
@@ -597,7 +612,7 @@ static void gp_brush_grab_stroke_free(void *ptr)
   /* free arrays */
   MEM_SAFE_FREE(data->points);
   MEM_SAFE_FREE(data->weights);
-  MEM_SAFE_FREE(data->rot);
+  MEM_SAFE_FREE(data->rot_eval);
 
   /* ... and this item itself, since it was also allocated */
   MEM_freeN(data);
@@ -608,7 +623,7 @@ static void gp_brush_grab_stroke_free(void *ptr)
 /* NOTE: Depends on gp_brush_grab_calc_dvec() */
 static bool gp_brush_push_apply(tGP_BrushEditData *gso,
                                 bGPDstroke *gps,
-                                float UNUSED(rot),
+                                float UNUSED(rot_eval),
                                 int pt_index,
                                 const int radius,
                                 const int co[2])
@@ -676,7 +691,7 @@ static void gp_brush_calc_midpoint(tGP_BrushEditData *gso)
 /* Shrink distance between midpoint and this point... */
 static bool gp_brush_pinch_apply(tGP_BrushEditData *gso,
                                  bGPDstroke *gps,
-                                 float UNUSED(rot),
+                                 float UNUSED(rot_eval),
                                  int pt_index,
                                  const int radius,
                                  const int co[2])
@@ -721,7 +736,7 @@ static bool gp_brush_pinch_apply(tGP_BrushEditData *gso,
   /* compute lock axis */
   gpsculpt_compute_lock_axis(gso, pt, save_pt);
 
-  gps->flag |= GP_STROKE_RECALC_GEOMETRY;
+  gp_recalc_geometry(gps);
 
   /* done */
   return true;
@@ -735,7 +750,7 @@ static bool gp_brush_pinch_apply(tGP_BrushEditData *gso,
 
 static bool gp_brush_twist_apply(tGP_BrushEditData *gso,
                                  bGPDstroke *gps,
-                                 float UNUSED(rot),
+                                 float UNUSED(rot_eval),
                                  int pt_index,
                                  const int radius,
                                  const int co[2])
@@ -809,7 +824,7 @@ static bool gp_brush_twist_apply(tGP_BrushEditData *gso,
     }
   }
 
-  gps->flag |= GP_STROKE_RECALC_GEOMETRY;
+  gp_recalc_geometry(gps);
 
   /* done */
   return true;
@@ -820,7 +835,7 @@ static bool gp_brush_twist_apply(tGP_BrushEditData *gso,
 /* Apply some random jitter to the point */
 static bool gp_brush_randomize_apply(tGP_BrushEditData *gso,
                                      bGPDstroke *gps,
-                                     float UNUSED(rot),
+                                     float UNUSED(rot_eval),
                                      int pt_index,
                                      const int radius,
                                      const int co[2])
@@ -929,7 +944,7 @@ static bool gp_brush_randomize_apply(tGP_BrushEditData *gso,
     CLAMP(pt->uv_rot, -M_PI_2, M_PI_2);
   }
 
-  gps->flag |= GP_STROKE_RECALC_GEOMETRY;
+  gp_recalc_geometry(gps);
 
   /* done */
   return true;
@@ -939,7 +954,7 @@ static bool gp_brush_randomize_apply(tGP_BrushEditData *gso,
 /* Change weight paint for vertex groups */
 static bool gp_brush_weight_apply(tGP_BrushEditData *gso,
                                   bGPDstroke *gps,
-                                  float UNUSED(rot),
+                                  float UNUSED(rot_eval),
                                   int pt_index,
                                   const int radius,
                                   const int co[2])
@@ -1175,7 +1190,7 @@ static void gp_brush_clone_adjust(tGP_BrushEditData *gso)
   size_t snum;
 
   /* Compute the amount of movement to apply (overwrites dvec) */
-  gso->rot = 0.0f;
+  gso->rot_eval = 0.0f;
   gp_brush_grab_calc_dvec(gso);
 
   /* For each of the stored strokes, apply the offset to each point */
@@ -1323,6 +1338,9 @@ static bool gpsculpt_brush_init(bContext *C, wmOperator *op)
   gso->sa = CTX_wm_area(C);
   gso->ar = CTX_wm_region(C);
 
+  /* save mask */
+  gso->mask = ts->gpencil_selectmode_sculpt;
+
   /* multiframe settings */
   gso->is_multiframe = (bool)GPENCIL_MULTIEDIT_SESSIONS_ON(gso->gpd);
   gso->use_multiframe_falloff = (ts->gp_sculpt.flag & GP_SCULPT_SETT_FLAG_FRAME_FALLOFF) != 0;
@@ -1486,35 +1504,38 @@ static void gpsculpt_brush_init_stroke(tGP_BrushEditData *gso)
 
 /* Apply ----------------------------------------------- */
 
-/* Get angle of the segment relative to the original segment before any transformation */
-static float gpsculpt_transform_rot_get(GP_SpaceConversion *gsc,
-                                        bGPDstroke *gps_derived,
-                                        bGPDspoint *pt_derived,
-                                        int idx_derived)
+/* Get angle of the segment relative to the original segment before any transformation
+ * For strokes with one point only this is impossible to calculate because there isn't a
+ * valid reference point.
+ */
+static float gpsculpt_rotation_eval_get(GP_SpaceConversion *gsc,
+                                        bGPDstroke *gps_eval,
+                                        bGPDspoint *pt_eval,
+                                        int idx_eval)
 {
 
-  bGPDstroke *gps_orig = gps_derived->runtime.gps_orig;
-  bGPDspoint *pt_orig = &gps_orig->points[pt_derived->runtime.idx_orig];
-  bGPDspoint *pt_derived_prev = NULL;
+  bGPDstroke *gps_orig = gps_eval->runtime.gps_orig;
+  bGPDspoint *pt_orig = &gps_orig->points[pt_eval->runtime.idx_orig];
+  bGPDspoint *pt_prev_eval = NULL;
   bGPDspoint *pt_orig_prev = NULL;
-  if (idx_derived != 0) {
-    pt_derived_prev = &gps_derived->points[idx_derived - 1];
+  if (idx_eval != 0) {
+    pt_prev_eval = &gps_eval->points[idx_eval - 1];
   }
   else {
-    if (gps_derived->totpoints > 1) {
-      pt_derived_prev = &gps_derived->points[idx_derived + 1];
+    if (gps_eval->totpoints > 1) {
+      pt_prev_eval = &gps_eval->points[idx_eval + 1];
     }
     else {
       return 0.0f;
     }
   }
 
-  if (pt_derived->runtime.idx_orig != 0) {
-    pt_orig_prev = &gps_orig->points[pt_derived->runtime.idx_orig - 1];
+  if (pt_eval->runtime.idx_orig != 0) {
+    pt_orig_prev = &gps_orig->points[pt_eval->runtime.idx_orig - 1];
   }
   else {
     if (gps_orig->totpoints > 1) {
-      pt_orig_prev = &gps_orig->points[pt_derived->runtime.idx_orig + 1];
+      pt_orig_prev = &gps_orig->points[pt_eval->runtime.idx_orig + 1];
     }
     else {
       return 0.0f;
@@ -1522,17 +1543,17 @@ static float gpsculpt_transform_rot_get(GP_SpaceConversion *gsc,
   }
 
   /* create 2D vectors of the stroke segments */
-  float v_orig_a[2], v_orig_b[2], v_derived_a[2], v_derived_b[2];
+  float v_orig_a[2], v_orig_b[2], v_eval_a[2], v_eval_b[2];
 
   gp_point_3d_to_xy(gsc, GP_STROKE_3DSPACE, &pt_orig->x, v_orig_a);
   gp_point_3d_to_xy(gsc, GP_STROKE_3DSPACE, &pt_orig_prev->x, v_orig_b);
   sub_v2_v2(v_orig_a, v_orig_b);
 
-  gp_point_3d_to_xy(gsc, GP_STROKE_3DSPACE, &pt_derived->x, v_derived_a);
-  gp_point_3d_to_xy(gsc, GP_STROKE_3DSPACE, &pt_derived_prev->x, v_derived_b);
-  sub_v2_v2(v_derived_a, v_derived_b);
+  gp_point_3d_to_xy(gsc, GP_STROKE_3DSPACE, &pt_eval->x, v_eval_a);
+  gp_point_3d_to_xy(gsc, GP_STROKE_3DSPACE, &pt_prev_eval->x, v_eval_b);
+  sub_v2_v2(v_eval_a, v_eval_b);
 
-  return angle_v2v2(v_orig_a, v_derived_a);
+  return angle_v2v2(v_orig_a, v_eval_a);
 }
 
 /* Apply brush operation to points in this stroke */
@@ -1555,7 +1576,7 @@ static bool gpsculpt_brush_do_stroke(tGP_BrushEditData *gso,
   int i;
   bool include_last = false;
   bool changed = false;
-  float rot = 0.0f;
+  float rot_eval = 0.0f;
   if (gps->totpoints == 1) {
     bGPDspoint pt_temp;
     pt = &gps->points[0];
@@ -1570,8 +1591,8 @@ static bool gpsculpt_brush_do_stroke(tGP_BrushEditData *gso,
       if (len_v2v2_int(mval_i, pc1) <= radius) {
         /* apply operation to this point */
         if (pt->runtime.pt_orig != NULL) {
-          rot = gpsculpt_transform_rot_get(&gso->gsc, gps, pt, 0);
-          changed = apply(gso, gps->runtime.gps_orig, rot, pt->runtime.idx_orig, radius, pc1);
+          rot_eval = gpsculpt_rotation_eval_get(&gso->gsc, gps, pt, 0);
+          changed = apply(gso, gps->runtime.gps_orig, rot_eval, pt->runtime.idx_orig, radius, pc1);
         }
       }
     }
@@ -1587,7 +1608,7 @@ static bool gpsculpt_brush_do_stroke(tGP_BrushEditData *gso,
 
       /* Skip if neither one is selected
        * (and we are only allowed to edit/consider selected points) */
-      if ((gso->settings->flag & GP_SCULPT_SETT_FLAG_SELECT_MASK) && (!gso->is_weight_mode)) {
+      if ((GPENCIL_ANY_SCULPT_MASK(gso->mask)) && (!gso->is_weight_mode)) {
         if (!(pt1->flag & GP_SPOINT_SELECT) && !(pt2->flag & GP_SPOINT_SELECT)) {
           include_last = false;
           continue;
@@ -1615,8 +1636,8 @@ static bool gpsculpt_brush_do_stroke(tGP_BrushEditData *gso,
           /* To each point individually... */
           pt = &gps->points[i];
           if (pt->runtime.pt_orig != NULL) {
-            rot = gpsculpt_transform_rot_get(&gso->gsc, gps, pt, i);
-            ok = apply(gso, gps->runtime.gps_orig, rot, pt->runtime.idx_orig, radius, pc1);
+            rot_eval = gpsculpt_rotation_eval_get(&gso->gsc, gps, pt, i);
+            ok = apply(gso, gps->runtime.gps_orig, rot_eval, pt->runtime.idx_orig, radius, pc1);
           }
 
           /* Only do the second point if this is the last segment,
@@ -1630,8 +1651,8 @@ static bool gpsculpt_brush_do_stroke(tGP_BrushEditData *gso,
           if (i + 1 == gps->totpoints - 1) {
             pt = &gps->points[i + 1];
             if (pt->runtime.pt_orig != NULL) {
-              rot = gpsculpt_transform_rot_get(&gso->gsc, gps, pt, i + 1);
-              ok |= apply(gso, gps->runtime.gps_orig, rot, pt->runtime.idx_orig, radius, pc2);
+              rot_eval = gpsculpt_rotation_eval_get(&gso->gsc, gps, pt, i + 1);
+              ok |= apply(gso, gps->runtime.gps_orig, rot_eval, pt->runtime.idx_orig, radius, pc2);
               include_last = false;
             }
           }
@@ -1649,8 +1670,9 @@ static bool gpsculpt_brush_do_stroke(tGP_BrushEditData *gso,
            */
           pt = &gps->points[i];
           if (pt->runtime.pt_orig != NULL) {
-            rot = gpsculpt_transform_rot_get(&gso->gsc, gps, pt, i);
-            changed |= apply(gso, gps->runtime.gps_orig, rot, pt->runtime.idx_orig, radius, pc1);
+            rot_eval = gpsculpt_rotation_eval_get(&gso->gsc, gps, pt, i);
+            changed |= apply(
+                gso, gps->runtime.gps_orig, rot_eval, pt->runtime.idx_orig, radius, pc1);
             include_last = false;
           }
         }
@@ -1752,10 +1774,7 @@ static bool gpsculpt_brush_do_frame(
         break;
     }
     /* Triangulation must be calculated if changed */
-    if (changed) {
-      gps->flag |= GP_STROKE_RECALC_GEOMETRY;
-      gps->tot_triangles = 0;
-    }
+    gp_recalc_geometry(gps);
   }
 
   return changed;
@@ -1777,7 +1796,7 @@ static bool gpsculpt_brush_apply_standard(bContext *C, tGP_BrushEditData *gso)
     case GP_SCULPT_TYPE_PUSH: /* Push points */
     {
       /* calculate amount of displacement to apply */
-      gso->rot = 0.0f;
+      gso->rot_eval = 0.0f;
       gp_brush_grab_calc_dvec(gso);
       break;
     }
@@ -1793,7 +1812,7 @@ static bool gpsculpt_brush_apply_standard(bContext *C, tGP_BrushEditData *gso)
     case GP_SCULPT_TYPE_RANDOMIZE: /* Random jitter */
     {
       /* compute the displacement vector for the cursor (in data space) */
-      gso->rot = 0.0f;
+      gso->rot_eval = 0.0f;
       gp_brush_grab_calc_dvec(gso);
       break;
     }
@@ -1808,10 +1827,10 @@ static bool gpsculpt_brush_apply_standard(bContext *C, tGP_BrushEditData *gso)
     if (gpl->actframe == NULL) {
       continue;
     }
-    /* Get derived frames array data */
-    int derived_idx = BLI_findindex(&gpd->layers, gpl);
-    bGPDframe *derived_gpf = &ob_eval->runtime.derived_frames[derived_idx];
-    if (derived_gpf == NULL) {
+    /* Get evaluated frames array data */
+    int idx_eval = BLI_findindex(&gpd->layers, gpl);
+    bGPDframe *gpf_eval = &ob_eval->runtime.gpencil_evaluated_frames[idx_eval];
+    if (gpf_eval == NULL) {
       continue;
     }
 
@@ -1845,14 +1864,14 @@ static bool gpsculpt_brush_apply_standard(bContext *C, tGP_BrushEditData *gso)
 
           /* affect strokes in this frame */
           changed |= gpsculpt_brush_do_frame(
-              C, gso, gpl, (gpf == gpl->actframe) ? derived_gpf : gpf, diff_mat);
+              C, gso, gpl, (gpf == gpl->actframe) ? gpf_eval : gpf, diff_mat);
         }
       }
     }
     else {
       /* Apply to active frame's strokes */
       gso->mf_falloff = 1.0f;
-      changed |= gpsculpt_brush_do_frame(C, gso, gpl, derived_gpf, diff_mat);
+      changed |= gpsculpt_brush_do_frame(C, gso, gpl, gpf_eval, diff_mat);
     }
   }
   CTX_DATA_END;
diff --git a/source/blender/editors/gpencil/gpencil_data.c b/source/blender/editors/gpencil/gpencil_data.c
index 0928913aad5..e763eec1b0d 100644
--- a/source/blender/editors/gpencil/gpencil_data.c
+++ b/source/blender/editors/gpencil/gpencil_data.c
@@ -105,7 +105,7 @@ static bool gp_data_add_poll(bContext *C)
 /* add new datablock - wrapper around API */
 static int gp_data_add_exec(bContext *C, wmOperator *op)
 {
-  PointerRNA gpd_owner = {{NULL}};
+  PointerRNA gpd_owner = {NULL};
   bGPdata **gpd_ptr = ED_gpencil_data_get_pointers(C, &gpd_owner);
   bool is_annotation = ED_gpencil_data_owner_is_annotation(&gpd_owner);
 
@@ -231,7 +231,7 @@ void GPENCIL_OT_data_unlink(wmOperatorType *ot)
 /* add new layer - wrapper around API */
 static int gp_layer_add_exec(bContext *C, wmOperator *op)
 {
-  PointerRNA gpd_owner = {{NULL}};
+  PointerRNA gpd_owner = {NULL};
   bGPdata **gpd_ptr = ED_gpencil_data_get_pointers(C, &gpd_owner);
   bool is_annotation = ED_gpencil_data_owner_is_annotation(&gpd_owner);
 
@@ -538,10 +538,12 @@ static int gp_layer_duplicate_object_exec(bContext *C, wmOperator *op)
        * otherwise add the slot with the material
        */
       Material *ma_src = give_current_material(ob_src, gps_src->mat_nr + 1);
-      int idx = BKE_gpencil_object_material_ensure(bmain, ob_dst, ma_src);
+      if (ma_src != NULL) {
+        int idx = BKE_gpencil_object_material_ensure(bmain, ob_dst, ma_src);
 
-      /* Reassign the stroke material to the right slot in destination object. */
-      gps_dst->mat_nr = idx;
+        /* Reassign the stroke material to the right slot in destination object. */
+        gps_dst->mat_nr = idx;
+      }
 
       /* add new stroke to frame */
       BLI_addtail(&gpf_dst->strokes, gps_dst);
@@ -840,6 +842,10 @@ static int gp_hide_exec(bContext *C, wmOperator *op)
       if (gpl != layer) {
         gpl->flag |= GP_LAYER_HIDE;
       }
+      else {
+        /* Be sure the active layer is unhidden. */
+        gpl->flag &= ~GP_LAYER_HIDE;
+      }
     }
   }
   else {
@@ -1419,7 +1425,7 @@ static int gp_stroke_change_color_exec(bContext *C, wmOperator *op)
   bGPdata *gpd = ED_gpencil_data_get_active(C);
   Object *ob = CTX_data_active_object(C);
   if (name[0] == '\0') {
-    ma = give_current_material(ob, ob->actcol);
+    ma = BKE_material_gpencil_get(ob, ob->actcol);
   }
   else {
     ma = (Material *)BKE_libblock_find_name(bmain, ID_MA, name);
@@ -1539,9 +1545,10 @@ static int gp_stroke_lock_color_exec(bContext *C, wmOperator *UNUSED(op))
           }
           /* unlock color */
           Material *tmp_ma = give_current_material(ob, gps->mat_nr + 1);
-
-          tmp_ma->gp_style->flag &= ~GP_STYLE_COLOR_LOCKED;
-          DEG_id_tag_update(&tmp_ma->id, ID_RECALC_COPY_ON_WRITE);
+          if (tmp_ma) {
+            tmp_ma->gp_style->flag &= ~GP_STYLE_COLOR_LOCKED;
+            DEG_id_tag_update(&tmp_ma->id, ID_RECALC_COPY_ON_WRITE);
+          }
         }
       }
     }
@@ -2291,7 +2298,7 @@ int ED_gpencil_join_objects_exec(bContext *C, wmOperator *op)
         short *totcol = give_totcolp(ob_src);
 
         for (short i = 0; i < *totcol; i++) {
-          Material *tmp_ma = give_current_material(ob_src, i + 1);
+          Material *tmp_ma = BKE_material_gpencil_get(ob_src, i + 1);
           BKE_gpencil_object_material_ensure(bmain, ob_dst, tmp_ma);
         }
 
@@ -2325,7 +2332,7 @@ int ED_gpencil_join_objects_exec(bContext *C, wmOperator *op)
             for (bGPDstroke *gps = gpf->strokes.first; gps; gps = gps->next) {
 
               /* Reassign material. Look old material and try to find in destination. */
-              ma_src = give_current_material(ob_src, gps->mat_nr + 1);
+              ma_src = BKE_material_gpencil_get(ob_src, gps->mat_nr + 1);
               gps->mat_nr = BKE_gpencil_object_material_ensure(bmain, ob_dst, ma_src);
 
               bGPDspoint *pt;
@@ -2433,7 +2440,7 @@ static int gpencil_lock_layer_exec(bContext *C, wmOperator *UNUSED(op))
   }
 
   for (short i = 0; i < *totcol; i++) {
-    ma = give_current_material(ob, i + 1);
+    ma = BKE_material_gpencil_get(ob, i + 1);
     if (ma) {
       gp_style = ma->gp_style;
       gp_style->flag |= GP_STYLE_COLOR_LOCKED;
@@ -2453,7 +2460,7 @@ static int gpencil_lock_layer_exec(bContext *C, wmOperator *UNUSED(op))
           continue;
         }
 
-        ma = give_current_material(ob, gps->mat_nr + 1);
+        ma = BKE_material_gpencil_get(ob, gps->mat_nr + 1);
         DEG_id_tag_update(&ma->id, ID_RECALC_COPY_ON_WRITE);
 
         gp_style = ma->gp_style;
@@ -2496,7 +2503,7 @@ static int gpencil_color_isolate_exec(bContext *C, wmOperator *op)
 {
   bGPdata *gpd = ED_gpencil_data_get_active(C);
   Object *ob = CTX_data_active_object(C);
-  Material *active_ma = give_current_material(ob, ob->actcol);
+  Material *active_ma = BKE_material_gpencil_get(ob, ob->actcol);
   MaterialGPencilStyle *active_color = BKE_material_gpencil_settings_get(ob, ob->actcol);
   MaterialGPencilStyle *gp_style;
 
@@ -2516,7 +2523,7 @@ static int gpencil_color_isolate_exec(bContext *C, wmOperator *op)
   Material *ma = NULL;
   short *totcol = give_totcolp(ob);
   for (short i = 0; i < *totcol; i++) {
-    ma = give_current_material(ob, i + 1);
+    ma = BKE_material_gpencil_get(ob, i + 1);
     /* Skip if this is the active one */
     if ((ma == NULL) || (ma == active_ma)) {
       continue;
@@ -2536,7 +2543,7 @@ static int gpencil_color_isolate_exec(bContext *C, wmOperator *op)
   if (isolate) {
     /* Set flags on all "other" colors */
     for (short i = 0; i < *totcol; i++) {
-      ma = give_current_material(ob, i + 1);
+      ma = BKE_material_gpencil_get(ob, i + 1);
       if (ma == NULL) {
         continue;
       }
@@ -2553,7 +2560,7 @@ static int gpencil_color_isolate_exec(bContext *C, wmOperator *op)
   else {
     /* Clear flags - Restore everything else */
     for (short i = 0; i < *totcol; i++) {
-      ma = give_current_material(ob, i + 1);
+      ma = BKE_material_gpencil_get(ob, i + 1);
       if (ma == NULL) {
         continue;
       }
@@ -2618,7 +2625,7 @@ static int gpencil_color_hide_exec(bContext *C, wmOperator *op)
     /* hide unselected */
     MaterialGPencilStyle *color = NULL;
     for (short i = 0; i < *totcol; i++) {
-      ma = give_current_material(ob, i + 1);
+      ma = BKE_material_gpencil_get(ob, i + 1);
       if (ma) {
         color = ma->gp_style;
         if (active_color != color) {
@@ -2681,7 +2688,7 @@ static int gpencil_color_reveal_exec(bContext *C, wmOperator *UNUSED(op))
   MaterialGPencilStyle *gp_style = NULL;
 
   for (short i = 0; i < *totcol; i++) {
-    ma = give_current_material(ob, i + 1);
+    ma = BKE_material_gpencil_get(ob, i + 1);
     if (ma) {
       gp_style = ma->gp_style;
       gp_style->flag &= ~GP_STYLE_COLOR_HIDE;
@@ -2734,7 +2741,7 @@ static int gpencil_color_lock_all_exec(bContext *C, wmOperator *UNUSED(op))
   MaterialGPencilStyle *gp_style = NULL;
 
   for (short i = 0; i < *totcol; i++) {
-    ma = give_current_material(ob, i + 1);
+    ma = BKE_material_gpencil_get(ob, i + 1);
     if (ma) {
       gp_style = ma->gp_style;
       gp_style->flag |= GP_STYLE_COLOR_LOCKED;
@@ -2787,7 +2794,7 @@ static int gpencil_color_unlock_all_exec(bContext *C, wmOperator *UNUSED(op))
   MaterialGPencilStyle *gp_style = NULL;
 
   for (short i = 0; i < *totcol; i++) {
-    ma = give_current_material(ob, i + 1);
+    ma = BKE_material_gpencil_get(ob, i + 1);
     if (ma) {
       gp_style = ma->gp_style;
       gp_style->flag &= ~GP_STYLE_COLOR_LOCKED;
diff --git a/source/blender/editors/gpencil/gpencil_edit.c b/source/blender/editors/gpencil/gpencil_edit.c
index cc229fcb383..12b3792e36e 100644
--- a/source/blender/editors/gpencil/gpencil_edit.c
+++ b/source/blender/editors/gpencil/gpencil_edit.c
@@ -222,6 +222,17 @@ void GPENCIL_OT_editmode_toggle(wmOperatorType *ot)
 }
 
 /* set select mode */
+static bool gpencil_selectmode_toggle_poll(bContext *C)
+{
+  /* edit only supported with grease pencil objects */
+  Object *ob = CTX_data_active_object(C);
+  if ((ob == NULL) || (ob->type != OB_GPENCIL) || (ob->mode != OB_MODE_EDIT_GPENCIL)) {
+    return false;
+  }
+
+  return ED_operator_view3d_active(C);
+}
+
 static int gpencil_selectmode_toggle_exec(bContext *C, wmOperator *op)
 {
   Scene *scene = CTX_data_scene(C);
@@ -229,7 +240,7 @@ static int gpencil_selectmode_toggle_exec(bContext *C, wmOperator *op)
   const int mode = RNA_int_get(op->ptr, "mode");
 
   /* Just set mode */
-  ts->gpencil_selectmode = mode;
+  ts->gpencil_selectmode_edit = mode;
 
   WM_main_add_notifier(NC_SCENE | ND_TOOLSETTINGS, NULL);
   DEG_id_tag_update(&scene->id, ID_RECALC_COPY_ON_WRITE);
@@ -248,7 +259,7 @@ void GPENCIL_OT_selectmode_toggle(wmOperatorType *ot)
 
   /* callbacks */
   ot->exec = gpencil_selectmode_toggle_exec;
-  ot->poll = gp_strokes_edit3d_poll;
+  ot->poll = gpencil_selectmode_toggle_poll;
 
   /* flags */
   ot->flag = OPTYPE_UNDO | OPTYPE_REGISTER;
@@ -1180,10 +1191,15 @@ static int gp_strokes_copy_exec(bContext *C, wmOperator *op)
     GHash *ma_to_name = gp_strokes_copypastebuf_colors_material_to_name_create(bmain);
     for (bGPDstroke *gps = gp_strokes_copypastebuf.first; gps; gps = gps->next) {
       if (ED_gpencil_stroke_can_use(C, gps)) {
+        Material *ma = give_current_material(ob, gps->mat_nr + 1);
+        /* Avoid default material. */
+        if (ma == NULL) {
+          continue;
+        }
+
         char **ma_name_val;
         if (!BLI_ghash_ensure_p(
                 gp_strokes_copypastebuf_colors, &gps->mat_nr, (void ***)&ma_name_val)) {
-          Material *ma = give_current_material(ob, gps->mat_nr + 1);
           char *ma_name = BLI_ghash_lookup(ma_to_name, ma);
           *ma_name_val = MEM_dupallocN(ma_name);
         }
@@ -1230,8 +1246,8 @@ static bool gp_strokes_paste_poll(bContext *C)
 }
 
 typedef enum eGP_PasteMode {
-  GP_COPY_ONLY = -1,
-  GP_COPY_MERGE = 1,
+  GP_COPY_BY_LAYER = -1,
+  GP_COPY_TO_ACTIVE = 1,
 } eGP_PasteMode;
 
 static int gp_strokes_paste_exec(bContext *C, wmOperator *op)
@@ -1264,7 +1280,7 @@ static int gp_strokes_paste_exec(bContext *C, wmOperator *op)
     /* no active layer - let's just create one */
     gpl = BKE_gpencil_layer_addnew(gpd, DATA_("GP_Layer"), true);
   }
-  else if ((gpencil_layer_is_editable(gpl) == false) && (type == GP_COPY_MERGE)) {
+  else if ((gpencil_layer_is_editable(gpl) == false) && (type == GP_COPY_TO_ACTIVE)) {
     BKE_report(
         op->reports, RPT_ERROR, "Can not paste strokes when active layer is hidden or locked");
     return OPERATOR_CANCELLED;
@@ -1317,7 +1333,7 @@ static int gp_strokes_paste_exec(bContext *C, wmOperator *op)
   for (bGPDstroke *gps = gp_strokes_copypastebuf.first; gps; gps = gps->next) {
     if (ED_gpencil_stroke_can_use(C, gps)) {
       /* Need to verify if layer exists */
-      if (type != GP_COPY_MERGE) {
+      if (type != GP_COPY_TO_ACTIVE) {
         gpl = BLI_findstring(&gpd->layers, gps->runtime.tmp_layerinfo, offsetof(bGPDlayer, info));
         if (gpl == NULL) {
           /* no layer - use active (only if layer deleted before paste) */
@@ -1350,7 +1366,7 @@ static int gp_strokes_paste_exec(bContext *C, wmOperator *op)
         /* Remap material */
         Material *ma = BLI_ghash_lookup(new_colors, POINTER_FROM_INT(new_stroke->mat_nr));
         new_stroke->mat_nr = BKE_gpencil_object_material_get_index(ob, ma);
-        BLI_assert(new_stroke->mat_nr >= 0); /* have to add the material first */
+        CLAMP_MIN(new_stroke->mat_nr, 0);
       }
     }
   }
@@ -1368,15 +1384,15 @@ static int gp_strokes_paste_exec(bContext *C, wmOperator *op)
 void GPENCIL_OT_paste(wmOperatorType *ot)
 {
   static const EnumPropertyItem copy_type[] = {
-      {GP_COPY_ONLY, "COPY", 0, "Copy", ""},
-      {GP_COPY_MERGE, "MERGE", 0, "Merge", ""},
+      {GP_COPY_TO_ACTIVE, "ACTIVE", 0, "Paste to Active", ""},
+      {GP_COPY_BY_LAYER, "LAYER", 0, "Paste by Layer", ""},
       {0, NULL, 0, NULL, NULL},
   };
 
   /* identifiers */
   ot->name = "Paste Strokes";
   ot->idname = "GPENCIL_OT_paste";
-  ot->description = "Paste previously copied strokes or copy and merge in active layer";
+  ot->description = "Paste previously copied strokes to active layer or to original layer";
 
   /* callbacks */
   ot->exec = gp_strokes_paste_exec;
@@ -1386,33 +1402,18 @@ void GPENCIL_OT_paste(wmOperatorType *ot)
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 
   /* properties */
-  ot->prop = RNA_def_enum(ot->srna, "type", copy_type, 0, "Type", "");
+  ot->prop = RNA_def_enum(ot->srna, "type", copy_type, GP_COPY_TO_ACTIVE, "Type", "");
 }
 
 /* ******************* Move To Layer ****************************** */
 
-static int gp_move_to_layer_invoke(bContext *C, wmOperator *op, const wmEvent *UNUSED(evt))
-{
-  uiPopupMenu *pup;
-  uiLayout *layout;
-
-  /* call the menu, which will call this operator again, hence the canceled */
-  pup = UI_popup_menu_begin(C, op->type->name, ICON_NONE);
-  layout = UI_popup_menu_layout(pup);
-  uiItemsEnumO(layout, "GPENCIL_OT_move_to_layer", "layer");
-  UI_popup_menu_end(C, pup);
-
-  return OPERATOR_INTERFACE;
-}
-
-// FIXME: allow moving partial strokes
 static int gp_move_to_layer_exec(bContext *C, wmOperator *op)
 {
   bGPdata *gpd = CTX_data_gpencil_data(C);
   Scene *scene = CTX_data_scene(C);
   bGPDlayer *target_layer = NULL;
   ListBase strokes = {NULL, NULL};
-  int layer_num = RNA_enum_get(op->ptr, "layer");
+  int layer_num = RNA_int_get(op->ptr, "layer");
   const bool use_autolock = (bool)(gpd->flag & GP_DATA_AUTOLOCK_LAYERS);
 
   if (GPENCIL_MULTIEDIT_SESSIONS_ON(gpd)) {
@@ -1425,23 +1426,16 @@ static int gp_move_to_layer_exec(bContext *C, wmOperator *op)
     gpd->flag &= ~GP_DATA_AUTOLOCK_LAYERS;
   }
 
-  /* Get layer or create new one */
-  if (layer_num == -1) {
-    /* Create layer */
-    target_layer = BKE_gpencil_layer_addnew(gpd, DATA_("GP_Layer"), true);
-  }
-  else {
-    /* Try to get layer */
-    target_layer = BLI_findlink(&gpd->layers, layer_num);
+  /* Try to get layer */
+  target_layer = BLI_findlink(&gpd->layers, layer_num);
 
-    if (target_layer == NULL) {
-      /* back autolock status */
-      if (use_autolock) {
-        gpd->flag |= GP_DATA_AUTOLOCK_LAYERS;
-      }
-      BKE_reportf(op->reports, RPT_ERROR, "There is no layer number %d", layer_num);
-      return OPERATOR_CANCELLED;
+  if (target_layer == NULL) {
+    /* back autolock status */
+    if (use_autolock) {
+      gpd->flag |= GP_DATA_AUTOLOCK_LAYERS;
     }
+    BKE_reportf(op->reports, RPT_ERROR, "There is no layer number %d", layer_num);
+    return OPERATOR_CANCELLED;
   }
 
   /* Extract all strokes to move to this layer
@@ -1509,16 +1503,14 @@ void GPENCIL_OT_move_to_layer(wmOperatorType *ot)
       "Move selected strokes to another layer";  // XXX: allow moving individual points too?
 
   /* callbacks */
-  ot->invoke = gp_move_to_layer_invoke;
   ot->exec = gp_move_to_layer_exec;
   ot->poll = gp_stroke_edit_poll;  // XXX?
 
   /* flags */
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 
-  /* gp layer to use (dynamic enum) */
-  ot->prop = RNA_def_enum(ot->srna, "layer", DummyRNA_DEFAULT_items, 0, "Grease Pencil Layer", "");
-  RNA_def_enum_funcs(ot->prop, ED_gpencil_layers_with_new_enum_itemf);
+  /* GPencil layer to use. */
+  ot->prop = RNA_def_int(ot->srna, "layer", 0, 0, INT_MAX, "Grease Pencil Layer", "", 0, INT_MAX);
 }
 
 /* ********************* Add Blank Frame *************************** */
@@ -3333,7 +3325,7 @@ typedef enum eGP_ReprojectModes {
   GP_REPROJECT_FRONT = 0,
   GP_REPROJECT_SIDE,
   GP_REPROJECT_TOP,
-  /* On same plane, parallel to viewplane */
+  /* On same plane, parallel to view-plane. */
   GP_REPROJECT_VIEW,
   /* Reprojected on to the scene geometry */
   GP_REPROJECT_SURFACE,
@@ -3344,11 +3336,14 @@ static int gp_strokes_reproject_exec(bContext *C, wmOperator *op)
 {
   bGPdata *gpd = ED_gpencil_data_get_active(C);
   Scene *scene = CTX_data_scene(C);
+  Main *bmain = CTX_data_main(C);
   ToolSettings *ts = CTX_data_tool_settings(C);
   Depsgraph *depsgraph = CTX_data_ensure_evaluated_depsgraph(C);
   Object *ob = CTX_data_active_object(C);
   ARegion *ar = CTX_wm_region(C);
   RegionView3D *rv3d = ar->regiondata;
+  SnapObjectContext *sctx = NULL;
+  int oldframe = (int)DEG_get_ctime(depsgraph);
 
   GP_SpaceConversion gsc = {NULL};
   eGP_ReprojectModes mode = RNA_enum_get(op->ptr, "type");
@@ -3358,28 +3353,33 @@ static int gp_strokes_reproject_exec(bContext *C, wmOperator *op)
   /* init space conversion stuff */
   gp_point_conversion_init(C, &gsc);
 
-  /* init autodist for geometry projection */
-  if (mode == GP_REPROJECT_SURFACE) {
-    view3d_region_operator_needs_opengl(CTX_wm_window(C), gsc.ar);
-    ED_view3d_autodist_init(depsgraph, gsc.ar, CTX_wm_view3d(C), 0);
-  }
-
-  // TODO: For deforming geometry workflow, create new frames?
+  int cfra_prv = INT_MIN;
+  /* init snap context for geometry projection */
+  sctx = ED_transform_snap_object_context_create_view3d(
+      bmain, scene, depsgraph, 0, ar, CTX_wm_view3d(C));
 
   /* Go through each editable + selected stroke, adjusting each of its points one by one... */
   GP_EDITABLE_STROKES_BEGIN (gpstroke_iter, C, gpl, gps) {
     if (gps->flag & GP_STROKE_SELECT) {
+
+      /* update frame to get the new location of objects */
+      if ((mode == GP_REPROJECT_SURFACE) && (cfra_prv != gpf_->framenum)) {
+        cfra_prv = gpf_->framenum;
+        CFRA = gpf_->framenum;
+        BKE_scene_graph_update_for_newframe(depsgraph, bmain);
+      }
+
       bGPDspoint *pt;
       int i;
       /* Adjust each point */
       for (i = 0, pt = gps->points; i < gps->totpoints; i++, pt++) {
         float xy[2];
 
-        /* 3D to Screenspace */
-        /* Note: We can't use gp_point_to_xy() here because that uses ints for the screenspace
-         *       coordinates, resulting in lost precision, which in turn causes stairstepping
-         *       artifacts in the final points.
-         */
+        /* 3D to Screen-space */
+        /* Note: We can't use gp_point_to_xy() here because that uses ints for the screen-space
+         * coordinates, resulting in lost precision, which in turn causes stair-stepping
+         * artifacts in the final points. */
+
         bGPDspoint pt2;
         gp_point_to_parent_space(pt, gpstroke_iter.diff_mat, &pt2);
         gp_point_to_xy_fl(&gsc, gps, &pt2, &xy[0], &xy[1]);
@@ -3428,26 +3428,34 @@ static int gp_strokes_reproject_exec(bContext *C, wmOperator *op)
           /* apply parent again */
           gp_apply_parent_point(depsgraph, ob, gpd, gpl, pt);
         }
-        /* Project screenspace back to 3D space (from current perspective)
-         * so that all points have been treated the same way
-         */
+        /* Project screen-space back to 3D space (from current perspective)
+         * so that all points have been treated the same way. */
         else if (mode == GP_REPROJECT_VIEW) {
-          /* Planar - All on same plane parallel to the viewplane */
+          /* Planar - All on same plane parallel to the view-plane. */
           gp_point_xy_to_3d(&gsc, scene, xy, &pt->x);
         }
         else {
           /* Geometry - Snap to surfaces of visible geometry */
-          /* XXX: There will be precision loss (possible stairstep artifacts)
-           * from this conversion to satisfy the API's */
-          const int screen_co[2] = {(int)xy[0], (int)xy[1]};
-
-          int depth_margin = 0;  // XXX: 4 for strokes, 0 for normal
-          float depth;
-
-          /* XXX: The proper procedure computes the depths into an array,
-           * to have smooth transitions when all else fails... */
-          if (ED_view3d_autodist_depth(gsc.ar, screen_co, depth_margin, &depth)) {
-            ED_view3d_autodist_simple(gsc.ar, screen_co, &pt->x, 0, &depth);
+          float ray_start[3];
+          float ray_normal[3];
+          /* magic value for initial depth copied from the default
+           * value of Python's Scene.ray_cast function
+           */
+          float depth = 1.70141e+38f;
+          float location[3] = {0.0f, 0.0f, 0.0f};
+          float normal[3] = {0.0f, 0.0f, 0.0f};
+
+          ED_view3d_win_to_ray(ar, xy, &ray_start[0], &ray_normal[0]);
+          if (ED_transform_snap_object_project_ray(sctx,
+                                                   &(const struct SnapObjectParams){
+                                                       .snap_select = SNAP_ALL,
+                                                   },
+                                                   &ray_start[0],
+                                                   &ray_normal[0],
+                                                   &depth,
+                                                   &location[0],
+                                                   &normal[0])) {
+            copy_v3_v3(&pt->x, location);
           }
           else {
             /* Default to planar */
@@ -3464,6 +3472,15 @@ static int gp_strokes_reproject_exec(bContext *C, wmOperator *op)
   }
   GP_EDITABLE_STROKES_END(gpstroke_iter);
 
+  /* return frame state and DB to original state */
+  CFRA = oldframe;
+  BKE_scene_graph_update_for_newframe(depsgraph, bmain);
+
+  if (sctx != NULL) {
+    ED_transform_snap_object_context_destroy(sctx);
+  }
+
+  /* update changed data */
   DEG_id_tag_update(&gpd->id, ID_RECALC_TRANSFORM | ID_RECALC_GEOMETRY);
   WM_event_add_notifier(C, NC_GPENCIL | ND_DATA | NA_EDITED, NULL);
   return OPERATOR_FINISHED;
@@ -3549,7 +3566,7 @@ static void gp_smooth_stroke(bContext *C, wmOperator *op)
           }
           if (smooth_thickness) {
             /* thickness need to repeat process several times */
-            for (int r2 = 0; r2 < r * 10; r2++) {
+            for (int r2 = 0; r2 < r * 20; r2++) {
               BKE_gpencil_smooth_stroke_thickness(gps, i, factor);
             }
           }
@@ -4055,7 +4072,7 @@ static int gp_stroke_separate_exec(bContext *C, wmOperator *op)
               /* add duplicate materials */
 
               /* XXX same material can be in multiple slots. */
-              ma = give_current_material(ob, gps->mat_nr + 1);
+              ma = BKE_material_gpencil_get(ob, gps->mat_nr + 1);
 
               idx = BKE_gpencil_object_material_ensure(bmain, ob_dst, ma);
 
@@ -4128,7 +4145,7 @@ static int gp_stroke_separate_exec(bContext *C, wmOperator *op)
           if (ED_gpencil_stroke_can_use(C, gps) == false) {
             continue;
           }
-          ma = give_current_material(ob, gps->mat_nr + 1);
+          ma = BKE_material_gpencil_get(ob, gps->mat_nr + 1);
           gps->mat_nr = BKE_gpencil_object_material_ensure(bmain, ob_dst, ma);
         }
       }
@@ -4305,7 +4322,7 @@ void GPENCIL_OT_stroke_smooth(wmOperatorType *ot)
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 
   /* properties */
-  prop = RNA_def_int(ot->srna, "repeat", 1, 1, 10, "Repeat", "", 1, 5);
+  prop = RNA_def_int(ot->srna, "repeat", 1, 1, 50, "Repeat", "", 1, 20);
   RNA_def_property_flag(prop, PROP_SKIP_SAVE);
 
   RNA_def_float(ot->srna, "factor", 0.5f, 0.0f, 2.0f, "Factor", "", 0.0f, 2.0f);
@@ -4441,7 +4458,7 @@ static int gpencil_cutter_lasso_select(bContext *C,
       if ((pt->flag & GP_SPOINT_SELECT) || (pt->flag & GP_SPOINT_TAG)) {
         continue;
       }
-      /* convert point coords to screenspace */
+      /* convert point coords to screen-space */
       const bool is_inside = is_inside_fn(gps, pt, &gsc, gpstroke_iter.diff_mat, user_data);
       if (is_inside) {
         tot_inside++;
diff --git a/source/blender/editors/gpencil/gpencil_fill.c b/source/blender/editors/gpencil/gpencil_fill.c
index 5637e755198..372973e1040 100644
--- a/source/blender/editors/gpencil/gpencil_fill.c
+++ b/source/blender/editors/gpencil/gpencil_fill.c
@@ -48,6 +48,7 @@
 #include "BKE_context.h"
 #include "BKE_screen.h"
 #include "BKE_paint.h"
+#include "BKE_report.h"
 
 #include "ED_gpencil.h"
 #include "ED_screen.h"
@@ -1038,7 +1039,15 @@ static void gpencil_stroke_from_buffer(tGPDfill *tgpf)
   gps->flag |= GP_STROKE_CYCLIC;
   gps->flag |= GP_STROKE_3DSPACE;
 
-  gps->mat_nr = BKE_gpencil_object_material_ensure(tgpf->bmain, tgpf->ob, tgpf->mat);
+  gps->mat_nr = BKE_gpencil_object_material_get_index_from_brush(tgpf->ob, brush);
+  if (gps->mat_nr < 0) {
+    if (tgpf->ob->actcol - 1 < 0) {
+      gps->mat_nr = 0;
+    }
+    else {
+      gps->mat_nr = tgpf->ob->actcol - 1;
+    }
+  }
 
   /* allocate memory for storage points */
   gps->totpoints = tgpf->sbuffer_used;
@@ -1346,8 +1355,28 @@ static int gpencil_fill_init(bContext *C, wmOperator *op)
 /* start of interactive part of operator */
 static int gpencil_fill_invoke(bContext *C, wmOperator *op, const wmEvent *UNUSED(event))
 {
+  Object *ob = CTX_data_active_object(C);
+  ToolSettings *ts = CTX_data_tool_settings(C);
+  Brush *brush = BKE_paint_brush(&ts->gp_paint->paint);
   tGPDfill *tgpf = NULL;
 
+  /* Fill tool needs a material (cannot use default material) */
+  bool valid = true;
+  if ((brush) && (brush->gpencil_settings->flag & GP_BRUSH_MATERIAL_PINNED)) {
+    if (brush->gpencil_settings->material == NULL) {
+      valid = false;
+    }
+  }
+  else {
+    if (give_current_material(ob, ob->actcol) == NULL) {
+      valid = false;
+    }
+  }
+  if (!valid) {
+    BKE_report(op->reports, RPT_ERROR, "Fill tool needs active material.");
+    return OPERATOR_CANCELLED;
+  }
+
   /* try to initialize context data needed */
   if (!gpencil_fill_init(C, op)) {
     gpencil_fill_exit(C, op);
diff --git a/source/blender/editors/gpencil/gpencil_intern.h b/source/blender/editors/gpencil/gpencil_intern.h
index 7f36eb50802..f7fc7e34460 100644
--- a/source/blender/editors/gpencil/gpencil_intern.h
+++ b/source/blender/editors/gpencil/gpencil_intern.h
@@ -637,8 +637,12 @@ struct GP_EditableStrokes_Iter {
   } \
   (void)0
 
+#define GPENCIL_ANY_SCULPT_MASK(flag) \
+  ((flag & (GP_SCULPT_MASK_SELECTMODE_POINT | GP_SCULPT_MASK_SELECTMODE_STROKE | \
+            GP_SCULPT_MASK_SELECTMODE_SEGMENT)))
+
 /**
- * Iterate over all editable strokes using derived data in the current context,
+ * Iterate over all editable strokes using evaluated data in the current context,
  * stopping on each usable layer + stroke pair (i.e. gpl and gps)
  * to perform some operations on the stroke.
  *
@@ -647,15 +651,15 @@ struct GP_EditableStrokes_Iter {
  * \param gps: The identifier to use for current stroke being processed.
  *                    Choose a suitable value to avoid name clashes.
  */
-#define GP_DERIVED_STROKES_BEGIN(gpstroke_iter, C, gpl, gps) \
+#define GP_EVALUATED_STROKES_BEGIN(gpstroke_iter, C, gpl, gps) \
   { \
     struct GP_EditableStrokes_Iter gpstroke_iter = {{{0}}}; \
-    Depsgraph *depsgraph_ = CTX_data_depsgraph_pointer(C); \
+    Depsgraph *depsgraph_ = CTX_data_ensure_evaluated_depsgraph(C); \
     Object *obact_ = CTX_data_active_object(C); \
     Object *obeval_ = DEG_get_evaluated_object(depsgraph_, obact_); \
     bGPdata *gpd_ = CTX_data_gpencil_data(C); \
     const bool is_multiedit_ = (bool)GPENCIL_MULTIEDIT_SESSIONS_ON(gpd_); \
-    int derived_idx = 0; \
+    int idx_eval = 0; \
     for (bGPDlayer *gpl = gpd->layers.first; gpl; gpl = gpl->next) { \
       if (gpencil_layer_is_editable(gpl)) { \
         bGPDframe *init_gpf_ = gpl->actframe; \
@@ -665,10 +669,11 @@ struct GP_EditableStrokes_Iter {
         for (bGPDframe *gpf_ = init_gpf_; gpf_; gpf_ = gpf_->next) { \
           if ((gpf_ == gpl->actframe) || ((gpf_->flag & GP_FRAME_SELECT) && is_multiedit_)) { \
             ED_gpencil_parent_location(depsgraph_, obact_, gpd_, gpl, gpstroke_iter.diff_mat); \
-            /* get derived frame with modifiers applied */ \
-            bGPDframe *derived_gpf_ = &obeval_->runtime.derived_frames[derived_idx]; \
+            invert_m4_m4(gpstroke_iter.inverse_diff_mat, gpstroke_iter.diff_mat); \
+            /* get evaluated frame with modifiers applied */ \
+            bGPDframe *gpf_eval_ = &obeval_->runtime.gpencil_evaluated_frames[idx_eval]; \
             /* loop over strokes */ \
-            for (bGPDstroke *gps = derived_gpf_->strokes.first; gps; gps = gps->next) { \
+            for (bGPDstroke *gps = gpf_eval_->strokes.first; gps; gps = gps->next) { \
               /* skip strokes that are invalid for current view */ \
               if (ED_gpencil_stroke_can_use(C, gps) == false) \
                 continue; \
@@ -677,7 +682,7 @@ struct GP_EditableStrokes_Iter {
                 continue; \
     /* ... Do Stuff With Strokes ...  */
 
-#define GP_DERIVED_STROKES_END(gpstroke_iter) \
+#define GP_EVALUATED_STROKES_END(gpstroke_iter) \
   } \
   } \
   if (!is_multiedit_) { \
@@ -685,7 +690,7 @@ struct GP_EditableStrokes_Iter {
   } \
   } \
   } \
-  derived_idx++; \
+  idx_eval++; \
   } \
   } \
   (void)0
diff --git a/source/blender/editors/gpencil/gpencil_merge.c b/source/blender/editors/gpencil/gpencil_merge.c
index cb11bb4cd63..930911ffac5 100644
--- a/source/blender/editors/gpencil/gpencil_merge.c
+++ b/source/blender/editors/gpencil/gpencil_merge.c
@@ -449,7 +449,7 @@ static bool gp_strokes_merge_poll(bContext *C)
 
   /* check material */
   Material *ma = NULL;
-  ma = give_current_material(ob, ob->actcol);
+  ma = BKE_material_gpencil_get(ob, ob->actcol);
   if ((ma == NULL) || (ma->gp_style == NULL)) {
     return false;
   }
diff --git a/source/blender/editors/gpencil/gpencil_paint.c b/source/blender/editors/gpencil/gpencil_paint.c
index a5425d64c2e..3217c94eebd 100644
--- a/source/blender/editors/gpencil/gpencil_paint.c
+++ b/source/blender/editors/gpencil/gpencil_paint.c
@@ -423,7 +423,7 @@ static void gp_stroke_convertcoords(tGPsdata *p, const float mval[2], float out[
 
     /* add small offset to keep stroke over the surface */
     if ((depth) && (gpd->zdepth_offset > 0.0f) && (*p->align_flag & GP_PROJECT_DEPTH_VIEW)) {
-      *depth *= (1.0f - gpd->zdepth_offset);
+      *depth *= (1.0f - (gpd->zdepth_offset / 1000.0f));
     }
 
     int mval_i[2];
@@ -563,7 +563,7 @@ static void gp_brush_angle(bGPdata *gpd, Brush *brush, tGPspoint *pt, const floa
 static void gp_smooth_buffer(tGPsdata *p, float inf, int idx)
 {
   bGPdata *gpd = p->gpd;
-  short num_points = gpd->runtime.sbuffer_used;
+  const short num_points = gpd->runtime.sbuffer_used;
 
   /* Do nothing if not enough points to smooth out */
   if ((num_points < 3) || (idx < 3) || (inf == 0.0f)) {
@@ -571,10 +571,7 @@ static void gp_smooth_buffer(tGPsdata *p, float inf, int idx)
   }
 
   tGPspoint *points = (tGPspoint *)gpd->runtime.sbuffer;
-  float steps = 4.0f;
-  if (idx < 4) {
-    steps--;
-  }
+  const float steps = (idx < 4) ? 3.0f : 4.0f;
 
   tGPspoint *pta = idx >= 4 ? &points[idx - 4] : NULL;
   tGPspoint *ptb = idx >= 3 ? &points[idx - 3] : NULL;
@@ -583,29 +580,36 @@ static void gp_smooth_buffer(tGPsdata *p, float inf, int idx)
 
   float sco[2] = {0.0f};
   float a[2], b[2], c[2], d[2];
+  float pressure = 0.0f;
   const float average_fac = 1.0f / steps;
 
   /* Compute smoothed coordinate by taking the ones nearby */
   if (pta) {
     copy_v2_v2(a, &pta->x);
     madd_v2_v2fl(sco, a, average_fac);
+    pressure += pta->pressure * average_fac;
   }
   if (ptb) {
     copy_v2_v2(b, &ptb->x);
     madd_v2_v2fl(sco, b, average_fac);
+    pressure += ptb->pressure * average_fac;
   }
   if (ptc) {
     copy_v2_v2(c, &ptc->x);
     madd_v2_v2fl(sco, c, average_fac);
+    pressure += ptc->pressure * average_fac;
   }
   if (ptd) {
     copy_v2_v2(d, &ptd->x);
     madd_v2_v2fl(sco, d, average_fac);
+    pressure += ptd->pressure * average_fac;
   }
 
-  /* Based on influence factor, blend between original and optimal smoothed coordinate */
+  /* Based on influence factor, blend between original and optimal smoothed coordinate. */
   interp_v2_v2v2(c, c, sco, inf);
   copy_v2_v2(&ptc->x, c);
+  /* Interpolate pressure. */
+  ptc->pressure = interpf(ptc->pressure, pressure, inf);
 }
 
 /* add current stroke-point to buffer (returns whether point was successfully added) */
@@ -1240,6 +1244,14 @@ static void gp_stroke_newfrombuffer(tGPsdata *p)
 
   /* Save material index */
   gps->mat_nr = BKE_gpencil_object_material_get_index_from_brush(p->ob, p->brush);
+  if (gps->mat_nr < 0) {
+    if (p->ob->actcol - 1 < 0) {
+      gps->mat_nr = 0;
+    }
+    else {
+      gps->mat_nr = p->ob->actcol - 1;
+    }
+  }
 
   /* calculate UVs along the stroke */
   ED_gpencil_calc_stroke_uv(obact, gps);
@@ -2128,15 +2140,6 @@ static void gp_paint_initstroke(tGPsdata *p, eGPencil_PaintModes paintmode, Deps
     /* Ensure this gets set... */
     p->gpf = p->gpl->actframe;
 
-    /* Restrict eraser to only affecting selected strokes, if the "selection mask" is on
-     * (though this is only available in editmode)
-     */
-    if (p->gpd->flag & GP_DATA_STROKE_EDITMODE) {
-      if (ts->gp_sculpt.flag & GP_SCULPT_SETT_FLAG_SELECT_MASK) {
-        p->flags |= GP_PAINTFLAG_SELECTMASK;
-      }
-    }
-
     if (has_layer_to_erase == false) {
       p->status = GP_STATUS_ERROR;
       return;
diff --git a/source/blender/editors/gpencil/gpencil_primitive.c b/source/blender/editors/gpencil/gpencil_primitive.c
index 25913fe821a..8d4c75d2e8c 100644
--- a/source/blender/editors/gpencil/gpencil_primitive.c
+++ b/source/blender/editors/gpencil/gpencil_primitive.c
@@ -330,7 +330,15 @@ static void gp_primitive_set_initdata(bContext *C, tGPDprimitive *tgpi)
 
   gps->flag |= GP_STROKE_3DSPACE;
 
-  gps->mat_nr = BKE_gpencil_object_material_get_index(tgpi->ob, tgpi->mat);
+  gps->mat_nr = BKE_gpencil_object_material_get_index_from_brush(tgpi->ob, tgpi->brush);
+  if (gps->mat_nr < 0) {
+    if (tgpi->ob->actcol - 1 < 0) {
+      gps->mat_nr = 0;
+    }
+    else {
+      gps->mat_nr = tgpi->ob->actcol - 1;
+    }
+  }
 
   /* allocate memory for storage points, but keep empty */
   gps->totpoints = 0;
@@ -959,7 +967,7 @@ static void gp_primitive_update_strokes(bContext *C, tGPDprimitive *tgpi)
 
     /* add small offset to keep stroke over the surface */
     if ((depth_arr) && (gpd->zdepth_offset > 0.0f)) {
-      depth_arr[i] *= (1.0f - gpd->zdepth_offset);
+      depth_arr[i] *= (1.0f - (gpd->zdepth_offset / 1000.0f));
     }
 
     /* convert screen-coordinates to 3D coordinates */
diff --git a/source/blender/editors/gpencil/gpencil_select.c b/source/blender/editors/gpencil/gpencil_select.c
index 01a2ac757d6..f4484624d5a 100644
--- a/source/blender/editors/gpencil/gpencil_select.c
+++ b/source/blender/editors/gpencil/gpencil_select.c
@@ -67,13 +67,30 @@
 /** \name Shared Utilities
  * \{ */
 
+/* Convert sculpt mask mode to Select mode */
+static int gpencil_select_mode_from_sculpt(eGP_Sculpt_SelectMaskFlag mode)
+{
+  if (mode & GP_SCULPT_MASK_SELECTMODE_POINT) {
+    return GP_SELECTMODE_POINT;
+  }
+  else if (mode & GP_SCULPT_MASK_SELECTMODE_STROKE) {
+    return GP_SELECTMODE_STROKE;
+  }
+  else if (mode & GP_SCULPT_MASK_SELECTMODE_SEGMENT) {
+    return GP_SELECTMODE_SEGMENT;
+  }
+  else {
+    return GP_SELECTMODE_POINT;
+  }
+}
+
 static bool gpencil_select_poll(bContext *C)
 {
   bGPdata *gpd = ED_gpencil_data_get_active(C);
 
   if (GPENCIL_SCULPT_MODE(gpd)) {
     ToolSettings *ts = CTX_data_tool_settings(C);
-    if ((ts->gp_sculpt.flag & GP_SCULPT_SETT_FLAG_SELECT_MASK) == 0) {
+    if (!(GPENCIL_ANY_SCULPT_MASK(ts->gpencil_selectmode_sculpt))) {
       return false;
     }
   }
@@ -94,6 +111,19 @@ static bool gpencil_select_poll(bContext *C)
 /* -------------------------------------------------------------------- */
 /** \name Select All Operator
  * \{ */
+static bool gpencil_select_all_poll(bContext *C)
+{
+  bGPdata *gpd = ED_gpencil_data_get_active(C);
+
+  /* we just need some visible strokes, and to be in editmode or other modes only to catch event */
+  if (GPENCIL_ANY_MODE(gpd)) {
+    if (gpd->layers.first) {
+      return true;
+    }
+  }
+
+  return false;
+}
 
 static int gpencil_select_all_exec(bContext *C, wmOperator *op)
 {
@@ -110,6 +140,14 @@ static int gpencil_select_all_exec(bContext *C, wmOperator *op)
     return OPERATOR_CANCELLED;
   }
 
+  /* For sculpt mode, if mask is disable, only allows deselect */
+  if (GPENCIL_SCULPT_MODE(gpd)) {
+    ToolSettings *ts = CTX_data_tool_settings(C);
+    if ((!(GPENCIL_ANY_SCULPT_MASK(ts->gpencil_selectmode_sculpt))) && (action != SEL_DESELECT)) {
+      return OPERATOR_CANCELLED;
+    }
+  }
+
   ED_gpencil_select_toggle_all(C, action);
 
   /* updates */
@@ -132,7 +170,7 @@ void GPENCIL_OT_select_all(wmOperatorType *ot)
 
   /* callbacks */
   ot->exec = gpencil_select_all_exec;
-  ot->poll = gpencil_select_poll;
+  ot->poll = gpencil_select_all_poll;
 
   /* flags */
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
@@ -928,7 +966,11 @@ static int gpencil_circle_select_exec(bContext *C, wmOperator *op)
 {
   bGPdata *gpd = ED_gpencil_data_get_active(C);
   ToolSettings *ts = CTX_data_tool_settings(C);
-  const int selectmode = ts->gpencil_selectmode;
+  Object *ob = CTX_data_active_object(C);
+
+  const int selectmode = (ob && ob->mode == OB_MODE_SCULPT_GPENCIL) ?
+                             gpencil_select_mode_from_sculpt(ts->gpencil_selectmode_sculpt) :
+                             ts->gpencil_selectmode_edit;
   const float scale = ts->gp_sculpt.isect_threshold;
 
   /* if not edit/sculpt mode, the event is catched but not processed */
@@ -972,12 +1014,12 @@ static int gpencil_circle_select_exec(bContext *C, wmOperator *op)
   rect.ymax = my + radius;
 
   /* find visible strokes, and select if hit */
-  GP_DERIVED_STROKES_BEGIN(gpstroke_iter, C, gpl, gps)
+  GP_EVALUATED_STROKES_BEGIN(gpstroke_iter, C, gpl, gps)
   {
     changed |= gp_stroke_do_circle_sel(
         gpl, gps, &gsc, mx, my, radius, select, &rect, gpstroke_iter.diff_mat, selectmode, scale);
   }
-  GP_DERIVED_STROKES_END(gpstroke_iter);
+  GP_EVALUATED_STROKES_END(gpstroke_iter);
 
   /* updates */
   if (changed) {
@@ -1035,12 +1077,18 @@ static int gpencil_generic_select_exec(bContext *C,
                                        GPencilTestFn is_inside_fn,
                                        void *user_data)
 {
+  Object *ob = CTX_data_active_object(C);
   bGPdata *gpd = ED_gpencil_data_get_active(C);
   ToolSettings *ts = CTX_data_tool_settings(C);
   ScrArea *sa = CTX_wm_area(C);
-  const bool strokemode = ((ts->gpencil_selectmode == GP_SELECTMODE_STROKE) &&
+
+  const short selectmode = (ob && ob->mode == OB_MODE_SCULPT_GPENCIL) ?
+                               gpencil_select_mode_from_sculpt(ts->gpencil_selectmode_sculpt) :
+                               ts->gpencil_selectmode_edit;
+
+  const bool strokemode = ((selectmode == GP_SELECTMODE_STROKE) &&
                            ((gpd->flag & GP_DATA_STROKE_PAINTMODE) == 0));
-  const bool segmentmode = ((ts->gpencil_selectmode == GP_SELECTMODE_SEGMENT) &&
+  const bool segmentmode = ((selectmode == GP_SELECTMODE_SEGMENT) &&
                             ((gpd->flag & GP_DATA_STROKE_PAINTMODE) == 0));
   const eSelectOp sel_op = RNA_enum_get(op->ptr, "mode");
   const float scale = ts->gp_sculpt.isect_threshold;
@@ -1075,7 +1123,7 @@ static int gpencil_generic_select_exec(bContext *C,
   }
 
   /* select/deselect points */
-  GP_DERIVED_STROKES_BEGIN(gpstroke_iter, C, gpl, gps)
+  GP_EVALUATED_STROKES_BEGIN(gpstroke_iter, C, gpl, gps)
   {
 
     bGPDspoint *pt;
@@ -1088,7 +1136,6 @@ static int gpencil_generic_select_exec(bContext *C,
 
       /* convert point coords to screenspace */
       const bool is_inside = is_inside_fn(gps, pt, &gsc, gpstroke_iter.diff_mat, user_data);
-
       if (strokemode == false) {
         const bool is_select = (pt->runtime.pt_orig->flag & GP_SPOINT_SELECT) != 0;
         const int sel_op_result = ED_select_op_action_deselected(sel_op, is_select, is_inside);
@@ -1143,7 +1190,7 @@ static int gpencil_generic_select_exec(bContext *C,
     /* Ensure that stroke selection is in sync with its points */
     BKE_gpencil_stroke_sync_selection(gps->runtime.gps_orig);
   }
-  GP_DERIVED_STROKES_END(gpstroke_iter);
+  GP_EVALUATED_STROKES_END(gpstroke_iter);
 
   /* if paint mode,delete selected points */
   if (gpd->flag & GP_DATA_STROKE_PAINTMODE) {
@@ -1318,6 +1365,7 @@ static void deselect_all_selected(bContext *C)
 static int gpencil_select_exec(bContext *C, wmOperator *op)
 {
   ScrArea *sa = CTX_wm_area(C);
+  Object *ob = CTX_data_active_object(C);
   bGPdata *gpd = ED_gpencil_data_get_active(C);
   ToolSettings *ts = CTX_data_tool_settings(C);
   const float scale = ts->gp_sculpt.isect_threshold;
@@ -1348,8 +1396,12 @@ static int gpencil_select_exec(bContext *C, wmOperator *op)
   }
 
   /* if select mode is stroke, use whole stroke */
-  if (ts->gpencil_selectmode == GP_SELECTMODE_STROKE) {
-    whole = true;
+  if ((ob) && (ob->mode == OB_MODE_SCULPT_GPENCIL)) {
+    whole = (bool)(gpencil_select_mode_from_sculpt(ts->gpencil_selectmode_sculpt) ==
+                   GP_SELECTMODE_STROKE);
+  }
+  else {
+    whole = (bool)(ts->gpencil_selectmode_edit == GP_SELECTMODE_STROKE);
   }
 
   /* init space conversion stuff */
@@ -1360,7 +1412,7 @@ static int gpencil_select_exec(bContext *C, wmOperator *op)
 
   /* First Pass: Find stroke point which gets hit */
   /* XXX: maybe we should go from the top of the stack down instead... */
-  GP_DERIVED_STROKES_BEGIN(gpstroke_iter, C, gpl, gps)
+  GP_EVALUATED_STROKES_BEGIN(gpstroke_iter, C, gpl, gps)
   {
     bGPDspoint *pt;
     int i;
@@ -1393,7 +1445,7 @@ static int gpencil_select_exec(bContext *C, wmOperator *op)
       }
     }
   }
-  GP_DERIVED_STROKES_END(gpstroke_iter);
+  GP_EVALUATED_STROKES_END(gpstroke_iter);
 
   /* Abort if nothing hit... */
   if (ELEM(NULL, hit_stroke, hit_point)) {
@@ -1454,7 +1506,11 @@ static int gpencil_select_exec(bContext *C, wmOperator *op)
       hit_stroke->flag |= GP_STROKE_SELECT;
 
       /* expand selection to segment */
-      if (ts->gpencil_selectmode == GP_SELECTMODE_SEGMENT) {
+      const short selectmode = (ob && ob->mode == OB_MODE_SCULPT_GPENCIL) ?
+                                   gpencil_select_mode_from_sculpt(ts->gpencil_selectmode_sculpt) :
+                                   ts->gpencil_selectmode_edit;
+
+      if (selectmode == GP_SELECTMODE_SEGMENT) {
         float r_hita[3], r_hitb[3];
         bool hit_select = (bool)(hit_point->flag & GP_SPOINT_SELECT);
         ED_gpencil_select_stroke_segment(
diff --git a/source/blender/editors/gpencil/gpencil_utils.c b/source/blender/editors/gpencil/gpencil_utils.c
index 2a0f16a4bbf..d7492b5be4d 100644
--- a/source/blender/editors/gpencil/gpencil_utils.c
+++ b/source/blender/editors/gpencil/gpencil_utils.c
@@ -401,17 +401,7 @@ const EnumPropertyItem *ED_gpencil_layers_with_new_enum_itemf(bContext *C,
 
   /* Create new layer */
   /* TODO: have some way of specifying that we don't want this? */
-  {
-    /* "New Layer" entry */
-    item_tmp.identifier = "__CREATE__";
-    item_tmp.name = "New Layer";
-    item_tmp.value = -1;
-    item_tmp.icon = ICON_ADD;
-    RNA_enum_item_add(&item, &totitem, &item_tmp);
 
-    /* separator */
-    RNA_enum_item_add_separator(&item, &totitem);
-  }
   const int tot = BLI_listbase_count(&gpd->layers);
   /* Existing layers */
   for (gpl = gpd->layers.last, i = 0; gpl; gpl = gpl->prev, i++) {
@@ -428,6 +418,17 @@ const EnumPropertyItem *ED_gpencil_layers_with_new_enum_itemf(bContext *C,
 
     RNA_enum_item_add(&item, &totitem, &item_tmp);
   }
+  {
+    /* separator */
+    RNA_enum_item_add_separator(&item, &totitem);
+
+    /* "New Layer" entry */
+    item_tmp.identifier = "__CREATE__";
+    item_tmp.name = "New Layer";
+    item_tmp.value = -1;
+    item_tmp.icon = ICON_ADD;
+    RNA_enum_item_add(&item, &totitem, &item_tmp);
+  }
 
   RNA_enum_item_end(&item, &totitem);
   *r_free = true;
@@ -2063,7 +2064,7 @@ void ED_gpencil_update_color_uv(Main *bmain, Material *mat)
               if (ED_gpencil_stroke_color_use(ob, gpl, gps) == false) {
                 continue;
               }
-              gps_ma = give_current_material(ob, gps->mat_nr + 1);
+              gps_ma = BKE_material_gpencil_get(ob, gps->mat_nr + 1);
               /* update */
               if ((gps_ma) && (gps_ma == mat)) {
                 ED_gpencil_calc_stroke_uv(ob, gps);
diff --git a/source/blender/editors/include/BIF_glutil.h b/source/blender/editors/include/BIF_glutil.h
index 101a65d151a..a49f02d28fe 100644
--- a/source/blender/editors/include/BIF_glutil.h
+++ b/source/blender/editors/include/BIF_glutil.h
@@ -24,7 +24,6 @@
 #ifndef __BIF_GLUTIL_H__
 #define __BIF_GLUTIL_H__
 
-struct rctf;
 struct rcti;
 
 struct ColorManagedDisplaySettings;
diff --git a/source/blender/editors/include/ED_armature.h b/source/blender/editors/include/ED_armature.h
index 6629eed8328..48d0a5fe8be 100644
--- a/source/blender/editors/include/ED_armature.h
+++ b/source/blender/editors/include/ED_armature.h
@@ -42,7 +42,6 @@ struct ReportList;
 struct Scene;
 struct UndoType;
 struct View3D;
-struct ViewContext;
 struct ViewLayer;
 struct bArmature;
 struct bContext;
@@ -75,12 +74,14 @@ typedef struct EditBone {
   int flag;
   int layer;
 
+  /* Envelope distance & weight */
   float dist, weight;
   /** put them in order! transform uses this as scale */
   float xwidth, length, zwidth;
   float rad_head, rad_tail;
 
   /* Bendy-Bone parameters */
+  short segments;
   float roll1, roll2;
   float curve_in_x, curve_in_y;
   float curve_out_x, curve_out_y;
@@ -91,8 +92,6 @@ typedef struct EditBone {
   /** for envelope scaling */
   float oldlength;
 
-  short segments;
-
   /** Type of next/prev bone handles */
   char bbone_prev_type;
   char bbone_next_type;
@@ -229,15 +228,9 @@ void ED_armature_edit_transform_mirror_update(struct Object *obedit);
 void ED_armature_origin_set(
     struct Main *bmain, struct Object *ob, const float cursor[3], int centermode, int around);
 
-void ED_armature_transform_bones(struct bArmature *arm, float mat[4][4], const bool do_props);
-void ED_armature_transform_apply(struct Main *bmain,
-                                 struct Object *ob,
-                                 float mat[4][4],
-                                 const bool do_props);
-void ED_armature_transform(struct Main *bmain,
-                           struct bArmature *arm,
-                           float mat[4][4],
-                           const bool do_props);
+void ED_armature_edit_transform(struct bArmature *arm, const float mat[4][4], const bool do_props);
+
+void ED_armature_transform(struct bArmature *arm, const float mat[4][4], const bool do_props);
 
 #define ARM_GROUPS_NAME 1
 #define ARM_GROUPS_ENVELOPE 2
diff --git a/source/blender/editors/include/ED_clip.h b/source/blender/editors/include/ED_clip.h
index 7cee5a358f4..22bf22e04d3 100644
--- a/source/blender/editors/include/ED_clip.h
+++ b/source/blender/editors/include/ED_clip.h
@@ -29,7 +29,6 @@ struct ImBuf;
 struct Main;
 struct Mask;
 struct MovieClip;
-struct Scene;
 struct SpaceClip;
 struct bContext;
 struct bScreen;
diff --git a/source/blender/editors/include/ED_fileselect.h b/source/blender/editors/include/ED_fileselect.h
index 7273f857a41..dc42237d804 100644
--- a/source/blender/editors/include/ED_fileselect.h
+++ b/source/blender/editors/include/ED_fileselect.h
@@ -34,17 +34,32 @@ struct wmWindowManager;
 #define FILE_LAYOUT_HOR 1
 #define FILE_LAYOUT_VER 2
 
-#define MAX_FILE_COLUMN 4
-
-typedef enum FileListColumns {
+typedef enum FileAttributeColumnType {
+  COLUMN_NONE = -1,
   COLUMN_NAME = 0,
-  COLUMN_DATE,
-  COLUMN_TIME,
+  COLUMN_DATETIME,
   COLUMN_SIZE,
-} FileListColumns;
+
+  ATTRIBUTE_COLUMN_MAX
+} FileAttributeColumnType;
+
+typedef struct FileAttributeColumn {
+  /** UI name for this column */
+  const char *name;
+
+  float width;
+  /* The sort type to use when sorting by this column. */
+  int sort_type; /* eFileSortType */
+
+  /* Alignment of column texts, header text is always left aligned */
+  int text_align; /* eFontStyle_Align */
+} FileAttributeColumn;
 
 typedef struct FileLayout {
   /* view settings - XXX - move into own struct */
+  int offset_top;
+  /* Height of the header for the different FileAttributeColumn's. */
+  int attribute_column_header_h;
   int prv_w;
   int prv_h;
   int tile_w;
@@ -54,13 +69,17 @@ typedef struct FileLayout {
   int prv_border_x;
   int prv_border_y;
   int rows;
-  int columns;
+  /* Those are the major layout columns the files are distributed across, not to be confused with
+   * 'attribute_columns' array below. */
+  int flow_columns;
   int width;
   int height;
   int flag;
   int dirty;
   int textheight;
-  float column_widths[MAX_FILE_COLUMN];
+  /* The columns for each item (name, modification date/time, size). Not to be confused with the
+   * 'flow_columns' above. */
+  FileAttributeColumn attribute_columns[ATTRIBUTE_COLUMN_MAX];
 
   /* When we change display size, we may have to update static strings like size of files... */
   short curr_size;
@@ -72,6 +91,7 @@ typedef struct FileSelection {
 } FileSelection;
 
 struct rcti;
+struct View2D;
 
 struct FileSelectParams *ED_fileselect_get_params(struct SpaceFile *sfile);
 
@@ -87,6 +107,17 @@ int ED_fileselect_layout_numfiles(FileLayout *layout, struct ARegion *ar);
 int ED_fileselect_layout_offset(FileLayout *layout, int x, int y);
 FileSelection ED_fileselect_layout_offset_rect(FileLayout *layout, const struct rcti *rect);
 
+void ED_fileselect_layout_maskrect(const FileLayout *layout,
+                                   const struct View2D *v2d,
+                                   struct rcti *r_rect);
+bool ED_fileselect_layout_is_inside_pt(const FileLayout *layout,
+                                       const struct View2D *v2d,
+                                       int x,
+                                       int y);
+bool ED_fileselect_layout_isect_rect(const FileLayout *layout,
+                                     const struct View2D *v2d,
+                                     const struct rcti *rect,
+                                     struct rcti *r_dst);
 void ED_fileselect_layout_tilepos(FileLayout *layout, int tile, int *x, int *y);
 
 void ED_operatormacros_file(void);
diff --git a/source/blender/editors/include/ED_gizmo_library.h b/source/blender/editors/include/ED_gizmo_library.h
index f14648bc026..a3e2e643509 100644
--- a/source/blender/editors/include/ED_gizmo_library.h
+++ b/source/blender/editors/include/ED_gizmo_library.h
@@ -43,10 +43,8 @@ void ED_gizmotypes_value_2d(void);
 void ED_gizmogrouptypes_value_2d(void);
 
 struct Object;
-struct Scene;
 struct bContext;
 struct wmGizmo;
-struct wmGizmoGroup;
 
 /* -------------------------------------------------------------------- */
 /* Shape Presets
diff --git a/source/blender/editors/include/ED_gpencil.h b/source/blender/editors/include/ED_gpencil.h
index 6410e6630a7..470eb58c72b 100644
--- a/source/blender/editors/include/ED_gpencil.h
+++ b/source/blender/editors/include/ED_gpencil.h
@@ -27,7 +27,6 @@
 struct ID;
 struct ListBase;
 struct PointerRNA;
-struct rcti;
 
 struct Brush;
 struct bGPDframe;
@@ -39,13 +38,11 @@ struct tGPspoint;
 
 struct ARegion;
 struct Depsgraph;
-struct EvaluationContext;
 struct Main;
 struct RegionView3D;
 struct ReportList;
 struct Scene;
 struct ScrArea;
-struct ToolSettings;
 struct View3D;
 struct ViewLayer;
 struct bContext;
@@ -58,8 +55,6 @@ struct bAnimContext;
 
 struct wmKeyConfig;
 struct wmOperator;
-struct wmWindow;
-struct wmWindowManager;
 
 /* ------------- Grease-Pencil Runtime Data ---------------- */
 
diff --git a/source/blender/editors/include/ED_image.h b/source/blender/editors/include/ED_image.h
index a09e1d579fd..7d69f86dbf8 100644
--- a/source/blender/editors/include/ED_image.h
+++ b/source/blender/editors/include/ED_image.h
@@ -31,7 +31,6 @@ struct ImageUser;
 struct ReportList;
 struct Scene;
 struct SpaceImage;
-struct ToolSettings;
 struct ViewLayer;
 struct bContext;
 struct wmWindowManager;
diff --git a/source/blender/editors/include/ED_keyframes_draw.h b/source/blender/editors/include/ED_keyframes_draw.h
index 05f641abe91..2630feaa9d3 100644
--- a/source/blender/editors/include/ED_keyframes_draw.h
+++ b/source/blender/editors/include/ED_keyframes_draw.h
@@ -31,7 +31,6 @@ struct FCurve;
 struct ListBase;
 struct MaskLayer;
 struct Object;
-struct Palette;
 struct Scene;
 struct View2D;
 struct bAction;
diff --git a/source/blender/editors/include/ED_keyframing.h b/source/blender/editors/include/ED_keyframing.h
index bbeeeade822..42e5add2ef0 100644
--- a/source/blender/editors/include/ED_keyframing.h
+++ b/source/blender/editors/include/ED_keyframing.h
@@ -44,8 +44,6 @@ struct bPoseChannel;
 struct ReportList;
 struct bContext;
 
-struct Depsgraph;
-
 struct EnumPropertyItem;
 struct PointerRNA;
 struct PropertyRNA;
diff --git a/source/blender/editors/include/ED_markers.h b/source/blender/editors/include/ED_markers.h
index a883187327f..a9e57ecc13d 100644
--- a/source/blender/editors/include/ED_markers.h
+++ b/source/blender/editors/include/ED_markers.h
@@ -29,7 +29,6 @@ struct TimeMarker;
 struct bAnimContext;
 struct bContext;
 struct wmKeyConfig;
-struct wmKeyMap;
 
 /* Drawing API ------------------------------ */
 
diff --git a/source/blender/editors/include/ED_mesh.h b/source/blender/editors/include/ED_mesh.h
index 5ebba4e5d25..ee4798430ef 100644
--- a/source/blender/editors/include/ED_mesh.h
+++ b/source/blender/editors/include/ED_mesh.h
@@ -45,7 +45,6 @@ struct Mesh;
 struct Object;
 struct ReportList;
 struct Scene;
-struct ToolSettings;
 struct UndoType;
 struct UvMapVert;
 struct UvMapVert;
@@ -55,7 +54,6 @@ struct View3D;
 struct ViewContext;
 struct bContext;
 struct bDeformGroup;
-struct rcti;
 struct wmKeyConfig;
 struct wmOperator;
 
@@ -105,7 +103,9 @@ bool EDBM_vert_color_check(struct BMEditMesh *em);
 bool EDBM_mesh_hide(struct BMEditMesh *em, bool swap);
 bool EDBM_mesh_reveal(struct BMEditMesh *em, bool select);
 
-void EDBM_update_generic(struct BMEditMesh *em, const bool do_tessface, const bool is_destructive);
+void EDBM_update_generic(struct BMEditMesh *em,
+                         const bool do_tessellation,
+                         const bool is_destructive);
 
 struct UvElementMap *BM_uv_element_map_create(struct BMesh *bm,
                                               const bool selected,
@@ -138,13 +138,21 @@ bool BMBVH_EdgeVisible(struct BMBVHTree *tree,
                        struct View3D *v3d,
                        struct Object *obedit);
 
+/* editmesh_automerge.c */
+void EDBM_automerge(struct Object *ob, bool update, const char hflag, const float dist);
+void EDBM_automerge_and_split(struct Object *ob,
+                              bool split_edges,
+                              bool split_faces,
+                              bool update,
+                              const char hflag,
+                              const float dist);
+
 /* editmesh_undo.c */
 void ED_mesh_undosys_type(struct UndoType *ut);
 
 /* editmesh_select.c */
 void EDBM_select_mirrored(
     struct BMEditMesh *em, const int axis, const bool extend, int *r_totmirr, int *r_totfail);
-void EDBM_automerge(struct Scene *scene, struct Object *ob, bool update, const char hflag);
 
 struct BMVert *EDBM_vert_find_nearest_ex(struct ViewContext *vc,
                                          float *r_dist,
@@ -188,8 +196,11 @@ bool EDBM_unified_findnearest(struct ViewContext *vc,
 bool EDBM_unified_findnearest_from_raycast(struct ViewContext *vc,
                                            struct Base **bases,
                                            const uint bases_len,
-                                           bool use_boundary,
-                                           int *r_base_index,
+                                           bool use_boundary_vertices,
+                                           bool use_boundary_edges,
+                                           int *r_base_index_vert,
+                                           int *r_base_index_edge,
+                                           int *r_base_index_face,
                                            struct BMVert **r_eve,
                                            struct BMEdge **r_eed,
                                            struct BMFace **r_efa);
@@ -239,15 +250,30 @@ void EDBM_preselect_edgering_update_from_edge(struct EditMesh_PreSelEdgeRing *ps
 
 /* editmesh_preselect_elem.c */
 struct EditMesh_PreSelElem;
+typedef enum eEditMesh_PreSelPreviewAction {
+  PRESELECT_ACTION_TRANSFORM = 1,
+  PRESELECT_ACTION_CREATE = 2,
+  PRESELECT_ACTION_DELETE = 3,
+} eEditMesh_PreSelPreviewAction;
+
 struct EditMesh_PreSelElem *EDBM_preselect_elem_create(void);
 void EDBM_preselect_elem_destroy(struct EditMesh_PreSelElem *psel);
 void EDBM_preselect_elem_clear(struct EditMesh_PreSelElem *psel);
+void EDBM_preselect_preview_clear(struct EditMesh_PreSelElem *psel);
 void EDBM_preselect_elem_draw(struct EditMesh_PreSelElem *psel, const float matrix[4][4]);
 void EDBM_preselect_elem_update_from_single(struct EditMesh_PreSelElem *psel,
                                             struct BMesh *bm,
                                             struct BMElem *ele,
                                             const float (*coords)[3]);
 
+void EDBM_preselect_elem_update_preview(struct EditMesh_PreSelElem *psel,
+                                        struct ViewContext *vc,
+                                        struct BMesh *bm,
+                                        struct BMElem *ele,
+                                        const int mval[2]);
+void EDBM_preselect_action_set(struct EditMesh_PreSelElem *psel,
+                               eEditMesh_PreSelPreviewAction action);
+eEditMesh_PreSelPreviewAction EDBM_preselect_action_get(struct EditMesh_PreSelElem *psel);
 /* mesh_ops.c */
 void ED_operatortypes_mesh(void);
 void ED_operatormacros_mesh(void);
@@ -358,21 +384,14 @@ void ED_mesh_geometry_add(
     struct Mesh *mesh, struct ReportList *reports, int verts, int edges, int faces);
 #endif
 void ED_mesh_polys_add(struct Mesh *mesh, struct ReportList *reports, int count);
-void ED_mesh_tessfaces_add(struct Mesh *mesh, struct ReportList *reports, int count);
 void ED_mesh_edges_add(struct Mesh *mesh, struct ReportList *reports, int count);
 void ED_mesh_loops_add(struct Mesh *mesh, struct ReportList *reports, int count);
 void ED_mesh_vertices_add(struct Mesh *mesh, struct ReportList *reports, int count);
 
-void ED_mesh_faces_remove(struct Mesh *mesh, struct ReportList *reports, int count);
 void ED_mesh_edges_remove(struct Mesh *mesh, struct ReportList *reports, int count);
 void ED_mesh_vertices_remove(struct Mesh *mesh, struct ReportList *reports, int count);
 
-void ED_mesh_calc_tessface(struct Mesh *mesh, bool free_mpoly);
-void ED_mesh_update(struct Mesh *mesh,
-                    struct bContext *C,
-                    bool calc_edges,
-                    bool calc_edges_loose,
-                    bool calc_tessface);
+void ED_mesh_update(struct Mesh *mesh, struct bContext *C, bool calc_edges, bool calc_edges_loose);
 
 void ED_mesh_uv_texture_ensure(struct Mesh *me, const char *name);
 int ED_mesh_uv_texture_add(struct Mesh *me,
diff --git a/source/blender/editors/include/ED_object.h b/source/blender/editors/include/ED_object.h
index 7b9a96e4d07..c481c19a552 100644
--- a/source/blender/editors/include/ED_object.h
+++ b/source/blender/editors/include/ED_object.h
@@ -34,7 +34,6 @@ struct EnumPropertyItem;
 struct EnumPropertyItem;
 struct ID;
 struct Main;
-struct Menu;
 struct ModifierData;
 struct Object;
 struct PointerRNA;
@@ -44,16 +43,15 @@ struct Scene;
 struct ShaderFxData;
 struct View3D;
 struct ViewLayer;
+struct XFormObjectData;
 struct bConstraint;
 struct bContext;
 struct bFaceMap;
 struct bPoseChannel;
 struct uiLayout;
 struct wmKeyConfig;
-struct wmKeyMap;
 struct wmOperator;
 struct wmOperatorType;
-struct wmWindow;
 struct wmWindowManager;
 
 #include "DNA_object_enums.h"
@@ -403,6 +401,12 @@ bool ED_object_jump_to_bone(struct bContext *C,
 void ED_object_facemap_face_add(struct Object *ob, struct bFaceMap *fmap, int facenum);
 void ED_object_facemap_face_remove(struct Object *ob, struct bFaceMap *fmap, int facenum);
 
+/* object_data_transform.c */
+struct XFormObjectData *ED_object_data_xform_create(struct ID *id);
+void ED_object_data_xform_destroy(struct XFormObjectData *xod);
+
+void ED_object_data_xform_by_mat4(struct XFormObjectData *xod, const float mat[4][4]);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/source/blender/editors/include/ED_outliner.h b/source/blender/editors/include/ED_outliner.h
index e94aedc2b2b..3015d8d9cdc 100644
--- a/source/blender/editors/include/ED_outliner.h
+++ b/source/blender/editors/include/ED_outliner.h
@@ -30,4 +30,18 @@ bool ED_outliner_collections_editor_poll(struct bContext *C);
 
 void ED_outliner_selected_objects_get(const struct bContext *C, struct ListBase *objects);
 
+Base *ED_outliner_give_base_under_cursor(struct bContext *C, const int mval[2]);
+
+void ED_outliner_select_sync_from_object_tag(struct bContext *C);
+void ED_outliner_select_sync_from_edit_bone_tag(struct bContext *C);
+void ED_outliner_select_sync_from_pose_bone_tag(struct bContext *C);
+void ED_outliner_select_sync_from_sequence_tag(struct bContext *C);
+void ED_outliner_select_sync_from_all_tag(struct bContext *C);
+
+bool ED_outliner_select_sync_is_dirty(const struct bContext *C);
+
+void ED_outliner_select_sync_from_outliner(struct bContext *C, struct SpaceOutliner *soops);
+
+void ED_outliner_select_sync_flag_outliners(const struct bContext *C);
+
 #endif /*  __ED_OUTLINER_H__ */
diff --git a/source/blender/editors/include/ED_screen.h b/source/blender/editors/include/ED_screen.h
index c7ee7be49b5..d0fab134dcc 100644
--- a/source/blender/editors/include/ED_screen.h
+++ b/source/blender/editors/include/ED_screen.h
@@ -39,10 +39,8 @@ struct Depsgraph;
 struct IDProperty;
 struct Main;
 struct MenuType;
-struct PropertyRNA;
 struct Scene;
 struct SpaceLink;
-struct ViewLayer;
 struct WorkSpace;
 struct WorkSpaceInstanceHook;
 struct bContext;
@@ -50,7 +48,6 @@ struct bScreen;
 struct rcti;
 struct uiBlock;
 struct uiLayout;
-struct wmEvent;
 struct wmKeyConfig;
 struct wmMsgBus;
 struct wmMsgSubscribeKey;
diff --git a/source/blender/editors/include/ED_sculpt.h b/source/blender/editors/include/ED_sculpt.h
index 034e002f86a..d907ba4e581 100644
--- a/source/blender/editors/include/ED_sculpt.h
+++ b/source/blender/editors/include/ED_sculpt.h
@@ -25,10 +25,7 @@
 #define __ED_SCULPT_H__
 
 struct ARegion;
-struct ListBase;
 struct Object;
-struct RegionView3D;
-struct UndoStep;
 struct UndoType;
 struct ViewContext;
 struct bContext;
diff --git a/source/blender/editors/include/ED_time_scrub_ui.h b/source/blender/editors/include/ED_time_scrub_ui.h
index b43e674224c..f010c45d939 100644
--- a/source/blender/editors/include/ED_time_scrub_ui.h
+++ b/source/blender/editors/include/ED_time_scrub_ui.h
@@ -24,7 +24,6 @@
 #ifndef __ED_TIME_SCRUB_UI_H__
 #define __ED_TIME_SCRUB_UI_H__
 
-struct View2DGrid;
 struct bContext;
 struct bDopeSheet;
 struct wmEvent;
diff --git a/source/blender/editors/include/ED_transform.h b/source/blender/editors/include/ED_transform.h
index 3605a245187..d8b65aa5975 100644
--- a/source/blender/editors/include/ED_transform.h
+++ b/source/blender/editors/include/ED_transform.h
@@ -26,18 +26,11 @@
 
 /* ******************* Registration Function ********************** */
 
-struct ARegion;
-struct ListBase;
-struct Main;
 struct Object;
 struct SnapObjectContext;
 struct SnapObjectParams;
-struct View3D;
-struct WorkSpace;
 struct bContext;
-struct wmEvent;
 struct wmKeyConfig;
-struct wmKeyMap;
 struct wmOperatorType;
 
 void ED_keymap_transform(struct wmKeyConfig *keyconf);
@@ -97,6 +90,10 @@ enum TfmMode {
 #define CTX_PAINT_CURVE (1 << 8)
 #define CTX_GPENCIL_STROKES (1 << 9)
 #define CTX_CURSOR (1 << 10)
+/** When transforming object's, adjust the object data so it stays in the same place. */
+#define CTX_OBMODE_XFORM_OBDATA (1 << 11)
+/** Transform object parents without moving their children. */
+#define CTX_OBMODE_XFORM_SKIP_CHILDREN (1 << 12)
 
 /* Standalone call to get the transformation center corresponding to the current situation
  * returns 1 if successful, 0 otherwise (usually means there's no selection)
@@ -112,7 +109,6 @@ struct Scene;
 struct TransInfo;
 struct wmGizmoGroup;
 struct wmGizmoGroupType;
-struct wmOperator;
 
 /* UNUSED */
 // int BIF_snappingSupported(struct Object *obedit);
@@ -197,12 +193,12 @@ bool peelObjectsSnapContext(struct SnapObjectContext *sctx,
                             float r_no[3],
                             float *r_thickness);
 
-bool snapObjectsTransform(struct TransInfo *t,
-                          const float mval[2],
-                          float *dist_px,
-                          /* return args */
-                          float r_loc[3],
-                          float r_no[3]);
+short snapObjectsTransform(struct TransInfo *t,
+                           const float mval[2],
+                           float *dist_px,
+                           /* return args */
+                           float r_loc[3],
+                           float r_no[3]);
 bool snapNodesTransform(struct TransInfo *t,
                         const int mval[2],
                         /* return args */
diff --git a/source/blender/editors/include/ED_transform_snap_object_context.h b/source/blender/editors/include/ED_transform_snap_object_context.h
index f27523bb1f8..40e0005b487 100644
--- a/source/blender/editors/include/ED_transform_snap_object_context.h
+++ b/source/blender/editors/include/ED_transform_snap_object_context.h
@@ -32,8 +32,6 @@ struct Main;
 struct Object;
 struct Scene;
 struct View3D;
-struct ViewLayer;
-struct bContext;
 
 /* transform_snap_object.c */
 
@@ -125,6 +123,7 @@ short ED_transform_snap_object_project_view3d_ex(struct SnapObjectContext *sctx,
                                                  const unsigned short snap_to,
                                                  const struct SnapObjectParams *params,
                                                  const float mval[2],
+                                                 const float prev_co[3],
                                                  float *dist_px,
                                                  float r_loc[3],
                                                  float r_no[3],
@@ -135,6 +134,7 @@ bool ED_transform_snap_object_project_view3d(struct SnapObjectContext *sctx,
                                              const unsigned short snap_to,
                                              const struct SnapObjectParams *params,
                                              const float mval[2],
+                                             const float prev_co[3],
                                              float *dist_px,
                                              /* return args */
                                              float r_loc[3],
diff --git a/source/blender/editors/include/ED_uvedit.h b/source/blender/editors/include/ED_uvedit.h
index 8dd2dab5209..1856ad8454b 100644
--- a/source/blender/editors/include/ED_uvedit.h
+++ b/source/blender/editors/include/ED_uvedit.h
@@ -37,7 +37,6 @@ struct Object;
 struct Scene;
 struct SpaceImage;
 struct ToolSettings;
-struct View3D;
 struct ViewLayer;
 struct bNode;
 struct wmKeyConfig;
diff --git a/source/blender/editors/include/ED_view3d.h b/source/blender/editors/include/ED_view3d.h
index 67dfb184d19..139b306b533 100644
--- a/source/blender/editors/include/ED_view3d.h
+++ b/source/blender/editors/include/ED_view3d.h
@@ -38,8 +38,6 @@ struct Camera;
 struct CustomData_MeshMasks;
 struct Depsgraph;
 struct EditBone;
-struct GPUFX;
-struct GPUFXSettings;
 struct GPUOffScreen;
 struct GPUViewport;
 struct ID;
@@ -58,7 +56,6 @@ struct View3D;
 struct View3DShading;
 struct ViewContext;
 struct ViewLayer;
-struct WorkSpace;
 struct bContext;
 struct bPoseChannel;
 struct bScreen;
@@ -690,6 +687,10 @@ bool ED_view3d_distance_set_from_location(struct RegionView3D *rv3d,
 
 float ED_scene_grid_scale(struct Scene *scene, const char **grid_unit);
 float ED_view3d_grid_scale(struct Scene *scene, struct View3D *v3d, const char **grid_unit);
+void ED_view3d_grid_steps(struct Scene *scene,
+                          struct View3D *v3d,
+                          struct RegionView3D *rv3d,
+                          float *r_grid_steps);
 float ED_view3d_grid_view_scale(struct Scene *scene,
                                 struct View3D *v3d,
                                 struct RegionView3D *rv3d,
diff --git a/source/blender/editors/include/UI_icons.h b/source/blender/editors/include/UI_icons.h
index 03a49943ed3..41ac1b6b452 100644
--- a/source/blender/editors/include/UI_icons.h
+++ b/source/blender/editors/include/UI_icons.h
@@ -483,7 +483,7 @@ DEF_ICON(UNDERLINE)
 DEF_ICON(SMALL_CAPS)
 DEF_ICON_BLANK(742)
 DEF_ICON_BLANK(743)
-DEF_ICON_BLANK(744)
+DEF_ICON_MODIFIER(CON_ACTION)
 DEF_ICON_BLANK(745)
 DEF_ICON_BLANK(746)
 DEF_ICON_BLANK(747)
@@ -667,8 +667,8 @@ DEF_ICON(PARTICLE_PATH)
 /* EDITING */
 DEF_ICON_BLANK(669)
 DEF_ICON_BLANK(670)
-DEF_ICON_BLANK(671)
-DEF_ICON_BLANK(672)
+DEF_ICON(SNAP_PERPENDICULAR)
+DEF_ICON(SNAP_MIDPOINT)
 DEF_ICON(SNAP_OFF)
 DEF_ICON(SNAP_ON)
 DEF_ICON(SNAP_NORMAL)
@@ -866,8 +866,8 @@ DEF_ICON(SEQ_LUMA_WAVEFORM)
 DEF_ICON(SEQ_CHROMA_SCOPE)
 DEF_ICON(SEQ_HISTOGRAM)
 DEF_ICON(SEQ_SPLITVIEW)
-DEF_ICON_BLANK(870)
-DEF_ICON_BLANK(871)
+DEF_ICON(SEQ_STRIP_META)
+DEF_ICON(SEQ_STRIP_DUPLICATE)
 DEF_ICON_BLANK(872)
 DEF_ICON(IMAGE_RGB)  // XXX CHANGE TO STRAIGHT ALPHA, Z ETC
 DEF_ICON(IMAGE_RGB_ALPHA)
diff --git a/source/blender/editors/include/UI_interface.h b/source/blender/editors/include/UI_interface.h
index 5ef3e5d8987..aa1044647c8 100644
--- a/source/blender/editors/include/UI_interface.h
+++ b/source/blender/editors/include/UI_interface.h
@@ -31,7 +31,6 @@
 /* Struct Declarations */
 
 struct ARegion;
-struct ARegionType;
 struct AutoComplete;
 struct ID;
 struct IDProperty;
@@ -90,7 +89,7 @@ typedef struct uiPopupBlockHandle uiPopupBlockHandle;
  * For #ARegion.overlap regions, pass events though if they don't overlap
  * the regions contents (the usable part of the #View2D and buttons).
  *
- * The margin is needed so it's not possible to accidentally click inbetween buttons.
+ * The margin is needed so it's not possible to accidentally click in between buttons.
  */
 #define UI_REGION_OVERLAP_MARGIN (U.widget_unit / 3)
 
diff --git a/source/blender/editors/include/UI_resources.h b/source/blender/editors/include/UI_resources.h
index 1db584f0f59..a73e2655b7d 100644
--- a/source/blender/editors/include/UI_resources.h
+++ b/source/blender/editors/include/UI_resources.h
@@ -255,6 +255,7 @@ typedef enum ThemeColorID {
 
   TH_MATCH,            /* highlight color for search matches */
   TH_SELECT_HIGHLIGHT, /* highlight color for selected outliner item */
+  TH_SELECT_ACTIVE,    /* highlight color for active outliner item */
   TH_SELECTED_OBJECT,  /* selected object color for outliner */
   TH_ACTIVE_OBJECT,    /* active object color for outliner */
   TH_EDITED_OBJECT,    /* edited object color for outliner */
diff --git a/source/blender/editors/interface/interface.c b/source/blender/editors/interface/interface.c
index 7a123599be5..ee354df3a25 100644
--- a/source/blender/editors/interface/interface.c
+++ b/source/blender/editors/interface/interface.c
@@ -908,12 +908,12 @@ void UI_but_execute(const bContext *C, ARegion *ar, uiBut *but)
  * returns false if undo needs to be disabled. */
 static bool ui_but_is_rna_undo(const uiBut *but)
 {
-  if (but->rnapoin.id.data) {
+  if (but->rnapoin.owner_id) {
     /* avoid undo push for buttons who's ID are screen or wm level
      * we could disable undo for buttons with no ID too but may have
      * unforeseen consequences, so best check for ID's we _know_ are not
      * handled by undo - campbell */
-    ID *id = but->rnapoin.id.data;
+    ID *id = but->rnapoin.owner_id;
     if (ID_CHECK_UNDO(id) == false) {
       return false;
     }
@@ -1226,8 +1226,8 @@ static bool ui_but_event_property_operator_string(const bContext *C,
      */
     char *data_path = NULL;
 
-    if (ptr->id.data) {
-      ID *id = ptr->id.data;
+    if (ptr->owner_id) {
+      ID *id = ptr->owner_id;
 
       if (GS(id->name) == ID_SCR) {
         /* screen/editor property
@@ -1731,7 +1731,7 @@ static void ui_block_message_subscribe(ARegion *ar, struct wmMsgBus *mbus, uiBlo
       if ((but_prev && (but_prev->rnaprop == but->rnaprop) &&
            (but_prev->rnapoin.type == but->rnapoin.type) &&
            (but_prev->rnapoin.data == but->rnapoin.data) &&
-           (but_prev->rnapoin.id.data == but->rnapoin.id.data)) == false) {
+           (but_prev->rnapoin.owner_id == but->rnapoin.owner_id)) == false) {
         /* TODO: could make this into utility function. */
         WM_msg_subscribe_rna(mbus,
                              &but->rnapoin,
@@ -2055,7 +2055,7 @@ bool ui_but_is_compatible(const uiBut *but_a, const uiBut *but_b)
   }
 
   if (but_a->rnaprop) {
-    /* skip 'rnapoin.data', 'rnapoin.id.data'
+    /* skip 'rnapoin.data', 'rnapoin.owner_id'
      * allow different data to have the same props edited at once */
     if (but_a->rnapoin.type != but_b->rnapoin.type) {
       return false;
@@ -2479,7 +2479,7 @@ void ui_but_string_get_ex(uiBut *but,
 
       /* uiBut.custom_data points to data this tab represents (e.g. workspace).
        * uiBut.rnapoin/prop store an active value (e.g. active workspace). */
-      RNA_pointer_create(but->rnapoin.id.data, ptr_type, but->custom_data, &ptr);
+      RNA_pointer_create(but->rnapoin.owner_id, ptr_type, but->custom_data, &ptr);
       buf = RNA_struct_name_get_alloc(&ptr, str, maxlen, &buf_len);
     }
     else if (type == PROP_STRING) {
@@ -2822,7 +2822,7 @@ bool ui_but_string_set(bContext *C, uiBut *but, const char *str)
 
       /* uiBut.custom_data points to data this tab represents (e.g. workspace).
        * uiBut.rnapoin/prop store an active value (e.g. active workspace). */
-      RNA_pointer_create(but->rnapoin.id.data, ptr_type, but->custom_data, &ptr);
+      RNA_pointer_create(but->rnapoin.owner_id, ptr_type, but->custom_data, &ptr);
       prop = RNA_struct_name_property(ptr_type);
       if (RNA_property_editable(&ptr, prop)) {
         RNA_property_string_set(&ptr, prop, str);
@@ -3586,6 +3586,14 @@ static uiBut *ui_def_but(uiBlock *block,
                 (a1 != 0.0f && a1 != 1.0f)) == false);
   }
 
+  /* Number buttons must have a click-step,
+   * assert instead of correcting the value to ensure the caller knows what they're doing.  */
+  if ((type & BUTTYPE) == UI_BTYPE_NUM) {
+    if (ELEM((type & UI_BUT_POIN_TYPES), UI_BUT_POIN_CHAR, UI_BUT_POIN_SHORT, UI_BUT_POIN_INT)) {
+      BLI_assert((int)a1 > 0);
+    }
+  }
+
   if (type & UI_BUT_POIN_TYPES) { /* a pointer is required */
     if (poin == NULL) {
       BLI_assert(0);
diff --git a/source/blender/editors/interface/interface_anim.c b/source/blender/editors/interface/interface_anim.c
index 4d87cc22ef2..168c6051327 100644
--- a/source/blender/editors/interface/interface_anim.c
+++ b/source/blender/editors/interface/interface_anim.c
@@ -228,7 +228,7 @@ bool ui_but_anim_expression_create(uiBut *but, const char *str)
   /* make sure we have animdata for this */
   /* FIXME: until materials can be handled by depsgraph,
    * don't allow drivers to be created for them */
-  id = (ID *)but->rnapoin.id.data;
+  id = but->rnapoin.owner_id;
   if ((id == NULL) || (GS(id->name) == ID_MA) || (GS(id->name) == ID_TE)) {
     if (G.debug & G_DEBUG) {
       printf("ERROR: create expression failed - invalid data-block for adding drivers (%p)\n", id);
@@ -314,7 +314,7 @@ void ui_but_anim_autokey(bContext *C, uiBut *but, Scene *scene, float cfra)
     }
   }
   else {
-    id = but->rnapoin.id.data;
+    id = but->rnapoin.owner_id;
 
     /* TODO: this should probably respect the keyingset only option for anim */
     if (autokeyframe_cfra_can_key(scene, id)) {
diff --git a/source/blender/editors/interface/interface_context_menu.c b/source/blender/editors/interface/interface_context_menu.c
index 22b75da4968..580ff527bf6 100644
--- a/source/blender/editors/interface/interface_context_menu.c
+++ b/source/blender/editors/interface/interface_context_menu.c
@@ -176,11 +176,21 @@ static uiBlock *menu_change_shortcut(bContext *C, ARegion *ar, void *arg)
   UI_block_flag_enable(block, UI_BLOCK_MOVEMOUSE_QUIT);
   UI_block_direction_set(block, UI_DIR_CENTER_Y);
 
-  layout = UI_block_layout(block, UI_LAYOUT_VERTICAL, UI_LAYOUT_PANEL, 0, 0, 200, 20, 0, style);
-
+  layout = UI_block_layout(block,
+                           UI_LAYOUT_VERTICAL,
+                           UI_LAYOUT_PANEL,
+                           0,
+                           0,
+                           U.widget_unit * 10,
+                           U.widget_unit * 2,
+                           0,
+                           style);
+
+  uiItemL(layout, CTX_IFACE_(BLT_I18NCONTEXT_OPERATOR_DEFAULT, "Change Shortcut"), ICON_HAND);
   uiItemR(layout, &ptr, "type", UI_ITEM_R_FULL_EVENT | UI_ITEM_R_IMMEDIATE, "", ICON_NONE);
 
-  UI_block_bounds_set_popup(block, 6, (const int[2]){-50, 26});
+  UI_block_bounds_set_popup(
+      block, 6 * U.dpi_fac, (const int[2]){-100 * U.dpi_fac, 36 * U.dpi_fac});
 
   shortcut_free_operator_property(prop);
 
@@ -227,11 +237,21 @@ static uiBlock *menu_add_shortcut(bContext *C, ARegion *ar, void *arg)
   UI_block_func_handle_set(block, but_shortcut_name_func, but);
   UI_block_direction_set(block, UI_DIR_CENTER_Y);
 
-  layout = UI_block_layout(block, UI_LAYOUT_VERTICAL, UI_LAYOUT_PANEL, 0, 0, 200, 20, 0, style);
-
+  layout = UI_block_layout(block,
+                           UI_LAYOUT_VERTICAL,
+                           UI_LAYOUT_PANEL,
+                           0,
+                           0,
+                           U.widget_unit * 10,
+                           U.widget_unit * 2,
+                           0,
+                           style);
+
+  uiItemL(layout, CTX_IFACE_(BLT_I18NCONTEXT_OPERATOR_DEFAULT, "Assign Shortcut"), ICON_HAND);
   uiItemR(layout, &ptr, "type", UI_ITEM_R_FULL_EVENT | UI_ITEM_R_IMMEDIATE, "", ICON_NONE);
 
-  UI_block_bounds_set_popup(block, 6, (const int[2]){-50, 26});
+  UI_block_bounds_set_popup(
+      block, 6 * U.dpi_fac, (const int[2]){-100 * U.dpi_fac, 36 * U.dpi_fac});
 
 #ifdef USE_KEYMAP_ADD_HACK
   g_kmi_id_hack = kmi_id;
@@ -470,8 +490,9 @@ static void ui_but_menu_add_path_operators(uiLayout *layout, PointerRNA *ptr, Pr
 
 bool ui_popup_context_menu_for_button(bContext *C, uiBut *but)
 {
-  /* having this menu for some buttons makes no sense */
-  if (but->type == UI_BTYPE_IMAGE) {
+  /* ui_but_is_interactive() may let some buttons through that should not get a context menu - it
+   * doesn't make sense for them. */
+  if (ELEM(but->type, UI_BTYPE_LABEL, UI_BTYPE_IMAGE)) {
     return false;
   }
 
@@ -904,7 +925,7 @@ bool ui_popup_context_menu_for_button(bContext *C, uiBut *but)
             ICON_NONE,
             "UI_OT_copy_data_path_button");
 
-    if (ptr->id.data && !is_whole_array &&
+    if (ptr->owner_id && !is_whole_array &&
         ELEM(type, PROP_BOOLEAN, PROP_INT, PROP_FLOAT, PROP_ENUM)) {
       uiItemO(layout,
               CTX_IFACE_(BLT_I18NCONTEXT_OPERATOR_DEFAULT, "Copy As New Driver"),
@@ -1222,7 +1243,7 @@ void ui_popup_context_menu_for_panel(bContext *C, ARegion *ar, Panel *pa)
                  sizeof(tmpstr),
                  "%s" UI_SEP_CHAR_S "%s",
                  IFACE_("Pin"),
-                 IFACE_("Shift+Left Mouse"));
+                 IFACE_("Shift Left Mouse"));
     uiItemR(layout, &ptr, "use_pin", 0, tmpstr, ICON_NONE);
 
     /* evil, force shortcut flag */
@@ -1230,6 +1251,7 @@ void ui_popup_context_menu_for_panel(bContext *C, ARegion *ar, Panel *pa)
       uiBlock *block = uiLayoutGetBlock(layout);
       uiBut *but = block->buttons.last;
       but->flag |= UI_BUT_HAS_SEP_CHAR;
+      but->drawflag |= UI_BUT_HAS_SHORTCUT;
     }
   }
   UI_popup_menu_end(C, pup);
diff --git a/source/blender/editors/interface/interface_draw.c b/source/blender/editors/interface/interface_draw.c
index 76630de96db..b7a0812c5f2 100644
--- a/source/blender/editors/interface/interface_draw.c
+++ b/source/blender/editors/interface/interface_draw.c
@@ -750,7 +750,7 @@ void ui_draw_but_IMAGE(ARegion *UNUSED(ar),
 /**
  * Draw title and text safe areas.
  *
- * \Note This functionn is to be used with the 2D dashed shader enabled.
+ * \note This function is to be used with the 2D dashed shader enabled.
  *
  * \param pos: is a PRIM_FLOAT, 2, GPU_FETCH_FLOAT vertex attribute.
  * \param line_origin: is a PRIM_FLOAT, 2, GPU_FETCH_FLOAT vertex attribute.
diff --git a/source/blender/editors/interface/interface_eyedropper_datablock.c b/source/blender/editors/interface/interface_eyedropper_datablock.c
index 658aa4f67f9..336fae45895 100644
--- a/source/blender/editors/interface/interface_eyedropper_datablock.c
+++ b/source/blender/editors/interface/interface_eyedropper_datablock.c
@@ -51,6 +51,7 @@
 #include "ED_space_api.h"
 #include "ED_screen.h"
 #include "ED_view3d.h"
+#include "ED_outliner.h"
 
 #include "interface_intern.h"
 #include "interface_eyedropper_intern.h"
@@ -67,6 +68,7 @@ typedef struct DataDropper {
 
   ID *init_id; /* for resetting on cancel */
 
+  ScrArea *cursor_area; /* Area under the cursor */
   ARegionType *art;
   void *draw_handle_pixel;
   char name[200];
@@ -103,6 +105,7 @@ static int datadropper_init(bContext *C, wmOperator *op)
 
   ddr->is_undo = UI_but_flag_is_set(but, UI_BUT_UNDO);
 
+  ddr->cursor_area = CTX_wm_area(C);
   ddr->art = art;
   ddr->draw_handle_pixel = ED_region_draw_cb_activate(
       art, datadropper_draw_cb, ddr, REGION_DRAW_POST_PIXEL);
@@ -115,7 +118,7 @@ static int datadropper_init(bContext *C, wmOperator *op)
   ddr->idcode_name = TIP_(BKE_idcode_to_name(ddr->idcode));
 
   PointerRNA ptr = RNA_property_pointer_get(&ddr->ptr, ddr->prop);
-  ddr->init_id = ptr.id.data;
+  ddr->init_id = ptr.owner_id;
 
   return true;
 }
@@ -141,7 +144,7 @@ static void datadropper_exit(bContext *C, wmOperator *op)
 
 /* *** datadropper id helper functions *** */
 /**
- * \brief get the ID from the screen.
+ * \brief get the ID from the 3D view or outliner.
  */
 static void datadropper_id_sample_pt(bContext *C, DataDropper *ddr, int mx, int my, ID **r_id)
 {
@@ -155,7 +158,7 @@ static void datadropper_id_sample_pt(bContext *C, DataDropper *ddr, int mx, int
   ddr->name[0] = '\0';
 
   if (sa) {
-    if (sa->spacetype == SPACE_VIEW3D) {
+    if (ELEM(sa->spacetype, SPACE_VIEW3D, SPACE_OUTLINER)) {
       ARegion *ar = BKE_area_find_region_xy(sa, RGN_TYPE_WINDOW, mx, my);
       if (ar) {
         const int mval[2] = {mx - ar->winrct.xmin, my - ar->winrct.ymin};
@@ -167,7 +170,13 @@ static void datadropper_id_sample_pt(bContext *C, DataDropper *ddr, int mx, int
         /* grr, always draw else we leave stale text */
         ED_region_tag_redraw(ar);
 
-        base = ED_view3d_give_base_under_cursor(C, mval);
+        if (sa->spacetype == SPACE_VIEW3D) {
+          base = ED_view3d_give_base_under_cursor(C, mval);
+        }
+        else {
+          base = ED_outliner_give_base_under_cursor(C, mval);
+        }
+
         if (base) {
           Object *ob = base->object;
           ID *id = NULL;
@@ -213,7 +222,7 @@ static bool datadropper_id_set(bContext *C, DataDropper *ddr, ID *id)
 
   ptr_value = RNA_property_pointer_get(&ddr->ptr, ddr->prop);
 
-  return (ptr_value.id.data == id);
+  return (ptr_value.owner_id == id);
 }
 
 /* single point sample & set */
@@ -232,6 +241,36 @@ static void datadropper_cancel(bContext *C, wmOperator *op)
   datadropper_exit(C, op);
 }
 
+/* To switch the draw callback when region under mouse event changes */
+static void datadropper_set_draw_callback_region(bContext *C,
+                                                 DataDropper *ddr,
+                                                 const int mx,
+                                                 const int my)
+{
+  bScreen *screen = CTX_wm_screen(C);
+  ScrArea *sa = BKE_screen_find_area_xy(screen, -1, mx, my);
+
+  if (sa) {
+    /* If spacetype changed */
+    if (sa->spacetype != ddr->cursor_area->spacetype) {
+      /* Remove old callback */
+      ED_region_draw_cb_exit(ddr->art, ddr->draw_handle_pixel);
+
+      /* Redraw old area */
+      ARegion *ar = BKE_area_find_region_type(ddr->cursor_area, RGN_TYPE_WINDOW);
+      ED_region_tag_redraw(ar);
+
+      /* Set draw callback in new region */
+      ARegionType *art = BKE_regiontype_from_id(sa->type, RGN_TYPE_WINDOW);
+
+      ddr->cursor_area = sa;
+      ddr->art = art;
+      ddr->draw_handle_pixel = ED_region_draw_cb_activate(
+          art, datadropper_draw_cb, ddr, REGION_DRAW_POST_PIXEL);
+    }
+  }
+}
+
 /* main modal status check */
 static int datadropper_modal(bContext *C, wmOperator *op, const wmEvent *event)
 {
@@ -260,6 +299,10 @@ static int datadropper_modal(bContext *C, wmOperator *op, const wmEvent *event)
   }
   else if (event->type == MOUSEMOVE) {
     ID *id = NULL;
+
+    /* Set the region for eyedropper cursor text drawing */
+    datadropper_set_draw_callback_region(C, ddr, event->x, event->y);
+
     datadropper_id_sample_pt(C, ddr, event->x, event->y, &id);
   }
 
diff --git a/source/blender/editors/interface/interface_eyedropper_driver.c b/source/blender/editors/interface/interface_eyedropper_driver.c
index c1aee190cd3..e6fc52bc3bc 100644
--- a/source/blender/editors/interface/interface_eyedropper_driver.c
+++ b/source/blender/editors/interface/interface_eyedropper_driver.c
@@ -116,10 +116,10 @@ static void driverdropper_sample(bContext *C, wmOperator *op, const wmEvent *eve
     /* Now create driver(s) */
     if (target_path && dst_path) {
       int success = ANIM_add_driver_with_target(op->reports,
-                                                ddr->ptr.id.data,
+                                                ddr->ptr.owner_id,
                                                 dst_path,
                                                 ddr->index,
-                                                target_ptr->id.data,
+                                                target_ptr->owner_id,
                                                 target_path,
                                                 target_index,
                                                 flag,
@@ -130,7 +130,7 @@ static void driverdropper_sample(bContext *C, wmOperator *op, const wmEvent *eve
         /* send updates */
         UI_context_update_anim_flag(C);
         DEG_relations_tag_update(CTX_data_main(C));
-        DEG_id_tag_update(ddr->ptr.id.data, ID_RECALC_TRANSFORM | ID_RECALC_GEOMETRY);
+        DEG_id_tag_update(ddr->ptr.owner_id, ID_RECALC_TRANSFORM | ID_RECALC_GEOMETRY);
         WM_event_add_notifier(C, NC_ANIMATION | ND_FCURVES_ORDER, NULL);  // XXX
       }
     }
diff --git a/source/blender/editors/interface/interface_handlers.c b/source/blender/editors/interface/interface_handlers.c
index 0cb0dbdc85c..55980099116 100644
--- a/source/blender/editors/interface/interface_handlers.c
+++ b/source/blender/editors/interface/interface_handlers.c
@@ -757,7 +757,7 @@ static void ui_apply_but_undo(uiBut *but)
     }
 
     /* Optionally override undo when undo system doesn't support storing properties. */
-    if (but->rnapoin.id.data) {
+    if (but->rnapoin.owner_id) {
       /* Exception for renaming ID data, we always need undo pushes in this case,
        * because undo systems track data by their ID, see: T67002. */
       extern PropertyRNA rna_ID_name;
@@ -765,7 +765,7 @@ static void ui_apply_but_undo(uiBut *but)
         /* pass */
       }
       else {
-        ID *id = but->rnapoin.id.data;
+        ID *id = but->rnapoin.owner_id;
         if (!ED_undo_is_legacy_compatible_for_property(but->block->evil_C, id)) {
           str = "";
         }
@@ -1558,7 +1558,7 @@ static bool ui_selectcontext_begin(bContext *C, uiBut *but, uiSelectContextStore
           if (use_path_from_id) {
             /* Path relative to ID. */
             lprop = NULL;
-            RNA_id_pointer_create(link->ptr.id.data, &idptr);
+            RNA_id_pointer_create(link->ptr.owner_id, &idptr);
             RNA_path_resolve_property(&idptr, path, &lptr, &lprop);
           }
           else if (path) {
@@ -2760,7 +2760,7 @@ static void ui_textedit_set_cursor_pos(uiBut *but, uiHandleButtonData *data, con
       startx += UI_DPI_ICON_SIZE / aspect;
     }
   }
-  /* but this extra .05 makes clicks inbetween characters feel nicer */
+  /* But this extra .05 makes clicks in between characters feel nicer. */
   startx += ((UI_TEXT_MARGIN_X + 0.05f) * U.widget_unit) / aspect;
 
   /* mouse dragged outside the widget to the left */
@@ -4800,18 +4800,19 @@ static int ui_do_but_NUM(
   if (click) {
     /* we can click on the side arrows to increment/decrement,
      * or click inside to edit the value directly */
-    const float softmin = but->softmin;
-    const float softmax = but->softmax;
 
     if (!ui_but_is_float(but)) {
       /* Integer Value. */
       if (but->drawflag & (UI_BUT_ACTIVE_LEFT | UI_BUT_ACTIVE_RIGHT)) {
         button_activate_state(C, but, BUTTON_STATE_NUM_EDITING);
+
         const int value_step = (int)but->a1;
         BLI_assert(value_step > 0);
+        const int softmin = round_fl_to_int_clamp(but->softmin);
+        const int softmax = round_fl_to_int_clamp(but->softmax);
         const double value_test = (but->drawflag & UI_BUT_ACTIVE_LEFT) ?
-                                      (double)max_ii((int)softmin, (int)data->value - value_step) :
-                                      (double)min_ii((int)softmax, (int)data->value + value_step);
+                                      (double)max_ii(softmin, (int)data->value - value_step) :
+                                      (double)min_ii(softmax, (int)data->value + value_step);
         if (value_test != data->value) {
           data->value = (double)value_test;
         }
@@ -4828,11 +4829,14 @@ static int ui_do_but_NUM(
       /* Float Value. */
       if (but->drawflag & (UI_BUT_ACTIVE_LEFT | UI_BUT_ACTIVE_RIGHT)) {
         button_activate_state(C, but, BUTTON_STATE_NUM_EDITING);
+
         const double value_step = (double)but->a1 * UI_PRECISION_FLOAT_SCALE;
         BLI_assert(value_step > 0.0f);
         const double value_test = (but->drawflag & UI_BUT_ACTIVE_LEFT) ?
-                                      (double)max_ff(softmin, (float)(data->value - value_step)) :
-                                      (double)min_ff(softmax, (float)(data->value + value_step));
+                                      (double)max_ff(but->softmin,
+                                                     (float)(data->value - value_step)) :
+                                      (double)min_ff(but->softmax,
+                                                     (float)(data->value + value_step));
         if (value_test != data->value) {
           data->value = value_test;
         }
@@ -5484,7 +5488,7 @@ static bool ui_numedit_but_UNITVEC(
 static void ui_palette_set_active(uiBut *but)
 {
   if ((int)(but->a1) == UI_PALETTE_COLOR) {
-    Palette *palette = but->rnapoin.id.data;
+    Palette *palette = (Palette *)but->rnapoin.owner_id;
     PaletteColor *color = but->rnapoin.data;
     palette->active_color = BLI_findindex(&palette->colors, color);
   }
@@ -5547,7 +5551,7 @@ static int ui_do_but_COLOR(bContext *C, uiBut *but, uiHandleButtonData *data, co
     }
     else if ((int)(but->a1) == UI_PALETTE_COLOR && event->type == DELKEY &&
              event->val == KM_PRESS) {
-      Palette *palette = but->rnapoin.id.data;
+      Palette *palette = (Palette *)but->rnapoin.owner_id;
       PaletteColor *color = but->rnapoin.data;
 
       BKE_palette_color_remove(palette, color);
@@ -8184,7 +8188,7 @@ static int ui_handle_button_event(bContext *C, const wmEvent *event, uiBut *but)
           else {
             /* Do this so we can still mouse-up, closing the menu and running the button.
              * This is nice to support but there are times when the button gets left pressed.
-             * Keep disavled for now. */
+             * Keep disabled for now. */
             WM_event_remove_timer(data->wm, data->window, data->hold_action_timer);
             data->hold_action_timer = NULL;
           }
diff --git a/source/blender/editors/interface/interface_icons.c b/source/blender/editors/interface/interface_icons.c
index e9aa18394fa..7ada4755a64 100644
--- a/source/blender/editors/interface/interface_icons.c
+++ b/source/blender/editors/interface/interface_icons.c
@@ -154,8 +154,8 @@ typedef struct IconType {
 } IconType;
 
 /* ******************* STATIC LOCAL VARS ******************* */
-/* static here to cache results of icon directory scan, so it's not
- * scanning the filesystem each time the menu is drawn */
+/* Static here to cache results of icon directory scan, so it's not
+ * scanning the file-system each time the menu is drawn. */
 static struct ListBase iconfilelist = {NULL, NULL};
 static IconTexture icongltex = {{0, 0}, 0, 0, 0, 0.0f, 0.0f};
 
@@ -317,8 +317,9 @@ static void vicon_keytype_draw_wrapper(
       format, "outlineColor", GPU_COMP_U8, 4, GPU_FETCH_INT_TO_FLOAT_UNIT);
   uint flags_id = GPU_vertformat_attr_add(format, "flags", GPU_COMP_U32, 1, GPU_FETCH_INT);
 
-  immBindBuiltinProgram(GPU_SHADER_KEYFRAME_DIAMOND);
   GPU_program_point_size(true);
+  immBindBuiltinProgram(GPU_SHADER_KEYFRAME_DIAMOND);
+  immUniform1f("outline_scale", 1.0f);
   immUniform2f("ViewportSize", -1.0f, -1.0f);
   immBegin(GPU_PRIM_POINTS, 1);
 
@@ -2136,7 +2137,7 @@ int UI_rnaptr_icon_get(bContext *C, PointerRNA *ptr, int rnaicon, const bool big
 
   /* try ID, material, texture or dynapaint slot */
   if (RNA_struct_is_ID(ptr->type)) {
-    id = ptr->id.data;
+    id = ptr->owner_id;
   }
   else if (RNA_struct_is_a(ptr->type, &RNA_MaterialSlot)) {
     id = RNA_pointer_get(ptr, "material").data;
@@ -2243,6 +2244,8 @@ int UI_idcode_icon_get(const int idcode)
       return ICON_FONT_DATA;
     case ID_WO:
       return ICON_WORLD_DATA;
+    case ID_WS:
+      return ICON_WORKSPACE;
     default:
       return ICON_NONE;
   }
diff --git a/source/blender/editors/interface/interface_layout.c b/source/blender/editors/interface/interface_layout.c
index 78eed98eb77..529cb0712df 100644
--- a/source/blender/editors/interface/interface_layout.c
+++ b/source/blender/editors/interface/interface_layout.c
@@ -1180,11 +1180,6 @@ static uiBut *uiItemFullO_ptr_ex(uiLayout *layout,
 
   assert(but->optype != NULL);
 
-  /* text alignment for toolbar buttons */
-  if ((layout->root->type == UI_LAYOUT_TOOLBAR) && !icon) {
-    but->drawflag |= UI_BUT_TEXT_LEFT;
-  }
-
   if (flag & UI_ITEM_R_NO_BG) {
     layout->emboss = prev_emboss;
   }
@@ -1878,7 +1873,8 @@ void uiItemFullR(uiLayout *layout,
     uiBut *but;
   } ui_decorate = {
       .use_prop_decorate = (((layout->item.flag & UI_ITEM_PROP_DECORATE) != 0) &&
-                            (use_prop_sep && ptr->id.data && id_can_have_animdata(ptr->id.data))),
+                            (use_prop_sep && ptr->owner_id &&
+                             id_can_have_animdata(ptr->owner_id))),
   };
 #endif /* UI_PROP_DECORATE */
 
@@ -2951,7 +2947,7 @@ void uiItemLDrag(uiLayout *layout, PointerRNA *ptr, const char *name, int icon)
 
   if (ptr && ptr->type) {
     if (RNA_struct_is_ID(ptr->type)) {
-      UI_but_drag_set_id(but, ptr->id.data);
+      UI_but_drag_set_id(but, ptr->owner_id);
     }
   }
 }
diff --git a/source/blender/editors/interface/interface_ops.c b/source/blender/editors/interface/interface_ops.c
index 6051bf5ca40..68d21e88211 100644
--- a/source/blender/editors/interface/interface_ops.c
+++ b/source/blender/editors/interface/interface_ops.c
@@ -27,7 +27,7 @@
 
 #include "DNA_armature_types.h"
 #include "DNA_screen_types.h"
-#include "DNA_text_types.h"   /* for UI_OT_reports_to_text */
+#include "DNA_text_types.h"
 #include "DNA_object_types.h" /* for OB_DATA_SUPPORT_ID */
 
 #include "BLI_blenlib.h"
@@ -45,7 +45,7 @@
 #include "BKE_node.h"
 #include "BKE_report.h"
 #include "BKE_screen.h"
-#include "BKE_text.h" /* for UI_OT_reports_to_text */
+#include "BKE_text.h"
 
 #include "IMB_colormanagement.h"
 
@@ -86,7 +86,7 @@ static bool copy_data_path_button_poll(bContext *C)
 
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (ptr.id.data && ptr.data && prop) {
+  if (ptr.owner_id && ptr.data && prop) {
     path = RNA_path_from_ID_to_property(&ptr, prop);
 
     if (path) {
@@ -100,29 +100,33 @@ static bool copy_data_path_button_poll(bContext *C)
 
 static int copy_data_path_button_exec(bContext *C, wmOperator *op)
 {
+  Main *bmain = CTX_data_main(C);
   PointerRNA ptr;
   PropertyRNA *prop;
   char *path;
   int index;
+  ID *id;
 
   const bool full_path = RNA_boolean_get(op->ptr, "full_path");
 
   /* try to create driver using property retrieved from UI */
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (ptr.id.data != NULL) {
-
+  if (ptr.owner_id != NULL) {
     if (full_path) {
-
       if (prop) {
-        path = RNA_path_full_property_py_ex(&ptr, prop, index, true);
+        path = RNA_path_full_property_py_ex(bmain, &ptr, prop, index, true);
       }
       else {
-        path = RNA_path_full_struct_py(&ptr);
+        path = RNA_path_full_struct_py(bmain, &ptr);
       }
     }
     else {
-      path = RNA_path_from_ID_to_property(&ptr, prop);
+      path = RNA_path_from_real_ID_to_property_index(bmain, &ptr, prop, 0, -1, &id);
+
+      if (!path) {
+        path = RNA_path_from_ID_to_property(&ptr, prop);
+      }
     }
 
     if (path) {
@@ -171,7 +175,7 @@ static bool copy_as_driver_button_poll(bContext *C)
 
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (ptr.id.data && ptr.data && prop &&
+  if (ptr.owner_id && ptr.data && prop &&
       ELEM(RNA_property_type(prop), PROP_BOOLEAN, PROP_INT, PROP_FLOAT, PROP_ENUM) &&
       (index >= 0 || !RNA_property_array_check(prop))) {
     path = RNA_path_from_ID_to_property(&ptr, prop);
@@ -185,8 +189,9 @@ static bool copy_as_driver_button_poll(bContext *C)
   return 0;
 }
 
-static int copy_as_driver_button_exec(bContext *C, wmOperator *UNUSED(op))
+static int copy_as_driver_button_exec(bContext *C, wmOperator *op)
 {
+  Main *bmain = CTX_data_main(C);
   PointerRNA ptr;
   PropertyRNA *prop;
   int index;
@@ -194,15 +199,20 @@ static int copy_as_driver_button_exec(bContext *C, wmOperator *UNUSED(op))
   /* try to create driver using property retrieved from UI */
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  if (ptr.id.data && ptr.data && prop) {
+  if (ptr.owner_id && ptr.data && prop) {
+    ID *id;
     int dim = RNA_property_array_dimension(&ptr, prop, NULL);
-    char *path = RNA_path_from_ID_to_property_index(&ptr, prop, dim, index);
+    char *path = RNA_path_from_real_ID_to_property_index(bmain, &ptr, prop, dim, index, &id);
 
     if (path) {
-      ANIM_copy_as_driver(ptr.id.data, path, RNA_property_identifier(prop));
+      ANIM_copy_as_driver(id, path, RNA_property_identifier(prop));
       MEM_freeN(path);
       return OPERATOR_FINISHED;
     }
+    else {
+      BKE_reportf(op->reports, RPT_ERROR, "Could not compute a valid data path");
+      return OPERATOR_CANCELLED;
+    }
   }
 
   return OPERATOR_CANCELLED;
@@ -287,7 +297,7 @@ static void UI_OT_copy_python_command_button(wmOperatorType *ot)
 
 static int operator_button_property_finish(bContext *C, PointerRNA *ptr, PropertyRNA *prop)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   /* perform updates required for this property */
   RNA_property_update(C, ptr, prop);
@@ -542,7 +552,7 @@ static int override_type_set_button_exec(bContext *C, wmOperator *op)
   /* try to reset the nominated setting to its default value */
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  BLI_assert(ptr.id.data != NULL);
+  BLI_assert(ptr.owner_id != NULL);
 
   if (all) {
     index = -1;
@@ -605,7 +615,7 @@ static bool override_remove_button_poll(bContext *C)
 
   const int override_status = RNA_property_override_library_status(&ptr, prop, index);
 
-  return (ptr.data && ptr.id.data && prop && (override_status & RNA_OVERRIDE_STATUS_OVERRIDDEN));
+  return (ptr.data && ptr.owner_id && prop && (override_status & RNA_OVERRIDE_STATUS_OVERRIDDEN));
 }
 
 static int override_remove_button_exec(bContext *C, wmOperator *op)
@@ -619,7 +629,7 @@ static int override_remove_button_exec(bContext *C, wmOperator *op)
   /* try to reset the nominated setting to its default value */
   UI_context_active_but_prop_get(C, &ptr, &prop, &index);
 
-  ID *id = ptr.id.data;
+  ID *id = ptr.owner_id;
   IDOverrideLibraryProperty *oprop = RNA_property_override_property_find(&ptr, prop);
   BLI_assert(oprop != NULL);
   BLI_assert(id != NULL && id->override_library != NULL);
@@ -629,8 +639,9 @@ static int override_remove_button_exec(bContext *C, wmOperator *op)
   /* We need source (i.e. linked data) to restore values of deleted overrides...
    * If this is an override template, we obviously do not need to restore anything. */
   if (!is_template) {
+    PropertyRNA *src_prop;
     RNA_id_pointer_create(id->override_library->reference, &id_refptr);
-    if (!RNA_path_resolve(&id_refptr, oprop->rna_path, &src, NULL)) {
+    if (!RNA_path_resolve_property(&id_refptr, oprop->rna_path, &src, &src_prop)) {
       BLI_assert(0 && "Failed to create matching source (linked data) RNA pointer");
     }
   }
@@ -720,7 +731,7 @@ bool UI_context_copy_to_selected_list(bContext *C,
       CollectionPointerLink *link;
       for (link = lb.first; link; link = link->next) {
         bPoseChannel *pchan = link->ptr.data;
-        RNA_pointer_create(link->ptr.id.data, &RNA_Bone, pchan->bone, &link->ptr);
+        RNA_pointer_create(link->ptr.owner_id, &RNA_Bone, pchan->bone, &link->ptr);
       }
     }
 
@@ -739,7 +750,7 @@ bool UI_context_copy_to_selected_list(bContext *C,
 
     /* Get the node we're editing */
     if (RNA_struct_is_a(ptr->type, &RNA_NodeSocket)) {
-      bNodeTree *ntree = ptr->id.data;
+      bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
       bNodeSocket *sock = ptr->data;
       if (nodeFindNode(ntree, sock, &node, NULL)) {
         if ((path = RNA_path_resolve_from_type_to_property(ptr, prop, &RNA_Node)) != NULL) {
@@ -773,8 +784,8 @@ bool UI_context_copy_to_selected_list(bContext *C,
     *r_lb = lb;
     *r_path = path;
   }
-  else if (ptr->id.data) {
-    ID *id = ptr->id.data;
+  else if (ptr->owner_id) {
+    ID *id = ptr->owner_id;
 
     if (GS(id->name) == ID_OB) {
       *r_lb = CTX_data_collection_get(C, "selected_editable_objects");
@@ -792,7 +803,7 @@ bool UI_context_copy_to_selected_list(bContext *C,
         CollectionPointerLink *link, *link_next;
 
         for (link = lb.first; link; link = link->next) {
-          Object *ob = link->ptr.id.data;
+          Object *ob = (Object *)link->ptr.owner_id;
           if (ob->data) {
             ID *id_data = ob->data;
             id_data->tag |= LIB_TAG_DOIT;
@@ -800,7 +811,7 @@ bool UI_context_copy_to_selected_list(bContext *C,
         }
 
         for (link = lb.first; link; link = link_next) {
-          Object *ob = link->ptr.id.data;
+          Object *ob = (Object *)link->ptr.owner_id;
           ID *id_data = ob->data;
           link_next = link->next;
 
@@ -872,7 +883,7 @@ static bool copy_to_selected_button(bContext *C, bool all, bool poll)
           if (use_path_from_id) {
             /* Path relative to ID. */
             lprop = NULL;
-            RNA_id_pointer_create(link->ptr.id.data, &idptr);
+            RNA_id_pointer_create(link->ptr.owner_id, &idptr);
             RNA_path_resolve_property(&idptr, path, &lptr, &lprop);
           }
           else if (path) {
@@ -982,12 +993,12 @@ static bool jump_to_target_ptr(bContext *C, PointerRNA ptr, const bool poll)
   /* Find the containing Object. */
   ViewLayer *view_layer = CTX_data_view_layer(C);
   Base *base = NULL;
-  const short id_type = GS(((ID *)ptr.id.data)->name);
+  const short id_type = GS(ptr.owner_id->name);
   if (id_type == ID_OB) {
-    base = BKE_view_layer_base_find(view_layer, ptr.id.data);
+    base = BKE_view_layer_base_find(view_layer, (Object *)ptr.owner_id);
   }
   else if (OB_DATA_SUPPORT_ID(id_type)) {
-    base = ED_object_find_first_by_data_id(view_layer, ptr.id.data);
+    base = ED_object_find_first_by_data_id(view_layer, ptr.owner_id);
   }
 
   bool ok = false;
@@ -1096,60 +1107,6 @@ static void UI_OT_jump_to_target_button(wmOperatorType *ot)
 /** \} */
 
 /* -------------------------------------------------------------------- */
-/** \name Reports to Textblock Operator
- * \{ */
-
-/* FIXME: this is just a temporary operator so that we can see all the reports somewhere
- * when there are too many to display...
- */
-
-static bool reports_to_text_poll(bContext *C)
-{
-  return CTX_wm_reports(C) != NULL;
-}
-
-static int reports_to_text_exec(bContext *C, wmOperator *UNUSED(op))
-{
-  ReportList *reports = CTX_wm_reports(C);
-  Main *bmain = CTX_data_main(C);
-  Text *txt;
-  char *str;
-
-  /* create new text-block to write to */
-  txt = BKE_text_add(bmain, "Recent Reports");
-
-  /* convert entire list to a display string, and add this to the text-block
-   * - if commandline debug option enabled, show debug reports too
-   * - otherwise, up to info (which is what users normally see)
-   */
-  str = BKE_reports_string(reports, (G.debug & G_DEBUG) ? RPT_DEBUG : RPT_INFO);
-
-  if (str) {
-    BKE_text_write(txt, str);
-    MEM_freeN(str);
-
-    return OPERATOR_FINISHED;
-  }
-  else {
-    return OPERATOR_CANCELLED;
-  }
-}
-
-static void UI_OT_reports_to_textblock(wmOperatorType *ot)
-{
-  /* identifiers */
-  ot->name = "Reports to Text Block";
-  ot->idname = "UI_OT_reports_to_textblock";
-  ot->description = "Write the reports ";
-
-  /* callbacks */
-  ot->poll = reports_to_text_poll;
-  ot->exec = reports_to_text_exec;
-}
-
-/** \} */
-
-/* -------------------------------------------------------------------- */
 /** \name Edit Python Source Operator
  * \{ */
 
@@ -1750,7 +1707,6 @@ void ED_operatortypes_ui(void)
   WM_operatortype_append(UI_OT_override_remove_button);
   WM_operatortype_append(UI_OT_copy_to_selected_button);
   WM_operatortype_append(UI_OT_jump_to_target_button);
-  WM_operatortype_append(UI_OT_reports_to_textblock); /* XXX: temp? */
   WM_operatortype_append(UI_OT_drop_color);
 #ifdef WITH_PYTHON
   WM_operatortype_append(UI_OT_editsource);
diff --git a/source/blender/editors/interface/interface_region_popup.c b/source/blender/editors/interface/interface_region_popup.c
index f3e18cc2f24..2073117d51c 100644
--- a/source/blender/editors/interface/interface_region_popup.c
+++ b/source/blender/editors/interface/interface_region_popup.c
@@ -690,9 +690,11 @@ uiBlock *ui_popup_block_refresh(bContext *C,
     /* Avoid menu moving down and losing cursor focus by keeping it at
      * the same height. */
     if (handle->refresh && handle->prev_block_rect.ymax > block->rect.ymax) {
-      float offset = handle->prev_block_rect.ymax - block->rect.ymax;
-      UI_block_translate(block, 0, offset);
-      block->rect.ymin = handle->prev_block_rect.ymin;
+      if (block->bounds_type != UI_BLOCK_BOUNDS_POPUP_CENTER) {
+        float offset = handle->prev_block_rect.ymax - block->rect.ymax;
+        UI_block_translate(block, 0, offset);
+        block->rect.ymin = handle->prev_block_rect.ymin;
+      }
     }
 
     handle->prev_block_rect = block->rect;
diff --git a/source/blender/editors/interface/interface_region_tooltip.c b/source/blender/editors/interface/interface_region_tooltip.c
index 6aad9e41e7d..7387fe5eb1c 100644
--- a/source/blender/editors/interface/interface_region_tooltip.c
+++ b/source/blender/editors/interface/interface_region_tooltip.c
@@ -765,8 +765,8 @@ static uiTooltipData *ui_tooltip_data_from_button(bContext *C, uiBut *but)
       }
     }
 
-    if (but->rnapoin.id.data) {
-      const ID *id = but->rnapoin.id.data;
+    if (but->rnapoin.owner_id) {
+      const ID *id = but->rnapoin.owner_id;
       if (ID_IS_LINKED(id)) {
         uiTooltipField *field = text_field_add(data,
                                                &(uiTooltipFormat){
@@ -847,7 +847,7 @@ static uiTooltipData *ui_tooltip_data_from_button(bContext *C, uiBut *but)
       }
     }
 
-    if (but->rnapoin.id.data) {
+    if (but->rnapoin.owner_id) {
       uiTooltipField *field = text_field_add(data,
                                              &(uiTooltipFormat){
                                                  .style = UI_TIP_STYLE_MONO,
@@ -860,10 +860,10 @@ static uiTooltipData *ui_tooltip_data_from_button(bContext *C, uiBut *but)
       /* move ownership (no need for re-alloc) */
       if (but->rnaprop) {
         field->text = RNA_path_full_property_py_ex(
-            &but->rnapoin, but->rnaprop, but->rnaindex, true);
+            CTX_data_main(C), &but->rnapoin, but->rnaprop, but->rnaindex, true);
       }
       else {
-        field->text = RNA_path_full_struct_py(&but->rnapoin);
+        field->text = RNA_path_full_struct_py(CTX_data_main(C), &but->rnapoin);
       }
     }
   }
diff --git a/source/blender/editors/interface/interface_templates.c b/source/blender/editors/interface/interface_templates.c
index 37eb1770f68..f53bef877c4 100644
--- a/source/blender/editors/interface/interface_templates.c
+++ b/source/blender/editors/interface/interface_templates.c
@@ -121,7 +121,7 @@ static void template_add_button_search_menu(const bContext *C,
 {
   PointerRNA active_ptr = RNA_property_pointer_get(ptr, prop);
   ID *id = (active_ptr.data && RNA_struct_is_ID(active_ptr.type)) ? active_ptr.data : NULL;
-  const ID *idfrom = ptr->id.data;
+  const ID *idfrom = ptr->owner_id;
   const StructRNA *type = active_ptr.type ? active_ptr.type : RNA_property_pointer_type(ptr, prop);
   uiBut *but;
 
@@ -310,7 +310,7 @@ static bool id_search_add(const bContext *C,
                           uiSearchItems *items,
                           ID *id)
 {
-  ID *id_from = template_ui->ptr.id.data;
+  ID *id_from = template_ui->ptr.owner_id;
 
   if (!((flag & PROP_ID_SELF_CHECK) && id == id_from)) {
 
@@ -401,7 +401,7 @@ static void id_search_cb_objects_from_scene(const bContext *C,
   TemplateID *template_ui = (TemplateID *)arg_template;
   ListBase *lb = template_ui->idlb;
   Scene *scene = NULL;
-  ID *id_from = template_ui->ptr.id.data;
+  ID *id_from = template_ui->ptr.owner_id;
 
   if (id_from && GS(id_from->name) == ID_SCE) {
     scene = (Scene *)id_from;
@@ -538,7 +538,7 @@ static void template_id_cb(bContext *C, void *arg_litem, void *arg_event)
       break;
     case UI_ID_OVERRIDE:
       if (id && id->override_library) {
-        BKE_override_library_free(&id->override_library);
+        BKE_override_library_free(&id->override_library, true);
         /* reassign to get get proper updates/notifiers */
         idptr = RNA_property_pointer_get(&template_ui->ptr, template_ui->prop);
         RNA_property_pointer_set(&template_ui->ptr, template_ui->prop, idptr, NULL);
@@ -667,7 +667,7 @@ static uiBut *template_id_def_new_but(uiBlock *block,
                                       const bool use_tab_but,
                                       int but_height)
 {
-  ID *idfrom = template_ui->ptr.id.data;
+  ID *idfrom = template_ui->ptr.owner_id;
   uiBut *but;
   const int w = id ? UI_UNIT_X : id_open ? UI_UNIT_X * 3 : UI_UNIT_X * 6;
   const int but_type = use_tab_but ? UI_BTYPE_TAB : UI_BTYPE_BUT;
@@ -769,7 +769,7 @@ static void template_ID(bContext *C,
 
   idptr = RNA_property_pointer_get(&template_ui->ptr, template_ui->prop);
   id = idptr.data;
-  idfrom = template_ui->ptr.id.data;
+  idfrom = template_ui->ptr.owner_id;
   // lb = template_ui->idlb;
 
   block = uiLayoutGetBlock(layout);
@@ -2037,7 +2037,7 @@ uiLayout *uiTemplateModifier(uiLayout *layout, bContext *C, PointerRNA *ptr)
     return NULL;
   }
 
-  ob = ptr->id.data;
+  ob = (Object *)ptr->owner_id;
   md = ptr->data;
 
   if (!ob || !(GS(ob->id.name) == ID_OB)) {
@@ -2179,7 +2179,7 @@ uiLayout *uiTemplateGpencilModifier(uiLayout *layout, bContext *UNUSED(C), Point
     return NULL;
   }
 
-  ob = ptr->id.data;
+  ob = (Object *)ptr->owner_id;
   md = ptr->data;
 
   if (!ob || !(GS(ob->id.name) == ID_OB)) {
@@ -2297,7 +2297,7 @@ uiLayout *uiTemplateShaderFx(uiLayout *layout, bContext *UNUSED(C), PointerRNA *
     return NULL;
   }
 
-  ob = ptr->id.data;
+  ob = (Object *)ptr->owner_id;
   fx = ptr->data;
 
   if (!ob || !(GS(ob->id.name) == ID_OB)) {
@@ -2523,7 +2523,7 @@ static uiLayout *draw_constraint(uiLayout *layout, Object *ob, bConstraint *con)
 
     /* enabled */
     UI_block_emboss_set(block, UI_EMBOSS_NONE);
-    uiItemR(row, &ptr, "mute", 0, "", (con->flag & CONSTRAINT_OFF) ? ICON_HIDE_ON : ICON_HIDE_OFF);
+    uiItemR(row, &ptr, "mute", 0, "", 0);
     UI_block_emboss_set(block, UI_EMBOSS);
 
     uiLayoutSetOperatorContext(row, WM_OP_INVOKE_DEFAULT);
@@ -2579,7 +2579,7 @@ uiLayout *uiTemplateConstraint(uiLayout *layout, PointerRNA *ptr)
     return NULL;
   }
 
-  ob = ptr->id.data;
+  ob = (Object *)ptr->owner_id;
   con = ptr->data;
 
   if (!ob || !(GS(ob->id.name) == ID_OB)) {
@@ -3098,7 +3098,7 @@ static void colorband_buttons_layout(uiLayout *layout,
   float ys = butr->ymin;
   PointerRNA ptr;
 
-  RNA_pointer_create(cb->ptr.id.data, &RNA_ColorRamp, coba, &ptr);
+  RNA_pointer_create(cb->ptr.owner_id, &RNA_ColorRamp, coba, &ptr);
 
   split = uiLayoutSplit(layout, 0.4f, false);
 
@@ -3190,7 +3190,7 @@ static void colorband_buttons_layout(uiLayout *layout,
   if (coba->tot) {
     CBData *cbd = coba->data + coba->cur;
 
-    RNA_pointer_create(cb->ptr.id.data, &RNA_ColorRampElement, cbd, &ptr);
+    RNA_pointer_create(cb->ptr.owner_id, &RNA_ColorRampElement, cbd, &ptr);
 
     if (!expand) {
       split = uiLayoutSplit(layout, 0.3f, false);
@@ -3207,7 +3207,7 @@ static void colorband_buttons_layout(uiLayout *layout,
                 &coba->cur,
                 0.0,
                 (float)(MAX2(0, coba->tot - 1)),
-                0,
+                1,
                 0,
                 TIP_("Choose active color stop"));
       row = uiLayoutRow(split, false);
@@ -3237,7 +3237,7 @@ static void colorband_buttons_layout(uiLayout *layout,
                 &coba->cur,
                 0.0,
                 (float)(MAX2(0, coba->tot - 1)),
-                0,
+                1,
                 0,
                 TIP_("Choose active color stop"));
       row = uiLayoutRow(subsplit, false);
@@ -3283,7 +3283,7 @@ void uiTemplateColorRamp(uiLayout *layout, PointerRNA *ptr, const char *propname
 
   block = uiLayoutAbsoluteBlock(layout);
 
-  id = cptr.id.data;
+  id = cptr.owner_id;
   UI_block_lock_set(block, (id && ID_IS_LINKED(id)), ERROR_LIBDATA_MESSAGE);
 
   colorband_buttons_layout(layout, block, cptr.data, &rect, cb, expand);
@@ -3745,12 +3745,11 @@ static uiBlock *curvemap_clipping_func(bContext *C, ARegion *ar, void *cumap_v)
   float width = 8 * UI_UNIT_X;
 
   block = UI_block_begin(C, ar, __func__, UI_EMBOSS);
-
-  /* use this for a fake extra empty space around the buttons */
-  uiDefBut(block, UI_BTYPE_LABEL, 0, "", -4, 16, width + 8, 6 * UI_UNIT_Y, NULL, 0, 0, 0, 0, "");
+  UI_block_flag_enable(block, UI_BLOCK_KEEP_OPEN | UI_BLOCK_MOVEMOUSE_QUIT);
+  UI_block_theme_style_set(block, UI_BLOCK_THEME_STYLE_POPUP);
 
   bt = uiDefButBitI(block,
-                    UI_BTYPE_TOGGLE,
+                    UI_BTYPE_CHECKBOX,
                     CUMA_DO_CLIP,
                     1,
                     IFACE_("Use Clipping"),
@@ -3770,7 +3769,7 @@ static uiBlock *curvemap_clipping_func(bContext *C, ARegion *ar, void *cumap_v)
   uiDefButF(block,
             UI_BTYPE_NUM,
             0,
-            IFACE_("Min X "),
+            IFACE_("Min X:"),
             0,
             4 * UI_UNIT_Y,
             width,
@@ -3784,7 +3783,7 @@ static uiBlock *curvemap_clipping_func(bContext *C, ARegion *ar, void *cumap_v)
   uiDefButF(block,
             UI_BTYPE_NUM,
             0,
-            IFACE_("Min Y "),
+            IFACE_("Min Y:"),
             0,
             3 * UI_UNIT_Y,
             width,
@@ -3798,7 +3797,7 @@ static uiBlock *curvemap_clipping_func(bContext *C, ARegion *ar, void *cumap_v)
   uiDefButF(block,
             UI_BTYPE_NUM,
             0,
-            IFACE_("Max X "),
+            IFACE_("Max X:"),
             0,
             2 * UI_UNIT_Y,
             width,
@@ -3812,7 +3811,7 @@ static uiBlock *curvemap_clipping_func(bContext *C, ARegion *ar, void *cumap_v)
   uiDefButF(block,
             UI_BTYPE_NUM,
             0,
-            IFACE_("Max Y "),
+            IFACE_("Max Y:"),
             0,
             UI_UNIT_Y,
             width,
@@ -3824,7 +3823,8 @@ static uiBlock *curvemap_clipping_func(bContext *C, ARegion *ar, void *cumap_v)
             2,
             "");
 
-  UI_block_direction_set(block, UI_DIR_RIGHT);
+  UI_block_bounds_set_normal(block, 0.3f * U.widget_unit);
+  UI_block_direction_set(block, UI_DIR_DOWN);
 
   return block;
 }
@@ -4405,7 +4405,7 @@ void uiTemplateCurveMapping(uiLayout *layout,
   cb->ptr = *ptr;
   cb->prop = prop;
 
-  id = cptr.id.data;
+  id = cptr.owner_id;
   UI_block_lock_set(block, (id && ID_IS_LINKED(id)), ERROR_LIBDATA_MESSAGE);
 
   curvemap_buttons_layout(layout, &cptr, type, levels, brush, neg_slope, tone, cb);
@@ -6378,7 +6378,7 @@ void uiTemplateReportsBanner(uiLayout *layout, bContext *C)
   if (reports->list.first != reports->list.last) {
     uiDefIconButO(block,
                   UI_BTYPE_BUT,
-                  "UI_OT_reports_to_textblock",
+                  "SCREEN_OT_info_log_show",
                   WM_OP_INVOKE_REGION_WIN,
                   icon,
                   2,
@@ -6392,22 +6392,17 @@ void uiTemplateReportsBanner(uiLayout *layout, bContext *C)
         block, UI_BTYPE_LABEL, 0, icon, 2, 0, UI_UNIT_X, UI_UNIT_Y, NULL, 0.0f, 0.0f, 0, 0, "");
   }
 
-  UI_block_emboss_set(block, UI_EMBOSS);
-
-  uiDefBut(block,
-           UI_BTYPE_LABEL,
-           0,
-           report->message,
-           UI_UNIT_X + 5,
-           0,
-           UI_UNIT_X + width,
-           UI_UNIT_Y,
-           NULL,
-           0.0f,
-           0.0f,
-           0,
-           0,
-           "");
+  but = uiDefButO(block,
+                  UI_BTYPE_BUT,
+                  "SCREEN_OT_info_log_show",
+                  WM_OP_INVOKE_REGION_WIN,
+                  report->message,
+                  UI_UNIT_X + 5,
+                  0,
+                  UI_UNIT_X + width,
+                  UI_UNIT_Y,
+                  "Show in Info Log");
+  rgba_float_to_uchar(but->col, rti->col);
 }
 
 void uiTemplateInputStatus(uiLayout *layout, struct bContext *C)
@@ -6627,7 +6622,7 @@ static uiBlock *component_menu(bContext *C, ARegion *ar, void *args_v)
 
   uiItemR(layout, &args->ptr, args->propname, UI_ITEM_R_EXPAND, "", ICON_NONE);
 
-  UI_block_bounds_set_normal(block, 6);
+  UI_block_bounds_set_normal(block, 0.3f * U.widget_unit);
   UI_block_direction_set(block, UI_DIR_DOWN);
 
   return block;
diff --git a/source/blender/editors/interface/interface_utils.c b/source/blender/editors/interface/interface_utils.c
index 9c8787d002f..63e382e2280 100644
--- a/source/blender/editors/interface/interface_utils.c
+++ b/source/blender/editors/interface/interface_utils.c
@@ -405,7 +405,7 @@ void ui_rna_collection_search_cb(const struct bContext *C,
   RNA_PROP_BEGIN (&data->search_ptr, itemptr, data->search_prop) {
 
     if (flag & PROP_ID_SELF_CHECK) {
-      if (itemptr.data == data->target_ptr.id.data) {
+      if (itemptr.data == data->target_ptr.owner_id) {
         continue;
       }
     }
@@ -566,7 +566,7 @@ int UI_calc_float_precision(int prec, double value)
 
 bool UI_but_online_manual_id(const uiBut *but, char *r_str, size_t maxlength)
 {
-  if (but->rnapoin.id.data && but->rnapoin.data && but->rnaprop) {
+  if (but->rnapoin.owner_id && but->rnapoin.data && but->rnaprop) {
     BLI_snprintf(r_str,
                  maxlength,
                  "%s.%s",
diff --git a/source/blender/editors/interface/interface_widgets.c b/source/blender/editors/interface/interface_widgets.c
index 9c4d628cec5..8d22c5b85bb 100644
--- a/source/blender/editors/interface/interface_widgets.c
+++ b/source/blender/editors/interface/interface_widgets.c
@@ -2188,7 +2188,7 @@ static void widget_draw_text(const uiFontStyle *fstyle,
     drawstr_right = strrchr(drawstr + but->ofs, ':');
     if (drawstr_right) {
       drawstr_right++;
-      drawstr_left_len = (drawstr_right - drawstr);
+      drawstr_left_len = (drawstr_right - drawstr - 1);
 
       while (*drawstr_right == ' ') {
         drawstr_right++;
@@ -2587,6 +2587,10 @@ static void widget_state(uiWidgetType *wt, int state, int drawflag)
     }
   }
   else {
+    if (state & UI_BUT_ACTIVE_DEFAULT) {
+      copy_v4_v4_uchar(wt->wcol.inner, wt->wcol.inner_sel);
+      copy_v4_v4_uchar(wt->wcol.text, wt->wcol.text_sel);
+    }
     if (color_blend != NULL) {
       widget_state_blend(wt->wcol.inner, color_blend, wcol_state->blend);
     }
@@ -2709,10 +2713,10 @@ static void widget_state_menu_item(uiWidgetType *wt, int state, int UNUSED(drawf
 
   /* active and disabled (not so common) */
   if ((state & UI_BUT_DISABLED) && (state & UI_ACTIVE)) {
-    widget_state_blend(wt->wcol.text, wt->wcol.text_sel, 0.5f);
     /* draw the backdrop at low alpha, helps navigating with keys
      * when disabled items are active */
-    copy_v4_v4_uchar(wt->wcol.inner, wt->wcol.inner_sel);
+    wt->wcol.text[3] = 128;
+    widget_state_blend(wt->wcol.inner, wt->wcol.text, 0.5f);
     wt->wcol.inner[3] = 64;
   }
   else {
@@ -3268,20 +3272,6 @@ static void ui_draw_but_HSV_v(uiBut *but, const rcti *rect)
   ui_hsv_cursor(x, y);
 }
 
-/* Generic round-box drawing. */
-static void ui_draw_roundbox(const rcti *rect, const float rad, const uiWidgetColors *wcol)
-{
-  uiWidgetBase wtb;
-  widget_init(&wtb);
-  round_box_edges(&wtb, UI_CNR_ALL, rect, rad);
-  widgetbase_draw(&wtb, wcol);
-
-  /* We are drawing on top of widget bases. Flush cache. */
-  GPU_blend(true);
-  UI_widgetbase_draw_cache_flush();
-  GPU_blend(false);
-}
-
 /* ************ separator, for menus etc ***************** */
 static void ui_draw_separator(const rcti *rect, const uiWidgetColors *wcol)
 {
@@ -3802,7 +3792,7 @@ static void widget_swatch(
 
   widgetbase_draw_ex(&wtb, wcol, show_alpha_checkers);
   if (but->a1 == UI_PALETTE_COLOR &&
-      ((Palette *)but->rnapoin.id.data)->active_color == (int)but->a2) {
+      ((Palette *)but->rnapoin.owner_id)->active_color == (int)but->a2) {
     float width = rect->xmax - rect->xmin;
     float height = rect->ymax - rect->ymin;
     /* find color luminance and change it slightly */
@@ -4497,10 +4487,6 @@ void ui_draw_but(const bContext *C, ARegion *ar, uiStyle *style, uiBut *but, rct
   const uiFontStyle *fstyle = &style->widget;
   uiWidgetType *wt = NULL;
 
-#ifdef USE_UI_POPOVER_ONCE
-  const rcti rect_orig = *rect;
-#endif
-
   /* handle menus separately */
   if (but->dt == UI_EMBOSS_PULLDOWN) {
     switch (but->type) {
@@ -4786,6 +4772,14 @@ void ui_draw_but(const bContext *C, ARegion *ar, uiStyle *style, uiBut *but, rct
       ui_widget_color_disabled(wt);
     }
 
+#ifdef USE_UI_POPOVER_ONCE
+    if (but->block->flag & UI_BLOCK_POPOVER_ONCE) {
+      if ((state & UI_ACTIVE) && ui_but_is_popover_once_compat(but)) {
+        state |= UI_BUT_ACTIVE_DEFAULT;
+      }
+    }
+#endif
+
     wt->state(wt, state, drawflag);
     if (wt->custom) {
       wt->custom(but, &wt->wcol, rect, state, roundboxalign);
@@ -4798,42 +4792,10 @@ void ui_draw_but(const bContext *C, ARegion *ar, uiStyle *style, uiBut *but, rct
       GPU_blend(true);
     }
 
-    bool show_semi_highlight = false;
-
-#ifdef USE_UI_POPOVER_ONCE
-    if (but->block->flag & UI_BLOCK_POPOVER_ONCE) {
-      if ((state & UI_ACTIVE) && ui_but_is_popover_once_compat(but)) {
-        show_semi_highlight = true;
-      }
-    }
-#endif
-    if (but->flag & UI_BUT_ACTIVE_DEFAULT) {
-      show_semi_highlight = true;
-    }
-
-    if (show_semi_highlight) {
-      uiWidgetType wt_back = *wt;
-      uiWidgetType *wt_temp = widget_type(UI_WTYPE_MENU_ITEM);
-      wt_temp->state(wt_temp, state, drawflag);
-      copy_v4_v4_uchar(wt->wcol.inner, wt->wcol.inner_sel);
-      wt->wcol.inner[3] = 128;
-      wt->wcol.roundness = 0.5f;
-      ui_draw_roundbox(&rect_orig,
-                       0.25f * min_ff(BLI_rcti_size_x(&rect_orig), BLI_rcti_size_y(&rect_orig)),
-                       &wt_temp->wcol);
-      *wt = wt_back;
-    }
-
     wt->text(fstyle, &wt->wcol, but, rect);
     if (disabled) {
       GPU_blend(false);
     }
-
-    //      if (state & (UI_BUT_DISABLED | UI_BUT_INACTIVE)) {
-    //          if (but->dt != UI_EMBOSS_PULLDOWN) {
-    //              widget_disabled(&disablerect);
-    //          }
-    //      }
   }
 }
 
diff --git a/source/blender/editors/interface/resources.c b/source/blender/editors/interface/resources.c
index 1e74ee50dab..93a93fb6918 100644
--- a/source/blender/editors/interface/resources.c
+++ b/source/blender/editors/interface/resources.c
@@ -794,6 +794,10 @@ const uchar *UI_ThemeGetColorPtr(bTheme *btheme, int spacetype, int colorid)
           cp = ts->selected_highlight;
           break;
 
+        case TH_SELECT_ACTIVE:
+          cp = ts->active;
+          break;
+
         case TH_SELECTED_OBJECT:
           cp = ts->selected_object;
           break;
diff --git a/source/blender/editors/interface/view2d_ops.c b/source/blender/editors/interface/view2d_ops.c
index f32fcffabd4..032fb7e4cc2 100644
--- a/source/blender/editors/interface/view2d_ops.c
+++ b/source/blender/editors/interface/view2d_ops.c
@@ -1844,7 +1844,7 @@ static void scroller_activate_init(bContext *C,
    */
   scrollers = UI_view2d_scrollers_calc(v2d, NULL);
 
-  /* use a union of 'cur' & 'tot' incase the current view is far outside 'tot'. In this cases
+  /* Use a union of 'cur' & 'tot' in case the current view is far outside 'tot'. In this cases
    * moving the scroll bars has far too little effect and the view can get stuck T31476. */
   tot_cur_union = v2d->tot;
   BLI_rctf_union(&tot_cur_union, &v2d->cur);
diff --git a/source/blender/editors/mask/mask_shapekey.c b/source/blender/editors/mask/mask_shapekey.c
index edc22943124..1abe805192d 100644
--- a/source/blender/editors/mask/mask_shapekey.c
+++ b/source/blender/editors/mask/mask_shapekey.c
@@ -81,6 +81,7 @@ void MASK_OT_shape_key_insert(wmOperatorType *ot)
 {
   /* identifiers */
   ot->name = "Insert Shape Key";
+  ot->description = "Insert mask shape keyframe for active mask layer at the current frame";
   ot->idname = "MASK_OT_shape_key_insert";
 
   /* api callbacks */
@@ -129,6 +130,7 @@ void MASK_OT_shape_key_clear(wmOperatorType *ot)
 {
   /* identifiers */
   ot->name = "Clear Shape Key";
+  ot->description = "Remove mask shape keyframe for active mask layer at the current frame";
   ot->idname = "MASK_OT_shape_key_clear";
 
   /* api callbacks */
diff --git a/source/blender/editors/mesh/CMakeLists.txt b/source/blender/editors/mesh/CMakeLists.txt
index 9a779db4812..d7d020ae19d 100644
--- a/source/blender/editors/mesh/CMakeLists.txt
+++ b/source/blender/editors/mesh/CMakeLists.txt
@@ -43,6 +43,7 @@ set(SRC
   editface.c
   editmesh_add.c
   editmesh_add_gizmo.c
+  editmesh_automerge.c
   editmesh_bevel.c
   editmesh_bisect.c
   editmesh_extrude.c
diff --git a/source/blender/editors/mesh/editmesh_automerge.c b/source/blender/editors/mesh/editmesh_automerge.c
new file mode 100644
index 00000000000..82f53aafad8
--- /dev/null
+++ b/source/blender/editors/mesh/editmesh_automerge.c
@@ -0,0 +1,517 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * The Original Code is Copyright (C) 2019 Blender Foundation.
+ * All rights reserved.
+ */
+
+/** \file
+ * \ingroup edmesh
+ *
+ * Utility functions for merging geometry once transform has finished:
+ *
+ * - #EDBM_automerge
+ * - #EDBM_automerge_and_split
+ */
+
+#include "MEM_guardedalloc.h"
+
+#include "BLI_bitmap.h"
+#include "BLI_math.h"
+#include "BLI_sort.h"
+
+#include "BKE_bvhutils.h"
+#include "BKE_editmesh.h"
+
+#include "WM_api.h"
+
+#include "ED_mesh.h"
+#include "ED_screen.h"
+#include "ED_transform.h"
+
+#include "DNA_object_types.h"
+
+#include "DEG_depsgraph.h"
+
+/* use bmesh operator flags for a few operators */
+#define BMO_ELE_TAG 1
+
+/* -------------------------------------------------------------------- */
+/** \name Auto-Merge Selection
+ *
+ * Used after transform operations.
+ * \{ */
+
+void EDBM_automerge(Object *obedit, bool update, const char hflag, const float dist)
+{
+  BMEditMesh *em = BKE_editmesh_from_object(obedit);
+  BMesh *bm = em->bm;
+  int totvert_prev = bm->totvert;
+
+  BMOperator findop, weldop;
+
+  /* Search for doubles among all vertices, but only merge non-VERT_KEEP
+   * vertices into VERT_KEEP vertices. */
+  BMO_op_initf(bm,
+               &findop,
+               BMO_FLAG_DEFAULTS,
+               "find_doubles verts=%av keep_verts=%Hv dist=%f",
+               hflag,
+               dist);
+
+  BMO_op_exec(bm, &findop);
+
+  /* weld the vertices */
+  BMO_op_init(bm, &weldop, BMO_FLAG_DEFAULTS, "weld_verts");
+  BMO_slot_copy(&findop, slots_out, "targetmap.out", &weldop, slots_in, "targetmap");
+  BMO_op_exec(bm, &weldop);
+
+  BMO_op_finish(bm, &findop);
+  BMO_op_finish(bm, &weldop);
+
+  if ((totvert_prev != bm->totvert) && update) {
+    EDBM_update_generic(em, true, true);
+  }
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Auto-Merge & Split Selection
+ *
+ * Used after transform operations.
+ * \{ */
+
+struct EDBMSplitEdge {
+  BMVert *v;
+  BMEdge *e;
+  float lambda;
+};
+
+struct EDBMSplitBestFaceData {
+  BMEdge **edgenet;
+  int edgenet_len;
+
+  /**
+   * Track the range of vertices in edgenet along the faces normal,
+   * find the lowest since it's most likely to be most co-planar with the face.
+   */
+  float best_face_range_on_normal_axis;
+  BMFace *r_best_face;
+};
+
+struct EDBMSplitEdgeData {
+  BMesh *bm;
+
+  BMEdge *r_edge;
+  float r_lambda;
+};
+
+static bool edbm_vert_pair_share_best_splittable_face_cb(BMFace *f,
+                                                         BMLoop *UNUSED(l_a),
+                                                         BMLoop *UNUSED(l_b),
+                                                         void *userdata)
+{
+  struct EDBMSplitBestFaceData *data = userdata;
+  float no[3], min = FLT_MAX, max = -FLT_MAX;
+  copy_v3_v3(no, f->no);
+
+  BMVert *verts[2] = {NULL};
+  BMEdge **e_iter = &data->edgenet[0];
+  for (int i = data->edgenet_len; i--; e_iter++) {
+    BMIter iter;
+    BMVert *v;
+    BM_ITER_ELEM (v, &iter, *e_iter, BM_VERTS_OF_EDGE) {
+      if (!ELEM(v, verts[0], verts[1])) {
+        float dot = dot_v3v3(v->co, no);
+        if (dot < min) {
+          min = dot;
+        }
+        if (dot > max) {
+          max = dot;
+        }
+      }
+    }
+    verts[0] = (*e_iter)->v1;
+    verts[1] = (*e_iter)->v2;
+  }
+
+  const float test_face_range_on_normal_axis = max - min;
+  if (test_face_range_on_normal_axis < data->best_face_range_on_normal_axis) {
+    data->best_face_range_on_normal_axis = test_face_range_on_normal_axis;
+    data->r_best_face = f;
+  }
+
+  return false;
+}
+
+/* find the best splittable face between the two vertices. */
+static bool edbm_vert_pair_share_splittable_face_cb(BMFace *UNUSED(f),
+                                                    BMLoop *l_a,
+                                                    BMLoop *l_b,
+                                                    void *userdata)
+{
+  float(*data)[3] = userdata;
+  float *v_a_co = data[0];
+  float *v_a_b_dir = data[1];
+
+  float lambda;
+  if (isect_ray_seg_v3(v_a_co, v_a_b_dir, l_a->prev->v->co, l_a->next->v->co, &lambda)) {
+    if (IN_RANGE(lambda, 0.0f, 1.0f)) {
+      return true;
+    }
+    else if (isect_ray_seg_v3(v_a_co, v_a_b_dir, l_b->prev->v->co, l_b->next->v->co, &lambda)) {
+      return IN_RANGE(lambda, 0.0f, 1.0f);
+    }
+  }
+  return false;
+}
+
+static void edbm_automerge_weld_linked_wire_edges_into_linked_faces(BMesh *bm,
+                                                                    BMVert *v,
+                                                                    BMEdge **r_edgenet[],
+                                                                    int *r_edgenet_alloc_len)
+{
+  BMEdge **edgenet = *r_edgenet;
+  int edgenet_alloc_len = *r_edgenet_alloc_len;
+
+  BMIter iter;
+  BMEdge *e;
+  BM_ITER_ELEM (e, &iter, v, BM_EDGES_OF_VERT) {
+    int edgenet_len = 0;
+    BMVert *v_other = v;
+    while (BM_edge_is_wire(e)) {
+      if (edgenet_alloc_len == edgenet_len) {
+        edgenet_alloc_len = (edgenet_alloc_len + 1) * 2;
+        edgenet = MEM_reallocN(edgenet, (edgenet_alloc_len) * sizeof(*edgenet));
+      }
+      edgenet[edgenet_len++] = e;
+      v_other = BM_edge_other_vert(e, v_other);
+      if (v_other == v) {
+        /* Endless loop. */
+        break;
+      }
+
+      BMEdge *e_next = BM_DISK_EDGE_NEXT(e, v_other);
+      if (e_next == e) {
+        /* Vert is wire_endpoint */
+        edgenet_len = 0;
+        break;
+      }
+      e = e_next;
+    }
+
+    BMLoop *dummy;
+    BMFace *best_face;
+    if (edgenet_len == 0) {
+      /* Nothing to do. */
+      continue;
+    }
+    if (edgenet_len == 1) {
+      float data[2][3];
+      copy_v3_v3(data[0], v_other->co);
+      sub_v3_v3v3(data[1], v->co, data[0]);
+      best_face = BM_vert_pair_shared_face_cb(
+          v_other, v, true, edbm_vert_pair_share_splittable_face_cb, &data, &dummy, &dummy);
+    }
+    else {
+      struct EDBMSplitBestFaceData data = {
+          .edgenet = edgenet,
+          .edgenet_len = edgenet_len,
+          .best_face_range_on_normal_axis = FLT_MAX,
+          .r_best_face = NULL,
+      };
+      BM_vert_pair_shared_face_cb(
+          v_other, v, true, edbm_vert_pair_share_best_splittable_face_cb, &data, &dummy, &dummy);
+
+      best_face = data.r_best_face;
+    }
+
+    if (best_face) {
+      BM_face_split_edgenet(bm, best_face, edgenet, edgenet_len, NULL, NULL);
+    }
+  }
+
+  *r_edgenet = edgenet;
+  *r_edgenet_alloc_len = edgenet_alloc_len;
+}
+
+static void ebbm_automerge_and_split_find_duplicate_cb(void *userdata,
+                                                       int index,
+                                                       const float co[3],
+                                                       BVHTreeNearest *nearest)
+{
+  struct EDBMSplitEdgeData *data = userdata;
+  BMEdge *e = BM_edge_at_index(data->bm, index);
+  float lambda = line_point_factor_v3_ex(co, e->v1->co, e->v2->co, 0.0f, -1.0f);
+  if (IN_RANGE(lambda, 0.0f, 1.0f)) {
+    float near_co[3];
+    interp_v3_v3v3(near_co, e->v1->co, e->v2->co, lambda);
+    float dist_sq = len_squared_v3v3(near_co, co);
+    if (dist_sq < nearest->dist_sq) {
+      nearest->dist_sq = dist_sq;
+      nearest->index = index;
+
+      data->r_edge = e;
+      data->r_lambda = lambda;
+    }
+  }
+}
+
+static int edbm_automerge_and_split_sort_cmp_by_keys_cb(const void *index1_v,
+                                                        const void *index2_v,
+                                                        void *keys_v)
+{
+  const struct EDBMSplitEdge *cuts = keys_v;
+  const int *index1 = (int *)index1_v;
+  const int *index2 = (int *)index2_v;
+
+  if (cuts[*index1].lambda > cuts[*index2].lambda) {
+    return 1;
+  }
+  else {
+    return -1;
+  }
+}
+
+void EDBM_automerge_and_split(Object *obedit,
+                              bool split_edges,
+                              bool split_faces,
+                              bool update,
+                              const char hflag,
+                              const float dist)
+{
+  bool ok = false;
+
+  BMEditMesh *em = BKE_editmesh_from_object(obedit);
+  BMesh *bm = em->bm;
+  BMOperator findop, weldop;
+  BMOpSlot *slot_targetmap;
+  BMIter iter;
+  BMVert *v;
+
+  /* tag and count the verts to be tested. */
+  BM_mesh_elem_toolflags_ensure(bm);
+  int verts_len = 0;
+  BM_ITER_MESH (v, &iter, bm, BM_VERTS_OF_MESH) {
+    if (BM_elem_flag_test(v, hflag)) {
+      BM_elem_flag_enable(v, BM_ELEM_TAG);
+      BMO_vert_flag_enable(bm, v, BMO_ELE_TAG);
+      verts_len++;
+    }
+    else {
+      BM_elem_flag_disable(v, BM_ELEM_TAG);
+    }
+  }
+
+  /* Search for doubles among all vertices, but only merge non-BMO_ELE_TAG
+   * vertices into BMO_ELE_TAG vertices. */
+  BMO_op_initf(bm, &findop, 0, "find_doubles verts=%av keep_verts=%Fv dist=%f", BMO_ELE_TAG, dist);
+  BMO_op_exec(bm, &findop);
+
+  /* Init weld_verts operator to later fill the targetmap. */
+  BMO_op_init(bm, &weldop, 0, "weld_verts");
+  BMO_slot_copy(&findop, slots_out, "targetmap.out", &weldop, slots_in, "targetmap");
+
+  slot_targetmap = BMO_slot_get(weldop.slots_in, "targetmap");
+
+  /* Remove duplicate vertices from the split edge test and check and split faces. */
+  GHashIterator gh_iter;
+  GHash *ghash_targetmap = BMO_SLOT_AS_GHASH(slot_targetmap);
+  GHASH_ITER (gh_iter, ghash_targetmap) {
+    v = BLI_ghashIterator_getKey(&gh_iter);
+    BMVert *v_dst = BLI_ghashIterator_getValue(&gh_iter);
+    if (!BM_elem_flag_test(v, BM_ELEM_TAG)) {
+      /* Should this happen? */
+      SWAP(BMVert *, v, v_dst);
+    }
+    BLI_assert(BM_elem_flag_test(v, BM_ELEM_TAG));
+    BM_elem_flag_disable(v, BM_ELEM_TAG);
+
+    ok = true;
+    verts_len--;
+  }
+
+  int totedge = bm->totedge;
+  if (totedge == 0 || verts_len == 0) {
+    split_edges = false;
+  }
+
+  if (split_edges) {
+    /* Count and tag edges. */
+    BMEdge *e;
+    int edges_len = 0;
+    BM_ITER_MESH (e, &iter, bm, BM_EDGES_OF_MESH) {
+      if (!BM_elem_flag_test(e, BM_ELEM_HIDDEN) && !BM_elem_flag_test(e->v1, BM_ELEM_TAG) &&
+          !BM_elem_flag_test(e->v2, BM_ELEM_TAG)) {
+        BM_elem_flag_enable(e, BM_ELEM_TAG);
+        edges_len++;
+      }
+      else {
+        BM_elem_flag_disable(e, BM_ELEM_TAG);
+      }
+    }
+
+    if (edges_len) {
+      /* Use `e->head.index` to count intersections. */
+      bm->elem_index_dirty &= ~BM_EDGE;
+
+      /* Create a BVHTree of edges with `dist` as epsilon. */
+      BVHTree *tree_edges = BLI_bvhtree_new(edges_len, dist, 2, 6);
+      int i;
+      BM_ITER_MESH_INDEX (e, &iter, bm, BM_EDGES_OF_MESH, i) {
+        if (BM_elem_flag_test(e, BM_ELEM_TAG)) {
+          float co[2][3];
+          copy_v3_v3(co[0], e->v1->co);
+          copy_v3_v3(co[1], e->v2->co);
+
+          BLI_bvhtree_insert(tree_edges, i, co[0], 2);
+
+          e->head.index = 0;
+        }
+      }
+      BLI_bvhtree_balance(tree_edges);
+
+      struct EDBMSplitEdge *cuts_iter, *cuts;
+
+      /* Store all intersections in this array. */
+      cuts = MEM_mallocN(verts_len * sizeof(*cuts), __func__);
+      cuts_iter = &cuts[0];
+
+      int cuts_len = 0;
+      int cut_edges_len = 0;
+      float dist_sq = SQUARE(dist);
+      struct EDBMSplitEdgeData data = {bm};
+
+      /* Start the search for intersections. */
+      BM_ITER_MESH (v, &iter, bm, BM_VERTS_OF_MESH) {
+        if (BM_elem_flag_test(v, BM_ELEM_TAG)) {
+          float co[3];
+          copy_v3_v3(co, v->co);
+          int e_index = BLI_bvhtree_find_nearest_first(
+              tree_edges, co, dist_sq, ebbm_automerge_and_split_find_duplicate_cb, &data);
+
+          if (e_index != -1) {
+            e = data.r_edge;
+            e->head.index++;
+
+            cuts_iter->v = v;
+            cuts_iter->e = e;
+            cuts_iter->lambda = data.r_lambda;
+            cuts_iter++;
+            cuts_len++;
+
+            if (BM_elem_flag_test(e, BM_ELEM_TAG)) {
+              BM_elem_flag_disable(e, BM_ELEM_TAG);
+              cut_edges_len++;
+            }
+          }
+        }
+      }
+      BLI_bvhtree_free(tree_edges);
+
+      if (cuts_len) {
+        /* Map intersections per edge. */
+        union {
+          struct {
+            int cuts_len;
+            int cuts_index[];
+          };
+          int as_int[0];
+        } * e_map_iter, *e_map;
+
+        e_map = MEM_mallocN((cut_edges_len * sizeof(*e_map)) +
+                                (cuts_len * sizeof(*(e_map->cuts_index))),
+                            __func__);
+
+        int map_len = 0;
+        cuts_iter = &cuts[0];
+        for (i = 0; i < cuts_len; i++, cuts_iter++) {
+          e = cuts_iter->e;
+          if (!BM_elem_flag_test(e, BM_ELEM_TAG)) {
+            BM_elem_flag_enable(e, BM_ELEM_TAG);
+            int e_cuts_len = e->head.index;
+
+            e_map_iter = (void *)&e_map->as_int[map_len];
+            e_map_iter->cuts_len = e_cuts_len;
+            e_map_iter->cuts_index[0] = i;
+
+            /* Use `e->head.index` to indicate which slot to fill with the `cuts` index. */
+            e->head.index = map_len + 1;
+            map_len += 1 + e_cuts_len;
+          }
+          else {
+            e_map->as_int[++e->head.index] = i;
+          }
+        }
+
+        /* Split Edges and Faces. */
+        for (i = 0; i < map_len;
+             e_map_iter = (void *)&e_map->as_int[i], i += 1 + e_map_iter->cuts_len) {
+
+          /* sort by lambda. */
+          BLI_qsort_r(e_map_iter->cuts_index,
+                      e_map_iter->cuts_len,
+                      sizeof(*(e_map->cuts_index)),
+                      edbm_automerge_and_split_sort_cmp_by_keys_cb,
+                      cuts);
+
+          float lambda, lambda_prev = 0.0f;
+          for (int j = 0; j < e_map_iter->cuts_len; j++) {
+            cuts_iter = &cuts[e_map_iter->cuts_index[j]];
+            lambda = (cuts_iter->lambda - lambda_prev) / (1.0f - lambda_prev);
+            lambda_prev = cuts_iter->lambda;
+            v = cuts_iter->v;
+            e = cuts_iter->e;
+
+            BMVert *v_new = BM_edge_split(bm, e, e->v1, NULL, lambda);
+
+            BMO_slot_map_elem_insert(&weldop, slot_targetmap, v_new, v);
+          }
+        }
+
+        ok = true;
+        MEM_freeN(e_map);
+      }
+
+      MEM_freeN(cuts);
+    }
+  }
+
+  BMO_op_exec(bm, &weldop);
+
+  BMEdge **edgenet = NULL;
+  int edgenet_alloc_len = 0;
+  if (split_faces) {
+    GHASH_ITER (gh_iter, ghash_targetmap) {
+      v = BLI_ghashIterator_getValue(&gh_iter);
+      BLI_assert(BM_elem_flag_test(v, hflag) || hflag == BM_ELEM_TAG);
+      edbm_automerge_weld_linked_wire_edges_into_linked_faces(bm, v, &edgenet, &edgenet_alloc_len);
+    }
+  }
+
+  if (edgenet) {
+    MEM_freeN(edgenet);
+  }
+
+  BMO_op_finish(bm, &findop);
+  BMO_op_finish(bm, &weldop);
+
+  if (LIKELY(ok) && update) {
+    EDBM_update_generic(em, true, true);
+  }
+}
+
+/** \} */
diff --git a/source/blender/editors/mesh/editmesh_extrude_spin_gizmo.c b/source/blender/editors/mesh/editmesh_extrude_spin_gizmo.c
index 74700e59e99..7155348fed5 100644
--- a/source/blender/editors/mesh/editmesh_extrude_spin_gizmo.c
+++ b/source/blender/editors/mesh/editmesh_extrude_spin_gizmo.c
@@ -1008,8 +1008,17 @@ static void gizmo_mesh_spin_redo_setup(const bContext *C, wmGizmoGroup *gzgroup)
                                       });
   }
 
-  /* Become modal as soon as it's started. */
-  gizmo_mesh_spin_redo_modal_from_setup(C, gzgroup);
+  wmWindow *win = CTX_wm_window(C);
+  if (win && win->active) {
+    bScreen *screen = WM_window_get_active_screen(win);
+    if (screen->active_region) {
+      ARegion *ar = CTX_wm_region(C);
+      if (screen->active_region == ar) {
+        /* Become modal as soon as it's started. */
+        gizmo_mesh_spin_redo_modal_from_setup(C, gzgroup);
+      }
+    }
+  }
 }
 
 static void gizmo_mesh_spin_redo_draw_prepare(const bContext *UNUSED(C), wmGizmoGroup *gzgroup)
diff --git a/source/blender/editors/mesh/editmesh_knife.c b/source/blender/editors/mesh/editmesh_knife.c
index bb584094580..61f9dc43c0f 100644
--- a/source/blender/editors/mesh/editmesh_knife.c
+++ b/source/blender/editors/mesh/editmesh_knife.c
@@ -2658,7 +2658,7 @@ static void knifetool_init_bmbvh(KnifeTool_OpData *kcd)
   Object *obedit_eval = (Object *)DEG_get_evaluated_id(kcd->vc.depsgraph, &kcd->em->ob->id);
   BMEditMesh *em_eval = BKE_editmesh_from_object(obedit_eval);
 
-  kcd->cagecos = (const float(*)[3])BKE_editmesh_vertexCos_get(
+  kcd->cagecos = (const float(*)[3])BKE_editmesh_vert_coords_alloc(
       kcd->vc.depsgraph, em_eval, scene_eval, NULL);
 
   kcd->bmbvh = BKE_bmbvh_new_from_editmesh(
@@ -2949,7 +2949,7 @@ static int knifetool_modal(bContext *C, wmOperator *op, const wmEvent *event)
       case KNF_MODAL_ADD_CUT_CLOSED:
         if (kcd->mode == MODE_DRAGGING) {
 
-          /* shouldn't be possible with default key-layout, just incase... */
+          /* Shouldn't be possible with default key-layout, just in case. */
           if (kcd->is_drag_hold) {
             kcd->is_drag_hold = false;
             knifetool_update_mval(kcd, kcd->curr.mval);
diff --git a/source/blender/editors/mesh/editmesh_polybuild.c b/source/blender/editors/mesh/editmesh_polybuild.c
index 088d1672cc9..21c850160dd 100644
--- a/source/blender/editors/mesh/editmesh_polybuild.c
+++ b/source/blender/editors/mesh/editmesh_polybuild.c
@@ -122,15 +122,160 @@ static bool edbm_preselect_or_active_init_viewcontext(bContext *C,
   return ok;
 }
 
+static int edbm_polybuild_transform_at_cursor_invoke(bContext *C,
+                                                     wmOperator *UNUSED(op),
+                                                     const wmEvent *UNUSED(event))
+{
+  ViewContext vc;
+  Base *basact = NULL;
+  BMElem *ele_act = NULL;
+  edbm_preselect_or_active_init_viewcontext(C, &vc, &basact, &ele_act);
+  BMEditMesh *em = vc.em;
+  BMesh *bm = em->bm;
+
+  invert_m4_m4(vc.obedit->imat, vc.obedit->obmat);
+  ED_view3d_init_mats_rv3d(vc.obedit, vc.rv3d);
+
+  if (!ele_act) {
+    return OPERATOR_CANCELLED;
+  }
+
+  edbm_selectmode_ensure(vc.scene, vc.em, SCE_SELECT_VERTEX);
+
+  edbm_flag_disable_all_multi(vc.view_layer, vc.v3d, BM_ELEM_SELECT);
+
+  if (ele_act->head.htype == BM_VERT) {
+    BM_vert_select_set(bm, (BMVert *)ele_act, true);
+  }
+  if (ele_act->head.htype == BM_EDGE) {
+    BM_edge_select_set(bm, (BMEdge *)ele_act, true);
+  }
+  if (ele_act->head.htype == BM_FACE) {
+    BM_face_select_set(bm, (BMFace *)ele_act, true);
+  }
+
+  EDBM_mesh_normals_update(em);
+  EDBM_update_generic(em, true, true);
+  if (basact != NULL) {
+    if (vc.view_layer->basact != basact) {
+      ED_object_base_activate(C, basact);
+    }
+  }
+  BM_select_history_store(bm, ele_act);
+  WM_event_add_mousemove(C);
+  return OPERATOR_FINISHED;
+}
+
+void MESH_OT_polybuild_transform_at_cursor(wmOperatorType *ot)
+{
+  /* identifiers */
+  ot->name = "Poly Build Transform at Cursor";
+  ot->idname = "MESH_OT_polybuild_transform_at_cursor";
+
+  /* api callbacks */
+  ot->invoke = edbm_polybuild_transform_at_cursor_invoke;
+  ot->poll = EDBM_view3d_poll;
+
+  /* flags */
+  ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
+
+  /* to give to transform */
+  Transform_Properties(ot, P_PROPORTIONAL | P_MIRROR_DUMMY);
+}
+
+static int edbm_polybuild_delete_at_cursor_invoke(bContext *C,
+                                                  wmOperator *op,
+                                                  const wmEvent *UNUSED(event))
+{
+  bool changed = false;
+
+  ViewContext vc;
+  Base *basact = NULL;
+  BMElem *ele_act = NULL;
+  edbm_preselect_or_active_init_viewcontext(C, &vc, &basact, &ele_act);
+  BMEditMesh *em = vc.em;
+  BMesh *bm = em->bm;
+
+  invert_m4_m4(vc.obedit->imat, vc.obedit->obmat);
+  ED_view3d_init_mats_rv3d(vc.obedit, vc.rv3d);
+
+  if (!ele_act) {
+    return OPERATOR_CANCELLED;
+  }
+
+  edbm_selectmode_ensure(vc.scene, vc.em, SCE_SELECT_VERTEX);
+
+  if (ele_act->head.htype == BM_FACE) {
+    BMFace *f_act = (BMFace *)ele_act;
+    EDBM_flag_disable_all(em, BM_ELEM_TAG);
+    BM_elem_flag_enable(f_act, BM_ELEM_TAG);
+    if (!EDBM_op_callf(em, op, "delete geom=%hf context=%i", BM_ELEM_TAG, DEL_FACES)) {
+      return OPERATOR_CANCELLED;
+    }
+    changed = true;
+  }
+  if (ele_act->head.htype == BM_VERT) {
+    BMVert *v_act = (BMVert *)ele_act;
+    if (BM_vert_is_edge_pair(v_act)) {
+      BM_edge_collapse(bm, v_act->e, v_act, true, true);
+      changed = true;
+    }
+    else {
+      EDBM_flag_disable_all(em, BM_ELEM_TAG);
+      BM_elem_flag_enable(v_act, BM_ELEM_TAG);
+
+      if (!EDBM_op_callf(em,
+                         op,
+                         "dissolve_verts verts=%hv use_face_split=%b use_boundary_tear=%b",
+                         BM_ELEM_TAG,
+                         false,
+                         false)) {
+        return OPERATOR_CANCELLED;
+      }
+      changed = true;
+    }
+  }
+
+  if (changed) {
+    EDBM_mesh_normals_update(em);
+    EDBM_update_generic(em, true, true);
+    if (basact != NULL) {
+      if (vc.view_layer->basact != basact) {
+        ED_object_base_activate(C, basact);
+      }
+    }
+    WM_event_add_mousemove(C);
+    return OPERATOR_FINISHED;
+  }
+  else {
+    return OPERATOR_CANCELLED;
+  }
+}
+
+void MESH_OT_polybuild_delete_at_cursor(wmOperatorType *ot)
+{
+  /* identifiers */
+  ot->name = "Poly Build Delete at Cursor";
+  ot->idname = "MESH_OT_polybuild_delete_at_cursor";
+
+  /* api callbacks */
+  ot->invoke = edbm_polybuild_delete_at_cursor_invoke;
+  ot->poll = EDBM_view3d_poll;
+
+  /* flags */
+  ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
+
+  /* to give to transform */
+  Transform_Properties(ot, P_PROPORTIONAL | P_MIRROR_DUMMY);
+}
+
 /** \} */
 
 /* -------------------------------------------------------------------- */
 /** \name Face at Cursor
  * \{ */
 
-static int edbm_polybuild_face_at_cursor_invoke(bContext *C,
-                                                wmOperator *UNUSED(op),
-                                                const wmEvent *event)
+static int edbm_polybuild_face_at_cursor_invoke(bContext *C, wmOperator *op, const wmEvent *event)
 {
   float center[3];
   bool changed = false;
@@ -168,20 +313,27 @@ static int edbm_polybuild_face_at_cursor_invoke(bContext *C,
     mul_m4_v3(vc.obedit->obmat, center);
     ED_view3d_win_to_3d_int(vc.v3d, vc.ar, center, event->mval, center);
     mul_m4_v3(vc.obedit->imat, center);
-
-    BMVert *v_tri[3];
-    v_tri[0] = e_act->v1;
-    v_tri[1] = e_act->v2;
-    v_tri[2] = BM_vert_create(bm, center, NULL, BM_CREATE_NOP);
-    if (e_act->l && e_act->l->v == v_tri[0]) {
-      SWAP(BMVert *, v_tri[0], v_tri[1]);
+    if (f_reference->len == 3 && RNA_boolean_get(op->ptr, "create_quads")) {
+      const float fac = line_point_factor_v3(center, e_act->v1->co, e_act->v2->co);
+      BMVert *v_new = BM_edge_split(bm, e_act, e_act->v1, NULL, CLAMPIS(fac, 0.0f, 1.0f));
+      copy_v3_v3(v_new->co, center);
+      edbm_flag_disable_all_multi(vc.view_layer, vc.v3d, BM_ELEM_SELECT);
+      BM_vert_select_set(bm, v_new, true);
+      BM_select_history_store(bm, v_new);
+    }
+    else {
+      BMVert *v_tri[3];
+      v_tri[0] = e_act->v1;
+      v_tri[1] = e_act->v2;
+      v_tri[2] = BM_vert_create(bm, center, NULL, BM_CREATE_NOP);
+      if (e_act->l && e_act->l->v == v_tri[0]) {
+        SWAP(BMVert *, v_tri[0], v_tri[1]);
+      }
+      BM_face_create_verts(bm, v_tri, 3, f_reference, BM_CREATE_NOP, true);
+      edbm_flag_disable_all_multi(vc.view_layer, vc.v3d, BM_ELEM_SELECT);
+      BM_vert_select_set(bm, v_tri[2], true);
+      BM_select_history_store(bm, v_tri[2]);
     }
-    // BMFace *f_new =
-    BM_face_create_verts(bm, v_tri, 3, f_reference, BM_CREATE_NOP, true);
-
-    edbm_flag_disable_all_multi(vc.view_layer, vc.v3d, BM_ELEM_SELECT);
-    BM_vert_select_set(bm, v_tri[2], true);
-    BM_select_history_store(bm, v_tri[2]);
     changed = true;
   }
   else if (ele_act->head.htype == BM_VERT) {
@@ -281,6 +433,11 @@ void MESH_OT_polybuild_face_at_cursor(wmOperatorType *ot)
   /* flags */
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 
+  RNA_def_boolean(ot->srna,
+                  "create_quads",
+                  true,
+                  "Create quads",
+                  "Automatically split edges in triangles to maintain quad topology");
   /* to give to transform */
   Transform_Properties(ot, P_PROPORTIONAL | P_MIRROR_DUMMY);
 }
diff --git a/source/blender/editors/mesh/editmesh_preselect_elem.c b/source/blender/editors/mesh/editmesh_preselect_elem.c
index a3e684a5493..05c4da68355 100644
--- a/source/blender/editors/mesh/editmesh_preselect_elem.c
+++ b/source/blender/editors/mesh/editmesh_preselect_elem.c
@@ -75,20 +75,49 @@ struct EditMesh_PreSelElem {
 
   float (*verts)[3];
   int verts_len;
+
+  float (*preview_tris)[3][3];
+  int preview_tris_len;
+  float (*preview_lines)[2][3];
+  int preview_lines_len;
+
+  eEditMesh_PreSelPreviewAction preview_action;
 };
 
+void EDBM_preselect_action_set(struct EditMesh_PreSelElem *psel,
+                               eEditMesh_PreSelPreviewAction action)
+{
+  psel->preview_action = action;
+}
+
+eEditMesh_PreSelPreviewAction EDBM_preselect_action_get(struct EditMesh_PreSelElem *psel)
+{
+  return psel->preview_action;
+}
+
 struct EditMesh_PreSelElem *EDBM_preselect_elem_create(void)
 {
   struct EditMesh_PreSelElem *psel = MEM_callocN(sizeof(*psel), __func__);
+  psel->preview_action = PRESELECT_ACTION_TRANSFORM;
   return psel;
 }
 
 void EDBM_preselect_elem_destroy(struct EditMesh_PreSelElem *psel)
 {
   EDBM_preselect_elem_clear(psel);
+  EDBM_preselect_preview_clear(psel);
   MEM_freeN(psel);
 }
 
+void EDBM_preselect_preview_clear(struct EditMesh_PreSelElem *psel)
+{
+  MEM_SAFE_FREE(psel->preview_tris);
+  psel->preview_tris_len = 0;
+
+  MEM_SAFE_FREE(psel->preview_lines);
+  psel->preview_lines_len = 0;
+}
+
 void EDBM_preselect_elem_clear(struct EditMesh_PreSelElem *psel)
 {
   MEM_SAFE_FREE(psel->edges);
@@ -112,9 +141,42 @@ void EDBM_preselect_elem_draw(struct EditMesh_PreSelElem *psel, const float matr
   uint pos = GPU_vertformat_attr_add(immVertexFormat(), "pos", GPU_COMP_F32, 3, GPU_FETCH_FLOAT);
 
   immBindBuiltinProgram(GPU_SHADER_3D_UNIFORM_COLOR);
-  immUniformColor3ub(255, 0, 255);
+
+  immUniformColor4ub(141, 171, 186, 100);
+  if (psel->preview_action != PRESELECT_ACTION_TRANSFORM) {
+    if (psel->preview_tris_len > 0) {
+      immBegin(GPU_PRIM_TRIS, psel->preview_tris_len * 3);
+
+      for (int i = 0; i < psel->preview_tris_len; i++) {
+        immVertex3fv(pos, psel->preview_tris[i][0]);
+        immVertex3fv(pos, psel->preview_tris[i][1]);
+        immVertex3fv(pos, psel->preview_tris[i][2]);
+      }
+      immEnd();
+    }
+
+    if (psel->preview_lines_len > 0) {
+
+      immUniformColor4ub(3, 161, 252, 200);
+      GPU_line_width(2.0f);
+      immBegin(GPU_PRIM_LINES, psel->preview_lines_len * 2);
+      for (int i = 0; i < psel->preview_lines_len; i++) {
+        immVertex3fv(pos, psel->preview_lines[i][0]);
+        immVertex3fv(pos, psel->preview_lines[i][1]);
+      }
+      immEnd();
+    }
+  }
+
+  if (psel->preview_action == PRESELECT_ACTION_DELETE) {
+    immUniformColor4ub(252, 49, 10, 200);
+  }
+  else {
+    immUniformColor4ub(3, 161, 252, 200);
+  }
 
   if (psel->edges_len > 0) {
+    GPU_line_width(3.0f);
     immBegin(GPU_PRIM_LINES, psel->edges_len * 2);
 
     for (int i = 0; i < psel->edges_len; i++) {
@@ -126,7 +188,7 @@ void EDBM_preselect_elem_draw(struct EditMesh_PreSelElem *psel, const float matr
   }
 
   if (psel->verts_len > 0) {
-    GPU_point_size(3.0f);
+    GPU_point_size(4.0f);
 
     immBegin(GPU_PRIM_POINTS, psel->verts_len);
 
@@ -167,6 +229,122 @@ static void view3d_preselect_mesh_elem_update_from_edge(struct EditMesh_PreSelEl
   psel->edges_len = 1;
 }
 
+static void view3d_preselect_update_preview_triangle_from_vert(struct EditMesh_PreSelElem *psel,
+                                                               ViewContext *vc,
+                                                               BMesh *UNUSED(bm),
+                                                               BMVert *eed,
+                                                               const int mval[2])
+{
+  BMVert *v_act = eed;
+  BMEdge *e_pair[2] = {NULL};
+  float center[3];
+
+  if (v_act->e != NULL) {
+    for (uint allow_wire = 0; allow_wire < 2 && (e_pair[1] == NULL); allow_wire++) {
+      int i = 0;
+      BMEdge *e_iter = v_act->e;
+      do {
+        if ((BM_elem_flag_test(e_iter, BM_ELEM_HIDDEN) == false) &&
+            (allow_wire ? BM_edge_is_wire(e_iter) : BM_edge_is_boundary(e_iter))) {
+          if (i == 2) {
+            e_pair[0] = e_pair[1] = NULL;
+            break;
+          }
+          e_pair[i++] = e_iter;
+        }
+      } while ((e_iter = BM_DISK_EDGE_NEXT(e_iter, v_act)) != v_act->e);
+    }
+  }
+
+  if (e_pair[1] != NULL) {
+    mul_v3_m4v3(center, vc->obedit->obmat, v_act->co);
+    ED_view3d_win_to_3d_int(vc->v3d, vc->ar, center, mval, center);
+    mul_m4_v3(vc->obedit->imat, center);
+
+    psel->preview_tris = MEM_mallocN(sizeof(*psel->preview_tris) * 2, __func__);
+    psel->preview_lines = MEM_mallocN(sizeof(*psel->preview_lines) * 4, __func__);
+
+    copy_v3_v3(psel->preview_tris[0][0], e_pair[0]->v1->co);
+    copy_v3_v3(psel->preview_tris[0][1], e_pair[0]->v2->co);
+    copy_v3_v3(psel->preview_tris[0][2], center);
+
+    copy_v3_v3(psel->preview_tris[1][0], e_pair[1]->v1->co);
+    copy_v3_v3(psel->preview_tris[1][1], e_pair[1]->v2->co);
+    copy_v3_v3(psel->preview_tris[1][2], center);
+
+    copy_v3_v3(psel->preview_lines[0][0], e_pair[0]->v1->co);
+    copy_v3_v3(psel->preview_lines[0][1], e_pair[0]->v2->co);
+
+    copy_v3_v3(psel->preview_lines[1][0], e_pair[1]->v1->co);
+    copy_v3_v3(psel->preview_lines[1][1], e_pair[1]->v2->co);
+
+    copy_v3_v3(psel->preview_lines[2][0], center);
+    if (e_pair[0]->v1 == v_act) {
+      copy_v3_v3(psel->preview_lines[2][1], e_pair[0]->v2->co);
+    }
+    else {
+      copy_v3_v3(psel->preview_lines[2][1], e_pair[0]->v1->co);
+    }
+
+    copy_v3_v3(psel->preview_lines[3][0], center);
+    if (e_pair[1]->v1 == v_act) {
+      copy_v3_v3(psel->preview_lines[3][1], e_pair[1]->v2->co);
+    }
+    else {
+      copy_v3_v3(psel->preview_lines[3][1], e_pair[1]->v1->co);
+    }
+    psel->preview_tris_len = 2;
+    psel->preview_lines_len = 4;
+  }
+}
+
+static void view3d_preselect_update_preview_triangle_from_face(struct EditMesh_PreSelElem *psel,
+                                                               ViewContext *UNUSED(vc),
+                                                               BMesh *UNUSED(bm),
+                                                               BMFace *efa,
+                                                               const int UNUSED(mval[2]))
+{
+  float(*preview_lines)[2][3] = MEM_mallocN(sizeof(*psel->edges) * efa->len, __func__);
+  BMLoop *l_iter, *l_first;
+  l_iter = l_first = BM_FACE_FIRST_LOOP(efa);
+  int i = 0;
+  do {
+    vcos_get_pair(&l_iter->e->v1, preview_lines[i++], NULL);
+  } while ((l_iter = l_iter->next) != l_first);
+  psel->preview_lines = preview_lines;
+  psel->preview_lines_len = efa->len;
+}
+
+static void view3d_preselect_update_preview_triangle_from_edge(struct EditMesh_PreSelElem *psel,
+                                                               ViewContext *vc,
+                                                               BMesh *UNUSED(bm),
+                                                               BMEdge *eed,
+                                                               const int mval[2])
+{
+  float center[3];
+  psel->preview_tris = MEM_mallocN(sizeof(*psel->preview_tris), __func__);
+  psel->preview_lines = MEM_mallocN(sizeof(*psel->preview_lines) * 3, __func__);
+  mid_v3_v3v3(center, eed->v1->co, eed->v2->co);
+  mul_m4_v3(vc->obedit->obmat, center);
+  ED_view3d_win_to_3d_int(vc->v3d, vc->ar, center, mval, center);
+  mul_m4_v3(vc->obedit->imat, center);
+
+  copy_v3_v3(psel->preview_tris[0][0], eed->v1->co);
+  copy_v3_v3(psel->preview_tris[0][1], eed->v2->co);
+  copy_v3_v3(psel->preview_tris[0][2], center);
+
+  copy_v3_v3(psel->preview_lines[0][0], eed->v1->co);
+  copy_v3_v3(psel->preview_lines[0][1], eed->v2->co);
+
+  copy_v3_v3(psel->preview_lines[1][0], eed->v2->co);
+  copy_v3_v3(psel->preview_lines[1][1], center);
+
+  copy_v3_v3(psel->preview_lines[2][0], center);
+  copy_v3_v3(psel->preview_lines[2][1], eed->v1->co);
+  psel->preview_tris_len = 1;
+  psel->preview_lines_len = 3;
+}
+
 static void view3d_preselect_mesh_elem_update_from_face(struct EditMesh_PreSelElem *psel,
                                                         BMesh *UNUSED(bm),
                                                         BMFace *efa,
@@ -209,4 +387,28 @@ void EDBM_preselect_elem_update_from_single(struct EditMesh_PreSelElem *psel,
   }
 }
 
+void EDBM_preselect_elem_update_preview(struct EditMesh_PreSelElem *psel,
+                                        struct ViewContext *vc,
+                                        struct BMesh *bm,
+                                        struct BMElem *ele,
+                                        const int mval[2])
+{
+  EDBM_preselect_preview_clear(psel);
+
+  switch (ele->head.htype) {
+    case BM_VERT:
+      if (EDBM_preselect_action_get(psel) == PRESELECT_ACTION_CREATE) {
+        view3d_preselect_update_preview_triangle_from_vert(psel, vc, bm, (BMVert *)ele, mval);
+      }
+      break;
+    case BM_EDGE:
+      view3d_preselect_update_preview_triangle_from_edge(psel, vc, bm, (BMEdge *)ele, mval);
+      break;
+    case BM_FACE:
+      view3d_preselect_update_preview_triangle_from_face(psel, vc, bm, (BMFace *)ele, mval);
+      break;
+    default:
+      BLI_assert(0);
+  }
+}
 /** \} */
diff --git a/source/blender/editors/mesh/editmesh_select.c b/source/blender/editors/mesh/editmesh_select.c
index 94ffd9a34d6..c0bd9e9f14c 100644
--- a/source/blender/editors/mesh/editmesh_select.c
+++ b/source/blender/editors/mesh/editmesh_select.c
@@ -24,7 +24,6 @@
 #include "MEM_guardedalloc.h"
 
 #include "BLI_bitmap.h"
-#include "BLI_bitmap_draw_2d.h"
 #include "BLI_listbase.h"
 #include "BLI_linklist.h"
 #include "BLI_linklist_stack.h"
@@ -35,13 +34,9 @@
 
 #include "BKE_context.h"
 #include "BKE_report.h"
-#include "BKE_paint.h"
 #include "BKE_editmesh.h"
 #include "BKE_layer.h"
 
-#include "IMB_imbuf_types.h"
-#include "IMB_imbuf.h"
-
 #include "WM_api.h"
 #include "WM_types.h"
 
@@ -67,7 +62,6 @@
 #include "DEG_depsgraph.h"
 #include "DEG_depsgraph_query.h"
 
-#include "DRW_engine.h"
 #include "DRW_select_buffer.h"
 
 #include "mesh_intern.h" /* own include */
@@ -172,30 +166,6 @@ void EDBM_select_mirrored(
 /** \} */
 
 /* -------------------------------------------------------------------- */
-/** \name Select Auto-Merge
- *
- * Used after transform operations.
- * \{ */
-
-void EDBM_automerge(Scene *scene, Object *obedit, bool update, const char hflag)
-{
-  bool ok;
-  BMEditMesh *em = BKE_editmesh_from_object(obedit);
-
-  ok = BMO_op_callf(em->bm,
-                    BMO_FLAG_DEFAULTS,
-                    "automerge verts=%hv dist=%f",
-                    hflag,
-                    scene->toolsettings->doublimit);
-
-  if (LIKELY(ok) && update) {
-    EDBM_update_generic(em, true, true);
-  }
-}
-
-/** \} */
-
-/* -------------------------------------------------------------------- */
 /** \name Back-Buffer OpenGL Selection
  * \{ */
 
@@ -1031,8 +1001,11 @@ bool EDBM_unified_findnearest(ViewContext *vc,
 bool EDBM_unified_findnearest_from_raycast(ViewContext *vc,
                                            Base **bases,
                                            const uint bases_len,
-                                           bool use_boundary,
-                                           int *r_base_index,
+                                           bool use_boundary_vertices,
+                                           bool use_boundary_edges,
+                                           int *r_base_index_vert,
+                                           int *r_base_index_edge,
+                                           int *r_base_index_face,
                                            struct BMVert **r_eve,
                                            struct BMEdge **r_eed,
                                            struct BMFace **r_efa)
@@ -1045,10 +1018,30 @@ bool EDBM_unified_findnearest_from_raycast(ViewContext *vc,
     uint base_index;
     BMElem *ele;
   } best = {0, NULL};
+  /* Currently unused, keep since we may want to pick the best. */
+  UNUSED_VARS(best);
+
+  struct {
+    uint base_index;
+    BMElem *ele;
+  } best_vert = {0, NULL};
+
+  struct {
+    uint base_index;
+    BMElem *ele;
+  } best_edge = {0, NULL};
+
+  struct {
+    uint base_index;
+    BMElem *ele;
+  } best_face = {0, NULL};
 
   if (ED_view3d_win_to_ray_clipped(
           vc->depsgraph, vc->ar, vc->v3d, mval_fl, ray_origin, ray_direction, true)) {
     float dist_sq_best = FLT_MAX;
+    float dist_sq_best_vert = FLT_MAX;
+    float dist_sq_best_edge = FLT_MAX;
+    float dist_sq_best_face = FLT_MAX;
 
     const bool use_vert = (r_eve != NULL);
     const bool use_edge = (r_eed != NULL);
@@ -1078,18 +1071,23 @@ bool EDBM_unified_findnearest_from_raycast(ViewContext *vc,
         BM_mesh_elem_index_ensure(bm, BM_VERT);
       }
 
-      if (use_boundary && (use_vert || use_edge)) {
+      if ((use_boundary_vertices || use_boundary_edges) && (use_vert || use_edge)) {
         BMEdge *e;
         BMIter eiter;
         BM_ITER_MESH (e, &eiter, bm, BM_EDGES_OF_MESH) {
           if ((BM_elem_flag_test(e, BM_ELEM_HIDDEN) == false) && (BM_edge_is_boundary(e))) {
-            if (use_vert) {
+            if (use_vert && use_boundary_vertices) {
               for (uint j = 0; j < 2; j++) {
                 BMVert *v = *((&e->v1) + j);
                 float point[3];
                 mul_v3_m4v3(point, obedit->obmat, coords ? coords[BM_elem_index_get(v)] : v->co);
                 const float dist_sq_test = dist_squared_to_ray_v3_normalized(
                     ray_origin, ray_direction, point);
+                if (dist_sq_test < dist_sq_best_vert) {
+                  dist_sq_best_vert = dist_sq_test;
+                  best_vert.base_index = base_index;
+                  best_vert.ele = (BMElem *)v;
+                }
                 if (dist_sq_test < dist_sq_best) {
                   dist_sq_best = dist_sq_test;
                   best.base_index = base_index;
@@ -1098,7 +1096,7 @@ bool EDBM_unified_findnearest_from_raycast(ViewContext *vc,
               }
             }
 
-            if (use_edge) {
+            if (use_edge && use_boundary_edges) {
               float point[3];
 #if 0
               const float dist_sq_test = dist_squared_ray_to_seg_v3(
@@ -1114,6 +1112,11 @@ bool EDBM_unified_findnearest_from_raycast(ViewContext *vc,
               mul_m4_v3(obedit->obmat, point);
               const float dist_sq_test = dist_squared_to_ray_v3_normalized(
                   ray_origin, ray_direction, point);
+              if (dist_sq_test < dist_sq_best_edge) {
+                dist_sq_best_edge = dist_sq_test;
+                best_edge.base_index = base_index;
+                best_edge.ele = (BMElem *)e;
+              }
               if (dist_sq_test < dist_sq_best) {
                 dist_sq_best = dist_sq_test;
                 best.base_index = base_index;
@@ -1124,46 +1127,55 @@ bool EDBM_unified_findnearest_from_raycast(ViewContext *vc,
           }
         }
       }
-      else {
-        /* Non boundary case. */
-        if (use_vert) {
-          BMVert *v;
-          BMIter viter;
-          BM_ITER_MESH (v, &viter, bm, BM_VERTS_OF_MESH) {
-            if (BM_elem_flag_test(v, BM_ELEM_HIDDEN) == false) {
-              float point[3];
-              mul_v3_m4v3(point, obedit->obmat, v->co);
-              const float dist_sq_test = dist_squared_to_ray_v3_normalized(
-                  ray_origin, ray_direction, v->co);
-              if (dist_sq_test < dist_sq_best) {
-                dist_sq_best = dist_sq_test;
-                best.base_index = base_index;
-                best.ele = (BMElem *)v;
-              }
+      /* Non boundary case. */
+      if (use_vert && !use_boundary_vertices) {
+        BMVert *v;
+        BMIter viter;
+        BM_ITER_MESH (v, &viter, bm, BM_VERTS_OF_MESH) {
+          if (BM_elem_flag_test(v, BM_ELEM_HIDDEN) == false) {
+            float point[3];
+            mul_v3_m4v3(point, obedit->obmat, coords ? coords[BM_elem_index_get(v)] : v->co);
+            const float dist_sq_test = dist_squared_to_ray_v3_normalized(
+                ray_origin, ray_direction, point);
+            if (dist_sq_test < dist_sq_best_vert) {
+              dist_sq_best_vert = dist_sq_test;
+              best_vert.base_index = base_index;
+              best_vert.ele = (BMElem *)v;
+            }
+            if (dist_sq_test < dist_sq_best) {
+              dist_sq_best = dist_sq_test;
+              best.base_index = base_index;
+              best.ele = (BMElem *)v;
             }
           }
         }
-        if (use_edge) {
-          BMEdge *e;
-          BMIter eiter;
-          BM_ITER_MESH (e, &eiter, bm, BM_EDGES_OF_MESH) {
-            if (BM_elem_flag_test(e, BM_ELEM_HIDDEN) == false) {
-              float point[3];
-              if (coords) {
-                mid_v3_v3v3(
-                    point, coords[BM_elem_index_get(e->v1)], coords[BM_elem_index_get(e->v2)]);
-              }
-              else {
-                mid_v3_v3v3(point, e->v1->co, e->v2->co);
-              }
-              mul_m4_v3(obedit->obmat, point);
-              const float dist_sq_test = dist_squared_to_ray_v3_normalized(
-                  ray_origin, ray_direction, point);
-              if (dist_sq_test < dist_sq_best) {
-                dist_sq_best = dist_sq_test;
-                best.base_index = base_index;
-                best.ele = (BMElem *)e;
-              }
+      }
+
+      if (use_edge && !use_boundary_edges) {
+        BMEdge *e;
+        BMIter eiter;
+        BM_ITER_MESH (e, &eiter, bm, BM_EDGES_OF_MESH) {
+          if (BM_elem_flag_test(e, BM_ELEM_HIDDEN) == false) {
+            float point[3];
+            if (coords) {
+              mid_v3_v3v3(
+                  point, coords[BM_elem_index_get(e->v1)], coords[BM_elem_index_get(e->v2)]);
+            }
+            else {
+              mid_v3_v3v3(point, e->v1->co, e->v2->co);
+            }
+            mul_m4_v3(obedit->obmat, point);
+            const float dist_sq_test = dist_squared_to_ray_v3_normalized(
+                ray_origin, ray_direction, point);
+            if (dist_sq_test < dist_sq_best_edge) {
+              dist_sq_best_edge = dist_sq_test;
+              best_edge.base_index = base_index;
+              best_edge.ele = (BMElem *)e;
+            }
+            if (dist_sq_test < dist_sq_best) {
+              dist_sq_best = dist_sq_test;
+              best.base_index = base_index;
+              best.ele = (BMElem *)e;
             }
           }
         }
@@ -1184,6 +1196,11 @@ bool EDBM_unified_findnearest_from_raycast(ViewContext *vc,
             mul_m4_v3(obedit->obmat, point);
             const float dist_sq_test = dist_squared_to_ray_v3_normalized(
                 ray_origin, ray_direction, point);
+            if (dist_sq_test < dist_sq_best_face) {
+              dist_sq_best_face = dist_sq_test;
+              best_face.base_index = base_index;
+              best_face.ele = (BMElem *)f;
+            }
             if (dist_sq_test < dist_sq_best) {
               dist_sq_best = dist_sq_test;
               best.base_index = base_index;
@@ -1195,7 +1212,10 @@ bool EDBM_unified_findnearest_from_raycast(ViewContext *vc,
     }
   }
 
-  *r_base_index = best.base_index;
+  *r_base_index_vert = best_vert.base_index;
+  *r_base_index_edge = best_edge.base_index;
+  *r_base_index_face = best_face.base_index;
+
   if (r_eve) {
     *r_eve = NULL;
   }
@@ -1206,22 +1226,17 @@ bool EDBM_unified_findnearest_from_raycast(ViewContext *vc,
     *r_efa = NULL;
   }
 
-  if (best.ele) {
-    switch (best.ele->head.htype) {
-      case BM_VERT:
-        *r_eve = (BMVert *)best.ele;
-        break;
-      case BM_EDGE:
-        *r_eed = (BMEdge *)best.ele;
-        break;
-      case BM_FACE:
-        *r_efa = (BMFace *)best.ele;
-        break;
-      default:
-        BLI_assert(0);
-    }
+  if (best_vert.ele) {
+    *r_eve = (BMVert *)best_vert.ele;
+  }
+  if (best_edge.ele) {
+    *r_eed = (BMEdge *)best_edge.ele;
   }
-  return (best.ele != NULL);
+  if (best_face.ele) {
+    *r_efa = (BMFace *)best_face.ele;
+  }
+
+  return (best_vert.ele != NULL || best_edge.ele != NULL || best_face.ele != NULL);
 }
 
 /** \} */
diff --git a/source/blender/editors/mesh/editmesh_tools.c b/source/blender/editors/mesh/editmesh_tools.c
index 65c058556f5..0c4db012786 100644
--- a/source/blender/editors/mesh/editmesh_tools.c
+++ b/source/blender/editors/mesh/editmesh_tools.c
@@ -437,6 +437,7 @@ void EDBM_project_snap_verts(bContext *C, Depsgraph *depsgraph, ARegion *ar, BME
                                                     },
                                                     mval,
                                                     NULL,
+                                                    NULL,
                                                     co_proj,
                                                     NULL)) {
           mul_v3_m4v3(eve->co, obedit->imat, co_proj);
@@ -3130,12 +3131,7 @@ static int edbm_remove_doubles_exec(bContext *C, wmOperator *op)
     BM_mesh_elem_hflag_enable_test(em->bm, htype_select, BM_ELEM_TAG, true, true, BM_ELEM_SELECT);
 
     if (use_unselected) {
-      EDBM_op_init(em, &bmop, op, "automerge verts=%hv dist=%f", BM_ELEM_SELECT, threshold);
-      BMO_op_exec(em->bm, &bmop);
-
-      if (!EDBM_op_finish(em, &bmop, op, true)) {
-        continue;
-      }
+      EDBM_automerge(obedit, false, BM_ELEM_SELECT, threshold);
     }
     else {
       EDBM_op_init(em, &bmop, op, "find_doubles verts=%hv dist=%f", BM_ELEM_SELECT, threshold);
@@ -4040,7 +4036,7 @@ static void mesh_separate_material_assign_mat_nr(Main *bmain, Object *ob, const
       ma_obdata = NULL;
     }
 
-    BKE_material_clear_id(bmain, obdata, true);
+    BKE_material_clear_id(bmain, obdata);
     BKE_material_resize_object(bmain, ob, 1, true);
     BKE_material_resize_id(bmain, obdata, 1, true);
 
@@ -4051,7 +4047,7 @@ static void mesh_separate_material_assign_mat_nr(Main *bmain, Object *ob, const
     id_us_plus((ID *)ma_obdata);
   }
   else {
-    BKE_material_clear_id(bmain, obdata, true);
+    BKE_material_clear_id(bmain, obdata);
     BKE_material_resize_object(bmain, ob, 0, true);
     BKE_material_resize_id(bmain, obdata, 0, true);
   }
@@ -5710,7 +5706,7 @@ void MESH_OT_dissolve_limited(wmOperatorType *ot)
                   "use_dissolve_boundaries",
                   false,
                   "All Boundaries",
-                  "Dissolve all vertices inbetween face boundaries");
+                  "Dissolve all vertices in between face boundaries");
   RNA_def_enum_flag(ot->srna,
                     "delimit",
                     rna_enum_mesh_delimit_mode_items,
diff --git a/source/blender/editors/mesh/editmesh_utils.c b/source/blender/editors/mesh/editmesh_utils.c
index d7ed14184fa..522c2f32d27 100644
--- a/source/blender/editors/mesh/editmesh_utils.c
+++ b/source/blender/editors/mesh/editmesh_utils.c
@@ -85,7 +85,7 @@ void EDBM_redo_state_restore(BMBackup backup, BMEditMesh *em, int recalctess)
   tmpbm = NULL;
 
   if (recalctess) {
-    BKE_editmesh_tessface_calc(em);
+    BKE_editmesh_looptri_calc(em);
   }
 }
 
@@ -105,7 +105,7 @@ void EDBM_redo_state_free(BMBackup *backup, BMEditMesh *em, int recalctess)
   backup->bmcopy = NULL;
 
   if (recalctess && em) {
-    BKE_editmesh_tessface_calc(em);
+    BKE_editmesh_looptri_calc(em);
   }
 }
 
@@ -162,7 +162,7 @@ bool EDBM_op_finish(BMEditMesh *em, BMOperator *bmop, wmOperator *op, const bool
     /* when copying, tessellation isn't to for faster copying,
      * but means we need to re-tessellate here */
     if (em->looptris == NULL) {
-      BKE_editmesh_tessface_calc(em);
+      BKE_editmesh_looptri_calc(em);
     }
 
     if (em->ob) {
@@ -292,10 +292,6 @@ void EDBM_mesh_make(Object *ob, const int select_mode, const bool add_key_index)
   Mesh *me = ob->data;
   BMesh *bm;
 
-  if (UNLIKELY(!me->mpoly && me->totface)) {
-    BKE_mesh_convert_mfaces_to_mpolys(me);
-  }
-
   bm = BKE_mesh_to_bmesh(me,
                          ob,
                          add_key_index,
@@ -347,10 +343,6 @@ void EDBM_mesh_load(Main *bmain, Object *ob)
                        .calc_object_remap = true,
                    }));
 
-#ifdef USE_TESSFACE_DEFAULT
-  BKE_mesh_tessface_calc(me);
-#endif
-
   /* Free derived mesh. usually this would happen through depsgraph but there
    * are exceptions like file save that will not cause this, and we want to
    * avoid ending up with an invalid derived mesh then.
@@ -1397,15 +1389,15 @@ void EDBM_stats_update(BMEditMesh *em)
 
 /* so many tools call these that we better make it a generic function.
  */
-void EDBM_update_generic(BMEditMesh *em, const bool do_tessface, const bool is_destructive)
+void EDBM_update_generic(BMEditMesh *em, const bool do_tessellation, const bool is_destructive)
 {
   Object *ob = em->ob;
   /* order of calling isn't important */
   DEG_id_tag_update(ob->data, ID_RECALC_GEOMETRY);
   WM_main_add_notifier(NC_GEOM | ND_DATA, ob->data);
 
-  if (do_tessface) {
-    BKE_editmesh_tessface_calc(em);
+  if (do_tessellation) {
+    BKE_editmesh_looptri_calc(em);
   }
 
   if (is_destructive) {
diff --git a/source/blender/editors/mesh/mesh_data.c b/source/blender/editors/mesh/mesh_data.c
index ee8de9d8ea9..569994bead1 100644
--- a/source/blender/editors/mesh/mesh_data.c
+++ b/source/blender/editors/mesh/mesh_data.c
@@ -289,18 +289,14 @@ int ED_mesh_uv_texture_add(Mesh *me, const char *name, const bool active_set, co
     if (me->mloopuv && do_init) {
       CustomData_add_layer_named(
           &me->ldata, CD_MLOOPUV, CD_DUPLICATE, me->mloopuv, me->totloop, name);
-      CustomData_add_layer_named(
-          &me->fdata, CD_MTFACE, CD_DUPLICATE, me->mtface, me->totface, name);
       is_init = true;
     }
     else {
       CustomData_add_layer_named(&me->ldata, CD_MLOOPUV, CD_DEFAULT, NULL, me->totloop, name);
-      CustomData_add_layer_named(&me->fdata, CD_MTFACE, CD_DEFAULT, NULL, me->totface, name);
     }
 
     if (active_set || layernum_dst == 0) {
       CustomData_set_layer_active(&me->ldata, CD_MLOOPUV, layernum_dst);
-      CustomData_set_layer_active(&me->fdata, CD_MTFACE, layernum_dst);
     }
 
     BKE_mesh_update_customdata_pointers(me, true);
@@ -418,16 +414,13 @@ int ED_mesh_color_add(Mesh *me, const char *name, const bool active_set, const b
     if (me->mloopcol && do_init) {
       CustomData_add_layer_named(
           &me->ldata, CD_MLOOPCOL, CD_DUPLICATE, me->mloopcol, me->totloop, name);
-      CustomData_add_layer_named(&me->fdata, CD_MCOL, CD_DUPLICATE, me->mcol, me->totface, name);
     }
     else {
       CustomData_add_layer_named(&me->ldata, CD_MLOOPCOL, CD_DEFAULT, NULL, me->totloop, name);
-      CustomData_add_layer_named(&me->fdata, CD_MCOL, CD_DEFAULT, NULL, me->totface, name);
     }
 
     if (active_set || layernum == 0) {
       CustomData_set_layer_active(&me->ldata, CD_MLOOPCOL, layernum);
-      CustomData_set_layer_active(&me->fdata, CD_MCOL, layernum);
     }
 
     BKE_mesh_update_customdata_pointers(me, true);
@@ -885,18 +878,8 @@ void MESH_OT_customdata_custom_splitnormals_clear(wmOperatorType *ot)
 
 /************************** Add Geometry Layers *************************/
 
-void ED_mesh_update(
-    Mesh *mesh, bContext *C, bool calc_edges, bool calc_edges_loose, bool calc_tessface)
+void ED_mesh_update(Mesh *mesh, bContext *C, bool calc_edges, bool calc_edges_loose)
 {
-  bool tessface_input = false;
-
-  if (mesh->totface > 0 && mesh->totpoly == 0) {
-    BKE_mesh_convert_mfaces_to_mpolys(mesh);
-
-    /* would only be converting back again, don't bother */
-    tessface_input = true;
-  }
-
   if (calc_edges_loose && mesh->totedge) {
     BKE_mesh_calc_edges_loose(mesh);
   }
@@ -905,15 +888,8 @@ void ED_mesh_update(
     BKE_mesh_calc_edges(mesh, calc_edges, true);
   }
 
-  if (calc_tessface) {
-    if (tessface_input == false) {
-      BKE_mesh_tessface_calc(mesh);
-    }
-  }
-  else {
-    /* default state is not to have tessface's so make sure this is the case */
-    BKE_mesh_tessface_clear(mesh);
-  }
+  /* Default state is not to have tessface's so make sure this is the case. */
+  BKE_mesh_tessface_clear(mesh);
 
   BKE_mesh_calc_normals(mesh);
 
@@ -988,39 +964,6 @@ static void mesh_add_edges(Mesh *mesh, int len)
   mesh->totedge = totedge;
 }
 
-static void mesh_add_tessfaces(Mesh *mesh, int len)
-{
-  CustomData fdata;
-  MFace *mface;
-  int i, totface;
-
-  if (len == 0) {
-    return;
-  }
-
-  totface = mesh->totface + len; /* new face count */
-
-  /* update customdata */
-  CustomData_copy(&mesh->fdata, &fdata, CD_MASK_MESH.fmask, CD_DEFAULT, totface);
-  CustomData_copy_data(&mesh->fdata, &fdata, 0, 0, mesh->totface);
-
-  if (!CustomData_has_layer(&fdata, CD_MFACE)) {
-    CustomData_add_layer(&fdata, CD_MFACE, CD_CALLOC, NULL, totface);
-  }
-
-  CustomData_free(&mesh->fdata, mesh->totface);
-  mesh->fdata = fdata;
-  BKE_mesh_update_customdata_pointers(mesh, true);
-
-  /* set default flags */
-  mface = &mesh->mface[mesh->totface];
-  for (i = 0; i < len; i++, mface++) {
-    mface->flag = ME_FACE_SEL;
-  }
-
-  mesh->totface = totface;
-}
-
 static void mesh_add_loops(Mesh *mesh, int len)
 {
   CustomData ldata;
@@ -1109,20 +1052,6 @@ static void mesh_remove_edges(Mesh *mesh, int len)
   mesh->totedge = totedge;
 }
 
-static void mesh_remove_faces(Mesh *mesh, int len)
-{
-  int totface;
-
-  if (len == 0) {
-    return;
-  }
-
-  totface = mesh->totface - len; /* new face count */
-  CustomData_free_elem(&mesh->fdata, totface, len);
-
-  mesh->totface = totface;
-}
-
 #if 0
 void ED_mesh_geometry_add(Mesh *mesh, ReportList *reports, int verts, int edges, int faces)
 {
@@ -1143,21 +1072,6 @@ void ED_mesh_geometry_add(Mesh *mesh, ReportList *reports, int verts, int edges,
 }
 #endif
 
-void ED_mesh_tessfaces_add(Mesh *mesh, ReportList *reports, int count)
-{
-  if (mesh->edit_mesh) {
-    BKE_report(reports, RPT_ERROR, "Cannot add tessfaces in edit mode");
-    return;
-  }
-
-  if (mesh->mpoly) {
-    BKE_report(reports, RPT_ERROR, "Cannot add tessfaces to a mesh that already has polygons");
-    return;
-  }
-
-  mesh_add_tessfaces(mesh, count);
-}
-
 void ED_mesh_edges_add(Mesh *mesh, ReportList *reports, int count)
 {
   if (mesh->edit_mesh) {
@@ -1178,20 +1092,6 @@ void ED_mesh_vertices_add(Mesh *mesh, ReportList *reports, int count)
   mesh_add_verts(mesh, count);
 }
 
-void ED_mesh_faces_remove(Mesh *mesh, ReportList *reports, int count)
-{
-  if (mesh->edit_mesh) {
-    BKE_report(reports, RPT_ERROR, "Cannot remove faces in edit mode");
-    return;
-  }
-  else if (count > mesh->totface) {
-    BKE_report(reports, RPT_ERROR, "Cannot remove more faces than the mesh contains");
-    return;
-  }
-
-  mesh_remove_faces(mesh, count);
-}
-
 void ED_mesh_edges_remove(Mesh *mesh, ReportList *reports, int count)
 {
   if (mesh->edit_mesh) {
@@ -1240,26 +1140,6 @@ void ED_mesh_polys_add(Mesh *mesh, ReportList *reports, int count)
   mesh_add_polys(mesh, count);
 }
 
-void ED_mesh_calc_tessface(Mesh *mesh, bool free_mpoly)
-{
-  if (mesh->edit_mesh) {
-    BKE_editmesh_tessface_calc(mesh->edit_mesh);
-  }
-  else {
-    BKE_mesh_tessface_calc(mesh);
-  }
-  if (free_mpoly) {
-    CustomData_free(&mesh->ldata, mesh->totloop);
-    CustomData_free(&mesh->pdata, mesh->totpoly);
-    mesh->totloop = 0;
-    mesh->totpoly = 0;
-    mesh->mloop = NULL;
-    mesh->mloopcol = NULL;
-    mesh->mloopuv = NULL;
-    mesh->mpoly = NULL;
-  }
-}
-
 void ED_mesh_report_mirror_ex(wmOperator *op, int totmirr, int totfail, char selectmode)
 {
   const char *elem_type;
diff --git a/source/blender/editors/mesh/mesh_intern.h b/source/blender/editors/mesh/mesh_intern.h
index 8d585977105..3558a07c6fb 100644
--- a/source/blender/editors/mesh/mesh_intern.h
+++ b/source/blender/editors/mesh/mesh_intern.h
@@ -122,6 +122,8 @@ void MESH_GGT_spin_redo(struct wmGizmoGroupType *gzgt);
 void MESH_OT_polybuild_face_at_cursor(struct wmOperatorType *ot);
 void MESH_OT_polybuild_split_at_cursor(struct wmOperatorType *ot);
 void MESH_OT_polybuild_dissolve_at_cursor(struct wmOperatorType *ot);
+void MESH_OT_polybuild_transform_at_cursor(struct wmOperatorType *ot);
+void MESH_OT_polybuild_delete_at_cursor(struct wmOperatorType *ot);
 
 /* *** editmesh_inset.c *** */
 void MESH_OT_inset(struct wmOperatorType *ot);
diff --git a/source/blender/editors/mesh/mesh_ops.c b/source/blender/editors/mesh/mesh_ops.c
index c3387dcfc09..9b6e991a9f5 100644
--- a/source/blender/editors/mesh/mesh_ops.c
+++ b/source/blender/editors/mesh/mesh_ops.c
@@ -145,6 +145,8 @@ void ED_operatortypes_mesh(void)
   WM_operatortype_append(MESH_OT_polybuild_face_at_cursor);
   WM_operatortype_append(MESH_OT_polybuild_split_at_cursor);
   WM_operatortype_append(MESH_OT_polybuild_dissolve_at_cursor);
+  WM_operatortype_append(MESH_OT_polybuild_transform_at_cursor);
+  WM_operatortype_append(MESH_OT_polybuild_delete_at_cursor);
 
   WM_operatortype_append(MESH_OT_uv_texture_add);
   WM_operatortype_append(MESH_OT_uv_texture_remove);
@@ -331,6 +333,25 @@ void ED_operatormacros_mesh(void)
   otmacro = WM_operatortype_macro_define(ot, "TRANSFORM_OT_translate");
   RNA_boolean_set(otmacro->ptr, "use_proportional_edit", false);
   RNA_boolean_set(otmacro->ptr, "mirror", false);
+
+  ot = WM_operatortype_append_macro("MESH_OT_polybuild_transform_at_cursor_move",
+                                    "Transform at Cursor Move",
+                                    "",
+                                    OPTYPE_UNDO | OPTYPE_REGISTER);
+  WM_operatortype_macro_define(ot, "MESH_OT_polybuild_transform_at_cursor");
+  otmacro = WM_operatortype_macro_define(ot, "TRANSFORM_OT_translate");
+  RNA_boolean_set(otmacro->ptr, "use_proportional_edit", false);
+  RNA_boolean_set(otmacro->ptr, "mirror", false);
+
+  ot = WM_operatortype_append_macro("MESH_OT_polybuild_extrude_at_cursor_move",
+                                    "Extrude at Cursor Move",
+                                    "",
+                                    OPTYPE_UNDO | OPTYPE_REGISTER);
+  WM_operatortype_macro_define(ot, "MESH_OT_polybuild_transform_at_cursor");
+  otmacro = WM_operatortype_macro_define(ot, "MESH_OT_extrude_edges_indiv");
+  otmacro = WM_operatortype_macro_define(ot, "TRANSFORM_OT_translate");
+  RNA_boolean_set(otmacro->ptr, "use_proportional_edit", false);
+  RNA_boolean_set(otmacro->ptr, "mirror", false);
 }
 
 /* note mesh keymap also for other space? */
diff --git a/source/blender/editors/object/CMakeLists.txt b/source/blender/editors/object/CMakeLists.txt
index 2490f88b5eb..aabfa78cf58 100644
--- a/source/blender/editors/object/CMakeLists.txt
+++ b/source/blender/editors/object/CMakeLists.txt
@@ -48,6 +48,7 @@ set(SRC
   object_collection.c
   object_constraint.c
   object_data_transfer.c
+  object_data_transform.c
   object_edit.c
   object_facemap_ops.c
   object_gpencil_modifier.c
diff --git a/source/blender/editors/object/object_add.c b/source/blender/editors/object/object_add.c
index db1aa6ca37f..3d5ec3d4ed5 100644
--- a/source/blender/editors/object/object_add.c
+++ b/source/blender/editors/object/object_add.c
@@ -103,6 +103,7 @@
 #include "ED_mesh.h"
 #include "ED_node.h"
 #include "ED_object.h"
+#include "ED_outliner.h"
 #include "ED_physics.h"
 #include "ED_render.h"
 #include "ED_screen.h"
@@ -502,6 +503,8 @@ Object *ED_object_add_type(bContext *C,
   /* TODO(sergey): Use proper flag for tagging here. */
   DEG_id_tag_update(&scene->id, 0);
 
+  ED_outliner_select_sync_from_object_tag(C);
+
   return ob;
 }
 
@@ -2383,13 +2386,22 @@ static int convert_exec(bContext *C, wmOperator *op)
         BKE_object_free_curve_cache(newob);
       }
       else if (target == OB_GPENCIL) {
-        /* Create a new grease pencil object */
-        if (gpencil_ob == NULL) {
-          const float *cur = scene->cursor.location;
-          ushort local_view_bits = (v3d && v3d->localvd) ? v3d->local_view_uuid : 0;
-          gpencil_ob = ED_gpencil_add_object(C, scene, cur, local_view_bits);
+        if (ob->type != OB_CURVE) {
+          BKE_report(
+              op->reports, RPT_ERROR, "Convert Surfaces to Grease Pencil is not supported.");
+        }
+        else {
+          /* Create a new grease pencil object only if it was not created before.
+           * All curves selected are converted as strokes of the same grease pencil object.
+           * Nurbs Surface are not supported.
+           */
+          if (gpencil_ob == NULL) {
+            const float *cur = scene->cursor.location;
+            ushort local_view_bits = (v3d && v3d->localvd) ? v3d->local_view_uuid : 0;
+            gpencil_ob = ED_gpencil_add_object(C, scene, cur, local_view_bits);
+          }
+          BKE_gpencil_convert_curve(bmain, scene, gpencil_ob, ob, false, false, true);
         }
-        BKE_gpencil_convert_curve(bmain, scene, gpencil_ob, ob, gpencil_lines, use_collections);
       }
     }
     else if (ob->type == OB_MBALL && target == OB_MESH) {
diff --git a/source/blender/editors/object/object_constraint.c b/source/blender/editors/object/object_constraint.c
index 7343dba420f..8981221cb9c 100644
--- a/source/blender/editors/object/object_constraint.c
+++ b/source/blender/editors/object/object_constraint.c
@@ -493,8 +493,12 @@ static void test_constraint(
                CONSTRAINT_TYPE_CLAMPTO,
                CONSTRAINT_TYPE_SPLINEIK)) {
         if (ct->tar) {
+          /* The object type check is only needed here in case we have a placeholder
+           * object assigned (because the library containing the curve is missing).
+           *
+           * In other cases it should be impossible to have a type mismatch.
+           */
           if (ct->tar->type != OB_CURVE) {
-            ct->tar = NULL;
             con->flag |= CONSTRAINT_DISABLE;
           }
           else {
@@ -507,8 +511,12 @@ static void test_constraint(
       }
       else if (con->type == CONSTRAINT_TYPE_ARMATURE) {
         if (ct->tar) {
+          /* The object type check is only needed here in case we have a placeholder
+           * object assigned (because the library containing the armature is missing).
+           *
+           * In other cases it should be impossible to have a type mismatch.
+           */
           if (ct->tar->type != OB_ARMATURE) {
-            ct->tar = NULL;
             con->flag |= CONSTRAINT_DISABLE;
           }
           else if (!BKE_armature_find_bone_name(BKE_armature_from_object(ct->tar),
@@ -636,7 +644,7 @@ static const EnumPropertyItem constraint_owner_items[] = {
 static bool edit_constraint_poll_generic(bContext *C, StructRNA *rna_type)
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "constraint", rna_type);
-  Object *ob = (ptr.id.data) ? ptr.id.data : ED_object_active_context(C);
+  Object *ob = (ptr.owner_id) ? (Object *)ptr.owner_id : ED_object_active_context(C);
 
   if (!ptr.data) {
     CTX_wm_operator_poll_msg_set(C, "Context missing 'constraint'");
@@ -648,7 +656,7 @@ static bool edit_constraint_poll_generic(bContext *C, StructRNA *rna_type)
     return 0;
   }
 
-  if (ID_IS_LINKED(ob) || (ptr.id.data && ID_IS_LINKED(ptr.id.data))) {
+  if (ID_IS_LINKED(ob) || (ptr.owner_id && ID_IS_LINKED(ptr.owner_id))) {
     CTX_wm_operator_poll_msg_set(C, "Cannot edit library data");
     return 0;
   }
@@ -680,7 +688,7 @@ static void edit_constraint_properties(wmOperatorType *ot)
 static int edit_constraint_invoke_properties(bContext *C, wmOperator *op)
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "constraint", &RNA_Constraint);
-  Object *ob = (ptr.id.data) ? ptr.id.data : ED_object_active_context(C);
+  Object *ob = (ptr.owner_id) ? (Object *)ptr.owner_id : ED_object_active_context(C);
   bConstraint *con;
   ListBase *list;
 
@@ -1422,14 +1430,14 @@ void ED_object_constraint_dependency_tag_update(Main *bmain, Object *ob, bConstr
 static bool constraint_poll(bContext *C)
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "constraint", &RNA_Constraint);
-  return (ptr.id.data && ptr.data);
+  return (ptr.owner_id && ptr.data);
 }
 
 static int constraint_delete_exec(bContext *C, wmOperator *UNUSED(op))
 {
   Main *bmain = CTX_data_main(C);
   PointerRNA ptr = CTX_data_pointer_get_type(C, "constraint", &RNA_Constraint);
-  Object *ob = ptr.id.data;
+  Object *ob = (Object *)ptr.owner_id;
   bConstraint *con = ptr.data;
   ListBase *lb = get_constraint_lb(ob, con, NULL);
 
diff --git a/source/blender/editors/object/object_data_transform.c b/source/blender/editors/object/object_data_transform.c
new file mode 100644
index 00000000000..ee86c79ead5
--- /dev/null
+++ b/source/blender/editors/object/object_data_transform.c
@@ -0,0 +1,333 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * The Original Code is Copyright (C) 2001-2002 by NaN Holding BV.
+ * All rights reserved.
+ */
+
+/** \file
+ * \ingroup edobj
+ *
+ * Use to transform object origins only.
+ *
+ * This is a small API to store & apply transformations to object data,
+ * where a transformation matrix can be continually applied ontop of the original values
+ * so we don't loose precision over time.
+ */
+
+#include <stdlib.h>
+#include <string.h>
+
+#include "DNA_anim_types.h"
+#include "DNA_armature_types.h"
+#include "DNA_mesh_types.h"
+#include "DNA_meta_types.h"
+#include "DNA_object_types.h"
+#include "DNA_scene_types.h"
+#include "DNA_collection_types.h"
+#include "DNA_lattice_types.h"
+
+#include "BLI_math.h"
+#include "BLI_listbase.h"
+#include "BLI_utildefines.h"
+
+#include "BKE_curve.h"
+#include "BKE_mball.h"
+#include "BKE_mesh.h"
+#include "BKE_armature.h"
+#include "BKE_lattice.h"
+
+#include "DEG_depsgraph.h"
+
+#include "WM_types.h"
+
+#include "ED_object.h"
+
+#include "MEM_guardedalloc.h"
+
+/* -------------------------------------------------------------------- */
+/** \name Internal Transform Get/Apply
+ *
+ * Some object data types don't have utility functions to access their transformation data.
+ * Define these locally.
+ *
+ * \{ */
+
+/* Armature */
+
+struct ElemData_Armature {
+  float tail[3];
+  float head[3];
+  float roll;
+  float arm_tail[3];
+  float arm_head[3];
+  float arm_roll;
+  float rad_tail;
+  float rad_head;
+  float dist;
+  float xwidth;
+  float zwidth;
+};
+
+static struct ElemData_Armature *armature_coords_and_quats_get_recurse(
+    const ListBase *bone_base, struct ElemData_Armature *elem_array)
+{
+  struct ElemData_Armature *elem = elem_array;
+  for (const Bone *bone = bone_base->first; bone; bone = bone->next) {
+
+#define COPY_PTR(member) memcpy(elem->member, bone->member, sizeof(bone->member))
+#define COPY_VAL(member) memcpy(&elem->member, &bone->member, sizeof(bone->member))
+    COPY_PTR(head);
+    COPY_PTR(tail);
+    COPY_VAL(roll);
+    COPY_PTR(arm_head);
+    COPY_PTR(arm_tail);
+    COPY_VAL(arm_roll);
+    COPY_VAL(rad_tail);
+    COPY_VAL(rad_head);
+    COPY_VAL(dist);
+    COPY_VAL(xwidth);
+    COPY_VAL(zwidth);
+#undef COPY_PTR
+#undef COPY_VAL
+
+    elem = armature_coords_and_quats_get_recurse(&bone->childbase, elem + 1);
+  }
+  return elem;
+}
+
+static void armature_coords_and_quats_get(const bArmature *arm,
+                                          struct ElemData_Armature *elem_array)
+{
+  armature_coords_and_quats_get_recurse(&arm->bonebase, elem_array);
+}
+
+static const struct ElemData_Armature *armature_coords_and_quats_apply_with_mat4_recurse(
+    ListBase *bone_base, const struct ElemData_Armature *elem_array, const float mat[4][4])
+{
+  const struct ElemData_Armature *elem = elem_array;
+  for (Bone *bone = bone_base->first; bone; bone = bone->next) {
+
+#define COPY_PTR(member) memcpy(bone->member, elem->member, sizeof(bone->member))
+#define COPY_VAL(member) memcpy(&bone->member, &elem->member, sizeof(bone->member))
+    COPY_PTR(head);
+    COPY_PTR(tail);
+    COPY_VAL(roll);
+    COPY_PTR(arm_head);
+    COPY_PTR(arm_tail);
+    COPY_VAL(arm_roll);
+    COPY_VAL(rad_tail);
+    COPY_VAL(rad_head);
+    COPY_VAL(dist);
+    COPY_VAL(xwidth);
+    COPY_VAL(zwidth);
+#undef COPY_PTR
+#undef COPY_VAL
+
+    elem = armature_coords_and_quats_apply_with_mat4_recurse(&bone->childbase, elem + 1, mat);
+  }
+  return elem;
+}
+
+static void armature_coords_and_quats_apply_with_mat4(bArmature *arm,
+                                                      const struct ElemData_Armature *elem_array,
+                                                      const float mat[4][4])
+{
+  armature_coords_and_quats_apply_with_mat4_recurse(&arm->bonebase, elem_array, mat);
+  BKE_armature_transform(arm, mat, true);
+}
+
+/* MetaBall */
+
+struct ElemData_MetaBall {
+  float co[3];
+  float quat[4];
+  float exp[3];
+  float rad;
+};
+
+static void metaball_coords_and_quats_get(const MetaBall *mb, struct ElemData_MetaBall *elem_array)
+{
+  struct ElemData_MetaBall *elem = elem_array;
+  for (const MetaElem *ml = mb->elems.first; ml; ml = ml->next, elem++) {
+    copy_v3_v3(elem->co, &ml->x);
+    copy_qt_qt(elem->quat, ml->quat);
+    copy_v3_v3(elem->exp, &ml->expx);
+    elem->rad = ml->rad;
+  }
+}
+
+static void metaball_coords_and_quats_apply_with_mat4(MetaBall *mb,
+                                                      const struct ElemData_MetaBall *elem_array,
+                                                      const float mat[4][4])
+{
+  const struct ElemData_MetaBall *elem = elem_array;
+  for (MetaElem *ml = mb->elems.first; ml; ml = ml->next, elem++) {
+    copy_v3_v3(&ml->x, elem->co);
+    copy_qt_qt(ml->quat, elem->quat);
+    copy_v3_v3(&ml->expx, elem->exp);
+    ml->rad = elem->rad;
+  }
+  BKE_mball_transform(mb, mat, true);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Public Object Data Storage API
+ *
+ * Used for interactively transforming object data.
+ *
+ * Store object data transformation in an opaque struct.
+ * \{ */
+
+struct XFormObjectData {
+  ID *id;
+};
+
+struct XFormObjectData_Mesh {
+  struct XFormObjectData base;
+  float elem_array[0][3];
+};
+
+struct XFormObjectData_Lattice {
+  struct XFormObjectData base;
+  float elem_array[0][3];
+};
+
+struct XFormObjectData_Curve {
+  struct XFormObjectData base;
+  float elem_array[0][3];
+};
+
+struct XFormObjectData_Armature {
+  struct XFormObjectData base;
+  struct ElemData_Armature elem_array[0];
+};
+
+struct XFormObjectData_MetaBall {
+  struct XFormObjectData base;
+  struct ElemData_MetaBall elem_array[0];
+};
+
+struct XFormObjectData *ED_object_data_xform_create(ID *id)
+{
+  struct XFormObjectData *xod_base = NULL;
+  switch (GS(id->name)) {
+    case ID_ME: {
+      Mesh *me = (Mesh *)id;
+      const int elem_array_len = me->totvert;
+      struct XFormObjectData_Mesh *xod = MEM_mallocN(
+          sizeof(*xod) + (sizeof(*xod->elem_array) * elem_array_len), __func__);
+      BKE_mesh_vert_coords_get(me, xod->elem_array);
+      xod_base = &xod->base;
+      break;
+    }
+    case ID_LT: {
+      Lattice *lt = (Lattice *)id;
+      const int elem_array_len = lt->pntsu * lt->pntsv * lt->pntsw;
+      struct XFormObjectData_Lattice *xod = MEM_mallocN(
+          sizeof(*xod) + (sizeof(*xod->elem_array) * elem_array_len), __func__);
+      BKE_lattice_vert_coords_get(lt, xod->elem_array);
+      xod_base = &xod->base;
+      break;
+    }
+    case ID_CU: {
+      Curve *cu = (Curve *)id;
+      const short ob_type = BKE_curve_type_get(cu);
+      if (ob_type == OB_FONT) {
+        /* We could support translation. */
+        break;
+      }
+      const int elem_array_len = BKE_nurbList_verts_count(&cu->nurb);
+      struct XFormObjectData_Curve *xod = MEM_mallocN(
+          sizeof(*xod) + (sizeof(*xod->elem_array) * elem_array_len), __func__);
+      BKE_curve_nurbs_vert_coords_get(&cu->nurb, xod->elem_array, elem_array_len);
+      xod_base = &xod->base;
+      break;
+    }
+    case ID_AR: {
+      bArmature *arm = (bArmature *)id;
+      const int elem_array_len = BKE_armature_bonelist_count(&arm->bonebase);
+      struct XFormObjectData_Armature *xod = MEM_mallocN(
+          sizeof(*xod) + (sizeof(*xod->elem_array) * elem_array_len), __func__);
+      armature_coords_and_quats_get(arm, xod->elem_array);
+      xod_base = &xod->base;
+      break;
+    }
+    case ID_MB: {
+      MetaBall *mb = (MetaBall *)id;
+      const int elem_array_len = BLI_listbase_count(&mb->elems);
+      struct XFormObjectData_MetaBall *xod = MEM_mallocN(
+          sizeof(*xod) + (sizeof(*xod->elem_array) * elem_array_len), __func__);
+      metaball_coords_and_quats_get(mb, xod->elem_array);
+      xod_base = &xod->base;
+      break;
+    }
+    default: {
+      break;
+    }
+  }
+  if (xod_base) {
+    xod_base->id = id;
+  }
+  return xod_base;
+}
+
+void ED_object_data_xform_destroy(struct XFormObjectData *xod)
+{
+  MEM_freeN(xod);
+}
+
+void ED_object_data_xform_by_mat4(struct XFormObjectData *xod_base, const float mat[4][4])
+{
+  switch (GS(xod_base->id->name)) {
+    case ID_ME: {
+      Mesh *me = (Mesh *)xod_base->id;
+      struct XFormObjectData_Mesh *xod = (struct XFormObjectData_Mesh *)xod_base;
+      BKE_mesh_vert_coords_apply_with_mat4(me, xod->elem_array, mat);
+      break;
+    }
+    case ID_LT: {
+      Lattice *lt = (Lattice *)xod_base->id;
+      struct XFormObjectData_Lattice *xod = (struct XFormObjectData_Lattice *)xod_base;
+      BKE_lattice_vert_coords_apply_with_mat4(lt, xod->elem_array, mat);
+      break;
+    }
+    case ID_CU: {
+      Curve *cu = (Curve *)xod_base->id;
+      struct XFormObjectData_Curve *xod = (struct XFormObjectData_Curve *)xod_base;
+      BKE_curve_nurbs_vert_coords_apply_with_mat4(&cu->nurb, xod->elem_array, mat, true);
+      break;
+    }
+    case ID_AR: {
+      bArmature *arm = (bArmature *)xod_base->id;
+      struct XFormObjectData_Armature *xod = (struct XFormObjectData_Armature *)xod_base;
+      armature_coords_and_quats_apply_with_mat4(arm, xod->elem_array, mat);
+      break;
+    }
+    case ID_MB: {
+      MetaBall *mb = (MetaBall *)xod_base->id;
+      struct XFormObjectData_MetaBall *xod = (struct XFormObjectData_MetaBall *)xod_base;
+      metaball_coords_and_quats_apply_with_mat4(mb, xod->elem_array, mat);
+      break;
+    }
+    default: {
+      break;
+    }
+  }
+}
+
+/** \} */
diff --git a/source/blender/editors/object/object_edit.c b/source/blender/editors/object/object_edit.c
index ed40a4eb948..33c7ffefb8b 100644
--- a/source/blender/editors/object/object_edit.c
+++ b/source/blender/editors/object/object_edit.c
@@ -116,6 +116,7 @@
 typedef struct MoveToCollectionData MoveToCollectionData;
 static void move_to_collection_menus_items(struct uiLayout *layout,
                                            struct MoveToCollectionData *menu);
+static ListBase selected_objects_get(bContext *C);
 
 /* ************* XXX **************** */
 static void error(const char *UNUSED(arg))
@@ -593,7 +594,7 @@ bool ED_object_editmode_enter_ex(Main *bmain, Scene *scene, Object *ob, int flag
     if (LIKELY(em)) {
       /* order doesn't matter */
       EDBM_mesh_normals_update(em);
-      BKE_editmesh_tessface_calc(em);
+      BKE_editmesh_looptri_calc(em);
     }
 
     WM_main_add_notifier(NC_SCENE | ND_MODE | NS_EDITMODE_MESH, NULL);
@@ -1199,7 +1200,7 @@ static int shade_smooth_exec(bContext *C, wmOperator *op)
     }
 
     if (ob->type == OB_MESH) {
-      BKE_mesh_smooth_flag_set(ob, !clear);
+      BKE_mesh_smooth_flag_set(ob->data, !clear);
 
       BKE_mesh_batch_cache_dirty_tag(ob->data, BKE_MESH_BATCH_DIRTY_ALL);
       DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
@@ -1460,6 +1461,23 @@ void OBJECT_OT_mode_set_or_submode(wmOperatorType *ot)
   RNA_def_property_flag(prop, PROP_SKIP_SAVE);
 }
 
+static ListBase selected_objects_get(bContext *C)
+{
+  ListBase objects = {NULL};
+
+  if (CTX_wm_space_outliner(C) != NULL) {
+    ED_outliner_selected_objects_get(C, &objects);
+  }
+  else {
+    CTX_DATA_BEGIN (C, Object *, ob, selected_objects) {
+      BLI_addtail(&objects, BLI_genericNodeN(ob));
+    }
+    CTX_DATA_END;
+  }
+
+  return objects;
+}
+
 static bool move_to_collection_poll(bContext *C)
 {
   if (CTX_wm_space_outliner(C) != NULL) {
@@ -1472,7 +1490,7 @@ static bool move_to_collection_poll(bContext *C)
       return false;
     }
 
-    return ED_operator_object_active_editable(C);
+    return ED_operator_objectmode(C);
   }
 }
 
@@ -1498,15 +1516,7 @@ static int move_to_collection_exec(bContext *C, wmOperator *op)
     return OPERATOR_CANCELLED;
   }
 
-  if (CTX_wm_space_outliner(C) != NULL) {
-    ED_outliner_selected_objects_get(C, &objects);
-  }
-  else {
-    CTX_DATA_BEGIN (C, Object *, ob, selected_objects) {
-      BLI_addtail(&objects, BLI_genericNodeN(ob));
-    }
-    CTX_DATA_END;
-  }
+  objects = selected_objects_get(C);
 
   if (is_new) {
     char new_collection_name[MAX_NAME];
@@ -1650,6 +1660,13 @@ static int move_to_collection_invoke(bContext *C, wmOperator *op, const wmEvent
 {
   Scene *scene = CTX_data_scene(C);
 
+  ListBase objects = selected_objects_get(C);
+  if (BLI_listbase_is_empty(&objects)) {
+    BKE_report(op->reports, RPT_ERROR, "No objects selected");
+    return OPERATOR_CANCELLED;
+  }
+  BLI_freelistN(&objects);
+
   /* Reset the menus data for the current master collection, and free previously allocated data. */
   move_to_collection_menus_free(&master_collection_menu);
 
@@ -1674,7 +1691,7 @@ static int move_to_collection_invoke(bContext *C, wmOperator *op, const wmEvent
     return move_to_collection_exec(C, op);
   }
 
-  Collection *master_collection = BKE_collection_master(scene);
+  Collection *master_collection = scene->master_collection;
 
   /* We need the data to be allocated so it's available during menu drawing.
    * Technically we could use wmOperator->customdata. However there is no free callback
diff --git a/source/blender/editors/object/object_gpencil_modifier.c b/source/blender/editors/object/object_gpencil_modifier.c
index 075cd5acad6..9138e65dd2f 100644
--- a/source/blender/editors/object/object_gpencil_modifier.c
+++ b/source/blender/editors/object/object_gpencil_modifier.c
@@ -390,7 +390,7 @@ void OBJECT_OT_gpencil_modifier_add(wmOperatorType *ot)
 static int gpencil_edit_modifier_poll_generic(bContext *C, StructRNA *rna_type, int obtype_flag)
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "modifier", rna_type);
-  Object *ob = (ptr.id.data) ? ptr.id.data : ED_object_active_context(C);
+  Object *ob = (ptr.owner_id) ? (Object *)ptr.owner_id : ED_object_active_context(C);
 
   if (!ob || ID_IS_LINKED(ob)) {
     return 0;
@@ -398,7 +398,7 @@ static int gpencil_edit_modifier_poll_generic(bContext *C, StructRNA *rna_type,
   if (obtype_flag && ((1 << ob->type) & obtype_flag) == 0) {
     return 0;
   }
-  if (ptr.id.data && ID_IS_LINKED(ptr.id.data)) {
+  if (ptr.owner_id && ID_IS_LINKED(ptr.owner_id)) {
     return 0;
   }
 
diff --git a/source/blender/editors/object/object_hook.c b/source/blender/editors/object/object_hook.c
index 15c5ea40991..bcbf7ddf9a6 100644
--- a/source/blender/editors/object/object_hook.c
+++ b/source/blender/editors/object/object_hook.c
@@ -352,7 +352,7 @@ static bool object_hook_index_array(Main *bmain,
       em = me->edit_mesh;
 
       EDBM_mesh_normals_update(em);
-      BKE_editmesh_tessface_calc(em);
+      BKE_editmesh_looptri_calc(em);
 
       /* check selected vertices first */
       if (return_editmesh_indexar(em, r_tot, r_indexar, r_cent) == 0) {
@@ -436,7 +436,7 @@ static void object_hook_from_context(
   HookModifierData *hmd;
 
   if (ptr->data) { /* if modifier context is available, use that */
-    ob = ptr->id.data;
+    ob = (Object *)ptr->owner_id;
     hmd = ptr->data;
   }
   else { /* use the provided property */
diff --git a/source/blender/editors/object/object_modifier.c b/source/blender/editors/object/object_modifier.c
index 8818fd71190..88d01936882 100644
--- a/source/blender/editors/object/object_modifier.c
+++ b/source/blender/editors/object/object_modifier.c
@@ -52,6 +52,7 @@
 #include "BKE_editmesh.h"
 #include "BKE_effect.h"
 #include "BKE_global.h"
+#include "BKE_gpencil_modifier.h"
 #include "BKE_key.h"
 #include "BKE_lattice.h"
 #include "BKE_library.h"
@@ -110,6 +111,9 @@ static void object_force_modifier_update_for_bind(Depsgraph *depsgraph, Object *
   else if (ELEM(ob->type, OB_CURVE, OB_SURF, OB_FONT)) {
     BKE_displist_make_curveTypes(depsgraph, scene_eval, ob_eval, false, false);
   }
+  else if (ob->type == OB_GPENCIL) {
+    BKE_gpencil_modifiers_calc(depsgraph, scene_eval, ob_eval);
+  }
 }
 
 static void object_force_modifier_bind_simple_options(Depsgraph *depsgraph,
@@ -714,9 +718,9 @@ static int modifier_apply_obdata(
                RPT_INFO,
                "Applied modifier only changed CV points, not tessellated/bevel vertices");
 
-    vertexCos = BKE_curve_nurbs_vertexCos_get(&curve_eval->nurb, &numVerts);
+    vertexCos = BKE_curve_nurbs_vert_coords_alloc(&curve_eval->nurb, &numVerts);
     mti->deformVerts(md_eval, &mectx, NULL, vertexCos, numVerts);
-    BK_curve_nurbs_vertexCos_apply(&curve->nurb, vertexCos);
+    BKE_curve_nurbs_vert_coords_apply(&curve->nurb, vertexCos, false);
 
     MEM_freeN(vertexCos);
 
@@ -918,7 +922,7 @@ bool edit_modifier_poll_generic(bContext *C,
                                 const bool is_editmode_allowed)
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "modifier", rna_type);
-  Object *ob = (ptr.id.data) ? ptr.id.data : ED_object_active_context(C);
+  Object *ob = (ptr.owner_id) ? (Object *)ptr.owner_id : ED_object_active_context(C);
 
   if (!ob || ID_IS_LINKED(ob)) {
     return 0;
@@ -926,7 +930,7 @@ bool edit_modifier_poll_generic(bContext *C,
   if (obtype_flag && ((1 << ob->type) & obtype_flag) == 0) {
     return 0;
   }
-  if (ptr.id.data && ID_IS_LINKED(ptr.id.data)) {
+  if (ptr.owner_id && ID_IS_LINKED(ptr.owner_id)) {
     return 0;
   }
 
diff --git a/source/blender/editors/object/object_relations.c b/source/blender/editors/object/object_relations.c
index a69f4872e72..06c360ed1cd 100644
--- a/source/blender/editors/object/object_relations.c
+++ b/source/blender/editors/object/object_relations.c
@@ -149,7 +149,7 @@ static int vertex_parent_set_exec(bContext *C, wmOperator *op)
     em = me->edit_mesh;
 
     EDBM_mesh_normals_update(em);
-    BKE_editmesh_tessface_calc(em);
+    BKE_editmesh_looptri_calc(em);
 
     /* Make sure the evaluated mesh is updated.
      *
@@ -1434,7 +1434,7 @@ static int make_links_scene_exec(bContext *C, wmOperator *op)
     return OPERATOR_CANCELLED;
   }
 
-  Collection *collection_to = BKE_collection_master(scene_to);
+  Collection *collection_to = scene_to->master_collection;
   CTX_DATA_BEGIN (C, Base *, base, selected_bases) {
     BKE_collection_object_add(bmain, collection_to, base->object);
   }
@@ -1744,7 +1744,7 @@ static Collection *single_object_users_collection(Main *bmain,
   }
 
   /* Since master collection has already be duplicated as part of scene copy,
-   * we do not duplictae it here.
+   * we do not duplicate it here.
    * However, this means its children need to be re-added manually here,
    * otherwise their parent lists are empty (which will lead to crashes, see T63101). */
   CollectionChild *child_next, *child = collection->children.first;
@@ -1771,7 +1771,7 @@ static void single_object_users(
     Main *bmain, Scene *scene, View3D *v3d, const int flag, const bool copy_collections)
 {
   /* duplicate all the objects of the scene (and matching collections, if required). */
-  Collection *master_collection = BKE_collection_master(scene);
+  Collection *master_collection = scene->master_collection;
   single_object_users_collection(bmain, scene, master_collection, flag, copy_collections, true);
 
   /* duplicate collections that consist entirely of duplicated objects */
@@ -2426,6 +2426,8 @@ static int make_override_library_exec(bContext *C, wmOperator *op)
 
   if (!ID_IS_LINKED(obact) && obact->instance_collection != NULL &&
       ID_IS_LINKED(obact->instance_collection)) {
+    BKE_main_id_tag_all(bmain, LIB_TAG_DOIT, false);
+
     Object *obcollection = obact;
     Collection *collection = obcollection->instance_collection;
 
@@ -2471,7 +2473,11 @@ static int make_override_library_exec(bContext *C, wmOperator *op)
           DEG_id_tag_update_ex(bmain, &new_ob->id, ID_RECALC_TRANSFORM | ID_RECALC_BASE_FLAGS);
         }
         /* parent to 'collection' empty */
-        if (new_ob->parent == NULL) {
+        /* Disabled for now, according to some artist this is probably not really useful anyway.
+         * And it breaks things like objects parented to bones
+         * (most likely due to missing proper setting of inverse parent matrix?)... */
+        /* Note: we might even actually want to get rid of that instanciating empty... */
+        if (0 && new_ob->parent == NULL) {
           new_ob->parent = obcollection;
         }
         if (new_ob == (Object *)obact->id.newid) {
@@ -2499,7 +2505,7 @@ static int make_override_library_exec(bContext *C, wmOperator *op)
 
     /* Cleanup. */
     BKE_main_id_clear_newpoins(bmain);
-    BKE_main_id_tag_listbase(&bmain->objects, LIB_TAG_DOIT, false);
+    BKE_main_id_tag_all(bmain, LIB_TAG_DOIT, false);
   }
   /* Else, poll func ensures us that ID_IS_LINKED(obact) is true. */
   else if (obact->type == OB_ARMATURE) {
@@ -2518,7 +2524,7 @@ static int make_override_library_exec(bContext *C, wmOperator *op)
 
     /* Cleanup. */
     BKE_main_id_clear_newpoins(bmain);
-    BKE_main_id_tag_listbase(&bmain->objects, LIB_TAG_DOIT, false);
+    BKE_main_id_tag_all(bmain, LIB_TAG_DOIT, false);
   }
   /* TODO: probably more cases where we want to do automated smart things in the future! */
   else {
@@ -2706,7 +2712,7 @@ static int object_unlink_data_exec(bContext *C, wmOperator *op)
     return OPERATOR_CANCELLED;
   }
 
-  id = pprop.ptr.id.data;
+  id = pprop.ptr.owner_id;
 
   if (GS(id->name) == ID_OB) {
     Object *ob = (Object *)id;
diff --git a/source/blender/editors/object/object_remesh.c b/source/blender/editors/object/object_remesh.c
index 1bcac9b4739..5f464084a9b 100644
--- a/source/blender/editors/object/object_remesh.c
+++ b/source/blender/editors/object/object_remesh.c
@@ -108,7 +108,7 @@ static int voxel_remesh_exec(bContext *C, wmOperator *op)
     return OPERATOR_CANCELLED;
   }
 
-  Mesh *obj_mesh_copy;
+  Mesh *obj_mesh_copy = NULL;
   if (mesh->flag & ME_REMESH_REPROJECT_PAINT_MASK) {
     obj_mesh_copy = BKE_mesh_new_nomain_from_template(mesh, mesh->totvert, 0, 0, 0, 0);
     CustomData_copy(
@@ -126,7 +126,7 @@ static int voxel_remesh_exec(bContext *C, wmOperator *op)
   }
 
   if (mesh->flag & ME_REMESH_SMOOTH_NORMALS) {
-    BKE_mesh_smooth_flag_set(ob, true);
+    BKE_mesh_smooth_flag_set(ob->data, true);
   }
 
   if (ob->mode == OB_MODE_SCULPT) {
diff --git a/source/blender/editors/object/object_select.c b/source/blender/editors/object/object_select.c
index da06707ebac..28242b986f1 100644
--- a/source/blender/editors/object/object_select.c
+++ b/source/blender/editors/object/object_select.c
@@ -69,6 +69,7 @@
 
 #include "ED_armature.h"
 #include "ED_object.h"
+#include "ED_outliner.h"
 #include "ED_screen.h"
 #include "ED_select_utils.h"
 #include "ED_keyframing.h"
@@ -436,6 +437,8 @@ static int object_select_by_type_exec(bContext *C, wmOperator *op)
   DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
   WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
 
+  ED_outliner_select_sync_from_object_tag(C);
+
   return OPERATOR_FINISHED;
 }
 
@@ -717,6 +720,7 @@ static int object_select_linked_exec(bContext *C, wmOperator *op)
   if (changed) {
     DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
     WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
+    ED_outliner_select_sync_from_object_tag(C);
     return OPERATOR_FINISHED;
   }
 
@@ -1100,6 +1104,7 @@ static int object_select_grouped_exec(bContext *C, wmOperator *op)
   if (changed) {
     DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
     WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
+    ED_outliner_select_sync_from_object_tag(C);
     return OPERATOR_FINISHED;
   }
 
@@ -1150,6 +1155,8 @@ static int object_select_all_exec(bContext *C, wmOperator *op)
     DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
     WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
 
+    ED_outliner_select_sync_from_object_tag(C);
+
     return OPERATOR_FINISHED;
   }
   else if (any_visible == false) {
@@ -1218,6 +1225,8 @@ static int object_select_same_collection_exec(bContext *C, wmOperator *op)
   DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
   WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
 
+  ED_outliner_select_sync_from_object_tag(C);
+
   return OPERATOR_FINISHED;
 }
 
@@ -1281,6 +1290,8 @@ static int object_select_mirror_exec(bContext *C, wmOperator *op)
   DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
   WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
 
+  ED_outliner_select_sync_from_object_tag(C);
+
   return OPERATOR_FINISHED;
 }
 
@@ -1369,6 +1380,9 @@ static int object_select_more_exec(bContext *C, wmOperator *UNUSED(op))
     Scene *scene = CTX_data_scene(C);
     DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
     WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
+
+    ED_outliner_select_sync_from_object_tag(C);
+
     return OPERATOR_FINISHED;
   }
   else {
@@ -1399,6 +1413,9 @@ static int object_select_less_exec(bContext *C, wmOperator *UNUSED(op))
     Scene *scene = CTX_data_scene(C);
     DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
     WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
+
+    ED_outliner_select_sync_from_object_tag(C);
+
     return OPERATOR_FINISHED;
   }
   else {
@@ -1448,6 +1465,8 @@ static int object_select_random_exec(bContext *C, wmOperator *op)
   DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
   WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
 
+  ED_outliner_select_sync_from_object_tag(C);
+
   return OPERATOR_FINISHED;
 }
 
diff --git a/source/blender/editors/object/object_shader_fx.c b/source/blender/editors/object/object_shader_fx.c
index 457d2421253..db0c8c54186 100644
--- a/source/blender/editors/object/object_shader_fx.c
+++ b/source/blender/editors/object/object_shader_fx.c
@@ -294,7 +294,7 @@ void OBJECT_OT_shaderfx_add(wmOperatorType *ot)
 static bool edit_shaderfx_poll_generic(bContext *C, StructRNA *rna_type, int obtype_flag)
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "shaderfx", rna_type);
-  Object *ob = (ptr.id.data) ? ptr.id.data : ED_object_active_context(C);
+  Object *ob = (ptr.owner_id) ? (Object *)ptr.owner_id : ED_object_active_context(C);
 
   if (!ptr.data) {
     CTX_wm_operator_poll_msg_set(C, "Context missing 'shaderfx'");
@@ -307,7 +307,7 @@ static bool edit_shaderfx_poll_generic(bContext *C, StructRNA *rna_type, int obt
   if (obtype_flag && ((1 << ob->type) & obtype_flag) == 0) {
     return 0;
   }
-  if (ptr.id.data && ID_IS_LINKED(ptr.id.data)) {
+  if (ptr.owner_id && ID_IS_LINKED(ptr.owner_id)) {
     return 0;
   }
 
diff --git a/source/blender/editors/object/object_transform.c b/source/blender/editors/object/object_transform.c
index 975aa0f5bac..2a8b306b085 100644
--- a/source/blender/editors/object/object_transform.c
+++ b/source/blender/editors/object/object_transform.c
@@ -77,7 +77,9 @@
 
 #include "object_intern.h"
 
-/*************************** Clear Transformation ****************************/
+/* -------------------------------------------------------------------- */
+/** \name Clear Transformation Utilities
+ * \{ */
 
 /* clear location of object */
 static void object_clear_loc(Object *ob, const bool clear_delta)
@@ -284,8 +286,6 @@ static void object_clear_scale(Object *ob, const bool clear_delta)
   }
 }
 
-/* --------------- */
-
 /* generic exec for clear-transform operators */
 static int object_clear_transform_generic_exec(bContext *C,
                                                wmOperator *op,
@@ -329,7 +329,11 @@ static int object_clear_transform_generic_exec(bContext *C,
   return OPERATOR_FINISHED;
 }
 
-/* --------------- */
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Clear Location Operator
+ * \{ */
 
 static int object_location_clear_exec(bContext *C, wmOperator *op)
 {
@@ -359,6 +363,12 @@ void OBJECT_OT_location_clear(wmOperatorType *ot)
       "Clear delta location in addition to clearing the normal location transform");
 }
 
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Clear Rotation Operator
+ * \{ */
+
 static int object_rotation_clear_exec(bContext *C, wmOperator *op)
 {
   return object_clear_transform_generic_exec(C, op, object_clear_rot, ANIM_KS_ROTATION_ID);
@@ -387,6 +397,12 @@ void OBJECT_OT_rotation_clear(wmOperatorType *ot)
       "Clear delta rotation in addition to clearing the normal rotation transform");
 }
 
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Clear Scale Operator
+ * \{ */
+
 static int object_scale_clear_exec(bContext *C, wmOperator *op)
 {
   return object_clear_transform_generic_exec(C, op, object_clear_scale, ANIM_KS_SCALING_ID);
@@ -415,7 +431,11 @@ void OBJECT_OT_scale_clear(wmOperatorType *ot)
       "Clear delta scale in addition to clearing the normal scale transform");
 }
 
-/* --------------- */
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Clear Origin Operator
+ * \{ */
 
 static int object_origin_clear_exec(bContext *C, wmOperator *UNUSED(op))
 {
@@ -457,7 +477,11 @@ void OBJECT_OT_origin_clear(wmOperatorType *ot)
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 }
 
-/*************************** Apply Transformation ****************************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Apply Transformation Operator
+ * \{ */
 
 /* use this when the loc/size/rot of the parent has changed but the children
  * should stay in the same place, e.g. for apply-size-rot or object center */
@@ -744,7 +768,8 @@ static int apply_objects_internal(bContext *C,
       BKE_mesh_calc_normals(me);
     }
     else if (ob->type == OB_ARMATURE) {
-      ED_armature_transform_apply(bmain, ob, mat, do_props);
+      bArmature *arm = ob->data;
+      BKE_armature_transform(arm, mat, do_props);
     }
     else if (ob->type == OB_LATTICE) {
       Lattice *lt = ob->data;
@@ -960,7 +985,11 @@ void OBJECT_OT_transform_apply(wmOperatorType *ot)
                   "Modify properties such as curve vertex radius, font size and bone envelope");
 }
 
-/********************* Set Object Center ************************/
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Set Object Center Operator
+ * \{ */
 
 enum {
   GEOMETRY_TO_ORIGIN = 0,
@@ -1152,7 +1181,8 @@ static int object_origin_set_exec(bContext *C, wmOperator *op)
       else if (ELEM(ob->type, OB_CURVE, OB_SURF)) {
         Curve *cu = ob->data;
 
-        if (centermode == ORIGIN_TO_CURSOR) { /* done */
+        if (centermode == ORIGIN_TO_CURSOR) {
+          /* done */
         }
         else if (around == V3D_AROUND_CENTER_MEDIAN) {
           BKE_curve_center_median(cu, cent);
@@ -1244,7 +1274,8 @@ static int object_origin_set_exec(bContext *C, wmOperator *op)
       else if (ob->type == OB_MBALL) {
         MetaBall *mb = ob->data;
 
-        if (centermode == ORIGIN_TO_CURSOR) { /* done */
+        if (centermode == ORIGIN_TO_CURSOR) {
+          /* done */
         }
         else if (around == V3D_AROUND_CENTER_MEDIAN) {
           BKE_mball_center_median(mb, cent);
@@ -1270,7 +1301,8 @@ static int object_origin_set_exec(bContext *C, wmOperator *op)
       else if (ob->type == OB_LATTICE) {
         Lattice *lt = ob->data;
 
-        if (centermode == ORIGIN_TO_CURSOR) { /* done */
+        if (centermode == ORIGIN_TO_CURSOR) {
+          /* done */
         }
         else if (around == V3D_AROUND_CENTER_MEDIAN) {
           BKE_lattice_center_median(lt, cent);
@@ -1323,11 +1355,6 @@ static int object_origin_set_exec(bContext *C, wmOperator *op)
               invert_m4_m4(inverse_diff_mat, diff_mat);
               for (bGPDframe *gpf = gpl->frames.first; gpf; gpf = gpf->next) {
                 for (bGPDstroke *gps = gpf->strokes.first; gps; gps = gps->next) {
-                  /* skip strokes that are invalid for current view */
-                  if (ED_gpencil_stroke_can_use(C, gps) == false) {
-                    continue;
-                  }
-
                   for (i = 0, pt = gps->points; i < gps->totpoints; i++, pt++) {
                     float mpt[3];
                     mul_v3_m4v3(mpt, inverse_diff_mat, &pt->x);
@@ -1509,6 +1536,8 @@ void OBJECT_OT_origin_set(wmOperatorType *ot)
   RNA_def_enum(ot->srna, "center", prop_set_bounds_types, V3D_AROUND_CENTER_MEDIAN, "Center", "");
 }
 
+/** \} */
+
 /* -------------------------------------------------------------------- */
 /** \name Transform Axis Target
  *
diff --git a/source/blender/editors/object/object_vgroup.c b/source/blender/editors/object/object_vgroup.c
index 08fe5e818b2..ce385b12608 100644
--- a/source/blender/editors/object/object_vgroup.c
+++ b/source/blender/editors/object/object_vgroup.c
@@ -3794,7 +3794,7 @@ static int vgroup_sort_name(const void *def_a_ptr, const void *def_b_ptr)
   const bDeformGroup *def_a = def_a_ptr;
   const bDeformGroup *def_b = def_b_ptr;
 
-  return BLI_natstrcmp(def_a->name, def_b->name);
+  return BLI_strcasecmp_natural(def_a->name, def_b->name);
 }
 
 /**
diff --git a/source/blender/editors/physics/particle_object.c b/source/blender/editors/physics/particle_object.c
index 0999d1f3588..fbaf02b7b3f 100644
--- a/source/blender/editors/physics/particle_object.c
+++ b/source/blender/editors/physics/particle_object.c
@@ -186,7 +186,7 @@ static int new_particle_settings_exec(bContext *C, wmOperator *UNUSED(op))
     part = BKE_particlesettings_add(bmain, "ParticleSettings");
   }
 
-  ob = ptr.id.data;
+  ob = (Object *)ptr.owner_id;
 
   if (psys->part) {
     id_us_min(&psys->part->id);
@@ -226,7 +226,7 @@ static int new_particle_target_exec(bContext *C, wmOperator *UNUSED(op))
   Main *bmain = CTX_data_main(C);
   PointerRNA ptr = CTX_data_pointer_get_type(C, "particle_system", &RNA_ParticleSystem);
   ParticleSystem *psys = ptr.data;
-  Object *ob = ptr.id.data;
+  Object *ob = (Object *)ptr.owner_id;
 
   ParticleTarget *pt;
 
@@ -273,7 +273,7 @@ static int remove_particle_target_exec(bContext *C, wmOperator *UNUSED(op))
   Main *bmain = CTX_data_main(C);
   PointerRNA ptr = CTX_data_pointer_get_type(C, "particle_system", &RNA_ParticleSystem);
   ParticleSystem *psys = ptr.data;
-  Object *ob = ptr.id.data;
+  Object *ob = (Object *)ptr.owner_id;
 
   ParticleTarget *pt;
 
@@ -323,7 +323,7 @@ static int target_move_up_exec(bContext *C, wmOperator *UNUSED(op))
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "particle_system", &RNA_ParticleSystem);
   ParticleSystem *psys = ptr.data;
-  Object *ob = ptr.id.data;
+  Object *ob = (Object *)ptr.owner_id;
   ParticleTarget *pt;
 
   if (!psys) {
@@ -363,7 +363,7 @@ static int target_move_down_exec(bContext *C, wmOperator *UNUSED(op))
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "particle_system", &RNA_ParticleSystem);
   ParticleSystem *psys = ptr.data;
-  Object *ob = ptr.id.data;
+  Object *ob = (Object *)ptr.owner_id;
   ParticleTarget *pt;
 
   if (!psys) {
@@ -1140,6 +1140,7 @@ static bool copy_particle_systems_to_object(const bContext *C,
 
     /* append to the object */
     BLI_addtail(&ob_to->particlesystem, psys);
+    psys_unique_name(ob_to, psys, psys->name);
 
     /* add a particle system modifier for each system */
     md = modifier_new(eModifierType_ParticleSystem);
diff --git a/source/blender/editors/physics/physics_intern.h b/source/blender/editors/physics/physics_intern.h
index 9ebbba07fdf..fc2f3d21bb6 100644
--- a/source/blender/editors/physics/physics_intern.h
+++ b/source/blender/editors/physics/physics_intern.h
@@ -30,7 +30,6 @@ struct PTCacheEdit;
 struct ParticleSystem;
 struct PointCache;
 struct Scene;
-struct ViewLayer;
 struct wmOperatorType;
 
 /* particle_edit.c */
diff --git a/source/blender/editors/physics/physics_pointcache.c b/source/blender/editors/physics/physics_pointcache.c
index bc2f1d6cef6..700a94e4f93 100644
--- a/source/blender/editors/physics/physics_pointcache.c
+++ b/source/blender/editors/physics/physics_pointcache.c
@@ -57,7 +57,7 @@ static bool ptcache_bake_all_poll(bContext *C)
 static bool ptcache_poll(bContext *C)
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "point_cache", &RNA_PointCache);
-  return (ptr.data && ptr.id.data);
+  return (ptr.data && ptr.owner_id);
 }
 
 typedef struct PointCacheJob {
@@ -165,7 +165,7 @@ static PTCacheBaker *ptcache_baker_create(bContext *C, wmOperator *op, bool all)
 
   if (!all) {
     PointerRNA ptr = CTX_data_pointer_get_type(C, "point_cache", &RNA_PointCache);
-    Object *ob = ptr.id.data;
+    Object *ob = (Object *)ptr.owner_id;
     PointCache *cache = ptr.data;
     baker->pid = BKE_ptcache_id_find(ob, baker->scene, cache);
   }
@@ -300,7 +300,7 @@ static int ptcache_free_bake_exec(bContext *C, wmOperator *UNUSED(op))
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "point_cache", &RNA_PointCache);
   PointCache *cache = ptr.data;
-  Object *ob = ptr.id.data;
+  Object *ob = (Object *)ptr.owner_id;
 
   ptcache_free_bake(cache);
 
@@ -312,7 +312,7 @@ static int ptcache_bake_from_cache_exec(bContext *C, wmOperator *UNUSED(op))
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "point_cache", &RNA_PointCache);
   PointCache *cache = ptr.data;
-  Object *ob = ptr.id.data;
+  Object *ob = (Object *)ptr.owner_id;
 
   cache->flag |= PTCACHE_BAKED;
 
@@ -372,7 +372,7 @@ static int ptcache_add_new_exec(bContext *C, wmOperator *UNUSED(op))
 {
   Scene *scene = CTX_data_scene(C);
   PointerRNA ptr = CTX_data_pointer_get_type(C, "point_cache", &RNA_PointCache);
-  Object *ob = ptr.id.data;
+  Object *ob = (Object *)ptr.owner_id;
   PointCache *cache = ptr.data;
   PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
 
@@ -392,7 +392,7 @@ static int ptcache_remove_exec(bContext *C, wmOperator *UNUSED(op))
 {
   PointerRNA ptr = CTX_data_pointer_get_type(C, "point_cache", &RNA_PointCache);
   Scene *scene = CTX_data_scene(C);
-  Object *ob = ptr.id.data;
+  Object *ob = (Object *)ptr.owner_id;
   PointCache *cache = ptr.data;
   PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
 
diff --git a/source/blender/editors/render/render_intern.h b/source/blender/editors/render/render_intern.h
index 50f0b53c037..88b913b84ca 100644
--- a/source/blender/editors/render/render_intern.h
+++ b/source/blender/editors/render/render_intern.h
@@ -24,7 +24,6 @@
 #ifndef __RENDER_INTERN_H__
 #define __RENDER_INTERN_H__
 
-struct RenderEngine;
 struct ScrArea;
 struct bContext;
 struct wmOperatorType;
diff --git a/source/blender/editors/render/render_shading.c b/source/blender/editors/render/render_shading.c
index 01740c13c9d..9f13431f25a 100644
--- a/source/blender/editors/render/render_shading.c
+++ b/source/blender/editors/render/render_shading.c
@@ -1836,7 +1836,7 @@ void SCENE_OT_freestyle_stroke_material_create(wmOperatorType *ot)
 
 static int texture_slot_move_exec(bContext *C, wmOperator *op)
 {
-  ID *id = CTX_data_pointer_get_type(C, "texture_slot", &RNA_TextureSlot).id.data;
+  ID *id = CTX_data_pointer_get_type(C, "texture_slot", &RNA_TextureSlot).owner_id;
 
   if (id) {
     MTex **mtex_ar, *mtexswap;
@@ -2031,7 +2031,7 @@ static void paste_mtex_copybuf(ID *id)
 
 static int copy_mtex_exec(bContext *C, wmOperator *UNUSED(op))
 {
-  ID *id = CTX_data_pointer_get_type(C, "texture_slot", &RNA_TextureSlot).id.data;
+  ID *id = CTX_data_pointer_get_type(C, "texture_slot", &RNA_TextureSlot).owner_id;
 
   if (id == NULL) {
     /* copying empty slot */
@@ -2046,7 +2046,7 @@ static int copy_mtex_exec(bContext *C, wmOperator *UNUSED(op))
 
 static bool copy_mtex_poll(bContext *C)
 {
-  ID *id = CTX_data_pointer_get_type(C, "texture_slot", &RNA_TextureSlot).id.data;
+  ID *id = CTX_data_pointer_get_type(C, "texture_slot", &RNA_TextureSlot).owner_id;
 
   return (id != NULL);
 }
@@ -2069,7 +2069,7 @@ void TEXTURE_OT_slot_copy(wmOperatorType *ot)
 
 static int paste_mtex_exec(bContext *C, wmOperator *UNUSED(op))
 {
-  ID *id = CTX_data_pointer_get_type(C, "texture_slot", &RNA_TextureSlot).id.data;
+  ID *id = CTX_data_pointer_get_type(C, "texture_slot", &RNA_TextureSlot).owner_id;
 
   if (id == NULL) {
     Material *ma = CTX_data_pointer_get_type(C, "material", &RNA_Material).data;
diff --git a/source/blender/editors/render/render_view.c b/source/blender/editors/render/render_view.c
index cd5edcdc3f4..3154d5d0985 100644
--- a/source/blender/editors/render/render_view.c
+++ b/source/blender/editors/render/render_view.c
@@ -223,8 +223,8 @@ ScrArea *render_view_open(bContext *C, int mx, int my, ReportList *reports)
   /* get the correct image, and scale it */
   sima->image = BKE_image_verify_viewer(bmain, IMA_TYPE_R_RESULT, "Render Result");
 
-  /* if we're rendering to full screen, set appropriate hints on image editor
-   * so it can restore properly on pressing esc */
+  /* If we're rendering to full screen, set appropriate hints on image editor
+   * so it can restore properly on pressing escape. */
   if (sa->full) {
     sima->flag |= SI_FULLWINDOW;
 
diff --git a/source/blender/editors/screen/area.c b/source/blender/editors/screen/area.c
index 1775a0c55a2..b0c9e3b9378 100644
--- a/source/blender/editors/screen/area.c
+++ b/source/blender/editors/screen/area.c
@@ -1052,16 +1052,11 @@ static void region_azones_scrollbars_initialize(ScrArea *sa, ARegion *ar)
 }
 
 /* *************************************************************** */
-
-static void region_azones_add(const bScreen *screen, ScrArea *sa, ARegion *ar, const int alignment)
+static void region_azones_add_edge(ScrArea *sa,
+                                   ARegion *ar,
+                                   const int alignment,
+                                   const bool is_fullscreen)
 {
-  const bool is_fullscreen = screen->state == SCREENFULL;
-
-  /* Only display tab or icons when the header region is hidden
-   * (not the tool header - they overlap). */
-  if (ar->regiontype == RGN_TYPE_TOOL_HEADER) {
-    return;
-  }
 
   /* edge code (t b l r) is along which area edge azone will be drawn */
   if (alignment == RGN_ALIGN_TOP) {
@@ -1076,6 +1071,25 @@ static void region_azones_add(const bScreen *screen, ScrArea *sa, ARegion *ar, c
   else if (alignment == RGN_ALIGN_LEFT) {
     region_azone_edge_initialize(sa, ar, AE_RIGHT_TO_TOPLEFT, is_fullscreen);
   }
+}
+
+static void region_azones_add(const bScreen *screen, ScrArea *sa, ARegion *ar)
+{
+  const bool is_fullscreen = screen->state == SCREENFULL;
+
+  /* Only display tab or icons when the header region is hidden
+   * (not the tool header - they overlap). */
+  if (ar->regiontype == RGN_TYPE_TOOL_HEADER) {
+    return;
+  }
+
+  region_azones_add_edge(sa, ar, RGN_ALIGN_ENUM_FROM_MASK(ar->alignment), is_fullscreen);
+
+  /* For a split region also continue the azone edge from the next region if this region is aligned
+   * with the next */
+  if ((ar->alignment & RGN_SPLIT_PREV) && ar->prev) {
+    region_azones_add_edge(sa, ar, RGN_ALIGN_ENUM_FROM_MASK(ar->prev->alignment), is_fullscreen);
+  }
 
   if (is_fullscreen) {
     fullscreen_azone_initialize(sa, ar);
@@ -1251,9 +1265,6 @@ static void region_rect_recursive(
   else if (ED_area_is_global(sa)) {
     prefsizey = ED_region_global_size_y();
   }
-  else if (ar->regiontype == RGN_TYPE_UI && sa->spacetype == SPACE_FILE) {
-    prefsizey = UI_UNIT_Y * 2 + (UI_UNIT_Y / 2);
-  }
   else {
     prefsizey = UI_DPI_FAC * (ar->sizey > 1 ? ar->sizey + 0.5f : ar->type->prefsizey);
   }
@@ -1695,7 +1706,7 @@ void ED_area_update_region_sizes(wmWindowManager *wm, wmWindow *win, ScrArea *ar
     }
 
     /* Some AZones use View2D data which is only updated in region init, so call that first! */
-    region_azones_add(screen, area, ar, RGN_ALIGN_ENUM_FROM_MASK(ar->alignment));
+    region_azones_add(screen, area, ar);
   }
   ED_area_azones_update(area, &win->eventstate->x);
 
@@ -1766,7 +1777,7 @@ void ED_area_initialize(wmWindowManager *wm, wmWindow *win, ScrArea *sa)
     }
 
     /* Some AZones use View2D data which is only updated in region init, so call that first! */
-    region_azones_add(screen, sa, ar, RGN_ALIGN_ENUM_FROM_MASK(ar->alignment));
+    region_azones_add(screen, sa, ar);
   }
 
   /* Avoid re-initializing tools while resizing the window. */
diff --git a/source/blender/editors/screen/screen_context.c b/source/blender/editors/screen/screen_context.c
index 2eb7f732534..6f8b25f782b 100644
--- a/source/blender/editors/screen/screen_context.c
+++ b/source/blender/editors/screen/screen_context.c
@@ -533,7 +533,7 @@ int ed_screen_context(const bContext *C, const char *member, bContextDataResult
     gpd_ptr = ED_gpencil_data_get_pointers_direct((ID *)sc, sa, scene, obact, &ptr);
 
     if (gpd_ptr) {
-      CTX_data_pointer_set(result, ptr.id.data, ptr.type, ptr.data);
+      CTX_data_pointer_set(result, ptr.owner_id, ptr.type, ptr.data);
       return 1;
     }
   }
diff --git a/source/blender/editors/screen/screen_edit.c b/source/blender/editors/screen/screen_edit.c
index dddc33e3ad0..b37aa47aba6 100644
--- a/source/blender/editors/screen/screen_edit.c
+++ b/source/blender/editors/screen/screen_edit.c
@@ -292,33 +292,45 @@ void screen_new_activate_prepare(const wmWindow *win, bScreen *screen_new)
 /* used with join operator */
 int area_getorientation(ScrArea *sa, ScrArea *sb)
 {
-  ScrVert *sav1, *sav2, *sav3, *sav4;
-  ScrVert *sbv1, *sbv2, *sbv3, *sbv4;
-
   if (sa == NULL || sb == NULL) {
     return -1;
   }
 
-  sav1 = sa->v1;
-  sav2 = sa->v2;
-  sav3 = sa->v3;
-  sav4 = sa->v4;
-  sbv1 = sb->v1;
-  sbv2 = sb->v2;
-  sbv3 = sb->v3;
-  sbv4 = sb->v4;
+  ScrVert *saBL = sa->v1;
+  ScrVert *saTL = sa->v2;
+  ScrVert *saTR = sa->v3;
+  ScrVert *saBR = sa->v4;
 
-  if (sav1 == sbv4 && sav2 == sbv3) { /* sa to right of sb = W */
-    return 0;
+  ScrVert *sbBL = sb->v1;
+  ScrVert *sbTL = sb->v2;
+  ScrVert *sbTR = sb->v3;
+  ScrVert *sbBR = sb->v4;
+
+  int tolerance = U.pixelsize * 4;
+
+  if (saBL->vec.x == sbBR->vec.x && saTL->vec.x == sbTR->vec.x) { /* sa to right of sb = W */
+    if ((ABS(saBL->vec.y - sbBR->vec.y) <= tolerance) &&
+        (ABS(saTL->vec.y - sbTR->vec.y) <= tolerance)) {
+      return 0;
+    }
   }
-  else if (sav2 == sbv1 && sav3 == sbv4) { /* sa to bottom of sb = N */
-    return 1;
+  else if (saTL->vec.y == sbBL->vec.y && saTR->vec.y == sbBR->vec.y) { /* sa to bottom of sb = N */
+    if ((ABS(saTL->vec.x - sbBL->vec.x) <= tolerance) &&
+        (ABS(saTR->vec.x - sbBR->vec.x) <= tolerance)) {
+      return 1;
+    }
   }
-  else if (sav3 == sbv2 && sav4 == sbv1) { /* sa to left of sb = E */
-    return 2;
+  else if (saTR->vec.x == sbTL->vec.x && saBR->vec.x == sbBL->vec.x) { /* sa to left of sb = E */
+    if ((ABS(saTR->vec.y - sbTL->vec.y) <= tolerance) &&
+        (ABS(saBR->vec.y - sbBL->vec.y) <= tolerance)) {
+      return 2;
+    }
   }
-  else if (sav1 == sbv2 && sav4 == sbv3) { /* sa on top of sb = S*/
-    return 3;
+  else if (saBL->vec.y == sbTL->vec.y && saBR->vec.y == sbTR->vec.y) { /* sa on top of sb = S*/
+    if ((ABS(saBL->vec.x - sbTL->vec.x) <= tolerance) &&
+        (ABS(saBR->vec.x - sbTR->vec.x) <= tolerance)) {
+      return 3;
+    }
   }
 
   return -1;
@@ -329,36 +341,50 @@ int area_getorientation(ScrArea *sa, ScrArea *sb)
  */
 int screen_area_join(bContext *C, bScreen *scr, ScrArea *sa1, ScrArea *sa2)
 {
-  int dir;
-
-  dir = area_getorientation(sa1, sa2);
-  /*printf("dir is : %i\n", dir);*/
+  int dir = area_getorientation(sa1, sa2);
 
   if (dir == -1) {
     return 0;
   }
 
-  if (dir == 0) {
-    sa1->v1 = sa2->v1;
-    sa1->v2 = sa2->v2;
+  /* Align areas if they are not. Do sanity checking before getting here. */
+
+  if (dir == 0 || dir == 2) {
+    /* horizontal join, so vertically align source vert to target */
+    sa2->v1->vec.y = sa1->v1->vec.y; /* vertical align sa1 BL */
+    sa2->v2->vec.y = sa1->v2->vec.y; /* vertical align sa1 TL */
+    sa2->v3->vec.y = sa1->v3->vec.y; /* vertical align sa1 TR */
+    sa2->v4->vec.y = sa1->v4->vec.y; /* vertical align sa1 BR */
+  }
+  else {
+    /* vertical join, so horizontally align source verts to target */
+    sa2->v1->vec.x = sa1->v1->vec.x; /* vertical align sa1 BL */
+    sa2->v2->vec.x = sa1->v2->vec.x; /* vertical align sa1 TL */
+    sa2->v3->vec.x = sa1->v3->vec.x; /* vertical align sa1 TR */
+    sa2->v4->vec.x = sa1->v4->vec.x; /* vertical align sa1 BR */
+  }
+
+  if (dir == 0) {      /* sa1 to right of sa2 = W */
+    sa1->v1 = sa2->v1; /* BL */
+    sa1->v2 = sa2->v2; /* TL */
     screen_geom_edge_add(scr, sa1->v2, sa1->v3);
     screen_geom_edge_add(scr, sa1->v1, sa1->v4);
   }
-  else if (dir == 1) {
-    sa1->v2 = sa2->v2;
-    sa1->v3 = sa2->v3;
+  else if (dir == 1) { /* sa1 to bottom of sa2 = N */
+    sa1->v2 = sa2->v2; /* TL */
+    sa1->v3 = sa2->v3; /* TR */
     screen_geom_edge_add(scr, sa1->v1, sa1->v2);
     screen_geom_edge_add(scr, sa1->v3, sa1->v4);
   }
-  else if (dir == 2) {
-    sa1->v3 = sa2->v3;
-    sa1->v4 = sa2->v4;
+  else if (dir == 2) { /* sa1 to left of sa2 = E */
+    sa1->v3 = sa2->v3; /* TR */
+    sa1->v4 = sa2->v4; /* BR */
     screen_geom_edge_add(scr, sa1->v2, sa1->v3);
     screen_geom_edge_add(scr, sa1->v1, sa1->v4);
   }
-  else if (dir == 3) {
-    sa1->v1 = sa2->v1;
-    sa1->v4 = sa2->v4;
+  else if (dir == 3) { /* sa1 on top of sa2 = S */
+    sa1->v1 = sa2->v1; /* BL */
+    sa1->v4 = sa2->v4; /* BR */
     screen_geom_edge_add(scr, sa1->v1, sa1->v2);
     screen_geom_edge_add(scr, sa1->v3, sa1->v4);
   }
@@ -1116,7 +1142,7 @@ void ED_screen_full_prevspace(bContext *C, ScrArea *sa)
 
 void ED_screen_restore_temp_type(bContext *C, ScrArea *sa)
 {
-  /* incase nether functions below run */
+  /* In case nether functions below run. */
   ED_area_tag_redraw(sa);
 
   if (sa->flag & AREA_FLAG_TEMP_TYPE) {
diff --git a/source/blender/editors/screen/screen_ops.c b/source/blender/editors/screen/screen_ops.c
index 8377fd0e128..50e5597ac0c 100644
--- a/source/blender/editors/screen/screen_ops.c
+++ b/source/blender/editors/screen/screen_ops.c
@@ -1121,6 +1121,47 @@ static void SCREEN_OT_actionzone(wmOperatorType *ot)
 /** \} */
 
 /* -------------------------------------------------------------------- */
+/** \name Area edge detection utility
+ * \{ */
+
+static ScrEdge *screen_area_edge_from_cursor(const bContext *C,
+                                             const int cursor[2],
+                                             ScrArea **r_sa1,
+                                             ScrArea **r_sa2)
+{
+  wmWindow *win = CTX_wm_window(C);
+  bScreen *sc = CTX_wm_screen(C);
+  ScrEdge *actedge;
+  rcti window_rect;
+  WM_window_rect_calc(win, &window_rect);
+  actedge = screen_geom_area_map_find_active_scredge(
+      AREAMAP_FROM_SCREEN(sc), &window_rect, cursor[0], cursor[1]);
+  *r_sa1 = NULL;
+  *r_sa2 = NULL;
+  if (actedge == NULL) {
+    return NULL;
+  }
+  int borderwidth = (4 * UI_DPI_FAC);
+  ScrArea *sa1, *sa2;
+  if (screen_geom_edge_is_horizontal(actedge)) {
+    sa1 = BKE_screen_find_area_xy(sc, SPACE_TYPE_ANY, cursor[0], cursor[1] + borderwidth);
+    sa2 = BKE_screen_find_area_xy(sc, SPACE_TYPE_ANY, cursor[0], cursor[1] - borderwidth);
+  }
+  else {
+    sa1 = BKE_screen_find_area_xy(sc, SPACE_TYPE_ANY, cursor[0] + borderwidth, cursor[1]);
+    sa2 = BKE_screen_find_area_xy(sc, SPACE_TYPE_ANY, cursor[0] - borderwidth, cursor[1]);
+  }
+  bool isGlobal = ((sa1 && ED_area_is_global(sa1)) || (sa2 && ED_area_is_global(sa2)));
+  if (!isGlobal) {
+    *r_sa1 = sa1;
+    *r_sa2 = sa2;
+  }
+  return actedge;
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Swap Area Operator
  * \{ */
 
@@ -1139,6 +1180,7 @@ static void SCREEN_OT_actionzone(wmOperatorType *ot)
  * callbacks:
  *
  * invoke() gets called on shift+lmb drag in action-zone
+ * exec()   execute without any user interaction, based on properties
  * call init(), add handler
  *
  * modal()  accept modal events while doing it
@@ -1229,6 +1271,19 @@ static int area_swap_modal(bContext *C, wmOperator *op, const wmEvent *event)
   return OPERATOR_RUNNING_MODAL;
 }
 
+static int area_swap_exec(bContext *C, wmOperator *op)
+{
+  ScrArea *sa1, *sa2;
+  int cursor[2];
+  RNA_int_get_array(op->ptr, "cursor", cursor);
+  screen_area_edge_from_cursor(C, cursor, &sa1, &sa2);
+  if (sa1 == NULL || sa2 == NULL) {
+    return OPERATOR_CANCELLED;
+  }
+  ED_area_swapspace(C, sa1, sa2);
+  return OPERATOR_FINISHED;
+}
+
 static void SCREEN_OT_area_swap(wmOperatorType *ot)
 {
   ot->name = "Swap Areas";
@@ -1237,10 +1292,15 @@ static void SCREEN_OT_area_swap(wmOperatorType *ot)
 
   ot->invoke = area_swap_invoke;
   ot->modal = area_swap_modal;
-  ot->poll = ED_operator_areaactive;
+  ot->exec = area_swap_exec;
+  ot->poll = screen_active_editable;
   ot->cancel = area_swap_cancel;
 
   ot->flag = OPTYPE_BLOCKING;
+
+  /* rna */
+  RNA_def_int_vector(
+      ot->srna, "cursor", 2, NULL, INT_MIN, INT_MAX, "Cursor", "", INT_MIN, INT_MAX);
 }
 
 /** \} */
@@ -2458,6 +2518,14 @@ static int area_max_regionsize(ScrArea *sa, ARegion *scalear, AZEdge edge)
   return dist;
 }
 
+static bool is_split_edge(const int alignment, const AZEdge edge)
+{
+  return ((alignment == RGN_ALIGN_BOTTOM) && (edge == AE_TOP_TO_BOTTOMRIGHT)) ||
+         ((alignment == RGN_ALIGN_TOP) && (edge == AE_BOTTOM_TO_TOPLEFT)) ||
+         ((alignment == RGN_ALIGN_LEFT) && (edge == AE_RIGHT_TO_TOPLEFT)) ||
+         ((alignment == RGN_ALIGN_RIGHT) && (edge == AE_LEFT_TO_TOPRIGHT));
+}
+
 static int region_scale_invoke(bContext *C, wmOperator *op, const wmEvent *event)
 {
   sActionzoneData *sad = event->customdata;
@@ -2476,7 +2544,16 @@ static int region_scale_invoke(bContext *C, wmOperator *op, const wmEvent *event
     op->customdata = rmd;
 
     rmd->az = az;
-    rmd->ar = az->ar;
+    /* special case for region within region - this allows the scale of
+     * the parent region if the azone edge is not the edge splitting
+     * both regions */
+    if ((az->ar->alignment & RGN_SPLIT_PREV) && az->ar->prev &&
+        !is_split_edge(RGN_ALIGN_ENUM_FROM_MASK(az->ar->alignment), az->edge)) {
+      rmd->ar = az->ar->prev;
+    }
+    else {
+      rmd->ar = az->ar;
+    }
     rmd->sa = sad->sa1;
     rmd->edge = az->edge;
     rmd->origx = event->x;
@@ -3163,40 +3240,19 @@ static void area_join_draw_cb(const struct wmWindow *UNUSED(win), void *userdata
 
 /* validate selection inside screen, set variables OK */
 /* return 0: init failed */
-/* XXX todo: find edge based on (x,y) and set other area? */
-static int area_join_init(bContext *C, wmOperator *op)
+static int area_join_init(bContext *C, wmOperator *op, ScrArea *sa1, ScrArea *sa2)
 {
-  const wmWindow *win = CTX_wm_window(C);
-  bScreen *screen = CTX_wm_screen(C);
-  ScrArea *sa1, *sa2;
-  sAreaJoinData *jd = NULL;
-  int x1, y1;
-  int x2, y2;
-
-  /* required properties, make negative to get return 0 if not set by caller */
-  x1 = RNA_int_get(op->ptr, "min_x");
-  y1 = RNA_int_get(op->ptr, "min_y");
-  x2 = RNA_int_get(op->ptr, "max_x");
-  y2 = RNA_int_get(op->ptr, "max_y");
-
-  sa1 = BKE_screen_find_area_xy(screen, SPACE_TYPE_ANY, x1, y1);
-  if (sa1 == NULL) {
-    sa1 = BKE_screen_area_map_find_area_xy(&win->global_areas, SPACE_TYPE_ANY, x1, y1);
-  }
-  sa2 = BKE_screen_find_area_xy(screen, SPACE_TYPE_ANY, x2, y2);
-  if (sa2 == NULL) {
-    sa2 = BKE_screen_area_map_find_area_xy(&win->global_areas, SPACE_TYPE_ANY, x2, y2);
-  }
-  if ((sa1 && ED_area_is_global(sa1)) || (sa2 && ED_area_is_global(sa2))) {
-    BKE_report(
-        op->reports, RPT_ERROR, "Global areas (Top Bar, Status Bar) do not support joining");
-    return 0;
+  if (sa1 == NULL || sa2 == NULL) {
+    /* Get areas from cursor location if not specified. */
+    int cursor[2];
+    RNA_int_get_array(op->ptr, "cursor", cursor);
+    screen_area_edge_from_cursor(C, cursor, &sa1, &sa2);
   }
-  else if (sa1 == NULL || sa2 == NULL || sa1 == sa2) {
+  if (sa1 == NULL || sa2 == NULL) {
     return 0;
   }
 
-  jd = (sAreaJoinData *)MEM_callocN(sizeof(sAreaJoinData), "op_area_join");
+  sAreaJoinData *jd = MEM_callocN(sizeof(sAreaJoinData), "op_area_join");
 
   jd->sa1 = sa1;
   jd->sa2 = sa2;
@@ -3249,7 +3305,7 @@ static void area_join_exit(bContext *C, wmOperator *op)
 
 static int area_join_exec(bContext *C, wmOperator *op)
 {
-  if (!area_join_init(C, op)) {
+  if (!area_join_init(C, op, NULL, NULL)) {
     return OPERATOR_CANCELLED;
   }
 
@@ -3279,16 +3335,11 @@ static int area_join_invoke(bContext *C, wmOperator *op, const wmEvent *event)
     if (sad->sa1 == sad->sa2) {
       return OPERATOR_PASS_THROUGH;
     }
-
-    /* prepare operator state vars */
-    RNA_int_set(op->ptr, "min_x", sad->sa1->totrct.xmin);
-    RNA_int_set(op->ptr, "min_y", sad->sa1->totrct.ymin);
-    RNA_int_set(op->ptr, "max_x", sad->sa2->totrct.xmin);
-    RNA_int_set(op->ptr, "max_y", sad->sa2->totrct.ymin);
-  }
-
-  if (!area_join_init(C, op)) {
-    return OPERATOR_CANCELLED;
+    else {
+      if (!area_join_init(C, op, sad->sa1, sad->sa2)) {
+        return OPERATOR_CANCELLED;
+      }
+    }
   }
 
   /* add temp handler */
@@ -3309,7 +3360,14 @@ static int area_join_modal(bContext *C, wmOperator *op, const wmEvent *event)
 {
   bScreen *sc = CTX_wm_screen(C);
   wmWindow *win = CTX_wm_window(C);
-  sAreaJoinData *jd = (sAreaJoinData *)op->customdata;
+  sAreaJoinData *jd;
+
+  if (op->customdata == NULL) {
+    if (!area_join_init(C, op, NULL, NULL)) {
+      return OPERATOR_CANCELLED;
+    }
+  }
+  jd = (sAreaJoinData *)op->customdata;
 
   /* execute the events */
   switch (event->type) {
@@ -3419,10 +3477,8 @@ static void SCREEN_OT_area_join(wmOperatorType *ot)
   ot->flag = OPTYPE_BLOCKING | OPTYPE_INTERNAL;
 
   /* rna */
-  RNA_def_int(ot->srna, "min_x", -100, INT_MIN, INT_MAX, "X 1", "", INT_MIN, INT_MAX);
-  RNA_def_int(ot->srna, "min_y", -100, INT_MIN, INT_MAX, "Y 1", "", INT_MIN, INT_MAX);
-  RNA_def_int(ot->srna, "max_x", -100, INT_MIN, INT_MAX, "X 2", "", INT_MIN, INT_MAX);
-  RNA_def_int(ot->srna, "max_y", -100, INT_MIN, INT_MAX, "Y 2", "", INT_MIN, INT_MAX);
+  RNA_def_int_vector(
+      ot->srna, "cursor", 2, NULL, INT_MIN, INT_MAX, "Cursor", "", INT_MIN, INT_MAX);
 }
 
 /** \} */
@@ -3433,36 +3489,74 @@ static void SCREEN_OT_area_join(wmOperatorType *ot)
 
 static int screen_area_options_invoke(bContext *C, wmOperator *op, const wmEvent *event)
 {
-  const wmWindow *win = CTX_wm_window(C);
-  const bScreen *sc = CTX_wm_screen(C);
   uiPopupMenu *pup;
   uiLayout *layout;
   PointerRNA ptr;
-  ScrEdge *actedge;
-  rcti window_rect;
 
-  WM_window_rect_calc(win, &window_rect);
-  actedge = screen_geom_area_map_find_active_scredge(
-      AREAMAP_FROM_SCREEN(sc), &window_rect, event->x, event->y);
+  ScrArea *sa1, *sa2;
 
-  if (actedge == NULL) {
+  if (screen_area_edge_from_cursor(C, &event->x, &sa1, &sa2) == NULL) {
     return OPERATOR_CANCELLED;
   }
 
   pup = UI_popup_menu_begin(C, WM_operatortype_name(op->type, op->ptr), ICON_NONE);
   layout = UI_popup_menu_layout(pup);
 
-  uiItemFullO(
-      layout, "SCREEN_OT_area_split", NULL, ICON_NONE, NULL, WM_OP_INVOKE_DEFAULT, 0, &ptr);
-  /* store initial mouse cursor position */
+  /* Vertical Split */
+  uiItemFullO(layout,
+              "SCREEN_OT_area_split",
+              IFACE_("Vertical Split"),
+              ICON_NONE,
+              NULL,
+              WM_OP_INVOKE_DEFAULT,
+              0,
+              &ptr);
+  /* store initial mouse cursor position. */
+  RNA_int_set_array(&ptr, "cursor", &event->x);
+  RNA_enum_set(&ptr, "direction", 'v');
+
+  /* Horizontal Split */
+  uiItemFullO(layout,
+              "SCREEN_OT_area_split",
+              IFACE_("Horizontal Split"),
+              ICON_NONE,
+              NULL,
+              WM_OP_INVOKE_DEFAULT,
+              0,
+              &ptr);
+  /* store initial mouse cursor position. */
   RNA_int_set_array(&ptr, "cursor", &event->x);
+  RNA_enum_set(&ptr, "direction", 'h');
 
-  uiItemFullO(layout, "SCREEN_OT_area_join", NULL, ICON_NONE, NULL, WM_OP_INVOKE_DEFAULT, 0, &ptr);
-  /* mouse cursor on edge, '4' can fail on wide edges... */
-  RNA_int_set(&ptr, "min_x", event->x + 4);
-  RNA_int_set(&ptr, "min_y", event->y + 4);
-  RNA_int_set(&ptr, "max_x", event->x - 4);
-  RNA_int_set(&ptr, "max_y", event->y - 4);
+  if (sa1 && sa2) {
+    uiItemS(layout);
+  }
+
+  /* Join needs two very similar areas. */
+  if (sa1 && sa2 && (area_getorientation(sa1, sa2) != -1)) {
+    uiItemFullO(layout,
+                "SCREEN_OT_area_join",
+                IFACE_("Join Areas"),
+                ICON_NONE,
+                NULL,
+                WM_OP_INVOKE_DEFAULT,
+                0,
+                &ptr);
+    RNA_int_set_array(&ptr, "cursor", &event->x);
+  }
+
+  /* Swap just needs two areas. */
+  if (sa1 && sa2) {
+    uiItemFullO(layout,
+                "SCREEN_OT_area_swap",
+                IFACE_("Swap Areas"),
+                ICON_NONE,
+                NULL,
+                WM_OP_EXEC_DEFAULT,
+                0,
+                &ptr);
+    RNA_int_set_array(&ptr, "cursor", &event->x);
+  }
 
   UI_popup_menu_end(C, pup);
 
@@ -3660,7 +3754,7 @@ static void SCREEN_OT_redo_last(wmOperatorType *ot)
 {
   /* identifiers */
   ot->name = "Redo Last";
-  ot->description = "Display menu for last action performed";
+  ot->description = "Display parameters for last action performed";
   ot->idname = "SCREEN_OT_redo_last";
 
   /* api callbacks */
@@ -3711,7 +3805,7 @@ static int region_quadview_exec(bContext *C, wmOperator *op)
 
   /* some rules... */
   if (ar->regiontype != RGN_TYPE_WINDOW) {
-    BKE_report(op->reports, RPT_ERROR, "Only window region can be 4-splitted");
+    BKE_report(op->reports, RPT_ERROR, "Only window region can be 4-split");
   }
   else if (ar->alignment == RGN_ALIGN_QSPLIT) {
     /* Exit quad-view */
@@ -3742,7 +3836,7 @@ static int region_quadview_exec(bContext *C, wmOperator *op)
       rv3d->viewlock = 0;
       rv3d->rflag &= ~RV3D_CLIPPING;
 
-      /* accumulate locks, incase they're mixed */
+      /* Accumulate locks, in case they're mixed. */
       for (ar_iter = sa->regionbase.first; ar_iter; ar_iter = ar_iter->next) {
         if (ar_iter->regiontype == RGN_TYPE_WINDOW) {
           RegionView3D *rv3d_iter = ar_iter->regiondata;
@@ -3764,7 +3858,7 @@ static int region_quadview_exec(bContext *C, wmOperator *op)
     WM_event_add_notifier(C, NC_SCREEN | NA_EDITED, NULL);
   }
   else if (ar->next) {
-    BKE_report(op->reports, RPT_ERROR, "Only last region can be 4-splitted");
+    BKE_report(op->reports, RPT_ERROR, "Only last region can be 4-split");
   }
   else {
     /* Enter quad-view */
@@ -3842,6 +3936,65 @@ static void SCREEN_OT_region_quadview(wmOperatorType *ot)
 /** \} */
 
 /* -------------------------------------------------------------------- */
+/** \name Region Toggle Operator
+ * \{ */
+
+static int region_toggle_exec(bContext *C, wmOperator *op)
+{
+  PropertyRNA *prop = RNA_struct_find_property(op->ptr, "region_type");
+  ARegion *region;
+
+  if (RNA_property_is_set(op->ptr, prop)) {
+    region = BKE_area_find_region_type(CTX_wm_area(C), RNA_property_enum_get(op->ptr, prop));
+  }
+  else {
+    region = CTX_wm_region(C);
+  }
+
+  if (region) {
+    ED_region_toggle_hidden(C, region);
+  }
+  ED_region_tag_redraw(region);
+
+  return OPERATOR_FINISHED;
+}
+
+static bool region_toggle_poll(bContext *C)
+{
+  ScrArea *area = CTX_wm_area(C);
+
+  /* don't flip anything around in topbar */
+  if (area && area->spacetype == SPACE_TOPBAR) {
+    CTX_wm_operator_poll_msg_set(C, "Toggling regions in the Top-bar is not allowed");
+    return 0;
+  }
+
+  return ED_operator_areaactive(C);
+}
+
+static void SCREEN_OT_region_toggle(wmOperatorType *ot)
+{
+  /* identifiers */
+  ot->name = "Toggle Region";
+  ot->idname = "SCREEN_OT_region_toggle";
+  ot->description = "Hide or unhide the region";
+
+  /* api callbacks */
+  ot->exec = region_toggle_exec;
+  ot->poll = region_toggle_poll;
+  ot->flag = 0;
+
+  RNA_def_enum(ot->srna,
+               "region_type",
+               rna_enum_region_type_items,
+               0,
+               "Region Type",
+               "Type of the region to toggle");
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Region Flip Operator
  * \{ */
 
@@ -4714,7 +4867,7 @@ static void SCREEN_OT_userpref_show(struct wmOperatorType *ot)
 
 static int drivers_editor_show_invoke(bContext *C, wmOperator *op, const wmEvent *event)
 {
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   int index = -1;
   uiBut *but = NULL;
@@ -4775,6 +4928,40 @@ static void SCREEN_OT_drivers_editor_show(struct wmOperatorType *ot)
 /** \} */
 
 /* -------------------------------------------------------------------- */
+/** \name Show Info Log Operator
+ * \{ */
+
+static int info_log_show_invoke(bContext *C, wmOperator *op, const wmEvent *event)
+{
+  int sizex = 900 * UI_DPI_FAC;
+  int sizey = 580 * UI_DPI_FAC;
+  int shift_y = 480;
+
+  /* changes context! */
+  if (WM_window_open_temp(C, event->x, event->y + shift_y, sizex, sizey, WM_WINDOW_INFO) != NULL) {
+    return OPERATOR_FINISHED;
+  }
+  else {
+    BKE_report(op->reports, RPT_ERROR, "Failed to open window!");
+    return OPERATOR_CANCELLED;
+  }
+}
+
+static void SCREEN_OT_info_log_show(struct wmOperatorType *ot)
+{
+  /* identifiers */
+  ot->name = "Show Info Log";
+  ot->description = "Show info log in a separate window";
+  ot->idname = "SCREEN_OT_info_log_show";
+
+  /* api callbacks */
+  ot->invoke = info_log_show_invoke;
+  ot->poll = ED_operator_screenactive;
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name New Screen Operator
  * \{ */
 
@@ -4903,7 +5090,9 @@ static void region_blend_end(bContext *C, ARegion *ar, const bool is_running)
   WM_event_remove_timer(CTX_wm_manager(C), NULL, ar->regiontimer); /* frees rgi */
   ar->regiontimer = NULL;
 }
-/* assumes that *ar itself is not a splitted version from previous region */
+/**
+ * \note Assumes that \a ar itself is not a split version from previous region.
+ */
 void ED_region_visibility_change_update_animated(bContext *C, ScrArea *sa, ARegion *ar)
 {
   wmWindowManager *wm = CTX_wm_manager(C);
@@ -5231,6 +5420,7 @@ void ED_operatortypes_screen(void)
   WM_operatortype_append(SCREEN_OT_area_swap);
   WM_operatortype_append(SCREEN_OT_region_quadview);
   WM_operatortype_append(SCREEN_OT_region_scale);
+  WM_operatortype_append(SCREEN_OT_region_toggle);
   WM_operatortype_append(SCREEN_OT_region_flip);
   WM_operatortype_append(SCREEN_OT_header_toggle_menus);
   WM_operatortype_append(SCREEN_OT_region_context_menu);
@@ -5241,6 +5431,7 @@ void ED_operatortypes_screen(void)
   WM_operatortype_append(SCREEN_OT_screenshot);
   WM_operatortype_append(SCREEN_OT_userpref_show);
   WM_operatortype_append(SCREEN_OT_drivers_editor_show);
+  WM_operatortype_append(SCREEN_OT_info_log_show);
   WM_operatortype_append(SCREEN_OT_region_blend);
   WM_operatortype_append(SCREEN_OT_space_type_set_or_cycle);
   WM_operatortype_append(SCREEN_OT_space_context_cycle);
diff --git a/source/blender/editors/sculpt_paint/paint_cursor.c b/source/blender/editors/sculpt_paint/paint_cursor.c
index 65e10f98753..4a552fb3744 100644
--- a/source/blender/editors/sculpt_paint/paint_cursor.c
+++ b/source/blender/editors/sculpt_paint/paint_cursor.c
@@ -45,13 +45,17 @@
 #include "BKE_node.h"
 #include "BKE_paint.h"
 #include "BKE_colortools.h"
+#include "BKE_object.h"
 
 #include "WM_api.h"
+#include "wm_cursors.h"
 
 #include "IMB_imbuf_types.h"
 
 #include "ED_view3d.h"
 
+#include "DEG_depsgraph.h"
+
 #include "GPU_draw.h"
 #include "GPU_immediate.h"
 #include "GPU_immediate_util.h"
@@ -602,7 +606,7 @@ static bool sculpt_get_brush_geometry(bContext *C,
 
 /* Draw an overlay that shows what effect the brush's texture will
  * have on brush strength */
-static void paint_draw_tex_overlay(UnifiedPaintSettings *ups,
+static bool paint_draw_tex_overlay(UnifiedPaintSettings *ups,
                                    Brush *brush,
                                    ViewContext *vc,
                                    int x,
@@ -622,7 +626,7 @@ static void paint_draw_tex_overlay(UnifiedPaintSettings *ups,
   if (!(mtex->tex) ||
       !((mtex->brush_map_mode == MTEX_MAP_MODE_STENCIL) ||
         (valid && ELEM(mtex->brush_map_mode, MTEX_MAP_MODE_VIEW, MTEX_MAP_MODE_TILED)))) {
-    return;
+    return false;
   }
 
   if (load_tex(brush, vc, zoom, col, primary)) {
@@ -728,18 +732,19 @@ static void paint_draw_tex_overlay(UnifiedPaintSettings *ups,
       GPU_matrix_pop();
     }
   }
+  return true;
 }
 
 /* Draw an overlay that shows what effect the brush's texture will
  * have on brush strength */
-static void paint_draw_cursor_overlay(
+static bool paint_draw_cursor_overlay(
     UnifiedPaintSettings *ups, Brush *brush, ViewContext *vc, int x, int y, float zoom)
 {
   rctf quad;
   /* check for overlay mode */
 
   if (!(brush->overlay_flags & BRUSH_OVERLAY_CURSOR)) {
-    return;
+    return false;
   }
 
   if (load_tex_cursor(brush, vc, zoom)) {
@@ -811,9 +816,10 @@ static void paint_draw_cursor_overlay(
       GPU_matrix_pop();
     }
   }
+  return true;
 }
 
-static void paint_draw_alpha_overlay(UnifiedPaintSettings *ups,
+static bool paint_draw_alpha_overlay(UnifiedPaintSettings *ups,
                                      Brush *brush,
                                      ViewContext *vc,
                                      int x,
@@ -824,6 +830,9 @@ static void paint_draw_alpha_overlay(UnifiedPaintSettings *ups,
   /* Color means that primary brush texture is colored and
    * secondary is used for alpha/mask control. */
   bool col = ELEM(mode, PAINT_MODE_TEXTURE_3D, PAINT_MODE_TEXTURE_2D, PAINT_MODE_VERTEX);
+
+  bool alpha_overlay_active = false;
+
   eOverlayControlFlags flags = BKE_paint_get_overlay_flags();
   gpuPushAttr(GPU_DEPTH_BUFFER_BIT | GPU_BLEND_BIT);
 
@@ -836,26 +845,28 @@ static void paint_draw_alpha_overlay(UnifiedPaintSettings *ups,
   /* Colored overlay should be drawn separately. */
   if (col) {
     if (!(flags & PAINT_OVERLAY_OVERRIDE_PRIMARY)) {
-      paint_draw_tex_overlay(ups, brush, vc, x, y, zoom, true, true);
+      alpha_overlay_active = paint_draw_tex_overlay(ups, brush, vc, x, y, zoom, true, true);
     }
     if (!(flags & PAINT_OVERLAY_OVERRIDE_SECONDARY)) {
-      paint_draw_tex_overlay(ups, brush, vc, x, y, zoom, false, false);
+      alpha_overlay_active = paint_draw_tex_overlay(ups, brush, vc, x, y, zoom, false, false);
     }
     if (!(flags & PAINT_OVERLAY_OVERRIDE_CURSOR)) {
-      paint_draw_cursor_overlay(ups, brush, vc, x, y, zoom);
+      alpha_overlay_active = paint_draw_cursor_overlay(ups, brush, vc, x, y, zoom);
     }
   }
   else {
     if (!(flags & PAINT_OVERLAY_OVERRIDE_PRIMARY) && (mode != PAINT_MODE_WEIGHT)) {
-      paint_draw_tex_overlay(ups, brush, vc, x, y, zoom, false, true);
+      alpha_overlay_active = paint_draw_tex_overlay(ups, brush, vc, x, y, zoom, false, true);
     }
     if (!(flags & PAINT_OVERLAY_OVERRIDE_CURSOR)) {
-      paint_draw_cursor_overlay(ups, brush, vc, x, y, zoom);
+      alpha_overlay_active = paint_draw_cursor_overlay(ups, brush, vc, x, y, zoom);
     }
   }
 
   GPU_matrix_pop();
   gpuPopAttr();
+
+  return alpha_overlay_active;
 }
 
 BLI_INLINE void draw_tri_point(
@@ -1074,6 +1085,98 @@ static bool ommit_cursor_drawing(Paint *paint, ePaintMode mode, Brush *brush)
   return true;
 }
 
+static void cursor_draw_point_screen_space(const uint gpuattr,
+                                           const ARegion *ar,
+                                           float true_location[3],
+                                           float obmat[4][4])
+{
+  float translation_vertex_cursor[3], location[3];
+  copy_v3_v3(location, true_location);
+  mul_m4_v3(obmat, location);
+  ED_view3d_project(ar, location, translation_vertex_cursor);
+  imm_draw_circle_fill_3d(
+      gpuattr, translation_vertex_cursor[0], translation_vertex_cursor[1], 3, 10);
+}
+
+static void cursor_draw_tiling_preview(const uint gpuattr,
+                                       const ARegion *ar,
+                                       float true_location[3],
+                                       Sculpt *sd,
+                                       Object *ob,
+                                       float radius)
+{
+  BoundBox *bb = BKE_object_boundbox_get(ob);
+  float orgLoc[3], location[3];
+  int dim, tile_pass = 0;
+  int start[3];
+  int end[3];
+  int cur[3];
+  const float *bbMin = bb->vec[0];
+  const float *bbMax = bb->vec[6];
+  const float *step = sd->paint.tile_offset;
+
+  copy_v3_v3(orgLoc, true_location);
+  for (dim = 0; dim < 3; ++dim) {
+    if ((sd->paint.symmetry_flags & (PAINT_TILE_X << dim)) && step[dim] > 0) {
+      start[dim] = (bbMin[dim] - orgLoc[dim] - radius) / step[dim];
+      end[dim] = (bbMax[dim] - orgLoc[dim] + radius) / step[dim];
+    }
+    else
+      start[dim] = end[dim] = 0;
+  }
+  copy_v3_v3_int(cur, start);
+  for (cur[0] = start[0]; cur[0] <= end[0]; cur[0]++) {
+    for (cur[1] = start[1]; cur[1] <= end[1]; cur[1]++) {
+      for (cur[2] = start[2]; cur[2] <= end[2]; cur[2]++) {
+        if (!cur[0] && !cur[1] && !cur[2])
+          continue; /* skip tile at orgLoc, this was already handled before all others */
+        tile_pass++;
+        for (dim = 0; dim < 3; dim++) {
+          location[dim] = cur[dim] * step[dim] + orgLoc[dim];
+        }
+        cursor_draw_point_screen_space(gpuattr, ar, location, ob->obmat);
+      }
+    }
+  }
+}
+
+static void cursor_draw_point_with_symmetry(const uint gpuattr,
+                                            const ARegion *ar,
+                                            const float true_location[3],
+                                            Sculpt *sd,
+                                            Object *ob,
+                                            float radius)
+{
+  const char symm = sd->paint.symmetry_flags & PAINT_SYMM_AXIS_ALL;
+  float location[3], symm_rot_mat[4][4];
+
+  for (int i = 0; i <= symm; ++i) {
+    if (i == 0 || (symm & i && (symm != 5 || i != 3) && (symm != 6 || (i != 3 && i != 5)))) {
+
+      /* Axis Symmetry */
+      flip_v3_v3(location, true_location, (char)i);
+      cursor_draw_point_screen_space(gpuattr, ar, location, ob->obmat);
+
+      /* Tiling */
+      cursor_draw_tiling_preview(gpuattr, ar, location, sd, ob, radius);
+
+      /* Radial Symmetry */
+      for (char raxis = 0; raxis < 3; raxis++) {
+        for (int r = 1; r < sd->radial_symm[raxis]; r++) {
+          float angle = 2 * M_PI * r / sd->radial_symm[(int)raxis];
+          flip_v3_v3(location, true_location, (char)i);
+          unit_m4(symm_rot_mat);
+          rotate_m4(symm_rot_mat, raxis + 'X', angle);
+          mul_m4_v3(symm_rot_mat, location);
+
+          cursor_draw_tiling_preview(gpuattr, ar, location, sd, ob, radius);
+          cursor_draw_point_screen_space(gpuattr, ar, location, ob->obmat);
+        }
+      }
+    }
+  }
+}
+
 static void paint_draw_cursor(bContext *C, int x, int y, void *UNUSED(unused))
 {
   Scene *scene = CTX_data_scene(C);
@@ -1121,7 +1224,7 @@ static void paint_draw_cursor(bContext *C, int x, int y, void *UNUSED(unused))
   }
 
   /* draw overlay */
-  paint_draw_alpha_overlay(ups, brush, &vc, x, y, zoomx, mode);
+  bool alpha_overlay_active = paint_draw_alpha_overlay(ups, brush, &vc, x, y, zoomx, mode);
 
   /* TODO: as sculpt and other paint modes are unified, this
    * special mode of drawing will go away */
@@ -1158,12 +1261,12 @@ static void paint_draw_cursor(bContext *C, int x, int y, void *UNUSED(unused))
   }
 
   /* make lines pretty */
-  GPU_line_width(1.0f);
+  GPU_line_width(2.0f);
   GPU_blend(true); /* TODO: also set blend mode? */
   GPU_line_smooth(true);
 
-  uint pos = GPU_vertformat_attr_add(immVertexFormat(), "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
-  immBindBuiltinProgram(GPU_SHADER_2D_UNIFORM_COLOR);
+  uint pos = GPU_vertformat_attr_add(immVertexFormat(), "pos", GPU_COMP_F32, 3, GPU_FETCH_FLOAT);
+  immBindBuiltinProgram(GPU_SHADER_3D_UNIFORM_COLOR);
 
   /* set brush color */
   immUniformColor3fvAlpha(outline_col, outline_alpha);
@@ -1176,7 +1279,103 @@ static void paint_draw_cursor(bContext *C, int x, int y, void *UNUSED(unused))
     /* outer at half alpha */
     immUniformColor3fvAlpha(outline_col, outline_alpha * 0.5f);
   }
-  imm_draw_circle_wire_2d(pos, translation[0], translation[1], final_radius, 40);
+
+  /* Only sculpt mode cursor for now */
+
+  /* Disable for PBVH_GRIDS */
+  SculptSession *ss = vc.obact->sculpt;
+  bool is_multires = ss && ss->pbvh && BKE_pbvh_type(ss->pbvh) == PBVH_GRIDS;
+
+  if ((mode == PAINT_MODE_SCULPT) && ss && !is_multires &&
+      !(brush->falloff_shape & BRUSH_AIRBRUSH)) {
+    Sculpt *sd = CTX_data_tool_settings(C)->sculpt;
+    wmWindow *win = CTX_wm_window(C);
+
+    /* Update WM mouse cursor, disable when the 3D brush cursor is enabled */
+    if (sd->paint.brush->overlay_flags & BRUSH_OVERLAY_CURSOR) {
+      WM_cursor_set(win, CURSOR_STD);
+    }
+    else {
+      WM_cursor_set(win, CURSOR_EDIT);
+    }
+
+    if (!ups->stroke_active) {
+      SculptCursorGeometryInfo gi;
+      float mouse[2] = {x - ar->winrct.xmin, y - ar->winrct.ymin};
+      if (sculpt_cursor_geometry_info_update(C, &gi, mouse, true) && !alpha_overlay_active) {
+
+        float rds;
+        if (!BKE_brush_use_locked_size(scene, brush)) {
+          rds = paint_calc_object_space_radius(&vc, gi.location, BKE_brush_size_get(scene, brush));
+        }
+        else {
+          rds = BKE_brush_unprojected_radius_get(scene, brush);
+        }
+
+        wmViewport(&ar->winrct);
+
+        /* Draw 3D active vertex preview with symmetry*/
+        if (len_v3v3(gi.active_vertex_co, gi.location) < rds) {
+          cursor_draw_point_with_symmetry(pos, ar, gi.active_vertex_co, sd, vc.obact, rds);
+        }
+
+        /* Draw 3D brush cursor */
+        GPU_matrix_push_projection();
+        ED_view3d_draw_setup_view(CTX_wm_window(C),
+                                  CTX_data_depsgraph_pointer(C),
+                                  CTX_data_scene(C),
+                                  ar,
+                                  CTX_wm_view3d(C),
+                                  NULL,
+                                  NULL,
+                                  NULL);
+
+        float cursor_trans[4][4], cursor_rot[4][4];
+        float z_axis[4] = {0.0f, 0.0f, 1.0f, 0.0f};
+        float quat[4];
+
+        copy_m4_m4(cursor_trans, vc.obact->obmat);
+        translate_m4(cursor_trans, gi.location[0], gi.location[1], gi.location[2]);
+        rotation_between_vecs_to_quat(quat, z_axis, gi.normal);
+        quat_to_mat4(cursor_rot, quat);
+
+        GPU_matrix_push();
+        GPU_matrix_mul(cursor_trans);
+        GPU_matrix_mul(cursor_rot);
+        imm_draw_circle_wire_3d(pos, 0, 0, rds, 40);
+        GPU_matrix_pop();
+
+        GPU_matrix_pop_projection();
+
+        wmWindowViewport(win);
+      }
+      else {
+        /* Draw default cursor when the mouse is not over the mesh or there are no supported
+         * overlays active */
+        GPU_line_width(1.0f);
+        imm_draw_circle_wire_3d(pos, translation[0], translation[1], final_radius, 40);
+      }
+    }
+    else {
+      if (vc.obact->sculpt->cache && !vc.obact->sculpt->cache->first_time) {
+        /* Draw cursor location preview when the stroke is active using the data from StrokeCache
+         */
+        float cursor_location[3];
+        wmViewport(&ar->winrct);
+        copy_v3_v3(cursor_location, ss->cache->true_location);
+        if (ss->cache->brush->sculpt_tool == SCULPT_TOOL_GRAB) {
+          add_v3_v3(cursor_location, ss->cache->grab_delta);
+        }
+        cursor_draw_point_with_symmetry(pos, ar, cursor_location, sd, vc.obact, ss->cache->radius);
+        wmWindowViewport(win);
+      }
+    }
+  }
+  else {
+    /* Draw default cursor in unsupported modes */
+    GPU_line_width(1.0f);
+    imm_draw_circle_wire_3d(pos, translation[0], translation[1], final_radius, 40);
+  }
 
   immUnbindProgram();
 
diff --git a/source/blender/editors/sculpt_paint/paint_image_proj.c b/source/blender/editors/sculpt_paint/paint_image_proj.c
index 342d0b6e820..397b2981ace 100644
--- a/source/blender/editors/sculpt_paint/paint_image_proj.c
+++ b/source/blender/editors/sculpt_paint/paint_image_proj.c
@@ -1719,9 +1719,9 @@ static float project_paint_uvpixel_mask(const ProjPaintState *ps,
       normalize_v3(no);
     }
     else {
-      /* incase the */
 #if 1
-      /* normalizing per pixel isn't optimal, we could cache or check ps->*/
+      /* In case the normalizing per pixel isn't optimal,
+       * we could cache or access from evaluated mesh. */
       normal_tri_v3(no,
                     ps->mvert_eval[lt_vtri[0]].co,
                     ps->mvert_eval[lt_vtri[1]].co,
diff --git a/source/blender/editors/sculpt_paint/paint_intern.h b/source/blender/editors/sculpt_paint/paint_intern.h
index ea4814857ba..5efedf69fe4 100644
--- a/source/blender/editors/sculpt_paint/paint_intern.h
+++ b/source/blender/editors/sculpt_paint/paint_intern.h
@@ -33,7 +33,6 @@ struct ListBase;
 struct MTex;
 struct Object;
 struct Paint;
-struct PaintCurve;
 struct PaintStroke;
 struct PointerRNA;
 struct RegionView3D;
diff --git a/source/blender/editors/sculpt_paint/paint_stroke.c b/source/blender/editors/sculpt_paint/paint_stroke.c
index 694dae49d30..b4b1ae3b0af 100644
--- a/source/blender/editors/sculpt_paint/paint_stroke.c
+++ b/source/blender/editors/sculpt_paint/paint_stroke.c
@@ -57,6 +57,7 @@
 #include "IMB_imbuf_types.h"
 
 #include "paint_intern.h"
+#include "sculpt_intern.h"
 
 #include <float.h>
 #include <math.h>
@@ -93,6 +94,8 @@ typedef struct PaintStroke {
   int cur_sample;
 
   float last_mouse_position[2];
+  float last_scene_space_position[3];
+  bool stroke_over_mesh;
   /* space distance covered so far */
   float stroke_distance;
 
@@ -233,6 +236,17 @@ static bool paint_tool_require_location(Brush *brush, ePaintMode mode)
   return true;
 }
 
+static bool paint_stroke_use_scene_spacing(Brush *brush, ePaintMode mode)
+{
+  switch (mode) {
+    case PAINT_MODE_SCULPT:
+      return brush->flag & BRUSH_SCENE_SPACING;
+    default:
+      break;
+  }
+  return false;
+}
+
 static bool paint_tool_require_inbetween_mouse_events(Brush *brush, ePaintMode mode)
 {
   switch (mode) {
@@ -523,6 +537,10 @@ static void paint_brush_stroke_add_step(bContext *C,
   copy_v2_v2(stroke->last_mouse_position, mouse_in);
   stroke->last_pressure = pressure;
 
+  if (paint_stroke_use_scene_spacing(brush, mode)) {
+    sculpt_stroke_get_location(C, stroke->last_scene_space_position, stroke->last_mouse_position);
+  }
+
   if (paint_stroke_use_jitter(mode, brush, stroke->stroke_mode == BRUSH_STROKE_INVERT)) {
     float delta[2];
     float factor = stroke->zoom_2d;
@@ -600,14 +618,32 @@ static bool paint_smooth_stroke(PaintStroke *stroke,
   return true;
 }
 
-static float paint_space_stroke_spacing(const Scene *scene,
+static float paint_space_stroke_spacing(bContext *C,
+                                        const Scene *scene,
                                         PaintStroke *stroke,
                                         float size_pressure,
                                         float spacing_pressure)
 {
-  /* brushes can have a minimum size of 1.0 but with pressure it can be smaller then a pixel
-   * causing very high step sizes, hanging blender [#32381] */
-  const float size_clamp = max_ff(1.0f, BKE_brush_size_get(scene, stroke->brush) * size_pressure);
+  Paint *paint = BKE_paint_get_active_from_context(C);
+  ePaintMode mode = BKE_paintmode_get_active_from_context(C);
+  Brush *brush = BKE_paint_brush(paint);
+  float size_clamp = 0.0f;
+  float size = BKE_brush_size_get(scene, stroke->brush) * size_pressure;
+  if (paint_stroke_use_scene_spacing(brush, mode)) {
+    if (!BKE_brush_use_locked_size(scene, brush)) {
+      size_clamp = paint_calc_object_space_radius(
+          &stroke->vc, stroke->last_scene_space_position, size);
+    }
+    else {
+      size_clamp = BKE_brush_unprojected_radius_get(scene, brush) * size_pressure;
+    }
+  }
+  else {
+    /* brushes can have a minimum size of 1.0 but with pressure it can be smaller then a pixel
+     * causing very high step sizes, hanging blender [#32381] */
+    size_clamp = max_ff(1.0f, size);
+  }
+
   float spacing = stroke->brush->spacing;
 
   /* apply spacing pressure */
@@ -619,7 +655,12 @@ static float paint_space_stroke_spacing(const Scene *scene,
    * the fact that brush can be scaled there. */
   spacing *= stroke->zoom_2d;
 
-  return max_ff(1.0, size_clamp * spacing / 50.0f);
+  if (paint_stroke_use_scene_spacing(brush, mode)) {
+    return size_clamp * spacing / 50.0f;
+  }
+  else {
+    return max_ff(1.0, size_clamp * spacing / 50.0f);
+  }
 }
 
 static float paint_stroke_overlapped_curve(Brush *br, float x, float spacing)
@@ -677,14 +718,18 @@ static float paint_stroke_integrate_overlap(Brush *br, float factor)
   }
 }
 
-static float paint_space_stroke_spacing_variable(
-    const Scene *scene, PaintStroke *stroke, float pressure, float dpressure, float length)
+static float paint_space_stroke_spacing_variable(bContext *C,
+                                                 const Scene *scene,
+                                                 PaintStroke *stroke,
+                                                 float pressure,
+                                                 float dpressure,
+                                                 float length)
 {
   if (BKE_brush_use_size_pressure(scene, stroke->brush)) {
     /* use pressure to modify size. set spacing so that at 100%, the circles
      * are aligned nicely with no overlap. for this the spacing needs to be
      * the average of the previous and next size. */
-    float s = paint_space_stroke_spacing(scene, stroke, 1.0f, pressure);
+    float s = paint_space_stroke_spacing(C, scene, stroke, 1.0f, pressure);
     float q = s * dpressure / (2.0f * length);
     float pressure_fac = (1.0f + q) / (1.0f - q);
 
@@ -692,14 +737,15 @@ static float paint_space_stroke_spacing_variable(
     float new_size_pressure = stroke->last_pressure * pressure_fac;
 
     /* average spacing */
-    float last_spacing = paint_space_stroke_spacing(scene, stroke, last_size_pressure, pressure);
-    float new_spacing = paint_space_stroke_spacing(scene, stroke, new_size_pressure, pressure);
+    float last_spacing = paint_space_stroke_spacing(
+        C, scene, stroke, last_size_pressure, pressure);
+    float new_spacing = paint_space_stroke_spacing(C, scene, stroke, new_size_pressure, pressure);
 
     return 0.5f * (last_spacing + new_spacing);
   }
   else {
     /* no size pressure */
-    return paint_space_stroke_spacing(scene, stroke, 1.0f, pressure);
+    return paint_space_stroke_spacing(C, scene, stroke, 1.0f, pressure);
   }
 }
 
@@ -711,29 +757,57 @@ static int paint_space_stroke(bContext *C,
                               float final_pressure)
 {
   const Scene *scene = CTX_data_scene(C);
+  ARegion *ar = CTX_wm_region(C);
   PaintStroke *stroke = op->customdata;
   UnifiedPaintSettings *ups = stroke->ups;
+  Paint *paint = BKE_paint_get_active_from_context(C);
+  ePaintMode mode = BKE_paintmode_get_active_from_context(C);
+  Brush *brush = BKE_paint_brush(paint);
   int cnt = 0;
 
   float pressure, dpressure;
   float mouse[2], dmouse[2];
+  float scene_space_position[3], d_scene_space_position[3], final_scene_space_position[3];
   float length;
-  float no_pressure_spacing = paint_space_stroke_spacing(scene, stroke, 1.0f, 1.0f);
-
-  sub_v2_v2v2(dmouse, final_mouse, stroke->last_mouse_position);
-
+  float no_pressure_spacing = paint_space_stroke_spacing(C, scene, stroke, 1.0f, 1.0f);
   pressure = stroke->last_pressure;
   dpressure = final_pressure - stroke->last_pressure;
-
+  sub_v2_v2v2(dmouse, final_mouse, stroke->last_mouse_position);
   length = normalize_v2(dmouse);
 
+  if (paint_stroke_use_scene_spacing(brush, mode)) {
+    bool hit = sculpt_stroke_get_location(C, scene_space_position, final_mouse);
+    if (hit && stroke->stroke_over_mesh) {
+      sub_v3_v3v3(d_scene_space_position, scene_space_position, stroke->last_scene_space_position);
+      length = len_v3(d_scene_space_position);
+      stroke->stroke_over_mesh = true;
+    }
+    else {
+      length = 0.0f;
+      stroke->stroke_over_mesh = hit;
+      if (stroke->stroke_over_mesh) {
+        copy_v3_v3(stroke->last_scene_space_position, scene_space_position);
+      }
+    }
+  }
+
   while (length > 0.0f) {
     float spacing = paint_space_stroke_spacing_variable(
-        scene, stroke, pressure, dpressure, length);
+        C, scene, stroke, pressure, dpressure, length);
 
     if (length >= spacing) {
-      mouse[0] = stroke->last_mouse_position[0] + dmouse[0] * spacing;
-      mouse[1] = stroke->last_mouse_position[1] + dmouse[1] * spacing;
+      if (paint_stroke_use_scene_spacing(brush, mode)) {
+        normalize_v3(d_scene_space_position);
+        mul_v3_v3fl(final_scene_space_position, d_scene_space_position, spacing);
+        add_v3_v3v3(final_scene_space_position,
+                    stroke->last_scene_space_position,
+                    final_scene_space_position);
+        ED_view3d_project(ar, final_scene_space_position, mouse);
+      }
+      else {
+        mouse[0] = stroke->last_mouse_position[0] + dmouse[0] * spacing;
+        mouse[1] = stroke->last_mouse_position[1] + dmouse[1] * spacing;
+      }
       pressure = stroke->last_pressure + (spacing / length) * dpressure;
 
       ups->overlap_factor = paint_stroke_integrate_overlap(stroke->brush,
@@ -1079,7 +1153,7 @@ static bool paint_stroke_curve_end(bContext *C, wmOperator *op, PaintStroke *str
   if (br->flag & BRUSH_CURVE) {
     UnifiedPaintSettings *ups = &CTX_data_tool_settings(C)->unified_paint_settings;
     const Scene *scene = CTX_data_scene(C);
-    const float spacing = paint_space_stroke_spacing(scene, stroke, 1.0f, 1.0f);
+    const float spacing = paint_space_stroke_spacing(C, scene, stroke, 1.0f, 1.0f);
     PaintCurve *pc = br->paint_curve;
     PaintCurvePoint *pcp;
     float length_residue = 0.0f;
@@ -1250,6 +1324,10 @@ int paint_stroke_modal(bContext *C, wmOperator *op, const wmEvent *event)
   if (!stroke->stroke_started) {
     stroke->last_pressure = sample_average.pressure;
     copy_v2_v2(stroke->last_mouse_position, sample_average.mouse);
+    if (paint_stroke_use_scene_spacing(br, mode)) {
+      stroke->stroke_over_mesh = sculpt_stroke_get_location(
+          C, stroke->last_scene_space_position, sample_average.mouse);
+    }
     stroke->stroke_started = stroke->test_start(C, op, sample_average.mouse);
     BLI_assert((stroke->stroke_started & ~1) == 0); /* 0/1 */
 
diff --git a/source/blender/editors/sculpt_paint/sculpt.c b/source/blender/editors/sculpt_paint/sculpt.c
index 440c4d42cae..5aa913ad006 100644
--- a/source/blender/editors/sculpt_paint/sculpt.c
+++ b/source/blender/editors/sculpt_paint/sculpt.c
@@ -96,7 +96,18 @@
 
 /* Do not use these functions while working with PBVH_GRIDS data in SculptSession */
 
-/* TODO: why is this kept, should it be removed? */
+static float *sculpt_vertex_co_get(SculptSession *ss, int index)
+{
+  switch (BKE_pbvh_type(ss->pbvh)) {
+    case PBVH_FACES:
+      return ss->mvert[index].co;
+    case PBVH_BMESH:
+      return BM_vert_at_index(BKE_pbvh_get_bmesh(ss->pbvh), index)->co;
+    default:
+      return NULL;
+  }
+}
+
 #if 0 /* UNUSED */
 
 static int sculpt_active_vertex_get(SculptSession *ss)
@@ -136,18 +147,6 @@ static void sculpt_vertex_normal_get(SculptSession *ss, int index, float no[3])
   }
 }
 
-static float *sculpt_vertex_co_get(SculptSession *ss, int index)
-{
-  switch (BKE_pbvh_type(ss->pbvh)) {
-    case PBVH_FACES:
-      return ss->mvert[index].co;
-    case PBVH_BMESH:
-      return BM_vert_at_index(BKE_pbvh_get_bmesh(ss->pbvh), index)->co;
-    default:
-      return NULL;
-  }
-}
-
 static void sculpt_vertex_co_set(SculptSession *ss, int index, float co[3])
 {
   switch (BKE_pbvh_type(ss->pbvh)) {
@@ -748,17 +747,26 @@ void ED_sculpt_redraw_planes_get(float planes[4][4], ARegion *ar, Object *ob)
 
 void sculpt_brush_test_init(SculptSession *ss, SculptBrushTest *test)
 {
-  RegionView3D *rv3d = ss->cache->vc->rv3d;
+  RegionView3D *rv3d = ss->cache ? ss->cache->vc->rv3d : ss->rv3d;
+
+  test->radius_squared = ss->cache ? ss->cache->radius_squared :
+                                     ss->cursor_radius * ss->cursor_radius;
+  if (ss->cache) {
+    copy_v3_v3(test->location, ss->cache->location);
+    test->mirror_symmetry_pass = ss->cache->mirror_symmetry_pass;
+  }
+  else {
+    copy_v3_v3(test->location, ss->cursor_location);
+    test->mirror_symmetry_pass = 0;
+  }
 
-  test->radius_squared = ss->cache->radius_squared;
-  copy_v3_v3(test->location, ss->cache->location);
   test->dist = 0.0f; /* just for initialize */
 
   /* Only for 2D projection. */
   zero_v4(test->plane_view);
   zero_v4(test->plane_tool);
 
-  test->mirror_symmetry_pass = ss->cache->mirror_symmetry_pass;
+  test->mirror_symmetry_pass = ss->cache ? ss->cache->mirror_symmetry_pass : 0;
 
   if (rv3d->rflag & RV3D_CLIPPING) {
     test->clip_rv3d = rv3d;
@@ -1050,7 +1058,7 @@ static void calc_area_normal_and_center_task_cb(void *__restrict userdata,
   int private_count[2] = {0};
   bool use_original = false;
 
-  if (ss->cache->original) {
+  if (ss->cache && ss->cache->original) {
     unode = sculpt_undo_push_node(data->ob, data->nodes[n], SCULPT_UNDO_COORDS);
     use_original = (unode->co || unode->bm_entry);
   }
@@ -1059,6 +1067,13 @@ static void calc_area_normal_and_center_task_cb(void *__restrict userdata,
   SculptBrushTestFn sculpt_brush_test_sq_fn = sculpt_brush_test_init_with_falloff_shape(
       ss, &test, data->brush->falloff_shape);
 
+  /* Update the test radius to sample the normal using the normal radius of the brush */
+  if (data->brush->ob_mode == OB_MODE_SCULPT) {
+    float test_radius = sqrtf(test.radius_squared);
+    test_radius *= data->brush->normal_radius_factor;
+    test.radius_squared = test_radius * test_radius;
+  }
+
   /* when the mesh is edited we can't rely on original coords
    * (original mesh may not even have verts in brush radius) */
   if (use_original && data->has_bm_orco) {
@@ -1120,6 +1135,8 @@ static void calc_area_normal_and_center_task_cb(void *__restrict userdata,
         const float *no;
         int flip_index;
 
+        data->any_vertex_sampled = true;
+
         if (use_original) {
           normal_short_to_float_v3(no_buf, no_s);
           no = no_buf;
@@ -1134,7 +1151,8 @@ static void calc_area_normal_and_center_task_cb(void *__restrict userdata,
           }
         }
 
-        flip_index = (dot_v3v3(ss->cache->view_normal, no) <= 0.0f);
+        flip_index = (dot_v3v3(ss->cache ? ss->cache->view_normal : ss->cursor_view_normal, no) <=
+                      0.0f);
         if (area_cos) {
           add_v3_v3(private_co[flip_index], co);
         }
@@ -1223,7 +1241,7 @@ static void calc_area_normal(
 }
 
 /* expose 'calc_area_normal' externally. */
-void sculpt_pbvh_calc_area_normal(const Brush *brush,
+bool sculpt_pbvh_calc_area_normal(const Brush *brush,
                                   Object *ob,
                                   PBVHNode **nodes,
                                   int totnode,
@@ -1249,6 +1267,7 @@ void sculpt_pbvh_calc_area_normal(const Brush *brush,
       .area_cos = NULL,
       .area_nos = area_nos,
       .count = count,
+      .any_vertex_sampled = false,
   };
   BLI_mutex_init(&data.mutex);
 
@@ -1265,6 +1284,8 @@ void sculpt_pbvh_calc_area_normal(const Brush *brush,
       break;
     }
   }
+
+  return data.any_vertex_sampled;
 }
 
 /* this calculates flatten center and area normal together,
@@ -1508,7 +1529,8 @@ float tex_strength(SculptSession *ss,
 bool sculpt_search_sphere_cb(PBVHNode *node, void *data_v)
 {
   SculptSearchSphereData *data = data_v;
-  float *center = data->ss->cache->location, nearest[3];
+  float *center, nearest[3];
+  center = data->ss->cache ? data->ss->cache->location : data->ss->cursor_location;
   float t[3], bb_min[3], bb_max[3];
   int i;
 
@@ -1585,12 +1607,13 @@ static PBVHNode **sculpt_pbvh_gather_generic(Object *ob,
   SculptSession *ss = ob->sculpt;
   PBVHNode **nodes = NULL;
 
-  /* Build a list of all nodes that are potentially within the brush's area of influence */
+  /* Build a list of all nodes that are potentially within the cursor or brush's area of influence
+   */
   if (brush->falloff_shape == PAINT_FALLOFF_SHAPE_SPHERE) {
     SculptSearchSphereData data = {
         .ss = ss,
         .sd = sd,
-        .radius_squared = SQUARE(ss->cache->radius * radius_scale),
+        .radius_squared = ss->cache ? SQUARE(ss->cache->radius * radius_scale) : ss->cursor_radius,
         .original = use_original,
     };
     BKE_pbvh_search_gather(ss->pbvh, sculpt_search_sphere_cb, &data, &nodes, r_totnode);
@@ -1602,7 +1625,7 @@ static PBVHNode **sculpt_pbvh_gather_generic(Object *ob,
     SculptSearchCircleData data = {
         .ss = ss,
         .sd = sd,
-        .radius_squared = SQUARE(ss->cache->radius * radius_scale),
+        .radius_squared = ss->cache ? SQUARE(ss->cache->radius * radius_scale) : ss->cursor_radius,
         .original = use_original,
         .dist_ray_to_aabb_precalc = &dist_ray_to_aabb_precalc,
     };
@@ -1964,10 +1987,14 @@ typedef struct SculptDoBrushSmoothGridDataChunk {
 typedef struct {
   SculptSession *ss;
   const float *ray_start;
+  const float *ray_normal;
   bool hit;
   float depth;
   bool original;
 
+  int active_vertex_index;
+  float *face_normal;
+
   struct IsectRayPrecalc isect_precalc;
 } SculptRaycastData;
 
@@ -3987,7 +4014,7 @@ static void do_gravity(Sculpt *sd, Object *ob, PBVHNode **nodes, int totnode, fl
   BLI_task_parallel_range(0, totnode, &data, do_gravity_task_cb_ex, &settings);
 }
 
-void sculpt_vertcos_to_key(Object *ob, KeyBlock *kb, float (*vertCos)[3])
+void sculpt_vertcos_to_key(Object *ob, KeyBlock *kb, const float (*vertCos)[3])
 {
   Mesh *me = (Mesh *)ob->data;
   float(*ofs)[3] = NULL;
@@ -4349,7 +4376,7 @@ static void sculpt_update_keyblock(Object *ob)
     vertCos = ss->orig_cos;
   }
   else {
-    vertCos = BKE_pbvh_get_vertCos(ss->pbvh);
+    vertCos = BKE_pbvh_vert_coords_alloc(ss->pbvh);
   }
 
   if (vertCos) {
@@ -4513,7 +4540,7 @@ static void do_tiled(
   float orgLoc[3]; /* position of the "prototype" stroke for tiling */
   copy_v3_v3(orgLoc, cache->location);
 
-  for (dim = 0; dim < 3; ++dim) {
+  for (dim = 0; dim < 3; dim++) {
     if ((sd->paint.symmetry_flags & (PAINT_TILE_X << dim)) && step[dim] > 0) {
       start[dim] = (bbMin[dim] - orgLoc[dim] - radius) / step[dim];
       end[dim] = (bbMax[dim] - orgLoc[dim] + radius) / step[dim];
@@ -4529,16 +4556,16 @@ static void do_tiled(
 
   /* now do it for all the tiles */
   copy_v3_v3_int(cur, start);
-  for (cur[0] = start[0]; cur[0] <= end[0]; ++cur[0]) {
-    for (cur[1] = start[1]; cur[1] <= end[1]; ++cur[1]) {
-      for (cur[2] = start[2]; cur[2] <= end[2]; ++cur[2]) {
+  for (cur[0] = start[0]; cur[0] <= end[0]; cur[0]++) {
+    for (cur[1] = start[1]; cur[1] <= end[1]; cur[1]++) {
+      for (cur[2] = start[2]; cur[2] <= end[2]; cur[2]++) {
         if (!cur[0] && !cur[1] && !cur[2]) {
           continue; /* skip tile at orgLoc, this was already handled before all others */
         }
 
         ++cache->tile_pass;
 
-        for (dim = 0; dim < 3; ++dim) {
+        for (dim = 0; dim < 3; dim++) {
           cache->location[dim] = cur[dim] * step[dim] + orgLoc[dim];
           cache->plane_offset[dim] = cur[dim] * step[dim];
         }
@@ -5154,10 +5181,11 @@ static bool sculpt_any_smooth_mode(const Brush *brush, StrokeCache *cache, int s
 static void sculpt_stroke_modifiers_check(const bContext *C, Object *ob, const Brush *brush)
 {
   SculptSession *ss = ob->sculpt;
+  View3D *v3d = CTX_wm_view3d(C);
 
-  if (ss->kb || ss->modifiers_active) {
+  bool need_pmap = sculpt_any_smooth_mode(brush, ss->cache, 0);
+  if (ss->kb || ss->modifiers_active || (!BKE_sculptsession_use_pbvh_draw(ob, v3d) && need_pmap)) {
     Depsgraph *depsgraph = CTX_data_depsgraph_pointer(C);
-    bool need_pmap = sculpt_any_smooth_mode(brush, ss->cache, 0);
     BKE_sculpt_update_object_for_edit(depsgraph, ob, need_pmap, false);
   }
 }
@@ -5186,8 +5214,11 @@ static void sculpt_raycast_cb(PBVHNode *node, void *data_v, float *tmin)
                               origco,
                               use_origco,
                               srd->ray_start,
+                              srd->ray_normal,
                               &srd->isect_precalc,
-                              &srd->depth)) {
+                              &srd->depth,
+                              &srd->active_vertex_index,
+                              srd->face_normal)) {
       srd->hit = 1;
       *tmin = srd->depth;
     }
@@ -5275,6 +5306,120 @@ static float sculpt_raycast_init(ViewContext *vc,
   return dist;
 }
 
+/* Gets the normal, location and active vertex location of the geometry under the cursor. This also
+ * updates
+ * the active vertex and cursor related data of the SculptSession using the mouse position */
+bool sculpt_cursor_geometry_info_update(bContext *C,
+                                        SculptCursorGeometryInfo *out,
+                                        const float mouse[2],
+                                        bool use_sampled_normal)
+{
+  Scene *scene = CTX_data_scene(C);
+  Sculpt *sd = scene->toolsettings->sculpt;
+  Object *ob;
+  SculptSession *ss;
+  ViewContext vc;
+  const Brush *brush = BKE_paint_brush(BKE_paint_get_active_from_context(C));
+  float ray_start[3], ray_end[3], ray_normal[3], depth, face_normal[3], sampled_normal[3],
+      mat[3][3];
+  float viewDir[3] = {0.0f, 0.0f, 1.0f};
+  int totnode;
+  bool original = false, hit = false;
+
+  ED_view3d_viewcontext_init(C, &vc);
+
+  ob = vc.obact;
+  ss = ob->sculpt;
+
+  if (!ss->pbvh) {
+    copy_v3_fl(out->location, 0.0f);
+    copy_v3_fl(out->normal, 0.0f);
+    copy_v3_fl(out->active_vertex_co, 0.0f);
+    return false;
+  }
+
+  /* PBVH raycast to get active vertex and face normal */
+  depth = sculpt_raycast_init(&vc, mouse, ray_start, ray_end, ray_normal, original);
+  sculpt_stroke_modifiers_check(C, ob, brush);
+
+  SculptRaycastData srd = {
+      .original = original,
+      .ss = ob->sculpt,
+      .hit = 0,
+      .ray_start = ray_start,
+      .ray_normal = ray_normal,
+      .depth = depth,
+      .face_normal = face_normal,
+  };
+  isect_ray_tri_watertight_v3_precalc(&srd.isect_precalc, ray_normal);
+  BKE_pbvh_raycast(ss->pbvh, sculpt_raycast_cb, &srd, ray_start, ray_normal, srd.original);
+
+  /* Cursor is not over the mesh, return default values */
+  if (!srd.hit) {
+    copy_v3_fl(out->location, 0.0f);
+    copy_v3_fl(out->normal, 0.0f);
+    copy_v3_fl(out->active_vertex_co, 0.0f);
+    return false;
+  }
+
+  /* Update the active vertex of the SculptSession */
+  ss->active_vertex_index = srd.active_vertex_index;
+
+  copy_v3_v3(out->active_vertex_co, sculpt_vertex_co_get(ss, srd.active_vertex_index));
+  copy_v3_v3(out->location, ray_normal);
+  mul_v3_fl(out->location, srd.depth);
+  add_v3_v3(out->location, ray_start);
+
+  /* Option to return the face normal directly for performance o accuracy reasons */
+  if (!use_sampled_normal) {
+    copy_v3_v3(out->normal, srd.face_normal);
+    return hit;
+  }
+
+  /* Sampled normal calculation */
+  const float radius_scale = 1.0f;
+  float radius;
+
+  /* Update cursor data in SculptSession */
+  invert_m4_m4(ob->imat, ob->obmat);
+  copy_m3_m4(mat, vc.rv3d->viewinv);
+  mul_m3_v3(mat, viewDir);
+  copy_m3_m4(mat, ob->imat);
+  mul_m3_v3(mat, viewDir);
+  normalize_v3_v3(ss->cursor_view_normal, viewDir);
+  copy_v3_v3(ss->cursor_normal, srd.face_normal);
+  copy_v3_v3(ss->cursor_location, out->location);
+  ss->rv3d = vc.rv3d;
+
+  if (!BKE_brush_use_locked_size(scene, brush)) {
+    radius = paint_calc_object_space_radius(&vc, out->location, BKE_brush_size_get(scene, brush));
+  }
+  else {
+    radius = BKE_brush_unprojected_radius_get(scene, brush);
+  }
+  ss->cursor_radius = radius;
+
+  PBVHNode **nodes = sculpt_pbvh_gather_generic(ob, sd, brush, original, radius_scale, &totnode);
+
+  /* In case there are no nodes under the cursor, return the face normal */
+  if (!totnode) {
+    MEM_freeN(nodes);
+    copy_v3_v3(out->normal, srd.face_normal);
+    return true;
+  }
+
+  /* Calculate the sampled normal */
+  if (sculpt_pbvh_calc_area_normal(brush, ob, nodes, totnode, true, sampled_normal)) {
+    copy_v3_v3(out->normal, sampled_normal);
+  }
+  else {
+    /* Use face normal when there are no vertices to sample inside the cursor radius */
+    copy_v3_v3(out->normal, srd.face_normal);
+  }
+  MEM_freeN(nodes);
+  return true;
+}
+
 /* Do a raycast in the tree to find the 3d brush location
  * (This allows us to ignore the GL depth buffer)
  * Returns 0 if the ray doesn't hit the mesh, non-zero otherwise
@@ -5284,7 +5429,7 @@ bool sculpt_stroke_get_location(bContext *C, float out[3], const float mouse[2])
   Object *ob;
   SculptSession *ss;
   StrokeCache *cache;
-  float ray_start[3], ray_end[3], ray_normal[3], depth;
+  float ray_start[3], ray_end[3], ray_normal[3], depth, face_normal[3];
   bool original;
   ViewContext vc;
 
@@ -5302,14 +5447,21 @@ bool sculpt_stroke_get_location(bContext *C, float out[3], const float mouse[2])
 
   depth = sculpt_raycast_init(&vc, mouse, ray_start, ray_end, ray_normal, original);
 
+  if (BKE_pbvh_type(ss->pbvh) == PBVH_BMESH) {
+    BM_mesh_elem_table_ensure(ss->bm, BM_VERT);
+    BM_mesh_elem_index_ensure(ss->bm, BM_VERT);
+  }
+
   bool hit = false;
   {
     SculptRaycastData srd;
     srd.ss = ob->sculpt;
     srd.ray_start = ray_start;
+    srd.ray_normal = ray_normal;
     srd.hit = 0;
     srd.depth = depth;
     srd.original = original;
+    srd.face_normal = face_normal;
     isect_ray_tri_watertight_v3_precalc(&srd.isect_precalc, ray_normal);
 
     BKE_pbvh_raycast(ss->pbvh, sculpt_raycast_cb, &srd, ray_start, ray_normal, srd.original);
@@ -5577,7 +5729,6 @@ static void sculpt_stroke_update_step(bContext *C,
   }
 
   do_symmetrical_brush_actions(sd, ob, do_brush_action, ups);
-
   sculpt_combine_proxies(sd, ob);
 
   /* hack to fix noise texture tearing mesh */
diff --git a/source/blender/editors/sculpt_paint/sculpt_intern.h b/source/blender/editors/sculpt_paint/sculpt_intern.h
index e646accf108..9f1cb7a53a4 100644
--- a/source/blender/editors/sculpt_paint/sculpt_intern.h
+++ b/source/blender/editors/sculpt_paint/sculpt_intern.h
@@ -34,9 +34,7 @@
 #include "BKE_pbvh.h"
 
 struct KeyBlock;
-struct Main;
 struct Object;
-struct SculptOrigVertData;
 struct SculptUndoNode;
 struct bContext;
 
@@ -47,7 +45,18 @@ bool sculpt_poll(struct bContext *C);
 bool sculpt_poll_view3d(struct bContext *C);
 
 /* Stroke */
+
+typedef struct SculptCursorGeometryInfo {
+  float location[3];
+  float normal[3];
+  float active_vertex_co[3];
+} SculptCursorGeometryInfo;
+
 bool sculpt_stroke_get_location(struct bContext *C, float out[3], const float mouse[2]);
+bool sculpt_cursor_geometry_info_update(bContext *C,
+                                        SculptCursorGeometryInfo *out,
+                                        const float mouse[2],
+                                        bool use_sampled_normal);
 
 /* Dynamic topology */
 void sculpt_pbvh_clear(Object *ob);
@@ -164,6 +173,7 @@ typedef struct SculptThreadedTaskData {
   float (*area_cos)[3];
   float (*area_nos)[3];
   int *count;
+  bool any_vertex_sampled;
 
   ThreadMutex mutex;
 
@@ -228,7 +238,7 @@ float tex_strength(struct SculptSession *ss,
                    const int thread_id);
 
 /* just for vertex paint. */
-void sculpt_pbvh_calc_area_normal(const struct Brush *brush,
+bool sculpt_pbvh_calc_area_normal(const struct Brush *brush,
                                   Object *ob,
                                   PBVHNode **nodes,
                                   int totnode,
@@ -342,7 +352,7 @@ SculptUndoNode *sculpt_undo_get_node(PBVHNode *node);
 void sculpt_undo_push_begin(const char *name);
 void sculpt_undo_push_end(void);
 
-void sculpt_vertcos_to_key(Object *ob, KeyBlock *kb, float (*vertCos)[3]);
+void sculpt_vertcos_to_key(Object *ob, KeyBlock *kb, const float (*vertCos)[3]);
 
 void sculpt_update_object_bounding_box(struct Object *ob);
 
diff --git a/source/blender/editors/sculpt_paint/sculpt_undo.c b/source/blender/editors/sculpt_paint/sculpt_undo.c
index 3a3487227a3..cb8afd5d7fa 100644
--- a/source/blender/editors/sculpt_paint/sculpt_undo.c
+++ b/source/blender/editors/sculpt_paint/sculpt_undo.c
@@ -189,7 +189,7 @@ static bool sculpt_undo_restore_coords(bContext *C, Depsgraph *depsgraph, Sculpt
 
       /* pbvh uses it's own mvert array, so coords should be */
       /* propagated to pbvh here */
-      BKE_pbvh_apply_vertCos(ss->pbvh, vertCos, ss->kb->totelem);
+      BKE_pbvh_vert_coords_apply(ss->pbvh, vertCos, ss->kb->totelem);
 
       MEM_freeN(vertCos);
     }
diff --git a/source/blender/editors/sound/sound_ops.c b/source/blender/editors/sound/sound_ops.c
index 51231ccf634..4e710d31cbb 100644
--- a/source/blender/editors/sound/sound_ops.c
+++ b/source/blender/editors/sound/sound_ops.c
@@ -710,7 +710,7 @@ static void SOUND_OT_mixdown(wmOperatorType *ot)
                                  FILE_TYPE_FOLDER | FILE_TYPE_SOUND,
                                  FILE_SPECIAL,
                                  FILE_SAVE,
-                                 WM_FILESEL_FILEPATH | WM_FILESEL_RELPATH,
+                                 WM_FILESEL_FILEPATH | WM_FILESEL_RELPATH | WM_FILESEL_SHOW_PROPS,
                                  FILE_DEFAULTDISPLAY,
                                  FILE_SORT_ALPHA);
 #ifdef WITH_AUDASPACE
diff --git a/source/blender/editors/space_action/action_data.c b/source/blender/editors/space_action/action_data.c
index bc7f8a0f79d..5ceaefd6309 100644
--- a/source/blender/editors/space_action/action_data.c
+++ b/source/blender/editors/space_action/action_data.c
@@ -220,7 +220,7 @@ static int action_new_exec(bContext *C, wmOperator *UNUSED(op))
     PointerRNA oldptr;
 
     oldptr = RNA_property_pointer_get(&ptr, prop);
-    oldact = (bAction *)oldptr.id.data;
+    oldact = (bAction *)oldptr.owner_id;
 
     /* stash the old action to prevent it from being lost */
     if (ptr.type == &RNA_AnimData) {
diff --git a/source/blender/editors/space_action/action_draw.c b/source/blender/editors/space_action/action_draw.c
index a7b6f399187..0c57113956d 100644
--- a/source/blender/editors/space_action/action_draw.c
+++ b/source/blender/editors/space_action/action_draw.c
@@ -352,179 +352,231 @@ void draw_channel_strips(bAnimContext *ac, SpaceAction *saction, ARegion *ar)
 /* ************************************************************************* */
 /* Timeline - Caches */
 
-void timeline_draw_cache(SpaceAction *saction, Object *ob, Scene *scene)
+static bool timeline_cache_is_hidden_by_setting(SpaceAction *saction, PTCacheID *pid)
 {
-  PTCacheID *pid;
-  ListBase pidlist;
-  const float cache_draw_height = (4.0f * UI_DPI_FAC * U.pixelsize);
-  float yoffs = 0.f;
-
-  if (!(saction->cache_display & TIME_CACHE_DISPLAY) || (!ob)) {
-    return;
+  switch (pid->type) {
+    case PTCACHE_TYPE_SOFTBODY:
+      if ((saction->cache_display & TIME_CACHE_SOFTBODY) == 0) {
+        return true;
+      }
+      break;
+    case PTCACHE_TYPE_PARTICLES:
+      if ((saction->cache_display & TIME_CACHE_PARTICLES) == 0) {
+        return true;
+      }
+      break;
+    case PTCACHE_TYPE_CLOTH:
+      if ((saction->cache_display & TIME_CACHE_CLOTH) == 0) {
+        return true;
+      }
+      break;
+    case PTCACHE_TYPE_SMOKE_DOMAIN:
+    case PTCACHE_TYPE_SMOKE_HIGHRES:
+      if ((saction->cache_display & TIME_CACHE_SMOKE) == 0) {
+        return true;
+      }
+      break;
+    case PTCACHE_TYPE_DYNAMICPAINT:
+      if ((saction->cache_display & TIME_CACHE_DYNAMICPAINT) == 0) {
+        return true;
+      }
+      break;
+    case PTCACHE_TYPE_RIGIDBODY:
+      if ((saction->cache_display & TIME_CACHE_RIGIDBODY) == 0) {
+        return true;
+      }
+      break;
   }
+  return false;
+}
 
-  BKE_ptcache_ids_from_object(&pidlist, ob, scene, 0);
+static void timeline_cache_color_get(PTCacheID *pid, float color[4])
+{
+  switch (pid->type) {
+    case PTCACHE_TYPE_SOFTBODY:
+      color[0] = 1.0;
+      color[1] = 0.4;
+      color[2] = 0.02;
+      color[3] = 0.1;
+      break;
+    case PTCACHE_TYPE_PARTICLES:
+      color[0] = 1.0;
+      color[1] = 0.1;
+      color[2] = 0.02;
+      color[3] = 0.1;
+      break;
+    case PTCACHE_TYPE_CLOTH:
+      color[0] = 0.1;
+      color[1] = 0.1;
+      color[2] = 0.75;
+      color[3] = 0.1;
+      break;
+    case PTCACHE_TYPE_SMOKE_DOMAIN:
+    case PTCACHE_TYPE_SMOKE_HIGHRES:
+      color[0] = 0.2;
+      color[1] = 0.2;
+      color[2] = 0.2;
+      color[3] = 0.1;
+      break;
+    case PTCACHE_TYPE_DYNAMICPAINT:
+      color[0] = 1.0;
+      color[1] = 0.1;
+      color[2] = 0.75;
+      color[3] = 0.1;
+      break;
+    case PTCACHE_TYPE_RIGIDBODY:
+      color[0] = 1.0;
+      color[1] = 0.6;
+      color[2] = 0.0;
+      color[3] = 0.1;
+      break;
+    default:
+      color[0] = 1.0;
+      color[1] = 0.0;
+      color[2] = 1.0;
+      color[3] = 0.1;
+      BLI_assert(0);
+      break;
+  }
+}
 
-  uint pos = GPU_vertformat_attr_add(immVertexFormat(), "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
-  immBindBuiltinProgram(GPU_SHADER_2D_UNIFORM_COLOR);
+static void timeline_cache_modify_color_based_on_state(PointCache *cache, float color[4])
+{
+  if (cache->flag & PTCACHE_BAKED) {
+    color[0] -= 0.4f;
+    color[1] -= 0.4f;
+    color[2] -= 0.4f;
+  }
+  else if (cache->flag & PTCACHE_OUTDATED) {
+    color[0] += 0.4f;
+    color[1] += 0.4f;
+    color[2] += 0.4f;
+  }
+}
 
-  /* iterate over pointcaches on the active object, and draw each one's range */
-  for (pid = pidlist.first; pid; pid = pid->next) {
-    float col[4];
+static bool timeline_cache_find_next_cached_segment(PointCache *cache,
+                                                    int search_start_frame,
+                                                    int *r_segment_start,
+                                                    int *r_segment_end)
+{
+  int offset = cache->startframe;
+  int current = search_start_frame;
 
-    switch (pid->type) {
-      case PTCACHE_TYPE_SOFTBODY:
-        if (!(saction->cache_display & TIME_CACHE_SOFTBODY)) {
-          continue;
-        }
-        break;
-      case PTCACHE_TYPE_PARTICLES:
-        if (!(saction->cache_display & TIME_CACHE_PARTICLES)) {
-          continue;
-        }
-        break;
-      case PTCACHE_TYPE_CLOTH:
-        if (!(saction->cache_display & TIME_CACHE_CLOTH)) {
-          continue;
-        }
-        break;
-      case PTCACHE_TYPE_SMOKE_DOMAIN:
-      case PTCACHE_TYPE_SMOKE_HIGHRES:
-        if (!(saction->cache_display & TIME_CACHE_SMOKE)) {
-          continue;
-        }
-        break;
-      case PTCACHE_TYPE_DYNAMICPAINT:
-        if (!(saction->cache_display & TIME_CACHE_DYNAMICPAINT)) {
-          continue;
-        }
-        break;
-      case PTCACHE_TYPE_RIGIDBODY:
-        if (!(saction->cache_display & TIME_CACHE_RIGIDBODY)) {
-          continue;
-        }
-        break;
+  /* Find segment start frame. */
+  while (true) {
+    if (current > cache->endframe) {
+      return false;
     }
-
-    if (pid->cache->cached_frames == NULL) {
-      continue;
+    if (cache->cached_frames[current - offset]) {
+      *r_segment_start = current;
+      break;
     }
+    current++;
+  }
 
-    GPU_matrix_push();
-    GPU_matrix_translate_2f(0.0, (float)V2D_SCROLL_HANDLE_HEIGHT + yoffs);
-    GPU_matrix_scale_2f(1.0, cache_draw_height);
-
-    switch (pid->type) {
-      case PTCACHE_TYPE_SOFTBODY:
-        col[0] = 1.0;
-        col[1] = 0.4;
-        col[2] = 0.02;
-        col[3] = 0.1;
-        break;
-      case PTCACHE_TYPE_PARTICLES:
-        col[0] = 1.0;
-        col[1] = 0.1;
-        col[2] = 0.02;
-        col[3] = 0.1;
-        break;
-      case PTCACHE_TYPE_CLOTH:
-        col[0] = 0.1;
-        col[1] = 0.1;
-        col[2] = 0.75;
-        col[3] = 0.1;
-        break;
-      case PTCACHE_TYPE_SMOKE_DOMAIN:
-      case PTCACHE_TYPE_SMOKE_HIGHRES:
-        col[0] = 0.2;
-        col[1] = 0.2;
-        col[2] = 0.2;
-        col[3] = 0.1;
-        break;
-      case PTCACHE_TYPE_DYNAMICPAINT:
-        col[0] = 1.0;
-        col[1] = 0.1;
-        col[2] = 0.75;
-        col[3] = 0.1;
-        break;
-      case PTCACHE_TYPE_RIGIDBODY:
-        col[0] = 1.0;
-        col[1] = 0.6;
-        col[2] = 0.0;
-        col[3] = 0.1;
-        break;
-      default:
-        col[0] = 1.0;
-        col[1] = 0.0;
-        col[2] = 1.0;
-        col[3] = 0.1;
-        BLI_assert(0);
-        break;
+  /* Find segment end frame. */
+  while (true) {
+    if (current > cache->endframe) {
+      *r_segment_end = current - 1;
+      return true;
+    }
+    if (!cache->cached_frames[current - offset]) {
+      *r_segment_end = current - 1;
+      return true;
     }
+    current++;
+  }
+}
+
+static uint timeline_cache_segments_count(PointCache *cache)
+{
+  uint count = 0;
+
+  int current = cache->startframe;
+  int segment_start;
+  int segment_end;
+  while (timeline_cache_find_next_cached_segment(cache, current, &segment_start, &segment_end)) {
+    count++;
+    current = segment_end + 1;
+  }
+
+  return count;
+}
 
-    const int sta = pid->cache->startframe, end = pid->cache->endframe;
+static void timeline_cache_draw_cached_segments(PointCache *cache, uint pos_id)
+{
+  uint segments_count = timeline_cache_segments_count(cache);
+  if (segments_count == 0) {
+    return;
+  }
 
-    GPU_blend(true);
+  immBeginAtMost(GPU_PRIM_TRIS, segments_count * 6);
 
-    immUniformColor4fv(col);
-    immRectf(pos, (float)sta, 0.0, (float)end, 1.0);
+  int current = cache->startframe;
+  int segment_start;
+  int segment_end;
+  while (timeline_cache_find_next_cached_segment(cache, current, &segment_start, &segment_end)) {
+    immRectf_fast(pos_id, segment_start - 0.5f, 0, segment_end + 0.5f, 1.0f);
+    current = segment_end + 1;
+  }
 
-    col[3] = 0.4f;
-    if (pid->cache->flag & PTCACHE_BAKED) {
-      col[0] -= 0.4f;
-      col[1] -= 0.4f;
-      col[2] -= 0.4f;
-    }
-    else if (pid->cache->flag & PTCACHE_OUTDATED) {
-      col[0] += 0.4f;
-      col[1] += 0.4f;
-      col[2] += 0.4f;
-    }
+  immEnd();
+}
 
-    immUniformColor4fv(col);
+static void timeline_cache_draw_single(PTCacheID *pid, float y_offset, float height, uint pos_id)
+{
+  GPU_matrix_push();
+  GPU_matrix_translate_2f(0.0, (float)V2D_SCROLL_HANDLE_HEIGHT + y_offset);
+  GPU_matrix_scale_2f(1.0, height);
 
-    {
-      /* draw a quad for each chunk of consecutive cached frames */
-      const int chunk_tot = 32;
-      int chunk_len = 0;
-      int ista = 0, iend = -1;
+  float color[4];
+  timeline_cache_color_get(pid, color);
 
-      for (int i = sta; i <= end; i++) {
-        if (pid->cache->cached_frames[i - sta]) {
-          if (chunk_len == 0) {
-            immBeginAtMost(GPU_PRIM_TRIS, chunk_tot * 6);
-          }
-          if (ista > iend) {
-            chunk_len++;
-            ista = i;
-          }
-          iend = i;
-        }
-        else {
-          if (ista <= iend) {
-            immRectf_fast(pos, (float)ista - 0.5f, 0.0f, (float)iend + 0.5f, 1.0f);
-            iend = ista - 1;
-          }
-          if (chunk_len >= chunk_tot) {
-            immEnd();
-            chunk_len = 0;
-          }
-        }
-      }
-      if (ista <= iend) {
-        immRectf_fast(pos, (float)ista - 0.5f, 0.0f, (float)iend + 0.5f, 1.0f);
-      }
-      if (chunk_len != 0) {
-        immEnd();
-      }
+  immUniformColor4fv(color);
+  immRectf(pos_id, (float)pid->cache->startframe, 0.0, (float)pid->cache->endframe, 1.0);
+
+  color[3] = 0.4f;
+  timeline_cache_modify_color_based_on_state(pid->cache, color);
+  immUniformColor4fv(color);
+
+  timeline_cache_draw_cached_segments(pid->cache, pos_id);
+
+  GPU_matrix_pop();
+}
+
+void timeline_draw_cache(SpaceAction *saction, Object *ob, Scene *scene)
+{
+  if ((saction->cache_display & TIME_CACHE_DISPLAY) == 0 || ob == NULL) {
+    return;
+  }
+
+  ListBase pidlist;
+  BKE_ptcache_ids_from_object(&pidlist, ob, scene, 0);
+
+  uint pos_id = GPU_vertformat_attr_add(
+      immVertexFormat(), "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
+  immBindBuiltinProgram(GPU_SHADER_2D_UNIFORM_COLOR);
+
+  GPU_blend(true);
+
+  /* Iterate over pointcaches on the active object, and draw each one's range. */
+  float y_offset = 0.0f;
+  const float cache_draw_height = 4.0f * UI_DPI_FAC * U.pixelsize;
+  for (PTCacheID *pid = pidlist.first; pid; pid = pid->next) {
+    if (timeline_cache_is_hidden_by_setting(saction, pid)) {
+      continue;
     }
 
-    GPU_blend(false);
+    if (pid->cache->cached_frames == NULL) {
+      continue;
+    }
 
-    GPU_matrix_pop();
+    timeline_cache_draw_single(pid, y_offset, cache_draw_height, pos_id);
 
-    yoffs += cache_draw_height;
+    y_offset += cache_draw_height;
   }
 
+  GPU_blend(false);
   immUnbindProgram();
 
   BLI_freelistN(&pidlist);
diff --git a/source/blender/editors/space_action/action_intern.h b/source/blender/editors/space_action/action_intern.h
index eaca7968a34..c227b794ae7 100644
--- a/source/blender/editors/space_action/action_intern.h
+++ b/source/blender/editors/space_action/action_intern.h
@@ -28,9 +28,7 @@ struct ARegion;
 struct ARegionType;
 struct Object;
 struct Scene;
-struct ScrArea;
 struct SpaceAction;
-struct View2D;
 struct bAnimContext;
 struct bContext;
 struct wmOperatorType;
diff --git a/source/blender/editors/space_action/action_select.c b/source/blender/editors/space_action/action_select.c
index 9e2634b183a..cbf4d0628e6 100644
--- a/source/blender/editors/space_action/action_select.c
+++ b/source/blender/editors/space_action/action_select.c
@@ -1717,22 +1717,26 @@ static void mouse_action_keys(bAnimContext *ac,
 
       /* Highlight GPencil Layer */
       if (ale != NULL && ale->data != NULL && ale->type == ANIMTYPE_GPLAYER) {
+        bGPdata *gpd = (bGPdata *)ale->id;
         bGPDlayer *gpl = ale->data;
 
         gpl->flag |= GP_LAYER_SELECT;
-        // gpencil_layer_setactive(gpd, gpl);
+        /* Update other layer status. */
+        if (BKE_gpencil_layer_getactive(gpd) != gpl) {
+          BKE_gpencil_layer_setactive(gpd, gpl);
+          BKE_gpencil_layer_autolock_set(gpd);
+          WM_main_add_notifier(NC_GPENCIL | ND_DATA | NA_EDITED, NULL);
+        }
       }
     }
     else if (ac->datatype == ANIMCONT_MASK) {
       /* deselect all other channels first */
       ANIM_deselect_anim_channels(ac, ac->data, ac->datatype, 0, ACHANNEL_SETFLAG_CLEAR);
 
-      /* Highlight GPencil Layer */
       if (ale != NULL && ale->data != NULL && ale->type == ANIMTYPE_MASKLAYER) {
         MaskLayer *masklay = ale->data;
 
         masklay->flag |= MASK_LAYERFLAG_SELECT;
-        // gpencil_layer_setactive(gpd, gpl);
       }
     }
   }
diff --git a/source/blender/editors/space_action/space_action.c b/source/blender/editors/space_action/space_action.c
index 3c879b03126..1685852dd02 100644
--- a/source/blender/editors/space_action/space_action.c
+++ b/source/blender/editors/space_action/space_action.c
@@ -373,7 +373,7 @@ static void saction_channel_region_message_subscribe(const struct bContext *UNUS
    * so just whitelist the entire structs for updates
    */
   {
-    wmMsgParams_RNA msg_key_params = {{{0}}};
+    wmMsgParams_RNA msg_key_params = {{0}};
     StructRNA *type_array[] = {
         &RNA_DopeSheet, /* dopesheet filters */
 
diff --git a/source/blender/editors/space_buttons/buttons_context.c b/source/blender/editors/space_buttons/buttons_context.c
index 78d5ed0f98a..56a0f472010 100644
--- a/source/blender/editors/space_buttons/buttons_context.c
+++ b/source/blender/editors/space_buttons/buttons_context.c
@@ -76,7 +76,7 @@ static int set_pointer_type(ButsContextPath *path, bContextDataResult *result, S
     ptr = &path->ptr[a];
 
     if (RNA_struct_is_a(ptr->type, type)) {
-      CTX_data_pointer_set(result, ptr->id.data, ptr->type, ptr->data);
+      CTX_data_pointer_set(result, ptr->owner_id, ptr->type, ptr->data);
       return 1;
     }
   }
@@ -930,7 +930,7 @@ int buttons_context(const bContext *C, const char *member, bContextDataResult *r
 
     if (ct->user && ct->user->ptr.data) {
       ButsTextureUser *user = ct->user;
-      CTX_data_pointer_set(result, user->ptr.id.data, user->ptr.type, user->ptr.data);
+      CTX_data_pointer_set(result, user->ptr.owner_id, user->ptr.type, user->ptr.data);
     }
 
     return 1;
@@ -1018,7 +1018,7 @@ int buttons_context(const bContext *C, const char *member, bContextDataResult *r
     PointerRNA *ptr = get_pointer_type(path, &RNA_ParticleSettings);
 
     if (ptr && ptr->data) {
-      CTX_data_pointer_set(result, ptr->id.data, &RNA_ParticleSettings, ptr->data);
+      CTX_data_pointer_set(result, ptr->owner_id, &RNA_ParticleSettings, ptr->data);
       return 1;
     }
     else {
@@ -1027,7 +1027,7 @@ int buttons_context(const bContext *C, const char *member, bContextDataResult *r
 
       if (ptr && ptr->data) {
         ParticleSettings *part = ((ParticleSystem *)ptr->data)->part;
-        CTX_data_pointer_set(result, ptr->id.data, &RNA_ParticleSettings, part);
+        CTX_data_pointer_set(result, ptr->owner_id, &RNA_ParticleSettings, part);
         return 1;
       }
     }
@@ -1285,8 +1285,8 @@ ID *buttons_context_id_path(const bContext *C)
         }
       }
 
-      if (ptr->id.data) {
-        return ptr->id.data;
+      if (ptr->owner_id) {
+        return ptr->owner_id;
       }
     }
   }
diff --git a/source/blender/editors/space_buttons/buttons_ops.c b/source/blender/editors/space_buttons/buttons_ops.c
index 500efe4bb4d..3ea6d183969 100644
--- a/source/blender/editors/space_buttons/buttons_ops.c
+++ b/source/blender/editors/space_buttons/buttons_ops.c
@@ -103,7 +103,7 @@ static int file_browse_exec(bContext *C, wmOperator *op)
   /* add slash for directories, important for some properties */
   if (RNA_property_subtype(fbo->prop) == PROP_DIRPATH) {
     const bool is_relative = RNA_boolean_get(op->ptr, "relative_path");
-    id = fbo->ptr.id.data;
+    id = fbo->ptr.owner_id;
 
     BLI_strncpy(path, str, FILE_MAX);
     BLI_path_abs(path, id ? ID_BLEND_PATH(bmain, id) : BKE_main_blendfile_path(bmain));
diff --git a/source/blender/editors/space_buttons/buttons_texture.c b/source/blender/editors/space_buttons/buttons_texture.c
index 87ea011e0a7..03c8fbb4e66 100644
--- a/source/blender/editors/space_buttons/buttons_texture.c
+++ b/source/blender/editors/space_buttons/buttons_texture.c
@@ -385,7 +385,7 @@ static void template_texture_select(bContext *C, void *user_p, void *UNUSED(arg)
     if (user->ptr.type == &RNA_ParticleSettingsTextureSlot) {
       /* stupid exception for particle systems which still uses influence
        * from the old texture system, set the active texture slots as well */
-      ParticleSettings *part = user->ptr.id.data;
+      ParticleSettings *part = (ParticleSettings *)user->ptr.owner_id;
       int a;
 
       for (a = 0; a < MAX_MTEX; a++) {
diff --git a/source/blender/editors/space_buttons/space_buttons.c b/source/blender/editors/space_buttons/space_buttons.c
index b3690bf2fe3..6008d06e376 100644
--- a/source/blender/editors/space_buttons/space_buttons.c
+++ b/source/blender/editors/space_buttons/space_buttons.c
@@ -404,8 +404,6 @@ static void buttons_area_listener(wmWindow *UNUSED(win),
           buttons_area_redraw(sa, BCONTEXT_DATA); /* autotexpace flag */
           break;
         case ND_POSE:
-          buttons_area_redraw(sa, BCONTEXT_DATA);
-          break;
         case ND_BONE_ACTIVE:
         case ND_BONE_SELECT:
           buttons_area_redraw(sa, BCONTEXT_BONE);
@@ -567,7 +565,7 @@ static void buttons_id_remap(ScrArea *UNUSED(sa), SpaceLink *slink, ID *old_id,
     int i;
 
     for (i = 0; i < path->len; i++) {
-      if (path->ptr[i].id.data == old_id) {
+      if (path->ptr[i].owner_id == old_id) {
         break;
       }
     }
diff --git a/source/blender/editors/space_clip/clip_dopesheet_draw.c b/source/blender/editors/space_clip/clip_dopesheet_draw.c
index 599a92ff77f..3c2d3eb1d97 100644
--- a/source/blender/editors/space_clip/clip_dopesheet_draw.c
+++ b/source/blender/editors/space_clip/clip_dopesheet_draw.c
@@ -222,8 +222,9 @@ void clip_draw_dopesheet_main(SpaceClip *sc, ARegion *ar, Scene *scene)
           format, "outlineColor", GPU_COMP_U8, 4, GPU_FETCH_INT_TO_FLOAT_UNIT);
       uint flags_id = GPU_vertformat_attr_add(format, "flags", GPU_COMP_U32, 1, GPU_FETCH_INT);
 
-      immBindBuiltinProgram(GPU_SHADER_KEYFRAME_DIAMOND);
       GPU_program_point_size(true);
+      immBindBuiltinProgram(GPU_SHADER_KEYFRAME_DIAMOND);
+      immUniform1f("outline_scale", 1.0f);
       immUniform2f(
           "ViewportSize", BLI_rcti_size_x(&v2d->mask) + 1, BLI_rcti_size_y(&v2d->mask) + 1);
       immBegin(GPU_PRIM_POINTS, keyframe_len);
diff --git a/source/blender/editors/space_console/space_console.c b/source/blender/editors/space_console/space_console.c
index 999255aef88..f5c02dbd724 100644
--- a/source/blender/editors/space_console/space_console.c
+++ b/source/blender/editors/space_console/space_console.c
@@ -26,6 +26,7 @@
 #include "BLI_blenlib.h"
 #include "BLI_utildefines.h"
 
+#include "BKE_global.h"
 #include "BKE_context.h"
 #include "BKE_screen.h"
 
@@ -173,7 +174,7 @@ static void id_drop_copy(wmDrag *drag, wmDropBox *drop)
   ID *id = WM_drag_ID(drag, 0);
 
   /* copy drag path to properties */
-  char *text = RNA_path_full_ID_py(id);
+  char *text = RNA_path_full_ID_py(G_MAIN, id);
   RNA_string_set(drop->ptr, "text", text);
   MEM_freeN(text);
 }
diff --git a/source/blender/editors/space_file/file_draw.c b/source/blender/editors/space_file/file_draw.c
index 3a6d59c1dbf..0083fc244d8 100644
--- a/source/blender/editors/space_file/file_draw.c
+++ b/source/blender/editors/space_file/file_draw.c
@@ -26,6 +26,7 @@
 #include <errno.h>
 
 #include "BLI_blenlib.h"
+#include "BLI_fileops_types.h"
 #include "BLI_utildefines.h"
 #include "BLI_math.h"
 
@@ -77,234 +78,6 @@ static char *file_draw_tooltip_func(bContext *UNUSED(C), void *argN, const char
   return BLI_strdup(dyn_tooltip);
 }
 
-/* Note: This function uses pixelspace (0, 0, winx, winy), not view2d.
- * The controls are laid out as follows:
- *
- * -------------------------------------------
- * | Directory input               | execute |
- * -------------------------------------------
- * | Filename input        | + | - | cancel  |
- * -------------------------------------------
- *
- * The input widgets will stretch to fill any excess space.
- * When there isn't enough space for all controls to be shown, they are
- * hidden in this order: x/-, execute/cancel, input widgets.
- */
-void file_draw_buttons(const bContext *C, ARegion *ar)
-{
-  /* Button layout. */
-  const int max_x = ar->winx - 10;
-  const int line1_y = ar->winy - (IMASEL_BUTTONS_HEIGHT / 2 + IMASEL_BUTTONS_MARGIN);
-  const int line2_y = line1_y - (IMASEL_BUTTONS_HEIGHT / 2 + IMASEL_BUTTONS_MARGIN);
-  const int input_minw = 20;
-  const int btn_h = UI_UNIT_Y;
-  const int btn_fn_w = UI_UNIT_X;
-  const int btn_minw = 80;
-  const int btn_margin = 20;
-  const int separator = 4;
-
-  /* Additional locals. */
-  char uiblockstr[32];
-  int loadbutton;
-  int fnumbuttons;
-  int min_x = 10;
-  int chan_offs = 0;
-  int available_w = max_x - min_x;
-  int line1_w = available_w;
-  int line2_w = available_w;
-
-  uiBut *but;
-  uiBlock *block;
-  SpaceFile *sfile = CTX_wm_space_file(C);
-  FileSelectParams *params = ED_fileselect_get_params(sfile);
-  ARegion *artmp;
-  const bool is_browse_only = (sfile->op == NULL);
-
-  /* Initialize UI block. */
-  BLI_snprintf(uiblockstr, sizeof(uiblockstr), "win %p", (void *)ar);
-  block = UI_block_begin(C, ar, uiblockstr, UI_EMBOSS);
-
-  /* exception to make space for collapsed region icon */
-  for (artmp = CTX_wm_area(C)->regionbase.first; artmp; artmp = artmp->next) {
-    if (artmp->regiontype == RGN_TYPE_TOOLS && artmp->flag & RGN_FLAG_HIDDEN) {
-      chan_offs = 16;
-      min_x += chan_offs;
-      available_w -= chan_offs;
-    }
-  }
-
-  /* Is there enough space for the execute / cancel buttons? */
-
-  if (is_browse_only) {
-    loadbutton = 0;
-  }
-  else {
-    const uiFontStyle *fstyle = UI_FSTYLE_WIDGET;
-    loadbutton = UI_fontstyle_string_width(fstyle, params->title) + btn_margin;
-    CLAMP_MIN(loadbutton, btn_minw);
-    if (available_w <= loadbutton + separator + input_minw) {
-      loadbutton = 0;
-    }
-  }
-
-  if (loadbutton) {
-    line1_w -= (loadbutton + separator);
-    line2_w = line1_w;
-  }
-
-  /* Is there enough space for file number increment/decrement buttons? */
-  fnumbuttons = 2 * btn_fn_w;
-  if (!loadbutton || line2_w <= fnumbuttons + separator + input_minw) {
-    fnumbuttons = 0;
-  }
-  else {
-    line2_w -= (fnumbuttons + separator);
-  }
-
-  /* Text input fields for directory and file. */
-  if (available_w > 0) {
-    const struct FileDirEntry *file = sfile->files ?
-                                          filelist_file(sfile->files, params->active_file) :
-                                          NULL;
-    int overwrite_alert = file_draw_check_exists(sfile);
-    const bool is_active_dir = file && (file->typeflag & FILE_TYPE_FOLDER);
-
-    /* callbacks for operator check functions */
-    UI_block_func_set(block, file_draw_check_cb, NULL, NULL);
-
-    but = uiDefBut(block,
-                   UI_BTYPE_TEXT,
-                   -1,
-                   "",
-                   min_x,
-                   line1_y,
-                   line1_w - chan_offs,
-                   btn_h,
-                   params->dir,
-                   0.0,
-                   (float)FILE_MAX,
-                   0,
-                   0,
-                   TIP_("File path"));
-    UI_but_func_complete_set(but, autocomplete_directory, NULL);
-    UI_but_flag_enable(but, UI_BUT_NO_UTF8);
-    UI_but_flag_disable(but, UI_BUT_UNDO);
-    UI_but_funcN_set(but, file_directory_enter_handle, NULL, but);
-
-    /* TODO, directory editing is non-functional while a library is loaded
-     * until this is properly supported just disable it. */
-    if (sfile->files && filelist_lib(sfile->files)) {
-      UI_but_flag_enable(but, UI_BUT_DISABLED);
-    }
-
-    if ((params->flag & FILE_DIRSEL_ONLY) == 0) {
-      but = uiDefBut(
-          block,
-          UI_BTYPE_TEXT,
-          -1,
-          "",
-          min_x,
-          line2_y,
-          line2_w - chan_offs,
-          btn_h,
-          is_active_dir ? (char *)"" : params->file,
-          0.0,
-          (float)FILE_MAXFILE,
-          0,
-          0,
-          TIP_(overwrite_alert ? N_("File name, overwrite existing") : N_("File name")));
-      UI_but_func_complete_set(but, autocomplete_file, NULL);
-      UI_but_flag_enable(but, UI_BUT_NO_UTF8);
-      UI_but_flag_disable(but, UI_BUT_UNDO);
-      /* silly workaround calling NFunc to ensure this does not get called
-       * immediate ui_apply_but_func but only after button deactivates */
-      UI_but_funcN_set(but, file_filename_enter_handle, NULL, but);
-
-      /* check if this overrides a file and if the operator option is used */
-      if (overwrite_alert) {
-        UI_but_flag_enable(but, UI_BUT_REDALERT);
-      }
-    }
-
-    /* clear func */
-    UI_block_func_set(block, NULL, NULL, NULL);
-  }
-
-  /* Filename number increment / decrement buttons. */
-  if (fnumbuttons && (params->flag & FILE_DIRSEL_ONLY) == 0) {
-    UI_block_align_begin(block);
-    but = uiDefIconButO(block,
-                        UI_BTYPE_BUT,
-                        "FILE_OT_filenum",
-                        0,
-                        ICON_REMOVE,
-                        min_x + line2_w + separator - chan_offs,
-                        line2_y,
-                        btn_fn_w,
-                        btn_h,
-                        TIP_("Decrement the filename number"));
-    RNA_int_set(UI_but_operator_ptr_get(but), "increment", -1);
-
-    but = uiDefIconButO(block,
-                        UI_BTYPE_BUT,
-                        "FILE_OT_filenum",
-                        0,
-                        ICON_ADD,
-                        min_x + line2_w + separator + btn_fn_w - chan_offs,
-                        line2_y,
-                        btn_fn_w,
-                        btn_h,
-                        TIP_("Increment the filename number"));
-    RNA_int_set(UI_but_operator_ptr_get(but), "increment", 1);
-    UI_block_align_end(block);
-  }
-
-  /* Execute / cancel buttons. */
-  if (loadbutton) {
-    const struct FileDirEntry *file = sfile->files ?
-                                          filelist_file(sfile->files, params->active_file) :
-                                          NULL;
-    char const *str_exec;
-
-    if (file && FILENAME_IS_PARENT(file->relpath)) {
-      str_exec = IFACE_("Parent Directory");
-    }
-    else if (file && file->typeflag & FILE_TYPE_DIR) {
-      str_exec = IFACE_("Open Directory");
-    }
-    else {
-      str_exec = params->title; /* params->title is already translated! */
-    }
-
-    but = uiDefButO(block,
-                    UI_BTYPE_BUT,
-                    "FILE_OT_execute",
-                    WM_OP_EXEC_REGION_WIN,
-                    str_exec,
-                    max_x - loadbutton,
-                    line1_y,
-                    loadbutton,
-                    btn_h,
-                    "");
-    /* Just a display hint. */
-    UI_but_flag_enable(but, UI_BUT_ACTIVE_DEFAULT);
-
-    uiDefButO(block,
-              UI_BTYPE_BUT,
-              "FILE_OT_cancel",
-              WM_OP_EXEC_REGION_WIN,
-              IFACE_("Cancel"),
-              max_x - loadbutton,
-              line2_y,
-              loadbutton,
-              btn_h,
-              "");
-  }
-
-  UI_block_end(C, block);
-  UI_block_draw(C, block);
-}
-
 static void draw_tile(int sx, int sy, int width, int height, int colorid, int shade)
 {
   float color[4];
@@ -349,7 +122,7 @@ static void file_draw_string(int sx,
   rcti rect;
   char fname[FILE_MAXFILE];
 
-  if (string[0] == '\0') {
+  if (string[0] == '\0' || width < 1) {
     return;
   }
 
@@ -362,7 +135,7 @@ static void file_draw_string(int sx,
   /* no text clipping needed, UI_fontstyle_draw does it but is a bit too strict
    * (for buttons it works) */
   rect.xmin = sx;
-  rect.xmax = (int)(sx + ceil(width + 5.0f / UI_DPI_FAC));
+  rect.xmax = sx + round_fl_to_int(width);
   rect.ymin = sy - height;
   rect.ymax = sy;
 
@@ -404,8 +177,8 @@ static void file_draw_preview(uiBlock *block,
   float scaledx, scaledy;
   float scale;
   int ex, ey;
-  bool use_dropshadow = !is_icon && (typeflags & FILE_TYPE_IMAGE);
-  float col[4] = {1.0f, 1.0f, 1.0f, 1.0f};
+  bool use_dropshadow = !is_icon &&
+                        (typeflags & (FILE_TYPE_IMAGE | FILE_TYPE_MOVIE | FILE_TYPE_BLENDER));
 
   BLI_assert(imb != NULL);
 
@@ -442,13 +215,27 @@ static void file_draw_preview(uiBlock *block,
 
   /* shadow */
   if (use_dropshadow) {
-    UI_draw_box_shadow(220, (float)xco, (float)yco, (float)(xco + ex), (float)(yco + ey));
+    UI_draw_box_shadow(128, (float)xco, (float)yco, (float)(xco + ex), (float)(yco + ey));
   }
 
   GPU_blend(true);
 
-  /* the image */
-  if (!is_icon && typeflags & FILE_TYPE_FTFONT) {
+  /* the large image */
+
+  float col[4] = {1.0f, 1.0f, 1.0f, 1.0f};
+  if (is_icon) {
+    /*  File and Folder icons draw with lowered opacity until we add themes */
+    col[3] = 0.6f;
+    /*  Use dark images if background is light */
+    float bg[3];
+    UI_GetThemeColor3fv(TH_BACK, bg);
+    if (rgb_to_grayscale(bg) > 0.5f) {
+      col[0] = 0;
+      col[1] = 0;
+      col[2] = 0;
+    }
+  }
+  else if (typeflags & FILE_TYPE_FTFONT) {
     UI_GetThemeColor4fv(TH_TEXT, col);
   }
 
@@ -477,30 +264,61 @@ static void file_draw_preview(uiBlock *block,
   GPU_blend_set_func_separate(
       GPU_SRC_ALPHA, GPU_ONE_MINUS_SRC_ALPHA, GPU_ONE, GPU_ONE_MINUS_SRC_ALPHA);
 
-  if (icon) {
-    UI_icon_draw_ex((float)xco + (7 * UI_DPI_FAC),
-                    (float)yco + (7 * UI_DPI_FAC),
-                    icon,
-                    icon_aspect,
-                    1.0f,
-                    0.0f,
-                    NULL,
-                    false);
+  if (icon && (icon != ICON_FILE_FONT)) {
+    /* size of center icon is scaled to fit container and UI scale */
+    float icon_x, icon_y;
+
+    if (is_icon) {
+      const float icon_size = 16.0f / icon_aspect * U.dpi_fac;
+      float icon_opacity = MIN2(icon_aspect, 0.7);
+      uchar icon_color[4] = {255, 255, 255, 255};
+      float bg[3];
+      /*  base this off theme color of file or folder later */
+      UI_GetThemeColor3fv(TH_BACK, bg);
+      if (rgb_to_grayscale(bg) > 0.5f) {
+        icon_color[0] = 0;
+        icon_color[1] = 0;
+        icon_color[2] = 0;
+      }
+      icon_x = xco + (ex / 2.0f) - (icon_size / 2.0f);
+      icon_y = yco + (ey / 2.0f) - (icon_size * ((typeflags & FILE_TYPE_DIR) ? 0.78f : 0.65f));
+      UI_icon_draw_ex(
+          icon_x, icon_y, icon, icon_aspect / U.dpi_fac, icon_opacity, 0.0f, icon_color, false);
+    }
+    else {
+      const uchar dark[4] = {0, 0, 0, 255};
+      const uchar light[4] = {255, 255, 255, 255};
+
+      /* Smaller, fainter icon for preview image thumbnail. */
+      icon_x = xco + (2.0f * UI_DPI_FAC);
+      icon_y = yco + (2.0f * UI_DPI_FAC);
+
+      UI_icon_draw_ex(icon_x + 1, icon_y - 1, icon, 1.0f / U.dpi_fac, 0.2f, 0.0f, dark, false);
+      UI_icon_draw_ex(icon_x, icon_y, icon, 1.0f / U.dpi_fac, 0.6f, 0.0f, light, false);
+    }
   }
 
   /* border */
   if (use_dropshadow) {
     GPUVertFormat *format = immVertexFormat();
-    uint pos = GPU_vertformat_attr_add(format, "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
+    uint pos_attr = GPU_vertformat_attr_add(format, "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
+    uint col_attr = GPU_vertformat_attr_add(format, "color", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
 
-    immBindBuiltinProgram(GPU_SHADER_2D_UNIFORM_COLOR);
-    immUniformColor4f(0.0f, 0.0f, 0.0f, 0.4f);
-    imm_draw_box_wire_2d(pos, (float)xco, (float)yco, (float)(xco + ex), (float)(yco + ey));
+    immBindBuiltinProgram(GPU_SHADER_2D_FLAT_COLOR);
+    immBegin(GPU_PRIM_LINE_LOOP, 4);
+    immAttr4f(col_attr, 1.0f, 1.0f, 1.0f, 0.15f);
+    immVertex2f(pos_attr, (float)xco + 1, (float)(yco + ey));
+    immAttr4f(col_attr, 1.0f, 1.0f, 1.0f, 0.2f);
+    immVertex2f(pos_attr, (float)(xco + ex), (float)(yco + ey));
+    immAttr4f(col_attr, 0.0f, 0.0f, 0.0f, 0.2f);
+    immVertex2f(pos_attr, (float)(xco + ex), (float)yco + 1);
+    immAttr4f(col_attr, 0.0f, 0.0f, 0.0f, 0.3f);
+    immVertex2f(pos_attr, (float)xco + 1, (float)yco + 1);
+    immEnd();
     immUnbindProgram();
   }
 
   but = uiDefBut(block, UI_BTYPE_LABEL, 0, "", xco, yco, ex, ey, NULL, 0.0, 0.0, 0, 0, NULL);
-  UI_but_func_tooltip_set(but, file_draw_tooltip_func, BLI_strdup(path));
 
   /* dragregion */
   if (drag) {
@@ -557,6 +375,7 @@ static void renamebutton_cb(bContext *C, void *UNUSED(arg1), char *oldname)
 
 static void draw_background(FileLayout *layout, View2D *v2d)
 {
+  const int item_height = layout->tile_h + (2 * layout->tile_border_y);
   int i;
   int sy;
 
@@ -565,9 +384,11 @@ static void draw_background(FileLayout *layout, View2D *v2d)
   immUniformThemeColorShade(TH_BACK, -7);
 
   /* alternating flat shade background */
-  for (i = 0; (i <= layout->rows); i += 2) {
-    sy = (int)v2d->cur.ymax - i * (layout->tile_h + 2 * layout->tile_border_y) -
-         layout->tile_border_y;
+  for (i = 2; (i <= layout->rows + 1); i += 2) {
+    sy = (int)v2d->cur.ymax - layout->offset_top - i * item_height - layout->tile_border_y;
+
+    /* Offsett pattern slightly to add scroll effect. */
+    sy += round_fl_to_int(item_height * (v2d->tot.ymax - v2d->cur.ymax) / item_height);
 
     immRectf(pos,
              v2d->cur.xmin,
@@ -632,6 +453,176 @@ static void draw_dividers(FileLayout *layout, View2D *v2d)
   }
 }
 
+static void draw_columnheader_background(const FileLayout *layout, const View2D *v2d)
+{
+  uint pos = GPU_vertformat_attr_add(immVertexFormat(), "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
+
+  immBindBuiltinProgram(GPU_SHADER_2D_UNIFORM_COLOR);
+  immUniformThemeColorShade(TH_BACK, 11);
+
+  immRectf(pos,
+           v2d->cur.xmin,
+           v2d->cur.ymax - layout->attribute_column_header_h,
+           v2d->cur.xmax,
+           v2d->cur.ymax);
+
+  immUnbindProgram();
+}
+
+static void draw_columnheader_columns(const FileSelectParams *params,
+                                      FileLayout *layout,
+                                      const View2D *v2d,
+                                      const uchar text_col[4])
+{
+  const float divider_pad = 0.2 * layout->attribute_column_header_h;
+  int sx = v2d->cur.xmin, sy = v2d->cur.ymax;
+
+  for (FileAttributeColumnType column_type = 0; column_type < ATTRIBUTE_COLUMN_MAX;
+       column_type++) {
+    if (!file_attribute_column_type_enabled(params, column_type)) {
+      continue;
+    }
+    const FileAttributeColumn *column = &layout->attribute_columns[column_type];
+
+    /* Active sort type triangle */
+    if (params->sort == column->sort_type) {
+      float tri_color[4];
+
+      rgba_uchar_to_float(tri_color, text_col);
+      UI_draw_icon_tri(sx + column->width - (0.3f * U.widget_unit) -
+                           ATTRIBUTE_COLUMN_PADDING / 2.0f,
+                       sy + (0.1f * U.widget_unit) - (layout->attribute_column_header_h / 2),
+                       (params->flag & FILE_SORT_INVERT) ? 't' : 'v',
+                       tri_color);
+    }
+
+    file_draw_string(sx + ATTRIBUTE_COLUMN_PADDING,
+                     sy - layout->tile_border_y,
+                     IFACE_(column->name),
+                     column->width - 2 * ATTRIBUTE_COLUMN_PADDING,
+                     layout->attribute_column_header_h - layout->tile_border_y,
+                     UI_STYLE_TEXT_LEFT,
+                     text_col);
+
+    /* Separator line */
+    if (column_type != COLUMN_NAME) {
+      uint pos = GPU_vertformat_attr_add(
+          immVertexFormat(), "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
+
+      immBindBuiltinProgram(GPU_SHADER_2D_UNIFORM_COLOR);
+      immUniformThemeColorShade(TH_BACK, -10);
+      immBegin(GPU_PRIM_LINES, 2);
+      immVertex2f(pos, sx - 1, sy - divider_pad);
+      immVertex2f(pos, sx - 1, sy - layout->attribute_column_header_h + divider_pad);
+      immEnd();
+      immUnbindProgram();
+    }
+
+    sx += column->width;
+  }
+
+  /* Vertical separator lines line */
+  {
+    uint pos = GPU_vertformat_attr_add(immVertexFormat(), "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
+    immBindBuiltinProgram(GPU_SHADER_2D_UNIFORM_COLOR);
+    immUniformThemeColorShade(TH_BACK, -10);
+    immBegin(GPU_PRIM_LINES, 4);
+    immVertex2f(pos, v2d->cur.xmin, sy);
+    immVertex2f(pos, v2d->cur.xmax, sy);
+    immVertex2f(pos, v2d->cur.xmin, sy - layout->attribute_column_header_h);
+    immVertex2f(pos, v2d->cur.xmax, sy - layout->attribute_column_header_h);
+    immEnd();
+    immUnbindProgram();
+  }
+}
+
+/**
+ * Updates the stat string stored in file->entry if necessary.
+ */
+static const char *filelist_get_details_column_string(FileAttributeColumnType column,
+                                                      const FileDirEntry *file,
+                                                      const bool small_size,
+                                                      const bool update_stat_strings)
+{
+  switch (column) {
+    case COLUMN_DATETIME:
+      if (!(file->typeflag & FILE_TYPE_BLENDERLIB) && !FILENAME_IS_CURRPAR(file->relpath)) {
+        if ((file->entry->datetime_str[0] == '\0') || update_stat_strings) {
+          char date[FILELIST_DIRENTRY_DATE_LEN], time[FILELIST_DIRENTRY_TIME_LEN];
+          bool is_today, is_yesterday;
+
+          BLI_filelist_entry_datetime_to_string(
+              NULL, file->entry->time, small_size, time, date, &is_today, &is_yesterday);
+
+          if (is_today || is_yesterday) {
+            BLI_strncpy(date, is_today ? N_("Today") : N_("Yesterday"), sizeof(date));
+          }
+          BLI_snprintf(
+              file->entry->datetime_str, sizeof(file->entry->datetime_str), "%s %s", date, time);
+        }
+
+        return file->entry->datetime_str;
+      }
+      break;
+    case COLUMN_SIZE:
+      if ((file->typeflag & (FILE_TYPE_BLENDER | FILE_TYPE_BLENDER_BACKUP)) ||
+          !(file->typeflag & (FILE_TYPE_DIR | FILE_TYPE_BLENDERLIB))) {
+        if ((file->entry->size_str[0] == '\0') || update_stat_strings) {
+          BLI_filelist_entry_size_to_string(
+              NULL, file->entry->size, small_size, file->entry->size_str);
+        }
+
+        return file->entry->size_str;
+      }
+      break;
+    default:
+      break;
+  }
+
+  return NULL;
+}
+
+static void draw_details_columns(const FileSelectParams *params,
+                                 const FileLayout *layout,
+                                 const FileDirEntry *file,
+                                 const int pos_x,
+                                 const int pos_y,
+                                 const uchar text_col[4])
+{
+  const bool small_size = SMALL_SIZE_CHECK(params->thumbnail_size);
+  const bool update_stat_strings = small_size != SMALL_SIZE_CHECK(layout->curr_size);
+  int sx = pos_x - layout->tile_border_x - (UI_UNIT_X * 0.1f), sy = pos_y;
+
+  for (FileAttributeColumnType column_type = 0; column_type < ATTRIBUTE_COLUMN_MAX;
+       column_type++) {
+    const FileAttributeColumn *column = &layout->attribute_columns[column_type];
+
+    /* Name column is not a detail column (should already be drawn), always skip here. */
+    if (column_type == COLUMN_NAME) {
+      sx += column->width;
+      continue;
+    }
+    if (!file_attribute_column_type_enabled(params, column_type)) {
+      continue;
+    }
+
+    const char *str = filelist_get_details_column_string(
+        column_type, file, small_size, update_stat_strings);
+
+    if (str) {
+      file_draw_string(sx + ATTRIBUTE_COLUMN_PADDING,
+                       sy - layout->tile_border_y,
+                       IFACE_(str),
+                       column->width - 2 * ATTRIBUTE_COLUMN_PADDING,
+                       layout->tile_h,
+                       column->text_align,
+                       text_col);
+    }
+
+    sx += column->width;
+  }
+}
+
 void file_draw_list(const bContext *C, ARegion *ar)
 {
   SpaceFile *sfile = CTX_wm_space_file(C);
@@ -652,18 +643,14 @@ void file_draw_list(const bContext *C, ARegion *ar)
   bool is_icon;
   eFontStyle_Align align;
   bool do_drag;
-  int column_space = 0.6f * UI_UNIT_X;
   unsigned char text_col[4];
-  const bool small_size = SMALL_SIZE_CHECK(params->thumbnail_size);
-  const bool update_stat_strings = small_size != SMALL_SIZE_CHECK(layout->curr_size);
-  const float thumb_icon_aspect = sqrtf(64.0f / (float)(params->thumbnail_size));
+  const bool draw_columnheader = (params->display == FILE_VERTICALDISPLAY);
+  const float thumb_icon_aspect = MIN2(64.0f / (float)(params->thumbnail_size), 1.0f);
 
   numfiles = filelist_files_ensure(files);
 
   if (params->display != FILE_IMGDISPLAY) {
-
     draw_background(layout, v2d);
-
     draw_dividers(layout, v2d);
   }
 
@@ -679,13 +666,14 @@ void file_draw_list(const bContext *C, ARegion *ar)
     numfiles_layout += layout->rows;
   }
   else {
-    numfiles_layout += layout->columns;
+    numfiles_layout += layout->flow_columns;
   }
 
   filelist_file_cache_slidingwindow_set(files, numfiles_layout);
 
-  textwidth = (FILE_IMGDISPLAY == params->display) ? layout->tile_w :
-                                                     (int)layout->column_widths[COLUMN_NAME];
+  textwidth = (FILE_IMGDISPLAY == params->display) ?
+                  layout->tile_w :
+                  round_fl_to_int(layout->attribute_columns[COLUMN_NAME].width);
   textheight = (int)(layout->textheight * 3.0 / 2.0 + 0.5);
 
   align = (FILE_IMGDISPLAY == params->display) ? UI_STYLE_TEXT_CENTER : UI_STYLE_TEXT_LEFT;
@@ -719,11 +707,16 @@ void file_draw_list(const bContext *C, ARegion *ar)
 
   BLF_batch_draw_begin();
 
+  UI_GetThemeColor4ubv(TH_TEXT, text_col);
+
   for (i = offset; (i < numfiles) && (i < offset + numfiles_layout); i++) {
     unsigned int file_selflag;
     char path[FILE_MAX_LIBEXTRA];
+    int padx = 0.1f * UI_UNIT_X;
+    int icon_ofs = 0;
+
     ED_fileselect_layout_tilepos(layout, i, &sx, &sy);
-    sx += (int)(v2d->tot.xmin + 0.1f * UI_UNIT_X);
+    sx += (int)(v2d->tot.xmin + padx);
     sy = (int)(v2d->tot.ymax - sy);
 
     file = filelist_file(files, i);
@@ -737,15 +730,14 @@ void file_draw_list(const bContext *C, ARegion *ar)
         int colorid = (file_selflag & FILE_SEL_SELECTED) ? TH_HILITE : TH_BACK;
         int shade = (params->highlight_file == i) || (file_selflag & FILE_SEL_HIGHLIGHTED) ? 35 :
                                                                                              0;
+        const short width = ELEM(params->display, FILE_VERTICALDISPLAY, FILE_HORIZONTALDISPLAY) ?
+                                layout->tile_w - (2 * padx) :
+                                layout->tile_w;
 
         BLI_assert(i == 0 || !FILENAME_IS_CURRPAR(file->relpath));
 
-        draw_tile(sx,
-                  sy - 1,
-                  layout->tile_w + 4,
-                  sfile->layout->tile_h + layout->tile_border_y,
-                  colorid,
-                  shade);
+        draw_tile(
+            sx, sy - 1, width, sfile->layout->tile_h + layout->tile_border_y, colorid, shade);
       }
     }
     UI_draw_roundbox_corner_set(UI_CNR_NONE);
@@ -778,38 +770,28 @@ void file_draw_list(const bContext *C, ARegion *ar)
       file_draw_icon(block,
                      path,
                      sx,
-                     sy - (UI_UNIT_Y / 6),
+                     sy - layout->tile_border_y,
                      filelist_geticon(files, i, true),
                      ICON_DEFAULT_WIDTH_SCALE,
                      ICON_DEFAULT_HEIGHT_SCALE,
                      do_drag);
-      sx += ICON_DEFAULT_WIDTH_SCALE + 0.2f * UI_UNIT_X;
+      icon_ofs += ICON_DEFAULT_WIDTH_SCALE + 0.2f * UI_UNIT_X;
     }
 
-    UI_GetThemeColor4ubv(TH_TEXT, text_col);
-
     if (file_selflag & FILE_SEL_EDITING) {
       uiBut *but;
-      short width;
-
-      if (params->display == FILE_SHORTDISPLAY) {
-        width = layout->tile_w - (ICON_DEFAULT_WIDTH_SCALE + 0.2f * UI_UNIT_X);
-      }
-      else if (params->display == FILE_LONGDISPLAY) {
-        width = layout->column_widths[COLUMN_NAME] + (column_space * 3.5f);
-      }
-      else {
-        BLI_assert(params->display == FILE_IMGDISPLAY);
-        width = textwidth;
-      }
+      const short width = (params->display == FILE_IMGDISPLAY) ?
+                              textwidth :
+                              layout->attribute_columns[COLUMN_NAME].width -
+                                  ATTRIBUTE_COLUMN_PADDING;
 
       but = uiDefBut(block,
                      UI_BTYPE_TEXT,
                      1,
                      "",
-                     sx,
+                     sx + icon_ofs,
                      sy - layout->tile_h - 0.15f * UI_UNIT_X,
-                     width,
+                     width - icon_ofs,
                      textheight,
                      sfile->params->renamefile,
                      1.0f,
@@ -825,74 +807,19 @@ void file_draw_list(const bContext *C, ARegion *ar)
             sfile->files, file, FILE_SEL_REMOVE, FILE_SEL_EDITING, CHECK_ALL);
       }
     }
-
-    if (!(file_selflag & FILE_SEL_EDITING)) {
-      int tpos = (FILE_IMGDISPLAY == params->display) ? sy - layout->tile_h + layout->textheight :
-                                                        sy;
-      file_draw_string(sx + 1, tpos, file->name, (float)textwidth, textheight, align, text_col);
-    }
-
-    sx += (int)layout->column_widths[COLUMN_NAME] + column_space;
-    if (params->display == FILE_SHORTDISPLAY) {
-      if ((file->typeflag & (FILE_TYPE_BLENDER | FILE_TYPE_BLENDER_BACKUP)) ||
-          !(file->typeflag & (FILE_TYPE_DIR | FILE_TYPE_BLENDERLIB))) {
-        if ((file->entry->size_str[0] == '\0') || update_stat_strings) {
-          BLI_filelist_entry_size_to_string(
-              NULL, file->entry->size, small_size, file->entry->size_str);
-        }
-        file_draw_string(sx,
-                         sy,
-                         file->entry->size_str,
-                         layout->column_widths[COLUMN_SIZE],
-                         layout->tile_h,
-                         align,
-                         text_col);
-      }
-      sx += (int)layout->column_widths[COLUMN_SIZE] + column_space;
+    else {
+      const int txpos = (params->display == FILE_IMGDISPLAY) ? sx : sx + 1 + icon_ofs;
+      const int typos = (params->display == FILE_IMGDISPLAY) ?
+                            sy - layout->tile_h + layout->textheight :
+                            sy - layout->tile_border_y;
+      const int twidth = (params->display == FILE_IMGDISPLAY) ?
+                             textwidth :
+                             textwidth - 1 - icon_ofs - padx - layout->tile_border_x;
+      file_draw_string(txpos, typos, file->name, (float)twidth, textheight, align, text_col);
     }
-    else if (params->display == FILE_LONGDISPLAY) {
-      if (!(file->typeflag & FILE_TYPE_BLENDERLIB) && !FILENAME_IS_CURRPAR(file->relpath)) {
-        if ((file->entry->date_str[0] == '\0') || update_stat_strings) {
-          BLI_filelist_entry_datetime_to_string(
-              NULL, file->entry->time, small_size, file->entry->time_str, file->entry->date_str);
-        }
-        file_draw_string(sx,
-                         sy,
-                         file->entry->date_str,
-                         layout->column_widths[COLUMN_DATE],
-                         layout->tile_h,
-                         align,
-                         text_col);
-        sx += (int)layout->column_widths[COLUMN_DATE] + column_space;
-        file_draw_string(sx,
-                         sy,
-                         file->entry->time_str,
-                         layout->column_widths[COLUMN_TIME],
-                         layout->tile_h,
-                         align,
-                         text_col);
-        sx += (int)layout->column_widths[COLUMN_TIME] + column_space;
-      }
-      else {
-        sx += (int)layout->column_widths[COLUMN_DATE] + column_space;
-        sx += (int)layout->column_widths[COLUMN_TIME] + column_space;
-      }
 
-      if ((file->typeflag & (FILE_TYPE_BLENDER | FILE_TYPE_BLENDER_BACKUP)) ||
-          !(file->typeflag & (FILE_TYPE_DIR | FILE_TYPE_BLENDERLIB))) {
-        if ((file->entry->size_str[0] == '\0') || update_stat_strings) {
-          BLI_filelist_entry_size_to_string(
-              NULL, file->entry->size, small_size, file->entry->size_str);
-        }
-        file_draw_string(sx,
-                         sy,
-                         file->entry->size_str,
-                         layout->column_widths[COLUMN_SIZE],
-                         layout->tile_h,
-                         align,
-                         text_col);
-      }
-      sx += (int)layout->column_widths[COLUMN_SIZE] + column_space;
+    if (params->display != FILE_IMGDISPLAY) {
+      draw_details_columns(params, layout, file, sx, sy, text_col);
     }
   }
 
@@ -901,5 +828,11 @@ void file_draw_list(const bContext *C, ARegion *ar)
   UI_block_end(C, block);
   UI_block_draw(C, block);
 
+  /* Draw last, on top of file list. */
+  if (draw_columnheader) {
+    draw_columnheader_background(layout, v2d);
+    draw_columnheader_columns(params, layout, v2d, text_col);
+  }
+
   layout->curr_size = params->thumbnail_size;
 }
diff --git a/source/blender/editors/space_file/file_intern.h b/source/blender/editors/space_file/file_intern.h
index bad25511dd5..61f13098783 100644
--- a/source/blender/editors/space_file/file_intern.h
+++ b/source/blender/editors/space_file/file_intern.h
@@ -30,9 +30,11 @@ struct ARegion;
 struct ARegionType;
 struct FileSelectParams;
 struct SpaceFile;
+struct View2D;
 
 /* file_ops.c */
 struct ARegion *file_tools_region(struct ScrArea *sa);
+struct ARegion *file_tool_props_region(struct ScrArea *sa);
 
 /* file_draw.c */
 #define TILE_BORDER_X (UI_UNIT_X / 4)
@@ -42,9 +44,10 @@ struct ARegion *file_tools_region(struct ScrArea *sa);
 #define IMASEL_BUTTONS_HEIGHT (UI_UNIT_Y * 2)
 #define IMASEL_BUTTONS_MARGIN (UI_UNIT_Y / 6)
 
+#define ATTRIBUTE_COLUMN_PADDING (0.5f * UI_UNIT_X)
+
 #define SMALL_SIZE_CHECK(_size) ((_size) < 64) /* Related to FileSelectParams.thumbnail_size. */
 
-void file_draw_buttons(const bContext *C, ARegion *ar);
 void file_calc_previews(const bContext *C, ARegion *ar);
 void file_draw_list(const bContext *C, ARegion *ar);
 
@@ -64,6 +67,7 @@ typedef enum WalkSelectDirection {
 } WalkSelectDirections;
 
 void FILE_OT_highlight(struct wmOperatorType *ot);
+void FILE_OT_sort_column_ui_context(struct wmOperatorType *ot);
 void FILE_OT_select(struct wmOperatorType *ot);
 void FILE_OT_select_walk(struct wmOperatorType *ot);
 void FILE_OT_select_all(struct wmOperatorType *ot);
@@ -112,6 +116,16 @@ void file_operator_to_sfile(bContext *C, struct SpaceFile *sfile, struct wmOpera
 
 /* filesel.c */
 void fileselect_file_set(SpaceFile *sfile, const int index);
+bool file_attribute_column_type_enabled(const FileSelectParams *params,
+                                        FileAttributeColumnType column);
+bool file_attribute_column_header_is_inside(const struct View2D *v2d,
+                                            const FileLayout *layout,
+                                            int x,
+                                            int y);
+FileAttributeColumnType file_attribute_column_type_find_isect(const View2D *v2d,
+                                                              const FileSelectParams *params,
+                                                              FileLayout *layout,
+                                                              int x);
 float file_string_width(const char *str);
 
 float file_font_pointsize(void);
diff --git a/source/blender/editors/space_file/file_ops.c b/source/blender/editors/space_file/file_ops.c
index eb5f02b6e13..5168300d21e 100644
--- a/source/blender/editors/space_file/file_ops.c
+++ b/source/blender/editors/space_file/file_ops.c
@@ -78,7 +78,12 @@ static FileSelection find_file_mouse_rect(SpaceFile *sfile, ARegion *ar, const r
 
   BLI_rctf_rcti_copy(&rect_region_fl, rect_region);
 
+  /* Okay, manipulating v2d rects here is hacky...  */
+  v2d->mask.ymax -= sfile->layout->offset_top;
+  v2d->cur.ymax -= sfile->layout->offset_top;
   UI_view2d_region_to_view_rctf(v2d, &rect_region_fl, &rect_view_fl);
+  v2d->mask.ymax += sfile->layout->offset_top;
+  v2d->cur.ymax += sfile->layout->offset_top;
 
   BLI_rcti_init(&rect_view,
                 (int)(v2d->tot.xmin + rect_view_fl.xmin),
@@ -190,7 +195,6 @@ static FileSelect file_select_do(bContext *C, int selected_idx, bool do_diropen)
       const bool is_parent_dir = FILENAME_IS_PARENT(file->relpath);
 
       if (do_diropen == false) {
-        params->file[0] = '\0';
         retval = FILE_SELECT_DIR;
       }
       /* the path is too long and we are not going up! */
@@ -262,8 +266,8 @@ static void file_ensure_inside_viewbounds(ARegion *ar, SpaceFile *sfile, const i
     cur->ymax = cur->ymin + ar->winy;
   }
   /* up */
-  else if (cur->ymax < rect.ymax) {
-    cur->ymax = rect.ymax + layout->tile_border_y;
+  else if ((cur->ymax - layout->offset_top) < rect.ymax) {
+    cur->ymax = rect.ymax + layout->tile_border_y + layout->offset_top;
     cur->ymin = cur->ymax - ar->winy;
   }
   /* left - also use if tile is wider than viewbounds so view is aligned to file name */
@@ -278,7 +282,7 @@ static void file_ensure_inside_viewbounds(ARegion *ar, SpaceFile *sfile, const i
   }
   else {
     BLI_assert(cur->xmin <= rect.xmin && cur->xmax >= rect.xmax && cur->ymin <= rect.ymin &&
-               cur->ymax >= rect.ymax);
+               (cur->ymax - layout->offset_top) >= rect.ymax);
     changed = false;
   }
 
@@ -384,7 +388,7 @@ static int file_box_select_modal(bContext *C, wmOperator *op, const wmEvent *eve
   if (result == OPERATOR_RUNNING_MODAL) {
     WM_operator_properties_border_to_rcti(op, &rect);
 
-    BLI_rcti_isect(&(ar->v2d.mask), &rect, &rect);
+    ED_fileselect_layout_isect_rect(sfile->layout, &ar->v2d, &rect, &rect);
 
     sel = file_selection_get(C, &rect, 0);
     if ((sel.first != params->sel_first) || (sel.last != params->sel_last)) {
@@ -440,13 +444,13 @@ static int file_box_select_exec(bContext *C, wmOperator *op)
     file_deselect_all(sfile, FILE_SEL_SELECTED);
   }
 
-  BLI_rcti_isect(&(ar->v2d.mask), &rect, &rect);
+  ED_fileselect_layout_isect_rect(sfile->layout, &ar->v2d, &rect, &rect);
 
   ret = file_select(C, &rect, select ? FILE_SEL_ADD : FILE_SEL_REMOVE, false, false);
 
   /* unselect '..' parent entry - it's not supposed to be selected if more than
    * one file is selected */
-  filelist_entry_select_index_set(sfile->files, 0, FILE_SEL_REMOVE, FILE_SEL_SELECTED, CHECK_ALL);
+  filelist_entry_parent_select_set(sfile->files, FILE_SEL_REMOVE, FILE_SEL_SELECTED, CHECK_ALL);
 
   if (FILE_SELECT_DIR == ret) {
     WM_event_add_notifier(C, NC_SPACE | ND_SPACE_FILE_LIST, NULL);
@@ -493,7 +497,7 @@ static int file_select_invoke(bContext *C, wmOperator *op, const wmEvent *event)
   rect.xmin = rect.xmax = event->mval[0];
   rect.ymin = rect.ymax = event->mval[1];
 
-  if (!BLI_rcti_isect_pt(&ar->v2d.mask, rect.xmin, rect.ymin)) {
+  if (!ED_fileselect_layout_is_inside_pt(sfile->layout, &ar->v2d, rect.xmin, rect.ymin)) {
     return OPERATOR_CANCELLED;
   }
 
@@ -514,8 +518,7 @@ static int file_select_invoke(bContext *C, wmOperator *op, const wmEvent *event)
   if (extend) {
     /* unselect '..' parent entry - it's not supposed to be selected if more
      * than one file is selected */
-    filelist_entry_select_index_set(
-        sfile->files, 0, FILE_SEL_REMOVE, FILE_SEL_SELECTED, CHECK_ALL);
+    filelist_entry_parent_select_set(sfile->files, FILE_SEL_REMOVE, FILE_SEL_SELECTED, CHECK_ALL);
   }
 
   if (FILE_SELECT_DIR == ret) {
@@ -580,6 +583,11 @@ static bool file_walk_select_selection_set(bContext *C,
 
   BLI_assert(params);
 
+  if (numfiles == 0) {
+    /* No files visible, nothing to do. */
+    return false;
+  }
+
   if (has_selection) {
     if (extend && filelist_entry_select_index_get(files, active_old, CHECK_ALL) &&
         filelist_entry_select_index_get(files, active_new, CHECK_ALL)) {
@@ -609,7 +617,7 @@ static bool file_walk_select_selection_set(bContext *C,
     }
     /* select first file */
     else if (ELEM(direction, FILE_SELECT_WALK_DOWN, FILE_SELECT_WALK_RIGHT)) {
-      params->active_file = active = extend ? 1 : 0;
+      params->active_file = active = 0;
     }
     else {
       BLI_assert(0);
@@ -626,7 +634,7 @@ static bool file_walk_select_selection_set(bContext *C,
 
     /* unselect '..' parent entry - it's not supposed to be selected if more
      * than one file is selected */
-    filelist_entry_select_index_set(files, 0, FILE_SEL_REMOVE, FILE_SEL_SELECTED, CHECK_ALL);
+    filelist_entry_parent_select_set(files, FILE_SEL_REMOVE, FILE_SEL_SELECTED, CHECK_ALL);
   }
   else {
     /* deselect all first */
@@ -641,11 +649,6 @@ static bool file_walk_select_selection_set(bContext *C,
   if (fill) {
     FileSelection sel = {MIN2(active, last_sel), MAX2(active, last_sel)};
 
-    /* clamping selection to not include '..' parent entry */
-    if (sel.first == 0) {
-      sel.first = 1;
-    }
-
     /* fill selection between last and first selected file */
     filelist_entries_select_index_range_set(
         files, &sel, deselect ? FILE_SEL_REMOVE : FILE_SEL_ADD, FILE_SEL_SELECTED, CHECK_ALL);
@@ -653,6 +656,12 @@ static bool file_walk_select_selection_set(bContext *C,
     if (deselect) {
       filelist_entry_select_index_set(files, active, FILE_SEL_ADD, FILE_SEL_SELECTED, CHECK_ALL);
     }
+
+    /* unselect '..' parent entry - it's not supposed to be selected if more
+     * than one file is selected */
+    if ((sel.last - sel.first) > 1) {
+      filelist_entry_parent_select_set(files, FILE_SEL_REMOVE, FILE_SEL_SELECTED, CHECK_ALL);
+    }
   }
   else {
     filelist_entry_select_index_set(
@@ -688,10 +697,15 @@ static bool file_walk_select_do(bContext *C,
 
   /* *** get all needed files for handling selection *** */
 
+  if (numfiles == 0) {
+    /* No files visible, nothing to do. */
+    return false;
+  }
+
   if (has_selection) {
     ARegion *ar = CTX_wm_region(C);
     FileLayout *layout = ED_fileselect_get_layout(sfile, ar);
-    const int idx_shift = (layout->flag & FILE_LAYOUT_HOR) ? layout->rows : layout->columns;
+    const int idx_shift = (layout->flag & FILE_LAYOUT_HOR) ? layout->rows : layout->flow_columns;
 
     if ((layout->flag & FILE_LAYOUT_HOR && direction == FILE_SELECT_WALK_UP) ||
         (layout->flag & FILE_LAYOUT_VER && direction == FILE_SELECT_WALK_LEFT)) {
@@ -718,7 +732,7 @@ static bool file_walk_select_do(bContext *C,
       BLI_assert(0);
     }
 
-    if (!IN_RANGE(active_new, 0, numfiles)) {
+    if (!IN_RANGE(active_new, -1, numfiles)) {
       if (extend) {
         /* extend to invalid file -> abort */
         return false;
@@ -1185,7 +1199,7 @@ int file_highlight_set(SpaceFile *sfile, ARegion *ar, int mx, int my)
   mx -= ar->winrct.xmin;
   my -= ar->winrct.ymin;
 
-  if (BLI_rcti_isect_pt(&ar->v2d.mask, mx, my)) {
+  if (ED_fileselect_layout_is_inside_pt(sfile->layout, v2d, mx, my)) {
     float fx, fy;
     int highlight_file;
 
@@ -1234,6 +1248,53 @@ void FILE_OT_highlight(struct wmOperatorType *ot)
   ot->poll = ED_operator_file_active;
 }
 
+static int file_column_sort_ui_context_invoke(bContext *C,
+                                              wmOperator *UNUSED(op),
+                                              const wmEvent *event)
+{
+  const ARegion *ar = CTX_wm_region(C);
+  SpaceFile *sfile = CTX_wm_space_file(C);
+
+  if (file_attribute_column_header_is_inside(
+          &ar->v2d, sfile->layout, event->mval[0], event->mval[1])) {
+    const FileAttributeColumnType column_type = file_attribute_column_type_find_isect(
+        &ar->v2d, sfile->params, sfile->layout, event->mval[0]);
+
+    if (column_type != COLUMN_NONE) {
+      const FileAttributeColumn *column = &sfile->layout->attribute_columns[column_type];
+
+      if (column->sort_type != FILE_SORT_NONE) {
+        if (sfile->params->sort == column->sort_type) {
+          /* Already sorting by selected column -> toggle sort invert (three state logic). */
+          sfile->params->flag ^= FILE_SORT_INVERT;
+        }
+        else {
+          sfile->params->sort = column->sort_type;
+          sfile->params->flag &= ~FILE_SORT_INVERT;
+        }
+
+        WM_event_add_notifier(C, NC_SPACE | ND_SPACE_FILE_PARAMS, NULL);
+      }
+    }
+  }
+
+  return OPERATOR_PASS_THROUGH;
+}
+
+void FILE_OT_sort_column_ui_context(wmOperatorType *ot)
+{
+  /* identifiers */
+  ot->name = "Sort from Column";
+  ot->description = "Change sorting to use column under cursor";
+  ot->idname = "FILE_OT_sort_column_ui_context";
+
+  /* api callbacks */
+  ot->invoke = file_column_sort_ui_context_invoke;
+  ot->poll = ED_operator_file_active;
+
+  ot->flag = OPTYPE_INTERNAL;
+}
+
 int file_cancel_exec(bContext *C, wmOperator *UNUSED(unused))
 {
   wmWindowManager *wm = CTX_wm_manager(C);
@@ -1713,7 +1774,7 @@ static int file_smoothscroll_invoke(bContext *C, wmOperator *UNUSED(op), const w
   /* Number of items in a block (i.e. lines in a column in horizontal layout, or columns in a line
    * in vertical layout).
    */
-  const int items_block_size = is_horizontal ? sfile->layout->rows : sfile->layout->columns;
+  const int items_block_size = is_horizontal ? sfile->layout->rows : sfile->layout->flow_columns;
 
   /* Scroll offset is the first file in the row/column we are editing in. */
   if (sfile->scroll_offset == 0) {
@@ -1998,7 +2059,6 @@ void FILE_OT_directory_new(struct wmOperatorType *ot)
   ot->idname = "FILE_OT_directory_new";
 
   /* api callbacks */
-  ot->invoke = WM_operator_confirm;
   ot->exec = file_directory_new_exec;
   ot->poll = ED_operator_file_active; /* <- important, handler is on window level */
 
@@ -2260,10 +2320,29 @@ ARegion *file_tools_region(ScrArea *sa)
   arnew->regiontype = RGN_TYPE_TOOLS;
   arnew->alignment = RGN_ALIGN_LEFT;
 
-  ar = MEM_callocN(sizeof(ARegion), "tool props for file");
-  BLI_insertlinkafter(&sa->regionbase, arnew, ar);
-  ar->regiontype = RGN_TYPE_TOOL_PROPS;
-  ar->alignment = RGN_ALIGN_BOTTOM | RGN_SPLIT_PREV;
+  return arnew;
+}
+
+ARegion *file_tool_props_region(ScrArea *sa)
+{
+  ARegion *ar, *arnew;
+
+  if ((ar = BKE_area_find_region_type(sa, RGN_TYPE_TOOL_PROPS)) != NULL) {
+    return ar;
+  }
+
+  /* add subdiv level; after execute region */
+  ar = BKE_area_find_region_type(sa, RGN_TYPE_EXECUTE);
+
+  /* is error! */
+  if (ar == NULL) {
+    return NULL;
+  }
+
+  arnew = MEM_callocN(sizeof(ARegion), "tool props for file");
+  BLI_insertlinkafter(&sa->regionbase, ar, arnew);
+  arnew->regiontype = RGN_TYPE_TOOL_PROPS;
+  arnew->alignment = RGN_ALIGN_RIGHT;
 
   return arnew;
 }
@@ -2292,6 +2371,17 @@ void FILE_OT_bookmark_toggle(struct wmOperatorType *ot)
   ot->poll = ED_operator_file_active; /* <- important, handler is on window level */
 }
 
+static bool file_filenum_poll(bContext *C)
+{
+  SpaceFile *sfile = CTX_wm_space_file(C);
+
+  if (!ED_operator_file_active(C)) {
+    return false;
+  }
+
+  return sfile->params && (sfile->params->action_type == FILE_SAVE);
+}
+
 /**
  * Looks for a string of digits within name (using BLI_stringdec) and adjusts it by add.
  */
@@ -2349,7 +2439,7 @@ void FILE_OT_filenum(struct wmOperatorType *ot)
 
   /* api callbacks */
   ot->exec = file_filenum_exec;
-  ot->poll = ED_operator_file_active; /* <- important, handler is on window level */
+  ot->poll = file_filenum_poll;
 
   /* props */
   RNA_def_int(ot->srna, "increment", 1, -100, 100, "Increment", "", -100, 100);
diff --git a/source/blender/editors/space_file/file_panels.c b/source/blender/editors/space_file/file_panels.c
index d9a6e70121f..b41358f575f 100644
--- a/source/blender/editors/space_file/file_panels.c
+++ b/source/blender/editors/space_file/file_panels.c
@@ -103,6 +103,7 @@ void file_panels_register(ARegionType *art)
   strcpy(pt->idname, "FILE_PT_operator");
   strcpy(pt->label, N_("Operator"));
   strcpy(pt->translation_context, BLT_I18NCONTEXT_DEFAULT_BPYRNA);
+  pt->flag = PNL_NO_HEADER;
   pt->poll = file_panel_operator_poll;
   pt->draw_header = file_panel_operator_header;
   pt->draw = file_panel_operator;
diff --git a/source/blender/editors/space_file/filelist.c b/source/blender/editors/space_file/filelist.c
index f7dda1defe8..46302e0d087 100644
--- a/source/blender/editors/space_file/filelist.c
+++ b/source/blender/editors/space_file/filelist.c
@@ -329,25 +329,20 @@ enum {
   FL_IS_PENDING = 1 << 2,
   FL_NEED_SORTING = 1 << 3,
   FL_NEED_FILTERING = 1 << 4,
+  FL_SORT_INVERT = 1 << 5,
 };
 
-#define SPECIAL_IMG_SIZE 48
-#define SPECIAL_IMG_ROWS 4
-#define SPECIAL_IMG_COLS 4
+#define SPECIAL_IMG_SIZE 256
+#define SPECIAL_IMG_ROWS 1
+#define SPECIAL_IMG_COLS 6
 
 enum {
-  SPECIAL_IMG_FOLDER = 0,
-  SPECIAL_IMG_PARENT = 1,
-  SPECIAL_IMG_REFRESH = 2,
-  SPECIAL_IMG_BLENDFILE = 3,
-  SPECIAL_IMG_SOUNDFILE = 4,
-  SPECIAL_IMG_MOVIEFILE = 5,
-  SPECIAL_IMG_PYTHONFILE = 6,
-  SPECIAL_IMG_TEXTFILE = 7,
-  SPECIAL_IMG_FONTFILE = 8,
-  SPECIAL_IMG_UNKNOWNFILE = 9,
-  SPECIAL_IMG_LOADING = 10,
-  SPECIAL_IMG_BACKUP = 11,
+  SPECIAL_IMG_DOCUMENT = 0,
+  SPECIAL_IMG_FOLDER = 1,
+  SPECIAL_IMG_PARENT = 2,
+  SPECIAL_IMG_DRIVE_FIXED = 3,
+  SPECIAL_IMG_DRIVE_ATTACHED = 4,
+  SPECIAL_IMG_DRIVE_REMOTE = 5,
   SPECIAL_IMG_MAX,
 };
 
@@ -369,6 +364,19 @@ static void filelist_cache_clear(FileListEntryCache *cache, size_t new_size);
 
 /* ********** Sort helpers ********** */
 
+struct FileSortData {
+  bool inverted;
+};
+
+static int compare_apply_inverted(int val, const struct FileSortData *sort_data)
+{
+  return sort_data->inverted ? -val : val;
+}
+
+/**
+ * Handles inverted sorting itself (currently there's nothing to invert), so if this returns non-0,
+ * it should be used as-is and not inverted.
+ */
 static int compare_direntry_generic(const FileListInternEntry *entry1,
                                     const FileListInternEntry *entry2)
 {
@@ -420,10 +428,11 @@ static int compare_direntry_generic(const FileListInternEntry *entry1,
   return 0;
 }
 
-static int compare_name(void *UNUSED(user_data), const void *a1, const void *a2)
+static int compare_name(void *user_data, const void *a1, const void *a2)
 {
   const FileListInternEntry *entry1 = a1;
   const FileListInternEntry *entry2 = a2;
+  const struct FileSortData *sort_data = user_data;
   char *name1, *name2;
   int ret;
 
@@ -434,13 +443,14 @@ static int compare_name(void *UNUSED(user_data), const void *a1, const void *a2)
   name1 = entry1->name;
   name2 = entry2->name;
 
-  return BLI_natstrcmp(name1, name2);
+  return compare_apply_inverted(BLI_strcasecmp_natural(name1, name2), sort_data);
 }
 
-static int compare_date(void *UNUSED(user_data), const void *a1, const void *a2)
+static int compare_date(void *user_data, const void *a1, const void *a2)
 {
   const FileListInternEntry *entry1 = a1;
   const FileListInternEntry *entry2 = a2;
+  const struct FileSortData *sort_data = user_data;
   char *name1, *name2;
   int64_t time1, time2;
   int ret;
@@ -452,22 +462,23 @@ static int compare_date(void *UNUSED(user_data), const void *a1, const void *a2)
   time1 = (int64_t)entry1->st.st_mtime;
   time2 = (int64_t)entry2->st.st_mtime;
   if (time1 < time2) {
-    return 1;
+    return compare_apply_inverted(1, sort_data);
   }
   if (time1 > time2) {
-    return -1;
+    return compare_apply_inverted(-1, sort_data);
   }
 
   name1 = entry1->name;
   name2 = entry2->name;
 
-  return BLI_natstrcmp(name1, name2);
+  return compare_apply_inverted(BLI_strcasecmp_natural(name1, name2), sort_data);
 }
 
-static int compare_size(void *UNUSED(user_data), const void *a1, const void *a2)
+static int compare_size(void *user_data, const void *a1, const void *a2)
 {
   const FileListInternEntry *entry1 = a1;
   const FileListInternEntry *entry2 = a2;
+  const struct FileSortData *sort_data = user_data;
   char *name1, *name2;
   uint64_t size1, size2;
   int ret;
@@ -479,22 +490,23 @@ static int compare_size(void *UNUSED(user_data), const void *a1, const void *a2)
   size1 = entry1->st.st_size;
   size2 = entry2->st.st_size;
   if (size1 < size2) {
-    return 1;
+    return compare_apply_inverted(1, sort_data);
   }
   if (size1 > size2) {
-    return -1;
+    return compare_apply_inverted(-1, sort_data);
   }
 
   name1 = entry1->name;
   name2 = entry2->name;
 
-  return BLI_natstrcmp(name1, name2);
+  return compare_apply_inverted(BLI_strcasecmp_natural(name1, name2), sort_data);
 }
 
-static int compare_extension(void *UNUSED(user_data), const void *a1, const void *a2)
+static int compare_extension(void *user_data, const void *a1, const void *a2)
 {
   const FileListInternEntry *entry1 = a1;
   const FileListInternEntry *entry2 = a2;
+  const struct FileSortData *sort_data = user_data;
   char *name1, *name2;
   int ret;
 
@@ -516,10 +528,10 @@ static int compare_extension(void *UNUSED(user_data), const void *a1, const void
       return -1;
     }
     if (entry1->blentype < entry2->blentype) {
-      return -1;
+      return compare_apply_inverted(-1, sort_data);
     }
     if (entry1->blentype > entry2->blentype) {
-      return 1;
+      return compare_apply_inverted(1, sort_data);
     }
   }
   else {
@@ -539,48 +551,58 @@ static int compare_extension(void *UNUSED(user_data), const void *a1, const void
     }
 
     if ((ret = BLI_strcasecmp(sufix1, sufix2))) {
-      return ret;
+      return compare_apply_inverted(ret, sort_data);
     }
   }
 
   name1 = entry1->name;
   name2 = entry2->name;
 
-  return BLI_natstrcmp(name1, name2);
+  return compare_apply_inverted(BLI_strcasecmp_natural(name1, name2), sort_data);
 }
 
 void filelist_sort(struct FileList *filelist)
 {
   if ((filelist->flags & FL_NEED_SORTING) && (filelist->sort != FILE_SORT_NONE)) {
+    void *sort_cb = NULL;
+
     switch (filelist->sort) {
       case FILE_SORT_ALPHA:
-        BLI_listbase_sort_r(&filelist->filelist_intern.entries, compare_name, NULL);
+        sort_cb = compare_name;
         break;
       case FILE_SORT_TIME:
-        BLI_listbase_sort_r(&filelist->filelist_intern.entries, compare_date, NULL);
+        sort_cb = compare_date;
         break;
       case FILE_SORT_SIZE:
-        BLI_listbase_sort_r(&filelist->filelist_intern.entries, compare_size, NULL);
+        sort_cb = compare_size;
         break;
       case FILE_SORT_EXTENSION:
-        BLI_listbase_sort_r(&filelist->filelist_intern.entries, compare_extension, NULL);
+        sort_cb = compare_extension;
         break;
       case FILE_SORT_NONE: /* Should never reach this point! */
       default:
         BLI_assert(0);
         break;
     }
+    BLI_listbase_sort_r(
+        &filelist->filelist_intern.entries,
+        sort_cb,
+        &(struct FileSortData){.inverted = (filelist->flags & FL_SORT_INVERT) != 0});
 
     filelist_filter_clear(filelist);
     filelist->flags &= ~FL_NEED_SORTING;
   }
 }
 
-void filelist_setsorting(struct FileList *filelist, const short sort)
+void filelist_setsorting(struct FileList *filelist, const short sort, bool invert_sort)
 {
-  if (filelist->sort != sort) {
+  const bool was_invert_sort = filelist->flags & FL_SORT_INVERT;
+
+  if ((filelist->sort != sort) || (was_invert_sort != invert_sort)) {
     filelist->sort = sort;
     filelist->flags |= FL_NEED_SORTING;
+    filelist->flags = invert_sort ? (filelist->flags | FL_SORT_INVERT) :
+                                    (filelist->flags & ~FL_SORT_INVERT);
   }
 }
 
@@ -635,9 +657,9 @@ static bool is_filtered_file(FileListInternEntry *file,
 {
   bool is_filtered = !is_hidden_file(file->relpath, filter);
 
-  if (is_filtered && (filter->flags & FLF_DO_FILTER) && !FILENAME_IS_CURRPAR(file->relpath)) {
+  if (is_filtered && !FILENAME_IS_CURRPAR(file->relpath)) {
     /* We only check for types if some type are enabled in filtering. */
-    if (filter->filter) {
+    if (filter->filter && (filter->flags & FLF_DO_FILTER)) {
       if (file->typeflag & FILE_TYPE_DIR) {
         if (file->typeflag &
             (FILE_TYPE_BLENDERLIB | FILE_TYPE_BLENDER | FILE_TYPE_BLENDER_BACKUP)) {
@@ -657,6 +679,7 @@ static bool is_filtered_file(FileListInternEntry *file,
         }
       }
     }
+    /* If there's a filter string, apply it as filter even if FLF_DO_FILTER is not set. */
     if (is_filtered && (filter->filter_search[0] != '\0')) {
       if (fnmatch(filter->filter_search, file->relpath, FNM_CASEFOLD) != 0) {
         is_filtered = false;
@@ -676,9 +699,9 @@ static bool is_filtered_lib(FileListInternEntry *file, const char *root, FileLis
 
   if (BLO_library_path_explode(path, dir, &group, &name)) {
     is_filtered = !is_hidden_file(file->relpath, filter);
-    if (is_filtered && (filter->flags & FLF_DO_FILTER) && !FILENAME_IS_CURRPAR(file->relpath)) {
+    if (is_filtered && !FILENAME_IS_CURRPAR(file->relpath)) {
       /* We only check for types if some type are enabled in filtering. */
-      if (filter->filter || filter->filter_id) {
+      if ((filter->filter || filter->filter_id) && (filter->flags & FLF_DO_FILTER)) {
         if (file->typeflag & FILE_TYPE_DIR) {
           if (file->typeflag &
               (FILE_TYPE_BLENDERLIB | FILE_TYPE_BLENDER | FILE_TYPE_BLENDER_BACKUP)) {
@@ -704,6 +727,7 @@ static bool is_filtered_lib(FileListInternEntry *file, const char *root, FileLis
           }
         }
       }
+      /* If there's a filter string, apply it as filter even if FLF_DO_FILTER is not set. */
       if (is_filtered && (filter->filter_search[0] != '\0')) {
         if (fnmatch(filter->filter_search, file->relpath, FNM_CASEFOLD) != 0) {
           is_filtered = false;
@@ -904,42 +928,12 @@ static ImBuf *filelist_geticon_image_ex(const unsigned int typeflag, const char
     if (FILENAME_IS_PARENT(relpath)) {
       ibuf = gSpecialFileImages[SPECIAL_IMG_PARENT];
     }
-    else if (FILENAME_IS_CURRENT(relpath)) {
-      ibuf = gSpecialFileImages[SPECIAL_IMG_REFRESH];
-    }
     else {
       ibuf = gSpecialFileImages[SPECIAL_IMG_FOLDER];
     }
   }
-  else if (typeflag & FILE_TYPE_BLENDER) {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_BLENDFILE];
-  }
-  else if (typeflag & FILE_TYPE_BLENDERLIB) {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_UNKNOWNFILE];
-  }
-  else if (typeflag & (FILE_TYPE_MOVIE)) {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_MOVIEFILE];
-  }
-  else if (typeflag & FILE_TYPE_SOUND) {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_SOUNDFILE];
-  }
-  else if (typeflag & FILE_TYPE_PYSCRIPT) {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_PYTHONFILE];
-  }
-  else if (typeflag & FILE_TYPE_FTFONT) {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_FONTFILE];
-  }
-  else if (typeflag & FILE_TYPE_TEXT) {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_TEXTFILE];
-  }
-  else if (typeflag & FILE_TYPE_IMAGE) {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_LOADING];
-  }
-  else if (typeflag & FILE_TYPE_BLENDER_BACKUP) {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_BACKUP];
-  }
   else {
-    ibuf = gSpecialFileImages[SPECIAL_IMG_UNKNOWNFILE];
+    ibuf = gSpecialFileImages[SPECIAL_IMG_DOCUMENT];
   }
 
   return ibuf;
@@ -1001,10 +995,13 @@ static int filelist_geticon_ex(const int typeflag,
     return ICON_FILE_BLANK;
   }
   else if (typeflag & FILE_TYPE_COLLADA) {
-    return ICON_FILE_BLANK;
+    return ICON_FILE_3D;
   }
   else if (typeflag & FILE_TYPE_ALEMBIC) {
-    return ICON_FILE_BLANK;
+    return ICON_FILE_3D;
+  }
+  else if (typeflag & FILE_TYPE_OBJECT_IO) {
+    return ICON_FILE_3D;
   }
   else if (typeflag & FILE_TYPE_TEXT) {
     return ICON_FILE_TEXT;
@@ -1243,7 +1240,8 @@ static void filelist_cache_previews_clear(FileListEntryCache *cache)
     BLI_task_pool_cancel(cache->previews_pool);
 
     while ((preview = BLI_thread_queue_pop_timeout(cache->previews_done, 0))) {
-      // printf("%s: DONE %d - %s - %p\n", __func__, preview->index, preview->path, preview->img);
+      // printf("%s: DONE %d - %s - %p\n", __func__, preview->index, preview->path,
+      // preview->img);
       if (preview->img) {
         IMB_freeImBuf(preview->img);
       }
@@ -2128,6 +2126,9 @@ int ED_path_extension_type(const char *path)
   else if (BLI_path_extension_check(path, ".abc")) {
     return FILE_TYPE_ALEMBIC;
   }
+  else if (BLI_path_extension_check_n(path, ".obj", ".3ds", ".fbx", ".glb", ".gltf", NULL)) {
+    return FILE_TYPE_OBJECT_IO;
+  }
   else if (BLI_path_extension_check_array(path, imb_ext_image)) {
     return FILE_TYPE_IMAGE;
   }
@@ -2177,9 +2178,9 @@ int ED_file_extension_icon(const char *path)
     case FILE_TYPE_BTX:
       return ICON_FILE_BLANK;
     case FILE_TYPE_COLLADA:
-      return ICON_FILE_BLANK;
     case FILE_TYPE_ALEMBIC:
-      return ICON_FILE_BLANK;
+    case FILE_TYPE_OBJECT_IO:
+      return ICON_FILE_3D;
     case FILE_TYPE_TEXT:
       return ICON_FILE_TEXT;
     default:
@@ -2298,6 +2299,19 @@ unsigned int filelist_entry_select_index_get(FileList *filelist,
   return 0;
 }
 
+/**
+ * Set selection of the '..' parent entry, but only if it's actually visible.
+ */
+void filelist_entry_parent_select_set(FileList *filelist,
+                                      FileSelType select,
+                                      unsigned int flag,
+                                      FileCheckType check)
+{
+  if ((filelist->filter_data.flags & FLF_HIDE_PARENT) == 0) {
+    filelist_entry_select_index_set(filelist, 0, select, flag, check);
+  }
+}
+
 /* WARNING! dir must be FILE_MAX_LIBEXTRA long! */
 bool filelist_islibrary(struct FileList *filelist, char *dir, char **group)
 {
diff --git a/source/blender/editors/space_file/filelist.h b/source/blender/editors/space_file/filelist.h
index caf77246797..9b1107294ff 100644
--- a/source/blender/editors/space_file/filelist.h
+++ b/source/blender/editors/space_file/filelist.h
@@ -55,7 +55,7 @@ void folderlist_pushdir(struct ListBase *folderlist, const char *dir);
 const char *folderlist_peeklastdir(struct ListBase *folderdist);
 int folderlist_clear_next(struct SpaceFile *sfile);
 
-void filelist_setsorting(struct FileList *filelist, const short sort);
+void filelist_setsorting(struct FileList *filelist, const short sort, bool invert_sort);
 void filelist_sort(struct FileList *filelist);
 
 void filelist_setfilter_options(struct FileList *filelist,
@@ -117,6 +117,10 @@ unsigned int filelist_entry_select_get(struct FileList *filelist,
 unsigned int filelist_entry_select_index_get(struct FileList *filelist,
                                              const int index,
                                              FileCheckType check);
+void filelist_entry_parent_select_set(struct FileList *filelist,
+                                      FileSelType select,
+                                      unsigned int flag,
+                                      FileCheckType check);
 
 void filelist_setrecursion(struct FileList *filelist, const int recursion_level);
 
diff --git a/source/blender/editors/space_file/filesel.c b/source/blender/editors/space_file/filesel.c
index db42d007b8e..3223fe0c6ce 100644
--- a/source/blender/editors/space_file/filesel.c
+++ b/source/blender/editors/space_file/filesel.c
@@ -49,6 +49,8 @@
 #include "BLI_utildefines.h"
 #include "BLI_fnmatch.h"
 
+#include "BLT_translation.h"
+
 #include "BKE_appdir.h"
 #include "BKE_context.h"
 #include "BKE_main.h"
@@ -69,6 +71,8 @@
 #include "file_intern.h"
 #include "filelist.h"
 
+#define VERTLIST_MAJORCOLUMN_WIDTH (25 * UI_UNIT_X)
+
 FileSelectParams *ED_fileselect_get_params(struct SpaceFile *sfile)
 {
   if (!sfile->params) {
@@ -99,6 +103,8 @@ short ED_fileselect_set_params(SpaceFile *sfile)
     sfile->params->filter_glob[0] = '\0';
     /* set the default thumbnails size */
     sfile->params->thumbnail_size = 128;
+    /* Show size column by default. */
+    sfile->params->details_flags = FILE_DETAILS_SIZE | FILE_DETAILS_DATETIME;
   }
 
   params = sfile->params;
@@ -161,6 +167,10 @@ short ED_fileselect_set_params(SpaceFile *sfile)
       params->flag &= ~FILE_DIRSEL_ONLY;
     }
 
+    if ((prop = RNA_struct_find_property(op->ptr, "hide_props_region"))) {
+      params->flag |= RNA_property_boolean_get(op->ptr, prop) ? FILE_HIDE_TOOL_PROPS : 0;
+    }
+
     params->filter = 0;
     if ((prop = RNA_struct_find_property(op->ptr, "filter_blender"))) {
       params->filter |= RNA_property_boolean_get(op->ptr, prop) ? FILE_TYPE_BLENDER : 0;
@@ -261,6 +271,10 @@ short ED_fileselect_set_params(SpaceFile *sfile)
       params->sort = FILE_SORT_ALPHA;
     }
 
+    if ((prop = RNA_struct_find_property(op->ptr, "action_type"))) {
+      params->action_type = RNA_property_enum_get(op->ptr, prop);
+    }
+
     if (params->display == FILE_DEFAULTDISPLAY) {
       if (params->display_previous == FILE_DEFAULTDISPLAY) {
         if (U.uiflag & USER_SHOW_THUMBNAILS) {
@@ -268,11 +282,11 @@ short ED_fileselect_set_params(SpaceFile *sfile)
             params->display = FILE_IMGDISPLAY;
           }
           else {
-            params->display = FILE_SHORTDISPLAY;
+            params->display = FILE_VERTICALDISPLAY;
           }
         }
         else {
-          params->display = FILE_SHORTDISPLAY;
+          params->display = FILE_VERTICALDISPLAY;
         }
       }
       else {
@@ -293,7 +307,7 @@ short ED_fileselect_set_params(SpaceFile *sfile)
     params->type = FILE_UNIX;
     params->flag |= FILE_HIDE_DOT;
     params->flag &= ~FILE_DIRSEL_ONLY;
-    params->display = FILE_SHORTDISPLAY;
+    params->display = FILE_VERTICALDISPLAY;
     params->display_previous = FILE_DEFAULTDISPLAY;
     params->sort = FILE_SORT_ALPHA;
     params->filter = 0;
@@ -344,7 +358,7 @@ void ED_fileselect_reset_params(SpaceFile *sfile)
 void fileselect_file_set(SpaceFile *sfile, const int index)
 {
   const struct FileDirEntry *file = filelist_file(sfile->files, index);
-  if (file && file->relpath && file->relpath[0] && !(file->typeflag & FILE_TYPE_FOLDER)) {
+  if (file && file->relpath && file->relpath[0] && !(file->typeflag & FILE_TYPE_DIR)) {
     BLI_strncpy(sfile->params->file, file->relpath, FILE_MAXFILE);
   }
 }
@@ -372,10 +386,10 @@ int ED_fileselect_layout_numfiles(FileLayout *layout, ARegion *ar)
   }
   else {
     const int y_item = layout->tile_h + (2 * layout->tile_border_y);
-    const int y_view = (int)(BLI_rctf_size_y(&ar->v2d.cur));
+    const int y_view = (int)(BLI_rctf_size_y(&ar->v2d.cur)) - layout->offset_top;
     const int y_over = y_item - (y_view % y_item);
     numfiles = (int)((float)(y_view + y_over) / (float)(y_item));
-    return numfiles * layout->columns;
+    return numfiles * layout->flow_columns;
   }
 }
 
@@ -395,19 +409,19 @@ FileSelection ED_fileselect_layout_offset_rect(FileLayout *layout, const rcti *r
   }
 
   colmin = (rect->xmin) / (layout->tile_w + 2 * layout->tile_border_x);
-  rowmin = (rect->ymin) / (layout->tile_h + 2 * layout->tile_border_y);
+  rowmin = (rect->ymin - layout->offset_top) / (layout->tile_h + 2 * layout->tile_border_y);
   colmax = (rect->xmax) / (layout->tile_w + 2 * layout->tile_border_x);
-  rowmax = (rect->ymax) / (layout->tile_h + 2 * layout->tile_border_y);
+  rowmax = (rect->ymax - layout->offset_top) / (layout->tile_h + 2 * layout->tile_border_y);
 
-  if (is_inside(colmin, rowmin, layout->columns, layout->rows) ||
-      is_inside(colmax, rowmax, layout->columns, layout->rows)) {
-    CLAMP(colmin, 0, layout->columns - 1);
+  if (is_inside(colmin, rowmin, layout->flow_columns, layout->rows) ||
+      is_inside(colmax, rowmax, layout->flow_columns, layout->rows)) {
+    CLAMP(colmin, 0, layout->flow_columns - 1);
     CLAMP(rowmin, 0, layout->rows - 1);
-    CLAMP(colmax, 0, layout->columns - 1);
+    CLAMP(colmax, 0, layout->flow_columns - 1);
     CLAMP(rowmax, 0, layout->rows - 1);
   }
 
-  if ((colmin > layout->columns - 1) || (rowmin > layout->rows - 1)) {
+  if ((colmin > layout->flow_columns - 1) || (rowmin > layout->rows - 1)) {
     sel.first = -1;
   }
   else {
@@ -415,10 +429,10 @@ FileSelection ED_fileselect_layout_offset_rect(FileLayout *layout, const rcti *r
       sel.first = layout->rows * colmin + rowmin;
     }
     else {
-      sel.first = colmin + layout->columns * rowmin;
+      sel.first = colmin + layout->flow_columns * rowmin;
     }
   }
-  if ((colmax > layout->columns - 1) || (rowmax > layout->rows - 1)) {
+  if ((colmax > layout->flow_columns - 1) || (rowmax > layout->rows - 1)) {
     sel.last = -1;
   }
   else {
@@ -426,7 +440,7 @@ FileSelection ED_fileselect_layout_offset_rect(FileLayout *layout, const rcti *r
       sel.last = layout->rows * colmax + rowmax;
     }
     else {
-      sel.last = colmax + layout->columns * rowmax;
+      sel.last = colmax + layout->flow_columns * rowmax;
     }
   }
 
@@ -443,9 +457,9 @@ int ED_fileselect_layout_offset(FileLayout *layout, int x, int y)
   }
 
   offsetx = (x) / (layout->tile_w + 2 * layout->tile_border_x);
-  offsety = (y) / (layout->tile_h + 2 * layout->tile_border_y);
+  offsety = (y - layout->offset_top) / (layout->tile_h + 2 * layout->tile_border_y);
 
-  if (offsetx > layout->columns - 1) {
+  if (offsetx > layout->flow_columns - 1) {
     return -1;
   }
   if (offsety > layout->rows - 1) {
@@ -456,25 +470,121 @@ int ED_fileselect_layout_offset(FileLayout *layout, int x, int y)
     active_file = layout->rows * offsetx + offsety;
   }
   else {
-    active_file = offsetx + layout->columns * offsety;
+    active_file = offsetx + layout->flow_columns * offsety;
   }
   return active_file;
 }
 
+/**
+ * Get the currently visible bounds of the layout in screen space. Matches View2D.mask minus the
+ * top column-header row.
+ */
+void ED_fileselect_layout_maskrect(const FileLayout *layout, const View2D *v2d, rcti *r_rect)
+{
+  *r_rect = v2d->mask;
+  r_rect->ymax -= layout->offset_top;
+}
+
+bool ED_fileselect_layout_is_inside_pt(const FileLayout *layout, const View2D *v2d, int x, int y)
+{
+  rcti maskrect;
+  ED_fileselect_layout_maskrect(layout, v2d, &maskrect);
+  return BLI_rcti_isect_pt(&maskrect, x, y);
+}
+
+bool ED_fileselect_layout_isect_rect(const FileLayout *layout,
+                                     const View2D *v2d,
+                                     const rcti *rect,
+                                     rcti *r_dst)
+{
+  rcti maskrect;
+  ED_fileselect_layout_maskrect(layout, v2d, &maskrect);
+  return BLI_rcti_isect(&maskrect, rect, r_dst);
+}
+
 void ED_fileselect_layout_tilepos(FileLayout *layout, int tile, int *x, int *y)
 {
   if (layout->flag == FILE_LAYOUT_HOR) {
     *x = layout->tile_border_x +
          (tile / layout->rows) * (layout->tile_w + 2 * layout->tile_border_x);
-    *y = layout->tile_border_y +
+    *y = layout->offset_top + layout->tile_border_y +
          (tile % layout->rows) * (layout->tile_h + 2 * layout->tile_border_y);
   }
   else {
     *x = layout->tile_border_x +
-         ((tile) % layout->columns) * (layout->tile_w + 2 * layout->tile_border_x);
-    *y = layout->tile_border_y +
-         ((tile) / layout->columns) * (layout->tile_h + 2 * layout->tile_border_y);
+         ((tile) % layout->flow_columns) * (layout->tile_w + 2 * layout->tile_border_x);
+    *y = layout->offset_top + layout->tile_border_y +
+         ((tile) / layout->flow_columns) * (layout->tile_h + 2 * layout->tile_border_y);
+  }
+}
+
+/**
+ * Check if the region coordinate defined by \a x and \a y are inside the column header.
+ */
+bool file_attribute_column_header_is_inside(const View2D *v2d,
+                                            const FileLayout *layout,
+                                            int x,
+                                            int y)
+{
+  rcti header_rect = v2d->mask;
+  header_rect.ymin = header_rect.ymax - layout->attribute_column_header_h;
+  return BLI_rcti_isect_pt(&header_rect, x, y);
+}
+
+bool file_attribute_column_type_enabled(const FileSelectParams *params,
+                                        FileAttributeColumnType column)
+{
+  switch (column) {
+    case COLUMN_NAME:
+      /* Always enabled */
+      return true;
+    case COLUMN_DATETIME:
+      return (params->details_flags & FILE_DETAILS_DATETIME) != 0;
+    case COLUMN_SIZE:
+      return (params->details_flags & FILE_DETAILS_SIZE) != 0;
+    default:
+      return false;
+  }
+}
+
+/**
+ * Find the column type at region coordinate given by \a x (y doesn't matter for this).
+ */
+FileAttributeColumnType file_attribute_column_type_find_isect(const View2D *v2d,
+                                                              const FileSelectParams *params,
+                                                              FileLayout *layout,
+                                                              int x)
+{
+  float mx, my;
+  int offset_tile;
+
+  UI_view2d_region_to_view(v2d, x, v2d->mask.ymax - layout->offset_top - 1, &mx, &my);
+  offset_tile = ED_fileselect_layout_offset(
+      layout, (int)(v2d->tot.xmin + mx), (int)(v2d->tot.ymax - my));
+  if (offset_tile > -1) {
+    int tile_x, tile_y;
+    int pos_x = 0;
+    int rel_x; /* x relative to the hovered tile */
+
+    ED_fileselect_layout_tilepos(layout, offset_tile, &tile_x, &tile_y);
+    /* Column header drawing doesn't use left tile border, so subtract it. */
+    rel_x = mx - (tile_x - layout->tile_border_x);
+
+    for (FileAttributeColumnType column = 0; column < ATTRIBUTE_COLUMN_MAX; column++) {
+      if (!file_attribute_column_type_enabled(params, column)) {
+        continue;
+      }
+      const int width = layout->attribute_columns[column].width;
+
+      if (IN_RANGE(rel_x, pos_x, pos_x + width)) {
+        return column;
+      }
+
+      pos_x += width;
+    }
   }
+
+  return COLUMN_NONE;
 }
 
 float file_string_width(const char *str)
@@ -512,20 +622,52 @@ float file_font_pointsize(void)
 #endif
 }
 
-static void column_widths(FileSelectParams *params, struct FileLayout *layout)
+static void file_attribute_columns_widths(const FileSelectParams *params, FileLayout *layout)
 {
-  int i;
+  FileAttributeColumn *columns = layout->attribute_columns;
   const bool small_size = SMALL_SIZE_CHECK(params->thumbnail_size);
+  const int pad = small_size ? 0 : ATTRIBUTE_COLUMN_PADDING * 2;
 
-  for (i = 0; i < MAX_FILE_COLUMN; ++i) {
-    layout->column_widths[i] = 0;
+  for (int i = 0; i < ATTRIBUTE_COLUMN_MAX; ++i) {
+    layout->attribute_columns[i].width = 0;
   }
 
-  layout->column_widths[COLUMN_NAME] = ((float)params->thumbnail_size / 8.0f) * UI_UNIT_X;
   /* Biggest possible reasonable values... */
-  layout->column_widths[COLUMN_DATE] = file_string_width(small_size ? "23/08/89" : "23-Dec-89");
-  layout->column_widths[COLUMN_TIME] = file_string_width("23:59");
-  layout->column_widths[COLUMN_SIZE] = file_string_width(small_size ? "98.7 M" : "98.7 MiB");
+  columns[COLUMN_DATETIME].width = file_string_width(small_size ? "23/08/89" :
+                                                                  "23 Dec 6789, 23:59") +
+                                   pad;
+  columns[COLUMN_SIZE].width = file_string_width(small_size ? "98.7 M" : "098.7 MB") + pad;
+  if (params->display == FILE_IMGDISPLAY) {
+    columns[COLUMN_NAME].width = ((float)params->thumbnail_size / 8.0f) * UI_UNIT_X;
+  }
+  /* Name column uses remaining width */
+  else {
+    int remwidth = layout->tile_w;
+    for (FileAttributeColumnType column_type = ATTRIBUTE_COLUMN_MAX - 1; column_type >= 0;
+         column_type--) {
+      if ((column_type == COLUMN_NAME) ||
+          !file_attribute_column_type_enabled(params, column_type)) {
+        continue;
+      }
+      remwidth -= columns[column_type].width;
+    }
+    columns[COLUMN_NAME].width = remwidth;
+  }
+}
+
+static void file_attribute_columns_init(const FileSelectParams *params, FileLayout *layout)
+{
+  file_attribute_columns_widths(params, layout);
+
+  layout->attribute_columns[COLUMN_NAME].name = N_("Name");
+  layout->attribute_columns[COLUMN_NAME].sort_type = FILE_SORT_ALPHA;
+  layout->attribute_columns[COLUMN_NAME].text_align = UI_STYLE_TEXT_LEFT;
+  layout->attribute_columns[COLUMN_DATETIME].name = N_("Date Modified");
+  layout->attribute_columns[COLUMN_DATETIME].sort_type = FILE_SORT_TIME;
+  layout->attribute_columns[COLUMN_DATETIME].text_align = UI_STYLE_TEXT_LEFT;
+  layout->attribute_columns[COLUMN_SIZE].name = N_("Size");
+  layout->attribute_columns[COLUMN_SIZE].sort_type = FILE_SORT_SIZE;
+  layout->attribute_columns[COLUMN_SIZE].text_align = UI_STYLE_TEXT_RIGHT;
 }
 
 void ED_fileselect_init_layout(struct SpaceFile *sfile, ARegion *ar)
@@ -533,7 +675,6 @@ void ED_fileselect_init_layout(struct SpaceFile *sfile, ARegion *ar)
   FileSelectParams *params = ED_fileselect_get_params(sfile);
   FileLayout *layout = NULL;
   View2D *v2d = &ar->v2d;
-  int maxlen = 0;
   int numfiles;
   int textheight;
 
@@ -560,57 +701,66 @@ void ED_fileselect_init_layout(struct SpaceFile *sfile, ARegion *ar)
     layout->tile_w = layout->prv_w + 2 * layout->prv_border_x;
     layout->tile_h = layout->prv_h + 2 * layout->prv_border_y + textheight;
     layout->width = (int)(BLI_rctf_size_x(&v2d->cur) - 2 * layout->tile_border_x);
-    layout->columns = layout->width / (layout->tile_w + 2 * layout->tile_border_x);
-    if (layout->columns > 0) {
-      layout->rows = numfiles / layout->columns + 1;  // XXX dirty, modulo is zero
+    layout->flow_columns = layout->width / (layout->tile_w + 2 * layout->tile_border_x);
+    layout->attribute_column_header_h = 0;
+    layout->offset_top = 0;
+    if (layout->flow_columns > 0) {
+      layout->rows = numfiles / layout->flow_columns + 1;  // XXX dirty, modulo is zero
     }
     else {
-      layout->columns = 1;
+      layout->flow_columns = 1;
       layout->rows = numfiles + 1;  // XXX dirty, modulo is zero
     }
     layout->height = sfile->layout->rows * (layout->tile_h + 2 * layout->tile_border_y) +
-                     layout->tile_border_y * 2;
+                     layout->tile_border_y * 2 - layout->offset_top;
     layout->flag = FILE_LAYOUT_VER;
   }
-  else {
-    int column_space = 0.6f * UI_UNIT_X;
-    int column_icon_space = 0.2f * UI_UNIT_X;
+  else if (params->display == FILE_VERTICALDISPLAY) {
+    int rowcount;
 
-    layout->prv_w = 0;
-    layout->prv_h = 0;
+    layout->prv_w = ((float)params->thumbnail_size / 20.0f) * UI_UNIT_X;
+    layout->prv_h = ((float)params->thumbnail_size / 20.0f) * UI_UNIT_Y;
+    layout->tile_border_x = 0.4f * UI_UNIT_X;
+    layout->tile_border_y = 0.1f * UI_UNIT_Y;
+    layout->tile_h = textheight * 3 / 2;
+    layout->width = (int)(BLI_rctf_size_x(&v2d->cur) - 2 * layout->tile_border_x);
+    layout->tile_w = layout->width;
+    layout->flow_columns = 1;
+    layout->attribute_column_header_h = layout->tile_h * 1.2f + 2 * layout->tile_border_y;
+    layout->offset_top = layout->attribute_column_header_h;
+    rowcount = (int)(BLI_rctf_size_y(&v2d->cur) - layout->offset_top - 2 * layout->tile_border_y) /
+               (layout->tile_h + 2 * layout->tile_border_y);
+    file_attribute_columns_init(params, layout);
+
+    layout->rows = MAX2(rowcount, numfiles);
+    BLI_assert(layout->rows != 0);
+    layout->height = sfile->layout->rows * (layout->tile_h + 2 * layout->tile_border_y) +
+                     layout->tile_border_y * 2 + layout->offset_top;
+    layout->flag = FILE_LAYOUT_VER;
+  }
+  else if (params->display == FILE_HORIZONTALDISPLAY) {
+    layout->prv_w = ((float)params->thumbnail_size / 20.0f) * UI_UNIT_X;
+    layout->prv_h = ((float)params->thumbnail_size / 20.0f) * UI_UNIT_Y;
     layout->tile_border_x = 0.4f * UI_UNIT_X;
     layout->tile_border_y = 0.1f * UI_UNIT_Y;
-    layout->prv_border_x = 0;
-    layout->prv_border_y = 0;
     layout->tile_h = textheight * 3 / 2;
+    layout->attribute_column_header_h = 0;
+    layout->offset_top = layout->attribute_column_header_h;
     layout->height = (int)(BLI_rctf_size_y(&v2d->cur) - 2 * layout->tile_border_y);
     /* Padding by full scrollbar H is too much, can overlap tile border Y. */
     layout->rows = (layout->height - V2D_SCROLL_HEIGHT + layout->tile_border_y) /
                    (layout->tile_h + 2 * layout->tile_border_y);
+    layout->tile_w = VERTLIST_MAJORCOLUMN_WIDTH;
+    file_attribute_columns_init(params, layout);
 
-    column_widths(params, layout);
-
-    if (params->display == FILE_SHORTDISPLAY) {
-      maxlen = ICON_DEFAULT_WIDTH_SCALE + column_icon_space +
-               (int)layout->column_widths[COLUMN_NAME] + column_space +
-               (int)layout->column_widths[COLUMN_SIZE] + column_space;
-    }
-    else {
-      maxlen = ICON_DEFAULT_WIDTH_SCALE + column_icon_space +
-               (int)layout->column_widths[COLUMN_NAME] + column_space +
-               (int)layout->column_widths[COLUMN_DATE] + column_space +
-               (int)layout->column_widths[COLUMN_TIME] + column_space +
-               (int)layout->column_widths[COLUMN_SIZE] + column_space;
-    }
-    layout->tile_w = maxlen;
     if (layout->rows > 0) {
-      layout->columns = numfiles / layout->rows + 1;  // XXX dirty, modulo is zero
+      layout->flow_columns = numfiles / layout->rows + 1;  // XXX dirty, modulo is zero
     }
     else {
       layout->rows = 1;
-      layout->columns = numfiles + 1;  // XXX dirty, modulo is zero
+      layout->flow_columns = numfiles + 1;  // XXX dirty, modulo is zero
     }
-    layout->width = sfile->layout->columns * (layout->tile_w + 2 * layout->tile_border_x) +
+    layout->width = sfile->layout->flow_columns * (layout->tile_w + 2 * layout->tile_border_x) +
                     layout->tile_border_x * 2;
     layout->flag = FILE_LAYOUT_HOR;
   }
diff --git a/source/blender/editors/space_file/space_file.c b/source/blender/editors/space_file/space_file.c
index 1fd878e4662..1befdd52d7d 100644
--- a/source/blender/editors/space_file/space_file.c
+++ b/source/blender/editors/space_file/space_file.c
@@ -72,23 +72,40 @@ static SpaceLink *file_new(const ScrArea *UNUSED(area), const Scene *UNUSED(scen
   /* Ignore user preference "USER_HEADER_BOTTOM" here (always show top for new types). */
   ar->alignment = RGN_ALIGN_TOP;
 
+  /* ui list region */
+  ar = MEM_callocN(sizeof(ARegion), "ui region for file");
+  BLI_addtail(&sfile->regionbase, ar);
+  ar->regiontype = RGN_TYPE_UI;
+  ar->alignment = RGN_ALIGN_TOP;
+  ar->flag |= RGN_FLAG_DYNAMIC_SIZE;
+
   /* Tools region */
   ar = MEM_callocN(sizeof(ARegion), "tools region for file");
   BLI_addtail(&sfile->regionbase, ar);
   ar->regiontype = RGN_TYPE_TOOLS;
   ar->alignment = RGN_ALIGN_LEFT;
+  /* Tools region (lower split region) */
+  ar = MEM_callocN(sizeof(ARegion), "lower tools region for file");
+  BLI_addtail(&sfile->regionbase, ar);
+  ar->regiontype = RGN_TYPE_TOOLS;
+  ar->alignment = RGN_ALIGN_BOTTOM | RGN_SPLIT_PREV;
+  ar->flag |= RGN_FLAG_DYNAMIC_SIZE;
+
+  /* Execute region */
+  ar = MEM_callocN(sizeof(ARegion), "execute region for file");
+  BLI_addtail(&sfile->regionbase, ar);
+  ar->regiontype = RGN_TYPE_EXECUTE;
+  ar->alignment = RGN_ALIGN_BOTTOM;
+  ar->flag |= RGN_FLAG_DYNAMIC_SIZE;
 
+  /* Tool props region is added as needed. */
+#if 0
   /* Tool props (aka operator) region */
   ar = MEM_callocN(sizeof(ARegion), "tool props region for file");
   BLI_addtail(&sfile->regionbase, ar);
   ar->regiontype = RGN_TYPE_TOOL_PROPS;
-  ar->alignment = RGN_ALIGN_BOTTOM | RGN_SPLIT_PREV;
-
-  /* ui list region */
-  ar = MEM_callocN(sizeof(ARegion), "ui region for file");
-  BLI_addtail(&sfile->regionbase, ar);
-  ar->regiontype = RGN_TYPE_UI;
-  ar->alignment = RGN_ALIGN_TOP;
+  ar->alignment = RGN_ALIGN_RIGHT;
+#endif
 
   /* main region */
   ar = MEM_callocN(sizeof(ARegion), "main region for file");
@@ -204,6 +221,7 @@ static SpaceLink *file_duplicate(SpaceLink *sl)
 static void file_refresh(const bContext *C, ScrArea *sa)
 {
   wmWindowManager *wm = CTX_wm_manager(C);
+  wmWindow *win = CTX_wm_window(C);
   SpaceFile *sfile = CTX_wm_space_file(C);
   FileSelectParams *params = ED_fileselect_get_params(sfile);
   struct FSMenu *fsmenu = ED_fsmenu_get();
@@ -217,15 +235,16 @@ static void file_refresh(const bContext *C, ScrArea *sa)
   }
   filelist_setdir(sfile->files, params->dir);
   filelist_setrecursion(sfile->files, params->recursion_level);
-  filelist_setsorting(sfile->files, params->sort);
-  filelist_setfilter_options(sfile->files,
-                             (params->flag & FILE_FILTER) != 0,
-                             (params->flag & FILE_HIDE_DOT) != 0,
-                             false, /* TODO hide_parent, should be controllable? */
-                             params->filter,
-                             params->filter_id,
-                             params->filter_glob,
-                             params->filter_search);
+  filelist_setsorting(sfile->files, params->sort, params->flag & FILE_SORT_INVERT);
+  filelist_setfilter_options(
+      sfile->files,
+      (params->flag & FILE_FILTER) != 0,
+      (params->flag & FILE_HIDE_DOT) != 0,
+      true, /* Just always hide parent, prefer to not add an extra user option for this. */
+      params->filter,
+      params->filter_id,
+      params->filter_glob,
+      params->filter_search);
 
   /* Update the active indices of bookmarks & co. */
   sfile->systemnr = fsmenu_get_active_indices(fsmenu, FS_CATEGORY_SYSTEM, params->dir);
@@ -254,7 +273,7 @@ static void file_refresh(const bContext *C, ScrArea *sa)
   else {
     filelist_cache_previews_set(sfile->files, false);
     if (sfile->previews_timer) {
-      WM_event_remove_timer_notifier(wm, CTX_wm_window(C), sfile->previews_timer);
+      WM_event_remove_timer_notifier(wm, win, sfile->previews_timer);
       sfile->previews_timer = NULL;
     }
   }
@@ -269,10 +288,20 @@ static void file_refresh(const bContext *C, ScrArea *sa)
 
   /* Might be called with NULL sa, see file_main_region_draw() below. */
   if (sa && BKE_area_find_region_type(sa, RGN_TYPE_TOOLS) == NULL) {
-    /* Create TOOLS/TOOL_PROPS regions. */
+    /* Create TOOLS region. */
     file_tools_region(sa);
 
-    ED_area_initialize(wm, CTX_wm_window(C), sa);
+    ED_area_initialize(wm, win, sa);
+  }
+  if (sa && sfile->op && BKE_area_find_region_type(sa, RGN_TYPE_TOOL_PROPS) == NULL) {
+    /* Create TOOL_PROPS region. */
+    ARegion *region_props = file_tool_props_region(sa);
+
+    if (params->flag & FILE_HIDE_TOOL_PROPS) {
+      region_props->flag |= RGN_FLAG_HIDDEN;
+    }
+
+    ED_area_initialize(wm, win, sa);
   }
 
   ED_area_tag_redraw(sa);
@@ -406,6 +435,11 @@ static void file_main_region_draw(const bContext *C, ARegion *ar)
     v2d->keepofs &= ~V2D_LOCKOFS_Y;
     v2d->keepofs |= V2D_LOCKOFS_X;
   }
+  else if (params->display == FILE_VERTICALDISPLAY) {
+    v2d->scroll = V2D_SCROLL_RIGHT;
+    v2d->keepofs &= ~V2D_LOCKOFS_Y;
+    v2d->keepofs |= V2D_LOCKOFS_X;
+  }
   else {
     v2d->scroll = V2D_SCROLL_BOTTOM;
     v2d->keepofs &= ~V2D_LOCKOFS_X;
@@ -439,7 +473,9 @@ static void file_main_region_draw(const bContext *C, ARegion *ar)
   UI_view2d_view_restore(C);
 
   /* scrollers */
-  scrollers = UI_view2d_scrollers_calc(v2d, NULL);
+  rcti view_rect;
+  ED_fileselect_layout_maskrect(sfile->layout, v2d, &view_rect);
+  scrollers = UI_view2d_scrollers_calc(v2d, &view_rect);
   UI_view2d_scrollers_draw(v2d, scrollers);
   UI_view2d_scrollers_free(scrollers);
 }
@@ -452,6 +488,7 @@ static void file_operatortypes(void)
   WM_operatortype_append(FILE_OT_select_box);
   WM_operatortype_append(FILE_OT_select_bookmark);
   WM_operatortype_append(FILE_OT_highlight);
+  WM_operatortype_append(FILE_OT_sort_column_ui_context);
   WM_operatortype_append(FILE_OT_execute);
   WM_operatortype_append(FILE_OT_cancel);
   WM_operatortype_append(FILE_OT_parent);
@@ -538,7 +575,8 @@ static void file_ui_region_init(wmWindowManager *wm, ARegion *ar)
 {
   wmKeyMap *keymap;
 
-  UI_view2d_region_reinit(&ar->v2d, V2D_COMMONVIEW_HEADER, ar->winx, ar->winy);
+  ED_region_panels_init(wm, ar);
+  ar->v2d.keepzoom |= V2D_LOCKZOOM_X | V2D_LOCKZOOM_Y;
 
   /* own keymap */
   keymap = WM_keymap_ensure(wm->defaultconf, "File Browser", SPACE_FILE, 0);
@@ -550,22 +588,18 @@ static void file_ui_region_init(wmWindowManager *wm, ARegion *ar)
 
 static void file_ui_region_draw(const bContext *C, ARegion *ar)
 {
-  float col[3];
-  /* clear */
-  UI_GetThemeColor3fv(TH_BACK, col);
-  GPU_clear_color(col[0], col[1], col[2], 0.0);
-  GPU_clear(GPU_COLOR_BIT);
-
-  /* scrolling here is just annoying, disable it */
-  ar->v2d.cur.ymax = BLI_rctf_size_y(&ar->v2d.cur);
-  ar->v2d.cur.ymin = 0;
-
-  /* set view2d view matrix for scrolling (without scrollers) */
-  UI_view2d_view_ortho(&ar->v2d);
+  ED_region_panels(C, ar);
+}
 
-  file_draw_buttons(C, ar);
+static void file_execution_region_init(wmWindowManager *wm, ARegion *ar)
+{
+  ED_region_panels_init(wm, ar);
+  ar->v2d.keepzoom |= V2D_LOCKZOOM_X | V2D_LOCKZOOM_Y;
+}
 
-  UI_view2d_view_restore(C);
+static void file_execution_region_draw(const bContext *C, ARegion *ar)
+{
+  ED_region_panels(C, ar);
 }
 
 static void file_ui_region_listener(wmWindow *UNUSED(win),
@@ -656,13 +690,21 @@ void ED_spacetype_file(void)
   /* regions: ui */
   art = MEM_callocN(sizeof(ARegionType), "spacetype file region");
   art->regionid = RGN_TYPE_UI;
-  art->prefsizey = 60;
   art->keymapflag = ED_KEYMAP_UI;
   art->listener = file_ui_region_listener;
   art->init = file_ui_region_init;
   art->draw = file_ui_region_draw;
   BLI_addhead(&st->regiontypes, art);
 
+  /* regions: execution */
+  art = MEM_callocN(sizeof(ARegionType), "spacetype file region");
+  art->regionid = RGN_TYPE_EXECUTE;
+  art->keymapflag = ED_KEYMAP_UI;
+  art->listener = file_ui_region_listener;
+  art->init = file_execution_region_init;
+  art->draw = file_execution_region_draw;
+  BLI_addhead(&st->regiontypes, art);
+
   /* regions: channels (directories) */
   art = MEM_callocN(sizeof(ARegionType), "spacetype file region");
   art->regionid = RGN_TYPE_TOOLS;
@@ -677,8 +719,8 @@ void ED_spacetype_file(void)
   /* regions: tool properties */
   art = MEM_callocN(sizeof(ARegionType), "spacetype file operator region");
   art->regionid = RGN_TYPE_TOOL_PROPS;
-  art->prefsizex = 0;
-  art->prefsizey = 360;
+  art->prefsizex = 240;
+  art->prefsizey = 60;
   art->keymapflag = ED_KEYMAP_UI;
   art->listener = file_tools_region_listener;
   art->init = file_tools_region_init;
diff --git a/source/blender/editors/space_graph/graph_buttons.c b/source/blender/editors/space_graph/graph_buttons.c
index c727c5225c9..708d91a82bb 100644
--- a/source/blender/editors/space_graph/graph_buttons.c
+++ b/source/blender/editors/space_graph/graph_buttons.c
@@ -861,6 +861,11 @@ static void graph_panel_driverVar__transChan(uiLayout *layout, ID *id, DriverVar
 
   sub = uiLayoutColumn(layout, true);
   uiItemR(sub, &dtar_ptr, "transform_type", 0, NULL, ICON_NONE);
+
+  if (ELEM(dtar->transChan, DTAR_TRANSCHAN_ROTX, DTAR_TRANSCHAN_ROTY, DTAR_TRANSCHAN_ROTZ)) {
+    uiItemR(sub, &dtar_ptr, "rotation_mode", 0, IFACE_("Mode"), ICON_NONE);
+  }
+
   uiItemR(sub, &dtar_ptr, "transform_space", 0, IFACE_("Space"), ICON_NONE);
 }
 
@@ -1236,7 +1241,7 @@ static void graph_panel_drivers_popover(const bContext *C, Panel *pa)
 {
   uiLayout *layout = pa->layout;
 
-  PointerRNA ptr = {{NULL}};
+  PointerRNA ptr = {NULL};
   PropertyRNA *prop = NULL;
   int index = -1;
   uiBut *but = NULL;
@@ -1258,7 +1263,7 @@ static void graph_panel_drivers_popover(const bContext *C, Panel *pa)
 
     /* Populate Panel - With a combination of the contents of the Driven and Driver panels */
     if (fcu && fcu->driver) {
-      ID *id = ptr.id.data;
+      ID *id = ptr.owner_id;
 
       PointerRNA ptr_fcurve;
       RNA_pointer_create(id, &RNA_FCurve, fcu, &ptr_fcurve);
diff --git a/source/blender/editors/space_graph/graph_draw.c b/source/blender/editors/space_graph/graph_draw.c
index e7ba498fc11..e18a440f7c0 100644
--- a/source/blender/editors/space_graph/graph_draw.c
+++ b/source/blender/editors/space_graph/graph_draw.c
@@ -205,7 +205,7 @@ static void draw_fcurve_keyframe_vertices(FCurve *fcu, View2D *v2d, bool edit, u
 {
   immBindBuiltinProgram(GPU_SHADER_2D_POINT_UNIFORM_SIZE_UNIFORM_COLOR_AA);
 
-  immUniform1f("size", UI_GetThemeValuef(TH_VERTEX_SIZE) * U.pixelsize);
+  immUniform1f("size", UI_GetThemeValuef(TH_VERTEX_SIZE) * U.dpi_fac);
 
   draw_fcurve_selected_keyframe_vertices(fcu, v2d, edit, false, pos);
   draw_fcurve_selected_keyframe_vertices(fcu, v2d, edit, true, pos);
@@ -269,8 +269,8 @@ static void draw_fcurve_handle_vertices(FCurve *fcu,
   immBindBuiltinProgram(GPU_SHADER_2D_POINT_UNIFORM_SIZE_UNIFORM_COLOR_OUTLINE_AA);
 
   /* set handle size */
-  immUniform1f("size", (1.4f * UI_GetThemeValuef(TH_HANDLE_VERTEX_SIZE)) * U.pixelsize);
-  immUniform1f("outlineWidth", 1.5f * U.pixelsize);
+  immUniform1f("size", (1.4f * UI_GetThemeValuef(TH_HANDLE_VERTEX_SIZE)) * U.dpi_fac);
+  immUniform1f("outlineWidth", 1.5f * U.dpi_fac);
 
   draw_fcurve_selected_handle_vertices(fcu, v2d, false, sel_handle_only, pos);
   draw_fcurve_selected_handle_vertices(fcu, v2d, true, sel_handle_only, pos);
diff --git a/source/blender/editors/space_graph/graph_edit.c b/source/blender/editors/space_graph/graph_edit.c
index b624e21937f..329067de545 100644
--- a/source/blender/editors/space_graph/graph_edit.c
+++ b/source/blender/editors/space_graph/graph_edit.c
@@ -1538,7 +1538,7 @@ void GRAPH_OT_sound_bake(wmOperatorType *ot)
                                  FILE_TYPE_FOLDER | FILE_TYPE_SOUND | FILE_TYPE_MOVIE,
                                  FILE_SPECIAL,
                                  FILE_OPENFILE,
-                                 WM_FILESEL_FILEPATH,
+                                 WM_FILESEL_FILEPATH | WM_FILESEL_SHOW_PROPS,
                                  FILE_DEFAULTDISPLAY,
                                  FILE_SORT_ALPHA);
   RNA_def_float(ot->srna,
diff --git a/source/blender/editors/space_graph/graph_intern.h b/source/blender/editors/space_graph/graph_intern.h
index 168c38b66a4..320240221b5 100644
--- a/source/blender/editors/space_graph/graph_intern.h
+++ b/source/blender/editors/space_graph/graph_intern.h
@@ -26,7 +26,6 @@
 
 struct ARegion;
 struct ARegionType;
-struct ScrArea;
 struct SpaceGraph;
 struct bAnimContext;
 struct bAnimListElem;
diff --git a/source/blender/editors/space_graph/space_graph.c b/source/blender/editors/space_graph/space_graph.c
index 6c5ebf77bf4..91e5ab61dd9 100644
--- a/source/blender/editors/space_graph/space_graph.c
+++ b/source/blender/editors/space_graph/space_graph.c
@@ -527,7 +527,7 @@ static void graph_region_message_subscribe(const struct bContext *UNUSED(C),
    * so just whitelist the entire structs for updates
    */
   {
-    wmMsgParams_RNA msg_key_params = {{{0}}};
+    wmMsgParams_RNA msg_key_params = {{0}};
     StructRNA *type_array[] = {
         &RNA_DopeSheet, /* dopesheet filters */
 
diff --git a/source/blender/editors/space_image/image_buttons.c b/source/blender/editors/space_image/image_buttons.c
index f1a29a1542d..64e1c02590e 100644
--- a/source/blender/editors/space_image/image_buttons.c
+++ b/source/blender/editors/space_image/image_buttons.c
@@ -991,7 +991,7 @@ void uiTemplateImage(uiLayout *layout,
 void uiTemplateImageSettings(uiLayout *layout, PointerRNA *imfptr, bool color_management)
 {
   ImageFormatData *imf = imfptr->data;
-  ID *id = imfptr->id.data;
+  ID *id = imfptr->owner_id;
   PointerRNA display_settings_ptr;
   PropertyRNA *prop;
   const int depth_ok = BKE_imtype_valid_depths(imf->imtype);
diff --git a/source/blender/editors/space_image/image_intern.h b/source/blender/editors/space_image/image_intern.h
index 2c723f45e94..1abb6715fdb 100644
--- a/source/blender/editors/space_image/image_intern.h
+++ b/source/blender/editors/space_image/image_intern.h
@@ -27,7 +27,6 @@
 /* internal exports only */
 struct ARegion;
 struct ARegionType;
-struct ScrArea;
 struct SpaceImage;
 struct bContext;
 struct bNodeTree;
diff --git a/source/blender/editors/space_image/image_ops.c b/source/blender/editors/space_image/image_ops.c
index 05ba82b8bde..e338a450db6 100644
--- a/source/blender/editors/space_image/image_ops.c
+++ b/source/blender/editors/space_image/image_ops.c
@@ -1431,7 +1431,7 @@ static int image_open_invoke(bContext *C, wmOperator *op, const wmEvent *UNUSED(
       Image *oldima;
 
       oldptr = RNA_property_pointer_get(&ptr, prop);
-      oldima = (Image *)oldptr.id.data;
+      oldima = (Image *)oldptr.owner_id;
       /* unlikely to fail but better avoid strange crash */
       if (oldima && GS(oldima->id.name) == ID_IM) {
         ima = oldima;
@@ -1773,6 +1773,7 @@ static int image_save_options_init(Main *bmain,
       }
       else {
         BLI_snprintf(opts->filepath, sizeof(opts->filepath), "//%s", ima->id.name + 2);
+        BLI_path_make_safe(opts->filepath);
         BLI_path_abs(opts->filepath, is_prev_save ? G.ima : BKE_main_blendfile_path(bmain));
       }
     }
diff --git a/source/blender/editors/space_image/space_image.c b/source/blender/editors/space_image/space_image.c
index 33e77d3623e..5fa4fe3e077 100644
--- a/source/blender/editors/space_image/space_image.c
+++ b/source/blender/editors/space_image/space_image.c
@@ -586,7 +586,7 @@ static void image_main_region_draw(const bContext *C, ARegion *ar)
   float col[3];
 
   /* XXX This is in order to draw UI batches with the DRW
-   * olg context since we now use it for drawing the entire area */
+   * old context since we now use it for drawing the entire area. */
   gpu_batch_presets_reset();
 
   GPUViewport *viewport =
diff --git a/source/blender/editors/space_info/info_stats.c b/source/blender/editors/space_info/info_stats.c
index 946274de882..106edc290d5 100644
--- a/source/blender/editors/space_info/info_stats.c
+++ b/source/blender/editors/space_info/info_stats.c
@@ -574,7 +574,7 @@ void ED_info_stats_clear(ViewLayer *view_layer)
 
 const char *ED_info_stats_string(Main *bmain, Scene *scene, ViewLayer *view_layer)
 {
-  /* Loopin through dependency graph when interface is locked in not safe.
+  /* Looping through dependency graph when interface is locked in not safe.
    * Thew interface is marked as locked when jobs wants to modify the
    * dependency graph. */
   wmWindowManager *wm = bmain->wm.first;
diff --git a/source/blender/editors/space_info/textview.c b/source/blender/editors/space_info/textview.c
index 24d3008b1e7..97d5faa9c13 100644
--- a/source/blender/editors/space_info/textview.c
+++ b/source/blender/editors/space_info/textview.c
@@ -313,7 +313,7 @@ int textview_draw(
   /* constants for the sequencer context */
   cdc.font_id = font_id;
   cdc.cwidth = (int)BLF_fixed_width(font_id);
-  assert(cdc.cwidth > 0);
+  BLI_assert(cdc.cwidth > 0);
   cdc.lheight = tvc->lheight;
   cdc.lofs = -BLF_descender(font_id);
   /* note, scroll bar must be already subtracted () */
diff --git a/source/blender/editors/space_nla/nla_buttons.c b/source/blender/editors/space_nla/nla_buttons.c
index 719efc7eeac..126b20a028e 100644
--- a/source/blender/editors/space_nla/nla_buttons.c
+++ b/source/blender/editors/space_nla/nla_buttons.c
@@ -268,8 +268,8 @@ static void nla_panel_animdata(const bContext *C, Panel *pa)
   /* icon + id-block name of block where AnimData came from to prevent
    * accidentally changing the properties of the wrong action
    */
-  if (adt_ptr.id.data) {
-    ID *id = adt_ptr.id.data;
+  if (adt_ptr.owner_id) {
+    ID *id = adt_ptr.owner_id;
     PointerRNA id_ptr;
 
     RNA_id_pointer_create(id, &id_ptr);
@@ -513,7 +513,7 @@ static void nla_panel_modifiers(const bContext *C, Panel *pa)
   for (fcm = strip->modifiers.first; fcm; fcm = fcm->next) {
     col = uiLayoutColumn(pa->layout, true);
 
-    ANIM_uiTemplate_fmodifier_draw(col, strip_ptr.id.data, &strip->modifiers, fcm);
+    ANIM_uiTemplate_fmodifier_draw(col, strip_ptr.owner_id, &strip->modifiers, fcm);
   }
 }
 
diff --git a/source/blender/editors/space_nla/nla_channels.c b/source/blender/editors/space_nla/nla_channels.c
index e5c116e85de..456eb783706 100644
--- a/source/blender/editors/space_nla/nla_channels.c
+++ b/source/blender/editors/space_nla/nla_channels.c
@@ -446,7 +446,7 @@ static int nlachannels_pushdown_exec(bContext *C, wmOperator *op)
 
   /* get anim-channel to use (or more specifically, the animdata block behind it) */
   if (channel_index == -1) {
-    PointerRNA adt_ptr = {{NULL}};
+    PointerRNA adt_ptr = {NULL};
 
     /* active animdata block */
     if (nla_panel_context(C, &adt_ptr, NULL, NULL) == 0 || (adt_ptr.data == NULL)) {
@@ -458,7 +458,7 @@ static int nlachannels_pushdown_exec(bContext *C, wmOperator *op)
       return OPERATOR_CANCELLED;
     }
     else {
-      id = adt_ptr.id.data;
+      id = adt_ptr.owner_id;
       adt = adt_ptr.data;
     }
   }
@@ -581,7 +581,7 @@ static int nla_action_unlink_exec(bContext *C, wmOperator *op)
   /* do unlinking */
   if (adt && adt->action) {
     bool force_delete = RNA_boolean_get(op->ptr, "force_delete");
-    ED_animedit_unlink_action(C, adt_ptr.id.data, adt, adt->action, op->reports, force_delete);
+    ED_animedit_unlink_action(C, adt_ptr.owner_id, adt, adt->action, op->reports, force_delete);
   }
 
   return OPERATOR_FINISHED;
diff --git a/source/blender/editors/space_nla/nla_draw.c b/source/blender/editors/space_nla/nla_draw.c
index 5cf9646210e..9d6ccd6fe35 100644
--- a/source/blender/editors/space_nla/nla_draw.c
+++ b/source/blender/editors/space_nla/nla_draw.c
@@ -141,8 +141,10 @@ static void nla_action_draw_keyframes(
     uint outline_color_id = GPU_vertformat_attr_add(
         format, "outlineColor", GPU_COMP_U8, 4, GPU_FETCH_INT_TO_FLOAT_UNIT);
     uint flags_id = GPU_vertformat_attr_add(format, "flags", GPU_COMP_U32, 1, GPU_FETCH_INT);
-    immBindBuiltinProgram(GPU_SHADER_KEYFRAME_DIAMOND);
+
     GPU_program_point_size(true);
+    immBindBuiltinProgram(GPU_SHADER_KEYFRAME_DIAMOND);
+    immUniform1f("outline_scale", 1.0f);
     immUniform2f("ViewportSize", BLI_rcti_size_x(&v2d->mask) + 1, BLI_rcti_size_y(&v2d->mask) + 1);
     immBegin(GPU_PRIM_POINTS, key_len);
 
diff --git a/source/blender/editors/space_nla/space_nla.c b/source/blender/editors/space_nla/space_nla.c
index 3bdd2804efc..f274f3c93ec 100644
--- a/source/blender/editors/space_nla/space_nla.c
+++ b/source/blender/editors/space_nla/space_nla.c
@@ -522,7 +522,7 @@ static void nla_channel_region_message_subscribe(const struct bContext *UNUSED(C
    * so just whitelist the entire struct for updates
    */
   {
-    wmMsgParams_RNA msg_key_params = {{{0}}};
+    wmMsgParams_RNA msg_key_params = {{0}};
     StructRNA *type_array[] = {
         &RNA_DopeSheet,
     };
diff --git a/source/blender/editors/space_node/drawnode.c b/source/blender/editors/space_node/drawnode.c
index 748e485f614..b89f163f579 100644
--- a/source/blender/editors/space_node/drawnode.c
+++ b/source/blender/editors/space_node/drawnode.c
@@ -24,6 +24,7 @@
 
 #include "BLI_blenlib.h"
 #include "BLI_math.h"
+#include "BLI_system.h"
 
 #include "DNA_node_types.h"
 #include "DNA_object_types.h"
@@ -90,7 +91,7 @@ static void node_buts_value(uiLayout *layout, bContext *UNUSED(C), PointerRNA *p
   /* first output stores value */
   bNodeSocket *output = node->outputs.first;
   PointerRNA sockptr;
-  RNA_pointer_create(ptr->id.data, &RNA_NodeSocket, output, &sockptr);
+  RNA_pointer_create(ptr->owner_id, &RNA_NodeSocket, output, &sockptr);
 
   uiItemR(layout, &sockptr, "default_value", 0, "", ICON_NONE);
 }
@@ -102,7 +103,7 @@ static void node_buts_rgb(uiLayout *layout, bContext *UNUSED(C), PointerRNA *ptr
   bNodeSocket *output = node->outputs.first;
   PointerRNA sockptr;
   uiLayout *col;
-  RNA_pointer_create(ptr->id.data, &RNA_NodeSocket, output, &sockptr);
+  RNA_pointer_create(ptr->owner_id, &RNA_NodeSocket, output, &sockptr);
 
   col = uiLayoutColumn(layout, false);
   uiTemplateColorPicker(col, &sockptr, "default_value", 1, 0, 0, 0);
@@ -113,7 +114,7 @@ static void node_buts_mix_rgb(uiLayout *layout, bContext *UNUSED(C), PointerRNA
 {
   uiLayout *row, *col;
 
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
 
   col = uiLayoutColumn(layout, false);
   row = uiLayoutRow(col, true);
@@ -193,7 +194,7 @@ static void node_buts_normal(uiLayout *layout, bContext *UNUSED(C), PointerRNA *
   /* first output stores normal */
   bNodeSocket *output = node->outputs.first;
   PointerRNA sockptr;
-  RNA_pointer_create(ptr->id.data, &RNA_NodeSocket, output, &sockptr);
+  RNA_pointer_create(ptr->owner_id, &RNA_NodeSocket, output, &sockptr);
 
   uiItemR(layout, &sockptr, "default_value", 0, "", ICON_NONE);
 }
@@ -1186,6 +1187,11 @@ static void node_shader_buts_ambient_occlusion(uiLayout *layout,
   uiItemR(layout, ptr, "only_local", 0, NULL, ICON_NONE);
 }
 
+static void node_shader_buts_white_noise(uiLayout *layout, bContext *UNUSED(C), PointerRNA *ptr)
+{
+  uiItemR(layout, ptr, "dimensions", 0, "", ICON_NONE);
+}
+
 /* only once called */
 static void node_shader_set_butfunc(bNodeType *ntype)
 {
@@ -1220,7 +1226,7 @@ static void node_shader_set_butfunc(bNodeType *ntype)
     case SH_NODE_MATH:
       ntype->draw_buttons = node_buts_math;
       break;
-    case SH_NODE_VECT_MATH:
+    case SH_NODE_VECTOR_MATH:
       ntype->draw_buttons = node_shader_buts_vect_math;
       break;
     case SH_NODE_VECT_TRANSFORM:
@@ -1330,6 +1336,9 @@ static void node_shader_set_butfunc(bNodeType *ntype)
     case SH_NODE_AMBIENT_OCCLUSION:
       ntype->draw_buttons = node_shader_buts_ambient_occlusion;
       break;
+    case SH_NODE_TEX_WHITE_NOISE:
+      ntype->draw_buttons = node_shader_buts_white_noise;
+      break;
   }
 }
 
@@ -1363,7 +1372,7 @@ static void node_composit_buts_image(uiLayout *layout, bContext *C, PointerRNA *
   bNode *node = ptr->data;
   PointerRNA imaptr, iuserptr;
 
-  RNA_pointer_create((ID *)ptr->id.data, &RNA_ImageUser, node->storage, &iuserptr);
+  RNA_pointer_create(ptr->owner_id, &RNA_ImageUser, node->storage, &iuserptr);
   uiLayoutSetContextPointer(layout, "image_user", &iuserptr);
   uiTemplateID(layout,
                C,
@@ -1390,7 +1399,7 @@ static void node_composit_buts_image_ex(uiLayout *layout, bContext *C, PointerRN
   bNode *node = ptr->data;
   PointerRNA iuserptr;
 
-  RNA_pointer_create((ID *)ptr->id.data, &RNA_ImageUser, node->storage, &iuserptr);
+  RNA_pointer_create(ptr->owner_id, &RNA_ImageUser, node->storage, &iuserptr);
   uiLayoutSetContextPointer(layout, "image_user", &iuserptr);
   uiTemplateImage(layout, C, ptr, "image", &iuserptr, 0, 1);
 }
@@ -1951,7 +1960,7 @@ static void node_composit_buts_file_output_ex(uiLayout *layout, bContext *C, Poi
   }
   /* XXX collection lookup does not return the ID part of the pointer,
    * setting this manually here */
-  active_input_ptr.id.data = ptr->id.data;
+  active_input_ptr.owner_id = ptr->owner_id;
 
   col = uiLayoutColumn(row, true);
   ot = WM_operatortype_find("NODE_OT_output_file_move_active_socket", false);
@@ -2695,6 +2704,10 @@ static void node_composit_buts_denoise(uiLayout *layout, bContext *UNUSED(C), Po
 {
 #ifndef WITH_OPENIMAGEDENOISE
   uiItemL(layout, IFACE_("Disabled, built without OpenImageDenoise"), ICON_ERROR);
+#else
+  if (!BLI_cpu_support_sse41()) {
+    uiItemL(layout, IFACE_("Disabled, CPU with SSE4.1 is required"), ICON_ERROR);
+  }
 #endif
 
   uiItemR(layout, ptr, "use_hdr", 0, NULL, ICON_NONE);
@@ -2959,7 +2972,7 @@ static void node_texture_buts_proc(uiLayout *layout, bContext *UNUSED(C), Pointe
 {
   PointerRNA tex_ptr;
   bNode *node = ptr->data;
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   Tex *tex = (Tex *)node->storage;
   uiLayout *col, *row;
 
@@ -3053,7 +3066,7 @@ static void node_texture_buts_image_ex(uiLayout *layout, bContext *C, PointerRNA
   bNode *node = ptr->data;
   PointerRNA iuserptr;
 
-  RNA_pointer_create((ID *)ptr->id.data, &RNA_ImageUser, node->storage, &iuserptr);
+  RNA_pointer_create(ptr->owner_id, &RNA_ImageUser, node->storage, &iuserptr);
   uiTemplateImage(layout, C, ptr, "image", &iuserptr, 0, 0);
 }
 
@@ -3115,7 +3128,7 @@ static void node_texture_set_butfunc(bNodeType *ntype)
 
 static void node_property_update_default(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = ptr->data;
   ED_node_tag_update_nodetree(bmain, ntree, node);
 }
@@ -3285,7 +3298,7 @@ static void node_file_output_socket_draw(bContext *C,
                                          PointerRNA *ptr,
                                          PointerRNA *node_ptr)
 {
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNodeSocket *sock = ptr->data;
   uiLayout *row;
   PointerRNA inputptr, imfptr;
@@ -3358,8 +3371,13 @@ static void std_node_socket_draw(
         uiTemplateComponentMenu(layout, ptr, "default_value", text);
       }
       else {
-        uiLayout *column = uiLayoutColumn(layout, true);
-        uiItemR(column, ptr, "default_value", 0, text, 0);
+        if (sock->typeinfo->subtype == PROP_DIRECTION) {
+          uiItemR(layout, ptr, "default_value", 0, "", ICON_NONE);
+        }
+        else {
+          uiLayout *column = uiLayoutColumn(layout, true);
+          uiItemR(column, ptr, "default_value", 0, text, ICON_NONE);
+        }
       }
       break;
     case SOCK_RGBA:
diff --git a/source/blender/editors/space_node/node_draw.c b/source/blender/editors/space_node/node_draw.c
index 1c3544077c4..6dc2182b684 100644
--- a/source/blender/editors/space_node/node_draw.c
+++ b/source/blender/editors/space_node/node_draw.c
@@ -713,15 +713,28 @@ static void node_draw_mute_line(View2D *v2d, SpaceNode *snode, bNode *node)
   GPU_blend(false);
 }
 
-static void node_socket_circle_draw(const bContext *C,
-                                    bNodeTree *ntree,
-                                    PointerRNA node_ptr,
-                                    bNodeSocket *sock,
-                                    unsigned pos,
-                                    unsigned col)
+/* flags used in gpu_shader_keyframe_diamond_frag.glsl */
+#define MARKER_SHAPE_DIAMOND 0x1
+#define MARKER_SHAPE_SQUARE 0xC
+#define MARKER_SHAPE_CIRCLE 0x2
+#define MARKER_SHAPE_INNER_DOT 0x10
+
+static void node_socket_draw(const bContext *C,
+                             bNodeTree *ntree,
+                             PointerRNA node_ptr,
+                             bNodeSocket *sock,
+                             unsigned pos_id,
+                             unsigned col_id,
+                             unsigned shape_id,
+                             unsigned size_id,
+                             unsigned outline_col_id,
+                             float size,
+                             bool selected)
 {
   PointerRNA ptr;
   float color[4];
+  float outline_color[4];
+  unsigned int flags = 0;
 
   RNA_pointer_create((ID *)ntree, &RNA_NodeSocket, sock, &ptr);
   sock->typeinfo->draw_color((bContext *)C, &ptr, &node_ptr, color);
@@ -731,8 +744,44 @@ static void node_socket_circle_draw(const bContext *C,
     color[3] *= 0.25f;
   }
 
-  immAttr4fv(col, color);
-  immVertex2f(pos, sock->locx, sock->locy);
+  if (selected) {
+    UI_GetThemeColor4fv(TH_TEXT_HI, outline_color);
+    outline_color[3] = 0.9f;
+  }
+  else {
+    copy_v4_fl(outline_color, 0.0f);
+    outline_color[3] = 0.6f;
+  }
+
+  /* sets shape flags */
+  switch (sock->display_shape) {
+    case SOCK_DISPLAY_SHAPE_DIAMOND:
+    case SOCK_DISPLAY_SHAPE_DIAMOND_DOT:
+      flags = MARKER_SHAPE_DIAMOND;
+      break;
+    case SOCK_DISPLAY_SHAPE_SQUARE:
+    case SOCK_DISPLAY_SHAPE_SQUARE_DOT:
+      flags = MARKER_SHAPE_SQUARE;
+      break;
+    default:
+    case SOCK_DISPLAY_SHAPE_CIRCLE:
+    case SOCK_DISPLAY_SHAPE_CIRCLE_DOT:
+      flags = MARKER_SHAPE_CIRCLE;
+      break;
+  }
+
+  if (ELEM(sock->display_shape,
+           SOCK_DISPLAY_SHAPE_DIAMOND_DOT,
+           SOCK_DISPLAY_SHAPE_SQUARE_DOT,
+           SOCK_DISPLAY_SHAPE_CIRCLE_DOT)) {
+    flags |= MARKER_SHAPE_INNER_DOT;
+  }
+
+  immAttr4fv(col_id, color);
+  immAttr1u(shape_id, flags);
+  immAttr1f(size_id, size);
+  immAttr4fv(outline_col_id, outline_color);
+  immVertex2f(pos_id, sock->locx, sock->locy);
 }
 
 /* **************  Socket callbacks *********** */
@@ -888,26 +937,28 @@ void node_draw_sockets(View2D *v2d,
   PointerRNA node_ptr;
   RNA_pointer_create((ID *)ntree, &RNA_Node, node, &node_ptr);
 
-  float scale;
-  UI_view2d_scale_get(v2d, &scale, NULL);
+  bool selected = false;
 
   GPUVertFormat *format = immVertexFormat();
-  uint pos = GPU_vertformat_attr_add(format, "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
-  uint col = GPU_vertformat_attr_add(format, "color", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
+  uint pos_id = GPU_vertformat_attr_add(format, "pos", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
+  uint col_id = GPU_vertformat_attr_add(format, "color", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
+  uint shape_id = GPU_vertformat_attr_add(format, "flags", GPU_COMP_U32, 1, GPU_FETCH_INT);
+  uint size_id = GPU_vertformat_attr_add(format, "size", GPU_COMP_F32, 1, GPU_FETCH_FLOAT);
+  uint outline_col_id = GPU_vertformat_attr_add(
+      format, "outlineColor", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
 
   GPU_blend(true);
   GPU_program_point_size(true);
-
-  immBindBuiltinProgram(GPU_SHADER_2D_POINT_UNIFORM_SIZE_VARYING_COLOR_OUTLINE_AA);
+  immBindBuiltinProgram(GPU_SHADER_KEYFRAME_DIAMOND);
+  immUniform1f("outline_scale", 0.7f);
+  immUniform2f("ViewportSize", -1.0f, -1.0f);
 
   /* set handle size */
-  immUniform1f("size", 2.0f * NODE_SOCKSIZE * scale); /* 2 * size to have diameter */
+  float scale;
+  UI_view2d_scale_get(v2d, &scale, NULL);
+  scale *= 2.25f * NODE_SOCKSIZE;
 
   if (!select_all) {
-    /* outline for unselected sockets */
-    immUniform1f("outlineWidth", 1.0f);
-    immUniform4f("outlineColor", 0.0f, 0.0f, 0.0f, 0.6f);
-
     immBeginAtMost(GPU_PRIM_POINTS, total_input_len + total_output_len);
   }
 
@@ -923,7 +974,17 @@ void node_draw_sockets(View2D *v2d,
       continue;
     }
 
-    node_socket_circle_draw(C, ntree, node_ptr, sock, pos, col);
+    node_socket_draw(C,
+                     ntree,
+                     node_ptr,
+                     sock,
+                     pos_id,
+                     col_id,
+                     shape_id,
+                     size_id,
+                     outline_col_id,
+                     scale,
+                     selected);
   }
 
   /* socket outputs */
@@ -938,7 +999,17 @@ void node_draw_sockets(View2D *v2d,
         continue;
       }
 
-      node_socket_circle_draw(C, ntree, node_ptr, sock, pos, col);
+      node_socket_draw(C,
+                       ntree,
+                       node_ptr,
+                       sock,
+                       pos_id,
+                       col_id,
+                       shape_id,
+                       size_id,
+                       outline_col_id,
+                       scale,
+                       selected);
     }
   }
 
@@ -949,10 +1020,8 @@ void node_draw_sockets(View2D *v2d,
   /* go back and draw selected sockets */
   if (selected_input_len + selected_output_len > 0) {
     /* outline for selected sockets */
-    float c[3];
-    UI_GetThemeColor3fv(TH_TEXT_HI, c);
-    immUniform4f("outlineColor", c[0], c[1], c[2], 1.0f);
-    immUniform1f("outlineWidth", 1.5f);
+
+    selected = true;
 
     immBegin(GPU_PRIM_POINTS, selected_input_len + selected_output_len);
 
@@ -963,7 +1032,17 @@ void node_draw_sockets(View2D *v2d,
           continue;
         }
         if (select_all || (sock->flag & SELECT)) {
-          node_socket_circle_draw(C, ntree, node_ptr, sock, pos, col);
+          node_socket_draw(C,
+                           ntree,
+                           node_ptr,
+                           sock,
+                           pos_id,
+                           col_id,
+                           shape_id,
+                           size_id,
+                           outline_col_id,
+                           scale,
+                           selected);
           if (--selected_input_len == 0) {
             break; /* stop as soon as last one is drawn */
           }
@@ -978,7 +1057,17 @@ void node_draw_sockets(View2D *v2d,
           continue;
         }
         if (select_all || (sock->flag & SELECT)) {
-          node_socket_circle_draw(C, ntree, node_ptr, sock, pos, col);
+          node_socket_draw(C,
+                           ntree,
+                           node_ptr,
+                           sock,
+                           pos_id,
+                           col_id,
+                           shape_id,
+                           size_id,
+                           outline_col_id,
+                           scale,
+                           selected);
           if (--selected_output_len == 0) {
             break; /* stop as soon as last one is drawn */
           }
diff --git a/source/blender/editors/space_node/node_edit.c b/source/blender/editors/space_node/node_edit.c
index d31256a1425..dcf901216f5 100644
--- a/source/blender/editors/space_node/node_edit.c
+++ b/source/blender/editors/space_node/node_edit.c
@@ -1838,7 +1838,7 @@ static int node_output_file_add_socket_exec(bContext *C, wmOperator *op)
 
   if (ptr.data) {
     node = ptr.data;
-    ntree = ptr.id.data;
+    ntree = (bNodeTree *)ptr.owner_id;
   }
   else if (snode && snode->edittree) {
     ntree = snode->edittree;
@@ -1886,7 +1886,7 @@ static int node_output_file_remove_active_socket_exec(bContext *C, wmOperator *U
 
   if (ptr.data) {
     node = ptr.data;
-    ntree = ptr.id.data;
+    ntree = (bNodeTree *)ptr.owner_id;
   }
   else if (snode && snode->edittree) {
     ntree = snode->edittree;
@@ -2535,7 +2535,7 @@ static int node_shader_script_update_exec(bContext *C, wmOperator *op)
 
   /* get node */
   if (nodeptr.data) {
-    ntree_base = nodeptr.id.data;
+    ntree_base = (bNodeTree *)nodeptr.owner_id;
     node = nodeptr.data;
   }
   else if (snode && snode->edittree) {
@@ -2734,7 +2734,7 @@ static int node_cryptomatte_add_socket_exec(bContext *C, wmOperator *UNUSED(op))
 
   if (ptr.data) {
     node = ptr.data;
-    ntree = ptr.id.data;
+    ntree = (bNodeTree *)ptr.owner_id;
   }
   else if (snode && snode->edittree) {
     ntree = snode->edittree;
@@ -2778,7 +2778,7 @@ static int node_cryptomatte_remove_socket_exec(bContext *C, wmOperator *UNUSED(o
 
   if (ptr.data) {
     node = ptr.data;
-    ntree = ptr.id.data;
+    ntree = (bNodeTree *)ptr.owner_id;
   }
   else if (snode && snode->edittree) {
     ntree = snode->edittree;
diff --git a/source/blender/editors/space_node/node_select.c b/source/blender/editors/space_node/node_select.c
index 78f36719880..450cf28cce1 100644
--- a/source/blender/editors/space_node/node_select.c
+++ b/source/blender/editors/space_node/node_select.c
@@ -1234,8 +1234,18 @@ static uiBlock *node_find_menu(bContext *C, ARegion *ar, void *arg_op)
   UI_block_flag_enable(block, UI_BLOCK_LOOP | UI_BLOCK_MOVEMOUSE_QUIT | UI_BLOCK_SEARCH_MENU);
   UI_block_theme_style_set(block, UI_BLOCK_THEME_STYLE_POPUP);
 
-  but = uiDefSearchBut(
-      block, search, 0, ICON_VIEWZOOM, sizeof(search), 10, 10, 9 * UI_UNIT_X, UI_UNIT_Y, 0, 0, "");
+  but = uiDefSearchBut(block,
+                       search,
+                       0,
+                       ICON_VIEWZOOM,
+                       sizeof(search),
+                       10,
+                       10,
+                       UI_searchbox_size_x(),
+                       UI_UNIT_Y,
+                       0,
+                       0,
+                       "");
   UI_but_func_search_set(but, NULL, node_find_cb, op->type, false, node_find_call_cb, NULL);
   UI_but_flag_enable(but, UI_BUT_ACTIVATE_ON_INIT);
 
@@ -1256,7 +1266,7 @@ static uiBlock *node_find_menu(bContext *C, ARegion *ar, void *arg_op)
            NULL);
 
   /* Move it downwards, mouse over button. */
-  UI_block_bounds_set_popup(block, 6, (const int[2]){0, -UI_UNIT_Y});
+  UI_block_bounds_set_popup(block, 0.3f * U.widget_unit, (const int[2]){0, -UI_UNIT_Y});
 
   return block;
 }
diff --git a/source/blender/editors/space_node/node_templates.c b/source/blender/editors/space_node/node_templates.c
index 8cc57a82fe0..423dec13c69 100644
--- a/source/blender/editors/space_node/node_templates.c
+++ b/source/blender/editors/space_node/node_templates.c
@@ -441,7 +441,7 @@ static int ui_node_item_name_compare(const void *a, const void *b)
 {
   const bNodeType *type_a = *(const bNodeType **)a;
   const bNodeType *type_b = *(const bNodeType **)b;
-  return BLI_natstrcmp(type_a->ui_name, type_b->ui_name);
+  return BLI_strcasecmp_natural(type_a->ui_name, type_b->ui_name);
 }
 
 static bool ui_node_item_special_poll(const bNodeTree *UNUSED(ntree), const bNodeType *ntype)
diff --git a/source/blender/editors/space_outliner/CMakeLists.txt b/source/blender/editors/space_outliner/CMakeLists.txt
index d235dd47136..616915dbc2c 100644
--- a/source/blender/editors/space_outliner/CMakeLists.txt
+++ b/source/blender/editors/space_outliner/CMakeLists.txt
@@ -41,6 +41,7 @@ set(SRC
   outliner_edit.c
   outliner_ops.c
   outliner_select.c
+  outliner_sync.c
   outliner_tools.c
   outliner_tree.c
   outliner_utils.c
diff --git a/source/blender/editors/space_outliner/outliner_collections.c b/source/blender/editors/space_outliner/outliner_collections.c
index 4740c412083..309446db83b 100644
--- a/source/blender/editors/space_outliner/outliner_collections.c
+++ b/source/blender/editors/space_outliner/outliner_collections.c
@@ -89,7 +89,7 @@ Collection *outliner_collection_from_tree_element(const TreeElement *te)
   }
   else if (ELEM(tselem->type, TSE_SCENE_COLLECTION_BASE, TSE_VIEW_COLLECTION_BASE)) {
     Scene *scene = (Scene *)tselem->id;
-    return BKE_collection_master(scene);
+    return scene->master_collection;
   }
   else if (tselem->type == 0 && te->idcode == ID_GR) {
     return (Collection *)tselem->id;
@@ -199,7 +199,7 @@ static int collection_new_exec(bContext *C, wmOperator *op)
   }
 
   if (data.collection == NULL || ID_IS_LINKED(data.collection)) {
-    data.collection = BKE_collection_master(scene);
+    data.collection = scene->master_collection;
   }
 
   if (ID_IS_LINKED(scene)) {
@@ -514,14 +514,14 @@ static int collection_duplicate_exec(bContext *C, wmOperator *op)
    * This can happen when a whole scene is linked e.g. */
   if (parent != NULL && ID_IS_LINKED(parent)) {
     Scene *scene = CTX_data_scene(C);
-    parent = ID_IS_LINKED(scene) ? NULL : BKE_collection_master(scene);
+    parent = ID_IS_LINKED(scene) ? NULL : scene->master_collection;
   }
   else if (parent != NULL && (parent->flag & COLLECTION_IS_MASTER) != 0) {
     Scene *scene = BKE_collection_master_scene_search(bmain, parent);
     BLI_assert(scene != NULL);
     if (ID_IS_LINKED(scene)) {
       scene = CTX_data_scene(C);
-      parent = ID_IS_LINKED(scene) ? NULL : BKE_collection_master(scene);
+      parent = ID_IS_LINKED(scene) ? NULL : scene->master_collection;
     }
   }
 
diff --git a/source/blender/editors/space_outliner/outliner_dragdrop.c b/source/blender/editors/space_outliner/outliner_dragdrop.c
index d8276aa2bbc..6e30157d216 100644
--- a/source/blender/editors/space_outliner/outliner_dragdrop.c
+++ b/source/blender/editors/space_outliner/outliner_dragdrop.c
@@ -326,38 +326,57 @@ static bool parent_drop_poll(bContext *C,
   return false;
 }
 
-static int parent_drop_exec(bContext *C, wmOperator *op)
+static void parent_drop_set_parents(
+    bContext *C, ReportList *reports, wmDragID *drag, Object *parent, short parent_type)
 {
-  Object *par = NULL, *ob = NULL;
   Main *bmain = CTX_data_main(C);
-  Scene *scene = CTX_data_scene(C);
-  int partype = -1;
-  char parname[MAX_NAME], childname[MAX_NAME];
+  SpaceOutliner *soops = CTX_wm_space_outliner(C);
 
-  partype = RNA_enum_get(op->ptr, "type");
-  RNA_string_get(op->ptr, "parent", parname);
-  par = (Object *)BKE_libblock_find_name(bmain, ID_OB, parname);
-  RNA_string_get(op->ptr, "child", childname);
-  ob = (Object *)BKE_libblock_find_name(bmain, ID_OB, childname);
+  TreeElement *te = outliner_find_id(soops, &soops->tree, &parent->id);
+  Scene *scene = (Scene *)outliner_search_back(soops, te, ID_SCE);
 
-  if (ID_IS_LINKED(ob)) {
-    BKE_report(op->reports, RPT_INFO, "Can't edit library linked object");
-    return OPERATOR_CANCELLED;
+  if (scene == NULL) {
+    /* currently outliner organized in a way, that if there's no parent scene
+     * element for object it means that all displayed objects belong to
+     * active scene and parenting them is allowed (sergey)
+     */
+
+    scene = CTX_data_scene(C);
   }
 
-  ED_object_parent_set(op->reports, C, scene, ob, par, partype, false, false, NULL);
+  bool parent_set = false;
+  bool linked_objects = false;
 
-  DEG_relations_tag_update(bmain);
-  WM_event_add_notifier(C, NC_OBJECT | ND_TRANSFORM, NULL);
-  WM_event_add_notifier(C, NC_OBJECT | ND_PARENT, NULL);
+  for (wmDragID *drag_id = drag; drag_id; drag_id = drag_id->next) {
+    if (GS(drag_id->id->name) == ID_OB) {
+      Object *object = (Object *)drag_id->id;
 
-  return OPERATOR_FINISHED;
+      /* Do nothing to linked data */
+      if (ID_IS_LINKED(object)) {
+        linked_objects = true;
+        continue;
+      }
+
+      if (ED_object_parent_set(
+              reports, C, scene, object, parent, parent_type, false, false, NULL)) {
+        parent_set = true;
+      }
+    }
+  }
+
+  if (linked_objects) {
+    BKE_report(reports, RPT_INFO, "Can't edit library linked object(s)");
+  }
+
+  if (parent_set) {
+    DEG_relations_tag_update(bmain);
+    WM_event_add_notifier(C, NC_OBJECT | ND_TRANSFORM, NULL);
+    WM_event_add_notifier(C, NC_OBJECT | ND_PARENT, NULL);
+  }
 }
 
 static int parent_drop_invoke(bContext *C, wmOperator *op, const wmEvent *event)
 {
-  Main *bmain = CTX_data_main(C);
-  SpaceOutliner *soops = CTX_wm_space_outliner(C);
   TreeElement *te = outliner_drop_find(C, event);
   TreeStoreElem *tselem = te ? TREESTORE(te) : NULL;
 
@@ -374,107 +393,15 @@ static int parent_drop_invoke(bContext *C, wmOperator *op, const wmEvent *event)
   if (ob == par) {
     return OPERATOR_CANCELLED;
   }
-  if (ID_IS_LINKED(ob)) {
-    BKE_report(op->reports, RPT_INFO, "Can't edit library linked object");
-    return OPERATOR_CANCELLED;
-  }
-
-  char childname[MAX_NAME];
-  char parname[MAX_NAME];
-  STRNCPY(childname, ob->id.name + 2);
-  STRNCPY(parname, par->id.name + 2);
-  RNA_string_set(op->ptr, "child", childname);
-  RNA_string_set(op->ptr, "parent", parname);
 
-  Scene *scene = (Scene *)outliner_search_back(soops, te, ID_SCE);
-
-  if (scene == NULL) {
-    /* currently outlier organized in a way, that if there's no parent scene
-     * element for object it means that all displayed objects belong to
-     * active scene and parenting them is allowed (sergey)
-     */
-
-    scene = CTX_data_scene(C);
+  if (event->custom != EVT_DATA_DRAGDROP) {
+    return OPERATOR_CANCELLED;
   }
 
-  if ((par->type != OB_ARMATURE) && (par->type != OB_CURVE) && (par->type != OB_LATTICE)) {
-    int partype = 0;
-    if (ED_object_parent_set(op->reports, C, scene, ob, par, partype, false, false, NULL)) {
-      DEG_relations_tag_update(bmain);
-      WM_event_add_notifier(C, NC_OBJECT | ND_TRANSFORM, NULL);
-      WM_event_add_notifier(C, NC_OBJECT | ND_PARENT, NULL);
-    }
-  }
-  else {
-    /* Menu creation */
-    wmOperatorType *ot = WM_operatortype_find("OUTLINER_OT_parent_drop", false);
-    uiPopupMenu *pup = UI_popup_menu_begin(C, IFACE_("Set Parent To"), ICON_NONE);
-    uiLayout *layout = UI_popup_menu_layout(pup);
-    PointerRNA ptr;
-
-    /* Cannot use uiItemEnumO()... have multiple properties to set. */
-    uiItemFullO_ptr(layout, ot, IFACE_("Object"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-    RNA_string_set(&ptr, "parent", parname);
-    RNA_string_set(&ptr, "child", childname);
-    RNA_enum_set(&ptr, "type", PAR_OBJECT);
-
-    /* par becomes parent, make the associated menus */
-    if (par->type == OB_ARMATURE) {
-      uiItemFullO_ptr(layout, ot, IFACE_("Armature Deform"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-      RNA_string_set(&ptr, "parent", parname);
-      RNA_string_set(&ptr, "child", childname);
-      RNA_enum_set(&ptr, "type", PAR_ARMATURE);
-
-      uiItemFullO_ptr(
-          layout, ot, IFACE_("   With Empty Groups"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-      RNA_string_set(&ptr, "parent", parname);
-      RNA_string_set(&ptr, "child", childname);
-      RNA_enum_set(&ptr, "type", PAR_ARMATURE_NAME);
-
-      uiItemFullO_ptr(
-          layout, ot, IFACE_("   With Envelope Weights"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-      RNA_string_set(&ptr, "parent", parname);
-      RNA_string_set(&ptr, "child", childname);
-      RNA_enum_set(&ptr, "type", PAR_ARMATURE_ENVELOPE);
-
-      uiItemFullO_ptr(
-          layout, ot, IFACE_("   With Automatic Weights"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-      RNA_string_set(&ptr, "parent", parname);
-      RNA_string_set(&ptr, "child", childname);
-      RNA_enum_set(&ptr, "type", PAR_ARMATURE_AUTO);
-
-      uiItemFullO_ptr(layout, ot, IFACE_("Bone"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-      RNA_string_set(&ptr, "parent", parname);
-      RNA_string_set(&ptr, "child", childname);
-      RNA_enum_set(&ptr, "type", PAR_BONE);
-    }
-    else if (par->type == OB_CURVE) {
-      uiItemFullO_ptr(layout, ot, IFACE_("Curve Deform"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-      RNA_string_set(&ptr, "parent", parname);
-      RNA_string_set(&ptr, "child", childname);
-      RNA_enum_set(&ptr, "type", PAR_CURVE);
-
-      uiItemFullO_ptr(layout, ot, IFACE_("Follow Path"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-      RNA_string_set(&ptr, "parent", parname);
-      RNA_string_set(&ptr, "child", childname);
-      RNA_enum_set(&ptr, "type", PAR_FOLLOW);
-
-      uiItemFullO_ptr(layout, ot, IFACE_("Path Constraint"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-      RNA_string_set(&ptr, "parent", parname);
-      RNA_string_set(&ptr, "child", childname);
-      RNA_enum_set(&ptr, "type", PAR_PATH_CONST);
-    }
-    else if (par->type == OB_LATTICE) {
-      uiItemFullO_ptr(layout, ot, IFACE_("Lattice Deform"), 0, NULL, WM_OP_EXEC_DEFAULT, 0, &ptr);
-      RNA_string_set(&ptr, "parent", parname);
-      RNA_string_set(&ptr, "child", childname);
-      RNA_enum_set(&ptr, "type", PAR_LATTICE);
-    }
-
-    UI_popup_menu_end(C, pup);
+  ListBase *lb = event->customdata;
+  wmDrag *drag = lb->first;
 
-    return OPERATOR_INTERFACE;
-  }
+  parent_drop_set_parents(C, op->reports, drag->ids.first, par, PAR_OBJECT);
 
   return OPERATOR_FINISHED;
 }
@@ -488,17 +415,11 @@ void OUTLINER_OT_parent_drop(wmOperatorType *ot)
 
   /* api callbacks */
   ot->invoke = parent_drop_invoke;
-  ot->exec = parent_drop_exec;
 
   ot->poll = ED_operator_outliner_active;
 
   /* flags */
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO | OPTYPE_INTERNAL;
-
-  /* properties */
-  RNA_def_string(ot->srna, "child", "Object", MAX_NAME, "Child", "Child Object");
-  RNA_def_string(ot->srna, "parent", "Object", MAX_NAME, "Parent", "Parent Object");
-  RNA_def_enum(ot->srna, "type", prop_make_parent_types, 0, "Type", "");
 }
 
 /* ******************** Parent Clear Operator *********************** */
@@ -549,13 +470,21 @@ static bool parent_clear_poll(bContext *C,
 static int parent_clear_invoke(bContext *C, wmOperator *UNUSED(op), const wmEvent *event)
 {
   Main *bmain = CTX_data_main(C);
-  Object *ob = (Object *)WM_drag_ID_from_event(event, ID_OB);
 
-  if (ob == NULL) {
+  if (event->custom != EVT_DATA_DRAGDROP) {
     return OPERATOR_CANCELLED;
   }
 
-  ED_object_parent_clear(ob, 0);
+  ListBase *lb = event->customdata;
+  wmDrag *drag = lb->first;
+
+  for (wmDragID *drag_id = drag->ids.first; drag_id; drag_id = drag_id->next) {
+    if (GS(drag_id->id->name) == ID_OB) {
+      Object *object = (Object *)drag_id->id;
+
+      ED_object_parent_clear(object, 0);
+    }
+  }
 
   DEG_relations_tag_update(bmain);
   WM_event_add_notifier(C, NC_OBJECT | ND_TRANSFORM, NULL);
@@ -608,7 +537,7 @@ static int scene_drop_invoke(bContext *C, wmOperator *UNUSED(op), const wmEvent
   Collection *collection;
   if (scene != CTX_data_scene(C)) {
     /* when linking to an inactive scene link to the master collection */
-    collection = BKE_collection_master(scene);
+    collection = scene->master_collection;
   }
   else {
     collection = CTX_data_collection(C);
@@ -966,6 +895,12 @@ static int outliner_item_drag_drop_invoke(bContext *C,
     return (OPERATOR_CANCELLED | OPERATOR_PASS_THROUGH);
   }
 
+  float view_mval[2];
+  UI_view2d_region_to_view(&ar->v2d, event->mval[0], event->mval[1], &view_mval[0], &view_mval[1]);
+  if (outliner_item_is_co_within_close_toggle(te, view_mval[0])) {
+    return (OPERATOR_CANCELLED | OPERATOR_PASS_THROUGH);
+  }
+
   wmDrag *drag = WM_event_start_drag(C, data.icon, WM_DRAG_ID, NULL, 0.0, WM_DRAG_NOP);
 
   if (ELEM(GS(data.drag_id->name), ID_OB, ID_GR)) {
@@ -1032,7 +967,7 @@ static int outliner_item_drag_drop_invoke(bContext *C,
       }
       else {
         Scene *scene = CTX_data_scene(C);
-        parent = BKE_collection_master(scene);
+        parent = scene->master_collection;
       }
 
       WM_drag_add_ID(drag, id, &parent->id);
diff --git a/source/blender/editors/space_outliner/outliner_draw.c b/source/blender/editors/space_outliner/outliner_draw.c
index e4881a6f13d..a2ca3254b30 100644
--- a/source/blender/editors/space_outliner/outliner_draw.c
+++ b/source/blender/editors/space_outliner/outliner_draw.c
@@ -31,6 +31,7 @@
 #include "DNA_object_types.h"
 #include "DNA_scene_types.h"
 #include "DNA_sequence_types.h"
+#include "DNA_constraint_types.h"
 #include "DNA_object_force_types.h"
 
 #include "BLI_math.h"
@@ -60,6 +61,7 @@
 #include "ED_armature.h"
 #include "ED_keyframing.h"
 #include "ED_object.h"
+#include "ED_outliner.h"
 #include "ED_screen.h"
 
 #include "WM_api.h"
@@ -175,8 +177,9 @@ static void restrictbutton_r_lay_cb(bContext *C, void *poin, void *UNUSED(poin2)
   WM_event_add_notifier(C, NC_SCENE | ND_RENDER_OPTIONS, poin);
 }
 
-static void restrictbutton_bone_visibility_cb(bContext *C, void *UNUSED(poin), void *poin2)
+static void restrictbutton_bone_visibility_cb(bContext *C, void *poin, void *poin2)
 {
+  bArmature *arm = (bArmature *)poin;
   Bone *bone = (Bone *)poin2;
   if (bone->flag & BONE_HIDDEN_P) {
     bone->flag &= ~(BONE_SELECTED | BONE_TIPSEL | BONE_ROOTSEL);
@@ -187,6 +190,7 @@ static void restrictbutton_bone_visibility_cb(bContext *C, void *UNUSED(poin), v
   }
 
   WM_event_add_notifier(C, NC_OBJECT | ND_POSE, NULL);
+  DEG_id_tag_update(&arm->id, ID_RECALC_COPY_ON_WRITE);
 }
 
 static void restrictbutton_bone_select_cb(bContext *C, void *UNUSED(poin), void *poin2)
@@ -848,6 +852,7 @@ typedef struct RestrictProperties {
   PropertyRNA *layer_collection_holdout, *layer_collection_indirect_only,
       *layer_collection_hide_viewport;
   PropertyRNA *modifier_show_viewport, *modifier_show_render;
+  PropertyRNA *constraint_enable;
 } RestrictProperties;
 
 /* We don't care about the value of the property
@@ -865,6 +870,7 @@ typedef struct RestrictPropertiesActive {
   bool layer_collection_hide_viewport;
   bool modifier_show_viewport;
   bool modifier_show_render;
+  bool constraint_enable;
 } RestrictPropertiesActive;
 
 static void outliner_restrict_properties_enable_collection_set(
@@ -878,6 +884,7 @@ static void outliner_restrict_properties_enable_collection_set(
       props_active->layer_collection_indirect_only = false;
       props_active->object_hide_render = false;
       props_active->modifier_show_render = false;
+      props_active->constraint_enable = false;
     }
   }
 
@@ -891,6 +898,7 @@ static void outliner_restrict_properties_enable_collection_set(
       props_active->object_hide_viewport = false;
       props_active->base_hide_viewport = false;
       props_active->modifier_show_viewport = false;
+      props_active->constraint_enable = false;
     }
   }
 
@@ -995,6 +1003,8 @@ static void outliner_draw_restrictbuts(uiBlock *block,
     props.modifier_show_viewport = RNA_struct_type_find_property(&RNA_Modifier, "show_viewport");
     props.modifier_show_render = RNA_struct_type_find_property(&RNA_Modifier, "show_render");
 
+    props.constraint_enable = RNA_struct_type_find_property(&RNA_Constraint, "mute");
+
     props.initialized = true;
   }
 
@@ -1181,6 +1191,35 @@ static void outliner_draw_restrictbuts(uiBlock *block,
           }
         }
       }
+      else if (tselem->type == TSE_CONSTRAINT) {
+        bConstraint *con = (bConstraint *)te->directdata;
+
+        PointerRNA ptr;
+        RNA_pointer_create(tselem->id, &RNA_Constraint, con, &ptr);
+
+        if (soops->show_restrict_flags & SO_RESTRICT_HIDE) {
+          bt = uiDefIconButR_prop(block,
+                                  UI_BTYPE_ICON_TOGGLE,
+                                  0,
+                                  0,
+                                  (int)(ar->v2d.cur.xmax - restrict_offsets.hide),
+                                  te->ys,
+                                  UI_UNIT_X,
+                                  UI_UNIT_Y,
+                                  &ptr,
+                                  props.constraint_enable,
+                                  -1,
+                                  0,
+                                  0,
+                                  -1,
+                                  -1,
+                                  NULL);
+          UI_but_flag_enable(bt, UI_BUT_DRAG_LOCK);
+          if (!props_active.constraint_enable) {
+            UI_but_flag_enable(bt, UI_BUT_INACTIVE);
+          }
+        }
+      }
       else if (tselem->type == TSE_MODIFIER) {
         ModifierData *md = (ModifierData *)te->directdata;
 
@@ -1243,7 +1282,7 @@ static void outliner_draw_restrictbuts(uiBlock *block,
                                 UI_BTYPE_ICON_TOGGLE,
                                 BONE_HIDDEN_P,
                                 0,
-                                ICON_HIDE_OFF,
+                                ICON_RESTRICT_VIEW_OFF,
                                 (int)(ar->v2d.cur.xmax - restrict_offsets.viewport),
                                 te->ys,
                                 UI_UNIT_X,
@@ -1878,6 +1917,9 @@ TreeElementIcon tree_element_get_icon(TreeStoreElem *tselem, TreeElement *te)
       case TSE_DEFGROUP_BASE:
         data.icon = ICON_GROUP_VERTEX;
         break;
+      case TSE_DEFGROUP:
+        data.icon = ICON_GROUP_VERTEX;
+        break;
       case TSE_BONE:
       case TSE_EBONE:
         data.icon = ICON_BONE_DATA;
@@ -1885,6 +1927,100 @@ TreeElementIcon tree_element_get_icon(TreeStoreElem *tselem, TreeElement *te)
       case TSE_CONSTRAINT_BASE:
         data.icon = ICON_CONSTRAINT;
         break;
+      case TSE_CONSTRAINT: {
+        bConstraint *con = te->directdata;
+        switch ((eBConstraint_Types)con->type) {
+          case CONSTRAINT_TYPE_CAMERASOLVER:
+            data.icon = ICON_CON_CAMERASOLVER;
+            break;
+          case CONSTRAINT_TYPE_FOLLOWTRACK:
+            data.icon = ICON_CON_FOLLOWTRACK;
+            break;
+          case CONSTRAINT_TYPE_OBJECTSOLVER:
+            data.icon = ICON_CON_OBJECTSOLVER;
+            break;
+          case CONSTRAINT_TYPE_LOCLIKE:
+            data.icon = ICON_CON_LOCLIKE;
+            break;
+          case CONSTRAINT_TYPE_ROTLIKE:
+            data.icon = ICON_CON_ROTLIKE;
+            break;
+          case CONSTRAINT_TYPE_SIZELIKE:
+            data.icon = ICON_CON_SIZELIKE;
+            break;
+          case CONSTRAINT_TYPE_TRANSLIKE:
+            data.icon = ICON_CON_TRANSLIKE;
+            break;
+          case CONSTRAINT_TYPE_DISTLIMIT:
+            data.icon = ICON_CON_DISTLIMIT;
+            break;
+          case CONSTRAINT_TYPE_LOCLIMIT:
+            data.icon = ICON_CON_LOCLIMIT;
+            break;
+          case CONSTRAINT_TYPE_ROTLIMIT:
+            data.icon = ICON_CON_ROTLIMIT;
+            break;
+          case CONSTRAINT_TYPE_SIZELIMIT:
+            data.icon = ICON_CON_SIZELIMIT;
+            break;
+          case CONSTRAINT_TYPE_SAMEVOL:
+            data.icon = ICON_CON_SAMEVOL;
+            break;
+          case CONSTRAINT_TYPE_TRANSFORM:
+            data.icon = ICON_CON_TRANSFORM;
+            break;
+          case CONSTRAINT_TYPE_TRANSFORM_CACHE:
+            data.icon = ICON_CON_TRANSFORM_CACHE;
+            break;
+          case CONSTRAINT_TYPE_CLAMPTO:
+            data.icon = ICON_CON_CLAMPTO;
+            break;
+          case CONSTRAINT_TYPE_DAMPTRACK:
+            data.icon = ICON_CON_TRACKTO;
+            break;
+          case CONSTRAINT_TYPE_KINEMATIC:
+            data.icon = ICON_CON_KINEMATIC;
+            break;
+          case CONSTRAINT_TYPE_LOCKTRACK:
+            data.icon = ICON_CON_LOCKTRACK;
+            break;
+          case CONSTRAINT_TYPE_SPLINEIK:
+            data.icon = ICON_CON_SPLINEIK;
+            break;
+          case CONSTRAINT_TYPE_STRETCHTO:
+            data.icon = ICON_CON_STRETCHTO;
+            break;
+          case CONSTRAINT_TYPE_TRACKTO:
+            data.icon = ICON_CON_TRACKTO;
+            break;
+          case CONSTRAINT_TYPE_ACTION:
+            data.icon = ICON_CON_ACTION;
+            break;
+          case CONSTRAINT_TYPE_ARMATURE:
+            data.icon = ICON_CON_ARMATURE;
+            break;
+          case CONSTRAINT_TYPE_CHILDOF:
+            data.icon = ICON_CON_CHILDOF;
+            break;
+          case CONSTRAINT_TYPE_MINMAX:
+            data.icon = ICON_CON_FLOOR;
+            break;
+          case CONSTRAINT_TYPE_FOLLOWPATH:
+            data.icon = ICON_CON_FOLLOWPATH;
+            break;
+          case CONSTRAINT_TYPE_PIVOT:
+            data.icon = ICON_CON_PIVOT;
+            break;
+          case CONSTRAINT_TYPE_SHRINKWRAP:
+            data.icon = ICON_CON_SHRINKWRAP;
+            break;
+
+          default:
+            data.icon = ICON_DOT;
+            break;
+        }
+        break;
+      }
       case TSE_MODIFIER_BASE:
         data.icon = ICON_MODIFIER_DATA;
         break;
@@ -2137,30 +2273,64 @@ TreeElementIcon tree_element_get_icon(TreeStoreElem *tselem, TreeElement *te)
         data.icon = ICON_GROUP_BONE;
         break;
       case TSE_SEQUENCE:
-        if (te->idcode == SEQ_TYPE_MOVIE) {
-          data.icon = ICON_SEQUENCE;
-        }
-        else if (te->idcode == SEQ_TYPE_META) {
-          data.icon = ICON_DOT;
-        }
-        else if (te->idcode == SEQ_TYPE_SCENE) {
-          data.icon = ICON_SCENE;
-        }
-        else if (te->idcode == SEQ_TYPE_SOUND_RAM) {
-          data.icon = ICON_SOUND;
-        }
-        else if (te->idcode == SEQ_TYPE_IMAGE) {
-          data.icon = ICON_IMAGE;
-        }
-        else {
-          data.icon = ICON_PARTICLES;
+        switch (te->idcode) {
+          case SEQ_TYPE_SCENE:
+            data.icon = ICON_SCENE_DATA;
+            break;
+          case SEQ_TYPE_MOVIECLIP:
+            data.icon = ICON_TRACKER;
+            break;
+          case SEQ_TYPE_MASK:
+            data.icon = ICON_MOD_MASK;
+            break;
+          case SEQ_TYPE_MOVIE:
+            data.icon = ICON_FILE_MOVIE;
+            break;
+          case SEQ_TYPE_SOUND_RAM:
+            data.icon = ICON_SOUND;
+            break;
+          case SEQ_TYPE_IMAGE:
+            data.icon = ICON_FILE_IMAGE;
+            break;
+          case SEQ_TYPE_COLOR:
+          case SEQ_TYPE_ADJUSTMENT:
+            data.icon = ICON_COLOR;
+            break;
+          case SEQ_TYPE_TEXT:
+            data.icon = ICON_FONT_DATA;
+            break;
+          case SEQ_TYPE_ADD:
+          case SEQ_TYPE_SUB:
+          case SEQ_TYPE_MUL:
+          case SEQ_TYPE_OVERDROP:
+          case SEQ_TYPE_ALPHAOVER:
+          case SEQ_TYPE_ALPHAUNDER:
+          case SEQ_TYPE_COLORMIX:
+          case SEQ_TYPE_MULTICAM:
+          case SEQ_TYPE_TRANSFORM:
+          case SEQ_TYPE_SPEED:
+          case SEQ_TYPE_GLOW:
+          case SEQ_TYPE_GAUSSIAN_BLUR:
+            data.icon = ICON_SHADERFX;
+            break;
+          case SEQ_TYPE_CROSS:
+          case SEQ_TYPE_GAMCROSS:
+          case SEQ_TYPE_WIPE:
+            data.icon = ICON_ARROW_LEFTRIGHT;
+            break;
+          case SEQ_TYPE_META:
+            data.icon = ICON_SEQ_STRIP_META;
+            break;
+          default:
+            data.icon = ICON_DOT;
+            break;
         }
         break;
       case TSE_SEQ_STRIP:
         data.icon = ICON_LIBRARY_DATA_DIRECT;
         break;
       case TSE_SEQUENCE_DUP:
-        data.icon = ICON_OBJECT_DATA;
+        data.icon = ICON_SEQ_STRIP_DUPLICATE;
         break;
       case TSE_RNA_STRUCT:
         if (RNA_struct_is_ID(te->rnaptr.type)) {
@@ -2459,7 +2629,11 @@ static void tselem_draw_icon(uiBlock *block,
     return;
   }
 
+  /* Icon is covered by restrict buttons */
   if (!is_clickable || x >= xmax) {
+    /* Reduce alpha to match icon buttons */
+    alpha *= 0.8f;
+
     /* placement of icons, copied from interface_widgets.c */
     float aspect = (0.8f * UI_UNIT_Y) / ICON_DEFAULT_HEIGHT;
     x += 2.0f * aspect;
@@ -2567,7 +2741,6 @@ static void outliner_draw_iconrow_doit(uiBlock *block,
     float ufac = UI_UNIT_X / 20.0f;
     float icon_color[4], icon_border[4];
     outliner_icon_background_colors(icon_color, icon_border);
-    icon_color[3] *= alpha_fac;
     if (active == OL_DRAWSEL_ACTIVE) {
       UI_GetThemeColor4fv(TH_EDITED_OBJECT, icon_color);
       icon_border[3] = 0.3f;
@@ -2592,6 +2765,9 @@ static void outliner_draw_iconrow_doit(uiBlock *block,
     GPU_blend(true); /* Roundbox disables. */
   }
 
+  if (tselem->flag & TSE_HIGHLIGHTED) {
+    alpha_fac += 0.5;
+  }
   tselem_draw_icon(block, xmax, (float)*offsx, (float)ys, tselem, te, alpha_fac, false);
   te->xs = *offsx;
   te->ys = ys;
@@ -2599,7 +2775,12 @@ static void outliner_draw_iconrow_doit(uiBlock *block,
 
   if (num_elements > 1) {
     outliner_draw_iconrow_number(fstyle, *offsx, ys, num_elements);
+    te->flag |= TE_ICONROW_MERGED;
   }
+  else {
+    te->flag |= TE_ICONROW;
+  }
+
   (*offsx) += UI_UNIT_X;
 }
 
@@ -2609,7 +2790,7 @@ static void outliner_draw_iconrow_doit(uiBlock *block,
  * We use a continuum of indices until we get to the object data-blocks
  * and we then make room for the object types.
  */
-static int tree_element_id_type_to_index(TreeElement *te)
+int tree_element_id_type_to_index(TreeElement *te)
 {
   TreeStoreElem *tselem = TREESTORE(te);
 
@@ -2739,7 +2920,7 @@ static void outliner_set_coord_tree_element(TreeElement *te, int startx, int sta
   TreeElement *ten;
 
   /* closed items may be displayed in row of parent, don't change their coordinate! */
-  if ((te->flag & TE_ICONROW) == 0) {
+  if ((te->flag & TE_ICONROW) == 0 && (te->flag & TE_ICONROW_MERGED) == 0) {
     /* store coord and continue, we need coordinates for elements outside view too */
     te->xs = startx;
     te->ys = starty;
@@ -3193,6 +3374,7 @@ static void outliner_draw_highlights_recursive(unsigned pos,
                                                const SpaceOutliner *soops,
                                                const ListBase *lb,
                                                const float col_selection[4],
+                                               const float col_active[4],
                                                const float col_highlight[4],
                                                const float col_searchmatch[4],
                                                int start_x,
@@ -3206,7 +3388,11 @@ static void outliner_draw_highlights_recursive(unsigned pos,
     const int start_y = *io_start_y;
 
     /* selection status */
-    if (tselem->flag & TSE_SELECTED) {
+    if ((tselem->flag & TSE_ACTIVE) && (tselem->flag & TSE_SELECTED)) {
+      immUniformColor4fv(col_active);
+      immRecti(pos, 0, start_y, (int)ar->v2d.cur.xmax, start_y + UI_UNIT_Y);
+    }
+    else if (tselem->flag & TSE_SELECTED) {
       immUniformColor4fv(col_selection);
       immRecti(pos, 0, start_y, (int)ar->v2d.cur.xmax, start_y + UI_UNIT_Y);
     }
@@ -3260,6 +3446,7 @@ static void outliner_draw_highlights_recursive(unsigned pos,
                                          soops,
                                          &te->subtree,
                                          col_selection,
+                                         col_active,
                                          col_highlight,
                                          col_searchmatch,
                                          start_x + UI_UNIT_X,
@@ -3271,10 +3458,12 @@ static void outliner_draw_highlights_recursive(unsigned pos,
 static void outliner_draw_highlights(ARegion *ar, SpaceOutliner *soops, int startx, int *starty)
 {
   const float col_highlight[4] = {1.0f, 1.0f, 1.0f, 0.13f};
-  float col_selection[4], col_searchmatch[4];
+  float col_selection[4], col_active[4], col_searchmatch[4];
 
   UI_GetThemeColor3fv(TH_SELECT_HIGHLIGHT, col_selection);
   col_selection[3] = 1.0f; /* no alpha */
+  UI_GetThemeColor3fv(TH_SELECT_ACTIVE, col_active);
+  col_active[3] = 1.0f; /* no alpha */
   UI_GetThemeColor4fv(TH_MATCH, col_searchmatch);
   col_searchmatch[3] = 0.5f;
 
@@ -3282,8 +3471,16 @@ static void outliner_draw_highlights(ARegion *ar, SpaceOutliner *soops, int star
   GPUVertFormat *format = immVertexFormat();
   uint pos = GPU_vertformat_attr_add(format, "pos", GPU_COMP_I32, 2, GPU_FETCH_INT_TO_FLOAT);
   immBindBuiltinProgram(GPU_SHADER_2D_UNIFORM_COLOR);
-  outliner_draw_highlights_recursive(
-      pos, ar, soops, &soops->tree, col_selection, col_highlight, col_searchmatch, startx, starty);
+  outliner_draw_highlights_recursive(pos,
+                                     ar,
+                                     soops,
+                                     &soops->tree,
+                                     col_selection,
+                                     col_active,
+                                     col_highlight,
+                                     col_searchmatch,
+                                     startx,
+                                     starty);
   immUnbindProgram();
   GPU_blend(false);
 }
@@ -3439,6 +3636,17 @@ void draw_outliner(const bContext *C)
 
   outliner_build_tree(mainvar, scene, view_layer, soops, ar);  // always
 
+  /* If global sync select is dirty, flag other outliners */
+  if (ED_outliner_select_sync_is_dirty(C)) {
+    ED_outliner_select_sync_flag_outliners(C);
+  }
+
+  /* Sync selection state from view layer */
+  if (!ELEM(soops->outlinevis, SO_LIBRARIES, SO_DATA_API, SO_ID_ORPHANS) &&
+      soops->flag & SO_SYNC_SELECT) {
+    outliner_sync_selection(C, soops);
+  }
+
   /* force display to pixel coords */
   v2d->flag |= (V2D_PIXELOFS_X | V2D_PIXELOFS_Y);
   /* set matrix for 2d-view controls */
diff --git a/source/blender/editors/space_outliner/outliner_edit.c b/source/blender/editors/space_outliner/outliner_edit.c
index de6e89e47c4..943993cb810 100644
--- a/source/blender/editors/space_outliner/outliner_edit.c
+++ b/source/blender/editors/space_outliner/outliner_edit.c
@@ -101,9 +101,15 @@ static int outliner_highlight_update(bContext *C, wmOperator *UNUSED(op), const
 
   ARegion *ar = CTX_wm_region(C);
   SpaceOutliner *soops = CTX_wm_space_outliner(C);
-  const float my = UI_view2d_region_to_view_y(&ar->v2d, event->mval[1]);
 
-  TreeElement *hovered_te = outliner_find_item_at_y(soops, &soops->tree, my);
+  float view_mval[2];
+  UI_view2d_region_to_view(&ar->v2d, event->mval[0], event->mval[1], &view_mval[0], &view_mval[1]);
+
+  TreeElement *hovered_te = outliner_find_item_at_y(soops, &soops->tree, view_mval[1]);
+
+  if (hovered_te) {
+    hovered_te = outliner_find_item_at_x_in_row(soops, hovered_te, view_mval[0], NULL);
+  }
   bool changed = false;
 
   if (!hovered_te || !(hovered_te->store_elem->flag & TSE_HIGHLIGHTED)) {
@@ -134,59 +140,108 @@ void OUTLINER_OT_highlight_update(wmOperatorType *ot)
 
 /* Toggle Open/Closed ------------------------------------------- */
 
-static int do_outliner_item_openclose(
-    bContext *C, SpaceOutliner *soops, TreeElement *te, const bool all, const float mval[2])
+/* Open or close a tree element, optionally toggling all children recursively */
+void outliner_item_openclose(TreeElement *te, bool open, bool toggle_all)
 {
+  TreeStoreElem *tselem = TREESTORE(te);
 
-  if (mval[1] > te->ys && mval[1] < te->ys + UI_UNIT_Y) {
-    TreeStoreElem *tselem = TREESTORE(te);
+  if (open) {
+    tselem->flag &= ~TSE_CLOSED;
+  }
+  else {
+    tselem->flag |= TSE_CLOSED;
+  }
 
-    /* all below close/open? */
-    if (all) {
-      tselem->flag &= ~TSE_CLOSED;
-      outliner_flag_set(
-          &te->subtree, TSE_CLOSED, !outliner_flag_is_any_test(&te->subtree, TSE_CLOSED, 1));
-    }
-    else {
-      if (tselem->flag & TSE_CLOSED) {
-        tselem->flag &= ~TSE_CLOSED;
-      }
-      else {
-        tselem->flag |= TSE_CLOSED;
+  if (toggle_all) {
+    outliner_flag_set(&te->subtree, TSE_CLOSED, !open);
+  }
+}
+
+typedef struct OpenCloseData {
+  TreeStoreElem *prev_tselem;
+  bool open;
+  int x_location;
+} OpenCloseData;
+
+static int outliner_item_openclose_modal(bContext *C, wmOperator *op, const wmEvent *event)
+{
+  ARegion *ar = CTX_wm_region(C);
+  SpaceOutliner *soops = CTX_wm_space_outliner(C);
+
+  float view_mval[2];
+  UI_view2d_region_to_view(&ar->v2d, event->mval[0], event->mval[1], &view_mval[0], &view_mval[1]);
+
+  if (event->type == MOUSEMOVE) {
+    TreeElement *te = outliner_find_item_at_y(soops, &soops->tree, view_mval[1]);
+
+    OpenCloseData *data = (OpenCloseData *)op->customdata;
+
+    /* Only openclose if mouse is not over the previously toggled element */
+    if (te && TREESTORE(te) != data->prev_tselem) {
+
+      /* Only toggle openclose on the same level as the first clicked element */
+      if (te->xs == data->x_location) {
+        outliner_item_openclose(te, data->open, false);
+        ED_region_tag_redraw(ar);
       }
     }
 
-    return 1;
+    if (te) {
+      data->prev_tselem = TREESTORE(te);
+    }
+    else {
+      data->prev_tselem = NULL;
+    }
   }
+  else if (event->val == KM_RELEASE) {
+    MEM_freeN(op->customdata);
 
-  for (te = te->subtree.first; te; te = te->next) {
-    if (do_outliner_item_openclose(C, soops, te, all, mval)) {
-      return 1;
-    }
+    return OPERATOR_FINISHED;
   }
-  return 0;
+
+  return OPERATOR_RUNNING_MODAL;
 }
 
-/* event can enterkey, then it opens/closes */
-static int outliner_item_openclose(bContext *C, wmOperator *op, const wmEvent *event)
+static int outliner_item_openclose_invoke(bContext *C, wmOperator *op, const wmEvent *event)
 {
   ARegion *ar = CTX_wm_region(C);
   SpaceOutliner *soops = CTX_wm_space_outliner(C);
-  TreeElement *te;
-  float fmval[2];
-  const bool all = RNA_boolean_get(op->ptr, "all");
 
-  UI_view2d_region_to_view(&ar->v2d, event->mval[0], event->mval[1], &fmval[0], &fmval[1]);
+  const bool toggle_all = RNA_boolean_get(op->ptr, "all");
 
-  for (te = soops->tree.first; te; te = te->next) {
-    if (do_outliner_item_openclose(C, soops, te, all, fmval)) {
-      break;
+  float view_mval[2];
+  UI_view2d_region_to_view(&ar->v2d, event->mval[0], event->mval[1], &view_mval[0], &view_mval[1]);
+
+  TreeElement *te = outliner_find_item_at_y(soops, &soops->tree, view_mval[1]);
+
+  if (te && outliner_item_is_co_within_close_toggle(te, view_mval[0])) {
+    TreeStoreElem *tselem = TREESTORE(te);
+
+    const bool open = (tselem->flag & TSE_CLOSED) ||
+                      (toggle_all && (outliner_flag_is_any_test(&te->subtree, TSE_CLOSED, 1)));
+
+    outliner_item_openclose(te, open, toggle_all);
+    ED_region_tag_redraw(ar);
+
+    /* Only toggle once for single click toggling */
+    if (event->type == LEFTMOUSE) {
+      return OPERATOR_FINISHED;
     }
-  }
 
-  ED_region_tag_redraw(ar);
+    /* Store last expanded tselem and x coordinate of disclosure triangle */
+    OpenCloseData *toggle_data = MEM_callocN(sizeof(OpenCloseData), "open_close_data");
+    toggle_data->prev_tselem = tselem;
+    toggle_data->open = open;
+    toggle_data->x_location = te->xs;
 
-  return OPERATOR_FINISHED;
+    /* Store the first clicked on element */
+    op->customdata = toggle_data;
+
+    WM_event_add_modal_handler(C, op);
+    return OPERATOR_RUNNING_MODAL;
+  }
+
+  return OPERATOR_CANCELLED | OPERATOR_PASS_THROUGH;
 }
 
 void OUTLINER_OT_item_openclose(wmOperatorType *ot)
@@ -195,11 +250,12 @@ void OUTLINER_OT_item_openclose(wmOperatorType *ot)
   ot->idname = "OUTLINER_OT_item_openclose";
   ot->description = "Toggle whether item under cursor is enabled or closed";
 
-  ot->invoke = outliner_item_openclose;
+  ot->invoke = outliner_item_openclose_invoke;
+  ot->modal = outliner_item_openclose_modal;
 
   ot->poll = ED_operator_outliner_active;
 
-  RNA_def_boolean(ot->srna, "all", 1, "All", "Close or open all items");
+  RNA_def_boolean(ot->srna, "all", false, "All", "Close or open all items");
 }
 
 /* -------------------------------------------------------------------- */
@@ -330,10 +386,10 @@ void item_rename_cb(bContext *C,
   do_item_rename(ar, te, tselem, reports);
 }
 
-static int do_outliner_item_rename(ReportList *reports,
-                                   ARegion *ar,
-                                   TreeElement *te,
-                                   const float mval[2])
+static void do_outliner_item_rename(ReportList *reports,
+                                    ARegion *ar,
+                                    TreeElement *te,
+                                    const float mval[2])
 {
   if (mval[1] > te->ys && mval[1] < te->ys + UI_UNIT_Y) {
     TreeStoreElem *tselem = TREESTORE(te);
@@ -341,17 +397,12 @@ static int do_outliner_item_rename(ReportList *reports,
     /* click on name */
     if (mval[0] > te->xs + UI_UNIT_X * 2 && mval[0] < te->xend) {
       do_item_rename(ar, te, tselem, reports);
-      return 1;
     }
-    return 0;
   }
 
   for (te = te->subtree.first; te; te = te->next) {
-    if (do_outliner_item_rename(reports, ar, te, mval)) {
-      return 1;
-    }
+    do_outliner_item_rename(reports, ar, te, mval);
   }
-  return 0;
 }
 
 static int outliner_item_rename(bContext *C, wmOperator *op, const wmEvent *event)
@@ -360,25 +411,34 @@ static int outliner_item_rename(bContext *C, wmOperator *op, const wmEvent *even
   SpaceOutliner *soops = CTX_wm_space_outliner(C);
   TreeElement *te;
   float fmval[2];
-  bool changed = false;
 
-  UI_view2d_region_to_view(&ar->v2d, event->mval[0], event->mval[1], &fmval[0], &fmval[1]);
+  /* Rename active element if key pressed, otherwise rename element at cursor coordinates */
+  if (event->val == KM_PRESS) {
+    TreeElement *active_element = outliner_find_element_with_flag(&soops->tree, TSE_ACTIVE);
 
-  for (te = soops->tree.first; te; te = te->next) {
-    if (do_outliner_item_rename(op->reports, ar, te, fmval)) {
-      changed = true;
-      break;
+    if (active_element) {
+      do_item_rename(ar, active_element, TREESTORE(active_element), op->reports);
+    }
+    else {
+      BKE_report(op->reports, RPT_WARNING, "No active item to rename");
     }
   }
+  else {
+    UI_view2d_region_to_view(&ar->v2d, event->mval[0], event->mval[1], &fmval[0], &fmval[1]);
 
-  return changed ? OPERATOR_FINISHED : OPERATOR_PASS_THROUGH;
+    for (te = soops->tree.first; te; te = te->next) {
+      do_outliner_item_rename(op->reports, ar, te, fmval);
+    }
+  }
+
+  return OPERATOR_FINISHED;
 }
 
 void OUTLINER_OT_item_rename(wmOperatorType *ot)
 {
   ot->name = "Rename";
   ot->idname = "OUTLINER_OT_item_rename";
-  ot->description = "Rename item under cursor";
+  ot->description = "Rename the active element";
 
   ot->invoke = outliner_item_rename;
 
@@ -1103,6 +1163,10 @@ static int outliner_select_all_exec(bContext *C, wmOperator *op)
       break;
   }
 
+  if (soops->flag & SO_SYNC_SELECT) {
+    ED_outliner_select_sync_from_outliner(C, soops);
+  }
+
   DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
   WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
   ED_region_tag_redraw_no_rebuild(ar);
@@ -1179,20 +1243,17 @@ static int outliner_open_back(TreeElement *te)
   return retval;
 }
 
-static int outliner_show_active_exec(bContext *C, wmOperator *UNUSED(op))
+/* Return element representing the active base or bone in the outliner, or NULL if none exists */
+static TreeElement *outliner_show_active_get_element(bContext *C,
+                                                     SpaceOutliner *so,
+                                                     ViewLayer *view_layer)
 {
-  SpaceOutliner *so = CTX_wm_space_outliner(C);
-  ViewLayer *view_layer = CTX_data_view_layer(C);
-  ARegion *ar = CTX_wm_region(C);
-  View2D *v2d = &ar->v2d;
-
   TreeElement *te;
-  int xdelta, ytop;
 
   Object *obact = OBACT(view_layer);
 
   if (!obact) {
-    return OPERATOR_CANCELLED;
+    return NULL;
   }
 
   te = outliner_find_id(so, &so->tree, &obact->id);
@@ -1215,25 +1276,50 @@ static int outliner_show_active_exec(bContext *C, wmOperator *UNUSED(op))
     }
   }
 
-  if (te) {
-    /* open up tree to active object/bone */
+  return te;
+}
+
+static void outliner_show_active(SpaceOutliner *so, ARegion *ar, TreeElement *te, ID *id)
+{
+  /* open up tree to active object/bone */
+  if (TREESTORE(te)->id == id) {
     if (outliner_open_back(te)) {
       outliner_set_coordinates(ar, so);
     }
+    return;
+  }
+
+  for (TreeElement *ten = te->subtree.first; ten; ten = ten->next) {
+    outliner_show_active(so, ar, ten, id);
+  }
+}
+
+static int outliner_show_active_exec(bContext *C, wmOperator *UNUSED(op))
+{
+  SpaceOutliner *so = CTX_wm_space_outliner(C);
+  ViewLayer *view_layer = CTX_data_view_layer(C);
+  ARegion *ar = CTX_wm_region(C);
+  View2D *v2d = &ar->v2d;
+
+  TreeElement *active_element = outliner_show_active_get_element(C, so, view_layer);
 
-    /* make te->ys center of view */
-    ytop = te->ys + BLI_rcti_size_y(&v2d->mask) / 2;
-    if (ytop > 0) {
-      ytop = 0;
+  if (active_element) {
+    ID *id = TREESTORE(active_element)->id;
+
+    /* Expand all elements in the outliner with matching ID */
+    for (TreeElement *te = so->tree.first; te; te = te->next) {
+      outliner_show_active(so, ar, te, id);
     }
 
-    v2d->cur.ymax = (float)ytop;
-    v2d->cur.ymin = (float)(ytop - BLI_rcti_size_y(&v2d->mask));
+    /* Center view on first element found */
+    int size_y = BLI_rcti_size_y(&v2d->mask) + 1;
+    int ytop = (active_element->ys + (size_y / 2));
+    int delta_y = ytop - v2d->cur.ymax;
 
-    /* make te->xs ==> te->xend center of view */
-    xdelta = (int)(te->xs - v2d->cur.xmin);
-    v2d->cur.xmin += xdelta;
-    v2d->cur.xmax += xdelta;
+    outliner_scroll_view(ar, delta_y);
+  }
+  else {
+    return OPERATOR_CANCELLED;
   }
 
   ED_region_tag_redraw_no_rebuild(ar);
@@ -1259,18 +1345,15 @@ void OUTLINER_OT_show_active(wmOperatorType *ot)
 static int outliner_scroll_page_exec(bContext *C, wmOperator *op)
 {
   ARegion *ar = CTX_wm_region(C);
-  int dy = BLI_rcti_size_y(&ar->v2d.mask);
-  int up = 0;
+  int size_y = BLI_rcti_size_y(&ar->v2d.mask) + 1;
 
-  if (RNA_boolean_get(op->ptr, "up")) {
-    up = 1;
-  }
+  bool up = RNA_boolean_get(op->ptr, "up");
 
-  if (up == 0) {
-    dy = -dy;
+  if (!up) {
+    size_y = -size_y;
   }
-  ar->v2d.cur.ymin += dy;
-  ar->v2d.cur.ymax += dy;
+
+  outliner_scroll_view(ar, size_y);
 
   ED_region_tag_redraw_no_rebuild(ar);
 
@@ -1697,7 +1780,7 @@ static void tree_element_to_path(TreeElement *te,
       /* no ID, so check if entry is RNA-struct,
        * and if that RNA-struct is an ID datablock to extract info from. */
       if (tse->type == TSE_RNA_STRUCT) {
-        /* ptr->data not ptr->id.data seems to be the one we want,
+        /* ptr->data not ptr->owner_id seems to be the one we want,
          * since ptr->data is sometimes the owner of this ID? */
         if (RNA_struct_is_ID(ptr->type)) {
           *id = (ID *)ptr->data;
diff --git a/source/blender/editors/space_outliner/outliner_intern.h b/source/blender/editors/space_outliner/outliner_intern.h
index fa28d119244..95e37dea249 100644
--- a/source/blender/editors/space_outliner/outliner_intern.h
+++ b/source/blender/editors/space_outliner/outliner_intern.h
@@ -40,7 +40,6 @@ struct TreeStoreElem;
 struct ViewLayer;
 struct bContext;
 struct bPoseChannel;
-struct wmEvent;
 struct wmKeyConfig;
 struct wmOperatorType;
 
@@ -50,6 +49,14 @@ typedef enum TreeElementInsertType {
   TE_INSERT_INTO,
 } TreeElementInsertType;
 
+/* Use generic walk select after D4771 is committed */
+typedef enum WalkSelectDirection {
+  OUTLINER_SELECT_WALK_UP,
+  OUTLINER_SELECT_WALK_DOWN,
+  OUTLINER_SELECT_WALK_LEFT,
+  OUTLINER_SELECT_WALK_RIGHT,
+} WalkSelectDirection;
+
 typedef enum TreeTraversalAction {
   /* Continue traversal regularly, don't skip children. */
   TRAVERSE_CONTINUE = 0,
@@ -131,6 +138,9 @@ enum {
   TE_DISABLED = (1 << 4),
   TE_DRAGGING = (1 << 5),
   TE_CHILD_NOT_IN_COLLECTION = (1 << 6),
+  /* Child elements of the same type in the icon-row are drawn merged as one icon.
+   * This flag is set for an element that is part of these merged child icons. */
+  TE_ICONROW_MERGED = (1 << 7),
 };
 
 /* button events */
@@ -223,6 +233,8 @@ void outliner_collection_isolate_flag(struct Scene *scene,
                                       const char *propname,
                                       const bool value);
 
+int tree_element_id_type_to_index(TreeElement *te);
+
 /* outliner_select.c -------------------------------------------- */
 eOLDrawState tree_element_type_active(struct bContext *C,
                                       struct Scene *scene,
@@ -253,6 +265,10 @@ void outliner_object_mode_toggle(struct bContext *C,
                                  ViewLayer *view_layer,
                                  Base *base);
 
+void outliner_element_activate(struct SpaceOutliner *soops, struct TreeStoreElem *tselem);
+
+bool outliner_item_is_co_within_close_toggle(TreeElement *te, float view_co_x);
+
 /* outliner_edit.c ---------------------------------------------- */
 typedef void (*outliner_operation_cb)(struct bContext *C,
                                       struct ReportList *,
@@ -337,6 +353,8 @@ void item_object_mode_exit_cb(struct bContext *C,
 
 void outliner_set_coordinates(struct ARegion *ar, struct SpaceOutliner *soops);
 
+void outliner_item_openclose(TreeElement *te, bool open, bool toggle_all);
+
 /* outliner_dragdrop.c */
 void outliner_dropboxes(void);
 
@@ -364,6 +382,7 @@ void OUTLINER_OT_show_active(struct wmOperatorType *ot);
 void OUTLINER_OT_show_hierarchy(struct wmOperatorType *ot);
 
 void OUTLINER_OT_select_box(struct wmOperatorType *ot);
+void OUTLINER_OT_select_walk(struct wmOperatorType *ot);
 
 void OUTLINER_OT_select_all(struct wmOperatorType *ot);
 void OUTLINER_OT_expanded_toggle(struct wmOperatorType *ot);
@@ -380,6 +399,10 @@ void OUTLINER_OT_orphans_purge(struct wmOperatorType *ot);
 
 /* outliner_tools.c ---------------------------------------------- */
 
+void merged_element_search_menu_invoke(struct bContext *C,
+                                       TreeElement *parent_te,
+                                       TreeElement *activate_te);
+
 void OUTLINER_OT_operation(struct wmOperatorType *ot);
 void OUTLINER_OT_scene_operation(struct wmOperatorType *ot);
 void OUTLINER_OT_object_operation(struct wmOperatorType *ot);
@@ -439,7 +462,8 @@ TreeElement *outliner_find_item_at_y(const SpaceOutliner *soops,
                                      float view_co_y);
 TreeElement *outliner_find_item_at_x_in_row(const SpaceOutliner *soops,
                                             const TreeElement *parent_te,
-                                            float view_co_x);
+                                            float view_co_x,
+                                            bool *multiple_objects);
 TreeElement *outliner_find_tse(struct SpaceOutliner *soops, const TreeStoreElem *tse);
 TreeElement *outliner_find_tree_element(ListBase *lb, const TreeStoreElem *store_elem);
 TreeElement *outliner_find_parent_element(ListBase *lb,
@@ -456,5 +480,12 @@ bool outliner_tree_traverse(const SpaceOutliner *soops,
                             TreeTraversalFunc func,
                             void *customdata);
 float outliner_restrict_columns_width(const struct SpaceOutliner *soops);
+TreeElement *outliner_find_element_with_flag(const ListBase *lb, short flag);
+bool outliner_is_element_visible(const TreeElement *te);
+void outliner_scroll_view(struct ARegion *ar, int delta_y);
+
+/* outliner_sync.c ---------------------------------------------- */
+
+void outliner_sync_selection(const struct bContext *C, struct SpaceOutliner *soops);
 
 #endif /* __OUTLINER_INTERN_H__ */
diff --git a/source/blender/editors/space_outliner/outliner_ops.c b/source/blender/editors/space_outliner/outliner_ops.c
index f155a2d5f89..4b57d4ad771 100644
--- a/source/blender/editors/space_outliner/outliner_ops.c
+++ b/source/blender/editors/space_outliner/outliner_ops.c
@@ -50,6 +50,7 @@ void outliner_operatortypes(void)
   WM_operatortype_append(OUTLINER_OT_highlight_update);
   WM_operatortype_append(OUTLINER_OT_item_activate);
   WM_operatortype_append(OUTLINER_OT_select_box);
+  WM_operatortype_append(OUTLINER_OT_select_walk);
   WM_operatortype_append(OUTLINER_OT_item_openclose);
   WM_operatortype_append(OUTLINER_OT_item_rename);
   WM_operatortype_append(OUTLINER_OT_item_drag_drop);
diff --git a/source/blender/editors/space_outliner/outliner_select.c b/source/blender/editors/space_outliner/outliner_select.c
index 7f45c4d22fa..44e67fa1508 100644
--- a/source/blender/editors/space_outliner/outliner_select.c
+++ b/source/blender/editors/space_outliner/outliner_select.c
@@ -51,14 +51,16 @@
 #include "BKE_workspace.h"
 
 #include "DEG_depsgraph.h"
+#include "DEG_depsgraph_build.h"
 
 #include "ED_armature.h"
+#include "ED_gpencil.h"
 #include "ED_object.h"
+#include "ED_outliner.h"
 #include "ED_screen.h"
 #include "ED_select_utils.h"
 #include "ED_sequencer.h"
 #include "ED_undo.h"
-#include "ED_gpencil.h"
 
 #include "WM_api.h"
 #include "WM_types.h"
@@ -251,9 +253,7 @@ static eOLDrawState active_viewlayer(bContext *C,
 }
 
 /**
- * Select object tree:
- * CTRL+LMB: Select/Deselect object and all children.
- * CTRL+SHIFT+LMB: Add/Remove object and all children.
+ * Select object tree
  */
 static void do_outliner_object_select_recursive(ViewLayer *view_layer,
                                                 Object *ob_parent,
@@ -450,9 +450,9 @@ static eOLDrawState tree_element_active_material(bContext *C,
   return OL_DRAWSEL_NONE;
 }
 
-static eOLDrawState tree_element_active_camera(bContext *UNUSED(C),
+static eOLDrawState tree_element_active_camera(bContext *C,
                                                Scene *scene,
-                                               ViewLayer *UNUSED(sl),
+                                               ViewLayer *UNUSED(view_layer),
                                                SpaceOutliner *soops,
                                                TreeElement *te,
                                                const eOLSetState set)
@@ -460,10 +460,21 @@ static eOLDrawState tree_element_active_camera(bContext *UNUSED(C),
   Object *ob = (Object *)outliner_search_back(soops, te, ID_OB);
 
   if (set != OL_SETSEL_NONE) {
+    scene->camera = ob;
+
+    Main *bmain = CTX_data_main(C);
+    wmWindowManager *wm = bmain->wm.first;
+
+    WM_windows_scene_data_sync(&wm->windows, scene);
+    DEG_id_tag_update(&scene->id, ID_RECALC_COPY_ON_WRITE);
+    DEG_relations_tag_update(bmain);
+    WM_event_add_notifier(C, NC_SCENE | NA_EDITED, NULL);
+
     return OL_DRAWSEL_NONE;
   }
-
-  return scene->camera == ob;
+  else {
+    return scene->camera == ob;
+  }
 }
 
 static eOLDrawState tree_element_active_world(bContext *C,
@@ -1083,6 +1094,13 @@ eOLDrawState tree_element_type_active(bContext *C,
 
 /* ================================================ */
 
+/* Activate a tree store element and set the walk navigation start element */
+void outliner_element_activate(SpaceOutliner *soops, TreeStoreElem *tselem)
+{
+  outliner_flag_set(&soops->tree, TSE_ACTIVE | TSE_ACTIVE_WALK, false);
+  tselem->flag |= TSE_ACTIVE | TSE_ACTIVE_WALK;
+}
+
 /**
  * Action when clicking to activate an item (typically under the mouse cursor),
  * but don't do any cursor intersection checks.
@@ -1114,7 +1132,8 @@ static void do_outliner_item_activate_tree_element(bContext *C,
   else if (tselem->type == TSE_POSE_BASE) {
     /* Support pose mode toggle, keeping the active object as is. */
   }
-  else {
+  else if (soops->flag & SO_SYNC_SELECT) {
+    /* Only activate when synced selection is enabled */
     tree_element_set_active_object(C,
                                    scene,
                                    view_layer,
@@ -1125,6 +1144,9 @@ static void do_outliner_item_activate_tree_element(bContext *C,
                                    recursive && tselem->type == 0);
   }
 
+  /* Mark as active in the outliner */
+  outliner_element_activate(soops, tselem);
+
   if (tselem->type == 0) {  // the lib blocks
     /* editmode? */
     if (te->idcode == ID_SCE) {
@@ -1189,7 +1211,7 @@ static void do_outliner_item_activate_tree_element(bContext *C,
       tree_element_active(C, scene, view_layer, soops, te, OL_SETSEL_NORMAL, false);
     }
   }
-  else {
+  else if (soops->flag & SO_SYNC_SELECT) {
     tree_element_type_active(C,
                              scene,
                              view_layer,
@@ -1211,7 +1233,8 @@ void outliner_item_select(SpaceOutliner *soops,
                           const bool toggle)
 {
   TreeStoreElem *tselem = TREESTORE(te);
-  const short new_flag = toggle ? (tselem->flag ^ TSE_SELECTED) : (tselem->flag | TSE_SELECTED);
+  const short new_flag = (toggle && (tselem->flag & TSE_ACTIVE)) ? (tselem->flag ^ TSE_SELECTED) :
+                                                                   (tselem->flag | TSE_SELECTED);
 
   if (extend == false) {
     outliner_flag_set(&soops->tree, TSE_SELECTED, false);
@@ -1219,24 +1242,72 @@ void outliner_item_select(SpaceOutliner *soops,
   tselem->flag = new_flag;
 }
 
-static void outliner_item_toggle_closed(TreeElement *te, const bool toggle_children)
+static bool do_outliner_range_select_recursive(ListBase *lb,
+                                               TreeElement *active,
+                                               TreeElement *cursor,
+                                               bool selecting)
 {
-  TreeStoreElem *tselem = TREESTORE(te);
-  if (toggle_children) {
-    tselem->flag &= ~TSE_CLOSED;
+  for (TreeElement *te = lb->first; te; te = te->next) {
+    TreeStoreElem *tselem = TREESTORE(te);
 
-    const bool all_opened = !outliner_flag_is_any_test(&te->subtree, TSE_CLOSED, 1);
-    outliner_flag_set(&te->subtree, TSE_CLOSED, all_opened);
-  }
-  else {
-    tselem->flag ^= TSE_CLOSED;
+    if (selecting) {
+      tselem->flag |= TSE_SELECTED;
+    }
+
+    /* Set state for selection */
+    if (te == active || te == cursor) {
+      selecting = !selecting;
+    }
+
+    if (selecting) {
+      tselem->flag |= TSE_SELECTED;
+    }
+
+    /* Don't look inside closed elements */
+    if (!(tselem->flag & TSE_CLOSED)) {
+      selecting = do_outliner_range_select_recursive(&te->subtree, active, cursor, selecting);
+    }
   }
+
+  return selecting;
 }
 
-static bool outliner_item_is_co_within_close_toggle(TreeElement *te, float view_co_x)
+/* Select a range of items between cursor and active element */
+static void do_outliner_range_select(bContext *C,
+                                     SpaceOutliner *soops,
+                                     TreeElement *cursor,
+                                     const bool extend)
 {
-  return ((te->flag & TE_ICONROW) == 0) && (view_co_x > te->xs) &&
-         (view_co_x < te->xs + UI_UNIT_X);
+  TreeElement *active = outliner_find_element_with_flag(&soops->tree, TSE_ACTIVE);
+
+  /* If no active element exists, activate the element under the cursor */
+  if (!active) {
+    outliner_item_select(soops, cursor, false, false);
+    outliner_item_do_activate_from_tree_element(C, cursor, TREESTORE(cursor), false, false);
+    return;
+  }
+
+  TreeStoreElem *tselem = TREESTORE(active);
+  const bool active_selected = (tselem->flag & TSE_SELECTED);
+
+  if (!extend) {
+    outliner_flag_set(&soops->tree, TSE_SELECTED, false);
+  }
+
+  /* Select active if under cursor */
+  if (active == cursor) {
+    TREESTORE(cursor)->flag |= TSE_SELECTED;
+    return;
+  }
+
+  /* If active is not selected, select the element under the cursor */
+  if (!active_selected || !outliner_is_element_visible(active)) {
+    outliner_item_select(soops, cursor, false, false);
+    outliner_item_do_activate_from_tree_element(C, cursor, TREESTORE(cursor), false, false);
+    return;
+  }
+
+  do_outliner_range_select_recursive(&soops->tree, active, cursor, false);
 }
 
 static bool outliner_is_co_within_restrict_columns(const SpaceOutliner *soops,
@@ -1247,7 +1318,7 @@ static bool outliner_is_co_within_restrict_columns(const SpaceOutliner *soops,
 }
 
 /**
- * A version of #outliner_item_do_acticate_from_cursor that takes the tree element directly.
+ * A version of #outliner_item_do_activate_from_cursor that takes the tree element directly.
  * and doesn't depend on the pointer position.
  *
  * This allows us to simulate clicking on an item without dealing with the mouse cursor.
@@ -1271,10 +1342,11 @@ void outliner_item_do_activate_from_tree_element(
 static int outliner_item_do_activate_from_cursor(bContext *C,
                                                  const int mval[2],
                                                  const bool extend,
-                                                 const bool recursive,
+                                                 const bool use_range,
                                                  const bool deselect_all)
 {
   ARegion *ar = CTX_wm_region(C);
+  Scene *scene = CTX_data_scene(C);
   SpaceOutliner *soops = CTX_wm_space_outliner(C);
   TreeElement *te;
   float view_mval[2];
@@ -1292,28 +1364,36 @@ static int outliner_item_do_activate_from_cursor(bContext *C,
       changed = true;
     }
   }
-  else if (outliner_item_is_co_within_close_toggle(te, view_mval[0])) {
-    outliner_item_toggle_closed(te, extend);
-    changed = true;
-    rebuild_tree = true;
+  /* Don't allow toggle on scene collection */
+  else if ((TREESTORE(te)->type != TSE_VIEW_COLLECTION_BASE) &&
+           outliner_item_is_co_within_close_toggle(te, view_mval[0])) {
+    return OPERATOR_CANCELLED | OPERATOR_PASS_THROUGH;
   }
   else {
-    Scene *scene = CTX_data_scene(C);
     ViewLayer *view_layer = CTX_data_view_layer(C);
-    /* the row may also contain children, if one is hovered we want this instead of current te */
-    TreeElement *activate_te = outliner_find_item_at_x_in_row(soops, te, view_mval[0]);
-    TreeStoreElem *activate_tselem = TREESTORE(activate_te);
 
-    outliner_item_select(soops, activate_te, extend, extend);
+    /* The row may also contain children, if one is hovered we want this instead of current te */
+    bool merged_elements = false;
+    TreeElement *activate_te = outliner_find_item_at_x_in_row(
+        soops, te, view_mval[0], &merged_elements);
 
-    /* Only change modes when clicking on the icon/text,
-     * otherwise we can't easily select without changing modes. */
-    if ((te->flag & TE_ICONROW) == 0) {
-      if (view_mval[0] >= te->xs && view_mval[0] <= te->xend) {
-        do_outliner_item_activate_tree_element(
-            C, scene, view_layer, soops, activate_te, activate_tselem, extend, recursive);
-      }
+    /* If the selected icon was an aggregate of multiple elements, run the search popup */
+    if (merged_elements) {
+      merged_element_search_menu_invoke(C, te, activate_te);
+      return OPERATOR_CANCELLED;
+    }
+
+    TreeStoreElem *activate_tselem = TREESTORE(activate_te);
+
+    if (use_range) {
+      do_outliner_range_select(C, soops, activate_te, extend);
+    }
+    else {
+      outliner_item_select(soops, activate_te, extend, extend);
+      do_outliner_item_activate_tree_element(
+          C, scene, view_layer, soops, activate_te, activate_tselem, extend, false);
     }
+
     changed = true;
   }
 
@@ -1324,7 +1404,10 @@ static int outliner_item_do_activate_from_cursor(bContext *C,
     else {
       ED_region_tag_redraw_no_rebuild(ar);
     }
-    ED_undo_push(C, "Outliner selection change");
+
+    if (soops->flag & SO_SYNC_SELECT) {
+      ED_outliner_select_sync_from_outliner(C, soops);
+    }
   }
 
   return OPERATOR_FINISHED;
@@ -1334,9 +1417,9 @@ static int outliner_item_do_activate_from_cursor(bContext *C,
 static int outliner_item_activate_invoke(bContext *C, wmOperator *op, const wmEvent *event)
 {
   const bool extend = RNA_boolean_get(op->ptr, "extend");
-  const bool recursive = RNA_boolean_get(op->ptr, "recursive");
+  const bool use_range = RNA_boolean_get(op->ptr, "extend_range");
   const bool deselect_all = RNA_boolean_get(op->ptr, "deselect_all");
-  return outliner_item_do_activate_from_cursor(C, event->mval, extend, recursive, deselect_all);
+  return outliner_item_do_activate_from_cursor(C, event->mval, extend, use_range, deselect_all);
 }
 
 void OUTLINER_OT_item_activate(wmOperatorType *ot)
@@ -1349,9 +1432,14 @@ void OUTLINER_OT_item_activate(wmOperatorType *ot)
 
   ot->poll = ED_operator_outliner_active;
 
+  ot->flag |= OPTYPE_REGISTER | OPTYPE_UNDO;
+
   PropertyRNA *prop;
   RNA_def_boolean(ot->srna, "extend", true, "Extend", "Extend selection for activation");
-  RNA_def_boolean(ot->srna, "recursive", false, "Recursive", "Select Objects and their children");
+  prop = RNA_def_boolean(
+      ot->srna, "extend_range", false, "Extend Range", "Select a range from active element");
+  RNA_def_property_flag(prop, PROP_SKIP_SAVE);
+
   prop = RNA_def_boolean(ot->srna,
                          "deselect_all",
                          false,
@@ -1409,9 +1497,44 @@ static int outliner_box_select_exec(bContext *C, wmOperator *op)
   WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
   ED_region_tag_redraw(ar);
 
+  if (soops->flag & SO_SYNC_SELECT) {
+    ED_outliner_select_sync_from_outliner(C, soops);
+  }
+
   return OPERATOR_FINISHED;
 }
 
+/* Find if x coordinate is over an icon or name */
+static bool outliner_item_is_co_over_name_icons(TreeElement *te, float view_co_x)
+{
+  /* Special case: count area left of Scene Collection as empty space */
+  bool outside_left = (TREESTORE(te)->type == TSE_VIEW_COLLECTION_BASE) ?
+                          (view_co_x > te->xs + UI_UNIT_X) :
+                          (view_co_x > te->xs);
+
+  return outside_left && (view_co_x < te->xend);
+}
+
+static int outliner_box_select_invoke(bContext *C, wmOperator *op, const wmEvent *event)
+{
+  SpaceOutliner *soops = CTX_wm_space_outliner(C);
+  ARegion *ar = CTX_wm_region(C);
+  float view_mval[2];
+  const bool tweak = RNA_boolean_get(op->ptr, "tweak");
+
+  UI_view2d_region_to_view(&ar->v2d, event->mval[0], event->mval[1], &view_mval[0], &view_mval[1]);
+
+  /* Find element clicked on */
+  TreeElement *te = outliner_find_item_at_y(soops, &soops->tree, view_mval[1]);
+
+  /* Pass through if click is over name or icons, or not tweak event */
+  if (te && tweak && outliner_item_is_co_over_name_icons(te, view_mval[0])) {
+    return OPERATOR_CANCELLED | OPERATOR_PASS_THROUGH;
+  }
+
+  return WM_gesture_box_invoke(C, op, event);
+}
+
 void OUTLINER_OT_select_box(wmOperatorType *ot)
 {
   /* identifiers */
@@ -1420,7 +1543,7 @@ void OUTLINER_OT_select_box(wmOperatorType *ot)
   ot->description = "Use box selection to select tree elements";
 
   /* api callbacks */
-  ot->invoke = WM_gesture_box_invoke;
+  ot->invoke = outliner_box_select_invoke;
   ot->exec = outliner_box_select_exec;
   ot->modal = WM_gesture_box_modal;
   ot->cancel = WM_gesture_box_cancel;
@@ -1431,8 +1554,242 @@ void OUTLINER_OT_select_box(wmOperatorType *ot)
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 
   /* properties */
+  PropertyRNA *prop;
+
+  prop = RNA_def_boolean(
+      ot->srna, "tweak", false, "Tweak", "Tweak gesture from empty space for box selection");
+  RNA_def_property_flag(prop, PROP_SKIP_SAVE);
+
   WM_operator_properties_gesture_box(ot);
   WM_operator_properties_select_operation_simple(ot);
 }
 
 /* ****************************************************** */
+
+/* **************** Walk Select Tool ****************** */
+
+/* Given a tree element return the rightmost child that is visible in the outliner */
+static TreeElement *outliner_find_rightmost_visible_child(SpaceOutliner *soops, TreeElement *te)
+{
+  while (te->subtree.last) {
+    if (TSELEM_OPEN(TREESTORE(te), soops)) {
+      te = te->subtree.last;
+    }
+    else {
+      break;
+    }
+  }
+  return te;
+}
+
+/* Find previous visible element in the tree  */
+static TreeElement *outliner_find_previous_element(SpaceOutliner *soops, TreeElement *walk_element)
+{
+  if (walk_element->prev) {
+    walk_element = outliner_find_rightmost_visible_child(soops, walk_element->prev);
+  }
+  else if (walk_element->parent) {
+    /* Use parent if at beginning of list */
+    walk_element = walk_element->parent;
+  }
+
+  return walk_element;
+}
+
+/* Recursively search up the tree until a successor to a given element is found */
+static TreeElement *outliner_element_find_successor_in_parents(TreeElement *te)
+{
+  TreeElement *successor = te;
+  while (successor->parent) {
+    if (successor->parent->next) {
+      te = successor->parent->next;
+      break;
+    }
+    else {
+      successor = successor->parent;
+    }
+  }
+
+  return te;
+}
+
+/* Find next visible element in the tree */
+static TreeElement *outliner_find_next_element(SpaceOutliner *soops, TreeElement *walk_element)
+{
+  TreeStoreElem *tselem = TREESTORE(walk_element);
+
+  if (TSELEM_OPEN(tselem, soops) && walk_element->subtree.first) {
+    walk_element = walk_element->subtree.first;
+  }
+  else if (walk_element->next) {
+    walk_element = walk_element->next;
+  }
+  else {
+    walk_element = outliner_element_find_successor_in_parents(walk_element);
+  }
+
+  return walk_element;
+}
+
+static TreeElement *do_outliner_select_walk(SpaceOutliner *soops,
+                                            TreeElement *walk_element,
+                                            const int direction,
+                                            const bool extend,
+                                            const bool toggle_all)
+{
+  TreeStoreElem *tselem = TREESTORE(walk_element);
+
+  if (!extend) {
+    outliner_flag_set(&soops->tree, TSE_SELECTED, false);
+  }
+  tselem->flag &= ~TSE_ACTIVE_WALK;
+
+  switch (direction) {
+    case OUTLINER_SELECT_WALK_UP:
+      walk_element = outliner_find_previous_element(soops, walk_element);
+      break;
+    case OUTLINER_SELECT_WALK_DOWN:
+      walk_element = outliner_find_next_element(soops, walk_element);
+      break;
+    case OUTLINER_SELECT_WALK_LEFT:
+      outliner_item_openclose(walk_element, false, toggle_all);
+      break;
+    case OUTLINER_SELECT_WALK_RIGHT:
+      outliner_item_openclose(walk_element, true, toggle_all);
+      break;
+  }
+
+  TreeStoreElem *tselem_new = TREESTORE(walk_element);
+
+  /* If new element is already selected, deselect the previous element */
+  if (extend) {
+    tselem->flag = (tselem_new->flag & TSE_SELECTED) ? (tselem->flag & ~TSE_SELECTED) :
+                                                       (tselem->flag | TSE_SELECTED);
+  }
+
+  tselem_new->flag |= TSE_SELECTED | TSE_ACTIVE_WALK;
+
+  return walk_element;
+}
+
+/* Find walk select element, or set it if it does not exist.
+ * Changed is set to true if walk element is found, false if it was set */
+static TreeElement *find_walk_select_start_element(SpaceOutliner *soops, bool *changed)
+{
+  TreeElement *walk_element = outliner_find_element_with_flag(&soops->tree, TSE_ACTIVE_WALK);
+
+  *changed = false;
+
+  /* If no walk element exists, start from active */
+  if (!walk_element) {
+    TreeElement *active_element = outliner_find_element_with_flag(&soops->tree, TSE_ACTIVE);
+
+    /* If no active element exists, use the first element in the tree */
+    if (!active_element) {
+      walk_element = soops->tree.first;
+    }
+    else {
+      walk_element = active_element;
+    }
+
+    *changed = true;
+  }
+
+  /* If walk element is not visible, set that element's first visible parent as walk element */
+  if (!outliner_is_element_visible(walk_element)) {
+    TREESTORE(walk_element)->flag &= ~TSE_ACTIVE_WALK;
+
+    while (!outliner_is_element_visible(walk_element)) {
+      walk_element = walk_element->parent;
+    }
+    *changed = true;
+  }
+
+  return walk_element;
+}
+
+/* Scroll the outliner when the walk element reaches the top or bottom boundary */
+static void outliner_walk_scroll(ARegion *ar, TreeElement *te)
+{
+  /* Account for the header height */
+  int y_max = ar->v2d.cur.ymax - UI_UNIT_Y;
+  int y_min = ar->v2d.cur.ymin;
+
+  /* Scroll if walked position is beyond the border */
+  if (te->ys > y_max) {
+    outliner_scroll_view(ar, te->ys - y_max);
+  }
+  else if (te->ys < y_min) {
+    outliner_scroll_view(ar, -(y_min - te->ys));
+  }
+}
+
+static int outliner_walk_select_invoke(bContext *C, wmOperator *op, const wmEvent *UNUSED(event))
+{
+  SpaceOutliner *soops = CTX_wm_space_outliner(C);
+  ARegion *ar = CTX_wm_region(C);
+
+  const short direction = RNA_enum_get(op->ptr, "direction");
+  const bool extend = RNA_boolean_get(op->ptr, "extend");
+  const bool toggle_all = RNA_boolean_get(op->ptr, "toggle_all");
+
+  bool changed;
+  TreeElement *walk_element = find_walk_select_start_element(soops, &changed);
+
+  /* If finding the starting walk select element did not move the element, proceed to walk */
+  if (!changed) {
+    walk_element = do_outliner_select_walk(soops, walk_element, direction, extend, toggle_all);
+  }
+  else {
+    TREESTORE(walk_element)->flag |= TSE_SELECTED | TSE_ACTIVE_WALK;
+  }
+
+  /* Scroll outliner to focus on walk element */
+  outliner_walk_scroll(ar, walk_element);
+
+  if (soops->flag & SO_SYNC_SELECT) {
+    ED_outliner_select_sync_from_outliner(C, soops);
+  }
+  ED_region_tag_redraw(ar);
+
+  return OPERATOR_FINISHED;
+}
+
+void OUTLINER_OT_select_walk(wmOperatorType *ot)
+{
+  static const EnumPropertyItem direction_items[] = {
+      {OUTLINER_SELECT_WALK_UP, "UP", 0, "Up", ""},
+      {OUTLINER_SELECT_WALK_DOWN, "DOWN", 0, "Down", ""},
+      {OUTLINER_SELECT_WALK_LEFT, "LEFT", 0, "Left", ""},
+      {OUTLINER_SELECT_WALK_RIGHT, "RIGHT", 0, "Right", ""},
+      {0, NULL, 0, NULL, NULL},
+  };
+
+  /* identifiers */
+  ot->name = "Walk Select";
+  ot->idname = "OUTLINER_OT_select_walk";
+  ot->description = "Use walk navigation to select tree elements";
+
+  /* api callbacks */
+  ot->invoke = outliner_walk_select_invoke;
+  ot->poll = ED_operator_outliner_active;
+
+  ot->flag |= OPTYPE_REGISTER | OPTYPE_UNDO;
+
+  /* properties */
+  PropertyRNA *prop;
+  prop = RNA_def_enum(ot->srna,
+                      "direction",
+                      direction_items,
+                      0,
+                      "Walk Direction",
+                      "Select element in this direction");
+  RNA_def_property_flag(prop, PROP_SKIP_SAVE);
+  prop = RNA_def_boolean(ot->srna, "extend", false, "Extend", "Extend selection on walk");
+  RNA_def_property_flag(prop, PROP_SKIP_SAVE);
+  prop = RNA_def_boolean(
+      ot->srna, "toggle_all", false, "Toggle All", "Toggle open/close hierarchy");
+  RNA_def_property_flag(prop, PROP_SKIP_SAVE);
+}
+
+/* ****************************************************** */
diff --git a/source/blender/editors/space_outliner/outliner_sync.c b/source/blender/editors/space_outliner/outliner_sync.c
new file mode 100644
index 00000000000..29c820bce92
--- /dev/null
+++ b/source/blender/editors/space_outliner/outliner_sync.c
@@ -0,0 +1,575 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * The Original Code is Copyright (C) 2004 Blender Foundation.
+ * All rights reserved.
+ */
+
+/** \file
+ * \ingroup spoutliner
+ */
+
+#include <stdio.h>
+
+#include "DNA_armature_types.h"
+#include "DNA_layer_types.h"
+#include "DNA_outliner_types.h"
+#include "DNA_screen_types.h"
+#include "DNA_sequence_types.h"
+#include "DNA_space_types.h"
+
+#include "BLI_compiler_compat.h"
+#include "BLI_ghash.h"
+
+#include "BKE_armature.h"
+#include "BKE_context.h"
+#include "BKE_layer.h"
+#include "BKE_main.h"
+#include "BKE_sequencer.h"
+
+#include "DEG_depsgraph.h"
+
+#include "ED_armature.h"
+#include "ED_object.h"
+#include "ED_outliner.h"
+
+#include "WM_api.h"
+#include "WM_types.h"
+
+#include "outliner_intern.h"
+
+/* Functions for tagging outliner selection syncing is dirty from operators */
+void ED_outliner_select_sync_from_object_tag(bContext *C)
+{
+  wmWindowManager *wm = CTX_wm_manager(C);
+  wm->outliner_sync_select_dirty |= WM_OUTLINER_SYNC_SELECT_FROM_OBJECT;
+}
+
+void ED_outliner_select_sync_from_edit_bone_tag(bContext *C)
+{
+  wmWindowManager *wm = CTX_wm_manager(C);
+  wm->outliner_sync_select_dirty |= WM_OUTLINER_SYNC_SELECT_FROM_EDIT_BONE;
+}
+
+void ED_outliner_select_sync_from_pose_bone_tag(bContext *C)
+{
+  wmWindowManager *wm = CTX_wm_manager(C);
+  wm->outliner_sync_select_dirty |= WM_OUTLINER_SYNC_SELECT_FROM_POSE_BONE;
+}
+
+void ED_outliner_select_sync_from_sequence_tag(bContext *C)
+{
+  wmWindowManager *wm = CTX_wm_manager(C);
+  wm->outliner_sync_select_dirty |= WM_OUTLINER_SYNC_SELECT_FROM_SEQUENCE;
+}
+
+void ED_outliner_select_sync_from_all_tag(bContext *C)
+{
+  wmWindowManager *wm = CTX_wm_manager(C);
+  wm->outliner_sync_select_dirty |= WM_OUTLINER_SYNC_SELECT_FROM_ALL;
+}
+
+bool ED_outliner_select_sync_is_dirty(const bContext *C)
+{
+  wmWindowManager *wm = CTX_wm_manager(C);
+  return wm->outliner_sync_select_dirty & WM_OUTLINER_SYNC_SELECT_FROM_ALL;
+}
+
+/* Copy sync select dirty flag from window manager to all outliners to be synced lazily on draw */
+void ED_outliner_select_sync_flag_outliners(const bContext *C)
+{
+  Main *bmain = CTX_data_main(C);
+  wmWindowManager *wm = CTX_wm_manager(C);
+
+  for (bScreen *screen = bmain->screens.first; screen; screen = screen->id.next) {
+    for (ScrArea *sa = screen->areabase.first; sa; sa = sa->next) {
+      for (SpaceLink *sl = sa->spacedata.first; sl; sl = sl->next) {
+        if (sl->spacetype == SPACE_OUTLINER) {
+          SpaceOutliner *soutliner = (SpaceOutliner *)sl;
+
+          soutliner->sync_select_dirty |= wm->outliner_sync_select_dirty;
+        }
+      }
+    }
+  }
+
+  /* Clear global sync flag */
+  wm->outliner_sync_select_dirty = 0;
+}
+
+/**
+ * Outliner sync select dirty flags are not enough to determine which types to sync,
+ * outliner display mode also needs to be considered. This stores the types of data
+ * to sync to increase code clarity.
+ */
+typedef struct SyncSelectTypes {
+  bool object;
+  bool edit_bone;
+  bool pose_bone;
+  bool sequence;
+} SyncSelectTypes;
+
+/**
+ * Set which types of data to sync when syncing selection from the outliner based on object
+ * interaction mode and outliner display mode
+ */
+static void outliner_sync_select_from_outliner_set_types(bContext *C,
+                                                         SpaceOutliner *soops,
+                                                         SyncSelectTypes *sync_types)
+{
+  Object *obact = CTX_data_active_object(C);
+  Object *obedit = CTX_data_edit_object(C);
+
+  const bool sequence_view = soops->outlinevis == SO_SEQUENCE;
+
+  sync_types->object = !sequence_view;
+  sync_types->edit_bone = !sequence_view && (obedit && obedit->type == OB_ARMATURE);
+  sync_types->pose_bone = !sequence_view && (obact && obact->mode == OB_MODE_POSE);
+  sync_types->sequence = sequence_view;
+}
+
+/**
+ * Current dirty flags and outliner display mode determine which type of syncing should occur.
+ * This is to ensure sync flag data is not lost on sync in the wrong display mode.
+ * Returns true if a sync is needed.
+ */
+static bool outliner_sync_select_to_outliner_set_types(const bContext *C,
+                                                       SpaceOutliner *soops,
+                                                       SyncSelectTypes *sync_types)
+{
+  Object *obact = CTX_data_active_object(C);
+  Object *obedit = CTX_data_edit_object(C);
+
+  const bool sequence_view = soops->outlinevis == SO_SEQUENCE;
+
+  sync_types->object = !sequence_view &&
+                       (soops->sync_select_dirty & WM_OUTLINER_SYNC_SELECT_FROM_OBJECT);
+  sync_types->edit_bone = !sequence_view && (obedit && obedit->type == OB_ARMATURE) &&
+                          (soops->sync_select_dirty & WM_OUTLINER_SYNC_SELECT_FROM_EDIT_BONE);
+  sync_types->pose_bone = !sequence_view && (obact && obact->mode == OB_MODE_POSE) &&
+                          (soops->sync_select_dirty & WM_OUTLINER_SYNC_SELECT_FROM_POSE_BONE);
+  sync_types->sequence = sequence_view &&
+                         (soops->sync_select_dirty & WM_OUTLINER_SYNC_SELECT_FROM_SEQUENCE);
+
+  return sync_types->object || sync_types->edit_bone || sync_types->pose_bone ||
+         sync_types->sequence;
+}
+
+/**
+ * Stores items selected from a sync from the outliner. Prevents syncing the selection
+ * state of the last instance of an object linked in multiple collections.
+ */
+typedef struct SelectedItems {
+  GSet *objects;
+  GSet *edit_bones;
+  GSet *pose_bones;
+} SelectedItems;
+
+static void selected_items_init(SelectedItems *selected_items)
+{
+  selected_items->objects = BLI_gset_new(BLI_ghashutil_ptrhash, BLI_ghashutil_ptrcmp, __func__);
+  selected_items->edit_bones = BLI_gset_new(BLI_ghashutil_ptrhash, BLI_ghashutil_ptrcmp, __func__);
+  selected_items->pose_bones = BLI_gset_new(BLI_ghashutil_ptrhash, BLI_ghashutil_ptrcmp, __func__);
+}
+
+static void selected_items_free(SelectedItems *selected_items)
+{
+  BLI_gset_free(selected_items->objects, NULL);
+  BLI_gset_free(selected_items->edit_bones, NULL);
+  BLI_gset_free(selected_items->pose_bones, NULL);
+}
+
+/* Check if an instance of this object been selected by the sync */
+static bool is_object_selected(GSet *selected_objects, Base *base)
+{
+  return BLI_gset_haskey(selected_objects, base);
+}
+
+/* Check if an instance of this edit bone been selected by the sync */
+static bool is_edit_bone_selected(GSet *selected_ebones, EditBone *ebone)
+{
+  return BLI_gset_haskey(selected_ebones, ebone);
+}
+
+/* Check if an instance of this pose bone been selected by the sync */
+static bool is_pose_bone_selected(GSet *selected_pbones, bPoseChannel *pchan)
+{
+  return BLI_gset_haskey(selected_pbones, pchan);
+}
+
+/* Add element's data to selected item set */
+static void add_selected_item(GSet *selected, void *data)
+{
+  BLI_gset_add(selected, data);
+}
+
+static void outliner_select_sync_to_object(ViewLayer *view_layer,
+                                           TreeElement *te,
+                                           TreeStoreElem *tselem,
+                                           GSet *selected_objects)
+{
+  Object *ob = (Object *)tselem->id;
+  Base *base = (te->directdata) ? (Base *)te->directdata :
+                                  BKE_view_layer_base_find(view_layer, ob);
+
+  if (base && (base->flag & BASE_SELECTABLE)) {
+    if (tselem->flag & TSE_SELECTED) {
+      ED_object_base_select(base, BA_SELECT);
+
+      add_selected_item(selected_objects, base);
+    }
+    else if (!is_object_selected(selected_objects, base)) {
+      ED_object_base_select(base, BA_DESELECT);
+    }
+  }
+}
+
+static void outliner_select_sync_to_edit_bone(ViewLayer *view_layer,
+                                              TreeElement *te,
+                                              TreeStoreElem *tselem,
+                                              GSet *selected_ebones)
+{
+  bArmature *arm = (bArmature *)tselem->id;
+  EditBone *ebone = (EditBone *)te->directdata;
+
+  short bone_flag = ebone->flag;
+
+  if (EBONE_SELECTABLE(arm, ebone)) {
+    if (tselem->flag & TSE_SELECTED) {
+      ebone->flag |= (BONE_SELECTED | BONE_TIPSEL | BONE_ROOTSEL);
+
+      add_selected_item(selected_ebones, ebone);
+    }
+    else if (!is_edit_bone_selected(selected_ebones, ebone)) {
+      ebone->flag &= ~(BONE_SELECTED | BONE_TIPSEL | BONE_ROOTSEL);
+    }
+  }
+
+  /* Tag if selection changed */
+  if (bone_flag != ebone->flag) {
+    Object *obedit = OBEDIT_FROM_VIEW_LAYER(view_layer);
+    DEG_id_tag_update(&arm->id, ID_RECALC_SELECT);
+    WM_main_add_notifier(NC_OBJECT | ND_BONE_SELECT, obedit);
+  }
+}
+
+static void outliner_select_sync_to_pose_bone(TreeElement *te,
+                                              TreeStoreElem *tselem,
+                                              GSet *selected_pbones)
+{
+  Object *ob = (Object *)tselem->id;
+  bArmature *arm = ob->data;
+  bPoseChannel *pchan = (bPoseChannel *)te->directdata;
+
+  short bone_flag = pchan->bone->flag;
+
+  if (PBONE_SELECTABLE(arm, pchan->bone)) {
+    if (tselem->flag & TSE_SELECTED) {
+      pchan->bone->flag |= BONE_SELECTED;
+
+      add_selected_item(selected_pbones, pchan);
+    }
+    else if (!is_pose_bone_selected(selected_pbones, pchan)) {
+      pchan->bone->flag &= ~BONE_SELECTED;
+    }
+  }
+
+  /* Tag if selection changed */
+  if (bone_flag != pchan->bone->flag) {
+    DEG_id_tag_update(&arm->id, ID_RECALC_SELECT);
+    WM_main_add_notifier(NC_OBJECT | ND_BONE_SELECT, ob);
+  }
+}
+
+static void outliner_select_sync_to_sequence(Scene *scene, TreeStoreElem *tselem)
+{
+  Sequence *seq = (Sequence *)tselem->id;
+
+  if (tselem->flag & TSE_ACTIVE) {
+    BKE_sequencer_active_set(scene, seq);
+  }
+
+  if (tselem->flag & TSE_SELECTED) {
+    seq->flag |= SELECT;
+  }
+  else {
+    seq->flag &= ~SELECT;
+  }
+}
+
+/** Sync select and active flags from outliner to active view layer, bones, and sequencer. */
+static void outliner_sync_selection_from_outliner(Scene *scene,
+                                                  ViewLayer *view_layer,
+                                                  ListBase *tree,
+                                                  const SyncSelectTypes *sync_types,
+                                                  SelectedItems *selected_items)
+{
+
+  for (TreeElement *te = tree->first; te; te = te->next) {
+    TreeStoreElem *tselem = TREESTORE(te);
+
+    if (tselem->type == 0 && te->idcode == ID_OB) {
+      if (sync_types->object) {
+        outliner_select_sync_to_object(view_layer, te, tselem, selected_items->objects);
+      }
+    }
+    else if (tselem->type == TSE_EBONE) {
+      if (sync_types->edit_bone) {
+        outliner_select_sync_to_edit_bone(view_layer, te, tselem, selected_items->edit_bones);
+      }
+    }
+    else if (tselem->type == TSE_POSE_CHANNEL) {
+      if (sync_types->pose_bone) {
+        outliner_select_sync_to_pose_bone(te, tselem, selected_items->pose_bones);
+      }
+    }
+    else if (tselem->type == TSE_SEQUENCE) {
+      if (sync_types->sequence) {
+        outliner_select_sync_to_sequence(scene, tselem);
+      }
+    }
+
+    outliner_sync_selection_from_outliner(
+        scene, view_layer, &te->subtree, sync_types, selected_items);
+  }
+}
+
+/* Set clean outliner and mark other outliners for syncing */
+void ED_outliner_select_sync_from_outliner(bContext *C, SpaceOutliner *soops)
+{
+  /* Don't sync in certain outliner display modes */
+  if (ELEM(soops->outlinevis, SO_LIBRARIES, SO_DATA_API, SO_ID_ORPHANS)) {
+    return;
+  }
+
+  Scene *scene = CTX_data_scene(C);
+  ViewLayer *view_layer = CTX_data_view_layer(C);
+
+  SyncSelectTypes sync_types;
+  outliner_sync_select_from_outliner_set_types(C, soops, &sync_types);
+
+  /* To store elements that have been selected to prevent linked object sync errors */
+  SelectedItems selected_items;
+
+  selected_items_init(&selected_items);
+
+  outliner_sync_selection_from_outliner(
+      scene, view_layer, &soops->tree, &sync_types, &selected_items);
+
+  selected_items_free(&selected_items);
+
+  /* Tag for updates and clear dirty flag toprevent a sync to the outliner on draw */
+  if (sync_types.object) {
+    soops->sync_select_dirty &= ~WM_OUTLINER_SYNC_SELECT_FROM_OBJECT;
+    DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
+    WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
+  }
+  else if (sync_types.edit_bone) {
+    soops->sync_select_dirty &= ~WM_OUTLINER_SYNC_SELECT_FROM_EDIT_BONE;
+  }
+  else if (sync_types.pose_bone) {
+    soops->sync_select_dirty &= ~WM_OUTLINER_SYNC_SELECT_FROM_POSE_BONE;
+  }
+  if (sync_types.sequence) {
+    soops->sync_select_dirty &= ~WM_OUTLINER_SYNC_SELECT_FROM_SEQUENCE;
+    WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
+  }
+}
+
+static void outliner_select_sync_from_object(ViewLayer *view_layer,
+                                             SpaceOutliner *soops,
+                                             Object *obact,
+                                             TreeElement *te,
+                                             TreeStoreElem *tselem)
+{
+  Object *ob = (Object *)tselem->id;
+  Base *base = (te->directdata) ? (Base *)te->directdata :
+                                  BKE_view_layer_base_find(view_layer, ob);
+  const bool is_selected = (base != NULL) && ((base->flag & BASE_SELECTED) != 0);
+
+  if (base && (ob == obact)) {
+    outliner_element_activate(soops, tselem);
+  }
+  else {
+    tselem->flag &= ~TSE_ACTIVE;
+  }
+
+  if (is_selected) {
+    tselem->flag |= TSE_SELECTED;
+  }
+  else {
+    tselem->flag &= ~TSE_SELECTED;
+  }
+}
+
+static void outliner_select_sync_from_edit_bone(SpaceOutliner *soops,
+                                                EditBone *ebone_active,
+                                                TreeElement *te,
+                                                TreeStoreElem *tselem)
+{
+  EditBone *ebone = (EditBone *)te->directdata;
+
+  if (ebone == ebone_active) {
+    outliner_element_activate(soops, tselem);
+  }
+  else {
+    tselem->flag &= ~TSE_ACTIVE;
+  }
+
+  if (ebone->flag & BONE_SELECTED) {
+    tselem->flag |= TSE_SELECTED;
+  }
+  else {
+    tselem->flag &= ~TSE_SELECTED;
+  }
+}
+
+static void outliner_select_sync_from_pose_bone(SpaceOutliner *soops,
+                                                bPoseChannel *pchan_active,
+                                                TreeElement *te,
+                                                TreeStoreElem *tselem)
+{
+  bPoseChannel *pchan = (bPoseChannel *)te->directdata;
+  Bone *bone = pchan->bone;
+
+  if (pchan == pchan_active) {
+    outliner_element_activate(soops, tselem);
+  }
+  else {
+    tselem->flag &= ~TSE_ACTIVE;
+  }
+
+  if (bone->flag & BONE_SELECTED) {
+    tselem->flag |= TSE_SELECTED;
+  }
+  else {
+    tselem->flag &= ~TSE_SELECTED;
+  }
+}
+
+static void outliner_select_sync_from_sequence(SpaceOutliner *soops,
+                                               Sequence *sequence_active,
+                                               TreeStoreElem *tselem)
+{
+  Sequence *seq = (Sequence *)tselem->id;
+
+  if (seq == sequence_active) {
+    outliner_element_activate(soops, tselem);
+  }
+  else {
+    tselem->flag &= ~TSE_ACTIVE;
+  }
+
+  if (seq->flag & SELECT) {
+    tselem->flag |= TSE_SELECTED;
+  }
+  else {
+    tselem->flag &= ~TSE_SELECTED;
+  }
+}
+
+/**
+ * Contains active object, bones, and sequence for syncing to prevent getting active data
+ * repeatedly throughout syncing to the outliner.
+ */
+typedef struct SyncSelectActiveData {
+  Object *object;
+  EditBone *edit_bone;
+  bPoseChannel *pose_channel;
+  Sequence *sequence;
+} SyncSelectActiveData;
+
+/** Sync select and active flags from active view layer, bones, and sequences to the outliner. */
+static void outliner_sync_selection_to_outliner(ViewLayer *view_layer,
+                                                SpaceOutliner *soops,
+                                                ListBase *tree,
+                                                SyncSelectActiveData *active_data,
+                                                const SyncSelectTypes *sync_types)
+{
+  for (TreeElement *te = tree->first; te; te = te->next) {
+    TreeStoreElem *tselem = TREESTORE(te);
+
+    if (tselem->type == 0 && te->idcode == ID_OB) {
+      if (sync_types->object) {
+        outliner_select_sync_from_object(view_layer, soops, active_data->object, te, tselem);
+      }
+    }
+    else if (tselem->type == TSE_EBONE) {
+      if (sync_types->edit_bone) {
+        outliner_select_sync_from_edit_bone(soops, active_data->edit_bone, te, tselem);
+      }
+    }
+    else if (tselem->type == TSE_POSE_CHANNEL) {
+      if (sync_types->pose_bone) {
+        outliner_select_sync_from_pose_bone(soops, active_data->pose_channel, te, tselem);
+      }
+    }
+    else if (tselem->type == TSE_SEQUENCE) {
+      if (sync_types->sequence) {
+        outliner_select_sync_from_sequence(soops, active_data->sequence, tselem);
+      }
+    }
+    else {
+      tselem->flag &= ~TSE_SELECTED;
+    }
+
+    /* Sync subtree elements */
+    outliner_sync_selection_to_outliner(view_layer, soops, &te->subtree, active_data, sync_types);
+  }
+}
+
+/* Get active data from context */
+static void get_sync_select_active_data(const bContext *C, SyncSelectActiveData *active_data)
+{
+  Scene *scene = CTX_data_scene(C);
+  ViewLayer *view_layer = CTX_data_view_layer(C);
+  active_data->object = OBACT(view_layer);
+  active_data->edit_bone = CTX_data_active_bone(C);
+  active_data->pose_channel = CTX_data_active_pose_bone(C);
+  active_data->sequence = BKE_sequencer_active_get(scene);
+}
+
+/* If outliner is dirty sync selection from view layer and sequwncer */
+void outliner_sync_selection(const bContext *C, SpaceOutliner *soops)
+{
+  /* Set which types of data to sync from sync dirty flag and outliner display mode */
+  SyncSelectTypes sync_types;
+  const bool sync_required = outliner_sync_select_to_outliner_set_types(C, soops, &sync_types);
+
+  if (sync_required) {
+    ViewLayer *view_layer = CTX_data_view_layer(C);
+
+    /* Store active object, bones, and sequence */
+    SyncSelectActiveData active_data;
+    get_sync_select_active_data(C, &active_data);
+
+    outliner_sync_selection_to_outliner(
+        view_layer, soops, &soops->tree, &active_data, &sync_types);
+
+    /* Keep any unsynced data in the dirty flag */
+    if (sync_types.object) {
+      soops->sync_select_dirty &= ~WM_OUTLINER_SYNC_SELECT_FROM_OBJECT;
+    }
+    if (sync_types.edit_bone) {
+      soops->sync_select_dirty &= ~WM_OUTLINER_SYNC_SELECT_FROM_EDIT_BONE;
+    }
+    if (sync_types.pose_bone) {
+      soops->sync_select_dirty &= ~WM_OUTLINER_SYNC_SELECT_FROM_POSE_BONE;
+    }
+    if (sync_types.sequence) {
+      soops->sync_select_dirty &= ~WM_OUTLINER_SYNC_SELECT_FROM_SEQUENCE;
+    }
+  }
+}
diff --git a/source/blender/editors/space_outliner/outliner_tools.c b/source/blender/editors/space_outliner/outliner_tools.c
index f9905cc4fcd..a2d988f1142 100644
--- a/source/blender/editors/space_outliner/outliner_tools.c
+++ b/source/blender/editors/space_outliner/outliner_tools.c
@@ -63,6 +63,7 @@
 
 #include "ED_armature.h"
 #include "ED_object.h"
+#include "ED_outliner.h"
 #include "ED_scene.h"
 #include "ED_screen.h"
 #include "ED_sequencer.h"
@@ -298,7 +299,7 @@ static void unlink_collection_cb(bContext *C,
     }
     else if (GS(tsep->id->name) == ID_SCE) {
       Scene *scene = (Scene *)tsep->id;
-      Collection *parent = BKE_collection_master(scene);
+      Collection *parent = scene->master_collection;
       id_fake_user_set(&collection->id);
       BKE_collection_child_remove(bmain, parent, collection);
       DEG_id_tag_update(&scene->id, ID_RECALC_COPY_ON_WRITE);
@@ -337,7 +338,7 @@ static void unlink_object_cb(bContext *C,
       }
       else if (GS(tsep->id->name) == ID_SCE) {
         Scene *scene = (Scene *)tsep->id;
-        Collection *parent = BKE_collection_master(scene);
+        Collection *parent = scene->master_collection;
         BKE_collection_object_remove(bmain, parent, ob, true);
         DEG_id_tag_update(&scene->id, ID_RECALC_COPY_ON_WRITE);
         DEG_relations_tag_update(bmain);
@@ -478,6 +479,129 @@ void OUTLINER_OT_scene_operation(wmOperatorType *ot)
 }
 /* ******************************************** */
 
+/**
+ * Stores the parent and a child element of a merged icon-row icon for
+ * the merged select popup menu. The sub-tree of the parent is searched and
+ * the child is needed to only show elements of the same type in the popup.
+ */
+typedef struct MergedSearchData {
+  TreeElement *parent_element;
+  TreeElement *select_element;
+} MergedSearchData;
+
+static void merged_element_search_cb_recursive(
+    const ListBase *tree, short tselem_type, short type, const char *str, uiSearchItems *items)
+{
+  char name[64];
+  int iconid;
+
+  for (TreeElement *te = tree->first; te; te = te->next) {
+    TreeStoreElem *tselem = TREESTORE(te);
+
+    if (tree_element_id_type_to_index(te) == type && tselem_type == tselem->type) {
+      if (BLI_strcasestr(te->name, str)) {
+        BLI_strncpy(name, te->name, 64);
+
+        iconid = tree_element_get_icon(tselem, te).icon;
+
+        /* Don't allow duplicate named items */
+        if (UI_search_items_find_index(items, name) == -1) {
+          if (!UI_search_item_add(items, name, te, iconid)) {
+            break;
+          }
+        }
+      }
+    }
+
+    merged_element_search_cb_recursive(&te->subtree, tselem_type, type, str, items);
+  }
+}
+
+/* Get a list of elements that match the search string */
+static void merged_element_search_cb(const bContext *UNUSED(C),
+                                     void *data,
+                                     const char *str,
+                                     uiSearchItems *items)
+{
+  MergedSearchData *search_data = (MergedSearchData *)data;
+  TreeElement *parent = search_data->parent_element;
+  TreeElement *te = search_data->select_element;
+
+  int type = tree_element_id_type_to_index(te);
+
+  merged_element_search_cb_recursive(&parent->subtree, TREESTORE(te)->type, type, str, items);
+}
+
+/* Activate an element from the merged element search menu */
+static void merged_element_search_call_cb(struct bContext *C, void *UNUSED(arg1), void *element)
+{
+  SpaceOutliner *soops = CTX_wm_space_outliner(C);
+  TreeElement *te = (TreeElement *)element;
+
+  outliner_item_select(soops, te, false, false);
+  outliner_item_do_activate_from_tree_element(C, te, te->store_elem, false, false);
+
+  if (soops->flag & SO_SYNC_SELECT) {
+    ED_outliner_select_sync_from_outliner(C, soops);
+  }
+}
+
+/** Merged element search menu
+ * Created on activation of a merged or aggregated icon-row icon.
+ */
+static uiBlock *merged_element_search_menu(bContext *C, ARegion *ar, void *data)
+{
+  static char search[64] = "";
+  uiBlock *block;
+  uiBut *but;
+
+  /* Clear search on each menu creation */
+  *search = '\0';
+
+  block = UI_block_begin(C, ar, __func__, UI_EMBOSS);
+  UI_block_flag_enable(block, UI_BLOCK_LOOP | UI_BLOCK_MOVEMOUSE_QUIT | UI_BLOCK_SEARCH_MENU);
+  UI_block_theme_style_set(block, UI_BLOCK_THEME_STYLE_POPUP);
+
+  short menu_width = 10 * UI_UNIT_X;
+  but = uiDefSearchBut(
+      block, search, 0, ICON_VIEWZOOM, sizeof(search), 10, 10, menu_width, UI_UNIT_Y, 0, 0, "");
+  UI_but_func_search_set(
+      but, NULL, merged_element_search_cb, data, false, merged_element_search_call_cb, NULL);
+  UI_but_flag_enable(but, UI_BUT_ACTIVATE_ON_INIT);
+
+  /* Fake button to hold space for search items */
+  uiDefBut(block,
+           UI_BTYPE_LABEL,
+           0,
+           "",
+           10,
+           10 - UI_searchbox_size_y(),
+           menu_width,
+           UI_searchbox_size_y(),
+           NULL,
+           0,
+           0,
+           0,
+           0,
+           NULL);
+
+  /* Center the menu on the cursor */
+  UI_block_bounds_set_popup(block, 6, (const int[2]){-(menu_width / 2), 0});
+
+  return block;
+}
+
+void merged_element_search_menu_invoke(bContext *C,
+                                       TreeElement *parent_te,
+                                       TreeElement *activate_te)
+{
+  MergedSearchData *select_data = MEM_callocN(sizeof(MergedSearchData), "merge_search_data");
+  select_data->parent_element = parent_te;
+  select_data->select_element = activate_te;
+
+  UI_popup_block_invoke(C, merged_element_search_menu, select_data, MEM_freeN);
+}
+
 static void object_select_cb(bContext *C,
                              ReportList *UNUSED(reports),
                              Scene *UNUSED(scene),
@@ -556,12 +680,7 @@ static void object_delete_cb(bContext *C,
     if (ob == CTX_data_edit_object(C)) {
       ED_object_editmode_exit(C, EM_FREEDATA);
     }
-    ED_object_base_free_and_unlink(CTX_data_main(C), scene, ob);
-    /* leave for ED_outliner_id_unref to handle */
-#if 0
-    te->directdata = NULL;
-    tselem->id = NULL;
-#endif
+    BKE_id_delete(bmain, ob);
   }
 }
 
@@ -654,7 +773,7 @@ static void singleuser_action_cb(bContext *C,
 
   if (id) {
     IdAdtTemplate *iat = (IdAdtTemplate *)tsep->id;
-    PointerRNA ptr = {{NULL}};
+    PointerRNA ptr = {NULL};
     PropertyRNA *prop;
 
     RNA_pointer_create(&iat->id, &RNA_AnimData, iat->adt, &ptr);
@@ -677,7 +796,7 @@ static void singleuser_world_cb(bContext *C,
   /* need to use parent scene not just scene, otherwise may end up getting wrong one */
   if (id) {
     Scene *parscene = (Scene *)tsep->id;
-    PointerRNA ptr = {{NULL}};
+    PointerRNA ptr = {NULL};
     PropertyRNA *prop;
 
     RNA_id_pointer_create(&parscene->id, &ptr);
@@ -1080,11 +1199,6 @@ static void object_delete_hierarchy_cb(bContext *C,
     }
 
     outline_delete_hierarchy(C, reports, scene, base);
-    /* leave for ED_outliner_id_unref to handle */
-#if 0
-    te->directdata = NULL;
-    tselem->id = NULL;
-#endif
   }
 
   DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
@@ -1168,11 +1282,6 @@ static void object_batch_delete_hierarchy_cb(bContext *C,
     }
 
     outline_batch_delete_hierarchy(reports, CTX_data_main(C), view_layer, scene, base);
-    /* leave for ED_outliner_id_unref to handle */
-#if 0
-    te->directdata = NULL;
-    tselem->id = NULL;
-#endif
   }
 }
 
diff --git a/source/blender/editors/space_outliner/outliner_tree.c b/source/blender/editors/space_outliner/outliner_tree.c
index cc062467dbe..fd6a052b84d 100644
--- a/source/blender/editors/space_outliner/outliner_tree.c
+++ b/source/blender/editors/space_outliner/outliner_tree.c
@@ -297,7 +297,7 @@ static void outliner_add_scene_contents(SpaceOutliner *soops,
 
   ViewLayer *view_layer;
   for (view_layer = sce->view_layers.first; view_layer; view_layer = view_layer->next) {
-    TreeElement *tenlay = outliner_add_element(soops, &ten->subtree, sce, te, TSE_R_LAYER, 0);
+    TreeElement *tenlay = outliner_add_element(soops, &ten->subtree, sce, ten, TSE_R_LAYER, 0);
     tenlay->name = view_layer->name;
     tenlay->directdata = view_layer;
   }
@@ -314,7 +314,7 @@ static void outliner_add_scene_contents(SpaceOutliner *soops,
   ten = outliner_add_element(soops, lb, sce, te, TSE_SCENE_OBJECTS_BASE, 0);
   ten->name = IFACE_("Objects");
   FOREACH_SCENE_OBJECT_BEGIN (sce, ob) {
-    outliner_add_element(soops, &ten->subtree, ob, NULL, 0, 0);
+    outliner_add_element(soops, &ten->subtree, ob, ten, 0, 0);
   }
   FOREACH_SCENE_OBJECT_END;
   outliner_make_object_parent_hierarchy(&ten->subtree);
@@ -766,7 +766,7 @@ static TreeElement *outliner_add_element(
   ID *id = idv;
 
   if (ELEM(type, TSE_RNA_STRUCT, TSE_RNA_PROPERTY, TSE_RNA_ARRAY_ELEM)) {
-    id = ((PointerRNA *)idv)->id.data;
+    id = ((PointerRNA *)idv)->owner_id;
     if (!id) {
       id = ((PointerRNA *)idv)->data;
     }
@@ -1601,7 +1601,7 @@ typedef struct tTreeSort {
   short idcode;
 } tTreeSort;
 
-/* alphabetical comparator, tryping to put objects first */
+/* alphabetical comparator, trying to put objects first */
 static int treesort_alpha_ob(const void *v1, const void *v2)
 {
   const tTreeSort *x1 = v1, *x2 = v2;
@@ -1627,7 +1627,7 @@ static int treesort_alpha_ob(const void *v1, const void *v2)
       return (x1->te->flag & TE_CHILD_NOT_IN_COLLECTION) ? 1 : -1;
     }
 
-    comp = strcmp(x1->name, x2->name);
+    comp = BLI_strcasecmp_natural(x1->name, x2->name);
 
     if (comp > 0) {
       return 1;
@@ -1659,7 +1659,7 @@ static int treesort_alpha(const void *v1, const void *v2)
   const tTreeSort *x1 = v1, *x2 = v2;
   int comp;
 
-  comp = strcmp(x1->name, x2->name);
+  comp = BLI_strcasecmp_natural(x1->name, x2->name);
 
   if (comp > 0) {
     return 1;
@@ -1697,7 +1697,7 @@ static int treesort_obtype_alpha(const void *v1, const void *v2)
       }
     }
     else {
-      int comp = strcmp(x1->name, x2->name);
+      int comp = BLI_strcasecmp_natural(x1->name, x2->name);
 
       if (comp > 0) {
         return 1;
@@ -2008,6 +2008,9 @@ static int outliner_exclude_filter_get(SpaceOutliner *soops)
     case SO_FILTER_OB_VISIBLE:
       exclude_filter |= SO_FILTER_OB_STATE_VISIBLE;
       break;
+    case SO_FILTER_OB_HIDDEN:
+      exclude_filter |= SO_FILTER_OB_STATE_HIDDEN;
+      break;
     case SO_FILTER_OB_SELECTED:
       exclude_filter |= SO_FILTER_OB_STATE_SELECTED;
       break;
@@ -2086,6 +2089,11 @@ static bool outliner_element_visible_get(ViewLayer *view_layer,
           return false;
         }
       }
+      else if (exclude_filter & SO_FILTER_OB_STATE_HIDDEN) {
+        if ((base->flag & BASE_VISIBLE) != 0) {
+          return false;
+        }
+      }
       else if (exclude_filter & SO_FILTER_OB_STATE_SELECTED) {
         if ((base->flag & BASE_SELECTED) == 0) {
           return false;
@@ -2339,7 +2347,8 @@ void outliner_build_tree(
       te = outliner_add_element(soops, &soops->tree, sce, NULL, 0, 0);
       tselem = TREESTORE(te);
 
-      if (sce == scene && show_opened) {
+      /* New scene elements open by default */
+      if ((sce == scene && show_opened) || !tselem->used) {
         tselem->flag &= ~TSE_CLOSED;
       }
 
diff --git a/source/blender/editors/space_outliner/outliner_utils.c b/source/blender/editors/space_outliner/outliner_utils.c
index f57dce97b38..5dfdf6f129b 100644
--- a/source/blender/editors/space_outliner/outliner_utils.c
+++ b/source/blender/editors/space_outliner/outliner_utils.c
@@ -24,11 +24,15 @@
 #include "BLI_utildefines.h"
 
 #include "DNA_action_types.h"
+#include "DNA_screen_types.h"
 #include "DNA_space_types.h"
 
+#include "BKE_context.h"
 #include "BKE_outliner_treehash.h"
+#include "BKE_layer.h"
 
 #include "ED_armature.h"
+#include "ED_outliner.h"
 
 #include "UI_interface.h"
 #include "UI_view2d.h"
@@ -62,6 +66,38 @@ TreeElement *outliner_find_item_at_y(const SpaceOutliner *soops,
   return NULL;
 }
 
+static TreeElement *outliner_find_item_at_x_in_row_recursive(const TreeElement *parent_te,
+                                                             float view_co_x,
+                                                             bool *r_merged)
+{
+  TreeElement *child_te = parent_te->subtree.first;
+
+  bool over_element = false;
+
+  while (child_te) {
+    over_element = (view_co_x > child_te->xs) && (view_co_x < child_te->xend);
+    if ((child_te->flag & TE_ICONROW) && over_element) {
+      return child_te;
+    }
+    else if ((child_te->flag & TE_ICONROW_MERGED) && over_element) {
+      if (r_merged) {
+        *r_merged = true;
+      }
+      return child_te;
+    }
+
+    TreeElement *te = outliner_find_item_at_x_in_row_recursive(child_te, view_co_x, r_merged);
+    if (te != child_te) {
+      return te;
+    }
+
+    child_te = child_te->next;
+  }
+
+  /* return parent if no child is hovered */
+  return (TreeElement *)parent_te;
+}
+
 /**
  * Collapsed items can show their children as click-able icons. This function tries to find
  * such an icon that represents the child item at x-coordinate \a view_co_x (view-space).
@@ -70,24 +106,14 @@ TreeElement *outliner_find_item_at_y(const SpaceOutliner *soops,
  */
 TreeElement *outliner_find_item_at_x_in_row(const SpaceOutliner *soops,
                                             const TreeElement *parent_te,
-                                            float view_co_x)
+                                            float view_co_x,
+                                            bool *r_merged)
 {
-  /* if parent_te is opened, it doesn't show childs in row */
+  /* if parent_te is opened, it doesn't show children in row */
   if (!TSELEM_OPEN(TREESTORE(parent_te), soops)) {
-    /* no recursion, items can only display their direct children in the row */
-    for (TreeElement *child_te = parent_te->subtree.first;
-         /* don't look further if co_x is smaller than child position*/
-         child_te && view_co_x >= child_te->xs;
-
-         child_te = child_te->next) {
-      if ((child_te->flag & TE_ICONROW) && (view_co_x > child_te->xs) &&
-          (view_co_x < child_te->xend)) {
-        return child_te;
-      }
-    }
+    return outliner_find_item_at_x_in_row_recursive(parent_te, view_co_x, r_merged);
   }
 
-  /* return parent if no child is hovered */
   return (TreeElement *)parent_te;
 }
 
@@ -300,3 +326,89 @@ float outliner_restrict_columns_width(const SpaceOutliner *soops)
   }
   return (num_columns * UI_UNIT_X + V2D_SCROLL_WIDTH);
 }
+
+/* Find first tree element in tree with matching treestore flag */
+TreeElement *outliner_find_element_with_flag(const ListBase *lb, short flag)
+{
+  for (TreeElement *te = lb->first; te; te = te->next) {
+    if ((TREESTORE(te)->flag & flag) == flag) {
+      return te;
+    }
+    TreeElement *active_element = outliner_find_element_with_flag(&te->subtree, flag);
+    if (active_element) {
+      return active_element;
+    }
+  }
+  return NULL;
+}
+
+/* Find if element is visible in the outliner tree */
+bool outliner_is_element_visible(const TreeElement *te)
+{
+  TreeStoreElem *tselem;
+
+  while (te->parent) {
+    tselem = TREESTORE(te->parent);
+
+    if (tselem->flag & TSE_CLOSED) {
+      return false;
+    }
+    else {
+      te = te->parent;
+    }
+  }
+
+  return true;
+}
+
+/* Find if x coordinate is over element disclosure toggle */
+bool outliner_item_is_co_within_close_toggle(TreeElement *te, float view_co_x)
+{
+  return (view_co_x > te->xs) && (view_co_x < te->xs + UI_UNIT_X);
+}
+
+/* Scroll view vertically while keeping within total bounds */
+void outliner_scroll_view(ARegion *ar, int delta_y)
+{
+  int y_min = MIN2(ar->v2d.cur.ymin, ar->v2d.tot.ymin);
+
+  ar->v2d.cur.ymax += delta_y;
+  ar->v2d.cur.ymin += delta_y;
+
+  /* Adjust view if delta placed view outside total area */
+  int offset;
+  if (ar->v2d.cur.ymax > -UI_UNIT_Y) {
+    offset = ar->v2d.cur.ymax;
+    ar->v2d.cur.ymax -= offset;
+    ar->v2d.cur.ymin -= offset;
+  }
+  else if (ar->v2d.cur.ymin < y_min) {
+    offset = y_min - ar->v2d.cur.ymin;
+    ar->v2d.cur.ymax += offset;
+    ar->v2d.cur.ymin += offset;
+  }
+}
+
+/* Get base of object under cursor. Used for eyedropper tool */
+Base *ED_outliner_give_base_under_cursor(bContext *C, const int mval[2])
+{
+  ARegion *ar = CTX_wm_region(C);
+  ViewLayer *view_layer = CTX_data_view_layer(C);
+  SpaceOutliner *soops = CTX_wm_space_outliner(C);
+  TreeElement *te;
+  Base *base = NULL;
+  float view_mval[2];
+
+  UI_view2d_region_to_view(&ar->v2d, mval[0], mval[1], &view_mval[0], &view_mval[1]);
+
+  te = outliner_find_item_at_y(soops, &soops->tree, view_mval[1]);
+  if (te) {
+    TreeStoreElem *tselem = TREESTORE(te);
+    if (tselem->type == 0) {
+      Object *ob = (Object *)tselem->id;
+      base = (te->directdata) ? (Base *)te->directdata : BKE_view_layer_base_find(view_layer, ob);
+    }
+  }
+
+  return base;
+}
diff --git a/source/blender/editors/space_outliner/space_outliner.c b/source/blender/editors/space_outliner/space_outliner.c
index 091efc56c09..79880c68120 100644
--- a/source/blender/editors/space_outliner/space_outliner.c
+++ b/source/blender/editors/space_outliner/space_outliner.c
@@ -131,6 +131,9 @@ static void outliner_main_region_listener(wmWindow *UNUSED(win),
           ED_region_tag_redraw(ar);
           break;
       }
+      if (wmn->action & NA_EDITED) {
+        ED_region_tag_redraw(ar);
+      }
       break;
     case NC_OBJECT:
       switch (wmn->data) {
@@ -145,13 +148,8 @@ static void outliner_main_region_listener(wmWindow *UNUSED(win),
           ED_region_tag_redraw(ar);
           break;
         case ND_CONSTRAINT:
-          switch (wmn->action) {
-            case NA_ADDED:
-            case NA_REMOVED:
-            case NA_RENAME:
-              ED_region_tag_redraw(ar);
-              break;
-          }
+          /* all constraint actions now, for reordering */
+          ED_region_tag_redraw(ar);
           break;
         case ND_MODIFIER:
           /* all modifier actions now */
@@ -304,6 +302,8 @@ static SpaceLink *outliner_new(const ScrArea *UNUSED(area), const Scene *UNUSED(
   soutliner->filter_id_type = ID_GR;
   soutliner->show_restrict_flags = SO_RESTRICT_ENABLE | SO_RESTRICT_HIDE;
   soutliner->outlinevis = SO_VIEW_LAYER;
+  soutliner->sync_select_dirty |= WM_OUTLINER_SYNC_SELECT_FROM_ALL;
+  soutliner->flag |= SO_SYNC_SELECT;
 
   /* header */
   ar = MEM_callocN(sizeof(ARegion), "header for outliner");
@@ -349,6 +349,9 @@ static SpaceLink *outliner_duplicate(SpaceLink *sl)
   soutlinern->treestore = NULL;
   soutlinern->treehash = NULL;
 
+  soutlinern->flag |= (soutliner->flag & SO_SYNC_SELECT);
+  soutlinern->sync_select_dirty = WM_OUTLINER_SYNC_SELECT_FROM_ALL;
+
   return (SpaceLink *)soutlinern;
 }
 
@@ -415,7 +418,7 @@ void ED_spacetype_outliner(void)
   /* regions: main window */
   art = MEM_callocN(sizeof(ARegionType), "spacetype outliner region");
   art->regionid = RGN_TYPE_WINDOW;
-  art->keymapflag = ED_KEYMAP_UI | ED_KEYMAP_VIEW2D | ED_KEYMAP_FRAMES;
+  art->keymapflag = ED_KEYMAP_UI | ED_KEYMAP_VIEW2D;
 
   art->init = outliner_main_region_init;
   art->draw = outliner_main_region_draw;
@@ -428,7 +431,7 @@ void ED_spacetype_outliner(void)
   art = MEM_callocN(sizeof(ARegionType), "spacetype outliner header region");
   art->regionid = RGN_TYPE_HEADER;
   art->prefsizey = HEADERY;
-  art->keymapflag = ED_KEYMAP_UI | ED_KEYMAP_VIEW2D | ED_KEYMAP_FRAMES | ED_KEYMAP_HEADER;
+  art->keymapflag = ED_KEYMAP_UI | ED_KEYMAP_VIEW2D | ED_KEYMAP_HEADER;
 
   art->init = outliner_header_region_init;
   art->draw = outliner_header_region_draw;
diff --git a/source/blender/editors/space_sequencer/sequencer_draw.c b/source/blender/editors/space_sequencer/sequencer_draw.c
index 07350b5269e..b15acb12d00 100644
--- a/source/blender/editors/space_sequencer/sequencer_draw.c
+++ b/source/blender/editors/space_sequencer/sequencer_draw.c
@@ -1045,9 +1045,8 @@ ImBuf *sequencer_ibuf_get(struct Main *bmain,
       bmain, depsgraph, scene, rectx, recty, proxy_size, false, &context);
   context.view_id = BKE_scene_multiview_view_id_get(&scene->r, viewname);
 
-  /* sequencer could start rendering, in this case we need to be sure it wouldn't be canceled
-   * by Esc pressed somewhere in the past
-   */
+  /* Sequencer could start rendering, in this case we need to be sure it wouldn't be canceled
+   * by Escape pressed somewhere in the past. */
   G.is_break = false;
 
   /* Rendering can change OGL context. Save & Restore framebuffer. */
diff --git a/source/blender/editors/space_sequencer/sequencer_edit.c b/source/blender/editors/space_sequencer/sequencer_edit.c
index ed0303564c6..22b73c32bfe 100644
--- a/source/blender/editors/space_sequencer/sequencer_edit.c
+++ b/source/blender/editors/space_sequencer/sequencer_edit.c
@@ -3405,6 +3405,7 @@ void SEQUENCER_OT_copy(wmOperatorType *ot)
   /* identifiers */
   ot->name = "Copy";
   ot->idname = "SEQUENCER_OT_copy";
+  ot->description = "Copy selected strips to clipboard";
 
   /* api callbacks */
   ot->exec = sequencer_copy_exec;
@@ -3470,6 +3471,7 @@ void SEQUENCER_OT_paste(wmOperatorType *ot)
   /* identifiers */
   ot->name = "Paste";
   ot->idname = "SEQUENCER_OT_paste";
+  ot->description = "Paste strips from clipboard";
 
   /* api callbacks */
   ot->exec = sequencer_paste_exec;
diff --git a/source/blender/editors/space_sequencer/sequencer_intern.h b/source/blender/editors/space_sequencer/sequencer_intern.h
index dd04260d80e..851d3b5f3aa 100644
--- a/source/blender/editors/space_sequencer/sequencer_intern.h
+++ b/source/blender/editors/space_sequencer/sequencer_intern.h
@@ -34,7 +34,6 @@ struct ARegionType;
 struct Depsgraph;
 struct Main;
 struct Scene;
-struct ScrArea;
 struct Sequence;
 struct SpaceSeq;
 struct StripElem;
diff --git a/source/blender/editors/space_sequencer/sequencer_select.c b/source/blender/editors/space_sequencer/sequencer_select.c
index 57f86059d9d..affb6d3fd88 100644
--- a/source/blender/editors/space_sequencer/sequencer_select.c
+++ b/source/blender/editors/space_sequencer/sequencer_select.c
@@ -41,6 +41,7 @@
 
 /* for menu/popup icons etc etc*/
 
+#include "ED_outliner.h"
 #include "ED_screen.h"
 #include "ED_sequencer.h"
 #include "ED_select_utils.h"
@@ -49,6 +50,7 @@
 
 /* own include */
 #include "sequencer_intern.h"
+
 static void *find_nearest_marker(int UNUSED(d1), int UNUSED(d2))
 {
   return NULL;
@@ -254,6 +256,8 @@ static int sequencer_de_select_all_exec(bContext *C, wmOperator *op)
     }
   }
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   return OPERATOR_FINISHED;
@@ -293,6 +297,8 @@ static int sequencer_select_inverse_exec(bContext *C, wmOperator *UNUSED(op))
     }
   }
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   return OPERATOR_FINISHED;
@@ -542,6 +548,8 @@ static int sequencer_select_invoke(bContext *C, wmOperator *op, const wmEvent *e
     }
   }
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   /* allowing tweaks */
@@ -668,6 +676,8 @@ static int sequencer_select_more_exec(bContext *C, wmOperator *UNUSED(op))
     return OPERATOR_CANCELLED;
   }
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   return OPERATOR_FINISHED;
@@ -699,6 +709,8 @@ static int sequencer_select_less_exec(bContext *C, wmOperator *UNUSED(op))
     return OPERATOR_CANCELLED;
   }
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   return OPERATOR_FINISHED;
@@ -750,6 +762,8 @@ static int sequencer_select_linked_pick_invoke(bContext *C, wmOperator *op, cons
     selected = select_more_less_seq__internal(scene, 1, 1);
   }
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   return OPERATOR_FINISHED;
@@ -784,6 +798,8 @@ static int sequencer_select_linked_exec(bContext *C, wmOperator *UNUSED(op))
     selected = select_more_less_seq__internal(scene, true, true);
   }
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   return OPERATOR_FINISHED;
@@ -832,6 +848,8 @@ static int sequencer_select_handles_exec(bContext *C, wmOperator *op)
     }
   }
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   return OPERATOR_FINISHED;
@@ -876,6 +894,8 @@ static int sequencer_select_active_side_exec(bContext *C, wmOperator *op)
   select_active_side(
       ed->seqbasep, RNA_enum_get(op->ptr, "side"), seq_act->machine, seq_act->startdisp);
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   return OPERATOR_FINISHED;
@@ -934,6 +954,8 @@ static int sequencer_box_select_exec(bContext *C, wmOperator *op)
     }
   }
 
+  ED_outliner_select_sync_from_sequence_tag(C);
+
   WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
 
   return OPERATOR_FINISHED;
@@ -1311,6 +1333,7 @@ static int sequencer_select_grouped_exec(bContext *C, wmOperator *op)
   }
 
   if (changed) {
+    ED_outliner_select_sync_from_sequence_tag(C);
     WM_event_add_notifier(C, NC_SCENE | ND_SEQUENCER | NA_SELECTED, scene);
     return OPERATOR_FINISHED;
   }
diff --git a/source/blender/editors/space_sequencer/space_sequencer.c b/source/blender/editors/space_sequencer/space_sequencer.c
index 9aa9d14cbc8..e1cf6d00b90 100644
--- a/source/blender/editors/space_sequencer/space_sequencer.c
+++ b/source/blender/editors/space_sequencer/space_sequencer.c
@@ -597,7 +597,7 @@ static void sequencer_main_region_message_subscribe(const struct bContext *UNUSE
         &RNA_SequenceModifier,
         &RNA_SequenceColorBalanceData,
     };
-    wmMsgParams_RNA msg_key_params = {{{0}}};
+    wmMsgParams_RNA msg_key_params = {{0}};
     for (int i = 0; i < ARRAY_SIZE(type_array); i++) {
       msg_key_params.ptr.type = type_array[i];
       WM_msg_subscribe_rna_params(
diff --git a/source/blender/editors/space_text/space_text.c b/source/blender/editors/space_text/space_text.c
index c1a3c79b0d8..9f39313b9ab 100644
--- a/source/blender/editors/space_text/space_text.c
+++ b/source/blender/editors/space_text/space_text.c
@@ -29,6 +29,7 @@
 
 #include "BLI_blenlib.h"
 
+#include "BKE_global.h"
 #include "BKE_context.h"
 #include "BKE_library.h"
 #include "BKE_screen.h"
@@ -356,7 +357,7 @@ static void text_drop_paste(wmDrag *drag, wmDropBox *drop)
   ID *id = WM_drag_ID(drag, 0);
 
   /* copy drag path to properties */
-  text = RNA_path_full_ID_py(id);
+  text = RNA_path_full_ID_py(G_MAIN, id);
   RNA_string_set(drop->ptr, "text", text);
   MEM_freeN(text);
 }
diff --git a/source/blender/editors/space_text/text_ops.c b/source/blender/editors/space_text/text_ops.c
index e1550deb659..caefb5070fb 100644
--- a/source/blender/editors/space_text/text_ops.c
+++ b/source/blender/editors/space_text/text_ops.c
@@ -142,7 +142,7 @@ static char *buf_tabs_to_spaces(const char *in_buf, const int tab_size)
 
 BLI_INLINE int text_pixel_x_to_column(SpaceText *st, const int x)
 {
-  /* add half the char width so mouse cursor selection is inbetween letters */
+  /* Add half the char width so mouse cursor selection is in between letters. */
   return (x + (st->cwidth / 2)) / st->cwidth;
 }
 
diff --git a/source/blender/editors/space_view3d/drawobject.c b/source/blender/editors/space_view3d/drawobject.c
index 38e8d285c77..b412a72cce1 100644
--- a/source/blender/editors/space_view3d/drawobject.c
+++ b/source/blender/editors/space_view3d/drawobject.c
@@ -145,7 +145,7 @@ void ED_draw_object_facemap(Depsgraph *depsgraph,
 
     facemap_data = CustomData_get_layer(&me->pdata, CD_FACEMAP);
 
-    /* use gawain immediate mode fore now */
+    /* Make a batch and free it each time for now. */
     const int looptris_len = poly_to_tri_count(mpoly_len, mloop_len);
     const int vbo_len_capacity = looptris_len * 3;
     int vbo_len_used = 0;
diff --git a/source/blender/editors/space_view3d/space_view3d.c b/source/blender/editors/space_view3d/space_view3d.c
index 7acff998e2b..8c4f2f62602 100644
--- a/source/blender/editors/space_view3d/space_view3d.c
+++ b/source/blender/editors/space_view3d/space_view3d.c
@@ -992,7 +992,7 @@ static void view3d_main_region_message_subscribe(const struct bContext *C,
    * accepting some redundant redraws.
    *
    * For other space types we might try avoid this, keep the 3D view as an exceptional case! */
-  wmMsgParams_RNA msg_key_params = {{{0}}};
+  wmMsgParams_RNA msg_key_params = {{0}};
 
   /* Only subscribe to types. */
   StructRNA *type_array[] = {
@@ -1177,7 +1177,7 @@ static void view3d_header_region_message_subscribe(const struct bContext *UNUSED
                                                    struct ARegion *ar,
                                                    struct wmMsgBus *mbus)
 {
-  wmMsgParams_RNA msg_key_params = {{{0}}};
+  wmMsgParams_RNA msg_key_params = {{0}};
 
   /* Only subscribe to types. */
   StructRNA *type_array[] = {
diff --git a/source/blender/editors/space_view3d/view3d_camera_control.c b/source/blender/editors/space_view3d/view3d_camera_control.c
index 0045094542f..260546738f4 100644
--- a/source/blender/editors/space_view3d/view3d_camera_control.c
+++ b/source/blender/editors/space_view3d/view3d_camera_control.c
@@ -87,8 +87,7 @@ typedef struct View3DCameraControl {
   float ofs_backup[3];
   /* backup the views offset in case the user cancels flying in non camera mode */
 
-  /* backup the views quat in case the user cancels flying in non camera mode.
-   * (quat for view, eul for camera) */
+  /* backup the views quat in case the user cancels flying in non camera mode. */
   float rot_backup[4];
   /* remember if were ortho or not, only used for restoring the view if it was a ortho view */
   char persp_backup;
@@ -209,8 +208,8 @@ void ED_view3d_cameracontrol_update(View3DCameraControl *vctrl,
                                     const bool do_rotate,
                                     const bool do_translate)
 {
-  /* we are in camera view so apply the view ofs and quat to the view matrix and set the camera
-   * to the view */
+  /* We are in camera view so apply the view offset and rotation to the view matrix
+   * and set the camera to the view. */
 
   Scene *scene = vctrl->ctx_scene;
   View3D *v3d = vctrl->ctx_v3d;
diff --git a/source/blender/editors/space_view3d/view3d_draw.c b/source/blender/editors/space_view3d/view3d_draw.c
index 6c534ee1b98..f6afa112f08 100644
--- a/source/blender/editors/space_view3d/view3d_draw.c
+++ b/source/blender/editors/space_view3d/view3d_draw.c
@@ -859,6 +859,51 @@ float ED_view3d_grid_scale(Scene *scene, View3D *v3d, const char **grid_unit)
   return v3d->grid * ED_scene_grid_scale(scene, grid_unit);
 }
 
+#define STEPS_LEN 8
+void ED_view3d_grid_steps(Scene *scene,
+                          View3D *v3d,
+                          RegionView3D *rv3d,
+                          float r_grid_steps[STEPS_LEN])
+{
+  const void *usys;
+  int i, len;
+  bUnit_GetSystem(scene->unit.system, B_UNIT_LENGTH, &usys, &len);
+  float grid_scale = v3d->grid;
+
+  if (usys) {
+    if (rv3d->view == RV3D_VIEW_USER) {
+      /* Skip steps */
+      len = bUnit_GetBaseUnit(usys) + 1;
+    }
+
+    grid_scale /= scene->unit.scale_length;
+
+    for (i = 0; i < len; i++) {
+      r_grid_steps[i] = (float)bUnit_GetScaler(usys, len - 1 - i) * grid_scale;
+    }
+    for (; i < STEPS_LEN; i++) {
+      /* Fill last slots */
+      r_grid_steps[i] = 10.0f * r_grid_steps[i - 1];
+    }
+  }
+  else {
+    if (rv3d->view != RV3D_VIEW_USER) {
+      /* Allow 3 more subdivisions. */
+      grid_scale /= powf(v3d->gridsubdiv, 3);
+    }
+    int subdiv = 1;
+    for (i = 0;; i++) {
+      r_grid_steps[i] = grid_scale * subdiv;
+
+      if (i == STEPS_LEN - 1) {
+        break;
+      }
+      subdiv *= v3d->gridsubdiv;
+    }
+  }
+}
+#undef STEPS_LEN
+
 /* Simulates the grid scale that is actually viewed.
  * The actual code is seen in `object_grid_frag.glsl` (see `grid_res`).
  * Currently the simulation is only done when RV3D_VIEW_IS_AXIS. */
@@ -867,24 +912,35 @@ float ED_view3d_grid_view_scale(Scene *scene,
                                 RegionView3D *rv3d,
                                 const char **grid_unit)
 {
-  float grid_scale = ED_view3d_grid_scale(scene, v3d, grid_unit);
+  float grid_scale;
   if (!rv3d->is_persp && RV3D_VIEW_IS_AXIS(rv3d->view)) {
     /* Decrease the distance between grid snap points depending on zoom. */
-    float grid_subdiv = v3d->gridsubdiv;
-    if (grid_subdiv > 1) {
-      /* Allow 3 more subdivisions (see OBJECT_engine_init). */
-      grid_scale /= powf(grid_subdiv, 3);
-
-      /* `3.0` was a value obtained by trial and error in order to get
-       * a nice snap distance.*/
-      float grid_res = 3.0 * (rv3d->dist / v3d->lens);
-      float lvl = (logf(grid_res / grid_scale) / logf(grid_subdiv));
+    /* `0.38` was a value visually obtained in order to get a snap distance
+     * that matches previous versions Blender.*/
+    float min_dist = 0.38f * (rv3d->dist / v3d->lens);
+    float grid_steps[8];
+    ED_view3d_grid_steps(scene, v3d, rv3d, grid_steps);
+    int i;
+    for (i = 0; i < ARRAY_SIZE(grid_steps); i++) {
+      grid_scale = grid_steps[i];
+      if (grid_scale > min_dist) {
+        break;
+      }
+    }
 
-      CLAMP_MIN(lvl, 0.0f);
+    if (grid_unit) {
+      const void *usys;
+      int len;
+      bUnit_GetSystem(scene->unit.system, B_UNIT_LENGTH, &usys, &len);
 
-      grid_scale *= pow(grid_subdiv, (int)lvl);
+      if (usys) {
+        *grid_unit = bUnit_GetNameDisplay(usys, len - i - 1);
+      }
     }
   }
+  else {
+    grid_scale = ED_view3d_grid_scale(scene, v3d, grid_unit);
+  }
 
   return grid_scale;
 }
@@ -1360,6 +1416,27 @@ static void draw_selected_name(
   BLF_disable(font_id, BLF_SHADOW);
 }
 
+static void draw_grid_unit_name(
+    Scene *scene, RegionView3D *rv3d, View3D *v3d, int xoffset, int *yoffset)
+{
+  if (!rv3d->is_persp && RV3D_VIEW_IS_AXIS(rv3d->view)) {
+    const char *grid_unit = NULL;
+    ED_view3d_grid_view_scale(scene, v3d, rv3d, &grid_unit);
+
+    if (grid_unit) {
+      char numstr[32] = "";
+      UI_FontThemeColor(BLF_default(), TH_TEXT_HI);
+      if (v3d->grid != 1.0f) {
+        BLI_snprintf(numstr, sizeof(numstr), "%s x %.4g", grid_unit, v3d->grid);
+      }
+
+      *yoffset -= U.widget_unit;
+      BLF_draw_default_ascii(
+          xoffset, *yoffset, 0.0f, numstr[0] ? numstr : grid_unit, sizeof(numstr));
+    }
+  }
+}
+
 /**
  * Information drawn on top of the solid plates and composed data
  */
@@ -1421,19 +1498,10 @@ void view3d_draw_region_info(const bContext *C, ARegion *ar)
       draw_selected_name(scene, view_layer, ob, xoffset, &yoffset);
     }
 
-#if 0 /* TODO */
-    if (grid_unit) { /* draw below the viewport name */
-      char numstr[32] = "";
-
-      UI_FontThemeColor(BLF_default(), TH_TEXT_HI);
-      if (v3d->grid != 1.0f) {
-        BLI_snprintf(numstr, sizeof(numstr), "%s x %.4g", grid_unit, v3d->grid);
-      }
-
-      *yoffset -= U.widget_unit;
-      BLF_draw_default_ascii(xoffset, *yoffset, numstr[0] ? numstr : grid_unit, sizeof(numstr));
+    if (v3d->gridflag & (V3D_SHOW_FLOOR | V3D_SHOW_X | V3D_SHOW_Y | V3D_SHOW_Z)) {
+      /* draw below the viewport name */
+      draw_grid_unit_name(scene, rv3d, v3d, xoffset, &yoffset);
     }
-#endif
   }
 
   if ((v3d->overlay.flag & V3D_OVERLAY_HIDE_TEXT) == 0) {
@@ -1492,7 +1560,7 @@ void view3d_main_region_draw(const bContext *C, ARegion *ar)
   GPU_pass_cache_garbage_collect();
 
   /* XXX This is in order to draw UI batches with the DRW
-   * olg context since we now use it for drawing the entire area */
+   * old context since we now use it for drawing the entire area. */
   gpu_batch_presets_reset();
 
   /* No depth test for drawing action zones afterwards. */
diff --git a/source/blender/editors/space_view3d/view3d_draw_legacy.c b/source/blender/editors/space_view3d/view3d_draw_legacy.c
index 02ad481e6ce..d5772e5052a 100644
--- a/source/blender/editors/space_view3d/view3d_draw_legacy.c
+++ b/source/blender/editors/space_view3d/view3d_draw_legacy.c
@@ -162,6 +162,7 @@ static void validate_object_select_id(
   Object *obact_eval = DEG_get_evaluated_object(depsgraph, obact);
 
   BLI_assert(ar->regiontype == RGN_TYPE_WINDOW);
+  UNUSED_VARS_NDEBUG(ar);
 
   if (obact_eval && (obact_eval->mode & (OB_MODE_VERTEX_PAINT | OB_MODE_WEIGHT_PAINT) ||
                      BKE_paint_select_face_test(obact_eval))) {
diff --git a/source/blender/editors/space_view3d/view3d_edit.c b/source/blender/editors/space_view3d/view3d_edit.c
index 9a9fe81a13e..e76ef2b0458 100644
--- a/source/blender/editors/space_view3d/view3d_edit.c
+++ b/source/blender/editors/space_view3d/view3d_edit.c
@@ -404,9 +404,7 @@ static void viewops_data_create(bContext *C,
   if (viewops_flag & VIEWOPS_FLAG_PERSP_ENSURE) {
     if (ED_view3d_persp_ensure(depsgraph, vod->v3d, vod->ar)) {
       /* If we're switching from camera view to the perspective one,
-       * need to tag viewport update, so camera vuew and borders
-       * are properly updated.
-       */
+       * need to tag viewport update, so camera view and borders are properly updated. */
       ED_region_tag_redraw(vod->ar);
     }
   }
@@ -4946,6 +4944,7 @@ void ED_view3d_cursor3d_position_rotation(bContext *C,
                                                        .use_object_edit_cage = false,
                                                    },
                                                    mval_fl,
+                                                   NULL,
                                                    &dist_px,
                                                    ray_co,
                                                    ray_no,
@@ -5052,7 +5051,7 @@ void ED_view3d_cursor3d_update(bContext *C,
 
   {
     struct wmMsgBus *mbus = CTX_wm_message_bus(C);
-    wmMsgParams_RNA msg_key_params = {{{0}}};
+    wmMsgParams_RNA msg_key_params = {{0}};
     RNA_pointer_create(&scene->id, &RNA_View3DCursor, &scene->cursor, &msg_key_params.ptr);
     WM_msg_publish_rna_params(mbus, &msg_key_params);
   }
diff --git a/source/blender/editors/space_view3d/view3d_gizmo_camera.c b/source/blender/editors/space_view3d/view3d_gizmo_camera.c
index 1bb10697936..42931d5abb5 100644
--- a/source/blender/editors/space_view3d/view3d_gizmo_camera.c
+++ b/source/blender/editors/space_view3d/view3d_gizmo_camera.c
@@ -100,7 +100,7 @@ static void WIDGETGROUP_camera_setup(const bContext *C, wmGizmoGroup *gzgroup)
     wmGizmo *gz;
     gz = cagzgroup->dop_dist = WM_gizmo_new_ptr(gzt_arrow, gzgroup, NULL);
     RNA_enum_set(gz->ptr, "draw_style", ED_GIZMO_ARROW_STYLE_CROSS);
-    WM_gizmo_set_flag(gz, WM_GIZMO_DRAW_HOVER, true);
+    WM_gizmo_set_flag(gz, WM_GIZMO_DRAW_HOVER | WM_GIZMO_DRAW_NO_SCALE, true);
 
     UI_GetThemeColor3fv(TH_GIZMO_A, gz->color);
     UI_GetThemeColor3fv(TH_GIZMO_HI, gz->color_hi);
@@ -152,9 +152,11 @@ static void WIDGETGROUP_camera_refresh(const bContext *C, wmGizmoGroup *gzgroup)
     WM_gizmo_set_scale(cagzgroup->dop_dist, ca->drawsize);
     WM_gizmo_set_flag(cagzgroup->dop_dist, WM_GIZMO_HIDDEN, false);
 
-    /* need to set property here for undo. TODO would prefer to do this in _init */
+    /* Need to set property here for undo. TODO would prefer to do this in _init */
+    PointerRNA camera_dof_ptr;
+    RNA_pointer_create(&ca->id, &RNA_CameraDOFSettings, &ca->dof, &camera_dof_ptr);
     WM_gizmo_target_property_def_rna(
-        cagzgroup->dop_dist, "offset", &camera_ptr, "dof.focus_distance", -1);
+        cagzgroup->dop_dist, "offset", &camera_dof_ptr, "focus_distance", -1);
   }
   else {
     WM_gizmo_set_flag(cagzgroup->dop_dist, WM_GIZMO_HIDDEN, true);
diff --git a/source/blender/editors/space_view3d/view3d_gizmo_preselect_type.c b/source/blender/editors/space_view3d/view3d_gizmo_preselect_type.c
index a984e339305..68159e2d684 100644
--- a/source/blender/editors/space_view3d/view3d_gizmo_preselect_type.c
+++ b/source/blender/editors/space_view3d/view3d_gizmo_preselect_type.c
@@ -77,7 +77,20 @@ static void gizmo_preselect_elem_draw(const bContext *UNUSED(C), wmGizmo *gz)
 
 static int gizmo_preselect_elem_test_select(bContext *C, wmGizmo *gz, const int mval[2])
 {
+  wmEvent *event = CTX_wm_window(C)->eventstate;
   MeshElemGizmo3D *gz_ele = (MeshElemGizmo3D *)gz;
+
+  /* Hack: Switch action mode based on key input */
+  const bool is_ctrl_pressed = WM_event_modifier_flag(event) & KM_CTRL;
+  const bool is_shift_pressed = WM_event_modifier_flag(event) & KM_SHIFT;
+  EDBM_preselect_action_set(gz_ele->psel, PRESELECT_ACTION_TRANSFORM);
+  if (is_ctrl_pressed && !is_shift_pressed) {
+    EDBM_preselect_action_set(gz_ele->psel, PRESELECT_ACTION_CREATE);
+  }
+  if (!is_ctrl_pressed && is_shift_pressed) {
+    EDBM_preselect_action_set(gz_ele->psel, PRESELECT_ACTION_DELETE);
+  }
+
   struct {
     Object *ob;
     BMElem *ele;
@@ -87,18 +100,6 @@ static int gizmo_preselect_elem_test_select(bContext *C, wmGizmo *gz, const int
       .dist = ED_view3d_select_dist_px(),
   };
 
-  struct {
-    int base_index;
-    int vert_index;
-    int edge_index;
-    int face_index;
-  } prev = {
-      .base_index = gz_ele->base_index,
-      .vert_index = gz_ele->vert_index,
-      .edge_index = gz_ele->edge_index,
-      .face_index = gz_ele->face_index,
-  };
-
   {
     ViewLayer *view_layer = CTX_data_view_layer(C);
     View3D *v3d = CTX_wm_view3d(C);
@@ -115,32 +116,66 @@ static int gizmo_preselect_elem_test_select(bContext *C, wmGizmo *gz, const int
 
   {
     /* TODO: support faces. */
-    int base_index = -1;
+    int base_index_vert = -1;
+    int base_index_edge = -1;
+    int base_index_face = -1;
     BMVert *eve_test;
     BMEdge *eed_test;
+    BMFace *efa_test;
 
     if (EDBM_unified_findnearest_from_raycast(&vc,
                                               gz_ele->bases,
                                               gz_ele->bases_len,
+                                              false,
                                               true,
-                                              &base_index,
+                                              &base_index_vert,
+                                              &base_index_edge,
+                                              &base_index_face,
                                               &eve_test,
                                               &eed_test,
-                                              NULL)) {
-      Base *base = gz_ele->bases[base_index];
-      best.ob = base->object;
-      if (eve_test) {
-        best.ele = (BMElem *)eve_test;
-      }
-      else if (eed_test) {
-        best.ele = (BMElem *)eed_test;
+                                              &efa_test)) {
+      if (EDBM_preselect_action_get(gz_ele->psel) == PRESELECT_ACTION_DELETE) {
+        /* Delete action */
+        if (efa_test) {
+          best.ele = (BMElem *)efa_test;
+          best.base_index = base_index_face;
+        }
       }
+
       else {
-        BLI_assert(0);
+        /* Transform and create action */
+        if (eed_test) {
+          best.ele = (BMElem *)eed_test;
+          best.base_index = base_index_edge;
+        }
+      }
+
+      /* All actions use same vertex pre-selection. */
+      /* Re-topology should always prioritize edge pre-selection.
+       * Only pre-select a vertex when the cursor is really close to it. */
+      if (eve_test) {
+        BMVert *vert = (BMVert *)eve_test;
+        float vert_p_co[3], vert_co[3];
+        float mval_f[2] = {UNPACK2(vc.mval)};
+        mul_v3_m4v3(vert_co, gz_ele->bases[base_index_vert]->object->obmat, vert->co);
+        ED_view3d_project(vc.ar, vert_co, vert_p_co);
+        float len = len_v2v2(vert_p_co, mval_f);
+        if (len < 35) {
+          best.ele = (BMElem *)eve_test;
+          best.base_index = base_index_vert;
+        }
+        if (!BM_vert_is_boundary(vert) &&
+            EDBM_preselect_action_get(gz_ele->psel) != PRESELECT_ACTION_DELETE) {
+          best.ele = (BMElem *)eve_test;
+          best.base_index = base_index_vert;
+        }
       }
-      best.base_index = base_index;
+
       /* Check above should never fail, if it does it's an internal error. */
       BLI_assert(best.base_index != -1);
+
+      Base *base = gz_ele->bases[best.base_index];
+      best.ob = base->object;
     }
   }
 
@@ -167,32 +202,30 @@ static int gizmo_preselect_elem_test_select(bContext *C, wmGizmo *gz, const int
     }
   }
 
-  if ((prev.base_index == gz_ele->base_index) && (prev.vert_index == gz_ele->vert_index) &&
-      (prev.edge_index == gz_ele->edge_index) && (prev.face_index == gz_ele->face_index)) {
-    /* pass (only recalculate on change) */
-  }
-  else {
-    if (best.ele) {
-      const float(*coords)[3] = NULL;
-      {
-        Object *ob = gz_ele->bases[gz_ele->base_index]->object;
-        Depsgraph *depsgraph = CTX_data_ensure_evaluated_depsgraph(C);
-        Mesh *me_eval = (Mesh *)DEG_get_evaluated_id(depsgraph, ob->data);
-        if (me_eval->runtime.edit_data) {
-          coords = me_eval->runtime.edit_data->vertexCos;
-        }
+  if (best.ele) {
+    const float(*coords)[3] = NULL;
+    {
+      Object *ob = gz_ele->bases[gz_ele->base_index]->object;
+      Depsgraph *depsgraph = CTX_data_ensure_evaluated_depsgraph(C);
+      Mesh *me_eval = (Mesh *)DEG_get_evaluated_id(depsgraph, ob->data);
+      if (me_eval->runtime.edit_data) {
+        coords = me_eval->runtime.edit_data->vertexCos;
       }
-      EDBM_preselect_elem_update_from_single(gz_ele->psel, bm, best.ele, coords);
-    }
-    else {
-      EDBM_preselect_elem_clear(gz_ele->psel);
     }
+    EDBM_preselect_elem_update_from_single(gz_ele->psel, bm, best.ele, coords);
+    EDBM_preselect_elem_update_preview(gz_ele->psel, &vc, bm, best.ele, mval);
+  }
+  else {
+    EDBM_preselect_elem_clear(gz_ele->psel);
+    EDBM_preselect_preview_clear(gz_ele->psel);
+  }
 
-    RNA_int_set(gz->ptr, "object_index", gz_ele->base_index);
-    RNA_int_set(gz->ptr, "vert_index", gz_ele->vert_index);
-    RNA_int_set(gz->ptr, "edge_index", gz_ele->edge_index);
-    RNA_int_set(gz->ptr, "face_index", gz_ele->face_index);
+  RNA_int_set(gz->ptr, "object_index", gz_ele->base_index);
+  RNA_int_set(gz->ptr, "vert_index", gz_ele->vert_index);
+  RNA_int_set(gz->ptr, "edge_index", gz_ele->edge_index);
+  RNA_int_set(gz->ptr, "face_index", gz_ele->face_index);
 
+  if (best.ele) {
     ARegion *ar = CTX_wm_region(C);
     ED_region_tag_redraw(ar);
   }
@@ -471,5 +504,4 @@ void ED_view3d_gizmo_mesh_preselect_get_active(bContext *C,
     }
   }
 }
-
 /** \} */
diff --git a/source/blender/editors/space_view3d/view3d_gizmo_ruler.c b/source/blender/editors/space_view3d/view3d_gizmo_ruler.c
index 97a12c7100e..f4e3dc85447 100644
--- a/source/blender/editors/space_view3d/view3d_gizmo_ruler.c
+++ b/source/blender/editors/space_view3d/view3d_gizmo_ruler.c
@@ -343,6 +343,7 @@ static bool view3d_ruler_item_mousemove(RulerInfo *ruler_info,
                                                       .use_object_edit_cage = true,
                                                   },
                                                   mval_fl,
+                                                  NULL,
                                                   &dist_px,
                                                   co,
                                                   ray_normal)) {
@@ -363,16 +364,31 @@ static bool view3d_ruler_item_mousemove(RulerInfo *ruler_info,
     }
     else if (do_snap) {
       const float mval_fl[2] = {UNPACK2(mval)};
+      float *prev_point = NULL;
+
+      if (inter->co_index != 1) {
+        if (ruler_item->flag & RULERITEM_USE_ANGLE) {
+          prev_point = ruler_item->co[1];
+        }
+        else if (inter->co_index == 0) {
+          prev_point = ruler_item->co[2];
+        }
+        else {
+          prev_point = ruler_item->co[0];
+        }
+      }
 
       if (ED_transform_snap_object_project_view3d(
               ruler_info->snap_context,
-              (SCE_SNAP_MODE_VERTEX | SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_FACE),
+              (SCE_SNAP_MODE_VERTEX | SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_FACE |
+               SCE_SNAP_MODE_EDGE_MIDPOINT | SCE_SNAP_MODE_EDGE_PERPENDICULAR),
               &(const struct SnapObjectParams){
                   .snap_select = SNAP_ALL,
                   .use_object_edit_cage = true,
                   .use_occlusion_test = true,
               },
               mval_fl,
+              prev_point,
               &dist_px,
               co,
               NULL)) {
diff --git a/source/blender/editors/space_view3d/view3d_intern.h b/source/blender/editors/space_view3d/view3d_intern.h
index 3a6825aacb4..6b5c27b68f4 100644
--- a/source/blender/editors/space_view3d/view3d_intern.h
+++ b/source/blender/editors/space_view3d/view3d_intern.h
@@ -30,18 +30,11 @@
 
 struct ARegion;
 struct ARegionType;
-struct Base;
 struct BoundBox;
 struct Depsgraph;
-struct GPUBatch;
-struct Mesh;
 struct Object;
-struct SmokeDomainSettings;
 struct ViewLayer;
-struct bAnimVizSettings;
 struct bContext;
-struct bMotionPath;
-struct bPoseChannel;
 struct wmGizmoGroupType;
 struct wmGizmoType;
 struct wmKeyConfig;
diff --git a/source/blender/editors/space_view3d/view3d_select.c b/source/blender/editors/space_view3d/view3d_select.c
index 725f655c8ce..1c9ce142165 100644
--- a/source/blender/editors/space_view3d/view3d_select.c
+++ b/source/blender/editors/space_view3d/view3d_select.c
@@ -88,6 +88,7 @@
 #include "ED_particle.h"
 #include "ED_mesh.h"
 #include "ED_object.h"
+#include "ED_outliner.h"
 #include "ED_screen.h"
 #include "ED_select_utils.h"
 #include "ED_sculpt.h"
@@ -1280,9 +1281,15 @@ static bool view3d_lasso_select(
     }
     else if (ob && (ob->mode & OB_MODE_POSE)) {
       changed_multi |= do_lasso_select_pose(vc, mcords, moves, sel_op);
+      if (changed_multi) {
+        ED_outliner_select_sync_from_pose_bone_tag(C);
+      }
     }
     else {
       changed_multi |= do_lasso_select_objects(vc, mcords, moves, sel_op);
+      if (changed_multi) {
+        ED_outliner_select_sync_from_object_tag(C);
+      }
     }
   }
   else { /* Edit Mode */
@@ -1303,12 +1310,15 @@ static bool view3d_lasso_select(
           break;
         case OB_ARMATURE:
           changed = do_lasso_select_armature(vc, mcords, moves, sel_op);
+          if (changed) {
+            ED_outliner_select_sync_from_edit_bone_tag(C);
+          }
           break;
         case OB_MBALL:
           changed = do_lasso_select_meta(vc, mcords, moves, sel_op);
           break;
         default:
-          assert(!"lasso select on incorrect object type");
+          BLI_assert(!"lasso select on incorrect object type");
           break;
       }
 
@@ -1488,6 +1498,9 @@ static int object_select_menu_exec(bContext *C, wmOperator *op)
     Scene *scene = CTX_data_scene(C);
     DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
     WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, scene);
+
+    ED_outliner_select_sync_from_object_tag(C);
+
     return OPERATOR_FINISHED;
   }
   else {
@@ -2350,6 +2363,9 @@ static int view3d_select_exec(bContext *C, wmOperator *op)
       if (!retval && deselect_all) {
         retval = ED_armature_edit_deselect_all_visible_multi(C);
       }
+      if (retval) {
+        ED_outliner_select_sync_from_edit_bone_tag(C);
+      }
     }
     else if (obedit->type == OB_LATTICE) {
       retval = ED_lattice_select_pick(C, location, extend, deselect, toggle);
@@ -2410,6 +2426,15 @@ static int view3d_select_exec(bContext *C, wmOperator *op)
         DEG_id_tag_update(&scene->id, ID_RECALC_SELECT);
       }
     }
+
+    if (retval) {
+      if (obact && obact->mode & OB_MODE_POSE) {
+        ED_outliner_select_sync_from_pose_bone_tag(C);
+      }
+      else {
+        ED_outliner_select_sync_from_object_tag(C);
+      }
+    }
   }
 
   /* Pass-through allows tweaks
@@ -3230,6 +3255,7 @@ static int view3d_box_select_exec(bContext *C, wmOperator *op)
           if (changed) {
             DEG_id_tag_update(&vc.obedit->id, ID_RECALC_SELECT);
             WM_event_add_notifier(C, NC_OBJECT | ND_BONE_SELECT, vc.obedit);
+            ED_outliner_select_sync_from_edit_bone_tag(C);
           }
           break;
         case OB_LATTICE:
@@ -3240,7 +3266,7 @@ static int view3d_box_select_exec(bContext *C, wmOperator *op)
           }
           break;
         default:
-          assert(!"box select on incorrect object type");
+          BLI_assert(!"box select on incorrect object type");
           break;
       }
       changed_multi |= changed;
@@ -3264,9 +3290,15 @@ static int view3d_box_select_exec(bContext *C, wmOperator *op)
     }
     else if (vc.obact && vc.obact->mode & OB_MODE_POSE) {
       changed_multi = do_pose_box_select(C, &vc, &rect, sel_op);
+      if (changed_multi) {
+        ED_outliner_select_sync_from_pose_bone_tag(C);
+      }
     }
     else { /* object mode with none active */
       changed_multi = do_object_box_select(C, &vc, &rect, sel_op);
+      if (changed_multi) {
+        ED_outliner_select_sync_from_object_tag(C);
+      }
     }
   }
 
@@ -3890,7 +3922,8 @@ static bool mball_circle_select(ViewContext *vc,
 
 /** Callbacks for circle selection in Editmode */
 
-static bool obedit_circle_select(ViewContext *vc,
+static bool obedit_circle_select(bContext *C,
+                                 ViewContext *vc,
                                  wmGenericUserData *wm_userdata,
                                  const eSelectOp sel_op,
                                  const int mval[2],
@@ -3911,6 +3944,9 @@ static bool obedit_circle_select(ViewContext *vc,
       break;
     case OB_ARMATURE:
       changed = armature_circle_select(vc, sel_op, mval, rad);
+      if (changed) {
+        ED_outliner_select_sync_from_edit_bone_tag(C);
+      }
       break;
     case OB_MBALL:
       changed = mball_circle_select(vc, sel_op, mval, rad);
@@ -3999,7 +4035,7 @@ static int view3d_circle_select_exec(bContext *C, wmOperator *op)
       obedit = vc.obedit;
 
       if (obedit) {
-        obedit_circle_select(&vc, wm_userdata, sel_op, mval, (float)radius);
+        obedit_circle_select(C, &vc, wm_userdata, sel_op, mval, (float)radius);
       }
       else if (BKE_paint_select_face_test(obact)) {
         paint_facesel_circle_select(&vc, wm_userdata, sel_op, mval, (float)radius);
@@ -4009,6 +4045,7 @@ static int view3d_circle_select_exec(bContext *C, wmOperator *op)
       }
       else if (obact->mode & OB_MODE_POSE) {
         pose_circle_select(&vc, sel_op, mval, (float)radius);
+        ED_outliner_select_sync_from_pose_bone_tag(C);
       }
       else {
         BLI_assert(0);
@@ -4029,6 +4066,8 @@ static int view3d_circle_select_exec(bContext *C, wmOperator *op)
     if (object_circle_select(&vc, sel_op, mval, (float)radius)) {
       DEG_id_tag_update(&vc.scene->id, ID_RECALC_SELECT);
       WM_event_add_notifier(C, NC_SCENE | ND_OB_SELECT, vc.scene);
+
+      ED_outliner_select_sync_from_object_tag(C);
     }
   }
 
diff --git a/source/blender/editors/space_view3d/view3d_utils.c b/source/blender/editors/space_view3d/view3d_utils.c
index 7f930f1d876..1af94e3ade5 100644
--- a/source/blender/editors/space_view3d/view3d_utils.c
+++ b/source/blender/editors/space_view3d/view3d_utils.c
@@ -622,8 +622,8 @@ bool ED_view3d_camera_autokey(
 /**
  * Call after modifying a locked view.
  *
- * \note Not every view edit currently auto-keys (numpad for eg),
- * this is complicated because of smoothview.
+ * \note Not every view edit currently auto-keys (num-pad for eg),
+ * this is complicated because of smooth-view.
  */
 bool ED_view3d_camera_lock_autokey(View3D *v3d,
                                    RegionView3D *rv3d,
diff --git a/source/blender/editors/transform/transform.c b/source/blender/editors/transform/transform.c
index 14995144c5c..c344b6c8649 100644
--- a/source/blender/editors/transform/transform.c
+++ b/source/blender/editors/transform/transform.c
@@ -3520,7 +3520,7 @@ static void initShear_mouseInputMode(TransInfo *t)
     copy_v3_v3(dir, t->orient_matrix[t->orient_axis_ortho]);
   }
   else {
-    cross_v3_v3v3(dir, t->orient_matrix[t->orient_axis_ortho], t->orient_matrix[t->orient_axis]);
+    cross_v3_v3v3(dir, t->orient_matrix[t->orient_axis], t->orient_matrix[t->orient_axis_ortho]);
   }
 
   /* Without this, half the gizmo handles move in the opposite direction. */
@@ -6601,7 +6601,7 @@ static void slide_origdata_interp_data_vert(SlideOrigData *sod,
      * and we do not want to mess up other shape keys */
     BM_loop_interp_from_face(bm, l, f_copy, false, false);
 
-    /* make sure face-attributes are correct (e.g. MTexPoly) */
+    /* make sure face-attributes are correct (e.g. #MLoopUV, #MLoopCol) */
     BM_elem_attrs_copy_ex(sod->bm_origfaces, bm, f_copy, l->f, 0x0, CD_MASK_NORMAL);
 
     /* weight the loop */
diff --git a/source/blender/editors/transform/transform.h b/source/blender/editors/transform/transform.h
index b023199fa1b..45efde24d83 100644
--- a/source/blender/editors/transform/transform.h
+++ b/source/blender/editors/transform/transform.h
@@ -42,7 +42,6 @@ struct EditBone;
 struct NumInput;
 struct Object;
 struct RNG;
-struct RenderEngineType;
 struct ReportList;
 struct Scene;
 struct ScrArea;
@@ -52,7 +51,6 @@ struct TransDataContainer;
 struct TransInfo;
 struct TransSnap;
 struct TransformOrientation;
-struct View3D;
 struct ViewLayer;
 struct bConstraint;
 struct bContext;
@@ -85,7 +83,9 @@ typedef struct TransSnap {
   bool snap_self;
   bool peel;
   bool snap_spatial_grid;
-  short status;
+  char status;
+  /* Snapped Element Type (currently for objects only). */
+  char snapElem;
   /** snapping from this point (in global-space). */
   float snapPoint[3];
   /** to this point (in global-space). */
@@ -375,6 +375,7 @@ typedef struct VertSlideParams {
 typedef struct BoneInitData {
   struct EditBone *bone;
   float tail[3];
+  float rad_head;
   float rad_tail;
   float roll;
   float head[3];
@@ -1164,4 +1165,7 @@ bool checkUseAxisMatrix(TransInfo *t);
        th != tc_end; \
        th++, i++)
 
+void trans_obdata_in_obmode_update_all(struct TransInfo *t);
+void trans_obchild_in_obmode_update_all(struct TransInfo *t);
+
 #endif
diff --git a/source/blender/editors/transform/transform_conversions.c b/source/blender/editors/transform/transform_conversions.c
index ef9d23d1db8..611992c1d3c 100644
--- a/source/blender/editors/transform/transform_conversions.c
+++ b/source/blender/editors/transform/transform_conversions.c
@@ -52,6 +52,7 @@
 #include "BLI_string.h"
 #include "BLI_bitmap.h"
 #include "BLI_rect.h"
+#include "BLI_kdtree.h"
 
 #include "BKE_action.h"
 #include "BKE_animsys.h"
@@ -235,8 +236,9 @@ static void sort_trans_data_selected_first(TransInfo *t)
   }
 }
 
-/* distance calculated from not-selected vertex to nearest selected vertex
- * warning; this is loops inside loop, has minor N^2 issues, but by sorting list it is OK */
+/**
+ * Distance calculated from not-selected vertex to nearest selected vertex.
+ */
 static void set_prop_dist(TransInfo *t, const bool with_dist)
 {
   int a;
@@ -255,54 +257,124 @@ static void set_prop_dist(TransInfo *t, const bool with_dist)
     }
   }
 
+  /* Count number of selected. */
+  int td_table_len = 0;
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *tob = tc->data;
-    for (a = 0; a < tc->data_len; a++, tob++) {
+    TransData *td = tc->data;
+    for (a = 0; a < tc->data_len; a++, td++) {
+      if (td->flag & TD_SELECTED) {
+        td_table_len++;
+      }
+      else {
+        /* By definition transform-data has selected items in beginning. */
+        break;
+      }
+    }
+  }
 
-      tob->rdist = 0.0f;  // init, it was mallocced
+  /* Pointers to selected's #TransData.
+   * Used to find #TransData from the index returned by #BLI_kdtree_find_nearest. */
+  TransData **td_table = MEM_mallocN(sizeof(*td_table) * td_table_len, __func__);
 
-      if ((tob->flag & TD_SELECTED) == 0) {
-        TransData *td;
-        int i;
-        float dist_sq, vec[3];
+  /* Create and fill kd-tree of selected's positions - in global or proj_vec space. */
+  KDTree_3d *td_tree = BLI_kdtree_3d_new(td_table_len);
 
-        tob->rdist = -1.0f;  // signal for next loop
+  int td_table_index = 0;
+  FOREACH_TRANS_DATA_CONTAINER (t, tc) {
+    TransData *td = tc->data;
+    for (a = 0; a < tc->data_len; a++, td++) {
+      if (td->flag & TD_SELECTED) {
+        /* Initialize, it was mallocced. */
+        float vec[3];
+        td->rdist = 0.0f;
+
+        if (use_island) {
+          if (tc->use_local_mat) {
+            mul_v3_m4v3(vec, tc->mat, td->iloc);
+          }
+          else {
+            mul_v3_m3v3(vec, td->mtx, td->iloc);
+          }
+        }
+        else {
+          if (tc->use_local_mat) {
+            mul_v3_m4v3(vec, tc->mat, td->center);
+          }
+          else {
+            mul_v3_m3v3(vec, td->mtx, td->center);
+          }
+        }
 
-        for (i = 0, td = tc->data; i < tc->data_len; i++, td++) {
-          if (td->flag & TD_SELECTED) {
-            if (use_island) {
-              sub_v3_v3v3(vec, tob->iloc, td->iloc);
-            }
-            else {
-              sub_v3_v3v3(vec, tob->center, td->center);
-            }
-            mul_m3_v3(tob->mtx, vec);
+        if (proj_vec) {
+          float vec_p[3];
+          project_v3_v3v3(vec_p, vec, proj_vec);
+          sub_v3_v3(vec, vec_p);
+        }
 
-            if (proj_vec) {
-              float vec_p[3];
-              project_v3_v3v3(vec_p, vec, proj_vec);
-              sub_v3_v3(vec, vec_p);
-            }
+        BLI_kdtree_3d_insert(td_tree, td_table_index, vec);
+        td_table[td_table_index++] = td;
+      }
+      else {
+        /* By definition transform-data has selected items in beginning. */
+        break;
+      }
+    }
+  }
+  BLI_assert(td_table_index == td_table_len);
 
-            dist_sq = len_squared_v3(vec);
-            if ((tob->rdist == -1.0f) || (dist_sq < SQUARE(tob->rdist))) {
-              tob->rdist = sqrtf(dist_sq);
-              if (use_island) {
-                copy_v3_v3(tob->center, td->center);
-                copy_m3_m3(tob->axismtx, td->axismtx);
-              }
-            }
+  BLI_kdtree_3d_balance(td_tree);
+
+  /* For each non-selected vertex, find distance to the nearest selected vertex. */
+  FOREACH_TRANS_DATA_CONTAINER (t, tc) {
+    TransData *td = tc->data;
+    for (a = 0; a < tc->data_len; a++, td++) {
+      if ((td->flag & TD_SELECTED) == 0) {
+        float vec[3];
+
+        if (use_island) {
+          if (tc->use_local_mat) {
+            mul_v3_m4v3(vec, tc->mat, td->iloc);
           }
           else {
-            break; /* by definition transdata has selected items in beginning */
+            mul_v3_m3v3(vec, td->mtx, td->iloc);
           }
         }
+        else {
+          if (tc->use_local_mat) {
+            mul_v3_m4v3(vec, tc->mat, td->center);
+          }
+          else {
+            mul_v3_m3v3(vec, td->mtx, td->center);
+          }
+        }
+
+        if (proj_vec) {
+          float vec_p[3];
+          project_v3_v3v3(vec_p, vec, proj_vec);
+          sub_v3_v3(vec, vec_p);
+        }
+
+        KDTreeNearest_3d nearest;
+        const int td_index = BLI_kdtree_3d_find_nearest(td_tree, vec, &nearest);
+
+        td->rdist = -1.0f;
+        if (td_index != -1) {
+          td->rdist = nearest.dist;
+          if (use_island) {
+            copy_v3_v3(td->center, td_table[td_index]->center);
+            copy_m3_m3(td->axismtx, td_table[td_index]->axismtx);
+          }
+        }
+
         if (with_dist) {
-          tob->dist = tob->rdist;
+          td->dist = td->rdist;
         }
       }
     }
   }
+
+  BLI_kdtree_3d_free(td_tree);
+  MEM_freeN(td_table);
 }
 
 /* ************************** CONVERSIONS ************************* */
@@ -1445,6 +1517,7 @@ void restoreBones(TransDataContainer *tc)
     ebo = bid->bone;
 
     ebo->dist = bid->dist;
+    ebo->rad_head = bid->rad_head;
     ebo->rad_tail = bid->rad_tail;
     ebo->roll = bid->roll;
     ebo->xwidth = bid->xwidth;
@@ -1710,6 +1783,7 @@ static void createTransArmatureVerts(TransInfo *t)
         if (eboflip) {
           bid[i].bone = eboflip;
           bid[i].dist = eboflip->dist;
+          bid[i].rad_head = eboflip->rad_head;
           bid[i].rad_tail = eboflip->rad_tail;
           bid[i].roll = eboflip->roll;
           bid[i].xwidth = eboflip->xwidth;
@@ -6462,11 +6536,14 @@ static void flush_trans_object_base_deps_flag(Depsgraph *depsgraph, Object *obje
       depsgraph, &object->id, DEG_OB_COMP_TRANSFORM, set_trans_object_base_deps_flag_cb, NULL);
 }
 
-static void trans_object_base_deps_flag_finish(ViewLayer *view_layer)
+static void trans_object_base_deps_flag_finish(const TransInfo *t, ViewLayer *view_layer)
 {
-  for (Base *base = view_layer->object_bases.first; base; base = base->next) {
-    if (base->object->id.tag & LIB_TAG_DOIT) {
-      base->flag_legacy |= BA_SNAP_FIX_DEPS_FIASCO;
+
+  if ((t->options & CTX_OBMODE_XFORM_OBDATA) == 0) {
+    for (Base *base = view_layer->object_bases.first; base; base = base->next) {
+      if (base->object->id.tag & LIB_TAG_DOIT) {
+        base->flag_legacy |= BA_SNAP_FIX_DEPS_FIASCO;
+      }
     }
   }
 }
@@ -6495,7 +6572,7 @@ static void set_trans_object_base_flags(TransInfo *t)
   trans_object_base_deps_flag_prepare(view_layer);
   /* Traverse all bases and set all possible flags. */
   for (Base *base = view_layer->object_bases.first; base; base = base->next) {
-    base->flag_legacy &= ~BA_WAS_SEL;
+    base->flag_legacy &= ~(BA_WAS_SEL | BA_TRANSFORM_LOCKED_IN_PLACE);
     if (BASE_SELECTED_EDITABLE(v3d, base)) {
       Object *ob = base->object;
       Object *parsel = ob->parent;
@@ -6528,7 +6605,7 @@ static void set_trans_object_base_flags(TransInfo *t)
   /* Store temporary bits in base indicating that base is being modified
    * (directly or indirectly) by transforming objects.
    */
-  trans_object_base_deps_flag_finish(view_layer);
+  trans_object_base_deps_flag_finish(t, view_layer);
 }
 
 static bool mark_children(Object *ob)
@@ -6596,7 +6673,7 @@ static int count_proportional_objects(TransInfo *t)
   /* Store temporary bits in base indicating that base is being modified
    * (directly or indirectly) by transforming objects.
    */
-  trans_object_base_deps_flag_finish(view_layer);
+  trans_object_base_deps_flag_finish(t, view_layer);
   return total;
 }
 
@@ -6611,7 +6688,8 @@ static void clear_trans_object_base_flags(TransInfo *t)
     }
 
     base->flag_legacy &= ~(BA_WAS_SEL | BA_SNAP_FIX_DEPS_FIASCO | BA_TEMP_TAG |
-                           BA_TRANSFORM_CHILD | BA_TRANSFORM_PARENT);
+                           BA_TRANSFORM_CHILD | BA_TRANSFORM_PARENT |
+                           BA_TRANSFORM_LOCKED_IN_PLACE);
   }
 }
 
@@ -7049,7 +7127,15 @@ static void special_aftertrans_update__mesh(bContext *UNUSED(C), TransInfo *t)
         hflag = BM_ELEM_SELECT;
       }
 
-      EDBM_automerge(t->scene, tc->obedit, true, hflag);
+      if (t->scene->toolsettings->automerge & AUTO_MERGE) {
+        if (t->scene->toolsettings->automerge & AUTO_MERGE_AND_SPLIT) {
+          EDBM_automerge_and_split(
+              tc->obedit, true, true, true, hflag, t->scene->toolsettings->doublimit);
+        }
+        else {
+          EDBM_automerge(tc->obedit, true, hflag, t->scene->toolsettings->doublimit);
+        }
+      }
 
       /* Special case, this is needed or faces won't re-select.
        * Flush selected edges to faces. */
@@ -7433,7 +7519,7 @@ void special_aftertrans_update(bContext *C, TransInfo *t)
   else if (t->flag & T_POSE && (t->mode == TFM_BONESIZE)) {
     /* Handle the exception where for TFM_BONESIZE in edit mode we pretend to be
      * in pose mode (to use bone orientation matrix),
-     * in that case we don't do operations like autokeyframing. */
+     * in that case we don't do operations like auto-keyframing. */
     FOREACH_TRANS_DATA_CONTAINER (t, tc) {
       ob = tc->poseobj;
       DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
@@ -7604,6 +7690,278 @@ int special_transform_moving(TransInfo *t)
   return 0;
 }
 
+/* -------------------------------------------------------------------- */
+/** \name Object Mode Custom Data
+ * \{ */
+
+typedef struct TransDataObject {
+
+  /**
+   * Object to object data transform table.
+   * Don't add these to transform data because we may want to include child objects
+   * which aren't being transformed.
+   * - The key is object data #ID.
+   * - The value is #XFormObjectData_Extra.
+   */
+  struct GHash *obdata_in_obmode_map;
+
+  /**
+   * Transform
+   * - The key is object data #Object.
+   * - The value is #XFormObjectSkipChild.
+   */
+  struct GHash *obchild_in_obmode_map;
+
+} TransDataObject;
+
+/** \} */
+
+static void trans_obdata_in_obmode_free_all(TransDataObject *tdo);
+static void trans_obchild_in_obmode_free_all(TransDataObject *tdo);
+
+static void freeTransObjectCustomData(TransInfo *t,
+                                      TransDataContainer *UNUSED(tc),
+                                      TransCustomData *custom_data)
+{
+  TransDataObject *tdo = custom_data->data;
+  custom_data->data = NULL;
+
+  if (t->options & CTX_OBMODE_XFORM_OBDATA) {
+    trans_obdata_in_obmode_free_all(tdo);
+  }
+
+  if (t->options & CTX_OBMODE_XFORM_SKIP_CHILDREN) {
+    trans_obchild_in_obmode_free_all(tdo);
+  }
+  MEM_freeN(tdo);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Object Data in Object Mode
+ *
+ * Use to implement 'Affect Only Origins' feature.
+ * We need this to be detached from transform data because,
+ * unlike transforming regular objects, we need to transform the children.
+ *
+ * \{ */
+
+struct XFormObjectData_Extra {
+  Object *ob;
+  float obmat_orig[4][4];
+  struct XFormObjectData *xod;
+};
+
+static void trans_obdata_in_obmode_ensure_object(TransDataObject *tdo, Object *ob)
+{
+  if (tdo->obdata_in_obmode_map == NULL) {
+    tdo->obdata_in_obmode_map = BLI_ghash_ptr_new(__func__);
+  }
+
+  void **xf_p;
+  if (!BLI_ghash_ensure_p(tdo->obdata_in_obmode_map, ob->data, &xf_p)) {
+    struct XFormObjectData_Extra *xf = MEM_mallocN(sizeof(*xf), __func__);
+    copy_m4_m4(xf->obmat_orig, ob->obmat);
+    xf->ob = ob;
+    /* Result may be NULL, that's OK. */
+    xf->xod = ED_object_data_xform_create(ob->data);
+    *xf_p = xf;
+  }
+}
+
+void trans_obdata_in_obmode_update_all(TransInfo *t)
+{
+  TransDataObject *tdo = t->custom.type.data;
+  if (tdo->obdata_in_obmode_map == NULL) {
+    return;
+  }
+
+  struct Main *bmain = CTX_data_main(t->context);
+  BKE_scene_graph_evaluated_ensure(t->depsgraph, bmain);
+
+  GHashIterator gh_iter;
+  GHASH_ITER (gh_iter, tdo->obdata_in_obmode_map) {
+    ID *id = BLI_ghashIterator_getKey(&gh_iter);
+    struct XFormObjectData_Extra *xf = BLI_ghashIterator_getValue(&gh_iter);
+    if (xf->xod == NULL) {
+      continue;
+    }
+
+    Object *ob_eval = DEG_get_evaluated_object(t->depsgraph, xf->ob);
+    float imat[4][4], dmat[4][4];
+    invert_m4_m4(imat, xf->obmat_orig);
+    mul_m4_m4m4(dmat, imat, ob_eval->obmat);
+    invert_m4(dmat);
+
+    ED_object_data_xform_by_mat4(xf->xod, dmat);
+    DEG_id_tag_update(id, ID_RECALC_GEOMETRY);
+  }
+}
+
+/** Callback for #GHash free. */
+static void trans_obdata_in_obmode_free_elem(void *xf_p)
+{
+  struct XFormObjectData_Extra *xf = xf_p;
+  if (xf->xod) {
+    ED_object_data_xform_destroy(xf->xod);
+  }
+  MEM_freeN(xf);
+}
+
+static void trans_obdata_in_obmode_free_all(TransDataObject *tdo)
+{
+  if (tdo->obdata_in_obmode_map != NULL) {
+    BLI_ghash_free(tdo->obdata_in_obmode_map, NULL, trans_obdata_in_obmode_free_elem);
+  }
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Object Child Skip
+ *
+ * Don't transform unselected children, this is done using the parent inverse matrix.
+ *
+ * \note The complex logic here is caused by mixed selection within a single selection chain,
+ * otherwise we only need #OB_SKIP_CHILD_PARENT_IS_XFORM for single objects.
+ *
+ * \{ */
+
+enum {
+  /**
+   * The parent is transformed, this is held in place.
+   */
+  OB_SKIP_CHILD_PARENT_IS_XFORM = 1,
+  /**
+   * The same as #OB_SKIP_CHILD_PARENT_IS_XFORM,
+   * however this objects parent isn't transformed directly.
+   */
+  OB_SKIP_CHILD_PARENT_IS_XFORM_INDIRECT = 3,
+  /**
+   * Use the parent invert matrix to apply transformation,
+   * this is needed, because breaks in the selection chain prevents this from being transformed.
+   * This is used to add the transform which would have been added
+   * if there weren't breaks in the parent/child chain.
+   */
+  OB_SKIP_CHILD_PARENT_APPLY_TRANSFORM = 2,
+};
+
+struct XFormObjectSkipChild {
+  float obmat_orig[4][4];
+  float parent_obmat_orig[4][4];
+  float parent_obmat_inv_orig[4][4];
+  float parent_recurse_obmat_orig[4][4];
+  float parentinv_orig[4][4];
+  Object *ob_parent_recurse;
+  int mode;
+};
+
+static void trans_obchild_in_obmode_ensure_object(TransDataObject *tdo,
+                                                  Object *ob,
+                                                  Object *ob_parent_recurse,
+                                                  int mode)
+{
+  if (tdo->obchild_in_obmode_map == NULL) {
+    tdo->obchild_in_obmode_map = BLI_ghash_ptr_new(__func__);
+  }
+
+  void **xf_p;
+  if (!BLI_ghash_ensure_p(tdo->obchild_in_obmode_map, ob, &xf_p)) {
+    struct XFormObjectSkipChild *xf = MEM_mallocN(sizeof(*xf), __func__);
+    copy_m4_m4(xf->parentinv_orig, ob->parentinv);
+    copy_m4_m4(xf->obmat_orig, ob->obmat);
+    copy_m4_m4(xf->parent_obmat_orig, ob->parent->obmat);
+    invert_m4_m4(xf->parent_obmat_inv_orig, ob->parent->obmat);
+    if (ob_parent_recurse) {
+      copy_m4_m4(xf->parent_recurse_obmat_orig, ob_parent_recurse->obmat);
+    }
+    xf->mode = mode;
+    xf->ob_parent_recurse = ob_parent_recurse;
+    *xf_p = xf;
+  }
+}
+
+void trans_obchild_in_obmode_update_all(TransInfo *t)
+{
+  TransDataObject *tdo = t->custom.type.data;
+  if (tdo->obchild_in_obmode_map == NULL) {
+    return;
+  }
+
+  struct Main *bmain = CTX_data_main(t->context);
+  BKE_scene_graph_evaluated_ensure(t->depsgraph, bmain);
+
+  GHashIterator gh_iter;
+  GHASH_ITER (gh_iter, tdo->obchild_in_obmode_map) {
+    Object *ob = BLI_ghashIterator_getKey(&gh_iter);
+    struct XFormObjectSkipChild *xf = BLI_ghashIterator_getValue(&gh_iter);
+
+    /* The following blocks below assign 'dmat'. */
+    float dmat[4][4];
+
+    if (xf->mode == OB_SKIP_CHILD_PARENT_IS_XFORM) {
+      /* Parent is transformed, this isn't so compensate. */
+      Object *ob_parent_eval = DEG_get_evaluated_object(t->depsgraph, ob->parent);
+      mul_m4_m4m4(dmat, xf->parent_obmat_inv_orig, ob_parent_eval->obmat);
+      invert_m4(dmat);
+    }
+    else if (xf->mode == OB_SKIP_CHILD_PARENT_IS_XFORM_INDIRECT) {
+      /* Calculate parent matrix (from the root transform). */
+      Object *ob_parent_recurse_eval = DEG_get_evaluated_object(t->depsgraph,
+                                                                xf->ob_parent_recurse);
+      float parent_recurse_obmat_inv[4][4];
+      invert_m4_m4(parent_recurse_obmat_inv, ob_parent_recurse_eval->obmat);
+      mul_m4_m4m4(dmat, xf->parent_recurse_obmat_orig, parent_recurse_obmat_inv);
+      invert_m4(dmat);
+      float parent_obmat_calc[4][4];
+      mul_m4_m4m4(parent_obmat_calc, dmat, xf->parent_obmat_orig);
+
+      /* Apply to the parent inverse matrix. */
+      mul_m4_m4m4(dmat, xf->parent_obmat_inv_orig, parent_obmat_calc);
+      invert_m4(dmat);
+    }
+    else {
+      BLI_assert(xf->mode == OB_SKIP_CHILD_PARENT_APPLY_TRANSFORM);
+      /* Transform this - without transform data. */
+      Object *ob_parent_recurse_eval = DEG_get_evaluated_object(t->depsgraph,
+                                                                xf->ob_parent_recurse);
+      float parent_recurse_obmat_inv[4][4];
+      invert_m4_m4(parent_recurse_obmat_inv, ob_parent_recurse_eval->obmat);
+      mul_m4_m4m4(dmat, xf->parent_recurse_obmat_orig, parent_recurse_obmat_inv);
+      invert_m4(dmat);
+      float obmat_calc[4][4];
+      mul_m4_m4m4(obmat_calc, dmat, xf->obmat_orig);
+      /* obmat_calc is just obmat. */
+
+      /* Get the matrices relative to the parent. */
+      float obmat_parent_relative_orig[4][4];
+      float obmat_parent_relative_calc[4][4];
+      float obmat_parent_relative_inv_orig[4][4];
+
+      mul_m4_m4m4(obmat_parent_relative_orig, xf->parent_obmat_inv_orig, xf->obmat_orig);
+      mul_m4_m4m4(obmat_parent_relative_calc, xf->parent_obmat_inv_orig, obmat_calc);
+      invert_m4_m4(obmat_parent_relative_inv_orig, obmat_parent_relative_orig);
+
+      /* Apply to the parent inverse matrix. */
+      mul_m4_m4m4(dmat, obmat_parent_relative_calc, obmat_parent_relative_inv_orig);
+    }
+
+    mul_m4_m4m4(ob->parentinv, dmat, xf->parentinv_orig);
+
+    DEG_id_tag_update(&ob->id, ID_RECALC_TRANSFORM);
+  }
+}
+
+static void trans_obchild_in_obmode_free_all(TransDataObject *tdo)
+{
+  if (tdo->obchild_in_obmode_map != NULL) {
+    BLI_ghash_free(tdo->obchild_in_obmode_map, NULL, MEM_freeN);
+  }
+}
+
+/** \} */
+
 static void createTransObject(bContext *C, TransInfo *t)
 {
   TransData *td = NULL;
@@ -7630,6 +7988,10 @@ static void createTransObject(bContext *C, TransInfo *t)
   td = tc->data = MEM_callocN(tc->data_len * sizeof(TransData), "TransOb");
   tx = tc->data_ext = MEM_callocN(tc->data_len * sizeof(TransDataExtension), "TransObExtension");
 
+  TransDataObject *tdo = MEM_callocN(sizeof(*tdo), __func__);
+  t->custom.type.data = tdo;
+  t->custom.type.free_cb = freeTransObjectCustomData;
+
   CTX_DATA_BEGIN (C, Base *, base, selected_bases) {
     Object *ob = base->object;
 
@@ -7648,6 +8010,24 @@ static void createTransObject(bContext *C, TransInfo *t)
       td->flag |= TD_SKIP;
     }
 
+    if (t->options & CTX_OBMODE_XFORM_OBDATA) {
+      ID *id = ob->data;
+      if (!id || id->lib) {
+        td->flag |= TD_SKIP;
+      }
+      else if (BKE_object_is_in_editmode(ob)) {
+        /* The object could have edit-mode data from another view-layer,
+         * it's such a corner-case it can be skipped for now - Campbell. */
+        td->flag |= TD_SKIP;
+      }
+    }
+
+    if (t->options & CTX_OBMODE_XFORM_OBDATA) {
+      if ((td->flag & TD_SKIP) == 0) {
+        trans_obdata_in_obmode_ensure_object(tdo, ob);
+      }
+    }
+
     ObjectToTransData(t, td, ob);
     td->val = NULL;
     td++;
@@ -7679,6 +8059,123 @@ static void createTransObject(bContext *C, TransInfo *t)
       }
     }
   }
+
+  if (t->options & CTX_OBMODE_XFORM_OBDATA) {
+    GSet *objects_in_transdata = BLI_gset_ptr_new_ex(__func__, tc->data_len);
+    td = tc->data;
+    for (int i = 0; i < tc->data_len; i++, td++) {
+      if ((td->flag & TD_SKIP) == 0) {
+        BLI_gset_add(objects_in_transdata, td->ob);
+      }
+    }
+
+    ViewLayer *view_layer = t->view_layer;
+    View3D *v3d = t->view;
+
+    for (Base *base = view_layer->object_bases.first; base; base = base->next) {
+      Object *ob = base->object;
+
+      /* if base is not selected, not a parent of selection
+       * or not a child of selection and it is editable and selectable */
+      if ((base->flag_legacy & BA_WAS_SEL) && (base->flag & BASE_SELECTED) == 0 &&
+          BASE_EDITABLE(v3d, base) && BASE_SELECTABLE(v3d, base)) {
+
+        Object *ob_parent = ob->parent;
+        if (ob_parent != NULL) {
+          if (!BLI_gset_haskey(objects_in_transdata, ob)) {
+            bool parent_in_transdata = false;
+            while (ob_parent != NULL) {
+              if (BLI_gset_haskey(objects_in_transdata, ob_parent)) {
+                parent_in_transdata = true;
+                break;
+              }
+              ob_parent = ob_parent->parent;
+            }
+            if (parent_in_transdata) {
+              trans_obdata_in_obmode_ensure_object(tdo, ob);
+            }
+          }
+        }
+      }
+    }
+    BLI_gset_free(objects_in_transdata, NULL);
+  }
+
+  if (t->options & CTX_OBMODE_XFORM_SKIP_CHILDREN) {
+
+#define BASE_XFORM_INDIRECT(base) \
+  ((base->flag_legacy & BA_WAS_SEL) && (base->flag & BASE_SELECTED) == 0)
+
+    GSet *objects_in_transdata = BLI_gset_ptr_new_ex(__func__, tc->data_len);
+    GHash *objects_parent_root = BLI_ghash_ptr_new_ex(__func__, tc->data_len);
+    td = tc->data;
+    for (int i = 0; i < tc->data_len; i++, td++) {
+      if ((td->flag & TD_SKIP) == 0) {
+        BLI_gset_add(objects_in_transdata, td->ob);
+      }
+    }
+
+    ViewLayer *view_layer = t->view_layer;
+
+    for (Base *base = view_layer->object_bases.first; base; base = base->next) {
+      Object *ob = base->object;
+      if (ob->parent != NULL) {
+        if (ob->parent && !BLI_gset_haskey(objects_in_transdata, ob->parent) &&
+            !BLI_gset_haskey(objects_in_transdata, ob)) {
+          if (((base->flag_legacy & BA_WAS_SEL) && (base->flag & BASE_SELECTED) == 0)) {
+            Base *base_parent = BKE_view_layer_base_find(view_layer, ob->parent);
+            if (base_parent && !BASE_XFORM_INDIRECT(base_parent)) {
+              Object *ob_parent_recurse = ob->parent;
+              if (ob_parent_recurse != NULL) {
+                while (ob_parent_recurse != NULL) {
+                  if (BLI_gset_haskey(objects_in_transdata, ob_parent_recurse)) {
+                    break;
+                  }
+                  ob_parent_recurse = ob_parent_recurse->parent;
+                }
+
+                if (ob_parent_recurse) {
+                  trans_obchild_in_obmode_ensure_object(
+                      tdo, ob, ob_parent_recurse, OB_SKIP_CHILD_PARENT_APPLY_TRANSFORM);
+                  BLI_ghash_insert(objects_parent_root, ob, ob_parent_recurse);
+                  base->flag_legacy |= BA_TRANSFORM_LOCKED_IN_PLACE;
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+
+    for (Base *base = view_layer->object_bases.first; base; base = base->next) {
+      Object *ob = base->object;
+
+      if (BASE_XFORM_INDIRECT(base) || BLI_gset_haskey(objects_in_transdata, ob)) {
+        /* pass. */
+      }
+      else if (ob->parent != NULL) {
+        Base *base_parent = BKE_view_layer_base_find(view_layer, ob->parent);
+        if (base_parent) {
+          if (BASE_XFORM_INDIRECT(base_parent) ||
+              BLI_gset_haskey(objects_in_transdata, ob->parent)) {
+            trans_obchild_in_obmode_ensure_object(tdo, ob, NULL, OB_SKIP_CHILD_PARENT_IS_XFORM);
+            base->flag_legacy |= BA_TRANSFORM_LOCKED_IN_PLACE;
+          }
+          else {
+            Object *ob_parent_recurse = BLI_ghash_lookup(objects_parent_root, ob->parent);
+            if (ob_parent_recurse) {
+              trans_obchild_in_obmode_ensure_object(
+                  tdo, ob, ob_parent_recurse, OB_SKIP_CHILD_PARENT_IS_XFORM_INDIRECT);
+            }
+          }
+        }
+      }
+    }
+    BLI_gset_free(objects_in_transdata, NULL);
+    BLI_ghash_free(objects_parent_root, NULL, NULL);
+
+#undef BASE_XFORM_INDIRECT
+  }
 }
 
 /* transcribe given node into TransData2D for Transforming */
@@ -9665,6 +10162,13 @@ void createTransData(bContext *C, TransInfo *t)
     /* Needed for correct Object.obmat after duplication, see: T62135. */
     BKE_scene_graph_evaluated_ensure(t->depsgraph, CTX_data_main(t->context));
 
+    if ((scene->toolsettings->transform_flag & SCE_XFORM_DATA_ORIGIN) != 0) {
+      t->options |= CTX_OBMODE_XFORM_OBDATA;
+    }
+    if ((scene->toolsettings->transform_flag & SCE_XFORM_SKIP_CHILDREN) != 0) {
+      t->options |= CTX_OBMODE_XFORM_SKIP_CHILDREN;
+    }
+
     createTransObject(C, t);
     countAndCleanTransDataContainer(t);
     t->flag |= T_OBJECT;
diff --git a/source/blender/editors/transform/transform_generics.c b/source/blender/editors/transform/transform_generics.c
index 969e2558abb..7ae140e1815 100644
--- a/source/blender/editors/transform/transform_generics.c
+++ b/source/blender/editors/transform/transform_generics.c
@@ -81,6 +81,7 @@
 #include "BKE_workspace.h"
 
 #include "DEG_depsgraph.h"
+#include "DEG_depsgraph_query.h"
 
 #include "ED_anim_api.h"
 #include "ED_armature.h"
@@ -900,7 +901,7 @@ static void recalcData_objects(TransInfo *t)
         DEG_id_tag_update(tc->obedit->data, 0); /* sets recalc flags */
         BMEditMesh *em = BKE_editmesh_from_object(tc->obedit);
         EDBM_mesh_normals_update(em);
-        BKE_editmesh_tessface_calc(em);
+        BKE_editmesh_looptri_calc(em);
       }
     }
     else if (t->obedit_type == OB_ARMATURE) { /* no recalc flag, does pose */
@@ -1149,6 +1150,14 @@ static void recalcData_objects(TransInfo *t)
       /* Update motion paths once for all transformed objects. */
       ED_objects_recalculate_paths(t->context, t->scene, true);
     }
+
+    if (t->options & CTX_OBMODE_XFORM_SKIP_CHILDREN) {
+      trans_obchild_in_obmode_update_all(t);
+    }
+
+    if (t->options & CTX_OBMODE_XFORM_OBDATA) {
+      trans_obdata_in_obmode_update_all(t);
+    }
   }
 }
 
diff --git a/source/blender/editors/transform/transform_gizmo_3d.c b/source/blender/editors/transform/transform_gizmo_3d.c
index fbfeb4f53de..355cf139abf 100644
--- a/source/blender/editors/transform/transform_gizmo_3d.c
+++ b/source/blender/editors/transform/transform_gizmo_3d.c
@@ -1392,7 +1392,8 @@ void drawDial3d(const TransInfo *t)
     scale *= ED_view3d_pixel_size_no_ui_scale(t->ar->regiondata, mat_final[3]);
     mul_mat3_m4_fl(mat_final, scale);
 
-    if ((t->tsnap.mode & (SCE_SNAP_MODE_INCREMENT | SCE_SNAP_MODE_GRID)) && activeSnap(t)) {
+    if (activeSnap(t) && (!transformModeUseSnap(t) ||
+                          (t->tsnap.mode & (SCE_SNAP_MODE_INCREMENT | SCE_SNAP_MODE_GRID)))) {
       increment = (t->modifiers & MOD_PRECISION) ? t->snap[2] : t->snap[1];
     }
     else {
diff --git a/source/blender/editors/transform/transform_input.c b/source/blender/editors/transform/transform_input.c
index 6ebed88878f..e771fe43bd8 100644
--- a/source/blender/editors/transform/transform_input.c
+++ b/source/blender/editors/transform/transform_input.c
@@ -299,7 +299,7 @@ static void calcSpringFactor(MouseInput *mi)
 
 void initMouseInputMode(TransInfo *t, MouseInput *mi, MouseInputMode mode)
 {
-  /* incase we allocate a new value */
+  /* In case we allocate a new value. */
   void *mi_data_prev = mi->data;
 
   mi->use_virtual_mval = true;
diff --git a/source/blender/editors/transform/transform_ops.c b/source/blender/editors/transform/transform_ops.c
index f8e33fe70ad..604c5bbc083 100644
--- a/source/blender/editors/transform/transform_ops.c
+++ b/source/blender/editors/transform/transform_ops.c
@@ -139,22 +139,22 @@ const EnumPropertyItem rna_enum_transform_mode_types[] = {
     {TFM_CREASE, "CREASE", 0, "Crease", ""},
     {TFM_MIRROR, "MIRROR", 0, "Mirror", ""},
     {TFM_BONESIZE, "BONE_SIZE", 0, "Bonesize", ""},
-    {TFM_BONE_ENVELOPE, "BONE_ENVELOPE", 0, "Bone_Envelope", ""},
-    {TFM_BONE_ENVELOPE_DIST, "BONE_ENVELOPE_DIST", 0, "Bone_Envelope_Distance", ""},
-    {TFM_CURVE_SHRINKFATTEN, "CURVE_SHRINKFATTEN", 0, "Curve_Shrinkfatten", ""},
-    {TFM_MASK_SHRINKFATTEN, "MASK_SHRINKFATTEN", 0, "Mask_Shrinkfatten", ""},
-    {TFM_GPENCIL_SHRINKFATTEN, "GPENCIL_SHRINKFATTEN", 0, "GPencil_Shrinkfatten", ""},
-    {TFM_BONE_ROLL, "BONE_ROLL", 0, "Bone_Roll", ""},
-    {TFM_TIME_TRANSLATE, "TIME_TRANSLATE", 0, "Time_Translate", ""},
-    {TFM_TIME_SLIDE, "TIME_SLIDE", 0, "Time_Slide", ""},
-    {TFM_TIME_SCALE, "TIME_SCALE", 0, "Time_Scale", ""},
-    {TFM_TIME_EXTEND, "TIME_EXTEND", 0, "Time_Extend", ""},
-    {TFM_BAKE_TIME, "BAKE_TIME", 0, "Bake_Time", ""},
+    {TFM_BONE_ENVELOPE, "BONE_ENVELOPE", 0, "Bone Envelope", ""},
+    {TFM_BONE_ENVELOPE_DIST, "BONE_ENVELOPE_DIST", 0, "Bone Envelope Distance", ""},
+    {TFM_CURVE_SHRINKFATTEN, "CURVE_SHRINKFATTEN", 0, "Curve Shrinkfatten", ""},
+    {TFM_MASK_SHRINKFATTEN, "MASK_SHRINKFATTEN", 0, "Mask Shrinkfatten", ""},
+    {TFM_GPENCIL_SHRINKFATTEN, "GPENCIL_SHRINKFATTEN", 0, "GPencil Shrinkfatten", ""},
+    {TFM_BONE_ROLL, "BONE_ROLL", 0, "Bone Roll", ""},
+    {TFM_TIME_TRANSLATE, "TIME_TRANSLATE", 0, "Time Translate", ""},
+    {TFM_TIME_SLIDE, "TIME_SLIDE", 0, "Time Slide", ""},
+    {TFM_TIME_SCALE, "TIME_SCALE", 0, "Time Scale", ""},
+    {TFM_TIME_EXTEND, "TIME_EXTEND", 0, "Time Extend", ""},
+    {TFM_BAKE_TIME, "BAKE_TIME", 0, "Bake Time", ""},
     {TFM_BWEIGHT, "BWEIGHT", 0, "Bweight", ""},
     {TFM_ALIGN, "ALIGN", 0, "Align", ""},
     {TFM_EDGE_SLIDE, "EDGESLIDE", 0, "Edge Slide", ""},
     {TFM_SEQ_SLIDE, "SEQSLIDE", 0, "Sequence Slide", ""},
-    {TFM_GPENCIL_OPACITY, "GPENCIL_OPACITY", 0, "GPencil_Opacity", ""},
+    {TFM_GPENCIL_OPACITY, "GPENCIL_OPACITY", 0, "GPencil Opacity", ""},
     {0, NULL, 0, NULL, NULL},
 };
 
@@ -573,7 +573,7 @@ void Transform_Properties(struct wmOperatorType *ot, int flags)
   if (flags & P_ORIENT_AXIS_ORTHO) {
     prop = RNA_def_property(ot->srna, "orient_axis_ortho", PROP_ENUM, PROP_NONE);
     RNA_def_property_ui_text(prop, "Axis Ortho", "");
-    RNA_def_property_enum_default(prop, 1);
+    RNA_def_property_enum_default(prop, 0);
     RNA_def_property_enum_items(prop, rna_enum_axis_xyz_items);
     RNA_def_property_flag(prop, PROP_SKIP_SAVE);
   }
diff --git a/source/blender/editors/transform/transform_snap.c b/source/blender/editors/transform/transform_snap.c
index dbcc6c1b04a..571ce7a6bc2 100644
--- a/source/blender/editors/transform/transform_snap.c
+++ b/source/blender/editors/transform/transform_snap.c
@@ -166,7 +166,10 @@ void drawSnapping(const struct bContext *C, TransInfo *t)
   activeCol[3] = 192;
 
   if (t->spacetype == SPACE_VIEW3D) {
-    if (validSnap(t)) {
+    bool draw_target = (t->tsnap.status & TARGET_INIT) &&
+                       (t->scene->toolsettings->snap_mode & SCE_SNAP_MODE_EDGE_PERPENDICULAR);
+
+    if (draw_target || validSnap(t)) {
       TransSnapPoint *p;
       RegionView3D *rv3d = CTX_wm_region_view3d(C);
       float imat[4][4];
@@ -214,6 +217,48 @@ void drawSnapping(const struct bContext *C, TransInfo *t)
         immEnd();
       }
 
+      if (draw_target) {
+        /* Draw snapTarget */
+        float targ_co[3], vx[3], vy[3], v1[3], v2[3], v3[3], v4[4];
+        copy_v3_v3(targ_co, t->tsnap.snapTarget);
+        float px_size = 0.75f * size * ED_view3d_pixel_size(rv3d, targ_co);
+
+        mul_v3_v3fl(vx, imat[0], px_size);
+        mul_v3_v3fl(vy, imat[1], px_size);
+
+        add_v3_v3v3(v1, vx, vy);
+        sub_v3_v3v3(v2, vx, vy);
+        negate_v3_v3(v3, v1);
+        negate_v3_v3(v4, v2);
+
+        add_v3_v3(v1, targ_co);
+        add_v3_v3(v2, targ_co);
+        add_v3_v3(v3, targ_co);
+        add_v3_v3(v4, targ_co);
+
+        immUniformColor4ubv(col);
+        immBegin(GPU_PRIM_LINES, 4);
+        immVertex3fv(pos, v3);
+        immVertex3fv(pos, v1);
+        immVertex3fv(pos, v4);
+        immVertex3fv(pos, v2);
+        immEnd();
+
+        if (t->tsnap.snapElem & SCE_SNAP_MODE_EDGE_PERPENDICULAR) {
+          immUnbindProgram();
+
+          immBindBuiltinProgram(GPU_SHADER_3D_LINE_DASHED_UNIFORM_COLOR);
+          immUniform1f("dash_width", 6.0f * U.pixelsize);
+          immUniform1f("dash_factor", 1.0f / 4.0f);
+          immUniformColor4ubv(col);
+
+          immBegin(GPU_PRIM_LINES, 2);
+          immVertex3fv(pos, targ_co);
+          immVertex3fv(pos, t->tsnap.snapPoint);
+          immEnd();
+        }
+      }
+
       immUnbindProgram();
 
       GPU_depth_test(true);
@@ -328,6 +373,7 @@ void applyProject(TransInfo *t)
                       .use_occlusion_test = false,
                   },
                   mval_fl,
+                  NULL,
                   0,
                   loc,
                   no)) {
@@ -586,12 +632,18 @@ static void initSnappingMode(TransInfo *t)
     else if (t->tsnap.applySnap != NULL &&  // A snapping function actually exist
              (obedit_type == -1))           // Object Mode
     {
-      /* In "Edit Strokes" mode,
-       * snap tool can perform snap to selected or active objects (see T49632)
-       * TODO: perform self snap in gpencil_strokes */
-      t->tsnap.modeSelect = (((t->options & (CTX_GPENCIL_STROKES | CTX_CURSOR)) != 0) ?
-                                 SNAP_ALL :
-                                 SNAP_NOT_SELECTED);
+
+      if (t->options & (CTX_GPENCIL_STROKES | CTX_CURSOR | CTX_OBMODE_XFORM_OBDATA)) {
+        /* In "Edit Strokes" mode,
+         * snap tool can perform snap to selected or active objects (see T49632)
+         * TODO: perform self snap in gpencil_strokes.
+         *
+         * When we're moving the origins, allow snapping onto our own geometry (see T69132). */
+        t->tsnap.modeSelect = SNAP_ALL;
+      }
+      else {
+        t->tsnap.modeSelect = SNAP_NOT_SELECTED;
+      }
     }
     else {
       /* Grid if snap is not possible */
@@ -993,18 +1045,25 @@ static void CalcSnapGeometry(TransInfo *t, float *UNUSED(vec))
     float no[3];
     float mval[2];
     bool found = false;
+    short snap_elem = 0;
     float dist_px = SNAP_MIN_DISTANCE;  // Use a user defined value here
 
     mval[0] = t->mval[0];
     mval[1] = t->mval[1];
 
-    if (t->tsnap.mode & (SCE_SNAP_MODE_VERTEX | SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_FACE)) {
+    if (t->tsnap.mode & (SCE_SNAP_MODE_VERTEX | SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_FACE |
+                         SCE_SNAP_MODE_EDGE_MIDPOINT | SCE_SNAP_MODE_EDGE_PERPENDICULAR)) {
       zero_v3(no); /* objects won't set this */
-      found = snapObjectsTransform(t, mval, &dist_px, loc, no);
+      snap_elem = snapObjectsTransform(t, mval, &dist_px, loc, no);
+      found = snap_elem != 0;
     }
     if ((found == false) && (t->tsnap.mode & SCE_SNAP_MODE_VOLUME)) {
       found = peelObjectsTransform(
           t, mval, (t->settings->snap_flag & SCE_SNAP_PEEL_OBJECT) != 0, loc, no, NULL);
+
+      if (found) {
+        snap_elem = SCE_SNAP_MODE_VOLUME;
+      }
     }
 
     if (found == true) {
@@ -1016,6 +1075,8 @@ static void CalcSnapGeometry(TransInfo *t, float *UNUSED(vec))
     else {
       t->tsnap.status &= ~POINT_INIT;
     }
+
+    t->tsnap.snapElem = (char)snap_elem;
   }
   else if (t->spacetype == SPACE_IMAGE && t->obedit_type == OB_MESH) {
     if (t->tsnap.mode & SCE_SNAP_MODE_VERTEX) {
@@ -1186,7 +1247,11 @@ static void TargetSnapClosest(TransInfo *t)
       FOREACH_TRANS_DATA_CONTAINER (t, tc) {
         TransData *td = tc->data;
         for (td = tc->data, i = 0; i < tc->data_len && td->flag & TD_SELECTED; i++, td++) {
-          struct BoundBox *bb = BKE_object_boundbox_get(td->ob);
+          const BoundBox *bb = NULL;
+
+          if ((t->options & CTX_OBMODE_XFORM_OBDATA) == 0) {
+            bb = BKE_object_boundbox_get(td->ob);
+          }
 
           /* use boundbox if possible */
           if (bb) {
@@ -1259,10 +1324,10 @@ static void TargetSnapClosest(TransInfo *t)
   }
 }
 
-bool snapObjectsTransform(
+short snapObjectsTransform(
     TransInfo *t, const float mval[2], float *dist_px, float r_loc[3], float r_no[3])
 {
-  return ED_transform_snap_object_project_view3d(
+  return ED_transform_snap_object_project_view3d_ex(
       t->tsnap.object_context,
       t->scene->toolsettings->snap_mode,
       &(const struct SnapObjectParams){
@@ -1271,9 +1336,13 @@ bool snapObjectsTransform(
           .use_occlusion_test = t->scene->toolsettings->snap_mode != SCE_SNAP_MODE_FACE,
       },
       mval,
+      t->tsnap.snapTarget,
       dist_px,
       r_loc,
-      r_no);
+      r_no,
+      NULL,
+      NULL,
+      NULL);
 }
 
 /******************** PEELING *********************************/
diff --git a/source/blender/editors/transform/transform_snap_object.c b/source/blender/editors/transform/transform_snap_object.c
index 67bd107ab49..d3cf52ce3fd 100644
--- a/source/blender/editors/transform/transform_snap_object.c
+++ b/source/blender/editors/transform/transform_snap_object.c
@@ -221,22 +221,40 @@ static void iter_snap_objects(SnapObjectContext *sctx,
 
   Base *base_act = view_layer->basact;
   for (Base *base = view_layer->object_bases.first; base != NULL; base = base->next) {
-    if ((BASE_VISIBLE(v3d, base)) && (base->flag_legacy & BA_SNAP_FIX_DEPS_FIASCO) == 0 &&
-        !((snap_select == SNAP_NOT_SELECTED &&
-           ((base->flag & BASE_SELECTED) || (base->flag_legacy & BA_WAS_SEL))) ||
-          (snap_select == SNAP_NOT_ACTIVE && base == base_act))) {
-      Object *obj_eval = DEG_get_evaluated_object(sctx->depsgraph, base->object);
-      if (obj_eval->transflag & OB_DUPLI) {
-        DupliObject *dupli_ob;
-        ListBase *lb = object_duplilist(sctx->depsgraph, sctx->scene, obj_eval);
-        for (dupli_ob = lb->first; dupli_ob; dupli_ob = dupli_ob->next) {
-          sob_callback(sctx, use_object_edit_cage, dupli_ob->ob, dupli_ob->mat, data);
-        }
-        free_object_duplilist(lb);
+
+    if (!BASE_VISIBLE(v3d, base)) {
+      continue;
+    }
+
+    if (base->flag_legacy & BA_TRANSFORM_LOCKED_IN_PLACE) {
+      /* pass */
+    }
+    else if (base->flag_legacy & BA_SNAP_FIX_DEPS_FIASCO) {
+      continue;
+    }
+
+    if (snap_select == SNAP_NOT_SELECTED) {
+      if ((base->flag & BASE_SELECTED) || (base->flag_legacy & BA_WAS_SEL)) {
+        continue;
+      }
+    }
+    else if (snap_select == SNAP_NOT_ACTIVE) {
+      if (base == base_act) {
+        continue;
       }
+    }
 
-      sob_callback(sctx, use_object_edit_cage, obj_eval, obj_eval->obmat, data);
+    Object *obj_eval = DEG_get_evaluated_object(sctx->depsgraph, base->object);
+    if (obj_eval->transflag & OB_DUPLI) {
+      DupliObject *dupli_ob;
+      ListBase *lb = object_duplilist(sctx->depsgraph, sctx->scene, obj_eval);
+      for (dupli_ob = lb->first; dupli_ob; dupli_ob = dupli_ob->next) {
+        sob_callback(sctx, use_object_edit_cage, dupli_ob->ob, dupli_ob->mat, data);
+      }
+      free_object_duplilist(lb);
     }
+
+    sob_callback(sctx, use_object_edit_cage, obj_eval, obj_eval->obmat, data);
   }
 }
 
@@ -583,37 +601,31 @@ static bool raycastEditMesh(SnapObjectContext *sctx,
   }
 
   if (treedata->tree == NULL) {
-    BVHCache **bvh_cache = NULL;
-    BLI_bitmap *elem_mask = NULL;
-    BMEditMesh *em_orig;
-    int looptri_num_active = -1;
-
     /* Get original version of the edit_mesh. */
-    em_orig = BKE_editmesh_from_object(DEG_get_original_object(ob));
+    BMEditMesh *em_orig = BKE_editmesh_from_object(DEG_get_original_object(ob));
 
     if (sctx->callbacks.edit_mesh.test_face_fn) {
       BMesh *bm = em_orig->bm;
       BLI_assert(poly_to_tri_count(bm->totface, bm->totloop) == em_orig->tottri);
 
-      elem_mask = BLI_BITMAP_NEW(em_orig->tottri, __func__);
-      looptri_num_active = BM_iter_mesh_bitmap_from_filter_tessface(
+      BLI_bitmap *elem_mask = BLI_BITMAP_NEW(em_orig->tottri, __func__);
+      int looptri_num_active = BM_iter_mesh_bitmap_from_filter_tessface(
           bm,
           elem_mask,
           sctx->callbacks.edit_mesh.test_face_fn,
           sctx->callbacks.edit_mesh.user_data);
+
+      bvhtree_from_editmesh_looptri_ex(
+          treedata, em_orig, elem_mask, looptri_num_active, 0.0f, 4, 6, 0, NULL);
+
+      MEM_freeN(elem_mask);
     }
     else {
       /* Only cache if bvhtree is created without a mask.
        * This helps keep a standardized bvhtree in cache. */
-      bvh_cache = em_bvh_cache;
+      BKE_bvhtree_from_editmesh_get(treedata, em_orig, 4, BVHTREE_FROM_EM_LOOPTRI, em_bvh_cache);
     }
 
-    bvhtree_from_editmesh_looptri_ex(
-        treedata, em_orig, elem_mask, looptri_num_active, 0.0f, 4, 6, bvh_cache);
-
-    if (elem_mask) {
-      MEM_freeN(elem_mask);
-    }
     if (treedata->tree == NULL) {
       return retval;
     }
@@ -1260,7 +1272,8 @@ static short snap_mesh_polygon(SnapObjectContext *sctx,
 
     const MPoly *mp = &((SnapObjectData_Mesh *)sod)->poly[*r_index];
     const MLoop *ml = &treedata->loop[mp->loopstart];
-    if (snapdata->snap_to_flag & SCE_SNAP_MODE_EDGE) {
+    if (snapdata->snap_to_flag &
+        (SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_EDGE_MIDPOINT | SCE_SNAP_MODE_EDGE_PERPENDICULAR)) {
       elem = SCE_SNAP_MODE_EDGE;
       BLI_assert(treedata->edge != NULL);
       for (int i = mp->totloop; i--; ml++) {
@@ -1297,7 +1310,8 @@ static short snap_mesh_polygon(SnapObjectContext *sctx,
     BMFace *f = BM_face_at_index(em->bm, *r_index);
     BMLoop *l_iter, *l_first;
     l_iter = l_first = BM_FACE_FIRST_LOOP(f);
-    if (snapdata->snap_to_flag & SCE_SNAP_MODE_EDGE) {
+    if (snapdata->snap_to_flag &
+        (SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_EDGE_MIDPOINT | SCE_SNAP_MODE_EDGE_PERPENDICULAR)) {
       elem = SCE_SNAP_MODE_EDGE;
       BM_mesh_elem_index_ensure(em->bm, BM_EDGE);
       BM_mesh_elem_table_ensure(em->bm, BM_VERT | BM_EDGE);
@@ -1352,6 +1366,7 @@ static short snap_mesh_edge_verts_mixed(SnapObjectContext *sctx,
                                         Object *ob,
                                         float obmat[4][4],
                                         float original_dist_px,
+                                        const float prev_co[3],
                                         /* read/write args */
                                         float *dist_px,
                                         /* return args */
@@ -1420,19 +1435,74 @@ static short snap_mesh_edge_verts_mixed(SnapObjectContext *sctx,
                         &lambda)) {
     /* do nothing */
   }
-  else if (lambda < 0.25f || 0.75f < lambda) {
-    int v_id = lambda < 0.5f ? 0 : 1;
+  else {
+    if (snapdata->snap_to_flag & SCE_SNAP_MODE_VERTEX) {
+      if (lambda < 0.25f || 0.75f < lambda) {
+        int v_id = lambda < 0.5f ? 0 : 1;
 
-    if (test_projected_vert_dist(&neasrest_precalc,
-                                 NULL,
-                                 0,
-                                 nearest2d.is_persp,
-                                 v_pair[v_id],
-                                 &nearest.dist_sq,
-                                 nearest.co)) {
-      nearest.index = vindex[v_id];
-      nearest2d.copy_vert_no(vindex[v_id], nearest.no, nearest2d.userdata);
-      elem = SCE_SNAP_MODE_VERTEX;
+        if (test_projected_vert_dist(&neasrest_precalc,
+                                     NULL,
+                                     0,
+                                     nearest2d.is_persp,
+                                     v_pair[v_id],
+                                     &nearest.dist_sq,
+                                     nearest.co)) {
+          nearest.index = vindex[v_id];
+          nearest2d.copy_vert_no(vindex[v_id], nearest.no, nearest2d.userdata);
+          elem = SCE_SNAP_MODE_VERTEX;
+        }
+      }
+    }
+
+    if (snapdata->snap_to_flag & SCE_SNAP_MODE_EDGE_MIDPOINT) {
+      if (0.375f < lambda && lambda < 0.625f) {
+        float vmid[3];
+        mid_v3_v3v3(vmid, v_pair[0], v_pair[1]);
+
+        if (test_projected_vert_dist(&neasrest_precalc,
+                                     NULL,
+                                     0,
+                                     nearest2d.is_persp,
+                                     vmid,
+                                     &nearest.dist_sq,
+                                     nearest.co)) {
+          float v_nor[2][3];
+          nearest2d.copy_vert_no(vindex[0], v_nor[0], nearest2d.userdata);
+          nearest2d.copy_vert_no(vindex[1], v_nor[1], nearest2d.userdata);
+          mid_v3_v3v3(nearest.no, v_nor[0], v_nor[1]);
+          nearest.index = *r_index;
+          elem = SCE_SNAP_MODE_EDGE_MIDPOINT;
+        }
+      }
+    }
+
+    if (prev_co && (snapdata->snap_to_flag & SCE_SNAP_MODE_EDGE_PERPENDICULAR)) {
+      float v_near[3], va_g[3], vb_g[3];
+
+      mul_v3_m4v3(va_g, obmat, v_pair[0]);
+      mul_v3_m4v3(vb_g, obmat, v_pair[1]);
+      lambda = line_point_factor_v3(prev_co, va_g, vb_g);
+
+      if (IN_RANGE(lambda, 0.0f, 1.0f)) {
+        interp_v3_v3v3(v_near, va_g, vb_g, lambda);
+
+        if ((len_squared_v3v3(prev_co, v_near) > FLT_EPSILON) &&
+            test_projected_vert_dist(&neasrest_precalc,
+                                     NULL,
+                                     0,
+                                     nearest2d.is_persp,
+                                     v_near,
+                                     &nearest.dist_sq,
+                                     nearest.co)) {
+          float v_nor[2][3];
+          nearest2d.copy_vert_no(vindex[0], v_nor[0], nearest2d.userdata);
+          nearest2d.copy_vert_no(vindex[1], v_nor[1], nearest2d.userdata);
+          mid_v3_v3v3(nearest.no, v_nor[0], v_nor[1]);
+
+          nearest.index = *r_index;
+          elem = SCE_SNAP_MODE_EDGE_PERPENDICULAR;
+        }
+      }
     }
   }
 
@@ -1440,7 +1510,9 @@ static short snap_mesh_edge_verts_mixed(SnapObjectContext *sctx,
     *dist_px = sqrtf(nearest.dist_sq);
 
     copy_v3_v3(r_loc, nearest.co);
-    mul_m4_v3(obmat, r_loc);
+    if (elem != SCE_SNAP_MODE_EDGE_PERPENDICULAR) {
+      mul_m4_v3(obmat, r_loc);
+    }
 
     if (r_no) {
       float imat[4][4];
@@ -1926,13 +1998,13 @@ static short snapMesh(SnapObjectContext *sctx,
 {
   BLI_assert(snapdata->snap_to_flag != SCE_SNAP_MODE_FACE);
 
-  if ((snapdata->snap_to_flag & ~SCE_SNAP_MODE_FACE) == SCE_SNAP_MODE_EDGE) {
-    if (me->totedge == 0) {
+  if ((snapdata->snap_to_flag & ~SCE_SNAP_MODE_FACE) == SCE_SNAP_MODE_VERTEX) {
+    if (me->totvert == 0) {
       return 0;
     }
   }
   else {
-    if (me->totvert == 0) {
+    if (me->totedge == 0) {
       return 0;
     }
   }
@@ -2066,7 +2138,8 @@ static short snapMesh(SnapObjectContext *sctx,
     last_index = nearest.index;
   }
 
-  if (snapdata->snap_to_flag & SCE_SNAP_MODE_EDGE) {
+  if (snapdata->snap_to_flag &
+      (SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_EDGE_MIDPOINT | SCE_SNAP_MODE_EDGE_PERPENDICULAR)) {
     if (bvhtree[0]) {
       /* snap to loose edges */
       BLI_bvhtree_find_nearest_projected(bvhtree[0],
@@ -2164,13 +2237,13 @@ static short snapEditMesh(SnapObjectContext *sctx,
 {
   BLI_assert(snapdata->snap_to_flag != SCE_SNAP_MODE_FACE);
 
-  if ((snapdata->snap_to_flag & ~SCE_SNAP_MODE_FACE) == SCE_SNAP_MODE_EDGE) {
-    if (em->bm->totedge == 0) {
+  if ((snapdata->snap_to_flag & ~SCE_SNAP_MODE_FACE) == SCE_SNAP_MODE_VERTEX) {
+    if (em->bm->totvert == 0) {
       return 0;
     }
   }
   else {
-    if (em->bm->totvert == 0) {
+    if (em->bm->totedge == 0) {
       return 0;
     }
   }
@@ -2223,16 +2296,17 @@ static short snapEditMesh(SnapObjectContext *sctx,
             sctx->callbacks.edit_mesh.user_data);
 
         bvhtree_from_editmesh_verts_ex(
-            treedata_vert, em, verts_mask, verts_num_active, 0.0f, 2, 6);
+            treedata_vert, em, verts_mask, verts_num_active, 0.0f, 2, 6, 0, NULL);
         MEM_freeN(verts_mask);
       }
       else {
-        bvhtree_from_editmesh_verts(treedata_vert, em, 0.0f, 2, 6, em_bvh_cache);
+        BKE_bvhtree_from_editmesh_get(treedata_vert, em, 2, BVHTREE_FROM_EM_VERTS, em_bvh_cache);
       }
     }
   }
 
-  if (snapdata->snap_to_flag & SCE_SNAP_MODE_EDGE) {
+  if (snapdata->snap_to_flag &
+      (SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_EDGE_MIDPOINT | SCE_SNAP_MODE_EDGE_PERPENDICULAR)) {
     if (sod->bvh_trees[1] == NULL) {
       sod->bvh_trees[1] = BLI_memarena_calloc(sctx->cache.mem_arena, sizeof(**sod->bvh_trees));
     }
@@ -2258,11 +2332,11 @@ static short snapEditMesh(SnapObjectContext *sctx,
             sctx->callbacks.edit_mesh.user_data);
 
         bvhtree_from_editmesh_edges_ex(
-            treedata_edge, em, edges_mask, edges_num_active, 0.0f, 2, 6);
+            treedata_edge, em, edges_mask, edges_num_active, 0.0f, 2, 6, 0, NULL);
         MEM_freeN(edges_mask);
       }
       else {
-        bvhtree_from_editmesh_edges(treedata_edge, em, 0.0f, 2, 6, em_bvh_cache);
+        BKE_bvhtree_from_editmesh_get(treedata_edge, em, 2, BVHTREE_FROM_EM_EDGES, em_bvh_cache);
       }
     }
   }
@@ -2279,7 +2353,6 @@ static short snapEditMesh(SnapObjectContext *sctx,
       .index = -1,
       .dist_sq = dist_px_sq,
   };
-  int last_index = nearest.index;
   short elem = SCE_SNAP_MODE_VERTEX;
 
   float tobmat[4][4], clip_planes_local[MAX_CLIPPLANE_LEN][4];
@@ -2300,11 +2373,12 @@ static short snapEditMesh(SnapObjectContext *sctx,
                                        &nearest,
                                        cb_snap_vert,
                                        &nearest2d);
-
-    last_index = nearest.index;
   }
 
-  if (treedata_edge && snapdata->snap_to_flag & SCE_SNAP_MODE_EDGE) {
+  if (treedata_edge && snapdata->snap_to_flag & (SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_EDGE_MIDPOINT |
+                                                 SCE_SNAP_MODE_EDGE_PERPENDICULAR)) {
+    int last_index = nearest.index;
+    nearest.index = -1;
     BM_mesh_elem_table_ensure(em->bm, BM_EDGE | BM_VERT);
     BLI_bvhtree_find_nearest_projected(treedata_edge->tree,
                                        lpmat,
@@ -2316,9 +2390,12 @@ static short snapEditMesh(SnapObjectContext *sctx,
                                        cb_snap_edge,
                                        &nearest2d);
 
-    if (last_index != nearest.index) {
+    if (nearest.index != -1) {
       elem = SCE_SNAP_MODE_EDGE;
     }
+    else {
+      nearest.index = last_index;
+    }
   }
 
   if (nearest.index != -1) {
@@ -2700,6 +2777,7 @@ static short transform_snap_context_project_view3d_mixed_impl(
     const unsigned short snap_to_flag,
     const struct SnapObjectParams *params,
     const float mval[2],
+    const float prev_co[3],
     float *dist_px,
     float r_loc[3],
     float r_no[3],
@@ -2707,7 +2785,8 @@ static short transform_snap_context_project_view3d_mixed_impl(
     Object **r_ob,
     float r_obmat[4][4])
 {
-  BLI_assert((snap_to_flag & (SCE_SNAP_MODE_VERTEX | SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_FACE)) !=
+  BLI_assert((snap_to_flag & (SCE_SNAP_MODE_VERTEX | SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_FACE |
+                              SCE_SNAP_MODE_EDGE_MIDPOINT | SCE_SNAP_MODE_EDGE_PERPENDICULAR)) !=
              0);
 
   short retval = 0;
@@ -2733,7 +2812,7 @@ static short transform_snap_context_project_view3d_mixed_impl(
                                          ray_normal,
                                          ray_start,
                                          true)) {
-      return false;
+      return 0;
     }
 
     float dummy_ray_depth = BVH_RAYCAST_DIST_MAX;
@@ -2746,7 +2825,8 @@ static short transform_snap_context_project_view3d_mixed_impl(
     }
   }
 
-  if (snap_to_flag & (SCE_SNAP_MODE_VERTEX | SCE_SNAP_MODE_EDGE)) {
+  if (snap_to_flag & (SCE_SNAP_MODE_VERTEX | SCE_SNAP_MODE_EDGE | SCE_SNAP_MODE_EDGE_MIDPOINT |
+                      SCE_SNAP_MODE_EDGE_PERPENDICULAR)) {
     short elem;
     float dist_px_tmp = *dist_px;
 
@@ -2799,11 +2879,18 @@ static short transform_snap_context_project_view3d_mixed_impl(
       retval = elem;
     }
 
-    if ((retval == SCE_SNAP_MODE_EDGE) && (snapdata.snap_to_flag & SCE_SNAP_MODE_VERTEX)) {
-      retval = snap_mesh_edge_verts_mixed(
-          sctx, &snapdata, ob, obmat, *dist_px, &dist_px_tmp, loc, no, &index);
+    if ((retval == SCE_SNAP_MODE_EDGE) &&
+        (snap_to_flag & (SCE_SNAP_MODE_VERTEX | SCE_SNAP_MODE_EDGE_MIDPOINT |
+                         SCE_SNAP_MODE_EDGE_PERPENDICULAR))) {
+      elem = snap_mesh_edge_verts_mixed(
+          sctx, &snapdata, ob, obmat, *dist_px, prev_co, &dist_px_tmp, loc, no, &index);
+    }
+
+    if (elem) {
+      retval = elem;
     }
 
+    retval &= snap_to_flag;
     *dist_px = dist_px_tmp;
   }
 
@@ -2831,6 +2918,7 @@ short ED_transform_snap_object_project_view3d_ex(SnapObjectContext *sctx,
                                                  const unsigned short snap_to,
                                                  const struct SnapObjectParams *params,
                                                  const float mval[2],
+                                                 const float prev_co[3],
                                                  float *dist_px,
                                                  float r_loc[3],
                                                  float r_no[3],
@@ -2839,7 +2927,7 @@ short ED_transform_snap_object_project_view3d_ex(SnapObjectContext *sctx,
                                                  float r_obmat[4][4])
 {
   return transform_snap_context_project_view3d_mixed_impl(
-             sctx, snap_to, params, mval, dist_px, r_loc, r_no, r_index, r_ob, r_obmat) != 0;
+      sctx, snap_to, params, mval, prev_co, dist_px, r_loc, r_no, r_index, r_ob, r_obmat);
 }
 
 /**
@@ -2849,6 +2937,7 @@ short ED_transform_snap_object_project_view3d_ex(SnapObjectContext *sctx,
  *
  * \param sctx: Snap context.
  * \param mval: Screenspace coordinate.
+ * \param prev_co: Coordinate for perpendicular point calculation (optional).
  * \param dist_px: Maximum distance to snap (in pixels).
  * \param r_co: hit location.
  * \param r_no: hit normal (optional).
@@ -2858,12 +2947,13 @@ bool ED_transform_snap_object_project_view3d(SnapObjectContext *sctx,
                                              const unsigned short snap_to,
                                              const struct SnapObjectParams *params,
                                              const float mval[2],
+                                             const float prev_co[3],
                                              float *dist_px,
                                              float r_loc[3],
                                              float r_no[3])
 {
   return ED_transform_snap_object_project_view3d_ex(
-             sctx, snap_to, params, mval, dist_px, r_loc, r_no, NULL, NULL, NULL) != 0;
+             sctx, snap_to, params, mval, prev_co, dist_px, r_loc, r_no, NULL, NULL, NULL) != 0;
 }
 
 /**
diff --git a/source/blender/editors/undo/ed_undo.c b/source/blender/editors/undo/ed_undo.c
index c84d7539cce..e0f4d1cf465 100644
--- a/source/blender/editors/undo/ed_undo.c
+++ b/source/blender/editors/undo/ed_undo.c
@@ -53,6 +53,7 @@
 #include "ED_gpencil.h"
 #include "ED_render.h"
 #include "ED_object.h"
+#include "ED_outliner.h"
 #include "ED_screen.h"
 #include "ED_undo.h"
 
@@ -390,6 +391,8 @@ static int ed_undo_exec(bContext *C, wmOperator *op)
     /* Keep button under the cursor active. */
     WM_event_add_mousemove(C);
   }
+
+  ED_outliner_select_sync_from_all_tag(C);
   return ret;
 }
 
@@ -417,6 +420,8 @@ static int ed_redo_exec(bContext *C, wmOperator *op)
     /* Keep button under the cursor active. */
     WM_event_add_mousemove(C);
   }
+
+  ED_outliner_select_sync_from_all_tag(C);
   return ret;
 }
 
diff --git a/source/blender/editors/uvedit/uvedit_draw.c b/source/blender/editors/uvedit/uvedit_draw.c
index d97714061c0..fe761f65702 100644
--- a/source/blender/editors/uvedit/uvedit_draw.c
+++ b/source/blender/editors/uvedit/uvedit_draw.c
@@ -205,14 +205,14 @@ static void draw_uvs_shadow(SpaceImage *UNUSED(sima),
                             Object *obedit,
                             Depsgraph *depsgraph)
 {
-  Object *eval_ob = DEG_get_evaluated_object(depsgraph, obedit);
-  Mesh *me = eval_ob->data;
+  Object *ob_eval = DEG_get_evaluated_object(depsgraph, obedit);
+  Mesh *me = ob_eval->data;
   float col[4];
   UI_GetThemeColor4fv(TH_UV_SHADOW, col);
 
   DRW_mesh_batch_cache_validate(me);
   GPUBatch *edges = DRW_mesh_batch_cache_get_uv_edges(me);
-  DRW_mesh_batch_cache_create_requested(eval_ob, me, scene, false, false);
+  DRW_mesh_batch_cache_create_requested(ob_eval, me, scene, false, false);
 
   if (edges) {
     GPU_batch_program_set_builtin(edges, GPU_SHADER_2D_UV_UNIFORM_COLOR);
@@ -223,8 +223,8 @@ static void draw_uvs_shadow(SpaceImage *UNUSED(sima),
 
 static void draw_uvs_texpaint(Scene *scene, Object *ob, Depsgraph *depsgraph)
 {
-  Object *eval_ob = DEG_get_evaluated_object(depsgraph, ob);
-  Mesh *me = eval_ob->data;
+  Object *ob_eval = DEG_get_evaluated_object(depsgraph, ob);
+  Mesh *me = ob_eval->data;
   ToolSettings *ts = scene->toolsettings;
   float col[4];
   UI_GetThemeColor4fv(TH_UV_SHADOW, col);
@@ -235,7 +235,7 @@ static void draw_uvs_texpaint(Scene *scene, Object *ob, Depsgraph *depsgraph)
 
   DRW_mesh_batch_cache_validate(me);
   GPUBatch *geom = DRW_mesh_batch_cache_get_uv_edges(me);
-  DRW_mesh_batch_cache_create_requested(eval_ob, me, scene, false, false);
+  DRW_mesh_batch_cache_create_requested(ob_eval, me, scene, false, false);
 
   GPU_batch_program_set_builtin(geom, GPU_SHADER_2D_UV_UNIFORM_COLOR);
   GPU_batch_uniform_4fv(geom, "color", col);
@@ -246,7 +246,7 @@ static void draw_uvs_texpaint(Scene *scene, Object *ob, Depsgraph *depsgraph)
     MPoly *mpoly = me->mpoly;
     uint draw_start = 0;
     uint idx = 0;
-    bool prev_ma_match = (mpoly->mat_nr == (eval_ob->actcol - 1));
+    bool prev_ma_match = (mpoly->mat_nr == (ob_eval->actcol - 1));
 
     GPU_matrix_bind(geom->interface);
     GPU_batch_bind(geom);
@@ -255,7 +255,7 @@ static void draw_uvs_texpaint(Scene *scene, Object *ob, Depsgraph *depsgraph)
      * we can use multi draw indirect drawcalls for this.
      * (not implemented in GPU module at the time of writing). */
     for (int a = 0; a < me->totpoly; a++, mpoly++) {
-      bool ma_match = (mpoly->mat_nr == (eval_ob->actcol - 1));
+      bool ma_match = (mpoly->mat_nr == (ob_eval->actcol - 1));
       if (ma_match != prev_ma_match) {
         if (ma_match == false) {
           GPU_batch_draw_advanced(geom, draw_start, idx - draw_start, 0, 0);
@@ -282,13 +282,13 @@ static void draw_uvs_texpaint(Scene *scene, Object *ob, Depsgraph *depsgraph)
 static void draw_uvs(SpaceImage *sima, Scene *scene, Object *obedit, Depsgraph *depsgraph)
 {
   GPUBatch *faces, *edges, *verts, *facedots;
-  Object *eval_ob = DEG_get_evaluated_object(depsgraph, obedit);
+  Object *ob_eval = DEG_get_evaluated_object(depsgraph, obedit);
   const ToolSettings *ts = scene->toolsettings;
   float col1[4], col2[4], col3[4], transparent[4] = {0.0f, 0.0f, 0.0f, 0.0f};
 
   if (sima->flag & SI_DRAWSHADOW) {
     bool is_cage_like_final_meshes = false;
-    Mesh *me = (Mesh *)eval_ob->data;
+    Mesh *me = (Mesh *)ob_eval->data;
     BMEditMesh *embm = me->edit_mesh;
     is_cage_like_final_meshes = embm && embm->mesh_eval_final &&
                                 embm->mesh_eval_final->runtime.is_original;
@@ -300,7 +300,7 @@ static void draw_uvs(SpaceImage *sima, Scene *scene, Object *obedit, Depsgraph *
     }
   }
 
-  uvedit_get_batches(eval_ob, sima, scene, &faces, &edges, &verts, &facedots);
+  uvedit_get_batches(ob_eval, sima, scene, &faces, &edges, &verts, &facedots);
 
   bool interpedges;
   bool draw_stretch = (sima->flag & SI_DRAW_STRETCH) != 0;
@@ -367,33 +367,33 @@ static void draw_uvs(SpaceImage *sima, Scene *scene, Object *obedit, Depsgraph *
         break;
       }
       case SI_UVDT_BLACK:
-      case SI_UVDT_WHITE: {
-        GPU_line_width(1.0f);
-        GPU_batch_program_set_builtin(edges, GPU_SHADER_2D_UNIFORM_COLOR);
-        if (sima->dt_uv == SI_UVDT_WHITE) {
-          GPU_batch_uniform_4f(edges, "color", 1.0f, 1.0f, 1.0f, 1.0f);
-        }
-        else {
-          GPU_batch_uniform_4f(edges, "color", 0.0f, 0.0f, 0.0f, 1.0f);
-        }
-        GPU_batch_draw(edges);
-        break;
-      }
+      case SI_UVDT_WHITE:
       case SI_UVDT_OUTLINE: {
         /* We could modify the vbo's data filling
          * instead of modifying the provoking vert. */
         glProvokingVertex(GL_FIRST_VERTEX_CONVENTION);
 
-        UI_GetThemeColor4fv(TH_WIRE_EDIT, col1);
         UI_GetThemeColor4fv(TH_EDGE_SELECT, col2);
 
         GPU_batch_program_set_builtin(
             edges, (interpedges) ? GPU_SHADER_2D_UV_EDGES_SMOOTH : GPU_SHADER_2D_UV_EDGES);
-        /* Black Outline. */
-        GPU_line_width(3.0f);
-        GPU_batch_uniform_4f(edges, "edgeColor", 0.0f, 0.0f, 0.0f, 1.0f);
-        GPU_batch_uniform_4f(edges, "selectColor", 0.0f, 0.0f, 0.0f, 1.0f);
-        GPU_batch_draw(edges);
+
+        if (sima->dt_uv == SI_UVDT_OUTLINE) {
+          /* Black Outline. */
+          GPU_line_width(3.0f);
+          GPU_batch_uniform_4f(edges, "edgeColor", 0.0f, 0.0f, 0.0f, 1.0f);
+          GPU_batch_uniform_4f(edges, "selectColor", 0.0f, 0.0f, 0.0f, 1.0f);
+          GPU_batch_draw(edges);
+
+          UI_GetThemeColor4fv(TH_WIRE_EDIT, col1);
+        }
+        else if (sima->dt_uv == SI_UVDT_WHITE) {
+          copy_v4_fl4(col1, 1.0f, 1.0f, 1.0f, 1.0f);
+        }
+        else {
+          copy_v4_fl4(col1, 0.0f, 0.0f, 0.0f, 1.0f);
+        }
+
         /* Inner Line. Use depth test to insure selection is drawn on top. */
         GPU_depth_test(true);
         GPU_line_width(1.0f);
diff --git a/source/blender/editors/uvedit/uvedit_intern.h b/source/blender/editors/uvedit/uvedit_intern.h
index 3f544ad90d1..7bc6b048585 100644
--- a/source/blender/editors/uvedit/uvedit_intern.h
+++ b/source/blender/editors/uvedit/uvedit_intern.h
@@ -24,7 +24,6 @@
 #ifndef __UVEDIT_INTERN_H__
 #define __UVEDIT_INTERN_H__
 
-struct BMEditMesh;
 struct BMFace;
 struct BMLoop;
 struct Image;
diff --git a/source/blender/editors/uvedit/uvedit_ops.c b/source/blender/editors/uvedit/uvedit_ops.c
index dcf1d04ffb3..cd7e5f9ba09 100644
--- a/source/blender/editors/uvedit/uvedit_ops.c
+++ b/source/blender/editors/uvedit/uvedit_ops.c
@@ -138,7 +138,7 @@ static int UNUSED_FUNCTION(ED_operator_uvmap_mesh)(bContext *C)
   if (ob && ob->type == OB_MESH) {
     Mesh *me = ob->data;
 
-    if (CustomData_get_layer(&me->fdata, CD_MTFACE) != NULL) {
+    if (CustomData_get_layer(&me->ldata, CD_MLOOPUV) != NULL) {
       return 1;
     }
   }
diff --git a/source/blender/editors/uvedit/uvedit_unwrap_ops.c b/source/blender/editors/uvedit/uvedit_unwrap_ops.c
index a42a6eba3ff..2d8f40ea5af 100644
--- a/source/blender/editors/uvedit/uvedit_unwrap_ops.c
+++ b/source/blender/editors/uvedit/uvedit_unwrap_ops.c
@@ -524,7 +524,7 @@ static ParamHandle *construct_param_handle_subsurfed(Scene *scene,
   smd.levels = smd_real->levels;
   smd.subdivType = smd_real->subdivType;
 
-  initialDerived = CDDM_from_editbmesh(em, false, false);
+  initialDerived = CDDM_from_editbmesh(em, false);
   derivedMesh = subsurf_make_derived_from_derived(
       initialDerived, &smd, scene, NULL, SUBSURF_IN_EDIT_MODE);
 
diff --git a/source/blender/freestyle/FRS_freestyle.h b/source/blender/freestyle/FRS_freestyle.h
index bad279ed9d9..dc8a0c79072 100644
--- a/source/blender/freestyle/FRS_freestyle.h
+++ b/source/blender/freestyle/FRS_freestyle.h
@@ -29,7 +29,6 @@ struct FreestyleConfig;
 struct FreestyleLineStyle;
 struct Material;
 struct Render;
-struct RenderLayer;
 
 struct FreestyleGlobals {
   struct Scene *scene;
diff --git a/source/blender/freestyle/intern/blender_interface/BlenderStrokeRenderer.cpp b/source/blender/freestyle/intern/blender_interface/BlenderStrokeRenderer.cpp
index 984a2d01a68..98da8e8f6d3 100644
--- a/source/blender/freestyle/intern/blender_interface/BlenderStrokeRenderer.cpp
+++ b/source/blender/freestyle/intern/blender_interface/BlenderStrokeRenderer.cpp
@@ -865,7 +865,7 @@ Object *BlenderStrokeRenderer::NewMesh() const
   BLI_snprintf(name, MAX_ID_NAME, "0%08xME", mesh_id);
   ob->data = BKE_mesh_add(freestyle_bmain, name);
 
-  Collection *collection_master = BKE_collection_master(freestyle_scene);
+  Collection *collection_master = freestyle_scene->master_collection;
   BKE_collection_object_add(freestyle_bmain, collection_master, ob);
   DEG_graph_tag_relations_update(freestyle_depsgraph);
 
diff --git a/source/blender/freestyle/intern/scene_graph/FrsMaterial.h b/source/blender/freestyle/intern/scene_graph/FrsMaterial.h
index c5a87c3baaf..80cd783f164 100644
--- a/source/blender/freestyle/intern/scene_graph/FrsMaterial.h
+++ b/source/blender/freestyle/intern/scene_graph/FrsMaterial.h
@@ -36,14 +36,14 @@ class FrsMaterial {
   /*! Default constructor */
   inline FrsMaterial();
 
-  /*! Builds a Material from its line, diffuse, ambiant, specular, emissive
+  /*! Builds a Material from its line, diffuse, ambient, specular, emissive
    *  colors, a shininess coefficient and line color priority.
    *    \param iLine:
    *      A 4 element float-array containing the line color.
    *    \param iDiffuse:
    *      A 4 element float-array containing the diffuse color.
    *    \param iAmbiant:
-   *      A 4 element float-array containing the ambiant color.
+   *      A 4 element float-array containing the ambient color.
    *    \param iSpecular:
    *      A 4 element float-array containing the specular color.
    *    \param iEmission:
@@ -159,31 +159,31 @@ class FrsMaterial {
     return Specular[3];
   }
 
-  /*! Returns the ambiant color as a 4 float array */
+  /*! Returns the ambient color as a 4 float array */
   inline const float *ambient() const
   {
     return Ambient;
   }
 
-  /*! Returns the red component of the ambiant color */
+  /*! Returns the red component of the ambient color */
   inline const float ambientR() const
   {
     return Ambient[0];
   }
 
-  /*! Returns the green component of the ambiant color */
+  /*! Returns the green component of the ambient color */
   inline const float ambientG() const
   {
     return Ambient[1];
   }
 
-  /*! Returns the blue component of the ambiant color */
+  /*! Returns the blue component of the ambient color */
   inline const float ambientB() const
   {
     return Ambient[2];
   }
 
-  /*! Returns the alpha component of the ambiant color */
+  /*! Returns the alpha component of the ambient color */
   inline const float ambientA() const
   {
     return Ambient[3];
@@ -267,7 +267,7 @@ class FrsMaterial {
    */
   inline void setSpecular(const float r, const float g, const float b, const float a);
 
-  /*! Sets the ambiant color.
+  /*! Sets the ambient color.
    *    \param r:
    *      Red component
    *    \param g:
diff --git a/source/blender/freestyle/intern/scene_graph/IndexedFaceSet.h b/source/blender/freestyle/intern/scene_graph/IndexedFaceSet.h
index d70794d9b78..f8dd25913c1 100644
--- a/source/blender/freestyle/intern/scene_graph/IndexedFaceSet.h
+++ b/source/blender/freestyle/intern/scene_graph/IndexedFaceSet.h
@@ -19,7 +19,7 @@
 
 /** \file
  * \ingroup freestyle
- * \brief A Set of indexed faces to represent a surfacic object
+ * \brief A Set of indexed faces to represent a surface object
  */
 
 #include <memory.h>
diff --git a/source/blender/freestyle/intern/scene_graph/NodeTransform.h b/source/blender/freestyle/intern/scene_graph/NodeTransform.h
index ddae714f9d4..1118417657f 100644
--- a/source/blender/freestyle/intern/scene_graph/NodeTransform.h
+++ b/source/blender/freestyle/intern/scene_graph/NodeTransform.h
@@ -48,7 +48,7 @@ class NodeTransform : public NodeGroup {
   /*! multiplies the current matrix by the x, y, z translation matrix. */
   void Translate(real x, real y, real z);
 
-  /*! multiplis the current matrix by a rotation matrix
+  /*! multiplies the current matrix by a rotation matrix
    *    iAngle
    *      The rotation angle
    *    x, y, z
@@ -56,13 +56,13 @@ class NodeTransform : public NodeGroup {
    */
   void Rotate(real iAngle, real x, real y, real z);
 
-  /*! multiplys the current matrix by a scaling matrix.
+  /*! multiplies the current matrix by a scaling matrix.
    *    x, y, z
    *      The scaling coefficients with respect to the x,y,z axis
    */
   void Scale(real x, real y, real z);
 
-  /*! Multiplys the current matrix by iMatrix */
+  /*! Multiplies the current matrix by iMatrix */
   void MultiplyMatrix(const Matrix44r &iMatrix);
 
   /*! Sets the current matrix to iMatrix */
diff --git a/source/blender/freestyle/intern/stroke/Curve.h b/source/blender/freestyle/intern/stroke/Curve.h
index 7eadfa830ce..8a233eef4ab 100644
--- a/source/blender/freestyle/intern/stroke/Curve.h
+++ b/source/blender/freestyle/intern/stroke/Curve.h
@@ -56,7 +56,7 @@ using namespace Geometry;
 /*! Class to represent a point of a curve.
  *  A CurvePoint can be any point of a 1D curve (it doesn't have to be a vertex of the curve).
  *  Any Interface1D is built upon ViewEdges, themselves built upon FEdges. Therefore, a curve is
- * basically a polyline made of a list SVertex. Thus, a CurvePoint is built by lineraly
+ * basically a polyline made of a list SVertex. Thus, a CurvePoint is built by linearly
  * interpolating two SVertex. CurvePoint can be used as virtual points while querying 0D
  * information along a curve at a given resolution.
  */
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencil_util.c b/source/blender/gpencil_modifiers/intern/MOD_gpencil_util.c
index c0ecd3bf0eb..72385fd2f2c 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencil_util.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencil_util.c
@@ -76,9 +76,10 @@ void gpencil_modifier_type_init(GpencilModifierTypeInfo *types[])
 #undef INIT_GP_TYPE
 }
 
-/* verify if valid layer and pass index */
+/* verify if valid layer, material and pass index */
 bool is_stroke_affected_by_modifier(Object *ob,
                                     char *mlayername,
+                                    char *mmaterialname,
                                     int mpassindex,
                                     int gpl_passindex,
                                     int minpoints,
@@ -86,9 +87,11 @@ bool is_stroke_affected_by_modifier(Object *ob,
                                     bGPDstroke *gps,
                                     bool inv1,
                                     bool inv2,
-                                    bool inv3)
+                                    bool inv3,
+                                    bool inv4)
 {
-  MaterialGPencilStyle *gp_style = BKE_material_gpencil_settings_get(ob, gps->mat_nr + 1);
+  Material *ma = BKE_material_gpencil_get(ob, gps->mat_nr + 1);
+  MaterialGPencilStyle *gp_style = ma->gp_style;
 
   /* omit if filter by layer */
   if (mlayername[0] != '\0') {
@@ -103,6 +106,19 @@ bool is_stroke_affected_by_modifier(Object *ob,
       }
     }
   }
+  /* omit if filter by material */
+  if (mmaterialname[0] != '\0') {
+    if (inv4 == false) {
+      if (!STREQ(mmaterialname, ma->id.name + 2)) {
+        return false;
+      }
+    }
+    else {
+      if (STREQ(mmaterialname, ma->id.name + 2)) {
+        return false;
+      }
+    }
+  }
   /* verify layer pass */
   if (gpl_passindex > 0) {
     if (inv3 == false) {
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencil_util.h b/source/blender/gpencil_modifiers/intern/MOD_gpencil_util.h
index c3dbd341e10..7f00e072cda 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencil_util.h
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencil_util.h
@@ -34,6 +34,7 @@ struct bGPDstroke;
 
 bool is_stroke_affected_by_modifier(struct Object *ob,
                                     char *mlayername,
+                                    char *mmaterialname,
                                     int mpassindex,
                                     int gpl_passindex,
                                     int minpoints,
@@ -41,7 +42,8 @@ bool is_stroke_affected_by_modifier(struct Object *ob,
                                     bGPDstroke *gps,
                                     bool inv1,
                                     bool inv2,
-                                    bool inv3);
+                                    bool inv3,
+                                    bool inv4);
 
 float get_modifier_point_weight(struct MDeformVert *dvert, bool inverse, int def_nr);
 
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilarmature.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilarmature.c
index 4297cbb545c..1f2f0554dd5 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilarmature.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilarmature.c
@@ -155,7 +155,11 @@ static bool isDisabled(GpencilModifierData *md, int UNUSED(userRenderParams))
 {
   ArmatureGpencilModifierData *mmd = (ArmatureGpencilModifierData *)md;
 
-  return !mmd->object;
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the armature is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch. */
+  return !mmd->object || mmd->object->type != OB_ARMATURE;
 }
 
 static void updateDepsgraph(GpencilModifierData *md, const ModifierUpdateDepsgraphContext *ctx)
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilarray.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilarray.c
index 0a391a6e63e..c311497ffbb 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilarray.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilarray.c
@@ -171,6 +171,7 @@ static void generate_geometry(GpencilModifierData *md,
      */
     if (is_stroke_affected_by_modifier(ob,
                                        mmd->layername,
+                                       mmd->materialname,
                                        mmd->pass_index,
                                        mmd->layer_pass,
                                        1,
@@ -178,7 +179,8 @@ static void generate_geometry(GpencilModifierData *md,
                                        gps,
                                        mmd->flag & GP_ARRAY_INVERT_LAYER,
                                        mmd->flag & GP_ARRAY_INVERT_PASS,
-                                       mmd->flag & GP_ARRAY_INVERT_LAYERPASS)) {
+                                       mmd->flag & GP_ARRAY_INVERT_LAYERPASS,
+                                       mmd->flag & GP_ARRAY_INVERT_MATERIAL)) {
       valid_strokes[idx] = true;
       num_valid++;
     }
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilcolor.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilcolor.c
index 13329e4176d..cb429b874a2 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilcolor.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilcolor.c
@@ -51,6 +51,7 @@ static void initData(GpencilModifierData *md)
   gpmd->pass_index = 0;
   ARRAY_SET_ITEMS(gpmd->hsv, 0.5f, 1.0f, 1.0f);
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   gpmd->flag |= GP_COLOR_CREATE_COLORS;
   gpmd->modify_color = GP_MODIFY_COLOR_BOTH;
 }
@@ -74,6 +75,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       1,
@@ -81,7 +83,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_COLOR_INVERT_LAYER,
                                       mmd->flag & GP_COLOR_INVERT_PASS,
-                                      mmd->flag & GP_COLOR_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_COLOR_INVERT_LAYERPASS,
+                                      mmd->flag & GP_COLOR_INVERT_MATERIAL)) {
     return;
   }
 
@@ -116,7 +119,7 @@ static void bakeModifier(Main *bmain, Depsgraph *depsgraph, GpencilModifierData
     for (bGPDframe *gpf = gpl->frames.first; gpf; gpf = gpf->next) {
       for (bGPDstroke *gps = gpf->strokes.first; gps; gps = gps->next) {
 
-        Material *mat = give_current_material(ob, gps->mat_nr + 1);
+        Material *mat = BKE_material_gpencil_get(ob, gps->mat_nr + 1);
         if (mat == NULL) {
           continue;
         }
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilhook.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilhook.c
index a3dbcdf23de..810a2ba7e25 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilhook.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilhook.c
@@ -77,6 +77,7 @@ static void initData(GpencilModifierData *md)
   HookGpencilModifierData *gpmd = (HookGpencilModifierData *)md;
   gpmd->pass_index = 0;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   gpmd->vgname[0] = '\0';
   gpmd->object = NULL;
   gpmd->force = 0.5f;
@@ -206,6 +207,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       1,
@@ -213,7 +215,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_HOOK_INVERT_LAYER,
                                       mmd->flag & GP_HOOK_INVERT_PASS,
-                                      mmd->flag & GP_HOOK_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_HOOK_INVERT_LAYERPASS,
+                                      mmd->flag & GP_HOOK_INVERT_MATERIAL)) {
     return;
   }
 
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencillattice.c b/source/blender/gpencil_modifiers/intern/MOD_gpencillattice.c
index aa5015bac35..6b74f96ce31 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencillattice.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencillattice.c
@@ -55,6 +55,7 @@ static void initData(GpencilModifierData *md)
   LatticeGpencilModifierData *gpmd = (LatticeGpencilModifierData *)md;
   gpmd->pass_index = 0;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   gpmd->vgname[0] = '\0';
   gpmd->object = NULL;
   gpmd->cache_data = NULL;
@@ -78,6 +79,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       1,
@@ -85,7 +87,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_LATTICE_INVERT_LAYER,
                                       mmd->flag & GP_LATTICE_INVERT_PASS,
-                                      mmd->flag & GP_LATTICE_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_LATTICE_INVERT_LAYERPASS,
+                                      mmd->flag & GP_LATTICE_INVERT_MATERIAL)) {
     return;
   }
 
@@ -166,7 +169,12 @@ static bool isDisabled(GpencilModifierData *md, int UNUSED(userRenderParams))
 {
   LatticeGpencilModifierData *mmd = (LatticeGpencilModifierData *)md;
 
-  return !mmd->object;
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the lattice is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  return !mmd->object || mmd->object->type != OB_LATTICE;
 }
 
 static void updateDepsgraph(GpencilModifierData *md, const ModifierUpdateDepsgraphContext *ctx)
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilmirror.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilmirror.c
index 058f60a69ac..112dfc2e9dc 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilmirror.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilmirror.c
@@ -57,6 +57,7 @@ static void initData(GpencilModifierData *md)
   MirrorGpencilModifierData *gpmd = (MirrorGpencilModifierData *)md;
   gpmd->pass_index = 0;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   gpmd->object = NULL;
   gpmd->flag |= GP_MIRROR_AXIS_X;
 }
@@ -134,6 +135,7 @@ static void generateStrokes(GpencilModifierData *md,
       for (i = 0, gps = gpf->strokes.first; i < tot_strokes; i++, gps = gps->next) {
         if (is_stroke_affected_by_modifier(ob,
                                            mmd->layername,
+                                           mmd->materialname,
                                            mmd->pass_index,
                                            mmd->layer_pass,
                                            1,
@@ -141,7 +143,8 @@ static void generateStrokes(GpencilModifierData *md,
                                            gps,
                                            mmd->flag & GP_MIRROR_INVERT_LAYER,
                                            mmd->flag & GP_MIRROR_INVERT_PASS,
-                                           mmd->flag & GP_MIRROR_INVERT_LAYERPASS)) {
+                                           mmd->flag & GP_MIRROR_INVERT_LAYERPASS,
+                                           mmd->flag & GP_MIRROR_INVERT_MATERIAL)) {
           gps_new = BKE_gpencil_stroke_duplicate(gps);
           update_position(ob, mmd, gps_new, xi);
           BLI_addtail(&gpf->strokes, gps_new);
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilnoise.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilnoise.c
index 2c4d99adeee..b286b55829e 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilnoise.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilnoise.c
@@ -55,6 +55,7 @@ static void initData(GpencilModifierData *md)
   gpmd->flag |= GP_NOISE_USE_RANDOM;
   gpmd->factor = 0.5f;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   gpmd->vgname[0] = '\0';
   gpmd->step = 1;
   gpmd->seed = 0;
@@ -92,6 +93,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       1,
@@ -99,7 +101,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_NOISE_INVERT_LAYER,
                                       mmd->flag & GP_NOISE_INVERT_PASS,
-                                      mmd->flag & GP_NOISE_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_NOISE_INVERT_LAYERPASS,
+                                      mmd->flag & GP_NOISE_INVERT_MATERIAL)) {
     return;
   }
 
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpenciloffset.c b/source/blender/gpencil_modifiers/intern/MOD_gpenciloffset.c
index 399fc407455..08f67eedc86 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpenciloffset.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpenciloffset.c
@@ -48,6 +48,7 @@ static void initData(GpencilModifierData *md)
   OffsetGpencilModifierData *gpmd = (OffsetGpencilModifierData *)md;
   gpmd->pass_index = 0;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   gpmd->vgname[0] = '\0';
   ARRAY_SET_ITEMS(gpmd->loc, 0.0f, 0.0f, 0.0f);
   ARRAY_SET_ITEMS(gpmd->rot, 0.0f, 0.0f, 0.0f);
@@ -75,6 +76,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       1,
@@ -82,7 +84,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_OFFSET_INVERT_LAYER,
                                       mmd->flag & GP_OFFSET_INVERT_PASS,
-                                      mmd->flag & GP_OFFSET_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_OFFSET_INVERT_LAYERPASS,
+                                      mmd->flag & GP_OFFSET_INVERT_MATERIAL)) {
     return;
   }
 
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilopacity.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilopacity.c
index 92c25f955fb..22610771045 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilopacity.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilopacity.c
@@ -52,6 +52,7 @@ static void initData(GpencilModifierData *md)
   gpmd->pass_index = 0;
   gpmd->factor = 1.0f;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   gpmd->vgname[0] = '\0';
   gpmd->flag |= GP_OPACITY_CREATE_COLORS;
   gpmd->modify_color = GP_MODIFY_COLOR_BOTH;
@@ -75,6 +76,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       1,
@@ -82,7 +84,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_OPACITY_INVERT_LAYER,
                                       mmd->flag & GP_OPACITY_INVERT_PASS,
-                                      mmd->flag & GP_OPACITY_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_OPACITY_INVERT_LAYERPASS,
+                                      mmd->flag & GP_OPACITY_INVERT_MATERIAL)) {
     return;
   }
 
@@ -109,22 +112,7 @@ static void deformStroke(GpencilModifierData *md,
     if (mmd->factor > 1.0f) {
       for (int i = 0; i < gps->totpoints; i++) {
         bGPDspoint *pt = &gps->points[i];
-        MDeformVert *dvert = gps->dvert != NULL ? &gps->dvert[i] : NULL;
-
-        /* verify vertex group */
-        float weight = get_modifier_point_weight(
-            dvert, (mmd->flag & GP_OPACITY_INVERT_VGROUP) != 0, def_nr);
-        if (weight < 0.0f) {
-          continue;
-        }
-        if (def_nr < 0) {
-          pt->strength += mmd->factor - 1.0f;
-        }
-        else {
-          CLAMP(weight, 0.0f, 1.0f);
-          weight = 1.0f - weight;
-          pt->strength += (mmd->factor * weight) - 1.0f;
-        }
+        pt->strength += mmd->factor - 1.0f;
         CLAMP(pt->strength, 0.0f, 1.0f);
       }
     }
@@ -145,14 +133,12 @@ static void deformStroke(GpencilModifierData *md,
         pt->strength += mmd->factor - 1.0f;
       }
       else {
-        CLAMP(weight, 0.0f, 1.0f);
-        weight = 1.0f - weight;
         /* High factor values, change weight too. */
         if ((mmd->factor > 1.0f) && (weight < 1.0f)) {
           weight += mmd->factor - 1.0f;
           CLAMP(weight, 0.0f, 1.0f);
         }
-        pt->strength += (mmd->factor * weight) - 1.0f;
+        pt->strength += (mmd->factor - 1) * weight;
       }
       CLAMP(pt->strength, 0.0f, 1.0f);
     }
@@ -169,7 +155,7 @@ static void bakeModifier(Main *bmain, Depsgraph *depsgraph, GpencilModifierData
     for (bGPDframe *gpf = gpl->frames.first; gpf; gpf = gpf->next) {
       for (bGPDstroke *gps = gpf->strokes.first; gps; gps = gps->next) {
 
-        Material *mat = give_current_material(ob, gps->mat_nr + 1);
+        Material *mat = BKE_material_gpencil_get(ob, gps->mat_nr + 1);
         if (mat == NULL) {
           continue;
         }
@@ -184,8 +170,10 @@ static void bakeModifier(Main *bmain, Depsgraph *depsgraph, GpencilModifierData
 
         deformStroke(md, depsgraph, ob, gpl, gpf, gps);
 
-        gpencil_apply_modifier_material(
-            bmain, ob, mat, gh_color, gps, (bool)(mmd->flag & GP_OPACITY_CREATE_COLORS));
+        if (mmd->opacity_mode == GP_OPACITY_MODE_MATERIAL) {
+          gpencil_apply_modifier_material(
+              bmain, ob, mat, gh_color, gps, (bool)(mmd->flag & GP_OPACITY_CREATE_COLORS));
+        }
       }
     }
   }
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilsimplify.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilsimplify.c
index 2168e7f07ec..25a56c4385c 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilsimplify.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilsimplify.c
@@ -48,6 +48,7 @@ static void initData(GpencilModifierData *md)
   gpmd->length = 0.1f;
   gpmd->distance = 0.1f;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
 }
 
 static void copyData(const GpencilModifierData *md, GpencilModifierData *target)
@@ -66,6 +67,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       mmd->mode == GP_SIMPLIFY_SAMPLE ? 3 : 4,
@@ -73,7 +75,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_SIMPLIFY_INVERT_LAYER,
                                       mmd->flag & GP_SIMPLIFY_INVERT_PASS,
-                                      mmd->flag & GP_SIMPLIFY_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_SIMPLIFY_INVERT_LAYERPASS,
+                                      mmd->flag & GP_SIMPLIFY_INVERT_MATERIAL)) {
     return;
   }
 
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilsmooth.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilsmooth.c
index 7c639bd8af0..5ec7fe4ff18 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilsmooth.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilsmooth.c
@@ -46,6 +46,7 @@ static void initData(GpencilModifierData *md)
   gpmd->flag |= GP_SMOOTH_MOD_LOCATION;
   gpmd->factor = 0.5f;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   gpmd->vgname[0] = '\0';
   gpmd->step = 1;
 }
@@ -68,6 +69,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       3,
@@ -75,7 +77,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_SMOOTH_INVERT_LAYER,
                                       mmd->flag & GP_SMOOTH_INVERT_PASS,
-                                      mmd->flag & GP_SMOOTH_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_SMOOTH_INVERT_LAYERPASS,
+                                      mmd->flag & GP_SMOOTH_INVERT_MATERIAL)) {
     return;
   }
 
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilsubdiv.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilsubdiv.c
index c8962cd58c9..741555722b5 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilsubdiv.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilsubdiv.c
@@ -47,6 +47,7 @@ static void initData(GpencilModifierData *md)
   gpmd->pass_index = 0;
   gpmd->level = 1;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
 }
 
 static void copyData(const GpencilModifierData *md, GpencilModifierData *target)
@@ -66,6 +67,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       3,
@@ -73,7 +75,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_SUBDIV_INVERT_LAYER,
                                       mmd->flag & GP_SUBDIV_INVERT_PASS,
-                                      mmd->flag & GP_SUBDIV_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_SUBDIV_INVERT_LAYERPASS,
+                                      mmd->flag & GP_SUBDIV_INVERT_MATERIAL)) {
     return;
   }
 
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpencilthick.c b/source/blender/gpencil_modifiers/intern/MOD_gpencilthick.c
index 53abfdf177c..cc38df141d1 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpencilthick.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpencilthick.c
@@ -47,6 +47,7 @@ static void initData(GpencilModifierData *md)
   gpmd->pass_index = 0;
   gpmd->thickness = 2;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   gpmd->vgname[0] = '\0';
   gpmd->curve_thickness = BKE_curvemapping_add(1, 0.0f, 0.0f, 1.0f, 1.0f);
   if (gpmd->curve_thickness) {
@@ -91,6 +92,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       1,
@@ -98,20 +100,46 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_THICK_INVERT_LAYER,
                                       mmd->flag & GP_THICK_INVERT_PASS,
-                                      mmd->flag & GP_THICK_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_THICK_INVERT_LAYERPASS,
+                                      mmd->flag & GP_THICK_INVERT_MATERIAL)) {
     return;
   }
 
-  /* if normalize, set stroke thickness */
+  /* Check to see if we normalize the whole stroke or only certain points along it. */
+  bool gps_has_affected_points = false;
+  bool gps_has_unaffected_points = false;
+
   if (mmd->flag & GP_THICK_NORMALIZE) {
+    for (int i = 0; i < gps->totpoints; i++) {
+      MDeformVert *dvert = gps->dvert != NULL ? &gps->dvert[i] : NULL;
+      const float weight = get_modifier_point_weight(
+          dvert, (mmd->flag & GP_THICK_INVERT_VGROUP) != 0, def_nr);
+      if (weight < 0.0f) {
+        gps_has_unaffected_points = true;
+      }
+      else {
+        gps_has_affected_points = true;
+      }
+
+      /* If both checks are true, we have what we need so we can stop looking. */
+      if (gps_has_affected_points && gps_has_unaffected_points) {
+        break;
+      }
+    }
+  }
+
+  /* If we are normalizing and all points of the stroke are affected, it's safe to reset thickness
+   */
+  if (mmd->flag & GP_THICK_NORMALIZE && gps_has_affected_points && !gps_has_unaffected_points) {
     gps->thickness = mmd->thickness;
   }
+  /* Without this check, modifier alters the thickness of strokes which have no points in scope */
 
   for (int i = 0; i < gps->totpoints; i++) {
     bGPDspoint *pt = &gps->points[i];
     MDeformVert *dvert = gps->dvert != NULL ? &gps->dvert[i] : NULL;
     float curvef = 1.0f;
-    /* verify vertex group */
+    /* Verify point is part of vertex group. */
     const float weight = get_modifier_point_weight(
         dvert, (mmd->flag & GP_THICK_INVERT_VGROUP) != 0, def_nr);
     if (weight < 0.0f) {
@@ -127,11 +155,21 @@ static void deformStroke(GpencilModifierData *md,
     float new_pressure = weight * curvef;
 
     if (mmd->flag & GP_THICK_NORMALIZE) {
-      pt->pressure = 1.0f * new_pressure;
+      if (gps_has_unaffected_points) {
+        /* Clamp value for very weird situations when stroke thickness can be zero. */
+        CLAMP_MIN(gps->thickness, 1);
+        /* Calculate pressure value to match the width of strokes with reset thickness and 1.0
+         * pressure. */
+        pt->pressure = (float)mmd->thickness / (float)gps->thickness;
+      }
+      else {
+        /* Reset point pressure values so only stroke thickness counts. */
+        pt->pressure = 1.0f;
+      }
     }
     else {
       if ((mmd->flag & GP_THICK_CUSTOM_CURVE) && (mmd->curve_thickness)) {
-        /* normalize value to evaluate curve */
+        /* Normalize value to evaluate curve. */
         float value = (float)i / (gps->totpoints - 1);
         curvef = BKE_curvemapping_evaluateF(mmd->curve_thickness, 0, value);
       }
diff --git a/source/blender/gpencil_modifiers/intern/MOD_gpenciltint.c b/source/blender/gpencil_modifiers/intern/MOD_gpenciltint.c
index d70d975c3a7..f6ddcf89bcf 100644
--- a/source/blender/gpencil_modifiers/intern/MOD_gpenciltint.c
+++ b/source/blender/gpencil_modifiers/intern/MOD_gpenciltint.c
@@ -50,6 +50,7 @@ static void initData(GpencilModifierData *md)
   gpmd->pass_index = 0;
   gpmd->factor = 0.5f;
   gpmd->layername[0] = '\0';
+  gpmd->materialname[0] = '\0';
   ARRAY_SET_ITEMS(gpmd->rgb, 1.0f, 1.0f, 1.0f);
   gpmd->flag |= GP_TINT_CREATE_COLORS;
   gpmd->modify_color = GP_MODIFY_COLOR_BOTH;
@@ -72,6 +73,7 @@ static void deformStroke(GpencilModifierData *md,
 
   if (!is_stroke_affected_by_modifier(ob,
                                       mmd->layername,
+                                      mmd->materialname,
                                       mmd->pass_index,
                                       mmd->layer_pass,
                                       1,
@@ -79,7 +81,8 @@ static void deformStroke(GpencilModifierData *md,
                                       gps,
                                       mmd->flag & GP_TINT_INVERT_LAYER,
                                       mmd->flag & GP_TINT_INVERT_PASS,
-                                      mmd->flag & GP_TINT_INVERT_LAYERPASS)) {
+                                      mmd->flag & GP_TINT_INVERT_LAYERPASS,
+                                      mmd->flag & GP_TINT_INVERT_MATERIAL)) {
     return;
   }
 
@@ -122,7 +125,7 @@ static void bakeModifier(Main *bmain, Depsgraph *depsgraph, GpencilModifierData
     for (bGPDframe *gpf = gpl->frames.first; gpf; gpf = gpf->next) {
       for (bGPDstroke *gps = gpf->strokes.first; gps; gps = gps->next) {
 
-        Material *mat = give_current_material(ob, gps->mat_nr + 1);
+        Material *mat = BKE_material_gpencil_get(ob, gps->mat_nr + 1);
         if (mat == NULL) {
           continue;
         }
diff --git a/source/blender/gpu/CMakeLists.txt b/source/blender/gpu/CMakeLists.txt
index fb7d3c1ace8..f11dcc9bcf0 100644
--- a/source/blender/gpu/CMakeLists.txt
+++ b/source/blender/gpu/CMakeLists.txt
@@ -116,6 +116,7 @@ set(SRC
   intern/gpu_batch_private.h
   intern/gpu_codegen.h
   intern/gpu_context_private.h
+  intern/gpu_material_library.h
   intern/gpu_matrix_private.h
   intern/gpu_primitive_private.h
   intern/gpu_private.h
@@ -231,7 +232,94 @@ data_to_c_simple(shaders/gpu_shader_keyframe_diamond_vert.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_keyframe_diamond_frag.glsl SRC)
 
 data_to_c_simple(shaders/gpu_shader_geometry.glsl SRC)
-data_to_c_simple(shaders/gpu_shader_material.glsl SRC)
+
+data_to_c_simple(shaders/material/gpu_shader_material_add_shader.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_ambient_occlusion.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_anisotropic.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_attribute.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_background.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_bevel.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_blackbody.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_bright_contrast.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_bump.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_camera.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_cell_noise.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_clamp.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_color_ramp.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_color_util.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_combine_hsv.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_combine_rgb.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_combine_xyz.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_diffuse.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_displacement.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_eevee_specular.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_emission.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_fractal_noise.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_fresnel.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_gamma.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_geometry.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_glass.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_glossy.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_hair_info.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_hash.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_holdout.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_hue_sat_val.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_invert.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_layer_weight.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_light_falloff.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_light_path.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_mapping.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_map_range.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_math.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_math_util.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_mix_rgb.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_mix_shader.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_noise.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_normal.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_normal_map.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_object_info.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_output_material.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_output_world.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_particle_info.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_principled.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_refraction.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_rgb_curves.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_rgb_to_bw.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_separate_hsv.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_separate_rgb.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_separate_xyz.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_set.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_shader_to_rgba.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_squeeze.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_subsurface_scattering.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tangent.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_brick.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_checker.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_environment.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_gradient.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_image.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_magic.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_musgrave.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_noise.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_sky.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_texture_coordinates.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_voronoi.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_wave.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_tex_white_noise.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_toon.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_translucent.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_transparent.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_uv_map.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_vector_curves.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_vector_displacement.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_vector_math.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_velvet.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_volume_absorption.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_volume_info.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_volume_principled.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_volume_scatter.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_wireframe.glsl SRC)
+data_to_c_simple(shaders/material/gpu_shader_material_world_normals.glsl SRC)
 
 data_to_c_simple(shaders/gpu_shader_gpencil_stroke_vert.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_gpencil_stroke_frag.glsl SRC)
diff --git a/source/blender/gpu/GPU_batch_presets.h b/source/blender/gpu/GPU_batch_presets.h
index 4b95783e6fa..55d2e461b30 100644
--- a/source/blender/gpu/GPU_batch_presets.h
+++ b/source/blender/gpu/GPU_batch_presets.h
@@ -27,9 +27,6 @@
 #ifndef __GPU_BATCH_PRESETS_H__
 #define __GPU_BATCH_PRESETS_H__
 
-struct GPUVertFormat;
-struct rctf;
-
 #include "BLI_compiler_attrs.h"
 #include "BLI_sys_types.h"
 
diff --git a/source/blender/gpu/GPU_draw.h b/source/blender/gpu/GPU_draw.h
index 507baa9531e..7814c723fec 100644
--- a/source/blender/gpu/GPU_draw.h
+++ b/source/blender/gpu/GPU_draw.h
@@ -28,17 +28,11 @@
 extern "C" {
 #endif
 
-struct DupliObject;
 struct ImBuf;
 struct Image;
 struct ImageUser;
 struct Main;
-struct Object;
-struct RegionView3D;
-struct Scene;
 struct SmokeModifierData;
-struct View3D;
-struct ViewLayer;
 
 #include "DNA_object_enums.h"
 
diff --git a/source/blender/gpu/GPU_material.h b/source/blender/gpu/GPU_material.h
index dd5292d9c58..4dc0019978a 100644
--- a/source/blender/gpu/GPU_material.h
+++ b/source/blender/gpu/GPU_material.h
@@ -45,10 +45,7 @@ struct ImageUser;
 struct ListBase;
 struct Main;
 struct Material;
-struct Object;
-struct PreviewImage;
 struct Scene;
-struct World;
 struct bNode;
 struct bNodeTree;
 
@@ -92,7 +89,7 @@ typedef enum eGPUBuiltin {
   GPU_INVERSE_OBJECT_MATRIX = (1 << 3),
   GPU_VIEW_POSITION = (1 << 4),
   GPU_VIEW_NORMAL = (1 << 5),
-  GPU_OBCOLOR = (1 << 6),
+  GPU_OBJECT_COLOR = (1 << 6),
   GPU_AUTO_BUMPSCALE = (1 << 7),
   GPU_CAMERA_TEXCO_FACTORS = (1 << 8),
   GPU_PARTICLE_SCALAR_PROPS = (1 << 9),
diff --git a/source/blender/gpu/GPU_shader_interface.h b/source/blender/gpu/GPU_shader_interface.h
index d3a9a18b392..ec97e1b085e 100644
--- a/source/blender/gpu/GPU_shader_interface.h
+++ b/source/blender/gpu/GPU_shader_interface.h
@@ -48,9 +48,10 @@ typedef enum {
   GPU_UNIFORM_ORCO,       /* vec3 OrcoTexCoFactors[] */
   GPU_UNIFORM_CLIPPLANES, /* vec4 WorldClipPlanes[] */
 
-  GPU_UNIFORM_COLOR,       /* vec4 color */
-  GPU_UNIFORM_CALLID,      /* int callId */
-  GPU_UNIFORM_OBJECT_INFO, /* vec3 objectInfo */
+  GPU_UNIFORM_COLOR,        /* vec4 color */
+  GPU_UNIFORM_CALLID,       /* int callId */
+  GPU_UNIFORM_OBJECT_INFO,  /* vec3 objectInfo */
+  GPU_UNIFORM_OBJECT_COLOR, /* vec4 objectColor */
 
   GPU_UNIFORM_CUSTOM, /* custom uniform, not one of the above built-ins */
 
diff --git a/source/blender/gpu/GPU_texture.h b/source/blender/gpu/GPU_texture.h
index 2b61d99e852..a8e8ca72023 100644
--- a/source/blender/gpu/GPU_texture.h
+++ b/source/blender/gpu/GPU_texture.h
@@ -36,7 +36,6 @@ struct ImageUser;
 struct MovieClip;
 struct MovieClipUser;
 struct PreviewImage;
-struct rcti;
 
 struct GPUFrameBuffer;
 typedef struct GPUTexture GPUTexture;
diff --git a/source/blender/gpu/intern/gpu_codegen.c b/source/blender/gpu/intern/gpu_codegen.c
index 0e15fdd000b..d655a43561e 100644
--- a/source/blender/gpu/intern/gpu_codegen.c
+++ b/source/blender/gpu/intern/gpu_codegen.c
@@ -51,15 +51,11 @@
 #include "BLI_sys_types.h" /* for intptr_t support */
 
 #include "gpu_codegen.h"
+#include "gpu_material_library.h"
 
 #include <string.h>
 #include <stdarg.h>
 
-extern char datatoc_gpu_shader_material_glsl[];
-extern char datatoc_gpu_shader_geometry_glsl[];
-
-static char *glsl_material_library = NULL;
-
 /* -------------------- GPUPass Cache ------------------ */
 /**
  * Internal shader cache: This prevent the shader recompilation / stall when
@@ -147,6 +143,7 @@ typedef struct GPUFunction {
   eGPUType paramtype[MAX_PARAMETER];
   GPUFunctionQual paramqual[MAX_PARAMETER];
   int totparam;
+  GPUMaterialLibrary *library;
 } GPUFunction;
 
 /* Indices match the eGPUType enum */
@@ -230,15 +227,17 @@ static char *gpu_str_skip_token(char *str, char *token, int max)
   return str;
 }
 
-static void gpu_parse_functions_string(GHash *hash, char *code)
+static void gpu_parse_material_library(GHash *hash, GPUMaterialLibrary *library)
 {
   GPUFunction *function;
   eGPUType type;
   GPUFunctionQual qual;
   int i;
+  char *code = library->code;
 
   while ((code = strstr(code, "void "))) {
     function = MEM_callocN(sizeof(GPUFunction), "GPUFunction");
+    function->library = library;
 
     code = gpu_str_skip_token(code, NULL, 0);
     code = gpu_str_skip_token(code, function->name, MAX_FUNCTION_NAME);
@@ -367,11 +366,6 @@ static char *gpu_generate_function_prototyps(GHash *hash)
 
 static GPUFunction *gpu_lookup_function(const char *name)
 {
-  if (!FUNCTION_HASH) {
-    FUNCTION_HASH = BLI_ghash_str_new("GPU_lookup_function gh");
-    gpu_parse_functions_string(FUNCTION_HASH, glsl_material_library);
-  }
-
   return BLI_ghash_lookup(FUNCTION_HASH, (const void *)name);
 }
 
@@ -395,11 +389,6 @@ void gpu_codegen_exit(void)
 
   GPU_shader_free_builtin_shaders();
 
-  if (glsl_material_library) {
-    MEM_freeN(glsl_material_library);
-    glsl_material_library = NULL;
-  }
-
 #if 0
   if (FUNCTION_PROTOTYPES) {
     MEM_freeN(FUNCTION_PROTOTYPES);
@@ -425,7 +414,7 @@ static void codegen_convert_datatype(DynStr *ds, int from, int to, const char *t
   }
   else if (to == GPU_FLOAT) {
     if (from == GPU_VEC4) {
-      BLI_dynstr_appendf(ds, "convert_rgba_to_float(%s)", name);
+      BLI_dynstr_appendf(ds, "dot(%s.rgb, vec3(0.2126, 0.7152, 0.0722))", name);
     }
     else if (from == GPU_VEC3) {
       BLI_dynstr_appendf(ds, "(%s.r + %s.g + %s.b) / 3.0", name, name, name);
@@ -542,8 +531,8 @@ const char *GPU_builtin_name(eGPUBuiltin builtin)
   else if (builtin == GPU_VIEW_NORMAL) {
     return "varnormal";
   }
-  else if (builtin == GPU_OBCOLOR) {
-    return "unfobcolor";
+  else if (builtin == GPU_OBJECT_COLOR) {
+    return "unfobjectcolor";
   }
   else if (builtin == GPU_AUTO_BUMPSCALE) {
     return "unfobautobumpscale";
@@ -1381,20 +1370,15 @@ static char *code_generate_geometry(ListBase *nodes, const char *geom_code, cons
 
 void GPU_code_generate_glsl_lib(void)
 {
-  DynStr *ds;
-
-  /* only initialize the library once */
-  if (glsl_material_library) {
+  /* Only parse GLSL shader files once. */
+  if (FUNCTION_HASH) {
     return;
   }
 
-  ds = BLI_dynstr_new();
-
-  BLI_dynstr_append(ds, datatoc_gpu_shader_material_glsl);
-
-  glsl_material_library = BLI_dynstr_get_cstring(ds);
-
-  BLI_dynstr_free(ds);
+  FUNCTION_HASH = BLI_ghash_str_new("GPU_lookup_function gh");
+  for (int i = 0; gpu_material_libraries[i]; i++) {
+    gpu_parse_material_library(FUNCTION_HASH, gpu_material_libraries[i]);
+  }
 }
 
 /* GPU pass binding/unbinding */
@@ -1796,6 +1780,22 @@ GPUNodeLink *GPU_builtin(eGPUBuiltin builtin)
   return link;
 }
 
+static void gpu_material_use_library_with_dependencies(GSet *used_libraries,
+                                                       GPUMaterialLibrary *library)
+{
+  if (BLI_gset_add(used_libraries, library->code)) {
+    for (int i = 0; library->dependencies[i]; i++) {
+      gpu_material_use_library_with_dependencies(used_libraries, library->dependencies[i]);
+    }
+  }
+}
+
+static void gpu_material_use_library(GPUMaterial *material, GPUMaterialLibrary *library)
+{
+  GSet *used_libraries = gpu_material_used_libraries(material);
+  gpu_material_use_library_with_dependencies(used_libraries, library);
+}
+
 bool GPU_link(GPUMaterial *mat, const char *name, ...)
 {
   GPUNode *node;
@@ -1810,6 +1810,8 @@ bool GPU_link(GPUMaterial *mat, const char *name, ...)
     return false;
   }
 
+  gpu_material_use_library(mat, function->library);
+
   node = GPU_node_begin(name);
 
   va_start(params, name);
@@ -1849,6 +1851,8 @@ bool GPU_stack_link(GPUMaterial *material,
     return false;
   }
 
+  gpu_material_use_library(material, function->library);
+
   node = GPU_node_begin(name);
   totin = 0;
   totout = 0;
@@ -1962,6 +1966,34 @@ static bool gpu_pass_is_valid(GPUPass *pass)
   return (pass->compiled == false || pass->shader != NULL);
 }
 
+static char *code_generate_material_library(GPUMaterial *material, const char *frag_lib)
+{
+  DynStr *ds = BLI_dynstr_new();
+
+  if (frag_lib) {
+    BLI_dynstr_append(ds, frag_lib);
+  }
+
+  GSet *used_libraries = gpu_material_used_libraries(material);
+
+  /* Always include those because they may be needed by the execution function. */
+  gpu_material_use_library_with_dependencies(used_libraries,
+                                             &gpu_shader_material_world_normals_library);
+
+  /* Add library code in order, for dependencies. */
+  for (int i = 0; gpu_material_libraries[i]; i++) {
+    GPUMaterialLibrary *library = gpu_material_libraries[i];
+    if (BLI_gset_haskey(used_libraries, library->code)) {
+      BLI_dynstr_append(ds, library->code);
+    }
+  }
+
+  char *result = BLI_dynstr_get_cstring(ds);
+  BLI_dynstr_free(ds);
+
+  return result;
+}
+
 GPUPass *GPU_generate_pass(GPUMaterial *material,
                            GPUNodeLink *frag_outlink,
                            struct GPUVertAttrLayers *attrs,
@@ -2000,7 +2032,7 @@ GPUPass *GPU_generate_pass(GPUMaterial *material,
 
   /* Either the shader is not compiled or there is a hash collision...
    * continue generating the shader strings. */
-  char *tmp = BLI_strdupcat(frag_lib, glsl_material_library);
+  char *tmp = code_generate_material_library(material, frag_lib);
 
   geometrycode = code_generate_geometry(nodes, geom_code, defines);
   vertexcode = code_generate_vertex(nodes, vert_code, (geometrycode != NULL));
diff --git a/source/blender/gpu/intern/gpu_codegen.h b/source/blender/gpu/intern/gpu_codegen.h
index ae34e21b9cd..4e09f16ebf8 100644
--- a/source/blender/gpu/intern/gpu_codegen.h
+++ b/source/blender/gpu/intern/gpu_codegen.h
@@ -33,8 +33,8 @@ struct GPUNode;
 struct GPUOutput;
 struct GPUShader;
 struct GPUVertAttrLayers;
+struct GSet;
 struct ListBase;
-struct PreviewImage;
 
 /* Pass Generation
  * - Takes a list of nodes and a desired output, and makes a pass. This
@@ -208,4 +208,6 @@ struct GPUTexture **gpu_material_ramp_texture_row_set(GPUMaterial *mat,
                                                       float *pixels,
                                                       float *row);
 
+struct GSet *gpu_material_used_libraries(struct GPUMaterial *material);
+
 #endif
diff --git a/source/blender/gpu/intern/gpu_draw.c b/source/blender/gpu/intern/gpu_draw.c
index be3655648f5..859b1e4a10d 100644
--- a/source/blender/gpu/intern/gpu_draw.c
+++ b/source/blender/gpu/intern/gpu_draw.c
@@ -348,7 +348,7 @@ static void gpu_texture_update_scaled(
   }
 
   /* Scale pixels. */
-  ImBuf *ibuf = IMB_allocFromBuffer((uint *)rect, rect_float, w, h);
+  ImBuf *ibuf = IMB_allocFromBuffer((uint *)rect, rect_float, w, h, 4);
   IMB_scaleImBuf(ibuf, sub_w, sub_h);
 
   if (ibuf->rect_float) {
@@ -648,13 +648,13 @@ void GPU_create_gl_tex(uint *bind,
     recth = smaller_power_of_2_limit(recth);
 
     if (frect) {
-      ibuf = IMB_allocFromBuffer(NULL, frect, tpx, tpy);
+      ibuf = IMB_allocFromBuffer(NULL, frect, tpx, tpy, 4);
       IMB_scaleImBuf(ibuf, rectw, recth);
 
       frect = ibuf->rect_float;
     }
     else {
-      ibuf = IMB_allocFromBuffer(rect, NULL, tpx, tpy);
+      ibuf = IMB_allocFromBuffer(rect, NULL, tpx, tpy, 4);
       IMB_scaleImBuf(ibuf, rectw, recth);
 
       rect = ibuf->rect;
diff --git a/source/blender/gpu/intern/gpu_element.c b/source/blender/gpu/intern/gpu_element.c
index 6c9331b4903..3f3f246d6d9 100644
--- a/source/blender/gpu/intern/gpu_element.c
+++ b/source/blender/gpu/intern/gpu_element.c
@@ -267,9 +267,9 @@ static uint index_range(const uint values[], uint value_len, uint *min_out, uint
     *max_out = 0;
     return 0;
   }
-  uint min_value = values[0];
-  uint max_value = values[0];
-  for (uint i = 1; i < value_len; ++i) {
+  uint min_value = RESTART_INDEX;
+  uint max_value = 0;
+  for (uint i = 0; i < value_len; ++i) {
     const uint value = values[i];
     if (value == RESTART_INDEX) {
       continue;
@@ -281,9 +281,16 @@ static uint index_range(const uint values[], uint value_len, uint *min_out, uint
       max_value = value;
     }
   }
-  *min_out = min_value;
-  *max_out = max_value;
-  return max_value - min_value;
+  if (min_value == RESTART_INDEX) {
+    *min_out = 0;
+    *max_out = 0;
+    return 0;
+  }
+  else {
+    *min_out = min_value;
+    *max_out = max_value;
+    return max_value - min_value;
+  }
 }
 
 static void squeeze_indices_short(GPUIndexBufBuilder *builder,
diff --git a/source/blender/gpu/intern/gpu_extensions.c b/source/blender/gpu/intern/gpu_extensions.c
index 99dbff05453..5839b34cd19 100644
--- a/source/blender/gpu/intern/gpu_extensions.c
+++ b/source/blender/gpu/intern/gpu_extensions.c
@@ -309,6 +309,7 @@ void gpu_extensions_init(void)
   }
   else if ((strstr(renderer, "Mesa DRI R")) ||
            (strstr(renderer, "Radeon") && strstr(vendor, "X.Org")) ||
+           (strstr(renderer, "AMD") && strstr(vendor, "X.Org")) ||
            (strstr(renderer, "Gallium ") && strstr(renderer, " on ATI ")) ||
            (strstr(renderer, "Gallium ") && strstr(renderer, " on AMD "))) {
     GG.device = GPU_DEVICE_ATI;
diff --git a/source/blender/gpu/intern/gpu_material.c b/source/blender/gpu/intern/gpu_material.c
index 20b91c0c95d..05d991105c0 100644
--- a/source/blender/gpu/intern/gpu_material.c
+++ b/source/blender/gpu/intern/gpu_material.c
@@ -37,6 +37,7 @@
 #include "BLI_utildefines.h"
 #include "BLI_string.h"
 #include "BLI_string_utils.h"
+#include "BLI_ghash.h"
 
 #include "BKE_main.h"
 #include "BKE_node.h"
@@ -102,6 +103,8 @@ struct GPUMaterial {
   GPUTexture *coba_tex; /* 1D Texture array containing all color bands. */
   GPUColorBandBuilder *coba_builder;
 
+  GSet *used_libraries;
+
 #ifndef NDEBUG
   char name[64];
 #endif
@@ -183,6 +186,8 @@ static void gpu_material_free_single(GPUMaterial *material)
   if (material->coba_tex != NULL) {
     GPU_texture_free(material->coba_tex);
   }
+
+  BLI_gset_free(material->used_libraries, NULL);
 }
 
 void GPU_material_free(ListBase *gpumaterial)
@@ -582,6 +587,11 @@ void gpu_material_add_node(GPUMaterial *material, GPUNode *node)
   BLI_addtail(&material->nodes, node);
 }
 
+GSet *gpu_material_used_libraries(GPUMaterial *material)
+{
+  return material->used_libraries;
+}
+
 /* Return true if the material compilation has not yet begin or begin. */
 eGPUMaterialStatus GPU_material_status(GPUMaterial *mat)
 {
@@ -659,6 +669,9 @@ GPUMaterial *GPU_material_from_nodetree(Scene *scene,
   UNUSED_VARS(name);
 #endif
 
+  mat->used_libraries = BLI_gset_new(
+      BLI_ghashutil_ptrhash, BLI_ghashutil_ptrcmp, "GPUMaterial.used_libraries");
+
   /* localize tree to create links for reroute and mute */
   bNodeTree *localtree = ntreeLocalize(ntree);
   ntreeGPUMaterialNodes(localtree, mat, &has_surface_output, &has_volume_output);
diff --git a/source/blender/gpu/intern/gpu_material_library.h b/source/blender/gpu/intern/gpu_material_library.h
new file mode 100644
index 00000000000..99a71094753
--- /dev/null
+++ b/source/blender/gpu/intern/gpu_material_library.h
@@ -0,0 +1,657 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * The Original Code is Copyright (C) 2005 Blender Foundation.
+ * All rights reserved.
+ */
+
+/** \file
+ * \ingroup gpu
+ *
+ * List of all gpu_shader_material_*.glsl files used by GLSL materials. These
+ * will be parsed to make all functions in them available to use for GPU_link().
+ *
+ * If a file uses functions from another file, it must be added to the list of
+ * dependencies, and be placed after that file in the list. */
+
+#ifndef __GPU_MATERIAL_LIBRARY_H__
+#define __GPU_MATERIAL_LIBRARY_H__
+
+typedef struct GPUMaterialLibrary {
+  char *code;
+  struct GPUMaterialLibrary *dependencies[8];
+} GPUMaterialLibrary;
+
+extern char datatoc_gpu_shader_material_add_shader_glsl[];
+extern char datatoc_gpu_shader_material_ambient_occlusion_glsl[];
+extern char datatoc_gpu_shader_material_anisotropic_glsl[];
+extern char datatoc_gpu_shader_material_attribute_glsl[];
+extern char datatoc_gpu_shader_material_background_glsl[];
+extern char datatoc_gpu_shader_material_bevel_glsl[];
+extern char datatoc_gpu_shader_material_blackbody_glsl[];
+extern char datatoc_gpu_shader_material_bright_contrast_glsl[];
+extern char datatoc_gpu_shader_material_bump_glsl[];
+extern char datatoc_gpu_shader_material_camera_glsl[];
+extern char datatoc_gpu_shader_material_cell_noise_glsl[];
+extern char datatoc_gpu_shader_material_clamp_glsl[];
+extern char datatoc_gpu_shader_material_color_ramp_glsl[];
+extern char datatoc_gpu_shader_material_color_util_glsl[];
+extern char datatoc_gpu_shader_material_combine_hsv_glsl[];
+extern char datatoc_gpu_shader_material_combine_rgb_glsl[];
+extern char datatoc_gpu_shader_material_combine_xyz_glsl[];
+extern char datatoc_gpu_shader_material_diffuse_glsl[];
+extern char datatoc_gpu_shader_material_displacement_glsl[];
+extern char datatoc_gpu_shader_material_eevee_specular_glsl[];
+extern char datatoc_gpu_shader_material_emission_glsl[];
+extern char datatoc_gpu_shader_material_fractal_noise_glsl[];
+extern char datatoc_gpu_shader_material_fresnel_glsl[];
+extern char datatoc_gpu_shader_material_gamma_glsl[];
+extern char datatoc_gpu_shader_material_geometry_glsl[];
+extern char datatoc_gpu_shader_material_glass_glsl[];
+extern char datatoc_gpu_shader_material_glossy_glsl[];
+extern char datatoc_gpu_shader_material_hair_info_glsl[];
+extern char datatoc_gpu_shader_material_hash_glsl[];
+extern char datatoc_gpu_shader_material_holdout_glsl[];
+extern char datatoc_gpu_shader_material_hue_sat_val_glsl[];
+extern char datatoc_gpu_shader_material_invert_glsl[];
+extern char datatoc_gpu_shader_material_layer_weight_glsl[];
+extern char datatoc_gpu_shader_material_light_falloff_glsl[];
+extern char datatoc_gpu_shader_material_light_path_glsl[];
+extern char datatoc_gpu_shader_material_mapping_glsl[];
+extern char datatoc_gpu_shader_material_map_range_glsl[];
+extern char datatoc_gpu_shader_material_math_glsl[];
+extern char datatoc_gpu_shader_material_math_util_glsl[];
+extern char datatoc_gpu_shader_material_mix_rgb_glsl[];
+extern char datatoc_gpu_shader_material_mix_shader_glsl[];
+extern char datatoc_gpu_shader_material_noise_glsl[];
+extern char datatoc_gpu_shader_material_normal_glsl[];
+extern char datatoc_gpu_shader_material_normal_map_glsl[];
+extern char datatoc_gpu_shader_material_object_info_glsl[];
+extern char datatoc_gpu_shader_material_output_material_glsl[];
+extern char datatoc_gpu_shader_material_output_world_glsl[];
+extern char datatoc_gpu_shader_material_particle_info_glsl[];
+extern char datatoc_gpu_shader_material_principled_glsl[];
+extern char datatoc_gpu_shader_material_refraction_glsl[];
+extern char datatoc_gpu_shader_material_rgb_curves_glsl[];
+extern char datatoc_gpu_shader_material_rgb_to_bw_glsl[];
+extern char datatoc_gpu_shader_material_separate_hsv_glsl[];
+extern char datatoc_gpu_shader_material_separate_rgb_glsl[];
+extern char datatoc_gpu_shader_material_separate_xyz_glsl[];
+extern char datatoc_gpu_shader_material_set_glsl[];
+extern char datatoc_gpu_shader_material_shader_to_rgba_glsl[];
+extern char datatoc_gpu_shader_material_squeeze_glsl[];
+extern char datatoc_gpu_shader_material_subsurface_scattering_glsl[];
+extern char datatoc_gpu_shader_material_tangent_glsl[];
+extern char datatoc_gpu_shader_material_tex_brick_glsl[];
+extern char datatoc_gpu_shader_material_tex_checker_glsl[];
+extern char datatoc_gpu_shader_material_tex_environment_glsl[];
+extern char datatoc_gpu_shader_material_tex_gradient_glsl[];
+extern char datatoc_gpu_shader_material_tex_image_glsl[];
+extern char datatoc_gpu_shader_material_tex_magic_glsl[];
+extern char datatoc_gpu_shader_material_tex_musgrave_glsl[];
+extern char datatoc_gpu_shader_material_tex_noise_glsl[];
+extern char datatoc_gpu_shader_material_tex_sky_glsl[];
+extern char datatoc_gpu_shader_material_texture_coordinates_glsl[];
+extern char datatoc_gpu_shader_material_tex_voronoi_glsl[];
+extern char datatoc_gpu_shader_material_tex_wave_glsl[];
+extern char datatoc_gpu_shader_material_tex_white_noise_glsl[];
+extern char datatoc_gpu_shader_material_toon_glsl[];
+extern char datatoc_gpu_shader_material_translucent_glsl[];
+extern char datatoc_gpu_shader_material_transparent_glsl[];
+extern char datatoc_gpu_shader_material_uv_map_glsl[];
+extern char datatoc_gpu_shader_material_vector_curves_glsl[];
+extern char datatoc_gpu_shader_material_vector_displacement_glsl[];
+extern char datatoc_gpu_shader_material_vector_math_glsl[];
+extern char datatoc_gpu_shader_material_velvet_glsl[];
+extern char datatoc_gpu_shader_material_volume_absorption_glsl[];
+extern char datatoc_gpu_shader_material_volume_info_glsl[];
+extern char datatoc_gpu_shader_material_volume_principled_glsl[];
+extern char datatoc_gpu_shader_material_volume_scatter_glsl[];
+extern char datatoc_gpu_shader_material_wireframe_glsl[];
+extern char datatoc_gpu_shader_material_world_normals_glsl[];
+
+static GPUMaterialLibrary gpu_shader_material_math_util_library = {
+    .code = datatoc_gpu_shader_material_math_util_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_color_util_library = {
+    .code = datatoc_gpu_shader_material_color_util_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_hash_library = {
+    .code = datatoc_gpu_shader_material_hash_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_noise_library = {
+    .code = datatoc_gpu_shader_material_noise_glsl,
+    .dependencies = {&gpu_shader_material_math_util_library,
+                     &gpu_shader_material_hash_library,
+                     NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_fractal_noise_library = {
+    .code = datatoc_gpu_shader_material_fractal_noise_glsl,
+    .dependencies = {&gpu_shader_material_noise_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_cell_noise_library = {
+    .code = datatoc_gpu_shader_material_cell_noise_glsl,
+    .dependencies = {&gpu_shader_material_math_util_library,
+                     &gpu_shader_material_hash_library,
+                     NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_add_shader_library = {
+    .code = datatoc_gpu_shader_material_add_shader_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_ambient_occlusion_library = {
+    .code = datatoc_gpu_shader_material_ambient_occlusion_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_glossy_library = {
+    .code = datatoc_gpu_shader_material_glossy_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_anisotropic_library = {
+    .code = datatoc_gpu_shader_material_anisotropic_glsl,
+    .dependencies = {&gpu_shader_material_glossy_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_attribute_library = {
+    .code = datatoc_gpu_shader_material_attribute_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_background_library = {
+    .code = datatoc_gpu_shader_material_background_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_bevel_library = {
+    .code = datatoc_gpu_shader_material_bevel_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_blackbody_library = {
+    .code = datatoc_gpu_shader_material_blackbody_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_bright_contrast_library = {
+    .code = datatoc_gpu_shader_material_bright_contrast_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_bump_library = {
+    .code = datatoc_gpu_shader_material_bump_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_camera_library = {
+    .code = datatoc_gpu_shader_material_camera_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_clamp_library = {
+    .code = datatoc_gpu_shader_material_clamp_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_color_ramp_library = {
+    .code = datatoc_gpu_shader_material_color_ramp_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_combine_hsv_library = {
+    .code = datatoc_gpu_shader_material_combine_hsv_glsl,
+    .dependencies = {&gpu_shader_material_color_util_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_combine_rgb_library = {
+    .code = datatoc_gpu_shader_material_combine_rgb_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_combine_xyz_library = {
+    .code = datatoc_gpu_shader_material_combine_xyz_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_diffuse_library = {
+    .code = datatoc_gpu_shader_material_diffuse_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_displacement_library = {
+    .code = datatoc_gpu_shader_material_displacement_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_eevee_specular_library = {
+    .code = datatoc_gpu_shader_material_eevee_specular_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_emission_library = {
+    .code = datatoc_gpu_shader_material_emission_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_fresnel_library = {
+    .code = datatoc_gpu_shader_material_fresnel_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_gamma_library = {
+    .code = datatoc_gpu_shader_material_gamma_glsl,
+    .dependencies = {&gpu_shader_material_math_util_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tangent_library = {
+    .code = datatoc_gpu_shader_material_tangent_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_geometry_library = {
+    .code = datatoc_gpu_shader_material_geometry_glsl,
+    .dependencies = {&gpu_shader_material_tangent_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_glass_library = {
+    .code = datatoc_gpu_shader_material_glass_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_hair_info_library = {
+    .code = datatoc_gpu_shader_material_hair_info_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_holdout_library = {
+    .code = datatoc_gpu_shader_material_holdout_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_hue_sat_val_library = {
+    .code = datatoc_gpu_shader_material_hue_sat_val_glsl,
+    .dependencies = {&gpu_shader_material_color_util_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_invert_library = {
+    .code = datatoc_gpu_shader_material_invert_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_layer_weight_library = {
+    .code = datatoc_gpu_shader_material_layer_weight_glsl,
+    .dependencies = {&gpu_shader_material_fresnel_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_light_falloff_library = {
+    .code = datatoc_gpu_shader_material_light_falloff_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_light_path_library = {
+    .code = datatoc_gpu_shader_material_light_path_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_mapping_library = {
+    .code = datatoc_gpu_shader_material_mapping_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_map_range_library = {
+    .code = datatoc_gpu_shader_material_map_range_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_math_library = {
+    .code = datatoc_gpu_shader_material_math_glsl,
+    .dependencies = {&gpu_shader_material_math_util_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_mix_rgb_library = {
+    .code = datatoc_gpu_shader_material_mix_rgb_glsl,
+    .dependencies = {&gpu_shader_material_color_util_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_mix_shader_library = {
+    .code = datatoc_gpu_shader_material_mix_shader_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_normal_library = {
+    .code = datatoc_gpu_shader_material_normal_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_normal_map_library = {
+    .code = datatoc_gpu_shader_material_normal_map_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_object_info_library = {
+    .code = datatoc_gpu_shader_material_object_info_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_output_material_library = {
+    .code = datatoc_gpu_shader_material_output_material_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_output_world_library = {
+    .code = datatoc_gpu_shader_material_output_world_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_particle_info_library = {
+    .code = datatoc_gpu_shader_material_particle_info_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_principled_library = {
+    .code = datatoc_gpu_shader_material_principled_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_refraction_library = {
+    .code = datatoc_gpu_shader_material_refraction_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_rgb_curves_library = {
+    .code = datatoc_gpu_shader_material_rgb_curves_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_rgb_to_bw_library = {
+    .code = datatoc_gpu_shader_material_rgb_to_bw_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_separate_hsv_library = {
+    .code = datatoc_gpu_shader_material_separate_hsv_glsl,
+    .dependencies = {&gpu_shader_material_color_util_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_separate_rgb_library = {
+    .code = datatoc_gpu_shader_material_separate_rgb_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_separate_xyz_library = {
+    .code = datatoc_gpu_shader_material_separate_xyz_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_set_library = {
+    .code = datatoc_gpu_shader_material_set_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_shader_to_rgba_library = {
+    .code = datatoc_gpu_shader_material_shader_to_rgba_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_squeeze_library = {
+    .code = datatoc_gpu_shader_material_squeeze_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_subsurface_scattering_library = {
+    .code = datatoc_gpu_shader_material_subsurface_scattering_glsl,
+    .dependencies = {&gpu_shader_material_diffuse_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_brick_library = {
+    .code = datatoc_gpu_shader_material_tex_brick_glsl,
+    .dependencies = {&gpu_shader_material_math_util_library,
+                     &gpu_shader_material_hash_library,
+                     NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_checker_library = {
+    .code = datatoc_gpu_shader_material_tex_checker_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_environment_library = {
+    .code = datatoc_gpu_shader_material_tex_environment_glsl,
+    .dependencies = {&gpu_shader_material_math_util_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_gradient_library = {
+    .code = datatoc_gpu_shader_material_tex_gradient_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_image_library = {
+    .code = datatoc_gpu_shader_material_tex_image_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_magic_library = {
+    .code = datatoc_gpu_shader_material_tex_magic_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_musgrave_library = {
+    .code = datatoc_gpu_shader_material_tex_musgrave_glsl,
+    .dependencies = {&gpu_shader_material_noise_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_noise_library = {
+    .code = datatoc_gpu_shader_material_tex_noise_glsl,
+    .dependencies = {&gpu_shader_material_fractal_noise_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_sky_library = {
+    .code = datatoc_gpu_shader_material_tex_sky_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_texture_coordinates_library = {
+    .code = datatoc_gpu_shader_material_texture_coordinates_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_voronoi_library = {
+    .code = datatoc_gpu_shader_material_tex_voronoi_glsl,
+    .dependencies = {&gpu_shader_material_math_util_library,
+                     &gpu_shader_material_cell_noise_library,
+                     NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_wave_library = {
+    .code = datatoc_gpu_shader_material_tex_wave_glsl,
+    .dependencies = {&gpu_shader_material_fractal_noise_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_tex_white_noise_library = {
+    .code = datatoc_gpu_shader_material_tex_white_noise_glsl,
+    .dependencies = {&gpu_shader_material_hash_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_toon_library = {
+    .code = datatoc_gpu_shader_material_toon_glsl,
+    .dependencies = {&gpu_shader_material_diffuse_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_translucent_library = {
+    .code = datatoc_gpu_shader_material_translucent_glsl,
+    .dependencies = {&gpu_shader_material_diffuse_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_transparent_library = {
+    .code = datatoc_gpu_shader_material_transparent_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_uv_map_library = {
+    .code = datatoc_gpu_shader_material_uv_map_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_vector_curves_library = {
+    .code = datatoc_gpu_shader_material_vector_curves_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_vector_displacement_library = {
+    .code = datatoc_gpu_shader_material_vector_displacement_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_vector_math_library = {
+    .code = datatoc_gpu_shader_material_vector_math_glsl,
+    .dependencies = {&gpu_shader_material_math_util_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_velvet_library = {
+    .code = datatoc_gpu_shader_material_velvet_glsl,
+    .dependencies = {&gpu_shader_material_diffuse_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_volume_absorption_library = {
+    .code = datatoc_gpu_shader_material_volume_absorption_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_volume_info_library = {
+    .code = datatoc_gpu_shader_material_volume_info_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_volume_principled_library = {
+    .code = datatoc_gpu_shader_material_volume_principled_glsl,
+    .dependencies = {&gpu_shader_material_blackbody_library, NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_volume_scatter_library = {
+    .code = datatoc_gpu_shader_material_volume_scatter_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_wireframe_library = {
+    .code = datatoc_gpu_shader_material_wireframe_glsl,
+    .dependencies = {NULL},
+};
+
+static GPUMaterialLibrary gpu_shader_material_world_normals_library = {
+    .code = datatoc_gpu_shader_material_world_normals_glsl,
+    .dependencies = {&gpu_shader_material_texture_coordinates_library, NULL},
+};
+
+static GPUMaterialLibrary *gpu_material_libraries[] = {
+    &gpu_shader_material_math_util_library,
+    &gpu_shader_material_color_util_library,
+    &gpu_shader_material_hash_library,
+    &gpu_shader_material_noise_library,
+    &gpu_shader_material_fractal_noise_library,
+    &gpu_shader_material_cell_noise_library,
+    &gpu_shader_material_add_shader_library,
+    &gpu_shader_material_ambient_occlusion_library,
+    &gpu_shader_material_glossy_library,
+    &gpu_shader_material_anisotropic_library,
+    &gpu_shader_material_attribute_library,
+    &gpu_shader_material_background_library,
+    &gpu_shader_material_bevel_library,
+    &gpu_shader_material_blackbody_library,
+    &gpu_shader_material_bright_contrast_library,
+    &gpu_shader_material_bump_library,
+    &gpu_shader_material_camera_library,
+    &gpu_shader_material_clamp_library,
+    &gpu_shader_material_color_ramp_library,
+    &gpu_shader_material_combine_hsv_library,
+    &gpu_shader_material_combine_rgb_library,
+    &gpu_shader_material_combine_xyz_library,
+    &gpu_shader_material_diffuse_library,
+    &gpu_shader_material_displacement_library,
+    &gpu_shader_material_eevee_specular_library,
+    &gpu_shader_material_emission_library,
+    &gpu_shader_material_fresnel_library,
+    &gpu_shader_material_gamma_library,
+    &gpu_shader_material_tangent_library,
+    &gpu_shader_material_geometry_library,
+    &gpu_shader_material_glass_library,
+    &gpu_shader_material_hair_info_library,
+    &gpu_shader_material_holdout_library,
+    &gpu_shader_material_hue_sat_val_library,
+    &gpu_shader_material_invert_library,
+    &gpu_shader_material_layer_weight_library,
+    &gpu_shader_material_light_falloff_library,
+    &gpu_shader_material_light_path_library,
+    &gpu_shader_material_mapping_library,
+    &gpu_shader_material_map_range_library,
+    &gpu_shader_material_math_library,
+    &gpu_shader_material_mix_rgb_library,
+    &gpu_shader_material_mix_shader_library,
+    &gpu_shader_material_normal_library,
+    &gpu_shader_material_normal_map_library,
+    &gpu_shader_material_object_info_library,
+    &gpu_shader_material_output_material_library,
+    &gpu_shader_material_output_world_library,
+    &gpu_shader_material_particle_info_library,
+    &gpu_shader_material_principled_library,
+    &gpu_shader_material_refraction_library,
+    &gpu_shader_material_rgb_curves_library,
+    &gpu_shader_material_rgb_to_bw_library,
+    &gpu_shader_material_separate_hsv_library,
+    &gpu_shader_material_separate_rgb_library,
+    &gpu_shader_material_separate_xyz_library,
+    &gpu_shader_material_set_library,
+    &gpu_shader_material_shader_to_rgba_library,
+    &gpu_shader_material_squeeze_library,
+    &gpu_shader_material_subsurface_scattering_library,
+    &gpu_shader_material_tex_brick_library,
+    &gpu_shader_material_tex_checker_library,
+    &gpu_shader_material_tex_environment_library,
+    &gpu_shader_material_tex_gradient_library,
+    &gpu_shader_material_tex_image_library,
+    &gpu_shader_material_tex_magic_library,
+    &gpu_shader_material_tex_musgrave_library,
+    &gpu_shader_material_tex_noise_library,
+    &gpu_shader_material_tex_sky_library,
+    &gpu_shader_material_texture_coordinates_library,
+    &gpu_shader_material_tex_voronoi_library,
+    &gpu_shader_material_tex_wave_library,
+    &gpu_shader_material_tex_white_noise_library,
+    &gpu_shader_material_toon_library,
+    &gpu_shader_material_translucent_library,
+    &gpu_shader_material_transparent_library,
+    &gpu_shader_material_uv_map_library,
+    &gpu_shader_material_vector_curves_library,
+    &gpu_shader_material_vector_displacement_library,
+    &gpu_shader_material_vector_math_library,
+    &gpu_shader_material_velvet_library,
+    &gpu_shader_material_volume_absorption_library,
+    &gpu_shader_material_volume_info_library,
+    &gpu_shader_material_volume_principled_library,
+    &gpu_shader_material_volume_scatter_library,
+    &gpu_shader_material_wireframe_library,
+    &gpu_shader_material_world_normals_library,
+    NULL};
+#endif
diff --git a/source/blender/gpu/intern/gpu_private.h b/source/blender/gpu/intern/gpu_private.h
index 4b90f8daa8c..43b2da13e28 100644
--- a/source/blender/gpu/intern/gpu_private.h
+++ b/source/blender/gpu/intern/gpu_private.h
@@ -21,9 +21,6 @@
 #ifndef __GPU_PRIVATE_H__
 #define __GPU_PRIVATE_H__
 
-struct GPUContext;
-struct GPUFrameBuffer;
-
 /* call this before running any of the functions below */
 void gpu_extensions_init(void);
 void gpu_extensions_exit(void);
diff --git a/source/blender/gpu/intern/gpu_shader_interface.c b/source/blender/gpu/intern/gpu_shader_interface.c
index e34c6e23024..f205ef31ed2 100644
--- a/source/blender/gpu/intern/gpu_shader_interface.c
+++ b/source/blender/gpu/intern/gpu_shader_interface.c
@@ -67,6 +67,7 @@ static const char *BuiltinUniform_name(GPUUniformBuiltin u)
       [GPU_UNIFORM_COLOR] = "color",
       [GPU_UNIFORM_CALLID] = "callId",
       [GPU_UNIFORM_OBJECT_INFO] = "unfobjectinfo",
+      [GPU_UNIFORM_OBJECT_COLOR] = "unfobjectcolor",
 
       [GPU_UNIFORM_CUSTOM] = NULL,
       [GPU_NUM_UNIFORMS] = NULL,
diff --git a/source/blender/gpu/intern/gpu_texture.c b/source/blender/gpu/intern/gpu_texture.c
index dab17fcd72a..955b11036ef 100644
--- a/source/blender/gpu/intern/gpu_texture.c
+++ b/source/blender/gpu/intern/gpu_texture.c
@@ -1598,7 +1598,7 @@ void GPU_texture_generate_mipmap(GPUTexture *tex)
 
   if (GPU_texture_depth(tex)) {
     /* Some drivers have bugs when using glGenerateMipmap with depth textures (see T56789).
-     * In this case we just create a complete texture with mipmaps manually without downsampling.
+     * In this case we just create a complete texture with mipmaps manually without down-sampling.
      * You must initialize the texture levels using other methods like
      * GPU_framebuffer_recursive_downsample(). */
     int levels = 1 + floor(log2(max_ii(tex->w, tex->h)));
diff --git a/source/blender/gpu/intern/gpu_vertex_format.c b/source/blender/gpu/intern/gpu_vertex_format.c
index 11df86c0b3a..66e5e254734 100644
--- a/source/blender/gpu/intern/gpu_vertex_format.c
+++ b/source/blender/gpu/intern/gpu_vertex_format.c
@@ -253,7 +253,7 @@ void GPU_vertformat_safe_attrib_name(const char *attrib_name,
     *(uint *)&data[4] = BLI_ghashutil_strhash_p_murmur(attrib_name + 4);
   }
   else {
-    /* Copy the whole name. Collision is barelly possible
+    /* Copy the whole name. Collision is barely possible
      * (hash would have to be equal to the last 4 bytes). */
     for (int i = 0; i < 8 && attrib_name[i] != '\0'; i++) {
       data[i] = attrib_name[i];
diff --git a/source/blender/gpu/shaders/gpu_shader_keyframe_diamond_vert.glsl b/source/blender/gpu/shaders/gpu_shader_keyframe_diamond_vert.glsl
index 8dc92a2b415..2ba89230d80 100644
--- a/source/blender/gpu/shaders/gpu_shader_keyframe_diamond_vert.glsl
+++ b/source/blender/gpu/shaders/gpu_shader_keyframe_diamond_vert.glsl
@@ -1,11 +1,11 @@
 
 uniform mat4 ModelViewProjectionMatrix;
 uniform vec2 ViewportSize = vec2(-1, -1);
+uniform float outline_scale = 1.0;
 
 const float line_falloff = 1.0;
 const float circle_scale = sqrt(2.0 / 3.1416);
 const float square_scale = sqrt(0.5);
-
 const float diagonal_scale = sqrt(0.5);
 
 in vec2 pos;
@@ -58,7 +58,7 @@ void main()
   float line_width = half_width + line_falloff;
 
   /* Outline thresholds. */
-  thresholds.xy = line_thresholds(line_width);
+  thresholds.xy = line_thresholds(line_width * outline_scale);
 
   /* Inner dot thresholds. */
   thresholds.zw = line_thresholds(line_width * 1.6);
diff --git a/source/blender/gpu/shaders/gpu_shader_material.glsl b/source/blender/gpu/shaders/gpu_shader_material.glsl
deleted file mode 100644
index f077ae55d8f..00000000000
--- a/source/blender/gpu/shaders/gpu_shader_material.glsl
+++ /dev/null
@@ -1,3615 +0,0 @@
-
-/* Converters */
-
-float convert_rgba_to_float(vec4 color)
-{
-  return dot(color.rgb, vec3(0.2126, 0.7152, 0.0722));
-}
-
-float exp_blender(float f)
-{
-  return pow(2.71828182846, f);
-}
-
-float compatible_pow(float x, float y)
-{
-  if (y == 0.0) { /* x^0 -> 1, including 0^0 */
-    return 1.0;
-  }
-
-  /* glsl pow doesn't accept negative x */
-  if (x < 0.0) {
-    if (mod(-y, 2.0) == 0.0) {
-      return pow(-x, y);
-    }
-    else {
-      return -pow(-x, y);
-    }
-  }
-  else if (x == 0.0) {
-    return 0.0;
-  }
-
-  return pow(x, y);
-}
-
-void rgb_to_hsv(vec4 rgb, out vec4 outcol)
-{
-  float cmax, cmin, h, s, v, cdelta;
-  vec3 c;
-
-  cmax = max(rgb[0], max(rgb[1], rgb[2]));
-  cmin = min(rgb[0], min(rgb[1], rgb[2]));
-  cdelta = cmax - cmin;
-
-  v = cmax;
-  if (cmax != 0.0) {
-    s = cdelta / cmax;
-  }
-  else {
-    s = 0.0;
-    h = 0.0;
-  }
-
-  if (s == 0.0) {
-    h = 0.0;
-  }
-  else {
-    c = (vec3(cmax) - rgb.xyz) / cdelta;
-
-    if (rgb.x == cmax) {
-      h = c[2] - c[1];
-    }
-    else if (rgb.y == cmax) {
-      h = 2.0 + c[0] - c[2];
-    }
-    else {
-      h = 4.0 + c[1] - c[0];
-    }
-
-    h /= 6.0;
-
-    if (h < 0.0) {
-      h += 1.0;
-    }
-  }
-
-  outcol = vec4(h, s, v, rgb.w);
-}
-
-void hsv_to_rgb(vec4 hsv, out vec4 outcol)
-{
-  float i, f, p, q, t, h, s, v;
-  vec3 rgb;
-
-  h = hsv[0];
-  s = hsv[1];
-  v = hsv[2];
-
-  if (s == 0.0) {
-    rgb = vec3(v, v, v);
-  }
-  else {
-    if (h == 1.0) {
-      h = 0.0;
-    }
-
-    h *= 6.0;
-    i = floor(h);
-    f = h - i;
-    rgb = vec3(f, f, f);
-    p = v * (1.0 - s);
-    q = v * (1.0 - (s * f));
-    t = v * (1.0 - (s * (1.0 - f)));
-
-    if (i == 0.0) {
-      rgb = vec3(v, t, p);
-    }
-    else if (i == 1.0) {
-      rgb = vec3(q, v, p);
-    }
-    else if (i == 2.0) {
-      rgb = vec3(p, v, t);
-    }
-    else if (i == 3.0) {
-      rgb = vec3(p, q, v);
-    }
-    else if (i == 4.0) {
-      rgb = vec3(t, p, v);
-    }
-    else {
-      rgb = vec3(v, p, q);
-    }
-  }
-
-  outcol = vec4(rgb, hsv.w);
-}
-
-void color_to_normal_new_shading(vec3 color, out vec3 normal)
-{
-  normal = vec3(2.0) * color - vec3(1.0);
-}
-
-void color_to_blender_normal_new_shading(vec3 color, out vec3 normal)
-{
-  normal = vec3(2.0, -2.0, -2.0) * color - vec3(1.0);
-}
-
-#ifndef M_PI
-#  define M_PI 3.14159265358979323846
-#endif
-#ifndef M_1_PI
-#  define M_1_PI 0.318309886183790671538
-#endif
-
-/*********** SHADER NODES ***************/
-
-void particle_info(vec4 sprops,
-                   vec4 loc,
-                   vec3 vel,
-                   vec3 avel,
-                   out float index,
-                   out float random,
-                   out float age,
-                   out float life_time,
-                   out vec3 location,
-                   out float size,
-                   out vec3 velocity,
-                   out vec3 angular_velocity)
-{
-  index = sprops.x;
-  random = loc.w;
-  age = sprops.y;
-  life_time = sprops.z;
-  size = sprops.w;
-
-  location = loc.xyz;
-  velocity = vel;
-  angular_velocity = avel;
-}
-
-void vect_normalize(vec3 vin, out vec3 vout)
-{
-  vout = normalize(vin);
-}
-
-void direction_transform_m4v3(vec3 vin, mat4 mat, out vec3 vout)
-{
-  vout = (mat * vec4(vin, 0.0)).xyz;
-}
-
-void normal_transform_transposed_m4v3(vec3 vin, mat4 mat, out vec3 vout)
-{
-  vout = transpose(mat3(mat)) * vin;
-}
-
-void point_transform_m4v3(vec3 vin, mat4 mat, out vec3 vout)
-{
-  vout = (mat * vec4(vin, 1.0)).xyz;
-}
-
-void point_texco_remap_square(vec3 vin, out vec3 vout)
-{
-  vout = vin * 2.0 - 1.0;
-}
-
-void point_texco_clamp(vec3 vin, sampler2D ima, out vec3 vout)
-{
-  vec2 half_texel_size = 0.5 / vec2(textureSize(ima, 0).xy);
-  vout = clamp(vin, half_texel_size.xyy, 1.0 - half_texel_size.xyy);
-}
-
-void point_map_to_sphere(vec3 vin, out vec3 vout)
-{
-  float len = length(vin);
-  float v, u;
-  if (len > 0.0) {
-    if (vin.x == 0.0 && vin.y == 0.0) {
-      u = 0.0;
-    }
-    else {
-      u = (1.0 - atan(vin.x, vin.y) / M_PI) / 2.0;
-    }
-
-    v = 1.0 - acos(vin.z / len) / M_PI;
-  }
-  else {
-    v = u = 0.0;
-  }
-
-  vout = vec3(u, v, 0.0);
-}
-
-void point_map_to_tube(vec3 vin, out vec3 vout)
-{
-  float u, v;
-  v = (vin.z + 1.0) * 0.5;
-  float len = sqrt(vin.x * vin.x + vin.y * vin[1]);
-  if (len > 0.0) {
-    u = (1.0 - (atan(vin.x / len, vin.y / len) / M_PI)) * 0.5;
-  }
-  else {
-    v = u = 0.0;
-  }
-
-  vout = vec3(u, v, 0.0);
-}
-
-void mapping(
-    vec3 vec, vec4 m0, vec4 m1, vec4 m2, vec4 m3, vec3 minvec, vec3 maxvec, out vec3 outvec)
-{
-  mat4 mat = mat4(m0, m1, m2, m3);
-  outvec = (mat * vec4(vec, 1.0)).xyz;
-  outvec = clamp(outvec, minvec, maxvec);
-}
-
-void camera(vec3 co, out vec3 outview, out float outdepth, out float outdist)
-{
-  outdepth = abs(co.z);
-  outdist = length(co);
-  outview = normalize(co);
-}
-
-void math_add(float val1, float val2, out float outval)
-{
-  outval = val1 + val2;
-}
-
-void math_subtract(float val1, float val2, out float outval)
-{
-  outval = val1 - val2;
-}
-
-void math_multiply(float val1, float val2, out float outval)
-{
-  outval = val1 * val2;
-}
-
-void math_divide(float val1, float val2, out float outval)
-{
-  if (val2 == 0.0) {
-    outval = 0.0;
-  }
-  else {
-    outval = val1 / val2;
-  }
-}
-
-void math_sine(float val, out float outval)
-{
-  outval = sin(val);
-}
-
-void math_cosine(float val, out float outval)
-{
-  outval = cos(val);
-}
-
-void math_tangent(float val, out float outval)
-{
-  outval = tan(val);
-}
-
-void math_asin(float val, out float outval)
-{
-  if (val <= 1.0 && val >= -1.0) {
-    outval = asin(val);
-  }
-  else {
-    outval = 0.0;
-  }
-}
-
-void math_acos(float val, out float outval)
-{
-  if (val <= 1.0 && val >= -1.0) {
-    outval = acos(val);
-  }
-  else {
-    outval = 0.0;
-  }
-}
-
-void math_atan(float val, out float outval)
-{
-  outval = atan(val);
-}
-
-void math_pow(float val1, float val2, out float outval)
-{
-  if (val1 >= 0.0) {
-    outval = compatible_pow(val1, val2);
-  }
-  else {
-    float val2_mod_1 = mod(abs(val2), 1.0);
-
-    if (val2_mod_1 > 0.999 || val2_mod_1 < 0.001) {
-      outval = compatible_pow(val1, floor(val2 + 0.5));
-    }
-    else {
-      outval = 0.0;
-    }
-  }
-}
-
-void math_log(float val1, float val2, out float outval)
-{
-  if (val1 > 0.0 && val2 > 0.0) {
-    outval = log2(val1) / log2(val2);
-  }
-  else {
-    outval = 0.0;
-  }
-}
-
-void math_max(float val1, float val2, out float outval)
-{
-  outval = max(val1, val2);
-}
-
-void math_min(float val1, float val2, out float outval)
-{
-  outval = min(val1, val2);
-}
-
-void math_round(float val, out float outval)
-{
-  outval = floor(val + 0.5);
-}
-
-void math_less_than(float val1, float val2, out float outval)
-{
-  if (val1 < val2) {
-    outval = 1.0;
-  }
-  else {
-    outval = 0.0;
-  }
-}
-
-void math_greater_than(float val1, float val2, out float outval)
-{
-  if (val1 > val2) {
-    outval = 1.0;
-  }
-  else {
-    outval = 0.0;
-  }
-}
-
-void math_modulo(float val1, float val2, out float outval)
-{
-  if (val2 == 0.0 || val1 == val2) {
-    outval = 0.0;
-  }
-  else {
-    /* change sign to match C convention, mod in GLSL will take absolute for negative numbers,
-     * see https://www.opengl.org/sdk/docs/man/html/mod.xhtml */
-    outval = sign(val1) * mod(abs(val1), val2);
-  }
-}
-
-void math_abs(float val1, out float outval)
-{
-  outval = abs(val1);
-}
-
-void math_atan2(float val1, float val2, out float outval)
-{
-  outval = atan(val1, val2);
-}
-
-void math_floor(float val, out float outval)
-{
-  outval = floor(val);
-}
-
-void math_ceil(float val, out float outval)
-{
-  outval = ceil(val);
-}
-
-void math_fract(float val, out float outval)
-{
-  outval = val - floor(val);
-}
-
-void math_sqrt(float val, out float outval)
-{
-  if (val > 0.0) {
-    outval = sqrt(val);
-  }
-  else {
-    outval = 0.0;
-  }
-}
-
-void squeeze(float val, float width, float center, out float outval)
-{
-  outval = 1.0 / (1.0 + pow(2.71828183, -((val - center) * width)));
-}
-
-void map_range(
-    float value, float fromMin, float fromMax, float toMin, float toMax, out float result)
-{
-  if (fromMax != fromMin) {
-    result = toMin + ((value - fromMin) / (fromMax - fromMin)) * (toMax - toMin);
-  }
-  else {
-    result = 0.0;
-  }
-}
-
-void vec_math_add(vec3 v1, vec3 v2, out vec3 outvec, out float outval)
-{
-  outvec = v1 + v2;
-  outval = (abs(outvec[0]) + abs(outvec[1]) + abs(outvec[2])) * 0.333333;
-}
-
-void vec_math_sub(vec3 v1, vec3 v2, out vec3 outvec, out float outval)
-{
-  outvec = v1 - v2;
-  outval = (abs(outvec[0]) + abs(outvec[1]) + abs(outvec[2])) * 0.333333;
-}
-
-void vec_math_average(vec3 v1, vec3 v2, out vec3 outvec, out float outval)
-{
-  outvec = v1 + v2;
-  outval = length(outvec);
-  outvec = normalize(outvec);
-}
-void vec_math_mix(float strength, vec3 v1, vec3 v2, out vec3 outvec)
-{
-  outvec = strength * v1 + (1 - strength) * v2;
-}
-
-void vec_math_dot(vec3 v1, vec3 v2, out vec3 outvec, out float outval)
-{
-  outvec = vec3(0);
-  outval = dot(v1, v2);
-}
-
-void vec_math_cross(vec3 v1, vec3 v2, out vec3 outvec, out float outval)
-{
-  outvec = cross(v1, v2);
-  outval = length(outvec);
-  outvec /= outval;
-}
-
-void vec_math_normalize(vec3 v, out vec3 outvec, out float outval)
-{
-  outval = length(v);
-  outvec = normalize(v);
-}
-
-void vec_math_negate(vec3 v, out vec3 outv)
-{
-  outv = -v;
-}
-
-void invert_z(vec3 v, out vec3 outv)
-{
-  v.z = -v.z;
-  outv = v;
-}
-
-void normal_new_shading(vec3 nor, vec3 dir, out vec3 outnor, out float outdot)
-{
-  outnor = dir;
-  outdot = dot(normalize(nor), dir);
-}
-
-void curves_vec(float fac, vec3 vec, sampler1DArray curvemap, float layer, out vec3 outvec)
-{
-  vec4 co = vec4(vec * 0.5 + 0.5, layer);
-  outvec.x = texture(curvemap, co.xw).x;
-  outvec.y = texture(curvemap, co.yw).y;
-  outvec.z = texture(curvemap, co.zw).z;
-  outvec = mix(vec, outvec, fac);
-}
-
-/* ext is vec4(in_x, in_dy, out_x, out_dy). */
-float curve_extrapolate(float x, float y, vec4 ext)
-{
-  if (x < 0.0) {
-    return y + x * ext.y;
-  }
-  else if (x > 1.0) {
-    return y + (x - 1.0) * ext.w;
-  }
-  else {
-    return y;
-  }
-}
-
-#define RANGE_RESCALE(x, min, range) ((x - min) * range)
-
-void curves_rgb(float fac,
-                vec4 col,
-                sampler1DArray curvemap,
-                float layer,
-                vec4 range,
-                vec4 ext_r,
-                vec4 ext_g,
-                vec4 ext_b,
-                vec4 ext_a,
-                out vec4 outcol)
-{
-  vec4 co = vec4(RANGE_RESCALE(col.rgb, ext_a.x, range.a), layer);
-  vec3 samp;
-  samp.r = texture(curvemap, co.xw).a;
-  samp.g = texture(curvemap, co.yw).a;
-  samp.b = texture(curvemap, co.zw).a;
-
-  samp.r = curve_extrapolate(co.x, samp.r, ext_a);
-  samp.g = curve_extrapolate(co.y, samp.g, ext_a);
-  samp.b = curve_extrapolate(co.z, samp.b, ext_a);
-
-  vec3 rgb_min = vec3(ext_r.x, ext_g.x, ext_b.x);
-  co.xyz = RANGE_RESCALE(samp.rgb, rgb_min, range.rgb);
-
-  samp.r = texture(curvemap, co.xw).r;
-  samp.g = texture(curvemap, co.yw).g;
-  samp.b = texture(curvemap, co.zw).b;
-
-  outcol.r = curve_extrapolate(co.x, samp.r, ext_r);
-  outcol.g = curve_extrapolate(co.y, samp.g, ext_g);
-  outcol.b = curve_extrapolate(co.z, samp.b, ext_b);
-  outcol.a = col.a;
-
-  outcol = mix(col, outcol, fac);
-}
-
-void curves_rgb_opti(float fac,
-                     vec4 col,
-                     sampler1DArray curvemap,
-                     float layer,
-                     vec4 range,
-                     vec4 ext_a,
-                     out vec4 outcol)
-{
-  vec4 co = vec4(RANGE_RESCALE(col.rgb, ext_a.x, range.a), layer);
-  vec3 samp;
-  samp.r = texture(curvemap, co.xw).a;
-  samp.g = texture(curvemap, co.yw).a;
-  samp.b = texture(curvemap, co.zw).a;
-
-  outcol.r = curve_extrapolate(co.x, samp.r, ext_a);
-  outcol.g = curve_extrapolate(co.y, samp.g, ext_a);
-  outcol.b = curve_extrapolate(co.z, samp.b, ext_a);
-  outcol.a = col.a;
-
-  outcol = mix(col, outcol, fac);
-}
-
-void set_value(float val, out float outval)
-{
-  outval = val;
-}
-
-void set_rgb(vec3 col, out vec3 outcol)
-{
-  outcol = col;
-}
-
-void set_rgba(vec4 col, out vec4 outcol)
-{
-  outcol = col;
-}
-
-void set_value_zero(out float outval)
-{
-  outval = 0.0;
-}
-
-void set_value_one(out float outval)
-{
-  outval = 1.0;
-}
-
-void set_rgb_zero(out vec3 outval)
-{
-  outval = vec3(0.0);
-}
-
-void set_rgb_one(out vec3 outval)
-{
-  outval = vec3(1.0);
-}
-
-void set_rgba_zero(out vec4 outval)
-{
-  outval = vec4(0.0);
-}
-
-void set_rgba_one(out vec4 outval)
-{
-  outval = vec4(1.0);
-}
-
-void brightness_contrast(vec4 col, float brightness, float contrast, out vec4 outcol)
-{
-  float a = 1.0 + contrast;
-  float b = brightness - contrast * 0.5;
-
-  outcol.r = max(a * col.r + b, 0.0);
-  outcol.g = max(a * col.g + b, 0.0);
-  outcol.b = max(a * col.b + b, 0.0);
-  outcol.a = col.a;
-}
-
-void mix_blend(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  outcol = mix(col1, col2, fac);
-  outcol.a = col1.a;
-}
-
-void mix_add(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  outcol = mix(col1, col1 + col2, fac);
-  outcol.a = col1.a;
-}
-
-void mix_mult(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  outcol = mix(col1, col1 * col2, fac);
-  outcol.a = col1.a;
-}
-
-void mix_screen(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  float facm = 1.0 - fac;
-
-  outcol = vec4(1.0) - (vec4(facm) + fac * (vec4(1.0) - col2)) * (vec4(1.0) - col1);
-  outcol.a = col1.a;
-}
-
-void mix_overlay(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  float facm = 1.0 - fac;
-
-  outcol = col1;
-
-  if (outcol.r < 0.5) {
-    outcol.r *= facm + 2.0 * fac * col2.r;
-  }
-  else {
-    outcol.r = 1.0 - (facm + 2.0 * fac * (1.0 - col2.r)) * (1.0 - outcol.r);
-  }
-
-  if (outcol.g < 0.5) {
-    outcol.g *= facm + 2.0 * fac * col2.g;
-  }
-  else {
-    outcol.g = 1.0 - (facm + 2.0 * fac * (1.0 - col2.g)) * (1.0 - outcol.g);
-  }
-
-  if (outcol.b < 0.5) {
-    outcol.b *= facm + 2.0 * fac * col2.b;
-  }
-  else {
-    outcol.b = 1.0 - (facm + 2.0 * fac * (1.0 - col2.b)) * (1.0 - outcol.b);
-  }
-}
-
-void mix_sub(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  outcol = mix(col1, col1 - col2, fac);
-  outcol.a = col1.a;
-}
-
-void mix_div(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  float facm = 1.0 - fac;
-
-  outcol = col1;
-
-  if (col2.r != 0.0) {
-    outcol.r = facm * outcol.r + fac * outcol.r / col2.r;
-  }
-  if (col2.g != 0.0) {
-    outcol.g = facm * outcol.g + fac * outcol.g / col2.g;
-  }
-  if (col2.b != 0.0) {
-    outcol.b = facm * outcol.b + fac * outcol.b / col2.b;
-  }
-}
-
-void mix_diff(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  outcol = mix(col1, abs(col1 - col2), fac);
-  outcol.a = col1.a;
-}
-
-void mix_dark(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  outcol.rgb = min(col1.rgb, col2.rgb * fac);
-  outcol.a = col1.a;
-}
-
-void mix_light(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  outcol.rgb = max(col1.rgb, col2.rgb * fac);
-  outcol.a = col1.a;
-}
-
-void mix_dodge(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  outcol = col1;
-
-  if (outcol.r != 0.0) {
-    float tmp = 1.0 - fac * col2.r;
-    if (tmp <= 0.0) {
-      outcol.r = 1.0;
-    }
-    else if ((tmp = outcol.r / tmp) > 1.0) {
-      outcol.r = 1.0;
-    }
-    else {
-      outcol.r = tmp;
-    }
-  }
-  if (outcol.g != 0.0) {
-    float tmp = 1.0 - fac * col2.g;
-    if (tmp <= 0.0) {
-      outcol.g = 1.0;
-    }
-    else if ((tmp = outcol.g / tmp) > 1.0) {
-      outcol.g = 1.0;
-    }
-    else {
-      outcol.g = tmp;
-    }
-  }
-  if (outcol.b != 0.0) {
-    float tmp = 1.0 - fac * col2.b;
-    if (tmp <= 0.0) {
-      outcol.b = 1.0;
-    }
-    else if ((tmp = outcol.b / tmp) > 1.0) {
-      outcol.b = 1.0;
-    }
-    else {
-      outcol.b = tmp;
-    }
-  }
-}
-
-void mix_burn(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  float tmp, facm = 1.0 - fac;
-
-  outcol = col1;
-
-  tmp = facm + fac * col2.r;
-  if (tmp <= 0.0) {
-    outcol.r = 0.0;
-  }
-  else if ((tmp = (1.0 - (1.0 - outcol.r) / tmp)) < 0.0) {
-    outcol.r = 0.0;
-  }
-  else if (tmp > 1.0) {
-    outcol.r = 1.0;
-  }
-  else {
-    outcol.r = tmp;
-  }
-
-  tmp = facm + fac * col2.g;
-  if (tmp <= 0.0) {
-    outcol.g = 0.0;
-  }
-  else if ((tmp = (1.0 - (1.0 - outcol.g) / tmp)) < 0.0) {
-    outcol.g = 0.0;
-  }
-  else if (tmp > 1.0) {
-    outcol.g = 1.0;
-  }
-  else {
-    outcol.g = tmp;
-  }
-
-  tmp = facm + fac * col2.b;
-  if (tmp <= 0.0) {
-    outcol.b = 0.0;
-  }
-  else if ((tmp = (1.0 - (1.0 - outcol.b) / tmp)) < 0.0) {
-    outcol.b = 0.0;
-  }
-  else if (tmp > 1.0) {
-    outcol.b = 1.0;
-  }
-  else {
-    outcol.b = tmp;
-  }
-}
-
-void mix_hue(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  float facm = 1.0 - fac;
-
-  outcol = col1;
-
-  vec4 hsv, hsv2, tmp;
-  rgb_to_hsv(col2, hsv2);
-
-  if (hsv2.y != 0.0) {
-    rgb_to_hsv(outcol, hsv);
-    hsv.x = hsv2.x;
-    hsv_to_rgb(hsv, tmp);
-
-    outcol = mix(outcol, tmp, fac);
-    outcol.a = col1.a;
-  }
-}
-
-void mix_sat(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  float facm = 1.0 - fac;
-
-  outcol = col1;
-
-  vec4 hsv, hsv2;
-  rgb_to_hsv(outcol, hsv);
-
-  if (hsv.y != 0.0) {
-    rgb_to_hsv(col2, hsv2);
-
-    hsv.y = facm * hsv.y + fac * hsv2.y;
-    hsv_to_rgb(hsv, outcol);
-  }
-}
-
-void mix_val(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  float facm = 1.0 - fac;
-
-  vec4 hsv, hsv2;
-  rgb_to_hsv(col1, hsv);
-  rgb_to_hsv(col2, hsv2);
-
-  hsv.z = facm * hsv.z + fac * hsv2.z;
-  hsv_to_rgb(hsv, outcol);
-}
-
-void mix_color(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  float facm = 1.0 - fac;
-
-  outcol = col1;
-
-  vec4 hsv, hsv2, tmp;
-  rgb_to_hsv(col2, hsv2);
-
-  if (hsv2.y != 0.0) {
-    rgb_to_hsv(outcol, hsv);
-    hsv.x = hsv2.x;
-    hsv.y = hsv2.y;
-    hsv_to_rgb(hsv, tmp);
-
-    outcol = mix(outcol, tmp, fac);
-    outcol.a = col1.a;
-  }
-}
-
-void mix_soft(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  float facm = 1.0 - fac;
-
-  vec4 one = vec4(1.0);
-  vec4 scr = one - (one - col2) * (one - col1);
-  outcol = facm * col1 + fac * ((one - col1) * col2 * col1 + col1 * scr);
-}
-
-void mix_linear(float fac, vec4 col1, vec4 col2, out vec4 outcol)
-{
-  fac = clamp(fac, 0.0, 1.0);
-
-  outcol = col1 + fac * (2.0 * (col2 - vec4(0.5)));
-}
-
-void valtorgb_opti_constant(
-    float fac, float edge, vec4 color1, vec4 color2, out vec4 outcol, out float outalpha)
-{
-  outcol = (fac > edge) ? color2 : color1;
-  outalpha = outcol.a;
-}
-
-void valtorgb_opti_linear(
-    float fac, vec2 mulbias, vec4 color1, vec4 color2, out vec4 outcol, out float outalpha)
-{
-  fac = clamp(fac * mulbias.x + mulbias.y, 0.0, 1.0);
-  outcol = mix(color1, color2, fac);
-  outalpha = outcol.a;
-}
-
-void valtorgb(float fac, sampler1DArray colormap, float layer, out vec4 outcol, out float outalpha)
-{
-  outcol = texture(colormap, vec2(fac, layer));
-  outalpha = outcol.a;
-}
-
-void valtorgb_nearest(
-    float fac, sampler1DArray colormap, float layer, out vec4 outcol, out float outalpha)
-{
-  fac = clamp(fac, 0.0, 1.0);
-  outcol = texelFetch(colormap, ivec2(fac * (textureSize(colormap, 0).x - 1), layer), 0);
-  outalpha = outcol.a;
-}
-
-void rgbtobw(vec4 color, out float outval)
-{
-  vec3 factors = vec3(0.2126, 0.7152, 0.0722);
-  outval = dot(color.rgb, factors);
-}
-
-void invert(float fac, vec4 col, out vec4 outcol)
-{
-  outcol.xyz = mix(col.xyz, vec3(1.0) - col.xyz, fac);
-  outcol.w = col.w;
-}
-
-void clamp_vec3(vec3 vec, vec3 min, vec3 max, out vec3 out_vec)
-{
-  out_vec = clamp(vec, min, max);
-}
-
-void clamp_value(float value, float min, float max, out float result)
-{
-  result = clamp(value, min, max);
-}
-
-void hue_sat(float hue, float sat, float value, float fac, vec4 col, out vec4 outcol)
-{
-  vec4 hsv;
-
-  rgb_to_hsv(col, hsv);
-
-  hsv[0] = fract(hsv[0] + hue + 0.5);
-  hsv[1] = clamp(hsv[1] * sat, 0.0, 1.0);
-  hsv[2] = hsv[2] * value;
-
-  hsv_to_rgb(hsv, outcol);
-
-  outcol = mix(col, outcol, fac);
-}
-
-void separate_rgb(vec4 col, out float r, out float g, out float b)
-{
-  r = col.r;
-  g = col.g;
-  b = col.b;
-}
-
-void combine_rgb(float r, float g, float b, out vec4 col)
-{
-  col = vec4(r, g, b, 1.0);
-}
-
-void separate_xyz(vec3 vec, out float x, out float y, out float z)
-{
-  x = vec.r;
-  y = vec.g;
-  z = vec.b;
-}
-
-void combine_xyz(float x, float y, float z, out vec3 vec)
-{
-  vec = vec3(x, y, z);
-}
-
-void separate_hsv(vec4 col, out float h, out float s, out float v)
-{
-  vec4 hsv;
-
-  rgb_to_hsv(col, hsv);
-  h = hsv[0];
-  s = hsv[1];
-  v = hsv[2];
-}
-
-void combine_hsv(float h, float s, float v, out vec4 col)
-{
-  hsv_to_rgb(vec4(h, s, v, 1.0), col);
-}
-
-void output_node(vec4 rgb, float alpha, out vec4 outrgb)
-{
-  outrgb = vec4(rgb.rgb, alpha);
-}
-
-/*********** TEXTURES ***************/
-
-void texco_norm(vec3 normal, out vec3 outnormal)
-{
-  /* corresponds to shi->orn, which is negated so cancels
-     out blender normal negation */
-  outnormal = normalize(normal);
-}
-
-vec3 mtex_2d_mapping(vec3 vec)
-{
-  return vec3(vec.xy * 0.5 + vec2(0.5), vec.z);
-}
-
-/** helper method to extract the upper left 3x3 matrix from a 4x4 matrix */
-mat3 to_mat3(mat4 m4)
-{
-  mat3 m3;
-  m3[0] = m4[0].xyz;
-  m3[1] = m4[1].xyz;
-  m3[2] = m4[2].xyz;
-  return m3;
-}
-
-/*********** NEW SHADER UTILITIES **************/
-
-float fresnel_dielectric_0(float eta)
-{
-  /* compute fresnel reflactance at normal incidence => cosi = 1.0 */
-  float A = (eta - 1.0) / (eta + 1.0);
-
-  return A * A;
-}
-
-float fresnel_dielectric_cos(float cosi, float eta)
-{
-  /* compute fresnel reflectance without explicitly computing
-   * the refracted direction */
-  float c = abs(cosi);
-  float g = eta * eta - 1.0 + c * c;
-  float result;
-
-  if (g > 0.0) {
-    g = sqrt(g);
-    float A = (g - c) / (g + c);
-    float B = (c * (g + c) - 1.0) / (c * (g - c) + 1.0);
-    result = 0.5 * A * A * (1.0 + B * B);
-  }
-  else {
-    result = 1.0; /* TIR (no refracted component) */
-  }
-
-  return result;
-}
-
-float fresnel_dielectric(vec3 Incoming, vec3 Normal, float eta)
-{
-  /* compute fresnel reflectance without explicitly computing
-   * the refracted direction */
-  return fresnel_dielectric_cos(dot(Incoming, Normal), eta);
-}
-
-float hypot(float x, float y)
-{
-  return sqrt(x * x + y * y);
-}
-
-void generated_from_orco(vec3 orco, out vec3 generated)
-{
-#ifdef VOLUMETRICS
-#  ifdef MESH_SHADER
-  generated = volumeObjectLocalCoord;
-#  else
-  generated = worldPosition;
-#  endif
-#else
-  generated = orco;
-#endif
-}
-
-int floor_to_int(float x)
-{
-  return int(floor(x));
-}
-
-int quick_floor(float x)
-{
-  return int(x) - ((x < 0) ? 1 : 0);
-}
-
-float integer_noise(int n)
-{
-  int nn;
-  n = (n + 1013) & 0x7fffffff;
-  n = (n >> 13) ^ n;
-  nn = (n * (n * n * 60493 + 19990303) + 1376312589) & 0x7fffffff;
-  return 0.5 * (float(nn) / 1073741824.0);
-}
-
-uint hash(uint kx, uint ky, uint kz)
-{
-#define rot(x, k) (((x) << (k)) | ((x) >> (32 - (k))))
-#define final(a, b, c) \
-  { \
-    c ^= b; \
-    c -= rot(b, 14); \
-    a ^= c; \
-    a -= rot(c, 11); \
-    b ^= a; \
-    b -= rot(a, 25); \
-    c ^= b; \
-    c -= rot(b, 16); \
-    a ^= c; \
-    a -= rot(c, 4); \
-    b ^= a; \
-    b -= rot(a, 14); \
-    c ^= b; \
-    c -= rot(b, 24); \
-  }
-  // now hash the data!
-  uint a, b, c, len = 3u;
-  a = b = c = 0xdeadbeefu + (len << 2u) + 13u;
-
-  c += kz;
-  b += ky;
-  a += kx;
-  final(a, b, c);
-
-  return c;
-#undef rot
-#undef final
-}
-
-uint hash(int kx, int ky, int kz)
-{
-  return hash(uint(kx), uint(ky), uint(kz));
-}
-
-float bits_to_01(uint bits)
-{
-  return (float(bits) / 4294967295.0);
-}
-
-float cellnoise(vec3 p)
-{
-  int ix = quick_floor(p.x);
-  int iy = quick_floor(p.y);
-  int iz = quick_floor(p.z);
-
-  return bits_to_01(hash(uint(ix), uint(iy), uint(iz)));
-}
-
-vec3 cellnoise_color(vec3 p)
-{
-  float r = cellnoise(p.xyz);
-  float g = cellnoise(p.yxz);
-  float b = cellnoise(p.yzx);
-
-  return vec3(r, g, b);
-}
-
-float floorfrac(float x, out int i)
-{
-  float x_floor = floor(x);
-  i = int(x_floor);
-  return x - x_floor;
-}
-
-/* bsdfs */
-
-vec3 tint_from_color(vec3 color)
-{
-  float lum = dot(color, vec3(0.3, 0.6, 0.1)); /* luminance approx. */
-  return (lum > 0) ? color / lum : vec3(1.0);  /* normalize lum. to isolate hue+sat */
-}
-
-void convert_metallic_to_specular_tinted(vec3 basecol,
-                                         vec3 basecol_tint,
-                                         float metallic,
-                                         float specular_fac,
-                                         float specular_tint,
-                                         out vec3 diffuse,
-                                         out vec3 f0)
-{
-  vec3 tmp_col = mix(vec3(1.0), basecol_tint, specular_tint);
-  f0 = mix((0.08 * specular_fac) * tmp_col, basecol, metallic);
-  diffuse = basecol * (1.0 - metallic);
-}
-
-vec3 principled_sheen(float NV, vec3 basecol_tint, float sheen_tint)
-{
-  float f = 1.0 - NV;
-  /* Temporary fix for T59784. Normal map seems to contain NaNs for tangent space normal maps,
-   * therefore we need to clamp value. */
-  f = clamp(f, 0.0, 1.0);
-  /* Empirical approximation (manual curve fitting). Can be refined. */
-  float sheen = f * f * f * 0.077 + f * 0.01 + 0.00026;
-  return sheen * mix(vec3(1.0), basecol_tint, sheen_tint);
-}
-
-#ifndef VOLUMETRICS
-void node_bsdf_diffuse(vec4 color, float roughness, vec3 N, out Closure result)
-{
-  N = normalize(N);
-  result = CLOSURE_DEFAULT;
-  eevee_closure_diffuse(N, color.rgb, 1.0, result.radiance);
-  closure_load_ssr_data(vec3(0.0), 0.0, N, viewCameraVec, -1, result);
-  result.radiance *= color.rgb;
-}
-
-void node_bsdf_glossy(vec4 color, float roughness, vec3 N, float ssr_id, out Closure result)
-{
-  N = normalize(N);
-  vec3 out_spec, ssr_spec;
-  eevee_closure_glossy(N, vec3(1.0), vec3(1.0), int(ssr_id), roughness, 1.0, out_spec, ssr_spec);
-  vec3 vN = mat3(ViewMatrix) * N;
-  result = CLOSURE_DEFAULT;
-  result.radiance = out_spec * color.rgb;
-  closure_load_ssr_data(ssr_spec * color.rgb, roughness, N, viewCameraVec, int(ssr_id), result);
-}
-
-void node_bsdf_anisotropic(vec4 color,
-                           float roughness,
-                           float anisotropy,
-                           float rotation,
-                           vec3 N,
-                           vec3 T,
-                           out Closure result)
-{
-  node_bsdf_glossy(color, roughness, N, -1, result);
-}
-
-void node_bsdf_glass(
-    vec4 color, float roughness, float ior, vec3 N, float ssr_id, out Closure result)
-{
-  N = normalize(N);
-  vec3 out_spec, out_refr, ssr_spec;
-  vec3 refr_color = (refractionDepth > 0.0) ? color.rgb * color.rgb :
-                                              color.rgb; /* Simulate 2 transmission event */
-  eevee_closure_glass(
-      N, vec3(1.0), vec3(1.0), int(ssr_id), roughness, 1.0, ior, out_spec, out_refr, ssr_spec);
-  out_refr *= refr_color;
-  out_spec *= color.rgb;
-  float fresnel = F_eta(ior, dot(N, cameraVec));
-  vec3 vN = mat3(ViewMatrix) * N;
-  result = CLOSURE_DEFAULT;
-  result.radiance = mix(out_refr, out_spec, fresnel);
-  closure_load_ssr_data(
-      ssr_spec * color.rgb * fresnel, roughness, N, viewCameraVec, int(ssr_id), result);
-}
-
-void node_bsdf_toon(vec4 color, float size, float tsmooth, vec3 N, out Closure result)
-{
-  node_bsdf_diffuse(color, 0.0, N, result);
-}
-
-void node_bsdf_principled(vec4 base_color,
-                          float subsurface,
-                          vec3 subsurface_radius,
-                          vec4 subsurface_color,
-                          float metallic,
-                          float specular,
-                          float specular_tint,
-                          float roughness,
-                          float anisotropic,
-                          float anisotropic_rotation,
-                          float sheen,
-                          float sheen_tint,
-                          float clearcoat,
-                          float clearcoat_roughness,
-                          float ior,
-                          float transmission,
-                          float transmission_roughness,
-                          vec4 emission,
-                          float alpha,
-                          vec3 N,
-                          vec3 CN,
-                          vec3 T,
-                          vec3 I,
-                          float ssr_id,
-                          float sss_id,
-                          vec3 sss_scale,
-                          out Closure result)
-{
-  N = normalize(N);
-  ior = max(ior, 1e-5);
-  metallic = saturate(metallic);
-  transmission = saturate(transmission);
-  float dielectric = 1.0 - metallic;
-  transmission *= dielectric;
-  sheen *= dielectric;
-  subsurface_color *= dielectric;
-
-  vec3 diffuse, f0, out_diff, out_spec, out_trans, out_refr, ssr_spec;
-  vec3 ctint = tint_from_color(base_color.rgb);
-  convert_metallic_to_specular_tinted(
-      base_color.rgb, ctint, metallic, specular, specular_tint, diffuse, f0);
-
-  float NV = dot(N, cameraVec);
-  vec3 out_sheen = sheen * principled_sheen(NV, ctint, sheen_tint);
-
-  /* Far from being accurate, but 2 glossy evaluation is too expensive.
-   * Most noticeable difference is at grazing angles since the bsdf lut
-   * f0 color interpolation is done on top of this interpolation. */
-  vec3 f0_glass = mix(vec3(1.0), base_color.rgb, specular_tint);
-  float fresnel = F_eta(ior, NV);
-  vec3 spec_col = F_color_blend(ior, fresnel, f0_glass) * fresnel;
-  f0 = mix(f0, spec_col, transmission);
-
-  vec3 f90 = mix(vec3(1.0), f0, (1.0 - specular) * metallic);
-
-  vec3 mixed_ss_base_color = mix(diffuse, subsurface_color.rgb, subsurface);
-
-  float sss_scalef = avg(sss_scale) * subsurface;
-  eevee_closure_principled(N,
-                           mixed_ss_base_color,
-                           f0,
-                           f90,
-                           int(ssr_id),
-                           roughness,
-                           CN,
-                           clearcoat * 0.25,
-                           clearcoat_roughness,
-                           1.0,
-                           sss_scalef,
-                           ior,
-                           out_diff,
-                           out_trans,
-                           out_spec,
-                           out_refr,
-                           ssr_spec);
-
-  vec3 refr_color = base_color.rgb;
-  refr_color *= (refractionDepth > 0.0) ? refr_color :
-                                          vec3(1.0); /* Simulate 2 transmission event */
-  out_refr *= refr_color * (1.0 - fresnel) * transmission;
-
-  result = CLOSURE_DEFAULT;
-  result.radiance = out_spec + out_refr;
-  result.radiance += out_diff * out_sheen; /* Coarse approx. */
-
-  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
-
-  vec3 sss_radiance = (out_diff + out_trans) * alpha;
-#  ifndef USE_SSS
-  result.radiance += sss_radiance * mixed_ss_base_color * (1.0 - transmission);
-#  else
-#    ifdef USE_SSS_ALBEDO
-  vec3 sss_albedo = mixed_ss_base_color;
-#    else
-  sss_radiance *= mixed_ss_base_color;
-#    endif
-  sss_radiance *= (1.0 - transmission);
-  closure_load_sss_data(sss_scalef,
-                        sss_radiance,
-#    ifdef USE_SSS_ALBEDO
-                        sss_albedo,
-#    endif
-                        int(sss_id),
-                        result);
-#  endif /* USE_SSS */
-
-  result.radiance += emission.rgb;
-  result.radiance *= alpha;
-  result.transmittance = vec3(1.0 - alpha);
-}
-
-void node_bsdf_principled_dielectric(vec4 base_color,
-                                     float subsurface,
-                                     vec3 subsurface_radius,
-                                     vec4 subsurface_color,
-                                     float metallic,
-                                     float specular,
-                                     float specular_tint,
-                                     float roughness,
-                                     float anisotropic,
-                                     float anisotropic_rotation,
-                                     float sheen,
-                                     float sheen_tint,
-                                     float clearcoat,
-                                     float clearcoat_roughness,
-                                     float ior,
-                                     float transmission,
-                                     float transmission_roughness,
-                                     vec4 emission,
-                                     float alpha,
-                                     vec3 N,
-                                     vec3 CN,
-                                     vec3 T,
-                                     vec3 I,
-                                     float ssr_id,
-                                     float sss_id,
-                                     vec3 sss_scale,
-                                     out Closure result)
-{
-  N = normalize(N);
-  metallic = saturate(metallic);
-  float dielectric = 1.0 - metallic;
-
-  vec3 diffuse, f0, out_diff, out_spec, ssr_spec;
-  vec3 ctint = tint_from_color(base_color.rgb);
-  convert_metallic_to_specular_tinted(
-      base_color.rgb, ctint, metallic, specular, specular_tint, diffuse, f0);
-
-  float NV = dot(N, cameraVec);
-  vec3 out_sheen = sheen * principled_sheen(NV, ctint, sheen_tint);
-
-  eevee_closure_default(
-      N, diffuse, f0, vec3(1.0), int(ssr_id), roughness, 1.0, out_diff, out_spec, ssr_spec);
-
-  result = CLOSURE_DEFAULT;
-  result.radiance = out_spec + out_diff * (diffuse + out_sheen);
-  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
-  result.radiance += emission.rgb;
-  result.radiance *= alpha;
-  result.transmittance = vec3(1.0 - alpha);
-}
-
-void node_bsdf_principled_metallic(vec4 base_color,
-                                   float subsurface,
-                                   vec3 subsurface_radius,
-                                   vec4 subsurface_color,
-                                   float metallic,
-                                   float specular,
-                                   float specular_tint,
-                                   float roughness,
-                                   float anisotropic,
-                                   float anisotropic_rotation,
-                                   float sheen,
-                                   float sheen_tint,
-                                   float clearcoat,
-                                   float clearcoat_roughness,
-                                   float ior,
-                                   float transmission,
-                                   float transmission_roughness,
-                                   vec4 emission,
-                                   float alpha,
-                                   vec3 N,
-                                   vec3 CN,
-                                   vec3 T,
-                                   vec3 I,
-                                   float ssr_id,
-                                   float sss_id,
-                                   vec3 sss_scale,
-                                   out Closure result)
-{
-  N = normalize(N);
-  vec3 out_spec, ssr_spec;
-
-  vec3 f90 = mix(vec3(1.0), base_color.rgb, (1.0 - specular) * metallic);
-
-  eevee_closure_glossy(N, base_color.rgb, f90, int(ssr_id), roughness, 1.0, out_spec, ssr_spec);
-
-  result = CLOSURE_DEFAULT;
-  result.radiance = out_spec;
-  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
-  result.radiance += emission.rgb;
-  result.radiance *= alpha;
-  result.transmittance = vec3(1.0 - alpha);
-}
-
-void node_bsdf_principled_clearcoat(vec4 base_color,
-                                    float subsurface,
-                                    vec3 subsurface_radius,
-                                    vec4 subsurface_color,
-                                    float metallic,
-                                    float specular,
-                                    float specular_tint,
-                                    float roughness,
-                                    float anisotropic,
-                                    float anisotropic_rotation,
-                                    float sheen,
-                                    float sheen_tint,
-                                    float clearcoat,
-                                    float clearcoat_roughness,
-                                    float ior,
-                                    float transmission,
-                                    float transmission_roughness,
-                                    vec4 emission,
-                                    float alpha,
-                                    vec3 N,
-                                    vec3 CN,
-                                    vec3 T,
-                                    vec3 I,
-                                    float ssr_id,
-                                    float sss_id,
-                                    vec3 sss_scale,
-                                    out Closure result)
-{
-  vec3 out_spec, ssr_spec;
-  N = normalize(N);
-
-  vec3 f90 = mix(vec3(1.0), base_color.rgb, (1.0 - specular) * metallic);
-
-  eevee_closure_clearcoat(N,
-                          base_color.rgb,
-                          f90,
-                          int(ssr_id),
-                          roughness,
-                          CN,
-                          clearcoat * 0.25,
-                          clearcoat_roughness,
-                          1.0,
-                          out_spec,
-                          ssr_spec);
-
-  result = CLOSURE_DEFAULT;
-  result.radiance = out_spec;
-  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
-  result.radiance += emission.rgb;
-  result.radiance *= alpha;
-  result.transmittance = vec3(1.0 - alpha);
-}
-
-void node_bsdf_principled_subsurface(vec4 base_color,
-                                     float subsurface,
-                                     vec3 subsurface_radius,
-                                     vec4 subsurface_color,
-                                     float metallic,
-                                     float specular,
-                                     float specular_tint,
-                                     float roughness,
-                                     float anisotropic,
-                                     float anisotropic_rotation,
-                                     float sheen,
-                                     float sheen_tint,
-                                     float clearcoat,
-                                     float clearcoat_roughness,
-                                     float ior,
-                                     float transmission,
-                                     float transmission_roughness,
-                                     vec4 emission,
-                                     float alpha,
-                                     vec3 N,
-                                     vec3 CN,
-                                     vec3 T,
-                                     vec3 I,
-                                     float ssr_id,
-                                     float sss_id,
-                                     vec3 sss_scale,
-                                     out Closure result)
-{
-  metallic = saturate(metallic);
-  N = normalize(N);
-
-  vec3 diffuse, f0, out_diff, out_spec, out_trans, ssr_spec;
-  vec3 ctint = tint_from_color(base_color.rgb);
-  convert_metallic_to_specular_tinted(
-      base_color.rgb, ctint, metallic, specular, specular_tint, diffuse, f0);
-
-  subsurface_color = subsurface_color * (1.0 - metallic);
-  vec3 mixed_ss_base_color = mix(diffuse, subsurface_color.rgb, subsurface);
-  float sss_scalef = avg(sss_scale) * subsurface;
-
-  float NV = dot(N, cameraVec);
-  vec3 out_sheen = sheen * principled_sheen(NV, ctint, sheen_tint);
-
-  vec3 f90 = mix(vec3(1.0), base_color.rgb, (1.0 - specular) * metallic);
-
-  eevee_closure_skin(N,
-                     mixed_ss_base_color,
-                     f0,
-                     f90,
-                     int(ssr_id),
-                     roughness,
-                     1.0,
-                     sss_scalef,
-                     out_diff,
-                     out_trans,
-                     out_spec,
-                     ssr_spec);
-
-  result = CLOSURE_DEFAULT;
-  result.radiance = out_spec;
-  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
-
-  vec3 sss_radiance = (out_diff + out_trans) * alpha;
-#  ifndef USE_SSS
-  result.radiance += sss_radiance * mixed_ss_base_color * (1.0 - transmission);
-#  else
-#    ifdef USE_SSS_ALBEDO
-  vec3 sss_albedo = mixed_ss_base_color;
-#    else
-  sss_radiance *= mixed_ss_base_color;
-#    endif
-  sss_radiance *= (1.0 - transmission);
-  closure_load_sss_data(sss_scalef,
-                        sss_radiance,
-#    ifdef USE_SSS_ALBEDO
-                        sss_albedo,
-#    endif
-                        int(sss_id),
-                        result);
-#  endif /* USE_SSS */
-
-  result.radiance += out_diff * out_sheen;
-  result.radiance += emission.rgb;
-  result.radiance *= alpha;
-  result.transmittance = vec3(1.0 - alpha);
-}
-
-void node_bsdf_principled_glass(vec4 base_color,
-                                float subsurface,
-                                vec3 subsurface_radius,
-                                vec4 subsurface_color,
-                                float metallic,
-                                float specular,
-                                float specular_tint,
-                                float roughness,
-                                float anisotropic,
-                                float anisotropic_rotation,
-                                float sheen,
-                                float sheen_tint,
-                                float clearcoat,
-                                float clearcoat_roughness,
-                                float ior,
-                                float transmission,
-                                float transmission_roughness,
-                                vec4 emission,
-                                float alpha,
-                                vec3 N,
-                                vec3 CN,
-                                vec3 T,
-                                vec3 I,
-                                float ssr_id,
-                                float sss_id,
-                                vec3 sss_scale,
-                                out Closure result)
-{
-  ior = max(ior, 1e-5);
-  N = normalize(N);
-
-  vec3 f0, out_spec, out_refr, ssr_spec;
-  f0 = mix(vec3(1.0), base_color.rgb, specular_tint);
-
-  eevee_closure_glass(
-      N, vec3(1.0), vec3(1.0), int(ssr_id), roughness, 1.0, ior, out_spec, out_refr, ssr_spec);
-
-  vec3 refr_color = base_color.rgb;
-  refr_color *= (refractionDepth > 0.0) ? refr_color :
-                                          vec3(1.0); /* Simulate 2 transmission events */
-  out_refr *= refr_color;
-
-  float fresnel = F_eta(ior, dot(N, cameraVec));
-  vec3 spec_col = F_color_blend(ior, fresnel, f0);
-  out_spec *= spec_col;
-  ssr_spec *= spec_col * fresnel;
-
-  result = CLOSURE_DEFAULT;
-  result.radiance = mix(out_refr, out_spec, fresnel);
-  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
-  result.radiance += emission.rgb;
-  result.radiance *= alpha;
-  result.transmittance = vec3(1.0 - alpha);
-}
-
-void node_bsdf_translucent(vec4 color, vec3 N, out Closure result)
-{
-  node_bsdf_diffuse(color, 0.0, -N, result);
-}
-
-void node_bsdf_transparent(vec4 color, out Closure result)
-{
-  result = CLOSURE_DEFAULT;
-  result.radiance = vec3(0.0);
-  result.transmittance = abs(color.rgb);
-}
-
-void node_bsdf_velvet(vec4 color, float sigma, vec3 N, out Closure result)
-{
-  node_bsdf_diffuse(color, 0.0, N, result);
-}
-
-void node_subsurface_scattering(vec4 color,
-                                float scale,
-                                vec3 radius,
-                                float sharpen,
-                                float texture_blur,
-                                vec3 N,
-                                float sss_id,
-                                out Closure result)
-{
-#  if defined(USE_SSS)
-  N = normalize(N);
-  vec3 out_diff, out_trans;
-  vec3 vN = mat3(ViewMatrix) * N;
-  result = CLOSURE_DEFAULT;
-  closure_load_ssr_data(vec3(0.0), 0.0, N, viewCameraVec, -1, result);
-
-  eevee_closure_subsurface(N, color.rgb, 1.0, scale, out_diff, out_trans);
-
-  vec3 sss_radiance = out_diff + out_trans;
-#    ifdef USE_SSS_ALBEDO
-  /* Not perfect for texture_blur not exactly equal to 0.0 or 1.0. */
-  vec3 sss_albedo = mix(color.rgb, vec3(1.0), texture_blur);
-  sss_radiance *= mix(vec3(1.0), color.rgb, texture_blur);
-#    else
-  sss_radiance *= color.rgb;
-#    endif
-  closure_load_sss_data(scale,
-                        sss_radiance,
-#    ifdef USE_SSS_ALBEDO
-                        sss_albedo,
-#    endif
-                        int(sss_id),
-                        result);
-#  else
-  node_bsdf_diffuse(color, 0.0, N, result);
-#  endif
-}
-
-void node_bsdf_refraction(vec4 color, float roughness, float ior, vec3 N, out Closure result)
-{
-  N = normalize(N);
-  vec3 out_refr;
-  color.rgb *= (refractionDepth > 0.0) ? color.rgb : vec3(1.0); /* Simulate 2 absorption event. */
-  eevee_closure_refraction(N, roughness, ior, out_refr);
-  vec3 vN = mat3(ViewMatrix) * N;
-  result = CLOSURE_DEFAULT;
-  result.ssr_normal = normal_encode(vN, viewCameraVec);
-  result.radiance = out_refr * color.rgb;
-}
-
-void node_ambient_occlusion(
-    vec4 color, float distance, vec3 normal, out vec4 result_color, out float result_ao)
-{
-  vec3 bent_normal;
-  vec4 rand = texelFetch(utilTex, ivec3(ivec2(gl_FragCoord.xy) % LUT_SIZE, 2.0), 0);
-  result_ao = occlusion_compute(normalize(normal), viewPosition, 1.0, rand, bent_normal);
-  result_color = result_ao * color;
-}
-
-void node_wireframe(float size, vec2 barycentric, vec3 barycentric_dist, out float fac)
-{
-  vec3 barys = barycentric.xyy;
-  barys.z = 1.0 - barycentric.x - barycentric.y;
-
-  size *= 0.5;
-  vec3 s = step(-size, -barys * barycentric_dist);
-
-  fac = max(s.x, max(s.y, s.z));
-}
-
-void node_wireframe_screenspace(float size, vec2 barycentric, out float fac)
-{
-  vec3 barys = barycentric.xyy;
-  barys.z = 1.0 - barycentric.x - barycentric.y;
-
-  size *= (1.0 / 3.0);
-  vec3 dx = dFdx(barys);
-  vec3 dy = dFdy(barys);
-  vec3 deltas = sqrt(dx * dx + dy * dy);
-
-  vec3 s = step(-deltas * size, -barys);
-
-  fac = max(s.x, max(s.y, s.z));
-}
-
-#else /* VOLUMETRICS */
-
-/* Stub all bsdf functions not compatible with volumetrics. */
-#  define node_bsdf_diffuse
-#  define node_bsdf_glossy
-#  define node_bsdf_anisotropic
-#  define node_bsdf_glass
-#  define node_bsdf_toon
-#  define node_bsdf_principled
-#  define node_bsdf_principled_dielectric
-#  define node_bsdf_principled_metallic
-#  define node_bsdf_principled_clearcoat
-#  define node_bsdf_principled_subsurface
-#  define node_bsdf_principled_glass
-#  define node_bsdf_translucent
-#  define node_bsdf_transparent
-#  define node_bsdf_velvet
-#  define node_subsurface_scattering
-#  define node_bsdf_refraction
-#  define node_ambient_occlusion
-#  define node_wireframe
-#  define node_wireframe_screenspace
-
-#endif /* VOLUMETRICS */
-
-/* emission */
-
-void node_emission(vec4 color, float strength, vec3 vN, out Closure result)
-{
-  result = CLOSURE_DEFAULT;
-#ifndef VOLUMETRICS
-  result.radiance = color.rgb * strength;
-  result.ssr_normal = normal_encode(vN, viewCameraVec);
-#else
-  result.emission = color.rgb * strength;
-#endif
-}
-
-/* background */
-
-void node_tex_environment_texco(vec3 viewvec, out vec3 worldvec)
-{
-#ifdef MESH_SHADER
-  worldvec = worldPosition;
-#else
-  vec4 v = (ProjectionMatrix[3][3] == 0.0) ? vec4(viewvec, 1.0) : vec4(0.0, 0.0, 1.0, 1.0);
-  vec4 co_homogenous = (ProjectionMatrixInverse * v);
-
-  vec3 co = co_homogenous.xyz / co_homogenous.w;
-#  if defined(WORLD_BACKGROUND) || defined(PROBE_CAPTURE)
-  worldvec = mat3(ViewMatrixInverse) * co;
-#  else
-  worldvec = mat3(ModelMatrixInverse) * (mat3(ViewMatrixInverse) * co);
-#  endif
-#endif
-}
-
-void node_background(vec4 color, float strength, out Closure result)
-{
-#ifndef VOLUMETRICS
-  color *= strength;
-  result = CLOSURE_DEFAULT;
-  result.radiance = color.rgb;
-  result.transmittance = vec3(0.0);
-#else
-  result = CLOSURE_DEFAULT;
-#endif
-}
-
-/* volumes */
-
-void node_volume_scatter(vec4 color, float density, float anisotropy, out Closure result)
-{
-#ifdef VOLUMETRICS
-  result = Closure(vec3(0.0), color.rgb * density, vec3(0.0), anisotropy);
-#else
-  result = CLOSURE_DEFAULT;
-#endif
-}
-
-void node_volume_absorption(vec4 color, float density, out Closure result)
-{
-#ifdef VOLUMETRICS
-  result = Closure((1.0 - color.rgb) * density, vec3(0.0), vec3(0.0), 0.0);
-#else
-  result = CLOSURE_DEFAULT;
-#endif
-}
-
-void node_blackbody(float temperature, sampler1DArray spectrummap, float layer, out vec4 color)
-{
-  if (temperature >= 12000.0) {
-    color = vec4(0.826270103, 0.994478524, 1.56626022, 1.0);
-  }
-  else if (temperature < 965.0) {
-    color = vec4(4.70366907, 0.0, 0.0, 1.0);
-  }
-  else {
-    float t = (temperature - 965.0) / (12000.0 - 965.0);
-    color = vec4(texture(spectrummap, vec2(t, layer)).rgb, 1.0);
-  }
-}
-
-void node_volume_principled(vec4 color,
-                            float density,
-                            float anisotropy,
-                            vec4 absorption_color,
-                            float emission_strength,
-                            vec4 emission_color,
-                            float blackbody_intensity,
-                            vec4 blackbody_tint,
-                            float temperature,
-                            float density_attribute,
-                            vec4 color_attribute,
-                            float temperature_attribute,
-                            sampler1DArray spectrummap,
-                            float layer,
-                            out Closure result)
-{
-#ifdef VOLUMETRICS
-  vec3 absorption_coeff = vec3(0.0);
-  vec3 scatter_coeff = vec3(0.0);
-  vec3 emission_coeff = vec3(0.0);
-
-  /* Compute density. */
-  density = max(density, 0.0);
-
-  if (density > 1e-5) {
-    density = max(density * density_attribute, 0.0);
-  }
-
-  if (density > 1e-5) {
-    /* Compute scattering and absorption coefficients. */
-    vec3 scatter_color = color.rgb * color_attribute.rgb;
-
-    scatter_coeff = scatter_color * density;
-    absorption_color.rgb = sqrt(max(absorption_color.rgb, 0.0));
-    absorption_coeff = max(1.0 - scatter_color, 0.0) * max(1.0 - absorption_color.rgb, 0.0) *
-                       density;
-  }
-
-  /* Compute emission. */
-  emission_strength = max(emission_strength, 0.0);
-
-  if (emission_strength > 1e-5) {
-    emission_coeff += emission_strength * emission_color.rgb;
-  }
-
-  if (blackbody_intensity > 1e-3) {
-    /* Add temperature from attribute. */
-    float T = max(temperature * max(temperature_attribute, 0.0), 0.0);
-
-    /* Stefan-Boltzman law. */
-    float T2 = T * T;
-    float T4 = T2 * T2;
-    float sigma = 5.670373e-8 * 1e-6 / M_PI;
-    float intensity = sigma * mix(1.0, T4, blackbody_intensity);
-
-    if (intensity > 1e-5) {
-      vec4 bb;
-      node_blackbody(T, spectrummap, layer, bb);
-      emission_coeff += bb.rgb * blackbody_tint.rgb * intensity;
-    }
-  }
-
-  result = Closure(absorption_coeff, scatter_coeff, emission_coeff, anisotropy);
-#else
-  result = CLOSURE_DEFAULT;
-#endif
-}
-
-void node_holdout(out Closure result)
-{
-  result = CLOSURE_DEFAULT;
-#ifndef VOLUMETRICS
-  result.holdout = 1.0;
-  result.flag = CLOSURE_HOLDOUT_FLAG;
-#endif
-}
-
-/* closures */
-
-void node_mix_shader(float fac, Closure shader1, Closure shader2, out Closure shader)
-{
-  shader = closure_mix(shader1, shader2, fac);
-}
-
-void node_add_shader(Closure shader1, Closure shader2, out Closure shader)
-{
-  shader = closure_add(shader1, shader2);
-}
-
-/* fresnel */
-
-void node_fresnel(float ior, vec3 N, vec3 I, out float result)
-{
-  N = normalize(N);
-  /* handle perspective/orthographic */
-  vec3 I_view = (ProjectionMatrix[3][3] == 0.0) ? normalize(I) : vec3(0.0, 0.0, -1.0);
-
-  float eta = max(ior, 0.00001);
-  result = fresnel_dielectric(I_view, N, (gl_FrontFacing) ? eta : 1.0 / eta);
-}
-
-/* layer_weight */
-
-void node_layer_weight(float blend, vec3 N, vec3 I, out float fresnel, out float facing)
-{
-  N = normalize(N);
-
-  /* fresnel */
-  float eta = max(1.0 - blend, 0.00001);
-  vec3 I_view = (ProjectionMatrix[3][3] == 0.0) ? normalize(I) : vec3(0.0, 0.0, -1.0);
-
-  fresnel = fresnel_dielectric(I_view, N, (gl_FrontFacing) ? 1.0 / eta : eta);
-
-  /* facing */
-  facing = abs(dot(I_view, N));
-  if (blend != 0.5) {
-    blend = clamp(blend, 0.0, 0.99999);
-    blend = (blend < 0.5) ? 2.0 * blend : 0.5 / (1.0 - blend);
-    facing = pow(facing, blend);
-  }
-  facing = 1.0 - facing;
-}
-
-/* gamma */
-
-void node_gamma(vec4 col, float gamma, out vec4 outcol)
-{
-  outcol = col;
-
-  if (col.r > 0.0) {
-    outcol.r = compatible_pow(col.r, gamma);
-  }
-  if (col.g > 0.0) {
-    outcol.g = compatible_pow(col.g, gamma);
-  }
-  if (col.b > 0.0) {
-    outcol.b = compatible_pow(col.b, gamma);
-  }
-}
-
-/* geometry */
-
-void node_attribute_volume_density(sampler3D tex, out vec4 outcol, out vec3 outvec, out float outf)
-{
-#if defined(MESH_SHADER) && defined(VOLUMETRICS)
-  vec3 cos = volumeObjectLocalCoord;
-#else
-  vec3 cos = vec3(0.0);
-#endif
-  outvec = texture(tex, cos).aaa;
-  outcol = vec4(outvec, 1.0);
-  outf = avg(outvec);
-}
-
-uniform vec3 volumeColor = vec3(1.0);
-
-void node_attribute_volume_color(sampler3D tex, out vec4 outcol, out vec3 outvec, out float outf)
-{
-#if defined(MESH_SHADER) && defined(VOLUMETRICS)
-  vec3 cos = volumeObjectLocalCoord;
-#else
-  vec3 cos = vec3(0.0);
-#endif
-
-  vec4 value = texture(tex, cos).rgba;
-  /* Density is premultiplied for interpolation, divide it out here. */
-  if (value.a > 1e-8) {
-    value.rgb /= value.a;
-  }
-
-  outvec = value.rgb * volumeColor;
-  outcol = vec4(outvec, 1.0);
-  outf = avg(outvec);
-}
-
-void node_attribute_volume_flame(sampler3D tex, out vec4 outcol, out vec3 outvec, out float outf)
-{
-#if defined(MESH_SHADER) && defined(VOLUMETRICS)
-  vec3 cos = volumeObjectLocalCoord;
-#else
-  vec3 cos = vec3(0.0);
-#endif
-  outf = texture(tex, cos).r;
-  outvec = vec3(outf, outf, outf);
-  outcol = vec4(outf, outf, outf, 1.0);
-}
-
-void node_attribute_volume_temperature(
-    sampler3D tex, vec2 temperature, out vec4 outcol, out vec3 outvec, out float outf)
-{
-#if defined(MESH_SHADER) && defined(VOLUMETRICS)
-  vec3 cos = volumeObjectLocalCoord;
-#else
-  vec3 cos = vec3(0.0);
-#endif
-  float flame = texture(tex, cos).r;
-
-  outf = (flame > 0.01) ? temperature.x + flame * (temperature.y - temperature.x) : 0.0;
-  outvec = vec3(outf, outf, outf);
-  outcol = vec4(outf, outf, outf, 1.0);
-}
-
-void node_attribute(vec3 attr, out vec4 outcol, out vec3 outvec, out float outf)
-{
-  outcol = vec4(attr, 1.0);
-  outvec = attr;
-  outf = avg(attr);
-}
-
-void node_uvmap(vec3 attr_uv, out vec3 outvec)
-{
-  outvec = attr_uv;
-}
-
-void tangent_orco_x(vec3 orco_in, out vec3 orco_out)
-{
-  orco_out = orco_in.xzy * vec3(0.0, -0.5, 0.5) + vec3(0.0, 0.25, -0.25);
-}
-
-void tangent_orco_y(vec3 orco_in, out vec3 orco_out)
-{
-  orco_out = orco_in.zyx * vec3(-0.5, 0.0, 0.5) + vec3(0.25, 0.0, -0.25);
-}
-
-void tangent_orco_z(vec3 orco_in, out vec3 orco_out)
-{
-  orco_out = orco_in.yxz * vec3(-0.5, 0.5, 0.0) + vec3(0.25, -0.25, 0.0);
-}
-
-void node_tangentmap(vec4 attr_tangent, out vec3 tangent)
-{
-  tangent = normalize(attr_tangent.xyz);
-}
-
-void node_tangent(vec3 N, vec3 orco, mat4 objmat, out vec3 T)
-{
-  T = (objmat * vec4(orco, 0.0)).xyz;
-  T = cross(N, normalize(cross(T, N)));
-}
-
-void node_geometry(vec3 I,
-                   vec3 N,
-                   vec3 orco,
-                   mat4 objmat,
-                   mat4 toworld,
-                   vec2 barycentric,
-                   out vec3 position,
-                   out vec3 normal,
-                   out vec3 tangent,
-                   out vec3 true_normal,
-                   out vec3 incoming,
-                   out vec3 parametric,
-                   out float backfacing,
-                   out float pointiness)
-{
-  /* handle perspective/orthographic */
-  vec3 I_view = (ProjectionMatrix[3][3] == 0.0) ? normalize(I) : vec3(0.0, 0.0, -1.0);
-  incoming = -(toworld * vec4(I_view, 0.0)).xyz;
-
-#if defined(WORLD_BACKGROUND) || defined(PROBE_CAPTURE)
-  position = -incoming;
-  true_normal = normal = incoming;
-  tangent = parametric = vec3(0.0);
-  vec3(0.0);
-  backfacing = 0.0;
-  pointiness = 0.0;
-#else
-
-  position = worldPosition;
-#  ifndef VOLUMETRICS
-  normal = normalize(N);
-  vec3 B = dFdx(worldPosition);
-  vec3 T = dFdy(worldPosition);
-  true_normal = normalize(cross(B, T));
-#  else
-  normal = (toworld * vec4(N, 0.0)).xyz;
-  true_normal = normal;
-#  endif
-  tangent_orco_z(orco, orco);
-  node_tangent(N, orco, objmat, tangent);
-
-  parametric = vec3(barycentric, 0.0);
-  backfacing = (gl_FrontFacing) ? 0.0 : 1.0;
-  pointiness = 0.5;
-#endif
-}
-
-void generated_texco(vec3 I, vec3 attr_orco, out vec3 generated)
-{
-  vec4 v = (ProjectionMatrix[3][3] == 0.0) ? vec4(I, 1.0) : vec4(0.0, 0.0, 1.0, 1.0);
-  vec4 co_homogenous = (ProjectionMatrixInverse * v);
-  vec4 co = vec4(co_homogenous.xyz / co_homogenous.w, 0.0);
-  co.xyz = normalize(co.xyz);
-#if defined(WORLD_BACKGROUND) || defined(PROBE_CAPTURE)
-  generated = (ViewMatrixInverse * co).xyz;
-#else
-  generated_from_orco(attr_orco, generated);
-#endif
-}
-
-void node_tex_coord(vec3 I,
-                    vec3 wN,
-                    mat4 obmatinv,
-                    vec4 camerafac,
-                    vec3 attr_orco,
-                    vec3 attr_uv,
-                    out vec3 generated,
-                    out vec3 normal,
-                    out vec3 uv,
-                    out vec3 object,
-                    out vec3 camera,
-                    out vec3 window,
-                    out vec3 reflection)
-{
-  generated = attr_orco;
-  normal = normalize(normal_world_to_object(wN));
-  uv = attr_uv;
-  object = (obmatinv * (ViewMatrixInverse * vec4(I, 1.0))).xyz;
-  camera = vec3(I.xy, -I.z);
-  vec4 projvec = ProjectionMatrix * vec4(I, 1.0);
-  window = vec3(mtex_2d_mapping(projvec.xyz / projvec.w).xy * camerafac.xy + camerafac.zw, 0.0);
-  reflection = -reflect(cameraVec, normalize(wN));
-}
-
-void node_tex_coord_background(vec3 I,
-                               vec3 N,
-                               mat4 obmatinv,
-                               vec4 camerafac,
-                               vec3 attr_orco,
-                               vec3 attr_uv,
-                               out vec3 generated,
-                               out vec3 normal,
-                               out vec3 uv,
-                               out vec3 object,
-                               out vec3 camera,
-                               out vec3 window,
-                               out vec3 reflection)
-{
-  vec4 v = (ProjectionMatrix[3][3] == 0.0) ? vec4(I, 1.0) : vec4(0.0, 0.0, 1.0, 1.0);
-  vec4 co_homogenous = (ProjectionMatrixInverse * v);
-
-  vec4 co = vec4(co_homogenous.xyz / co_homogenous.w, 0.0);
-
-  co = normalize(co);
-
-  vec3 coords = (ViewMatrixInverse * co).xyz;
-
-  generated = coords;
-  normal = -coords;
-  uv = vec3(attr_uv.xy, 0.0);
-  object = (obmatinv * vec4(coords, 1.0)).xyz;
-
-  camera = vec3(co.xy, -co.z);
-  window = vec3(mtex_2d_mapping(I).xy * camerafac.xy + camerafac.zw, 0.0);
-
-  reflection = -coords;
-}
-
-#if defined(WORLD_BACKGROUND) || (defined(PROBE_CAPTURE) && !defined(MESH_SHADER))
-#  define node_tex_coord node_tex_coord_background
-#endif
-
-/* textures */
-
-float calc_gradient(vec3 p, int gradient_type)
-{
-  float x, y, z;
-  x = p.x;
-  y = p.y;
-  z = p.z;
-  if (gradient_type == 0) { /* linear */
-    return x;
-  }
-  else if (gradient_type == 1) { /* quadratic */
-    float r = max(x, 0.0);
-    return r * r;
-  }
-  else if (gradient_type == 2) { /* easing */
-    float r = min(max(x, 0.0), 1.0);
-    float t = r * r;
-    return (3.0 * t - 2.0 * t * r);
-  }
-  else if (gradient_type == 3) { /* diagonal */
-    return (x + y) * 0.5;
-  }
-  else if (gradient_type == 4) { /* radial */
-    return atan(y, x) / (M_PI * 2) + 0.5;
-  }
-  else {
-    /* Bias a little bit for the case where p is a unit length vector,
-     * to get exactly zero instead of a small random value depending
-     * on float precision. */
-    float r = max(0.999999 - sqrt(x * x + y * y + z * z), 0.0);
-    if (gradient_type == 5) { /* quadratic sphere */
-      return r * r;
-    }
-    else if (gradient_type == 6) { /* sphere */
-      return r;
-    }
-  }
-  return 0.0;
-}
-
-void node_tex_gradient(vec3 co, float gradient_type, out vec4 color, out float fac)
-{
-  float f = calc_gradient(co, int(gradient_type));
-  f = clamp(f, 0.0, 1.0);
-
-  color = vec4(f, f, f, 1.0);
-  fac = f;
-}
-
-void node_tex_checker(
-    vec3 co, vec4 color1, vec4 color2, float scale, out vec4 color, out float fac)
-{
-  vec3 p = co * scale;
-
-  /* Prevent precision issues on unit coordinates. */
-  p = (p + 0.000001) * 0.999999;
-
-  int xi = int(abs(floor(p.x)));
-  int yi = int(abs(floor(p.y)));
-  int zi = int(abs(floor(p.z)));
-
-  bool check = ((mod(xi, 2) == mod(yi, 2)) == bool(mod(zi, 2)));
-
-  color = check ? color1 : color2;
-  fac = check ? 1.0 : 0.0;
-}
-
-vec2 calc_brick_texture(vec3 p,
-                        float mortar_size,
-                        float mortar_smooth,
-                        float bias,
-                        float brick_width,
-                        float row_height,
-                        float offset_amount,
-                        int offset_frequency,
-                        float squash_amount,
-                        int squash_frequency)
-{
-  int bricknum, rownum;
-  float offset = 0.0;
-  float x, y;
-
-  rownum = floor_to_int(p.y / row_height);
-
-  if (offset_frequency != 0 && squash_frequency != 0) {
-    brick_width *= (rownum % squash_frequency != 0) ? 1.0 : squash_amount;           /* squash */
-    offset = (rownum % offset_frequency != 0) ? 0.0 : (brick_width * offset_amount); /* offset */
-  }
-
-  bricknum = floor_to_int((p.x + offset) / brick_width);
-
-  x = (p.x + offset) - brick_width * bricknum;
-  y = p.y - row_height * rownum;
-
-  float tint = clamp((integer_noise((rownum << 16) + (bricknum & 0xFFFF)) + bias), 0.0, 1.0);
-
-  float min_dist = min(min(x, y), min(brick_width - x, row_height - y));
-  if (min_dist >= mortar_size) {
-    return vec2(tint, 0.0);
-  }
-  else if (mortar_smooth == 0.0) {
-    return vec2(tint, 1.0);
-  }
-  else {
-    min_dist = 1.0 - min_dist / mortar_size;
-    return vec2(tint, smoothstep(0.0, mortar_smooth, min_dist));
-  }
-}
-
-void node_tex_brick(vec3 co,
-                    vec4 color1,
-                    vec4 color2,
-                    vec4 mortar,
-                    float scale,
-                    float mortar_size,
-                    float mortar_smooth,
-                    float bias,
-                    float brick_width,
-                    float row_height,
-                    float offset_amount,
-                    float offset_frequency,
-                    float squash_amount,
-                    float squash_frequency,
-                    out vec4 color,
-                    out float fac)
-{
-  vec2 f2 = calc_brick_texture(co * scale,
-                               mortar_size,
-                               mortar_smooth,
-                               bias,
-                               brick_width,
-                               row_height,
-                               offset_amount,
-                               int(offset_frequency),
-                               squash_amount,
-                               int(squash_frequency));
-  float tint = f2.x;
-  float f = f2.y;
-  if (f != 1.0) {
-    float facm = 1.0 - tint;
-    color1 = facm * color1 + tint * color2;
-  }
-  color = mix(color1, mortar, f);
-  fac = f;
-}
-
-void node_tex_clouds(vec3 co, float size, out vec4 color, out float fac)
-{
-  color = vec4(1.0);
-  fac = 1.0;
-}
-
-void node_tex_environment_equirectangular(vec3 co, float clamp_size, sampler2D ima, out vec3 uv)
-{
-  vec3 nco = normalize(co);
-  uv.x = -atan(nco.y, nco.x) / (2.0 * M_PI) + 0.5;
-  uv.y = atan(nco.z, hypot(nco.x, nco.y)) / M_PI + 0.5;
-
-  /* Fix pole bleeding */
-  float half_height = clamp_size / float(textureSize(ima, 0).y);
-  uv.y = clamp(uv.y, half_height, 1.0 - half_height);
-  uv.z = 0.0;
-}
-
-void node_tex_environment_mirror_ball(vec3 co, out vec3 uv)
-{
-  vec3 nco = normalize(co);
-  nco.y -= 1.0;
-
-  float div = 2.0 * sqrt(max(-0.5 * nco.y, 0.0));
-  nco /= max(1e-8, div);
-
-  uv = 0.5 * nco.xzz + 0.5;
-}
-
-void node_tex_environment_empty(vec3 co, out vec4 color)
-{
-  color = vec4(1.0, 0.0, 1.0, 1.0);
-}
-
-/* 16bits floats limits. Higher/Lower values produce +/-inf. */
-#define safe_color(a) (clamp(a, -65520.0, 65520.0))
-
-void tex_color_alpha_clear(vec4 color, out vec4 result)
-{
-  result = vec4(color.rgb, 1.0);
-}
-
-void tex_color_alpha_premultiply(vec4 color, out vec4 result)
-{
-  result = vec4(color.rgb * color.a, 1.0);
-}
-
-void tex_color_alpha_unpremultiply(vec4 color, out vec4 result)
-{
-  if (color.a == 0.0 || color.a == 1.0) {
-    result = vec4(color.rgb, 1.0);
-  }
-  else {
-    result = vec4(color.rgb / color.a, 1.0);
-  }
-}
-
-void node_tex_image_linear(vec3 co, sampler2D ima, out vec4 color, out float alpha)
-{
-  color = safe_color(texture(ima, co.xy));
-  alpha = color.a;
-}
-
-void node_tex_image_linear_no_mip(vec3 co, sampler2D ima, out vec4 color, out float alpha)
-{
-  color = safe_color(textureLod(ima, co.xy, 0.0));
-  alpha = color.a;
-}
-
-void node_tex_image_nearest(vec3 co, sampler2D ima, out vec4 color, out float alpha)
-{
-  ivec2 pix = ivec2(fract(co.xy) * textureSize(ima, 0).xy);
-  color = safe_color(texelFetch(ima, pix, 0));
-  alpha = color.a;
-}
-
-/* @arg f: signed distance to texel center. */
-void cubic_bspline_coefs(vec2 f, out vec2 w0, out vec2 w1, out vec2 w2, out vec2 w3)
-{
-  vec2 f2 = f * f;
-  vec2 f3 = f2 * f;
-  /* Bspline coefs (optimized) */
-  w3 = f3 / 6.0;
-  w0 = -w3 + f2 * 0.5 - f * 0.5 + 1.0 / 6.0;
-  w1 = f3 * 0.5 - f2 * 1.0 + 2.0 / 3.0;
-  w2 = 1.0 - w0 - w1 - w3;
-}
-
-void node_tex_image_cubic_ex(
-    vec3 co, sampler2D ima, float do_extend, out vec4 color, out float alpha)
-{
-  vec2 tex_size = vec2(textureSize(ima, 0).xy);
-
-  co.xy *= tex_size;
-  /* texel center */
-  vec2 tc = floor(co.xy - 0.5) + 0.5;
-  vec2 w0, w1, w2, w3;
-  cubic_bspline_coefs(co.xy - tc, w0, w1, w2, w3);
-
-#if 1 /* Optimized version using 4 filtered tap. */
-  vec2 s0 = w0 + w1;
-  vec2 s1 = w2 + w3;
-
-  vec2 f0 = w1 / (w0 + w1);
-  vec2 f1 = w3 / (w2 + w3);
-
-  vec4 final_co;
-  final_co.xy = tc - 1.0 + f0;
-  final_co.zw = tc + 1.0 + f1;
-
-  if (do_extend == 1.0) {
-    final_co = clamp(final_co, vec4(0.5), tex_size.xyxy - 0.5);
-  }
-  final_co /= tex_size.xyxy;
-
-  color = safe_color(textureLod(ima, final_co.xy, 0.0)) * s0.x * s0.y;
-  color += safe_color(textureLod(ima, final_co.zy, 0.0)) * s1.x * s0.y;
-  color += safe_color(textureLod(ima, final_co.xw, 0.0)) * s0.x * s1.y;
-  color += safe_color(textureLod(ima, final_co.zw, 0.0)) * s1.x * s1.y;
-
-#else /* Reference bruteforce 16 tap. */
-  color = texelFetch(ima, ivec2(tc + vec2(-1.0, -1.0)), 0) * w0.x * w0.y;
-  color += texelFetch(ima, ivec2(tc + vec2(0.0, -1.0)), 0) * w1.x * w0.y;
-  color += texelFetch(ima, ivec2(tc + vec2(1.0, -1.0)), 0) * w2.x * w0.y;
-  color += texelFetch(ima, ivec2(tc + vec2(2.0, -1.0)), 0) * w3.x * w0.y;
-
-  color += texelFetch(ima, ivec2(tc + vec2(-1.0, 0.0)), 0) * w0.x * w1.y;
-  color += texelFetch(ima, ivec2(tc + vec2(0.0, 0.0)), 0) * w1.x * w1.y;
-  color += texelFetch(ima, ivec2(tc + vec2(1.0, 0.0)), 0) * w2.x * w1.y;
-  color += texelFetch(ima, ivec2(tc + vec2(2.0, 0.0)), 0) * w3.x * w1.y;
-
-  color += texelFetch(ima, ivec2(tc + vec2(-1.0, 1.0)), 0) * w0.x * w2.y;
-  color += texelFetch(ima, ivec2(tc + vec2(0.0, 1.0)), 0) * w1.x * w2.y;
-  color += texelFetch(ima, ivec2(tc + vec2(1.0, 1.0)), 0) * w2.x * w2.y;
-  color += texelFetch(ima, ivec2(tc + vec2(2.0, 1.0)), 0) * w3.x * w2.y;
-
-  color += texelFetch(ima, ivec2(tc + vec2(-1.0, 2.0)), 0) * w0.x * w3.y;
-  color += texelFetch(ima, ivec2(tc + vec2(0.0, 2.0)), 0) * w1.x * w3.y;
-  color += texelFetch(ima, ivec2(tc + vec2(1.0, 2.0)), 0) * w2.x * w3.y;
-  color += texelFetch(ima, ivec2(tc + vec2(2.0, 2.0)), 0) * w3.x * w3.y;
-#endif
-
-  alpha = color.a;
-}
-
-void node_tex_image_cubic(vec3 co, sampler2D ima, out vec4 color, out float alpha)
-{
-  node_tex_image_cubic_ex(co, ima, 0.0, color, alpha);
-}
-
-void node_tex_image_cubic_extend(vec3 co, sampler2D ima, out vec4 color, out float alpha)
-{
-  node_tex_image_cubic_ex(co, ima, 1.0, color, alpha);
-}
-
-void node_tex_image_smart(vec3 co, sampler2D ima, out vec4 color, out float alpha)
-{
-  /* use cubic for now */
-  node_tex_image_cubic_ex(co, ima, 0.0, color, alpha);
-}
-
-void tex_box_sample_linear(
-    vec3 texco, vec3 N, sampler2D ima, out vec4 color1, out vec4 color2, out vec4 color3)
-{
-  /* X projection */
-  vec2 uv = texco.yz;
-  if (N.x < 0.0) {
-    uv.x = 1.0 - uv.x;
-  }
-  color1 = texture(ima, uv);
-  /* Y projection */
-  uv = texco.xz;
-  if (N.y > 0.0) {
-    uv.x = 1.0 - uv.x;
-  }
-  color2 = texture(ima, uv);
-  /* Z projection */
-  uv = texco.yx;
-  if (N.z > 0.0) {
-    uv.x = 1.0 - uv.x;
-  }
-  color3 = texture(ima, uv);
-}
-
-void tex_box_sample_nearest(
-    vec3 texco, vec3 N, sampler2D ima, out vec4 color1, out vec4 color2, out vec4 color3)
-{
-  /* X projection */
-  vec2 uv = texco.yz;
-  if (N.x < 0.0) {
-    uv.x = 1.0 - uv.x;
-  }
-  ivec2 pix = ivec2(uv.xy * textureSize(ima, 0).xy);
-  color1 = texelFetch(ima, pix, 0);
-  /* Y projection */
-  uv = texco.xz;
-  if (N.y > 0.0) {
-    uv.x = 1.0 - uv.x;
-  }
-  pix = ivec2(uv.xy * textureSize(ima, 0).xy);
-  color2 = texelFetch(ima, pix, 0);
-  /* Z projection */
-  uv = texco.yx;
-  if (N.z > 0.0) {
-    uv.x = 1.0 - uv.x;
-  }
-  pix = ivec2(uv.xy * textureSize(ima, 0).xy);
-  color3 = texelFetch(ima, pix, 0);
-}
-
-void tex_box_sample_cubic(
-    vec3 texco, vec3 N, sampler2D ima, out vec4 color1, out vec4 color2, out vec4 color3)
-{
-  float alpha;
-  /* X projection */
-  vec2 uv = texco.yz;
-  if (N.x < 0.0) {
-    uv.x = 1.0 - uv.x;
-  }
-  node_tex_image_cubic_ex(uv.xyy, ima, 0.0, color1, alpha);
-  /* Y projection */
-  uv = texco.xz;
-  if (N.y > 0.0) {
-    uv.x = 1.0 - uv.x;
-  }
-  node_tex_image_cubic_ex(uv.xyy, ima, 0.0, color2, alpha);
-  /* Z projection */
-  uv = texco.yx;
-  if (N.z > 0.0) {
-    uv.x = 1.0 - uv.x;
-  }
-  node_tex_image_cubic_ex(uv.xyy, ima, 0.0, color3, alpha);
-}
-
-void tex_box_sample_smart(
-    vec3 texco, vec3 N, sampler2D ima, out vec4 color1, out vec4 color2, out vec4 color3)
-{
-  tex_box_sample_cubic(texco, N, ima, color1, color2, color3);
-}
-
-void node_tex_image_box(vec3 texco,
-                        vec3 N,
-                        vec4 color1,
-                        vec4 color2,
-                        vec4 color3,
-                        sampler2D ima,
-                        float blend,
-                        out vec4 color,
-                        out float alpha)
-{
-  /* project from direction vector to barycentric coordinates in triangles */
-  N = abs(N);
-  N /= dot(N, vec3(1.0));
-
-  /* basic idea is to think of this as a triangle, each corner representing
-   * one of the 3 faces of the cube. in the corners we have single textures,
-   * in between we blend between two textures, and in the middle we a blend
-   * between three textures.
-   *
-   * the Nxyz values are the barycentric coordinates in an equilateral
-   * triangle, which in case of blending, in the middle has a smaller
-   * equilateral triangle where 3 textures blend. this divides things into
-   * 7 zones, with an if () test for each zone
-   * EDIT: Now there is only 4 if's. */
-
-  float limit = 0.5 + 0.5 * blend;
-
-  vec3 weight;
-  weight = N.xyz / (N.xyx + N.yzz);
-  weight = clamp((weight - 0.5 * (1.0 - blend)) / max(1e-8, blend), 0.0, 1.0);
-
-  /* test for mixes between two textures */
-  if (N.z < (1.0 - limit) * (N.y + N.x)) {
-    weight.z = 0.0;
-    weight.y = 1.0 - weight.x;
-  }
-  else if (N.x < (1.0 - limit) * (N.y + N.z)) {
-    weight.x = 0.0;
-    weight.z = 1.0 - weight.y;
-  }
-  else if (N.y < (1.0 - limit) * (N.x + N.z)) {
-    weight.y = 0.0;
-    weight.x = 1.0 - weight.z;
-  }
-  else {
-    /* last case, we have a mix between three */
-    weight = ((2.0 - limit) * N + (limit - 1.0)) / max(1e-8, blend);
-  }
-
-  color = weight.x * color1 + weight.y * color2 + weight.z * color3;
-  alpha = color.a;
-}
-
-void tex_clip_linear(vec3 co, sampler2D ima, vec4 icolor, out vec4 color, out float alpha)
-{
-  vec2 tex_size = vec2(textureSize(ima, 0).xy);
-  vec2 minco = min(co.xy, 1.0 - co.xy);
-  minco = clamp(minco * tex_size + 0.5, 0.0, 1.0);
-  float fac = minco.x * minco.y;
-
-  color = mix(vec4(0.0), icolor, fac);
-  alpha = color.a;
-}
-
-void tex_clip_nearest(vec3 co, sampler2D ima, vec4 icolor, out vec4 color, out float alpha)
-{
-  vec4 minco = vec4(co.xy, 1.0 - co.xy);
-  color = (any(lessThan(minco, vec4(0.0)))) ? vec4(0.0) : icolor;
-  alpha = color.a;
-}
-
-void tex_clip_cubic(vec3 co, sampler2D ima, vec4 icolor, out vec4 color, out float alpha)
-{
-  vec2 tex_size = vec2(textureSize(ima, 0).xy);
-
-  co.xy *= tex_size;
-  /* texel center */
-  vec2 tc = floor(co.xy - 0.5) + 0.5;
-  vec2 w0, w1, w2, w3;
-  cubic_bspline_coefs(co.xy - tc, w0, w1, w2, w3);
-
-  /* TODO Optimize this part. I'm sure there is a smarter way to do that.
-   * Could do that when sampling? */
-#define CLIP_CUBIC_SAMPLE(samp, size) \
-  (float(all(greaterThan(samp, vec2(-0.5)))) * float(all(lessThan(ivec2(samp), itex_size))))
-  ivec2 itex_size = textureSize(ima, 0).xy;
-  float fac;
-  fac = CLIP_CUBIC_SAMPLE(tc + vec2(-1.0, -1.0), itex_size) * w0.x * w0.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(0.0, -1.0), itex_size) * w1.x * w0.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(1.0, -1.0), itex_size) * w2.x * w0.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(2.0, -1.0), itex_size) * w3.x * w0.y;
-
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(-1.0, 0.0), itex_size) * w0.x * w1.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(0.0, 0.0), itex_size) * w1.x * w1.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(1.0, 0.0), itex_size) * w2.x * w1.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(2.0, 0.0), itex_size) * w3.x * w1.y;
-
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(-1.0, 1.0), itex_size) * w0.x * w2.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(0.0, 1.0), itex_size) * w1.x * w2.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(1.0, 1.0), itex_size) * w2.x * w2.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(2.0, 1.0), itex_size) * w3.x * w2.y;
-
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(-1.0, 2.0), itex_size) * w0.x * w3.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(0.0, 2.0), itex_size) * w1.x * w3.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(1.0, 2.0), itex_size) * w2.x * w3.y;
-  fac += CLIP_CUBIC_SAMPLE(tc + vec2(2.0, 2.0), itex_size) * w3.x * w3.y;
-#undef CLIP_CUBIC_SAMPLE
-
-  color = mix(vec4(0.0), icolor, fac);
-  alpha = color.a;
-}
-
-void tex_clip_smart(vec3 co, sampler2D ima, vec4 icolor, out vec4 color, out float alpha)
-{
-  tex_clip_cubic(co, ima, icolor, color, alpha);
-}
-
-void node_tex_image_empty(vec3 co, out vec4 color, out float alpha)
-{
-  color = vec4(0.0);
-  alpha = 0.0;
-}
-
-void node_tex_magic(
-    vec3 co, float scale, float distortion, float depth, out vec4 color, out float fac)
-{
-  vec3 p = co * scale;
-  float x = sin((p.x + p.y + p.z) * 5.0);
-  float y = cos((-p.x + p.y - p.z) * 5.0);
-  float z = -cos((-p.x - p.y + p.z) * 5.0);
-
-  if (depth > 0) {
-    x *= distortion;
-    y *= distortion;
-    z *= distortion;
-    y = -cos(x - y + z);
-    y *= distortion;
-    if (depth > 1) {
-      x = cos(x - y - z);
-      x *= distortion;
-      if (depth > 2) {
-        z = sin(-x - y - z);
-        z *= distortion;
-        if (depth > 3) {
-          x = -cos(-x + y - z);
-          x *= distortion;
-          if (depth > 4) {
-            y = -sin(-x + y + z);
-            y *= distortion;
-            if (depth > 5) {
-              y = -cos(-x + y + z);
-              y *= distortion;
-              if (depth > 6) {
-                x = cos(x + y + z);
-                x *= distortion;
-                if (depth > 7) {
-                  z = sin(x + y - z);
-                  z *= distortion;
-                  if (depth > 8) {
-                    x = -cos(-x - y + z);
-                    x *= distortion;
-                    if (depth > 9) {
-                      y = -sin(x - y + z);
-                      y *= distortion;
-                    }
-                  }
-                }
-              }
-            }
-          }
-        }
-      }
-    }
-  }
-  if (distortion != 0.0) {
-    distortion *= 2.0;
-    x /= distortion;
-    y /= distortion;
-    z /= distortion;
-  }
-
-  color = vec4(0.5 - x, 0.5 - y, 0.5 - z, 1.0);
-  fac = (color.x + color.y + color.z) / 3.0;
-}
-
-float noise_fade(float t)
-{
-  return t * t * t * (t * (t * 6.0 - 15.0) + 10.0);
-}
-
-float noise_scale3(float result)
-{
-  return 0.9820 * result;
-}
-
-float noise_nerp(float t, float a, float b)
-{
-  return (1.0 - t) * a + t * b;
-}
-
-float noise_grad(uint hash, float x, float y, float z)
-{
-  uint h = hash & 15u;
-  float u = h < 8u ? x : y;
-  float vt = ((h == 12u) || (h == 14u)) ? x : z;
-  float v = h < 4u ? y : vt;
-  return (((h & 1u) != 0u) ? -u : u) + (((h & 2u) != 0u) ? -v : v);
-}
-
-float noise_perlin(float x, float y, float z)
-{
-  int X;
-  float fx = floorfrac(x, X);
-  int Y;
-  float fy = floorfrac(y, Y);
-  int Z;
-  float fz = floorfrac(z, Z);
-
-  float u = noise_fade(fx);
-  float v = noise_fade(fy);
-  float w = noise_fade(fz);
-
-  float noise_u[2], noise_v[2];
-
-  noise_u[0] = noise_nerp(
-      u, noise_grad(hash(X, Y, Z), fx, fy, fz), noise_grad(hash(X + 1, Y, Z), fx - 1.0, fy, fz));
-
-  noise_u[1] = noise_nerp(u,
-                          noise_grad(hash(X, Y + 1, Z), fx, fy - 1.0, fz),
-                          noise_grad(hash(X + 1, Y + 1, Z), fx - 1.0, fy - 1.0, fz));
-
-  noise_v[0] = noise_nerp(v, noise_u[0], noise_u[1]);
-
-  noise_u[0] = noise_nerp(u,
-                          noise_grad(hash(X, Y, Z + 1), fx, fy, fz - 1.0),
-                          noise_grad(hash(X + 1, Y, Z + 1), fx - 1.0, fy, fz - 1.0));
-
-  noise_u[1] = noise_nerp(u,
-                          noise_grad(hash(X, Y + 1, Z + 1), fx, fy - 1.0, fz - 1.0),
-                          noise_grad(hash(X + 1, Y + 1, Z + 1), fx - 1.0, fy - 1.0, fz - 1.0));
-
-  noise_v[1] = noise_nerp(v, noise_u[0], noise_u[1]);
-
-  float r = noise_scale3(noise_nerp(w, noise_v[0], noise_v[1]));
-
-  return (isinf(r)) ? 0.0 : r;
-}
-
-float noise(vec3 p)
-{
-  return 0.5 * noise_perlin(p.x, p.y, p.z) + 0.5;
-}
-
-float snoise(vec3 p)
-{
-  return noise_perlin(p.x, p.y, p.z);
-}
-
-float noise_turbulence(vec3 p, float octaves, int hard)
-{
-  float fscale = 1.0;
-  float amp = 1.0;
-  float sum = 0.0;
-  octaves = clamp(octaves, 0.0, 16.0);
-  int n = int(octaves);
-  for (int i = 0; i <= n; i++) {
-    float t = noise(fscale * p);
-    if (hard != 0) {
-      t = abs(2.0 * t - 1.0);
-    }
-    sum += t * amp;
-    amp *= 0.5;
-    fscale *= 2.0;
-  }
-  float rmd = octaves - floor(octaves);
-  if (rmd != 0.0) {
-    float t = noise(fscale * p);
-    if (hard != 0) {
-      t = abs(2.0 * t - 1.0);
-    }
-    float sum2 = sum + t * amp;
-    sum *= (float(1 << n) / float((1 << (n + 1)) - 1));
-    sum2 *= (float(1 << (n + 1)) / float((1 << (n + 2)) - 1));
-    return (1.0 - rmd) * sum + rmd * sum2;
-  }
-  else {
-    sum *= (float(1 << n) / float((1 << (n + 1)) - 1));
-    return sum;
-  }
-}
-
-void node_tex_noise(
-    vec3 co, float scale, float detail, float distortion, out vec4 color, out float fac)
-{
-  vec3 p = co * scale;
-  int hard = 0;
-  if (distortion != 0.0) {
-    vec3 r, offset = vec3(13.5, 13.5, 13.5);
-    r.x = noise(p + offset) * distortion;
-    r.y = noise(p) * distortion;
-    r.z = noise(p - offset) * distortion;
-    p += r;
-  }
-
-  fac = noise_turbulence(p, detail, hard);
-  color = vec4(fac,
-               noise_turbulence(vec3(p.y, p.x, p.z), detail, hard),
-               noise_turbulence(vec3(p.y, p.z, p.x), detail, hard),
-               1);
-}
-
-/* Musgrave fBm
- *
- * H: fractal increment parameter
- * lacunarity: gap between successive frequencies
- * octaves: number of frequencies in the fBm
- *
- * from "Texturing and Modelling: A procedural approach"
- */
-
-float noise_musgrave_fBm(vec3 p, float H, float lacunarity, float octaves)
-{
-  float rmd;
-  float value = 0.0;
-  float pwr = 1.0;
-  float pwHL = pow(lacunarity, -H);
-
-  for (int i = 0; i < int(octaves); i++) {
-    value += snoise(p) * pwr;
-    pwr *= pwHL;
-    p *= lacunarity;
-  }
-
-  rmd = octaves - floor(octaves);
-  if (rmd != 0.0) {
-    value += rmd * snoise(p) * pwr;
-  }
-
-  return value;
-}
-
-/* Musgrave Multifractal
- *
- * H: highest fractal dimension
- * lacunarity: gap between successive frequencies
- * octaves: number of frequencies in the fBm
- */
-
-float noise_musgrave_multi_fractal(vec3 p, float H, float lacunarity, float octaves)
-{
-  float rmd;
-  float value = 1.0;
-  float pwr = 1.0;
-  float pwHL = pow(lacunarity, -H);
-
-  for (int i = 0; i < int(octaves); i++) {
-    value *= (pwr * snoise(p) + 1.0);
-    pwr *= pwHL;
-    p *= lacunarity;
-  }
-
-  rmd = octaves - floor(octaves);
-  if (rmd != 0.0) {
-    value *= (rmd * pwr * snoise(p) + 1.0); /* correct? */
-  }
-
-  return value;
-}
-
-/* Musgrave Heterogeneous Terrain
- *
- * H: fractal dimension of the roughest area
- * lacunarity: gap between successive frequencies
- * octaves: number of frequencies in the fBm
- * offset: raises the terrain from `sea level'
- */
-
-float noise_musgrave_hetero_terrain(vec3 p, float H, float lacunarity, float octaves, float offset)
-{
-  float value, increment, rmd;
-  float pwHL = pow(lacunarity, -H);
-  float pwr = pwHL;
-
-  /* first unscaled octave of function; later octaves are scaled */
-  value = offset + snoise(p);
-  p *= lacunarity;
-
-  for (int i = 1; i < int(octaves); i++) {
-    increment = (snoise(p) + offset) * pwr * value;
-    value += increment;
-    pwr *= pwHL;
-    p *= lacunarity;
-  }
-
-  rmd = octaves - floor(octaves);
-  if (rmd != 0.0) {
-    increment = (snoise(p) + offset) * pwr * value;
-    value += rmd * increment;
-  }
-
-  return value;
-}
-
-/* Hybrid Additive/Multiplicative Multifractal Terrain
- *
- * H: fractal dimension of the roughest area
- * lacunarity: gap between successive frequencies
- * octaves: number of frequencies in the fBm
- * offset: raises the terrain from `sea level'
- */
-
-float noise_musgrave_hybrid_multi_fractal(
-    vec3 p, float H, float lacunarity, float octaves, float offset, float gain)
-{
-  float result, signal, weight, rmd;
-  float pwHL = pow(lacunarity, -H);
-  float pwr = pwHL;
-
-  result = snoise(p) + offset;
-  weight = gain * result;
-  p *= lacunarity;
-
-  for (int i = 1; (weight > 0.001f) && (i < int(octaves)); i++) {
-    if (weight > 1.0) {
-      weight = 1.0;
-    }
-
-    signal = (snoise(p) + offset) * pwr;
-    pwr *= pwHL;
-    result += weight * signal;
-    weight *= gain * signal;
-    p *= lacunarity;
-  }
-
-  rmd = octaves - floor(octaves);
-  if (rmd != 0.0) {
-    result += rmd * ((snoise(p) + offset) * pwr);
-  }
-
-  return result;
-}
-
-/* Ridged Multifractal Terrain
- *
- * H: fractal dimension of the roughest area
- * lacunarity: gap between successive frequencies
- * octaves: number of frequencies in the fBm
- * offset: raises the terrain from `sea level'
- */
-
-float noise_musgrave_ridged_multi_fractal(
-    vec3 p, float H, float lacunarity, float octaves, float offset, float gain)
-{
-  float result, signal, weight;
-  float pwHL = pow(lacunarity, -H);
-  float pwr = pwHL;
-
-  signal = offset - abs(snoise(p));
-  signal *= signal;
-  result = signal;
-  weight = 1.0;
-
-  for (int i = 1; i < int(octaves); i++) {
-    p *= lacunarity;
-    weight = clamp(signal * gain, 0.0, 1.0);
-    signal = offset - abs(snoise(p));
-    signal *= signal;
-    signal *= weight;
-    result += signal * pwr;
-    pwr *= pwHL;
-  }
-
-  return result;
-}
-
-float svm_musgrave(int type,
-                   float dimension,
-                   float lacunarity,
-                   float octaves,
-                   float offset,
-                   float intensity,
-                   float gain,
-                   vec3 p)
-{
-  if (type == 0 /* NODE_MUSGRAVE_MULTIFRACTAL */) {
-    return intensity * noise_musgrave_multi_fractal(p, dimension, lacunarity, octaves);
-  }
-  else if (type == 1 /* NODE_MUSGRAVE_FBM */) {
-    return intensity * noise_musgrave_fBm(p, dimension, lacunarity, octaves);
-  }
-  else if (type == 2 /* NODE_MUSGRAVE_HYBRID_MULTIFRACTAL */) {
-    return intensity *
-           noise_musgrave_hybrid_multi_fractal(p, dimension, lacunarity, octaves, offset, gain);
-  }
-  else if (type == 3 /* NODE_MUSGRAVE_RIDGED_MULTIFRACTAL */) {
-    return intensity *
-           noise_musgrave_ridged_multi_fractal(p, dimension, lacunarity, octaves, offset, gain);
-  }
-  else if (type == 4 /* NODE_MUSGRAVE_HETERO_TERRAIN */) {
-    return intensity * noise_musgrave_hetero_terrain(p, dimension, lacunarity, octaves, offset);
-  }
-  return 0.0;
-}
-
-void node_tex_musgrave(vec3 co,
-                       float scale,
-                       float detail,
-                       float dimension,
-                       float lacunarity,
-                       float offset,
-                       float gain,
-                       float type,
-                       out vec4 color,
-                       out float fac)
-{
-  fac = svm_musgrave(int(type), dimension, lacunarity, detail, offset, 1.0, gain, co *scale);
-
-  color = vec4(fac, fac, fac, 1.0);
-}
-
-void node_tex_sky(vec3 co, out vec4 color)
-{
-  color = vec4(1.0);
-}
-
-void node_tex_voronoi(vec3 co,
-                      float scale,
-                      float exponent,
-                      float coloring,
-                      float metric,
-                      float feature,
-                      out vec4 color,
-                      out float fac)
-{
-  vec3 p = co * scale;
-  int xx, yy, zz, xi, yi, zi;
-  vec4 da = vec4(1e10);
-  vec3 pa[4] = vec3[4](vec3(0.0), vec3(0.0), vec3(0.0), vec3(0.0));
-
-  xi = floor_to_int(p[0]);
-  yi = floor_to_int(p[1]);
-  zi = floor_to_int(p[2]);
-
-  for (xx = xi - 1; xx <= xi + 1; xx++) {
-    for (yy = yi - 1; yy <= yi + 1; yy++) {
-      for (zz = zi - 1; zz <= zi + 1; zz++) {
-        vec3 ip = vec3(xx, yy, zz);
-        vec3 vp = cellnoise_color(ip);
-        vec3 pd = p - (vp + ip);
-
-        float d = 0.0;
-        if (metric == 0.0) { /* SHD_VORONOI_DISTANCE 0 */
-          d = dot(pd, pd);
-        }
-        else if (metric == 1.0) { /* SHD_VORONOI_MANHATTAN 1 */
-          d = abs(pd[0]) + abs(pd[1]) + abs(pd[2]);
-        }
-        else if (metric == 2.0) { /* SHD_VORONOI_CHEBYCHEV 2 */
-          d = max(abs(pd[0]), max(abs(pd[1]), abs(pd[2])));
-        }
-        else if (metric == 3.0) { /* SHD_VORONOI_MINKOWSKI 3 */
-          d = pow(pow(abs(pd[0]), exponent) + pow(abs(pd[1]), exponent) +
-                      pow(abs(pd[2]), exponent),
-                  1.0 / exponent);
-        }
-
-        vp += vec3(xx, yy, zz);
-        if (d < da[0]) {
-          da.yzw = da.xyz;
-          da[0] = d;
-
-          pa[3] = pa[2];
-          pa[2] = pa[1];
-          pa[1] = pa[0];
-          pa[0] = vp;
-        }
-        else if (d < da[1]) {
-          da.zw = da.yz;
-          da[1] = d;
-
-          pa[3] = pa[2];
-          pa[2] = pa[1];
-          pa[1] = vp;
-        }
-        else if (d < da[2]) {
-          da[3] = da[2];
-          da[2] = d;
-
-          pa[3] = pa[2];
-          pa[2] = vp;
-        }
-        else if (d < da[3]) {
-          da[3] = d;
-          pa[3] = vp;
-        }
-      }
-    }
-  }
-
-  if (coloring == 0.0) {
-    /* Intensity output */
-    if (feature == 0.0) { /* F1 */
-      fac = abs(da[0]);
-    }
-    else if (feature == 1.0) { /* F2 */
-      fac = abs(da[1]);
-    }
-    else if (feature == 2.0) { /* F3 */
-      fac = abs(da[2]);
-    }
-    else if (feature == 3.0) { /* F4 */
-      fac = abs(da[3]);
-    }
-    else if (feature == 4.0) { /* F2F1 */
-      fac = abs(da[1] - da[0]);
-    }
-    color = vec4(fac, fac, fac, 1.0);
-  }
-  else {
-    /* Color output */
-    vec3 col = vec3(fac, fac, fac);
-    if (feature == 0.0) { /* F1 */
-      col = pa[0];
-    }
-    else if (feature == 1.0) { /* F2 */
-      col = pa[1];
-    }
-    else if (feature == 2.0) { /* F3 */
-      col = pa[2];
-    }
-    else if (feature == 3.0) { /* F4 */
-      col = pa[3];
-    }
-    else if (feature == 4.0) { /* F2F1 */
-      col = abs(pa[1] - pa[0]);
-    }
-
-    color = vec4(cellnoise_color(col), 1.0);
-    fac = (color.x + color.y + color.z) * (1.0 / 3.0);
-  }
-}
-
-float calc_wave(
-    vec3 p, float distortion, float detail, float detail_scale, int wave_type, int wave_profile)
-{
-  float n;
-
-  if (wave_type == 0) { /* type bands */
-    n = (p.x + p.y + p.z) * 10.0;
-  }
-  else { /* type rings */
-    n = length(p) * 20.0;
-  }
-
-  if (distortion != 0.0) {
-    n += distortion * noise_turbulence(p * detail_scale, detail, 0);
-  }
-
-  if (wave_profile == 0) { /* profile sin */
-    return 0.5 + 0.5 * sin(n);
-  }
-  else { /* profile saw */
-    n /= 2.0 * M_PI;
-    n -= int(n);
-    return (n < 0.0) ? n + 1.0 : n;
-  }
-}
-
-void node_tex_wave(vec3 co,
-                   float scale,
-                   float distortion,
-                   float detail,
-                   float detail_scale,
-                   float wave_type,
-                   float wave_profile,
-                   out vec4 color,
-                   out float fac)
-{
-  float f;
-  f = calc_wave(co * scale, distortion, detail, detail_scale, int(wave_type), int(wave_profile));
-
-  color = vec4(f, f, f, 1.0);
-  fac = f;
-}
-
-/* light path */
-
-void node_light_path(out float is_camera_ray,
-                     out float is_shadow_ray,
-                     out float is_diffuse_ray,
-                     out float is_glossy_ray,
-                     out float is_singular_ray,
-                     out float is_reflection_ray,
-                     out float is_transmission_ray,
-                     out float ray_length,
-                     out float ray_depth,
-                     out float diffuse_depth,
-                     out float glossy_depth,
-                     out float transparent_depth,
-                     out float transmission_depth)
-{
-  /* Supported. */
-  is_camera_ray = (rayType == EEVEE_RAY_CAMERA) ? 1.0 : 0.0;
-  is_shadow_ray = (rayType == EEVEE_RAY_SHADOW) ? 1.0 : 0.0;
-  is_diffuse_ray = (rayType == EEVEE_RAY_DIFFUSE) ? 1.0 : 0.0;
-  is_glossy_ray = (rayType == EEVEE_RAY_GLOSSY) ? 1.0 : 0.0;
-  /* Kind of supported. */
-  is_singular_ray = is_glossy_ray;
-  is_reflection_ray = is_glossy_ray;
-  is_transmission_ray = is_glossy_ray;
-  ray_depth = rayDepth;
-  diffuse_depth = (is_diffuse_ray == 1.0) ? rayDepth : 0.0;
-  glossy_depth = (is_glossy_ray == 1.0) ? rayDepth : 0.0;
-  transmission_depth = (is_transmission_ray == 1.0) ? glossy_depth : 0.0;
-  /* Not supported. */
-  ray_length = 1.0;
-  transparent_depth = 0.0;
-}
-
-void node_light_falloff(
-    float strength, float tsmooth, out float quadratic, out float linear, out float constant)
-{
-  quadratic = strength;
-  linear = strength;
-  constant = strength;
-}
-
-void node_object_info(mat4 obmat,
-                      vec4 info,
-                      float mat_index,
-                      out vec3 location,
-                      out float object_index,
-                      out float material_index,
-                      out float random)
-{
-  location = obmat[3].xyz;
-  object_index = info.x;
-  material_index = mat_index;
-  random = info.z;
-}
-
-void node_normal_map(vec4 info, vec4 tangent, vec3 normal, vec3 texnormal, out vec3 outnormal)
-{
-  if (all(equal(tangent, vec4(0.0, 0.0, 0.0, 1.0)))) {
-    outnormal = normal;
-    return;
-  }
-  tangent *= (gl_FrontFacing ? 1.0 : -1.0);
-  vec3 B = tangent.w * cross(normal, tangent.xyz) * info.w;
-
-  outnormal = texnormal.x * tangent.xyz + texnormal.y * B + texnormal.z * normal;
-  outnormal = normalize(outnormal);
-}
-
-void node_bump(
-    float strength, float dist, float height, vec3 N, vec3 surf_pos, float invert, out vec3 result)
-{
-  N = mat3(ViewMatrix) * normalize(N);
-  dist *= gl_FrontFacing ? invert : -invert;
-
-  vec3 dPdx = dFdx(surf_pos);
-  vec3 dPdy = dFdy(surf_pos);
-
-  /* Get surface tangents from normal. */
-  vec3 Rx = cross(dPdy, N);
-  vec3 Ry = cross(N, dPdx);
-
-  /* Compute surface gradient and determinant. */
-  float det = dot(dPdx, Rx);
-
-  float dHdx = dFdx(height);
-  float dHdy = dFdy(height);
-  vec3 surfgrad = dHdx * Rx + dHdy * Ry;
-
-  strength = max(strength, 0.0);
-
-  result = normalize(abs(det) * N - dist * sign(det) * surfgrad);
-  result = normalize(mix(N, result, strength));
-
-  result = mat3(ViewMatrixInverse) * result;
-}
-
-void node_bevel(float radius, vec3 N, out vec3 result)
-{
-  result = N;
-}
-
-void node_hair_info(out float is_strand,
-                    out float intercept,
-                    out float thickness,
-                    out vec3 tangent,
-                    out float random)
-{
-#ifdef HAIR_SHADER
-  is_strand = 1.0;
-  intercept = hairTime;
-  thickness = hairThickness;
-  tangent = normalize(worldNormal);
-  random = wang_hash_noise(
-      uint(hairStrandID)); /* TODO: could be precomputed per strand instead. */
-#else
-  is_strand = 0.0;
-  intercept = 0.0;
-  thickness = 0.0;
-  tangent = vec3(1.0);
-  random = 0.0;
-#endif
-}
-
-void node_displacement_object(
-    float height, float midlevel, float scale, vec3 N, mat4 obmat, out vec3 result)
-{
-  N = (vec4(N, 0.0) * obmat).xyz;
-  result = (height - midlevel) * scale * normalize(N);
-  result = (obmat * vec4(result, 0.0)).xyz;
-}
-
-void node_displacement_world(float height, float midlevel, float scale, vec3 N, out vec3 result)
-{
-  result = (height - midlevel) * scale * normalize(N);
-}
-
-void node_vector_displacement_tangent(vec4 vector,
-                                      float midlevel,
-                                      float scale,
-                                      vec4 tangent,
-                                      vec3 normal,
-                                      mat4 obmat,
-                                      mat4 viewmat,
-                                      out vec3 result)
-{
-  /* TODO(fclem) this is broken. revisit latter. */
-  vec3 N_object = normalize(((vec4(normal, 0.0) * viewmat) * obmat).xyz);
-  vec3 T_object = normalize(((vec4(tangent.xyz, 0.0) * viewmat) * obmat).xyz);
-  vec3 B_object = tangent.w * normalize(cross(N_object, T_object));
-
-  vec3 offset = (vector.xyz - vec3(midlevel)) * scale;
-  result = offset.x * T_object + offset.y * N_object + offset.z * B_object;
-  result = (obmat * vec4(result, 0.0)).xyz;
-}
-
-void node_vector_displacement_object(
-    vec4 vector, float midlevel, float scale, mat4 obmat, out vec3 result)
-{
-  result = (vector.xyz - vec3(midlevel)) * scale;
-  result = (obmat * vec4(result, 0.0)).xyz;
-}
-
-void node_vector_displacement_world(vec4 vector, float midlevel, float scale, out vec3 result)
-{
-  result = (vector.xyz - vec3(midlevel)) * scale;
-}
-
-/* output */
-
-void node_output_material(Closure surface, Closure volume, vec3 displacement, out Closure result)
-{
-#ifdef VOLUMETRICS
-  result = volume;
-#else
-  result = surface;
-#endif
-}
-
-uniform float backgroundAlpha;
-
-void node_output_world(Closure surface, Closure volume, out Closure result)
-{
-#ifndef VOLUMETRICS
-  result.radiance = surface.radiance * backgroundAlpha;
-  result.transmittance = vec3(1.0 - backgroundAlpha);
-#else
-  result = volume;
-#endif /* VOLUMETRICS */
-}
-
-/* TODO : clean this ifdef mess */
-/* EEVEE output */
-void world_normals_get(out vec3 N)
-{
-#ifndef VOLUMETRICS
-#  ifdef HAIR_SHADER
-  vec3 B = normalize(cross(worldNormal, hairTangent));
-  float cos_theta;
-  if (hairThicknessRes == 1) {
-    vec4 rand = texelFetch(utilTex, ivec3(ivec2(gl_FragCoord.xy) % LUT_SIZE, 2.0), 0);
-    /* Random cosine normal distribution on the hair surface. */
-    cos_theta = rand.x * 2.0 - 1.0;
-  }
-  else {
-    /* Shade as a cylinder. */
-    cos_theta = hairThickTime / hairThickness;
-  }
-  float sin_theta = sqrt(max(0.0, 1.0 - cos_theta * cos_theta));
-  N = normalize(worldNormal * sin_theta + B * cos_theta);
-#  else
-  N = gl_FrontFacing ? worldNormal : -worldNormal;
-#  endif
-#else
-  generated_from_orco(vec3(0.0), N);
-#endif
-}
-
-#ifndef VOLUMETRICS
-void node_eevee_specular(vec4 diffuse,
-                         vec4 specular,
-                         float roughness,
-                         vec4 emissive,
-                         float transp,
-                         vec3 normal,
-                         float clearcoat,
-                         float clearcoat_roughness,
-                         vec3 clearcoat_normal,
-                         float occlusion,
-                         float ssr_id,
-                         out Closure result)
-{
-  normal = normalize(normal);
-
-  vec3 out_diff, out_spec, ssr_spec;
-  eevee_closure_default_clearcoat(normal,
-                                  diffuse.rgb,
-                                  specular.rgb,
-                                  vec3(1.0),
-                                  int(ssr_id),
-                                  roughness,
-                                  clearcoat_normal,
-                                  clearcoat * 0.25,
-                                  clearcoat_roughness,
-                                  occlusion,
-                                  out_diff,
-                                  out_spec,
-                                  ssr_spec);
-
-  float alpha = 1.0 - transp;
-  result = CLOSURE_DEFAULT;
-  result.radiance = out_diff * diffuse.rgb + out_spec + emissive.rgb;
-  result.radiance *= alpha;
-  result.transmittance = vec3(transp);
-
-  closure_load_ssr_data(ssr_spec * alpha, roughness, normal, viewCameraVec, int(ssr_id), result);
-}
-
-void node_shader_to_rgba(Closure cl, out vec4 outcol, out float outalpha)
-{
-  vec4 spec_accum = vec4(0.0);
-  if (ssrToggle && FLAG_TEST(cl.flag, CLOSURE_SSR_FLAG)) {
-    vec3 V = cameraVec;
-    vec3 vN = normal_decode(cl.ssr_normal, viewCameraVec);
-    vec3 N = transform_direction(ViewMatrixInverse, vN);
-    float roughness = cl.ssr_data.a;
-    float roughnessSquared = max(1e-3, roughness * roughness);
-    fallback_cubemap(N, V, worldPosition, viewPosition, roughness, roughnessSquared, spec_accum);
-  }
-
-  outalpha = avg(cl.transmittance);
-  outcol = vec4((spec_accum.rgb * cl.ssr_data.rgb) + cl.radiance, 1.0);
-
-#  ifdef USE_SSS
-#    ifdef USE_SSS_ALBEDO
-  outcol.rgb += cl.sss_data.rgb * cl.sss_albedo;
-#    else
-  outcol.rgb += cl.sss_data.rgb;
-#    endif
-#  endif
-}
-
-#endif /* VOLUMETRICS */
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_add_shader.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_add_shader.glsl
new file mode 100644
index 00000000000..99117400c57
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_add_shader.glsl
@@ -0,0 +1,4 @@
+void node_add_shader(Closure shader1, Closure shader2, out Closure shader)
+{
+  shader = closure_add(shader1, shader2);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_ambient_occlusion.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_ambient_occlusion.glsl
new file mode 100644
index 00000000000..8f8ebebb5f1
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_ambient_occlusion.glsl
@@ -0,0 +1,13 @@
+#ifndef VOLUMETRICS
+void node_ambient_occlusion(
+    vec4 color, float distance, vec3 normal, out vec4 result_color, out float result_ao)
+{
+  vec3 bent_normal;
+  vec4 rand = texelFetch(utilTex, ivec3(ivec2(gl_FragCoord.xy) % LUT_SIZE, 2.0), 0);
+  result_ao = occlusion_compute(normalize(normal), viewPosition, 1.0, rand, bent_normal);
+  result_color = result_ao * color;
+}
+#else
+/* Stub ambient occlusion because it is not compatible with volumetrics. */
+#  define node_ambient_occlusion
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_anisotropic.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_anisotropic.glsl
new file mode 100644
index 00000000000..a8a900b40c6
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_anisotropic.glsl
@@ -0,0 +1,15 @@
+#ifndef VOLUMETRICS
+void node_bsdf_anisotropic(vec4 color,
+                           float roughness,
+                           float anisotropy,
+                           float rotation,
+                           vec3 N,
+                           vec3 T,
+                           out Closure result)
+{
+  node_bsdf_glossy(color, roughness, N, -1, result);
+}
+#else
+/* Stub anisotropic because it is not compatible with volumetrics. */
+#  define node_bsdf_anisotropic
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_attribute.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_attribute.glsl
new file mode 100644
index 00000000000..10e1b4563bc
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_attribute.glsl
@@ -0,0 +1,6 @@
+void node_attribute(vec3 attr, out vec4 outcol, out vec3 outvec, out float outf)
+{
+  outcol = vec4(attr, 1.0);
+  outvec = attr;
+  outf = avg(attr);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_background.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_background.glsl
new file mode 100644
index 00000000000..69ef4dcb7c7
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_background.glsl
@@ -0,0 +1,11 @@
+void node_background(vec4 color, float strength, out Closure result)
+{
+#ifndef VOLUMETRICS
+  color *= strength;
+  result = CLOSURE_DEFAULT;
+  result.radiance = color.rgb;
+  result.transmittance = vec3(0.0);
+#else
+  result = CLOSURE_DEFAULT;
+#endif
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_bevel.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_bevel.glsl
new file mode 100644
index 00000000000..0d99390c2f9
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_bevel.glsl
@@ -0,0 +1,4 @@
+void node_bevel(float radius, vec3 N, out vec3 result)
+{
+  result = N;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_blackbody.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_blackbody.glsl
new file mode 100644
index 00000000000..d0111aa3839
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_blackbody.glsl
@@ -0,0 +1,13 @@
+void node_blackbody(float temperature, sampler1DArray spectrummap, float layer, out vec4 color)
+{
+  if (temperature >= 12000.0) {
+    color = vec4(0.826270103, 0.994478524, 1.56626022, 1.0);
+  }
+  else if (temperature < 965.0) {
+    color = vec4(4.70366907, 0.0, 0.0, 1.0);
+  }
+  else {
+    float t = (temperature - 965.0) / (12000.0 - 965.0);
+    color = vec4(texture(spectrummap, vec2(t, layer)).rgb, 1.0);
+  }
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_bright_contrast.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_bright_contrast.glsl
new file mode 100644
index 00000000000..a5a10833065
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_bright_contrast.glsl
@@ -0,0 +1,10 @@
+void brightness_contrast(vec4 col, float brightness, float contrast, out vec4 outcol)
+{
+  float a = 1.0 + contrast;
+  float b = brightness - contrast * 0.5;
+
+  outcol.r = max(a * col.r + b, 0.0);
+  outcol.g = max(a * col.g + b, 0.0);
+  outcol.b = max(a * col.b + b, 0.0);
+  outcol.a = col.a;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_bump.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_bump.glsl
new file mode 100644
index 00000000000..1137e5acdc6
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_bump.glsl
@@ -0,0 +1,27 @@
+void node_bump(
+    float strength, float dist, float height, vec3 N, vec3 surf_pos, float invert, out vec3 result)
+{
+  N = mat3(ViewMatrix) * normalize(N);
+  dist *= gl_FrontFacing ? invert : -invert;
+
+  vec3 dPdx = dFdx(surf_pos);
+  vec3 dPdy = dFdy(surf_pos);
+
+  /* Get surface tangents from normal. */
+  vec3 Rx = cross(dPdy, N);
+  vec3 Ry = cross(N, dPdx);
+
+  /* Compute surface gradient and determinant. */
+  float det = dot(dPdx, Rx);
+
+  float dHdx = dFdx(height);
+  float dHdy = dFdy(height);
+  vec3 surfgrad = dHdx * Rx + dHdy * Ry;
+
+  strength = max(strength, 0.0);
+
+  result = normalize(abs(det) * N - dist * sign(det) * surfgrad);
+  result = normalize(mix(N, result, strength));
+
+  result = mat3(ViewMatrixInverse) * result;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_camera.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_camera.glsl
new file mode 100644
index 00000000000..03e61e9f472
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_camera.glsl
@@ -0,0 +1,6 @@
+void camera(vec3 co, out vec3 outview, out float outdepth, out float outdist)
+{
+  outdepth = abs(co.z);
+  outdist = length(co);
+  outview = normalize(co);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_cell_noise.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_cell_noise.glsl
new file mode 100644
index 00000000000..881f2386cd4
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_cell_noise.glsl
@@ -0,0 +1,17 @@
+float cellnoise(vec3 p)
+{
+  int ix = quick_floor(p.x);
+  int iy = quick_floor(p.y);
+  int iz = quick_floor(p.z);
+
+  return hash_uint3_to_float(uint(ix), uint(iy), uint(iz));
+}
+
+vec3 cellnoise_color(vec3 p)
+{
+  float r = cellnoise(p.xyz);
+  float g = cellnoise(p.yxz);
+  float b = cellnoise(p.yzx);
+
+  return vec3(r, g, b);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_clamp.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_clamp.glsl
new file mode 100644
index 00000000000..b8842064b6f
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_clamp.glsl
@@ -0,0 +1,4 @@
+void clamp_value(float value, float min, float max, out float result)
+{
+  result = clamp(value, min, max);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_color_ramp.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_color_ramp.glsl
new file mode 100644
index 00000000000..9fe45f91f45
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_color_ramp.glsl
@@ -0,0 +1,28 @@
+void valtorgb_opti_constant(
+    float fac, float edge, vec4 color1, vec4 color2, out vec4 outcol, out float outalpha)
+{
+  outcol = (fac > edge) ? color2 : color1;
+  outalpha = outcol.a;
+}
+
+void valtorgb_opti_linear(
+    float fac, vec2 mulbias, vec4 color1, vec4 color2, out vec4 outcol, out float outalpha)
+{
+  fac = clamp(fac * mulbias.x + mulbias.y, 0.0, 1.0);
+  outcol = mix(color1, color2, fac);
+  outalpha = outcol.a;
+}
+
+void valtorgb(float fac, sampler1DArray colormap, float layer, out vec4 outcol, out float outalpha)
+{
+  outcol = texture(colormap, vec2(fac, layer));
+  outalpha = outcol.a;
+}
+
+void valtorgb_nearest(
+    float fac, sampler1DArray colormap, float layer, out vec4 outcol, out float outalpha)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  outcol = texelFetch(colormap, ivec2(fac * (textureSize(colormap, 0).x - 1), layer), 0);
+  outalpha = outcol.a;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_color_util.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_color_util.glsl
new file mode 100644
index 00000000000..a5c3a990d90
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_color_util.glsl
@@ -0,0 +1,111 @@
+void rgb_to_hsv(vec4 rgb, out vec4 outcol)
+{
+  float cmax, cmin, h, s, v, cdelta;
+  vec3 c;
+
+  cmax = max(rgb[0], max(rgb[1], rgb[2]));
+  cmin = min(rgb[0], min(rgb[1], rgb[2]));
+  cdelta = cmax - cmin;
+
+  v = cmax;
+  if (cmax != 0.0) {
+    s = cdelta / cmax;
+  }
+  else {
+    s = 0.0;
+    h = 0.0;
+  }
+
+  if (s == 0.0) {
+    h = 0.0;
+  }
+  else {
+    c = (vec3(cmax) - rgb.xyz) / cdelta;
+
+    if (rgb.x == cmax) {
+      h = c[2] - c[1];
+    }
+    else if (rgb.y == cmax) {
+      h = 2.0 + c[0] - c[2];
+    }
+    else {
+      h = 4.0 + c[1] - c[0];
+    }
+
+    h /= 6.0;
+
+    if (h < 0.0) {
+      h += 1.0;
+    }
+  }
+
+  outcol = vec4(h, s, v, rgb.w);
+}
+
+void hsv_to_rgb(vec4 hsv, out vec4 outcol)
+{
+  float i, f, p, q, t, h, s, v;
+  vec3 rgb;
+
+  h = hsv[0];
+  s = hsv[1];
+  v = hsv[2];
+
+  if (s == 0.0) {
+    rgb = vec3(v, v, v);
+  }
+  else {
+    if (h == 1.0) {
+      h = 0.0;
+    }
+
+    h *= 6.0;
+    i = floor(h);
+    f = h - i;
+    rgb = vec3(f, f, f);
+    p = v * (1.0 - s);
+    q = v * (1.0 - (s * f));
+    t = v * (1.0 - (s * (1.0 - f)));
+
+    if (i == 0.0) {
+      rgb = vec3(v, t, p);
+    }
+    else if (i == 1.0) {
+      rgb = vec3(q, v, p);
+    }
+    else if (i == 2.0) {
+      rgb = vec3(p, v, t);
+    }
+    else if (i == 3.0) {
+      rgb = vec3(p, q, v);
+    }
+    else if (i == 4.0) {
+      rgb = vec3(t, p, v);
+    }
+    else {
+      rgb = vec3(v, p, q);
+    }
+  }
+
+  outcol = vec4(rgb, hsv.w);
+}
+
+void color_alpha_clear(vec4 color, out vec4 result)
+{
+  result = vec4(color.rgb, 1.0);
+}
+
+void color_alpha_premultiply(vec4 color, out vec4 result)
+{
+  result = vec4(color.rgb * color.a, 1.0);
+}
+
+void color_alpha_unpremultiply(vec4 color, out vec4 result)
+{
+  if (color.a == 0.0 || color.a == 1.0) {
+    result = vec4(color.rgb, 1.0);
+  }
+  else {
+    result = vec4(color.rgb / color.a, 1.0);
+  }
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_combine_hsv.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_combine_hsv.glsl
new file mode 100644
index 00000000000..2ce061da3cb
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_combine_hsv.glsl
@@ -0,0 +1,4 @@
+void combine_hsv(float h, float s, float v, out vec4 col)
+{
+  hsv_to_rgb(vec4(h, s, v, 1.0), col);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_combine_rgb.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_combine_rgb.glsl
new file mode 100644
index 00000000000..d9c882a048f
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_combine_rgb.glsl
@@ -0,0 +1,4 @@
+void combine_rgb(float r, float g, float b, out vec4 col)
+{
+  col = vec4(r, g, b, 1.0);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_combine_xyz.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_combine_xyz.glsl
new file mode 100644
index 00000000000..d8d132ff1f9
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_combine_xyz.glsl
@@ -0,0 +1,4 @@
+void combine_xyz(float x, float y, float z, out vec3 vec)
+{
+  vec = vec3(x, y, z);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_diffuse.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_diffuse.glsl
new file mode 100644
index 00000000000..14acf9925a2
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_diffuse.glsl
@@ -0,0 +1,13 @@
+#ifndef VOLUMETRICS
+void node_bsdf_diffuse(vec4 color, float roughness, vec3 N, out Closure result)
+{
+  N = normalize(N);
+  result = CLOSURE_DEFAULT;
+  eevee_closure_diffuse(N, color.rgb, 1.0, result.radiance);
+  closure_load_ssr_data(vec3(0.0), 0.0, N, viewCameraVec, -1, result);
+  result.radiance *= color.rgb;
+}
+#else
+/* Stub diffuse because it is not compatible with volumetrics. */
+#  define node_bsdf_diffuse
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_displacement.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_displacement.glsl
new file mode 100644
index 00000000000..0838b5c8b71
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_displacement.glsl
@@ -0,0 +1,12 @@
+void node_displacement_object(
+    float height, float midlevel, float scale, vec3 N, mat4 obmat, out vec3 result)
+{
+  N = (vec4(N, 0.0) * obmat).xyz;
+  result = (height - midlevel) * scale * normalize(N);
+  result = (obmat * vec4(result, 0.0)).xyz;
+}
+
+void node_displacement_world(float height, float midlevel, float scale, vec3 N, out vec3 result)
+{
+  result = (height - midlevel) * scale * normalize(N);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_eevee_specular.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_eevee_specular.glsl
new file mode 100644
index 00000000000..e69a53b6596
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_eevee_specular.glsl
@@ -0,0 +1,40 @@
+#ifndef VOLUMETRICS
+void node_eevee_specular(vec4 diffuse,
+                         vec4 specular,
+                         float roughness,
+                         vec4 emissive,
+                         float transp,
+                         vec3 normal,
+                         float clearcoat,
+                         float clearcoat_roughness,
+                         vec3 clearcoat_normal,
+                         float occlusion,
+                         float ssr_id,
+                         out Closure result)
+{
+  normal = normalize(normal);
+
+  vec3 out_diff, out_spec, ssr_spec;
+  eevee_closure_default_clearcoat(normal,
+                                  diffuse.rgb,
+                                  specular.rgb,
+                                  vec3(1.0),
+                                  int(ssr_id),
+                                  roughness,
+                                  clearcoat_normal,
+                                  clearcoat * 0.25,
+                                  clearcoat_roughness,
+                                  occlusion,
+                                  out_diff,
+                                  out_spec,
+                                  ssr_spec);
+
+  float alpha = 1.0 - transp;
+  result = CLOSURE_DEFAULT;
+  result.radiance = out_diff * diffuse.rgb + out_spec + emissive.rgb;
+  result.radiance *= alpha;
+  result.transmittance = vec3(transp);
+
+  closure_load_ssr_data(ssr_spec * alpha, roughness, normal, viewCameraVec, int(ssr_id), result);
+}
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_emission.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_emission.glsl
new file mode 100644
index 00000000000..092b9ed08bb
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_emission.glsl
@@ -0,0 +1,10 @@
+void node_emission(vec4 color, float strength, vec3 vN, out Closure result)
+{
+  result = CLOSURE_DEFAULT;
+#ifndef VOLUMETRICS
+  result.radiance = color.rgb * strength;
+  result.ssr_normal = normal_encode(vN, viewCameraVec);
+#else
+  result.emission = color.rgb * strength;
+#endif
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_fractal_noise.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_fractal_noise.glsl
new file mode 100644
index 00000000000..2a925c2a622
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_fractal_noise.glsl
@@ -0,0 +1,32 @@
+float noise_turbulence(vec3 p, float octaves, int hard)
+{
+  float fscale = 1.0;
+  float amp = 1.0;
+  float sum = 0.0;
+  octaves = clamp(octaves, 0.0, 16.0);
+  int n = int(octaves);
+  for (int i = 0; i <= n; i++) {
+    float t = noise(fscale * p);
+    if (hard != 0) {
+      t = abs(2.0 * t - 1.0);
+    }
+    sum += t * amp;
+    amp *= 0.5;
+    fscale *= 2.0;
+  }
+  float rmd = octaves - floor(octaves);
+  if (rmd != 0.0) {
+    float t = noise(fscale * p);
+    if (hard != 0) {
+      t = abs(2.0 * t - 1.0);
+    }
+    float sum2 = sum + t * amp;
+    sum *= (float(1 << n) / float((1 << (n + 1)) - 1));
+    sum2 *= (float(1 << (n + 1)) / float((1 << (n + 2)) - 1));
+    return (1.0 - rmd) * sum + rmd * sum2;
+  }
+  else {
+    sum *= (float(1 << n) / float((1 << (n + 1)) - 1));
+    return sum;
+  }
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_fresnel.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_fresnel.glsl
new file mode 100644
index 00000000000..7a4d28f2dd6
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_fresnel.glsl
@@ -0,0 +1,37 @@
+float fresnel_dielectric_cos(float cosi, float eta)
+{
+  /* compute fresnel reflectance without explicitly computing
+   * the refracted direction */
+  float c = abs(cosi);
+  float g = eta * eta - 1.0 + c * c;
+  float result;
+
+  if (g > 0.0) {
+    g = sqrt(g);
+    float A = (g - c) / (g + c);
+    float B = (c * (g + c) - 1.0) / (c * (g - c) + 1.0);
+    result = 0.5 * A * A * (1.0 + B * B);
+  }
+  else {
+    result = 1.0; /* TIR (no refracted component) */
+  }
+
+  return result;
+}
+
+float fresnel_dielectric(vec3 Incoming, vec3 Normal, float eta)
+{
+  /* compute fresnel reflectance without explicitly computing
+   * the refracted direction */
+  return fresnel_dielectric_cos(dot(Incoming, Normal), eta);
+}
+
+void node_fresnel(float ior, vec3 N, vec3 I, out float result)
+{
+  N = normalize(N);
+  /* handle perspective/orthographic */
+  vec3 I_view = (ProjectionMatrix[3][3] == 0.0) ? normalize(I) : vec3(0.0, 0.0, -1.0);
+
+  float eta = max(ior, 0.00001);
+  result = fresnel_dielectric(I_view, N, (gl_FrontFacing) ? eta : 1.0 / eta);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_gamma.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_gamma.glsl
new file mode 100644
index 00000000000..5733992f8dd
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_gamma.glsl
@@ -0,0 +1,14 @@
+void node_gamma(vec4 col, float gamma, out vec4 outcol)
+{
+  outcol = col;
+
+  if (col.r > 0.0) {
+    outcol.r = compatible_pow(col.r, gamma);
+  }
+  if (col.g > 0.0) {
+    outcol.g = compatible_pow(col.g, gamma);
+  }
+  if (col.b > 0.0) {
+    outcol.b = compatible_pow(col.b, gamma);
+  }
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_geometry.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_geometry.glsl
new file mode 100644
index 00000000000..79614495499
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_geometry.glsl
@@ -0,0 +1,46 @@
+void node_geometry(vec3 I,
+                   vec3 N,
+                   vec3 orco,
+                   mat4 objmat,
+                   mat4 toworld,
+                   vec2 barycentric,
+                   out vec3 position,
+                   out vec3 normal,
+                   out vec3 tangent,
+                   out vec3 true_normal,
+                   out vec3 incoming,
+                   out vec3 parametric,
+                   out float backfacing,
+                   out float pointiness)
+{
+  /* handle perspective/orthographic */
+  vec3 I_view = (ProjectionMatrix[3][3] == 0.0) ? normalize(I) : vec3(0.0, 0.0, -1.0);
+  incoming = -(toworld * vec4(I_view, 0.0)).xyz;
+
+#if defined(WORLD_BACKGROUND) || defined(PROBE_CAPTURE)
+  position = -incoming;
+  true_normal = normal = incoming;
+  tangent = parametric = vec3(0.0);
+  vec3(0.0);
+  backfacing = 0.0;
+  pointiness = 0.0;
+#else
+
+  position = worldPosition;
+#  ifndef VOLUMETRICS
+  normal = normalize(N);
+  vec3 B = dFdx(worldPosition);
+  vec3 T = dFdy(worldPosition);
+  true_normal = normalize(cross(B, T));
+#  else
+  normal = (toworld * vec4(N, 0.0)).xyz;
+  true_normal = normal;
+#  endif
+  tangent_orco_z(orco, orco);
+  node_tangent(N, orco, objmat, tangent);
+
+  parametric = vec3(barycentric, 0.0);
+  backfacing = (gl_FrontFacing) ? 0.0 : 1.0;
+  pointiness = 0.5;
+#endif
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_glass.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_glass.glsl
new file mode 100644
index 00000000000..8cc8ba2dd15
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_glass.glsl
@@ -0,0 +1,23 @@
+#ifndef VOLUMETRICS
+void node_bsdf_glass(
+    vec4 color, float roughness, float ior, vec3 N, float ssr_id, out Closure result)
+{
+  N = normalize(N);
+  vec3 out_spec, out_refr, ssr_spec;
+  vec3 refr_color = (refractionDepth > 0.0) ? color.rgb * color.rgb :
+                                              color.rgb; /* Simulate 2 transmission event */
+  eevee_closure_glass(
+      N, vec3(1.0), vec3(1.0), int(ssr_id), roughness, 1.0, ior, out_spec, out_refr, ssr_spec);
+  out_refr *= refr_color;
+  out_spec *= color.rgb;
+  float fresnel = F_eta(ior, dot(N, cameraVec));
+  vec3 vN = mat3(ViewMatrix) * N;
+  result = CLOSURE_DEFAULT;
+  result.radiance = mix(out_refr, out_spec, fresnel);
+  closure_load_ssr_data(
+      ssr_spec * color.rgb * fresnel, roughness, N, viewCameraVec, int(ssr_id), result);
+}
+#else
+/* Stub glass because it is not compatible with volumetrics. */
+#  define node_bsdf_glass
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_glossy.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_glossy.glsl
new file mode 100644
index 00000000000..4d582e63725
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_glossy.glsl
@@ -0,0 +1,15 @@
+#ifndef VOLUMETRICS
+void node_bsdf_glossy(vec4 color, float roughness, vec3 N, float ssr_id, out Closure result)
+{
+  N = normalize(N);
+  vec3 out_spec, ssr_spec;
+  eevee_closure_glossy(N, vec3(1.0), vec3(1.0), int(ssr_id), roughness, 1.0, out_spec, ssr_spec);
+  vec3 vN = mat3(ViewMatrix) * N;
+  result = CLOSURE_DEFAULT;
+  result.radiance = out_spec * color.rgb;
+  closure_load_ssr_data(ssr_spec * color.rgb, roughness, N, viewCameraVec, int(ssr_id), result);
+}
+#else
+/* Stub glossy because it is not compatible with volumetrics. */
+#  define node_bsdf_glossy
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_hair_info.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_hair_info.glsl
new file mode 100644
index 00000000000..3b23ac976ae
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_hair_info.glsl
@@ -0,0 +1,21 @@
+void node_hair_info(out float is_strand,
+                    out float intercept,
+                    out float thickness,
+                    out vec3 tangent,
+                    out float random)
+{
+#ifdef HAIR_SHADER
+  is_strand = 1.0;
+  intercept = hairTime;
+  thickness = hairThickness;
+  tangent = normalize(worldNormal);
+  random = wang_hash_noise(
+      uint(hairStrandID)); /* TODO: could be precomputed per strand instead. */
+#else
+  is_strand = 0.0;
+  intercept = 0.0;
+  thickness = 0.0;
+  tangent = vec3(1.0);
+  random = 0.0;
+#endif
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_hash.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_hash.glsl
new file mode 100644
index 00000000000..86191451e5f
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_hash.glsl
@@ -0,0 +1,217 @@
+/* ***** Jenkins Lookup3 Hash Functions ***** */
+
+/* Source: http://burtleburtle.net/bob/c/lookup3.c */
+
+#define rot(x, k) (((x) << (k)) | ((x) >> (32 - (k))))
+
+#define mix(a, b, c) \
+  { \
+    a -= c; \
+    a ^= rot(c, 4); \
+    c += b; \
+    b -= a; \
+    b ^= rot(a, 6); \
+    a += c; \
+    c -= b; \
+    c ^= rot(b, 8); \
+    b += a; \
+    a -= c; \
+    a ^= rot(c, 16); \
+    c += b; \
+    b -= a; \
+    b ^= rot(a, 19); \
+    a += c; \
+    c -= b; \
+    c ^= rot(b, 4); \
+    b += a; \
+  }
+
+#define final(a, b, c) \
+  { \
+    c ^= b; \
+    c -= rot(b, 14); \
+    a ^= c; \
+    a -= rot(c, 11); \
+    b ^= a; \
+    b -= rot(a, 25); \
+    c ^= b; \
+    c -= rot(b, 16); \
+    a ^= c; \
+    a -= rot(c, 4); \
+    b ^= a; \
+    b -= rot(a, 14); \
+    c ^= b; \
+    c -= rot(b, 24); \
+  }
+
+uint hash_uint(uint kx)
+{
+  uint a, b, c;
+  a = b = c = 0xdeadbeefu + (1u << 2u) + 13u;
+
+  a += kx;
+  final(a, b, c);
+
+  return c;
+}
+
+uint hash_uint2(uint kx, uint ky)
+{
+  uint a, b, c;
+  a = b = c = 0xdeadbeefu + (2u << 2u) + 13u;
+
+  b += ky;
+  a += kx;
+  final(a, b, c);
+
+  return c;
+}
+
+uint hash_uint3(uint kx, uint ky, uint kz)
+{
+  uint a, b, c;
+  a = b = c = 0xdeadbeefu + (3u << 2u) + 13u;
+
+  c += kz;
+  b += ky;
+  a += kx;
+  final(a, b, c);
+
+  return c;
+}
+
+uint hash_uint4(uint kx, uint ky, uint kz, uint kw)
+{
+  uint a, b, c;
+  a = b = c = 0xdeadbeefu + (4u << 2u) + 13u;
+
+  a += kx;
+  b += ky;
+  c += kz;
+  mix(a, b, c);
+
+  a += kw;
+  final(a, b, c);
+
+  return c;
+}
+
+#undef rot
+#undef final
+#undef mix
+
+uint hash_int(int kx)
+{
+  return hash_uint(uint(kx));
+}
+
+uint hash_int2(int kx, int ky)
+{
+  return hash_uint2(uint(kx), uint(ky));
+}
+
+uint hash_int3(int kx, int ky, int kz)
+{
+  return hash_uint3(uint(kx), uint(ky), uint(kz));
+}
+
+uint hash_int4(int kx, int ky, int kz, int kw)
+{
+  return hash_uint4(uint(kx), uint(ky), uint(kz), uint(kw));
+}
+
+/* Hashing uint or uint[234] into a float in the range [0, 1]. */
+
+float hash_uint_to_float(uint kx)
+{
+  return float(hash_uint(kx)) / float(0xFFFFFFFFu);
+}
+
+float hash_uint2_to_float(uint kx, uint ky)
+{
+  return float(hash_uint2(kx, ky)) / float(0xFFFFFFFFu);
+}
+
+float hash_uint3_to_float(uint kx, uint ky, uint kz)
+{
+  return float(hash_uint3(kx, ky, kz)) / float(0xFFFFFFFFu);
+}
+
+float hash_uint4_to_float(uint kx, uint ky, uint kz, uint kw)
+{
+  return float(hash_uint4(kx, ky, kz, kw)) / float(0xFFFFFFFFu);
+}
+
+/* Hashing float or vec[234] into a float in the range [0, 1]. */
+
+float hash_float_to_float(float k)
+{
+  return hash_uint_to_float(floatBitsToUint(k));
+}
+
+float hash_vec2_to_float(vec2 k)
+{
+  return hash_uint2_to_float(floatBitsToUint(k.x), floatBitsToUint(k.y));
+}
+
+float hash_vec3_to_float(vec3 k)
+{
+  return hash_uint3_to_float(floatBitsToUint(k.x), floatBitsToUint(k.y), floatBitsToUint(k.z));
+}
+
+float hash_vec4_to_float(vec4 k)
+{
+  return hash_uint4_to_float(
+      floatBitsToUint(k.x), floatBitsToUint(k.y), floatBitsToUint(k.z), floatBitsToUint(k.w));
+}
+
+/* Hashing vec[234] into vec[234] of components in the range [0, 1]. */
+
+vec2 hash_vec2_to_vec2(vec2 k)
+{
+  return vec2(hash_vec2_to_float(k), hash_vec3_to_float(vec3(k, 1.0)));
+}
+
+vec3 hash_vec3_to_vec3(vec3 k)
+{
+  return vec3(
+      hash_vec3_to_float(k), hash_vec4_to_float(vec4(k, 1.0)), hash_vec4_to_float(vec4(k, 2.0)));
+}
+
+vec4 hash_vec4_to_vec4(vec4 k)
+{
+  return vec4(hash_vec4_to_float(k.xyzw),
+              hash_vec4_to_float(k.wxyz),
+              hash_vec4_to_float(k.zwxy),
+              hash_vec4_to_float(k.yzwx));
+}
+
+/* Hashing float or vec[234] into vec3 of components in range [0, 1]. */
+
+vec3 hash_float_to_vec3(float k)
+{
+  return vec3(
+      hash_float_to_float(k), hash_vec2_to_float(vec2(k, 1.0)), hash_vec2_to_float(vec2(k, 2.0)));
+}
+
+vec3 hash_vec2_to_vec3(vec2 k)
+{
+  return vec3(
+      hash_vec2_to_float(k), hash_vec3_to_float(vec3(k, 1.0)), hash_vec3_to_float(vec3(k, 2.0)));
+}
+
+vec3 hash_vec4_to_vec3(vec4 k)
+{
+  return vec3(hash_vec4_to_float(k.xyzw), hash_vec4_to_float(k.zxwy), hash_vec4_to_float(k.wzyx));
+}
+
+/* Other Hash Functions */
+
+float integer_noise(int n)
+{
+  int nn;
+  n = (n + 1013) & 0x7fffffff;
+  n = (n >> 13) ^ n;
+  nn = (n * (n * n * 60493 + 19990303) + 1376312589) & 0x7fffffff;
+  return 0.5 * (float(nn) / 1073741824.0);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_holdout.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_holdout.glsl
new file mode 100644
index 00000000000..50ce2bf2ab8
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_holdout.glsl
@@ -0,0 +1,8 @@
+void node_holdout(out Closure result)
+{
+  result = CLOSURE_DEFAULT;
+#ifndef VOLUMETRICS
+  result.holdout = 1.0;
+  result.flag = CLOSURE_HOLDOUT_FLAG;
+#endif
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_hue_sat_val.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_hue_sat_val.glsl
new file mode 100644
index 00000000000..64ac73ecdf3
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_hue_sat_val.glsl
@@ -0,0 +1,14 @@
+void hue_sat(float hue, float sat, float value, float fac, vec4 col, out vec4 outcol)
+{
+  vec4 hsv;
+
+  rgb_to_hsv(col, hsv);
+
+  hsv[0] = fract(hsv[0] + hue + 0.5);
+  hsv[1] = clamp(hsv[1] * sat, 0.0, 1.0);
+  hsv[2] = hsv[2] * value;
+
+  hsv_to_rgb(hsv, outcol);
+
+  outcol = mix(col, outcol, fac);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_invert.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_invert.glsl
new file mode 100644
index 00000000000..5cf362a9947
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_invert.glsl
@@ -0,0 +1,5 @@
+void invert(float fac, vec4 col, out vec4 outcol)
+{
+  outcol.xyz = mix(col.xyz, vec3(1.0) - col.xyz, fac);
+  outcol.w = col.w;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_layer_weight.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_layer_weight.glsl
new file mode 100644
index 00000000000..588d295bcc4
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_layer_weight.glsl
@@ -0,0 +1,19 @@
+void node_layer_weight(float blend, vec3 N, vec3 I, out float fresnel, out float facing)
+{
+  N = normalize(N);
+
+  /* fresnel */
+  float eta = max(1.0 - blend, 0.00001);
+  vec3 I_view = (ProjectionMatrix[3][3] == 0.0) ? normalize(I) : vec3(0.0, 0.0, -1.0);
+
+  fresnel = fresnel_dielectric(I_view, N, (gl_FrontFacing) ? 1.0 / eta : eta);
+
+  /* facing */
+  facing = abs(dot(I_view, N));
+  if (blend != 0.5) {
+    blend = clamp(blend, 0.0, 0.99999);
+    blend = (blend < 0.5) ? 2.0 * blend : 0.5 / (1.0 - blend);
+    facing = pow(facing, blend);
+  }
+  facing = 1.0 - facing;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_light_falloff.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_light_falloff.glsl
new file mode 100644
index 00000000000..f3eae653f95
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_light_falloff.glsl
@@ -0,0 +1,7 @@
+void node_light_falloff(
+    float strength, float tsmooth, out float quadratic, out float linear, out float constant)
+{
+  quadratic = strength;
+  linear = strength;
+  constant = strength;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_light_path.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_light_path.glsl
new file mode 100644
index 00000000000..50c87e3f105
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_light_path.glsl
@@ -0,0 +1,31 @@
+void node_light_path(out float is_camera_ray,
+                     out float is_shadow_ray,
+                     out float is_diffuse_ray,
+                     out float is_glossy_ray,
+                     out float is_singular_ray,
+                     out float is_reflection_ray,
+                     out float is_transmission_ray,
+                     out float ray_length,
+                     out float ray_depth,
+                     out float diffuse_depth,
+                     out float glossy_depth,
+                     out float transparent_depth,
+                     out float transmission_depth)
+{
+  /* Supported. */
+  is_camera_ray = (rayType == EEVEE_RAY_CAMERA) ? 1.0 : 0.0;
+  is_shadow_ray = (rayType == EEVEE_RAY_SHADOW) ? 1.0 : 0.0;
+  is_diffuse_ray = (rayType == EEVEE_RAY_DIFFUSE) ? 1.0 : 0.0;
+  is_glossy_ray = (rayType == EEVEE_RAY_GLOSSY) ? 1.0 : 0.0;
+  /* Kind of supported. */
+  is_singular_ray = is_glossy_ray;
+  is_reflection_ray = is_glossy_ray;
+  is_transmission_ray = is_glossy_ray;
+  ray_depth = rayDepth;
+  diffuse_depth = (is_diffuse_ray == 1.0) ? rayDepth : 0.0;
+  glossy_depth = (is_glossy_ray == 1.0) ? rayDepth : 0.0;
+  transmission_depth = (is_transmission_ray == 1.0) ? glossy_depth : 0.0;
+  /* Not supported. */
+  ray_length = 1.0;
+  transparent_depth = 0.0;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_map_range.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_map_range.glsl
new file mode 100644
index 00000000000..a185774f4b3
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_map_range.glsl
@@ -0,0 +1,10 @@
+void map_range(
+    float value, float fromMin, float fromMax, float toMin, float toMax, out float result)
+{
+  if (fromMax != fromMin) {
+    result = toMin + ((value - fromMin) / (fromMax - fromMin)) * (toMax - toMin);
+  }
+  else {
+    result = 0.0;
+  }
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_mapping.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_mapping.glsl
new file mode 100644
index 00000000000..ef47ac2be98
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_mapping.glsl
@@ -0,0 +1,7 @@
+void mapping(
+    vec3 vec, vec4 m0, vec4 m1, vec4 m2, vec4 m3, vec3 minvec, vec3 maxvec, out vec3 outvec)
+{
+  mat4 mat = mat4(m0, m1, m2, m3);
+  outvec = (mat * vec4(vec, 1.0)).xyz;
+  outvec = clamp(outvec, minvec, maxvec);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_math.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_math.glsl
new file mode 100644
index 00000000000..4fac770e8fe
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_math.glsl
@@ -0,0 +1,130 @@
+void math_add(float a, float b, out float result)
+{
+  result = a + b;
+}
+
+void math_subtract(float a, float b, out float result)
+{
+  result = a - b;
+}
+
+void math_multiply(float a, float b, out float result)
+{
+  result = a * b;
+}
+
+void math_divide(float a, float b, out float result)
+{
+  result = safe_divide(a, b);
+}
+
+void math_power(float a, float b, out float result)
+{
+  if (a >= 0.0) {
+    result = compatible_pow(a, b);
+  }
+  else {
+    float fraction = mod(abs(b), 1.0);
+    if (fraction > 0.999 || fraction < 0.001) {
+      result = compatible_pow(a, floor(b + 0.5));
+    }
+    else {
+      result = 0.0;
+    }
+  }
+}
+
+void math_logarithm(float a, float b, out float result)
+{
+  result = (a > 0.0 && b > 0.0) ? log2(a) / log2(b) : 0.0;
+}
+
+void math_sqrt(float a, float b, out float result)
+{
+  result = (a > 0.0) ? sqrt(a) : 0.0;
+}
+
+void math_absolute(float a, float b, out float result)
+{
+  result = abs(a);
+}
+
+void math_minimum(float a, float b, out float result)
+{
+  result = min(a, b);
+}
+
+void math_maximum(float a, float b, out float result)
+{
+  result = max(a, b);
+}
+
+void math_less_than(float a, float b, out float result)
+{
+  result = (a < b) ? 1.0 : 0.0;
+}
+
+void math_greater_than(float a, float b, out float result)
+{
+  result = (a > b) ? 1.0 : 0.0;
+}
+
+void math_round(float a, float b, out float result)
+{
+  result = floor(a + 0.5);
+}
+
+void math_floor(float a, float b, out float result)
+{
+  result = floor(a);
+}
+
+void math_ceil(float a, float b, out float result)
+{
+  result = ceil(a);
+}
+
+void math_fraction(float a, float b, out float result)
+{
+  result = a - floor(a);
+}
+
+void math_modulo(float a, float b, out float result)
+{
+  result = c_mod(a, b);
+}
+
+void math_sine(float a, float b, out float result)
+{
+  result = sin(a);
+}
+
+void math_cosine(float a, float b, out float result)
+{
+  result = cos(a);
+}
+
+void math_tangent(float a, float b, out float result)
+{
+  result = tan(a);
+}
+
+void math_arcsine(float a, float b, out float result)
+{
+  result = (a <= 1.0 && a >= -1.0) ? asin(a) : 0.0;
+}
+
+void math_arccosine(float a, float b, out float result)
+{
+  result = (a <= 1.0 && a >= -1.0) ? acos(a) : 0.0;
+}
+
+void math_arctangent(float a, float b, out float result)
+{
+  result = atan(a);
+}
+
+void math_arctan2(float a, float b, out float result)
+{
+  result = atan(a, b);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_math_util.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_math_util.glsl
new file mode 100644
index 00000000000..0be7da0cc4c
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_math_util.glsl
@@ -0,0 +1,101 @@
+/* Float Math */
+
+float safe_divide(float a, float b)
+{
+  return (b != 0.0) ? a / b : 0.0;
+}
+
+/* Modulo with C sign convention. mod in GLSL will take absolute for negative numbers. */
+float c_mod(float a, float b)
+{
+  return (b != 0.0 && a != b) ? sign(a) * mod(abs(a), b) : 0.0;
+}
+
+float compatible_pow(float x, float y)
+{
+  if (y == 0.0) { /* x^0 -> 1, including 0^0 */
+    return 1.0;
+  }
+
+  /* glsl pow doesn't accept negative x */
+  if (x < 0.0) {
+    if (mod(-y, 2.0) == 0.0) {
+      return pow(-x, y);
+    }
+    else {
+      return -pow(-x, y);
+    }
+  }
+  else if (x == 0.0) {
+    return 0.0;
+  }
+
+  return pow(x, y);
+}
+
+float hypot(float x, float y)
+{
+  return sqrt(x * x + y * y);
+}
+
+int floor_to_int(float x)
+{
+  return int(floor(x));
+}
+
+int quick_floor(float x)
+{
+  return int(x) - ((x < 0) ? 1 : 0);
+}
+
+float floorfrac(float x, out int i)
+{
+  float x_floor = floor(x);
+  i = int(x_floor);
+  return x - x_floor;
+}
+
+/* Vector Math */
+
+vec3 safe_divide(vec3 a, vec3 b)
+{
+  return vec3(safe_divide(a.x, b.x), safe_divide(a.y, b.y), safe_divide(a.z, b.z));
+}
+
+vec3 c_mod(vec3 a, vec3 b)
+{
+  return vec3(c_mod(a.x, b.x), c_mod(a.y, b.y), c_mod(a.z, b.z));
+}
+
+void vector_mix(float strength, vec3 a, vec3 b, out vec3 outVector)
+{
+  outVector = strength * a + (1 - strength) * b;
+}
+
+void invert_z(vec3 v, out vec3 outv)
+{
+  v.z = -v.z;
+  outv = v;
+}
+
+void vector_normalize(vec3 normal, out vec3 outnormal)
+{
+  outnormal = normalize(normal);
+}
+
+/* Matirx Math */
+
+void direction_transform_m4v3(vec3 vin, mat4 mat, out vec3 vout)
+{
+  vout = (mat * vec4(vin, 0.0)).xyz;
+}
+
+void normal_transform_transposed_m4v3(vec3 vin, mat4 mat, out vec3 vout)
+{
+  vout = transpose(mat3(mat)) * vin;
+}
+
+void point_transform_m4v3(vec3 vin, mat4 mat, out vec3 vout)
+{
+  vout = (mat * vec4(vin, 1.0)).xyz;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_mix_rgb.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_mix_rgb.glsl
new file mode 100644
index 00000000000..abe6081489d
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_mix_rgb.glsl
@@ -0,0 +1,291 @@
+void mix_blend(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  outcol = mix(col1, col2, fac);
+  outcol.a = col1.a;
+}
+
+void mix_add(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  outcol = mix(col1, col1 + col2, fac);
+  outcol.a = col1.a;
+}
+
+void mix_mult(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  outcol = mix(col1, col1 * col2, fac);
+  outcol.a = col1.a;
+}
+
+void mix_screen(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  float facm = 1.0 - fac;
+
+  outcol = vec4(1.0) - (vec4(facm) + fac * (vec4(1.0) - col2)) * (vec4(1.0) - col1);
+  outcol.a = col1.a;
+}
+
+void mix_overlay(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  float facm = 1.0 - fac;
+
+  outcol = col1;
+
+  if (outcol.r < 0.5) {
+    outcol.r *= facm + 2.0 * fac * col2.r;
+  }
+  else {
+    outcol.r = 1.0 - (facm + 2.0 * fac * (1.0 - col2.r)) * (1.0 - outcol.r);
+  }
+
+  if (outcol.g < 0.5) {
+    outcol.g *= facm + 2.0 * fac * col2.g;
+  }
+  else {
+    outcol.g = 1.0 - (facm + 2.0 * fac * (1.0 - col2.g)) * (1.0 - outcol.g);
+  }
+
+  if (outcol.b < 0.5) {
+    outcol.b *= facm + 2.0 * fac * col2.b;
+  }
+  else {
+    outcol.b = 1.0 - (facm + 2.0 * fac * (1.0 - col2.b)) * (1.0 - outcol.b);
+  }
+}
+
+void mix_sub(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  outcol = mix(col1, col1 - col2, fac);
+  outcol.a = col1.a;
+}
+
+void mix_div(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  float facm = 1.0 - fac;
+
+  outcol = col1;
+
+  if (col2.r != 0.0) {
+    outcol.r = facm * outcol.r + fac * outcol.r / col2.r;
+  }
+  if (col2.g != 0.0) {
+    outcol.g = facm * outcol.g + fac * outcol.g / col2.g;
+  }
+  if (col2.b != 0.0) {
+    outcol.b = facm * outcol.b + fac * outcol.b / col2.b;
+  }
+}
+
+void mix_diff(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  outcol = mix(col1, abs(col1 - col2), fac);
+  outcol.a = col1.a;
+}
+
+void mix_dark(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  outcol.rgb = min(col1.rgb, col2.rgb * fac);
+  outcol.a = col1.a;
+}
+
+void mix_light(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  outcol.rgb = max(col1.rgb, col2.rgb * fac);
+  outcol.a = col1.a;
+}
+
+void mix_dodge(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  outcol = col1;
+
+  if (outcol.r != 0.0) {
+    float tmp = 1.0 - fac * col2.r;
+    if (tmp <= 0.0) {
+      outcol.r = 1.0;
+    }
+    else if ((tmp = outcol.r / tmp) > 1.0) {
+      outcol.r = 1.0;
+    }
+    else {
+      outcol.r = tmp;
+    }
+  }
+  if (outcol.g != 0.0) {
+    float tmp = 1.0 - fac * col2.g;
+    if (tmp <= 0.0) {
+      outcol.g = 1.0;
+    }
+    else if ((tmp = outcol.g / tmp) > 1.0) {
+      outcol.g = 1.0;
+    }
+    else {
+      outcol.g = tmp;
+    }
+  }
+  if (outcol.b != 0.0) {
+    float tmp = 1.0 - fac * col2.b;
+    if (tmp <= 0.0) {
+      outcol.b = 1.0;
+    }
+    else if ((tmp = outcol.b / tmp) > 1.0) {
+      outcol.b = 1.0;
+    }
+    else {
+      outcol.b = tmp;
+    }
+  }
+}
+
+void mix_burn(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  float tmp, facm = 1.0 - fac;
+
+  outcol = col1;
+
+  tmp = facm + fac * col2.r;
+  if (tmp <= 0.0) {
+    outcol.r = 0.0;
+  }
+  else if ((tmp = (1.0 - (1.0 - outcol.r) / tmp)) < 0.0) {
+    outcol.r = 0.0;
+  }
+  else if (tmp > 1.0) {
+    outcol.r = 1.0;
+  }
+  else {
+    outcol.r = tmp;
+  }
+
+  tmp = facm + fac * col2.g;
+  if (tmp <= 0.0) {
+    outcol.g = 0.0;
+  }
+  else if ((tmp = (1.0 - (1.0 - outcol.g) / tmp)) < 0.0) {
+    outcol.g = 0.0;
+  }
+  else if (tmp > 1.0) {
+    outcol.g = 1.0;
+  }
+  else {
+    outcol.g = tmp;
+  }
+
+  tmp = facm + fac * col2.b;
+  if (tmp <= 0.0) {
+    outcol.b = 0.0;
+  }
+  else if ((tmp = (1.0 - (1.0 - outcol.b) / tmp)) < 0.0) {
+    outcol.b = 0.0;
+  }
+  else if (tmp > 1.0) {
+    outcol.b = 1.0;
+  }
+  else {
+    outcol.b = tmp;
+  }
+}
+
+void mix_hue(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  float facm = 1.0 - fac;
+
+  outcol = col1;
+
+  vec4 hsv, hsv2, tmp;
+  rgb_to_hsv(col2, hsv2);
+
+  if (hsv2.y != 0.0) {
+    rgb_to_hsv(outcol, hsv);
+    hsv.x = hsv2.x;
+    hsv_to_rgb(hsv, tmp);
+
+    outcol = mix(outcol, tmp, fac);
+    outcol.a = col1.a;
+  }
+}
+
+void mix_sat(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  float facm = 1.0 - fac;
+
+  outcol = col1;
+
+  vec4 hsv, hsv2;
+  rgb_to_hsv(outcol, hsv);
+
+  if (hsv.y != 0.0) {
+    rgb_to_hsv(col2, hsv2);
+
+    hsv.y = facm * hsv.y + fac * hsv2.y;
+    hsv_to_rgb(hsv, outcol);
+  }
+}
+
+void mix_val(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  float facm = 1.0 - fac;
+
+  vec4 hsv, hsv2;
+  rgb_to_hsv(col1, hsv);
+  rgb_to_hsv(col2, hsv2);
+
+  hsv.z = facm * hsv.z + fac * hsv2.z;
+  hsv_to_rgb(hsv, outcol);
+}
+
+void mix_color(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  float facm = 1.0 - fac;
+
+  outcol = col1;
+
+  vec4 hsv, hsv2, tmp;
+  rgb_to_hsv(col2, hsv2);
+
+  if (hsv2.y != 0.0) {
+    rgb_to_hsv(outcol, hsv);
+    hsv.x = hsv2.x;
+    hsv.y = hsv2.y;
+    hsv_to_rgb(hsv, tmp);
+
+    outcol = mix(outcol, tmp, fac);
+    outcol.a = col1.a;
+  }
+}
+
+void mix_soft(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+  float facm = 1.0 - fac;
+
+  vec4 one = vec4(1.0);
+  vec4 scr = one - (one - col2) * (one - col1);
+  outcol = facm * col1 + fac * ((one - col1) * col2 * col1 + col1 * scr);
+}
+
+void mix_linear(float fac, vec4 col1, vec4 col2, out vec4 outcol)
+{
+  fac = clamp(fac, 0.0, 1.0);
+
+  outcol = col1 + fac * (2.0 * (col2 - vec4(0.5)));
+}
+
+void clamp_color(vec3 vec, vec3 min, vec3 max, out vec3 out_vec)
+{
+  out_vec = clamp(vec, min, max);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_mix_shader.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_mix_shader.glsl
new file mode 100644
index 00000000000..c303d21d7c1
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_mix_shader.glsl
@@ -0,0 +1,4 @@
+void node_mix_shader(float fac, Closure shader1, Closure shader2, out Closure shader)
+{
+  shader = closure_mix(shader1, shader2, fac);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_noise.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_noise.glsl
new file mode 100644
index 00000000000..8c6a10e3fe7
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_noise.glsl
@@ -0,0 +1,74 @@
+float noise_fade(float t)
+{
+  return t * t * t * (t * (t * 6.0 - 15.0) + 10.0);
+}
+
+float noise_scale3(float result)
+{
+  return 0.9820 * result;
+}
+
+float noise_nerp(float t, float a, float b)
+{
+  return (1.0 - t) * a + t * b;
+}
+
+float noise_grad(uint hash, float x, float y, float z)
+{
+  uint h = hash & 15u;
+  float u = h < 8u ? x : y;
+  float vt = ((h == 12u) || (h == 14u)) ? x : z;
+  float v = h < 4u ? y : vt;
+  return (((h & 1u) != 0u) ? -u : u) + (((h & 2u) != 0u) ? -v : v);
+}
+
+float noise_perlin(float x, float y, float z)
+{
+  int X;
+  float fx = floorfrac(x, X);
+  int Y;
+  float fy = floorfrac(y, Y);
+  int Z;
+  float fz = floorfrac(z, Z);
+
+  float u = noise_fade(fx);
+  float v = noise_fade(fy);
+  float w = noise_fade(fz);
+
+  float noise_u[2], noise_v[2];
+
+  noise_u[0] = noise_nerp(u,
+                          noise_grad(hash_int3(X, Y, Z), fx, fy, fz),
+                          noise_grad(hash_int3(X + 1, Y, Z), fx - 1.0, fy, fz));
+
+  noise_u[1] = noise_nerp(u,
+                          noise_grad(hash_int3(X, Y + 1, Z), fx, fy - 1.0, fz),
+                          noise_grad(hash_int3(X + 1, Y + 1, Z), fx - 1.0, fy - 1.0, fz));
+
+  noise_v[0] = noise_nerp(v, noise_u[0], noise_u[1]);
+
+  noise_u[0] = noise_nerp(u,
+                          noise_grad(hash_int3(X, Y, Z + 1), fx, fy, fz - 1.0),
+                          noise_grad(hash_int3(X + 1, Y, Z + 1), fx - 1.0, fy, fz - 1.0));
+
+  noise_u[1] = noise_nerp(
+      u,
+      noise_grad(hash_int3(X, Y + 1, Z + 1), fx, fy - 1.0, fz - 1.0),
+      noise_grad(hash_int3(X + 1, Y + 1, Z + 1), fx - 1.0, fy - 1.0, fz - 1.0));
+
+  noise_v[1] = noise_nerp(v, noise_u[0], noise_u[1]);
+
+  float r = noise_scale3(noise_nerp(w, noise_v[0], noise_v[1]));
+
+  return (isinf(r)) ? 0.0 : r;
+}
+
+float noise(vec3 p)
+{
+  return 0.5 * noise_perlin(p.x, p.y, p.z) + 0.5;
+}
+
+float snoise(vec3 p)
+{
+  return noise_perlin(p.x, p.y, p.z);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_normal.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_normal.glsl
new file mode 100644
index 00000000000..4f6e68909ad
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_normal.glsl
@@ -0,0 +1,5 @@
+void normal_new_shading(vec3 nor, vec3 dir, out vec3 outnor, out float outdot)
+{
+  outnor = dir;
+  outdot = dot(normalize(nor), dir);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_normal_map.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_normal_map.glsl
new file mode 100644
index 00000000000..6930e0c5dad
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_normal_map.glsl
@@ -0,0 +1,22 @@
+void node_normal_map(vec4 info, vec4 tangent, vec3 normal, vec3 texnormal, out vec3 outnormal)
+{
+  if (all(equal(tangent, vec4(0.0, 0.0, 0.0, 1.0)))) {
+    outnormal = normal;
+    return;
+  }
+  tangent *= (gl_FrontFacing ? 1.0 : -1.0);
+  vec3 B = tangent.w * cross(normal, tangent.xyz) * info.w;
+
+  outnormal = texnormal.x * tangent.xyz + texnormal.y * B + texnormal.z * normal;
+  outnormal = normalize(outnormal);
+}
+
+void color_to_normal_new_shading(vec3 color, out vec3 normal)
+{
+  normal = vec3(2.0) * color - vec3(1.0);
+}
+
+void color_to_blender_normal_new_shading(vec3 color, out vec3 normal)
+{
+  normal = vec3(2.0, -2.0, -2.0) * color - vec3(1.0);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_object_info.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_object_info.glsl
new file mode 100644
index 00000000000..ff77b0beea2
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_object_info.glsl
@@ -0,0 +1,16 @@
+void node_object_info(mat4 obmat,
+                      vec4 obcolor,
+                      vec4 info,
+                      float mat_index,
+                      out vec3 location,
+                      out vec4 color,
+                      out float object_index,
+                      out float material_index,
+                      out float random)
+{
+  location = obmat[3].xyz;
+  color = obcolor;
+  object_index = info.x;
+  material_index = mat_index;
+  random = info.z;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_output_material.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_output_material.glsl
new file mode 100644
index 00000000000..62f76d46088
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_output_material.glsl
@@ -0,0 +1,8 @@
+void node_output_material(Closure surface, Closure volume, vec3 displacement, out Closure result)
+{
+#ifdef VOLUMETRICS
+  result = volume;
+#else
+  result = surface;
+#endif
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_output_world.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_output_world.glsl
new file mode 100644
index 00000000000..ba391df185e
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_output_world.glsl
@@ -0,0 +1,11 @@
+uniform float backgroundAlpha;
+
+void node_output_world(Closure surface, Closure volume, out Closure result)
+{
+#ifndef VOLUMETRICS
+  result.radiance = surface.radiance * backgroundAlpha;
+  result.transmittance = vec3(1.0 - backgroundAlpha);
+#else
+  result = volume;
+#endif /* VOLUMETRICS */
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_particle_info.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_particle_info.glsl
new file mode 100644
index 00000000000..bdd60c20a81
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_particle_info.glsl
@@ -0,0 +1,23 @@
+void particle_info(vec4 sprops,
+                   vec4 loc,
+                   vec3 vel,
+                   vec3 avel,
+                   out float index,
+                   out float random,
+                   out float age,
+                   out float life_time,
+                   out vec3 location,
+                   out float size,
+                   out vec3 velocity,
+                   out vec3 angular_velocity)
+{
+  index = sprops.x;
+  random = loc.w;
+  age = sprops.y;
+  life_time = sprops.z;
+  size = sprops.w;
+
+  location = loc.xyz;
+  velocity = vel;
+  angular_velocity = avel;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_principled.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_principled.glsl
new file mode 100644
index 00000000000..c6b640c572d
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_principled.glsl
@@ -0,0 +1,439 @@
+#ifndef VOLUMETRICS
+vec3 tint_from_color(vec3 color)
+{
+  float lum = dot(color, vec3(0.3, 0.6, 0.1)); /* luminance approx. */
+  return (lum > 0) ? color / lum : vec3(1.0);  /* normalize lum. to isolate hue+sat */
+}
+
+void convert_metallic_to_specular_tinted(vec3 basecol,
+                                         vec3 basecol_tint,
+                                         float metallic,
+                                         float specular_fac,
+                                         float specular_tint,
+                                         out vec3 diffuse,
+                                         out vec3 f0)
+{
+  vec3 tmp_col = mix(vec3(1.0), basecol_tint, specular_tint);
+  f0 = mix((0.08 * specular_fac) * tmp_col, basecol, metallic);
+  diffuse = basecol * (1.0 - metallic);
+}
+
+vec3 principled_sheen(float NV, vec3 basecol_tint, float sheen_tint)
+{
+  float f = 1.0 - NV;
+  /* Temporary fix for T59784. Normal map seems to contain NaNs for tangent space normal maps,
+   * therefore we need to clamp value. */
+  f = clamp(f, 0.0, 1.0);
+  /* Empirical approximation (manual curve fitting). Can be refined. */
+  float sheen = f * f * f * 0.077 + f * 0.01 + 0.00026;
+  return sheen * mix(vec3(1.0), basecol_tint, sheen_tint);
+}
+
+void node_bsdf_principled(vec4 base_color,
+                          float subsurface,
+                          vec3 subsurface_radius,
+                          vec4 subsurface_color,
+                          float metallic,
+                          float specular,
+                          float specular_tint,
+                          float roughness,
+                          float anisotropic,
+                          float anisotropic_rotation,
+                          float sheen,
+                          float sheen_tint,
+                          float clearcoat,
+                          float clearcoat_roughness,
+                          float ior,
+                          float transmission,
+                          float transmission_roughness,
+                          vec4 emission,
+                          float alpha,
+                          vec3 N,
+                          vec3 CN,
+                          vec3 T,
+                          vec3 I,
+                          float ssr_id,
+                          float sss_id,
+                          vec3 sss_scale,
+                          out Closure result)
+{
+  N = normalize(N);
+  ior = max(ior, 1e-5);
+  metallic = saturate(metallic);
+  transmission = saturate(transmission);
+  float dielectric = 1.0 - metallic;
+  transmission *= dielectric;
+  sheen *= dielectric;
+  subsurface_color *= dielectric;
+
+  vec3 diffuse, f0, out_diff, out_spec, out_trans, out_refr, ssr_spec;
+  vec3 ctint = tint_from_color(base_color.rgb);
+  convert_metallic_to_specular_tinted(
+      base_color.rgb, ctint, metallic, specular, specular_tint, diffuse, f0);
+
+  float NV = dot(N, cameraVec);
+  vec3 out_sheen = sheen * principled_sheen(NV, ctint, sheen_tint);
+
+  /* Far from being accurate, but 2 glossy evaluation is too expensive.
+   * Most noticeable difference is at grazing angles since the bsdf lut
+   * f0 color interpolation is done on top of this interpolation. */
+  vec3 f0_glass = mix(vec3(1.0), base_color.rgb, specular_tint);
+  float fresnel = F_eta(ior, NV);
+  vec3 spec_col = F_color_blend(ior, fresnel, f0_glass) * fresnel;
+  f0 = mix(f0, spec_col, transmission);
+
+  vec3 f90 = mix(vec3(1.0), f0, (1.0 - specular) * metallic);
+
+  vec3 mixed_ss_base_color = mix(diffuse, subsurface_color.rgb, subsurface);
+
+  float sss_scalef = avg(sss_scale) * subsurface;
+  eevee_closure_principled(N,
+                           mixed_ss_base_color,
+                           f0,
+                           f90,
+                           int(ssr_id),
+                           roughness,
+                           CN,
+                           clearcoat * 0.25,
+                           clearcoat_roughness,
+                           1.0,
+                           sss_scalef,
+                           ior,
+                           out_diff,
+                           out_trans,
+                           out_spec,
+                           out_refr,
+                           ssr_spec);
+
+  vec3 refr_color = base_color.rgb;
+  refr_color *= (refractionDepth > 0.0) ? refr_color :
+                                          vec3(1.0); /* Simulate 2 transmission event */
+  out_refr *= refr_color * (1.0 - fresnel) * transmission;
+
+  result = CLOSURE_DEFAULT;
+  result.radiance = out_spec + out_refr;
+  result.radiance += out_diff * out_sheen; /* Coarse approx. */
+
+  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
+
+  vec3 sss_radiance = (out_diff + out_trans) * alpha;
+#  ifndef USE_SSS
+  result.radiance += sss_radiance * mixed_ss_base_color * (1.0 - transmission);
+#  else
+#    ifdef USE_SSS_ALBEDO
+  vec3 sss_albedo = mixed_ss_base_color;
+#    else
+  sss_radiance *= mixed_ss_base_color;
+#    endif
+  sss_radiance *= (1.0 - transmission);
+  closure_load_sss_data(sss_scalef,
+                        sss_radiance,
+#    ifdef USE_SSS_ALBEDO
+                        sss_albedo,
+#    endif
+                        int(sss_id),
+                        result);
+#  endif /* USE_SSS */
+
+  result.radiance += emission.rgb;
+  result.radiance *= alpha;
+  result.transmittance = vec3(1.0 - alpha);
+}
+
+void node_bsdf_principled_dielectric(vec4 base_color,
+                                     float subsurface,
+                                     vec3 subsurface_radius,
+                                     vec4 subsurface_color,
+                                     float metallic,
+                                     float specular,
+                                     float specular_tint,
+                                     float roughness,
+                                     float anisotropic,
+                                     float anisotropic_rotation,
+                                     float sheen,
+                                     float sheen_tint,
+                                     float clearcoat,
+                                     float clearcoat_roughness,
+                                     float ior,
+                                     float transmission,
+                                     float transmission_roughness,
+                                     vec4 emission,
+                                     float alpha,
+                                     vec3 N,
+                                     vec3 CN,
+                                     vec3 T,
+                                     vec3 I,
+                                     float ssr_id,
+                                     float sss_id,
+                                     vec3 sss_scale,
+                                     out Closure result)
+{
+  N = normalize(N);
+  metallic = saturate(metallic);
+  float dielectric = 1.0 - metallic;
+
+  vec3 diffuse, f0, out_diff, out_spec, ssr_spec;
+  vec3 ctint = tint_from_color(base_color.rgb);
+  convert_metallic_to_specular_tinted(
+      base_color.rgb, ctint, metallic, specular, specular_tint, diffuse, f0);
+
+  float NV = dot(N, cameraVec);
+  vec3 out_sheen = sheen * principled_sheen(NV, ctint, sheen_tint);
+
+  eevee_closure_default(
+      N, diffuse, f0, vec3(1.0), int(ssr_id), roughness, 1.0, out_diff, out_spec, ssr_spec);
+
+  result = CLOSURE_DEFAULT;
+  result.radiance = out_spec + out_diff * (diffuse + out_sheen);
+  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
+  result.radiance += emission.rgb;
+  result.radiance *= alpha;
+  result.transmittance = vec3(1.0 - alpha);
+}
+
+void node_bsdf_principled_metallic(vec4 base_color,
+                                   float subsurface,
+                                   vec3 subsurface_radius,
+                                   vec4 subsurface_color,
+                                   float metallic,
+                                   float specular,
+                                   float specular_tint,
+                                   float roughness,
+                                   float anisotropic,
+                                   float anisotropic_rotation,
+                                   float sheen,
+                                   float sheen_tint,
+                                   float clearcoat,
+                                   float clearcoat_roughness,
+                                   float ior,
+                                   float transmission,
+                                   float transmission_roughness,
+                                   vec4 emission,
+                                   float alpha,
+                                   vec3 N,
+                                   vec3 CN,
+                                   vec3 T,
+                                   vec3 I,
+                                   float ssr_id,
+                                   float sss_id,
+                                   vec3 sss_scale,
+                                   out Closure result)
+{
+  N = normalize(N);
+  vec3 out_spec, ssr_spec;
+
+  vec3 f90 = mix(vec3(1.0), base_color.rgb, (1.0 - specular) * metallic);
+
+  eevee_closure_glossy(N, base_color.rgb, f90, int(ssr_id), roughness, 1.0, out_spec, ssr_spec);
+
+  result = CLOSURE_DEFAULT;
+  result.radiance = out_spec;
+  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
+  result.radiance += emission.rgb;
+  result.radiance *= alpha;
+  result.transmittance = vec3(1.0 - alpha);
+}
+
+void node_bsdf_principled_clearcoat(vec4 base_color,
+                                    float subsurface,
+                                    vec3 subsurface_radius,
+                                    vec4 subsurface_color,
+                                    float metallic,
+                                    float specular,
+                                    float specular_tint,
+                                    float roughness,
+                                    float anisotropic,
+                                    float anisotropic_rotation,
+                                    float sheen,
+                                    float sheen_tint,
+                                    float clearcoat,
+                                    float clearcoat_roughness,
+                                    float ior,
+                                    float transmission,
+                                    float transmission_roughness,
+                                    vec4 emission,
+                                    float alpha,
+                                    vec3 N,
+                                    vec3 CN,
+                                    vec3 T,
+                                    vec3 I,
+                                    float ssr_id,
+                                    float sss_id,
+                                    vec3 sss_scale,
+                                    out Closure result)
+{
+  vec3 out_spec, ssr_spec;
+  N = normalize(N);
+
+  vec3 f90 = mix(vec3(1.0), base_color.rgb, (1.0 - specular) * metallic);
+
+  eevee_closure_clearcoat(N,
+                          base_color.rgb,
+                          f90,
+                          int(ssr_id),
+                          roughness,
+                          CN,
+                          clearcoat * 0.25,
+                          clearcoat_roughness,
+                          1.0,
+                          out_spec,
+                          ssr_spec);
+
+  result = CLOSURE_DEFAULT;
+  result.radiance = out_spec;
+  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
+  result.radiance += emission.rgb;
+  result.radiance *= alpha;
+  result.transmittance = vec3(1.0 - alpha);
+}
+
+void node_bsdf_principled_subsurface(vec4 base_color,
+                                     float subsurface,
+                                     vec3 subsurface_radius,
+                                     vec4 subsurface_color,
+                                     float metallic,
+                                     float specular,
+                                     float specular_tint,
+                                     float roughness,
+                                     float anisotropic,
+                                     float anisotropic_rotation,
+                                     float sheen,
+                                     float sheen_tint,
+                                     float clearcoat,
+                                     float clearcoat_roughness,
+                                     float ior,
+                                     float transmission,
+                                     float transmission_roughness,
+                                     vec4 emission,
+                                     float alpha,
+                                     vec3 N,
+                                     vec3 CN,
+                                     vec3 T,
+                                     vec3 I,
+                                     float ssr_id,
+                                     float sss_id,
+                                     vec3 sss_scale,
+                                     out Closure result)
+{
+  metallic = saturate(metallic);
+  N = normalize(N);
+
+  vec3 diffuse, f0, out_diff, out_spec, out_trans, ssr_spec;
+  vec3 ctint = tint_from_color(base_color.rgb);
+  convert_metallic_to_specular_tinted(
+      base_color.rgb, ctint, metallic, specular, specular_tint, diffuse, f0);
+
+  subsurface_color = subsurface_color * (1.0 - metallic);
+  vec3 mixed_ss_base_color = mix(diffuse, subsurface_color.rgb, subsurface);
+  float sss_scalef = avg(sss_scale) * subsurface;
+
+  float NV = dot(N, cameraVec);
+  vec3 out_sheen = sheen * principled_sheen(NV, ctint, sheen_tint);
+
+  vec3 f90 = mix(vec3(1.0), base_color.rgb, (1.0 - specular) * metallic);
+
+  eevee_closure_skin(N,
+                     mixed_ss_base_color,
+                     f0,
+                     f90,
+                     int(ssr_id),
+                     roughness,
+                     1.0,
+                     sss_scalef,
+                     out_diff,
+                     out_trans,
+                     out_spec,
+                     ssr_spec);
+
+  result = CLOSURE_DEFAULT;
+  result.radiance = out_spec;
+  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
+
+  vec3 sss_radiance = (out_diff + out_trans) * alpha;
+#  ifndef USE_SSS
+  result.radiance += sss_radiance * mixed_ss_base_color * (1.0 - transmission);
+#  else
+#    ifdef USE_SSS_ALBEDO
+  vec3 sss_albedo = mixed_ss_base_color;
+#    else
+  sss_radiance *= mixed_ss_base_color;
+#    endif
+  sss_radiance *= (1.0 - transmission);
+  closure_load_sss_data(sss_scalef,
+                        sss_radiance,
+#    ifdef USE_SSS_ALBEDO
+                        sss_albedo,
+#    endif
+                        int(sss_id),
+                        result);
+#  endif /* USE_SSS */
+
+  result.radiance += out_diff * out_sheen;
+  result.radiance += emission.rgb;
+  result.radiance *= alpha;
+  result.transmittance = vec3(1.0 - alpha);
+}
+
+void node_bsdf_principled_glass(vec4 base_color,
+                                float subsurface,
+                                vec3 subsurface_radius,
+                                vec4 subsurface_color,
+                                float metallic,
+                                float specular,
+                                float specular_tint,
+                                float roughness,
+                                float anisotropic,
+                                float anisotropic_rotation,
+                                float sheen,
+                                float sheen_tint,
+                                float clearcoat,
+                                float clearcoat_roughness,
+                                float ior,
+                                float transmission,
+                                float transmission_roughness,
+                                vec4 emission,
+                                float alpha,
+                                vec3 N,
+                                vec3 CN,
+                                vec3 T,
+                                vec3 I,
+                                float ssr_id,
+                                float sss_id,
+                                vec3 sss_scale,
+                                out Closure result)
+{
+  ior = max(ior, 1e-5);
+  N = normalize(N);
+
+  vec3 f0, out_spec, out_refr, ssr_spec;
+  f0 = mix(vec3(1.0), base_color.rgb, specular_tint);
+
+  eevee_closure_glass(
+      N, vec3(1.0), vec3(1.0), int(ssr_id), roughness, 1.0, ior, out_spec, out_refr, ssr_spec);
+
+  vec3 refr_color = base_color.rgb;
+  refr_color *= (refractionDepth > 0.0) ? refr_color :
+                                          vec3(1.0); /* Simulate 2 transmission events */
+  out_refr *= refr_color;
+
+  float fresnel = F_eta(ior, dot(N, cameraVec));
+  vec3 spec_col = F_color_blend(ior, fresnel, f0);
+  out_spec *= spec_col;
+  ssr_spec *= spec_col * fresnel;
+
+  result = CLOSURE_DEFAULT;
+  result.radiance = mix(out_refr, out_spec, fresnel);
+  closure_load_ssr_data(ssr_spec * alpha, roughness, N, viewCameraVec, int(ssr_id), result);
+  result.radiance += emission.rgb;
+  result.radiance *= alpha;
+  result.transmittance = vec3(1.0 - alpha);
+}
+#else
+/* Stub principled because it is not compatible with volumetrics. */
+#  define node_bsdf_principled
+#  define node_bsdf_principled_dielectric
+#  define node_bsdf_principled_metallic
+#  define node_bsdf_principled_clearcoat
+#  define node_bsdf_principled_subsurface
+#  define node_bsdf_principled_glass
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_refraction.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_refraction.glsl
new file mode 100644
index 00000000000..04394a9420b
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_refraction.glsl
@@ -0,0 +1,16 @@
+#ifndef VOLUMETRICS
+void node_bsdf_refraction(vec4 color, float roughness, float ior, vec3 N, out Closure result)
+{
+  N = normalize(N);
+  vec3 out_refr;
+  color.rgb *= (refractionDepth > 0.0) ? color.rgb : vec3(1.0); /* Simulate 2 absorption event. */
+  eevee_closure_refraction(N, roughness, ior, out_refr);
+  vec3 vN = mat3(ViewMatrix) * N;
+  result = CLOSURE_DEFAULT;
+  result.ssr_normal = normal_encode(vN, viewCameraVec);
+  result.radiance = out_refr * color.rgb;
+}
+#else
+/* Stub refraction because it is not compatible with volumetrics. */
+#  define node_bsdf_refraction
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_rgb_curves.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_rgb_curves.glsl
new file mode 100644
index 00000000000..054fdddf7c3
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_rgb_curves.glsl
@@ -0,0 +1,73 @@
+/* ext is vec4(in_x, in_dy, out_x, out_dy). */
+float curve_extrapolate(float x, float y, vec4 ext)
+{
+  if (x < 0.0) {
+    return y + x * ext.y;
+  }
+  else if (x > 1.0) {
+    return y + (x - 1.0) * ext.w;
+  }
+  else {
+    return y;
+  }
+}
+
+#define RANGE_RESCALE(x, min, range) ((x - min) * range)
+
+void curves_rgb(float fac,
+                vec4 col,
+                sampler1DArray curvemap,
+                float layer,
+                vec4 range,
+                vec4 ext_r,
+                vec4 ext_g,
+                vec4 ext_b,
+                vec4 ext_a,
+                out vec4 outcol)
+{
+  vec4 co = vec4(RANGE_RESCALE(col.rgb, ext_a.x, range.a), layer);
+  vec3 samp;
+  samp.r = texture(curvemap, co.xw).a;
+  samp.g = texture(curvemap, co.yw).a;
+  samp.b = texture(curvemap, co.zw).a;
+
+  samp.r = curve_extrapolate(co.x, samp.r, ext_a);
+  samp.g = curve_extrapolate(co.y, samp.g, ext_a);
+  samp.b = curve_extrapolate(co.z, samp.b, ext_a);
+
+  vec3 rgb_min = vec3(ext_r.x, ext_g.x, ext_b.x);
+  co.xyz = RANGE_RESCALE(samp.rgb, rgb_min, range.rgb);
+
+  samp.r = texture(curvemap, co.xw).r;
+  samp.g = texture(curvemap, co.yw).g;
+  samp.b = texture(curvemap, co.zw).b;
+
+  outcol.r = curve_extrapolate(co.x, samp.r, ext_r);
+  outcol.g = curve_extrapolate(co.y, samp.g, ext_g);
+  outcol.b = curve_extrapolate(co.z, samp.b, ext_b);
+  outcol.a = col.a;
+
+  outcol = mix(col, outcol, fac);
+}
+
+void curves_rgb_opti(float fac,
+                     vec4 col,
+                     sampler1DArray curvemap,
+                     float layer,
+                     vec4 range,
+                     vec4 ext_a,
+                     out vec4 outcol)
+{
+  vec4 co = vec4(RANGE_RESCALE(col.rgb, ext_a.x, range.a), layer);
+  vec3 samp;
+  samp.r = texture(curvemap, co.xw).a;
+  samp.g = texture(curvemap, co.yw).a;
+  samp.b = texture(curvemap, co.zw).a;
+
+  outcol.r = curve_extrapolate(co.x, samp.r, ext_a);
+  outcol.g = curve_extrapolate(co.y, samp.g, ext_a);
+  outcol.b = curve_extrapolate(co.z, samp.b, ext_a);
+  outcol.a = col.a;
+
+  outcol = mix(col, outcol, fac);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_rgb_to_bw.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_rgb_to_bw.glsl
new file mode 100644
index 00000000000..ceca02a2356
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_rgb_to_bw.glsl
@@ -0,0 +1,5 @@
+void rgbtobw(vec4 color, out float outval)
+{
+  vec3 factors = vec3(0.2126, 0.7152, 0.0722);
+  outval = dot(color.rgb, factors);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_separate_hsv.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_separate_hsv.glsl
new file mode 100644
index 00000000000..fb64e424c6c
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_separate_hsv.glsl
@@ -0,0 +1,9 @@
+void separate_hsv(vec4 col, out float h, out float s, out float v)
+{
+  vec4 hsv;
+
+  rgb_to_hsv(col, hsv);
+  h = hsv[0];
+  s = hsv[1];
+  v = hsv[2];
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_separate_rgb.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_separate_rgb.glsl
new file mode 100644
index 00000000000..4232b4c001e
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_separate_rgb.glsl
@@ -0,0 +1,6 @@
+void separate_rgb(vec4 col, out float r, out float g, out float b)
+{
+  r = col.r;
+  g = col.g;
+  b = col.b;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_separate_xyz.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_separate_xyz.glsl
new file mode 100644
index 00000000000..fac29ccc135
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_separate_xyz.glsl
@@ -0,0 +1,6 @@
+void separate_xyz(vec3 vec, out float x, out float y, out float z)
+{
+  x = vec.r;
+  y = vec.g;
+  z = vec.b;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_set.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_set.glsl
new file mode 100644
index 00000000000..dc2ecb05351
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_set.glsl
@@ -0,0 +1,44 @@
+void set_value(float val, out float outval)
+{
+  outval = val;
+}
+
+void set_rgb(vec3 col, out vec3 outcol)
+{
+  outcol = col;
+}
+
+void set_rgba(vec4 col, out vec4 outcol)
+{
+  outcol = col;
+}
+
+void set_value_zero(out float outval)
+{
+  outval = 0.0;
+}
+
+void set_value_one(out float outval)
+{
+  outval = 1.0;
+}
+
+void set_rgb_zero(out vec3 outval)
+{
+  outval = vec3(0.0);
+}
+
+void set_rgb_one(out vec3 outval)
+{
+  outval = vec3(1.0);
+}
+
+void set_rgba_zero(out vec4 outval)
+{
+  outval = vec4(0.0);
+}
+
+void set_rgba_one(out vec4 outval)
+{
+  outval = vec4(1.0);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_shader_to_rgba.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_shader_to_rgba.glsl
new file mode 100644
index 00000000000..0a587c7e471
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_shader_to_rgba.glsl
@@ -0,0 +1,25 @@
+#ifndef VOLUMETRICS
+void node_shader_to_rgba(Closure cl, out vec4 outcol, out float outalpha)
+{
+  vec4 spec_accum = vec4(0.0);
+  if (ssrToggle && FLAG_TEST(cl.flag, CLOSURE_SSR_FLAG)) {
+    vec3 V = cameraVec;
+    vec3 vN = normal_decode(cl.ssr_normal, viewCameraVec);
+    vec3 N = transform_direction(ViewMatrixInverse, vN);
+    float roughness = cl.ssr_data.a;
+    float roughnessSquared = max(1e-3, roughness * roughness);
+    fallback_cubemap(N, V, worldPosition, viewPosition, roughness, roughnessSquared, spec_accum);
+  }
+
+  outalpha = avg(cl.transmittance);
+  outcol = vec4((spec_accum.rgb * cl.ssr_data.rgb) + cl.radiance, 1.0);
+
+#  ifdef USE_SSS
+#    ifdef USE_SSS_ALBEDO
+  outcol.rgb += cl.sss_data.rgb * cl.sss_albedo;
+#    else
+  outcol.rgb += cl.sss_data.rgb;
+#    endif
+#  endif
+}
+#endif /* VOLUMETRICS */
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_squeeze.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_squeeze.glsl
new file mode 100644
index 00000000000..b73bdebf7f6
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_squeeze.glsl
@@ -0,0 +1,4 @@
+void squeeze(float val, float width, float center, out float outval)
+{
+  outval = 1.0 / (1.0 + pow(2.71828183, -((val - center) * width)));
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_subsurface_scattering.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_subsurface_scattering.glsl
new file mode 100644
index 00000000000..fc81f3c4674
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_subsurface_scattering.glsl
@@ -0,0 +1,42 @@
+#ifndef VOLUMETRICS
+void node_subsurface_scattering(vec4 color,
+                                float scale,
+                                vec3 radius,
+                                float sharpen,
+                                float texture_blur,
+                                vec3 N,
+                                float sss_id,
+                                out Closure result)
+{
+#  if defined(USE_SSS)
+  N = normalize(N);
+  vec3 out_diff, out_trans;
+  vec3 vN = mat3(ViewMatrix) * N;
+  result = CLOSURE_DEFAULT;
+  closure_load_ssr_data(vec3(0.0), 0.0, N, viewCameraVec, -1, result);
+
+  eevee_closure_subsurface(N, color.rgb, 1.0, scale, out_diff, out_trans);
+
+  vec3 sss_radiance = out_diff + out_trans;
+#    ifdef USE_SSS_ALBEDO
+  /* Not perfect for texture_blur not exactly equal to 0.0 or 1.0. */
+  vec3 sss_albedo = mix(color.rgb, vec3(1.0), texture_blur);
+  sss_radiance *= mix(vec3(1.0), color.rgb, texture_blur);
+#    else
+  sss_radiance *= color.rgb;
+#    endif
+  closure_load_sss_data(scale,
+                        sss_radiance,
+#    ifdef USE_SSS_ALBEDO
+                        sss_albedo,
+#    endif
+                        int(sss_id),
+                        result);
+#  else
+  node_bsdf_diffuse(color, 0.0, N, result);
+#  endif
+}
+#else
+/* Stub subsurface scattering because it is not compatible with volumetrics. */
+#  define node_subsurface_scattering
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tangent.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tangent.glsl
new file mode 100644
index 00000000000..ff2dbc7ead3
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tangent.glsl
@@ -0,0 +1,25 @@
+void tangent_orco_x(vec3 orco_in, out vec3 orco_out)
+{
+  orco_out = orco_in.xzy * vec3(0.0, -0.5, 0.5) + vec3(0.0, 0.25, -0.25);
+}
+
+void tangent_orco_y(vec3 orco_in, out vec3 orco_out)
+{
+  orco_out = orco_in.zyx * vec3(-0.5, 0.0, 0.5) + vec3(0.25, 0.0, -0.25);
+}
+
+void tangent_orco_z(vec3 orco_in, out vec3 orco_out)
+{
+  orco_out = orco_in.yxz * vec3(-0.5, 0.5, 0.0) + vec3(0.25, -0.25, 0.0);
+}
+
+void node_tangentmap(vec4 attr_tangent, out vec3 tangent)
+{
+  tangent = normalize(attr_tangent.xyz);
+}
+
+void node_tangent(vec3 N, vec3 orco, mat4 objmat, out vec3 T)
+{
+  T = (objmat * vec4(orco, 0.0)).xyz;
+  T = cross(N, normalize(cross(T, N)));
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_brick.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_brick.glsl
new file mode 100644
index 00000000000..e252e5ba726
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_brick.glsl
@@ -0,0 +1,78 @@
+vec2 calc_brick_texture(vec3 p,
+                        float mortar_size,
+                        float mortar_smooth,
+                        float bias,
+                        float brick_width,
+                        float row_height,
+                        float offset_amount,
+                        int offset_frequency,
+                        float squash_amount,
+                        int squash_frequency)
+{
+  int bricknum, rownum;
+  float offset = 0.0;
+  float x, y;
+
+  rownum = floor_to_int(p.y / row_height);
+
+  if (offset_frequency != 0 && squash_frequency != 0) {
+    brick_width *= (rownum % squash_frequency != 0) ? 1.0 : squash_amount;           /* squash */
+    offset = (rownum % offset_frequency != 0) ? 0.0 : (brick_width * offset_amount); /* offset */
+  }
+
+  bricknum = floor_to_int((p.x + offset) / brick_width);
+
+  x = (p.x + offset) - brick_width * bricknum;
+  y = p.y - row_height * rownum;
+
+  float tint = clamp((integer_noise((rownum << 16) + (bricknum & 0xFFFF)) + bias), 0.0, 1.0);
+
+  float min_dist = min(min(x, y), min(brick_width - x, row_height - y));
+  if (min_dist >= mortar_size) {
+    return vec2(tint, 0.0);
+  }
+  else if (mortar_smooth == 0.0) {
+    return vec2(tint, 1.0);
+  }
+  else {
+    min_dist = 1.0 - min_dist / mortar_size;
+    return vec2(tint, smoothstep(0.0, mortar_smooth, min_dist));
+  }
+}
+
+void node_tex_brick(vec3 co,
+                    vec4 color1,
+                    vec4 color2,
+                    vec4 mortar,
+                    float scale,
+                    float mortar_size,
+                    float mortar_smooth,
+                    float bias,
+                    float brick_width,
+                    float row_height,
+                    float offset_amount,
+                    float offset_frequency,
+                    float squash_amount,
+                    float squash_frequency,
+                    out vec4 color,
+                    out float fac)
+{
+  vec2 f2 = calc_brick_texture(co * scale,
+                               mortar_size,
+                               mortar_smooth,
+                               bias,
+                               brick_width,
+                               row_height,
+                               offset_amount,
+                               int(offset_frequency),
+                               squash_amount,
+                               int(squash_frequency));
+  float tint = f2.x;
+  float f = f2.y;
+  if (f != 1.0) {
+    float facm = 1.0 - tint;
+    color1 = facm * color1 + tint * color2;
+  }
+  color = mix(color1, mortar, f);
+  fac = f;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_checker.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_checker.glsl
new file mode 100644
index 00000000000..f534f3bddf3
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_checker.glsl
@@ -0,0 +1,17 @@
+void node_tex_checker(
+    vec3 co, vec4 color1, vec4 color2, float scale, out vec4 color, out float fac)
+{
+  vec3 p = co * scale;
+
+  /* Prevent precision issues on unit coordinates. */
+  p = (p + 0.000001) * 0.999999;
+
+  int xi = int(abs(floor(p.x)));
+  int yi = int(abs(floor(p.y)));
+  int zi = int(abs(floor(p.z)));
+
+  bool check = ((mod(xi, 2) == mod(yi, 2)) == bool(mod(zi, 2)));
+
+  color = check ? color1 : color2;
+  fac = check ? 1.0 : 0.0;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_environment.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_environment.glsl
new file mode 100644
index 00000000000..9bd36f8a757
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_environment.glsl
@@ -0,0 +1,44 @@
+void node_tex_environment_texco(vec3 viewvec, out vec3 worldvec)
+{
+#ifdef MESH_SHADER
+  worldvec = worldPosition;
+#else
+  vec4 v = (ProjectionMatrix[3][3] == 0.0) ? vec4(viewvec, 1.0) : vec4(0.0, 0.0, 1.0, 1.0);
+  vec4 co_homogenous = (ProjectionMatrixInverse * v);
+
+  vec3 co = co_homogenous.xyz / co_homogenous.w;
+#  if defined(WORLD_BACKGROUND) || defined(PROBE_CAPTURE)
+  worldvec = mat3(ViewMatrixInverse) * co;
+#  else
+  worldvec = mat3(ModelMatrixInverse) * (mat3(ViewMatrixInverse) * co);
+#  endif
+#endif
+}
+
+void node_tex_environment_equirectangular(vec3 co, float clamp_size, sampler2D ima, out vec3 uv)
+{
+  vec3 nco = normalize(co);
+  uv.x = -atan(nco.y, nco.x) / (2.0 * M_PI) + 0.5;
+  uv.y = atan(nco.z, hypot(nco.x, nco.y)) / M_PI + 0.5;
+
+  /* Fix pole bleeding */
+  float half_height = clamp_size / float(textureSize(ima, 0).y);
+  uv.y = clamp(uv.y, half_height, 1.0 - half_height);
+  uv.z = 0.0;
+}
+
+void node_tex_environment_mirror_ball(vec3 co, out vec3 uv)
+{
+  vec3 nco = normalize(co);
+  nco.y -= 1.0;
+
+  float div = 2.0 * sqrt(max(-0.5 * nco.y, 0.0));
+  nco /= max(1e-8, div);
+
+  uv = 0.5 * nco.xzz + 0.5;
+}
+
+void node_tex_environment_empty(vec3 co, out vec4 color)
+{
+  color = vec4(1.0, 0.0, 1.0, 1.0);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_gradient.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_gradient.glsl
new file mode 100644
index 00000000000..f2f7e615267
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_gradient.glsl
@@ -0,0 +1,47 @@
+float calc_gradient(vec3 p, int gradient_type)
+{
+  float x, y, z;
+  x = p.x;
+  y = p.y;
+  z = p.z;
+  if (gradient_type == 0) { /* linear */
+    return x;
+  }
+  else if (gradient_type == 1) { /* quadratic */
+    float r = max(x, 0.0);
+    return r * r;
+  }
+  else if (gradient_type == 2) { /* easing */
+    float r = min(max(x, 0.0), 1.0);
+    float t = r * r;
+    return (3.0 * t - 2.0 * t * r);
+  }
+  else if (gradient_type == 3) { /* diagonal */
+    return (x + y) * 0.5;
+  }
+  else if (gradient_type == 4) { /* radial */
+    return atan(y, x) / (M_PI * 2) + 0.5;
+  }
+  else {
+    /* Bias a little bit for the case where p is a unit length vector,
+     * to get exactly zero instead of a small random value depending
+     * on float precision. */
+    float r = max(0.999999 - sqrt(x * x + y * y + z * z), 0.0);
+    if (gradient_type == 5) { /* quadratic sphere */
+      return r * r;
+    }
+    else if (gradient_type == 6) { /* sphere */
+      return r;
+    }
+  }
+  return 0.0;
+}
+
+void node_tex_gradient(vec3 co, float gradient_type, out vec4 color, out float fac)
+{
+  float f = calc_gradient(co, int(gradient_type));
+  f = clamp(f, 0.0, 1.0);
+
+  color = vec4(f, f, f, 1.0);
+  fac = f;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_image.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_image.glsl
new file mode 100644
index 00000000000..c234e064d36
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_image.glsl
@@ -0,0 +1,355 @@
+void point_texco_remap_square(vec3 vin, out vec3 vout)
+{
+  vout = vin * 2.0 - 1.0;
+}
+
+void point_texco_clamp(vec3 vin, sampler2D ima, out vec3 vout)
+{
+  vec2 half_texel_size = 0.5 / vec2(textureSize(ima, 0).xy);
+  vout = clamp(vin, half_texel_size.xyy, 1.0 - half_texel_size.xyy);
+}
+
+void point_map_to_sphere(vec3 vin, out vec3 vout)
+{
+  float len = length(vin);
+  float v, u;
+  if (len > 0.0) {
+    if (vin.x == 0.0 && vin.y == 0.0) {
+      u = 0.0;
+    }
+    else {
+      u = (1.0 - atan(vin.x, vin.y) / M_PI) / 2.0;
+    }
+
+    v = 1.0 - acos(vin.z / len) / M_PI;
+  }
+  else {
+    v = u = 0.0;
+  }
+
+  vout = vec3(u, v, 0.0);
+}
+
+void point_map_to_tube(vec3 vin, out vec3 vout)
+{
+  float u, v;
+  v = (vin.z + 1.0) * 0.5;
+  float len = sqrt(vin.x * vin.x + vin.y * vin[1]);
+  if (len > 0.0) {
+    u = (1.0 - (atan(vin.x / len, vin.y / len) / M_PI)) * 0.5;
+  }
+  else {
+    v = u = 0.0;
+  }
+
+  vout = vec3(u, v, 0.0);
+}
+
+/* 16bits floats limits. Higher/Lower values produce +/-inf. */
+#define safe_color(a) (clamp(a, -65520.0, 65520.0))
+
+void node_tex_image_linear(vec3 co, sampler2D ima, out vec4 color, out float alpha)
+{
+  color = safe_color(texture(ima, co.xy));
+  alpha = color.a;
+}
+
+void node_tex_image_linear_no_mip(vec3 co, sampler2D ima, out vec4 color, out float alpha)
+{
+  color = safe_color(textureLod(ima, co.xy, 0.0));
+  alpha = color.a;
+}
+
+void node_tex_image_nearest(vec3 co, sampler2D ima, out vec4 color, out float alpha)
+{
+  ivec2 pix = ivec2(fract(co.xy) * textureSize(ima, 0).xy);
+  color = safe_color(texelFetch(ima, pix, 0));
+  alpha = color.a;
+}
+
+/* @arg f: signed distance to texel center. */
+void cubic_bspline_coefs(vec2 f, out vec2 w0, out vec2 w1, out vec2 w2, out vec2 w3)
+{
+  vec2 f2 = f * f;
+  vec2 f3 = f2 * f;
+  /* Bspline coefs (optimized) */
+  w3 = f3 / 6.0;
+  w0 = -w3 + f2 * 0.5 - f * 0.5 + 1.0 / 6.0;
+  w1 = f3 * 0.5 - f2 * 1.0 + 2.0 / 3.0;
+  w2 = 1.0 - w0 - w1 - w3;
+}
+
+void node_tex_image_cubic_ex(
+    vec3 co, sampler2D ima, float do_extend, out vec4 color, out float alpha)
+{
+  vec2 tex_size = vec2(textureSize(ima, 0).xy);
+
+  co.xy *= tex_size;
+  /* texel center */
+  vec2 tc = floor(co.xy - 0.5) + 0.5;
+  vec2 w0, w1, w2, w3;
+  cubic_bspline_coefs(co.xy - tc, w0, w1, w2, w3);
+
+#if 1 /* Optimized version using 4 filtered tap. */
+  vec2 s0 = w0 + w1;
+  vec2 s1 = w2 + w3;
+
+  vec2 f0 = w1 / (w0 + w1);
+  vec2 f1 = w3 / (w2 + w3);
+
+  vec4 final_co;
+  final_co.xy = tc - 1.0 + f0;
+  final_co.zw = tc + 1.0 + f1;
+
+  if (do_extend == 1.0) {
+    final_co = clamp(final_co, vec4(0.5), tex_size.xyxy - 0.5);
+  }
+  final_co /= tex_size.xyxy;
+
+  color = safe_color(textureLod(ima, final_co.xy, 0.0)) * s0.x * s0.y;
+  color += safe_color(textureLod(ima, final_co.zy, 0.0)) * s1.x * s0.y;
+  color += safe_color(textureLod(ima, final_co.xw, 0.0)) * s0.x * s1.y;
+  color += safe_color(textureLod(ima, final_co.zw, 0.0)) * s1.x * s1.y;
+
+#else /* Reference bruteforce 16 tap. */
+  color = texelFetch(ima, ivec2(tc + vec2(-1.0, -1.0)), 0) * w0.x * w0.y;
+  color += texelFetch(ima, ivec2(tc + vec2(0.0, -1.0)), 0) * w1.x * w0.y;
+  color += texelFetch(ima, ivec2(tc + vec2(1.0, -1.0)), 0) * w2.x * w0.y;
+  color += texelFetch(ima, ivec2(tc + vec2(2.0, -1.0)), 0) * w3.x * w0.y;
+
+  color += texelFetch(ima, ivec2(tc + vec2(-1.0, 0.0)), 0) * w0.x * w1.y;
+  color += texelFetch(ima, ivec2(tc + vec2(0.0, 0.0)), 0) * w1.x * w1.y;
+  color += texelFetch(ima, ivec2(tc + vec2(1.0, 0.0)), 0) * w2.x * w1.y;
+  color += texelFetch(ima, ivec2(tc + vec2(2.0, 0.0)), 0) * w3.x * w1.y;
+
+  color += texelFetch(ima, ivec2(tc + vec2(-1.0, 1.0)), 0) * w0.x * w2.y;
+  color += texelFetch(ima, ivec2(tc + vec2(0.0, 1.0)), 0) * w1.x * w2.y;
+  color += texelFetch(ima, ivec2(tc + vec2(1.0, 1.0)), 0) * w2.x * w2.y;
+  color += texelFetch(ima, ivec2(tc + vec2(2.0, 1.0)), 0) * w3.x * w2.y;
+
+  color += texelFetch(ima, ivec2(tc + vec2(-1.0, 2.0)), 0) * w0.x * w3.y;
+  color += texelFetch(ima, ivec2(tc + vec2(0.0, 2.0)), 0) * w1.x * w3.y;
+  color += texelFetch(ima, ivec2(tc + vec2(1.0, 2.0)), 0) * w2.x * w3.y;
+  color += texelFetch(ima, ivec2(tc + vec2(2.0, 2.0)), 0) * w3.x * w3.y;
+#endif
+
+  alpha = color.a;
+}
+
+void node_tex_image_cubic(vec3 co, sampler2D ima, out vec4 color, out float alpha)
+{
+  node_tex_image_cubic_ex(co, ima, 0.0, color, alpha);
+}
+
+void node_tex_image_cubic_extend(vec3 co, sampler2D ima, out vec4 color, out float alpha)
+{
+  node_tex_image_cubic_ex(co, ima, 1.0, color, alpha);
+}
+
+void node_tex_image_smart(vec3 co, sampler2D ima, out vec4 color, out float alpha)
+{
+  /* use cubic for now */
+  node_tex_image_cubic_ex(co, ima, 0.0, color, alpha);
+}
+
+void tex_box_sample_linear(
+    vec3 texco, vec3 N, sampler2D ima, out vec4 color1, out vec4 color2, out vec4 color3)
+{
+  /* X projection */
+  vec2 uv = texco.yz;
+  if (N.x < 0.0) {
+    uv.x = 1.0 - uv.x;
+  }
+  color1 = texture(ima, uv);
+  /* Y projection */
+  uv = texco.xz;
+  if (N.y > 0.0) {
+    uv.x = 1.0 - uv.x;
+  }
+  color2 = texture(ima, uv);
+  /* Z projection */
+  uv = texco.yx;
+  if (N.z > 0.0) {
+    uv.x = 1.0 - uv.x;
+  }
+  color3 = texture(ima, uv);
+}
+
+void tex_box_sample_nearest(
+    vec3 texco, vec3 N, sampler2D ima, out vec4 color1, out vec4 color2, out vec4 color3)
+{
+  /* X projection */
+  vec2 uv = texco.yz;
+  if (N.x < 0.0) {
+    uv.x = 1.0 - uv.x;
+  }
+  ivec2 pix = ivec2(uv.xy * textureSize(ima, 0).xy);
+  color1 = texelFetch(ima, pix, 0);
+  /* Y projection */
+  uv = texco.xz;
+  if (N.y > 0.0) {
+    uv.x = 1.0 - uv.x;
+  }
+  pix = ivec2(uv.xy * textureSize(ima, 0).xy);
+  color2 = texelFetch(ima, pix, 0);
+  /* Z projection */
+  uv = texco.yx;
+  if (N.z > 0.0) {
+    uv.x = 1.0 - uv.x;
+  }
+  pix = ivec2(uv.xy * textureSize(ima, 0).xy);
+  color3 = texelFetch(ima, pix, 0);
+}
+
+void tex_box_sample_cubic(
+    vec3 texco, vec3 N, sampler2D ima, out vec4 color1, out vec4 color2, out vec4 color3)
+{
+  float alpha;
+  /* X projection */
+  vec2 uv = texco.yz;
+  if (N.x < 0.0) {
+    uv.x = 1.0 - uv.x;
+  }
+  node_tex_image_cubic_ex(uv.xyy, ima, 0.0, color1, alpha);
+  /* Y projection */
+  uv = texco.xz;
+  if (N.y > 0.0) {
+    uv.x = 1.0 - uv.x;
+  }
+  node_tex_image_cubic_ex(uv.xyy, ima, 0.0, color2, alpha);
+  /* Z projection */
+  uv = texco.yx;
+  if (N.z > 0.0) {
+    uv.x = 1.0 - uv.x;
+  }
+  node_tex_image_cubic_ex(uv.xyy, ima, 0.0, color3, alpha);
+}
+
+void tex_box_sample_smart(
+    vec3 texco, vec3 N, sampler2D ima, out vec4 color1, out vec4 color2, out vec4 color3)
+{
+  tex_box_sample_cubic(texco, N, ima, color1, color2, color3);
+}
+
+void node_tex_image_box(vec3 texco,
+                        vec3 N,
+                        vec4 color1,
+                        vec4 color2,
+                        vec4 color3,
+                        sampler2D ima,
+                        float blend,
+                        out vec4 color,
+                        out float alpha)
+{
+  /* project from direction vector to barycentric coordinates in triangles */
+  N = abs(N);
+  N /= dot(N, vec3(1.0));
+
+  /* basic idea is to think of this as a triangle, each corner representing
+   * one of the 3 faces of the cube. in the corners we have single textures,
+   * in between we blend between two textures, and in the middle we a blend
+   * between three textures.
+   *
+   * the Nxyz values are the barycentric coordinates in an equilateral
+   * triangle, which in case of blending, in the middle has a smaller
+   * equilateral triangle where 3 textures blend. this divides things into
+   * 7 zones, with an if () test for each zone
+   * EDIT: Now there is only 4 if's. */
+
+  float limit = 0.5 + 0.5 * blend;
+
+  vec3 weight;
+  weight = N.xyz / (N.xyx + N.yzz);
+  weight = clamp((weight - 0.5 * (1.0 - blend)) / max(1e-8, blend), 0.0, 1.0);
+
+  /* test for mixes between two textures */
+  if (N.z < (1.0 - limit) * (N.y + N.x)) {
+    weight.z = 0.0;
+    weight.y = 1.0 - weight.x;
+  }
+  else if (N.x < (1.0 - limit) * (N.y + N.z)) {
+    weight.x = 0.0;
+    weight.z = 1.0 - weight.y;
+  }
+  else if (N.y < (1.0 - limit) * (N.x + N.z)) {
+    weight.y = 0.0;
+    weight.x = 1.0 - weight.z;
+  }
+  else {
+    /* last case, we have a mix between three */
+    weight = ((2.0 - limit) * N + (limit - 1.0)) / max(1e-8, blend);
+  }
+
+  color = weight.x * color1 + weight.y * color2 + weight.z * color3;
+  alpha = color.a;
+}
+
+void tex_clip_linear(vec3 co, sampler2D ima, vec4 icolor, out vec4 color, out float alpha)
+{
+  vec2 tex_size = vec2(textureSize(ima, 0).xy);
+  vec2 minco = min(co.xy, 1.0 - co.xy);
+  minco = clamp(minco * tex_size + 0.5, 0.0, 1.0);
+  float fac = minco.x * minco.y;
+
+  color = mix(vec4(0.0), icolor, fac);
+  alpha = color.a;
+}
+
+void tex_clip_nearest(vec3 co, sampler2D ima, vec4 icolor, out vec4 color, out float alpha)
+{
+  vec4 minco = vec4(co.xy, 1.0 - co.xy);
+  color = (any(lessThan(minco, vec4(0.0)))) ? vec4(0.0) : icolor;
+  alpha = color.a;
+}
+
+void tex_clip_cubic(vec3 co, sampler2D ima, vec4 icolor, out vec4 color, out float alpha)
+{
+  vec2 tex_size = vec2(textureSize(ima, 0).xy);
+
+  co.xy *= tex_size;
+  /* texel center */
+  vec2 tc = floor(co.xy - 0.5) + 0.5;
+  vec2 w0, w1, w2, w3;
+  cubic_bspline_coefs(co.xy - tc, w0, w1, w2, w3);
+
+  /* TODO Optimize this part. I'm sure there is a smarter way to do that.
+   * Could do that when sampling? */
+#define CLIP_CUBIC_SAMPLE(samp, size) \
+  (float(all(greaterThan(samp, vec2(-0.5)))) * float(all(lessThan(ivec2(samp), itex_size))))
+  ivec2 itex_size = textureSize(ima, 0).xy;
+  float fac;
+  fac = CLIP_CUBIC_SAMPLE(tc + vec2(-1.0, -1.0), itex_size) * w0.x * w0.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(0.0, -1.0), itex_size) * w1.x * w0.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(1.0, -1.0), itex_size) * w2.x * w0.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(2.0, -1.0), itex_size) * w3.x * w0.y;
+
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(-1.0, 0.0), itex_size) * w0.x * w1.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(0.0, 0.0), itex_size) * w1.x * w1.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(1.0, 0.0), itex_size) * w2.x * w1.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(2.0, 0.0), itex_size) * w3.x * w1.y;
+
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(-1.0, 1.0), itex_size) * w0.x * w2.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(0.0, 1.0), itex_size) * w1.x * w2.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(1.0, 1.0), itex_size) * w2.x * w2.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(2.0, 1.0), itex_size) * w3.x * w2.y;
+
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(-1.0, 2.0), itex_size) * w0.x * w3.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(0.0, 2.0), itex_size) * w1.x * w3.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(1.0, 2.0), itex_size) * w2.x * w3.y;
+  fac += CLIP_CUBIC_SAMPLE(tc + vec2(2.0, 2.0), itex_size) * w3.x * w3.y;
+#undef CLIP_CUBIC_SAMPLE
+
+  color = mix(vec4(0.0), icolor, fac);
+  alpha = color.a;
+}
+
+void tex_clip_smart(vec3 co, sampler2D ima, vec4 icolor, out vec4 color, out float alpha)
+{
+  tex_clip_cubic(co, ima, icolor, color, alpha);
+}
+
+void node_tex_image_empty(vec3 co, out vec4 color, out float alpha)
+{
+  color = vec4(0.0);
+  alpha = 0.0;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_magic.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_magic.glsl
new file mode 100644
index 00000000000..942c507cc38
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_magic.glsl
@@ -0,0 +1,61 @@
+void node_tex_magic(
+    vec3 co, float scale, float distortion, float depth, out vec4 color, out float fac)
+{
+  vec3 p = co * scale;
+  float x = sin((p.x + p.y + p.z) * 5.0);
+  float y = cos((-p.x + p.y - p.z) * 5.0);
+  float z = -cos((-p.x - p.y + p.z) * 5.0);
+
+  if (depth > 0) {
+    x *= distortion;
+    y *= distortion;
+    z *= distortion;
+    y = -cos(x - y + z);
+    y *= distortion;
+    if (depth > 1) {
+      x = cos(x - y - z);
+      x *= distortion;
+      if (depth > 2) {
+        z = sin(-x - y - z);
+        z *= distortion;
+        if (depth > 3) {
+          x = -cos(-x + y - z);
+          x *= distortion;
+          if (depth > 4) {
+            y = -sin(-x + y + z);
+            y *= distortion;
+            if (depth > 5) {
+              y = -cos(-x + y + z);
+              y *= distortion;
+              if (depth > 6) {
+                x = cos(x + y + z);
+                x *= distortion;
+                if (depth > 7) {
+                  z = sin(x + y - z);
+                  z *= distortion;
+                  if (depth > 8) {
+                    x = -cos(-x - y + z);
+                    x *= distortion;
+                    if (depth > 9) {
+                      y = -sin(x - y + z);
+                      y *= distortion;
+                    }
+                  }
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  if (distortion != 0.0) {
+    distortion *= 2.0;
+    x /= distortion;
+    y /= distortion;
+    z /= distortion;
+  }
+
+  color = vec4(0.5 - x, 0.5 - y, 0.5 - z, 1.0);
+  fac = (color.x + color.y + color.z) / 3.0;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_musgrave.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_musgrave.glsl
new file mode 100644
index 00000000000..52332c45c3d
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_musgrave.glsl
@@ -0,0 +1,208 @@
+/* Musgrave fBm
+ *
+ * H: fractal increment parameter
+ * lacunarity: gap between successive frequencies
+ * octaves: number of frequencies in the fBm
+ *
+ * from "Texturing and Modelling: A procedural approach"
+ */
+
+float noise_musgrave_fBm(vec3 p, float H, float lacunarity, float octaves)
+{
+  float rmd;
+  float value = 0.0;
+  float pwr = 1.0;
+  float pwHL = pow(lacunarity, -H);
+
+  for (int i = 0; i < int(octaves); i++) {
+    value += snoise(p) * pwr;
+    pwr *= pwHL;
+    p *= lacunarity;
+  }
+
+  rmd = octaves - floor(octaves);
+  if (rmd != 0.0) {
+    value += rmd * snoise(p) * pwr;
+  }
+
+  return value;
+}
+
+/* Musgrave Multifractal
+ *
+ * H: highest fractal dimension
+ * lacunarity: gap between successive frequencies
+ * octaves: number of frequencies in the fBm
+ */
+
+float noise_musgrave_multi_fractal(vec3 p, float H, float lacunarity, float octaves)
+{
+  float rmd;
+  float value = 1.0;
+  float pwr = 1.0;
+  float pwHL = pow(lacunarity, -H);
+
+  for (int i = 0; i < int(octaves); i++) {
+    value *= (pwr * snoise(p) + 1.0);
+    pwr *= pwHL;
+    p *= lacunarity;
+  }
+
+  rmd = octaves - floor(octaves);
+  if (rmd != 0.0) {
+    value *= (rmd * pwr * snoise(p) + 1.0); /* correct? */
+  }
+
+  return value;
+}
+
+/* Musgrave Heterogeneous Terrain
+ *
+ * H: fractal dimension of the roughest area
+ * lacunarity: gap between successive frequencies
+ * octaves: number of frequencies in the fBm
+ * offset: raises the terrain from `sea level'
+ */
+
+float noise_musgrave_hetero_terrain(vec3 p, float H, float lacunarity, float octaves, float offset)
+{
+  float value, increment, rmd;
+  float pwHL = pow(lacunarity, -H);
+  float pwr = pwHL;
+
+  /* first unscaled octave of function; later octaves are scaled */
+  value = offset + snoise(p);
+  p *= lacunarity;
+
+  for (int i = 1; i < int(octaves); i++) {
+    increment = (snoise(p) + offset) * pwr * value;
+    value += increment;
+    pwr *= pwHL;
+    p *= lacunarity;
+  }
+
+  rmd = octaves - floor(octaves);
+  if (rmd != 0.0) {
+    increment = (snoise(p) + offset) * pwr * value;
+    value += rmd * increment;
+  }
+
+  return value;
+}
+
+/* Hybrid Additive/Multiplicative Multifractal Terrain
+ *
+ * H: fractal dimension of the roughest area
+ * lacunarity: gap between successive frequencies
+ * octaves: number of frequencies in the fBm
+ * offset: raises the terrain from `sea level'
+ */
+
+float noise_musgrave_hybrid_multi_fractal(
+    vec3 p, float H, float lacunarity, float octaves, float offset, float gain)
+{
+  float result, signal, weight, rmd;
+  float pwHL = pow(lacunarity, -H);
+  float pwr = pwHL;
+
+  result = snoise(p) + offset;
+  weight = gain * result;
+  p *= lacunarity;
+
+  for (int i = 1; (weight > 0.001f) && (i < int(octaves)); i++) {
+    if (weight > 1.0) {
+      weight = 1.0;
+    }
+
+    signal = (snoise(p) + offset) * pwr;
+    pwr *= pwHL;
+    result += weight * signal;
+    weight *= gain * signal;
+    p *= lacunarity;
+  }
+
+  rmd = octaves - floor(octaves);
+  if (rmd != 0.0) {
+    result += rmd * ((snoise(p) + offset) * pwr);
+  }
+
+  return result;
+}
+
+/* Ridged Multifractal Terrain
+ *
+ * H: fractal dimension of the roughest area
+ * lacunarity: gap between successive frequencies
+ * octaves: number of frequencies in the fBm
+ * offset: raises the terrain from `sea level'
+ */
+
+float noise_musgrave_ridged_multi_fractal(
+    vec3 p, float H, float lacunarity, float octaves, float offset, float gain)
+{
+  float result, signal, weight;
+  float pwHL = pow(lacunarity, -H);
+  float pwr = pwHL;
+
+  signal = offset - abs(snoise(p));
+  signal *= signal;
+  result = signal;
+  weight = 1.0;
+
+  for (int i = 1; i < int(octaves); i++) {
+    p *= lacunarity;
+    weight = clamp(signal * gain, 0.0, 1.0);
+    signal = offset - abs(snoise(p));
+    signal *= signal;
+    signal *= weight;
+    result += signal * pwr;
+    pwr *= pwHL;
+  }
+
+  return result;
+}
+
+float svm_musgrave(int type,
+                   float dimension,
+                   float lacunarity,
+                   float octaves,
+                   float offset,
+                   float intensity,
+                   float gain,
+                   vec3 p)
+{
+  if (type == 0 /* NODE_MUSGRAVE_MULTIFRACTAL */) {
+    return intensity * noise_musgrave_multi_fractal(p, dimension, lacunarity, octaves);
+  }
+  else if (type == 1 /* NODE_MUSGRAVE_FBM */) {
+    return intensity * noise_musgrave_fBm(p, dimension, lacunarity, octaves);
+  }
+  else if (type == 2 /* NODE_MUSGRAVE_HYBRID_MULTIFRACTAL */) {
+    return intensity *
+           noise_musgrave_hybrid_multi_fractal(p, dimension, lacunarity, octaves, offset, gain);
+  }
+  else if (type == 3 /* NODE_MUSGRAVE_RIDGED_MULTIFRACTAL */) {
+    return intensity *
+           noise_musgrave_ridged_multi_fractal(p, dimension, lacunarity, octaves, offset, gain);
+  }
+  else if (type == 4 /* NODE_MUSGRAVE_HETERO_TERRAIN */) {
+    return intensity * noise_musgrave_hetero_terrain(p, dimension, lacunarity, octaves, offset);
+  }
+  return 0.0;
+}
+
+void node_tex_musgrave(vec3 co,
+                       float scale,
+                       float detail,
+                       float dimension,
+                       float lacunarity,
+                       float offset,
+                       float gain,
+                       float type,
+                       out vec4 color,
+                       out float fac)
+{
+  fac = svm_musgrave(int(type), dimension, lacunarity, detail, offset, 1.0, gain, co *scale);
+
+  color = vec4(fac, fac, fac, 1.0);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_noise.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_noise.glsl
new file mode 100644
index 00000000000..30e27da16e8
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_noise.glsl
@@ -0,0 +1,19 @@
+void node_tex_noise(
+    vec3 co, float scale, float detail, float distortion, out vec4 color, out float fac)
+{
+  vec3 p = co * scale;
+  int hard = 0;
+  if (distortion != 0.0) {
+    vec3 r, offset = vec3(13.5, 13.5, 13.5);
+    r.x = noise(p + offset) * distortion;
+    r.y = noise(p) * distortion;
+    r.z = noise(p - offset) * distortion;
+    p += r;
+  }
+
+  fac = noise_turbulence(p, detail, hard);
+  color = vec4(fac,
+               noise_turbulence(vec3(p.y, p.x, p.z), detail, hard),
+               noise_turbulence(vec3(p.y, p.z, p.x), detail, hard),
+               1);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_sky.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_sky.glsl
new file mode 100644
index 00000000000..981d17b4283
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_sky.glsl
@@ -0,0 +1,4 @@
+void node_tex_sky(vec3 co, out vec4 color)
+{
+  color = vec4(1.0);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_voronoi.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_voronoi.glsl
new file mode 100644
index 00000000000..018b74fd1a5
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_voronoi.glsl
@@ -0,0 +1,116 @@
+void node_tex_voronoi(vec3 co,
+                      float scale,
+                      float exponent,
+                      float coloring,
+                      float metric,
+                      float feature,
+                      out vec4 color,
+                      out float fac)
+{
+  vec3 p = co * scale;
+  int xx, yy, zz, xi, yi, zi;
+  vec4 da = vec4(1e10);
+  vec3 pa[4] = vec3[4](vec3(0.0), vec3(0.0), vec3(0.0), vec3(0.0));
+
+  xi = floor_to_int(p[0]);
+  yi = floor_to_int(p[1]);
+  zi = floor_to_int(p[2]);
+
+  for (xx = xi - 1; xx <= xi + 1; xx++) {
+    for (yy = yi - 1; yy <= yi + 1; yy++) {
+      for (zz = zi - 1; zz <= zi + 1; zz++) {
+        vec3 ip = vec3(xx, yy, zz);
+        vec3 vp = cellnoise_color(ip);
+        vec3 pd = p - (vp + ip);
+
+        float d = 0.0;
+        if (metric == 0.0) { /* SHD_VORONOI_DISTANCE 0 */
+          d = dot(pd, pd);
+        }
+        else if (metric == 1.0) { /* SHD_VORONOI_MANHATTAN 1 */
+          d = abs(pd[0]) + abs(pd[1]) + abs(pd[2]);
+        }
+        else if (metric == 2.0) { /* SHD_VORONOI_CHEBYCHEV 2 */
+          d = max(abs(pd[0]), max(abs(pd[1]), abs(pd[2])));
+        }
+        else if (metric == 3.0) { /* SHD_VORONOI_MINKOWSKI 3 */
+          d = pow(pow(abs(pd[0]), exponent) + pow(abs(pd[1]), exponent) +
+                      pow(abs(pd[2]), exponent),
+                  1.0 / exponent);
+        }
+
+        vp += vec3(xx, yy, zz);
+        if (d < da[0]) {
+          da.yzw = da.xyz;
+          da[0] = d;
+
+          pa[3] = pa[2];
+          pa[2] = pa[1];
+          pa[1] = pa[0];
+          pa[0] = vp;
+        }
+        else if (d < da[1]) {
+          da.zw = da.yz;
+          da[1] = d;
+
+          pa[3] = pa[2];
+          pa[2] = pa[1];
+          pa[1] = vp;
+        }
+        else if (d < da[2]) {
+          da[3] = da[2];
+          da[2] = d;
+
+          pa[3] = pa[2];
+          pa[2] = vp;
+        }
+        else if (d < da[3]) {
+          da[3] = d;
+          pa[3] = vp;
+        }
+      }
+    }
+  }
+
+  if (coloring == 0.0) {
+    /* Intensity output */
+    if (feature == 0.0) { /* F1 */
+      fac = abs(da[0]);
+    }
+    else if (feature == 1.0) { /* F2 */
+      fac = abs(da[1]);
+    }
+    else if (feature == 2.0) { /* F3 */
+      fac = abs(da[2]);
+    }
+    else if (feature == 3.0) { /* F4 */
+      fac = abs(da[3]);
+    }
+    else if (feature == 4.0) { /* F2F1 */
+      fac = abs(da[1] - da[0]);
+    }
+    color = vec4(fac, fac, fac, 1.0);
+  }
+  else {
+    /* Color output */
+    vec3 col = vec3(fac, fac, fac);
+    if (feature == 0.0) { /* F1 */
+      col = pa[0];
+    }
+    else if (feature == 1.0) { /* F2 */
+      col = pa[1];
+    }
+    else if (feature == 2.0) { /* F3 */
+      col = pa[2];
+    }
+    else if (feature == 3.0) { /* F4 */
+      col = pa[3];
+    }
+    else if (feature == 4.0) { /* F2F1 */
+      col = abs(pa[1] - pa[0]);
+    }
+
+    color = vec4(cellnoise_color(col), 1.0);
+    fac = (color.x + color.y + color.z) * (1.0 / 3.0);
+  }
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_wave.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_wave.glsl
new file mode 100644
index 00000000000..2196848a4ef
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_wave.glsl
@@ -0,0 +1,42 @@
+float calc_wave(
+    vec3 p, float distortion, float detail, float detail_scale, int wave_type, int wave_profile)
+{
+  float n;
+
+  if (wave_type == 0) { /* type bands */
+    n = (p.x + p.y + p.z) * 10.0;
+  }
+  else { /* type rings */
+    n = length(p) * 20.0;
+  }
+
+  if (distortion != 0.0) {
+    n += distortion * noise_turbulence(p * detail_scale, detail, 0);
+  }
+
+  if (wave_profile == 0) { /* profile sin */
+    return 0.5 + 0.5 * sin(n);
+  }
+  else { /* profile saw */
+    n /= 2.0 * M_PI;
+    n -= int(n);
+    return (n < 0.0) ? n + 1.0 : n;
+  }
+}
+
+void node_tex_wave(vec3 co,
+                   float scale,
+                   float distortion,
+                   float detail,
+                   float detail_scale,
+                   float wave_type,
+                   float wave_profile,
+                   out vec4 color,
+                   out float fac)
+{
+  float f;
+  f = calc_wave(co * scale, distortion, detail, detail_scale, int(wave_type), int(wave_profile));
+
+  color = vec4(f, f, f, 1.0);
+  fac = f;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_tex_white_noise.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_tex_white_noise.glsl
new file mode 100644
index 00000000000..fce511deb79
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_tex_white_noise.glsl
@@ -0,0 +1,21 @@
+/* White Noise */
+
+void node_white_noise_1d(vec3 vector, float w, out float value)
+{
+  value = hash_float_to_float(w);
+}
+
+void node_white_noise_2d(vec3 vector, float w, out float value)
+{
+  value = hash_vec2_to_float(vector.xy);
+}
+
+void node_white_noise_3d(vec3 vector, float w, out float value)
+{
+  value = hash_vec3_to_float(vector);
+}
+
+void node_white_noise_4d(vec3 vector, float w, out float value)
+{
+  value = hash_vec4_to_float(vec4(vector, w));
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_texture_coordinates.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_texture_coordinates.glsl
new file mode 100644
index 00000000000..24276156d55
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_texture_coordinates.glsl
@@ -0,0 +1,92 @@
+vec3 mtex_2d_mapping(vec3 vec)
+{
+  return vec3(vec.xy * 0.5 + vec2(0.5), vec.z);
+}
+
+void generated_from_orco(vec3 orco, out vec3 generated)
+{
+#ifdef VOLUMETRICS
+#  ifdef MESH_SHADER
+  generated = volumeObjectLocalCoord;
+#  else
+  generated = worldPosition;
+#  endif
+#else
+  generated = orco;
+#endif
+}
+
+void generated_texco(vec3 I, vec3 attr_orco, out vec3 generated)
+{
+  vec4 v = (ProjectionMatrix[3][3] == 0.0) ? vec4(I, 1.0) : vec4(0.0, 0.0, 1.0, 1.0);
+  vec4 co_homogenous = (ProjectionMatrixInverse * v);
+  vec4 co = vec4(co_homogenous.xyz / co_homogenous.w, 0.0);
+  co.xyz = normalize(co.xyz);
+#if defined(WORLD_BACKGROUND) || defined(PROBE_CAPTURE)
+  generated = (ViewMatrixInverse * co).xyz;
+#else
+  generated_from_orco(attr_orco, generated);
+#endif
+}
+
+void node_tex_coord(vec3 I,
+                    vec3 wN,
+                    mat4 obmatinv,
+                    vec4 camerafac,
+                    vec3 attr_orco,
+                    vec3 attr_uv,
+                    out vec3 generated,
+                    out vec3 normal,
+                    out vec3 uv,
+                    out vec3 object,
+                    out vec3 camera,
+                    out vec3 window,
+                    out vec3 reflection)
+{
+  generated = attr_orco;
+  normal = normalize(normal_world_to_object(wN));
+  uv = attr_uv;
+  object = (obmatinv * (ViewMatrixInverse * vec4(I, 1.0))).xyz;
+  camera = vec3(I.xy, -I.z);
+  vec4 projvec = ProjectionMatrix * vec4(I, 1.0);
+  window = vec3(mtex_2d_mapping(projvec.xyz / projvec.w).xy * camerafac.xy + camerafac.zw, 0.0);
+  reflection = -reflect(cameraVec, normalize(wN));
+}
+
+void node_tex_coord_background(vec3 I,
+                               vec3 N,
+                               mat4 obmatinv,
+                               vec4 camerafac,
+                               vec3 attr_orco,
+                               vec3 attr_uv,
+                               out vec3 generated,
+                               out vec3 normal,
+                               out vec3 uv,
+                               out vec3 object,
+                               out vec3 camera,
+                               out vec3 window,
+                               out vec3 reflection)
+{
+  vec4 v = (ProjectionMatrix[3][3] == 0.0) ? vec4(I, 1.0) : vec4(0.0, 0.0, 1.0, 1.0);
+  vec4 co_homogenous = (ProjectionMatrixInverse * v);
+
+  vec4 co = vec4(co_homogenous.xyz / co_homogenous.w, 0.0);
+
+  co = normalize(co);
+
+  vec3 coords = (ViewMatrixInverse * co).xyz;
+
+  generated = coords;
+  normal = -coords;
+  uv = vec3(attr_uv.xy, 0.0);
+  object = (obmatinv * vec4(coords, 1.0)).xyz;
+
+  camera = vec3(co.xy, -co.z);
+  window = vec3(mtex_2d_mapping(I).xy * camerafac.xy + camerafac.zw, 0.0);
+
+  reflection = -coords;
+}
+
+#if defined(WORLD_BACKGROUND) || (defined(PROBE_CAPTURE) && !defined(MESH_SHADER))
+#  define node_tex_coord node_tex_coord_background
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_toon.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_toon.glsl
new file mode 100644
index 00000000000..02d288d42bf
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_toon.glsl
@@ -0,0 +1,9 @@
+#ifndef VOLUMETRICS
+void node_bsdf_toon(vec4 color, float size, float tsmooth, vec3 N, out Closure result)
+{
+  node_bsdf_diffuse(color, 0.0, N, result);
+}
+#else
+/* Stub toon because it is not compatible with volumetrics. */
+#  define node_bsdf_toon
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_translucent.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_translucent.glsl
new file mode 100644
index 00000000000..ea41790e6aa
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_translucent.glsl
@@ -0,0 +1,9 @@
+#ifndef VOLUMETRICS
+void node_bsdf_translucent(vec4 color, vec3 N, out Closure result)
+{
+  node_bsdf_diffuse(color, 0.0, -N, result);
+}
+#else
+/* Stub translucent because it is not compatible with volumetrics. */
+#  define node_bsdf_translucent
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_transparent.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_transparent.glsl
new file mode 100644
index 00000000000..800d0f81d4a
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_transparent.glsl
@@ -0,0 +1,11 @@
+#ifndef VOLUMETRICS
+void node_bsdf_transparent(vec4 color, out Closure result)
+{
+  result = CLOSURE_DEFAULT;
+  result.radiance = vec3(0.0);
+  result.transmittance = abs(color.rgb);
+}
+#else
+/* Stub transparent because it is not compatible with volumetrics. */
+#  define node_bsdf_transparent
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_uv_map.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_uv_map.glsl
new file mode 100644
index 00000000000..d8fcbbfc361
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_uv_map.glsl
@@ -0,0 +1,4 @@
+void node_uvmap(vec3 attr_uv, out vec3 outvec)
+{
+  outvec = attr_uv;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_vector_curves.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_vector_curves.glsl
new file mode 100644
index 00000000000..35d2e903cf4
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_vector_curves.glsl
@@ -0,0 +1,8 @@
+void curves_vec(float fac, vec3 vec, sampler1DArray curvemap, float layer, out vec3 outvec)
+{
+  vec4 co = vec4(vec * 0.5 + 0.5, layer);
+  outvec.x = texture(curvemap, co.xw).x;
+  outvec.y = texture(curvemap, co.yw).y;
+  outvec.z = texture(curvemap, co.zw).z;
+  outvec = mix(vec, outvec, fac);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_vector_displacement.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_vector_displacement.glsl
new file mode 100644
index 00000000000..b6b955dcdb4
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_vector_displacement.glsl
@@ -0,0 +1,30 @@
+void node_vector_displacement_tangent(vec4 vector,
+                                      float midlevel,
+                                      float scale,
+                                      vec4 tangent,
+                                      vec3 normal,
+                                      mat4 obmat,
+                                      mat4 viewmat,
+                                      out vec3 result)
+{
+  /* TODO(fclem) this is broken. revisit latter. */
+  vec3 N_object = normalize(((vec4(normal, 0.0) * viewmat) * obmat).xyz);
+  vec3 T_object = normalize(((vec4(tangent.xyz, 0.0) * viewmat) * obmat).xyz);
+  vec3 B_object = tangent.w * normalize(cross(N_object, T_object));
+
+  vec3 offset = (vector.xyz - vec3(midlevel)) * scale;
+  result = offset.x * T_object + offset.y * N_object + offset.z * B_object;
+  result = (obmat * vec4(result, 0.0)).xyz;
+}
+
+void node_vector_displacement_object(
+    vec4 vector, float midlevel, float scale, mat4 obmat, out vec3 result)
+{
+  result = (vector.xyz - vec3(midlevel)) * scale;
+  result = (obmat * vec4(result, 0.0)).xyz;
+}
+
+void node_vector_displacement_world(vec4 vector, float midlevel, float scale, out vec3 result)
+{
+  result = (vector.xyz - vec3(midlevel)) * scale;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_vector_math.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_vector_math.glsl
new file mode 100644
index 00000000000..93132b6044f
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_vector_math.glsl
@@ -0,0 +1,100 @@
+void vector_math_add(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = a + b;
+}
+
+void vector_math_subtract(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = a - b;
+}
+
+void vector_math_multiply(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = a * b;
+}
+
+void vector_math_divide(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = safe_divide(a, b);
+}
+
+void vector_math_cross(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = cross(a, b);
+}
+
+void vector_math_project(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  float lenSquared = dot(b, b);
+  outVector = (lenSquared != 0.0) ? (dot(a, b) / lenSquared) * b : vec3(0.0);
+}
+
+void vector_math_reflect(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = reflect(a, normalize(b));
+}
+
+void vector_math_dot(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outValue = dot(a, b);
+}
+
+void vector_math_distance(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outValue = distance(a, b);
+}
+
+void vector_math_length(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outValue = length(a);
+}
+
+void vector_math_scale(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = a * scale;
+}
+
+void vector_math_normalize(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = normalize(a);
+}
+
+void vector_math_snap(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = floor(safe_divide(a, b)) * b;
+}
+
+void vector_math_floor(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = floor(a);
+}
+
+void vector_math_ceil(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = ceil(a);
+}
+
+void vector_math_modulo(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = c_mod(a, b);
+}
+
+void vector_math_fraction(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = fract(a);
+}
+
+void vector_math_absolute(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = abs(a);
+}
+
+void vector_math_minimum(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = min(a, b);
+}
+
+void vector_math_maximum(vec3 a, vec3 b, float scale, out vec3 outVector, out float outValue)
+{
+  outVector = max(a, b);
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_velvet.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_velvet.glsl
new file mode 100644
index 00000000000..9646ffff8ca
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_velvet.glsl
@@ -0,0 +1,9 @@
+#ifndef VOLUMETRICS
+void node_bsdf_velvet(vec4 color, float sigma, vec3 N, out Closure result)
+{
+  node_bsdf_diffuse(color, 0.0, N, result);
+}
+#else
+/* Stub velvet because it is not compatible with volumetrics. */
+#  define node_bsdf_velvet
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_volume_absorption.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_volume_absorption.glsl
new file mode 100644
index 00000000000..e6c0880cd07
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_volume_absorption.glsl
@@ -0,0 +1,8 @@
+void node_volume_absorption(vec4 color, float density, out Closure result)
+{
+#ifdef VOLUMETRICS
+  result = Closure((1.0 - color.rgb) * density, vec3(0.0), vec3(0.0), 0.0);
+#else
+  result = CLOSURE_DEFAULT;
+#endif
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_volume_info.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_volume_info.glsl
new file mode 100644
index 00000000000..501aeb6f34e
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_volume_info.glsl
@@ -0,0 +1,88 @@
+void node_attribute_volume_density(sampler3D tex, out vec4 outcol, out vec3 outvec, out float outf)
+{
+#if defined(MESH_SHADER) && defined(VOLUMETRICS)
+  vec3 cos = volumeObjectLocalCoord;
+#else
+  vec3 cos = vec3(0.0);
+#endif
+  outvec = texture(tex, cos).aaa;
+  outcol = vec4(outvec, 1.0);
+  outf = avg(outvec);
+}
+
+uniform vec3 volumeColor = vec3(1.0);
+
+void node_attribute_volume_color(sampler3D tex, out vec4 outcol, out vec3 outvec, out float outf)
+{
+#if defined(MESH_SHADER) && defined(VOLUMETRICS)
+  vec3 cos = volumeObjectLocalCoord;
+#else
+  vec3 cos = vec3(0.0);
+#endif
+
+  vec4 value = texture(tex, cos).rgba;
+  /* Density is premultiplied for interpolation, divide it out here. */
+  if (value.a > 1e-8) {
+    value.rgb /= value.a;
+  }
+
+  outvec = value.rgb * volumeColor;
+  outcol = vec4(outvec, 1.0);
+  outf = avg(outvec);
+}
+
+void node_attribute_volume_flame(sampler3D tex, out vec4 outcol, out vec3 outvec, out float outf)
+{
+#if defined(MESH_SHADER) && defined(VOLUMETRICS)
+  vec3 cos = volumeObjectLocalCoord;
+#else
+  vec3 cos = vec3(0.0);
+#endif
+  outf = texture(tex, cos).r;
+  outvec = vec3(outf, outf, outf);
+  outcol = vec4(outf, outf, outf, 1.0);
+}
+
+void node_attribute_volume_temperature(
+    sampler3D tex, vec2 temperature, out vec4 outcol, out vec3 outvec, out float outf)
+{
+#if defined(MESH_SHADER) && defined(VOLUMETRICS)
+  vec3 cos = volumeObjectLocalCoord;
+#else
+  vec3 cos = vec3(0.0);
+#endif
+  float flame = texture(tex, cos).r;
+
+  outf = (flame > 0.01) ? temperature.x + flame * (temperature.y - temperature.x) : 0.0;
+  outvec = vec3(outf, outf, outf);
+  outcol = vec4(outf, outf, outf, 1.0);
+}
+
+void node_volume_info(sampler3D densitySampler,
+                      sampler3D flameSampler,
+                      vec2 temperature,
+                      out vec4 outColor,
+                      out float outDensity,
+                      out float outFlame,
+                      out float outTemprature)
+{
+#if defined(MESH_SHADER) && defined(VOLUMETRICS)
+  vec3 p = volumeObjectLocalCoord;
+#else
+  vec3 p = vec3(0.0);
+#endif
+
+  vec4 density = texture(densitySampler, p);
+  outDensity = density.a;
+
+  /* Density is premultiplied for interpolation, divide it out here. */
+  if (density.a > 1e-8) {
+    density.rgb /= density.a;
+  }
+  outColor = vec4(density.rgb * volumeColor, 1.0);
+
+  float flame = texture(flameSampler, p).r;
+  outFlame = flame;
+
+  outTemprature = (flame > 0.01) ? temperature.x + flame * (temperature.y - temperature.x) : 0.0;
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_volume_principled.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_volume_principled.glsl
new file mode 100644
index 00000000000..884d5415c51
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_volume_principled.glsl
@@ -0,0 +1,67 @@
+void node_volume_principled(vec4 color,
+                            float density,
+                            float anisotropy,
+                            vec4 absorption_color,
+                            float emission_strength,
+                            vec4 emission_color,
+                            float blackbody_intensity,
+                            vec4 blackbody_tint,
+                            float temperature,
+                            float density_attribute,
+                            vec4 color_attribute,
+                            float temperature_attribute,
+                            sampler1DArray spectrummap,
+                            float layer,
+                            out Closure result)
+{
+#ifdef VOLUMETRICS
+  vec3 absorption_coeff = vec3(0.0);
+  vec3 scatter_coeff = vec3(0.0);
+  vec3 emission_coeff = vec3(0.0);
+
+  /* Compute density. */
+  density = max(density, 0.0);
+
+  if (density > 1e-5) {
+    density = max(density * density_attribute, 0.0);
+  }
+
+  if (density > 1e-5) {
+    /* Compute scattering and absorption coefficients. */
+    vec3 scatter_color = color.rgb * color_attribute.rgb;
+
+    scatter_coeff = scatter_color * density;
+    absorption_color.rgb = sqrt(max(absorption_color.rgb, 0.0));
+    absorption_coeff = max(1.0 - scatter_color, 0.0) * max(1.0 - absorption_color.rgb, 0.0) *
+                       density;
+  }
+
+  /* Compute emission. */
+  emission_strength = max(emission_strength, 0.0);
+
+  if (emission_strength > 1e-5) {
+    emission_coeff += emission_strength * emission_color.rgb;
+  }
+
+  if (blackbody_intensity > 1e-3) {
+    /* Add temperature from attribute. */
+    float T = max(temperature * max(temperature_attribute, 0.0), 0.0);
+
+    /* Stefan-Boltzman law. */
+    float T2 = T * T;
+    float T4 = T2 * T2;
+    float sigma = 5.670373e-8 * 1e-6 / M_PI;
+    float intensity = sigma * mix(1.0, T4, blackbody_intensity);
+
+    if (intensity > 1e-5) {
+      vec4 bb;
+      node_blackbody(T, spectrummap, layer, bb);
+      emission_coeff += bb.rgb * blackbody_tint.rgb * intensity;
+    }
+  }
+
+  result = Closure(absorption_coeff, scatter_coeff, emission_coeff, anisotropy);
+#else
+  result = CLOSURE_DEFAULT;
+#endif
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_volume_scatter.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_volume_scatter.glsl
new file mode 100644
index 00000000000..02c54658be5
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_volume_scatter.glsl
@@ -0,0 +1,8 @@
+void node_volume_scatter(vec4 color, float density, float anisotropy, out Closure result)
+{
+#ifdef VOLUMETRICS
+  result = Closure(vec3(0.0), color.rgb * density, vec3(0.0), anisotropy);
+#else
+  result = CLOSURE_DEFAULT;
+#endif
+}
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_wireframe.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_wireframe.glsl
new file mode 100644
index 00000000000..2fcf1b8d914
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_wireframe.glsl
@@ -0,0 +1,31 @@
+#ifndef VOLUMETRICS
+void node_wireframe(float size, vec2 barycentric, vec3 barycentric_dist, out float fac)
+{
+  vec3 barys = barycentric.xyy;
+  barys.z = 1.0 - barycentric.x - barycentric.y;
+
+  size *= 0.5;
+  vec3 s = step(-size, -barys * barycentric_dist);
+
+  fac = max(s.x, max(s.y, s.z));
+}
+
+void node_wireframe_screenspace(float size, vec2 barycentric, out float fac)
+{
+  vec3 barys = barycentric.xyy;
+  barys.z = 1.0 - barycentric.x - barycentric.y;
+
+  size *= (1.0 / 3.0);
+  vec3 dx = dFdx(barys);
+  vec3 dy = dFdy(barys);
+  vec3 deltas = sqrt(dx * dx + dy * dy);
+
+  vec3 s = step(-deltas * size, -barys);
+
+  fac = max(s.x, max(s.y, s.z));
+}
+#else
+/* Stub wireframe because it is not compatible with volumetrics. */
+#  define node_wireframe
+#  define node_wireframe_screenspace
+#endif
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_world_normals.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_world_normals.glsl
new file mode 100644
index 00000000000..f9691beee6f
--- /dev/null
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_world_normals.glsl
@@ -0,0 +1,25 @@
+/* TODO : clean this ifdef mess */
+void world_normals_get(out vec3 N)
+{
+#ifndef VOLUMETRICS
+#  ifdef HAIR_SHADER
+  vec3 B = normalize(cross(worldNormal, hairTangent));
+  float cos_theta;
+  if (hairThicknessRes == 1) {
+    vec4 rand = texelFetch(utilTex, ivec3(ivec2(gl_FragCoord.xy) % LUT_SIZE, 2.0), 0);
+    /* Random cosine normal distribution on the hair surface. */
+    cos_theta = rand.x * 2.0 - 1.0;
+  }
+  else {
+    /* Shade as a cylinder. */
+    cos_theta = hairThickTime / hairThickness;
+  }
+  float sin_theta = sqrt(max(0.0, 1.0 - cos_theta * cos_theta));
+  N = normalize(worldNormal * sin_theta + B * cos_theta);
+#  else
+  N = gl_FrontFacing ? worldNormal : -worldNormal;
+#  endif
+#else
+  generated_from_orco(vec3(0.0), N);
+#endif
+}
diff --git a/source/blender/ikplugin/intern/itasc_plugin.cpp b/source/blender/ikplugin/intern/itasc_plugin.cpp
index 883919d76ec..2ceedca59f7 100644
--- a/source/blender/ikplugin/intern/itasc_plugin.cpp
+++ b/source/blender/ikplugin/intern/itasc_plugin.cpp
@@ -879,7 +879,7 @@ static bool joint_callback(const iTaSC::Timestamp &timestamp,
   bPoseChannel *chan = ikchan->pchan;
   int dof;
 
-  // a channel can be splitted into multiple joints, so we get called multiple
+  // a channel can be split into multiple joints, so we get called multiple
   // times for one channel (this callback is only for 1 joint in the armature)
   // the IK_JointTarget structure is shared between multiple joint constraint
   // and the target joint values is computed only once, remember this in jointValid
diff --git a/source/blender/imbuf/IMB_imbuf.h b/source/blender/imbuf/IMB_imbuf.h
index 791e939db7f..7e7e489a6c3 100644
--- a/source/blender/imbuf/IMB_imbuf.h
+++ b/source/blender/imbuf/IMB_imbuf.h
@@ -46,8 +46,7 @@
  *
  * IMB needs:
  * - \ref DNA module
- *     The listbase types are used for handling the memory
- *     management.
+ *     The #ListBase types are used for handling the memory management.
  * - \ref blenlib module
  *     blenlib handles guarded memory management in blender-style.
  *     BLI_winstuff.h makes a few windows specific behaviors
@@ -146,7 +145,8 @@ bool IMB_initImBuf(
 struct ImBuf *IMB_allocFromBuffer(const unsigned int *rect,
                                   const float *rectf,
                                   unsigned int w,
-                                  unsigned int h);
+                                  unsigned int h,
+                                  unsigned int channels);
 
 /**
  *
diff --git a/source/blender/imbuf/intern/allocimbuf.c b/source/blender/imbuf/intern/allocimbuf.c
index 7d2f51af65e..75db3fd3c73 100644
--- a/source/blender/imbuf/intern/allocimbuf.c
+++ b/source/blender/imbuf/intern/allocimbuf.c
@@ -428,7 +428,8 @@ bool imb_addrectImBuf(ImBuf *ibuf)
 struct ImBuf *IMB_allocFromBuffer(const unsigned int *rect,
                                   const float *rectf,
                                   unsigned int w,
-                                  unsigned int h)
+                                  unsigned int h,
+                                  unsigned int channels)
 {
   ImBuf *ibuf = NULL;
 
@@ -438,6 +439,7 @@ struct ImBuf *IMB_allocFromBuffer(const unsigned int *rect,
 
   ibuf = IMB_allocImBuf(w, h, 32, 0);
 
+  ibuf->channels = channels;
   if (rectf) {
     ibuf->rect_float = MEM_dupallocN(rectf);
     ibuf->flags |= IB_rectfloat;
diff --git a/source/blender/imbuf/intern/scaling.c b/source/blender/imbuf/intern/scaling.c
index 28557277d72..46ddee25b0f 100644
--- a/source/blender/imbuf/intern/scaling.c
+++ b/source/blender/imbuf/intern/scaling.c
@@ -1629,8 +1629,8 @@ bool IMB_scaleImBuf(struct ImBuf *ibuf, unsigned int newx, unsigned int newy)
     return false;
   }
 
-  /* scaleup / scaledown functions below change ibuf->x and ibuf->y
-   * so we first scale the Z-buffer (if any) */
+  /* Scale-up / scale-down functions below change ibuf->x and ibuf->y
+   * so we first scale the Z-buffer (if any). */
   scalefast_Z_ImBuf(ibuf, newx, newy);
 
   /* try to scale common cases in a fast way */
diff --git a/source/blender/makesdna/DNA_ID.h b/source/blender/makesdna/DNA_ID.h
index ef9069acb78..ed7416e06c9 100644
--- a/source/blender/makesdna/DNA_ID.h
+++ b/source/blender/makesdna/DNA_ID.h
@@ -467,7 +467,11 @@ typedef enum ID_Type {
 
 /* id->flag (persitent). */
 enum {
+  /* Don't delete the datablock even if unused. */
   LIB_FAKEUSER = 1 << 9,
+  /* The datablock structure is a sub-object of a different one.
+   * Direct persistent references are not allowed. */
+  LIB_PRIVATE_DATA = 1 << 10,
 };
 
 /**
diff --git a/source/blender/makesdna/DNA_action_types.h b/source/blender/makesdna/DNA_action_types.h
index a4cd31a1357..b7834e2c7e0 100644
--- a/source/blender/makesdna/DNA_action_types.h
+++ b/source/blender/makesdna/DNA_action_types.h
@@ -847,8 +847,7 @@ typedef enum eSAction_Flag {
   SACTION_POSEMARKERS_SHOW = (1 << 6),
   /* don't draw action channels using group colors (where applicable) */
   SACTION_NODRAWGCOLORS = (1 << 7),
-  /* don't draw current frame number beside frame indicator */
-  SACTION_NODRAWCFRANUM = (1 << 8),
+  /* SACTION_NODRAWCFRANUM = (1 << 8), DEPRECATED */
   /* don't perform realtime updates */
   SACTION_NOREALTIMEUPDATES = (1 << 10),
   /* move markers as well as keyframes */
diff --git a/source/blender/makesdna/DNA_anim_types.h b/source/blender/makesdna/DNA_anim_types.h
index 1718aabc51d..70add4c156f 100644
--- a/source/blender/makesdna/DNA_anim_types.h
+++ b/source/blender/makesdna/DNA_anim_types.h
@@ -319,6 +319,10 @@ typedef struct DriverTarget {
   /** Transform channel index (for DVAR_TYPE_TRANSFORM_CHAN.)*/
   short transChan;
 
+  /** Rotation channel calculation type. */
+  char rotation_mode;
+  char _pad[7];
+
   /**
    * Flags for the validity of the target
    * (NOTE: these get reset every time the types change).
@@ -358,10 +362,36 @@ typedef enum eDriverTarget_TransformChannels {
   DTAR_TRANSCHAN_SCALEY,
   DTAR_TRANSCHAN_SCALEZ,
   DTAR_TRANSCHAN_SCALE_AVG,
+  DTAR_TRANSCHAN_ROTW,
 
   MAX_DTAR_TRANSCHAN_TYPES,
 } eDriverTarget_TransformChannels;
 
+/* Rotation channel mode for Driver Targets */
+typedef enum eDriverTarget_RotationMode {
+  /** Automatic euler mode. */
+  DTAR_ROTMODE_AUTO = 0,
+
+  /** Explicit euler rotation modes - must sync with BLI_math_rotation.h defines. */
+  DTAR_ROTMODE_EULER_XYZ = 1,
+  DTAR_ROTMODE_EULER_XZY,
+  DTAR_ROTMODE_EULER_YXZ,
+  DTAR_ROTMODE_EULER_YZX,
+  DTAR_ROTMODE_EULER_ZXY,
+  DTAR_ROTMODE_EULER_ZYX,
+
+  DTAR_ROTMODE_QUATERNION,
+
+  /** Implements the very common Damped Track + child trick to decompose
+   *  rotation into bending followed by twist around the remaining axis. */
+  DTAR_ROTMODE_SWING_TWIST_X,
+  DTAR_ROTMODE_SWING_TWIST_Y,
+  DTAR_ROTMODE_SWING_TWIST_Z,
+
+  DTAR_ROTMODE_EULER_MIN = DTAR_ROTMODE_EULER_XYZ,
+  DTAR_ROTMODE_EULER_MAX = DTAR_ROTMODE_EULER_ZYX,
+} eDriverTarget_RotationMode;
+
 /* --- */
 
 /* maximum number of driver targets per variable */
diff --git a/source/blender/makesdna/DNA_brush_types.h b/source/blender/makesdna/DNA_brush_types.h
index 93ce3d9769b..bae9c8f40ea 100644
--- a/source/blender/makesdna/DNA_brush_types.h
+++ b/source/blender/makesdna/DNA_brush_types.h
@@ -295,6 +295,8 @@ typedef struct Brush {
 
   float crease_pinch_factor;
 
+  float normal_radius_factor;
+
   float plane_trim;
   /** Affectable height of brush (layer height for layer tool, i.e.). */
   float height;
@@ -302,7 +304,6 @@ typedef struct Brush {
   float texture_sample_bias;
 
   int curve_preset;
-  char _pad1[4];
 
   /* overlay */
   int texture_overlay_alpha;
@@ -395,7 +396,7 @@ typedef enum eBrushFlags {
   BRUSH_LOCK_ALPHA = (1 << 14),
   BRUSH_ORIGINAL_NORMAL = (1 << 15),
   BRUSH_OFFSET_PRESSURE = (1 << 16),
-  BRUSH_FLAG_UNUSED_17 = (1 << 17), /* cleared */
+  BRUSH_SCENE_SPACING = (1 << 17),
   BRUSH_SPACE_ATTEN = (1 << 18),
   BRUSH_ADAPTIVE_SPACE = (1 << 19),
   BRUSH_LOCK_SIZE = (1 << 20),
diff --git a/source/blender/makesdna/DNA_constraint_types.h b/source/blender/makesdna/DNA_constraint_types.h
index eda52fdf14a..42d58cb34d0 100644
--- a/source/blender/makesdna/DNA_constraint_types.h
+++ b/source/blender/makesdna/DNA_constraint_types.h
@@ -271,7 +271,8 @@ typedef struct bTrackToConstraint {
 typedef struct bRotateLikeConstraint {
   struct Object *tar;
   int flag;
-  int reserved1;
+  char euler_order;
+  char _pad[3];
   /** MAX_ID_NAME-2. */
   char subtarget[64];
 } bRotateLikeConstraint;
@@ -441,6 +442,13 @@ typedef struct bTransformConstraint {
   /** Extrapolate motion? if 0, confine to ranges. */
   char expo;
 
+  /** Input rotation type - uses the same values as driver targets. */
+  char from_rotation_mode;
+  /** Output euler order override. */
+  char to_euler_order;
+
+  char _pad[6];
+
   /** From_min/max defines range of target transform. */
   float from_min[3];
   /** To map on to to_min/max range. */
@@ -715,6 +723,20 @@ typedef enum eConstraintChannel_Flags {
   CONSTRAINT_CHANNEL_PROTECTED = (1 << 1),
 } eConstraintChannel_Flags;
 
+/* Common enum for constraints that support override. */
+typedef enum eConstraint_EulerOrder {
+  /** Automatic euler mode. */
+  CONSTRAINT_EULER_AUTO = 0,
+
+  /** Explicit euler rotation modes - must sync with BLI_math_rotation.h defines. */
+  CONSTRAINT_EULER_XYZ = 1,
+  CONSTRAINT_EULER_XZY,
+  CONSTRAINT_EULER_YXZ,
+  CONSTRAINT_EULER_YZX,
+  CONSTRAINT_EULER_ZXY,
+  CONSTRAINT_EULER_ZYX,
+} eConstraint_EulerOrder;
+
 /* -------------------------------------- */
 
 /* bRotateLikeConstraint.flag */
@@ -748,6 +770,7 @@ typedef enum eCopyScale_Flags {
   SIZELIKE_Z = (1 << 2),
   SIZELIKE_OFFSET = (1 << 3),
   SIZELIKE_MULTIPLY = (1 << 4),
+  SIZELIKE_UNIFORM = (1 << 5),
 } eCopyScale_Flags;
 
 /* bTransformConstraint.to/from */
diff --git a/source/blender/makesdna/DNA_customdata_types.h b/source/blender/makesdna/DNA_customdata_types.h
index 5a169fe1d16..415eb03e26d 100644
--- a/source/blender/makesdna/DNA_customdata_types.h
+++ b/source/blender/makesdna/DNA_customdata_types.h
@@ -197,6 +197,9 @@ typedef enum CustomDataType {
 /** Data types that may be defined for all mesh elements types. */
 #define CD_MASK_GENERIC_DATA (CD_MASK_PROP_FLT | CD_MASK_PROP_INT | CD_MASK_PROP_STR)
 
+/** Multires loop data. */
+#define CD_MASK_MULTIRES_GRIDS (CD_MASK_MDISPS | CD_GRID_PAINT_MASK)
+
 typedef struct CustomData_MeshMasks {
   uint64_t vmask;
   uint64_t emask;
diff --git a/source/blender/makesdna/DNA_documentation.h b/source/blender/makesdna/DNA_documentation.h
index be11b3d9040..0251625292c 100644
--- a/source/blender/makesdna/DNA_documentation.h
+++ b/source/blender/makesdna/DNA_documentation.h
@@ -27,7 +27,7 @@
  * blender file. There is an executable that scans all files, looking
  * for struct-s to serialize (hence sdna: Struct \ref DNA). From this
  * information, it builds a file with numbers that encode the format,
- * the names of variables, and the plce to look for them.
+ * the names of variables, and the place to look for them.
  *
  * \section dnaissues Known issues with DNA
  *
diff --git a/source/blender/makesdna/DNA_gpencil_modifier_types.h b/source/blender/makesdna/DNA_gpencil_modifier_types.h
index 663e4dd8e44..f934d70fcdf 100644
--- a/source/blender/makesdna/DNA_gpencil_modifier_types.h
+++ b/source/blender/makesdna/DNA_gpencil_modifier_types.h
@@ -28,8 +28,6 @@
  * (ONLY ADD NEW ITEMS AT THE END)
  */
 
-struct RNG;
-
 typedef enum GpencilModifierType {
   eGpencilModifierType_None = 0,
   eGpencilModifierType_Noise = 1,
@@ -82,6 +80,8 @@ typedef struct NoiseGpencilModifierData {
   GpencilModifierData modifier;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Optional vertexgroup name, MAX_VGROUP_NAME. */
   char vgname[64];
   /** Custom index for passes. */
@@ -110,12 +110,15 @@ typedef enum eNoiseGpencil_Flag {
   GP_NOISE_INVERT_VGROUP = (1 << 8),
   GP_NOISE_MOD_UV = (1 << 9),
   GP_NOISE_INVERT_LAYERPASS = (1 << 10),
+  GP_NOISE_INVERT_MATERIAL = (1 << 11),
 } eNoiseGpencil_Flag;
 
 typedef struct SubdivGpencilModifierData {
   GpencilModifierData modifier;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Custom index for passes. */
   int pass_index;
   /** Flags. */
@@ -131,12 +134,15 @@ typedef enum eSubdivGpencil_Flag {
   GP_SUBDIV_INVERT_LAYER = (1 << 1),
   GP_SUBDIV_INVERT_PASS = (1 << 2),
   GP_SUBDIV_INVERT_LAYERPASS = (1 << 3),
+  GP_SUBDIV_INVERT_MATERIAL = (1 << 4),
 } eSubdivGpencil_Flag;
 
 typedef struct ThickGpencilModifierData {
   GpencilModifierData modifier;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Optional vertexgroup name, MAX_VGROUP_NAME. */
   char vgname[64];
   /** Custom index for passes. */
@@ -157,6 +163,7 @@ typedef enum eThickGpencil_Flag {
   GP_THICK_CUSTOM_CURVE = (1 << 3),
   GP_THICK_NORMALIZE = (1 << 4),
   GP_THICK_INVERT_LAYERPASS = (1 << 5),
+  GP_THICK_INVERT_MATERIAL = (1 << 6),
 } eThickGpencil_Flag;
 
 typedef struct TimeGpencilModifierData {
@@ -204,6 +211,8 @@ typedef struct TintGpencilModifierData {
   GpencilModifierData modifier;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Custom index for passes. */
   int pass_index;
   /** Flags. */
@@ -225,12 +234,15 @@ typedef enum eTintGpencil_Flag {
   GP_TINT_INVERT_LAYER = (1 << 1),
   GP_TINT_INVERT_PASS = (1 << 2),
   GP_TINT_INVERT_LAYERPASS = (1 << 3),
+  GP_TINT_INVERT_MATERIAL = (1 << 4),
 } eTintGpencil_Flag;
 
 typedef struct ColorGpencilModifierData {
   GpencilModifierData modifier;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Custom index for passes. */
   int pass_index;
   /** Flags. */
@@ -250,12 +262,15 @@ typedef enum eColorGpencil_Flag {
   GP_COLOR_INVERT_LAYER = (1 << 1),
   GP_COLOR_INVERT_PASS = (1 << 2),
   GP_COLOR_INVERT_LAYERPASS = (1 << 3),
+  GP_COLOR_INVERT_MATERIAL = (1 << 4),
 } eColorGpencil_Flag;
 
 typedef struct OpacityGpencilModifierData {
   GpencilModifierData modifier;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Optional vertexgroup name, MAX_VGROUP_NAME. */
   char vgname[64];
   /** Custom index for passes. */
@@ -280,6 +295,7 @@ typedef enum eOpacityGpencil_Flag {
   GP_OPACITY_INVERT_VGROUP = (1 << 2),
   GP_OPACITY_CREATE_COLORS = (1 << 3),
   GP_OPACITY_INVERT_LAYERPASS = (1 << 4),
+  GP_OPACITY_INVERT_MATERIAL = (1 << 5),
 } eOpacityGpencil_Flag;
 
 typedef struct ArrayGpencilModifierData {
@@ -309,6 +325,8 @@ typedef struct ArrayGpencilModifierData {
   int pass_index;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Material replace (0 keep default). */
   int mat_rpl;
   /** Custom index for passes. */
@@ -322,6 +340,7 @@ typedef enum eArrayGpencil_Flag {
   GP_ARRAY_INVERT_PASS = (1 << 3),
   GP_ARRAY_KEEP_ONTOP = (1 << 4),
   GP_ARRAY_INVERT_LAYERPASS = (1 << 5),
+  GP_ARRAY_INVERT_MATERIAL = (1 << 6),
 } eArrayGpencil_Flag;
 
 typedef struct BuildGpencilModifierData {
@@ -331,6 +350,9 @@ typedef struct BuildGpencilModifierData {
   char layername[64];
   int pass_index;
 
+  /** Material name. */
+  char materialname[64];
+
   /** Custom index for passes. */
   int layer_pass;
 
@@ -401,6 +423,8 @@ typedef struct LatticeGpencilModifierData {
   struct Object *object;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Optional vertexgroup name, MAX_VGROUP_NAME. */
   char vgname[64];
   /** Custom index for passes. */
@@ -419,6 +443,7 @@ typedef enum eLatticeGpencil_Flag {
   GP_LATTICE_INVERT_PASS = (1 << 1),
   GP_LATTICE_INVERT_VGROUP = (1 << 2),
   GP_LATTICE_INVERT_LAYERPASS = (1 << 3),
+  GP_LATTICE_INVERT_MATERIAL = (1 << 4),
 } eLatticeGpencil_Flag;
 
 typedef struct MirrorGpencilModifierData {
@@ -426,6 +451,8 @@ typedef struct MirrorGpencilModifierData {
   struct Object *object;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Custom index for passes. */
   int pass_index;
   /** Flags. */
@@ -443,6 +470,7 @@ typedef enum eMirrorGpencil_Flag {
   GP_MIRROR_AXIS_Y = (1 << 4),
   GP_MIRROR_AXIS_Z = (1 << 5),
   GP_MIRROR_INVERT_LAYERPASS = (1 << 6),
+  GP_MIRROR_INVERT_MATERIAL = (1 << 7),
 } eMirrorGpencil_Flag;
 
 typedef struct HookGpencilModifierData {
@@ -453,6 +481,8 @@ typedef struct HookGpencilModifierData {
   char subtarget[64];
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Optional vertexgroup name, MAX_VGROUP_NAME. */
   char vgname[64];
   /** Custom index for passes. */
@@ -481,6 +511,7 @@ typedef enum eHookGpencil_Flag {
   GP_HOOK_INVERT_VGROUP = (1 << 2),
   GP_HOOK_UNIFORM_SPACE = (1 << 3),
   GP_HOOK_INVERT_LAYERPASS = (1 << 4),
+  GP_HOOK_INVERT_MATERIAL = (1 << 5),
 } eHookGpencil_Flag;
 
 typedef enum eHookGpencil_Falloff {
@@ -499,6 +530,8 @@ typedef struct SimplifyGpencilModifierData {
   GpencilModifierData modifier;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Custom index for passes. */
   int pass_index;
   /** Flags. */
@@ -522,6 +555,7 @@ typedef enum eSimplifyGpencil_Flag {
   GP_SIMPLIFY_INVERT_LAYER = (1 << 0),
   GP_SIMPLIFY_INVERT_PASS = (1 << 1),
   GP_SIMPLIFY_INVERT_LAYERPASS = (1 << 2),
+  GP_SIMPLIFY_INVERT_MATERIAL = (1 << 3),
 } eSimplifyGpencil_Flag;
 
 typedef enum eSimplifyGpencil_Mode {
@@ -539,6 +573,8 @@ typedef struct OffsetGpencilModifierData {
   GpencilModifierData modifier;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Optional vertexgroup name, MAX_VGROUP_NAME. */
   char vgname[64];
   /** Custom index for passes. */
@@ -557,12 +593,15 @@ typedef enum eOffsetGpencil_Flag {
   GP_OFFSET_INVERT_PASS = (1 << 1),
   GP_OFFSET_INVERT_VGROUP = (1 << 2),
   GP_OFFSET_INVERT_LAYERPASS = (1 << 3),
+  GP_OFFSET_INVERT_MATERIAL = (1 << 4),
 } eOffsetGpencil_Flag;
 
 typedef struct SmoothGpencilModifierData {
   GpencilModifierData modifier;
   /** Layer name. */
   char layername[64];
+  /** Material name. */
+  char materialname[64];
   /** Optional vertexgroup name, MAX_VGROUP_NAME. */
   char vgname[64];
   /** Custom index for passes. */
@@ -587,6 +626,7 @@ typedef enum eSmoothGpencil_Flag {
   GP_SMOOTH_INVERT_VGROUP = (1 << 5),
   GP_SMOOTH_MOD_UV = (1 << 6),
   GP_SMOOTH_INVERT_LAYERPASS = (1 << 7),
+  GP_SMOOTH_INVERT_MATERIAL = (1 << 4),
 } eSmoothGpencil_Flag;
 
 typedef struct ArmatureGpencilModifierData {
diff --git a/source/blender/makesdna/DNA_gpencil_types.h b/source/blender/makesdna/DNA_gpencil_types.h
index 0de7d8bdd16..58962dd9469 100644
--- a/source/blender/makesdna/DNA_gpencil_types.h
+++ b/source/blender/makesdna/DNA_gpencil_types.h
@@ -30,8 +30,6 @@
 
 struct ARegion;
 struct AnimData;
-struct CurveMapping;
-struct GHash;
 struct MDeformVert;
 
 #define GP_DEFAULT_PIX_FACTOR 1.0f
diff --git a/source/blender/makesdna/DNA_light_types.h b/source/blender/makesdna/DNA_light_types.h
index 5e881053910..82cab5d1594 100644
--- a/source/blender/makesdna/DNA_light_types.h
+++ b/source/blender/makesdna/DNA_light_types.h
@@ -34,7 +34,6 @@
 struct AnimData;
 struct CurveMapping;
 struct Ipo;
-struct MTex;
 struct bNodeTree;
 
 typedef struct Light {
diff --git a/source/blender/makesdna/DNA_mesh_types.h b/source/blender/makesdna/DNA_mesh_types.h
index caa7c98335a..cb6991ce67a 100644
--- a/source/blender/makesdna/DNA_mesh_types.h
+++ b/source/blender/makesdna/DNA_mesh_types.h
@@ -40,7 +40,6 @@ struct MLoopCol;
 struct MLoopTri;
 struct MLoopUV;
 struct MPoly;
-struct MTexPoly;
 struct MVert;
 struct Material;
 struct Mesh;
@@ -270,10 +269,4 @@ enum {
 
 #define MESH_MAX_VERTS 2000000000L
 
-/* this is so we can save bmesh files that load in trunk, ignoring NGons
- * will eventually be removed */
-
-/* enable this so meshes get tessfaces calculated by default */
-/* #define USE_TESSFACE_DEFAULT */
-
 #endif
diff --git a/source/blender/makesdna/DNA_meshdata_types.h b/source/blender/makesdna/DNA_meshdata_types.h
index f924368de1d..50402bbdc87 100644
--- a/source/blender/makesdna/DNA_meshdata_types.h
+++ b/source/blender/makesdna/DNA_meshdata_types.h
@@ -27,8 +27,6 @@
 #include "DNA_customdata_types.h"
 #include "DNA_listBase.h"
 
-struct Image;
-
 /*tessellation face, see MLoop/MPoly for the real face data*/
 typedef struct MFace {
   unsigned int v1, v2, v3, v4;
@@ -188,10 +186,6 @@ typedef struct MVertTri {
   unsigned int tri[3];
 } MVertTri;
 
-// typedef struct MTexPoly {
-//  void *_pad;
-//} MTexPoly;
-
 typedef struct MLoopUV {
   float uv[2];
   int flag;
diff --git a/source/blender/makesdna/DNA_modifier_types.h b/source/blender/makesdna/DNA_modifier_types.h
index 37c0065b166..5d71d33733d 100644
--- a/source/blender/makesdna/DNA_modifier_types.h
+++ b/source/blender/makesdna/DNA_modifier_types.h
@@ -29,8 +29,6 @@
  */
 
 struct Mesh;
-struct Scene;
-struct Subdiv;
 
 typedef enum ModifierType {
   eModifierType_None = 0,
diff --git a/source/blender/makesdna/DNA_node_types.h b/source/blender/makesdna/DNA_node_types.h
index 3ad857ac7b7..af66add01f3 100644
--- a/source/blender/makesdna/DNA_node_types.h
+++ b/source/blender/makesdna/DNA_node_types.h
@@ -113,7 +113,7 @@ typedef struct bNodeSocket {
   short stack_index;
   /* XXX deprecated, kept for forward compatibility */
   short stack_type DNA_DEPRECATED;
-  char draw_shape;
+  char display_shape;
   char _pad[3];
 
   /** Cached data from execution. */
@@ -153,11 +153,14 @@ typedef enum eNodeSocketDatatype {
 } eNodeSocketDatatype;
 
 /* socket shape */
-typedef enum eNodeSocketDrawShape {
-  SOCK_DRAW_SHAPE_CIRCLE = 0,
-  SOCK_DRAW_SHAPE_SQUARE = 1,
-  SOCK_DRAW_SHAPE_DIAMOND = 2,
-} eNodeSocketDrawShape;
+typedef enum eNodeSocketDisplayShape {
+  SOCK_DISPLAY_SHAPE_CIRCLE = 0,
+  SOCK_DISPLAY_SHAPE_SQUARE = 1,
+  SOCK_DISPLAY_SHAPE_DIAMOND = 2,
+  SOCK_DISPLAY_SHAPE_CIRCLE_DOT = 3,
+  SOCK_DISPLAY_SHAPE_SQUARE_DOT = 4,
+  SOCK_DISPLAY_SHAPE_DIAMOND_DOT = 5,
+} eNodeSocketDisplayShape;
 
 /* socket side (input/output) */
 typedef enum eNodeSocketInOut {
@@ -1167,34 +1170,61 @@ typedef struct NodeDenoise {
 /* math node clamp */
 #define SHD_MATH_CLAMP 1
 
-/* Math node operation/ */
+/* Math node operations. */
 enum {
   NODE_MATH_ADD = 0,
-  NODE_MATH_SUB = 1,
-  NODE_MATH_MUL = 2,
+  NODE_MATH_SUBTRACT = 1,
+  NODE_MATH_MULTIPLY = 2,
   NODE_MATH_DIVIDE = 3,
-  NODE_MATH_SIN = 4,
-  NODE_MATH_COS = 5,
-  NODE_MATH_TAN = 6,
-  NODE_MATH_ASIN = 7,
-  NODE_MATH_ACOS = 8,
-  NODE_MATH_ATAN = 9,
-  NODE_MATH_POW = 10,
-  NODE_MATH_LOG = 11,
-  NODE_MATH_MIN = 12,
-  NODE_MATH_MAX = 13,
+  NODE_MATH_SINE = 4,
+  NODE_MATH_COSINE = 5,
+  NODE_MATH_TANGENT = 6,
+  NODE_MATH_ARCSINE = 7,
+  NODE_MATH_ARCCOSINE = 8,
+  NODE_MATH_ARCTANGENT = 9,
+  NODE_MATH_POWER = 10,
+  NODE_MATH_LOGARITHM = 11,
+  NODE_MATH_MINIMUM = 12,
+  NODE_MATH_MAXIMUM = 13,
   NODE_MATH_ROUND = 14,
-  NODE_MATH_LESS = 15,
-  NODE_MATH_GREATER = 16,
-  NODE_MATH_MOD = 17,
-  NODE_MATH_ABS = 18,
-  NODE_MATH_ATAN2 = 19,
+  NODE_MATH_LESS_THAN = 15,
+  NODE_MATH_GREATER_THAN = 16,
+  NODE_MATH_MODULO = 17,
+  NODE_MATH_ABSOLUTE = 18,
+  NODE_MATH_ARCTAN2 = 19,
   NODE_MATH_FLOOR = 20,
   NODE_MATH_CEIL = 21,
-  NODE_MATH_FRACT = 22,
+  NODE_MATH_FRACTION = 22,
   NODE_MATH_SQRT = 23,
 };
 
+/* Vector Math node operations. */
+enum {
+  NODE_VECTOR_MATH_ADD = 0,
+  NODE_VECTOR_MATH_SUBTRACT = 1,
+  NODE_VECTOR_MATH_MULTIPLY = 2,
+  NODE_VECTOR_MATH_DIVIDE = 3,
+
+  NODE_VECTOR_MATH_CROSS_PRODUCT = 4,
+  NODE_VECTOR_MATH_PROJECT = 5,
+  NODE_VECTOR_MATH_REFLECT = 6,
+  NODE_VECTOR_MATH_DOT_PRODUCT = 7,
+
+  NODE_VECTOR_MATH_DISTANCE = 8,
+  NODE_VECTOR_MATH_LENGTH = 9,
+  NODE_VECTOR_MATH_SCALE = 10,
+  NODE_VECTOR_MATH_NORMALIZE = 11,
+
+  NODE_VECTOR_MATH_SNAP = 12,
+  NODE_VECTOR_MATH_FLOOR = 13,
+  NODE_VECTOR_MATH_CEIL = 14,
+  NODE_VECTOR_MATH_MODULO = 15,
+  NODE_VECTOR_MATH_FRACTION = 16,
+  NODE_VECTOR_MATH_ABSOLUTE = 17,
+  NODE_VECTOR_MATH_MINIMUM = 18,
+  NODE_VECTOR_MATH_MAXIMUM = 19,
+};
+
 /* mix rgb node flags */
 #define SHD_MIXRGB_USE_ALPHA 1
 #define SHD_MIXRGB_CLAMP 2
diff --git a/source/blender/makesdna/DNA_object_fluidsim_types.h b/source/blender/makesdna/DNA_object_fluidsim_types.h
index 40502fc27c9..4e821ad5f07 100644
--- a/source/blender/makesdna/DNA_object_fluidsim_types.h
+++ b/source/blender/makesdna/DNA_object_fluidsim_types.h
@@ -32,7 +32,6 @@ extern "C" {
 #endif
 
 struct Ipo;
-struct Mesh;
 
 typedef struct FluidVertexVelocity {
   float vel[3];
diff --git a/source/blender/makesdna/DNA_object_types.h b/source/blender/makesdna/DNA_object_types.h
index beff286ab40..f307fc8fc2d 100644
--- a/source/blender/makesdna/DNA_object_types.h
+++ b/source/blender/makesdna/DNA_object_types.h
@@ -47,7 +47,6 @@ struct Material;
 struct Mesh;
 struct Object;
 struct PartDeflect;
-struct ParticleSystem;
 struct Path;
 struct RigidBodyOb;
 struct SculptSession;
@@ -163,12 +162,14 @@ typedef struct Object_Runtime {
   struct Mesh *mesh_eval;
   /**
    * Mesh structure created during object evaluation.
-   * It has deforemation only modifiers applied on it.
+   * It has deformation only modifiers applied on it.
    */
   struct Mesh *mesh_deform_eval;
 
-  /* This is a mesh representation of corresponding object.
-   * It created when Python calls `object.to_mesh()`. */
+  /**
+   * This is a mesh representation of corresponding object.
+   * It created when Python calls `object.to_mesh()`.
+   */
   struct Mesh *object_as_temp_mesh;
 
   /** Runtime evaluated curve-specific data, not stored in the file. */
@@ -176,11 +177,11 @@ typedef struct Object_Runtime {
 
   /** Runtime grease pencil drawing data */
   struct GpencilBatchCache *gpencil_cache;
-  /** Runtime grease pencil total layers used for derived data created by modifiers */
-  int tot_layers;
+  /** Runtime grease pencil total layers used for evaluated data created by modifiers */
+  int gpencil_tot_layers;
   char _pad4[4];
-  /** Runtime grease pencil derived data created by modifiers */
-  struct bGPDframe *derived_frames;
+  /** Runtime grease pencil evaluated data created by modifiers */
+  struct bGPDframe *gpencil_evaluated_frames;
 
   void *_pad2; /* Padding is here for win32s unconventional struct alignment rules. */
 } Object_Runtime;
@@ -515,7 +516,7 @@ enum {
   OB_DUPLIROT = 1 << 5,
   OB_TRANSFLAG_UNUSED_6 = 1 << 6, /* cleared */
   /* runtime, calculate derivedmesh for dupli before it's used */
-  OB_DUPLICALCDERIVED = 1 << 7,
+  OB_TRANSFLAG_UNUSED_7 = 1 << 7, /* dirty */
   OB_DUPLICOLLECTION = 1 << 8,
   OB_DUPLIFACES = 1 << 9,
   OB_DUPLIFACES_SCALE = 1 << 10,
@@ -619,7 +620,11 @@ enum {
 /* NOTE: this was used as a proper setting in past, so nullify before using */
 #define BA_TEMP_TAG (1 << 5)
 
-/* #define BA_FROMSET          (1 << 7) */ /*UNUSED*/
+/**
+ * Even if this is is tagged for transform, this flag means it's being locked in place.
+ * Use for #SCE_XFORM_SKIP_CHILDREN.
+ */
+#define BA_TRANSFORM_LOCKED_IN_PLACE (1 << 7)
 
 #define BA_TRANSFORM_CHILD (1 << 8)   /* child of a transformed object */
 #define BA_TRANSFORM_PARENT (1 << 13) /* parent of a transformed object */
@@ -698,6 +703,7 @@ enum {
   OB_EMPTY_IMAGE_HIDE_ORTHOGRAPHIC = 1 << 1,
   OB_EMPTY_IMAGE_HIDE_BACK = 1 << 2,
   OB_EMPTY_IMAGE_HIDE_FRONT = 1 << 3,
+  OB_EMPTY_IMAGE_HIDE_NON_AXIS_ALIGNED = 1 << 4,
 };
 
 /** #Object.empty_image_flag */
diff --git a/source/blender/makesdna/DNA_outliner_types.h b/source/blender/makesdna/DNA_outliner_types.h
index 1462d955f81..9776063f220 100644
--- a/source/blender/makesdna/DNA_outliner_types.h
+++ b/source/blender/makesdna/DNA_outliner_types.h
@@ -60,6 +60,10 @@ enum {
   TSE_DRAG_INTO = (1 << 6),
   TSE_DRAG_BEFORE = (1 << 7),
   TSE_DRAG_AFTER = (1 << 8),
+  /* Needed because outliner-only elements can be active */
+  TSE_ACTIVE = (1 << 9),
+  /* Needed because walk selection should not activate */
+  TSE_ACTIVE_WALK = (1 << 10),
   TSE_DRAG_ANY = (TSE_DRAG_INTO | TSE_DRAG_BEFORE | TSE_DRAG_AFTER),
 };
 
diff --git a/source/blender/makesdna/DNA_scene_types.h b/source/blender/makesdna/DNA_scene_types.h
index a6544a7f950..45d80029eac 100644
--- a/source/blender/makesdna/DNA_scene_types.h
+++ b/source/blender/makesdna/DNA_scene_types.h
@@ -58,8 +58,6 @@ struct MovieClip;
 struct Object;
 struct Scene;
 struct SceneCollection;
-struct SceneStats;
-struct Text;
 struct World;
 struct bGPdata;
 struct bNodeTree;
@@ -1125,7 +1123,7 @@ typedef struct GP_Sculpt_Settings {
 /* GP_Sculpt_Settings.flag */
 typedef enum eGP_Sculpt_SettingsFlag {
   /* only affect selected points */
-  GP_SCULPT_SETT_FLAG_SELECT_MASK = (1 << 0),
+  GP_SCULPT_SETT_FLAG_DEPRECATED = (1 << 0),
   /* apply brush to position */
   GP_SCULPT_SETT_FLAG_APPLY_POSITION = (1 << 1),
   /* apply brush to strength */
@@ -1142,6 +1140,16 @@ typedef enum eGP_Sculpt_SettingsFlag {
   GP_SCULPT_SETT_FLAG_PRIMITIVE_CURVE = (1 << 7),
 } eGP_Sculpt_SettingsFlag;
 
+/* GP_Sculpt_Settings.gpencil_selectmode_sculpt */
+typedef enum eGP_Sculpt_SelectMaskFlag {
+  /* only affect selected points */
+  GP_SCULPT_MASK_SELECTMODE_POINT = (1 << 0),
+  /* only affect selected strokes */
+  GP_SCULPT_MASK_SELECTMODE_STROKE = (1 << 1),
+  /* only affect selected segmenst */
+  GP_SCULPT_MASK_SELECTMODE_SEGMENT = (1 << 2),
+} eGP_Sculpt_SelectMaskFlag;
+
 /* Settings for GP Interpolation Operators */
 typedef struct GP_Interpolate_Settings {
   /** #eGP_Interpolate_SettingsFlag. */
@@ -1358,6 +1366,12 @@ typedef struct MeshStatVis {
 /* *************************************************************** */
 /* Tool Settings */
 
+/* CurvePaintSettings.surface_plane */
+enum {
+  AUTO_MERGE = 1 << 0,
+  AUTO_MERGE_AND_SPLIT = 1 << 1,
+};
+
 typedef struct ToolSettings {
   /** Vertex paint. */
   VPaint *vpaint;
@@ -1412,8 +1426,10 @@ typedef struct ToolSettings {
 
   /** Default stroke thickness for annotation strokes. */
   short annotate_thickness;
-  /** Stroke selection mode. */
-  short gpencil_selectmode;
+  /** Stroke selection mode for Edit. */
+  char gpencil_selectmode_edit;
+  /** Stroke selection mode for Sculpt. */
+  char gpencil_selectmode_sculpt;
 
   /* Grease Pencil Sculpt */
   struct GP_Sculpt_Settings gp_sculpt;
@@ -2072,6 +2088,8 @@ extern const char *RE_engine_id_CYCLES;
   (((workspace)->object_mode & OD_MODE_EDIT) ? OBACT(_view_layer) : NULL)
 #define OBEDIT_FROM_OBACT(ob) ((ob) ? (((ob)->mode & OB_MODE_EDIT) ? ob : NULL) : NULL)
 #define OBPOSE_FROM_OBACT(ob) ((ob) ? (((ob)->mode & OB_MODE_POSE) ? ob : NULL) : NULL)
+#define OBWEIGHTPAINT_FROM_OBACT(ob) \
+  ((ob) ? (((ob)->mode & OB_MODE_WEIGHT_PAINT) ? ob : NULL) : NULL)
 #define OBEDIT_FROM_VIEW_LAYER(view_layer) OBEDIT_FROM_OBACT(OBACT(view_layer))
 
 #define V3D_CAMERA_LOCAL(v3d) ((!(v3d)->scenelock && (v3d)->camera) ? (v3d)->camera : NULL)
@@ -2094,6 +2112,8 @@ extern const char *RE_engine_id_CYCLES;
 /* ToolSettings.transform_flag */
 enum {
   SCE_XFORM_AXIS_ALIGN = (1 << 0),
+  SCE_XFORM_DATA_ORIGIN = (1 << 1),
+  SCE_XFORM_SKIP_CHILDREN = (1 << 2),
 };
 
 /* ToolSettings.object_flag */
@@ -2121,11 +2141,15 @@ enum {
 #define SCE_SNAP_MODE_FACE (1 << 2)
 #define SCE_SNAP_MODE_VOLUME (1 << 3)
 #define SCE_SNAP_MODE_INCREMENT (1 << 4)
+#define SCE_SNAP_MODE_EDGE_MIDPOINT (1 << 5)
+#define SCE_SNAP_MODE_EDGE_PERPENDICULAR (1 << 6)
 
 /* ToolSettings.snap_node_mode */
-#define SCE_SNAP_MODE_GRID (1 << 5)
-#define SCE_SNAP_MODE_NODE_X (1 << 6)
-#define SCE_SNAP_MODE_NODE_Y (1 << 7)
+#define SCE_SNAP_MODE_NODE_X (1 << 5)
+#define SCE_SNAP_MODE_NODE_Y (1 << 6)
+
+/* ToolSettings.snap_mode and ToolSettings.snap_node_mode */
+#define SCE_SNAP_MODE_GRID (1 << 7)
 
 /** #ToolSettings.snap_transform_mode_flag */
 enum {
diff --git a/source/blender/makesdna/DNA_sequence_types.h b/source/blender/makesdna/DNA_sequence_types.h
index 5240de4aff9..087d30ce312 100644
--- a/source/blender/makesdna/DNA_sequence_types.h
+++ b/source/blender/makesdna/DNA_sequence_types.h
@@ -584,7 +584,7 @@ enum {
   SEQ_TYPE_LIGHTEN = 44,
   SEQ_TYPE_DODGE = 45,
   SEQ_TYPE_DARKEN = 46,
-  SEQ_TYPE_BURN = 47,
+  SEQ_TYPE_COLOR_BURN = 47,
   SEQ_TYPE_LINEAR_BURN = 48,
   SEQ_TYPE_OVERLAY = 49,
   SEQ_TYPE_HARD_LIGHT = 50,
diff --git a/source/blender/makesdna/DNA_space_types.h b/source/blender/makesdna/DNA_space_types.h
index 2a6a771bbeb..498ab290f31 100644
--- a/source/blender/makesdna/DNA_space_types.h
+++ b/source/blender/makesdna/DNA_space_types.h
@@ -246,7 +246,12 @@ typedef struct SpaceOutliner {
   char search_string[64];
   struct TreeStoreElem search_tse;
 
-  short flag, outlinevis, storeflag, search_flags;
+  short flag, outlinevis, storeflag;
+  char search_flags;
+
+  /** Selection syncing flag (#WM_OUTLINER_SYNC_SELECT_FROM_OBJECT and similar flags). */
+  char sync_select_dirty;
+
   int filter;
   char filter_state;
   char show_restrict_flags;
@@ -265,6 +270,7 @@ typedef enum eSpaceOutliner_Flag {
   SO_FLAG_UNUSED_1 = (1 << 2), /* cleared */
   SO_HIDE_KEYINGSETINFO = (1 << 3),
   SO_SKIP_SORT_ALPHA = (1 << 4),
+  SO_SYNC_SELECT = (1 << 5),
 } eSpaceOutliner_Flag;
 
 /* SpaceOutliner.filter */
@@ -285,11 +291,12 @@ typedef enum eSpaceOutliner_Filter {
 
   SO_FILTER_UNUSED_12 = (1 << 12),         /* cleared */
   SO_FILTER_OB_STATE_VISIBLE = (1 << 13),  /* Not set via DNA. */
-  SO_FILTER_OB_STATE_SELECTED = (1 << 14), /* Not set via DNA. */
-  SO_FILTER_OB_STATE_ACTIVE = (1 << 15),   /* Not set via DNA. */
-  SO_FILTER_NO_COLLECTION = (1 << 16),
+  SO_FILTER_OB_STATE_HIDDEN = (1 << 14),   /* Not set via DNA. */
+  SO_FILTER_OB_STATE_SELECTED = (1 << 15), /* Not set via DNA. */
+  SO_FILTER_OB_STATE_ACTIVE = (1 << 16),   /* Not set via DNA. */
+  SO_FILTER_NO_COLLECTION = (1 << 17),
 
-  SO_FILTER_ID_TYPE = (1 << 17),
+  SO_FILTER_ID_TYPE = (1 << 18),
 } eSpaceOutliner_Filter;
 
 #define SO_FILTER_OB_TYPE \
@@ -297,7 +304,8 @@ typedef enum eSpaceOutliner_Filter {
    SO_FILTER_NO_OB_LAMP | SO_FILTER_NO_OB_CAMERA | SO_FILTER_NO_OB_OTHERS)
 
 #define SO_FILTER_OB_STATE \
-  (SO_FILTER_OB_STATE_VISIBLE | SO_FILTER_OB_STATE_SELECTED | SO_FILTER_OB_STATE_ACTIVE)
+  (SO_FILTER_OB_STATE_VISIBLE | SO_FILTER_OB_STATE_HIDDEN | SO_FILTER_OB_STATE_SELECTED | \
+   SO_FILTER_OB_STATE_ACTIVE)
 
 #define SO_FILTER_ANY \
   (SO_FILTER_NO_OB_CONTENT | SO_FILTER_NO_CHILDREN | SO_FILTER_OB_TYPE | SO_FILTER_OB_STATE | \
@@ -307,8 +315,9 @@ typedef enum eSpaceOutliner_Filter {
 typedef enum eSpaceOutliner_StateFilter {
   SO_FILTER_OB_ALL = 0,
   SO_FILTER_OB_VISIBLE = 1,
-  SO_FILTER_OB_SELECTED = 2,
-  SO_FILTER_OB_ACTIVE = 3,
+  SO_FILTER_OB_HIDDEN = 2,
+  SO_FILTER_OB_SELECTED = 3,
+  SO_FILTER_OB_ACTIVE = 4,
 } eSpaceOutliner_StateFilter;
 
 /* SpaceOutliner.show_restrict_flags */
@@ -420,8 +429,7 @@ typedef enum eGraphEdit_Flag {
   SIPO_NOTRANSKEYCULL = (1 << 1),
   /* don't show any keyframe handles at all */
   SIPO_NOHANDLES = (1 << 2),
-  /* don't show current frame number beside indicator line */
-  SIPO_NODRAWCFRANUM = (1 << 3),
+  /* SIPO_NODRAWCFRANUM = (1 << 3), DEPRECATED */
   /* show timing in seconds instead of frames */
   SIPO_DRAWTIME = (1 << 4),
   /* only show keyframes for selected F-Curves */
@@ -496,8 +504,7 @@ typedef enum eSpaceNla_Flag {
   /* draw timing in seconds instead of frames */
   SNLA_DRAWTIME = (1 << 2),
   SNLA_FLAG_UNUSED_3 = (1 << 3),
-  /* don't draw frame number beside frame indicator */
-  SNLA_NODRAWCFRANUM = (1 << 4),
+  /* SNLA_NODRAWCFRANUM = (1 << 4), DEPRECATED */
   /* don't draw influence curves on strips */
   SNLA_NOSTRIPCURVES = (1 << 5),
   /* don't perform realtime updates */
@@ -581,7 +588,7 @@ typedef enum eSpaceSeq_Flag {
   SEQ_DRAW_COLOR_SEPARATED = (1 << 2),
   SEQ_SHOW_SAFE_MARGINS = (1 << 3),
   SEQ_SHOW_GPENCIL = (1 << 4),
-  SEQ_NO_DRAW_CFRANUM = (1 << 5),
+  /* SEQ_NO_DRAW_CFRANUM = (1 << 5), DEPRECATED */
   SEQ_USE_ALPHA = (1 << 6),     /* use RGBA display mode for preview */
   SEQ_ALL_WAVEFORMS = (1 << 7), /* draw all waveforms */
   SEQ_NO_WAVEFORMS = (1 << 8),  /* draw no waveforms */
@@ -672,8 +679,11 @@ typedef struct FileSelectParams {
   /** Display mode flag. */
   short display;
   short display_previous;
+  /** Details toggles (file size, creation date, etc.) */
+  char details_flags;
+  /* The type of file action (opening or saving) */
+  char action_type; /* eFileSel_Action */
   /** Filter when (flags & FILE_FILTER) is true. */
-  char _pad2[2];
   int filter;
 
   /** Max number of levels in dirtree to show at once, 0 to disable recursion. */
@@ -731,8 +741,8 @@ typedef struct SpaceFile {
 /* FileSelectParams.display */
 enum eFileDisplayType {
   FILE_DEFAULTDISPLAY = 0,
-  FILE_SHORTDISPLAY = 1,
-  FILE_LONGDISPLAY = 2,
+  FILE_VERTICALDISPLAY = 1,
+  FILE_HORIZONTALDISPLAY = 2,
   FILE_IMGDISPLAY = 3,
 };
 
@@ -745,6 +755,12 @@ enum eFileSortType {
   FILE_SORT_SIZE = 4,
 };
 
+/* FileSelectParams.details_flags */
+enum eFileDetails {
+  FILE_DETAILS_SIZE = (1 << 0),
+  FILE_DETAILS_DATETIME = (1 << 1),
+};
+
 /* these values need to be hardcoded in structs, dna does not recognize defines */
 /* also defined in BKE */
 #define FILE_MAXDIR 768
@@ -782,6 +798,8 @@ typedef enum eFileSel_Params_Flag {
   FILE_FILTER = (1 << 8),
   FILE_PARAMS_FLAG_UNUSED_9 = (1 << 9), /* cleared */
   FILE_GROUP_INSTANCE = (1 << 10),
+  FILE_SORT_INVERT = (1 << 11),
+  FILE_HIDE_TOOL_PROPS = (1 << 12)
 } eFileSel_Params_Flag;
 
 /* sfile->params->rename_flag */
@@ -819,6 +837,8 @@ typedef enum eFileSel_File_Types {
   FILE_TYPE_OPERATOR = (1 << 14),
   FILE_TYPE_APPLICATIONBUNDLE = (1 << 15),
   FILE_TYPE_ALEMBIC = (1 << 16),
+  /** For all kinds of recognized import/export formats. No need for specialized types. */
+  FILE_TYPE_OBJECT_IO = (1 << 17),
 
   /** An FS directory (i.e. S_ISDIR on its path is true). */
   FILE_TYPE_DIR = (1 << 30),
@@ -877,8 +897,7 @@ typedef struct FileDirEntryRevision {
   int64_t time;
   /* Temp caching of UI-generated strings... */
   char size_str[16];
-  char time_str[8];
-  char date_str[16];
+  char datetime_str[16 + 8];
 } FileDirEntryRevision;
 
 /* Container for a variant, only relevant in asset context.
@@ -1113,8 +1132,6 @@ typedef enum eSpaceImage_Flag {
   SI_SHOW_R = (1 << 27),
   SI_SHOW_G = (1 << 28),
   SI_SHOW_B = (1 << 29),
-
-  SI_NO_DRAWEDGES = (1 << 30),
 } eSpaceImage_Flag;
 
 /* SpaceImage.other_uv_filter */
diff --git a/source/blender/makesdna/DNA_texture_types.h b/source/blender/makesdna/DNA_texture_types.h
index 3fad1c57427..ead1bcc943d 100644
--- a/source/blender/makesdna/DNA_texture_types.h
+++ b/source/blender/makesdna/DNA_texture_types.h
@@ -35,7 +35,6 @@ extern "C" {
 struct AnimData;
 struct ColorBand;
 struct CurveMapping;
-struct ImBuf;
 struct Image;
 struct Ipo;
 struct Object;
diff --git a/source/blender/makesdna/DNA_userdef_types.h b/source/blender/makesdna/DNA_userdef_types.h
index b8914c7a74f..8eabbc0df0d 100644
--- a/source/blender/makesdna/DNA_userdef_types.h
+++ b/source/blender/makesdna/DNA_userdef_types.h
@@ -717,7 +717,7 @@ typedef struct UserDef {
   /** Overall sensitivity of 3D mouse. */
   float ndof_sensitivity;
   float ndof_orbit_sensitivity;
-  /** Deadzone of 3D mouse. */
+  /** Dead-zone of 3D mouse. */
   float ndof_deadzone;
   /** #eNdof_Flag, flags for 3D mouse. */
   int ndof_flag;
@@ -868,6 +868,7 @@ typedef enum eUserPref_Flag {
   USER_NONEGFRAMES = (1 << 24),
   USER_TXT_TABSTOSPACES_DISABLE = (1 << 25),
   USER_TOOLTIPS_PYTHON = (1 << 26),
+  USER_ADDONS_ENABLED_ONLY = (1 << 27),
 } eUserPref_Flag;
 
 typedef enum eUserPref_PrefFlag {
diff --git a/source/blender/makesdna/DNA_view3d_types.h b/source/blender/makesdna/DNA_view3d_types.h
index 3ba33cfe3d4..51655790fbd 100644
--- a/source/blender/makesdna/DNA_view3d_types.h
+++ b/source/blender/makesdna/DNA_view3d_types.h
@@ -25,11 +25,6 @@
 #define __DNA_VIEW3D_TYPES_H__
 
 struct BoundBox;
-struct GPUViewport;
-struct Image;
-struct Material;
-struct MovieClip;
-struct MovieClipUser;
 struct Object;
 struct RenderEngine;
 struct SmoothView3DStore;
diff --git a/source/blender/makesdna/DNA_windowmanager_types.h b/source/blender/makesdna/DNA_windowmanager_types.h
index 8dcae41aaa2..57c0a29382d 100644
--- a/source/blender/makesdna/DNA_windowmanager_types.h
+++ b/source/blender/makesdna/DNA_windowmanager_types.h
@@ -48,7 +48,6 @@ struct PointerRNA;
 struct Report;
 struct ReportList;
 struct Stereo3dFormat;
-struct UndoStep;
 struct bContext;
 struct bScreen;
 struct uiLayout;
@@ -131,12 +130,15 @@ typedef struct wmWindowManager {
   ListBase windows;
 
   /** Set on file read. */
-  int initialized;
+  short initialized;
   /** Indicator whether data was saved. */
   short file_saved;
   /** Operator stack depth to avoid nested undo pushes. */
   short op_undo_depth;
 
+  /** Set after selection to notify outliner to sync. Stores type of selection */
+  short outliner_sync_select_dirty;
+
   /** Operator registry. */
   ListBase operators;
 
@@ -186,6 +188,18 @@ enum {
   WM_KEYCONFIG_IS_INITIALIZED = (1 << 1),
 };
 
+/* wmWindowManager.outliner_sync_select_dirty */
+enum {
+  WM_OUTLINER_SYNC_SELECT_FROM_OBJECT = (1 << 0),
+  WM_OUTLINER_SYNC_SELECT_FROM_EDIT_BONE = (1 << 1),
+  WM_OUTLINER_SYNC_SELECT_FROM_POSE_BONE = (1 << 2),
+  WM_OUTLINER_SYNC_SELECT_FROM_SEQUENCE = (1 << 3),
+};
+
+#define WM_OUTLINER_SYNC_SELECT_FROM_ALL \
+  (WM_OUTLINER_SYNC_SELECT_FROM_OBJECT | WM_OUTLINER_SYNC_SELECT_FROM_EDIT_BONE | \
+   WM_OUTLINER_SYNC_SELECT_FROM_POSE_BONE | WM_OUTLINER_SYNC_SELECT_FROM_SEQUENCE)
+
 #define WM_KEYCONFIG_STR_DEFAULT "blender"
 
 /* IME is win32 only! */
diff --git a/source/blender/makesdna/DNA_world_types.h b/source/blender/makesdna/DNA_world_types.h
index 5143f667083..af1e5123dc8 100644
--- a/source/blender/makesdna/DNA_world_types.h
+++ b/source/blender/makesdna/DNA_world_types.h
@@ -29,7 +29,6 @@
 
 struct AnimData;
 struct Ipo;
-struct MTex;
 struct bNodeTree;
 
 #ifndef MAX_MTEX
diff --git a/source/blender/makesdna/intern/dna_genfile.c b/source/blender/makesdna/intern/dna_genfile.c
index 5252c8f3350..09a7a157996 100644
--- a/source/blender/makesdna/intern/dna_genfile.c
+++ b/source/blender/makesdna/intern/dna_genfile.c
@@ -339,7 +339,7 @@ static bool init_structDNA(SDNA *sdna, bool do_endian_swap, const char **r_error
 
   data = (int *)sdna->data;
 
-  /* clear pointers incase of error */
+  /* Clear pointers in case of error. */
   sdna->names = NULL;
   sdna->types = NULL;
   sdna->structs = NULL;
@@ -1102,7 +1102,7 @@ static void reconstruct_elem(const SDNA *newsdna,
    *          - cast type
    *      - name partially equal (array differs)
    *          - type equal: memcpy
-   *          - types casten
+   *          - type cast (per element).
    * (nzc 2-4-2001 I want the 'unsigned' bit to be parsed as well. Where
    * can I force this?)
    */
diff --git a/source/blender/makesrna/RNA_access.h b/source/blender/makesrna/RNA_access.h
index 613df728dc9..0b60d1fa344 100644
--- a/source/blender/makesrna/RNA_access.h
+++ b/source/blender/makesrna/RNA_access.h
@@ -1158,24 +1158,37 @@ struct PropertyElemRNA {
 };
 bool RNA_path_resolve_elements(PointerRNA *ptr, const char *path, struct ListBase *r_elements);
 
+struct ID *RNA_find_real_ID_and_path(struct Main *bmain, struct ID *id, const char **r_path);
+
 char *RNA_path_from_ID_to_struct(PointerRNA *ptr);
+
+char *RNA_path_from_real_ID_to_struct(struct Main *bmain, PointerRNA *ptr, struct ID **r_real);
+
 char *RNA_path_from_ID_to_property(PointerRNA *ptr, PropertyRNA *prop);
 char *RNA_path_from_ID_to_property_index(PointerRNA *ptr,
                                          PropertyRNA *prop,
                                          int array_dim,
                                          int index);
 
+char *RNA_path_from_real_ID_to_property_index(struct Main *bmain,
+                                              PointerRNA *ptr,
+                                              PropertyRNA *prop,
+                                              int array_dim,
+                                              int index,
+                                              struct ID **r_real_id);
+
 char *RNA_path_resolve_from_type_to_property(struct PointerRNA *ptr,
                                              struct PropertyRNA *prop,
                                              const struct StructRNA *type);
 
-char *RNA_path_full_ID_py(struct ID *id);
-char *RNA_path_full_struct_py(struct PointerRNA *ptr);
-char *RNA_path_full_property_py_ex(PointerRNA *ptr,
-                                   PropertyRNA *prop,
-                                   int index,
-                                   bool use_fallback);
-char *RNA_path_full_property_py(struct PointerRNA *ptr, struct PropertyRNA *prop, int index);
+char *RNA_path_full_ID_py(struct Main *bmain, struct ID *id);
+char *RNA_path_full_struct_py(struct Main *bmain, struct PointerRNA *ptr);
+char *RNA_path_full_property_py_ex(
+    struct Main *bmain, PointerRNA *ptr, PropertyRNA *prop, int index, bool use_fallback);
+char *RNA_path_full_property_py(struct Main *bmain,
+                                struct PointerRNA *ptr,
+                                struct PropertyRNA *prop,
+                                int index);
 char *RNA_path_struct_property_py(struct PointerRNA *ptr, struct PropertyRNA *prop, int index);
 char *RNA_path_property_py(struct PointerRNA *ptr, struct PropertyRNA *prop, int index);
 
@@ -1418,8 +1431,8 @@ StructRNA *ID_code_to_RNA_type(short idcode);
 #define RNA_POINTER_INVALIDATE(ptr) \
   { \
     /* this is checked for validity */ \
-    (ptr)->type = /* should not be needed but prevent bad pointer access, just in case */ \
-        (ptr)->id.data = NULL; \
+    (ptr)->type = NULL; /* should not be needed but prevent bad pointer access, just in case */ \
+    (ptr)->owner_id = NULL; \
   } \
   (void)0
 
@@ -1511,8 +1524,8 @@ bool RNA_struct_override_store(struct Main *bmain,
                                struct IDOverrideLibrary *override);
 
 void RNA_struct_override_apply(struct Main *bmain,
-                               struct PointerRNA *ptr_local,
-                               struct PointerRNA *ptr_override,
+                               struct PointerRNA *ptr_dst,
+                               struct PointerRNA *ptr_src,
                                struct PointerRNA *ptr_storage,
                                struct IDOverrideLibrary *override);
 
diff --git a/source/blender/makesrna/RNA_enum_types.h b/source/blender/makesrna/RNA_enum_types.h
index 02aaef44a1f..b3e1f22f413 100644
--- a/source/blender/makesrna/RNA_enum_types.h
+++ b/source/blender/makesrna/RNA_enum_types.h
@@ -91,6 +91,7 @@ extern const EnumPropertyItem rna_enum_beztriple_keyframe_type_items[];
 extern const EnumPropertyItem rna_enum_beztriple_interpolation_mode_items[];
 extern const EnumPropertyItem rna_enum_beztriple_interpolation_easing_items[];
 extern const EnumPropertyItem rna_enum_keyframe_handle_type_items[];
+extern const EnumPropertyItem rna_enum_driver_target_rotation_mode_items[];
 
 extern const EnumPropertyItem rna_enum_keyblock_type_items[];
 
diff --git a/source/blender/makesrna/RNA_types.h b/source/blender/makesrna/RNA_types.h
index 38631d1acf2..2a341249247 100644
--- a/source/blender/makesrna/RNA_types.h
+++ b/source/blender/makesrna/RNA_types.h
@@ -29,6 +29,7 @@ extern "C" {
 
 struct BlenderRNA;
 struct FunctionRNA;
+struct ID;
 struct Main;
 struct ParameterList;
 struct PropertyRNA;
@@ -44,10 +45,7 @@ struct bContext;
  * the properties and validate them. */
 
 typedef struct PointerRNA {
-  struct {
-    void *data;
-  } id;
-
+  struct ID *owner_id;
   struct StructRNA *type;
   void *data;
 } PointerRNA;
@@ -534,7 +532,7 @@ typedef struct ParameterDynAlloc {
 typedef enum FunctionFlag {
   /**
    * Pass ID owning 'self' data
-   * (i.e. ptr->id.data, might be same as self in case data is an ID...).
+   * (i.e. ptr->owner_id, might be same as self in case data is an ID...).
    */
   FUNC_USE_SELF_ID = (1 << 11),
 
diff --git a/source/blender/makesrna/intern/CMakeLists.txt b/source/blender/makesrna/intern/CMakeLists.txt
index ab13c6c9ed4..05408f1a41a 100644
--- a/source/blender/makesrna/intern/CMakeLists.txt
+++ b/source/blender/makesrna/intern/CMakeLists.txt
@@ -365,9 +365,11 @@ add_custom_command(
 # Build bf_rna
 set(SRC
   rna_access.c
+  rna_access_compare_override.c
   ${GENSRC}
 
   ${SRC_RNA_INC}
+  rna_access_internal.h
   rna_internal.h
   rna_internal_types.h
   rna_mesh_utils.h
diff --git a/source/blender/makesrna/intern/makesrna.c b/source/blender/makesrna/intern/makesrna.c
index f6c248acdb5..1617d85b6c2 100644
--- a/source/blender/makesrna/intern/makesrna.c
+++ b/source/blender/makesrna/intern/makesrna.c
@@ -389,7 +389,7 @@ static void rna_print_data_get(FILE *f, PropertyDefRNA *dp)
 
 static void rna_print_id_get(FILE *f, PropertyDefRNA *UNUSED(dp))
 {
-  fprintf(f, "    ID *id = ptr->id.data;\n");
+  fprintf(f, "    ID *id = ptr->owner_id;\n");
 }
 
 static void rna_construct_function_name(
@@ -2470,7 +2470,7 @@ static void rna_def_struct_function_call_impl_cpp(FILE *f, StructRNA *srna, Func
   dsrna = rna_find_struct_def(srna);
 
   if (func->flag & FUNC_USE_SELF_ID) {
-    WRITE_PARAM("(::ID *) ptr.id.data");
+    WRITE_PARAM("(::ID *) ptr.owner_id");
   }
 
   if ((func->flag & FUNC_NO_SELF) == 0) {
@@ -2579,7 +2579,7 @@ static void rna_def_struct_function_impl_cpp(FILE *f, StructRNA *srna, FunctionD
         }
         else {
           fprintf(f,
-                  "\t\tRNA_pointer_create((::ID *) ptr.id.data, &RNA_%s, retdata, &result);\n",
+                  "\t\tRNA_pointer_create((::ID *) ptr.owner_id, &RNA_%s, retdata, &result);\n",
                   (const char *)pprop->type);
         }
       }
@@ -2792,7 +2792,7 @@ static void rna_def_function_funcs(FILE *f, StructDefRNA *dsrna, FunctionDefRNA
 
   /* assign self */
   if (func->flag & FUNC_USE_SELF_ID) {
-    fprintf(f, "\t_selfid = (struct ID *)_ptr->id.data;\n");
+    fprintf(f, "\t_selfid = (struct ID *)_ptr->owner_id;\n");
   }
 
   if ((func->flag & FUNC_NO_SELF) == 0) {
diff --git a/source/blender/makesrna/intern/rna_ID.c b/source/blender/makesrna/intern/rna_ID.c
index 57cdbbadeb8..5a4b4a3fa3d 100644
--- a/source/blender/makesrna/intern/rna_ID.c
+++ b/source/blender/makesrna/intern/rna_ID.c
@@ -620,9 +620,9 @@ static int rna_IDPArray_length(PointerRNA *ptr)
 
 int rna_IDMaterials_assign_int(PointerRNA *ptr, int key, const PointerRNA *assign_ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   short *totcol = give_totcolp_id(id);
-  Material *mat_id = assign_ptr->id.data;
+  Material *mat_id = (Material *)assign_ptr->owner_id;
   if (totcol && (key >= 0 && key < *totcol)) {
     BLI_assert(BKE_id_is_in_global_main(id));
     BLI_assert(BKE_id_is_in_global_main(&mat_id->id));
@@ -642,8 +642,7 @@ static void rna_IDMaterials_append_id(ID *id, Main *bmain, Material *ma)
   WM_main_add_notifier(NC_OBJECT | ND_OB_SHADING, id);
 }
 
-static Material *rna_IDMaterials_pop_id(
-    ID *id, Main *bmain, ReportList *reports, int index_i, bool remove_material_slot)
+static Material *rna_IDMaterials_pop_id(ID *id, Main *bmain, ReportList *reports, int index_i)
 {
   Material *ma;
   short *totcol = give_totcolp_id(id);
@@ -657,7 +656,7 @@ static Material *rna_IDMaterials_pop_id(
     return NULL;
   }
 
-  ma = BKE_material_pop_id(bmain, id, index_i, remove_material_slot);
+  ma = BKE_material_pop_id(bmain, id, index_i);
 
   if (*totcol == totcol_orig) {
     BKE_report(reports, RPT_ERROR, "No material to removed");
@@ -671,9 +670,9 @@ static Material *rna_IDMaterials_pop_id(
   return ma;
 }
 
-static void rna_IDMaterials_clear_id(ID *id, Main *bmain, bool remove_material_slot)
+static void rna_IDMaterials_clear_id(ID *id, Main *bmain)
 {
-  BKE_material_clear_id(bmain, id, remove_material_slot);
+  BKE_material_clear_id(bmain, id);
 
   DEG_id_tag_update(id, ID_RECALC_GEOMETRY);
   WM_main_add_notifier(NC_OBJECT | ND_DRAW, id);
@@ -691,7 +690,7 @@ static void rna_Library_filepath_set(PointerRNA *ptr, const char *value)
 
 static void rna_ImagePreview_is_custom_set(PointerRNA *ptr, int value, enum eIconSizes size)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   PreviewImage *prv_img = (PreviewImage *)ptr->data;
 
   if (id != NULL) {
@@ -717,7 +716,7 @@ static void rna_ImagePreview_is_custom_set(PointerRNA *ptr, int value, enum eIco
 
 static void rna_ImagePreview_size_get(PointerRNA *ptr, int *values, enum eIconSizes size)
 {
-  ID *id = (ID *)ptr->id.data;
+  ID *id = ptr->owner_id;
   PreviewImage *prv_img = (PreviewImage *)ptr->data;
 
   if (id != NULL) {
@@ -732,7 +731,7 @@ static void rna_ImagePreview_size_get(PointerRNA *ptr, int *values, enum eIconSi
 
 static void rna_ImagePreview_size_set(PointerRNA *ptr, const int *values, enum eIconSizes size)
 {
-  ID *id = (ID *)ptr->id.data;
+  ID *id = ptr->owner_id;
   PreviewImage *prv_img = (PreviewImage *)ptr->data;
 
   if (id != NULL) {
@@ -755,7 +754,7 @@ static int rna_ImagePreview_pixels_get_length(PointerRNA *ptr,
                                               int length[RNA_MAX_ARRAY_DIMENSION],
                                               enum eIconSizes size)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   PreviewImage *prv_img = (PreviewImage *)ptr->data;
 
   if (id != NULL) {
@@ -771,7 +770,7 @@ static int rna_ImagePreview_pixels_get_length(PointerRNA *ptr,
 
 static void rna_ImagePreview_pixels_get(PointerRNA *ptr, int *values, enum eIconSizes size)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   PreviewImage *prv_img = (PreviewImage *)ptr->data;
 
   if (id != NULL) {
@@ -785,7 +784,7 @@ static void rna_ImagePreview_pixels_get(PointerRNA *ptr, int *values, enum eIcon
 
 static void rna_ImagePreview_pixels_set(PointerRNA *ptr, const int *values, enum eIconSizes size)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   PreviewImage *prv_img = (PreviewImage *)ptr->data;
 
   if (id != NULL) {
@@ -800,7 +799,7 @@ static int rna_ImagePreview_pixels_float_get_length(PointerRNA *ptr,
                                                     int length[RNA_MAX_ARRAY_DIMENSION],
                                                     enum eIconSizes size)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   PreviewImage *prv_img = (PreviewImage *)ptr->data;
 
   BLI_assert(sizeof(unsigned int) == 4);
@@ -818,7 +817,7 @@ static int rna_ImagePreview_pixels_float_get_length(PointerRNA *ptr,
 
 static void rna_ImagePreview_pixels_float_get(PointerRNA *ptr, float *values, enum eIconSizes size)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   PreviewImage *prv_img = (PreviewImage *)ptr->data;
 
   unsigned char *data = (unsigned char *)prv_img->rect[size];
@@ -842,7 +841,7 @@ static void rna_ImagePreview_pixels_float_set(PointerRNA *ptr,
                                               const float *values,
                                               enum eIconSizes size)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   PreviewImage *prv_img = (PreviewImage *)ptr->data;
 
   unsigned char *data = (unsigned char *)prv_img->rect[size];
@@ -959,7 +958,7 @@ static int rna_ImagePreview_icon_id_get(PointerRNA *ptr)
 {
   /* Using a callback here allows us to only generate icon matching
    * that preview when icon_id is requested. */
-  return BKE_icon_preview_ensure(ptr->id.data, (PreviewImage *)(ptr->data));
+  return BKE_icon_preview_ensure(ptr->owner_id, (PreviewImage *)(ptr->data));
 }
 static void rna_ImagePreview_icon_reload(PreviewImage *prv)
 {
@@ -1116,16 +1115,12 @@ static void rna_def_ID_materials(BlenderRNA *brna)
   RNA_def_function_ui_description(func, "Remove a material from the data-block");
   parm = RNA_def_int(
       func, "index", -1, -MAXMAT, MAXMAT, "", "Index of material to remove", 0, MAXMAT);
-  RNA_def_boolean(
-      func, "update_data", 0, "", "Update data by re-adjusting the material slots assigned");
   parm = RNA_def_pointer(func, "material", "Material", "", "Material to remove");
   RNA_def_function_return(func, parm);
 
   func = RNA_def_function(srna, "clear", "rna_IDMaterials_clear_id");
   RNA_def_function_flag(func, FUNC_USE_MAIN);
   RNA_def_function_ui_description(func, "Remove all materials from the data-block");
-  RNA_def_boolean(
-      func, "update_data", 0, "", "Update data by re-adjusting the material slots assigned");
 }
 
 static void rna_def_image_preview(BlenderRNA *brna)
@@ -1455,6 +1450,7 @@ static void rna_def_ID(BlenderRNA *brna)
       "Actual data-block from .blend file (Main database) that generated that evaluated one");
   RNA_def_property_pointer_funcs(prop, "rna_ID_original_get", NULL, NULL, NULL);
   RNA_def_property_clear_flag(prop, PROP_EDITABLE | PROP_PTR_NO_OWNERSHIP);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_NO_COMPARISON);
 
   prop = RNA_def_property(srna, "users", PROP_INT, PROP_UNSIGNED);
   RNA_def_property_int_sdna(prop, NULL, "us");
@@ -1483,6 +1479,7 @@ static void rna_def_ID(BlenderRNA *brna)
   prop = RNA_def_property(srna, "library", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "lib");
   RNA_def_property_clear_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_NO_COMPARISON);
   RNA_def_property_ui_text(prop, "Library", "Library file the data-block is linked from");
 
   prop = RNA_def_pointer(
@@ -1607,6 +1604,7 @@ static void rna_def_library(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "parent", PROP_POINTER, PROP_NONE);
   RNA_def_property_struct_type(prop, "Library");
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_NO_COMPARISON);
   RNA_def_property_ui_text(prop, "Parent", "");
 
   prop = RNA_def_property(srna, "packed_file", PROP_POINTER, PROP_NONE);
diff --git a/source/blender/makesrna/intern/rna_access.c b/source/blender/makesrna/intern/rna_access.c
index 330d45c2fa5..61634a84d41 100644
--- a/source/blender/makesrna/intern/rna_access.c
+++ b/source/blender/makesrna/intern/rna_access.c
@@ -37,21 +37,18 @@
 #include "BLI_ghash.h"
 #include "BLI_math.h"
 
-#ifdef DEBUG_OVERRIDE_TIMEIT
-#  include "PIL_time_utildefines.h"
-#endif
-
 #include "BLF_api.h"
 #include "BLT_translation.h"
 
 #include "BKE_animsys.h"
+#include "BKE_collection.h"
 #include "BKE_context.h"
 #include "BKE_idcode.h"
 #include "BKE_idprop.h"
 #include "BKE_fcurve.h"
-#include "BKE_library_override.h"
 #include "BKE_main.h"
 #include "BKE_report.h"
+#include "BKE_node.h"
 
 #include "DEG_depsgraph.h"
 
@@ -67,8 +64,9 @@
 #include "WM_types.h"
 
 #include "rna_internal.h"
+#include "rna_access_internal.h"
 
-const PointerRNA PointerRNA_NULL = {{NULL}};
+const PointerRNA PointerRNA_NULL = {NULL};
 
 /* Init/Exit */
 
@@ -116,7 +114,7 @@ void RNA_exit(void)
 
 void RNA_main_pointer_create(struct Main *main, PointerRNA *r_ptr)
 {
-  r_ptr->id.data = NULL;
+  r_ptr->owner_id = NULL;
   r_ptr->type = &RNA_BlendData;
   r_ptr->data = main;
 }
@@ -126,7 +124,7 @@ void RNA_id_pointer_create(ID *id, PointerRNA *r_ptr)
   StructRNA *type, *idtype = NULL;
 
   if (id) {
-    PointerRNA tmp = {{NULL}};
+    PointerRNA tmp = {NULL};
     tmp.data = id;
     idtype = rna_ID_refine(&tmp);
 
@@ -142,7 +140,7 @@ void RNA_id_pointer_create(ID *id, PointerRNA *r_ptr)
     }
   }
 
-  r_ptr->id.data = id;
+  r_ptr->owner_id = id;
   r_ptr->type = idtype;
   r_ptr->data = id;
 }
@@ -153,13 +151,13 @@ void RNA_pointer_create(ID *id, StructRNA *type, void *data, PointerRNA *r_ptr)
   StructRNA *idtype = NULL;
 
   if (id) {
-    PointerRNA tmp = {{0}};
+    PointerRNA tmp = {0};
     tmp.data = id;
     idtype = rna_ID_refine(&tmp);
   }
 #endif
 
-  r_ptr->id.data = id;
+  r_ptr->owner_id = id;
   r_ptr->type = type;
   r_ptr->data = data;
 
@@ -179,22 +177,22 @@ void RNA_pointer_create(ID *id, StructRNA *type, void *data, PointerRNA *r_ptr)
 
 bool RNA_pointer_is_null(const PointerRNA *ptr)
 {
-  return !((ptr->data != NULL) && (ptr->id.data != NULL) && (ptr->type != NULL));
+  return !((ptr->data != NULL) && (ptr->owner_id != NULL) && (ptr->type != NULL));
 }
 
 static void rna_pointer_inherit_id(StructRNA *type, PointerRNA *parent, PointerRNA *ptr)
 {
   if (type && type->flag & STRUCT_ID) {
-    ptr->id.data = ptr->data;
+    ptr->owner_id = ptr->data;
   }
   else {
-    ptr->id.data = parent->id.data;
+    ptr->owner_id = parent->owner_id;
   }
 }
 
 void RNA_blender_rna_pointer_create(PointerRNA *r_ptr)
 {
-  r_ptr->id.data = NULL;
+  r_ptr->owner_id = NULL;
   r_ptr->type = &RNA_BlenderRNA;
   r_ptr->data = &BLENDER_RNA;
 }
@@ -229,7 +227,7 @@ void RNA_pointer_recast(PointerRNA *ptr, PointerRNA *r_ptr)
 #if 0 /* works but this case if covered by more general code below. */
   if (RNA_struct_is_ID(ptr->type)) {
     /* simple case */
-    RNA_id_pointer_create(ptr->id.data, r_ptr);
+    RNA_id_pointer_create(ptr->owner_id, r_ptr);
   }
   else
 #endif
@@ -249,7 +247,7 @@ void RNA_pointer_recast(PointerRNA *ptr, PointerRNA *r_ptr)
 
 /* ID Properties */
 
-static void rna_idproperty_touch(IDProperty *idprop)
+void rna_idproperty_touch(IDProperty *idprop)
 {
   /* so the property is seen as 'set' by rna */
   idprop->flag &= ~IDP_FLAG_GHOST;
@@ -619,7 +617,7 @@ PropertyRNA *rna_ensure_property_realdata(PropertyRNA **prop, PointerRNA *ptr)
   return rna_idproperty_check_ex(prop, ptr, true);
 }
 
-static PropertyRNA *rna_ensure_property(PropertyRNA *prop)
+PropertyRNA *rna_ensure_property(PropertyRNA *prop)
 {
   /* the quick version if we don't need the idproperty */
 
@@ -1171,11 +1169,6 @@ int RNA_property_flag(PropertyRNA *prop)
   return rna_ensure_property(prop)->flag;
 }
 
-int RNA_property_override_flag(PropertyRNA *prop)
-{
-  return rna_ensure_property(prop)->flag_override;
-}
-
 /**
  * Get the tags set for \a prop as int bitfield.
  * \note Doesn't perform any validity check on the set bits. #RNA_def_property_tags does this
@@ -2067,7 +2060,7 @@ int RNA_property_ui_icon(PropertyRNA *prop)
 
 bool RNA_property_editable(PointerRNA *ptr, PropertyRNA *prop)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   int flag;
   const char *dummy_info;
 
@@ -2085,7 +2078,7 @@ bool RNA_property_editable(PointerRNA *ptr, PropertyRNA *prop)
  */
 bool RNA_property_editable_info(PointerRNA *ptr, PropertyRNA *prop, const char **r_info)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   int flag;
 
   prop = rna_ensure_property(prop);
@@ -2152,7 +2145,7 @@ bool RNA_property_editable_index(PointerRNA *ptr, PropertyRNA *prop, int index)
     flag &= prop->itemeditable(ptr, index);
   }
 
-  id = ptr->id.data;
+  id = ptr->owner_id;
 
   return (flag & PROP_EDITABLE) && (!id || !ID_IS_LINKED(id) || (prop->flag & PROP_LIB_EXCEPTION));
 }
@@ -2160,7 +2153,7 @@ bool RNA_property_editable_index(PointerRNA *ptr, PropertyRNA *prop, int index)
 bool RNA_property_animateable(PointerRNA *ptr, PropertyRNA *prop)
 {
   /* check that base ID-block can support animation data */
-  if (!id_can_have_animdata(ptr->id.data)) {
+  if (!id_can_have_animdata(ptr->owner_id)) {
     return false;
   }
 
@@ -2194,77 +2187,6 @@ bool RNA_property_animated(PointerRNA *ptr, PropertyRNA *prop)
 
   return false;
 }
-
-/** \note Does not take into account editable status, this has to be checked separately
- * (using #RNA_property_editable_flag() usually). */
-bool RNA_property_overridable_get(PointerRNA *ptr, PropertyRNA *prop)
-{
-  if (prop->magic == RNA_MAGIC) {
-    /* Special handling for insertions of constraints or modifiers... */
-    /* TODO Note We may want to add a more generic system to RNA
-     * (like a special property in struct of items)
-     * if we get more overrideable collections,
-     * for now we can live with those special-cases handling I think. */
-    if (RNA_struct_is_a(ptr->type, &RNA_Constraint)) {
-      bConstraint *con = ptr->data;
-      if (con->flag & CONSTRAINT_OVERRIDE_LIBRARY_LOCAL) {
-        return true;
-      }
-    }
-    else if (RNA_struct_is_a(ptr->type, &RNA_Modifier)) {
-      ModifierData *mod = ptr->data;
-      if (mod->flag & eModifierFlag_OverrideLibrary_Local) {
-        return true;
-      }
-    }
-    /* If this is a RNA-defined property (real or 'virtual' IDProp),
-     * we want to use RNA prop flag. */
-    return !(prop->flag_override & PROPOVERRIDE_NO_COMPARISON) &&
-           (prop->flag_override & PROPOVERRIDE_OVERRIDABLE_LIBRARY);
-  }
-  else {
-    /* If this is a real 'pure' IDProp (aka custom property), we want to use the IDProp flag. */
-    return !(prop->flag_override & PROPOVERRIDE_NO_COMPARISON) &&
-           (((IDProperty *)prop)->flag & IDP_FLAG_OVERRIDABLE_LIBRARY);
-  }
-}
-
-/* Should only be used for custom properties */
-bool RNA_property_overridable_library_set(PointerRNA *UNUSED(ptr),
-                                          PropertyRNA *prop,
-                                          const bool is_overridable)
-{
-  /* Only works for pure custom properties IDProps. */
-  if (prop->magic != RNA_MAGIC) {
-    IDProperty *idprop = (IDProperty *)prop;
-
-    idprop->flag = is_overridable ? (idprop->flag | IDP_FLAG_OVERRIDABLE_LIBRARY) :
-                                    (idprop->flag & ~IDP_FLAG_OVERRIDABLE_LIBRARY);
-    return true;
-  }
-
-  return false;
-}
-
-bool RNA_property_overridden(PointerRNA *ptr, PropertyRNA *prop)
-{
-  char *rna_path = RNA_path_from_ID_to_property(ptr, prop);
-  ID *id = ptr->id.data;
-
-  if (rna_path == NULL || id == NULL || id->override_library == NULL) {
-    return false;
-  }
-
-  return (BKE_override_library_property_find(id->override_library, rna_path) != NULL);
-}
-
-bool RNA_property_comparable(PointerRNA *UNUSED(ptr), PropertyRNA *prop)
-{
-  prop = rna_ensure_property(prop);
-
-  return !(prop->flag_override & PROPOVERRIDE_NO_COMPARISON);
-}
-
 /* this function is to check if its possible to create a valid path from the ID
  * its slow so don't call in a loop */
 bool RNA_property_path_from_ID_check(PointerRNA *ptr, PropertyRNA *prop)
@@ -2277,7 +2199,7 @@ bool RNA_property_path_from_ID_check(PointerRNA *ptr, PropertyRNA *prop)
     PointerRNA r_ptr;
     PropertyRNA *r_prop;
 
-    RNA_id_pointer_create(ptr->id.data, &id_ptr);
+    RNA_id_pointer_create(ptr->owner_id, &id_ptr);
     if (RNA_path_resolve(&id_ptr, path, &r_ptr, &r_prop) == true) {
       ret = (prop == r_prop);
     }
@@ -2316,7 +2238,7 @@ static void rna_property_update(
     /* TODO(campbell): Should eventually be replaced entirely by message bus (below)
      * for now keep since COW, bugs are hard to track when we have other missing updates. */
     if (prop->noteflag) {
-      WM_main_add_notifier(prop->noteflag, ptr->id.data);
+      WM_main_add_notifier(prop->noteflag, ptr->owner_id);
     }
 #endif
 
@@ -2327,10 +2249,10 @@ static void rna_property_update(
       /* we could add NULL check, for now don't */
       WM_msg_publish_rna(mbus, ptr, prop);
     }
-    if (ptr->id.data != NULL && ((prop->flag & PROP_NO_DEG_UPDATE) == 0)) {
-      const short id_type = GS(((ID *)ptr->id.data)->name);
+    if (ptr->owner_id != NULL && ((prop->flag & PROP_NO_DEG_UPDATE) == 0)) {
+      const short id_type = GS(ptr->owner_id->name);
       if (ID_TYPE_IS_COW(id_type)) {
-        DEG_id_tag_update(ptr->id.data, ID_RECALC_COPY_ON_WRITE);
+        DEG_id_tag_update(ptr->owner_id, ID_RECALC_COPY_ON_WRITE);
       }
     }
     /* End message bus. */
@@ -2339,14 +2261,14 @@ static void rna_property_update(
   if (!is_rna || (prop->flag & PROP_IDPROPERTY)) {
     /* WARNING! This is so property drivers update the display!
      * not especially nice  */
-    DEG_id_tag_update(ptr->id.data,
+    DEG_id_tag_update(ptr->owner_id,
                       ID_RECALC_TRANSFORM | ID_RECALC_GEOMETRY | ID_RECALC_PARAMETERS);
     WM_main_add_notifier(NC_WINDOW, NULL);
     /* Not nice as well, but the only way to make sure material preview
      * is updated with custom nodes.
      */
-    if ((prop->flag & PROP_IDPROPERTY) != 0 && (ptr->id.data != NULL) &&
-        (GS(((ID *)ptr->id.data)->name) == ID_NT)) {
+    if ((prop->flag & PROP_IDPROPERTY) != 0 && (ptr->owner_id != NULL) &&
+        (GS(ptr->owner_id->name) == ID_NT)) {
       WM_main_add_notifier(NC_MATERIAL | ND_SHADING, NULL);
     }
   }
@@ -2379,7 +2301,7 @@ void RNA_property_update_main(Main *bmain, Scene *scene, PointerRNA *ptr, Proper
  * property updates to actually work).
  *
  * The cache is structured with a dual-layer structure
- * - L1 = PointerRNA used as key; id.data is used (it should always be defined,
+ * - L1 = PointerRNA used as key; owner_id is used (it should always be defined,
  *        and most updates end up using just that anyways)
  * - L2 = Update functions to be called on those PointerRNA's
  */
@@ -2423,7 +2345,7 @@ void RNA_property_update_cache_add(PointerRNA *ptr, PropertyRNA *prop)
      * since most update calls that we'll encounter only really care about this. */
     /* TODO: later, the cache might need to have some nesting on L1 to cope better
      * with these problems + some tagging to indicate we need this */
-    if (uce->ptr.id.data == ptr->id.data) {
+    if (uce->ptr.owner_id == ptr->owner_id) {
       break;
     }
   }
@@ -2433,7 +2355,7 @@ void RNA_property_update_cache_add(PointerRNA *ptr, PropertyRNA *prop)
     BLI_addtail(&rna_updates_cache, uce);
 
     /* copy pointer */
-    RNA_pointer_create(ptr->id.data, ptr->type, ptr->data, &uce->ptr);
+    RNA_pointer_create(ptr->owner_id, ptr->type, ptr->data, &uce->ptr);
   }
 
   /* check on the update func */
@@ -3851,7 +3773,7 @@ void RNA_property_pointer_set(PointerRNA *ptr,
 
   /* RNA */
   if (pprop->set && !((prop->flag & PROP_NEVER_NULL) && ptr_value.data == NULL) &&
-      !((prop->flag & PROP_ID_SELF_CHECK) && ptr->id.data == ptr_value.id.data)) {
+      !((prop->flag & PROP_ID_SELF_CHECK) && ptr->owner_id == ptr_value.owner_id)) {
     pprop->set(ptr, ptr_value, reports);
   }
   /* IDProperty */
@@ -5440,6 +5362,10 @@ static bool rna_path_parse(PointerRNA *ptr,
  * Resolve the given RNA Path to find the pointer and/or property
  * indicated by fully resolving the path.
  *
+ * \warning Unlike \a RNA_path_resolve_property(), that one *will* try to follow RNAPointers,
+ * e.g. the path 'parent' applied to a RNAObject \a ptr will return the object.parent in \a r_ptr,
+ * and a NULL \a r_prop...
+ *
  * \note Assumes all pointers provided are valid
  * \return True if path can be resolved to a valid "pointer + property" OR "pointer only"
  */
@@ -5814,14 +5740,14 @@ static char *rna_path_from_ID_to_idpgroup(PointerRNA *ptr)
   IDProperty *haystack;
   IDProperty *needle;
 
-  BLI_assert(ptr->id.data != NULL);
+  BLI_assert(ptr->owner_id != NULL);
 
   /* TODO, Support Bones/PoseBones. no pointers stored to the bones from here, only the ID.
    *       See example in T25746.
    *       Unless this is added only way to find this is to also search
    *       all bones and pose bones of an armature or object.
    */
-  RNA_id_pointer_create(ptr->id.data, &id_ptr);
+  RNA_id_pointer_create(ptr->owner_id, &id_ptr);
 
   haystack = RNA_struct_idprops(&id_ptr, false);
   if (haystack) { /* can fail when called on bones */
@@ -5833,11 +5759,79 @@ static char *rna_path_from_ID_to_idpgroup(PointerRNA *ptr)
   }
 }
 
+/**
+ * Find the actual ID pointer and path from it to the given ID.
+ *
+ * \param id: ID reference to search the global owner for.
+ * \param[out] r_path: Path from the real ID to the initial ID.
+ * \return The ID pointer, or NULL in case of failure.
+ */
+ID *RNA_find_real_ID_and_path(Main *bmain, ID *id, const char **r_path)
+{
+  if (r_path) {
+    *r_path = "";
+  }
+
+  if ((id != NULL) && (id->flag & LIB_PRIVATE_DATA)) {
+    switch (GS(id->name)) {
+      case ID_NT:
+        if (r_path) {
+          *r_path = "node_tree";
+        }
+        return BKE_node_tree_find_owner_ID(bmain, (bNodeTree *)id);
+      case ID_GR:
+        if (r_path) {
+          *r_path = "collection";
+        }
+        return (ID *)BKE_collection_master_scene_search(bmain, (Collection *)id);
+
+      default:
+        return NULL;
+    }
+  }
+  else {
+    return id;
+  }
+}
+
+static char *rna_prepend_real_ID_path(Main *bmain, ID *id, char *path, ID **r_real_id)
+{
+  if (r_real_id != NULL) {
+    *r_real_id = NULL;
+  }
+
+  const char *prefix;
+  ID *real_id = RNA_find_real_ID_and_path(bmain, id, &prefix);
+
+  if (r_real_id != NULL) {
+    *r_real_id = real_id;
+  }
+
+  if (path != NULL) {
+    char *new_path = NULL;
+
+    if (real_id) {
+      if (prefix[0]) {
+        new_path = BLI_sprintfN("%s%s%s", prefix, path[0] == '[' ? "" : ".", path);
+      }
+      else {
+        return path;
+      }
+    }
+
+    MEM_freeN(path);
+    return new_path;
+  }
+  else {
+    return prefix[0] != '\0' ? BLI_strdup(prefix) : NULL;
+  }
+}
+
 char *RNA_path_from_ID_to_struct(PointerRNA *ptr)
 {
   char *ptrpath = NULL;
 
-  if (!ptr->id.data || !ptr->data) {
+  if (!ptr->owner_id || !ptr->data) {
     return NULL;
   }
 
@@ -5853,7 +5847,7 @@ char *RNA_path_from_ID_to_struct(PointerRNA *ptr)
       /* find the property in the struct we're nested in that references this struct, and
        * use its identifier as the first part of the path used...
        */
-      RNA_id_pointer_create(ptr->id.data, &parentptr);
+      RNA_id_pointer_create(ptr->owner_id, &parentptr);
       userprop = RNA_struct_find_nested(&parentptr, ptr->type);
 
       if (userprop) {
@@ -5875,6 +5869,13 @@ char *RNA_path_from_ID_to_struct(PointerRNA *ptr)
   return ptrpath;
 }
 
+char *RNA_path_from_real_ID_to_struct(Main *bmain, PointerRNA *ptr, struct ID **r_real)
+{
+  char *path = RNA_path_from_ID_to_struct(ptr);
+
+  return rna_prepend_real_ID_path(bmain, ptr->owner_id, path, r_real);
+}
+
 static void rna_path_array_multi_from_flat_index(const int dimsize[RNA_MAX_ARRAY_LENGTH],
                                                  const int totdims,
                                                  const int index_dim,
@@ -5929,7 +5930,7 @@ char *RNA_path_from_ID_to_property_index(PointerRNA *ptr,
   const char *propname;
   char *ptrpath, *path;
 
-  if (!ptr->id.data || !ptr->data) {
+  if (!ptr->owner_id || !ptr->data) {
     return NULL;
   }
 
@@ -5981,6 +5982,14 @@ char *RNA_path_from_ID_to_property(PointerRNA *ptr, PropertyRNA *prop)
   return RNA_path_from_ID_to_property_index(ptr, prop, 0, -1);
 }
 
+char *RNA_path_from_real_ID_to_property_index(
+    Main *bmain, PointerRNA *ptr, PropertyRNA *prop, int index_dim, int index, ID **r_real_id)
+{
+  char *path = RNA_path_from_ID_to_property_index(ptr, prop, index_dim, index);
+
+  return rna_prepend_real_ID_path(bmain, ptr->owner_id, path, r_real_id);
+}
+
 /**
  * \return the path to given ptr/prop from the closest ancestor of given type,
  * if any (else return NULL).
@@ -6000,7 +6009,7 @@ char *RNA_path_resolve_from_type_to_property(PointerRNA *ptr,
     return NULL;
   }
 
-  RNA_id_pointer_create(ptr->id.data, &idptr);
+  RNA_id_pointer_create(ptr->owner_id, &idptr);
 
   if (RNA_path_resolve_elements(&idptr, full_path, &path_elems)) {
     PropertyElemRNA *prop_elem;
@@ -6027,32 +6036,46 @@ char *RNA_path_resolve_from_type_to_property(PointerRNA *ptr,
  * Get the ID as a python representation, eg:
  *   bpy.data.foo["bar"]
  */
-char *RNA_path_full_ID_py(ID *id)
+char *RNA_path_full_ID_py(Main *bmain, ID *id)
 {
+  const char *path;
+  ID *id_real = RNA_find_real_ID_and_path(bmain, id, &path);
+
+  if (id_real) {
+    id = id_real;
+  }
+  else {
+    path = "";
+  }
+
   char id_esc[(sizeof(id->name) - 2) * 2];
 
   BLI_strescape(id_esc, id->name + 2, sizeof(id_esc));
 
-  return BLI_sprintfN("bpy.data.%s[\"%s\"]", BKE_idcode_to_name_plural(GS(id->name)), id_esc);
+  return BLI_sprintfN("bpy.data.%s[\"%s\"]%s%s",
+                      BKE_idcode_to_name_plural(GS(id->name)),
+                      id_esc,
+                      path[0] ? "." : "",
+                      path);
 }
 
 /**
  * Get the ID.struct as a python representation, eg:
  *   bpy.data.foo["bar"].some_struct
  */
-char *RNA_path_full_struct_py(struct PointerRNA *ptr)
+char *RNA_path_full_struct_py(Main *bmain, struct PointerRNA *ptr)
 {
   char *id_path;
   char *data_path;
 
   char *ret;
 
-  if (!ptr->id.data) {
+  if (!ptr->owner_id) {
     return NULL;
   }
 
   /* never fails */
-  id_path = RNA_path_full_ID_py(ptr->id.data);
+  id_path = RNA_path_full_ID_py(bmain, ptr->owner_id);
 
   data_path = RNA_path_from_ID_to_struct(ptr);
 
@@ -6072,10 +6095,8 @@ char *RNA_path_full_struct_py(struct PointerRNA *ptr)
  * Get the ID.struct.property as a python representation, eg:
  *   bpy.data.foo["bar"].some_struct.some_prop[10]
  */
-char *RNA_path_full_property_py_ex(PointerRNA *ptr,
-                                   PropertyRNA *prop,
-                                   int index,
-                                   bool use_fallback)
+char *RNA_path_full_property_py_ex(
+    Main *bmain, PointerRNA *ptr, PropertyRNA *prop, int index, bool use_fallback)
 {
   char *id_path;
   const char *data_delim;
@@ -6084,12 +6105,12 @@ char *RNA_path_full_property_py_ex(PointerRNA *ptr,
 
   char *ret;
 
-  if (!ptr->id.data) {
+  if (!ptr->owner_id) {
     return NULL;
   }
 
   /* never fails */
-  id_path = RNA_path_full_ID_py(ptr->id.data);
+  id_path = RNA_path_full_ID_py(bmain, ptr->owner_id);
 
   data_path = RNA_path_from_ID_to_property(ptr, prop);
   if (data_path) {
@@ -6122,9 +6143,9 @@ char *RNA_path_full_property_py_ex(PointerRNA *ptr,
   return ret;
 }
 
-char *RNA_path_full_property_py(PointerRNA *ptr, PropertyRNA *prop, int index)
+char *RNA_path_full_property_py(Main *bmain, PointerRNA *ptr, PropertyRNA *prop, int index)
 {
-  return RNA_path_full_property_py_ex(ptr, prop, index, false);
+  return RNA_path_full_property_py_ex(bmain, ptr, prop, index, false);
 }
 
 /**
@@ -6137,7 +6158,7 @@ char *RNA_path_struct_property_py(PointerRNA *ptr, PropertyRNA *prop, int index)
 
   char *ret;
 
-  if (!ptr->id.data) {
+  if (!ptr->owner_id) {
     return NULL;
   }
 
@@ -6722,16 +6743,16 @@ char *RNA_pointer_as_string_id(bContext *C, PointerRNA *ptr)
   return cstring;
 }
 
-static char *rna_pointer_as_string__bldata(PointerRNA *ptr)
+static char *rna_pointer_as_string__bldata(Main *bmain, PointerRNA *ptr)
 {
-  if (ptr->type == NULL || ptr->id.data == NULL) {
+  if (ptr->type == NULL || ptr->owner_id == NULL) {
     return BLI_strdup("None");
   }
   else if (RNA_struct_is_ID(ptr->type)) {
-    return RNA_path_full_ID_py(ptr->id.data);
+    return RNA_path_full_ID_py(bmain, ptr->owner_id);
   }
   else {
-    return RNA_path_full_struct_py(ptr);
+    return RNA_path_full_struct_py(bmain, ptr);
   }
 }
 
@@ -6748,7 +6769,7 @@ char *RNA_pointer_as_string(bContext *C,
     return RNA_pointer_as_string_id(C, ptr_prop);
   }
   else {
-    return rna_pointer_as_string__bldata(ptr_prop);
+    return rna_pointer_as_string__bldata(CTX_data_main(C), ptr_prop);
   }
 }
 
@@ -8115,59 +8136,6 @@ bool RNA_property_assign_default(PointerRNA *ptr, PropertyRNA *prop)
   }
 }
 
-static bool rna_property_override_operation_apply(Main *bmain,
-                                                  PointerRNA *ptr_local,
-                                                  PointerRNA *ptr_override,
-                                                  PointerRNA *ptr_storage,
-                                                  PropertyRNA *prop_local,
-                                                  PropertyRNA *prop_override,
-                                                  PropertyRNA *prop_storage,
-                                                  PointerRNA *ptr_item_local,
-                                                  PointerRNA *ptr_item_override,
-                                                  PointerRNA *ptr_item_storage,
-                                                  IDOverrideLibraryPropertyOperation *opop);
-
-bool RNA_property_copy(
-    Main *bmain, PointerRNA *ptr, PointerRNA *fromptr, PropertyRNA *prop, int index)
-{
-  if (!RNA_property_editable(ptr, prop)) {
-    return false;
-  }
-
-  PropertyRNA *prop_dst = prop;
-  PropertyRNA *prop_src = prop;
-
-  /* Ensure we get real property data,
-   * be it an actual RNA property, or an IDProperty in disguise. */
-  prop_dst = rna_ensure_property_realdata(&prop_dst, ptr);
-  prop_src = rna_ensure_property_realdata(&prop_src, fromptr);
-
-  /* IDprops: destination may not exist, if source does and is set, try to create it. */
-  /* Note: this is sort of quick hack/bandage to fix the issue,
-   * we need to rethink how IDProps are handled in 'diff' RNA code completely, imho... */
-  if (prop_src != NULL && prop_dst == NULL && RNA_property_is_set(fromptr, prop)) {
-    BLI_assert(prop_src->magic != RNA_MAGIC);
-    IDProperty *idp_dst = RNA_struct_idprops(ptr, true);
-    IDProperty *prop_idp_dst = IDP_CopyProperty((IDProperty *)prop_src);
-    IDP_AddToGroup(idp_dst, prop_idp_dst);
-    rna_idproperty_touch(prop_idp_dst);
-    /* Nothing else to do here... */
-    return true;
-  }
-
-  if (ELEM(NULL, prop_dst, prop_src)) {
-    return false;
-  }
-
-  IDOverrideLibraryPropertyOperation opop = {
-      .operation = IDOVERRIDE_LIBRARY_OP_REPLACE,
-      .subitem_reference_index = index,
-      .subitem_local_index = index,
-  };
-  return rna_property_override_operation_apply(
-      bmain, ptr, fromptr, NULL, prop_dst, prop_src, NULL, NULL, NULL, NULL, &opop);
-}
-
 /* use RNA_warning macro which includes __func__ suffix */
 void _RNA_warning(const char *format, ...)
 {
@@ -8190,865 +8158,6 @@ void _RNA_warning(const char *format, ...)
 #endif
 }
 
-static int rna_property_override_diff(Main *bmain,
-                                      PointerRNA *ptr_a,
-                                      PointerRNA *ptr_b,
-                                      PropertyRNA *prop,
-                                      PropertyRNA *prop_a,
-                                      PropertyRNA *prop_b,
-                                      const char *rna_path,
-                                      eRNACompareMode mode,
-                                      IDOverrideLibrary *override,
-                                      const int flags,
-                                      eRNAOverrideMatchResult *r_report_flags);
-
-bool RNA_property_equals(
-    Main *bmain, PointerRNA *ptr_a, PointerRNA *ptr_b, PropertyRNA *prop, eRNACompareMode mode)
-{
-  BLI_assert(ELEM(mode, RNA_EQ_STRICT, RNA_EQ_UNSET_MATCH_ANY, RNA_EQ_UNSET_MATCH_NONE));
-
-  return (rna_property_override_diff(
-              bmain, ptr_a, ptr_b, prop, NULL, NULL, NULL, mode, NULL, 0, NULL) == 0);
-}
-
-bool RNA_struct_equals(Main *bmain, PointerRNA *ptr_a, PointerRNA *ptr_b, eRNACompareMode mode)
-{
-  CollectionPropertyIterator iter;
-  PropertyRNA *iterprop;
-  bool equals = true;
-
-  if (ptr_a == NULL && ptr_b == NULL) {
-    return true;
-  }
-  else if (ptr_a == NULL || ptr_b == NULL) {
-    return false;
-  }
-  else if (ptr_a->type != ptr_b->type) {
-    return false;
-  }
-
-  iterprop = RNA_struct_iterator_property(ptr_a->type);
-
-  RNA_property_collection_begin(ptr_a, iterprop, &iter);
-  for (; iter.valid; RNA_property_collection_next(&iter)) {
-    PropertyRNA *prop = iter.ptr.data;
-
-    if (!RNA_property_equals(bmain, ptr_a, ptr_b, prop, mode)) {
-      equals = false;
-      break;
-    }
-  }
-  RNA_property_collection_end(&iter);
-
-  return equals;
-}
-
-/* Low-level functions, also used by non-override RNA API like copy or equality check. */
-
-/** Generic RNA property diff function.
- *
- * \note about \a prop and \a prop_a/prop_b parameters:
- * the former is expected to be an 'un-resolved' one,
- * while the two later are expected to be fully resolved ones
- * (i.e. to be the IDProps when they should be, etc.).
- * When \a prop is given, \a prop_a and \a prop_b should always be NULL, and vice-versa.
- * This is necessary, because we cannot perform 'set/unset' checks on resolved properties
- * (unset IDProps would merely be NULL then).
- *
- * \note When there is no equality,
- * but we cannot determine an order (greater than/lesser than), we return 1.
- */
-static int rna_property_override_diff(Main *bmain,
-                                      PointerRNA *ptr_a,
-                                      PointerRNA *ptr_b,
-                                      PropertyRNA *prop,
-                                      PropertyRNA *prop_a,
-                                      PropertyRNA *prop_b,
-                                      const char *rna_path,
-                                      eRNACompareMode mode,
-                                      IDOverrideLibrary *override,
-                                      const int flags,
-                                      eRNAOverrideMatchResult *r_report_flags)
-{
-  if (prop != NULL) {
-    BLI_assert(prop_a == NULL && prop_b == NULL);
-    prop_a = prop;
-    prop_b = prop;
-  }
-
-  if (ELEM(NULL, prop_a, prop_b)) {
-    return (prop_a == prop_b) ? 0 : 1;
-  }
-
-  if (!RNA_property_comparable(ptr_a, prop_a) || !RNA_property_comparable(ptr_b, prop_b)) {
-    return 0;
-  }
-
-  if (mode == RNA_EQ_UNSET_MATCH_ANY) {
-    /* uninitialized properties are assumed to match anything */
-    if (!RNA_property_is_set(ptr_a, prop_a) || !RNA_property_is_set(ptr_b, prop_b)) {
-      return 0;
-    }
-  }
-  else if (mode == RNA_EQ_UNSET_MATCH_NONE) {
-    /* unset properties never match set properties */
-    if (RNA_property_is_set(ptr_a, prop_a) != RNA_property_is_set(ptr_b, prop_b)) {
-      return 1;
-    }
-  }
-
-  if (prop != NULL) {
-    /* Ensure we get real property data, be it an actual RNA property,
-     * or an IDProperty in disguise. */
-    prop_a = rna_ensure_property_realdata(&prop_a, ptr_a);
-    prop_b = rna_ensure_property_realdata(&prop_b, ptr_b);
-
-    if (ELEM(NULL, prop_a, prop_b)) {
-      return (prop_a == prop_b) ? 0 : 1;
-    }
-  }
-
-  /* Check if we are working with arrays. */
-  const bool is_array_a = RNA_property_array_check(prop_a);
-  const bool is_array_b = RNA_property_array_check(prop_b);
-
-  if (is_array_a != is_array_b) {
-    /* Should probably never happen actually... */
-    BLI_assert(0);
-    return is_array_a ? 1 : -1;
-  }
-
-  /* Get the length of the array to work with. */
-  const int len_a = RNA_property_array_length(ptr_a, prop_a);
-  const int len_b = RNA_property_array_length(ptr_b, prop_b);
-
-  if (len_a != len_b) {
-    /* Do not handle override in that case,
-     * we do not support insertion/deletion from arrays for now. */
-    return len_a > len_b ? 1 : -1;
-  }
-
-  if (is_array_a && len_a == 0) {
-    /* Empty arrays, will happen in some case with dynamic ones. */
-    return 0;
-  }
-
-  RNAPropOverrideDiff override_diff = NULL;
-  /* Special case for IDProps, we use default callback then. */
-  if (prop_a->magic != RNA_MAGIC) {
-    override_diff = rna_property_override_diff_default;
-    if (prop_b->magic == RNA_MAGIC && prop_b->override_diff != override_diff) {
-      override_diff = NULL;
-    }
-  }
-  else if (prop_b->magic != RNA_MAGIC) {
-    override_diff = rna_property_override_diff_default;
-    if (prop_a->override_diff != override_diff) {
-      override_diff = NULL;
-    }
-  }
-  else if (prop_a->override_diff == prop_b->override_diff) {
-    override_diff = prop_a->override_diff;
-  }
-
-  if (override_diff == NULL) {
-#ifndef NDEBUG
-    printf("'%s' gives unmatching or NULL RNA diff callbacks, should not happen (%d vs. %d).\n",
-           rna_path ?
-               rna_path :
-               (prop_a->magic != RNA_MAGIC ? ((IDProperty *)prop_a)->name : prop_a->identifier),
-           prop_a->magic == RNA_MAGIC,
-           prop_b->magic == RNA_MAGIC);
-#endif
-    BLI_assert(0);
-    return 1;
-  }
-
-  bool override_changed = false;
-  int diff_flags = flags;
-  if (!RNA_property_overridable_get(ptr_a, prop_a)) {
-    diff_flags &= ~RNA_OVERRIDE_COMPARE_CREATE;
-  }
-  const int diff = override_diff(bmain,
-                                 ptr_a,
-                                 ptr_b,
-                                 prop_a,
-                                 prop_b,
-                                 len_a,
-                                 len_b,
-                                 mode,
-                                 override,
-                                 rna_path,
-                                 diff_flags,
-                                 &override_changed);
-  if (override_changed && r_report_flags) {
-    *r_report_flags |= RNA_OVERRIDE_MATCH_RESULT_CREATED;
-  }
-
-  return diff;
-}
-
-/* Modify local data-block to make it ready for override application
- * (only needed for diff operations, where we use
- * the local data-block's data as second operand). */
-static bool rna_property_override_operation_store(Main *bmain,
-                                                  PointerRNA *ptr_local,
-                                                  PointerRNA *ptr_reference,
-                                                  PointerRNA *ptr_storage,
-                                                  PropertyRNA *prop_local,
-                                                  PropertyRNA *prop_reference,
-                                                  PropertyRNA *prop_storage,
-                                                  IDOverrideLibraryProperty *op)
-{
-  int len_local, len_reference, len_storage = 0;
-  bool changed = false;
-
-  if (ptr_storage == NULL) {
-    return changed;
-  }
-
-  /* get the length of the array to work with */
-  len_local = RNA_property_array_length(ptr_local, prop_local);
-  len_reference = RNA_property_array_length(ptr_reference, prop_reference);
-  if (prop_storage) {
-    len_storage = RNA_property_array_length(ptr_storage, prop_storage);
-  }
-
-  if (len_local != len_reference || len_local != len_storage) {
-    /* Do not handle override in that case,
-     * we do not support insertion/deletion from arrays for now. */
-    return changed;
-  }
-
-  BLI_assert(prop_local->override_store == prop_reference->override_store &&
-             (!ptr_storage || prop_local->override_store == prop_storage->override_store) &&
-             prop_local->override_store != NULL);
-
-  for (IDOverrideLibraryPropertyOperation *opop = op->operations.first; opop; opop = opop->next) {
-    /* Only needed for diff operations. */
-    if (!ELEM(opop->operation,
-              IDOVERRIDE_LIBRARY_OP_ADD,
-              IDOVERRIDE_LIBRARY_OP_SUBTRACT,
-              IDOVERRIDE_LIBRARY_OP_MULTIPLY)) {
-      continue;
-    }
-
-    if (prop_local->override_store(bmain,
-                                   ptr_local,
-                                   ptr_reference,
-                                   ptr_storage,
-                                   prop_local,
-                                   prop_reference,
-                                   prop_storage,
-                                   len_local,
-                                   len_reference,
-                                   len_storage,
-                                   opop)) {
-      changed = true;
-    }
-  }
-
-  return changed;
-}
-
-static bool rna_property_override_operation_apply(Main *bmain,
-                                                  PointerRNA *ptr_local,
-                                                  PointerRNA *ptr_override,
-                                                  PointerRNA *ptr_storage,
-                                                  PropertyRNA *prop_local,
-                                                  PropertyRNA *prop_override,
-                                                  PropertyRNA *prop_storage,
-                                                  PointerRNA *ptr_item_local,
-                                                  PointerRNA *ptr_item_override,
-                                                  PointerRNA *ptr_item_storage,
-                                                  IDOverrideLibraryPropertyOperation *opop)
-{
-  int len_local, len_reference, len_storage = 0;
-
-  const short override_op = opop->operation;
-
-  if (override_op == IDOVERRIDE_LIBRARY_OP_NOOP) {
-    return true;
-  }
-
-  if (ELEM(override_op,
-           IDOVERRIDE_LIBRARY_OP_ADD,
-           IDOVERRIDE_LIBRARY_OP_SUBTRACT,
-           IDOVERRIDE_LIBRARY_OP_MULTIPLY) &&
-      !ptr_storage) {
-    /* We cannot apply 'diff' override operations without some reference storage.
-     * This should typically only happen at read time of .blend file... */
-    return false;
-  }
-
-  if (ELEM(override_op,
-           IDOVERRIDE_LIBRARY_OP_ADD,
-           IDOVERRIDE_LIBRARY_OP_SUBTRACT,
-           IDOVERRIDE_LIBRARY_OP_MULTIPLY) &&
-      !prop_storage) {
-    /* We cannot apply 'diff' override operations without some reference storage.
-     * This should typically only happen at read time of .blend file... */
-    return false;
-  }
-
-  RNAPropOverrideApply override_apply = NULL;
-  /* Special case for IDProps, we use default callback then. */
-  if (prop_local->magic != RNA_MAGIC) {
-    override_apply = rna_property_override_apply_default;
-    if (prop_override->magic == RNA_MAGIC && prop_override->override_apply != override_apply) {
-      override_apply = NULL;
-    }
-  }
-  else if (prop_override->magic != RNA_MAGIC) {
-    override_apply = rna_property_override_apply_default;
-    if (prop_local->override_apply != override_apply) {
-      override_apply = NULL;
-    }
-  }
-  else if (prop_local->override_apply == prop_override->override_apply) {
-    override_apply = prop_local->override_apply;
-  }
-
-  if (ptr_storage && prop_storage->magic == RNA_MAGIC &&
-      prop_storage->override_apply != override_apply) {
-    override_apply = NULL;
-  }
-
-  if (override_apply == NULL) {
-#ifndef NDEBUG
-    printf("'%s' gives unmatching or NULL RNA copy callbacks, should not happen (%d vs. %d).\n",
-           prop_local->magic != RNA_MAGIC ? ((IDProperty *)prop_local)->name :
-                                            prop_local->identifier,
-           prop_local->magic == RNA_MAGIC,
-           prop_override->magic == RNA_MAGIC);
-#endif
-    BLI_assert(0);
-    return false;
-  }
-
-  /* get the length of the array to work with */
-  len_local = RNA_property_array_length(ptr_local, prop_local);
-  len_reference = RNA_property_array_length(ptr_override, prop_override);
-  if (ptr_storage) {
-    len_storage = RNA_property_array_length(ptr_storage, prop_storage);
-  }
-
-  if (len_local != len_reference || (ptr_storage && len_local != len_storage)) {
-    /* Do not handle override in that case,
-     * we do not support insertion/deletion from arrays for now. */
-    return false;
-  }
-
-  /* get and set the default values as appropriate for the various types */
-  return override_apply(bmain,
-                        ptr_local,
-                        ptr_override,
-                        ptr_storage,
-                        prop_local,
-                        prop_override,
-                        prop_storage,
-                        len_local,
-                        len_reference,
-                        len_storage,
-                        ptr_item_local,
-                        ptr_item_override,
-                        ptr_item_storage,
-                        opop);
-}
-
-/**
- * Check whether reference and local overridden data match (are the same),
- * with respect to given restrictive sets of properties.
- * If requested, will generate needed new property overrides, and/or restore values from reference.
- *
- * \param r_report_flags: If given,
- * will be set with flags matching actions taken by the function on \a ptr_local.
- *
- * \return True if _resulting_ \a ptr_local does match \a ptr_reference.
- */
-bool RNA_struct_override_matches(Main *bmain,
-                                 PointerRNA *ptr_local,
-                                 PointerRNA *ptr_reference,
-                                 const char *root_path,
-                                 IDOverrideLibrary *override,
-                                 const eRNAOverrideMatch flags,
-                                 eRNAOverrideMatchResult *r_report_flags)
-{
-  CollectionPropertyIterator iter;
-  PropertyRNA *iterprop;
-  bool matching = true;
-
-  BLI_assert(ptr_local->type == ptr_reference->type);
-  BLI_assert(ptr_local->id.data && ptr_reference->id.data);
-
-  const bool ignore_non_overridable = (flags & RNA_OVERRIDE_COMPARE_IGNORE_NON_OVERRIDABLE) != 0;
-  const bool ignore_overridden = (flags & RNA_OVERRIDE_COMPARE_IGNORE_OVERRIDDEN) != 0;
-  const bool do_create = (flags & RNA_OVERRIDE_COMPARE_CREATE) != 0;
-  const bool do_restore = (flags & RNA_OVERRIDE_COMPARE_RESTORE) != 0;
-
-//#define DEBUG_OVERRIDE_TIMEIT
-#ifdef DEBUG_OVERRIDE_TIMEIT
-  static float _sum_time_global = 0.0f;
-  static float _num_time_global = 0.0f;
-  double _timeit_time_global;
-  static float _sum_time_diffing = 0.0f;
-  static float _delta_time_diffing = 0.0f;
-  static int _num_delta_time_diffing = 0.0f;
-  static float _num_time_diffing = 0.0f;
-  double _timeit_time_diffing;
-
-  if (!root_path) {
-    _delta_time_diffing = 0.0f;
-    _num_delta_time_diffing = 0;
-    _timeit_time_global = PIL_check_seconds_timer();
-  }
-#endif
-
-  iterprop = RNA_struct_iterator_property(ptr_local->type);
-
-  for (RNA_property_collection_begin(ptr_local, iterprop, &iter); iter.valid;
-       RNA_property_collection_next(&iter)) {
-    PropertyRNA *prop_local = iter.ptr.data;
-    PropertyRNA *prop_reference = iter.ptr.data;
-
-    /* Ensure we get real property data, be it an actual RNA property,
-     * or an IDProperty in disguise. */
-    prop_local = rna_ensure_property_realdata(&prop_local, ptr_local);
-    prop_reference = rna_ensure_property_realdata(&prop_reference, ptr_reference);
-
-    if (ELEM(NULL, prop_local, prop_reference)) {
-      continue;
-    }
-
-    if (ignore_non_overridable && !RNA_property_overridable_get(ptr_local, prop_local)) {
-      continue;
-    }
-
-#if 0 /* This actually makes things slower, since it has to check for animation paths etc! */
-    if (RNA_property_animated(ptr_local, prop_local)) {
-      /* We cannot do anything here really, animation is some kind of dynamic overrides that has
-       * precedence over static one... */
-      continue;
-    }
-#endif
-
-#define RNA_PATH_BUFFSIZE 8192
-#define RNA_PATH_PRINTF(_str, ...) \
-  if (BLI_snprintf(rna_path, RNA_PATH_BUFFSIZE, (_str), __VA_ARGS__) >= RNA_PATH_BUFFSIZE) { \
-    rna_path = BLI_sprintfN((_str), __VA_ARGS__); \
-  } \
-  (void)0
-#define RNA_PATH_FREE \
-  if (rna_path != rna_path_buffer) \
-  MEM_freeN(rna_path)
-
-    char rna_path_buffer[RNA_PATH_BUFFSIZE];
-    char *rna_path = rna_path_buffer;
-
-    /* XXX TODO this will have to be refined to handle collections insertions, and array items */
-    if (root_path) {
-      /* Inlined building, much much more efficient. */
-      if (prop_local->magic == RNA_MAGIC) {
-        RNA_PATH_PRINTF("%s.%s", root_path, RNA_property_identifier(prop_local));
-      }
-      else {
-        RNA_PATH_PRINTF("%s[\"%s\"]", root_path, RNA_property_identifier(prop_local));
-      }
-    }
-    else {
-      /* This is rather slow, but is not much called, so not really worth optimizing. */
-      rna_path = RNA_path_from_ID_to_property(ptr_local, prop_local);
-    }
-    if (rna_path == NULL) {
-      continue;
-    }
-
-    //      printf("Override Checking %s\n", rna_path);
-
-    if (ignore_overridden && BKE_override_library_property_find(override, rna_path) != NULL) {
-      RNA_PATH_FREE;
-      continue;
-    }
-
-#ifdef DEBUG_OVERRIDE_TIMEIT
-    if (!root_path) {
-      _timeit_time_diffing = PIL_check_seconds_timer();
-    }
-#endif
-
-    eRNAOverrideMatchResult report_flags = 0;
-    const int diff = rna_property_override_diff(bmain,
-                                                ptr_local,
-                                                ptr_reference,
-                                                NULL,
-                                                prop_local,
-                                                prop_reference,
-                                                rna_path,
-                                                RNA_EQ_STRICT,
-                                                override,
-                                                flags,
-                                                &report_flags);
-
-#ifdef DEBUG_OVERRIDE_TIMEIT
-    if (!root_path) {
-      const float _delta_time = (float)(PIL_check_seconds_timer() - _timeit_time_diffing);
-      _delta_time_diffing += _delta_time;
-      _num_delta_time_diffing++;
-    }
-#endif
-
-    matching = matching && diff == 0;
-    if (r_report_flags) {
-      *r_report_flags |= report_flags;
-    }
-
-    if (diff != 0) {
-      /* XXX TODO: refine this for per-item overriding of arrays... */
-      IDOverrideLibraryProperty *op = BKE_override_library_property_find(override, rna_path);
-      IDOverrideLibraryPropertyOperation *opop = op ? op->operations.first : NULL;
-
-      if (do_restore && (report_flags & RNA_OVERRIDE_MATCH_RESULT_CREATED) == 0) {
-        /* We are allowed to restore to reference's values. */
-        if (ELEM(NULL, op, opop) || opop->operation == IDOVERRIDE_LIBRARY_OP_NOOP) {
-          /* We should restore that property to its reference value */
-          if (RNA_property_editable(ptr_local, prop_local)) {
-            IDOverrideLibraryPropertyOperation opop_tmp = {
-                .operation = IDOVERRIDE_LIBRARY_OP_REPLACE,
-                .subitem_reference_index = -1,
-                .subitem_local_index = -1,
-            };
-            rna_property_override_operation_apply(bmain,
-                                                  ptr_local,
-                                                  ptr_reference,
-                                                  NULL,
-                                                  prop_local,
-                                                  prop_reference,
-                                                  NULL,
-                                                  NULL,
-                                                  NULL,
-                                                  NULL,
-                                                  &opop_tmp);
-            if (r_report_flags) {
-              *r_report_flags |= RNA_OVERRIDE_MATCH_RESULT_RESTORED;
-            }
-          }
-          else {
-            /* Too noisy for now, this triggers on runtime props like transform matrices etc. */
-#if 0
-            BLI_assert(!"We have differences between reference and "
-                       "overriding data on non-editable property.");
-#endif
-            matching = false;
-          }
-        }
-      }
-      else if ((report_flags & RNA_OVERRIDE_MATCH_RESULT_CREATED) == 0 && ELEM(NULL, op, opop)) {
-        /* This property is not overridden, and differs from reference, so we have no match. */
-        matching = false;
-        if (!(do_create || do_restore)) {
-          /* Since we have no 'changing' action allowed, we can break here. */
-          MEM_SAFE_FREE(rna_path);
-          break;
-        }
-      }
-    }
-
-    RNA_PATH_FREE;
-
-#undef RNA_PATH_BUFFSIZE
-#undef RNA_PATH_PRINTF
-#undef RNA_PATH_FREE
-  }
-  RNA_property_collection_end(&iter);
-
-#ifdef DEBUG_OVERRIDE_TIMEIT
-  if (!root_path) {
-    const float _delta_time = (float)(PIL_check_seconds_timer() - _timeit_time_global);
-    _sum_time_global += _delta_time;
-    _num_time_global++;
-    _sum_time_diffing += _delta_time_diffing;
-    _num_time_diffing++;
-    printf("ID: %s\n", ((ID *)ptr_local->id.data)->name);
-    printf("time end      (%s): %.6f\n", __func__, _delta_time);
-    printf("time averaged (%s): %.6f (total: %.6f, in %d runs)\n",
-           __func__,
-           (_sum_time_global / _num_time_global),
-           _sum_time_global,
-           (int)_num_time_global);
-    printf("diffing time end      (%s): %.6f (in %d runs)\n",
-           __func__,
-           _delta_time_diffing,
-           _num_delta_time_diffing);
-    printf("diffing time averaged (%s): %.6f (total: %.6f, in %d runs)\n",
-           __func__,
-           (_sum_time_diffing / _num_time_diffing),
-           _sum_time_diffing,
-           (int)_num_time_diffing);
-  }
-#endif
-
-  return matching;
-}
-
-/** Store needed second operands into \a storage data-block
- * for differential override operations. */
-bool RNA_struct_override_store(Main *bmain,
-                               PointerRNA *ptr_local,
-                               PointerRNA *ptr_reference,
-                               PointerRNA *ptr_storage,
-                               IDOverrideLibrary *override)
-{
-  bool changed = false;
-
-#ifdef DEBUG_OVERRIDE_TIMEIT
-  TIMEIT_START_AVERAGED(RNA_struct_override_store);
-#endif
-  for (IDOverrideLibraryProperty *op = override->properties.first; op; op = op->next) {
-    /* Simplified for now! */
-    PointerRNA data_reference, data_local;
-    PropertyRNA *prop_reference, *prop_local;
-
-    if (RNA_path_resolve_property(ptr_local, op->rna_path, &data_local, &prop_local) &&
-        RNA_path_resolve_property(ptr_reference, op->rna_path, &data_reference, &prop_reference)) {
-      PointerRNA data_storage;
-      PropertyRNA *prop_storage = NULL;
-
-      /* It is totally OK if this does not success,
-       * only a subset of override operations actually need storage. */
-      if (ptr_storage && (ptr_storage->id.data != NULL)) {
-        RNA_path_resolve_property(ptr_storage, op->rna_path, &data_storage, &prop_storage);
-      }
-
-      if (rna_property_override_operation_store(bmain,
-                                                &data_local,
-                                                &data_reference,
-                                                &data_storage,
-                                                prop_reference,
-                                                prop_local,
-                                                prop_storage,
-                                                op)) {
-        changed = true;
-      }
-    }
-  }
-#ifdef DEBUG_OVERRIDE_TIMEIT
-  TIMEIT_END_AVERAGED(RNA_struct_override_store);
-#endif
-
-  return changed;
-}
-
-static void rna_property_override_apply_ex(Main *bmain,
-                                           PointerRNA *ptr_local,
-                                           PointerRNA *ptr_override,
-                                           PointerRNA *ptr_storage,
-                                           PropertyRNA *prop_local,
-                                           PropertyRNA *prop_override,
-                                           PropertyRNA *prop_storage,
-                                           PointerRNA *ptr_item_local,
-                                           PointerRNA *ptr_item_override,
-                                           PointerRNA *ptr_item_storage,
-                                           IDOverrideLibraryProperty *op,
-                                           const bool do_insert)
-{
-  for (IDOverrideLibraryPropertyOperation *opop = op->operations.first; opop; opop = opop->next) {
-    if (!do_insert != !ELEM(opop->operation,
-                            IDOVERRIDE_LIBRARY_OP_INSERT_AFTER,
-                            IDOVERRIDE_LIBRARY_OP_INSERT_BEFORE)) {
-      if (!do_insert) {
-        printf("Skipping insert override operations in first pass (%s)!\n", op->rna_path);
-      }
-      continue;
-    }
-    if (!rna_property_override_operation_apply(bmain,
-                                               ptr_local,
-                                               ptr_override,
-                                               ptr_storage,
-                                               prop_local,
-                                               prop_override,
-                                               prop_storage,
-                                               ptr_item_local,
-                                               ptr_item_override,
-                                               ptr_item_storage,
-                                               opop)) {
-      /* TODO No assert here, would be much much better to just report as warning,
-       * failing override applications will probably be fairly common! */
-      BLI_assert(0);
-    }
-  }
-}
-
-/** Apply given \a override operations on \a ptr_local, using \a ptr_override
- * (and \a ptr_storage form differential ops) as source. */
-void RNA_struct_override_apply(Main *bmain,
-                               PointerRNA *ptr_local,
-                               PointerRNA *ptr_override,
-                               PointerRNA *ptr_storage,
-                               IDOverrideLibrary *override)
-{
-#ifdef DEBUG_OVERRIDE_TIMEIT
-  TIMEIT_START_AVERAGED(RNA_struct_override_apply);
-#endif
-  /* Note: Applying insert operations in a separate pass is mandatory.
-   * We could optimize this later, but for now, as inefficient as it is,
-   * don't think this is a critical point.
-   */
-  bool do_insert = false;
-  for (int i = 0; i < 2; i++, do_insert = true) {
-    for (IDOverrideLibraryProperty *op = override->properties.first; op; op = op->next) {
-      /* Simplified for now! */
-      PointerRNA data_override, data_local;
-      PointerRNA data_item_override, data_item_local;
-      PropertyRNA *prop_override, *prop_local;
-
-      if (RNA_path_resolve_property_and_item_pointer(
-              ptr_local, op->rna_path, &data_local, &prop_local, &data_item_local) &&
-          RNA_path_resolve_property_and_item_pointer(
-              ptr_override, op->rna_path, &data_override, &prop_override, &data_item_override)) {
-        PointerRNA data_storage, data_item_storage;
-        PropertyRNA *prop_storage = NULL;
-
-        /* It is totally OK if this does not success,
-         * only a subset of override operations actually need storage. */
-        if (ptr_storage && (ptr_storage->id.data != NULL)) {
-          RNA_path_resolve_property_and_item_pointer(
-              ptr_storage, op->rna_path, &data_storage, &prop_storage, &data_item_storage);
-        }
-
-        rna_property_override_apply_ex(bmain,
-                                       &data_local,
-                                       &data_override,
-                                       prop_storage ? &data_storage : NULL,
-                                       prop_local,
-                                       prop_override,
-                                       prop_storage,
-                                       &data_item_local,
-                                       &data_item_override,
-                                       prop_storage ? &data_item_storage : NULL,
-                                       op,
-                                       do_insert);
-      }
-#ifndef NDEBUG
-      else {
-        printf(
-            "Failed to apply library override operation to '%s.%s' "
-            "(could not resolve some properties, local:  %d, override: %d)\n",
-            ((ID *)ptr_override->id.data)->name,
-            op->rna_path,
-            RNA_path_resolve_property(ptr_local, op->rna_path, &data_local, &prop_local),
-            RNA_path_resolve_property(ptr_override, op->rna_path, &data_override, &prop_override));
-      }
-#endif
-    }
-  }
-#ifdef DEBUG_OVERRIDE_TIMEIT
-  TIMEIT_END_AVERAGED(RNA_struct_override_apply);
-#endif
-}
-
-IDOverrideLibraryProperty *RNA_property_override_property_find(PointerRNA *ptr, PropertyRNA *prop)
-{
-  ID *id = ptr->id.data;
-
-  if (!id || !id->override_library) {
-    return NULL;
-  }
-
-  char *rna_path = RNA_path_from_ID_to_property(ptr, prop);
-  if (rna_path) {
-    IDOverrideLibraryProperty *op = BKE_override_library_property_find(id->override_library,
-                                                                       rna_path);
-    MEM_freeN(rna_path);
-    return op;
-  }
-  return NULL;
-}
-
-IDOverrideLibraryProperty *RNA_property_override_property_get(PointerRNA *ptr,
-                                                              PropertyRNA *prop,
-                                                              bool *r_created)
-{
-  ID *id = ptr->id.data;
-
-  if (!id || !id->override_library) {
-    return NULL;
-  }
-
-  char *rna_path = RNA_path_from_ID_to_property(ptr, prop);
-  if (rna_path) {
-    IDOverrideLibraryProperty *op = BKE_override_library_property_get(
-        id->override_library, rna_path, r_created);
-    MEM_freeN(rna_path);
-    return op;
-  }
-  return NULL;
-}
-
-IDOverrideLibraryPropertyOperation *RNA_property_override_property_operation_find(
-    PointerRNA *ptr, PropertyRNA *prop, const int index, const bool strict, bool *r_strict)
-{
-  IDOverrideLibraryProperty *op = RNA_property_override_property_find(ptr, prop);
-
-  if (!op) {
-    return NULL;
-  }
-
-  return BKE_override_library_property_operation_find(
-      op, NULL, NULL, index, index, strict, r_strict);
-}
-
-IDOverrideLibraryPropertyOperation *RNA_property_override_property_operation_get(
-    PointerRNA *ptr,
-    PropertyRNA *prop,
-    const short operation,
-    const int index,
-    const bool strict,
-    bool *r_strict,
-    bool *r_created)
-{
-  IDOverrideLibraryProperty *op = RNA_property_override_property_get(ptr, prop, NULL);
-
-  if (!op) {
-    return NULL;
-  }
-
-  return BKE_override_library_property_operation_get(
-      op, operation, NULL, NULL, index, index, strict, r_strict, r_created);
-}
-
-eRNAOverrideStatus RNA_property_override_library_status(PointerRNA *ptr,
-                                                        PropertyRNA *prop,
-                                                        const int index)
-{
-  int override_status = 0;
-
-  if (!BKE_override_library_is_enabled()) {
-    return override_status;
-  }
-
-  if (!ptr || !prop || !ptr->id.data || !((ID *)ptr->id.data)->override_library) {
-    return override_status;
-  }
-
-  if (RNA_property_overridable_get(ptr, prop) && RNA_property_editable_flag(ptr, prop)) {
-    override_status |= RNA_OVERRIDE_STATUS_OVERRIDABLE;
-  }
-
-  IDOverrideLibraryPropertyOperation *opop = RNA_property_override_property_operation_find(
-      ptr, prop, index, false, NULL);
-  if (opop != NULL) {
-    override_status |= RNA_OVERRIDE_STATUS_OVERRIDDEN;
-    if (opop->flag & IDOVERRIDE_LIBRARY_FLAG_MANDATORY) {
-      override_status |= RNA_OVERRIDE_STATUS_MANDATORY;
-    }
-    if (opop->flag & IDOVERRIDE_LIBRARY_FLAG_LOCKED) {
-      override_status |= RNA_OVERRIDE_STATUS_LOCKED;
-    }
-  }
-
-  return override_status;
-}
-
 bool RNA_path_resolved_create(PointerRNA *ptr,
                               struct PropertyRNA *prop,
                               const int prop_index,
diff --git a/source/blender/makesrna/intern/rna_access_compare_override.c b/source/blender/makesrna/intern/rna_access_compare_override.c
new file mode 100644
index 00000000000..b0a83ea38c6
--- /dev/null
+++ b/source/blender/makesrna/intern/rna_access_compare_override.c
@@ -0,0 +1,1099 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+/** \file
+ * \ingroup RNA
+ */
+
+#include "MEM_guardedalloc.h"
+
+#include "DNA_ID.h"
+#include "DNA_constraint_types.h"
+#include "DNA_modifier_types.h"
+
+#include "BLI_utildefines.h"
+#include "BLI_string.h"
+
+#ifdef DEBUG_OVERRIDE_TIMEIT
+#  include "PIL_time_utildefines.h"
+#endif
+
+#include "BKE_idprop.h"
+#include "BKE_library_override.h"
+#include "BKE_main.h"
+
+#include "RNA_access.h"
+#include "RNA_define.h"
+#include "RNA_enum_types.h"
+
+#include "rna_internal.h"
+#include "rna_access_internal.h"
+
+int RNA_property_override_flag(PropertyRNA *prop)
+{
+  return rna_ensure_property(prop)->flag_override;
+}
+
+/** \note Does not take into account editable status, this has to be checked separately
+ * (using #RNA_property_editable_flag() usually). */
+bool RNA_property_overridable_get(PointerRNA *ptr, PropertyRNA *prop)
+{
+  if (prop->magic == RNA_MAGIC) {
+    /* Special handling for insertions of constraints or modifiers... */
+    /* TODO Note We may want to add a more generic system to RNA
+     * (like a special property in struct of items)
+     * if we get more overrideable collections,
+     * for now we can live with those special-cases handling I think. */
+    if (RNA_struct_is_a(ptr->type, &RNA_Constraint)) {
+      bConstraint *con = ptr->data;
+      if (con->flag & CONSTRAINT_OVERRIDE_LIBRARY_LOCAL) {
+        return true;
+      }
+    }
+    else if (RNA_struct_is_a(ptr->type, &RNA_Modifier)) {
+      ModifierData *mod = ptr->data;
+      if (mod->flag & eModifierFlag_OverrideLibrary_Local) {
+        return true;
+      }
+    }
+    /* If this is a RNA-defined property (real or 'virtual' IDProp),
+     * we want to use RNA prop flag. */
+    return !(prop->flag_override & PROPOVERRIDE_NO_COMPARISON) &&
+           (prop->flag_override & PROPOVERRIDE_OVERRIDABLE_LIBRARY);
+  }
+  else {
+    /* If this is a real 'pure' IDProp (aka custom property), we want to use the IDProp flag. */
+    return !(prop->flag_override & PROPOVERRIDE_NO_COMPARISON) &&
+           (((IDProperty *)prop)->flag & IDP_FLAG_OVERRIDABLE_LIBRARY);
+  }
+}
+
+/* Should only be used for custom properties */
+bool RNA_property_overridable_library_set(PointerRNA *UNUSED(ptr),
+                                          PropertyRNA *prop,
+                                          const bool is_overridable)
+{
+  /* Only works for pure custom properties IDProps. */
+  if (prop->magic != RNA_MAGIC) {
+    IDProperty *idprop = (IDProperty *)prop;
+
+    idprop->flag = is_overridable ? (idprop->flag | IDP_FLAG_OVERRIDABLE_LIBRARY) :
+                                    (idprop->flag & ~IDP_FLAG_OVERRIDABLE_LIBRARY);
+    return true;
+  }
+
+  return false;
+}
+
+bool RNA_property_overridden(PointerRNA *ptr, PropertyRNA *prop)
+{
+  char *rna_path = RNA_path_from_ID_to_property(ptr, prop);
+  ID *id = ptr->owner_id;
+
+  if (rna_path == NULL || id == NULL || id->override_library == NULL) {
+    return false;
+  }
+
+  return (BKE_override_library_property_find(id->override_library, rna_path) != NULL);
+}
+
+bool RNA_property_comparable(PointerRNA *UNUSED(ptr), PropertyRNA *prop)
+{
+  prop = rna_ensure_property(prop);
+
+  return !(prop->flag_override & PROPOVERRIDE_NO_COMPARISON);
+}
+
+static bool rna_property_override_operation_apply(Main *bmain,
+                                                  PointerRNA *ptr_local,
+                                                  PointerRNA *ptr_override,
+                                                  PointerRNA *ptr_storage,
+                                                  PropertyRNA *prop_local,
+                                                  PropertyRNA *prop_override,
+                                                  PropertyRNA *prop_storage,
+                                                  PointerRNA *ptr_item_local,
+                                                  PointerRNA *ptr_item_override,
+                                                  PointerRNA *ptr_item_storage,
+                                                  IDOverrideLibraryPropertyOperation *opop);
+
+bool RNA_property_copy(
+    Main *bmain, PointerRNA *ptr, PointerRNA *fromptr, PropertyRNA *prop, int index)
+{
+  if (!RNA_property_editable(ptr, prop)) {
+    return false;
+  }
+
+  PropertyRNA *prop_dst = prop;
+  PropertyRNA *prop_src = prop;
+
+  /* Ensure we get real property data,
+   * be it an actual RNA property, or an IDProperty in disguise. */
+  prop_dst = rna_ensure_property_realdata(&prop_dst, ptr);
+  prop_src = rna_ensure_property_realdata(&prop_src, fromptr);
+
+  /* IDprops: destination may not exist, if source does and is set, try to create it. */
+  /* Note: this is sort of quick hack/bandage to fix the issue,
+   * we need to rethink how IDProps are handled in 'diff' RNA code completely, imho... */
+  if (prop_src != NULL && prop_dst == NULL && RNA_property_is_set(fromptr, prop)) {
+    BLI_assert(prop_src->magic != RNA_MAGIC);
+    IDProperty *idp_dst = RNA_struct_idprops(ptr, true);
+    IDProperty *prop_idp_dst = IDP_CopyProperty((IDProperty *)prop_src);
+    IDP_AddToGroup(idp_dst, prop_idp_dst);
+    rna_idproperty_touch(prop_idp_dst);
+    /* Nothing else to do here... */
+    return true;
+  }
+
+  if (ELEM(NULL, prop_dst, prop_src)) {
+    return false;
+  }
+
+  IDOverrideLibraryPropertyOperation opop = {
+      .operation = IDOVERRIDE_LIBRARY_OP_REPLACE,
+      .subitem_reference_index = index,
+      .subitem_local_index = index,
+  };
+  return rna_property_override_operation_apply(
+      bmain, ptr, fromptr, NULL, prop_dst, prop_src, NULL, NULL, NULL, NULL, &opop);
+}
+
+static int rna_property_override_diff(Main *bmain,
+                                      PointerRNA *ptr_a,
+                                      PointerRNA *ptr_b,
+                                      PropertyRNA *prop,
+                                      PropertyRNA *prop_a,
+                                      PropertyRNA *prop_b,
+                                      const char *rna_path,
+                                      eRNACompareMode mode,
+                                      IDOverrideLibrary *override,
+                                      const int flags,
+                                      eRNAOverrideMatchResult *r_report_flags);
+
+bool RNA_property_equals(
+    Main *bmain, PointerRNA *ptr_a, PointerRNA *ptr_b, PropertyRNA *prop, eRNACompareMode mode)
+{
+  BLI_assert(ELEM(mode, RNA_EQ_STRICT, RNA_EQ_UNSET_MATCH_ANY, RNA_EQ_UNSET_MATCH_NONE));
+
+  return (rna_property_override_diff(
+              bmain, ptr_a, ptr_b, prop, NULL, NULL, NULL, mode, NULL, 0, NULL) == 0);
+}
+
+bool RNA_struct_equals(Main *bmain, PointerRNA *ptr_a, PointerRNA *ptr_b, eRNACompareMode mode)
+{
+  CollectionPropertyIterator iter;
+  PropertyRNA *iterprop;
+  bool equals = true;
+
+  if (ptr_a == NULL && ptr_b == NULL) {
+    return true;
+  }
+  else if (ptr_a == NULL || ptr_b == NULL) {
+    return false;
+  }
+  else if (ptr_a->type != ptr_b->type) {
+    return false;
+  }
+
+  iterprop = RNA_struct_iterator_property(ptr_a->type);
+
+  RNA_property_collection_begin(ptr_a, iterprop, &iter);
+  for (; iter.valid; RNA_property_collection_next(&iter)) {
+    PropertyRNA *prop = iter.ptr.data;
+
+    if (!RNA_property_equals(bmain, ptr_a, ptr_b, prop, mode)) {
+      equals = false;
+      break;
+    }
+  }
+  RNA_property_collection_end(&iter);
+
+  return equals;
+}
+
+/* Low-level functions, also used by non-override RNA API like copy or equality check. */
+
+/** Generic RNA property diff function.
+ *
+ * \note about \a prop and \a prop_a/prop_b parameters:
+ * the former is expected to be an 'un-resolved' one,
+ * while the two later are expected to be fully resolved ones
+ * (i.e. to be the IDProps when they should be, etc.).
+ * When \a prop is given, \a prop_a and \a prop_b should always be NULL, and vice-versa.
+ * This is necessary, because we cannot perform 'set/unset' checks on resolved properties
+ * (unset IDProps would merely be NULL then).
+ *
+ * \note When there is no equality,
+ * but we cannot determine an order (greater than/lesser than), we return 1.
+ */
+static int rna_property_override_diff(Main *bmain,
+                                      PointerRNA *ptr_a,
+                                      PointerRNA *ptr_b,
+                                      PropertyRNA *prop,
+                                      PropertyRNA *prop_a,
+                                      PropertyRNA *prop_b,
+                                      const char *rna_path,
+                                      eRNACompareMode mode,
+                                      IDOverrideLibrary *override,
+                                      const int flags,
+                                      eRNAOverrideMatchResult *r_report_flags)
+{
+  if (prop != NULL) {
+    BLI_assert(prop_a == NULL && prop_b == NULL);
+    prop_a = prop;
+    prop_b = prop;
+  }
+
+  if (ELEM(NULL, prop_a, prop_b)) {
+    return (prop_a == prop_b) ? 0 : 1;
+  }
+
+  if (!RNA_property_comparable(ptr_a, prop_a) || !RNA_property_comparable(ptr_b, prop_b)) {
+    return 0;
+  }
+
+  if (mode == RNA_EQ_UNSET_MATCH_ANY) {
+    /* uninitialized properties are assumed to match anything */
+    if (!RNA_property_is_set(ptr_a, prop_a) || !RNA_property_is_set(ptr_b, prop_b)) {
+      return 0;
+    }
+  }
+  else if (mode == RNA_EQ_UNSET_MATCH_NONE) {
+    /* unset properties never match set properties */
+    if (RNA_property_is_set(ptr_a, prop_a) != RNA_property_is_set(ptr_b, prop_b)) {
+      return 1;
+    }
+  }
+
+  if (prop != NULL) {
+    /* Ensure we get real property data, be it an actual RNA property,
+     * or an IDProperty in disguise. */
+    prop_a = rna_ensure_property_realdata(&prop_a, ptr_a);
+    prop_b = rna_ensure_property_realdata(&prop_b, ptr_b);
+
+    if (ELEM(NULL, prop_a, prop_b)) {
+      return (prop_a == prop_b) ? 0 : 1;
+    }
+  }
+
+  /* Check if we are working with arrays. */
+  const bool is_array_a = RNA_property_array_check(prop_a);
+  const bool is_array_b = RNA_property_array_check(prop_b);
+
+  if (is_array_a != is_array_b) {
+    /* Should probably never happen actually... */
+    BLI_assert(0);
+    return is_array_a ? 1 : -1;
+  }
+
+  /* Get the length of the array to work with. */
+  const int len_a = RNA_property_array_length(ptr_a, prop_a);
+  const int len_b = RNA_property_array_length(ptr_b, prop_b);
+
+  if (len_a != len_b) {
+    /* Do not handle override in that case,
+     * we do not support insertion/deletion from arrays for now. */
+    return len_a > len_b ? 1 : -1;
+  }
+
+  if (is_array_a && len_a == 0) {
+    /* Empty arrays, will happen in some case with dynamic ones. */
+    return 0;
+  }
+
+  RNAPropOverrideDiff override_diff = NULL;
+  /* Special case for IDProps, we use default callback then. */
+  if (prop_a->magic != RNA_MAGIC) {
+    override_diff = rna_property_override_diff_default;
+    if (prop_b->magic == RNA_MAGIC && prop_b->override_diff != override_diff) {
+      override_diff = NULL;
+    }
+  }
+  else if (prop_b->magic != RNA_MAGIC) {
+    override_diff = rna_property_override_diff_default;
+    if (prop_a->override_diff != override_diff) {
+      override_diff = NULL;
+    }
+  }
+  else if (prop_a->override_diff == prop_b->override_diff) {
+    override_diff = prop_a->override_diff;
+  }
+
+  if (override_diff == NULL) {
+#ifndef NDEBUG
+    printf("'%s' gives unmatching or NULL RNA diff callbacks, should not happen (%d vs. %d).\n",
+           rna_path ?
+               rna_path :
+               (prop_a->magic != RNA_MAGIC ? ((IDProperty *)prop_a)->name : prop_a->identifier),
+           prop_a->magic == RNA_MAGIC,
+           prop_b->magic == RNA_MAGIC);
+#endif
+    BLI_assert(0);
+    return 1;
+  }
+
+  bool override_changed = false;
+  int diff_flags = flags;
+  if (!RNA_property_overridable_get(ptr_a, prop_a)) {
+    diff_flags &= ~RNA_OVERRIDE_COMPARE_CREATE;
+  }
+  const int diff = override_diff(bmain,
+                                 ptr_a,
+                                 ptr_b,
+                                 prop_a,
+                                 prop_b,
+                                 len_a,
+                                 len_b,
+                                 mode,
+                                 override,
+                                 rna_path,
+                                 diff_flags,
+                                 &override_changed);
+  if (override_changed && r_report_flags) {
+    *r_report_flags |= RNA_OVERRIDE_MATCH_RESULT_CREATED;
+  }
+
+  return diff;
+}
+
+/* Modify local data-block to make it ready for override application
+ * (only needed for diff operations, where we use
+ * the local data-block's data as second operand). */
+static bool rna_property_override_operation_store(Main *bmain,
+                                                  PointerRNA *ptr_local,
+                                                  PointerRNA *ptr_reference,
+                                                  PointerRNA *ptr_storage,
+                                                  PropertyRNA *prop_local,
+                                                  PropertyRNA *prop_reference,
+                                                  PropertyRNA *prop_storage,
+                                                  IDOverrideLibraryProperty *op)
+{
+  int len_local, len_reference, len_storage = 0;
+  bool changed = false;
+
+  if (ptr_storage == NULL) {
+    return changed;
+  }
+
+  /* get the length of the array to work with */
+  len_local = RNA_property_array_length(ptr_local, prop_local);
+  len_reference = RNA_property_array_length(ptr_reference, prop_reference);
+  if (prop_storage) {
+    len_storage = RNA_property_array_length(ptr_storage, prop_storage);
+  }
+
+  if (len_local != len_reference || len_local != len_storage) {
+    /* Do not handle override in that case,
+     * we do not support insertion/deletion from arrays for now. */
+    return changed;
+  }
+
+  BLI_assert(prop_local->override_store == prop_reference->override_store &&
+             (!ptr_storage || prop_local->override_store == prop_storage->override_store) &&
+             prop_local->override_store != NULL);
+
+  for (IDOverrideLibraryPropertyOperation *opop = op->operations.first; opop; opop = opop->next) {
+    /* Only needed for diff operations. */
+    if (!ELEM(opop->operation,
+              IDOVERRIDE_LIBRARY_OP_ADD,
+              IDOVERRIDE_LIBRARY_OP_SUBTRACT,
+              IDOVERRIDE_LIBRARY_OP_MULTIPLY)) {
+      continue;
+    }
+
+    if (prop_local->override_store(bmain,
+                                   ptr_local,
+                                   ptr_reference,
+                                   ptr_storage,
+                                   prop_local,
+                                   prop_reference,
+                                   prop_storage,
+                                   len_local,
+                                   len_reference,
+                                   len_storage,
+                                   opop)) {
+      changed = true;
+    }
+  }
+
+  return changed;
+}
+
+static bool rna_property_override_operation_apply(Main *bmain,
+                                                  PointerRNA *ptr_dst,
+                                                  PointerRNA *ptr_src,
+                                                  PointerRNA *ptr_storage,
+                                                  PropertyRNA *prop_dst,
+                                                  PropertyRNA *prop_src,
+                                                  PropertyRNA *prop_storage,
+                                                  PointerRNA *ptr_item_dst,
+                                                  PointerRNA *ptr_item_src,
+                                                  PointerRNA *ptr_item_storage,
+                                                  IDOverrideLibraryPropertyOperation *opop)
+{
+  int len_dst, len_src, len_storage = 0;
+
+  const short override_op = opop->operation;
+
+  if (override_op == IDOVERRIDE_LIBRARY_OP_NOOP) {
+    return true;
+  }
+
+  if (ELEM(override_op,
+           IDOVERRIDE_LIBRARY_OP_ADD,
+           IDOVERRIDE_LIBRARY_OP_SUBTRACT,
+           IDOVERRIDE_LIBRARY_OP_MULTIPLY) &&
+      !ptr_storage) {
+    /* We cannot apply 'diff' override operations without some reference storage.
+     * This should typically only happen at read time of .blend file... */
+    return false;
+  }
+
+  if (ELEM(override_op,
+           IDOVERRIDE_LIBRARY_OP_ADD,
+           IDOVERRIDE_LIBRARY_OP_SUBTRACT,
+           IDOVERRIDE_LIBRARY_OP_MULTIPLY) &&
+      !prop_storage) {
+    /* We cannot apply 'diff' override operations without some reference storage.
+     * This should typically only happen at read time of .blend file... */
+    return false;
+  }
+
+  RNAPropOverrideApply override_apply = NULL;
+  /* Special case for IDProps, we use default callback then. */
+  if (prop_dst->magic != RNA_MAGIC) {
+    override_apply = rna_property_override_apply_default;
+    if (prop_src->magic == RNA_MAGIC && prop_src->override_apply != override_apply) {
+      override_apply = NULL;
+    }
+  }
+  else if (prop_src->magic != RNA_MAGIC) {
+    override_apply = rna_property_override_apply_default;
+    if (prop_dst->override_apply != override_apply) {
+      override_apply = NULL;
+    }
+  }
+  else if (prop_dst->override_apply == prop_src->override_apply) {
+    override_apply = prop_dst->override_apply;
+  }
+
+  if (ptr_storage && prop_storage->magic == RNA_MAGIC &&
+      prop_storage->override_apply != override_apply) {
+    override_apply = NULL;
+  }
+
+  if (override_apply == NULL) {
+#ifndef NDEBUG
+    printf("'%s' gives unmatching or NULL RNA copy callbacks, should not happen (%d vs. %d).\n",
+           prop_dst->magic != RNA_MAGIC ? ((IDProperty *)prop_dst)->name : prop_dst->identifier,
+           prop_dst->magic == RNA_MAGIC,
+           prop_src->magic == RNA_MAGIC);
+#endif
+    BLI_assert(0);
+    return false;
+  }
+
+  /* get the length of the array to work with */
+  len_dst = RNA_property_array_length(ptr_dst, prop_dst);
+  len_src = RNA_property_array_length(ptr_src, prop_src);
+  if (ptr_storage) {
+    len_storage = RNA_property_array_length(ptr_storage, prop_storage);
+  }
+
+  if (len_dst != len_src || (ptr_storage && len_dst != len_storage)) {
+    /* Do not handle override in that case,
+     * we do not support insertion/deletion from arrays for now. */
+    return false;
+  }
+
+  /* get and set the default values as appropriate for the various types */
+  return override_apply(bmain,
+                        ptr_dst,
+                        ptr_src,
+                        ptr_storage,
+                        prop_dst,
+                        prop_src,
+                        prop_storage,
+                        len_dst,
+                        len_src,
+                        len_storage,
+                        ptr_item_dst,
+                        ptr_item_src,
+                        ptr_item_storage,
+                        opop);
+}
+
+/**
+ * Check whether reference and local overridden data match (are the same),
+ * with respect to given restrictive sets of properties.
+ * If requested, will generate needed new property overrides, and/or restore values from reference.
+ *
+ * \param r_report_flags: If given,
+ * will be set with flags matching actions taken by the function on \a ptr_local.
+ *
+ * \return True if _resulting_ \a ptr_local does match \a ptr_reference.
+ */
+bool RNA_struct_override_matches(Main *bmain,
+                                 PointerRNA *ptr_local,
+                                 PointerRNA *ptr_reference,
+                                 const char *root_path,
+                                 IDOverrideLibrary *override,
+                                 const eRNAOverrideMatch flags,
+                                 eRNAOverrideMatchResult *r_report_flags)
+{
+  CollectionPropertyIterator iter;
+  PropertyRNA *iterprop;
+  bool matching = true;
+
+  BLI_assert(ptr_local->type == ptr_reference->type);
+  BLI_assert(ptr_local->owner_id && ptr_reference->owner_id);
+
+  const bool ignore_non_overridable = (flags & RNA_OVERRIDE_COMPARE_IGNORE_NON_OVERRIDABLE) != 0;
+  const bool ignore_overridden = (flags & RNA_OVERRIDE_COMPARE_IGNORE_OVERRIDDEN) != 0;
+  const bool do_create = (flags & RNA_OVERRIDE_COMPARE_CREATE) != 0;
+  const bool do_restore = (flags & RNA_OVERRIDE_COMPARE_RESTORE) != 0;
+
+//#define DEBUG_OVERRIDE_TIMEIT
+#ifdef DEBUG_OVERRIDE_TIMEIT
+  static float _sum_time_global = 0.0f;
+  static float _num_time_global = 0.0f;
+  double _timeit_time_global;
+  static float _sum_time_diffing = 0.0f;
+  static float _delta_time_diffing = 0.0f;
+  static int _num_delta_time_diffing = 0.0f;
+  static float _num_time_diffing = 0.0f;
+  double _timeit_time_diffing;
+
+  if (!root_path) {
+    _delta_time_diffing = 0.0f;
+    _num_delta_time_diffing = 0;
+    _timeit_time_global = PIL_check_seconds_timer();
+  }
+#endif
+
+  iterprop = RNA_struct_iterator_property(ptr_local->type);
+
+  for (RNA_property_collection_begin(ptr_local, iterprop, &iter); iter.valid;
+       RNA_property_collection_next(&iter)) {
+    PropertyRNA *prop_local = iter.ptr.data;
+    PropertyRNA *prop_reference = iter.ptr.data;
+
+    /* Ensure we get real property data, be it an actual RNA property,
+     * or an IDProperty in disguise. */
+    prop_local = rna_ensure_property_realdata(&prop_local, ptr_local);
+    prop_reference = rna_ensure_property_realdata(&prop_reference, ptr_reference);
+
+    if (ELEM(NULL, prop_local, prop_reference)) {
+      continue;
+    }
+
+    if (ignore_non_overridable && !RNA_property_overridable_get(ptr_local, prop_local)) {
+      continue;
+    }
+
+#if 0 /* This actually makes things slower, since it has to check for animation paths etc! */
+    if (RNA_property_animated(ptr_local, prop_local)) {
+      /* We cannot do anything here really, animation is some kind of dynamic overrides that has
+       * precedence over static one... */
+      continue;
+    }
+#endif
+
+#define RNA_PATH_BUFFSIZE 8192
+#define RNA_PATH_PRINTF(_str, ...) \
+  if (BLI_snprintf(rna_path, RNA_PATH_BUFFSIZE, (_str), __VA_ARGS__) >= RNA_PATH_BUFFSIZE) { \
+    rna_path = BLI_sprintfN((_str), __VA_ARGS__); \
+  } \
+  (void)0
+#define RNA_PATH_FREE \
+  if (rna_path != rna_path_buffer) \
+  MEM_freeN(rna_path)
+
+    char rna_path_buffer[RNA_PATH_BUFFSIZE];
+    char *rna_path = rna_path_buffer;
+
+    /* XXX TODO this will have to be refined to handle collections insertions, and array items */
+    if (root_path) {
+      /* Inlined building, much much more efficient. */
+      if (prop_local->magic == RNA_MAGIC) {
+        RNA_PATH_PRINTF("%s.%s", root_path, RNA_property_identifier(prop_local));
+      }
+      else {
+        RNA_PATH_PRINTF("%s[\"%s\"]", root_path, RNA_property_identifier(prop_local));
+      }
+    }
+    else {
+      /* This is rather slow, but is not much called, so not really worth optimizing. */
+      rna_path = RNA_path_from_ID_to_property(ptr_local, prop_local);
+    }
+    if (rna_path == NULL) {
+      continue;
+    }
+
+    //    printf("Override Checking %s\n", rna_path);
+
+    if (ignore_overridden && BKE_override_library_property_find(override, rna_path) != NULL) {
+      RNA_PATH_FREE;
+      continue;
+    }
+
+#ifdef DEBUG_OVERRIDE_TIMEIT
+    if (!root_path) {
+      _timeit_time_diffing = PIL_check_seconds_timer();
+    }
+#endif
+
+    eRNAOverrideMatchResult report_flags = 0;
+    const int diff = rna_property_override_diff(bmain,
+                                                ptr_local,
+                                                ptr_reference,
+                                                NULL,
+                                                prop_local,
+                                                prop_reference,
+                                                rna_path,
+                                                RNA_EQ_STRICT,
+                                                override,
+                                                flags,
+                                                &report_flags);
+
+#ifdef DEBUG_OVERRIDE_TIMEIT
+    if (!root_path) {
+      const float _delta_time = (float)(PIL_check_seconds_timer() - _timeit_time_diffing);
+      _delta_time_diffing += _delta_time;
+      _num_delta_time_diffing++;
+    }
+#endif
+
+    matching = matching && diff == 0;
+    if (r_report_flags) {
+      *r_report_flags |= report_flags;
+    }
+
+    if (diff != 0) {
+      /* XXX TODO: refine this for per-item overriding of arrays... */
+      IDOverrideLibraryProperty *op = BKE_override_library_property_find(override, rna_path);
+      IDOverrideLibraryPropertyOperation *opop = op ? op->operations.first : NULL;
+
+      if (do_restore && (report_flags & RNA_OVERRIDE_MATCH_RESULT_CREATED) == 0) {
+        /* We are allowed to restore to reference's values. */
+        if (ELEM(NULL, op, opop) || opop->operation == IDOVERRIDE_LIBRARY_OP_NOOP) {
+          /* We should restore that property to its reference value */
+          if (RNA_property_editable(ptr_local, prop_local)) {
+            IDOverrideLibraryPropertyOperation opop_tmp = {
+                .operation = IDOVERRIDE_LIBRARY_OP_REPLACE,
+                .subitem_reference_index = -1,
+                .subitem_local_index = -1,
+            };
+            rna_property_override_operation_apply(bmain,
+                                                  ptr_local,
+                                                  ptr_reference,
+                                                  NULL,
+                                                  prop_local,
+                                                  prop_reference,
+                                                  NULL,
+                                                  NULL,
+                                                  NULL,
+                                                  NULL,
+                                                  &opop_tmp);
+            if (r_report_flags) {
+              *r_report_flags |= RNA_OVERRIDE_MATCH_RESULT_RESTORED;
+            }
+          }
+          else {
+            /* Too noisy for now, this triggers on runtime props like transform matrices etc. */
+#if 0
+            BLI_assert(!"We have differences between reference and "
+                       "overriding data on non-editable property.");
+#endif
+            matching = false;
+          }
+        }
+      }
+      else if ((report_flags & RNA_OVERRIDE_MATCH_RESULT_CREATED) == 0 && ELEM(NULL, op, opop)) {
+        /* This property is not overridden, and differs from reference, so we have no match. */
+        matching = false;
+        if (!(do_create || do_restore)) {
+          /* Since we have no 'changing' action allowed, we can break here. */
+          MEM_SAFE_FREE(rna_path);
+          break;
+        }
+      }
+    }
+
+    RNA_PATH_FREE;
+
+#undef RNA_PATH_BUFFSIZE
+#undef RNA_PATH_PRINTF
+#undef RNA_PATH_FREE
+  }
+  RNA_property_collection_end(&iter);
+
+#ifdef DEBUG_OVERRIDE_TIMEIT
+  if (!root_path) {
+    const float _delta_time = (float)(PIL_check_seconds_timer() - _timeit_time_global);
+    _sum_time_global += _delta_time;
+    _num_time_global++;
+    _sum_time_diffing += _delta_time_diffing;
+    _num_time_diffing++;
+    printf("ID: %s\n", ((ID *)ptr_local->owner_id)->name);
+    printf("time end      (%s): %.6f\n", __func__, _delta_time);
+    printf("time averaged (%s): %.6f (total: %.6f, in %d runs)\n",
+           __func__,
+           (_sum_time_global / _num_time_global),
+           _sum_time_global,
+           (int)_num_time_global);
+    printf("diffing time end      (%s): %.6f (in %d runs)\n",
+           __func__,
+           _delta_time_diffing,
+           _num_delta_time_diffing);
+    printf("diffing time averaged (%s): %.6f (total: %.6f, in %d runs)\n",
+           __func__,
+           (_sum_time_diffing / _num_time_diffing),
+           _sum_time_diffing,
+           (int)_num_time_diffing);
+  }
+#endif
+
+  return matching;
+}
+
+/** Store needed second operands into \a storage data-block
+ * for differential override operations. */
+bool RNA_struct_override_store(Main *bmain,
+                               PointerRNA *ptr_local,
+                               PointerRNA *ptr_reference,
+                               PointerRNA *ptr_storage,
+                               IDOverrideLibrary *override)
+{
+  bool changed = false;
+
+#ifdef DEBUG_OVERRIDE_TIMEIT
+  TIMEIT_START_AVERAGED(RNA_struct_override_store);
+#endif
+  for (IDOverrideLibraryProperty *op = override->properties.first; op; op = op->next) {
+    /* Simplified for now! */
+    PointerRNA data_reference, data_local;
+    PropertyRNA *prop_reference, *prop_local;
+
+    if (RNA_path_resolve_property(ptr_local, op->rna_path, &data_local, &prop_local) &&
+        RNA_path_resolve_property(ptr_reference, op->rna_path, &data_reference, &prop_reference)) {
+      PointerRNA data_storage;
+      PropertyRNA *prop_storage = NULL;
+
+      /* It is totally OK if this does not success,
+       * only a subset of override operations actually need storage. */
+      if (ptr_storage && (ptr_storage->owner_id != NULL)) {
+        RNA_path_resolve_property(ptr_storage, op->rna_path, &data_storage, &prop_storage);
+      }
+
+      if (rna_property_override_operation_store(bmain,
+                                                &data_local,
+                                                &data_reference,
+                                                &data_storage,
+                                                prop_reference,
+                                                prop_local,
+                                                prop_storage,
+                                                op)) {
+        changed = true;
+      }
+    }
+  }
+#ifdef DEBUG_OVERRIDE_TIMEIT
+  TIMEIT_END_AVERAGED(RNA_struct_override_store);
+#endif
+
+  return changed;
+}
+
+static void rna_property_override_apply_ex(Main *bmain,
+                                           PointerRNA *ptr_dst,
+                                           PointerRNA *ptr_src,
+                                           PointerRNA *ptr_storage,
+                                           PropertyRNA *prop_dst,
+                                           PropertyRNA *prop_src,
+                                           PropertyRNA *prop_storage,
+                                           PointerRNA *ptr_item_dst,
+                                           PointerRNA *ptr_item_src,
+                                           PointerRNA *ptr_item_storage,
+                                           IDOverrideLibraryProperty *op,
+                                           const bool do_insert)
+{
+  for (IDOverrideLibraryPropertyOperation *opop = op->operations.first; opop; opop = opop->next) {
+    if (!do_insert != !ELEM(opop->operation,
+                            IDOVERRIDE_LIBRARY_OP_INSERT_AFTER,
+                            IDOVERRIDE_LIBRARY_OP_INSERT_BEFORE)) {
+      if (!do_insert) {
+        printf("Skipping insert override operations in first pass (%s)!\n", op->rna_path);
+      }
+      continue;
+    }
+
+    /* Note: will have to think about putting that logic into its own function maybe?
+     * Would be nice to have it in a single place...
+     * Note that here, src is the local saved ID, and dst is a copy of the linked ID (since we use
+     * local ID as storage to apply local changes on top of a clean copy of the linked data). */
+    PointerRNA private_ptr_item_dst, private_ptr_item_src, private_ptr_item_storage;
+    if (opop->subitem_local_name != NULL || opop->subitem_reference_name != NULL ||
+        opop->subitem_local_index != -1 || opop->subitem_reference_index != -1) {
+      RNA_POINTER_INVALIDATE(&private_ptr_item_dst);
+      RNA_POINTER_INVALIDATE(&private_ptr_item_src);
+      RNA_POINTER_INVALIDATE(&private_ptr_item_storage);
+      if (opop->subitem_local_name != NULL) {
+        RNA_property_collection_lookup_string(
+            ptr_src, prop_src, opop->subitem_local_name, &private_ptr_item_src);
+        if (opop->subitem_reference_name != NULL) {
+          RNA_property_collection_lookup_string(
+              ptr_dst, prop_dst, opop->subitem_reference_name, &private_ptr_item_dst);
+        }
+        else {
+          RNA_property_collection_lookup_string(
+              ptr_dst, prop_dst, opop->subitem_local_name, &private_ptr_item_dst);
+        }
+      }
+      else if (opop->subitem_reference_name != NULL) {
+        RNA_property_collection_lookup_string(
+            ptr_src, prop_src, opop->subitem_reference_name, &private_ptr_item_src);
+        RNA_property_collection_lookup_string(
+            ptr_dst, prop_dst, opop->subitem_reference_name, &private_ptr_item_dst);
+      }
+      else if (opop->subitem_local_index != -1) {
+        RNA_property_collection_lookup_int(
+            ptr_src, prop_src, opop->subitem_local_index, &private_ptr_item_src);
+        if (opop->subitem_reference_index != -1) {
+          RNA_property_collection_lookup_int(
+              ptr_dst, prop_dst, opop->subitem_reference_index, &private_ptr_item_dst);
+        }
+        else {
+          RNA_property_collection_lookup_int(
+              ptr_dst, prop_dst, opop->subitem_local_index, &private_ptr_item_dst);
+        }
+      }
+      else if (opop->subitem_reference_index != -1) {
+        RNA_property_collection_lookup_int(
+            ptr_src, prop_src, opop->subitem_reference_index, &private_ptr_item_src);
+        RNA_property_collection_lookup_int(
+            ptr_dst, prop_dst, opop->subitem_reference_index, &private_ptr_item_dst);
+      }
+      if (prop_storage != NULL) {
+        if (opop->subitem_local_name != NULL) {
+          RNA_property_collection_lookup_string(
+              ptr_storage, prop_storage, opop->subitem_local_name, &private_ptr_item_storage);
+        }
+        else if (opop->subitem_reference_name != NULL) {
+          RNA_property_collection_lookup_string(
+              ptr_storage, prop_storage, opop->subitem_reference_name, &private_ptr_item_storage);
+        }
+        else if (opop->subitem_local_index != -1) {
+          RNA_property_collection_lookup_int(
+              ptr_storage, prop_storage, opop->subitem_local_index, &private_ptr_item_storage);
+        }
+        else if (opop->subitem_reference_index != -1) {
+          RNA_property_collection_lookup_int(
+              ptr_storage, prop_storage, opop->subitem_reference_index, &private_ptr_item_storage);
+        }
+      }
+      ptr_item_dst = &private_ptr_item_dst;
+      ptr_item_src = &private_ptr_item_src;
+      ptr_item_storage = &private_ptr_item_storage;
+    }
+
+    if (!rna_property_override_operation_apply(bmain,
+                                               ptr_dst,
+                                               ptr_src,
+                                               ptr_storage,
+                                               prop_dst,
+                                               prop_src,
+                                               prop_storage,
+                                               ptr_item_dst,
+                                               ptr_item_src,
+                                               ptr_item_storage,
+                                               opop)) {
+      /* TODO No assert here, would be much much better to just report as warning,
+       * failing override applications will probably be fairly common! */
+      BLI_assert(0);
+    }
+  }
+}
+
+/** Apply given \a override operations on \a ptr_dst, using \a ptr_src
+ * (and \a ptr_storage for differential ops) as source. */
+void RNA_struct_override_apply(Main *bmain,
+                               PointerRNA *ptr_dst,
+                               PointerRNA *ptr_src,
+                               PointerRNA *ptr_storage,
+                               IDOverrideLibrary *override)
+{
+#ifdef DEBUG_OVERRIDE_TIMEIT
+  TIMEIT_START_AVERAGED(RNA_struct_override_apply);
+#endif
+  /* Note: Applying insert operations in a separate pass is mandatory.
+   * We could optimize this later, but for now, as inefficient as it is,
+   * don't think this is a critical point.
+   */
+  bool do_insert = false;
+  for (int i = 0; i < 2; i++, do_insert = true) {
+    for (IDOverrideLibraryProperty *op = override->properties.first; op; op = op->next) {
+      /* Simplified for now! */
+      PointerRNA data_src, data_dst;
+      PointerRNA data_item_src, data_item_dst;
+      PropertyRNA *prop_src, *prop_dst;
+
+      if (RNA_path_resolve_property_and_item_pointer(
+              ptr_dst, op->rna_path, &data_dst, &prop_dst, &data_item_dst) &&
+          RNA_path_resolve_property_and_item_pointer(
+              ptr_src, op->rna_path, &data_src, &prop_src, &data_item_src)) {
+        PointerRNA data_storage, data_item_storage;
+        PropertyRNA *prop_storage = NULL;
+
+        /* It is totally OK if this does not success,
+         * only a subset of override operations actually need storage. */
+        if (ptr_storage && (ptr_storage->owner_id != NULL)) {
+          RNA_path_resolve_property_and_item_pointer(
+              ptr_storage, op->rna_path, &data_storage, &prop_storage, &data_item_storage);
+        }
+
+        rna_property_override_apply_ex(bmain,
+                                       &data_dst,
+                                       &data_src,
+                                       prop_storage ? &data_storage : NULL,
+                                       prop_dst,
+                                       prop_src,
+                                       prop_storage,
+                                       &data_item_dst,
+                                       &data_item_src,
+                                       prop_storage ? &data_item_storage : NULL,
+                                       op,
+                                       do_insert);
+      }
+#ifndef NDEBUG
+      else {
+        printf(
+            "Failed to apply library override operation to '%s.%s' "
+            "(could not resolve some properties, local:  %d, override: %d)\n",
+            ((ID *)ptr_src->owner_id)->name,
+            op->rna_path,
+            RNA_path_resolve_property(ptr_dst, op->rna_path, &data_dst, &prop_dst),
+            RNA_path_resolve_property(ptr_src, op->rna_path, &data_src, &prop_src));
+      }
+#endif
+    }
+  }
+#ifdef DEBUG_OVERRIDE_TIMEIT
+  TIMEIT_END_AVERAGED(RNA_struct_override_apply);
+#endif
+}
+
+IDOverrideLibraryProperty *RNA_property_override_property_find(PointerRNA *ptr, PropertyRNA *prop)
+{
+  ID *id = ptr->owner_id;
+
+  if (!id || !id->override_library) {
+    return NULL;
+  }
+
+  char *rna_path = RNA_path_from_ID_to_property(ptr, prop);
+  if (rna_path) {
+    IDOverrideLibraryProperty *op = BKE_override_library_property_find(id->override_library,
+                                                                       rna_path);
+    MEM_freeN(rna_path);
+    return op;
+  }
+  return NULL;
+}
+
+IDOverrideLibraryProperty *RNA_property_override_property_get(PointerRNA *ptr,
+                                                              PropertyRNA *prop,
+                                                              bool *r_created)
+{
+  ID *id = ptr->owner_id;
+
+  if (!id || !id->override_library) {
+    return NULL;
+  }
+
+  char *rna_path = RNA_path_from_ID_to_property(ptr, prop);
+  if (rna_path) {
+    IDOverrideLibraryProperty *op = BKE_override_library_property_get(
+        id->override_library, rna_path, r_created);
+    MEM_freeN(rna_path);
+    return op;
+  }
+  return NULL;
+}
+
+IDOverrideLibraryPropertyOperation *RNA_property_override_property_operation_find(
+    PointerRNA *ptr, PropertyRNA *prop, const int index, const bool strict, bool *r_strict)
+{
+  IDOverrideLibraryProperty *op = RNA_property_override_property_find(ptr, prop);
+
+  if (!op) {
+    return NULL;
+  }
+
+  return BKE_override_library_property_operation_find(
+      op, NULL, NULL, index, index, strict, r_strict);
+}
+
+IDOverrideLibraryPropertyOperation *RNA_property_override_property_operation_get(
+    PointerRNA *ptr,
+    PropertyRNA *prop,
+    const short operation,
+    const int index,
+    const bool strict,
+    bool *r_strict,
+    bool *r_created)
+{
+  IDOverrideLibraryProperty *op = RNA_property_override_property_get(ptr, prop, NULL);
+
+  if (!op) {
+    return NULL;
+  }
+
+  return BKE_override_library_property_operation_get(
+      op, operation, NULL, NULL, index, index, strict, r_strict, r_created);
+}
+
+eRNAOverrideStatus RNA_property_override_library_status(PointerRNA *ptr,
+                                                        PropertyRNA *prop,
+                                                        const int index)
+{
+  int override_status = 0;
+
+  if (!BKE_override_library_is_enabled()) {
+    return override_status;
+  }
+
+  if (!ptr || !prop || !ptr->owner_id || !(ptr->owner_id)->override_library) {
+    return override_status;
+  }
+
+  if (RNA_property_overridable_get(ptr, prop) && RNA_property_editable_flag(ptr, prop)) {
+    override_status |= RNA_OVERRIDE_STATUS_OVERRIDABLE;
+  }
+
+  IDOverrideLibraryPropertyOperation *opop = RNA_property_override_property_operation_find(
+      ptr, prop, index, false, NULL);
+  if (opop != NULL) {
+    override_status |= RNA_OVERRIDE_STATUS_OVERRIDDEN;
+    if (opop->flag & IDOVERRIDE_LIBRARY_FLAG_MANDATORY) {
+      override_status |= RNA_OVERRIDE_STATUS_MANDATORY;
+    }
+    if (opop->flag & IDOVERRIDE_LIBRARY_FLAG_LOCKED) {
+      override_status |= RNA_OVERRIDE_STATUS_LOCKED;
+    }
+  }
+
+  return override_status;
+}
diff --git a/source/blender/makesrna/intern/rna_access_internal.h b/source/blender/makesrna/intern/rna_access_internal.h
new file mode 100644
index 00000000000..28ec504e376
--- /dev/null
+++ b/source/blender/makesrna/intern/rna_access_internal.h
@@ -0,0 +1,34 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+/** \file
+ * \ingroup RNA
+ */
+
+#ifndef __RNA_ACCESS_INTERNAL_H__
+#define __RNA_ACCESS_INTERNAL_H__
+
+#include "BLI_utildefines.h"
+
+#include "rna_internal_types.h"
+
+struct IDProperty;
+
+PropertyRNA *rna_ensure_property(PropertyRNA *prop);
+
+void rna_idproperty_touch(struct IDProperty *idprop);
+
+#endif /* __ACCESS_RNA_INTERNAL_H__ */
diff --git a/source/blender/makesrna/intern/rna_action.c b/source/blender/makesrna/intern/rna_action.c
index 4ad2aa55c75..24f8d5c4e3d 100644
--- a/source/blender/makesrna/intern/rna_action.c
+++ b/source/blender/makesrna/intern/rna_action.c
@@ -249,15 +249,15 @@ static void rna_Action_active_pose_marker_index_range(
 static void rna_Action_frame_range_get(PointerRNA *ptr, float *values)
 { /* don't include modifiers because they too easily can have very large
    * ranges: MINAFRAMEF to MAXFRAMEF. */
-  calc_action_range(ptr->id.data, values, values + 1, false);
+  calc_action_range((bAction *)ptr->owner_id, values, values + 1, false);
 }
 
 /* Used to check if an action (value pointer)
  * is suitable to be assigned to the ID-block that is ptr. */
 bool rna_Action_id_poll(PointerRNA *ptr, PointerRNA value)
 {
-  ID *srcId = (ID *)ptr->id.data;
-  bAction *act = (bAction *)value.id.data;
+  ID *srcId = ptr->owner_id;
+  bAction *act = (bAction *)value.owner_id;
 
   if (act) {
     /* there can still be actions that will have undefined id-root
@@ -280,7 +280,7 @@ bool rna_Action_id_poll(PointerRNA *ptr, PointerRNA value)
 bool rna_Action_actedit_assign_poll(PointerRNA *ptr, PointerRNA value)
 {
   SpaceAction *saction = (SpaceAction *)ptr->data;
-  bAction *act = (bAction *)value.id.data;
+  bAction *act = (bAction *)value.owner_id;
 
   if (act) {
     /* there can still be actions that will have undefined id-root
diff --git a/source/blender/makesrna/intern/rna_animation.c b/source/blender/makesrna/intern/rna_animation.c
index 3d14616bcc9..6157ec41f19 100644
--- a/source/blender/makesrna/intern/rna_animation.c
+++ b/source/blender/makesrna/intern/rna_animation.c
@@ -128,7 +128,7 @@ const EnumPropertyItem rna_enum_keying_flag_items_api[] = {
 
 static void rna_AnimData_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   ANIM_id_update(bmain, id);
 }
@@ -157,7 +157,7 @@ static void rna_AnimData_action_set(PointerRNA *ptr,
                                     PointerRNA value,
                                     struct ReportList *UNUSED(reports))
 {
-  ID *ownerId = (ID *)ptr->id.data;
+  ID *ownerId = ptr->owner_id;
 
   /* set action */
   BKE_animdata_set_action(NULL, ownerId, value.data);
@@ -303,7 +303,7 @@ static StructRNA *rna_KeyingSetInfo_register(Main *bmain,
 {
   KeyingSetInfo dummyksi = {NULL};
   KeyingSetInfo *ksi;
-  PointerRNA dummyptr = {{NULL}};
+  PointerRNA dummyptr = {NULL};
   int have_function[3];
 
   /* setup dummy type info to store static properties in */
@@ -727,12 +727,12 @@ bool rna_AnimaData_override_apply(Main *UNUSED(bmain),
 
   if (adt_dst == NULL && adt_src != NULL) {
     /* Copy anim data from reference into final local ID. */
-    BKE_animdata_copy_id(NULL, ptr_dst->id.data, ptr_src->id.data, 0);
+    BKE_animdata_copy_id(NULL, ptr_dst->owner_id, ptr_src->owner_id, 0);
     return true;
   }
   else if (adt_dst != NULL && adt_src == NULL) {
     /* Override has cleared/removed anim data from its reference. */
-    BKE_animdata_free(ptr_dst->id.data, true);
+    BKE_animdata_free(ptr_dst->owner_id, true);
     return true;
   }
 
diff --git a/source/blender/makesrna/intern/rna_armature.c b/source/blender/makesrna/intern/rna_armature.c
index 59e5c584811..7949db9055f 100644
--- a/source/blender/makesrna/intern/rna_armature.c
+++ b/source/blender/makesrna/intern/rna_armature.c
@@ -50,7 +50,7 @@
 
 static void rna_Armature_update_data(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   DEG_id_tag_update(id, 0);
   WM_main_add_notifier(NC_GEOM | ND_DATA, id);
@@ -59,7 +59,7 @@ static void rna_Armature_update_data(Main *UNUSED(bmain), Scene *UNUSED(scene),
 
 static void rna_Armature_dependency_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   DEG_relations_tag_update(bmain);
 
@@ -73,12 +73,12 @@ static void rna_Armature_act_bone_set(PointerRNA *ptr,
 {
   bArmature *arm = (bArmature *)ptr->data;
 
-  if (value.id.data == NULL && value.data == NULL) {
+  if (value.owner_id == NULL && value.data == NULL) {
     arm->act_bone = NULL;
   }
   else {
-    if (value.id.data != arm) {
-      Object *ob = (Object *)value.id.data;
+    if (value.owner_id != &arm->id) {
+      Object *ob = (Object *)value.owner_id;
 
       if (GS(ob->id.name) != ID_OB || (ob->data != arm)) {
         printf("ERROR: armature set active bone - new active doesn't come from this armature\n");
@@ -97,11 +97,11 @@ static void rna_Armature_act_edit_bone_set(PointerRNA *ptr,
 {
   bArmature *arm = (bArmature *)ptr->data;
 
-  if (value.id.data == NULL && value.data == NULL) {
+  if (value.owner_id == NULL && value.data == NULL) {
     arm->act_edbone = NULL;
   }
   else {
-    if (value.id.data != arm) {
+    if (value.owner_id != &arm->id) {
       /* raise an error! */
     }
     else {
@@ -151,7 +151,7 @@ static void rna_Armature_edit_bone_remove(bArmature *arm,
 
 static void rna_Armature_update_layers(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bArmature *arm = ptr->id.data;
+  bArmature *arm = (bArmature *)ptr->owner_id;
   Object *ob;
 
   /* proxy lib exception, store it here so we can restore layers on file
@@ -168,7 +168,7 @@ static void rna_Armature_update_layers(Main *bmain, Scene *UNUSED(scene), Pointe
 
 static void rna_Armature_redraw_data(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   DEG_id_tag_update(id, ID_RECALC_COPY_ON_WRITE);
   WM_main_add_notifier(NC_GEOM | ND_DATA, id);
@@ -177,7 +177,7 @@ static void rna_Armature_redraw_data(Main *UNUSED(bmain), Scene *UNUSED(scene),
 /* called whenever a bone is renamed */
 static void rna_Bone_update_renamed(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   /* redraw view */
   WM_main_add_notifier(NC_GEOM | ND_DATA, id);
@@ -188,7 +188,7 @@ static void rna_Bone_update_renamed(Main *UNUSED(bmain), Scene *UNUSED(scene), P
 
 static void rna_Bone_select_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   /* 1) special updates for cases where rigs try to hook into armature drawing stuff
    *    e.g. Mask Modifier - 'Armature' option
@@ -225,7 +225,7 @@ static void rna_Bone_select_update(Main *UNUSED(bmain), Scene *UNUSED(scene), Po
 
 static char *rna_Bone_path(PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   Bone *bone = (Bone *)ptr->data;
   char name_esc[sizeof(bone->name) * 2];
 
@@ -295,7 +295,7 @@ static void rna_bone_layer_set(int *layer, const bool *values)
 
 static void rna_Bone_layer_set(PointerRNA *ptr, const bool *values)
 {
-  bArmature *arm = (bArmature *)ptr->id.data;
+  bArmature *arm = (bArmature *)ptr->owner_id;
   Bone *bone = (Bone *)ptr->data;
 
   rna_bone_layer_set(&bone->layer, values);
@@ -331,7 +331,7 @@ static void rna_Armature_layer_set(PointerRNA *ptr, const bool *values)
 
 static void rna_EditBone_name_set(PointerRNA *ptr, const char *value)
 {
-  bArmature *arm = (bArmature *)ptr->id.data;
+  bArmature *arm = (bArmature *)ptr->owner_id;
   EditBone *ebone = (EditBone *)ptr->data;
   char oldname[sizeof(ebone->name)], newname[sizeof(ebone->name)];
 
@@ -345,7 +345,7 @@ static void rna_EditBone_name_set(PointerRNA *ptr, const char *value)
 
 static void rna_Bone_name_set(PointerRNA *ptr, const char *value)
 {
-  bArmature *arm = (bArmature *)ptr->id.data;
+  bArmature *arm = (bArmature *)ptr->owner_id;
   Bone *bone = (Bone *)ptr->data;
   char oldname[sizeof(bone->name)], newname[sizeof(bone->name)];
 
@@ -417,7 +417,7 @@ static void rna_EditBone_parent_set(PointerRNA *ptr,
   }
   else {
     /* within same armature */
-    if (value.id.data != ptr->id.data) {
+    if (value.owner_id != ptr->owner_id) {
       return;
     }
 
@@ -451,7 +451,7 @@ static void rna_EditBone_matrix_set(PointerRNA *ptr, const float *values)
 
 static void rna_Bone_bbone_handle_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  bArmature *arm = (bArmature *)ptr->id.data;
+  bArmature *arm = (bArmature *)ptr->owner_id;
   Bone *bone = (Bone *)ptr->data;
 
   /* Update all users of this armature after changing B-Bone handles. */
@@ -483,7 +483,7 @@ static void rna_EditBone_bbone_prev_set(PointerRNA *ptr,
   EditBone *hbone = (EditBone *)value.data;
 
   /* Within the same armature? */
-  if (hbone == NULL || value.id.data == ptr->id.data) {
+  if (hbone == NULL || value.owner_id == ptr->owner_id) {
     ebone->bbone_prev = hbone;
   }
 }
@@ -496,7 +496,7 @@ static void rna_Bone_bbone_prev_set(PointerRNA *ptr,
   Bone *hbone = (Bone *)value.data;
 
   /* Within the same armature? */
-  if (hbone == NULL || value.id.data == ptr->id.data) {
+  if (hbone == NULL || value.owner_id == ptr->owner_id) {
     bone->bbone_prev = hbone;
   }
 }
@@ -515,7 +515,7 @@ static void rna_EditBone_bbone_next_set(PointerRNA *ptr,
   EditBone *hbone = (EditBone *)value.data;
 
   /* Within the same armature? */
-  if (hbone == NULL || value.id.data == ptr->id.data) {
+  if (hbone == NULL || value.owner_id == ptr->owner_id) {
     ebone->bbone_next = hbone;
   }
 }
@@ -528,14 +528,14 @@ static void rna_Bone_bbone_next_set(PointerRNA *ptr,
   Bone *hbone = (Bone *)value.data;
 
   /* Within the same armature? */
-  if (hbone == NULL || value.id.data == ptr->id.data) {
+  if (hbone == NULL || value.owner_id == ptr->owner_id) {
     bone->bbone_next = hbone;
   }
 }
 
 static void rna_Armature_editbone_transform_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  bArmature *arm = (bArmature *)ptr->id.data;
+  bArmature *arm = (bArmature *)ptr->owner_id;
   EditBone *ebone = (EditBone *)ptr->data;
   EditBone *child;
 
@@ -590,7 +590,7 @@ static int rna_Armature_bones_lookup_string(PointerRNA *ptr, const char *key, Po
   bArmature *arm = (bArmature *)ptr->data;
   Bone *bone = BKE_armature_find_bone_name(arm, key);
   if (bone) {
-    RNA_pointer_create(ptr->id.data, &RNA_Bone, bone, r_ptr);
+    RNA_pointer_create(ptr->owner_id, &RNA_Bone, bone, r_ptr);
     return true;
   }
   else {
@@ -600,26 +600,29 @@ static int rna_Armature_bones_lookup_string(PointerRNA *ptr, const char *key, Po
 
 static bool rna_Armature_is_editmode_get(PointerRNA *ptr)
 {
-  bArmature *arm = (bArmature *)ptr->id.data;
+  bArmature *arm = (bArmature *)ptr->owner_id;
   return (arm->edbo != NULL);
 }
 
-static void rna_Armature_transform(struct bArmature *arm, Main *bmain, float *mat)
+static void rna_Armature_transform(bArmature *arm, float *mat)
 {
-  ED_armature_transform(bmain, arm, (float(*)[4])mat, true);
+  ED_armature_transform(arm, (const float(*)[4])mat, true);
 }
 
 #else
 
 /* Settings for curved bbone settings -
  * The posemode values get applied over the top of the editmode ones. */
-void rna_def_bone_curved_common(StructRNA *srna, bool is_posebone)
+void rna_def_bone_curved_common(StructRNA *srna, bool is_posebone, bool is_editbone)
 {
-#  define RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone) \
+#  define RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone) \
     { \
       if (is_posebone) { \
         RNA_def_property_update(prop, NC_OBJECT | ND_POSE, "rna_Pose_update"); \
       } \
+      else if (is_editbone) { \
+        RNA_def_property_update(prop, 0, "rna_Armature_editbone_transform_update"); \
+      } \
       else { \
         RNA_def_property_update(prop, 0, "rna_Armature_update_data"); \
       } \
@@ -634,14 +637,14 @@ void rna_def_bone_curved_common(StructRNA *srna, bool is_posebone)
   RNA_def_property_ui_range(prop, -M_PI * 2, M_PI * 2, 10, 2);
   RNA_def_property_ui_text(
       prop, "Roll In", "Roll offset for the start of the B-Bone, adjusts twist");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   prop = RNA_def_property(srna, "bbone_rollout", PROP_FLOAT, PROP_ANGLE);
   RNA_def_property_float_sdna(prop, NULL, "roll2");
   RNA_def_property_ui_range(prop, -M_PI * 2, M_PI * 2, 10, 2);
   RNA_def_property_ui_text(
       prop, "Roll Out", "Roll offset for the end of the B-Bone, adjusts twist");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   if (is_posebone == false) {
     prop = RNA_def_property(srna, "use_endroll_as_inroll", PROP_BOOLEAN, PROP_NONE);
@@ -658,28 +661,28 @@ void rna_def_bone_curved_common(StructRNA *srna, bool is_posebone)
   RNA_def_property_ui_range(prop, -FLT_MAX, FLT_MAX, 1, RNA_TRANSLATION_PREC_DEFAULT);
   RNA_def_property_ui_text(
       prop, "In X", "X-axis handle offset for start of the B-Bone's curve, adjusts curvature");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   prop = RNA_def_property(srna, "bbone_curveiny", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "curve_in_y");
   RNA_def_property_ui_range(prop, -FLT_MAX, FLT_MAX, 1, RNA_TRANSLATION_PREC_DEFAULT);
   RNA_def_property_ui_text(
       prop, "In Y", "Y-axis handle offset for start of the B-Bone's curve, adjusts curvature");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   prop = RNA_def_property(srna, "bbone_curveoutx", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "curve_out_x");
   RNA_def_property_ui_range(prop, -FLT_MAX, FLT_MAX, 1, RNA_TRANSLATION_PREC_DEFAULT);
   RNA_def_property_ui_text(
       prop, "Out X", "X-axis handle offset for end of the B-Bone's curve, adjusts curvature");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   prop = RNA_def_property(srna, "bbone_curveouty", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "curve_out_y");
   RNA_def_property_ui_range(prop, -FLT_MAX, FLT_MAX, 1, RNA_TRANSLATION_PREC_DEFAULT);
   RNA_def_property_ui_text(
       prop, "Out Y", "Y-axis handle offset for end of the B-Bone's curve, adjusts curvature");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   /* Ease In/Out */
   prop = RNA_def_property(srna, "bbone_easein", PROP_FLOAT, PROP_NONE);
@@ -687,14 +690,14 @@ void rna_def_bone_curved_common(StructRNA *srna, bool is_posebone)
   RNA_def_property_ui_range(prop, -5.0f, 5.0f, 1, 3);
   RNA_def_property_float_default(prop, 1.0f);
   RNA_def_property_ui_text(prop, "Ease In", "Length of first Bezier Handle (for B-Bones only)");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   prop = RNA_def_property(srna, "bbone_easeout", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "ease2");
   RNA_def_property_ui_range(prop, -5.0f, 5.0f, 1, 3);
   RNA_def_property_float_default(prop, 1.0f);
   RNA_def_property_ui_text(prop, "Ease Out", "Length of second Bezier Handle (for B-Bones only)");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   /* Scale In/Out */
   prop = RNA_def_property(srna, "bbone_scaleinx", PROP_FLOAT, PROP_NONE);
@@ -706,7 +709,7 @@ void rna_def_bone_curved_common(StructRNA *srna, bool is_posebone)
                            "Scale In X",
                            "X-axis scale factor for start of the B-Bone, "
                            "adjusts thickness (for tapering effects)");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   prop = RNA_def_property(srna, "bbone_scaleiny", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "scale_in_y");
@@ -717,7 +720,7 @@ void rna_def_bone_curved_common(StructRNA *srna, bool is_posebone)
                            "Scale In Y",
                            "Y-axis scale factor for start of the B-Bone, "
                            "adjusts thickness (for tapering effects)");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   prop = RNA_def_property(srna, "bbone_scaleoutx", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "scale_out_x");
@@ -728,7 +731,7 @@ void rna_def_bone_curved_common(StructRNA *srna, bool is_posebone)
                            "Scale Out X",
                            "X-axis scale factor for end of the B-Bone, "
                            "adjusts thickness (for tapering effects)");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
   prop = RNA_def_property(srna, "bbone_scaleouty", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "scale_out_y");
@@ -739,7 +742,7 @@ void rna_def_bone_curved_common(StructRNA *srna, bool is_posebone)
                            "Scale Out Y",
                            "Y-axis scale factor for end of the B-Bone, "
                            "adjusts thickness (for tapering effects)");
-  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone);
+  RNA_DEF_CURVEBONE_UPDATE(prop, is_posebone, is_editbone);
 
 #  undef RNA_DEF_CURVEBONE_UPDATE
 }
@@ -871,11 +874,16 @@ static void rna_def_bone_common(StructRNA *srna, int editbone)
   /* Number values */
   /* envelope deform settings */
   prop = RNA_def_property(srna, "envelope_distance", PROP_FLOAT, PROP_DISTANCE);
+  if (editbone) {
+    RNA_def_property_update(prop, 0, "rna_Armature_editbone_transform_update");
+  }
+  else {
+    RNA_def_property_update(prop, 0, "rna_Armature_update_data");
+  }
   RNA_def_property_float_sdna(prop, NULL, "dist");
   RNA_def_property_range(prop, 0.0f, 1000.0f);
   RNA_def_property_ui_text(
       prop, "Envelope Deform Distance", "Bone deformation distance (for Envelope deform only)");
-  RNA_def_property_update(prop, 0, "rna_Armature_update_data");
 
   prop = RNA_def_property(srna, "envelope_weight", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "weight");
@@ -914,23 +922,38 @@ static void rna_def_bone_common(StructRNA *srna, int editbone)
 
   /* b-bones deform settings */
   prop = RNA_def_property(srna, "bbone_segments", PROP_INT, PROP_NONE);
+  if (editbone) {
+    RNA_def_property_update(prop, 0, "rna_Armature_editbone_transform_update");
+  }
+  else {
+    RNA_def_property_update(prop, 0, "rna_Armature_dependency_update");
+  }
   RNA_def_property_int_sdna(prop, NULL, "segments");
   RNA_def_property_range(prop, 1, 32);
   RNA_def_property_ui_text(
       prop, "B-Bone Segments", "Number of subdivisions of bone (for B-Bones only)");
-  RNA_def_property_update(prop, 0, "rna_Armature_dependency_update");
 
   prop = RNA_def_property(srna, "bbone_x", PROP_FLOAT, PROP_NONE);
+  if (editbone) {
+    RNA_def_property_update(prop, 0, "rna_Armature_editbone_transform_update");
+  }
+  else {
+    RNA_def_property_update(prop, 0, "rna_Armature_update_data");
+  }
   RNA_def_property_float_sdna(prop, NULL, "xwidth");
   RNA_def_property_ui_range(prop, 0.0f, 1000.0f, 1, RNA_TRANSLATION_PREC_DEFAULT);
   RNA_def_property_ui_text(prop, "B-Bone Display X Width", "B-Bone X size");
-  RNA_def_property_update(prop, 0, "rna_Armature_update_data");
 
   prop = RNA_def_property(srna, "bbone_z", PROP_FLOAT, PROP_NONE);
+  if (editbone) {
+    RNA_def_property_update(prop, 0, "rna_Armature_editbone_transform_update");
+  }
+  else {
+    RNA_def_property_update(prop, 0, "rna_Armature_update_data");
+  }
   RNA_def_property_float_sdna(prop, NULL, "zwidth");
   RNA_def_property_ui_range(prop, 0.0f, 1000.0f, 1, RNA_TRANSLATION_PREC_DEFAULT);
   RNA_def_property_ui_text(prop, "B-Bone Display Z Width", "B-Bone Z size");
-  RNA_def_property_update(prop, 0, "rna_Armature_update_data");
 
   prop = RNA_def_property(srna, "bbone_handle_type_start", PROP_ENUM, PROP_NONE);
   RNA_def_property_enum_sdna(prop, NULL, "bbone_prev_type");
@@ -1010,7 +1033,7 @@ static void rna_def_bone(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Children", "Bones which are children of this bone");
 
   rna_def_bone_common(srna, 0);
-  rna_def_bone_curved_common(srna, 0);
+  rna_def_bone_curved_common(srna, false, false);
 
   /* XXX should we define this in PoseChannel wrapping code instead?
    *     But PoseChannels directly get some of their flags from here... */
@@ -1133,7 +1156,7 @@ static void rna_def_edit_bone(BlenderRNA *brna)
   RNA_def_property_update(prop, 0, "rna_Armature_editbone_transform_update");
 
   rna_def_bone_common(srna, 1);
-  rna_def_bone_curved_common(srna, 0);
+  rna_def_bone_curved_common(srna, false, true);
 
   prop = RNA_def_property(srna, "hide", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", BONE_HIDDEN_A);
@@ -1300,7 +1323,6 @@ static void rna_def_armature(BlenderRNA *brna)
   RNA_def_struct_sdna(srna, "bArmature");
 
   func = RNA_def_function(srna, "transform", "rna_Armature_transform");
-  RNA_def_function_flag(func, FUNC_USE_MAIN);
   RNA_def_function_ui_description(func, "Transform armature bones by a matrix");
   parm = RNA_def_float_matrix(func, "matrix", 4, 4, NULL, 0.0f, 0.0f, "", "Matrix", 0.0f, 0.0f);
   RNA_def_parameter_flags(parm, 0, PARM_REQUIRED);
diff --git a/source/blender/makesrna/intern/rna_boid.c b/source/blender/makesrna/intern/rna_boid.c
index 8f30a7e0a98..722b7b12271 100644
--- a/source/blender/makesrna/intern/rna_boid.c
+++ b/source/blender/makesrna/intern/rna_boid.c
@@ -130,10 +130,10 @@ static void rna_Boids_reset(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRN
 
     psys->recalc = ID_RECALC_PSYS_RESET;
 
-    DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
+    DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
   }
   else {
-    DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY | ID_RECALC_PSYS_RESET);
+    DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY | ID_RECALC_PSYS_RESET);
   }
 
   WM_main_add_notifier(NC_OBJECT | ND_PARTICLE | NA_EDITED, NULL);
@@ -145,10 +145,10 @@ static void rna_Boids_reset_deps(Main *bmain, Scene *UNUSED(scene), PointerRNA *
 
     psys->recalc = ID_RECALC_PSYS_RESET;
 
-    DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
+    DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
   }
   else {
-    DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY | ID_RECALC_PSYS_RESET);
+    DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY | ID_RECALC_PSYS_RESET);
   }
 
   DEG_relations_tag_update(bmain);
@@ -240,7 +240,7 @@ static void rna_BoidState_active_boid_rule_index_set(struct PointerRNA *ptr, int
 
 static int particle_id_check(PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   return (GS(id->name) == ID_PA);
 }
@@ -250,7 +250,7 @@ static char *rna_BoidSettings_path(PointerRNA *ptr)
   BoidSettings *boids = (BoidSettings *)ptr->data;
 
   if (particle_id_check(ptr)) {
-    ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+    ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
     if (part->boids == boids) {
       return BLI_strdup("boids");
diff --git a/source/blender/makesrna/intern/rna_brush.c b/source/blender/makesrna/intern/rna_brush.c
index 713ddfa0067..14aa9c5d8cf 100644
--- a/source/blender/makesrna/intern/rna_brush.c
+++ b/source/blender/makesrna/intern/rna_brush.c
@@ -434,27 +434,27 @@ static bool rna_BrushCapabilitiesImagePaint_has_radius_get(PointerRNA *ptr)
 
 static PointerRNA rna_Sculpt_tool_capabilities_get(PointerRNA *ptr)
 {
-  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilitiesSculpt, ptr->id.data);
+  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilitiesSculpt, ptr->owner_id);
 }
 
 static PointerRNA rna_Imapaint_tool_capabilities_get(PointerRNA *ptr)
 {
-  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilitiesImagePaint, ptr->id.data);
+  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilitiesImagePaint, ptr->owner_id);
 }
 
 static PointerRNA rna_Vertexpaint_tool_capabilities_get(PointerRNA *ptr)
 {
-  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilitiesVertexPaint, ptr->id.data);
+  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilitiesVertexPaint, ptr->owner_id);
 }
 
 static PointerRNA rna_Weightpaint_tool_capabilities_get(PointerRNA *ptr)
 {
-  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilitiesWeightPaint, ptr->id.data);
+  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilitiesWeightPaint, ptr->owner_id);
 }
 
 static PointerRNA rna_Brush_capabilities_get(PointerRNA *ptr)
 {
-  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilities, ptr->id.data);
+  return rna_pointer_inherit_refine(ptr, &RNA_BrushCapabilities, ptr->owner_id);
 }
 
 static void rna_Brush_reset_icon(Brush *br)
@@ -764,7 +764,7 @@ static void rna_BrushGpencilSettings_use_material_pin_update(bContext *C, Pointe
 {
   ViewLayer *view_layer = CTX_data_view_layer(C);
   Object *ob = OBACT(view_layer);
-  Brush *brush = ptr->id.data;
+  Brush *brush = (Brush *)ptr->owner_id;
 
   if (brush->gpencil_settings->flag & GP_BRUSH_MATERIAL_PINNED) {
     Material *material = give_current_material(ob, ob->actcol);
@@ -1571,6 +1571,16 @@ static void rna_def_brush(BlenderRNA *brna)
       {0, NULL, 0, NULL, NULL},
   };
 
+  static const EnumPropertyItem brush_spacing_unit_items[] = {
+      {0, "VIEW", 0, "View", "Calculate brush spacing relative to the view"},
+      {BRUSH_SCENE_SPACING,
+       "SCENE",
+       0,
+       "Scene",
+       "Calculate brush spacing relative to the scene using the stroke location"},
+      {0, NULL, 0, NULL, NULL},
+  };
+
   static const EnumPropertyItem brush_curve_preset_items[] = {
       {BRUSH_CURVE_CUSTOM, "CUSTOM", ICON_RNDCURVE, "Custom", ""},
       {BRUSH_CURVE_SMOOTH, "SMOOTH", ICON_SMOOTHCURVE, "Smooth", ""},
@@ -1835,6 +1845,17 @@ static void rna_def_brush(BlenderRNA *brna)
                            "Best used on low-poly meshes as it has a performance impact");
   RNA_def_property_update(prop, 0, "rna_Brush_update");
 
+  prop = RNA_def_property(srna, "normal_radius_factor", PROP_FLOAT, PROP_FACTOR);
+  RNA_def_property_float_sdna(prop, NULL, "normal_radius_factor");
+  RNA_def_property_float_default(prop, 0.5f);
+  RNA_def_property_range(prop, 0.0f, 1.0f);
+  RNA_def_property_ui_range(prop, 0.0f, 1.0f, 0.001, 3);
+  RNA_def_property_ui_text(prop,
+                           "Normal Radius",
+                           "Ratio between the brush radius and the radius that is going to be "
+                           "used to sample the normal");
+  RNA_def_property_update(prop, 0, "rna_Brush_update");
+
   prop = RNA_def_property(srna, "stencil_pos", PROP_FLOAT, PROP_XYZ);
   RNA_def_property_float_sdna(prop, NULL, "stencil_pos");
   RNA_def_property_array(prop, 2);
@@ -1920,6 +1941,13 @@ static void rna_def_brush(BlenderRNA *brna)
                            "When locked keep using normal of surface where stroke was initiated");
   RNA_def_property_update(prop, 0, "rna_Brush_update");
 
+  prop = RNA_def_property(srna, "use_scene_spacing", PROP_ENUM, PROP_NONE);
+  RNA_def_property_enum_bitflag_sdna(prop, NULL, "flag");
+  RNA_def_property_enum_items(prop, brush_spacing_unit_items);
+  RNA_def_property_ui_text(
+      prop, "Spacing distance", "Calculate the brush spacing using view or scene distance");
+  RNA_def_property_update(prop, 0, "rna_Brush_update");
+
   prop = RNA_def_property(srna, "use_pressure_strength", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", BRUSH_ALPHA_PRESSURE);
   RNA_def_property_ui_icon(prop, ICON_STYLUS_PRESSURE, 0);
diff --git a/source/blender/makesrna/intern/rna_camera.c b/source/blender/makesrna/intern/rna_camera.c
index bb236fc3ef4..88083f600f8 100644
--- a/source/blender/makesrna/intern/rna_camera.c
+++ b/source/blender/makesrna/intern/rna_camera.c
@@ -43,52 +43,52 @@
 
 static float rna_Camera_angle_get(PointerRNA *ptr)
 {
-  Camera *cam = ptr->id.data;
+  Camera *cam = (Camera *)ptr->owner_id;
   float sensor = BKE_camera_sensor_size(cam->sensor_fit, cam->sensor_x, cam->sensor_y);
   return focallength_to_fov(cam->lens, sensor);
 }
 
 static void rna_Camera_angle_set(PointerRNA *ptr, float value)
 {
-  Camera *cam = ptr->id.data;
+  Camera *cam = (Camera *)ptr->owner_id;
   float sensor = BKE_camera_sensor_size(cam->sensor_fit, cam->sensor_x, cam->sensor_y);
   cam->lens = fov_to_focallength(value, sensor);
 }
 
 static float rna_Camera_angle_x_get(PointerRNA *ptr)
 {
-  Camera *cam = ptr->id.data;
+  Camera *cam = (Camera *)ptr->owner_id;
   return focallength_to_fov(cam->lens, cam->sensor_x);
 }
 
 static void rna_Camera_angle_x_set(PointerRNA *ptr, float value)
 {
-  Camera *cam = ptr->id.data;
+  Camera *cam = (Camera *)ptr->owner_id;
   cam->lens = fov_to_focallength(value, cam->sensor_x);
 }
 
 static float rna_Camera_angle_y_get(PointerRNA *ptr)
 {
-  Camera *cam = ptr->id.data;
+  Camera *cam = (Camera *)ptr->owner_id;
   return focallength_to_fov(cam->lens, cam->sensor_y);
 }
 
 static void rna_Camera_angle_y_set(PointerRNA *ptr, float value)
 {
-  Camera *cam = ptr->id.data;
+  Camera *cam = (Camera *)ptr->owner_id;
   cam->lens = fov_to_focallength(value, cam->sensor_y);
 }
 
 static void rna_Camera_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Camera *camera = (Camera *)ptr->id.data;
+  Camera *camera = (Camera *)ptr->owner_id;
 
   DEG_id_tag_update(&camera->id, 0);
 }
 
 static void rna_Camera_dependency_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Camera *camera = (Camera *)ptr->id.data;
+  Camera *camera = (Camera *)ptr->owner_id;
   DEG_relations_tag_update(bmain);
   DEG_id_tag_update(&camera->id, 0);
 }
@@ -136,8 +136,8 @@ char *rna_CameraDOFSettings_path(PointerRNA *ptr)
    * since the name used is the name of the texture assigned, but the texture
    * may be used multiple times in the same stack
    */
-  if (ptr->id.data) {
-    if (GS(((ID *)ptr->id.data)->name) == ID_CA) {
+  if (ptr->owner_id) {
+    if (GS(ptr->owner_id->name) == ID_CA) {
       return BLI_strdup("dof");
     }
   }
@@ -203,12 +203,14 @@ static void rna_def_camera_background_image(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "ima");
   RNA_def_property_ui_text(prop, "Image", "Image displayed and edited in this space");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, NULL);
 
   prop = RNA_def_property(srna, "clip", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "clip");
   RNA_def_property_ui_text(prop, "MovieClip", "Movie clip displayed and edited in this space");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, NULL);
 
   prop = RNA_def_property(srna, "image_user", PROP_POINTER, PROP_NONE);
@@ -428,6 +430,7 @@ static void rna_def_camera_dof_settings_data(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_pointer_sdna(prop, NULL, "focus_object");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Focus Object", "Use this object to define the depth of field focal point");
   RNA_def_property_update(prop, NC_OBJECT | ND_DRAW, "rna_Camera_dependency_update");
diff --git a/source/blender/makesrna/intern/rna_cloth.c b/source/blender/makesrna/intern/rna_cloth.c
index 548ed656d7b..036aeebda52 100644
--- a/source/blender/makesrna/intern/rna_cloth.c
+++ b/source/blender/makesrna/intern/rna_cloth.c
@@ -45,7 +45,7 @@
 
 static void rna_cloth_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
   WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ob);
@@ -59,7 +59,7 @@ static void rna_cloth_dependency_update(Main *bmain, Scene *scene, PointerRNA *p
 
 static void rna_cloth_pinning_changed(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   /*  ClothSimSettings *settings = (ClothSimSettings *)ptr->data; */
   ClothModifierData *clmd = (ClothModifierData *)modifiers_findByType(ob, eModifierType_Cloth);
 
@@ -311,7 +311,7 @@ static void rna_CollSettings_selfcol_vgroup_set(PointerRNA *ptr, const char *val
 
 static PointerRNA rna_ClothSettings_rest_shape_key_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ClothSimSettings *sim = (ClothSimSettings *)ptr->data;
 
   return rna_object_shapekey_index_get(ob->data, sim->shapekey_rest);
@@ -321,7 +321,7 @@ static void rna_ClothSettings_rest_shape_key_set(PointerRNA *ptr,
                                                  PointerRNA value,
                                                  struct ReportList *UNUSED(reports))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ClothSimSettings *sim = (ClothSimSettings *)ptr->data;
 
   sim->shapekey_rest = rna_object_shapekey_index_set(ob->data, value, sim->shapekey_rest);
@@ -347,7 +347,7 @@ static void rna_ClothSettings_gravity_set(PointerRNA *ptr, const float *values)
 
 static char *rna_ClothSettings_path(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ModifierData *md = modifiers_findByType(ob, eModifierType_Cloth);
 
   if (md) {
@@ -362,7 +362,7 @@ static char *rna_ClothSettings_path(PointerRNA *ptr)
 
 static char *rna_ClothCollisionSettings_path(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ModifierData *md = modifiers_findByType(ob, eModifierType_Cloth);
 
   if (md) {
@@ -891,6 +891,7 @@ static void rna_def_cloth_collision_settings(BlenderRNA *brna)
   prop = RNA_def_property(srna, "collection", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "group");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Collision Collection", "Limit colliders to this Collection");
   RNA_def_property_update(prop, 0, "rna_cloth_dependency_update");
 
diff --git a/source/blender/makesrna/intern/rna_collection.c b/source/blender/makesrna/intern/rna_collection.c
index 12fdebbb8ca..e1c29dc906e 100644
--- a/source/blender/makesrna/intern/rna_collection.c
+++ b/source/blender/makesrna/intern/rna_collection.c
@@ -130,11 +130,11 @@ static bool rna_Collection_objects_override_apply(Main *bmain,
                                                   PointerRNA *UNUSED(ptr_item_storage),
                                                   IDOverrideLibraryPropertyOperation *opop)
 {
-  (void)opop;
   BLI_assert(opop->operation == IDOVERRIDE_LIBRARY_OP_REPLACE &&
              "Unsupported RNA override operation on collections' objects");
+  UNUSED_VARS_NDEBUG(opop);
 
-  Collection *coll_dst = ptr_dst->id.data;
+  Collection *coll_dst = (Collection *)ptr_dst->owner_id;
 
   if (ptr_item_dst->type == NULL || ptr_item_src->type == NULL) {
     BLI_assert(0 && "invalid source or destination object.");
@@ -233,11 +233,11 @@ static bool rna_Collection_children_override_apply(Main *bmain,
                                                    PointerRNA *UNUSED(ptr_item_storage),
                                                    IDOverrideLibraryPropertyOperation *opop)
 {
-  (void)opop;
   BLI_assert(opop->operation == IDOVERRIDE_LIBRARY_OP_REPLACE &&
-             "Unsupported RNA override operation on collections' objects");
+             "Unsupported RNA override operation on collections' children");
+  UNUSED_VARS_NDEBUG(opop);
 
-  Collection *coll_dst = ptr_dst->id.data;
+  Collection *coll_dst = (Collection *)ptr_dst->owner_id;
 
   if (ptr_item_dst->type == NULL || ptr_item_src->type == NULL) {
     BLI_assert(0 && "invalid source or destination sub-collection.");
@@ -482,6 +482,7 @@ void RNA_def_collections(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_ui_text(
       prop, "All Objects", "Objects that are in this collection and its child collections");
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_NO_COMPARISON);
   RNA_def_property_collection_funcs(prop,
                                     "rna_Collection_all_objects_begin",
                                     "rna_iterator_listbase_next",
diff --git a/source/blender/makesrna/intern/rna_color.c b/source/blender/makesrna/intern/rna_color.c
index 6bfd2b9f63b..011e373cc61 100644
--- a/source/blender/makesrna/intern/rna_color.c
+++ b/source/blender/makesrna/intern/rna_color.c
@@ -164,8 +164,8 @@ static char *rna_ColorRamp_path(PointerRNA *ptr)
   char *path = NULL;
 
   /* handle the cases where a single data-block may have 2 ramp types */
-  if (ptr->id.data) {
-    ID *id = ptr->id.data;
+  if (ptr->owner_id) {
+    ID *id = ptr->owner_id;
 
     switch (GS(id->name)) {
       case ID_NT: {
@@ -237,8 +237,8 @@ static char *rna_ColorRampElement_path(PointerRNA *ptr)
 
   /* determine the path from the ID-block to the ramp */
   /* FIXME: this is a very slow way to do it, but it will have to suffice... */
-  if (ptr->id.data) {
-    ID *id = ptr->id.data;
+  if (ptr->owner_id) {
+    ID *id = ptr->owner_id;
 
     switch (GS(id->name)) {
       case ID_NT: {
@@ -286,12 +286,12 @@ static char *rna_ColorRampElement_path(PointerRNA *ptr)
 
 static void rna_ColorRamp_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  if (ptr->id.data) {
-    ID *id = ptr->id.data;
+  if (ptr->owner_id) {
+    ID *id = ptr->owner_id;
 
     switch (GS(id->name)) {
       case ID_MA: {
-        Material *ma = ptr->id.data;
+        Material *ma = (Material *)ptr->owner_id;
 
         DEG_id_tag_update(&ma->id, 0);
         WM_main_add_notifier(NC_MATERIAL | ND_SHADING_DRAW, ma);
@@ -309,20 +309,20 @@ static void rna_ColorRamp_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *
         break;
       }
       case ID_TE: {
-        Tex *tex = ptr->id.data;
+        Tex *tex = (Tex *)ptr->owner_id;
 
         DEG_id_tag_update(&tex->id, 0);
         WM_main_add_notifier(NC_TEXTURE, tex);
         break;
       }
       case ID_LS: {
-        FreestyleLineStyle *linestyle = ptr->id.data;
+        FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->owner_id;
 
         WM_main_add_notifier(NC_LINESTYLE, linestyle);
         break;
       }
       case ID_PA: {
-        ParticleSettings *part = ptr->id.data;
+        ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
         DEG_id_tag_update(&part->id, ID_RECALC_GEOMETRY | ID_RECALC_PSYS_REDO);
         WM_main_add_notifier(NC_OBJECT | ND_PARTICLE | NA_EDITED, part);
@@ -417,7 +417,7 @@ static void rna_ColorManagedDisplaySettings_display_device_update(Main *bmain,
                                                                   Scene *UNUSED(scene),
                                                                   PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   if (!id) {
     return;
@@ -582,7 +582,7 @@ static void rna_ColorManagedColorspaceSettings_reload_update(Main *bmain,
                                                              Scene *UNUSED(scene),
                                                              PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   if (GS(id->name) == ID_IM) {
     Image *ima = (Image *)id;
@@ -657,7 +657,7 @@ static char *rna_ColorManagedInputColorspaceSettings_path(PointerRNA *UNUSED(ptr
 
 static void rna_ColorManagement_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   if (!id) {
     return;
diff --git a/source/blender/makesrna/intern/rna_constraint.c b/source/blender/makesrna/intern/rna_constraint.c
index 6e57d16df27..05154dcc20e 100644
--- a/source/blender/makesrna/intern/rna_constraint.c
+++ b/source/blender/makesrna/intern/rna_constraint.c
@@ -256,6 +256,17 @@ static const EnumPropertyItem track_axis_items[] = {
     {0, NULL, 0, NULL, NULL},
 };
 
+static const EnumPropertyItem euler_order_items[] = {
+    {CONSTRAINT_EULER_AUTO, "AUTO", 0, "Default", "Euler using the default rotation order"},
+    {CONSTRAINT_EULER_XYZ, "XYZ", 0, "XYZ Euler", "Euler using the XYZ rotation order"},
+    {CONSTRAINT_EULER_XZY, "XZY", 0, "XZY Euler", "Euler using the XZY rotation order"},
+    {CONSTRAINT_EULER_YXZ, "YXZ", 0, "YXZ Euler", "Euler using the YXZ rotation order"},
+    {CONSTRAINT_EULER_YZX, "YZX", 0, "YZX Euler", "Euler using the YZX rotation order"},
+    {CONSTRAINT_EULER_ZXY, "ZXY", 0, "ZXY Euler", "Euler using the ZXY rotation order"},
+    {CONSTRAINT_EULER_ZYX, "ZYX", 0, "ZYX Euler", "Euler using the ZYX rotation order"},
+    {0, NULL, 0, NULL, NULL},
+};
+
 #ifdef RNA_RUNTIME
 
 static const EnumPropertyItem space_object_items[] = {
@@ -376,8 +387,8 @@ static void rna_Constraint_name_set(PointerRNA *ptr, const char *value)
   BLI_strncpy_utf8(con->name, value, sizeof(con->name));
 
   /* make sure name is unique */
-  if (ptr->id.data) {
-    Object *ob = ptr->id.data;
+  if (ptr->owner_id) {
+    Object *ob = (Object *)ptr->owner_id;
     ListBase *list = get_constraint_lb(ob, con, NULL);
 
     /* if we have the list, check for unique name, otherwise give up */
@@ -418,7 +429,7 @@ static char *rna_Constraint_do_compute_path(Object *ob, bConstraint *con)
 
 static char *rna_Constraint_path(PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bConstraint *con = ptr->data;
 
   return rna_Constraint_do_compute_path(ob, con);
@@ -426,7 +437,7 @@ static char *rna_Constraint_path(PointerRNA *ptr)
 
 static bConstraint *rna_constraint_from_target(PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bConstraintTarget *tgt = ptr->data;
 
   return BKE_constraint_find_from_target(ob, tgt, NULL);
@@ -434,7 +445,7 @@ static bConstraint *rna_constraint_from_target(PointerRNA *ptr)
 
 static char *rna_ConstraintTarget_path(PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bConstraintTarget *tgt = ptr->data;
   bConstraint *con = rna_constraint_from_target(ptr);
   int index = -1;
@@ -469,29 +480,30 @@ static char *rna_ConstraintTarget_path(PointerRNA *ptr)
 
 static void rna_Constraint_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ED_object_constraint_tag_update(bmain, ptr->id.data, ptr->data);
+  ED_object_constraint_tag_update(bmain, (Object *)ptr->owner_id, ptr->data);
 }
 
 static void rna_Constraint_dependency_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ED_object_constraint_dependency_tag_update(bmain, ptr->id.data, ptr->data);
+  ED_object_constraint_dependency_tag_update(bmain, (Object *)ptr->owner_id, ptr->data);
 }
 
 static void rna_ConstraintTarget_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ED_object_constraint_tag_update(bmain, ptr->id.data, rna_constraint_from_target(ptr));
+  ED_object_constraint_tag_update(bmain, (Object *)ptr->owner_id, rna_constraint_from_target(ptr));
 }
 
 static void rna_ConstraintTarget_dependency_update(Main *bmain,
                                                    Scene *UNUSED(scene),
                                                    PointerRNA *ptr)
 {
-  ED_object_constraint_dependency_tag_update(bmain, ptr->id.data, rna_constraint_from_target(ptr));
+  ED_object_constraint_dependency_tag_update(
+      bmain, (Object *)ptr->owner_id, rna_constraint_from_target(ptr));
 }
 
 static void rna_Constraint_influence_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   if (ob->pose) {
     ob->pose->flag |= (POSE_LOCKED | POSE_DO_UNLOCK);
@@ -523,7 +535,7 @@ static const EnumPropertyItem *rna_Constraint_owner_space_itemf(bContext *UNUSED
                                                                 PropertyRNA *UNUSED(prop),
                                                                 bool *UNUSED(r_free))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bConstraint *con = (bConstraint *)ptr->data;
 
   if (BLI_findindex(&ob->constraints, con) == -1) {
@@ -671,7 +683,7 @@ static bool rna_Constraint_cameraObject_poll(PointerRNA *ptr, PointerRNA value)
   Object *ob = (Object *)value.data;
 
   if (ob) {
-    if (ob->type == OB_CAMERA && ob != (Object *)ptr->id.data) {
+    if (ob->type == OB_CAMERA && ob != (Object *)ptr->owner_id) {
       return 1;
     }
   }
@@ -688,7 +700,7 @@ static void rna_Constraint_followTrack_camera_set(PointerRNA *ptr,
   Object *ob = (Object *)value.data;
 
   if (ob) {
-    if (ob->type == OB_CAMERA && ob != (Object *)ptr->id.data) {
+    if (ob->type == OB_CAMERA && ob != (Object *)ptr->owner_id) {
       data->camera = ob;
       id_lib_extern((ID *)ob);
     }
@@ -707,7 +719,7 @@ static void rna_Constraint_followTrack_depthObject_set(PointerRNA *ptr,
   Object *ob = (Object *)value.data;
 
   if (ob) {
-    if (ob->type == OB_MESH && ob != (Object *)ptr->id.data) {
+    if (ob->type == OB_MESH && ob != (Object *)ptr->owner_id) {
       data->depth_ob = ob;
       id_lib_extern((ID *)ob);
     }
@@ -722,7 +734,7 @@ static bool rna_Constraint_followTrack_depthObject_poll(PointerRNA *ptr, Pointer
   Object *ob = (Object *)value.data;
 
   if (ob) {
-    if (ob->type == OB_MESH && ob != (Object *)ptr->id.data) {
+    if (ob->type == OB_MESH && ob != (Object *)ptr->owner_id) {
       return 1;
     }
   }
@@ -739,7 +751,7 @@ static void rna_Constraint_objectSolver_camera_set(PointerRNA *ptr,
   Object *ob = (Object *)value.data;
 
   if (ob) {
-    if (ob->type == OB_CAMERA && ob != (Object *)ptr->id.data) {
+    if (ob->type == OB_CAMERA && ob != (Object *)ptr->owner_id) {
       data->camera = ob;
       id_lib_extern((ID *)ob);
     }
@@ -1009,7 +1021,7 @@ static void rna_def_constraint_armature_deform(BlenderRNA *brna)
   RNA_def_struct_ui_text(
       srna, "Armature Constraint", "Applies transformations done by the Armature modifier");
   RNA_def_struct_sdna_from(srna, "bArmatureConstraint", "data");
-  RNA_def_struct_ui_icon(srna, ICON_MOD_ARMATURE);
+  RNA_def_struct_ui_icon(srna, ICON_CON_ARMATURE);
 
   prop = RNA_def_property(srna, "targets", PROP_COLLECTION, PROP_NONE);
   RNA_def_property_collection_sdna(prop, NULL, "targets", NULL);
@@ -1323,6 +1335,12 @@ static void rna_def_constraint_rotate_like(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Invert Z", "Invert the Z rotation");
   RNA_def_property_update(prop, NC_OBJECT | ND_CONSTRAINT, "rna_Constraint_update");
 
+  prop = RNA_def_property(srna, "euler_order", PROP_ENUM, PROP_NONE);
+  RNA_def_property_enum_sdna(prop, NULL, "euler_order");
+  RNA_def_property_enum_items(prop, euler_order_items);
+  RNA_def_property_ui_text(prop, "Euler Order", "Explicitly specify the euler rotation order");
+  RNA_def_property_update(prop, NC_OBJECT | ND_CONSTRAINT, "rna_Constraint_update");
+
   prop = RNA_def_property(srna, "use_offset", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", ROTLIKE_OFFSET);
   RNA_def_property_ui_text(prop, "Offset", "Add original rotation into copied rotation");
@@ -1363,6 +1381,14 @@ static void rna_def_constraint_size_like(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Power", "Raise the target's scale to the specified power");
   RNA_def_property_update(prop, NC_OBJECT | ND_CONSTRAINT, "rna_Constraint_update");
 
+  prop = RNA_def_property(srna, "use_make_uniform", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", SIZELIKE_UNIFORM);
+  RNA_def_property_ui_text(prop,
+                           "Make Uniform",
+                           "Redistribute the copied change in volume equally "
+                           "between the three axes of the owner");
+  RNA_def_property_update(prop, NC_OBJECT | ND_CONSTRAINT, "rna_Constraint_update");
+
   prop = RNA_def_property(srna, "use_offset", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", SIZELIKE_OFFSET);
   RNA_def_property_ui_text(prop, "Offset", "Combine original scale with copied scale");
@@ -1876,6 +1902,18 @@ static void rna_def_constraint_transform(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Extrapolate Motion", "Extrapolate ranges");
   RNA_def_property_update(prop, NC_OBJECT | ND_CONSTRAINT, "rna_Constraint_update");
 
+  prop = RNA_def_property(srna, "from_rotation_mode", PROP_ENUM, PROP_NONE);
+  RNA_def_property_enum_sdna(prop, NULL, "from_rotation_mode");
+  RNA_def_property_enum_items(prop, rna_enum_driver_target_rotation_mode_items);
+  RNA_def_property_ui_text(prop, "From Mode", "Specify the type of rotation channels to use");
+  RNA_def_property_update(prop, NC_OBJECT | ND_CONSTRAINT, "rna_Constraint_update");
+
+  prop = RNA_def_property(srna, "to_euler_order", PROP_ENUM, PROP_NONE);
+  RNA_def_property_enum_sdna(prop, NULL, "to_euler_order");
+  RNA_def_property_enum_items(prop, euler_order_items);
+  RNA_def_property_ui_text(prop, "To Order", "Explicitly specify the output euler rotation order");
+  RNA_def_property_update(prop, NC_OBJECT | ND_CONSTRAINT, "rna_Constraint_update");
+
   /* Loc */
   prop = RNA_def_property(srna, "from_min_x", PROP_FLOAT, PROP_DISTANCE);
   RNA_def_property_float_sdna(prop, NULL, "from_min[0]");
@@ -3026,6 +3064,7 @@ void RNA_def_constraint(BlenderRNA *brna)
   RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Disable", "Enable/Disable Constraint");
   RNA_def_property_update(prop, NC_OBJECT | ND_CONSTRAINT, "rna_Constraint_update");
+  RNA_def_property_ui_icon(prop, ICON_HIDE_OFF, -1);
 
   prop = RNA_def_property(srna, "show_expanded", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_flag(prop, PROP_NO_DEG_UPDATE);
diff --git a/source/blender/makesrna/intern/rna_context.c b/source/blender/makesrna/intern/rna_context.c
index 735cd1fd923..579dd92b703 100644
--- a/source/blender/makesrna/intern/rna_context.c
+++ b/source/blender/makesrna/intern/rna_context.c
@@ -178,14 +178,14 @@ static PointerRNA rna_Context_collection_get(PointerRNA *ptr)
 static PointerRNA rna_Context_layer_collection_get(PointerRNA *ptr)
 {
   bContext *C = (bContext *)ptr->data;
-  ptr->id.data = CTX_data_scene(C);
+  ptr->owner_id = &CTX_data_scene(C)->id;
   return rna_pointer_inherit_refine(ptr, &RNA_LayerCollection, CTX_data_layer_collection(C));
 }
 
 static PointerRNA rna_Context_tool_settings_get(PointerRNA *ptr)
 {
   bContext *C = (bContext *)ptr->data;
-  ptr->id.data = CTX_data_scene(C);
+  ptr->owner_id = &CTX_data_scene(C)->id;
   return rna_pointer_inherit_refine(ptr, &RNA_ToolSettings, CTX_data_tool_settings(C));
 }
 
diff --git a/source/blender/makesrna/intern/rna_curve.c b/source/blender/makesrna/intern/rna_curve.c
index 644d0b712a2..a7dac4100db 100644
--- a/source/blender/makesrna/intern/rna_curve.c
+++ b/source/blender/makesrna/intern/rna_curve.c
@@ -326,7 +326,7 @@ static void rna_Curve_texspace_size_set(PointerRNA *ptr, const float *values)
 static void rna_Curve_material_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   *min = 0;
   *max = max_ii(0, cu->totcol - 1);
 }
@@ -347,14 +347,14 @@ static void rna_ChariInfo_material_index_set(PointerRNA *ptr, int value)
 static void rna_Curve_active_textbox_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   *min = 0;
   *max = max_ii(0, cu->totbox - 1);
 }
 
 static void rna_Curve_dimension_set(PointerRNA *ptr, int value)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   if (value == CU_3D) {
     cu->flag |= CU_3D;
   }
@@ -370,7 +370,7 @@ static const EnumPropertyItem *rna_Curve_fill_mode_itemf(bContext *UNUSED(C),
                                                          PropertyRNA *UNUSED(prop),
                                                          bool *UNUSED(r_free))
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
 
   /* cast to quiet warning it IS a const still */
   return (EnumPropertyItem *)((cu->flag & CU_3D) ? curve3d_fill_mode_items :
@@ -388,7 +388,7 @@ static int rna_Nurb_length(PointerRNA *ptr)
 
 static void rna_Nurb_type_set(PointerRNA *ptr, int value)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   Nurb *nu = (Nurb *)ptr->data;
   const int pntsu_prev = nu->pntsu;
 
@@ -418,7 +418,7 @@ static void rna_Curve_update_data_id(Main *UNUSED(bmain), Scene *UNUSED(scene),
 
 static void rna_Curve_update_data(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  rna_Curve_update_data_id(bmain, scene, ptr->id.data);
+  rna_Curve_update_data_id(bmain, scene, ptr->owner_id);
 }
 
 static void rna_Curve_update_deps(Main *bmain, Scene *scene, PointerRNA *ptr)
@@ -429,7 +429,7 @@ static void rna_Curve_update_deps(Main *bmain, Scene *scene, PointerRNA *ptr)
 
 static void rna_Curve_update_points(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   Nurb *nu = curve_nurb_from_point(cu, ptr->data, NULL, NULL);
 
   if (nu) {
@@ -441,7 +441,7 @@ static void rna_Curve_update_points(Main *bmain, Scene *scene, PointerRNA *ptr)
 
 static PointerRNA rna_Curve_bevelObject_get(PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   Object *ob = cu->bevobj;
 
   if (ob) {
@@ -455,7 +455,7 @@ static void rna_Curve_bevelObject_set(PointerRNA *ptr,
                                       PointerRNA value,
                                       struct ReportList *UNUSED(reports))
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   Object *ob = (Object *)value.data;
 
   if (ob) {
@@ -473,7 +473,7 @@ static void rna_Curve_bevelObject_set(PointerRNA *ptr,
 
 static bool rna_Curve_otherObject_poll(PointerRNA *ptr, PointerRNA value)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   Object *ob = (Object *)value.data;
 
   if (ob) {
@@ -487,7 +487,7 @@ static bool rna_Curve_otherObject_poll(PointerRNA *ptr, PointerRNA value)
 
 static PointerRNA rna_Curve_taperObject_get(PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   Object *ob = cu->taperobj;
 
   if (ob) {
@@ -501,7 +501,7 @@ static void rna_Curve_taperObject_set(PointerRNA *ptr,
                                       PointerRNA value,
                                       struct ReportList *UNUSED(reports))
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   Object *ob = (Object *)value.data;
 
   if (ob) {
@@ -519,7 +519,7 @@ static void rna_Curve_taperObject_set(PointerRNA *ptr,
 
 static void rna_Curve_resolution_u_update_data(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   ListBase *nurbs = BKE_curve_nurbs_get(cu);
   Nurb *nu = nurbs->first;
 
@@ -533,7 +533,7 @@ static void rna_Curve_resolution_u_update_data(Main *bmain, Scene *scene, Pointe
 
 static void rna_Curve_resolution_v_update_data(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   ListBase *nurbs = BKE_curve_nurbs_get(cu);
   Nurb *nu = nurbs->first;
 
@@ -547,26 +547,26 @@ static void rna_Curve_resolution_v_update_data(Main *bmain, Scene *scene, Pointe
 
 static float rna_Curve_offset_get(PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   return cu->width - 1.0f;
 }
 
 static void rna_Curve_offset_set(PointerRNA *ptr, float value)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   cu->width = 1.0f + value;
 }
 
 static int rna_Curve_body_length(PointerRNA *ptr);
 static void rna_Curve_body_get(PointerRNA *ptr, char *value)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   BLI_strncpy(value, cu->str, rna_Curve_body_length(ptr) + 1);
 }
 
 static int rna_Curve_body_length(PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   return cu->len;
 }
 
@@ -576,7 +576,7 @@ static void rna_Curve_body_set(PointerRNA *ptr, const char *value)
   size_t len_bytes;
   size_t len_chars = BLI_strlen_utf8_ex(value, &len_bytes);
 
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
 
   cu->len_wchar = len_chars;
   cu->len = len_bytes;
@@ -770,7 +770,7 @@ static void rna_Curve_active_spline_set(PointerRNA *ptr,
 
 static char *rna_Curve_spline_path(PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   ListBase *nubase = BKE_curve_nurbs_get(cu);
   Nurb *nu = ptr->data;
   int index = BLI_findindex(nubase, nu);
@@ -786,7 +786,7 @@ static char *rna_Curve_spline_path(PointerRNA *ptr)
 /* use for both bezier and nurbs */
 static char *rna_Curve_spline_point_path(PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   Nurb *nu;
   void *point = ptr->data;
   int nu_index, pt_index;
@@ -808,7 +808,7 @@ static char *rna_Curve_spline_point_path(PointerRNA *ptr)
 
 static char *rna_TextBox_path(PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   TextBox *tb = ptr->data;
   int index = (int)(tb - cu->tb);
 
@@ -822,13 +822,13 @@ static char *rna_TextBox_path(PointerRNA *ptr)
 
 static void rna_Curve_splines_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   rna_iterator_listbase_begin(iter, BKE_curve_nurbs_get(cu), NULL);
 }
 
 static bool rna_Curve_is_editmode_get(PointerRNA *ptr)
 {
-  Curve *cu = (Curve *)ptr->id.data;
+  Curve *cu = (Curve *)ptr->owner_id;
   const short type = BKE_curve_type_get(cu);
   if (type == OB_FONT) {
     return (cu->editfont != NULL);
@@ -1229,6 +1229,7 @@ static void rna_def_font(BlenderRNA *UNUSED(brna), StructRNA *srna)
   RNA_def_property_pointer_sdna(prop, NULL, "textoncurve");
   RNA_def_property_pointer_funcs(prop, NULL, NULL, NULL, "rna_Curve_otherObject_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Text on Curve", "Curve deforming text object");
   RNA_def_property_update(prop, 0, "rna_Curve_update_deps");
 
@@ -1236,24 +1237,28 @@ static void rna_def_font(BlenderRNA *UNUSED(brna), StructRNA *srna)
   RNA_def_property_pointer_sdna(prop, NULL, "vfont");
   RNA_def_property_ui_text(prop, "Font", "");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Curve_update_data");
 
   prop = RNA_def_property(srna, "font_bold", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "vfontb");
   RNA_def_property_ui_text(prop, "Font Bold", "");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Curve_update_data");
 
   prop = RNA_def_property(srna, "font_italic", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "vfonti");
   RNA_def_property_ui_text(prop, "Font Italic", "");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Curve_update_data");
 
   prop = RNA_def_property(srna, "font_bold_italic", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "vfontbi");
   RNA_def_property_ui_text(prop, "Font Bold Italic", "");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Curve_update_data");
 
   prop = RNA_def_property(srna, "edit_format", PROP_POINTER, PROP_NONE);
@@ -1646,6 +1651,7 @@ static void rna_def_curve(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_pointer_sdna(prop, NULL, "bevobj");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Bevel Object", "Curve object name that defines the bevel shape");
   RNA_def_property_update(prop, 0, "rna_Curve_update_deps");
   RNA_def_property_pointer_funcs(prop,
@@ -1658,6 +1664,7 @@ static void rna_def_curve(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_pointer_sdna(prop, NULL, "taperobj");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Taper Object", "Curve object name that defines the taper (width)");
   RNA_def_property_update(prop, 0, "rna_Curve_update_deps");
diff --git a/source/blender/makesrna/intern/rna_dynamicpaint.c b/source/blender/makesrna/intern/rna_dynamicpaint.c
index 792c4450b9c..30a68a4919e 100644
--- a/source/blender/makesrna/intern/rna_dynamicpaint.c
+++ b/source/blender/makesrna/intern/rna_dynamicpaint.c
@@ -92,7 +92,7 @@ static void rna_DynamicPaint_redoModifier(Main *UNUSED(bmain),
                                           Scene *UNUSED(scene),
                                           PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
 }
 
 static void rna_DynamicPaintSurfaces_updateFrames(Main *UNUSED(bmain),
diff --git a/source/blender/makesrna/intern/rna_fcurve.c b/source/blender/makesrna/intern/rna_fcurve.c
index 222d2fc69da..254f3bc3710 100644
--- a/source/blender/makesrna/intern/rna_fcurve.c
+++ b/source/blender/makesrna/intern/rna_fcurve.c
@@ -132,6 +132,33 @@ const EnumPropertyItem rna_enum_beztriple_interpolation_easing_items[] = {
     {0, NULL, 0, NULL, NULL},
 };
 
+const EnumPropertyItem rna_enum_driver_target_rotation_mode_items[] = {
+    {DTAR_ROTMODE_AUTO, "AUTO", 0, "Auto Euler", "Euler using the rotation order of the target"},
+    {DTAR_ROTMODE_EULER_XYZ, "XYZ", 0, "XYZ Euler", "Euler using the XYZ rotation order"},
+    {DTAR_ROTMODE_EULER_XZY, "XZY", 0, "XZY Euler", "Euler using the XZY rotation order"},
+    {DTAR_ROTMODE_EULER_YXZ, "YXZ", 0, "YXZ Euler", "Euler using the YXZ rotation order"},
+    {DTAR_ROTMODE_EULER_YZX, "YZX", 0, "YZX Euler", "Euler using the YZX rotation order"},
+    {DTAR_ROTMODE_EULER_ZXY, "ZXY", 0, "ZXY Euler", "Euler using the ZXY rotation order"},
+    {DTAR_ROTMODE_EULER_ZYX, "ZYX", 0, "ZYX Euler", "Euler using the ZYX rotation order"},
+    {DTAR_ROTMODE_QUATERNION, "QUATERNION", 0, "Quaternion", "Quaternion rotation"},
+    {DTAR_ROTMODE_SWING_TWIST_X,
+     "SWING_TWIST_X",
+     0,
+     "Swing and X Twist",
+     "Decompose into a swing rotation to aim the X axis, followed by twist around it"},
+    {DTAR_ROTMODE_SWING_TWIST_Y,
+     "SWING_TWIST_Y",
+     0,
+     "Swing and Y Twist",
+     "Decompose into a swing rotation to aim the Y axis, followed by twist around it"},
+    {DTAR_ROTMODE_SWING_TWIST_Z,
+     "SWING_TWIST_Z",
+     0,
+     "Swing and Z Twist",
+     "Decompose into a swing rotation to aim the Z axis, followed by twist around it"},
+    {0, NULL, 0, NULL, NULL},
+};
+
 #ifdef RNA_RUNTIME
 
 #  include "WM_api.h"
@@ -183,7 +210,7 @@ static bool rna_ChannelDriver_is_simple_expression_get(PointerRNA *ptr)
 
 static void rna_ChannelDriver_update_data(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   ChannelDriver *driver = ptr->data;
 
   driver->flag &= ~DRIVER_FLAG_INVALID;
@@ -211,7 +238,7 @@ static void rna_DriverTarget_update_data(Main *bmain, Scene *scene, PointerRNA *
   PointerRNA driverptr;
   ChannelDriver *driver;
   FCurve *fcu;
-  AnimData *adt = BKE_animdata_from_id(ptr->id.data);
+  AnimData *adt = BKE_animdata_from_id(ptr->owner_id);
 
   /* find the driver this belongs to and update it */
   for (fcu = adt->drivers.first; fcu; fcu = fcu->next) {
@@ -221,7 +248,7 @@ static void rna_DriverTarget_update_data(Main *bmain, Scene *scene, PointerRNA *
     if (driver) {
       /* FIXME: need to be able to search targets for required one... */
       /*BLI_findindex(&driver->targets, ptr->data) != -1)  */
-      RNA_pointer_create(ptr->id.data, &RNA_Driver, driver, &driverptr);
+      RNA_pointer_create(ptr->owner_id, &RNA_Driver, driver, &driverptr);
       rna_ChannelDriver_update_data(bmain, scene, &driverptr);
       return;
     }
@@ -466,8 +493,8 @@ static void rna_FCurve_group_set(PointerRNA *ptr,
                                  PointerRNA value,
                                  struct ReportList *UNUSED(reports))
 {
-  ID *pid = (ID *)ptr->id.data;
-  ID *vid = (ID *)value.id.data;
+  ID *pid = ptr->owner_id;
+  ID *vid = value.owner_id;
   FCurve *fcu = ptr->data;
   bAction *act = NULL;
 
@@ -491,7 +518,7 @@ static void rna_FCurve_group_set(PointerRNA *ptr,
   }
   else {
     /* the ID given is the owner of the F-Curve (for drivers) */
-    AnimData *adt = BKE_animdata_from_id(ptr->id.data);
+    AnimData *adt = BKE_animdata_from_id(ptr->owner_id);
     act = (adt) ? adt->action : NULL;
   }
 
@@ -572,7 +599,7 @@ static void rna_FCurve_update_data_ex(ID *id, FCurve *fcu, Main *bmain)
 static void rna_FCurve_update_data(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
   BLI_assert(ptr->type == &RNA_FCurve);
-  rna_FCurve_update_data_ex((ID *)ptr->id.data, (FCurve *)ptr->data, bmain);
+  rna_FCurve_update_data_ex(ptr->owner_id, (FCurve *)ptr->data, bmain);
 }
 
 static void rna_FCurve_update_data_relations(Main *bmain,
@@ -587,7 +614,7 @@ static void rna_FCurve_update_data_relations(Main *bmain,
  */
 static void rna_FCurve_update_eval(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  rna_tag_animation_update(bmain, (ID *)ptr->id.data, true);
+  rna_tag_animation_update(bmain, ptr->owner_id, true);
 }
 
 static PointerRNA rna_FCurve_active_modifier_get(PointerRNA *ptr)
@@ -700,7 +727,7 @@ static void rna_FModifier_blending_range(
 
 static void rna_FModifier_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   FModifier *fcm = (FModifier *)ptr->data;
 
   if (fcm->curve && fcm->type == FMODIFIER_TYPE_CYCLES) {
@@ -1065,7 +1092,7 @@ static void rna_FModifierEnvelope_points_remove(
 
 static void rna_Keyframe_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  rna_tag_animation_update(bmain, (ID *)ptr->id.data, true);
+  rna_tag_animation_update(bmain, ptr->owner_id, true);
 }
 
 #else
@@ -1671,9 +1698,12 @@ static void rna_def_drivertarget(BlenderRNA *brna)
       {DTAR_TRANSCHAN_LOCX, "LOC_X", 0, "X Location", ""},
       {DTAR_TRANSCHAN_LOCY, "LOC_Y", 0, "Y Location", ""},
       {DTAR_TRANSCHAN_LOCZ, "LOC_Z", 0, "Z Location", ""},
+      {0, "", 0, NULL, NULL},
       {DTAR_TRANSCHAN_ROTX, "ROT_X", 0, "X Rotation", ""},
       {DTAR_TRANSCHAN_ROTY, "ROT_Y", 0, "Y Rotation", ""},
       {DTAR_TRANSCHAN_ROTZ, "ROT_Z", 0, "Z Rotation", ""},
+      {DTAR_TRANSCHAN_ROTW, "ROT_W", 0, "W Rotation", ""},
+      {0, "", 0, NULL, NULL},
       {DTAR_TRANSCHAN_SCALEX, "SCALE_X", 0, "X Scale", ""},
       {DTAR_TRANSCHAN_SCALEY, "SCALE_Y", 0, "Y Scale", ""},
       {DTAR_TRANSCHAN_SCALEZ, "SCALE_Z", 0, "Z Scale", ""},
@@ -1749,6 +1779,12 @@ static void rna_def_drivertarget(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Type", "Driver variable type");
   RNA_def_property_update(prop, 0, "rna_DriverTarget_update_data");
 
+  prop = RNA_def_property(srna, "rotation_mode", PROP_ENUM, PROP_NONE);
+  RNA_def_property_enum_sdna(prop, NULL, "rotation_mode");
+  RNA_def_property_enum_items(prop, rna_enum_driver_target_rotation_mode_items);
+  RNA_def_property_ui_text(prop, "Rotation Mode", "Mode for calculating rotation channel values");
+  RNA_def_property_update(prop, 0, "rna_DriverTarget_update_data");
+
   prop = RNA_def_property(srna, "transform_space", PROP_ENUM, PROP_NONE);
   RNA_def_property_enum_bitflag_sdna(prop, NULL, "flag");
   RNA_def_property_enum_items(prop, prop_local_space_items);
diff --git a/source/blender/makesrna/intern/rna_fluidsim.c b/source/blender/makesrna/intern/rna_fluidsim.c
index 45361702384..5d21e718934 100644
--- a/source/blender/makesrna/intern/rna_fluidsim.c
+++ b/source/blender/makesrna/intern/rna_fluidsim.c
@@ -75,7 +75,7 @@ static StructRNA *rna_FluidSettings_refine(struct PointerRNA *ptr)
 
 static void rna_fluid_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
   WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ob);
@@ -101,7 +101,7 @@ static int fluidsim_find_lastframe(Main *bmain, Object *ob, FluidsimSettings *fs
 
 static void rna_fluid_find_enframe(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   FluidsimModifierData *fluidmd = (FluidsimModifierData *)modifiers_findByType(
       ob, eModifierType_Fluidsim);
 
@@ -116,7 +116,7 @@ static void rna_fluid_find_enframe(Main *bmain, Scene *scene, PointerRNA *ptr)
 
 static void rna_FluidSettings_update_type(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   FluidsimModifierData *fluidmd;
   ParticleSystemModifierData *psmd;
   ParticleSystem *psys, *next_psys;
@@ -177,7 +177,7 @@ static void rna_DomainFluidSettings_memory_estimate_get(PointerRNA *ptr, char *v
   (void)ptr;
   value[0] = '\0';
 #  else
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   FluidsimSettings *fss = (FluidsimSettings *)ptr->data;
 
   fluid_estimate_memory(ob, fss, value);
diff --git a/source/blender/makesrna/intern/rna_gpencil.c b/source/blender/makesrna/intern/rna_gpencil.c
index 2671246b589..01ecf2f8cae 100644
--- a/source/blender/makesrna/intern/rna_gpencil.c
+++ b/source/blender/makesrna/intern/rna_gpencil.c
@@ -144,37 +144,14 @@ static EnumPropertyItem rna_enum_gpencil_caps_modes_items[] = {
 
 static void rna_GPencil_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
   WM_main_add_notifier(NC_GPENCIL | NA_EDITED, NULL);
 }
 
 static void rna_GPencil_autolock(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  bGPdata *gpd = (bGPdata *)ptr->id.data;
-  bGPDlayer *gpl = NULL;
-
-  if (gpd->flag & GP_DATA_AUTOLOCK_LAYERS) {
-    bGPDlayer *layer = BKE_gpencil_layer_getactive(gpd);
-
-    /* Lock all other layers */
-    for (gpl = gpd->layers.first; gpl; gpl = gpl->next) {
-      /* unlock active layer */
-      if (gpl == layer) {
-        gpl->flag &= ~GP_LAYER_LOCKED;
-      }
-      else {
-        gpl->flag |= GP_LAYER_LOCKED;
-      }
-    }
-  }
-  else {
-    /* If disable is better unlock all layers by default or it looks there is
-     * a problem in the UI because the user expects all layers will be unlocked
-     */
-    for (gpl = gpd->layers.first; gpl; gpl = gpl->next) {
-      gpl->flag &= ~GP_LAYER_LOCKED;
-    }
-  }
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
+  BKE_gpencil_layer_autolock_set(gpd);
 
   /* standard update */
   rna_GPencil_update(bmain, scene, ptr);
@@ -182,7 +159,7 @@ static void rna_GPencil_autolock(Main *bmain, Scene *scene, PointerRNA *ptr)
 
 static void rna_GPencil_editmode_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bGPdata *gpd = (bGPdata *)ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
   DEG_id_tag_update(&gpd->id, ID_RECALC_TRANSFORM | ID_RECALC_GEOMETRY);
 
   /* Notify all places where GPencil data lives that the editing state is different */
@@ -194,7 +171,7 @@ static void UNUSED_FUNCTION(rna_GPencil_onion_skinning_update)(Main *bmain,
                                                                Scene *scene,
                                                                PointerRNA *ptr)
 {
-  bGPdata *gpd = (bGPdata *)ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
   bGPDlayer *gpl;
   bool enabled = false;
 
@@ -362,7 +339,7 @@ static bool rna_GPencilLayer_is_parented_get(PointerRNA *ptr)
 
 static PointerRNA rna_GPencil_active_layer_get(PointerRNA *ptr)
 {
-  bGPdata *gpd = ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
 
   if (GS(gpd->id.name) == ID_GD) { /* why would this ever be not GD */
     bGPDlayer *gl;
@@ -385,7 +362,7 @@ static void rna_GPencil_active_layer_set(PointerRNA *ptr,
                                          PointerRNA value,
                                          struct ReportList *UNUSED(reports))
 {
-  bGPdata *gpd = ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
 
   /* Don't allow setting active layer to NULL if layers exist
    * as this breaks various tools. Tools should be used instead
@@ -414,7 +391,7 @@ static void rna_GPencil_active_layer_set(PointerRNA *ptr,
 
 static int rna_GPencil_active_layer_index_get(PointerRNA *ptr)
 {
-  bGPdata *gpd = (bGPdata *)ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
   bGPDlayer *gpl = BKE_gpencil_layer_getactive(gpd);
 
   return BLI_findindex(&gpd->layers, gpl);
@@ -422,7 +399,7 @@ static int rna_GPencil_active_layer_index_get(PointerRNA *ptr)
 
 static void rna_GPencil_active_layer_index_set(PointerRNA *ptr, int value)
 {
-  bGPdata *gpd = (bGPdata *)ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
   bGPDlayer *gpl = BLI_findlink(&gpd->layers, value);
 
   BKE_gpencil_layer_setactive(gpd, gpl);
@@ -435,7 +412,7 @@ static void rna_GPencil_active_layer_index_set(PointerRNA *ptr, int value)
 static void rna_GPencil_active_layer_index_range(
     PointerRNA *ptr, int *min, int *max, int *softmin, int *softmax)
 {
-  bGPdata *gpd = (bGPdata *)ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
 
   *min = 0;
   *max = max_ii(0, BLI_listbase_count(&gpd->layers) - 1);
@@ -449,7 +426,7 @@ static const EnumPropertyItem *rna_GPencil_active_layer_itemf(bContext *C,
                                                               PropertyRNA *UNUSED(prop),
                                                               bool *r_free)
 {
-  bGPdata *gpd = (bGPdata *)ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
   bGPDlayer *gpl;
   EnumPropertyItem *item = NULL, item_tmp = {0};
   int totitem = 0;
@@ -478,7 +455,7 @@ static const EnumPropertyItem *rna_GPencil_active_layer_itemf(bContext *C,
 
 static void rna_GPencilLayer_info_set(PointerRNA *ptr, const char *value)
 {
-  bGPdata *gpd = ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
   bGPDlayer *gpl = ptr->data;
 
   char oldname[128] = "";
@@ -539,7 +516,7 @@ static bGPDstroke *rna_GPencil_stroke_point_find_stroke(const bGPdata *gpd,
 
 static void rna_GPencil_stroke_point_select_set(PointerRNA *ptr, const bool value)
 {
-  bGPdata *gpd = ptr->id.data;
+  bGPdata *gpd = (bGPdata *)ptr->owner_id;
   bGPDspoint *pt = ptr->data;
   bGPDstroke *gps = NULL;
 
@@ -1510,6 +1487,7 @@ static void rna_def_gpencil_layer(BlenderRNA *brna)
   prop = RNA_def_property(srna, "parent", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_funcs(prop, NULL, "rna_GPencilLayer_parent_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Parent", "Parent Object");
   RNA_def_property_update(prop, NC_GPENCIL | ND_DATA, "rna_GPencil_update");
 
@@ -1648,7 +1626,7 @@ static void rna_def_gpencil_grid(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "lines", PROP_INT, PROP_NONE);
   RNA_def_property_int_sdna(prop, NULL, "lines");
-  RNA_def_property_range(prop, 0, INT_MAX);
+  RNA_def_property_range(prop, 0, SHRT_MAX);
   RNA_def_property_int_default(prop, GP_DEFAULT_GRID_LINES);
   RNA_def_property_ui_text(
       prop, "Grid Subdivisions", "Number of subdivisions in each side of symmetry line");
@@ -1913,7 +1891,7 @@ static void rna_def_gpencil_data(BlenderRNA *brna)
   prop = RNA_def_property(srna, "zdepth_offset", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "zdepth_offset");
   RNA_def_property_range(prop, 0.0f, 1.0f);
-  RNA_def_property_ui_range(prop, 0.0f, 1.0f, 0.001, 5);
+  RNA_def_property_ui_range(prop, 0.0f, 1.0f, 0.1f, 3);
   RNA_def_property_ui_text(prop, "Surface Offset", "Offset amount when drawing in surface mode");
   RNA_def_property_update(prop, NC_GPENCIL | ND_DATA, "rna_GPencil_update");
 
diff --git a/source/blender/makesrna/intern/rna_gpencil_modifier.c b/source/blender/makesrna/intern/rna_gpencil_modifier.c
index 04175d8cfd9..4a0cf4d7ce2 100644
--- a/source/blender/makesrna/intern/rna_gpencil_modifier.c
+++ b/source/blender/makesrna/intern/rna_gpencil_modifier.c
@@ -252,8 +252,8 @@ static void rna_GpencilModifier_name_set(PointerRNA *ptr, const char *value)
   BLI_strncpy_utf8(gmd->name, value, sizeof(gmd->name));
 
   /* make sure the name is truly unique */
-  if (ptr->id.data) {
-    Object *ob = ptr->id.data;
+  if (ptr->owner_id) {
+    Object *ob = (Object *)ptr->owner_id;
     BKE_gpencil_modifier_unique_name(&ob->greasepencil_modifiers, gmd);
   }
 
@@ -272,8 +272,8 @@ static char *rna_GpencilModifier_path(PointerRNA *ptr)
 
 static void rna_GpencilModifier_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
-  WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ptr->id.data);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
+  WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ptr->owner_id);
 }
 
 static void rna_GpencilModifier_dependency_update(Main *bmain, Scene *scene, PointerRNA *ptr)
@@ -324,7 +324,7 @@ static void greasepencil_modifier_object_set(Object *self,
         PointerRNA *ptr, PointerRNA value, struct ReportList *UNUSED(reports)) \
     { \
       _type##GpencilModifierData *tmd = (_type##GpencilModifierData *)ptr->data; \
-      greasepencil_modifier_object_set(ptr->id.data, &tmd->_prop, _obtype, value); \
+      greasepencil_modifier_object_set((Object *)ptr->owner_id, &tmd->_prop, _obtype, value); \
     }
 
 RNA_GP_MOD_OBJECT_SET(Armature, object, OB_ARMATURE);
@@ -384,6 +384,11 @@ static void rna_def_modifier_gpencilnoise(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
   RNA_def_property_string_sdna(prop, NULL, "vgname");
   RNA_def_property_ui_text(prop, "Vertex Group", "Vertex group name for modulating the deform");
@@ -458,6 +463,11 @@ static void rna_def_modifier_gpencilnoise(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_NOISE_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_NOISE_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -495,6 +505,11 @@ static void rna_def_modifier_gpencilsmooth(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
   RNA_def_property_string_sdna(prop, NULL, "vgname");
   RNA_def_property_ui_text(prop, "Vertex Group", "Vertex group name for modulating the deform");
@@ -549,6 +564,11 @@ static void rna_def_modifier_gpencilsmooth(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_SMOOTH_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_SMOOTH_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -586,6 +606,11 @@ static void rna_def_modifier_gpencilsubdiv(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "level", PROP_INT, PROP_NONE);
   RNA_def_property_int_sdna(prop, NULL, "level");
   RNA_def_property_range(prop, 0, 5);
@@ -608,6 +633,11 @@ static void rna_def_modifier_gpencilsubdiv(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_SUBDIV_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_SUBDIV_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -664,6 +694,11 @@ static void rna_def_modifier_gpencilsimplify(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "factor", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "factor");
   RNA_def_property_range(prop, 0, 100.0);
@@ -682,6 +717,11 @@ static void rna_def_modifier_gpencilsimplify(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_SIMPLIFY_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_SIMPLIFY_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -721,7 +761,7 @@ static void rna_def_modifier_gpencilsimplify(BlenderRNA *brna)
   prop = RNA_def_property(srna, "distance", PROP_FLOAT, PROP_NONE);
   RNA_def_property_float_sdna(prop, NULL, "distance");
   RNA_def_property_range(prop, 0, 100.0f);
-  RNA_def_property_ui_text(prop, "Distance", "Distance between vertex");
+  RNA_def_property_ui_text(prop, "Distance", "Distance between points");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 }
 
@@ -740,6 +780,11 @@ static void rna_def_modifier_gpencilthick(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
   RNA_def_property_string_sdna(prop, NULL, "vgname");
   RNA_def_property_ui_text(prop, "Vertex Group", "Vertex group name for modulating the deform");
@@ -763,6 +808,11 @@ static void rna_def_modifier_gpencilthick(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_THICK_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_THICK_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -815,6 +865,11 @@ static void rna_def_modifier_gpenciloffset(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
   RNA_def_property_string_sdna(prop, NULL, "vgname");
   RNA_def_property_ui_text(prop, "Vertex Group", "Vertex group name for modulating the deform");
@@ -832,6 +887,11 @@ static void rna_def_modifier_gpenciloffset(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_OFFSET_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_OFFSET_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -892,6 +952,11 @@ static void rna_def_modifier_gpenciltint(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "color", PROP_FLOAT, PROP_COLOR_GAMMA);
   RNA_def_property_range(prop, 0.0, 1.0);
   RNA_def_property_float_sdna(prop, NULL, "rgb");
@@ -921,6 +986,11 @@ static void rna_def_modifier_gpenciltint(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_TINT_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_TINT_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -977,7 +1047,7 @@ static void rna_def_modifier_gpenciltime(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "offset", PROP_INT, PROP_NONE);
   RNA_def_property_int_sdna(prop, NULL, "offset");
-  RNA_def_property_range(prop, -INT_MAX, INT_MAX);
+  RNA_def_property_range(prop, SHRT_MIN, SHRT_MAX);
   RNA_def_property_ui_text(
       prop, "Frame Offset", "Number of frames to offset original keyframe number or frame to fix");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
@@ -1039,6 +1109,11 @@ static void rna_def_modifier_gpencilcolor(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "hue", PROP_FLOAT, PROP_NONE);
   RNA_def_property_range(prop, 0.0, 1.0);
   RNA_def_property_ui_range(prop, 0.0, 1.0, 0.1, 3);
@@ -1076,6 +1151,11 @@ static void rna_def_modifier_gpencilcolor(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_COLOR_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_COLOR_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -1118,6 +1198,11 @@ static void rna_def_modifier_gpencilopacity(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
   RNA_def_property_string_sdna(prop, NULL, "vgname");
   RNA_def_property_ui_text(prop, "Vertex Group", "Vertex group name for modulating the deform");
@@ -1146,6 +1231,11 @@ static void rna_def_modifier_gpencilopacity(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_OPACITY_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_OPACITY_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -1183,6 +1273,11 @@ static void rna_def_modifier_gpencilinstance(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "pass_index", PROP_INT, PROP_NONE);
   RNA_def_property_int_sdna(prop, NULL, "pass_index");
   RNA_def_property_range(prop, 0, 100);
@@ -1190,7 +1285,7 @@ static void rna_def_modifier_gpencilinstance(BlenderRNA *brna)
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
   prop = RNA_def_property(srna, "count", PROP_INT, PROP_NONE);
-  RNA_def_property_range(prop, 1, INT_MAX);
+  RNA_def_property_range(prop, 1, SHRT_MAX);
   RNA_def_property_ui_range(prop, 1, 50, 1, -1);
   RNA_def_property_ui_text(prop, "Count", "Number of items");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
@@ -1204,6 +1299,7 @@ static void rna_def_modifier_gpencilinstance(BlenderRNA *brna)
       "Use the location and rotation of another object to determine the distance and "
       "rotational change between arrayed items");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_dependency_update");
 
   prop = RNA_def_property(srna, "offset", PROP_FLOAT, PROP_TRANSLATION);
@@ -1254,7 +1350,7 @@ static void rna_def_modifier_gpencilinstance(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "replace_material", PROP_INT, PROP_NONE);
   RNA_def_property_int_sdna(prop, NULL, "mat_rpl");
-  RNA_def_property_range(prop, 0, INT_MAX);
+  RNA_def_property_range(prop, 0, SHRT_MAX);
   RNA_def_property_ui_text(
       prop,
       "Material",
@@ -1266,6 +1362,11 @@ static void rna_def_modifier_gpencilinstance(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_ARRAY_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_ARRAY_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -1452,6 +1553,11 @@ static void rna_def_modifier_gpencillattice(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
   RNA_def_property_string_sdna(prop, NULL, "vgname");
   RNA_def_property_ui_text(prop, "Vertex Group", "Vertex group name for modulating the deform");
@@ -1469,6 +1575,11 @@ static void rna_def_modifier_gpencillattice(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_LATTICE_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_LATTICE_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -1495,6 +1606,7 @@ static void rna_def_modifier_gpencillattice(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_LatticeGpencilModifier_object_set", NULL, "rna_Lattice_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_dependency_update");
 
   prop = RNA_def_property(srna, "strength", PROP_FLOAT, PROP_NONE);
@@ -1520,6 +1632,11 @@ static void rna_def_modifier_gpencilmirror(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "pass_index", PROP_INT, PROP_NONE);
   RNA_def_property_int_sdna(prop, NULL, "pass_index");
   RNA_def_property_range(prop, 0, 100);
@@ -1531,6 +1648,11 @@ static void rna_def_modifier_gpencilmirror(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_MIRROR_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_MIRROR_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
@@ -1551,6 +1673,7 @@ static void rna_def_modifier_gpencilmirror(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Object", "Object used as center");
   RNA_def_property_pointer_funcs(prop, NULL, "rna_MirrorGpencilModifier_object_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_dependency_update");
 
   prop = RNA_def_property(srna, "use_clip", PROP_BOOLEAN, PROP_NONE);
@@ -1589,6 +1712,7 @@ static void rna_def_modifier_gpencilhook(BlenderRNA *brna)
   RNA_def_property_ui_text(
       prop, "Object", "Parent Object for hook, also recalculates and clears offset");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_pointer_funcs(prop, NULL, "rna_HookGpencilModifier_object_set", NULL, NULL);
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_dependency_update");
 
@@ -1605,6 +1729,11 @@ static void rna_def_modifier_gpencilhook(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Layer", "Layer name");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "material", PROP_STRING, PROP_NONE);
+  RNA_def_property_string_sdna(prop, NULL, "materialname");
+  RNA_def_property_ui_text(prop, "Material", "Material name");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
   RNA_def_property_string_sdna(prop, NULL, "vgname");
   RNA_def_property_ui_text(prop, "Vertex Group", "Vertex group name for modulating the deform");
@@ -1622,6 +1751,11 @@ static void rna_def_modifier_gpencilhook(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Inverse Layers", "Inverse filter");
   RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
 
+  prop = RNA_def_property(srna, "invert_materials", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_HOOK_INVERT_MATERIAL);
+  RNA_def_property_ui_text(prop, "Inverse Materials", "Inverse filter");
+  RNA_def_property_update(prop, 0, "rna_GpencilModifier_update");
+
   prop = RNA_def_property(srna, "invert_material_pass", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_HOOK_INVERT_PASS);
   RNA_def_property_ui_text(prop, "Inverse Pass", "Inverse filter");
diff --git a/source/blender/makesrna/intern/rna_image.c b/source/blender/makesrna/intern/rna_image.c
index 75867b2fe13..517bd2b7276 100644
--- a/source/blender/makesrna/intern/rna_image.c
+++ b/source/blender/makesrna/intern/rna_image.c
@@ -91,7 +91,7 @@ static bool rna_Image_dirty_get(PointerRNA *ptr)
 
 static void rna_Image_source_set(PointerRNA *ptr, int value)
 {
-  Image *ima = ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
 
   if (value != ima->source) {
     ima->source = value;
@@ -105,7 +105,7 @@ static void rna_Image_source_set(PointerRNA *ptr, int value)
 
 static void rna_Image_reload_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Image *ima = ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
   BKE_image_signal(bmain, ima, NULL, IMA_SIGNAL_RELOAD);
   WM_main_add_notifier(NC_IMAGE | NA_EDITED, &ima->id);
   DEG_id_tag_update(&ima->id, 0);
@@ -114,13 +114,13 @@ static void rna_Image_reload_update(Main *bmain, Scene *UNUSED(scene), PointerRN
 
 static void rna_Image_generated_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Image *ima = ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
   BKE_image_signal(bmain, ima, NULL, IMA_SIGNAL_FREE);
 }
 
 static void rna_Image_colormanage_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Image *ima = ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
   BKE_image_signal(bmain, ima, NULL, IMA_SIGNAL_COLORMANAGE);
   DEG_id_tag_update(&ima->id, 0);
   DEG_id_tag_update(&ima->id, ID_RECALC_EDITORS);
@@ -130,7 +130,7 @@ static void rna_Image_colormanage_update(Main *bmain, Scene *UNUSED(scene), Poin
 
 static void rna_Image_views_format_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Image *ima = ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
   ImBuf *ibuf;
   void *lock;
 
@@ -148,7 +148,7 @@ static void rna_Image_views_format_update(Main *bmain, Scene *scene, PointerRNA
 static void rna_ImageUser_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
   ImageUser *iuser = ptr->data;
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   BKE_image_user_frame_calc(NULL, iuser, scene->r.cfra);
 
@@ -173,10 +173,10 @@ static void rna_ImageUser_relations_update(Main *bmain, Scene *scene, PointerRNA
 
 static char *rna_ImageUser_path(PointerRNA *ptr)
 {
-  if (ptr->id.data) {
+  if (ptr->owner_id) {
     /* ImageUser *iuser = ptr->data; */
 
-    switch (GS(((ID *)ptr->id.data)->name)) {
+    switch (GS(ptr->owner_id->name)) {
       case ID_OB:
       case ID_TE: {
         return BLI_strdup("image_user");
@@ -331,7 +331,7 @@ static int rna_Image_depth_get(PointerRNA *ptr)
 
 static int rna_Image_frame_duration_get(PointerRNA *ptr)
 {
-  Image *ima = ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
   int duration = 1;
 
   if (BKE_image_has_anim(ima)) {
@@ -352,7 +352,7 @@ static int rna_Image_frame_duration_get(PointerRNA *ptr)
 
 static int rna_Image_pixels_get_length(PointerRNA *ptr, int length[RNA_MAX_ARRAY_DIMENSION])
 {
-  Image *ima = ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
   ImBuf *ibuf;
   void *lock;
 
@@ -372,7 +372,7 @@ static int rna_Image_pixels_get_length(PointerRNA *ptr, int length[RNA_MAX_ARRAY
 
 static void rna_Image_pixels_get(PointerRNA *ptr, float *values)
 {
-  Image *ima = ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
   ImBuf *ibuf;
   void *lock;
   int i, size;
@@ -397,7 +397,7 @@ static void rna_Image_pixels_get(PointerRNA *ptr, float *values)
 
 static void rna_Image_pixels_set(PointerRNA *ptr, const float *values)
 {
-  Image *ima = ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
   ImBuf *ibuf;
   void *lock;
   int i, size;
@@ -463,7 +463,7 @@ static bool rna_Image_is_float_get(PointerRNA *ptr)
 
 static PointerRNA rna_Image_packed_file_get(PointerRNA *ptr)
 {
-  Image *ima = (Image *)ptr->id.data;
+  Image *ima = (Image *)ptr->owner_id;
 
   if (BKE_image_has_packedfile(ima)) {
     ImagePackedFile *imapf = ima->packedfiles.first;
@@ -485,7 +485,7 @@ static void rna_RenderSlot_clear(ID *id, RenderSlot *slot, ImageUser *iuser)
 
 static PointerRNA rna_render_slots_active_get(PointerRNA *ptr)
 {
-  Image *image = (Image *)ptr->id.data;
+  Image *image = (Image *)ptr->owner_id;
   RenderSlot *render_slot = BKE_image_get_renderslot(image, image->render_slot);
 
   return rna_pointer_inherit_refine(ptr, &RNA_RenderSlot, render_slot);
@@ -495,8 +495,8 @@ static void rna_render_slots_active_set(PointerRNA *ptr,
                                         PointerRNA value,
                                         struct ReportList *UNUSED(reports))
 {
-  Image *image = (Image *)ptr->id.data;
-  if (value.id.data == image) {
+  Image *image = (Image *)ptr->owner_id;
+  if (value.owner_id == &image->id) {
     RenderSlot *slot = (RenderSlot *)value.data;
     int index = BLI_findindex(&image->renderslots, slot);
     if (index != -1) {
@@ -507,13 +507,13 @@ static void rna_render_slots_active_set(PointerRNA *ptr,
 
 static int rna_render_slots_active_index_get(PointerRNA *ptr)
 {
-  Image *image = (Image *)ptr->id.data;
+  Image *image = (Image *)ptr->owner_id;
   return image->render_slot;
 }
 
 static void rna_render_slots_active_index_set(PointerRNA *ptr, int value)
 {
-  Image *image = (Image *)ptr->id.data;
+  Image *image = (Image *)ptr->owner_id;
   int num_slots = BLI_listbase_count(&image->renderslots);
   image->render_slot = value;
   CLAMP(image->render_slot, 0, num_slots - 1);
@@ -522,7 +522,7 @@ static void rna_render_slots_active_index_set(PointerRNA *ptr, int value)
 static void rna_render_slots_active_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Image *image = (Image *)ptr->id.data;
+  Image *image = (Image *)ptr->owner_id;
   *min = 0;
   *max = max_ii(0, BLI_listbase_count(&image->renderslots) - 1);
 }
diff --git a/source/blender/makesrna/intern/rna_internal.h b/source/blender/makesrna/intern/rna_internal.h
index 96cdc06d901..e5058626180 100644
--- a/source/blender/makesrna/intern/rna_internal.h
+++ b/source/blender/makesrna/intern/rna_internal.h
@@ -29,15 +29,12 @@
 
 #define RNA_MAGIC ((int)~0)
 
-struct Depsgraph;
 struct FreestyleSettings;
 struct ID;
 struct IDOverrideLibrary;
-struct IDOverrideLibraryProperty;
 struct IDOverrideLibraryPropertyOperation;
 struct IDProperty;
 struct Main;
-struct Mesh;
 struct Object;
 struct ReportList;
 struct SDNA;
@@ -227,7 +224,7 @@ bool rna_AnimaData_override_apply(struct Main *bmain,
 void rna_def_animviz_common(struct StructRNA *srna);
 void rna_def_motionpath_common(struct StructRNA *srna);
 
-void rna_def_bone_curved_common(struct StructRNA *srna, bool is_posebone);
+void rna_def_bone_curved_common(struct StructRNA *srna, bool is_posebone, bool is_editbone);
 
 void rna_def_texmat_common(struct StructRNA *srna, const char *texspace_editable);
 void rna_def_mtex_common(struct BlenderRNA *brna,
diff --git a/source/blender/makesrna/intern/rna_internal_types.h b/source/blender/makesrna/intern/rna_internal_types.h
index 60cf443383e..7c6a471b1d7 100644
--- a/source/blender/makesrna/intern/rna_internal_types.h
+++ b/source/blender/makesrna/intern/rna_internal_types.h
@@ -31,7 +31,6 @@ struct ContainerRNA;
 struct FunctionRNA;
 struct GHash;
 struct IDOverrideLibrary;
-struct IDOverrideLibraryProperty;
 struct IDOverrideLibraryPropertyOperation;
 struct IDProperty;
 struct Main;
diff --git a/source/blender/makesrna/intern/rna_key.c b/source/blender/makesrna/intern/rna_key.c
index 8ab2127d898..46c3adbd6ef 100644
--- a/source/blender/makesrna/intern/rna_key.c
+++ b/source/blender/makesrna/intern/rna_key.c
@@ -87,8 +87,8 @@ static void rna_ShapeKey_name_set(PointerRNA *ptr, const char *value)
   BLI_strncpy_utf8(kb->name, value, sizeof(kb->name));
 
   /* make sure the name is truly unique */
-  if (ptr->id.data) {
-    Key *key = rna_ShapeKey_find_key(ptr->id.data);
+  if (ptr->owner_id) {
+    Key *key = rna_ShapeKey_find_key(ptr->owner_id);
     BLI_uniquename(&key->block,
                    kb,
                    CTX_DATA_(BLT_I18NCONTEXT_ID_SHAPEKEY, "Key"),
@@ -174,7 +174,7 @@ static void rna_ShapeKey_slider_max_set(PointerRNA *ptr, float value)
 
 static Mesh *rna_KeyBlock_normals_get_mesh(PointerRNA *ptr, ID *id)
 {
-  Key *key = rna_ShapeKey_find_key((id == NULL && ptr != NULL) ? ptr->id.data : id);
+  Key *key = rna_ShapeKey_find_key((id == NULL && ptr != NULL) ? ptr->owner_id : id);
   id = key ? key->from : NULL;
 
   if (id != NULL) {
@@ -315,7 +315,7 @@ static PointerRNA rna_ShapeKey_relative_key_get(PointerRNA *ptr)
 {
   KeyBlock *kb = (KeyBlock *)ptr->data;
 
-  return rna_object_shapekey_index_get(ptr->id.data, kb->relative);
+  return rna_object_shapekey_index_get(ptr->owner_id, kb->relative);
 }
 
 static void rna_ShapeKey_relative_key_set(PointerRNA *ptr,
@@ -324,7 +324,7 @@ static void rna_ShapeKey_relative_key_set(PointerRNA *ptr,
 {
   KeyBlock *kb = (KeyBlock *)ptr->data;
 
-  kb->relative = rna_object_shapekey_index_set(ptr->id.data, value, kb->relative);
+  kb->relative = rna_object_shapekey_index_set(ptr->owner_id, value, kb->relative);
 }
 
 static void rna_ShapeKeyPoint_co_get(PointerRNA *ptr, float *values)
@@ -565,7 +565,7 @@ static void rna_ShapeKey_data_begin_mixed(CollectionPropertyIterator *iter,
 
 static void rna_ShapeKey_data_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  Key *key = rna_ShapeKey_find_key(ptr->id.data);
+  Key *key = rna_ShapeKey_find_key(ptr->owner_id);
   KeyBlock *kb = (KeyBlock *)ptr->data;
   int tot = kb->totelem, size = key->elemsize;
 
@@ -602,7 +602,7 @@ static void rna_ShapeKey_data_begin(CollectionPropertyIterator *iter, PointerRNA
 
 static int rna_ShapeKey_data_length(PointerRNA *ptr)
 {
-  Key *key = rna_ShapeKey_find_key(ptr->id.data);
+  Key *key = rna_ShapeKey_find_key(ptr->owner_id);
   KeyBlock *kb = (KeyBlock *)ptr->data;
   int tot = kb->totelem;
 
@@ -615,7 +615,7 @@ static int rna_ShapeKey_data_length(PointerRNA *ptr)
 
 static PointerRNA rna_ShapeKey_data_get(CollectionPropertyIterator *iter)
 {
-  Key *key = rna_ShapeKey_find_key(iter->parent.id.data);
+  Key *key = rna_ShapeKey_find_key(iter->parent.owner_id);
   void *ptr = rna_iterator_array_get(iter);
   StructRNA *type = &RNA_ShapeKeyPoint;
 
@@ -637,7 +637,7 @@ static PointerRNA rna_ShapeKey_data_get(CollectionPropertyIterator *iter)
 
 int rna_ShapeKey_data_lookup_int(PointerRNA *ptr, int index, PointerRNA *r_ptr)
 {
-  Key *key = rna_ShapeKey_find_key(ptr->id.data);
+  Key *key = rna_ShapeKey_find_key(ptr->owner_id);
   KeyBlock *kb = (KeyBlock *)ptr->data;
   int elemsize = key->elemsize;
   char *databuf = kb->data;
@@ -672,7 +672,7 @@ int rna_ShapeKey_data_lookup_int(PointerRNA *ptr, int index, PointerRNA *r_ptr)
 static char *rna_ShapeKey_path(PointerRNA *ptr)
 {
   KeyBlock *kb = (KeyBlock *)ptr->data;
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   char name_esc[sizeof(kb->name) * 2];
 
   BLI_strescape(name_esc, kb->name, sizeof(name_esc));
@@ -687,7 +687,7 @@ static char *rna_ShapeKey_path(PointerRNA *ptr)
 
 static void rna_Key_update_data(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Key *key = ptr->id.data;
+  Key *key = (Key *)ptr->owner_id;
   Object *ob;
 
   for (ob = bmain->objects.first; ob; ob = ob->id.next) {
@@ -756,8 +756,8 @@ static int rna_ShapeKeyPoint_get_index(Key *key, KeyBlock *kb, float *point)
 
 static char *rna_ShapeKeyPoint_path(PointerRNA *ptr)
 {
-  ID *id = (ID *)ptr->id.data;
-  Key *key = rna_ShapeKey_find_key(ptr->id.data);
+  ID *id = ptr->owner_id;
+  Key *key = rna_ShapeKey_find_key(ptr->owner_id);
   KeyBlock *kb;
   float *point = (float *)ptr->data;
 
@@ -1034,6 +1034,7 @@ static void rna_def_key(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "user", PROP_POINTER, PROP_NONE);
   RNA_def_property_flag(prop, PROP_NEVER_NULL);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_NO_COMPARISON);
   RNA_def_property_pointer_sdna(prop, NULL, "from");
   RNA_def_property_ui_text(prop, "User", "Data-block using these shape keys");
 
diff --git a/source/blender/makesrna/intern/rna_lattice.c b/source/blender/makesrna/intern/rna_lattice.c
index 5a8223af28a..34f5b6caee1 100644
--- a/source/blender/makesrna/intern/rna_lattice.c
+++ b/source/blender/makesrna/intern/rna_lattice.c
@@ -49,7 +49,7 @@
 
 static void rna_LatticePoint_co_get(PointerRNA *ptr, float *values)
 {
-  Lattice *lt = (Lattice *)ptr->id.data;
+  Lattice *lt = (Lattice *)ptr->owner_id;
   BPoint *bp = (BPoint *)ptr->data;
   int index = bp - lt->def;
   int u, v, w;
@@ -63,7 +63,7 @@ static void rna_LatticePoint_co_get(PointerRNA *ptr, float *values)
 
 static void rna_LatticePoint_groups_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  Lattice *lt = (Lattice *)ptr->id.data;
+  Lattice *lt = (Lattice *)ptr->owner_id;
 
   if (lt->dvert) {
     BPoint *bp = (BPoint *)ptr->data;
@@ -95,7 +95,7 @@ static void rna_Lattice_points_begin(CollectionPropertyIterator *iter, PointerRN
 
 static void rna_Lattice_update_data(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   DEG_id_tag_update(id, 0);
   WM_main_add_notifier(NC_GEOM | ND_DATA, id);
@@ -109,8 +109,8 @@ static void rna_Lattice_update_data_editlatt(Main *UNUSED(bmain),
                                              Scene *UNUSED(scene),
                                              PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
-  Lattice *lt = (Lattice *)ptr->id.data;
+  ID *id = ptr->owner_id;
+  Lattice *lt = (Lattice *)ptr->owner_id;
 
   if (lt->editlatt) {
     Lattice *lt_em = lt->editlatt->latt;
@@ -127,7 +127,7 @@ static void rna_Lattice_update_data_editlatt(Main *UNUSED(bmain),
 
 static void rna_Lattice_update_size(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Lattice *lt = ptr->id.data;
+  Lattice *lt = (Lattice *)ptr->owner_id;
   Object *ob;
   int newu, newv, neww;
 
@@ -224,7 +224,7 @@ static void rna_Lattice_vg_name_set(PointerRNA *ptr, const char *value)
 /* annoying, but is a consequence of RNA structures... */
 static char *rna_LatticePoint_path(PointerRNA *ptr)
 {
-  Lattice *lt = (Lattice *)ptr->id.data;
+  Lattice *lt = (Lattice *)ptr->owner_id;
   void *point = ptr->data;
   BPoint *points = NULL;
 
@@ -251,7 +251,7 @@ static char *rna_LatticePoint_path(PointerRNA *ptr)
 
 static bool rna_Lattice_is_editmode_get(PointerRNA *ptr)
 {
-  Lattice *lt = (Lattice *)ptr->id.data;
+  Lattice *lt = (Lattice *)ptr->owner_id;
   return (lt->editlatt != NULL);
 }
 
diff --git a/source/blender/makesrna/intern/rna_layer.c b/source/blender/makesrna/intern/rna_layer.c
index c9b51707759..1ad006cbc37 100644
--- a/source/blender/makesrna/intern/rna_layer.c
+++ b/source/blender/makesrna/intern/rna_layer.c
@@ -152,7 +152,7 @@ static int rna_ViewLayer_objects_selected_skip(CollectionPropertyIterator *iter,
 
 static PointerRNA rna_ViewLayer_depsgraph_get(PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   if (GS(id->name) == ID_SCE) {
     Scene *scene = (Scene *)id;
     ViewLayer *view_layer = (ViewLayer *)ptr->data;
@@ -274,7 +274,7 @@ static void rna_LayerCollection_exclude_update_recursive(ListBase *lb, const boo
 
 static void rna_LayerCollection_exclude_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   LayerCollection *lc = (LayerCollection *)ptr->data;
   ViewLayer *view_layer = BKE_view_layer_find_from_collection(scene, lc);
 
@@ -294,7 +294,7 @@ static void rna_LayerCollection_exclude_update(Main *bmain, Scene *UNUSED(scene)
 
 static void rna_LayerCollection_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   LayerCollection *lc = (LayerCollection *)ptr->data;
   ViewLayer *view_layer = BKE_view_layer_find_from_collection(scene, lc);
 
diff --git a/source/blender/makesrna/intern/rna_light.c b/source/blender/makesrna/intern/rna_light.c
index 79ba5970007..8640c35f1b8 100644
--- a/source/blender/makesrna/intern/rna_light.c
+++ b/source/blender/makesrna/intern/rna_light.c
@@ -77,7 +77,7 @@ static StructRNA *rna_Light_refine(struct PointerRNA *ptr)
 
 static void rna_Light_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Light *la = ptr->id.data;
+  Light *la = (Light *)ptr->owner_id;
 
   DEG_id_tag_update(&la->id, 0);
   WM_main_add_notifier(NC_LAMP | ND_LIGHTING, la);
@@ -85,7 +85,7 @@ static void rna_Light_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerR
 
 static void rna_Light_draw_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Light *la = ptr->id.data;
+  Light *la = (Light *)ptr->owner_id;
 
   DEG_id_tag_update(&la->id, 0);
   WM_main_add_notifier(NC_LAMP | ND_LIGHTING_DRAW, la);
diff --git a/source/blender/makesrna/intern/rna_lightprobe.c b/source/blender/makesrna/intern/rna_lightprobe.c
index b6d6fac587e..463cefdf3f0 100644
--- a/source/blender/makesrna/intern/rna_lightprobe.c
+++ b/source/blender/makesrna/intern/rna_lightprobe.c
@@ -43,7 +43,7 @@
 
 static void rna_LightProbe_recalc(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
 }
 
 #else
@@ -212,6 +212,7 @@ static void rna_def_lightprobe(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "Collection");
   RNA_def_property_pointer_sdna(prop, NULL, "visibility_grp");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Visibility Collection", "Restrict objects visible for this probe");
   RNA_def_property_update(prop, NC_MATERIAL | ND_SHADING, "rna_LightProbe_recalc");
diff --git a/source/blender/makesrna/intern/rna_linestyle.c b/source/blender/makesrna/intern/rna_linestyle.c
index 936a5e6a53d..f6cfc33c82e 100644
--- a/source/blender/makesrna/intern/rna_linestyle.c
+++ b/source/blender/makesrna/intern/rna_linestyle.c
@@ -287,7 +287,7 @@ static char *rna_LineStyle_geometry_modifier_path(PointerRNA *ptr)
 
 static void rna_LineStyleColorModifier_name_set(PointerRNA *ptr, const char *value)
 {
-  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->id.data;
+  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->owner_id;
   LineStyleModifier *m = (LineStyleModifier *)ptr->data;
 
   BLI_strncpy_utf8(m->name, value, sizeof(m->name));
@@ -301,7 +301,7 @@ static void rna_LineStyleColorModifier_name_set(PointerRNA *ptr, const char *val
 
 static void rna_LineStyleAlphaModifier_name_set(PointerRNA *ptr, const char *value)
 {
-  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->id.data;
+  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->owner_id;
   LineStyleModifier *m = (LineStyleModifier *)ptr->data;
 
   BLI_strncpy_utf8(m->name, value, sizeof(m->name));
@@ -315,7 +315,7 @@ static void rna_LineStyleAlphaModifier_name_set(PointerRNA *ptr, const char *val
 
 static void rna_LineStyleThicknessModifier_name_set(PointerRNA *ptr, const char *value)
 {
-  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->id.data;
+  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->owner_id;
   LineStyleModifier *m = (LineStyleModifier *)ptr->data;
 
   BLI_strncpy_utf8(m->name, value, sizeof(m->name));
@@ -329,7 +329,7 @@ static void rna_LineStyleThicknessModifier_name_set(PointerRNA *ptr, const char
 
 static void rna_LineStyleGeometryModifier_name_set(PointerRNA *ptr, const char *value)
 {
-  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->id.data;
+  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->owner_id;
   LineStyleModifier *m = (LineStyleModifier *)ptr->data;
 
   BLI_strncpy_utf8(m->name, value, sizeof(m->name));
@@ -343,13 +343,13 @@ static void rna_LineStyleGeometryModifier_name_set(PointerRNA *ptr, const char *
 
 static void rna_LineStyle_mtex_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->id.data;
+  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->owner_id;
   rna_iterator_array_begin(iter, (void *)linestyle->mtex, sizeof(MTex *), MAX_MTEX, 0, NULL);
 }
 
 static PointerRNA rna_LineStyle_active_texture_get(PointerRNA *ptr)
 {
-  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->id.data;
+  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->owner_id;
   Tex *tex;
 
   tex = give_current_linestyle_texture(linestyle);
@@ -360,14 +360,14 @@ static void rna_LineStyle_active_texture_set(PointerRNA *ptr,
                                              PointerRNA value,
                                              struct ReportList *UNUSED(reports))
 {
-  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->id.data;
+  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->owner_id;
 
   set_current_linestyle_texture(linestyle, value.data);
 }
 
 static void rna_LineStyle_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  FreestyleLineStyle *linestyle = ptr->id.data;
+  FreestyleLineStyle *linestyle = (FreestyleLineStyle *)ptr->owner_id;
 
   DEG_id_tag_update(&linestyle->id, 0);
   WM_main_add_notifier(NC_LINESTYLE, linestyle);
diff --git a/source/blender/makesrna/intern/rna_mask.c b/source/blender/makesrna/intern/rna_mask.c
index abbbd04143c..adc041174aa 100644
--- a/source/blender/makesrna/intern/rna_mask.c
+++ b/source/blender/makesrna/intern/rna_mask.c
@@ -58,7 +58,7 @@
 
 static void rna_Mask_update_data(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Mask *mask = ptr->id.data;
+  Mask *mask = (Mask *)ptr->owner_id;
 
   WM_main_add_notifier(NC_MASK | ND_DATA, mask);
   DEG_id_tag_update(&mask->id, 0);
@@ -147,21 +147,21 @@ static void rna_MaskParent_id_type_set(PointerRNA *ptr, int value)
 
 static void rna_Mask_layers_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  Mask *mask = (Mask *)ptr->id.data;
+  Mask *mask = (Mask *)ptr->owner_id;
 
   rna_iterator_listbase_begin(iter, &mask->masklayers, NULL);
 }
 
 static int rna_Mask_layer_active_index_get(PointerRNA *ptr)
 {
-  Mask *mask = (Mask *)ptr->id.data;
+  Mask *mask = (Mask *)ptr->owner_id;
 
   return mask->masklay_act;
 }
 
 static void rna_Mask_layer_active_index_set(PointerRNA *ptr, int value)
 {
-  Mask *mask = (Mask *)ptr->id.data;
+  Mask *mask = (Mask *)ptr->owner_id;
 
   mask->masklay_act = value;
 }
@@ -169,7 +169,7 @@ static void rna_Mask_layer_active_index_set(PointerRNA *ptr, int value)
 static void rna_Mask_layer_active_index_range(
     PointerRNA *ptr, int *min, int *max, int *softmin, int *softmax)
 {
-  Mask *mask = (Mask *)ptr->id.data;
+  Mask *mask = (Mask *)ptr->owner_id;
 
   *min = 0;
   *max = max_ii(0, mask->masklay_tot - 1);
@@ -188,7 +188,7 @@ static char *rna_MaskLayer_path(PointerRNA *ptr)
 
 static PointerRNA rna_Mask_layer_active_get(PointerRNA *ptr)
 {
-  Mask *mask = (Mask *)ptr->id.data;
+  Mask *mask = (Mask *)ptr->owner_id;
   MaskLayer *masklay = BKE_mask_layer_active(mask);
 
   return rna_pointer_inherit_refine(ptr, &RNA_MaskLayer, masklay);
@@ -198,7 +198,7 @@ static void rna_Mask_layer_active_set(PointerRNA *ptr,
                                       PointerRNA value,
                                       struct ReportList *UNUSED(reports))
 {
-  Mask *mask = (Mask *)ptr->id.data;
+  Mask *mask = (Mask *)ptr->owner_id;
   MaskLayer *masklay = (MaskLayer *)value.data;
 
   BKE_mask_layer_active_set(mask, masklay);
@@ -213,7 +213,7 @@ static void rna_MaskLayer_splines_begin(CollectionPropertyIterator *iter, Pointe
 
 static void rna_MaskLayer_name_set(PointerRNA *ptr, const char *value)
 {
-  Mask *mask = (Mask *)ptr->id.data;
+  Mask *mask = (Mask *)ptr->owner_id;
   MaskLayer *masklay = (MaskLayer *)ptr->data;
   char oldname[sizeof(masklay->name)], newname[sizeof(masklay->name)];
 
@@ -351,7 +351,7 @@ static void rna_MaskSplinePoint_handle_type_set(PointerRNA *ptr, int value)
 {
   MaskSplinePoint *point = (MaskSplinePoint *)ptr->data;
   BezTriple *bezt = &point->bezt;
-  MaskSpline *spline = mask_spline_from_point((Mask *)ptr->id.data, point);
+  MaskSpline *spline = mask_spline_from_point((Mask *)ptr->owner_id, point);
 
   bezt->h1 = bezt->h2 = value;
   mask_point_check_stick(point);
@@ -370,7 +370,7 @@ static void rna_MaskSplinePoint_handle_left_type_set(PointerRNA *ptr, int value)
 {
   MaskSplinePoint *point = (MaskSplinePoint *)ptr->data;
   BezTriple *bezt = &point->bezt;
-  MaskSpline *spline = mask_spline_from_point((Mask *)ptr->id.data, point);
+  MaskSpline *spline = mask_spline_from_point((Mask *)ptr->owner_id, point);
 
   bezt->h1 = value;
   mask_point_check_stick(point);
@@ -389,7 +389,7 @@ static void rna_MaskSplinePoint_handle_right_type_set(PointerRNA *ptr, int value
 {
   MaskSplinePoint *point = (MaskSplinePoint *)ptr->data;
   BezTriple *bezt = &point->bezt;
-  MaskSpline *spline = mask_spline_from_point((Mask *)ptr->id.data, point);
+  MaskSpline *spline = mask_spline_from_point((Mask *)ptr->owner_id, point);
 
   bezt->h2 = value;
   mask_point_check_stick(point);
diff --git a/source/blender/makesrna/intern/rna_material.c b/source/blender/makesrna/intern/rna_material.c
index 3a7e58fbaa4..885df13724f 100644
--- a/source/blender/makesrna/intern/rna_material.c
+++ b/source/blender/makesrna/intern/rna_material.c
@@ -39,7 +39,7 @@ const EnumPropertyItem rna_enum_ramp_blend_items[] = {
     {0, "", ICON_NONE, NULL, NULL},
     {MA_RAMP_DARK, "DARKEN", 0, "Darken", ""},
     {MA_RAMP_MULT, "MULTIPLY", 0, "Multiply", ""},
-    {MA_RAMP_BURN, "BURN", 0, "Burn", ""},
+    {MA_RAMP_BURN, "BURN", 0, "Color Burn", ""},
     {0, "", ICON_NONE, NULL, NULL},
     {MA_RAMP_LIGHT, "LIGHTEN", 0, "Lighten", ""},
     {MA_RAMP_SCREEN, "SCREEN", 0, "Screen", ""},
@@ -91,7 +91,7 @@ const EnumPropertyItem rna_enum_ramp_blend_items[] = {
 
 static void rna_Material_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Material *ma = ptr->id.data;
+  Material *ma = (Material *)ptr->owner_id;
 
   DEG_id_tag_update(&ma->id, ID_RECALC_SHADING);
   WM_main_add_notifier(NC_MATERIAL | ND_SHADING, ma);
@@ -101,7 +101,7 @@ static void rna_Material_update_previews(Main *UNUSED(bmain),
                                          Scene *UNUSED(scene),
                                          PointerRNA *ptr)
 {
-  Material *ma = ptr->id.data;
+  Material *ma = (Material *)ptr->owner_id;
 
   if (ma->nodetree) {
     BKE_node_preview_clear_tree(ma->nodetree);
@@ -112,7 +112,7 @@ static void rna_Material_update_previews(Main *UNUSED(bmain),
 
 static void rna_MaterialGpencil_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Material *ma = ptr->id.data;
+  Material *ma = (Material *)ptr->owner_id;
 
   rna_Material_update(bmain, scene, ptr);
   WM_main_add_notifier(NC_GPENCIL | ND_DATA, ma);
@@ -120,7 +120,7 @@ static void rna_MaterialGpencil_update(Main *bmain, Scene *scene, PointerRNA *pt
 
 static void rna_MaterialGpencil_nopreview_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Material *ma = ptr->id.data;
+  Material *ma = (Material *)ptr->owner_id;
 
   rna_Material_update(bmain, scene, ptr);
   WM_main_add_notifier(NC_GPENCIL | ND_DATA, ma);
@@ -128,7 +128,7 @@ static void rna_MaterialGpencil_nopreview_update(Main *bmain, Scene *scene, Poin
 
 static void rna_Material_draw_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Material *ma = ptr->id.data;
+  Material *ma = (Material *)ptr->owner_id;
 
   DEG_id_tag_update(&ma->id, ID_RECALC_SHADING);
   WM_main_add_notifier(NC_MATERIAL | ND_SHADING_DRAW, ma);
@@ -146,7 +146,7 @@ static void rna_Material_active_paint_texture_index_update(Main *bmain,
                                                            PointerRNA *ptr)
 {
   bScreen *sc;
-  Material *ma = ptr->id.data;
+  Material *ma = (Material *)ptr->owner_id;
 
   if (ma->use_nodes && ma->nodetree) {
     struct bNode *node = BKE_texpaint_slot_material_find_node(ma, ma->paint_active_slot);
@@ -305,7 +305,7 @@ static bool rna_is_grease_pencil_get(PointerRNA *ptr)
 static void rna_gpcolordata_uv_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
   /* update all uv strokes of this color */
-  Material *ma = ptr->id.data;
+  Material *ma = (Material *)ptr->owner_id;
   ED_gpencil_update_color_uv(bmain, ma);
 
   rna_MaterialGpencil_update(bmain, scene, ptr);
@@ -421,12 +421,14 @@ static void rna_def_material_display(StructRNA *srna)
   prop = RNA_def_property(srna, "diffuse_color", PROP_FLOAT, PROP_COLOR);
   RNA_def_property_float_sdna(prop, NULL, "r");
   RNA_def_property_array(prop, 4);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Diffuse Color", "Diffuse color of the material");
   RNA_def_property_update(prop, 0, "rna_Material_draw_update");
 
   prop = RNA_def_property(srna, "specular_color", PROP_FLOAT, PROP_COLOR);
   RNA_def_property_float_sdna(prop, NULL, "specr");
   RNA_def_property_array(prop, 3);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Specular Color", "Specular color of the material");
   RNA_def_property_update(prop, 0, "rna_Material_draw_update");
 
@@ -434,6 +436,7 @@ static void rna_def_material_display(StructRNA *srna)
   RNA_def_property_float_sdna(prop, NULL, "roughness");
   RNA_def_property_float_default(prop, 0.25f);
   RNA_def_property_range(prop, 0, 1);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Roughness", "Roughness of the material");
   RNA_def_property_update(prop, 0, "rna_Material_draw_update");
 
@@ -441,12 +444,14 @@ static void rna_def_material_display(StructRNA *srna)
   RNA_def_property_float_sdna(prop, NULL, "spec");
   RNA_def_property_float_default(prop, 0.5f);
   RNA_def_property_range(prop, 0, 1);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Specular", "How intense (bright) the specular reflection is");
   RNA_def_property_update(prop, 0, "rna_Material_draw_update");
 
   prop = RNA_def_property(srna, "metallic", PROP_FLOAT, PROP_FACTOR);
   RNA_def_property_float_sdna(prop, NULL, "metallic");
   RNA_def_property_range(prop, 0.0f, 1.0f);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Metallic", "Amount of mirror reflection for raytrace");
   RNA_def_property_update(prop, 0, "rna_Material_update");
 
@@ -778,6 +783,7 @@ static void rna_def_material_greasepencil(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "sima");
   RNA_def_property_pointer_funcs(prop, NULL, "rna_GpencilColorData_stroke_image_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Image", "");
   RNA_def_property_update(prop, NC_GPENCIL | ND_SHADING, "rna_MaterialGpencil_update");
 
@@ -800,6 +806,7 @@ static void rna_def_material_greasepencil(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "ima");
   RNA_def_property_pointer_funcs(prop, NULL, "rna_GpencilColorData_fill_image_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Image", "");
   RNA_def_property_update(prop, NC_GPENCIL | ND_SHADING, "rna_MaterialGpencil_update");
 
diff --git a/source/blender/makesrna/intern/rna_mesh.c b/source/blender/makesrna/intern/rna_mesh.c
index 99428fd872b..704bb11fc8a 100644
--- a/source/blender/makesrna/intern/rna_mesh.c
+++ b/source/blender/makesrna/intern/rna_mesh.c
@@ -80,7 +80,7 @@ const EnumPropertyItem rna_enum_mesh_delimit_mode_items[] = {
 
 static Mesh *rna_mesh(PointerRNA *ptr)
 {
-  Mesh *me = (Mesh *)ptr->id.data;
+  Mesh *me = (Mesh *)ptr->owner_id;
   return me;
 }
 
@@ -146,7 +146,7 @@ static void rna_cd_layer_name_set(CustomData *cdata, CustomDataLayer *cdl, const
 static CustomData *rna_cd_from_layer(PointerRNA *ptr, CustomDataLayer *cdl)
 {
   /* find out where we come from by */
-  Mesh *me = ptr->id.data;
+  Mesh *me = (Mesh *)ptr->owner_id;
   CustomData *cd;
 
   /* rely on negative values wrapping */
@@ -202,7 +202,7 @@ static bool rna_Mesh_has_custom_normals_get(PointerRNA *ptr)
 
 static void rna_Mesh_update_data(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   /* cheating way for importers to avoid slow updates */
   if (id->us > 0) {
@@ -226,7 +226,7 @@ static void rna_Mesh_update_data_edit_active_color(Main *bmain, Scene *scene, Po
 }
 static void rna_Mesh_update_select(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   /* cheating way for importers to avoid slow updates */
   if (id->us > 0) {
     WM_main_add_notifier(NC_GEOM | ND_SELECT, id);
@@ -235,7 +235,7 @@ static void rna_Mesh_update_select(Main *UNUSED(bmain), Scene *UNUSED(scene), Po
 
 void rna_Mesh_update_draw(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   /* cheating way for importers to avoid slow updates */
   if (id->us > 0) {
     WM_main_add_notifier(NC_GEOM | ND_DATA, id);
@@ -403,7 +403,7 @@ static void rna_MeshPolygon_center_get(PointerRNA *ptr, float *values)
 
 static float rna_MeshPolygon_area_get(PointerRNA *ptr)
 {
-  Mesh *me = (Mesh *)ptr->id.data;
+  Mesh *me = (Mesh *)ptr->owner_id;
   MPoly *mp = (MPoly *)ptr->data;
 
   return BKE_mesh_calc_poly_area(mp, me->mloop + mp->loopstart, me->mvert);
@@ -571,7 +571,7 @@ static int rna_CustomDataLayer_clone_get(PointerRNA *ptr, CustomData *data, int
 static void rna_CustomDataLayer_active_set(
     PointerRNA *ptr, CustomData *data, int value, int type, int render)
 {
-  Mesh *me = ptr->id.data;
+  Mesh *me = (Mesh *)ptr->owner_id;
   int n = (((CustomDataLayer *)ptr->data) - data->layers) - CustomData_get_layer_index(data, type);
 
   if (value == 0) {
@@ -651,14 +651,6 @@ static void rna_MPoly_freestyle_face_mark_set(PointerRNA *ptr, int value)
   }
 }
 
-/* Generic UV rename! */
-static void rna_MeshUVLayer_name_set(PointerRNA *ptr, const char *name)
-{
-  char buf[MAX_CUSTOMDATA_LAYER_NAME];
-  BLI_strncpy_utf8(buf, name, MAX_CUSTOMDATA_LAYER_NAME);
-  BKE_mesh_uv_cdlayer_rename(rna_mesh(ptr), ((CustomDataLayer *)ptr->data)->name, buf, true);
-}
-
 /* uv_layers */
 
 DEFINE_CUSTOMDATA_LAYER_COLLECTION(uv_layer, ldata, CD_MLOOPUV)
@@ -1378,7 +1370,7 @@ static int rna_MeshPolygonStringPropertyLayer_data_length(PointerRNA *ptr)
 }
 
 /* XXX, we dont have proper byte string support yet, so for now use the (bytes + 1)
- * bmesh API exposes correct python/bytestring access */
+ * bmesh API exposes correct python/byte-string access. */
 void rna_MeshStringProperty_s_get(PointerRNA *ptr, char *value)
 {
   MStringProperty *ms = (MStringProperty *)ptr->data;
@@ -1941,7 +1933,7 @@ static void rna_def_mloopuv(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "name", PROP_STRING, PROP_NONE);
   RNA_def_struct_name_property(srna, prop);
-  RNA_def_property_string_funcs(prop, NULL, NULL, "rna_MeshUVLayer_name_set");
+  RNA_def_property_string_funcs(prop, NULL, NULL, "rna_MeshLoopLayer_name_set");
   RNA_def_property_ui_text(prop, "Name", "Name of UV map");
   RNA_def_property_update(prop, 0, "rna_Mesh_update_data");
 
@@ -2789,6 +2781,7 @@ static void rna_def_mesh(BlenderRNA *brna)
   prop = RNA_def_property(srna, "texture_mesh", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "texcomesh");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop,
       "Texture Mesh",
@@ -3053,6 +3046,7 @@ static void rna_def_mesh(BlenderRNA *brna)
   prop = RNA_def_property(srna, "texco_mesh", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "texcomesh");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Texture Space Mesh", "Derive texture coordinates from another mesh");
 
diff --git a/source/blender/makesrna/intern/rna_mesh_api.c b/source/blender/makesrna/intern/rna_mesh_api.c
index 283590fc529..0b3a7802082 100644
--- a/source/blender/makesrna/intern/rna_mesh_api.c
+++ b/source/blender/makesrna/intern/rna_mesh_api.c
@@ -311,11 +311,6 @@ void RNA_api_mesh(StructRNA *srna)
                   0,
                   "Calculate Loose Edges",
                   "Calculate the loose state of each edge");
-  RNA_def_boolean(func,
-                  "calc_loop_triangles",
-                  0,
-                  "Calculate Triangules",
-                  "Force recalculation of triangle tessellation");
   RNA_def_function_flag(func, FUNC_USE_CONTEXT);
 
   RNA_def_function(srna, "update_gpu_tag", "rna_Mesh_update_gpu_tag");
diff --git a/source/blender/makesrna/intern/rna_meta.c b/source/blender/makesrna/intern/rna_meta.c
index 743bac2b709..451bdda45f5 100644
--- a/source/blender/makesrna/intern/rna_meta.c
+++ b/source/blender/makesrna/intern/rna_meta.c
@@ -89,7 +89,7 @@ static void rna_Meta_texspace_size_set(PointerRNA *ptr, const float *values)
 
 static void rna_MetaBall_update_data(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  MetaBall *mb = ptr->id.data;
+  MetaBall *mb = (MetaBall *)ptr->owner_id;
   Object *ob;
 
   /* cheating way for importers to avoid slow updates */
@@ -158,13 +158,13 @@ static void rna_MetaBall_elements_clear(MetaBall *mb)
 
 static bool rna_Meta_is_editmode_get(PointerRNA *ptr)
 {
-  MetaBall *mb = ptr->id.data;
+  MetaBall *mb = (MetaBall *)ptr->owner_id;
   return (mb->editelems != NULL);
 }
 
 static char *rna_MetaElement_path(PointerRNA *ptr)
 {
-  MetaBall *mb = ptr->id.data;
+  MetaBall *mb = (MetaBall *)ptr->owner_id;
   MetaElem *ml = ptr->data;
   int index = -1;
 
diff --git a/source/blender/makesrna/intern/rna_modifier.c b/source/blender/makesrna/intern/rna_modifier.c
index cae9e2a8c9b..18cf57c5624 100644
--- a/source/blender/makesrna/intern/rna_modifier.c
+++ b/source/blender/makesrna/intern/rna_modifier.c
@@ -609,8 +609,8 @@ static void rna_Modifier_name_set(PointerRNA *ptr, const char *value)
   BLI_strncpy_utf8(md->name, value, sizeof(md->name));
 
   /* make sure the name is truly unique */
-  if (ptr->id.data) {
-    Object *ob = ptr->id.data;
+  if (ptr->owner_id) {
+    Object *ob = (Object *)ptr->owner_id;
     modifier_unique_name(&ob->modifiers, md);
   }
 
@@ -629,8 +629,8 @@ static char *rna_Modifier_path(PointerRNA *ptr)
 
 static void rna_Modifier_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
-  WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ptr->id.data);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
+  WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ptr->owner_id);
 }
 
 static void rna_Modifier_dependency_update(Main *bmain, Scene *scene, PointerRNA *ptr)
@@ -737,7 +737,7 @@ static void modifier_object_set(Object *self, Object **ob_p, int type, PointerRN
         PointerRNA *ptr, PointerRNA value, struct ReportList *UNUSED(reports)) \
     { \
       _type##ModifierData *tmd = (_type##ModifierData *)ptr->data; \
-      modifier_object_set(ptr->id.data, &tmd->_prop, _obtype, value); \
+      modifier_object_set((Object *)ptr->owner_id, &tmd->_prop, _obtype, value); \
     }
 
 RNA_MOD_OBJECT_SET(Armature, object, OB_ARMATURE);
@@ -760,7 +760,7 @@ static void rna_HookModifier_object_set(PointerRNA *ptr,
                                         PointerRNA value,
                                         struct ReportList *UNUSED(reports))
 {
-  Object *owner = (Object *)ptr->id.data;
+  Object *owner = (Object *)ptr->owner_id;
   HookModifierData *hmd = ptr->data;
   Object *ob = (Object *)value.data;
 
@@ -769,9 +769,50 @@ static void rna_HookModifier_object_set(PointerRNA *ptr,
   BKE_object_modifier_hook_reset(owner, hmd);
 }
 
+static bool rna_HookModifier_object_override_apply(Main *UNUSED(bmain),
+                                                   PointerRNA *ptr_dst,
+                                                   PointerRNA *ptr_src,
+                                                   PointerRNA *ptr_storage,
+                                                   PropertyRNA *prop_dst,
+                                                   PropertyRNA *prop_src,
+                                                   PropertyRNA *UNUSED(prop_storage),
+                                                   const int len_dst,
+                                                   const int len_src,
+                                                   const int len_storage,
+                                                   PointerRNA *UNUSED(ptr_item_dst),
+                                                   PointerRNA *UNUSED(ptr_item_src),
+                                                   PointerRNA *UNUSED(ptr_item_storage),
+                                                   IDOverrideLibraryPropertyOperation *opop)
+{
+  BLI_assert(len_dst == len_src && (!ptr_storage || len_dst == len_storage) && len_dst == 0);
+  BLI_assert(opop->operation == IDOVERRIDE_LIBRARY_OP_REPLACE &&
+             "Unsupported RNA override operation on Hook modifier target objet pointer");
+  UNUSED_VARS_NDEBUG(ptr_storage, len_dst, len_src, len_storage, opop);
+
+  /* We need a special handling here because setting hook target resets invert parent matrix,
+   * which is evil in our case. */
+  HookModifierData *hmd = ptr_dst->data;
+  Object *owner = (Object *)ptr_dst->owner_id;
+  Object *target_dst = RNA_property_pointer_get(ptr_dst, prop_dst).data;
+  Object *target_src = RNA_property_pointer_get(ptr_src, prop_src).data;
+
+  BLI_assert(target_dst == hmd->object);
+
+  if (target_src == target_dst) {
+    return false;
+  }
+
+  hmd->object = target_src;
+  if (target_src == NULL) {
+    /* The only case where we do want default behavior (with matrix reset). */
+    BKE_object_modifier_hook_reset(owner, hmd);
+  }
+  return true;
+}
+
 static void rna_HookModifier_subtarget_set(PointerRNA *ptr, const char *value)
 {
-  Object *owner = (Object *)ptr->id.data;
+  Object *owner = (Object *)ptr->owner_id;
   HookModifierData *hmd = ptr->data;
 
   BLI_strncpy(hmd->subtarget, value, sizeof(hmd->subtarget));
@@ -858,7 +899,7 @@ static void rna_UVProjector_object_set(PointerRNA *ptr,
 static void rna_Smoke_set_type(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
   SmokeModifierData *smd = (SmokeModifierData *)ptr->data;
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   /* nothing changed */
   if ((smd->type & MOD_SMOKE_TYPE_DOMAIN) && smd->domain) {
@@ -885,7 +926,7 @@ static void rna_Smoke_set_type(Main *bmain, Scene *scene, PointerRNA *ptr)
 
 static void rna_MultiresModifier_type_set(PointerRNA *ptr, int value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   MultiresModifierData *mmd = (MultiresModifierData *)ptr->data;
 
   multires_force_update(ob);
@@ -903,7 +944,7 @@ static void rna_MultiresModifier_level_range(
 
 static bool rna_MultiresModifier_external_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Mesh *me = ob->data;
 
   return CustomData_external_test(&me->ldata, CD_MDISPS);
@@ -911,7 +952,7 @@ static bool rna_MultiresModifier_external_get(PointerRNA *ptr)
 
 static void rna_MultiresModifier_filepath_get(PointerRNA *ptr, char *value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   CustomDataExternal *external = ((Mesh *)ob->data)->ldata.external;
 
   BLI_strncpy(value, (external) ? external->filename : "", sizeof(external->filename));
@@ -919,7 +960,7 @@ static void rna_MultiresModifier_filepath_get(PointerRNA *ptr, char *value)
 
 static void rna_MultiresModifier_filepath_set(PointerRNA *ptr, const char *value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   CustomDataExternal *external = ((Mesh *)ob->data)->ldata.external;
 
   if (external && !STREQ(external->filename, value)) {
@@ -930,7 +971,7 @@ static void rna_MultiresModifier_filepath_set(PointerRNA *ptr, const char *value
 
 static int rna_MultiresModifier_filepath_length(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   CustomDataExternal *external = ((Mesh *)ob->data)->ldata.external;
 
   return strlen((external) ? external->filename : "");
@@ -955,19 +996,19 @@ static bool rna_MeshDeformModifier_is_bound_get(PointerRNA *ptr)
 
 static PointerRNA rna_SoftBodyModifier_settings_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   return rna_pointer_inherit_refine(ptr, &RNA_SoftBodySettings, ob->soft);
 }
 
 static PointerRNA rna_SoftBodyModifier_point_cache_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   return rna_pointer_inherit_refine(ptr, &RNA_PointCache, ob->soft->shared->pointcache);
 }
 
 static PointerRNA rna_CollisionModifier_settings_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   return rna_pointer_inherit_refine(ptr, &RNA_CollisionSettings, ob->pd);
 }
 
@@ -1574,6 +1615,7 @@ static void rna_def_modifier_generic_map_info(StructRNA *srna)
   prop = RNA_def_property(srna, "texture", PROP_POINTER, PROP_NONE);
   RNA_def_property_ui_text(prop, "Texture", "");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "texture_coords", PROP_ENUM, PROP_NONE);
@@ -1593,6 +1635,7 @@ static void rna_def_modifier_generic_map_info(StructRNA *srna)
   RNA_def_property_ui_text(
       prop, "Texture Coordinate Object", "Object to set the texture coordinates");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 }
 
@@ -1609,11 +1652,13 @@ static void rna_def_modifier_warp(BlenderRNA *brna)
   prop = RNA_def_property(srna, "object_from", PROP_POINTER, PROP_NONE);
   RNA_def_property_ui_text(prop, "From", "Object to transform from");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "object_to", PROP_POINTER, PROP_NONE);
   RNA_def_property_ui_text(prop, "To", "Object to transform to");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "strength", PROP_FLOAT, PROP_NONE);
@@ -1729,6 +1774,7 @@ static void rna_def_modifier_lattice(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_LatticeModifier_object_set", NULL, "rna_Lattice_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
@@ -1772,6 +1818,7 @@ static void rna_def_modifier_curve(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_CurveModifier_object_set", NULL, "rna_Curve_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_CurveModifier_dependency_update");
 
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
@@ -1930,6 +1977,7 @@ static void rna_def_modifier_mirror(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "mirror_ob");
   RNA_def_property_ui_text(prop, "Mirror Object", "Object to use as mirror");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 }
 
@@ -2028,7 +2076,7 @@ static void rna_def_modifier_decimate(BlenderRNA *brna)
   prop = RNA_def_property(srna, "use_dissolve_boundaries", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", MOD_DECIM_FLAG_ALL_BOUNDARY_VERTS);
   RNA_def_property_ui_text(
-      prop, "All Boundaries", "Dissolve all vertices inbetween face boundaries (planar only)");
+      prop, "All Boundaries", "Dissolve all vertices in between face boundaries (planar only)");
   RNA_def_property_update(prop, 0, "rna_Modifier_update");
 
   prop = RNA_def_property(srna, "delimit", PROP_ENUM, PROP_NONE);
@@ -2139,6 +2187,7 @@ static void rna_def_modifier_wave(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "objectcenter");
   RNA_def_property_ui_text(prop, "Start Position Object", "Object which defines the wave center");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
@@ -2294,6 +2343,8 @@ static void rna_def_modifier_hook(BlenderRNA *brna)
   RNA_def_property_ui_text(
       prop, "Object", "Parent Object for hook, also recalculates and clears offset");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
+  RNA_def_property_override_funcs(prop, NULL, NULL, "rna_HookModifier_object_override_apply");
   RNA_def_property_pointer_funcs(prop, NULL, "rna_HookModifier_object_set", NULL, NULL);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
@@ -2394,6 +2445,7 @@ static void rna_def_modifier_boolean(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_BooleanModifier_object_set", NULL, "rna_Mesh_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "operation", PROP_ENUM, PROP_NONE);
@@ -2482,6 +2534,7 @@ static void rna_def_modifier_array(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_ArrayModifier_curve_ob_set", NULL, "rna_Curve_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_ArrayModifier_dependency_update");
 
   /* Offset parameters */
@@ -2545,6 +2598,7 @@ static void rna_def_modifier_array(BlenderRNA *brna)
       "Use the location and rotation of another object to determine the distance and "
       "rotational change between arrayed items");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   /* Caps */
@@ -2553,6 +2607,7 @@ static void rna_def_modifier_array(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_ArrayModifier_start_cap_set", NULL, "rna_Mesh_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_update");
 
   prop = RNA_def_property(srna, "end_cap", PROP_POINTER, PROP_NONE);
@@ -2560,6 +2615,7 @@ static void rna_def_modifier_array(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_ArrayModifier_end_cap_set", NULL, "rna_Mesh_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "offset_u", PROP_FLOAT, PROP_FACTOR);
@@ -2772,6 +2828,7 @@ static void rna_def_modifier_uvproject(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, "rna_UVProjector_object_get", "rna_UVProjector_object_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Object", "Object to use as projector transform");
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 }
@@ -3016,6 +3073,7 @@ static void rna_def_modifier_cast(BlenderRNA *brna)
       "Control object: if available, its location determines the center of the effect");
   RNA_def_property_pointer_funcs(prop, NULL, "rna_CastModifier_object_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "use_x", PROP_BOOLEAN, PROP_NONE);
@@ -3098,6 +3156,7 @@ static void rna_def_modifier_meshdeform(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_MeshDeformModifier_object_set", NULL, "rna_Mesh_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "is_bound", PROP_BOOLEAN, PROP_NONE);
@@ -3183,6 +3242,7 @@ static void rna_def_modifier_particleinstance(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(prop, NULL, NULL, NULL, "rna_Mesh_object_poll");
   RNA_def_property_ui_text(prop, "Object", "Object that has the particle system");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "particle_system_index", PROP_INT, PROP_NONE);
@@ -3750,6 +3810,7 @@ static void rna_def_modifier_shrinkwrap(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_ShrinkwrapModifier_target_set", NULL, "rna_Mesh_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "auxiliary_target", PROP_POINTER, PROP_NONE);
@@ -3758,6 +3819,7 @@ static void rna_def_modifier_shrinkwrap(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_ShrinkwrapModifier_auxTarget_set", NULL, "rna_Mesh_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
@@ -3875,6 +3937,7 @@ static void rna_def_modifier_mask(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_MaskModifier_ob_arm_set", NULL, "rna_Armature_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "vertex_group", PROP_STRING, PROP_NONE);
@@ -3953,6 +4016,7 @@ static void rna_def_modifier_simpledeform(BlenderRNA *brna)
   prop = RNA_def_property(srna, "origin", PROP_POINTER, PROP_NONE);
   RNA_def_property_ui_text(prop, "Origin", "Offset the origin and orientation of the deformation");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "factor", PROP_FLOAT, PROP_NONE);
@@ -4146,6 +4210,7 @@ static void rna_def_modifier_screw(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "ob_axis");
   RNA_def_property_ui_text(prop, "Object", "Object to define the screw axis");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "steps", PROP_INT, PROP_UNSIGNED);
@@ -4272,6 +4337,7 @@ static void rna_def_modifier_uvwarp(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "object_src");
   RNA_def_property_ui_text(prop, "Object From", "Object defining offset");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "bone_from", PROP_STRING, PROP_NONE);
@@ -4283,6 +4349,7 @@ static void rna_def_modifier_uvwarp(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "object_dst");
   RNA_def_property_ui_text(prop, "Object To", "Object defining offset");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "bone_to", PROP_STRING, PROP_NONE);
@@ -4350,6 +4417,7 @@ static void rna_def_modifier_weightvg_mask(BlenderRNA *UNUSED(brna),
   prop = RNA_def_property(srna, "mask_texture", PROP_POINTER, PROP_NONE);
   RNA_def_property_ui_text(prop, "Masking Tex", "Masking texture");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "mask_tex_use_channel", PROP_ENUM, PROP_NONE);
@@ -4378,6 +4446,7 @@ static void rna_def_modifier_weightvg_mask(BlenderRNA *UNUSED(brna),
                            "Which object to take texture "
                            "coordinates from");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 }
 
@@ -4647,6 +4716,7 @@ static void rna_def_modifier_weightvgproximity(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "proximity_ob_target");
   RNA_def_property_ui_text(prop, "Target Object", "Object to calculate vertices distances from");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "min_dist", PROP_FLOAT, PROP_DISTANCE);
@@ -5210,6 +5280,7 @@ static void rna_def_modifier_meshseqcache(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "CacheFile");
   RNA_def_property_ui_text(prop, "Cache File", "");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "object_path", PROP_STRING, PROP_NONE);
@@ -5411,6 +5482,7 @@ static void rna_def_modifier_datatransfer(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "ob_source");
   RNA_def_property_ui_text(prop, "Source Object", "Object to transfer data from");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_DataTransferModifier_ob_source_set", NULL, "rna_Mesh_object_poll");
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
@@ -5816,6 +5888,7 @@ static void rna_def_modifier_normaledit(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Target", "Target object used to affect normals");
   RNA_def_property_pointer_funcs(prop, NULL, "rna_NormalEditModifier_target_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "use_direction_parallel", PROP_BOOLEAN, PROP_NONE);
@@ -5843,6 +5916,7 @@ static void rna_def_modifier_surfacedeform(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_SurfaceDeformModifier_target_set", NULL, "rna_Mesh_object_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_Modifier_dependency_update");
 
   prop = RNA_def_property(srna, "falloff", PROP_FLOAT, PROP_NONE);
diff --git a/source/blender/makesrna/intern/rna_movieclip.c b/source/blender/makesrna/intern/rna_movieclip.c
index 10e19b92dda..9bfa93ceba5 100644
--- a/source/blender/makesrna/intern/rna_movieclip.c
+++ b/source/blender/makesrna/intern/rna_movieclip.c
@@ -53,14 +53,14 @@
 
 static void rna_MovieClip_reload_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   DEG_id_tag_update(&clip->id, ID_RECALC_SOURCE);
 }
 
 static void rna_MovieClip_size_get(PointerRNA *ptr, int *values)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   values[0] = clip->lastsize[0];
   values[1] = clip->lastsize[1];
@@ -68,13 +68,13 @@ static void rna_MovieClip_size_get(PointerRNA *ptr, int *values)
 
 static float rna_MovieClip_fps_get(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   return BKE_movieclip_get_fps(clip);
 }
 
 static void rna_MovieClip_use_proxy_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   BKE_movieclip_clear_cache(clip);
   BKE_sequence_invalidate_movieclip_strips(bmain, clip);
 }
@@ -83,7 +83,7 @@ static void rna_MovieClipUser_proxy_render_settings_update(Main *bmain,
                                                            Scene *UNUSED(scene),
                                                            PointerRNA *ptr)
 {
-  ID *id = (ID *)ptr->id.data;
+  ID *id = ptr->owner_id;
   MovieClipUser *user = (MovieClipUser *)ptr->data;
 
   /* when changing render settings of space clip user
diff --git a/source/blender/makesrna/intern/rna_nla.c b/source/blender/makesrna/intern/rna_nla.c
index 6f55fb4e24c..97cab783aed 100644
--- a/source/blender/makesrna/intern/rna_nla.c
+++ b/source/blender/makesrna/intern/rna_nla.c
@@ -65,8 +65,8 @@ static void rna_NlaStrip_name_set(PointerRNA *ptr, const char *value)
   BLI_strncpy_utf8(data->name, value, sizeof(data->name));
 
   /* validate if there's enough info to do so */
-  if (ptr->id.data) {
-    AnimData *adt = BKE_animdata_from_id(ptr->id.data);
+  if (ptr->owner_id) {
+    AnimData *adt = BKE_animdata_from_id(ptr->owner_id);
     BKE_nlastrip_validate_name(adt, data);
   }
 }
@@ -74,7 +74,7 @@ static void rna_NlaStrip_name_set(PointerRNA *ptr, const char *value)
 static char *rna_NlaStrip_path(PointerRNA *ptr)
 {
   NlaStrip *strip = (NlaStrip *)ptr->data;
-  AnimData *adt = BKE_animdata_from_id(ptr->id.data);
+  AnimData *adt = BKE_animdata_from_id(ptr->owner_id);
 
   /* if we're attached to AnimData, try to resolve path back to AnimData */
   if (adt) {
@@ -103,7 +103,7 @@ static char *rna_NlaStrip_path(PointerRNA *ptr)
 
 static void rna_NlaStrip_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   ANIM_id_update(bmain, id);
 }
@@ -122,9 +122,9 @@ static void rna_NlaStrip_transform_update(Main *bmain, Scene *scene, PointerRNA
   BKE_nlameta_flush_transforms(strip);
 
   /* set the flag */
-  if ((strip->flag & NLASTRIP_FLAG_AUTO_BLENDS) && ptr->id.data) {
+  if ((strip->flag & NLASTRIP_FLAG_AUTO_BLENDS) && ptr->owner_id) {
     /* validate state to ensure that auto-blend gets applied immediately */
-    IdAdtTemplate *iat = (IdAdtTemplate *)ptr->id.data;
+    IdAdtTemplate *iat = (IdAdtTemplate *)ptr->owner_id;
 
     if (iat->adt) {
       BKE_nla_validate_state(iat->adt);
@@ -274,8 +274,8 @@ static void rna_NlaStrip_use_auto_blend_set(PointerRNA *ptr, bool value)
     data->flag |= NLASTRIP_FLAG_AUTO_BLENDS;
 
     /* validate state to ensure that auto-blend gets applied immediately */
-    if (ptr->id.data) {
-      IdAdtTemplate *iat = (IdAdtTemplate *)ptr->id.data;
+    if (ptr->owner_id) {
+      IdAdtTemplate *iat = (IdAdtTemplate *)ptr->owner_id;
 
       if (iat->adt) {
         BKE_nla_validate_state(iat->adt);
@@ -298,8 +298,8 @@ static int rna_NlaStrip_action_editable(PointerRNA *ptr, const char **UNUSED(r_i
   NlaStrip *strip = (NlaStrip *)ptr->data;
 
   /* strip actions shouldn't be editable if NLA tweakmode is on */
-  if (ptr->id.data) {
-    AnimData *adt = BKE_animdata_from_id(ptr->id.data);
+  if (ptr->owner_id) {
+    AnimData *adt = BKE_animdata_from_id(ptr->owner_id);
 
     if (adt) {
       /* active action is only editable when it is not a tweaking strip */
@@ -475,7 +475,7 @@ static void rna_NlaStrip_remove(
 static void rna_NlaTrack_solo_set(PointerRNA *ptr, bool value)
 {
   NlaTrack *data = (NlaTrack *)ptr->data;
-  AnimData *adt = BKE_animdata_from_id(ptr->id.data);
+  AnimData *adt = BKE_animdata_from_id(ptr->owner_id);
   NlaTrack *nt;
 
   if (data == NULL) {
diff --git a/source/blender/makesrna/intern/rna_nodetree.c b/source/blender/makesrna/intern/rna_nodetree.c
index c9815a6cc37..0909aa42a4d 100644
--- a/source/blender/makesrna/intern/rna_nodetree.c
+++ b/source/blender/makesrna/intern/rna_nodetree.c
@@ -63,10 +63,13 @@ const EnumPropertyItem rna_enum_node_socket_in_out_items[] = {
     {SOCK_IN, "IN", 0, "Input", ""}, {SOCK_OUT, "OUT", 0, "Output", ""}, {0, NULL, 0, NULL, NULL}};
 
 #ifndef RNA_RUNTIME
-static const EnumPropertyItem rna_enum_node_socket_draw_shape_items[] = {
-    {SOCK_DRAW_SHAPE_CIRCLE, "CIRCLE", 0, "Circle", ""},
-    {SOCK_DRAW_SHAPE_SQUARE, "SQUARE", 0, "Square", ""},
-    {SOCK_DRAW_SHAPE_DIAMOND, "DIAMOND", 0, "Diamond", ""},
+static const EnumPropertyItem rna_enum_node_socket_display_shape_items[] = {
+    {SOCK_DISPLAY_SHAPE_CIRCLE, "CIRCLE", 0, "Circle", ""},
+    {SOCK_DISPLAY_SHAPE_SQUARE, "SQUARE", 0, "Square", ""},
+    {SOCK_DISPLAY_SHAPE_DIAMOND, "DIAMOND", 0, "Diamond", ""},
+    {SOCK_DISPLAY_SHAPE_CIRCLE_DOT, "CIRCLE_DOT", 0, "Circle with inner dot", ""},
+    {SOCK_DISPLAY_SHAPE_SQUARE_DOT, "SQUARE_DOT", 0, "Square with inner dot", ""},
+    {SOCK_DISPLAY_SHAPE_DIAMOND_DOT, "DIAMOND_DOT", 0, "Diamond with inner dot", ""},
     {0, NULL, 0, NULL, NULL}};
 
 static const EnumPropertyItem node_socket_type_items[] = {
@@ -100,44 +103,81 @@ static const EnumPropertyItem node_chunksize_items[] = {
 #endif
 
 const EnumPropertyItem rna_enum_node_math_items[] = {
-    {NODE_MATH_ADD, "ADD", 0, "Add", ""},
-    {NODE_MATH_SUB, "SUBTRACT", 0, "Subtract", ""},
-    {NODE_MATH_MUL, "MULTIPLY", 0, "Multiply", ""},
-    {NODE_MATH_DIVIDE, "DIVIDE", 0, "Divide", ""},
+    {NODE_MATH_ADD, "ADD", 0, "Add", "A + B"},
+    {NODE_MATH_SUBTRACT, "SUBTRACT", 0, "Subtract", "A - B"},
+    {NODE_MATH_MULTIPLY, "MULTIPLY", 0, "Multiply", "A * B"},
+    {NODE_MATH_DIVIDE, "DIVIDE", 0, "Divide", "A / B"},
     {0, "", ICON_NONE, NULL, NULL},
-    {NODE_MATH_POW, "POWER", 0, "Power", ""},
-    {NODE_MATH_LOG, "LOGARITHM", 0, "Logarithm", ""},
-    {NODE_MATH_SQRT, "SQRT", 0, "Square Root", ""},
-    {NODE_MATH_ABS, "ABSOLUTE", 0, "Absolute", ""},
+    {NODE_MATH_POWER, "POWER", 0, "Power", "A power B"},
+    {NODE_MATH_LOGARITHM, "LOGARITHM", 0, "Logarithm", "Logarithm A base B"},
+    {NODE_MATH_SQRT, "SQRT", 0, "Square Root", "Square root of A"},
+    {NODE_MATH_ABSOLUTE, "ABSOLUTE", 0, "Absolute", "Magnitude of A"},
     {0, "", ICON_NONE, NULL, NULL},
-    {NODE_MATH_MIN, "MINIMUM", 0, "Minimum", ""},
-    {NODE_MATH_MAX, "MAXIMUM", 0, "Maximum", ""},
-    {NODE_MATH_LESS, "LESS_THAN", 0, "Less Than", ""},
-    {NODE_MATH_GREATER, "GREATER_THAN", 0, "Greater Than", ""},
+    {NODE_MATH_MINIMUM, "MINIMUM", 0, "Minimum", "The minimum from A and B"},
+    {NODE_MATH_MAXIMUM, "MAXIMUM", 0, "Maximum", "The maximum from A and B"},
+    {NODE_MATH_LESS_THAN, "LESS_THAN", 0, "Less Than", "1 if A < B else 0"},
+    {NODE_MATH_GREATER_THAN, "GREATER_THAN", 0, "Greater Than", "1 if A > B else 0"},
     {0, "", ICON_NONE, NULL, NULL},
-    {NODE_MATH_ROUND, "ROUND", 0, "Round", ""},
-    {NODE_MATH_FLOOR, "FLOOR", 0, "Floor", ""},
-    {NODE_MATH_CEIL, "CEIL", 0, "Ceil", ""},
-    {NODE_MATH_FRACT, "FRACT", 0, "Fract", ""},
-    {NODE_MATH_MOD, "MODULO", 0, "Modulo", ""},
+    {NODE_MATH_ROUND,
+     "ROUND",
+     0,
+     "Round",
+     "Round A to the nearest integer. Round upward if the fraction part is 0.5"},
+    {NODE_MATH_FLOOR, "FLOOR", 0, "Floor", "The largest integer smaller than or equal A"},
+    {NODE_MATH_CEIL, "CEIL", 0, "Ceil", "The smallest integer greater than or equal A"},
+    {NODE_MATH_FRACTION, "FRACT", 0, "Fraction", "The fraction part of A"},
+    {NODE_MATH_MODULO, "MODULO", 0, "Modulo", "A mod B"},
     {0, "", ICON_NONE, NULL, NULL},
-    {NODE_MATH_SIN, "SINE", 0, "Sine", ""},
-    {NODE_MATH_COS, "COSINE", 0, "Cosine", ""},
-    {NODE_MATH_TAN, "TANGENT", 0, "Tangent", ""},
-    {NODE_MATH_ASIN, "ARCSINE", 0, "Arcsine", ""},
-    {NODE_MATH_ACOS, "ARCCOSINE", 0, "Arccosine", ""},
-    {NODE_MATH_ATAN, "ARCTANGENT", 0, "Arctangent", ""},
-    {NODE_MATH_ATAN2, "ARCTAN2", 0, "Arctan2", ""},
+    {NODE_MATH_SINE, "SINE", 0, "Sine", "sin(A)"},
+    {NODE_MATH_COSINE, "COSINE", 0, "Cosine", "cos(A)"},
+    {NODE_MATH_TANGENT, "TANGENT", 0, "Tangent", "tan(A)"},
+    {NODE_MATH_ARCSINE, "ARCSINE", 0, "Arcsine", "arcsin(A)"},
+    {NODE_MATH_ARCCOSINE, "ARCCOSINE", 0, "Arccosine", "arccos(A)"},
+    {NODE_MATH_ARCTANGENT, "ARCTANGENT", 0, "Arctangent", "arctan(A)"},
+    {NODE_MATH_ARCTAN2, "ARCTAN2", 0, "Arctan2", "The signed angle arctan(A / B)"},
     {0, NULL, 0, NULL, NULL},
 };
 
 const EnumPropertyItem rna_enum_node_vec_math_items[] = {
-    {0, "ADD", 0, "Add", ""},
-    {1, "SUBTRACT", 0, "Subtract", ""},
-    {2, "AVERAGE", 0, "Average", ""},
-    {3, "DOT_PRODUCT", 0, "Dot Product", ""},
-    {4, "CROSS_PRODUCT", 0, "Cross Product", ""},
-    {5, "NORMALIZE", 0, "Normalize", ""},
+    {NODE_VECTOR_MATH_ADD, "ADD", 0, "Add", "A + B"},
+    {NODE_VECTOR_MATH_SUBTRACT, "SUBTRACT", 0, "Subtract", "A - B"},
+    {NODE_VECTOR_MATH_MULTIPLY, "MULTIPLY", 0, "Multiply", "Entrywise multiply"},
+    {NODE_VECTOR_MATH_DIVIDE, "DIVIDE", 0, "Divide", "Entrywise divide"},
+    {0, "", ICON_NONE, NULL, NULL},
+    {NODE_VECTOR_MATH_CROSS_PRODUCT, "CROSS_PRODUCT", 0, "Cross Product", "A cross B"},
+    {NODE_VECTOR_MATH_PROJECT, "PROJECT", 0, "Project", "Project A onto B"},
+    {NODE_VECTOR_MATH_REFLECT,
+     "REFLECT",
+     0,
+     "Reflect",
+     "Reflect A around the normal B. B needn't be normalized"},
+    {NODE_VECTOR_MATH_DOT_PRODUCT, "DOT_PRODUCT", 0, "Dot Product", "A dot B"},
+    {0, "", ICON_NONE, NULL, NULL},
+    {NODE_VECTOR_MATH_DISTANCE, "DISTANCE", 0, "Distance", "Distance between A and B"},
+    {NODE_VECTOR_MATH_LENGTH, "LENGTH", 0, "Length", "Length of A"},
+    {NODE_VECTOR_MATH_SCALE, "SCALE", 0, "Scale", "A multiplied by Scale"},
+    {NODE_VECTOR_MATH_NORMALIZE, "NORMALIZE", 0, "Normalize", "Normalize A"},
+    {0, "", ICON_NONE, NULL, NULL},
+    {NODE_VECTOR_MATH_SNAP,
+     "SNAP",
+     0,
+     "Snap",
+     "Round A to the largest integer multiple of B less than or equal A"},
+    {NODE_VECTOR_MATH_FLOOR, "FLOOR", 0, "Floor", "Entrywise floor"},
+    {NODE_VECTOR_MATH_CEIL, "CEIL", 0, "Ceil", "Entrywise ceil"},
+    {NODE_VECTOR_MATH_MODULO, "MODULO", 0, "Modulo", "Entrywise modulo"},
+    {NODE_VECTOR_MATH_FRACTION, "FRACTION", 0, "Fraction", "The fraction part of A entrywise"},
+    {NODE_VECTOR_MATH_ABSOLUTE, "ABSOLUTE", 0, "Absolute", "Entrywise absolute"},
+    {NODE_VECTOR_MATH_MINIMUM, "MINIMUM", 0, "Minimum", "Entrywise minimum"},
+    {NODE_VECTOR_MATH_MAXIMUM, "MAXIMUM", 0, "Maximum", "Entrywise maximum"},
+    {0, NULL, 0, NULL, NULL},
+};
+
+static const EnumPropertyItem rna_enum_node_tex_dimensions_items[] = {
+    {1, "1D", 0, "1D", "Use the scalar value W as input"},
+    {2, "2D", 0, "2D", "Use the 2D vector (x, y) as input. The z component is ignored"},
+    {3, "3D", 0, "3D", "Use the 3D vector Vector as input"},
+    {4, "4D", 0, "4D", "Use the 4D vector (x, y, z, w) as input"},
     {0, NULL, 0, NULL, NULL},
 };
 
@@ -720,7 +760,7 @@ static bool rna_NodeTree_check(bNodeTree *ntree, ReportList *reports)
 
 static void rna_NodeTree_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = (bNode *)ptr->data;
 
   WM_main_add_notifier(NC_NODE | NA_EDITED, NULL);
@@ -1201,7 +1241,7 @@ static char *rna_Node_path(PointerRNA *ptr)
 
 char *rna_Node_ImageUser_path(PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node;
   char name_esc[sizeof(node->name) * 2];
 
@@ -1706,7 +1746,7 @@ static bool rna_Node_parent_poll(PointerRNA *ptr, PointerRNA value)
 
 static void rna_Node_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = (bNode *)ptr->data;
   ED_node_tag_update_nodetree(bmain, ntree, node);
 }
@@ -1730,7 +1770,7 @@ static void rna_Node_select_set(PointerRNA *ptr, bool value)
 
 static void rna_Node_name_set(PointerRNA *ptr, const char *value)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = (bNode *)ptr->data;
   char oldname[sizeof(node->name)];
 
@@ -2082,7 +2122,7 @@ static StructRNA *rna_NodeSocket_refine(PointerRNA *ptr)
 
 static char *rna_NodeSocket_path(PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNodeSocket *sock = (bNodeSocket *)ptr->data;
   bNode *node;
   int socketindex;
@@ -2116,7 +2156,7 @@ static IDProperty *rna_NodeSocket_idprops(PointerRNA *ptr, bool create)
 
 static PointerRNA rna_NodeSocket_node_get(PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNodeSocket *sock = (bNodeSocket *)ptr->data;
   bNode *node;
   PointerRNA r_ptr;
@@ -2129,7 +2169,7 @@ static PointerRNA rna_NodeSocket_node_get(PointerRNA *ptr)
 
 static void rna_NodeSocket_type_set(PointerRNA *ptr, int value)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNodeSocket *sock = (bNodeSocket *)ptr->data;
   bNode *node;
   nodeFindNode(ntree, sock, &node, NULL);
@@ -2138,7 +2178,7 @@ static void rna_NodeSocket_type_set(PointerRNA *ptr, int value)
 
 static void rna_NodeSocket_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNodeSocket *sock = (bNodeSocket *)ptr->data;
   bNode *node;
   if (nodeFindNode(ntree, sock, &node, NULL)) {
@@ -2398,7 +2438,7 @@ static StructRNA *rna_NodeSocketInterface_refine(PointerRNA *ptr)
 
 static char *rna_NodeSocketInterface_path(PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNodeSocket *sock = (bNodeSocket *)ptr->data;
   int socketindex;
 
@@ -2429,7 +2469,7 @@ static IDProperty *rna_NodeSocketInterface_idprops(PointerRNA *ptr, bool create)
 
 static void rna_NodeSocketInterface_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNodeSocket *stemp = ptr->data;
 
   if (!stemp->typeinfo) {
@@ -2545,7 +2585,7 @@ static void rna_NodeSocketStandard_value_update(struct bContext *C, PointerRNA *
   /* try to use node from context, faster */
   node = CTX_data_pointer_get(C, "node").data;
   if (!node) {
-    bNodeTree *ntree = ptr->id.data;
+    bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
     bNodeSocket *sock = ptr->data;
 
     /* fall back to searching node in the tree */
@@ -2763,7 +2803,7 @@ static void rna_CompositorNode_tag_need_exec(bNode *node)
 
 static void rna_Node_tex_image_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = (bNode *)ptr->data;
 
   ED_node_tag_update_nodetree(bmain, ntree, node);
@@ -2772,7 +2812,7 @@ static void rna_Node_tex_image_update(Main *bmain, Scene *UNUSED(scene), Pointer
 
 static void rna_NodeGroup_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = (bNode *)ptr->data;
 
   if (node->id) {
@@ -2786,7 +2826,7 @@ static void rna_NodeGroup_node_tree_set(PointerRNA *ptr,
                                         const PointerRNA value,
                                         struct ReportList *UNUSED(reports))
 {
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = ptr->data;
   bNodeTree *ngroup = value.data;
 
@@ -2804,7 +2844,7 @@ static void rna_NodeGroup_node_tree_set(PointerRNA *ptr,
 
 static bool rna_NodeGroup_node_tree_poll(PointerRNA *ptr, const PointerRNA value)
 {
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNodeTree *ngroup = value.data;
 
   /* only allow node trees of the same type as the group node's tree */
@@ -2831,7 +2871,7 @@ static StructRNA *rna_NodeGroup_interface_typef(PointerRNA *ptr)
 
 static StructRNA *rna_NodeGroupInputOutput_interface_typef(PointerRNA *ptr)
 {
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
 
   if (ntree) {
     StructRNA *srna = ntreeInterfaceTypeGet(ntree, true);
@@ -3166,7 +3206,7 @@ static const EnumPropertyItem *rna_Node_channel_itemf(bContext *UNUSED(C),
 
 static void rna_Image_Node_update_id(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = (bNode *)ptr->data;
 
   node->update |= NODE_UPDATE_ID;
@@ -3190,19 +3230,19 @@ static PointerRNA rna_NodeOutputFile_slot_file_get(CollectionPropertyIterator *i
 {
   PointerRNA ptr;
   bNodeSocket *sock = rna_iterator_listbase_get(iter);
-  RNA_pointer_create(iter->parent.id.data, &RNA_NodeOutputFileSlotFile, sock->storage, &ptr);
+  RNA_pointer_create(iter->parent.owner_id, &RNA_NodeOutputFileSlotFile, sock->storage, &ptr);
   return ptr;
 }
 
 static void rna_NodeColorBalance_update_lgg(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  ntreeCompositColorBalanceSyncFromLGG(ptr->id.data, ptr->data);
+  ntreeCompositColorBalanceSyncFromLGG((bNodeTree *)ptr->owner_id, ptr->data);
   rna_Node_update(bmain, scene, ptr);
 }
 
 static void rna_NodeColorBalance_update_cdl(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  ntreeCompositColorBalanceSyncFromCDL(ptr->id.data, ptr->data);
+  ntreeCompositColorBalanceSyncFromCDL((bNodeTree *)ptr->owner_id, ptr->data);
   rna_Node_update(bmain, scene, ptr);
 }
 
@@ -3241,13 +3281,13 @@ static void rna_NodeCryptomatte_matte_set(PointerRNA *ptr, const char *value)
 
 static void rna_NodeCryptomatte_update_add(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  ntreeCompositCryptomatteSyncFromAdd(ptr->id.data, ptr->data);
+  ntreeCompositCryptomatteSyncFromAdd((bNodeTree *)ptr->owner_id, ptr->data);
   rna_Node_update(bmain, scene, ptr);
 }
 
 static void rna_NodeCryptomatte_update_remove(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  ntreeCompositCryptomatteSyncFromRemove(ptr->id.data, ptr->data);
+  ntreeCompositCryptomatteSyncFromRemove((bNodeTree *)ptr->owner_id, ptr->data);
   rna_Node_update(bmain, scene, ptr);
 }
 
@@ -3257,7 +3297,7 @@ static PointerRNA rna_NodeOutputFile_slot_layer_get(CollectionPropertyIterator *
 {
   PointerRNA ptr;
   bNodeSocket *sock = rna_iterator_listbase_get(iter);
-  RNA_pointer_create(iter->parent.id.data, &RNA_NodeOutputFileSlotLayer, sock->storage, &ptr);
+  RNA_pointer_create(iter->parent.owner_id, &RNA_NodeOutputFileSlotLayer, sock->storage, &ptr);
   return ptr;
 }
 
@@ -3287,7 +3327,7 @@ static int rna_NodeOutputFileSocket_find_node(bNodeTree *ntree,
 
 static void rna_NodeOutputFileSlotFile_path_set(PointerRNA *ptr, const char *value)
 {
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   NodeImageMultiFileSocket *sockdata = ptr->data;
   bNode *node;
   bNodeSocket *sock;
@@ -3299,7 +3339,7 @@ static void rna_NodeOutputFileSlotFile_path_set(PointerRNA *ptr, const char *val
 
 static void rna_NodeOutputFileSlotLayer_name_set(PointerRNA *ptr, const char *value)
 {
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   NodeImageMultiFileSocket *sockdata = ptr->data;
   bNode *node;
   bNodeSocket *sock;
@@ -3420,7 +3460,7 @@ static void rna_ShaderNodeScript_bytecode_set(PointerRNA *ptr, const char *value
 
 static void rna_ShaderNodeScript_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = (bNode *)ptr->data;
   RenderEngineType *engine_type = RE_engines_find(scene->r.engine);
 
@@ -3436,7 +3476,7 @@ static void rna_ShaderNodeScript_update(Main *bmain, Scene *scene, PointerRNA *p
 
 static void rna_ShaderNode_socket_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = (bNode *)ptr->data;
 
   nodeUpdate(ntree, node);
@@ -3445,7 +3485,7 @@ static void rna_ShaderNode_socket_update(Main *bmain, Scene *scene, PointerRNA *
 
 static void rna_CompositorNodeScale_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  bNodeTree *ntree = (bNodeTree *)ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = (bNode *)ptr->data;
 
   nodeUpdate(ntree, node);
@@ -3476,7 +3516,7 @@ static void rna_ShaderNodePointDensity_psys_set(PointerRNA *ptr,
   NodeShaderTexPointDensity *shader_point_density = node->storage;
   Object *ob = (Object *)node->id;
 
-  if (ob && value.id.data == ob) {
+  if (ob && value.owner_id == &ob->id) {
     shader_point_density->particle_system = BLI_findindex(&ob->particlesystem, value.data) + 1;
   }
   else {
@@ -3780,6 +3820,7 @@ static void def_group(StructRNA *srna)
   RNA_def_property_pointer_funcs(
       prop, NULL, "rna_NodeGroup_node_tree_set", NULL, "rna_NodeGroup_node_tree_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Node Tree", "");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_NodeGroup_update");
 
@@ -3852,7 +3893,7 @@ static void def_math(StructRNA *srna)
   RNA_def_property_enum_sdna(prop, NULL, "custom1");
   RNA_def_property_enum_items(prop, rna_enum_node_math_items);
   RNA_def_property_ui_text(prop, "Operation", "");
-  RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Node_update");
+  RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_ShaderNode_socket_update");
 
   prop = RNA_def_property(srna, "use_clamp", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "custom2", SHD_MATH_CLAMP);
@@ -3868,7 +3909,7 @@ static void def_vector_math(StructRNA *srna)
   RNA_def_property_enum_sdna(prop, NULL, "custom1");
   RNA_def_property_enum_items(prop, rna_enum_node_vec_math_items);
   RNA_def_property_ui_text(prop, "Operation", "");
-  RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Node_update");
+  RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_ShaderNode_socket_update");
 }
 
 static void def_rgb_curve(StructRNA *srna)
@@ -3954,6 +3995,7 @@ static void def_texture(StructRNA *srna)
   RNA_def_property_pointer_sdna(prop, NULL, "id");
   RNA_def_property_struct_type(prop, "Texture");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Texture", "");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Node_update");
 
@@ -4152,6 +4194,7 @@ static void def_sh_tex_environment(StructRNA *srna)
   RNA_def_property_pointer_sdna(prop, NULL, "id");
   RNA_def_property_struct_type(prop, "Image");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Image", "");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Node_tex_image_update");
 
@@ -4230,6 +4273,7 @@ static void def_sh_tex_image(StructRNA *srna)
   RNA_def_property_pointer_sdna(prop, NULL, "id");
   RNA_def_property_struct_type(prop, "Image");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Image", "");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Node_tex_image_update");
 
@@ -4464,6 +4508,18 @@ static void def_sh_tex_wave(StructRNA *srna)
   RNA_def_property_update(prop, 0, "rna_Node_update");
 }
 
+static void def_sh_tex_white_noise(StructRNA *srna)
+{
+  PropertyRNA *prop;
+
+  prop = RNA_def_property(srna, "dimensions", PROP_ENUM, PROP_NONE);
+  RNA_def_property_enum_sdna(prop, NULL, "custom1");
+  RNA_def_property_enum_items(prop, rna_enum_node_tex_dimensions_items);
+  RNA_def_property_ui_text(
+      prop, "Dimensions", "The number of dimensions to evaluate the noise in");
+  RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_ShaderNode_socket_update");
+}
+
 static void def_sh_tex_coord(StructRNA *srna)
 {
   PropertyRNA *prop;
@@ -4472,6 +4528,7 @@ static void def_sh_tex_coord(StructRNA *srna)
   RNA_def_property_pointer_sdna(prop, NULL, "id");
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_REFCOUNT);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Object", "Use coordinates from this object (for object texture coordinates output)");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Node_update_relations");
@@ -4606,6 +4663,7 @@ static void def_sh_tex_pointdensity(StructRNA *srna)
   RNA_def_property_pointer_sdna(prop, NULL, "id");
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_REFCOUNT);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Object", "Object to take point data from");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Node_update");
 
@@ -5459,6 +5517,7 @@ static void def_cmp_image(StructRNA *srna)
   RNA_def_property_pointer_sdna(prop, NULL, "id");
   RNA_def_property_struct_type(prop, "Image");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Image", "");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Image_Node_update_id");
 
@@ -5486,6 +5545,7 @@ static void def_cmp_render_layers(StructRNA *srna)
   RNA_def_property_pointer_funcs(prop, NULL, "rna_Node_scene_set", NULL, NULL);
   RNA_def_property_struct_type(prop, "Scene");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_REFCOUNT);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Scene", "");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Node_view_layer_update");
 
@@ -6172,6 +6232,7 @@ static void def_cmp_defocus(StructRNA *srna)
   RNA_def_property_pointer_funcs(prop, NULL, "rna_Node_scene_set", NULL, NULL);
   RNA_def_property_struct_type(prop, "Scene");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_REFCOUNT);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Scene", "Scene from which to select the active camera (render scene if undefined)");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_Node_update");
@@ -6672,7 +6733,7 @@ static void def_cmp_colorbalance(StructRNA *srna)
   RNA_def_property_float_sdna(prop, NULL, "offset");
   RNA_def_property_array(prop, 3);
   RNA_def_property_ui_range(prop, 0, 1, 0.1, 3);
-  RNA_def_property_ui_text(prop, "Offset", "Correction for Shadows");
+  RNA_def_property_ui_text(prop, "Offset", "Correction for entire tonal range");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_NodeColorBalance_update_cdl");
 
   prop = RNA_def_property(srna, "power", PROP_FLOAT, PROP_COLOR_GAMMA);
@@ -7800,10 +7861,10 @@ static void rna_def_node_socket(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Type", "Data type");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_NodeSocket_update");
 
-  prop = RNA_def_property(srna, "draw_shape", PROP_ENUM, PROP_NONE);
-  RNA_def_property_enum_sdna(prop, NULL, "draw_shape");
-  RNA_def_property_enum_items(prop, rna_enum_node_socket_draw_shape_items);
-  RNA_def_property_enum_default(prop, SOCK_DRAW_SHAPE_CIRCLE);
+  prop = RNA_def_property(srna, "display_shape", PROP_ENUM, PROP_NONE);
+  RNA_def_property_enum_sdna(prop, NULL, "display_shape");
+  RNA_def_property_enum_items(prop, rna_enum_node_socket_display_shape_items);
+  RNA_def_property_enum_default(prop, SOCK_DISPLAY_SHAPE_CIRCLE);
   RNA_def_property_ui_text(prop, "Shape", "Socket shape");
   RNA_def_property_update(prop, NC_NODE | NA_EDITED, "rna_NodeSocket_update");
 
diff --git a/source/blender/makesrna/intern/rna_object.c b/source/blender/makesrna/intern/rna_object.c
index a12f45732d6..c9061c7c087 100644
--- a/source/blender/makesrna/intern/rna_object.c
+++ b/source/blender/makesrna/intern/rna_object.c
@@ -299,27 +299,27 @@ const EnumPropertyItem rna_enum_object_axis_items[] = {
 
 static void rna_Object_internal_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_TRANSFORM);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_TRANSFORM);
 }
 
 static void rna_Object_internal_update_draw(Main *UNUSED(bmain),
                                             Scene *UNUSED(scene),
                                             PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_TRANSFORM);
-  WM_main_add_notifier(NC_OBJECT | ND_DRAW, ptr->id.data);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_TRANSFORM);
+  WM_main_add_notifier(NC_OBJECT | ND_DRAW, ptr->owner_id);
 }
 
 static void rna_Object_matrix_world_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
   /* don't use compat so we get predictable rotation */
-  BKE_object_apply_mat4(ptr->id.data, ((Object *)ptr->id.data)->obmat, false, true);
+  BKE_object_apply_mat4((Object *)ptr->owner_id, ((Object *)ptr->owner_id)->obmat, false, true);
   rna_Object_internal_update(bmain, scene, ptr);
 }
 
 static void rna_Object_hide_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   BKE_main_collection_sync_remap(bmain);
   DEG_id_tag_update(&ob->id, ID_RECALC_COPY_ON_WRITE);
   DEG_relations_tag_update(bmain);
@@ -328,7 +328,7 @@ static void rna_Object_hide_update(Main *bmain, Scene *UNUSED(scene), PointerRNA
 
 static void rna_MaterialIndex_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   if (ob && ob->type == OB_GPENCIL) {
     /* notifying material property in topbar */
     WM_main_add_notifier(NC_SPACE | ND_SPACE_VIEW3D, NULL);
@@ -337,13 +337,13 @@ static void rna_MaterialIndex_update(Main *UNUSED(bmain), Scene *UNUSED(scene),
 
 static void rna_Object_matrix_local_get(PointerRNA *ptr, float values[16])
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   BKE_object_matrix_local_get(ob, (float(*)[4])values);
 }
 
 static void rna_Object_matrix_local_set(PointerRNA *ptr, const float values[16])
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   float local_mat[4][4];
 
   /* Localspace matrix is truly relative to the parent,
@@ -365,20 +365,20 @@ static void rna_Object_matrix_local_set(PointerRNA *ptr, const float values[16])
 
 static void rna_Object_matrix_basis_get(PointerRNA *ptr, float values[16])
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   BKE_object_to_mat4(ob, (float(*)[4])values);
 }
 
 static void rna_Object_matrix_basis_set(PointerRNA *ptr, const float values[16])
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   BKE_object_apply_mat4(ob, (float(*)[4])values, false, false);
 }
 
 void rna_Object_internal_update_data(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
-  WM_main_add_notifier(NC_OBJECT | ND_DRAW, ptr->id.data);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
+  WM_main_add_notifier(NC_OBJECT | ND_DRAW, ptr->owner_id);
 }
 
 void rna_Object_internal_update_data_dependency(Main *bmain, Scene *scene, PointerRNA *ptr)
@@ -389,7 +389,7 @@ void rna_Object_internal_update_data_dependency(Main *bmain, Scene *scene, Point
 
 static void rna_Object_active_shape_update(bContext *C, PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Main *bmain = CTX_data_main(C);
   Scene *scene = CTX_data_scene(C);
 
@@ -403,7 +403,7 @@ static void rna_Object_active_shape_update(bContext *C, PointerRNA *ptr)
         DEG_id_tag_update(ob->data, 0);
 
         EDBM_mesh_normals_update(((Mesh *)ob->data)->edit_mesh);
-        BKE_editmesh_tessface_calc(((Mesh *)ob->data)->edit_mesh);
+        BKE_editmesh_looptri_calc(((Mesh *)ob->data)->edit_mesh);
         break;
       case OB_CURVE:
       case OB_SURF:
@@ -422,9 +422,9 @@ static void rna_Object_active_shape_update(bContext *C, PointerRNA *ptr)
 
 static void rna_Object_dependency_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_TRANSFORM);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_TRANSFORM);
   DEG_relations_tag_update(bmain);
-  WM_main_add_notifier(NC_OBJECT | ND_PARENT, ptr->id.data);
+  WM_main_add_notifier(NC_OBJECT | ND_PARENT, ptr->owner_id);
 }
 
 static void rna_Object_data_set(PointerRNA *ptr, PointerRNA value, struct ReportList *reports)
@@ -546,27 +546,27 @@ static void rna_Object_parent_set(PointerRNA *ptr,
   }
 }
 
-bool rna_Object_parent_override_apply(Main *UNUSED(bmain),
-                                      PointerRNA *ptr_dst,
-                                      PointerRNA *ptr_src,
-                                      PointerRNA *ptr_storage,
-                                      PropertyRNA *prop_dst,
-                                      PropertyRNA *prop_src,
-                                      PropertyRNA *UNUSED(prop_storage),
-                                      const int len_dst,
-                                      const int len_src,
-                                      const int len_storage,
-                                      PointerRNA *UNUSED(ptr_item_dst),
-                                      PointerRNA *UNUSED(ptr_item_src),
-                                      PointerRNA *UNUSED(ptr_item_storage),
-                                      IDOverrideLibraryPropertyOperation *opop)
+static bool rna_Object_parent_override_apply(Main *UNUSED(bmain),
+                                             PointerRNA *ptr_dst,
+                                             PointerRNA *ptr_src,
+                                             PointerRNA *ptr_storage,
+                                             PropertyRNA *prop_dst,
+                                             PropertyRNA *prop_src,
+                                             PropertyRNA *UNUSED(prop_storage),
+                                             const int len_dst,
+                                             const int len_src,
+                                             const int len_storage,
+                                             PointerRNA *UNUSED(ptr_item_dst),
+                                             PointerRNA *UNUSED(ptr_item_src),
+                                             PointerRNA *UNUSED(ptr_item_storage),
+                                             IDOverrideLibraryPropertyOperation *opop)
 {
   BLI_assert(len_dst == len_src && (!ptr_storage || len_dst == len_storage) && len_dst == 0);
   BLI_assert(opop->operation == IDOVERRIDE_LIBRARY_OP_REPLACE &&
-             "Unsupported RNA override operation on animdata pointer");
+             "Unsupported RNA override operation on object parent pointer");
   UNUSED_VARS_NDEBUG(ptr_storage, len_dst, len_src, len_storage, opop);
 
-  /* We need a special handling here because setting parent resets pinvert parent matrix,
+  /* We need a special handling here because setting parent resets invert parent matrix,
    * which is evil in our case. */
   Object *ob = (Object *)ptr_dst->data;
   Object *parent_dst = RNA_property_pointer_get(ptr_dst, prop_dst).data;
@@ -691,7 +691,7 @@ static void rna_Object_dup_collection_set(PointerRNA *ptr,
 
 static void rna_VertexGroup_name_set(PointerRNA *ptr, const char *value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bDeformGroup *dg = (bDeformGroup *)ptr->data;
   BLI_strncpy_utf8(dg->name, value, sizeof(dg->name));
   defgroup_unique_name(dg, ob);
@@ -699,34 +699,34 @@ static void rna_VertexGroup_name_set(PointerRNA *ptr, const char *value)
 
 static int rna_VertexGroup_index_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   return BLI_findindex(&ob->defbase, ptr->data);
 }
 
 static PointerRNA rna_Object_active_vertex_group_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   return rna_pointer_inherit_refine(
       ptr, &RNA_VertexGroup, BLI_findlink(&ob->defbase, ob->actdef - 1));
 }
 
 static int rna_Object_active_vertex_group_index_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   return ob->actdef - 1;
 }
 
 static void rna_Object_active_vertex_group_index_set(PointerRNA *ptr, int value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ob->actdef = value + 1;
 }
 
 static void rna_Object_active_vertex_group_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   *min = 0;
   *max = max_ii(0, BLI_listbase_count(&ob->defbase) - 1);
@@ -734,7 +734,7 @@ static void rna_Object_active_vertex_group_index_range(
 
 void rna_object_vgroup_name_index_get(PointerRNA *ptr, char *value, int index)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bDeformGroup *dg;
 
   dg = BLI_findlink(&ob->defbase, index - 1);
@@ -749,7 +749,7 @@ void rna_object_vgroup_name_index_get(PointerRNA *ptr, char *value, int index)
 
 int rna_object_vgroup_name_index_length(PointerRNA *ptr, int index)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bDeformGroup *dg;
 
   dg = BLI_findlink(&ob->defbase, index - 1);
@@ -758,13 +758,13 @@ int rna_object_vgroup_name_index_length(PointerRNA *ptr, int index)
 
 void rna_object_vgroup_name_index_set(PointerRNA *ptr, const char *value, short *index)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   *index = defgroup_name_index(ob, value) + 1;
 }
 
 void rna_object_vgroup_name_set(PointerRNA *ptr, const char *value, char *result, int maxlen)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bDeformGroup *dg = defgroup_find_name(ob, value);
   if (dg) {
     /* No need for BLI_strncpy_utf8, since this matches an existing group. */
@@ -777,7 +777,7 @@ void rna_object_vgroup_name_set(PointerRNA *ptr, const char *value, char *result
 
 static void rna_FaceMap_name_set(PointerRNA *ptr, const char *value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bFaceMap *fmap = (bFaceMap *)ptr->data;
   BLI_strncpy_utf8(fmap->name, value, sizeof(fmap->name));
   BKE_object_facemap_unique_name(ob, fmap);
@@ -785,33 +785,33 @@ static void rna_FaceMap_name_set(PointerRNA *ptr, const char *value)
 
 static int rna_FaceMap_index_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   return BLI_findindex(&ob->fmaps, ptr->data);
 }
 
 static PointerRNA rna_Object_active_face_map_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   return rna_pointer_inherit_refine(ptr, &RNA_FaceMap, BLI_findlink(&ob->fmaps, ob->actfmap - 1));
 }
 
 static int rna_Object_active_face_map_index_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   return ob->actfmap - 1;
 }
 
 static void rna_Object_active_face_map_index_set(PointerRNA *ptr, int value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ob->actfmap = value + 1;
 }
 
 static void rna_Object_active_face_map_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   *min = 0;
   *max = max_ii(0, BLI_listbase_count(&ob->fmaps) - 1);
@@ -819,7 +819,7 @@ static void rna_Object_active_face_map_index_range(
 
 void rna_object_BKE_object_facemap_name_index_get(PointerRNA *ptr, char *value, int index)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bFaceMap *fmap;
 
   fmap = BLI_findlink(&ob->fmaps, index - 1);
@@ -834,7 +834,7 @@ void rna_object_BKE_object_facemap_name_index_get(PointerRNA *ptr, char *value,
 
 int rna_object_BKE_object_facemap_name_index_length(PointerRNA *ptr, int index)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bFaceMap *fmap;
 
   fmap = BLI_findlink(&ob->fmaps, index - 1);
@@ -843,13 +843,13 @@ int rna_object_BKE_object_facemap_name_index_length(PointerRNA *ptr, int index)
 
 void rna_object_BKE_object_facemap_name_index_set(PointerRNA *ptr, const char *value, short *index)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   *index = BKE_object_facemap_name_index(ob, value) + 1;
 }
 
 void rna_object_fmap_name_set(PointerRNA *ptr, const char *value, char *result, int maxlen)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bFaceMap *fmap = BKE_object_facemap_find_name(ob, value);
   if (fmap) {
     /* No need for BLI_strncpy_utf8, since this matches an existing group. */
@@ -862,7 +862,7 @@ void rna_object_fmap_name_set(PointerRNA *ptr, const char *value, char *result,
 
 void rna_object_uvlayer_name_set(PointerRNA *ptr, const char *value, char *result, int maxlen)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Mesh *me;
   CustomDataLayer *layer;
   int a;
@@ -885,7 +885,7 @@ void rna_object_uvlayer_name_set(PointerRNA *ptr, const char *value, char *resul
 
 void rna_object_vcollayer_name_set(PointerRNA *ptr, const char *value, char *result, int maxlen)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Mesh *me;
   CustomDataLayer *layer;
   int a;
@@ -908,13 +908,13 @@ void rna_object_vcollayer_name_set(PointerRNA *ptr, const char *value, char *res
 
 static int rna_Object_active_material_index_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   return MAX2(ob->actcol - 1, 0);
 }
 
 static void rna_Object_active_material_index_set(PointerRNA *ptr, int value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ob->actcol = value + 1;
 
   if (ob->type == OB_MESH) {
@@ -929,7 +929,7 @@ static void rna_Object_active_material_index_set(PointerRNA *ptr, int value)
 static void rna_Object_active_material_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   *min = 0;
   *max = max_ii(ob->totcol - 1, 0);
 }
@@ -937,7 +937,7 @@ static void rna_Object_active_material_index_range(
 /* returns active base material */
 static PointerRNA rna_Object_active_material_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Material *ma;
 
   ma = (ob->totcol) ? give_current_material(ob, ob->actcol) : NULL;
@@ -948,7 +948,7 @@ static void rna_Object_active_material_set(PointerRNA *ptr,
                                            PointerRNA value,
                                            struct ReportList *UNUSED(reports))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   DEG_id_tag_update(value.data, 0);
   BLI_assert(BKE_id_is_in_global_main(&ob->id));
@@ -963,7 +963,7 @@ static void rna_Object_active_material_set(PointerRNA *ptr,
 
 static int rna_Object_active_material_editable(PointerRNA *ptr, const char **UNUSED(r_info))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bool is_editable;
 
   if ((ob->matbits == NULL) || (ob->actcol == 0) || ob->matbits[ob->actcol - 1]) {
@@ -979,20 +979,20 @@ static int rna_Object_active_material_editable(PointerRNA *ptr, const char **UNU
 static void rna_Object_active_particle_system_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   *min = 0;
   *max = max_ii(0, BLI_listbase_count(&ob->particlesystem) - 1);
 }
 
 static int rna_Object_active_particle_system_index_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   return psys_get_current_num(ob);
 }
 
 static void rna_Object_active_particle_system_index_set(PointerRNA *ptr, int value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   psys_set_current_num(ob, value);
 }
 
@@ -1000,7 +1000,7 @@ static void rna_Object_particle_update(Main *UNUSED(bmain), Scene *scene, Pointe
 {
   /* TODO: Disabled for now, because bContext is not available. */
 #  if 0
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   PE_current_changed(NULL, scene, ob);
 #  else
   (void)scene;
@@ -1137,7 +1137,7 @@ static int rna_Object_rotation_4d_editable(PointerRNA *ptr, int index)
 
 static int rna_MaterialSlot_material_editable(PointerRNA *ptr, const char **UNUSED(r_info))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   const int index = (Material **)ptr->data - ob->mat;
   bool is_editable;
 
@@ -1153,7 +1153,7 @@ static int rna_MaterialSlot_material_editable(PointerRNA *ptr, const char **UNUS
 
 static PointerRNA rna_MaterialSlot_material_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Material *ma;
   const int index = (Material **)ptr->data - ob->mat;
 
@@ -1165,7 +1165,7 @@ static void rna_MaterialSlot_material_set(PointerRNA *ptr,
                                           PointerRNA value,
                                           struct ReportList *UNUSED(reports))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   int index = (Material **)ptr->data - ob->mat;
 
   BLI_assert(BKE_id_is_in_global_main(&ob->id));
@@ -1175,7 +1175,7 @@ static void rna_MaterialSlot_material_set(PointerRNA *ptr,
 
 static bool rna_MaterialSlot_material_poll(PointerRNA *ptr, PointerRNA value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Material *ma = (Material *)value.data;
 
   if (ob->type == OB_GPENCIL) {
@@ -1190,7 +1190,7 @@ static bool rna_MaterialSlot_material_poll(PointerRNA *ptr, PointerRNA value)
 
 static int rna_MaterialSlot_link_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   int index = (Material **)ptr->data - ob->mat;
 
   return ob->matbits[index] != 0;
@@ -1198,7 +1198,7 @@ static int rna_MaterialSlot_link_get(PointerRNA *ptr)
 
 static void rna_MaterialSlot_link_set(PointerRNA *ptr, int value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   int index = (Material **)ptr->data - ob->mat;
 
   if (value) {
@@ -1213,7 +1213,7 @@ static void rna_MaterialSlot_link_set(PointerRNA *ptr, int value)
 
 static int rna_MaterialSlot_name_length(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Material *ma;
   int index = (Material **)ptr->data - ob->mat;
 
@@ -1228,7 +1228,7 @@ static int rna_MaterialSlot_name_length(PointerRNA *ptr)
 
 static void rna_MaterialSlot_name_get(PointerRNA *ptr, char *str)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Material *ma;
   int index = (Material **)ptr->data - ob->mat;
 
@@ -1246,14 +1246,14 @@ static void rna_MaterialSlot_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
   rna_Object_internal_update(bmain, scene, ptr);
 
-  WM_main_add_notifier(NC_OBJECT | ND_OB_SHADING, ptr->id.data);
+  WM_main_add_notifier(NC_OBJECT | ND_OB_SHADING, ptr->owner_id);
   WM_main_add_notifier(NC_MATERIAL | ND_SHADING_LINKS, NULL);
   DEG_relations_tag_update(bmain);
 }
 
 static char *rna_MaterialSlot_path(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   int index = (Material **)ptr->data - ob->mat;
 
   return BLI_sprintfN("material_slots[%d]", index);
@@ -1271,7 +1271,7 @@ static char *rna_ObjectDisplay_path(PointerRNA *UNUSED(ptr))
 
 static PointerRNA rna_Object_active_particle_system_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ParticleSystem *psys = psys_get_current(ob);
   return rna_pointer_inherit_refine(ptr, &RNA_ParticleSystem, psys);
 }
@@ -1279,7 +1279,7 @@ static PointerRNA rna_Object_active_particle_system_get(PointerRNA *ptr)
 static void rna_Object_active_shape_key_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Key *key = BKE_key_from_object(ob);
 
   *min = 0;
@@ -1296,21 +1296,21 @@ static void rna_Object_active_shape_key_index_range(
 
 static int rna_Object_active_shape_key_index_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   return MAX2(ob->shapenr - 1, 0);
 }
 
 static void rna_Object_active_shape_key_index_set(PointerRNA *ptr, int value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   ob->shapenr = value + 1;
 }
 
 static PointerRNA rna_Object_active_shape_key_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   Key *key = BKE_key_from_object(ob);
   KeyBlock *kb;
   PointerRNA keyptr;
@@ -1326,7 +1326,7 @@ static PointerRNA rna_Object_active_shape_key_get(PointerRNA *ptr)
 
 static PointerRNA rna_Object_field_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   /* weak */
   if (!ob->pd) {
@@ -1338,7 +1338,7 @@ static PointerRNA rna_Object_field_get(PointerRNA *ptr)
 
 static PointerRNA rna_Object_collision_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   if (ob->type != OB_MESH) {
     return PointerRNA_NULL;
@@ -1354,7 +1354,7 @@ static PointerRNA rna_Object_collision_get(PointerRNA *ptr)
 
 static PointerRNA rna_Object_active_constraint_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bConstraint *con = BKE_constraints_active_get(&ob->constraints);
   return rna_pointer_inherit_refine(ptr, &RNA_Constraint, con);
 }
@@ -1363,7 +1363,7 @@ static void rna_Object_active_constraint_set(PointerRNA *ptr,
                                              PointerRNA value,
                                              struct ReportList *UNUSED(reports))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   BKE_constraints_active_set(&ob->constraints, (bConstraint *)value.data);
 }
 
@@ -1428,12 +1428,12 @@ bool rna_Object_constraints_override_apply(Main *UNUSED(bmain),
   BLI_assert(opop->operation == IDOVERRIDE_LIBRARY_OP_INSERT_AFTER &&
              "Unsupported RNA override operation on constraints collection");
 
-  Object *ob_dst = (Object *)ptr_dst->id.data;
-  Object *ob_src = (Object *)ptr_src->id.data;
+  Object *ob_dst = (Object *)ptr_dst->owner_id;
+  Object *ob_src = (Object *)ptr_src->owner_id;
 
   /* Remember that insertion operations are defined and stored in correct order, which means that
-   * even if we insert several items in a row, we alays insert first one, then second one, etc.
-   * So we should always find 'anchor' constraint in both _src *and* _dst> */
+   * even if we insert several items in a row, we always insert first one, then second one, etc.
+   * So we should always find 'anchor' constraint in both _src *and* _dst. */
   bConstraint *con_anchor = NULL;
   if (opop->subitem_local_name && opop->subitem_local_name[0]) {
     con_anchor = BLI_findstring(
@@ -1516,12 +1516,12 @@ bool rna_Object_modifiers_override_apply(Main *UNUSED(bmain),
   BLI_assert(opop->operation == IDOVERRIDE_LIBRARY_OP_INSERT_AFTER &&
              "Unsupported RNA override operation on modifiers collection");
 
-  Object *ob_dst = (Object *)ptr_dst->id.data;
-  Object *ob_src = (Object *)ptr_src->id.data;
+  Object *ob_dst = (Object *)ptr_dst->owner_id;
+  Object *ob_src = (Object *)ptr_src->owner_id;
 
   /* Remember that insertion operations are defined and stored in correct order, which means that
-   * even if we insert several items in a row, we alays insert first one, then second one, etc.
-   * So we should always find 'anchor' constraint in both _src *and* _dst> */
+   * even if we insert several items in a row, we always insert first one, then second one, etc.
+   * So we should always find 'anchor' constraint in both _src *and* _dst. */
   ModifierData *mod_anchor = NULL;
   if (opop->subitem_local_name && opop->subitem_local_name[0]) {
     mod_anchor = BLI_findstring(
@@ -1619,7 +1619,7 @@ static void rna_Object_shaderfx_clear(Object *object, bContext *C)
 
 static void rna_Object_boundbox_get(PointerRNA *ptr, float *values)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   BoundBox *bb = BKE_object_boundbox_get(ob);
   if (bb) {
     memcpy(values, bb->vec, sizeof(bb->vec));
@@ -1791,42 +1791,42 @@ static void rna_FaceMap_face_remove(
 /* generic poll functions */
 bool rna_Lattice_object_poll(PointerRNA *UNUSED(ptr), PointerRNA value)
 {
-  return ((Object *)value.id.data)->type == OB_LATTICE;
+  return ((Object *)value.owner_id)->type == OB_LATTICE;
 }
 
 bool rna_Curve_object_poll(PointerRNA *UNUSED(ptr), PointerRNA value)
 {
-  return ((Object *)value.id.data)->type == OB_CURVE;
+  return ((Object *)value.owner_id)->type == OB_CURVE;
 }
 
 bool rna_Armature_object_poll(PointerRNA *UNUSED(ptr), PointerRNA value)
 {
-  return ((Object *)value.id.data)->type == OB_ARMATURE;
+  return ((Object *)value.owner_id)->type == OB_ARMATURE;
 }
 
 bool rna_Mesh_object_poll(PointerRNA *UNUSED(ptr), PointerRNA value)
 {
-  return ((Object *)value.id.data)->type == OB_MESH;
+  return ((Object *)value.owner_id)->type == OB_MESH;
 }
 
 bool rna_Camera_object_poll(PointerRNA *UNUSED(ptr), PointerRNA value)
 {
-  return ((Object *)value.id.data)->type == OB_CAMERA;
+  return ((Object *)value.owner_id)->type == OB_CAMERA;
 }
 
 bool rna_Light_object_poll(PointerRNA *UNUSED(ptr), PointerRNA value)
 {
-  return ((Object *)value.id.data)->type == OB_LAMP;
+  return ((Object *)value.owner_id)->type == OB_LAMP;
 }
 
 bool rna_GPencil_object_poll(PointerRNA *UNUSED(ptr), PointerRNA value)
 {
-  return ((Object *)value.id.data)->type == OB_GPENCIL;
+  return ((Object *)value.owner_id)->type == OB_GPENCIL;
 }
 
 int rna_Object_use_dynamic_topology_sculpting_get(PointerRNA *ptr)
 {
-  SculptSession *ss = ((Object *)ptr->id.data)->sculpt;
+  SculptSession *ss = ((Object *)ptr->owner_id)->sculpt;
   return (ss && ss->bm);
 }
 
@@ -2509,10 +2509,12 @@ static void rna_def_object(BlenderRNA *brna)
 
   /* proxy */
   prop = RNA_def_property(srna, "proxy", PROP_POINTER, PROP_NONE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_NO_COMPARISON);
   RNA_def_property_ui_text(prop, "Proxy", "Library object this proxy object controls");
 
   prop = RNA_def_property(srna, "proxy_collection", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "proxy_group");
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_NO_COMPARISON);
   RNA_def_property_ui_text(
       prop, "Proxy Collection", "Library collection duplicator object this proxy object controls");
 
@@ -2848,6 +2850,14 @@ static void rna_def_object(BlenderRNA *brna)
       prop, "Display in Orthographic Mode", "Display image in orthographic mode");
   RNA_def_property_update(prop, NC_OBJECT | ND_DRAW, NULL);
 
+  prop = RNA_def_property(srna, "show_empty_image_only_axis_aligned", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(
+      prop, NULL, "empty_image_visibility_flag", OB_EMPTY_IMAGE_HIDE_NON_AXIS_ALIGNED);
+  RNA_def_property_ui_text(prop,
+                           "Display Only Axis Aligned",
+                           "Only display the image when it is aligned with the view axis");
+  RNA_def_property_update(prop, NC_OBJECT | ND_DRAW, NULL);
+
   prop = RNA_def_property(srna, "use_empty_image_alpha", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "empty_image_flag", OB_EMPTY_IMAGE_USE_ALPHA_BLEND);
   RNA_def_property_ui_text(
@@ -2878,7 +2888,7 @@ static void rna_def_object(BlenderRNA *brna)
   prop = RNA_def_property(srna, "color", PROP_FLOAT, PROP_COLOR);
   RNA_def_property_ui_text(
       prop, "Color", "Object color and alpha, used when faces have the ObColor mode enabled");
-  RNA_def_property_update(prop, NC_OBJECT | ND_DRAW, NULL);
+  RNA_def_property_update(prop, NC_OBJECT | ND_DRAW, "rna_Object_internal_update_draw");
 
   /* physics */
   prop = RNA_def_property(srna, "field", PROP_POINTER, PROP_NONE);
@@ -2983,6 +2993,7 @@ static void rna_def_object(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "Collection");
   RNA_def_property_pointer_sdna(prop, NULL, "instance_collection");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_pointer_funcs(prop, NULL, "rna_Object_dup_collection_set", NULL, NULL);
   RNA_def_property_ui_text(prop, "Instance Collection", "Instance an existing collection");
   RNA_def_property_update(prop, NC_OBJECT | ND_DRAW, "rna_Object_dependency_update");
diff --git a/source/blender/makesrna/intern/rna_object_api.c b/source/blender/makesrna/intern/rna_object_api.c
index 3f0b5546a03..5bec6f0c7d3 100644
--- a/source/blender/makesrna/intern/rna_object_api.c
+++ b/source/blender/makesrna/intern/rna_object_api.c
@@ -244,7 +244,7 @@ static Base *rna_Object_local_view_property_helper(
 
 static bool rna_Object_local_view_get(Object *ob, ReportList *reports, PointerRNA *v3d_ptr)
 {
-  bScreen *sc = v3d_ptr->id.data;
+  bScreen *sc = (bScreen *)v3d_ptr->owner_id;
   View3D *v3d = v3d_ptr->data;
   Base *base = rna_Object_local_view_property_helper(sc, v3d, ob, reports, NULL);
   if (base == NULL) {
@@ -258,7 +258,7 @@ static void rna_Object_local_view_set(Object *ob,
                                       PointerRNA *v3d_ptr,
                                       bool state)
 {
-  bScreen *sc = v3d_ptr->id.data;
+  bScreen *sc = (bScreen *)v3d_ptr->owner_id;
   View3D *v3d = v3d_ptr->data;
   Scene *scene;
   Base *base = rna_Object_local_view_property_helper(sc, v3d, ob, reports, &scene);
@@ -704,7 +704,7 @@ bool rna_Object_generate_gpencil_strokes(Object *ob,
   Main *bmain = CTX_data_main(C);
   Scene *scene = CTX_data_scene(C);
 
-  BKE_gpencil_convert_curve(bmain, scene, ob_gpencil, ob, gpencil_lines, use_collections);
+  BKE_gpencil_convert_curve(bmain, scene, ob_gpencil, ob, gpencil_lines, use_collections, false);
 
   WM_main_add_notifier(NC_GPENCIL | ND_DATA, NULL);
 
diff --git a/source/blender/makesrna/intern/rna_object_force.c b/source/blender/makesrna/intern/rna_object_force.c
index 066e350ab3e..430059c7a18 100644
--- a/source/blender/makesrna/intern/rna_object_force.c
+++ b/source/blender/makesrna/intern/rna_object_force.c
@@ -87,15 +87,15 @@ static const EnumPropertyItem empty_shape_items[] = {
 static const EnumPropertyItem vortex_shape_items[] = {
     {PFIELD_SHAPE_POINT, "POINT", 0, "Point", ""},
     {PFIELD_SHAPE_PLANE, "PLANE", 0, "Plane", ""},
-    {PFIELD_SHAPE_SURFACE, "SURFACE", 0, "Surface falloff (New)", ""},
-    {PFIELD_SHAPE_POINTS, "POINTS", 0, "Every Point (New)", ""},
+    {PFIELD_SHAPE_SURFACE, "SURFACE", 0, "Surface", ""},
+    {PFIELD_SHAPE_POINTS, "POINTS", 0, "Every Point", ""},
     {0, NULL, 0, NULL, NULL},
 };
 
 static const EnumPropertyItem curve_vortex_shape_items[] = {
     {PFIELD_SHAPE_POINT, "POINT", 0, "Point", ""},
     {PFIELD_SHAPE_PLANE, "PLANE", 0, "Plane", ""},
-    {PFIELD_SHAPE_SURFACE, "SURFACE", 0, "Curve (New)", ""},
+    {PFIELD_SHAPE_SURFACE, "SURFACE", 0, "Curve", ""},
     {0, NULL, 0, NULL, NULL},
 };
 
@@ -120,18 +120,38 @@ static const EnumPropertyItem empty_vortex_shape_items[] = {
 
 #  include "ED_object.h"
 
+static bool rna_Cache_get_valid_owner_ID(PointerRNA *ptr, Object **ob, Scene **scene)
+{
+  switch (GS(ptr->owner_id->name)) {
+    case ID_OB:
+      *ob = (Object *)ptr->owner_id;
+      break;
+    case ID_SCE:
+      *scene = (Scene *)ptr->owner_id;
+      break;
+    default:
+      BLI_assert(!"Trying to get PTCacheID from an invalid ID type "
+                  "(Only scenes and objects are supported).");
+      break;
+  }
+
+  return (*ob != NULL || *scene != NULL);
+}
+
 static void rna_Cache_change(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
-  PointCache *cache = (PointCache *)ptr->data;
+  Object *ob = NULL;
+  Scene *scene = NULL;
 
-  if (!ob) {
+  if (!rna_Cache_get_valid_owner_ID(ptr, &ob, &scene)) {
     return;
   }
 
+  PointCache *cache = (PointCache *)ptr->data;
+
   cache->flag |= PTCACHE_OUTDATED;
 
-  PTCacheID pid = BKE_ptcache_id_find(ob, NULL, cache);
+  PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
 
   DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
 
@@ -146,14 +166,16 @@ static void rna_Cache_change(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerR
 
 static void rna_Cache_toggle_disk_cache(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
-  PointCache *cache = (PointCache *)ptr->data;
+  Object *ob = NULL;
+  Scene *scene = NULL;
 
-  if (!ob) {
+  if (!rna_Cache_get_valid_owner_ID(ptr, &ob, &scene)) {
     return;
   }
 
-  PTCacheID pid = BKE_ptcache_id_find(ob, NULL, cache);
+  PointCache *cache = (PointCache *)ptr->data;
+
+  PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
 
   /* smoke can only use disk cache */
   if (pid.cache && pid.type != PTCACHE_TYPE_SMOKE_DOMAIN) {
@@ -166,18 +188,20 @@ static void rna_Cache_toggle_disk_cache(Main *UNUSED(bmain), Scene *UNUSED(scene
 
 static void rna_Cache_idname_change(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
-  PointCache *cache = (PointCache *)ptr->data;
-  bool use_new_name = true;
+  Object *ob = NULL;
+  Scene *scene = NULL;
 
-  if (!ob) {
+  if (!rna_Cache_get_valid_owner_ID(ptr, &ob, &scene)) {
     return;
   }
 
+  PointCache *cache = (PointCache *)ptr->data;
+  bool use_new_name = true;
+
   /* TODO: check for proper characters */
 
   if (cache->flag & PTCACHE_EXTERNAL) {
-    PTCacheID pid = BKE_ptcache_id_find(ob, NULL, cache);
+    PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
 
     if (pid.cache) {
       BKE_ptcache_load_external(&pid);
@@ -190,7 +214,7 @@ static void rna_Cache_idname_change(Main *UNUSED(bmain), Scene *UNUSED(scene), P
     PTCacheID *pid = NULL, *pid2 = NULL;
     ListBase pidlist;
 
-    BKE_ptcache_ids_from_object(&pidlist, ob, NULL, 0);
+    BKE_ptcache_ids_from_object(&pidlist, ob, scene, 0);
 
     for (pid = pidlist.first; pid; pid = pid->next) {
       if (pid->cache == cache) {
@@ -240,13 +264,19 @@ static void rna_Cache_list_begin(CollectionPropertyIterator *iter, PointerRNA *p
 static void rna_Cache_active_point_cache_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Object *ob = ptr->id.data;
-  PointCache *cache = ptr->data;
-  PTCacheID pid = BKE_ptcache_id_find(ob, NULL, cache);
-
   *min = 0;
   *max = 0;
 
+  Object *ob = NULL;
+  Scene *scene = NULL;
+
+  if (!rna_Cache_get_valid_owner_ID(ptr, &ob, &scene)) {
+    return;
+  }
+
+  PointCache *cache = ptr->data;
+  PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
+
   if (pid.cache) {
     *max = max_ii(0, BLI_listbase_count(pid.ptcaches) - 1);
   }
@@ -254,11 +284,18 @@ static void rna_Cache_active_point_cache_index_range(
 
 static int rna_Cache_active_point_cache_index_get(PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
-  PointCache *cache = ptr->data;
-  PTCacheID pid = BKE_ptcache_id_find(ob, NULL, cache);
   int num = 0;
 
+  Object *ob = NULL;
+  Scene *scene = NULL;
+
+  if (!rna_Cache_get_valid_owner_ID(ptr, &ob, &scene)) {
+    return num;
+  }
+
+  PointCache *cache = ptr->data;
+  PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
+
   if (pid.cache) {
     num = BLI_findindex(pid.ptcaches, cache);
   }
@@ -268,9 +305,15 @@ static int rna_Cache_active_point_cache_index_get(PointerRNA *ptr)
 
 static void rna_Cache_active_point_cache_index_set(struct PointerRNA *ptr, int value)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = NULL;
+  Scene *scene = NULL;
+
+  if (!rna_Cache_get_valid_owner_ID(ptr, &ob, &scene)) {
+    return;
+  }
+
   PointCache *cache = ptr->data;
-  PTCacheID pid = BKE_ptcache_id_find(ob, NULL, cache);
+  PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
 
   if (pid.cache) {
     *(pid.cache_ptr) = BLI_findlink(pid.ptcaches, value);
@@ -280,13 +323,19 @@ static void rna_Cache_active_point_cache_index_set(struct PointerRNA *ptr, int v
 static void rna_PointCache_frame_step_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Object *ob = ptr->id.data;
-  PointCache *cache = ptr->data;
-  PTCacheID pid = BKE_ptcache_id_find(ob, NULL, cache);
-
   *min = 1;
   *max = 20;
 
+  Object *ob = NULL;
+  Scene *scene = NULL;
+
+  if (!rna_Cache_get_valid_owner_ID(ptr, &ob, &scene)) {
+    return;
+  }
+
+  PointCache *cache = ptr->data;
+  PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
+
   if (pid.cache) {
     *max = pid.max_step;
   }
@@ -294,14 +343,16 @@ static void rna_PointCache_frame_step_range(
 
 int rna_Cache_info_length(PointerRNA *ptr)
 {
-  PointCache *cache = (PointCache *)ptr->data;
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = NULL;
+  Scene *scene = NULL;
 
-  if (!ob) {
+  if (!rna_Cache_get_valid_owner_ID(ptr, &ob, &scene)) {
     return 0;
   }
 
-  PTCacheID pid = BKE_ptcache_id_find(ob, NULL, cache);
+  PointCache *cache = (PointCache *)ptr->data;
+
+  PTCacheID pid = BKE_ptcache_id_find(ob, scene, cache);
 
   if (cache->flag & PTCACHE_FLAG_INFO_DIRTY) {
     BKE_ptcache_update_info(&pid);
@@ -314,7 +365,7 @@ static char *rna_CollisionSettings_path(PointerRNA *UNUSED(ptr))
 {
   /* both methods work ok, but return the shorter path */
 #  if 0
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ModifierData *md = (ModifierData *)modifiers_findByType(ob, eModifierType_Collision);
 
   if (md) {
@@ -334,13 +385,13 @@ static char *rna_CollisionSettings_path(PointerRNA *UNUSED(ptr))
 
 static bool rna_SoftBodySettings_use_edges_get(PointerRNA *ptr)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   return (((data->softflag) & OB_SB_EDGES) != 0);
 }
 
 static void rna_SoftBodySettings_use_edges_set(PointerRNA *ptr, bool value)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   if (value) {
     data->softflag |= OB_SB_EDGES;
   }
@@ -351,13 +402,13 @@ static void rna_SoftBodySettings_use_edges_set(PointerRNA *ptr, bool value)
 
 static bool rna_SoftBodySettings_use_goal_get(PointerRNA *ptr)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   return (((data->softflag) & OB_SB_GOAL) != 0);
 }
 
 static void rna_SoftBodySettings_use_goal_set(PointerRNA *ptr, bool value)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   if (value) {
     data->softflag |= OB_SB_GOAL;
   }
@@ -368,13 +419,13 @@ static void rna_SoftBodySettings_use_goal_set(PointerRNA *ptr, bool value)
 
 static bool rna_SoftBodySettings_stiff_quads_get(PointerRNA *ptr)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   return (((data->softflag) & OB_SB_QUADS) != 0);
 }
 
 static void rna_SoftBodySettings_stiff_quads_set(PointerRNA *ptr, bool value)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   if (value) {
     data->softflag |= OB_SB_QUADS;
   }
@@ -385,13 +436,13 @@ static void rna_SoftBodySettings_stiff_quads_set(PointerRNA *ptr, bool value)
 
 static bool rna_SoftBodySettings_self_collision_get(PointerRNA *ptr)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   return (((data->softflag) & OB_SB_SELF) != 0);
 }
 
 static void rna_SoftBodySettings_self_collision_set(PointerRNA *ptr, bool value)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   if (value) {
     data->softflag |= OB_SB_SELF;
   }
@@ -402,7 +453,7 @@ static void rna_SoftBodySettings_self_collision_set(PointerRNA *ptr, bool value)
 
 static int rna_SoftBodySettings_new_aero_get(PointerRNA *ptr)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   if (data->softflag & OB_SB_AERO_ANGLE) {
     return 1;
   }
@@ -413,7 +464,7 @@ static int rna_SoftBodySettings_new_aero_get(PointerRNA *ptr)
 
 static void rna_SoftBodySettings_new_aero_set(PointerRNA *ptr, int value)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   if (value == 1) {
     data->softflag |= OB_SB_AERO_ANGLE;
   }
@@ -424,13 +475,13 @@ static void rna_SoftBodySettings_new_aero_set(PointerRNA *ptr, int value)
 
 static bool rna_SoftBodySettings_face_collision_get(PointerRNA *ptr)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   return (((data->softflag) & OB_SB_FACECOLL) != 0);
 }
 
 static void rna_SoftBodySettings_face_collision_set(PointerRNA *ptr, bool value)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   if (value) {
     data->softflag |= OB_SB_FACECOLL;
   }
@@ -441,13 +492,13 @@ static void rna_SoftBodySettings_face_collision_set(PointerRNA *ptr, bool value)
 
 static bool rna_SoftBodySettings_edge_collision_get(PointerRNA *ptr)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   return (((data->softflag) & OB_SB_EDGECOLL) != 0);
 }
 
 static void rna_SoftBodySettings_edge_collision_set(PointerRNA *ptr, bool value)
 {
-  Object *data = (Object *)(ptr->id.data);
+  Object *data = (Object *)(ptr->owner_id);
   if (value) {
     data->softflag |= OB_SB_EDGECOLL;
   }
@@ -488,7 +539,7 @@ static void rna_SoftBodySettings_spring_vgroup_set(PointerRNA *ptr, const char *
 
 static char *rna_SoftBodySettings_path(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ModifierData *md = (ModifierData *)modifiers_findByType(ob, eModifierType_Softbody);
   char name_esc[sizeof(md->name) * 2];
 
@@ -498,7 +549,7 @@ static char *rna_SoftBodySettings_path(PointerRNA *ptr)
 
 static int particle_id_check(PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   return (GS(id->name) == ID_PA);
 }
@@ -506,7 +557,7 @@ static int particle_id_check(PointerRNA *ptr)
 static void rna_FieldSettings_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
   if (particle_id_check(ptr)) {
-    ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+    ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
     if (part->pd->forcefield != PFIELD_TEXTURE && part->pd->tex) {
       id_us_min(&part->pd->tex->id);
@@ -524,7 +575,7 @@ static void rna_FieldSettings_update(Main *UNUSED(bmain), Scene *UNUSED(scene),
     WM_main_add_notifier(NC_OBJECT | ND_DRAW, NULL);
   }
   else {
-    Object *ob = (Object *)ptr->id.data;
+    Object *ob = (Object *)ptr->owner_id;
 
     if (ob->pd->forcefield != PFIELD_TEXTURE && ob->pd->tex) {
       id_us_min(&ob->pd->tex->id);
@@ -539,7 +590,7 @@ static void rna_FieldSettings_update(Main *UNUSED(bmain), Scene *UNUSED(scene),
 static void rna_FieldSettings_shape_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
   if (!particle_id_check(ptr)) {
-    Object *ob = (Object *)ptr->id.data;
+    Object *ob = (Object *)ptr->owner_id;
     ED_object_check_force_modifiers(bmain, scene, ob);
 
     DEG_id_tag_update(&ob->id, ID_RECALC_TRANSFORM);
@@ -555,7 +606,7 @@ static void rna_FieldSettings_type_set(PointerRNA *ptr, int value)
   part_deflect->forcefield = value;
 
   if (!particle_id_check(ptr)) {
-    Object *ob = (Object *)ptr->id.data;
+    Object *ob = (Object *)ptr->owner_id;
     ob->pd->forcefield = value;
     if (ELEM(value, PFIELD_WIND, PFIELD_VORTEX)) {
       ob->empty_drawtype = OB_SINGLE_ARROW;
@@ -571,12 +622,12 @@ static void rna_FieldSettings_dependency_update(Main *bmain, Scene *scene, Point
   DEG_relations_tag_update(bmain);
 
   if (particle_id_check(ptr)) {
-    DEG_id_tag_update((ID *)ptr->id.data,
+    DEG_id_tag_update(ptr->owner_id,
                       ID_RECALC_TRANSFORM | ID_RECALC_GEOMETRY | ID_RECALC_ANIMATION |
                           ID_RECALC_PSYS_RESET);
   }
   else {
-    Object *ob = (Object *)ptr->id.data;
+    Object *ob = (Object *)ptr->owner_id;
 
     rna_FieldSettings_shape_update(bmain, scene, ptr);
 
@@ -599,7 +650,7 @@ static char *rna_FieldSettings_path(PointerRNA *ptr)
 
   if (particle_id_check(ptr)) {
     /* particle system force field */
-    ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+    ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
     if (part->pd == pd) {
       return BLI_strdup("force_field_1");
@@ -610,7 +661,7 @@ static char *rna_FieldSettings_path(PointerRNA *ptr)
   }
   else {
     /* object force field */
-    Object *ob = (Object *)ptr->id.data;
+    Object *ob = (Object *)ptr->owner_id;
 
     if (ob->pd == pd) {
       return BLI_strdup("field");
@@ -621,7 +672,7 @@ static char *rna_FieldSettings_path(PointerRNA *ptr)
 
 static void rna_EffectorWeight_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   if (id && GS(id->name) == ID_SCE) {
     Scene *scene = (Scene *)id;
@@ -642,7 +693,7 @@ static void rna_EffectorWeight_dependency_update(Main *bmain,
 {
   DEG_relations_tag_update(bmain);
 
-  DEG_id_tag_update((ID *)ptr->id.data, ID_RECALC_GEOMETRY | ID_RECALC_PSYS_RESET);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY | ID_RECALC_PSYS_RESET);
 
   WM_main_add_notifier(NC_OBJECT | ND_DRAW, NULL);
 }
@@ -654,14 +705,14 @@ static char *rna_EffectorWeight_path(PointerRNA *ptr)
 
   if (particle_id_check(ptr)) {
     /* particle effector weights */
-    ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+    ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
     if (part->effector_weights == ew) {
       return BLI_strdup("effector_weights");
     }
   }
   else {
-    Object *ob = (Object *)ptr->id.data;
+    Object *ob = (Object *)ptr->owner_id;
     ModifierData *md;
 
     /* check softbody modifier */
@@ -727,7 +778,7 @@ static char *rna_EffectorWeight_path(PointerRNA *ptr)
 
 static void rna_CollisionSettings_dependency_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ModifierData *md = modifiers_findByType(ob, eModifierType_Collision);
 
   /* add/remove modifier as needed */
@@ -745,7 +796,7 @@ static void rna_CollisionSettings_update(Main *UNUSED(bmain),
                                          Scene *UNUSED(scene),
                                          PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   DEG_id_tag_update(&ob->id, ID_RECALC_TRANSFORM | ID_RECALC_GEOMETRY | ID_RECALC_ANIMATION);
   WM_main_add_notifier(NC_OBJECT | ND_DRAW, ob);
@@ -753,7 +804,7 @@ static void rna_CollisionSettings_update(Main *UNUSED(bmain),
 
 static void rna_softbody_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
   WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ob);
@@ -776,7 +827,7 @@ static const EnumPropertyItem *rna_Effector_shape_itemf(bContext *UNUSED(C),
     return empty_shape_items;
   }
 
-  ob = (Object *)ptr->id.data;
+  ob = (Object *)ptr->owner_id;
 
   if (ob->type == OB_CURVE) {
     if (ob->pd->forcefield == PFIELD_VORTEX) {
@@ -888,7 +939,7 @@ static void rna_def_pointcache_common(StructRNA *srna)
   prop = RNA_def_property(srna, "info", PROP_STRING, PROP_NONE);
   RNA_def_property_string_sdna(prop, NULL, "info");
   RNA_def_property_clear_flag(prop, PROP_EDITABLE);
-  /* Note that we do not actually need a getter here, `rna_Cache_info_length` will upate the info
+  /* Note that we do not actually need a getter here, `rna_Cache_info_length` will update the info
    * string just as well. */
   RNA_def_property_string_funcs(prop, NULL, "rna_Cache_info_length", NULL);
   RNA_def_property_string_maxlength(
@@ -1262,13 +1313,17 @@ static void rna_def_field(BlenderRNA *brna)
        ICON_FORCE_LENNARDJONES,
        "Lennard-Jones",
        "Forcefield based on the Lennard-Jones potential"},
-      {PFIELD_TEXTURE, "TEXTURE", ICON_FORCE_TEXTURE, "Texture", "Forcefield based on a texture"},
+      {PFIELD_TEXTURE, "TEXTURE", ICON_FORCE_TEXTURE, "Texture", "Force field based on a texture"},
       {PFIELD_GUIDE,
        "GUIDE",
        ICON_FORCE_CURVE,
        "Curve Guide",
        "Create a force along a curve object"},
-      {PFIELD_BOID, "BOID", ICON_FORCE_BOID, "Boid", ""},
+      {PFIELD_BOID,
+       "BOID",
+       ICON_FORCE_BOID,
+       "Boid",
+       "Create a force that acts as a boid's predators or target"},
       {PFIELD_TURBULENCE,
        "TURBULENCE",
        ICON_FORCE_TURBULENCE,
diff --git a/source/blender/makesrna/intern/rna_particle.c b/source/blender/makesrna/intern/rna_particle.c
index 86ce0ade96b..98ecb053641 100644
--- a/source/blender/makesrna/intern/rna_particle.c
+++ b/source/blender/makesrna/intern/rna_particle.c
@@ -146,7 +146,7 @@ static void rna_ParticleHairKey_location_object_info(PointerRNA *ptr,
                                                      ParticleData **pa_pt)
 {
   HairKey *hkey = (HairKey *)ptr->data;
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ModifierData *md;
   ParticleSystemModifierData *psmd = NULL;
   ParticleSystem *psys;
@@ -193,7 +193,7 @@ static void rna_ParticleHairKey_location_object_info(PointerRNA *ptr,
 static void rna_ParticleHairKey_location_object_get(PointerRNA *ptr, float *values)
 {
   HairKey *hkey = (HairKey *)ptr->data;
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ParticleSystemModifierData *psmd;
   ParticleData *pa;
 
@@ -221,7 +221,7 @@ static void rna_ParticleHairKey_location_object_get(PointerRNA *ptr, float *valu
 static void rna_ParticleHairKey_location_object_set(PointerRNA *ptr, const float *values)
 {
   HairKey *hkey = (HairKey *)ptr->data;
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ParticleSystemModifierData *psmd;
   ParticleData *pa;
 
@@ -618,7 +618,7 @@ static void rna_ParticleSystem_mcol_on_emitter(ParticleSystem *particlesystem,
 static void particle_recalc(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr, short flag)
 {
   if (ptr->type == &RNA_ParticleSystem) {
-    Object *ob = ptr->id.data;
+    Object *ob = (Object *)ptr->owner_id;
     ParticleSystem *psys = (ParticleSystem *)ptr->data;
 
     psys->recalc = flag;
@@ -626,7 +626,7 @@ static void particle_recalc(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRN
     DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
   }
   else {
-    DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY | flag);
+    DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY | flag);
   }
 
   WM_main_add_notifier(NC_OBJECT | ND_PARTICLE | NA_EDITED, NULL);
@@ -663,7 +663,7 @@ static void rna_Particle_reset_dependency(Main *bmain, Scene *scene, PointerRNA
 
 static void rna_Particle_change_type(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ParticleSettings *part = ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
   /* Iterating over all object is slow, but no better solution exists at the moment. */
   for (Object *ob = bmain->objects.first; ob; ob = ob->id.next) {
@@ -716,7 +716,7 @@ static void rna_Particle_redo_child(Main *bmain, Scene *scene, PointerRNA *ptr)
 
 static void rna_Particle_cloth_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
   WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ob);
@@ -741,7 +741,7 @@ static ParticleSystem *rna_particle_system_for_target(Object *ob, ParticleTarget
 static void rna_Particle_target_reset(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
   if (ptr->type == &RNA_ParticleTarget) {
-    Object *ob = (Object *)ptr->id.data;
+    Object *ob = (Object *)ptr->owner_id;
     ParticleTarget *pt = (ParticleTarget *)ptr->data;
     ParticleSystem *kpsys = NULL, *psys = rna_particle_system_for_target(ob, pt);
 
@@ -780,7 +780,7 @@ static void rna_Particle_target_reset(Main *bmain, Scene *UNUSED(scene), Pointer
 static void rna_Particle_target_redo(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
   if (ptr->type == &RNA_ParticleTarget) {
-    Object *ob = (Object *)ptr->id.data;
+    Object *ob = (Object *)ptr->owner_id;
     ParticleTarget *pt = (ParticleTarget *)ptr->data;
     ParticleSystem *psys = rna_particle_system_for_target(ob, pt);
 
@@ -793,7 +793,7 @@ static void rna_Particle_target_redo(Main *UNUSED(bmain), Scene *UNUSED(scene),
 
 static void rna_Particle_hair_dynamics_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ParticleSystem *psys = (ParticleSystem *)ptr->data;
 
   if (psys && !psys->clmd) {
@@ -823,7 +823,7 @@ static void rna_particle_settings_set(PointerRNA *ptr,
                                       PointerRNA value,
                                       struct ReportList *UNUSED(reports))
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ParticleSystem *psys = (ParticleSystem *)ptr->data;
   int old_type = 0;
 
@@ -1000,7 +1000,7 @@ static void rna_ParticleSettings_use_twist_curve_update(Main *bmain, Scene *scen
 
 static void rna_ParticleSystem_name_set(PointerRNA *ptr, const char *value)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ParticleSystem *part = (ParticleSystem *)ptr->data;
 
   /* copy the new name into the name slot */
@@ -1076,7 +1076,7 @@ static void rna_ParticleTarget_name_get(PointerRNA *ptr, char *str)
       psys = BLI_findlink(&pt->ob->particlesystem, pt->psys - 1);
     }
     else {
-      Object *ob = (Object *)ptr->id.data;
+      Object *ob = (Object *)ptr->owner_id;
       psys = BLI_findlink(&ob->particlesystem, pt->psys - 1);
     }
 
@@ -1108,7 +1108,7 @@ static int rna_ParticleTarget_name_length(PointerRNA *ptr)
 
 static int particle_id_check(PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   return (GS(id->name) == ID_PA);
 }
@@ -1118,7 +1118,7 @@ static char *rna_SPHFluidSettings_path(PointerRNA *ptr)
   SPHFluidSettings *fluid = (SPHFluidSettings *)ptr->data;
 
   if (particle_id_check(ptr)) {
-    ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+    ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
     if (part->fluid == fluid) {
       return BLI_strdup("fluid");
@@ -1152,7 +1152,7 @@ static bool rna_ParticleSystem_edited_get(PointerRNA *ptr)
 }
 static PointerRNA rna_ParticleDupliWeight_active_get(PointerRNA *ptr)
 {
-  ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
   ParticleDupliWeight *dw = part->instance_weights.first;
 
   for (; dw; dw = dw->next) {
@@ -1165,14 +1165,14 @@ static PointerRNA rna_ParticleDupliWeight_active_get(PointerRNA *ptr)
 static void rna_ParticleDupliWeight_active_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
   *min = 0;
   *max = max_ii(0, BLI_listbase_count(&part->instance_weights) - 1);
 }
 
 static int rna_ParticleDupliWeight_active_index_get(PointerRNA *ptr)
 {
-  ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
   ParticleDupliWeight *dw = part->instance_weights.first;
   int i = 0;
 
@@ -1187,7 +1187,7 @@ static int rna_ParticleDupliWeight_active_index_get(PointerRNA *ptr)
 
 static void rna_ParticleDupliWeight_active_index_set(struct PointerRNA *ptr, int value)
 {
-  ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
   ParticleDupliWeight *dw = part->instance_weights.first;
   int i = 0;
 
@@ -1203,7 +1203,7 @@ static void rna_ParticleDupliWeight_active_index_set(struct PointerRNA *ptr, int
 
 static void rna_ParticleDupliWeight_name_get(PointerRNA *ptr, char *str)
 {
-  ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
   psys_find_group_weights(part);
 
   ParticleDupliWeight *dw = ptr->data;
@@ -1241,7 +1241,7 @@ static const EnumPropertyItem *rna_Particle_dist_itemf(bContext *UNUSED(C),
                                                        PropertyRNA *UNUSED(prop),
                                                        bool *UNUSED(r_free))
 {
-  ParticleSettings *part = ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
   if (part->type == PART_HAIR) {
     return part_hair_dist_items;
@@ -1256,7 +1256,7 @@ static const EnumPropertyItem *rna_Particle_draw_as_itemf(bContext *UNUSED(C),
                                                           PropertyRNA *UNUSED(prop),
                                                           bool *UNUSED(r_free))
 {
-  ParticleSettings *part = ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
   if (part->type == PART_HAIR) {
     return part_hair_draw_as_items;
@@ -1271,7 +1271,7 @@ static const EnumPropertyItem *rna_Particle_ren_as_itemf(bContext *UNUSED(C),
                                                          PropertyRNA *UNUSED(prop),
                                                          bool *UNUSED(r_free))
 {
-  ParticleSettings *part = ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
   if (part->type == PART_HAIR) {
     return part_hair_ren_as_items;
@@ -1283,7 +1283,7 @@ static const EnumPropertyItem *rna_Particle_ren_as_itemf(bContext *UNUSED(C),
 
 static PointerRNA rna_Particle_field1_get(PointerRNA *ptr)
 {
-  ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
   /* weak */
   if (!part->pd) {
@@ -1295,7 +1295,7 @@ static PointerRNA rna_Particle_field1_get(PointerRNA *ptr)
 
 static PointerRNA rna_Particle_field2_get(PointerRNA *ptr)
 {
-  ParticleSettings *part = (ParticleSettings *)ptr->id.data;
+  ParticleSettings *part = (ParticleSettings *)ptr->owner_id;
 
   /* weak */
   if (!part->pd2) {
@@ -1307,7 +1307,7 @@ static PointerRNA rna_Particle_field2_get(PointerRNA *ptr)
 
 static void psys_vg_name_get__internal(PointerRNA *ptr, char *value, int index)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ParticleSystem *psys = (ParticleSystem *)ptr->data;
 
   if (psys->vgroup[index] > 0) {
@@ -1323,7 +1323,7 @@ static void psys_vg_name_get__internal(PointerRNA *ptr, char *value, int index)
 }
 static int psys_vg_name_len__internal(PointerRNA *ptr, int index)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ParticleSystem *psys = (ParticleSystem *)ptr->data;
 
   if (psys->vgroup[index] > 0) {
@@ -1337,7 +1337,7 @@ static int psys_vg_name_len__internal(PointerRNA *ptr, int index)
 }
 static void psys_vg_name_set__internal(PointerRNA *ptr, const char *value, int index)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   ParticleSystem *psys = (ParticleSystem *)ptr->data;
 
   if (value[0] == '\0') {
@@ -2041,6 +2041,7 @@ static void rna_def_particle_settings_mtex(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "object");
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Object", "Object to use for mapping with Object texture coordinates");
   RNA_def_property_update(prop, 0, "rna_Particle_reset_dependency");
@@ -2986,6 +2987,7 @@ static void rna_def_particle_settings(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "Collection");
   RNA_def_property_pointer_sdna(prop, NULL, "collision_group");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Collision Collection", "Limit colliders to this collection");
   RNA_def_property_update(prop, 0, "rna_Particle_reset_dependency");
 
@@ -3313,6 +3315,7 @@ static void rna_def_particle_settings(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "instance_collection");
   RNA_def_property_struct_type(prop, "Collection");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_REFCOUNT);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Dupli Collection", "Show Objects in this collection in place of particles");
   RNA_def_property_update(prop, 0, "rna_Particle_redo_count");
@@ -3338,6 +3341,7 @@ static void rna_def_particle_settings(BlenderRNA *brna)
   prop = RNA_def_property(srna, "instance_object", PROP_POINTER, PROP_NONE);
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Instance Object", "Show this Object in place of particles");
   RNA_def_property_update(prop, 0, "rna_Particle_redo_dependency");
 
@@ -3456,6 +3460,7 @@ static void rna_def_particle_target(BlenderRNA *brna)
   prop = RNA_def_property(srna, "object", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "ob");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop,
       "Target Object",
@@ -3570,6 +3575,7 @@ static void rna_def_particle_system(BlenderRNA *brna)
   prop = RNA_def_property(srna, "reactor_target_object", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "target_ob");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop,
                            "Reactor Target Object",
                            "For reactor systems, the object that has the target particle system "
@@ -3822,6 +3828,7 @@ static void rna_def_particle_system(BlenderRNA *brna)
   prop = RNA_def_property(srna, "parent", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "parent");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Parent", "Use this object's coordinate system instead of global coordinate system");
   RNA_def_property_update(prop, 0, "rna_Particle_redo");
diff --git a/source/blender/makesrna/intern/rna_pose.c b/source/blender/makesrna/intern/rna_pose.c
index d359726b3bd..33d7d7d99cf 100644
--- a/source/blender/makesrna/intern/rna_pose.c
+++ b/source/blender/makesrna/intern/rna_pose.c
@@ -99,25 +99,25 @@ static void rna_Pose_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRN
 {
   /* XXX when to use this? ob->pose->flag |= (POSE_LOCKED|POSE_DO_UNLOCK); */
 
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
-  WM_main_add_notifier(NC_OBJECT | ND_POSE, ptr->id.data);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
+  WM_main_add_notifier(NC_OBJECT | ND_POSE, ptr->owner_id);
 }
 
 static void rna_Pose_dependency_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
   DEG_relations_tag_update(bmain);
 
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
-  WM_main_add_notifier(NC_OBJECT | ND_POSE, ptr->id.data);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
+  WM_main_add_notifier(NC_OBJECT | ND_POSE, ptr->owner_id);
 }
 
 static void rna_Pose_IK_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
   /* XXX when to use this? ob->pose->flag |= (POSE_LOCKED|POSE_DO_UNLOCK); */
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
-  WM_main_add_notifier(NC_OBJECT | ND_POSE, ptr->id.data);
+  WM_main_add_notifier(NC_OBJECT | ND_POSE, ptr->owner_id);
 
   BIK_clear_data(ob->pose);
 }
@@ -183,7 +183,7 @@ bool rna_ActionGroup_is_custom_colorset_get(PointerRNA *ptr)
 
 static void rna_BoneGroup_name_set(PointerRNA *ptr, const char *value)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bActionGroup *agrp = ptr->data;
 
   /* copy the new name into the name slot */
@@ -227,7 +227,7 @@ static void rna_Pose_ik_solver_set(struct PointerRNA *ptr, int value)
 
 static void rna_Pose_ik_solver_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bPose *pose = ptr->data;
 
   BKE_pose_tag_recalc(bmain, pose); /* checks & sorts pose channels */
@@ -276,7 +276,7 @@ static void rna_PoseChannel_rotation_mode_set(PointerRNA *ptr, int value)
 
 static void rna_PoseChannel_name_set(PointerRNA *ptr, const char *value)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bPoseChannel *pchan = (bPoseChannel *)ptr->data;
   char oldname[sizeof(pchan->name)], newname[sizeof(pchan->name)];
 
@@ -291,19 +291,19 @@ static void rna_PoseChannel_name_set(PointerRNA *ptr, const char *value)
 
 static PointerRNA rna_PoseChannel_bone_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bPoseChannel *pchan = (bPoseChannel *)ptr->data;
   PointerRNA tmp_ptr = *ptr;
 
   /* Replace the id_data pointer with the Armature ID. */
-  tmp_ptr.id.data = ob->data;
+  tmp_ptr.owner_id = ob->data;
 
   return rna_pointer_inherit_refine(&tmp_ptr, &RNA_Bone, pchan->bone);
 }
 
 static bool rna_PoseChannel_has_ik_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bPoseChannel *pchan = (bPoseChannel *)ptr->data;
 
   return BKE_pose_channel_in_IK_chain(ob, pchan);
@@ -341,7 +341,7 @@ static StructRNA *rna_Pose_ikparam_typef(PointerRNA *ptr)
 
 static void rna_Itasc_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bItasc *itasc = ptr->data;
 
   /* verify values */
@@ -373,7 +373,7 @@ static void rna_Itasc_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerR
 
 static void rna_Itasc_update_rebuild(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bPose *pose = ob->pose;
 
   BKE_pose_tag_recalc(bmain, pose); /* checks & sorts pose channels */
@@ -398,7 +398,7 @@ static void rna_PoseChannel_bone_custom_set(PointerRNA *ptr,
 
 static PointerRNA rna_PoseChannel_bone_group_get(PointerRNA *ptr)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bPose *pose = (ob) ? ob->pose : NULL;
   bPoseChannel *pchan = (bPoseChannel *)ptr->data;
   bActionGroup *grp;
@@ -417,7 +417,7 @@ static void rna_PoseChannel_bone_group_set(PointerRNA *ptr,
                                            PointerRNA value,
                                            struct ReportList *UNUSED(reports))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bPose *pose = (ob) ? ob->pose : NULL;
   bPoseChannel *pchan = (bPoseChannel *)ptr->data;
 
@@ -444,7 +444,7 @@ static void rna_PoseChannel_bone_group_index_set(PointerRNA *ptr, int value)
 static void rna_PoseChannel_bone_group_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bPose *pose = (ob) ? ob->pose : NULL;
 
   *min = 0;
@@ -663,7 +663,7 @@ bool rna_PoseChannel_constraints_override_apply(Main *UNUSED(bmain),
 
 static int rna_PoseChannel_proxy_editable(PointerRNA *ptr, const char **r_info)
 {
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   bArmature *arm = ob->data;
   bPoseChannel *pchan = (bPoseChannel *)ptr->data;
 
@@ -762,7 +762,7 @@ static int rna_PoseBones_lookup_string(PointerRNA *ptr, const char *key, Pointer
   bPose *pose = (bPose *)ptr->data;
   bPoseChannel *pchan = BKE_pose_channel_find_name(pose, key);
   if (pchan) {
-    RNA_pointer_create(ptr->id.data, &RNA_PoseBone, pchan, r_ptr);
+    RNA_pointer_create(ptr->owner_id, &RNA_PoseBone, pchan, r_ptr);
     return true;
   }
   else {
@@ -785,7 +785,7 @@ static void rna_PoseChannel_matrix_basis_set(PointerRNA *ptr, const float *value
 static void rna_PoseChannel_matrix_set(PointerRNA *ptr, const float *values)
 {
   bPoseChannel *pchan = (bPoseChannel *)ptr->data;
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
   float tmat[4][4];
 
   BKE_armature_mat_pose_to_bone_ex(NULL, ob, pchan, (float(*)[4])values, tmat);
@@ -812,9 +812,9 @@ static void rna_PoseChannel_custom_shape_transform_set(PointerRNA *ptr,
                                                        struct ReportList *UNUSED(reports))
 {
   bPoseChannel *pchan = (bPoseChannel *)ptr->data;
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
-  pchan->custom_tx = rna_PoseChannel_ensure_own_pchan(ob, value.id.data, value.data);
+  pchan->custom_tx = rna_PoseChannel_ensure_own_pchan(ob, (Object *)value.owner_id, value.data);
 }
 
 #else
@@ -1044,7 +1044,7 @@ static void rna_def_pose_channel(BlenderRNA *brna)
   RNA_def_property_update(prop, NC_OBJECT | ND_POSE, "rna_Pose_update");
 
   /* Curved bones settings - Applied on top of restpose values */
-  rna_def_bone_curved_common(srna, true);
+  rna_def_bone_curved_common(srna, true, false);
 
   /* Custom BBone next/prev sources */
   prop = RNA_def_property(srna, "bbone_custom_handle_start", PROP_POINTER, PROP_NONE);
@@ -1256,6 +1256,7 @@ static void rna_def_pose_channel(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "custom");
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_NO_COMPARISON);
   RNA_def_property_pointer_funcs(prop, NULL, "rna_PoseChannel_bone_custom_set", NULL, NULL);
   RNA_def_property_ui_text(
       prop, "Custom Object", "Object that defines custom draw type for this bone");
diff --git a/source/blender/makesrna/intern/rna_rigidbody.c b/source/blender/makesrna/intern/rna_rigidbody.c
index cb6776c2c06..03937a5434e 100644
--- a/source/blender/makesrna/intern/rna_rigidbody.c
+++ b/source/blender/makesrna/intern/rna_rigidbody.c
@@ -215,7 +215,7 @@ static void rna_RigidBodyOb_reset(Main *UNUSED(bmain), Scene *scene, PointerRNA
 
 static void rna_RigidBodyOb_shape_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Object *ob = ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   rna_RigidBodyOb_reset(bmain, scene, ptr);
 
@@ -859,6 +859,7 @@ static void rna_def_rigidbody_world(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "Collection");
   RNA_def_property_pointer_sdna(prop, NULL, "group");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Collection", "Collection containing objects participating in this simulation");
   RNA_def_property_update(prop, NC_SCENE, "rna_RigidBodyWorld_objects_collection_update");
@@ -866,6 +867,7 @@ static void rna_def_rigidbody_world(BlenderRNA *brna)
   prop = RNA_def_property(srna, "constraints", PROP_POINTER, PROP_NONE);
   RNA_def_property_struct_type(prop, "Collection");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Constraints", "Collection containing rigid body constraint objects");
   RNA_def_property_update(prop, NC_SCENE, "rna_RigidBodyWorld_constraints_collection_update");
@@ -1224,15 +1226,15 @@ static void rna_def_rigidbody_constraint(BlenderRNA *brna)
   prop = RNA_def_property(srna, "object1", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "ob1");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Object 1", "First Rigid Body Object to be constrained");
-  RNA_def_property_flag(prop, PROP_EDITABLE);
   RNA_def_property_update(prop, NC_OBJECT | ND_POINTCACHE, "rna_RigidBodyOb_reset");
 
   prop = RNA_def_property(srna, "object2", PROP_POINTER, PROP_NONE);
   RNA_def_property_pointer_sdna(prop, NULL, "ob2");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Object 2", "Second Rigid Body Object to be constrained");
-  RNA_def_property_flag(prop, PROP_EDITABLE);
   RNA_def_property_update(prop, NC_OBJECT | ND_POINTCACHE, "rna_RigidBodyOb_reset");
 
   /* Breaking Threshold */
diff --git a/source/blender/makesrna/intern/rna_rna.c b/source/blender/makesrna/intern/rna_rna.c
index 631ad72e146..6e21d02c0b3 100644
--- a/source/blender/makesrna/intern/rna_rna.c
+++ b/source/blender/makesrna/intern/rna_rna.c
@@ -388,10 +388,10 @@ void rna_builtin_properties_begin(CollectionPropertyIterator *iter, PointerRNA *
   newptr.data = ptr->type;
 
   if (ptr->type->flag & STRUCT_ID) {
-    newptr.id.data = ptr->data;
+    newptr.owner_id = ptr->data;
   }
   else {
-    newptr.id.data = NULL;
+    newptr.owner_id = NULL;
   }
 
   iter->parent = newptr;
@@ -414,7 +414,7 @@ int rna_builtin_properties_lookup_string(PointerRNA *ptr, const char *key, Point
 {
   StructRNA *srna;
   PropertyRNA *prop;
-  PointerRNA propptr = {{NULL}};
+  PointerRNA propptr = {NULL};
 
   srna = ptr->type;
 
@@ -770,7 +770,8 @@ static void rna_IntProperty_default_array_get(PointerRNA *ptr, int *values)
   PropertyRNA *prop = (PropertyRNA *)ptr->data;
   rna_idproperty_check(&prop, ptr);
   if (prop->totarraylength > 0) {
-    RNA_property_int_get_default_array(ptr, prop, values);
+    PointerRNA null_ptr = PointerRNA_NULL;
+    RNA_property_int_get_default_array(&null_ptr, prop, values);
   }
 }
 
@@ -779,7 +780,8 @@ static void rna_BoolProperty_default_array_get(PointerRNA *ptr, bool *values)
   PropertyRNA *prop = (PropertyRNA *)ptr->data;
   rna_idproperty_check(&prop, ptr);
   if (prop->totarraylength > 0) {
-    RNA_property_boolean_get_default_array(ptr, prop, values);
+    PointerRNA null_ptr = PointerRNA_NULL;
+    RNA_property_boolean_get_default_array(&null_ptr, prop, values);
   }
 }
 
@@ -788,7 +790,8 @@ static void rna_FloatProperty_default_array_get(PointerRNA *ptr, float *values)
   PropertyRNA *prop = (PropertyRNA *)ptr->data;
   rna_idproperty_check(&prop, ptr);
   if (prop->totarraylength > 0) {
-    RNA_property_float_get_default_array(ptr, prop, values);
+    PointerRNA null_ptr = PointerRNA_NULL;
+    RNA_property_float_get_default_array(&null_ptr, prop, values);
   }
 }
 
@@ -1226,7 +1229,7 @@ static bool rna_property_override_diff_propptr_validate_diffing(PointerRNA *prop
   }
 
   if (*r_is_id) {
-    BLI_assert(propptr_a->data == propptr_a->id.data && propptr_b->data == propptr_b->id.data);
+    BLI_assert(propptr_a->data == propptr_a->owner_id && propptr_b->data == propptr_b->owner_id);
   }
 
   return is_valid_for_diffing;
@@ -1241,6 +1244,10 @@ static int rna_property_override_diff_propptr(Main *bmain,
                                               const bool no_prop_name,
                                               IDOverrideLibrary *override,
                                               const char *rna_path,
+                                              const char *rna_itemname_a,
+                                              const char *rna_itemname_b,
+                                              const int rna_itemindex_a,
+                                              const int rna_itemindex_b,
                                               const int flags,
                                               bool *r_override_changed)
 {
@@ -1282,9 +1289,18 @@ static int rna_property_override_diff_propptr(Main *bmain,
         IDOverrideLibraryProperty *op = BKE_override_library_property_get(
             override, rna_path, &created);
 
-        if (op != NULL && created) { /* If not yet overridden... */
-          BKE_override_library_property_operation_get(
-              op, IDOVERRIDE_LIBRARY_OP_REPLACE, NULL, NULL, -1, -1, true, NULL, NULL);
+        /* If not yet overridden, or if we are handling sub-items (inside a collection)... */
+        if (op != NULL && (created || rna_itemname_a != NULL || rna_itemname_b != NULL ||
+                           rna_itemindex_a != -1 || rna_itemindex_b != -1)) {
+          BKE_override_library_property_operation_get(op,
+                                                      IDOVERRIDE_LIBRARY_OP_REPLACE,
+                                                      rna_itemname_b,
+                                                      rna_itemname_a,
+                                                      rna_itemindex_b,
+                                                      rna_itemindex_a,
+                                                      true,
+                                                      NULL,
+                                                      &created);
           if (r_override_changed) {
             *r_override_changed = created;
           }
@@ -1294,12 +1310,56 @@ static int rna_property_override_diff_propptr(Main *bmain,
       return comp;
     }
     else {
+      /* In case we got some array/collection like items identifiers, now is the time to generate a
+       * proper rna path from those. */
+#  define RNA_PATH_BUFFSIZE 8192
+
+      char extended_rna_path_buffer[RNA_PATH_BUFFSIZE];
+      char *extended_rna_path = extended_rna_path_buffer;
+
+#  define RNA_PATH_PRINTF(_str, ...) \
+    if (BLI_snprintf(extended_rna_path_buffer, RNA_PATH_BUFFSIZE, (_str), __VA_ARGS__) >= \
+        RNA_PATH_BUFFSIZE - 1) { \
+      extended_rna_path = BLI_sprintfN((_str), __VA_ARGS__); \
+    } \
+    (void)0
+#  define RNA_PATH_FREE() \
+    if (extended_rna_path != extended_rna_path_buffer && extended_rna_path != rna_path) \
+    MEM_freeN(extended_rna_path)
+
+      /* There may be a propname defined in some cases, while no actual name set
+       * (e.g. happens with point cache), in that case too we want to fall back to index.
+       * Note that we do not need the RNA path for insertion operations. */
+      if (rna_path) {
+        if ((rna_itemname_a != NULL && rna_itemname_a[0] != '\0') &&
+            (rna_itemname_b != NULL && rna_itemname_b[0] != '\0')) {
+          BLI_assert(STREQ(rna_itemname_a, rna_itemname_b));
+          char esc_item_name[RNA_PATH_BUFFSIZE];
+          BLI_strescape(esc_item_name, rna_itemname_a, RNA_PATH_BUFFSIZE);
+          RNA_PATH_PRINTF("%s[\"%s\"]", rna_path, esc_item_name);
+        }
+        else if (rna_itemindex_a != -1) { /* Based on index... */
+          BLI_assert(rna_itemindex_a == rna_itemindex_b);
+          RNA_PATH_PRINTF("%s[%d]", rna_path, rna_itemindex_a);
+        }
+        else {
+          extended_rna_path = (char *)rna_path;
+        }
+      }
+
       eRNAOverrideMatchResult report_flags = 0;
       const bool match = RNA_struct_override_matches(
-          bmain, propptr_a, propptr_b, rna_path, override, flags, &report_flags);
+          bmain, propptr_a, propptr_b, extended_rna_path, override, flags, &report_flags);
       if (r_override_changed && (report_flags & RNA_OVERRIDE_MATCH_RESULT_CREATED) != 0) {
         *r_override_changed = true;
       }
+
+      RNA_PATH_FREE();
+
+#  undef RNA_PATH_BUFFSIZE
+#  undef RNA_PATH_PRINTF
+#  undef RNA_PATH_FREE
+
       return !match;
     }
   }
@@ -1616,6 +1676,10 @@ int rna_property_override_diff_default(Main *bmain,
                                                   no_prop_name,
                                                   override,
                                                   rna_path,
+                                                  NULL,
+                                                  NULL,
+                                                  -1,
+                                                  -1,
                                                   flags,
                                                   r_override_changed);
       }
@@ -1636,21 +1700,6 @@ int rna_property_override_diff_default(Main *bmain,
       int idx_a = 0;
       int idx_b = 0;
 
-#  define RNA_PATH_BUFFSIZE 8192
-
-      char extended_rna_path_buffer[RNA_PATH_BUFFSIZE];
-      char *extended_rna_path = extended_rna_path_buffer;
-
-#  define RNA_PATH_PRINTF(_str, ...) \
-    if (BLI_snprintf(extended_rna_path_buffer, RNA_PATH_BUFFSIZE, (_str), __VA_ARGS__) >= \
-        RNA_PATH_BUFFSIZE - 1) { \
-      extended_rna_path = BLI_sprintfN((_str), __VA_ARGS__); \
-    } \
-    (void)0
-#  define RNA_PATH_FREE() \
-    if (extended_rna_path != extended_rna_path_buffer) \
-    MEM_freeN(extended_rna_path)
-
       CollectionPropertyIterator iter_a, iter_b;
       RNA_property_collection_begin(ptr_a, prop_a, &iter_a);
       RNA_property_collection_begin(ptr_b, prop_b, &iter_b);
@@ -1730,41 +1779,15 @@ int rna_property_override_diff_default(Main *bmain,
 #  endif
 
           if (!(is_id || is_valid_for_diffing || is_valid_for_insertion)) {
-            /* Differences we cannot handle, we can break here
-             * (we do not support replacing ID pointers in collections e.g.). */
+            /* Differences we cannot handle, we can break here. */
             equals = false;
             abort = true;
             break;
           }
 
-          /* There may be a propname defined in some cases, while no actual name set
-           * (e.g. happens with point cache), in that case too we want to fall back to index.
-           * Note that we do not need the RNA path for insertion operations. */
-          if (is_id || is_valid_for_diffing) {
-            if ((propname_a != NULL && propname_a[0] != '\0') &&
-                (propname_b != NULL && propname_b[0] != '\0')) {
-              if (rna_path) {
-                /* In case of name, either it is valid for diffing, and _a and _b are identical,
-                 * or it is valid for insertion, and we need to use _a. */
-                char esc_item_name[RNA_PATH_BUFFSIZE];
-                BLI_strescape(esc_item_name, propname_a, RNA_PATH_BUFFSIZE);
-                RNA_PATH_PRINTF("%s[\"%s\"]", rna_path, esc_item_name);
-              }
-            }
-            else { /* Based on index... */
-              if (rna_path) {
-                /* In case of indices, we need _a one for insertion,
-                 * but _b ones for in-depth diffing.
-                 * Insertion always happen once all 'replace' operations have been done,
-                 * otherwise local and reference paths for those would have to be different! */
-                RNA_PATH_PRINTF("%s[%d]", rna_path, is_valid_for_insertion ? idx_a : idx_b);
-              }
-            }
-          }
-
-          /* Collections do not support replacement of their data
-           * (since they do not support removing), only in *some* cases, insertion.
-           * We also assume then that _a data is the one where things are inserted. */
+          /* Collections do not support replacement of their data (except for collections of ID
+           * pointers), since they do not support removing, only in *some* cases, insertion. We
+           * also assume then that _a data is the one where things are inserted. */
           if (is_valid_for_insertion && use_insertion) {
             bool created;
             IDOverrideLibraryProperty *op = BKE_override_library_property_get(
@@ -1812,7 +1835,11 @@ int rna_property_override_diff_default(Main *bmain,
                                                                 no_ownership,
                                                                 no_prop_name,
                                                                 override,
-                                                                extended_rna_path,
+                                                                rna_path,
+                                                                propname_a,
+                                                                propname_b,
+                                                                idx_a,
+                                                                idx_b,
                                                                 flags,
                                                                 r_override_changed);
               equals = equals && eq;
@@ -1839,7 +1866,6 @@ int rna_property_override_diff_default(Main *bmain,
             propname_b = buff_b;
           }
           propname_b[0] = '\0';
-          RNA_PATH_FREE();
 
           if (!do_create && !equals) {
             abort = true; /* Early out in case we do not want to loop over whole collection. */
@@ -1864,10 +1890,6 @@ int rna_property_override_diff_default(Main *bmain,
           RNA_property_collection_next(&iter_b);
           idx_b++;
         }
-
-#  undef RNA_PATH_BUFFSIZE
-#  undef RNA_PATH_PRINTF
-#  undef RNA_PATH_FREE
       }
 
       /* Not same number of items in both collections. */
diff --git a/source/blender/makesrna/intern/rna_scene.c b/source/blender/makesrna/intern/rna_scene.c
index 4397d204972..3d2d91d2334 100644
--- a/source/blender/makesrna/intern/rna_scene.c
+++ b/source/blender/makesrna/intern/rna_scene.c
@@ -168,6 +168,16 @@ const EnumPropertyItem rna_enum_snap_element_items[] = {
     {SCE_SNAP_MODE_EDGE, "EDGE", ICON_SNAP_EDGE, "Edge", "Snap to edges"},
     {SCE_SNAP_MODE_FACE, "FACE", ICON_SNAP_FACE, "Face", "Snap to faces"},
     {SCE_SNAP_MODE_VOLUME, "VOLUME", ICON_SNAP_VOLUME, "Volume", "Snap to volume"},
+    {SCE_SNAP_MODE_EDGE_MIDPOINT,
+     "EDGE_MIDPOINT",
+     ICON_SNAP_MIDPOINT,
+     "Edge Center",
+     "Snap to the middle of edges"},
+    {SCE_SNAP_MODE_EDGE_PERPENDICULAR,
+     "EDGE_PERPENDICULAR",
+     ICON_SNAP_PERPENDICULAR,
+     "Edge Perpendicular",
+     "Snap to the nearest point on an edge"},
     {0, NULL, 0, NULL, NULL},
 };
 
@@ -740,6 +750,65 @@ static void rna_GPencilInterpolateSettings_type_set(PointerRNA *ptr, int value)
   }
 }
 
+static void rna_Gpencil_selectmode_update(bContext *C, PointerRNA *ptr)
+{
+  ToolSettings *ts = (ToolSettings *)ptr->data;
+  /* If the mode is not Stroke, don't extend selection. */
+  if ((ts->gpencil_selectmode_edit & GP_SELECTMODE_STROKE) == 0) {
+    return;
+  }
+
+  /* Extend selection to all points in all selected strokes. */
+  ViewLayer *view_layer = CTX_data_view_layer(C);
+  Object *ob = OBACT(view_layer);
+  if ((ob) && (ob->type == OB_GPENCIL)) {
+    bGPdata *gpd = (bGPdata *)ob->data;
+    CTX_DATA_BEGIN (C, bGPDstroke *, gps, editable_gpencil_strokes) {
+      if ((gps->flag & GP_STROKE_SELECT) && (gps->totpoints > 1)) {
+        bGPDspoint *pt;
+        for (int i = 0; i < gps->totpoints; i++) {
+          pt = &gps->points[i];
+          pt->flag |= GP_SPOINT_SELECT;
+        }
+      }
+    }
+    CTX_DATA_END;
+
+    gpd->flag |= GP_DATA_CACHE_IS_DIRTY;
+    DEG_id_tag_update(&gpd->id, ID_RECALC_TRANSFORM | ID_RECALC_GEOMETRY);
+  }
+}
+
+static void rna_Gpencil_mask_point_update(Main *UNUSED(bmain),
+                                          Scene *UNUSED(scene),
+                                          PointerRNA *ptr)
+{
+  ToolSettings *ts = (ToolSettings *)ptr->data;
+
+  ts->gpencil_selectmode_sculpt &= ~GP_SCULPT_MASK_SELECTMODE_STROKE;
+  ts->gpencil_selectmode_sculpt &= ~GP_SCULPT_MASK_SELECTMODE_SEGMENT;
+}
+
+static void rna_Gpencil_mask_stroke_update(Main *UNUSED(bmain),
+                                           Scene *UNUSED(scene),
+                                           PointerRNA *ptr)
+{
+  ToolSettings *ts = (ToolSettings *)ptr->data;
+
+  ts->gpencil_selectmode_sculpt &= ~GP_SCULPT_MASK_SELECTMODE_POINT;
+  ts->gpencil_selectmode_sculpt &= ~GP_SCULPT_MASK_SELECTMODE_SEGMENT;
+}
+
+static void rna_Gpencil_mask_segment_update(Main *UNUSED(bmain),
+                                            Scene *UNUSED(scene),
+                                            PointerRNA *ptr)
+{
+  ToolSettings *ts = (ToolSettings *)ptr->data;
+
+  ts->gpencil_selectmode_sculpt &= ~GP_SCULPT_MASK_SELECTMODE_POINT;
+  ts->gpencil_selectmode_sculpt &= ~GP_SCULPT_MASK_SELECTMODE_STROKE;
+}
+
 /* Read-only Iterator of all the scene objects. */
 
 static void rna_Scene_objects_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
@@ -796,7 +865,7 @@ static void rna_Scene_set_set(PointerRNA *ptr,
 
 void rna_Scene_set_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   DEG_relations_tag_update(bmain);
   DEG_id_tag_update_ex(bmain, &scene->id, 0);
@@ -822,8 +891,8 @@ static void rna_Scene_fps_update(Main *bmain, Scene *scene, PointerRNA *UNUSED(p
 {
   DEG_id_tag_update(&scene->id, ID_RECALC_AUDIO_FPS | ID_RECALC_SEQUENCER_STRIPS);
   /* NOTE: Tag via dependency graph will take care of all the updates ion the evaluated domain,
-   * however, changes in FPS actually modifies an original stip length, so this we take care about
-   * here. */
+   * however, changes in FPS actually modifies an original skip length,
+   * so this we take care about here. */
   BKE_sequencer_refresh_sound_length(bmain, scene);
 }
 
@@ -834,7 +903,7 @@ static void rna_Scene_listener_update(Main *UNUSED(bmain), Scene *scene, Pointer
 
 static void rna_Scene_volume_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   DEG_id_tag_update(&scene->id, ID_RECALC_AUDIO_VOLUME | ID_RECALC_SEQUENCER_STRIPS);
 }
 
@@ -958,7 +1027,7 @@ static void rna_Scene_show_subframe_update(Main *UNUSED(bmain),
                                            Scene *UNUSED(current_scene),
                                            PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   scene->r.subframe = 0.0f;
 }
 
@@ -966,7 +1035,7 @@ static void rna_Scene_frame_update(Main *UNUSED(bmain),
                                    Scene *UNUSED(current_scene),
                                    PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   DEG_id_tag_update(&scene->id, ID_RECALC_AUDIO_SEEK);
   WM_main_add_notifier(NC_SCENE | ND_FRAME, scene);
 }
@@ -1079,7 +1148,7 @@ static char *rna_BakeSettings_path(PointerRNA *UNUSED(ptr))
 static char *rna_ImageFormatSettings_path(PointerRNA *ptr)
 {
   ImageFormatData *imf = (ImageFormatData *)ptr->data;
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   switch (GS(id->name)) {
     case ID_SCE: {
@@ -1150,7 +1219,7 @@ static bool rna_RenderSettings_is_movie_format_get(PointerRNA *ptr)
 static void rna_ImageFormatSettings_file_format_set(PointerRNA *ptr, int value)
 {
   ImageFormatData *imf = (ImageFormatData *)ptr->data;
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   imf->imtype = value;
 
   const bool is_render = (id && GS(id->name) == ID_SCE);
@@ -1191,7 +1260,7 @@ static void rna_ImageFormatSettings_file_format_set(PointerRNA *ptr, int value)
   }
 
   if (id && GS(id->name) == ID_SCE) {
-    Scene *scene = ptr->id.data;
+    Scene *scene = (Scene *)ptr->owner_id;
     RenderData *rd = &scene->r;
 #  ifdef WITH_FFMPEG
     BKE_ffmpeg_image_type_verify(rd, imf);
@@ -1205,7 +1274,7 @@ static const EnumPropertyItem *rna_ImageFormatSettings_file_format_itemf(bContex
                                                                          PropertyRNA *UNUSED(prop),
                                                                          bool *UNUSED(r_free))
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   if (id && GS(id->name) == ID_SCE) {
     return rna_enum_image_type_items;
   }
@@ -1220,7 +1289,7 @@ static const EnumPropertyItem *rna_ImageFormatSettings_color_mode_itemf(bContext
                                                                         bool *r_free)
 {
   ImageFormatData *imf = (ImageFormatData *)ptr->data;
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   const bool is_render = (id && GS(id->name) == ID_SCE);
 
   /* note, we need to act differently for render
@@ -1236,7 +1305,7 @@ static const EnumPropertyItem *rna_ImageFormatSettings_color_mode_itemf(bContext
    * the same MPEG format with QTRLE codec can easily handle alpha channel.
    * not sure how to deal with such cases in a nicer way (sergey) */
   if (is_render) {
-    Scene *scene = ptr->id.data;
+    Scene *scene = (Scene *)ptr->owner_id;
     RenderData *rd = &scene->r;
 
     if (BKE_ffmpeg_alpha_channel_is_supported(rd)) {
@@ -1405,7 +1474,7 @@ static void rna_SceneRender_file_ext_get(PointerRNA *ptr, char *str)
 #  ifdef WITH_FFMPEG
 static void rna_FFmpegSettings_lossless_output_set(PointerRNA *ptr, bool value)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   RenderData *rd = &scene->r;
 
   if (value) {
@@ -1422,7 +1491,7 @@ static void rna_FFmpegSettings_codec_settings_update(Main *UNUSED(bmain),
                                                      Scene *UNUSED(scene_unused),
                                                      PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   RenderData *rd = &scene->r;
 
   BKE_ffmpeg_codec_settings_verify(rd);
@@ -1521,7 +1590,7 @@ static void rna_RenderSettings_engine_set(PointerRNA *ptr, int value)
 
   if (type) {
     BLI_strncpy_utf8(rd->engine, type->idname, sizeof(rd->engine));
-    DEG_id_tag_update(ptr->id.data, ID_RECALC_COPY_ON_WRITE);
+    DEG_id_tag_update(ptr->owner_id, ID_RECALC_COPY_ON_WRITE);
   }
 }
 
@@ -1577,20 +1646,20 @@ static bool rna_RenderSettings_multiple_engines_get(PointerRNA *UNUSED(ptr))
 
 static bool rna_RenderSettings_use_spherical_stereo_get(PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   return BKE_scene_use_spherical_stereo(scene);
 }
 
 void rna_Scene_glsl_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   DEG_id_tag_update(&scene->id, 0);
 }
 
 static void rna_Scene_world_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Scene *sc = (Scene *)ptr->id.data;
+  Scene *sc = (Scene *)ptr->owner_id;
 
   rna_Scene_glsl_update(bmain, scene, ptr);
   WM_main_add_notifier(NC_WORLD | ND_WORLD, &sc->id);
@@ -1599,7 +1668,7 @@ static void rna_Scene_world_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 
 void rna_Scene_freestyle_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   DEG_id_tag_update(&scene->id, 0);
 }
@@ -1615,7 +1684,7 @@ void rna_Scene_use_view_map_cache_update(Main *UNUSED(bmain),
 
 void rna_ViewLayer_name_set(PointerRNA *ptr, const char *value)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   ViewLayer *view_layer = (ViewLayer *)ptr->data;
   BLI_assert(BKE_id_is_in_global_main(&scene->id));
   BKE_view_layer_rename(G_MAIN, scene, view_layer, value);
@@ -1623,7 +1692,7 @@ void rna_ViewLayer_name_set(PointerRNA *ptr, const char *value)
 
 static void rna_SceneRenderView_name_set(PointerRNA *ptr, const char *value)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   SceneRenderView *rv = (SceneRenderView *)ptr->data;
   BLI_strncpy_utf8(rv->name, value, sizeof(rv->name));
   BLI_uniquename(&scene->r.views,
@@ -1636,14 +1705,14 @@ static void rna_SceneRenderView_name_set(PointerRNA *ptr, const char *value)
 
 void rna_ViewLayer_material_override_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   rna_Scene_glsl_update(bmain, scene, ptr);
   DEG_relations_tag_update(bmain);
 }
 
 void rna_ViewLayer_pass_update(Main *bmain, Scene *activescene, PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   if (scene->nodetree) {
     ntreeCompositUpdateRLayers(scene->nodetree);
@@ -1676,7 +1745,7 @@ static void rna_Physics_relations_update(Main *bmain,
 
 static void rna_Physics_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   FOREACH_SCENE_OBJECT_BEGIN (scene, ob) {
     BKE_ptcache_object_reset(scene, ob, PTCACHE_RESET_DEPSGRAPH);
   }
@@ -1763,7 +1832,7 @@ static void object_simplify_update(Object *ob)
 
 static void rna_Scene_use_simplify_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *sce = ptr->id.data;
+  Scene *sce = (Scene *)ptr->owner_id;
   Scene *sce_iter;
   Base *base;
 
@@ -1784,7 +1853,7 @@ static void rna_Scene_use_simplify_update(Main *bmain, Scene *UNUSED(scene), Poi
 
 static void rna_Scene_simplify_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  Scene *sce = ptr->id.data;
+  Scene *sce = (Scene *)ptr->owner_id;
 
   if (sce->r.mode & R_SIMPLIFY) {
     rna_Scene_use_simplify_update(bmain, scene, ptr);
@@ -1795,7 +1864,7 @@ static void rna_Scene_use_persistent_data_update(Main *UNUSED(bmain),
                                                  Scene *UNUSED(scene),
                                                  PointerRNA *ptr)
 {
-  Scene *sce = ptr->id.data;
+  Scene *sce = (Scene *)ptr->owner_id;
 
   if (!(sce->r.mode & R_PERSISTENT_DATA)) {
     RE_FreePersistentData();
@@ -1806,7 +1875,7 @@ static void rna_Scene_use_persistent_data_update(Main *UNUSED(bmain),
 static void rna_Scene_transform_orientation_slots_begin(CollectionPropertyIterator *iter,
                                                         PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   TransformOrientationSlot *orient_slot = &scene->orientation_slots[0];
   rna_iterator_array_begin(
       iter, orient_slot, sizeof(*orient_slot), ARRAY_SIZE(scene->orientation_slots), 0, NULL);
@@ -1814,7 +1883,7 @@ static void rna_Scene_transform_orientation_slots_begin(CollectionPropertyIterat
 
 static int rna_Scene_transform_orientation_slots_length(PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   return ARRAY_SIZE(scene->orientation_slots);
 }
 
@@ -2062,7 +2131,7 @@ static void rna_Scene_update_active_object_data(bContext *C, PointerRNA *UNUSED(
 
 static void rna_SceneCamera_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Object *camera = scene->camera;
 
   BKE_sequence_invalidate_scene_strips(bmain, scene);
@@ -2203,7 +2272,7 @@ void rna_FreestyleSettings_module_remove(ID *id,
 
 static void rna_Stereo3dFormat_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   if (id && GS(id->name) == ID_IM) {
     Image *ima = (Image *)id;
@@ -2251,7 +2320,7 @@ static void rna_ViewLayer_remove(
 
 static int rna_TransformOrientationSlot_type_get(PointerRNA *ptr)
 {
-  Scene *scene = ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   TransformOrientationSlot *orient_slot = ptr->data;
   if (orient_slot != &scene->orientation_slots[SCE_ORIENT_DEFAULT]) {
     if ((orient_slot->flag & SELECT) == 0) {
@@ -2263,7 +2332,7 @@ static int rna_TransformOrientationSlot_type_get(PointerRNA *ptr)
 
 void rna_TransformOrientationSlot_type_set(PointerRNA *ptr, int value)
 {
-  Scene *scene = ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   TransformOrientationSlot *orient_slot = ptr->data;
 
   if (orient_slot != &scene->orientation_slots[SCE_ORIENT_DEFAULT]) {
@@ -2281,7 +2350,7 @@ void rna_TransformOrientationSlot_type_set(PointerRNA *ptr, int value)
 
 static PointerRNA rna_TransformOrientationSlot_get(PointerRNA *ptr)
 {
-  Scene *scene = ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   TransformOrientationSlot *orient_slot = ptr->data;
   TransformOrientation *orientation;
   if (orient_slot->type < V3D_ORIENT_CUSTOM) {
@@ -2339,8 +2408,8 @@ const EnumPropertyItem *rna_TransformOrientation_itemf(bContext *C,
                                                        bool *r_free)
 {
   Scene *scene;
-  if (ptr->id.data && (GS(((ID *)ptr->id.data)->name) == ID_SCE)) {
-    scene = ptr->id.data;
+  if (ptr->owner_id && (GS(ptr->owner_id->name) == ID_SCE)) {
+    scene = (Scene *)ptr->owner_id;
   }
   else {
     scene = CTX_data_scene(C);
@@ -2353,7 +2422,7 @@ const EnumPropertyItem *rna_TransformOrientation_with_scene_itemf(bContext *UNUS
                                                                   PropertyRNA *UNUSED(prop),
                                                                   bool *r_free)
 {
-  Scene *scene = ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   TransformOrientationSlot *orient_slot = ptr->data;
   bool include_default = (orient_slot != &scene->orientation_slots[SCE_ORIENT_DEFAULT]);
   return rna_TransformOrientation_impl_itemf(scene, include_default, r_free);
@@ -2954,14 +3023,29 @@ static void rna_def_tool_settings(BlenderRNA *brna)
   prop = RNA_def_property(srna, "use_transform_pivot_point_align", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "transform_flag", SCE_XFORM_AXIS_ALIGN);
   RNA_def_property_ui_text(
-      prop, "Only Origins", "Manipulate origins (object, pose and weight paint mode only)");
-  RNA_def_property_ui_icon(prop, ICON_CENTER_ONLY, 0);
-  RNA_def_property_update(prop, NC_SCENE, NULL);
+      prop, "Only Locations", "Manipulate origins (object, pose and weight paint mode only)");
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, NULL);
+
+  prop = RNA_def_property(srna, "use_transform_data_origin", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "transform_flag", SCE_XFORM_DATA_ORIGIN);
+  RNA_def_property_ui_text(prop, "Data Origins", "Manipulate object data");
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, NULL);
+
+  prop = RNA_def_property(srna, "use_transform_skip_children", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "transform_flag", SCE_XFORM_SKIP_CHILDREN);
+  RNA_def_property_ui_text(prop, "Skip Children", "Don't transform children");
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, NULL);
 
   prop = RNA_def_property(srna, "use_mesh_automerge", PROP_BOOLEAN, PROP_NONE);
-  RNA_def_property_boolean_sdna(prop, NULL, "automerge", 0);
+  RNA_def_property_boolean_sdna(prop, NULL, "automerge", AUTO_MERGE);
   RNA_def_property_ui_text(
-      prop, "Auto Merge", "Automatically merge vertices moved to the same location");
+      prop, "Auto Merge Vertices", "Automatically merge vertices moved to the same location");
+  RNA_def_property_ui_icon(prop, ICON_AUTOMERGE_OFF, 1);
+  RNA_def_property_update(prop, NC_SCENE | ND_TOOLSETTINGS, NULL); /* header redraw */
+
+  prop = RNA_def_property(srna, "use_mesh_automerge_and_split", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "automerge", AUTO_MERGE_AND_SPLIT);
+  RNA_def_property_ui_text(prop, "Split Edges & Faces", "Automatically split edges and faces");
   RNA_def_property_ui_icon(prop, ICON_AUTOMERGE_OFF, 1);
   RNA_def_property_update(prop, NC_SCENE | ND_TOOLSETTINGS, NULL); /* header redraw */
 
@@ -3117,12 +3201,40 @@ static void rna_def_tool_settings(BlenderRNA *brna)
       prop, "Only Endpoints", "Only use the first and last parts of the stroke for snapping");
   RNA_def_property_update(prop, NC_GPENCIL | ND_DATA, NULL);
 
-  /* Grease Pencil - Select mode */
-  prop = RNA_def_property(srna, "gpencil_selectmode", PROP_ENUM, PROP_NONE);
-  RNA_def_property_enum_sdna(prop, NULL, "gpencil_selectmode");
+  /* Grease Pencil - Select mode Edit */
+  prop = RNA_def_property(srna, "gpencil_selectmode_edit", PROP_ENUM, PROP_NONE);
+  RNA_def_property_enum_sdna(prop, NULL, "gpencil_selectmode_edit");
   RNA_def_property_enum_items(prop, gpencil_selectmode_items);
   RNA_def_property_ui_text(prop, "Select Mode", "");
   RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
+  RNA_def_property_flag(prop, PROP_CONTEXT_UPDATE);
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, "rna_Gpencil_selectmode_update");
+
+  /* Grease Pencil - Select mode Sculpt */
+  prop = RNA_def_property(srna, "use_gpencil_select_mask_point", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(
+      prop, NULL, "gpencil_selectmode_sculpt", GP_SCULPT_MASK_SELECTMODE_POINT);
+  RNA_def_property_ui_text(prop, "Selection Mask", "Only sculpt selected stroke points");
+  RNA_def_property_ui_icon(prop, ICON_GP_SELECT_POINTS, 0);
+  RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, "rna_Gpencil_mask_point_update");
+
+  prop = RNA_def_property(srna, "use_gpencil_select_mask_stroke", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(
+      prop, NULL, "gpencil_selectmode_sculpt", GP_SCULPT_MASK_SELECTMODE_STROKE);
+  RNA_def_property_ui_text(prop, "Selection Mask", "Only sculpt selected stroke");
+  RNA_def_property_ui_icon(prop, ICON_GP_SELECT_STROKES, 0);
+  RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, "rna_Gpencil_mask_stroke_update");
+
+  prop = RNA_def_property(srna, "use_gpencil_select_mask_segment", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(
+      prop, NULL, "gpencil_selectmode_sculpt", GP_SCULPT_MASK_SELECTMODE_SEGMENT);
+  RNA_def_property_ui_text(
+      prop, "Selection Mask", "Only sculpt selected stroke points between other strokes");
+  RNA_def_property_ui_icon(prop, ICON_GP_SELECT_BETWEEN_STROKES, 0);
+  RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, "rna_Gpencil_mask_segment_update");
 
   /* Annotations - 2D Views Stroke Placement */
   prop = RNA_def_property(srna, "annotation_stroke_placement_view2d", PROP_ENUM, PROP_NONE);
@@ -3708,6 +3820,7 @@ void rna_def_view_layer_common(StructRNA *srna, const bool scene)
     RNA_def_property_pointer_sdna(prop, NULL, "mat_override");
     RNA_def_property_struct_type(prop, "Material");
     RNA_def_property_flag(prop, PROP_EDITABLE);
+    RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
     RNA_def_property_ui_text(
         prop, "Material Override", "Material to override all other materials in this view layer");
     RNA_def_property_update(
@@ -5019,7 +5132,7 @@ static void rna_def_image_format_stereo3d_format(BlenderRNA *brna)
 /* use for render output and image save operator,
  * note: there are some cases where the members act differently when this is
  * used from a scene, video formats can only be selected for render output
- * for example, this is checked by seeing if the ptr->id.data is a Scene id */
+ * for example, this is checked by seeing if the ptr->owner_id is a Scene id */
 
 static void rna_def_scene_image_format_data(BlenderRNA *brna)
 {
@@ -7477,6 +7590,7 @@ void RNA_def_scene(BlenderRNA *brna)
   /* Global Settings */
   prop = RNA_def_property(srna, "camera", PROP_POINTER, PROP_NONE);
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_pointer_funcs(prop, NULL, NULL, NULL, "rna_Camera_object_poll");
   RNA_def_property_ui_text(prop, "Camera", "Active camera, used for rendering the scene");
   RNA_def_property_update(prop, NC_SCENE | NA_EDITED, "rna_Scene_camera_update");
@@ -7485,12 +7599,14 @@ void RNA_def_scene(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "set");
   RNA_def_property_struct_type(prop, "Scene");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_pointer_funcs(prop, NULL, "rna_Scene_set_set", NULL, NULL);
   RNA_def_property_ui_text(prop, "Background Scene", "Background set scene");
   RNA_def_property_update(prop, NC_SCENE | NA_EDITED, "rna_Scene_set_update");
 
   prop = RNA_def_property(srna, "world", PROP_POINTER, PROP_NONE);
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "World", "World used for rendering the scene");
   RNA_def_property_update(prop, NC_SCENE | ND_WORLD, "rna_Scene_world_update");
 
@@ -7851,6 +7967,7 @@ void RNA_def_scene(BlenderRNA *brna)
   RNA_def_property_pointer_funcs(
       prop, NULL, NULL, NULL, "rna_GPencil_datablocks_annotations_poll");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_REFCOUNT);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(
       prop, "Annotations", "Grease Pencil data-block used for annotations in the 3D view");
   RNA_def_property_update(prop, NC_GPENCIL | ND_DATA | NA_EDITED, NULL);
diff --git a/source/blender/makesrna/intern/rna_screen.c b/source/blender/makesrna/intern/rna_screen.c
index c868c79e968..728ef3fb706 100644
--- a/source/blender/makesrna/intern/rna_screen.c
+++ b/source/blender/makesrna/intern/rna_screen.c
@@ -149,7 +149,7 @@ static void rna_Area_type_set(PointerRNA *ptr, int value)
 
 static void rna_Area_type_update(bContext *C, PointerRNA *ptr)
 {
-  bScreen *sc = (bScreen *)ptr->id.data;
+  bScreen *sc = (bScreen *)ptr->owner_id;
   ScrArea *sa = (ScrArea *)ptr->data;
 
   /* Running update without having called 'set', see: T64049 */
diff --git a/source/blender/makesrna/intern/rna_sculpt_paint.c b/source/blender/makesrna/intern/rna_sculpt_paint.c
index 84f2cb7c1be..2e3f41d656b 100644
--- a/source/blender/makesrna/intern/rna_sculpt_paint.c
+++ b/source/blender/makesrna/intern/rna_sculpt_paint.c
@@ -319,7 +319,7 @@ static char *rna_ParticleEdit_path(PointerRNA *UNUSED(ptr))
 static bool rna_Brush_mode_poll(PointerRNA *ptr, PointerRNA value)
 {
   const Paint *paint = ptr->data;
-  Brush *brush = value.id.data;
+  Brush *brush = (Brush *)value.owner_id;
   const uint tool_offset = paint->runtime.tool_offset;
   const eObjectMode ob_mode = paint->runtime.ob_mode;
   UNUSED_VARS_NDEBUG(tool_offset);
@@ -350,10 +350,10 @@ static bool paint_contains_brush_slot(const Paint *paint, const PaintToolSlot *t
 
 static bool rna_Brush_mode_with_tool_poll(PointerRNA *ptr, PointerRNA value)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   const PaintToolSlot *tslot = ptr->data;
   ToolSettings *ts = scene->toolsettings;
-  Brush *brush = value.id.data;
+  Brush *brush = (Brush *)value.owner_id;
   int mode = 0;
   int slot_index = 0;
 
@@ -438,7 +438,7 @@ static char *rna_Sculpt_path(PointerRNA *UNUSED(ptr))
 
 static char *rna_VertexPaint_path(PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   ToolSettings *ts = scene->toolsettings;
   if (ptr->data == ts->vpaint) {
     return BLI_strdup("tool_settings.vertex_paint");
@@ -1420,13 +1420,6 @@ static void rna_def_gpencil_sculpt(BlenderRNA *brna)
   RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
   RNA_def_property_ui_text(prop, "Guide", "");
 
-  prop = RNA_def_property(srna, "use_select_mask", PROP_BOOLEAN, PROP_NONE);
-  RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_SCULPT_SETT_FLAG_SELECT_MASK);
-  RNA_def_property_ui_text(prop, "Selection Mask", "Only sculpt selected stroke points");
-  RNA_def_property_ui_icon(prop, ICON_GP_ONLY_SELECTED, 0);
-  RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
-  RNA_def_property_update(prop, NC_SCENE | ND_TOOLSETTINGS, NULL);
-
   prop = RNA_def_property(srna, "use_edit_position", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", GP_SCULPT_SETT_FLAG_APPLY_POSITION);
   RNA_def_property_ui_text(prop, "Affect Position", "The brush affects the position of the point");
diff --git a/source/blender/makesrna/intern/rna_sequencer.c b/source/blender/makesrna/intern/rna_sequencer.c
index 80a4defba00..efb2fd376b3 100644
--- a/source/blender/makesrna/intern/rna_sequencer.c
+++ b/source/blender/makesrna/intern/rna_sequencer.c
@@ -100,7 +100,7 @@ static void meta_tmp_ref(Sequence *seq_par, Sequence *seq)
 
 static void rna_SequenceElement_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
 
   if (ed) {
@@ -119,7 +119,7 @@ static void rna_Sequence_invalidate_raw_update(Main *UNUSED(bmain),
                                                Scene *UNUSED(scene),
                                                PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
 
   if (ed) {
@@ -133,7 +133,7 @@ static void rna_Sequence_invalidate_preprocessed_update(Main *UNUSED(bmain),
                                                         Scene *UNUSED(scene),
                                                         PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
 
   if (ed) {
@@ -147,7 +147,7 @@ static void rna_Sequence_invalidate_composite_update(Main *UNUSED(bmain),
                                                      Scene *UNUSED(scene),
                                                      PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
 
   if (ed) {
@@ -179,7 +179,7 @@ static void rna_Sequence_use_sequence(Main *bmain, Scene *scene, PointerRNA *ptr
 static void rna_SequenceEditor_sequences_all_begin(CollectionPropertyIterator *iter,
                                                    PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
 
   meta_tmp_ref(NULL, ed->seqbase.first);
@@ -281,7 +281,7 @@ static void do_sequence_frame_change_update(Scene *scene, Sequence *seq)
  */
 static void rna_Sequence_frame_change_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   do_sequence_frame_change_update(scene, (Sequence *)ptr->data);
   rna_Sequence_invalidate_preprocessed_update(bmain, scene, ptr);
 }
@@ -289,7 +289,7 @@ static void rna_Sequence_frame_change_update(Main *bmain, Scene *UNUSED(scene),
 static void rna_Sequence_start_frame_set(PointerRNA *ptr, int value)
 {
   Sequence *seq = (Sequence *)ptr->data;
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   BKE_sequence_translate(scene, seq, value - seq->start);
   do_sequence_frame_change_update(scene, seq);
@@ -298,7 +298,7 @@ static void rna_Sequence_start_frame_set(PointerRNA *ptr, int value)
 static void rna_Sequence_start_frame_final_set(PointerRNA *ptr, int value)
 {
   Sequence *seq = (Sequence *)ptr->data;
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   BKE_sequence_tx_set_final_left(seq, value);
   BKE_sequence_single_fix(seq);
@@ -308,7 +308,7 @@ static void rna_Sequence_start_frame_final_set(PointerRNA *ptr, int value)
 static void rna_Sequence_end_frame_final_set(PointerRNA *ptr, int value)
 {
   Sequence *seq = (Sequence *)ptr->data;
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   BKE_sequence_tx_set_final_right(seq, value);
   BKE_sequence_single_fix(seq);
@@ -318,7 +318,7 @@ static void rna_Sequence_end_frame_final_set(PointerRNA *ptr, int value)
 static void rna_Sequence_anim_startofs_final_set(PointerRNA *ptr, int value)
 {
   Sequence *seq = (Sequence *)ptr->data;
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   seq->anim_startofs = MIN2(value, seq->len + seq->anim_startofs);
 
@@ -329,7 +329,7 @@ static void rna_Sequence_anim_startofs_final_set(PointerRNA *ptr, int value)
 static void rna_Sequence_anim_endofs_final_set(PointerRNA *ptr, int value)
 {
   Sequence *seq = (Sequence *)ptr->data;
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   seq->anim_endofs = MIN2(value, seq->len + seq->anim_endofs);
 
@@ -340,7 +340,7 @@ static void rna_Sequence_anim_endofs_final_set(PointerRNA *ptr, int value)
 static void rna_Sequence_frame_length_set(PointerRNA *ptr, int value)
 {
   Sequence *seq = (Sequence *)ptr->data;
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   BKE_sequence_tx_set_final_right(seq, BKE_sequence_tx_get_final_left(seq, false) + value);
   do_sequence_frame_change_update(scene, seq);
@@ -362,7 +362,7 @@ static int rna_Sequence_frame_editable(PointerRNA *ptr, const char **UNUSED(r_in
 static void rna_Sequence_channel_set(PointerRNA *ptr, int value)
 {
   Sequence *seq = (Sequence *)ptr->data;
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   ListBase *seqbase = BKE_sequence_seqbase(&ed->seqbase, seq);
 
@@ -445,7 +445,7 @@ static Sequence *sequence_get_by_transform(Editing *ed, StripTransform *transfor
 
 static char *rna_SequenceTransform_path(PointerRNA *ptr)
 {
-  Scene *scene = ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_transform(ed, ptr->data);
 
@@ -464,7 +464,7 @@ static void rna_SequenceTransform_update(Main *UNUSED(bmain),
                                          Scene *UNUSED(scene),
                                          PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_transform(ed, ptr->data);
 
@@ -497,7 +497,7 @@ static Sequence *sequence_get_by_crop(Editing *ed, StripCrop *crop)
 
 static char *rna_SequenceCrop_path(PointerRNA *ptr)
 {
-  Scene *scene = ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_crop(ed, ptr->data);
 
@@ -514,7 +514,7 @@ static char *rna_SequenceCrop_path(PointerRNA *ptr)
 
 static void rna_SequenceCrop_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_crop(ed, ptr->data);
 
@@ -551,7 +551,7 @@ static int rna_Sequence_name_length(PointerRNA *ptr)
 
 static void rna_Sequence_name_set(PointerRNA *ptr, const char *value)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Sequence *seq = (Sequence *)ptr->data;
   char oldname[sizeof(seq->name)];
   AnimData *adt;
@@ -782,7 +782,7 @@ static void rna_SequenceElement_filename_set(PointerRNA *ptr, const char *value)
 
 static void rna_Sequence_reopen_files_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
 
   BKE_sequencer_free_imbuf(scene, &ed->seqbase, false);
@@ -795,14 +795,14 @@ static void rna_Sequence_reopen_files_update(Main *bmain, Scene *UNUSED(scene),
 
 static void rna_Sequence_mute_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
 
   DEG_id_tag_update(&scene->id, ID_RECALC_SEQUENCER_STRIPS);
 }
 
 static void rna_Sequence_filepath_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Sequence *seq = (Sequence *)(ptr->data);
   BKE_sequence_reload_new_file(bmain, scene, seq, true);
   BKE_sequence_calc(scene, seq);
@@ -838,7 +838,7 @@ static Sequence *sequence_get_by_proxy(Editing *ed, StripProxy *proxy)
 
 static void rna_Sequence_tcindex_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_proxy(ed, ptr->data);
 
@@ -848,7 +848,7 @@ static void rna_Sequence_tcindex_update(Main *bmain, Scene *UNUSED(scene), Point
 
 static void rna_SequenceProxy_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_proxy(ed, ptr->data);
 
@@ -911,7 +911,7 @@ static Sequence *sequence_get_by_colorbalance(Editing *ed,
 
 static char *rna_SequenceColorBalance_path(PointerRNA *ptr)
 {
-  Scene *scene = ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   SequenceModifierData *smd;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_colorbalance(ed, ptr->data, &smd);
@@ -944,7 +944,7 @@ static void rna_SequenceColorBalance_update(Main *UNUSED(bmain),
                                             Scene *UNUSED(scene),
                                             PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   SequenceModifierData *smd;
   Sequence *seq = sequence_get_by_colorbalance(ed, ptr->data, &smd);
@@ -954,7 +954,7 @@ static void rna_SequenceColorBalance_update(Main *UNUSED(bmain),
 
 static void rna_SequenceEditor_overlay_lock_set(PointerRNA *ptr, bool value)
 {
-  Scene *scene = ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
 
   if (ed == NULL) {
@@ -974,7 +974,7 @@ static void rna_SequenceEditor_overlay_lock_set(PointerRNA *ptr, bool value)
 
 static int rna_SequenceEditor_overlay_frame_get(PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
 
   if (ed == NULL) {
@@ -991,7 +991,7 @@ static int rna_SequenceEditor_overlay_frame_get(PointerRNA *ptr)
 
 static void rna_SequenceEditor_overlay_frame_set(PointerRNA *ptr, int value)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
 
   if (ed == NULL) {
@@ -1055,7 +1055,7 @@ static StructRNA *rna_SequenceModifier_refine(struct PointerRNA *ptr)
 
 static char *rna_SequenceModifier_path(PointerRNA *ptr)
 {
-  Scene *scene = ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   SequenceModifierData *smd = ptr->data;
   Sequence *seq = sequence_get_by_modifier(ed, smd);
@@ -1077,7 +1077,7 @@ static char *rna_SequenceModifier_path(PointerRNA *ptr)
 static void rna_SequenceModifier_name_set(PointerRNA *ptr, const char *value)
 {
   SequenceModifierData *smd = ptr->data;
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_modifier(ed, smd);
   AnimData *adt;
@@ -1106,7 +1106,7 @@ static void rna_SequenceModifier_name_set(PointerRNA *ptr, const char *value)
 static void rna_SequenceModifier_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
   /* strip from other scenes could be modified, so using active scene is not reliable */
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_modifier(ed, ptr->data);
 
@@ -1115,7 +1115,7 @@ static void rna_SequenceModifier_update(Main *UNUSED(bmain), Scene *UNUSED(scene
 
 static bool rna_SequenceModifier_otherSequence_poll(PointerRNA *ptr, PointerRNA value)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Editing *ed = BKE_sequencer_editing_get(scene, false);
   Sequence *seq = sequence_get_by_modifier(ed, ptr->data);
   Sequence *cur = (Sequence *)value.data;
@@ -1181,7 +1181,7 @@ static void rna_Sequence_modifier_clear(Sequence *seq, bContext *C)
 
 static float rna_Sequence_fps_get(PointerRNA *ptr)
 {
-  Scene *scene = (Scene *)ptr->id.data;
+  Scene *scene = (Scene *)ptr->owner_id;
   Sequence *seq = (Sequence *)(ptr->data);
   return BKE_sequence_get_fps(scene, seq);
 }
@@ -1457,7 +1457,7 @@ static const EnumPropertyItem blend_mode_items[] = {
     {0, "", ICON_NONE, NULL, NULL},
     {SEQ_TYPE_DARKEN, "DARKEN", 0, "Darken", ""},
     {SEQ_TYPE_MUL, "MULTIPLY", 0, "Multiply", ""},
-    {SEQ_TYPE_BURN, "BURN", 0, "Burn", ""},
+    {SEQ_TYPE_COLOR_BURN, "BURN", 0, "Color Burn", ""},
     {SEQ_TYPE_LINEAR_BURN, "LINEAR_BURN", 0, "Linear Burn", ""},
     {0, "", ICON_NONE, NULL, NULL},
     {SEQ_TYPE_LIGHTEN, "LIGHTEN", 0, "Lighten", ""},
@@ -2758,14 +2758,16 @@ static void rna_def_text(StructRNA *srna)
   prop = RNA_def_property(srna, "align_x", PROP_ENUM, PROP_NONE);
   RNA_def_property_enum_sdna(prop, NULL, "align");
   RNA_def_property_enum_items(prop, text_align_x_items);
-  RNA_def_property_ui_text(prop, "Align X", "Align the text along the X axis");
+  RNA_def_property_ui_text(
+      prop, "Align X", "Align the text along the X axis, relative to the text midpoint");
   RNA_def_property_update(
       prop, NC_SCENE | ND_SEQUENCER, "rna_Sequence_invalidate_preprocessed_update");
 
   prop = RNA_def_property(srna, "align_y", PROP_ENUM, PROP_NONE);
   RNA_def_property_enum_sdna(prop, NULL, "align_y");
   RNA_def_property_enum_items(prop, text_align_y_items);
-  RNA_def_property_ui_text(prop, "Align Y", "Align the image along the Y axis");
+  RNA_def_property_ui_text(
+      prop, "Align Y", "Align the image along the Y axis, relative to the text midpoint");
   RNA_def_property_update(
       prop, NC_SCENE | ND_SEQUENCER, "rna_Sequence_invalidate_preprocessed_update");
 
@@ -2792,7 +2794,7 @@ static void rna_def_color_mix(StructRNA *srna)
       {SEQ_TYPE_SCREEN, "SCREEN", 0, "Screen", ""},
       {SEQ_TYPE_OVERLAY, "OVERLAY", 0, "Overlay", ""},
       {SEQ_TYPE_DODGE, "DODGE", 0, "Dodge", ""},
-      {SEQ_TYPE_BURN, "BURN", 0, "Burn", ""},
+      {SEQ_TYPE_COLOR_BURN, "BURN", 0, "Color Burn", ""},
       {SEQ_TYPE_LINEAR_BURN, "LINEAR_BURN", 0, "Linear Burn", ""},
       {SEQ_TYPE_SOFT_LIGHT, "SOFT_LIGHT", 0, "Soft Light", ""},
       {SEQ_TYPE_HARD_LIGHT, "HARD_LIGHT", 0, "Hard Light", ""},
diff --git a/source/blender/makesrna/intern/rna_shader_fx.c b/source/blender/makesrna/intern/rna_shader_fx.c
index e48e682018d..acf525e1788 100644
--- a/source/blender/makesrna/intern/rna_shader_fx.c
+++ b/source/blender/makesrna/intern/rna_shader_fx.c
@@ -140,8 +140,8 @@ static void rna_ShaderFx_name_set(PointerRNA *ptr, const char *value)
   BLI_strncpy_utf8(gmd->name, value, sizeof(gmd->name));
 
   /* make sure the name is truly unique */
-  if (ptr->id.data) {
-    Object *ob = ptr->id.data;
+  if (ptr->owner_id) {
+    Object *ob = (Object *)ptr->owner_id;
     BKE_shaderfx_unique_name(&ob->shader_fx, gmd);
   }
 
@@ -160,8 +160,8 @@ static char *rna_ShaderFx_path(PointerRNA *ptr)
 
 static void rna_ShaderFx_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
-  WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ptr->id.data);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
+  WM_main_add_notifier(NC_OBJECT | ND_MODIFIER, ptr->owner_id);
 }
 
 static void rna_ShaderFx_dependency_update(Main *bmain, Scene *scene, PointerRNA *ptr)
@@ -189,7 +189,7 @@ static void shaderfx_object_set(Object *self, Object **ob_p, int type, PointerRN
         PointerRNA *ptr, PointerRNA value, struct ReportList *UNUSED(reports)) \
     { \
       _type##ShaderFxData *tmd = (_type##ShaderFxData *)ptr->data; \
-      shaderfx_object_set(ptr->id.data, &tmd->_prop, _obtype, value); \
+      shaderfx_object_set((Object *)ptr->owner_id, &tmd->_prop, _obtype, value); \
     }
 
 RNA_FX_OBJECT_SET(Light, object, OB_EMPTY);
@@ -212,7 +212,7 @@ static void rna_def_shader_fx_blur(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "factor", PROP_INT, PROP_PIXEL);
   RNA_def_property_int_sdna(prop, NULL, "radius");
-  RNA_def_property_range(prop, 0, INT_MAX);
+  RNA_def_property_range(prop, 0, SHRT_MAX);
   RNA_def_property_ui_text(prop, "Factor", "Factor of Blur");
   RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_ShaderFx_update");
 
@@ -329,7 +329,7 @@ static void rna_def_shader_fx_pixel(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "size", PROP_INT, PROP_PIXEL);
   RNA_def_property_int_sdna(prop, NULL, "size");
-  RNA_def_property_range(prop, 1, INT_MAX);
+  RNA_def_property_range(prop, 1, SHRT_MAX);
   RNA_def_property_array(prop, 2);
   RNA_def_property_ui_text(prop, "Size", "Pixel size");
   RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_ShaderFx_update");
@@ -359,7 +359,7 @@ static void rna_def_shader_fx_rim(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "offset", PROP_INT, PROP_PIXEL);
   RNA_def_property_int_sdna(prop, NULL, "offset");
-  RNA_def_property_range(prop, -INT_MAX, INT_MAX);
+  RNA_def_property_range(prop, SHRT_MIN, SHRT_MAX);
   RNA_def_property_ui_text(prop, "Offset", "Offset of the rim");
   RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_ShaderFx_update");
 
@@ -385,7 +385,7 @@ static void rna_def_shader_fx_rim(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "blur", PROP_INT, PROP_PIXEL);
   RNA_def_property_int_sdna(prop, NULL, "blur");
-  RNA_def_property_range(prop, 0, INT_MAX);
+  RNA_def_property_range(prop, 0, SHRT_MAX);
   RNA_def_property_ui_text(
       prop, "Blur", "Number of pixels for blurring rim (set to 0 to disable)");
   RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_ShaderFx_update");
@@ -418,11 +418,12 @@ static void rna_def_shader_fx_shadow(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Object", "Object to determine center of rotation");
   RNA_def_property_pointer_funcs(prop, NULL, "rna_ShadowShaderFx_object_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_ShaderFx_dependency_update");
 
   prop = RNA_def_property(srna, "offset", PROP_INT, PROP_PIXEL);
   RNA_def_property_int_sdna(prop, NULL, "offset");
-  RNA_def_property_range(prop, -INT_MAX, INT_MAX);
+  RNA_def_property_range(prop, SHRT_MIN, SHRT_MAX);
   RNA_def_property_ui_text(prop, "Offset", "Offset of the shadow");
   RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_ShaderFx_update");
 
@@ -472,7 +473,7 @@ static void rna_def_shader_fx_shadow(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "blur", PROP_INT, PROP_PIXEL);
   RNA_def_property_int_sdna(prop, NULL, "blur");
-  RNA_def_property_range(prop, 0, INT_MAX);
+  RNA_def_property_range(prop, 0, SHRT_MAX);
   RNA_def_property_ui_text(
       prop, "Blur", "Number of pixels for blurring shadow (set to 0 to disable)");
   RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_ShaderFx_update");
@@ -537,7 +538,7 @@ static void rna_def_shader_fx_glow(BlenderRNA *brna)
    * but only makes public first array element. */
   prop = RNA_def_property(srna, "radius", PROP_INT, PROP_PIXEL);
   RNA_def_property_int_sdna(prop, NULL, "blur[0]");
-  RNA_def_property_range(prop, 0, INT_MAX);
+  RNA_def_property_range(prop, 0, SHRT_MAX);
   RNA_def_property_ui_text(
       prop, "Radius", "Number of pixels for blurring glow (set to 0 to disable)");
   RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_ShaderFx_update");
@@ -568,7 +569,7 @@ static void rna_def_shader_fx_swirl(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "radius", PROP_INT, PROP_PIXEL);
   RNA_def_property_int_sdna(prop, NULL, "radius");
-  RNA_def_property_range(prop, 0, INT_MAX);
+  RNA_def_property_range(prop, 0, SHRT_MAX);
   RNA_def_property_ui_text(prop, "Radius", "Radius to apply");
   RNA_def_property_update(prop, NC_OBJECT | ND_MODIFIER, "rna_ShaderFx_update");
 
@@ -588,6 +589,7 @@ static void rna_def_shader_fx_swirl(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Object", "Object to determine center location");
   RNA_def_property_pointer_funcs(prop, NULL, "rna_SwirlShaderFx_object_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_ShaderFx_dependency_update");
 }
 
@@ -640,6 +642,7 @@ static void rna_def_shader_fx_light(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Object", "Object to determine light source location");
   RNA_def_property_pointer_funcs(prop, NULL, "rna_LightShaderFx_object_set", NULL, NULL);
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_update(prop, 0, "rna_ShaderFx_dependency_update");
 }
 
diff --git a/source/blender/makesrna/intern/rna_smoke.c b/source/blender/makesrna/intern/rna_smoke.c
index 5c04802ac3c..fa0297ac306 100644
--- a/source/blender/makesrna/intern/rna_smoke.c
+++ b/source/blender/makesrna/intern/rna_smoke.c
@@ -55,7 +55,7 @@
 
 static void rna_Smoke_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
 }
 
 static void rna_Smoke_dependency_update(Main *bmain, Scene *scene, PointerRNA *ptr)
@@ -70,13 +70,13 @@ static void rna_Smoke_resetCache(Main *UNUSED(bmain), Scene *UNUSED(scene), Poin
   if (settings->smd && settings->smd->domain) {
     settings->point_cache[0]->flag |= PTCACHE_OUTDATED;
   }
-  DEG_id_tag_update(ptr->id.data, ID_RECALC_GEOMETRY);
+  DEG_id_tag_update(ptr->owner_id, ID_RECALC_GEOMETRY);
 }
 
 static void rna_Smoke_cachetype_set(struct PointerRNA *ptr, int value)
 {
   SmokeDomainSettings *settings = (SmokeDomainSettings *)ptr->data;
-  Object *ob = (Object *)ptr->id.data;
+  Object *ob = (Object *)ptr->owner_id;
 
   if (value != settings->cache_file_format) {
     /* Clear old caches. */
diff --git a/source/blender/makesrna/intern/rna_space.c b/source/blender/makesrna/intern/rna_space.c
index 8108311319b..d866e1a98be 100644
--- a/source/blender/makesrna/intern/rna_space.c
+++ b/source/blender/makesrna/intern/rna_space.c
@@ -422,19 +422,19 @@ const EnumPropertyItem rna_enum_file_sort_items[] = {
     {FILE_SORT_ALPHA,
      "FILE_SORT_ALPHA",
      ICON_SORTALPHA,
-     "Sort alphabetically",
+     "Name",
      "Sort the file list alphabetically"},
     {FILE_SORT_EXTENSION,
      "FILE_SORT_EXTENSION",
      ICON_SORTBYEXT,
-     "Sort by extension",
+     "Extension",
      "Sort the file list by extension/type"},
     {FILE_SORT_TIME,
      "FILE_SORT_TIME",
      ICON_SORTTIME,
-     "Sort by time",
+     "Modified Date",
      "Sort files by modification time"},
-    {FILE_SORT_SIZE, "FILE_SORT_SIZE", ICON_SORTSIZE, "Sort by size", "Sort files by size"},
+    {FILE_SORT_SIZE, "FILE_SORT_SIZE", ICON_SORTSIZE, "Size", "Sort files by size"},
     {0, NULL, 0, NULL, NULL},
 };
 
@@ -524,7 +524,7 @@ static StructRNA *rna_Space_refine(struct PointerRNA *ptr)
 
 static ScrArea *rna_area_from_space(PointerRNA *ptr)
 {
-  bScreen *sc = (bScreen *)ptr->id.data;
+  bScreen *sc = (bScreen *)ptr->owner_id;
   SpaceLink *link = (SpaceLink *)ptr->data;
   return BKE_screen_find_area_from_space(sc, link);
 }
@@ -553,7 +553,7 @@ static void area_region_from_regiondata(bScreen *sc,
 
 static void rna_area_region_from_regiondata(PointerRNA *ptr, ScrArea **r_sa, ARegion **r_ar)
 {
-  bScreen *sc = (bScreen *)ptr->id.data;
+  bScreen *sc = (bScreen *)ptr->owner_id;
   void *regiondata = ptr->data;
 
   area_region_from_regiondata(sc, regiondata, r_sa, r_ar);
@@ -771,7 +771,7 @@ static void rna_Space_view2d_sync_update(Main *UNUSED(bmain),
   ar = BKE_area_find_region_type(sa, RGN_TYPE_WINDOW);
 
   if (ar) {
-    bScreen *sc = (bScreen *)ptr->id.data;
+    bScreen *sc = (bScreen *)ptr->owner_id;
     View2D *v2d = &ar->v2d;
 
     UI_view2d_sync(sc, sa, v2d, V2D_LOCK_SET);
@@ -804,7 +804,7 @@ static void rna_SpaceView3D_camera_update(Main *bmain, Scene *scene, PointerRNA
 static void rna_SpaceView3D_use_local_camera_set(PointerRNA *ptr, bool value)
 {
   View3D *v3d = (View3D *)(ptr->data);
-  bScreen *sc = (bScreen *)ptr->id.data;
+  bScreen *sc = (bScreen *)ptr->owner_id;
 
   v3d->scenelock = !value;
 
@@ -817,7 +817,7 @@ static void rna_SpaceView3D_use_local_camera_set(PointerRNA *ptr, bool value)
 static float rna_View3DOverlay_GridScaleUnit_get(PointerRNA *ptr)
 {
   View3D *v3d = (View3D *)(ptr->data);
-  bScreen *screen = ptr->id.data;
+  bScreen *screen = (bScreen *)ptr->owner_id;
   Scene *scene = ED_screen_scene_find(screen, G_MAIN->wm.first);
 
   return ED_view3d_grid_scale(scene, v3d, NULL);
@@ -934,7 +934,7 @@ static bool rna_RegionView3D_is_orthographic_side_view_get(PointerRNA *ptr)
 
 static void rna_3DViewShading_type_update(Main *bmain, Scene *scene, PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   if (GS(id->name) == ID_SCE) {
     return;
   }
@@ -951,7 +951,7 @@ static void rna_3DViewShading_type_update(Main *bmain, Scene *scene, PointerRNA
     }
   }
 
-  bScreen *screen = ptr->id.data;
+  bScreen *screen = (bScreen *)ptr->owner_id;
   for (ScrArea *sa = screen->areabase.first; sa; sa = sa->next) {
     for (SpaceLink *sl = sa->spacedata.first; sl; sl = sl->next) {
       if (sl->spacetype == SPACE_VIEW3D) {
@@ -968,12 +968,12 @@ static void rna_3DViewShading_type_update(Main *bmain, Scene *scene, PointerRNA
 static Scene *rna_3DViewShading_scene(PointerRNA *ptr)
 {
   /* Get scene, depends if using 3D view or OpenGL render settings. */
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   if (GS(id->name) == ID_SCE) {
     return (Scene *)id;
   }
   else {
-    bScreen *screen = ptr->id.data;
+    bScreen *screen = (bScreen *)ptr->owner_id;
     return WM_windows_scene_get_from_screen(G_MAIN->wm.first, screen);
   }
 }
@@ -1296,7 +1296,7 @@ static bool rna_SpaceImageEditor_show_paint_get(PointerRNA *ptr)
 static bool rna_SpaceImageEditor_show_uvedit_get(PointerRNA *ptr)
 {
   SpaceImage *sima = (SpaceImage *)(ptr->data);
-  bScreen *sc = (bScreen *)ptr->id.data;
+  bScreen *sc = (bScreen *)ptr->owner_id;
   wmWindow *win = ED_screen_window_find(sc, G_MAIN->wm.first);
   ViewLayer *view_layer = WM_window_get_active_view_layer(win);
   Object *obedit = OBEDIT_FROM_VIEW_LAYER(view_layer);
@@ -1306,7 +1306,7 @@ static bool rna_SpaceImageEditor_show_uvedit_get(PointerRNA *ptr)
 static bool rna_SpaceImageEditor_show_maskedit_get(PointerRNA *ptr)
 {
   SpaceImage *sima = (SpaceImage *)(ptr->data);
-  bScreen *sc = (bScreen *)ptr->id.data;
+  bScreen *sc = (bScreen *)ptr->owner_id;
   wmWindow *win = ED_screen_window_find(sc, G_MAIN->wm.first);
   ViewLayer *view_layer = WM_window_get_active_view_layer(win);
   return ED_space_image_check_show_maskedit(sima, view_layer);
@@ -1317,7 +1317,7 @@ static void rna_SpaceImageEditor_image_set(PointerRNA *ptr,
                                            struct ReportList *UNUSED(reports))
 {
   SpaceImage *sima = (SpaceImage *)(ptr->data);
-  bScreen *sc = (bScreen *)ptr->id.data;
+  bScreen *sc = (bScreen *)ptr->owner_id;
   wmWindow *win = ED_screen_window_find(sc, G_MAIN->wm.first);
   ViewLayer *view_layer = WM_window_get_active_view_layer(win);
   Object *obedit = OBEDIT_FROM_VIEW_LAYER(view_layer);
@@ -2106,7 +2106,7 @@ static void rna_SpaceClipEditor_clip_set(PointerRNA *ptr,
                                          struct ReportList *UNUSED(reports))
 {
   SpaceClip *sc = (SpaceClip *)(ptr->data);
-  bScreen *screen = (bScreen *)ptr->id.data;
+  bScreen *screen = (bScreen *)ptr->owner_id;
 
   ED_space_clip_set_clip(NULL, screen, sc, (MovieClip *)value.data);
 }
@@ -2156,6 +2156,18 @@ static void rna_SpaceClipEditor_view_type_update(Main *UNUSED(bmain),
 
 /* File browser. */
 
+int rna_FileSelectParams_filename_editable(struct PointerRNA *ptr, const char **r_info)
+{
+  FileSelectParams *params = ptr->data;
+
+  if (params && (params->flag & FILE_DIRSEL_ONLY)) {
+    *r_info = "Only directories can be chosen for the current operation.";
+    return 0;
+  }
+
+  return params ? PROP_EDITABLE : 0;
+}
+
 static bool rna_FileSelectParams_use_lib_get(PointerRNA *ptr)
 {
   FileSelectParams *params = ptr->data;
@@ -2725,11 +2737,6 @@ static void rna_def_space_image_uv(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Display Faces", "Display faces over the image");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_IMAGE, NULL);
 
-  prop = RNA_def_property(srna, "show_edges", PROP_BOOLEAN, PROP_NONE);
-  RNA_def_property_boolean_negative_sdna(prop, NULL, "flag", SI_NO_DRAWEDGES);
-  RNA_def_property_ui_text(prop, "Display Edges", "Display edges in vertex select mode");
-  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_IMAGE, NULL);
-
   /* todo: move edge and face drawing options here from G.f */
 
   prop = RNA_def_property(srna, "pixel_snap_mode", PROP_ENUM, PROP_NONE);
@@ -2791,6 +2798,7 @@ static void rna_def_space_outliner(BlenderRNA *brna)
   static const EnumPropertyItem filter_state_items[] = {
       {SO_FILTER_OB_ALL, "ALL", 0, "All", "Show all objects in the view layer"},
       {SO_FILTER_OB_VISIBLE, "VISIBLE", 0, "Visible", "Show visible objects"},
+      {SO_FILTER_OB_HIDDEN, "HIDDEN", 0, "Hidden", "Show hidden objects"},
       {SO_FILTER_OB_SELECTED, "SELECTED", 0, "Selected", "Show selected objects"},
       {SO_FILTER_OB_ACTIVE, "ACTIVE", 0, "Active", "Show only the active object"},
       {0, NULL, 0, NULL, NULL},
@@ -2829,6 +2837,12 @@ static void rna_def_space_outliner(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Sort Alphabetically", "");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_OUTLINER, NULL);
 
+  prop = RNA_def_property(srna, "use_sync_select", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", SO_SYNC_SELECT);
+  RNA_def_property_ui_text(
+      prop, "Sync Outliner Selection", "Sync outliner selection with other editors");
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_OUTLINER, NULL);
+
   /* Granular restriction column option. */
   prop = RNA_def_property(srna, "show_restrict_column_enable", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "show_restrict_flags", SO_RESTRICT_ENABLE);
@@ -3788,7 +3802,7 @@ static void rna_def_space_view3d(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "show_gizmo_navigate", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_negative_sdna(prop, NULL, "gizmo_flag", V3D_GIZMO_HIDE_NAVIGATE);
-  RNA_def_property_ui_text(prop, "Navigate Gizmo", "");
+  RNA_def_property_ui_text(prop, "Navigate Gizmo", "Viewport navigation gizmo");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_VIEW3D, NULL);
 
   prop = RNA_def_property(srna, "show_gizmo_context", PROP_BOOLEAN, PROP_NONE);
@@ -4440,13 +4454,6 @@ static void rna_def_space_sequencer(BlenderRNA *brna)
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_SEQUENCER, NULL);
 
   /* flags */
-  prop = RNA_def_property(srna, "show_frame_indicator", PROP_BOOLEAN, PROP_NONE);
-  RNA_def_property_boolean_negative_sdna(prop, NULL, "flag", SEQ_NO_DRAW_CFRANUM);
-  RNA_def_property_ui_text(prop,
-                           "Show Frame Number Indicator",
-                           "Show frame number beside the current frame indicator line");
-  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_SEQUENCER, NULL);
-
   prop = RNA_def_property(srna, "show_frames", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", SEQ_DRAWFRAMES);
   RNA_def_property_ui_text(prop, "Display Frames", "Display frames rather than seconds");
@@ -4734,13 +4741,6 @@ static void rna_def_space_dopesheet(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Show Seconds", "Show timing in seconds not frames");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_DOPESHEET, NULL);
 
-  prop = RNA_def_property(srna, "show_frame_indicator", PROP_BOOLEAN, PROP_NONE);
-  RNA_def_property_boolean_negative_sdna(prop, NULL, "flag", SACTION_NODRAWCFRANUM);
-  RNA_def_property_ui_text(prop,
-                           "Show Frame Number Indicator",
-                           "Show frame number beside the current frame indicator line");
-  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_DOPESHEET, NULL);
-
   prop = RNA_def_property(srna, "show_sliders", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", SACTION_SLIDERS);
   RNA_def_property_ui_text(prop, "Show Sliders", "Show sliders beside F-Curve channels");
@@ -4896,13 +4896,6 @@ static void rna_def_space_graph(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Show Seconds", "Show timing in seconds not frames");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_GRAPH, NULL);
 
-  prop = RNA_def_property(srna, "show_frame_indicator", PROP_BOOLEAN, PROP_NONE);
-  RNA_def_property_boolean_negative_sdna(prop, NULL, "flag", SIPO_NODRAWCFRANUM);
-  RNA_def_property_ui_text(prop,
-                           "Show Frame Number Indicator",
-                           "Show frame number beside the current frame indicator line");
-  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_GRAPH, NULL);
-
   prop = RNA_def_property(srna, "show_sliders", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", SIPO_SLIDERS);
   RNA_def_property_ui_text(prop, "Show Sliders", "Show sliders beside F-Curve channels");
@@ -5040,13 +5033,6 @@ static void rna_def_space_nla(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Show Seconds", "Show timing in seconds not frames");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_NLA, NULL);
 
-  prop = RNA_def_property(srna, "show_frame_indicator", PROP_BOOLEAN, PROP_NONE);
-  RNA_def_property_boolean_negative_sdna(prop, NULL, "flag", SNLA_NODRAWCFRANUM);
-  RNA_def_property_ui_text(prop,
-                           "Show Frame Number Indicator",
-                           "Show frame number beside the current frame indicator line");
-  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_NLA, NULL);
-
   prop = RNA_def_property(srna, "show_strip_curves", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_negative_sdna(prop, NULL, "flag", SNLA_NOSTRIPCURVES);
   RNA_def_property_ui_text(prop, "Show Control F-Curves", "Show influence F-Curves on strips");
@@ -5173,25 +5159,25 @@ static void rna_def_fileselect_params(BlenderRNA *brna)
   PropertyRNA *prop;
 
   static const EnumPropertyItem file_display_items[] = {
-      {FILE_SHORTDISPLAY,
-       "LIST_SHORT",
-       ICON_SHORTDISPLAY,
-       "Short List",
-       "Display files as short list"},
-      {FILE_LONGDISPLAY,
-       "LIST_LONG",
+      {FILE_VERTICALDISPLAY,
+       "LIST_VERTICAL",
        ICON_LONGDISPLAY,
-       "Long List",
-       "Display files as a detailed list"},
+       "Vertical List",
+       "Display files as a vertical list"},
+      {FILE_HORIZONTALDISPLAY,
+       "LIST_HORIZONTAL",
+       ICON_SHORTDISPLAY,
+       "Horizontal List",
+       "Display files as a horizontal list"},
       {FILE_IMGDISPLAY, "THUMBNAIL", ICON_IMGDISPLAY, "Thumbnails", "Display files as thumbnails"},
       {0, NULL, 0, NULL, NULL},
   };
 
   static const EnumPropertyItem display_size_items[] = {
-      {32, "TINY", 0, "Tiny", ""},
-      {64, "SMALL", 0, "Small", ""},
+      {64, "TINY", 0, "Tiny", ""},
+      {96, "SMALL", 0, "Small", ""},
       {128, "NORMAL", 0, "Regular", ""},
-      {256, "LARGE", 0, "Large", ""},
+      {192, "LARGE", 0, "Large", ""},
       {0, NULL, 0, NULL, NULL},
   };
 
@@ -5307,7 +5293,10 @@ static void rna_def_fileselect_params(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Title", "Title for the file browser");
   RNA_def_property_clear_flag(prop, PROP_EDITABLE);
 
-  prop = RNA_def_property(srna, "directory", PROP_STRING, PROP_DIRPATH);
+  /* Use BYTESTRING rather than DIRPATH as subtype so UI code doesn't add OT_directory_browse
+   * button when displaying this prop in the file browser (it would just open a file browser). That
+   * should be the only effective difference between the two. */
+  prop = RNA_def_property(srna, "directory", PROP_STRING, PROP_BYTESTRING);
   RNA_def_property_string_sdna(prop, NULL, "dir");
   RNA_def_property_ui_text(prop, "Directory", "Directory displayed in the file browser");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_FILE_PARAMS, NULL);
@@ -5315,6 +5304,7 @@ static void rna_def_fileselect_params(BlenderRNA *brna)
   prop = RNA_def_property(srna, "filename", PROP_STRING, PROP_FILENAME);
   RNA_def_property_string_sdna(prop, NULL, "file");
   RNA_def_property_ui_text(prop, "File Name", "Active file in the file browser");
+  RNA_def_property_editable_func(prop, "rna_FileSelectParams_filename_editable");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_FILE_PARAMS, NULL);
 
   prop = RNA_def_property(srna, "use_library_browsing", PROP_BOOLEAN, PROP_NONE);
@@ -5335,6 +5325,19 @@ static void rna_def_fileselect_params(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Recursion", "Numbers of dirtree levels to show simultaneously");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_FILE_PARAMS, NULL);
 
+  prop = RNA_def_property(srna, "show_details_size", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "details_flags", FILE_DETAILS_SIZE);
+  RNA_def_property_ui_text(prop, "File Size", "Draw a column listing the size of each file");
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_FILE_PARAMS, NULL);
+
+  prop = RNA_def_property(srna, "show_details_datetime", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "details_flags", FILE_DETAILS_DATETIME);
+  RNA_def_property_ui_text(
+      prop,
+      "File Modification Date",
+      "Draw a column listing the date and time of modification for each file");
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_FILE_PARAMS, NULL);
+
   prop = RNA_def_property(srna, "use_filter", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", FILE_FILTER);
   RNA_def_property_ui_text(prop, "Filter Files", "Enable filtering of files");
@@ -5351,6 +5354,12 @@ static void rna_def_fileselect_params(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Sort", "");
   RNA_def_property_update(prop, NC_SPACE | ND_SPACE_FILE_PARAMS, NULL);
 
+  prop = RNA_def_property(srna, "use_sort_invert", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", FILE_SORT_INVERT);
+  RNA_def_property_ui_text(
+      prop, "Reverse Sorting", "Sort items descending, from highest value to lowest");
+  RNA_def_property_update(prop, NC_SPACE | ND_SPACE_FILE_PARAMS, NULL);
+
   prop = RNA_def_property(srna, "use_filter_image", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "filter", FILE_TYPE_IMAGE);
   RNA_def_property_ui_text(prop, "Filter Images", "Show image files");
diff --git a/source/blender/makesrna/intern/rna_speaker.c b/source/blender/makesrna/intern/rna_speaker.c
index 0c63cf5eae8..8f156a67e08 100644
--- a/source/blender/makesrna/intern/rna_speaker.c
+++ b/source/blender/makesrna/intern/rna_speaker.c
@@ -61,6 +61,7 @@ static void rna_def_speaker(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "Sound");
   RNA_def_property_flag(prop, PROP_EDITABLE);
   RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Sound", "Sound data-block used by this speaker");
   /* RNA_def_property_float_funcs(prop, NULL, "rna_Speaker_sound_set", NULL); */
   /* RNA_def_property_update(prop, 0, "rna_Speaker_update"); */
diff --git a/source/blender/makesrna/intern/rna_texture.c b/source/blender/makesrna/intern/rna_texture.c
index 28d2d722460..2015356d071 100644
--- a/source/blender/makesrna/intern/rna_texture.c
+++ b/source/blender/makesrna/intern/rna_texture.c
@@ -186,10 +186,10 @@ static StructRNA *rna_Texture_refine(struct PointerRNA *ptr)
 
 static void rna_Texture_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   if (GS(id->name) == ID_TE) {
-    Tex *tex = ptr->id.data;
+    Tex *tex = (Tex *)ptr->owner_id;
 
     DEG_id_tag_update(&tex->id, 0);
     DEG_id_tag_update(&tex->id, ID_RECALC_EDITORS);
@@ -197,7 +197,7 @@ static void rna_Texture_update(Main *bmain, Scene *UNUSED(scene), PointerRNA *pt
     WM_main_add_notifier(NC_MATERIAL | ND_SHADING_DRAW, NULL);
   }
   else if (GS(id->name) == ID_NT) {
-    bNodeTree *ntree = ptr->id.data;
+    bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
     ED_node_tag_update_nodetree(bmain, ntree, NULL);
   }
 }
@@ -219,7 +219,7 @@ static void rna_Color_mapping_update(Main *UNUSED(bmain),
 /* Used for Texture Properties, used (also) for/in Nodes */
 static void rna_Texture_nodes_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  Tex *tex = ptr->id.data;
+  Tex *tex = (Tex *)ptr->owner_id;
 
   DEG_id_tag_update(&tex->id, 0);
   DEG_id_tag_update(&tex->id, ID_RECALC_EDITORS);
@@ -235,7 +235,7 @@ static void rna_Texture_type_set(PointerRNA *ptr, int value)
 
 void rna_TextureSlot_update(bContext *C, PointerRNA *ptr)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
 
   DEG_id_tag_update(id, 0);
 
@@ -290,8 +290,8 @@ char *rna_TextureSlot_path(PointerRNA *ptr)
    * since the name used is the name of the texture assigned, but the texture
    * may be used multiple times in the same stack
    */
-  if (ptr->id.data) {
-    if (GS(((ID *)ptr->id.data)->name) == ID_BR) {
+  if (ptr->owner_id) {
+    if (GS(ptr->owner_id->name) == ID_BR) {
       return BLI_strdup("texture_slot");
     }
     else {
@@ -299,7 +299,7 @@ char *rna_TextureSlot_path(PointerRNA *ptr)
       PropertyRNA *prop;
 
       /* find the 'textures' property of the ID-struct */
-      RNA_id_pointer_create(ptr->id.data, &id_ptr);
+      RNA_id_pointer_create(ptr->owner_id, &id_ptr);
       prop = RNA_struct_find_property(&id_ptr, "texture_slots");
 
       /* get an iterator for this property, and try to find the relevant index */
@@ -630,6 +630,7 @@ static void rna_def_mtex(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "tex");
   RNA_def_property_struct_type(prop, "Texture");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_CONTEXT_UPDATE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Texture", "Texture data-block used by this texture slot");
   RNA_def_property_update(prop, NC_MATERIAL | ND_SHADING_LINKS, "rna_TextureSlot_update");
 
@@ -834,7 +835,7 @@ static void rna_def_texture_clouds(BlenderRNA *brna)
   prop = RNA_def_property(srna, "noise_depth", PROP_INT, PROP_NONE);
   RNA_def_property_int_sdna(prop, NULL, "noisedepth");
   RNA_def_property_range(prop, 0, 30);
-  RNA_def_property_ui_range(prop, 0, 24, 0, 2);
+  RNA_def_property_ui_range(prop, 0, 24, 1, 2);
   RNA_def_property_ui_text(prop, "Noise Depth", "Depth of the cloud calculation");
   RNA_def_property_update(prop, 0, "rna_Texture_nodes_update");
 
@@ -973,7 +974,7 @@ static void rna_def_texture_marble(BlenderRNA *brna)
   prop = RNA_def_property(srna, "noise_depth", PROP_INT, PROP_NONE);
   RNA_def_property_int_sdna(prop, NULL, "noisedepth");
   RNA_def_property_range(prop, 0, 30);
-  RNA_def_property_ui_range(prop, 0, 24, 0, 2);
+  RNA_def_property_ui_range(prop, 0, 24, 1, 2);
   RNA_def_property_ui_text(prop, "Noise Depth", "Depth of the cloud calculation");
   RNA_def_property_update(prop, 0, "rna_Texture_update");
 
@@ -1027,7 +1028,7 @@ static void rna_def_texture_magic(BlenderRNA *brna)
   prop = RNA_def_property(srna, "noise_depth", PROP_INT, PROP_NONE);
   RNA_def_property_int_sdna(prop, NULL, "noisedepth");
   RNA_def_property_range(prop, 0, 30);
-  RNA_def_property_ui_range(prop, 0, 24, 0, 2);
+  RNA_def_property_ui_range(prop, 0, 24, 1, 2);
   RNA_def_property_ui_text(prop, "Noise Depth", "Depth of the noise");
   RNA_def_property_update(prop, 0, "rna_Texture_update");
 }
@@ -1277,6 +1278,7 @@ static void rna_def_texture_image(BlenderRNA *brna)
   RNA_def_property_pointer_sdna(prop, NULL, "ima");
   RNA_def_property_struct_type(prop, "Image");
   RNA_def_property_flag(prop, PROP_EDITABLE);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Image", "");
   RNA_def_property_update(prop, 0, "rna_Texture_update");
 
diff --git a/source/blender/makesrna/intern/rna_timeline.c b/source/blender/makesrna/intern/rna_timeline.c
index 3308a493212..ed17a509fc2 100644
--- a/source/blender/makesrna/intern/rna_timeline.c
+++ b/source/blender/makesrna/intern/rna_timeline.c
@@ -70,6 +70,7 @@ static void rna_def_timeline_marker(BlenderRNA *brna)
   prop = RNA_def_property(srna, "camera", PROP_POINTER, PROP_NONE);
   RNA_def_property_struct_type(prop, "Object");
   RNA_def_property_flag(prop, PROP_EDITABLE | PROP_ID_SELF_CHECK);
+  RNA_def_property_override_flag(prop, PROPOVERRIDE_OVERRIDABLE_LIBRARY);
   RNA_def_property_ui_text(prop, "Camera", "Camera that becomes active on this frame");
 #  endif
 }
diff --git a/source/blender/makesrna/intern/rna_tracking.c b/source/blender/makesrna/intern/rna_tracking.c
index 84bde91e654..0a824b3c67a 100644
--- a/source/blender/makesrna/intern/rna_tracking.c
+++ b/source/blender/makesrna/intern/rna_tracking.c
@@ -61,7 +61,7 @@ static void rna_tracking_defaultSettings_patternUpdate(Main *UNUSED(bmain),
                                                        Scene *UNUSED(scene),
                                                        PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTracking *tracking = &clip->tracking;
   MovieTrackingSettings *settings = &tracking->settings;
 
@@ -74,7 +74,7 @@ static void rna_tracking_defaultSettings_searchUpdate(Main *UNUSED(bmain),
                                                       Scene *UNUSED(scene),
                                                       PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTracking *tracking = &clip->tracking;
   MovieTrackingSettings *settings = &tracking->settings;
 
@@ -85,7 +85,7 @@ static void rna_tracking_defaultSettings_searchUpdate(Main *UNUSED(bmain),
 
 static char *rna_trackingTrack_path(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingTrack *track = (MovieTrackingTrack *)ptr->data;
   /* Escaped object name, escaped track name, rest of the path. */
   char rna_path[MAX_NAME * 4 + 64];
@@ -95,35 +95,35 @@ static char *rna_trackingTrack_path(PointerRNA *ptr)
 
 static void rna_trackingTracks_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   rna_iterator_listbase_begin(iter, &clip->tracking.tracks, NULL);
 }
 
 static void rna_trackingPlaneTracks_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   rna_iterator_listbase_begin(iter, &clip->tracking.plane_tracks, NULL);
 }
 
 static void rna_trackingObjects_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   rna_iterator_listbase_begin(iter, &clip->tracking.objects, NULL);
 }
 
 static int rna_tracking_active_object_index_get(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   return clip->tracking.objectnr;
 }
 
 static void rna_tracking_active_object_index_set(PointerRNA *ptr, int value)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   clip->tracking.objectnr = value;
   BKE_tracking_dopesheet_tag_update(&clip->tracking);
@@ -132,7 +132,7 @@ static void rna_tracking_active_object_index_set(PointerRNA *ptr, int value)
 static void rna_tracking_active_object_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   *min = 0;
   *max = max_ii(0, clip->tracking.tot_object - 1);
@@ -140,7 +140,7 @@ static void rna_tracking_active_object_index_range(
 
 static PointerRNA rna_tracking_active_track_get(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingTrack *act_track = BKE_tracking_track_get_active(&clip->tracking);
 
   return rna_pointer_inherit_refine(ptr, &RNA_MovieTrackingTrack, act_track);
@@ -150,7 +150,7 @@ static void rna_tracking_active_track_set(PointerRNA *ptr,
                                           PointerRNA value,
                                           struct ReportList *UNUSED(reports))
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingTrack *track = (MovieTrackingTrack *)value.data;
   ListBase *tracksbase = BKE_tracking_get_active_tracks(&clip->tracking);
   int index = BLI_findindex(tracksbase, track);
@@ -165,7 +165,7 @@ static void rna_tracking_active_track_set(PointerRNA *ptr,
 
 static PointerRNA rna_tracking_active_plane_track_get(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingPlaneTrack *act_plane_track = BKE_tracking_plane_track_get_active(&clip->tracking);
 
   return rna_pointer_inherit_refine(ptr, &RNA_MovieTrackingPlaneTrack, act_plane_track);
@@ -175,7 +175,7 @@ static void rna_tracking_active_plane_track_set(PointerRNA *ptr,
                                                 PointerRNA value,
                                                 struct ReportList *UNUSED(reports))
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingPlaneTrack *plane_track = (MovieTrackingPlaneTrack *)value.data;
   ListBase *plane_tracks_base = BKE_tracking_get_active_plane_tracks(&clip->tracking);
   int index = BLI_findindex(plane_tracks_base, plane_track);
@@ -190,7 +190,7 @@ static void rna_tracking_active_plane_track_set(PointerRNA *ptr,
 
 static void rna_trackingTrack_name_set(PointerRNA *ptr, const char *value)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingTrack *track = (MovieTrackingTrack *)ptr->data;
   ListBase *tracksbase = BKE_tracking_find_tracks_list_for_track(&clip->tracking, track);
   /* Store old name, for the animation fix later. */
@@ -233,7 +233,7 @@ static void rna_trackingTrack_select_set(PointerRNA *ptr, bool value)
 
 static void rna_trackingPlaneMarker_frame_set(PointerRNA *ptr, int value)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTracking *tracking = &clip->tracking;
   MovieTrackingPlaneMarker *plane_marker = (MovieTrackingPlaneMarker *)ptr->data;
   MovieTrackingObject *tracking_object;
@@ -268,7 +268,7 @@ static void rna_trackingPlaneMarker_frame_set(PointerRNA *ptr, int value)
 
 static char *rna_trackingPlaneTrack_path(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingPlaneTrack *plane_track = (MovieTrackingPlaneTrack *)ptr->data;
   /* Escaped object name, escaped track name, rest of the path. */
   char rna_path[MAX_NAME * 4 + 64];
@@ -279,7 +279,7 @@ static char *rna_trackingPlaneTrack_path(PointerRNA *ptr)
 
 static void rna_trackingPlaneTrack_name_set(PointerRNA *ptr, const char *value)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingPlaneTrack *plane_track = (MovieTrackingPlaneTrack *)ptr->data;
   ListBase *plane_tracks_base = BKE_tracking_find_tracks_list_for_plane_track(&clip->tracking,
                                                                               plane_track);
@@ -307,7 +307,7 @@ static char *rna_trackingCamera_path(PointerRNA *UNUSED(ptr))
 
 static float rna_trackingCamera_focal_mm_get(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingCamera *camera = &clip->tracking.camera;
   float val = camera->focal;
 
@@ -320,7 +320,7 @@ static float rna_trackingCamera_focal_mm_get(PointerRNA *ptr)
 
 static void rna_trackingCamera_focal_mm_set(PointerRNA *ptr, float value)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingCamera *camera = &clip->tracking.camera;
 
   if (clip->lastsize[0]) {
@@ -362,26 +362,26 @@ static int rna_track_2d_stabilization_rotation(CollectionPropertyIterator *UNUSE
 
 static void rna_tracking_stabTracks_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   rna_iterator_listbase_begin(iter, &clip->tracking.tracks, rna_track_2d_stabilization);
 }
 
 static int rna_tracking_stabTracks_active_index_get(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   return clip->tracking.stabilization.act_track;
 }
 
 static void rna_tracking_stabTracks_active_index_set(PointerRNA *ptr, int value)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   clip->tracking.stabilization.act_track = value;
 }
 
 static void rna_tracking_stabTracks_active_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   *min = 0;
   *max = max_ii(0, clip->tracking.stabilization.tot_track - 1);
@@ -389,26 +389,26 @@ static void rna_tracking_stabTracks_active_index_range(
 
 static void rna_tracking_stabRotTracks_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   rna_iterator_listbase_begin(iter, &clip->tracking.tracks, rna_track_2d_stabilization_rotation);
 }
 
 static int rna_tracking_stabRotTracks_active_index_get(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   return clip->tracking.stabilization.act_rot_track;
 }
 
 static void rna_tracking_stabRotTracks_active_index_set(PointerRNA *ptr, int value)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   clip->tracking.stabilization.act_rot_track = value;
 }
 
 static void rna_tracking_stabRotTracks_active_index_range(
     PointerRNA *ptr, int *min, int *max, int *UNUSED(softmin), int *UNUSED(softmax))
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   *min = 0;
   *max = max_ii(0, clip->tracking.stabilization.tot_rot_track - 1);
@@ -416,7 +416,7 @@ static void rna_tracking_stabRotTracks_active_index_range(
 
 static void rna_tracking_flushUpdate(Main *UNUSED(bmain), Scene *scene, PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   nodeUpdateID(scene->nodetree, &clip->id);
 
@@ -429,7 +429,7 @@ static void rna_tracking_resetIntrinsics(Main *UNUSED(bmain),
                                          Scene *UNUSED(scene),
                                          PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTracking *tracking = &clip->tracking;
 
   if (tracking->camera.intrinsics) {
@@ -443,7 +443,7 @@ static void rna_trackingObject_tracks_begin(CollectionPropertyIterator *iter, Po
   MovieTrackingObject *object = (MovieTrackingObject *)ptr->data;
 
   if (object->flag & TRACKING_OBJECT_CAMERA) {
-    MovieClip *clip = (MovieClip *)ptr->id.data;
+    MovieClip *clip = (MovieClip *)ptr->owner_id;
 
     rna_iterator_listbase_begin(iter, &clip->tracking.tracks, NULL);
   }
@@ -458,7 +458,7 @@ static void rna_trackingObject_plane_tracks_begin(CollectionPropertyIterator *it
   MovieTrackingObject *object = (MovieTrackingObject *)ptr->data;
 
   if (object->flag & TRACKING_OBJECT_CAMERA) {
-    MovieClip *clip = (MovieClip *)ptr->id.data;
+    MovieClip *clip = (MovieClip *)ptr->owner_id;
 
     rna_iterator_listbase_begin(iter, &clip->tracking.plane_tracks, NULL);
   }
@@ -472,7 +472,7 @@ static PointerRNA rna_trackingObject_reconstruction_get(PointerRNA *ptr)
   MovieTrackingObject *object = (MovieTrackingObject *)ptr->data;
 
   if (object->flag & TRACKING_OBJECT_CAMERA) {
-    MovieClip *clip = (MovieClip *)ptr->id.data;
+    MovieClip *clip = (MovieClip *)ptr->owner_id;
 
     return rna_pointer_inherit_refine(
         ptr, &RNA_MovieTrackingReconstruction, &clip->tracking.reconstruction);
@@ -485,7 +485,7 @@ static PointerRNA rna_trackingObject_reconstruction_get(PointerRNA *ptr)
 
 static PointerRNA rna_tracking_active_object_get(PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingObject *object = BLI_findlink(&clip->tracking.objects, clip->tracking.objectnr);
 
   return rna_pointer_inherit_refine(ptr, &RNA_MovieTrackingObject, object);
@@ -495,7 +495,7 @@ static void rna_tracking_active_object_set(PointerRNA *ptr,
                                            PointerRNA value,
                                            struct ReportList *UNUSED(reports))
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingObject *object = (MovieTrackingObject *)value.data;
   int index = BLI_findindex(&clip->tracking.objects, object);
 
@@ -509,7 +509,7 @@ static void rna_tracking_active_object_set(PointerRNA *ptr,
 
 static void rna_trackingObject_name_set(PointerRNA *ptr, const char *value)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingObject *object = (MovieTrackingObject *)ptr->data;
 
   BLI_strncpy(object->name, value, sizeof(object->name));
@@ -521,7 +521,7 @@ static void rna_trackingObject_flushUpdate(Main *UNUSED(bmain),
                                            Scene *UNUSED(scene),
                                            PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
 
   WM_main_add_notifier(NC_OBJECT | ND_TRANSFORM, NULL);
   DEG_id_tag_update(&clip->id, 0);
@@ -529,7 +529,7 @@ static void rna_trackingObject_flushUpdate(Main *UNUSED(bmain),
 
 static void rna_trackingMarker_frame_set(PointerRNA *ptr, int value)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTracking *tracking = &clip->tracking;
   MovieTrackingMarker *marker = (MovieTrackingMarker *)ptr->data;
   MovieTrackingObject *tracking_object;
@@ -594,7 +594,7 @@ static void rna_trackingDopesheet_tagUpdate(Main *UNUSED(bmain),
                                             Scene *UNUSED(scene),
                                             PointerRNA *ptr)
 {
-  MovieClip *clip = (MovieClip *)ptr->id.data;
+  MovieClip *clip = (MovieClip *)ptr->owner_id;
   MovieTrackingDopesheet *dopesheet = &clip->tracking.dopesheet;
 
   dopesheet->ok = 0;
diff --git a/source/blender/makesrna/intern/rna_ui.c b/source/blender/makesrna/intern/rna_ui.c
index 097e62a1fea..a7f0c8ff19c 100644
--- a/source/blender/makesrna/intern/rna_ui.c
+++ b/source/blender/makesrna/intern/rna_ui.c
@@ -1410,7 +1410,7 @@ static void rna_def_panel(BlenderRNA *brna)
 
   prop = RNA_def_property(srna, "use_pin", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flag", PNL_PIN);
-  RNA_def_property_ui_text(prop, "Pin", "");
+  RNA_def_property_ui_text(prop, "Pin", "Show the panel on all tabs");
   /* XXX, should only tag region for redraw */
   RNA_def_property_update(prop, NC_WINDOW, NULL);
 
diff --git a/source/blender/makesrna/intern/rna_userdef.c b/source/blender/makesrna/intern/rna_userdef.c
index a449ed850e9..64ed284eb1f 100644
--- a/source/blender/makesrna/intern/rna_userdef.c
+++ b/source/blender/makesrna/intern/rna_userdef.c
@@ -912,6 +912,13 @@ static bool rna_UserDef_studiolight_is_user_defined_get(PointerRNA *ptr)
   return (sl->flag & STUDIOLIGHT_USER_DEFINED) != 0;
 }
 
+/* StudioLight.is_user_defined */
+static bool rna_UserDef_studiolight_has_specular_highlight_pass_get(PointerRNA *ptr)
+{
+  StudioLight *sl = (StudioLight *)ptr->data;
+  return sl->flag & STUDIOLIGHT_SPECULAR_HIGHLIGHT_PASS;
+}
+
 /* StudioLight.type */
 
 static int rna_UserDef_studiolight_type_get(PointerRNA *ptr)
@@ -2297,6 +2304,11 @@ static void rna_def_userdef_theme_space_outliner(BlenderRNA *brna)
   RNA_def_property_ui_text(prop, "Selected Highlight", "");
   RNA_def_property_update(prop, 0, "rna_userdef_theme_update");
 
+  prop = RNA_def_property(srna, "active", PROP_FLOAT, PROP_COLOR_GAMMA);
+  RNA_def_property_array(prop, 3);
+  RNA_def_property_ui_text(prop, "Active Highlight", "");
+  RNA_def_property_update(prop, 0, "rna_userdef_theme_update");
+
   prop = RNA_def_property(srna, "selected_object", PROP_FLOAT, PROP_COLOR_GAMMA);
   RNA_def_property_array(prop, 3);
   RNA_def_property_ui_text(prop, "Selected Objects", "");
@@ -3695,6 +3707,15 @@ static void rna_def_userdef_studiolight(BlenderRNA *brna)
   RNA_def_property_clear_flag(prop, PROP_EDITABLE);
   RNA_def_property_ui_text(prop, "User Defined", "");
 
+  prop = RNA_def_property(srna, "has_specular_highlight_pass", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_funcs(
+      prop, "rna_UserDef_studiolight_has_specular_highlight_pass_get", NULL);
+  RNA_def_property_clear_flag(prop, PROP_EDITABLE);
+  RNA_def_property_ui_text(
+      prop,
+      "Has Specular Highlight",
+      "Studio light image file has separate \"diffuse\" and \"specular\" passes");
+
   prop = RNA_def_property(srna, "type", PROP_ENUM, PROP_NONE);
   RNA_def_property_enum_items(prop, rna_enum_studio_light_type_items);
   RNA_def_property_enum_funcs(prop, "rna_UserDef_studiolight_type_get", NULL, NULL);
@@ -4090,6 +4111,12 @@ static void rna_def_userdef_view(BlenderRNA *brna)
       "Show the frames per second screen refresh rate, while animation is played back");
   RNA_def_property_update(prop, 0, "rna_userdef_update");
 
+  prop = RNA_def_property(srna, "show_addons_enabled_only", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flag", USER_ADDONS_ENABLED_ONLY);
+  RNA_def_property_ui_text(prop,
+                           "Enabled Add-ons Only",
+                           "Only show enabled add-ons. Un-check to see all installed add-ons");
+
   static const EnumPropertyItem factor_display_items[] = {
       {USER_FACTOR_AS_FACTOR, "FACTOR", 0, "Factor", "Display factors as values between 0 and 1"},
       {USER_FACTOR_AS_PERCENTAGE, "PERCENTAGE", 0, "Percentage", "Display factors as percentages"},
diff --git a/source/blender/makesrna/intern/rna_vfont.c b/source/blender/makesrna/intern/rna_vfont.c
index cf61fe07c68..3e96b5816e5 100644
--- a/source/blender/makesrna/intern/rna_vfont.c
+++ b/source/blender/makesrna/intern/rna_vfont.c
@@ -40,7 +40,7 @@
 /* matching fnction in rna_ID.c */
 static int rna_VectorFont_filepath_editable(PointerRNA *ptr, const char **UNUSED(r_info))
 {
-  VFont *vfont = ptr->id.data;
+  VFont *vfont = (VFont *)ptr->owner_id;
   if (BKE_vfont_is_builtin(vfont)) {
     return 0;
   }
@@ -51,7 +51,7 @@ static void rna_VectorFont_reload_update(Main *UNUSED(bmain),
                                          Scene *UNUSED(scene),
                                          PointerRNA *ptr)
 {
-  VFont *vf = ptr->id.data;
+  VFont *vf = (VFont *)ptr->owner_id;
   BKE_vfont_free_data(vf);
 
   /* update */
diff --git a/source/blender/makesrna/intern/rna_wm.c b/source/blender/makesrna/intern/rna_wm.c
index 123ec9634c9..c3ffeaf6f6f 100644
--- a/source/blender/makesrna/intern/rna_wm.c
+++ b/source/blender/makesrna/intern/rna_wm.c
@@ -530,7 +530,7 @@ const EnumPropertyItem rna_enum_wm_report_items[] = {
 
 static wmOperator *rna_OperatorProperties_find_operator(PointerRNA *ptr)
 {
-  wmWindowManager *wm = ptr->id.data;
+  wmWindowManager *wm = (wmWindowManager *)ptr->owner_id;
 
   if (wm) {
     IDProperty *properties = (IDProperty *)ptr->data;
@@ -668,7 +668,7 @@ static PointerRNA rna_PopupMenu_layout_get(PointerRNA *ptr)
   uiLayout *layout = UI_popup_menu_layout(pup);
 
   PointerRNA rptr;
-  RNA_pointer_create(ptr->id.data, &RNA_UILayout, layout, &rptr);
+  RNA_pointer_create(ptr->owner_id, &RNA_UILayout, layout, &rptr);
 
   return rptr;
 }
@@ -679,7 +679,7 @@ static PointerRNA rna_PopoverMenu_layout_get(PointerRNA *ptr)
   uiLayout *layout = UI_popover_layout(pup);
 
   PointerRNA rptr;
-  RNA_pointer_create(ptr->id.data, &RNA_UILayout, layout, &rptr);
+  RNA_pointer_create(ptr->owner_id, &RNA_UILayout, layout, &rptr);
 
   return rptr;
 }
@@ -690,7 +690,7 @@ static PointerRNA rna_PieMenu_layout_get(PointerRNA *ptr)
   uiLayout *layout = UI_pie_menu_layout(pie);
 
   PointerRNA rptr;
-  RNA_pointer_create(ptr->id.data, &RNA_UILayout, layout, &rptr);
+  RNA_pointer_create(ptr->owner_id, &RNA_UILayout, layout, &rptr);
 
   return rptr;
 }
@@ -804,7 +804,7 @@ static void rna_Window_screen_set(PointerRNA *ptr,
 
 static bool rna_Window_screen_assign_poll(PointerRNA *UNUSED(ptr), PointerRNA value)
 {
-  bScreen *screen = value.id.data;
+  bScreen *screen = (bScreen *)value.owner_id;
   return !screen->temp;
 }
 
diff --git a/source/blender/makesrna/intern/rna_wm_gizmo.c b/source/blender/makesrna/intern/rna_wm_gizmo.c
index 534e3042768..ba26929d808 100644
--- a/source/blender/makesrna/intern/rna_wm_gizmo.c
+++ b/source/blender/makesrna/intern/rna_wm_gizmo.c
@@ -252,7 +252,7 @@ static void rna_Gizmo_update_redraw(Main *UNUSED(bmain), Scene *UNUSED(scene), P
 static wmGizmo *rna_GizmoProperties_find_operator(PointerRNA *ptr)
 {
 #  if 0
-  wmWindowManager *wm = ptr->id.data;
+  wmWindowManager *wm = (wmWindowManager *)ptr->owner_id;
 #  endif
 
   /* We could try workaruond this lookup, but not trivial. */
diff --git a/source/blender/makesrna/intern/rna_workspace.c b/source/blender/makesrna/intern/rna_workspace.c
index 6349b5206d6..47138653af1 100644
--- a/source/blender/makesrna/intern/rna_workspace.c
+++ b/source/blender/makesrna/intern/rna_workspace.c
@@ -58,7 +58,7 @@ static void rna_window_update_all(Main *UNUSED(bmain),
 
 void rna_workspace_screens_begin(CollectionPropertyIterator *iter, PointerRNA *ptr)
 {
-  WorkSpace *workspace = ptr->id.data;
+  WorkSpace *workspace = (WorkSpace *)ptr->owner_id;
   rna_iterator_listbase_begin(iter, BKE_workspace_layouts_get(workspace), NULL);
 }
 
diff --git a/source/blender/makesrna/intern/rna_world.c b/source/blender/makesrna/intern/rna_world.c
index 9b5611f30d7..07db2755523 100644
--- a/source/blender/makesrna/intern/rna_world.c
+++ b/source/blender/makesrna/intern/rna_world.c
@@ -48,17 +48,17 @@
 
 static PointerRNA rna_World_lighting_get(PointerRNA *ptr)
 {
-  return rna_pointer_inherit_refine(ptr, &RNA_WorldLighting, ptr->id.data);
+  return rna_pointer_inherit_refine(ptr, &RNA_WorldLighting, ptr->owner_id);
 }
 
 static PointerRNA rna_World_mist_get(PointerRNA *ptr)
 {
-  return rna_pointer_inherit_refine(ptr, &RNA_WorldMistSettings, ptr->id.data);
+  return rna_pointer_inherit_refine(ptr, &RNA_WorldMistSettings, ptr->owner_id);
 }
 
 static void rna_World_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  World *wo = ptr->id.data;
+  World *wo = (World *)ptr->owner_id;
 
   DEG_id_tag_update(&wo->id, 0);
   WM_main_add_notifier(NC_WORLD | ND_WORLD, wo);
@@ -67,7 +67,7 @@ static void rna_World_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerR
 #  if 0
 static void rna_World_draw_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  World *wo = ptr->id.data;
+  World *wo = (World *)ptr->owner_id;
 
   DEG_id_tag_update(&wo->id, 0);
   WM_main_add_notifier(NC_WORLD | ND_WORLD_DRAW, wo);
@@ -76,7 +76,7 @@ static void rna_World_draw_update(Main *UNUSED(bmain), Scene *UNUSED(scene), Poi
 
 static void rna_World_draw_update(Main *UNUSED(bmain), Scene *UNUSED(scene), PointerRNA *ptr)
 {
-  World *wo = ptr->id.data;
+  World *wo = (World *)ptr->owner_id;
 
   DEG_id_tag_update(&wo->id, 0);
   WM_main_add_notifier(NC_WORLD | ND_WORLD_DRAW, wo);
diff --git a/source/blender/modifiers/CMakeLists.txt b/source/blender/modifiers/CMakeLists.txt
index 3d8f96d1abc..8ff66de18b0 100644
--- a/source/blender/modifiers/CMakeLists.txt
+++ b/source/blender/modifiers/CMakeLists.txt
@@ -143,6 +143,10 @@ if(WITH_MOD_OCEANSIM)
   add_definitions(-DWITH_OCEANSIM)
 endif()
 
+if(WITH_OPENSUBDIV)
+  add_definitions(-DWITH_OPENSUBDIV)
+endif()
+
 if(WITH_BULLET)
   list(APPEND LIB
     extern_bullet
diff --git a/source/blender/modifiers/intern/MOD_armature.c b/source/blender/modifiers/intern/MOD_armature.c
index 7ae5fda7111..9f3802a4fa1 100644
--- a/source/blender/modifiers/intern/MOD_armature.c
+++ b/source/blender/modifiers/intern/MOD_armature.c
@@ -82,7 +82,7 @@ static bool isDisabled(const struct Scene *UNUSED(scene),
   /* The object type check is only needed here in case we have a placeholder
    * object assigned (because the library containing the armature is missing).
    *
-   * In other cases it should be impossible.
+   * In other cases it should be impossible to have a type mismatch.
    */
   return !amd->object || amd->object->type != OB_ARMATURE;
 }
diff --git a/source/blender/modifiers/intern/MOD_array.c b/source/blender/modifiers/intern/MOD_array.c
index fd6f58a21d4..9e368d2c050 100644
--- a/source/blender/modifiers/intern/MOD_array.c
+++ b/source/blender/modifiers/intern/MOD_array.c
@@ -395,7 +395,7 @@ static Mesh *arrayModifier_doArray(ArrayModifierData *amd,
   count = amd->count;
 
   Object *start_cap_ob = amd->start_cap;
-  if (start_cap_ob && start_cap_ob != ctx->object && start_cap_ob->type == OB_MESH) {
+  if (start_cap_ob && start_cap_ob != ctx->object) {
     vgroup_start_cap_remap = BKE_object_defgroup_index_map_create(
         start_cap_ob, ctx->object, &vgroup_start_cap_remap_len);
 
@@ -408,7 +408,7 @@ static Mesh *arrayModifier_doArray(ArrayModifierData *amd,
     }
   }
   Object *end_cap_ob = amd->end_cap;
-  if (end_cap_ob && end_cap_ob != ctx->object && end_cap_ob->type == OB_MESH) {
+  if (end_cap_ob && end_cap_ob != ctx->object) {
     vgroup_end_cap_remap = BKE_object_defgroup_index_map_create(
         end_cap_ob, ctx->object, &vgroup_end_cap_remap_len);
 
@@ -465,13 +465,10 @@ static Mesh *arrayModifier_doArray(ArrayModifierData *amd,
 
   if (amd->fit_type == MOD_ARR_FITCURVE && amd->curve_ob != NULL) {
     Object *curve_ob = amd->curve_ob;
-    Curve *cu = curve_ob->data;
-    if (cu) {
-      CurveCache *curve_cache = curve_ob->runtime.curve_cache;
-      if (curve_cache != NULL && curve_cache->path != NULL) {
-        float scale_fac = mat4_to_scale(curve_ob->obmat);
-        length = scale_fac * curve_cache->path->totdist;
-      }
+    CurveCache *curve_cache = curve_ob->runtime.curve_cache;
+    if (curve_cache != NULL && curve_cache->path != NULL) {
+      float scale_fac = mat4_to_scale(curve_ob->obmat);
+      length = scale_fac * curve_cache->path->totdist;
     }
   }
 
@@ -769,6 +766,31 @@ static Mesh *applyModifier(ModifierData *md, const ModifierEvalContext *ctx, Mes
   return arrayModifier_doArray(amd, ctx, mesh);
 }
 
+static bool isDisabled(const struct Scene *UNUSED(scene),
+                       ModifierData *md,
+                       bool UNUSED(useRenderParams))
+{
+  ArrayModifierData *amd = (ArrayModifierData *)md;
+
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the curve/mesh is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+
+  if (amd->curve_ob && amd->curve_ob->type != OB_CURVE) {
+    return true;
+  }
+  else if (amd->start_cap && amd->start_cap->type != OB_MESH) {
+    return true;
+  }
+  else if (amd->end_cap && amd->end_cap->type != OB_MESH) {
+    return true;
+  }
+
+  return false;
+}
+
 ModifierTypeInfo modifierType_Array = {
     /* name */ "Array",
     /* structName */ "ArrayModifierData",
@@ -789,7 +811,7 @@ ModifierTypeInfo modifierType_Array = {
     /* initData */ initData,
     /* requiredDataMask */ NULL,
     /* freeData */ NULL,
-    /* isDisabled */ NULL,
+    /* isDisabled */ isDisabled,
     /* updateDepsgraph */ updateDepsgraph,
     /* dependsOnTime */ NULL,
     /* dependsOnNormals */ NULL,
diff --git a/source/blender/modifiers/intern/MOD_boolean.c b/source/blender/modifiers/intern/MOD_boolean.c
index dc4898c83ff..e55eeddbfa5 100644
--- a/source/blender/modifiers/intern/MOD_boolean.c
+++ b/source/blender/modifiers/intern/MOD_boolean.c
@@ -71,7 +71,12 @@ static bool isDisabled(const struct Scene *UNUSED(scene),
 {
   BooleanModifierData *bmd = (BooleanModifierData *)md;
 
-  return !bmd->object;
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the mesh is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  return !bmd->object || bmd->object->type != OB_MESH;
 }
 
 static void foreachObjectLink(ModifierData *md, Object *ob, ObjectWalkFunc walk, void *userData)
diff --git a/source/blender/modifiers/intern/MOD_cloth.c b/source/blender/modifiers/intern/MOD_cloth.c
index 1a6d172d2f9..cc91e345c8f 100644
--- a/source/blender/modifiers/intern/MOD_cloth.c
+++ b/source/blender/modifiers/intern/MOD_cloth.c
@@ -114,7 +114,7 @@ static void deformVerts(ModifierData *md,
     }
   }
 
-  BKE_mesh_apply_vert_coords(mesh_src, vertexCos);
+  BKE_mesh_vert_coords_apply(mesh_src, vertexCos);
 
   clothModifier_do(clmd, ctx->depsgraph, scene, ctx->object, mesh_src, vertexCos);
 
diff --git a/source/blender/modifiers/intern/MOD_collision.c b/source/blender/modifiers/intern/MOD_collision.c
index 1bbc25643a1..3d43c6de88e 100644
--- a/source/blender/modifiers/intern/MOD_collision.c
+++ b/source/blender/modifiers/intern/MOD_collision.c
@@ -120,7 +120,7 @@ static void deformVerts(ModifierData *md,
     float current_time = 0;
     unsigned int mvert_num = 0;
 
-    BKE_mesh_apply_vert_coords(mesh_src, vertexCos);
+    BKE_mesh_vert_coords_apply(mesh_src, vertexCos);
     BKE_mesh_calc_normals(mesh_src);
 
     current_time = DEG_get_ctime(ctx->depsgraph);
diff --git a/source/blender/modifiers/intern/MOD_correctivesmooth.c b/source/blender/modifiers/intern/MOD_correctivesmooth.c
index be1580f0d70..a234f468e45 100644
--- a/source/blender/modifiers/intern/MOD_correctivesmooth.c
+++ b/source/blender/modifiers/intern/MOD_correctivesmooth.c
@@ -635,12 +635,12 @@ static void correctivesmooth_modifier_do(ModifierData *md,
     if (csmd->rest_source == MOD_CORRECTIVESMOOTH_RESTSOURCE_BIND) {
       /* caller needs to do sanity check here */
       csmd->bind_coords_num = numVerts;
-      rest_coords = (const float(*)[3])csmd->bind_coords;
+      rest_coords = csmd->bind_coords;
     }
     else {
       int me_numVerts;
-      rest_coords = (const float(*)[3])((em) ? BKE_editmesh_vertexCos_get_orco(em, &me_numVerts) :
-                                               BKE_mesh_vertexCos_get(ob->data, &me_numVerts));
+      rest_coords = em ? BKE_editmesh_vert_coords_alloc_orco(em, &me_numVerts) :
+                         BKE_mesh_vert_coords_alloc(ob->data, &me_numVerts);
 
       BLI_assert((unsigned int)me_numVerts == numVerts);
       is_rest_coords_alloc = true;
diff --git a/source/blender/modifiers/intern/MOD_curve.c b/source/blender/modifiers/intern/MOD_curve.c
index 7a07d9b28eb..f9137572d6f 100644
--- a/source/blender/modifiers/intern/MOD_curve.c
+++ b/source/blender/modifiers/intern/MOD_curve.c
@@ -66,7 +66,12 @@ static bool isDisabled(const Scene *UNUSED(scene), ModifierData *md, bool UNUSED
 {
   CurveModifierData *cmd = (CurveModifierData *)md;
 
-  return !cmd->object;
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the curve is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  return !cmd->object || cmd->object->type != OB_CURVE;
 }
 
 static void foreachObjectLink(ModifierData *md, Object *ob, ObjectWalkFunc walk, void *userData)
diff --git a/source/blender/modifiers/intern/MOD_datatransfer.c b/source/blender/modifiers/intern/MOD_datatransfer.c
index c994a680fb0..bb032f9725c 100644
--- a/source/blender/modifiers/intern/MOD_datatransfer.c
+++ b/source/blender/modifiers/intern/MOD_datatransfer.c
@@ -141,9 +141,14 @@ static bool isDisabled(const struct Scene *UNUSED(scene),
                        ModifierData *md,
                        bool UNUSED(useRenderParams))
 {
-  DataTransferModifierData *dtmd = (DataTransferModifierData *)md;
   /* If no source object, bypass. */
-  return (dtmd->ob_source == NULL);
+  DataTransferModifierData *dtmd = (DataTransferModifierData *)md;
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the mesh is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  return !dtmd->ob_source || dtmd->ob_source->type != OB_MESH;
 }
 
 #define HIGH_POLY_WARNING 10000
diff --git a/source/blender/modifiers/intern/MOD_fluidsim_util.c b/source/blender/modifiers/intern/MOD_fluidsim_util.c
index 3083d89d555..587aa108fd1 100644
--- a/source/blender/modifiers/intern/MOD_fluidsim_util.c
+++ b/source/blender/modifiers/intern/MOD_fluidsim_util.c
@@ -290,7 +290,7 @@ static Mesh *fluidsim_read_obj(const char *filename, const MPoly *mp_example)
   gzclose(gzf);
 
   BKE_mesh_calc_edges(mesh, false, false);
-  BKE_mesh_apply_vert_normals(mesh, (short(*)[3])normals);
+  BKE_mesh_vert_normals_apply(mesh, (short(*)[3])normals);
   MEM_freeN(normals);
 
   // CDDM_calc_normals(result);
diff --git a/source/blender/modifiers/intern/MOD_fluidsim_util.h b/source/blender/modifiers/intern/MOD_fluidsim_util.h
index 0d2be3e7074..da70568353e 100644
--- a/source/blender/modifiers/intern/MOD_fluidsim_util.h
+++ b/source/blender/modifiers/intern/MOD_fluidsim_util.h
@@ -27,8 +27,6 @@
 struct FluidsimModifierData;
 struct Mesh;
 struct ModifierEvalContext;
-struct Object;
-struct Scene;
 
 /* new fluid-modifier interface */
 void fluidsim_init(struct FluidsimModifierData *fluidmd);
diff --git a/source/blender/modifiers/intern/MOD_lattice.c b/source/blender/modifiers/intern/MOD_lattice.c
index b639e874a88..a73e96da975 100644
--- a/source/blender/modifiers/intern/MOD_lattice.c
+++ b/source/blender/modifiers/intern/MOD_lattice.c
@@ -64,7 +64,12 @@ static bool isDisabled(const struct Scene *UNUSED(scene),
 {
   LatticeModifierData *lmd = (LatticeModifierData *)md;
 
-  return !lmd->object;
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the lattice is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  return !lmd->object || lmd->object->type != OB_LATTICE;
 }
 
 static void foreachObjectLink(ModifierData *md, Object *ob, ObjectWalkFunc walk, void *userData)
diff --git a/source/blender/modifiers/intern/MOD_mask.c b/source/blender/modifiers/intern/MOD_mask.c
index 3417aaeeb5c..df8f68862e6 100644
--- a/source/blender/modifiers/intern/MOD_mask.c
+++ b/source/blender/modifiers/intern/MOD_mask.c
@@ -138,10 +138,10 @@ static Mesh *applyModifier(ModifierData *md, const ModifierEvalContext *ctx, Mes
       return mesh;
     }
 
-    /* Determine whether each vertexgroup is associated with a selected bone or not:
-     * - Each cell is a boolean saying whether bone corresponding to the ith group is selected.
+    /* Determine whether each vertex-group is associated with a selected bone or not:
+     * - Each cell is a boolean saying whether bone corresponding to the i'th group selected.
      * - Groups that don't match a bone are treated as not existing
-     *   (along with the corresponding ungrouped verts).
+     *   (along with the corresponding un-grouped verts).
      */
     bone_select_array = MEM_malloc_arrayN((size_t)defbase_tot, sizeof(char), "mask array");
 
@@ -344,6 +344,20 @@ static Mesh *applyModifier(ModifierData *md, const ModifierEvalContext *ctx, Mes
   return result;
 }
 
+static bool isDisabled(const struct Scene *UNUSED(scene),
+                       ModifierData *md,
+                       bool UNUSED(useRenderParams))
+{
+  MaskModifierData *mmd = (MaskModifierData *)md;
+
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the armature is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  return mmd->ob_arm && mmd->ob_arm->type != OB_ARMATURE;
+}
+
 ModifierTypeInfo modifierType_Mask = {
     /* name */ "Mask",
     /* structName */ "MaskModifierData",
@@ -363,7 +377,7 @@ ModifierTypeInfo modifierType_Mask = {
     /* initData */ NULL,
     /* requiredDataMask */ requiredDataMask,
     /* freeData */ NULL,
-    /* isDisabled */ NULL,
+    /* isDisabled */ isDisabled,
     /* updateDepsgraph */ updateDepsgraph,
     /* dependsOnTime */ NULL,
     /* dependsOnNormals */ NULL,
diff --git a/source/blender/modifiers/intern/MOD_meshdeform.c b/source/blender/modifiers/intern/MOD_meshdeform.c
index da261b4f835..408e38f43ab 100644
--- a/source/blender/modifiers/intern/MOD_meshdeform.c
+++ b/source/blender/modifiers/intern/MOD_meshdeform.c
@@ -139,7 +139,12 @@ static bool isDisabled(const struct Scene *UNUSED(scene),
 {
   MeshDeformModifierData *mmd = (MeshDeformModifierData *)md;
 
-  return !mmd->object;
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the mesh is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  return !mmd->object || mmd->object->type != OB_MESH;
 }
 
 static void foreachObjectLink(ModifierData *md, Object *ob, ObjectWalkFunc walk, void *userData)
@@ -401,7 +406,7 @@ static void meshdeformModifier_do(ModifierData *md,
   }
 
   /* setup deformation data */
-  cagecos = BKE_mesh_vertexCos_get(cagemesh, NULL);
+  cagecos = BKE_mesh_vert_coords_alloc(cagemesh, NULL);
   bindcagecos = (float(*)[3])mmd->bindcagecos;
 
   /* We allocate 1 element extra to make it possible to
diff --git a/source/blender/modifiers/intern/MOD_meshsequencecache.c b/source/blender/modifiers/intern/MOD_meshsequencecache.c
index 760830ffb24..0f57b759e38 100644
--- a/source/blender/modifiers/intern/MOD_meshsequencecache.c
+++ b/source/blender/modifiers/intern/MOD_meshsequencecache.c
@@ -114,10 +114,20 @@ static Mesh *applyModifier(ModifierData *md, const ModifierEvalContext *ctx, Mes
     }
   }
 
+  /* If this invocation is for the ORCO mesh, and the mesh in Alembic hasn't changed topology, we
+   * must return the mesh as-is instead of deforming it. */
+  if (ctx->flag & MOD_APPLY_ORCO &&
+      !ABC_mesh_topology_changed(mcmd->reader, ctx->object, mesh, time, &err_str)) {
+    return mesh;
+  }
+
   if (me != NULL) {
     MVert *mvert = mesh->mvert;
     MEdge *medge = mesh->medge;
     MPoly *mpoly = mesh->mpoly;
+
+    /* TODO(sybren+bastien): possibly check relevant custom data layers (UV/color depending on
+     * flags) and duplicate those too. */
     if ((me->mvert == mvert) || (me->medge == medge) || (me->mpoly == mpoly)) {
       /* We need to duplicate data here, otherwise we'll modify org mesh, see T51701. */
       BKE_id_copy_ex(NULL,
diff --git a/source/blender/modifiers/intern/MOD_multires.c b/source/blender/modifiers/intern/MOD_multires.c
index c64d9be1158..dd7c001931c 100644
--- a/source/blender/modifiers/intern/MOD_multires.c
+++ b/source/blender/modifiers/intern/MOD_multires.c
@@ -168,6 +168,10 @@ static Mesh *multires_as_ccg(MultiresModifierData *mmd,
 static Mesh *applyModifier(ModifierData *md, const ModifierEvalContext *ctx, Mesh *mesh)
 {
   Mesh *result = mesh;
+#if !defined(WITH_OPENSUBDIV)
+  modifier_setError(md, "Disabled, built without OpenSubdiv");
+  return result;
+#endif
   MultiresModifierData *mmd = (MultiresModifierData *)md;
   SubdivSettings subdiv_settings;
   BKE_multires_subdiv_settings_init(&subdiv_settings, mmd);
diff --git a/source/blender/modifiers/intern/MOD_particleinstance.c b/source/blender/modifiers/intern/MOD_particleinstance.c
index 71913378277..49bb8691764 100644
--- a/source/blender/modifiers/intern/MOD_particleinstance.c
+++ b/source/blender/modifiers/intern/MOD_particleinstance.c
@@ -80,7 +80,12 @@ static bool isDisabled(const struct Scene *scene, ModifierData *md, bool useRend
   ParticleSystem *psys;
   ModifierData *ob_md;
 
-  if (!pimd->ob) {
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the mesh is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  if (!pimd->ob || pimd->ob->type != OB_MESH) {
     return true;
   }
 
diff --git a/source/blender/modifiers/intern/MOD_particlesystem.c b/source/blender/modifiers/intern/MOD_particlesystem.c
index 021e61bd46d..a7c7c207cd6 100644
--- a/source/blender/modifiers/intern/MOD_particlesystem.c
+++ b/source/blender/modifiers/intern/MOD_particlesystem.c
@@ -152,7 +152,7 @@ static void deformVerts(ModifierData *md,
 
   /* make new mesh */
   psmd->mesh_final = BKE_mesh_copy_for_eval(mesh_src, false);
-  BKE_mesh_apply_vert_coords(psmd->mesh_final, vertexCos);
+  BKE_mesh_vert_coords_apply(psmd->mesh_final, vertexCos);
   BKE_mesh_calc_normals(psmd->mesh_final);
 
   BKE_mesh_tessface_ensure(psmd->mesh_final);
diff --git a/source/blender/modifiers/intern/MOD_remesh.c b/source/blender/modifiers/intern/MOD_remesh.c
index 23da1ec2754..631401d9d9e 100644
--- a/source/blender/modifiers/intern/MOD_remesh.c
+++ b/source/blender/modifiers/intern/MOD_remesh.c
@@ -105,7 +105,7 @@ static void dualcon_add_vert(void *output_v, const float co[3])
   DualConOutput *output = output_v;
   Mesh *mesh = output->mesh;
 
-  assert(output->curvert < mesh->totvert);
+  BLI_assert(output->curvert < mesh->totvert);
 
   copy_v3_v3(mesh->mvert[output->curvert].co, co);
   output->curvert++;
@@ -119,7 +119,7 @@ static void dualcon_add_quad(void *output_v, const int vert_indices[4])
   MPoly *cur_poly;
   int i;
 
-  assert(output->curface < mesh->totpoly);
+  BLI_assert(output->curface < mesh->totpoly);
 
   mloop = mesh->mloop;
   cur_poly = &mesh->mpoly[output->curface];
diff --git a/source/blender/modifiers/intern/MOD_shrinkwrap.c b/source/blender/modifiers/intern/MOD_shrinkwrap.c
index 07182d82fdc..4bca48852a2 100644
--- a/source/blender/modifiers/intern/MOD_shrinkwrap.c
+++ b/source/blender/modifiers/intern/MOD_shrinkwrap.c
@@ -75,7 +75,19 @@ static bool isDisabled(const struct Scene *UNUSED(scene),
                        bool UNUSED(useRenderParams))
 {
   ShrinkwrapModifierData *smd = (ShrinkwrapModifierData *)md;
-  return !smd->target;
+
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the mesh is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  if (!smd->target || smd->target->type != OB_MESH) {
+    return true;
+  }
+  else if (smd->auxTarget && smd->auxTarget->type != OB_MESH) {
+    return true;
+  }
+  return false;
 }
 
 static void foreachObjectLink(ModifierData *md, Object *ob, ObjectWalkFunc walk, void *userData)
diff --git a/source/blender/modifiers/intern/MOD_subsurf.c b/source/blender/modifiers/intern/MOD_subsurf.c
index 0b1249e263c..08a884fa879 100644
--- a/source/blender/modifiers/intern/MOD_subsurf.c
+++ b/source/blender/modifiers/intern/MOD_subsurf.c
@@ -208,6 +208,10 @@ static SubsurfRuntimeData *subsurf_ensure_runtime(SubsurfModifierData *smd)
 static Mesh *applyModifier(ModifierData *md, const ModifierEvalContext *ctx, Mesh *mesh)
 {
   Mesh *result = mesh;
+#if !defined(WITH_OPENSUBDIV)
+  modifier_setError(md, "Disabled, built without OpenSubdiv");
+  return result;
+#endif
   SubsurfModifierData *smd = (SubsurfModifierData *)md;
   SubdivSettings subdiv_settings;
   subdiv_settings_init(&subdiv_settings, smd);
diff --git a/source/blender/modifiers/intern/MOD_surface.c b/source/blender/modifiers/intern/MOD_surface.c
index dcc6371e024..840914aa313 100644
--- a/source/blender/modifiers/intern/MOD_surface.c
+++ b/source/blender/modifiers/intern/MOD_surface.c
@@ -130,7 +130,7 @@ static void deformVerts(ModifierData *md,
     float *vec;
     MVert *x, *v;
 
-    BKE_mesh_apply_vert_coords(surmd->mesh, vertexCos);
+    BKE_mesh_vert_coords_apply(surmd->mesh, vertexCos);
     BKE_mesh_calc_normals(surmd->mesh);
 
     numverts = surmd->mesh->totvert;
diff --git a/source/blender/modifiers/intern/MOD_surfacedeform.c b/source/blender/modifiers/intern/MOD_surfacedeform.c
index c428325e42b..b34688d27d6 100644
--- a/source/blender/modifiers/intern/MOD_surfacedeform.c
+++ b/source/blender/modifiers/intern/MOD_surfacedeform.c
@@ -1279,7 +1279,13 @@ static bool isDisabled(const Scene *UNUSED(scene), ModifierData *md, bool UNUSED
 {
   SurfaceDeformModifierData *smd = (SurfaceDeformModifierData *)md;
 
-  return smd->target == NULL && !(smd->verts != NULL && !(smd->flags & MOD_SDEF_BIND));
+  /* The object type check is only needed here in case we have a placeholder
+   * object assigned (because the library containing the mesh is missing).
+   *
+   * In other cases it should be impossible to have a type mismatch.
+   */
+  return (smd->target == NULL || smd->target->type != OB_MESH) &&
+         !(smd->verts != NULL && !(smd->flags & MOD_SDEF_BIND));
 }
 
 ModifierTypeInfo modifierType_SurfaceDeform = {
diff --git a/source/blender/modifiers/intern/MOD_util.c b/source/blender/modifiers/intern/MOD_util.c
index 452c9ff945d..71163388707 100644
--- a/source/blender/modifiers/intern/MOD_util.c
+++ b/source/blender/modifiers/intern/MOD_util.c
@@ -198,7 +198,7 @@ Mesh *MOD_deform_mesh_eval_get(Object *ob,
     /* TODO(sybren): after modifier conversion of DM to Mesh is done, check whether
      * we really need vertexCos here. */
     if (vertexCos) {
-      BKE_mesh_apply_vert_coords(mesh, vertexCos);
+      BKE_mesh_vert_coords_apply(mesh, vertexCos);
       mesh->runtime.cd_dirty_vert |= CD_MASK_NORMAL;
     }
 
diff --git a/source/blender/modifiers/intern/MOD_util.h b/source/blender/modifiers/intern/MOD_util.h
index ba3ca44c026..e1991de3bb8 100644
--- a/source/blender/modifiers/intern/MOD_util.h
+++ b/source/blender/modifiers/intern/MOD_util.h
@@ -31,8 +31,6 @@ struct Mesh;
 struct ModifierData;
 struct ModifierEvalContext;
 struct Object;
-struct Scene;
-struct Tex;
 
 void MOD_init_texture(struct MappingInfoModifierData *dmd, const struct ModifierEvalContext *ctx);
 void MOD_get_texture_coords(struct MappingInfoModifierData *dmd,
diff --git a/source/blender/modifiers/intern/MOD_uvproject.c b/source/blender/modifiers/intern/MOD_uvproject.c
index f9eb92cd132..5f1eae0297a 100644
--- a/source/blender/modifiers/intern/MOD_uvproject.c
+++ b/source/blender/modifiers/intern/MOD_uvproject.c
@@ -204,7 +204,7 @@ static Mesh *uvprojectModifier_do(UVProjectModifierData *umd,
   mloop_uv = CustomData_duplicate_referenced_layer_named(
       &mesh->ldata, CD_MLOOPUV, uvname, numLoops);
 
-  coords = BKE_mesh_vertexCos_get(mesh, &numVerts);
+  coords = BKE_mesh_vert_coords_alloc(mesh, &numVerts);
 
   /* convert coords to world space */
   for (i = 0, co = coords; i < numVerts; ++i, ++co) {
diff --git a/source/blender/modifiers/intern/MOD_weightvgproximity.c b/source/blender/modifiers/intern/MOD_weightvgproximity.c
index 86d1b310d0c..95b15b4a924 100644
--- a/source/blender/modifiers/intern/MOD_weightvgproximity.c
+++ b/source/blender/modifiers/intern/MOD_weightvgproximity.c
@@ -492,7 +492,7 @@ static Mesh *applyModifier(ModifierData *md, const ModifierEvalContext *ctx, Mes
 
   /* Get our vertex coordinates. */
   if (numIdx != numVerts) {
-    float(*tv_cos)[3] = BKE_mesh_vertexCos_get(mesh, NULL);
+    float(*tv_cos)[3] = BKE_mesh_vert_coords_alloc(mesh, NULL);
     v_cos = MEM_malloc_arrayN(numIdx, sizeof(float[3]), "WeightVGProximity Modifier, v_cos");
     for (i = 0; i < numIdx; i++) {
       copy_v3_v3(v_cos[i], tv_cos[indices[i]]);
@@ -500,7 +500,7 @@ static Mesh *applyModifier(ModifierData *md, const ModifierEvalContext *ctx, Mes
     MEM_freeN(tv_cos);
   }
   else {
-    v_cos = BKE_mesh_vertexCos_get(mesh, NULL);
+    v_cos = BKE_mesh_vert_coords_alloc(mesh, NULL);
   }
 
   /* Compute wanted distances. */
diff --git a/source/blender/nodes/CMakeLists.txt b/source/blender/nodes/CMakeLists.txt
index 284eaa8b70b..5cf5e1ac6d3 100644
--- a/source/blender/nodes/CMakeLists.txt
+++ b/source/blender/nodes/CMakeLists.txt
@@ -200,14 +200,16 @@ set(SRC
   shader/nodes/node_shader_tex_sky.c
   shader/nodes/node_shader_tex_voronoi.c
   shader/nodes/node_shader_tex_wave.c
+  shader/nodes/node_shader_tex_white_noise.c
   shader/nodes/node_shader_uvAlongStroke.c
   shader/nodes/node_shader_uvmap.c
   shader/nodes/node_shader_valToRgb.c
   shader/nodes/node_shader_value.c
-  shader/nodes/node_shader_vectMath.c
+  shader/nodes/node_shader_vector_math.c
   shader/nodes/node_shader_vectTransform.c
   shader/nodes/node_shader_vector_displacement.c
   shader/nodes/node_shader_volume_absorption.c
+  shader/nodes/node_shader_volume_info.c
   shader/nodes/node_shader_volume_principled.c
   shader/nodes/node_shader_volume_scatter.c
   shader/nodes/node_shader_wavelength.c
diff --git a/source/blender/nodes/NOD_shader.h b/source/blender/nodes/NOD_shader.h
index 9349a428021..30673e7cd31 100644
--- a/source/blender/nodes/NOD_shader.h
+++ b/source/blender/nodes/NOD_shader.h
@@ -81,6 +81,7 @@ void register_node_type_sh_layer_weight(void);
 void register_node_type_sh_tex_coord(void);
 void register_node_type_sh_particle_info(void);
 void register_node_type_sh_hair_info(void);
+void register_node_type_sh_volume_info(void);
 void register_node_type_sh_script(void);
 void register_node_type_sh_normal_map(void);
 void register_node_type_sh_tangent(void);
@@ -131,6 +132,7 @@ void register_node_type_sh_tex_noise(void);
 void register_node_type_sh_tex_checker(void);
 void register_node_type_sh_bump(void);
 void register_node_type_sh_tex_ies(void);
+void register_node_type_sh_tex_white_noise(void);
 
 void register_node_type_sh_custom_group(bNodeType *ntype);
 
diff --git a/source/blender/nodes/NOD_static_types.h b/source/blender/nodes/NOD_static_types.h
index c72e97642a2..d14511504f0 100644
--- a/source/blender/nodes/NOD_static_types.h
+++ b/source/blender/nodes/NOD_static_types.h
@@ -53,7 +53,7 @@ DefNode(ShaderNode,     SH_NODE_CAMERA,          0,                      "CAMERA
 DefNode(ShaderNode,     SH_NODE_MAP_RANGE,       def_map_range,          "MAP_RANGE",      MapRange,         "Map Range",         ""              )
 DefNode(ShaderNode,     SH_NODE_CLAMP,           0,                      "CLAMP",          Clamp,            "Clamp",             ""              )
 DefNode(ShaderNode,     SH_NODE_MATH,            def_math,               "MATH",           Math,             "Math",              ""              )
-DefNode(ShaderNode,     SH_NODE_VECT_MATH,       def_vector_math,        "VECT_MATH",      VectorMath,       "Vector Math",       ""              )
+DefNode(ShaderNode,     SH_NODE_VECTOR_MATH,     def_vector_math,        "VECT_MATH",      VectorMath,       "Vector Math",       ""              )
 DefNode(ShaderNode,     SH_NODE_SQUEEZE,         0,                      "SQUEEZE",        Squeeze,          "Squeeze Value",     ""              )
 DefNode(ShaderNode,     SH_NODE_INVERT,          0,                      "INVERT",         Invert,           "Invert",            ""              )
 DefNode(ShaderNode,     SH_NODE_SEPRGB,          0,                      "SEPRGB",         SeparateRGB,      "Separate RGB",      ""              )
@@ -96,6 +96,7 @@ DefNode(ShaderNode,     SH_NODE_LIGHT_FALLOFF,      0,                      "LIG
 DefNode(ShaderNode,     SH_NODE_OBJECT_INFO,        0,                      "OBJECT_INFO",        ObjectInfo,       "Object Info",       ""       )
 DefNode(ShaderNode,     SH_NODE_PARTICLE_INFO,      0,                      "PARTICLE_INFO",      ParticleInfo,     "Particle Info",     ""       )
 DefNode(ShaderNode,     SH_NODE_HAIR_INFO,          0,                      "HAIR_INFO",          HairInfo,         "Hair Info",         ""       )
+DefNode(ShaderNode,     SH_NODE_VOLUME_INFO,        0,                      "VOLUME_INFO",        VolumeInfo,       "Volume Info",       ""       )
 DefNode(ShaderNode,     SH_NODE_WIREFRAME,          def_sh_tex_wireframe,   "WIREFRAME",          Wireframe,        "Wireframe",         ""       )
 DefNode(ShaderNode,     SH_NODE_WAVELENGTH,         0,                      "WAVELENGTH",         Wavelength,       "Wavelength",        ""       )
 DefNode(ShaderNode,     SH_NODE_BLACKBODY,          0,                      "BLACKBODY",          Blackbody,        "Blackbody",         ""       )
@@ -127,6 +128,7 @@ DefNode(ShaderNode,     SH_NODE_BEVEL,              def_sh_bevel,           "BEV
 DefNode(ShaderNode,     SH_NODE_DISPLACEMENT,       def_sh_displacement,    "DISPLACEMENT",       Displacement,     "Displacement",      ""       )
 DefNode(ShaderNode,     SH_NODE_VECTOR_DISPLACEMENT,def_sh_vector_displacement,"VECTOR_DISPLACEMENT",VectorDisplacement,"Vector Displacement",""  )
 DefNode(ShaderNode,     SH_NODE_TEX_IES,            def_sh_tex_ies,         "TEX_IES",            TexIES,           "IES Texture",       ""       )
+DefNode(ShaderNode,     SH_NODE_TEX_WHITE_NOISE,    def_sh_tex_white_noise, "TEX_WHITE_NOISE",    TexWhiteNoise,    "White Noise",       ""       )
 
 DefNode(CompositorNode, CMP_NODE_VIEWER,         def_cmp_viewer,         "VIEWER",         Viewer,           "Viewer",            ""              )
 DefNode(CompositorNode, CMP_NODE_RGB,            0,                      "RGB",            RGB,              "RGB",               ""              )
diff --git a/source/blender/nodes/composite/nodes/node_composite_denoise.c b/source/blender/nodes/composite/nodes/node_composite_denoise.c
index e2fdb08816a..bb62869f470 100644
--- a/source/blender/nodes/composite/nodes/node_composite_denoise.c
+++ b/source/blender/nodes/composite/nodes/node_composite_denoise.c
@@ -33,8 +33,8 @@
 
 static bNodeSocketTemplate cmp_node_denoise_in[] = {
     {SOCK_RGBA, 1, N_("Image"), 1.0f, 1.0f, 1.0f, 1.0f, 0.0f, 1.0f},
-    {SOCK_RGBA, 1, N_("Albedo"), 1.0f, 1.0f, 1.0f, 1.0f, 0.0f, 1.0f},
     {SOCK_VECTOR, 0, N_("Normal"), 0.0f, 0.0f, 0.0f, 0.0f, -1.0f, 1.0f},
+    {SOCK_RGBA, 1, N_("Albedo"), 1.0f, 1.0f, 1.0f, 1.0f, 0.0f, 1.0f},
     {-1, 0, ""}};
 static bNodeSocketTemplate cmp_node_denoise_out[] = {{SOCK_RGBA, 0, N_("Image")}, {-1, 0, ""}};
 
diff --git a/source/blender/nodes/composite/nodes/node_composite_outputFile.c b/source/blender/nodes/composite/nodes/node_composite_outputFile.c
index faf9e20da51..af806a37341 100644
--- a/source/blender/nodes/composite/nodes/node_composite_outputFile.c
+++ b/source/blender/nodes/composite/nodes/node_composite_outputFile.c
@@ -191,7 +191,7 @@ void ntreeCompositOutputFileSetLayer(bNode *node, bNodeSocket *sock, const char
 static void init_output_file(const bContext *C, PointerRNA *ptr)
 {
   Scene *scene = CTX_data_scene(C);
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = ptr->data;
   NodeImageMultiFile *nimf = MEM_callocN(sizeof(NodeImageMultiFile), "node image multi file");
   ImageFormatData *format = NULL;
diff --git a/source/blender/nodes/composite/nodes/node_composite_switchview.c b/source/blender/nodes/composite/nodes/node_composite_switchview.c
index 92e06307afb..cecd8204f4a 100644
--- a/source/blender/nodes/composite/nodes/node_composite_switchview.c
+++ b/source/blender/nodes/composite/nodes/node_composite_switchview.c
@@ -116,7 +116,7 @@ static void cmp_node_switch_view_update(bNodeTree *ntree, bNode *node)
 static void init_switch_view(const bContext *C, PointerRNA *ptr)
 {
   Scene *scene = CTX_data_scene(C);
-  bNodeTree *ntree = ptr->id.data;
+  bNodeTree *ntree = (bNodeTree *)ptr->owner_id;
   bNode *node = ptr->data;
   SceneRenderView *srv;
   bNodeSocket *sock;
diff --git a/source/blender/nodes/intern/node_socket.c b/source/blender/nodes/intern/node_socket.c
index 8aeada7bde0..3485bd41660 100644
--- a/source/blender/nodes/intern/node_socket.c
+++ b/source/blender/nodes/intern/node_socket.c
@@ -39,8 +39,6 @@
 
 #include "NOD_socket.h"
 
-struct Main;
-
 struct bNodeSocket *node_add_socket_from_template(struct bNodeTree *ntree,
                                                   struct bNode *node,
                                                   struct bNodeSocketTemplate *stemp,
diff --git a/source/blender/nodes/intern/node_util.c b/source/blender/nodes/intern/node_util.c
index e9a825b5b3f..0d7f19fb67a 100644
--- a/source/blender/nodes/intern/node_util.c
+++ b/source/blender/nodes/intern/node_util.c
@@ -100,7 +100,7 @@ void node_math_label(bNodeTree *UNUSED(ntree), bNode *node, char *label, int max
   BLI_strncpy(label, IFACE_(name), maxlen);
 }
 
-void node_vect_math_label(bNodeTree *UNUSED(ntree), bNode *node, char *label, int maxlen)
+void node_vector_math_label(bNodeTree *UNUSED(ntree), bNode *node, char *label, int maxlen)
 {
   const char *name;
   RNA_enum_name(rna_enum_node_vec_math_items, node->custom1, &name);
diff --git a/source/blender/nodes/intern/node_util.h b/source/blender/nodes/intern/node_util.h
index 7eef70db498..49117b38aba 100644
--- a/source/blender/nodes/intern/node_util.h
+++ b/source/blender/nodes/intern/node_util.h
@@ -71,7 +71,7 @@ extern void *node_initexec_curves(struct bNodeExecContext *context,
 void node_blend_label(struct bNodeTree *ntree, struct bNode *node, char *label, int maxlen);
 void node_image_label(struct bNodeTree *ntree, struct bNode *node, char *label, int maxlen);
 void node_math_label(struct bNodeTree *ntree, struct bNode *node, char *label, int maxlen);
-void node_vect_math_label(struct bNodeTree *ntree, struct bNode *node, char *label, int maxlen);
+void node_vector_math_label(struct bNodeTree *ntree, struct bNode *node, char *label, int maxlen);
 void node_filter_label(struct bNodeTree *ntree, struct bNode *node, char *label, int maxlen);
 
 /*** Link Handling */
diff --git a/source/blender/nodes/shader/node_shader_tree.c b/source/blender/nodes/shader/node_shader_tree.c
index 4891fb323ad..41f13087f67 100644
--- a/source/blender/nodes/shader/node_shader_tree.c
+++ b/source/blender/nodes/shader/node_shader_tree.c
@@ -799,13 +799,15 @@ static void ntree_shader_relink_displacement(bNodeTree *ntree, bNode *output_nod
   nodeRemLink(ntree, displacement_link);
 
   /* Convert displacement vector to bump height. */
-  bNode *dot_node = nodeAddStaticNode(NULL, ntree, SH_NODE_VECT_MATH);
+  bNode *dot_node = nodeAddStaticNode(NULL, ntree, SH_NODE_VECTOR_MATH);
   bNode *geo_node = nodeAddStaticNode(NULL, ntree, SH_NODE_NEW_GEOMETRY);
   bNodeSocket *normal_socket = ntree_shader_node_find_output(geo_node, "Normal");
-  dot_node->custom1 = 3; /* dot product */
+  bNodeSocket *dot_input1 = dot_node->inputs.first;
+  bNodeSocket *dot_input2 = dot_input1->next;
+  dot_node->custom1 = NODE_VECTOR_MATH_DOT_PRODUCT;
 
-  nodeAddLink(ntree, displacement_node, displacement_socket, dot_node, dot_node->inputs.first);
-  nodeAddLink(ntree, geo_node, normal_socket, dot_node, dot_node->inputs.last);
+  nodeAddLink(ntree, displacement_node, displacement_socket, dot_node, dot_input1);
+  nodeAddLink(ntree, geo_node, normal_socket, dot_node, dot_input2);
   displacement_node = dot_node;
   displacement_socket = ntree_shader_node_find_output(dot_node, "Value");
 
diff --git a/source/blender/nodes/shader/node_shader_util.c b/source/blender/nodes/shader/node_shader_util.c
index 2e8f81979a8..65676a5ea91 100644
--- a/source/blender/nodes/shader/node_shader_util.c
+++ b/source/blender/nodes/shader/node_shader_util.c
@@ -283,7 +283,7 @@ void node_shader_gpu_tex_mapping(GPUMaterial *mat,
     GPU_link(mat, "mapping", in[0].link, tmat0, tmat1, tmat2, tmat3, tmin, tmax, &in[0].link);
 
     if (texmap->type == TEXMAP_TYPE_NORMAL) {
-      GPU_link(mat, "texco_norm", in[0].link, &in[0].link);
+      GPU_link(mat, "vector_normalize", in[0].link, &in[0].link);
     }
   }
 }
diff --git a/source/blender/nodes/shader/nodes/node_shader_mapping.c b/source/blender/nodes/shader/nodes/node_shader_mapping.c
index eca0d96f2c8..e58a5d72f28 100644
--- a/source/blender/nodes/shader/nodes/node_shader_mapping.c
+++ b/source/blender/nodes/shader/nodes/node_shader_mapping.c
@@ -115,7 +115,7 @@ static int gpu_shader_mapping(GPUMaterial *mat,
   GPU_stack_link(mat, node, "mapping", in, out, tmat0, tmat1, tmat2, tmat3, tmin, tmax);
 
   if (texmap->type == TEXMAP_TYPE_NORMAL) {
-    GPU_link(mat, "texco_norm", out[0].link, &out[0].link);
+    GPU_link(mat, "vector_normalize", out[0].link, &out[0].link);
   }
 
   return true;
diff --git a/source/blender/nodes/shader/nodes/node_shader_math.c b/source/blender/nodes/shader/nodes/node_shader_math.c
index 29c6d855eae..aaedc4aa1b7 100644
--- a/source/blender/nodes/shader/nodes/node_shader_math.c
+++ b/source/blender/nodes/shader/nodes/node_shader_math.c
@@ -31,272 +31,6 @@ static bNodeSocketTemplate sh_node_math_in[] = {
 
 static bNodeSocketTemplate sh_node_math_out[] = {{SOCK_FLOAT, 0, N_("Value")}, {-1, 0, ""}};
 
-static void node_shader_exec_math(void *UNUSED(data),
-                                  int UNUSED(thread),
-                                  bNode *node,
-                                  bNodeExecData *UNUSED(execdata),
-                                  bNodeStack **in,
-                                  bNodeStack **out)
-{
-  float a, b, r = 0.0f;
-
-  nodestack_get_vec(&a, SOCK_FLOAT, in[0]);
-  nodestack_get_vec(&b, SOCK_FLOAT, in[1]);
-
-  switch (node->custom1) {
-
-    case NODE_MATH_ADD:
-      r = a + b;
-      break;
-    case NODE_MATH_SUB:
-      r = a - b;
-      break;
-    case NODE_MATH_MUL:
-      r = a * b;
-      break;
-    case NODE_MATH_DIVIDE: {
-      if (b == 0) { /* We don't want to divide by zero. */
-        r = 0.0;
-      }
-      else {
-        r = a / b;
-      }
-      break;
-    }
-    case NODE_MATH_SIN: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        r = sinf(a);
-      }
-      else {
-        r = sinf(b);
-      }
-      break;
-    }
-    case NODE_MATH_COS: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        r = cosf(a);
-      }
-      else {
-        r = cosf(b);
-      }
-      break;
-    }
-    case NODE_MATH_TAN: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        r = tanf(a);
-      }
-      else {
-        r = tanf(b);
-      }
-      break;
-    }
-    case NODE_MATH_ASIN: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        /* Can't do the impossible... */
-        if (a <= 1 && a >= -1) {
-          r = asinf(a);
-        }
-        else {
-          r = 0.0;
-        }
-      }
-      else {
-        /* Can't do the impossible... */
-        if (b <= 1 && b >= -1) {
-          r = asinf(b);
-        }
-        else {
-          r = 0.0;
-        }
-      }
-      break;
-    }
-    case NODE_MATH_ACOS: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        /* Can't do the impossible... */
-        if (a <= 1 && a >= -1) {
-          r = acosf(a);
-        }
-        else {
-          r = 0.0;
-        }
-      }
-      else {
-        /* Can't do the impossible... */
-        if (b <= 1 && b >= -1) {
-          r = acosf(b);
-        }
-        else {
-          r = 0.0;
-        }
-      }
-      break;
-    }
-    case NODE_MATH_ATAN: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        r = atan(a);
-      }
-      else {
-        r = atan(b);
-      }
-      break;
-    }
-    case NODE_MATH_POW: {
-      /* Only raise negative numbers by full integers */
-      if (a >= 0) {
-        r = pow(a, b);
-      }
-      else {
-        float y_mod_1 = fabsf(fmodf(b, 1.0f));
-
-        /* if input value is not nearly an integer,
-         * fall back to zero, nicer than straight rounding. */
-        if (y_mod_1 > 0.999f || y_mod_1 < 0.001f) {
-          r = powf(a, floorf(b + 0.5f));
-        }
-        else {
-          r = 0.0f;
-        }
-      }
-
-      break;
-    }
-    case NODE_MATH_LOG: {
-      /* Don't want any imaginary numbers... */
-      if (a > 0 && b > 0) {
-        r = log(a) / log(b);
-      }
-      else {
-        r = 0.0;
-      }
-      break;
-    }
-    case NODE_MATH_MIN: {
-      if (a < b) {
-        r = a;
-      }
-      else {
-        r = b;
-      }
-      break;
-    }
-    case NODE_MATH_MAX: {
-      if (a > b) {
-        r = a;
-      }
-      else {
-        r = b;
-      }
-      break;
-    }
-    case NODE_MATH_ROUND: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        r = (a < 0) ? (int)(a - 0.5f) : (int)(a + 0.5f);
-      }
-      else {
-        r = (b < 0) ? (int)(b - 0.5f) : (int)(b + 0.5f);
-      }
-      break;
-    }
-    case NODE_MATH_LESS: {
-      if (a < b) {
-        r = 1.0f;
-      }
-      else {
-        r = 0.0f;
-      }
-      break;
-    }
-    case NODE_MATH_GREATER: {
-      if (a > b) {
-        r = 1.0f;
-      }
-      else {
-        r = 0.0f;
-      }
-      break;
-    }
-    case NODE_MATH_MOD: {
-      if (b == 0.0f) {
-        r = 0.0f;
-      }
-      else {
-        r = fmod(a, b);
-      }
-      break;
-    }
-    case NODE_MATH_ABS: {
-      r = fabsf(a);
-      break;
-    }
-    case NODE_MATH_ATAN2: {
-      r = atan2(a, b);
-      break;
-    }
-    case NODE_MATH_FLOOR: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        r = floorf(a);
-      }
-      else {
-        r = floorf(b);
-      }
-      break;
-    }
-    case NODE_MATH_CEIL: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        r = ceilf(a);
-      }
-      else {
-        r = ceilf(b);
-      }
-      break;
-    }
-    case NODE_MATH_FRACT: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        r = a - floorf(a);
-      }
-      else {
-        r = b - floorf(b);
-      }
-      break;
-    }
-    case NODE_MATH_SQRT: {
-      /* This one only takes one input, so we've got to choose. */
-      if (in[0]->hasinput || !in[1]->hasinput) {
-        if (a > 0) {
-          r = sqrt(a);
-        }
-        else {
-          r = 0.0;
-        }
-      }
-      else {
-        if (b > 0) {
-          r = sqrt(b);
-        }
-        else {
-          r = 0.0;
-        }
-      }
-      break;
-    }
-  }
-  if (node->custom2 & SHD_MATH_CLAMP) {
-    CLAMP(r, 0.0f, 1.0f);
-  }
-  out[0]->vec[0] = r;
-}
-
 static int gpu_shader_math(GPUMaterial *mat,
                            bNode *node,
                            bNodeExecData *UNUSED(execdata),
@@ -304,68 +38,65 @@ static int gpu_shader_math(GPUMaterial *mat,
                            GPUNodeStack *out)
 {
   static const char *names[] = {
-      "math_add",       "math_subtract",     "math_multiply", "math_divide", "math_sine",
-      "math_cosine",    "math_tangent",      "math_asin",     "math_acos",   "math_atan",
-      "math_pow",       "math_log",          "math_min",      "math_max",    "math_round",
-      "math_less_than", "math_greater_than", "math_modulo",   "math_abs",    "math_atan2",
-      "math_floor",     "math_ceil",         "math_fract",    "math_sqrt",
+      [NODE_MATH_ADD] = "math_add",
+      [NODE_MATH_SUBTRACT] = "math_subtract",
+      [NODE_MATH_MULTIPLY] = "math_multiply",
+      [NODE_MATH_DIVIDE] = "math_divide",
+
+      [NODE_MATH_POWER] = "math_power",
+      [NODE_MATH_LOGARITHM] = "math_logarithm",
+      [NODE_MATH_SQRT] = "math_sqrt",
+      [NODE_MATH_ABSOLUTE] = "math_absolute",
+
+      [NODE_MATH_MINIMUM] = "math_minimum",
+      [NODE_MATH_MAXIMUM] = "math_maximum",
+      [NODE_MATH_LESS_THAN] = "math_less_than",
+      [NODE_MATH_GREATER_THAN] = "math_greater_than",
+
+      [NODE_MATH_ROUND] = "math_round",
+      [NODE_MATH_FLOOR] = "math_floor",
+      [NODE_MATH_CEIL] = "math_ceil",
+      [NODE_MATH_FRACTION] = "math_fraction",
+      [NODE_MATH_MODULO] = "math_modulo",
+
+      [NODE_MATH_SINE] = "math_sine",
+      [NODE_MATH_COSINE] = "math_cosine",
+      [NODE_MATH_TANGENT] = "math_tangent",
+      [NODE_MATH_ARCSINE] = "math_arcsine",
+      [NODE_MATH_ARCCOSINE] = "math_arccosine",
+      [NODE_MATH_ARCTANGENT] = "math_arctangent",
+      [NODE_MATH_ARCTAN2] = "math_arctan2",
   };
 
-  switch (node->custom1) {
-    case NODE_MATH_ADD:
-    case NODE_MATH_SUB:
-    case NODE_MATH_MUL:
-    case NODE_MATH_DIVIDE:
-    case NODE_MATH_POW:
-    case NODE_MATH_LOG:
-    case NODE_MATH_MIN:
-    case NODE_MATH_MAX:
-    case NODE_MATH_LESS:
-    case NODE_MATH_GREATER:
-    case NODE_MATH_MOD:
-    case NODE_MATH_ATAN2:
-      GPU_stack_link(mat, node, names[node->custom1], in, out);
-      break;
-    case NODE_MATH_SIN:
-    case NODE_MATH_COS:
-    case NODE_MATH_TAN:
-    case NODE_MATH_ASIN:
-    case NODE_MATH_ACOS:
-    case NODE_MATH_ATAN:
-    case NODE_MATH_ROUND:
-    case NODE_MATH_ABS:
-    case NODE_MATH_FLOOR:
-    case NODE_MATH_FRACT:
-    case NODE_MATH_CEIL:
-    case NODE_MATH_SQRT:
-      if (in[0].hasinput || !in[1].hasinput) {
-        /* use only first item and terminator */
-        GPUNodeStack tmp_in[2];
-        memcpy(&tmp_in[0], &in[0], sizeof(GPUNodeStack));
-        memcpy(&tmp_in[1], &in[2], sizeof(GPUNodeStack));
-        GPU_stack_link(mat, node, names[node->custom1], tmp_in, out);
-      }
-      else {
-        /* use only second item and terminator */
-        GPUNodeStack tmp_in[2];
-        memcpy(&tmp_in[0], &in[1], sizeof(GPUNodeStack));
-        memcpy(&tmp_in[1], &in[2], sizeof(GPUNodeStack));
-        GPU_stack_link(mat, node, names[node->custom1], tmp_in, out);
-      }
-      break;
-    default:
-      return 0;
-  }
+  GPU_stack_link(mat, node, names[node->custom1], in, out);
 
   if (node->custom2 & SHD_MATH_CLAMP) {
     float min[3] = {0.0f, 0.0f, 0.0f};
     float max[3] = {1.0f, 1.0f, 1.0f};
     GPU_link(mat, "clamp_value", out[0].link, GPU_constant(min), GPU_constant(max), &out[0].link);
   }
-
   return 1;
 }
 
+static void node_shader_update_math(bNodeTree *UNUSED(ntree), bNode *node)
+{
+  bNodeSocket *sock = BLI_findlink(&node->inputs, 1);
+  nodeSetSocketAvailability(sock,
+                            !ELEM(node->custom1,
+                                  NODE_MATH_SQRT,
+                                  NODE_MATH_CEIL,
+                                  NODE_MATH_SINE,
+                                  NODE_MATH_ROUND,
+                                  NODE_MATH_FLOOR,
+                                  NODE_MATH_COSINE,
+                                  NODE_MATH_ARCSINE,
+                                  NODE_MATH_TANGENT,
+                                  NODE_MATH_ABSOLUTE,
+                                  NODE_MATH_FRACTION,
+                                  NODE_MATH_ARCCOSINE,
+                                  NODE_MATH_ARCTANGENT));
+}
+
 void register_node_type_sh_math(void)
 {
   static bNodeType ntype;
@@ -373,9 +104,8 @@ void register_node_type_sh_math(void)
   sh_node_type_base(&ntype, SH_NODE_MATH, "Math", NODE_CLASS_CONVERTOR, 0);
   node_type_socket_templates(&ntype, sh_node_math_in, sh_node_math_out);
   node_type_label(&ntype, node_math_label);
-  node_type_storage(&ntype, "", NULL, NULL);
-  node_type_exec(&ntype, NULL, NULL, node_shader_exec_math);
   node_type_gpu(&ntype, gpu_shader_math);
+  node_type_update(&ntype, node_shader_update_math);
 
   nodeRegisterType(&ntype);
 }
diff --git a/source/blender/nodes/shader/nodes/node_shader_mixRgb.c b/source/blender/nodes/shader/nodes/node_shader_mixRgb.c
index 872f4f9da9c..ae2184d8237 100644
--- a/source/blender/nodes/shader/nodes/node_shader_mixRgb.c
+++ b/source/blender/nodes/shader/nodes/node_shader_mixRgb.c
@@ -92,7 +92,7 @@ static int gpu_shader_mix_rgb(GPUMaterial *mat,
   if (ret && node->custom2 & SHD_MIXRGB_CLAMP) {
     float min[3] = {0.0f, 0.0f, 0.0f};
     float max[3] = {1.0f, 1.0f, 1.0f};
-    GPU_link(mat, "clamp_vec3", out[0].link, GPU_constant(min), GPU_constant(max), &out[0].link);
+    GPU_link(mat, "clamp_color", out[0].link, GPU_constant(min), GPU_constant(max), &out[0].link);
   }
   return ret;
 }
diff --git a/source/blender/nodes/shader/nodes/node_shader_normal.c b/source/blender/nodes/shader/nodes/node_shader_normal.c
index 074cc3dd87f..9dd89258446 100644
--- a/source/blender/nodes/shader/nodes/node_shader_normal.c
+++ b/source/blender/nodes/shader/nodes/node_shader_normal.c
@@ -25,7 +25,7 @@
 
 /* **************** NORMAL  ******************** */
 static bNodeSocketTemplate sh_node_normal_in[] = {
-    {SOCK_VECTOR, 1, N_("Normal"), 0.0f, 0.0f, 0.0f, 1.0f, -1.0f, 1.0f, PROP_DIRECTION},
+    {SOCK_VECTOR, 1, N_("Normal"), 0.0f, 0.0f, 1.0f, 0.0f, -1.0f, 1.0f, PROP_DIRECTION},
     {-1, 0, ""},
 };
 
diff --git a/source/blender/nodes/shader/nodes/node_shader_normal_map.c b/source/blender/nodes/shader/nodes/node_shader_normal_map.c
index 4976d038065..18015d94f03 100644
--- a/source/blender/nodes/shader/nodes/node_shader_normal_map.c
+++ b/source/blender/nodes/shader/nodes/node_shader_normal_map.c
@@ -58,7 +58,7 @@ static int gpu_shader_normal_map(GPUMaterial *mat,
   GPUNodeLink *realnorm;
   GPUNodeLink *strength;
 
-  float d[4] = {0, 0, 0, 0};
+  float strength_min[4] = {0, 0, 0, 0};
 
   if (in[0].link) {
     strength = in[0].link;
@@ -85,7 +85,7 @@ static int gpu_shader_normal_map(GPUMaterial *mat,
   }
 
   negnorm = GPU_builtin(GPU_WORLD_NORMAL);
-  GPU_link(mat, "math_max", strength, GPU_constant(d), &strength);
+  GPU_link(mat, "math_maximum", strength, GPU_constant(strength_min), &strength);
 
   const char *color_to_normal_fnc_name = "color_to_normal_new_shading";
   if (nm->space == SHD_SPACE_BLENDER_OBJECT || nm->space == SHD_SPACE_BLENDER_WORLD) {
@@ -114,8 +114,8 @@ static int gpu_shader_normal_map(GPUMaterial *mat,
       break;
   }
 
-  GPU_link(mat, "vec_math_mix", strength, realnorm, negnorm, &out[0].link);
-  GPU_link(mat, "vect_normalize", out[0].link, &out[0].link);
+  GPU_link(mat, "vector_mix", strength, realnorm, negnorm, &out[0].link);
+  GPU_link(mat, "vector_normalize", out[0].link, &out[0].link);
 
   return true;
 }
diff --git a/source/blender/nodes/shader/nodes/node_shader_object_info.c b/source/blender/nodes/shader/nodes/node_shader_object_info.c
index 118b8136693..58449a53706 100644
--- a/source/blender/nodes/shader/nodes/node_shader_object_info.c
+++ b/source/blender/nodes/shader/nodes/node_shader_object_info.c
@@ -23,6 +23,7 @@
 
 static bNodeSocketTemplate sh_node_object_info_out[] = {
     {SOCK_VECTOR, 0, N_("Location"), 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f},
+    {SOCK_RGBA, 0, N_("Color"), 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f},
     {SOCK_FLOAT, 0, N_("Object Index"), 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f},
     {SOCK_FLOAT, 0, N_("Material Index"), 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f},
     {SOCK_FLOAT, 0, N_("Random"), 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f},
@@ -36,38 +37,25 @@ static int node_shader_gpu_object_info(GPUMaterial *mat,
                                        GPUNodeStack *out)
 {
   Material *ma = GPU_material_get_material(mat);
-  /* Convert to float. */
-  float index = ma ? ma->index : 0;
+  float index = ma ? ma->index : 0.0f;
   return GPU_stack_link(mat,
                         node,
                         "node_object_info",
                         in,
                         out,
                         GPU_builtin(GPU_OBJECT_MATRIX),
+                        GPU_builtin(GPU_OBJECT_COLOR),
                         GPU_builtin(GPU_OBJECT_INFO),
                         GPU_constant(&index));
 }
 
-static void node_shader_exec_object_info(void *UNUSED(data),
-                                         int UNUSED(thread),
-                                         bNode *UNUSED(node),
-                                         bNodeExecData *UNUSED(execdata),
-                                         bNodeStack **UNUSED(in),
-                                         bNodeStack **UNUSED(out))
-{
-}
-
-/* node type definition */
 void register_node_type_sh_object_info(void)
 {
   static bNodeType ntype;
 
   sh_node_type_base(&ntype, SH_NODE_OBJECT_INFO, "Object Info", NODE_CLASS_INPUT, 0);
   node_type_socket_templates(&ntype, NULL, sh_node_object_info_out);
-  node_type_init(&ntype, NULL);
-  node_type_storage(&ntype, "", NULL, NULL);
   node_type_gpu(&ntype, node_shader_gpu_object_info);
-  node_type_exec(&ntype, NULL, NULL, node_shader_exec_object_info);
 
   nodeRegisterType(&ntype);
 }
diff --git a/source/blender/nodes/shader/nodes/node_shader_tex_environment.c b/source/blender/nodes/shader/nodes/node_shader_tex_environment.c
index bd8355ec885..a02262950a8 100644
--- a/source/blender/nodes/shader/nodes/node_shader_tex_environment.c
+++ b/source/blender/nodes/shader/nodes/node_shader_tex_environment.c
@@ -124,15 +124,15 @@ static int node_shader_gpu_tex_environment(GPUMaterial *mat,
     if (ELEM(ima->alpha_mode, IMA_ALPHA_IGNORE, IMA_ALPHA_CHANNEL_PACKED) ||
         IMB_colormanagement_space_name_is_data(ima->colorspace_settings.name)) {
       /* Don't let alpha affect color output in these cases. */
-      GPU_link(mat, "tex_color_alpha_clear", out[0].link, &out[0].link);
+      GPU_link(mat, "color_alpha_clear", out[0].link, &out[0].link);
     }
     else {
       /* Always output with premultiplied alpha. */
       if (ima->alpha_mode == IMA_ALPHA_PREMUL) {
-        GPU_link(mat, "tex_color_alpha_clear", out[0].link, &out[0].link);
+        GPU_link(mat, "color_alpha_clear", out[0].link, &out[0].link);
       }
       else {
-        GPU_link(mat, "tex_color_alpha_premultiply", out[0].link, &out[0].link);
+        GPU_link(mat, "color_alpha_premultiply", out[0].link, &out[0].link);
       }
     }
   }
diff --git a/source/blender/nodes/shader/nodes/node_shader_tex_image.c b/source/blender/nodes/shader/nodes/node_shader_tex_image.c
index 6f3614e357d..5eaf09659d8 100644
--- a/source/blender/nodes/shader/nodes/node_shader_tex_image.c
+++ b/source/blender/nodes/shader/nodes/node_shader_tex_image.c
@@ -183,7 +183,7 @@ static int node_shader_gpu_tex_image(GPUMaterial *mat,
     if (ELEM(ima->alpha_mode, IMA_ALPHA_IGNORE, IMA_ALPHA_CHANNEL_PACKED) ||
         IMB_colormanagement_space_name_is_data(ima->colorspace_settings.name)) {
       /* Don't let alpha affect color output in these cases. */
-      GPU_link(mat, "tex_color_alpha_clear", out[0].link, &out[0].link);
+      GPU_link(mat, "color_alpha_clear", out[0].link, &out[0].link);
     }
     else {
       /* Output premultiplied alpha depending on alpha socket usage. This makes
@@ -192,18 +192,18 @@ static int node_shader_gpu_tex_image(GPUMaterial *mat,
        * not, then there will be no artifacts from zero alpha areas. */
       if (ima->alpha_mode == IMA_ALPHA_PREMUL) {
         if (out[1].hasoutput) {
-          GPU_link(mat, "tex_color_alpha_unpremultiply", out[0].link, &out[0].link);
+          GPU_link(mat, "color_alpha_unpremultiply", out[0].link, &out[0].link);
         }
         else {
-          GPU_link(mat, "tex_color_alpha_clear", out[0].link, &out[0].link);
+          GPU_link(mat, "color_alpha_clear", out[0].link, &out[0].link);
         }
       }
       else {
         if (out[1].hasoutput) {
-          GPU_link(mat, "tex_color_alpha_clear", out[0].link, &out[0].link);
+          GPU_link(mat, "color_alpha_clear", out[0].link, &out[0].link);
         }
         else {
-          GPU_link(mat, "tex_color_alpha_premultiply", out[0].link, &out[0].link);
+          GPU_link(mat, "color_alpha_premultiply", out[0].link, &out[0].link);
         }
       }
     }
diff --git a/source/blender/nodes/shader/nodes/node_shader_tex_white_noise.c b/source/blender/nodes/shader/nodes/node_shader_tex_white_noise.c
new file mode 100644
index 00000000000..c0f9a033476
--- /dev/null
+++ b/source/blender/nodes/shader/nodes/node_shader_tex_white_noise.c
@@ -0,0 +1,76 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * The Original Code is Copyright (C) 2005 Blender Foundation.
+ * All rights reserved.
+ */
+
+#include "../node_shader_util.h"
+
+/* **************** WHITE NOISE **************** */
+
+static bNodeSocketTemplate sh_node_tex_white_noise_in[] = {
+    {SOCK_VECTOR, 1, N_("Vector"), 0.0f, 0.0f, 0.0f, 0.0f, -10000.0f, 10000.0f, PROP_NONE},
+    {SOCK_FLOAT, 1, N_("W"), 0.0f, 0.0f, 0.0f, 0.0f, -10000.0f, 10000.0f, PROP_NONE},
+    {-1, 0, ""}};
+
+static bNodeSocketTemplate sh_node_tex_white_noise_out[] = {
+    {SOCK_FLOAT, 0, N_("Value")},
+    {-1, 0, ""},
+};
+
+static void node_shader_init_tex_white_noise(bNodeTree *UNUSED(ntree), bNode *node)
+{
+  node->custom1 = 3;
+}
+
+static int gpu_shader_tex_white_noise(GPUMaterial *mat,
+                                      bNode *node,
+                                      bNodeExecData *UNUSED(execdata),
+                                      GPUNodeStack *in,
+                                      GPUNodeStack *out)
+{
+  static const char *names[] = {
+      "",
+      "node_white_noise_1d",
+      "node_white_noise_2d",
+      "node_white_noise_3d",
+      "node_white_noise_4d",
+  };
+
+  return GPU_stack_link(mat, node, names[node->custom1], in, out);
+}
+
+static void node_shader_update_tex_white_noise(bNodeTree *UNUSED(ntree), bNode *node)
+{
+  bNodeSocket *sockVector = nodeFindSocket(node, SOCK_IN, "Vector");
+  bNodeSocket *sockW = nodeFindSocket(node, SOCK_IN, "W");
+
+  nodeSetSocketAvailability(sockVector, node->custom1 != 1);
+  nodeSetSocketAvailability(sockW, node->custom1 == 1 || node->custom1 == 4);
+}
+
+void register_node_type_sh_tex_white_noise(void)
+{
+  static bNodeType ntype;
+
+  sh_node_type_base(&ntype, SH_NODE_TEX_WHITE_NOISE, "White Noise Texture", NODE_CLASS_TEXTURE, 0);
+  node_type_socket_templates(&ntype, sh_node_tex_white_noise_in, sh_node_tex_white_noise_out);
+  node_type_init(&ntype, node_shader_init_tex_white_noise);
+  node_type_gpu(&ntype, gpu_shader_tex_white_noise);
+  node_type_update(&ntype, node_shader_update_tex_white_noise);
+
+  nodeRegisterType(&ntype);
+}
diff --git a/source/blender/nodes/shader/nodes/node_shader_vectMath.c b/source/blender/nodes/shader/nodes/node_shader_vectMath.c
deleted file mode 100644
index 41273a6dc1d..00000000000
--- a/source/blender/nodes/shader/nodes/node_shader_vectMath.c
+++ /dev/null
@@ -1,155 +0,0 @@
-/*
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public License
- * as published by the Free Software Foundation; either version 2
- * of the License, or (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
- *
- * The Original Code is Copyright (C) 2005 Blender Foundation.
- * All rights reserved.
- */
-
-/** \file
- * \ingroup shdnodes
- */
-
-#include "node_shader_util.h"
-
-/* **************** VECTOR MATH ******************** */
-static bNodeSocketTemplate sh_node_vect_math_in[] = {
-    {SOCK_VECTOR, 1, N_("Vector"), 0.5f, 0.5f, 0.5f, 1.0f, -10000.0f, 10000.0f, PROP_NONE},
-    {SOCK_VECTOR, 1, N_("Vector"), 0.5f, 0.5f, 0.5f, 1.0f, -10000.0f, 10000.0f, PROP_NONE},
-    {-1, 0, ""}};
-
-static bNodeSocketTemplate sh_node_vect_math_out[] = {
-    {SOCK_VECTOR, 0, N_("Vector")}, {SOCK_FLOAT, 0, N_("Value")}, {-1, 0, ""}};
-
-static void node_shader_exec_vect_math(void *UNUSED(data),
-                                       int UNUSED(thread),
-                                       bNode *node,
-                                       bNodeExecData *UNUSED(execdata),
-                                       bNodeStack **in,
-                                       bNodeStack **out)
-{
-  float vec1[3], vec2[3];
-
-  nodestack_get_vec(vec1, SOCK_VECTOR, in[0]);
-  nodestack_get_vec(vec2, SOCK_VECTOR, in[1]);
-
-  if (node->custom1 == 0) { /* Add */
-    out[0]->vec[0] = vec1[0] + vec2[0];
-    out[0]->vec[1] = vec1[1] + vec2[1];
-    out[0]->vec[2] = vec1[2] + vec2[2];
-
-    out[1]->vec[0] = (fabsf(out[0]->vec[0]) + fabsf(out[0]->vec[1]) + fabsf(out[0]->vec[2])) /
-                     3.0f;
-  }
-  else if (node->custom1 == 1) { /* Subtract */
-    out[0]->vec[0] = vec1[0] - vec2[0];
-    out[0]->vec[1] = vec1[1] - vec2[1];
-    out[0]->vec[2] = vec1[2] - vec2[2];
-
-    out[1]->vec[0] = (fabsf(out[0]->vec[0]) + fabsf(out[0]->vec[1]) + fabsf(out[0]->vec[2])) /
-                     3.0f;
-  }
-  else if (node->custom1 == 2) { /* Average */
-    out[0]->vec[0] = vec1[0] + vec2[0];
-    out[0]->vec[1] = vec1[1] + vec2[1];
-    out[0]->vec[2] = vec1[2] + vec2[2];
-
-    out[1]->vec[0] = normalize_v3(out[0]->vec);
-  }
-  else if (node->custom1 == 3) { /* Dot product */
-    out[1]->vec[0] = (vec1[0] * vec2[0]) + (vec1[1] * vec2[1]) + (vec1[2] * vec2[2]);
-  }
-  else if (node->custom1 == 4) { /* Cross product */
-    out[0]->vec[0] = (vec1[1] * vec2[2]) - (vec1[2] * vec2[1]);
-    out[0]->vec[1] = (vec1[2] * vec2[0]) - (vec1[0] * vec2[2]);
-    out[0]->vec[2] = (vec1[0] * vec2[1]) - (vec1[1] * vec2[0]);
-
-    out[1]->vec[0] = normalize_v3(out[0]->vec);
-  }
-  else if (node->custom1 == 5) { /* Normalize */
-    /* This one only takes one input, so we've got to choose. */
-    if (in[0]->hasinput || !in[1]->hasinput) {
-      out[0]->vec[0] = vec1[0];
-      out[0]->vec[1] = vec1[1];
-      out[0]->vec[2] = vec1[2];
-    }
-    else {
-      out[0]->vec[0] = vec2[0];
-      out[0]->vec[1] = vec2[1];
-      out[0]->vec[2] = vec2[2];
-    }
-
-    out[1]->vec[0] = normalize_v3(out[0]->vec);
-  }
-}
-
-static int gpu_shader_vect_math(GPUMaterial *mat,
-                                bNode *node,
-                                bNodeExecData *UNUSED(execdata),
-                                GPUNodeStack *in,
-                                GPUNodeStack *out)
-{
-  static const char *names[] = {
-      "vec_math_add",
-      "vec_math_sub",
-      "vec_math_average",
-      "vec_math_dot",
-      "vec_math_cross",
-      "vec_math_normalize",
-  };
-
-  switch (node->custom1) {
-    case 0:
-    case 1:
-    case 2:
-    case 3:
-    case 4:
-      GPU_stack_link(mat, node, names[node->custom1], in, out);
-      break;
-    case 5:
-      if (in[0].hasinput || !in[1].hasinput) {
-        /* use only first item and terminator */
-        GPUNodeStack tmp_in[2];
-        memcpy(&tmp_in[0], &in[0], sizeof(GPUNodeStack));
-        memcpy(&tmp_in[1], &in[2], sizeof(GPUNodeStack));
-        GPU_stack_link(mat, node, names[node->custom1], tmp_in, out);
-      }
-      else {
-        /* use only second item and terminator */
-        GPUNodeStack tmp_in[2];
-        memcpy(&tmp_in[0], &in[1], sizeof(GPUNodeStack));
-        memcpy(&tmp_in[1], &in[2], sizeof(GPUNodeStack));
-        GPU_stack_link(mat, node, names[node->custom1], tmp_in, out);
-      }
-      break;
-    default:
-      return false;
-  }
-
-  return true;
-}
-
-void register_node_type_sh_vect_math(void)
-{
-  static bNodeType ntype;
-
-  sh_node_type_base(&ntype, SH_NODE_VECT_MATH, "Vector Math", NODE_CLASS_CONVERTOR, 0);
-  node_type_socket_templates(&ntype, sh_node_vect_math_in, sh_node_vect_math_out);
-  node_type_label(&ntype, node_vect_math_label);
-  node_type_storage(&ntype, "", NULL, NULL);
-  node_type_exec(&ntype, NULL, NULL, node_shader_exec_vect_math);
-  node_type_gpu(&ntype, gpu_shader_vect_math);
-
-  nodeRegisterType(&ntype);
-}
diff --git a/source/blender/nodes/shader/nodes/node_shader_vectTransform.c b/source/blender/nodes/shader/nodes/node_shader_vectTransform.c
index fe0e7b1d638..563ef89162b 100644
--- a/source/blender/nodes/shader/nodes/node_shader_vectTransform.c
+++ b/source/blender/nodes/shader/nodes/node_shader_vectTransform.c
@@ -132,7 +132,7 @@ static int gpu_shader_vect_transform(GPUMaterial *mat,
   }
 
   if (nodeprop->type == SHD_VECT_TRANSFORM_TYPE_NORMAL) {
-    GPU_link(mat, "vect_normalize", out[0].link, &out[0].link);
+    GPU_link(mat, "vector_normalize", out[0].link, &out[0].link);
   }
 
   return true;
diff --git a/source/blender/nodes/shader/nodes/node_shader_vector_math.c b/source/blender/nodes/shader/nodes/node_shader_vector_math.c
new file mode 100644
index 00000000000..03de83655b5
--- /dev/null
+++ b/source/blender/nodes/shader/nodes/node_shader_vector_math.c
@@ -0,0 +1,113 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * The Original Code is Copyright (C) 2005 Blender Foundation.
+ * All rights reserved.
+ */
+
+/** \file
+ * \ingroup shdnodes
+ */
+
+#include "node_shader_util.h"
+
+/* **************** VECTOR MATH ******************** */
+static bNodeSocketTemplate sh_node_vector_math_in[] = {
+    {SOCK_VECTOR, 1, N_("Vector"), 0.0f, 0.0f, 0.0f, 1.0f, -10000.0f, 10000.0f, PROP_NONE},
+    {SOCK_VECTOR, 1, N_("Vector"), 0.0f, 0.0f, 0.0f, 1.0f, -10000.0f, 10000.0f, PROP_NONE},
+    {SOCK_FLOAT, 1, N_("Scale"), 1.0f, 1.0f, 1.0f, 1.0f, -10000.0f, 10000.0f, PROP_NONE},
+    {-1, 0, ""}};
+
+static bNodeSocketTemplate sh_node_vector_math_out[] = {
+    {SOCK_VECTOR, 0, N_("Vector")}, {SOCK_FLOAT, 0, N_("Value")}, {-1, 0, ""}};
+
+static int gpu_shader_vector_math(GPUMaterial *mat,
+                                  bNode *node,
+                                  bNodeExecData *UNUSED(execdata),
+                                  GPUNodeStack *in,
+                                  GPUNodeStack *out)
+{
+  static const char *names[] = {
+      [NODE_VECTOR_MATH_ADD] = "vector_math_add",
+      [NODE_VECTOR_MATH_SUBTRACT] = "vector_math_subtract",
+      [NODE_VECTOR_MATH_MULTIPLY] = "vector_math_multiply",
+      [NODE_VECTOR_MATH_DIVIDE] = "vector_math_divide",
+
+      [NODE_VECTOR_MATH_CROSS_PRODUCT] = "vector_math_cross",
+      [NODE_VECTOR_MATH_PROJECT] = "vector_math_project",
+      [NODE_VECTOR_MATH_REFLECT] = "vector_math_reflect",
+      [NODE_VECTOR_MATH_DOT_PRODUCT] = "vector_math_dot",
+
+      [NODE_VECTOR_MATH_DISTANCE] = "vector_math_distance",
+      [NODE_VECTOR_MATH_LENGTH] = "vector_math_length",
+      [NODE_VECTOR_MATH_SCALE] = "vector_math_scale",
+      [NODE_VECTOR_MATH_NORMALIZE] = "vector_math_normalize",
+
+      [NODE_VECTOR_MATH_SNAP] = "vector_math_snap",
+      [NODE_VECTOR_MATH_FLOOR] = "vector_math_floor",
+      [NODE_VECTOR_MATH_CEIL] = "vector_math_ceil",
+      [NODE_VECTOR_MATH_MODULO] = "vector_math_modulo",
+      [NODE_VECTOR_MATH_FRACTION] = "vector_math_fraction",
+      [NODE_VECTOR_MATH_ABSOLUTE] = "vector_math_absolute",
+      [NODE_VECTOR_MATH_MINIMUM] = "vector_math_minimum",
+      [NODE_VECTOR_MATH_MAXIMUM] = "vector_math_maximum",
+  };
+
+  GPU_stack_link(mat, node, names[node->custom1], in, out);
+  return true;
+}
+
+static void node_shader_update_vector_math(bNodeTree *UNUSED(ntree), bNode *node)
+{
+  bNodeSocket *sockB = BLI_findlink(&node->inputs, 1);
+  bNodeSocket *sockScale = nodeFindSocket(node, SOCK_IN, "Scale");
+
+  bNodeSocket *sockVector = nodeFindSocket(node, SOCK_OUT, "Vector");
+  bNodeSocket *sockValue = nodeFindSocket(node, SOCK_OUT, "Value");
+
+  nodeSetSocketAvailability(sockB,
+                            !ELEM(node->custom1,
+                                  NODE_VECTOR_MATH_CEIL,
+                                  NODE_VECTOR_MATH_SCALE,
+                                  NODE_VECTOR_MATH_FLOOR,
+                                  NODE_VECTOR_MATH_LENGTH,
+                                  NODE_VECTOR_MATH_ABSOLUTE,
+                                  NODE_VECTOR_MATH_FRACTION,
+                                  NODE_VECTOR_MATH_NORMALIZE));
+  nodeSetSocketAvailability(sockScale, node->custom1 == NODE_VECTOR_MATH_SCALE);
+  nodeSetSocketAvailability(sockVector,
+                            !ELEM(node->custom1,
+                                  NODE_VECTOR_MATH_LENGTH,
+                                  NODE_VECTOR_MATH_DISTANCE,
+                                  NODE_VECTOR_MATH_DOT_PRODUCT));
+  nodeSetSocketAvailability(sockValue,
+                            ELEM(node->custom1,
+                                 NODE_VECTOR_MATH_LENGTH,
+                                 NODE_VECTOR_MATH_DISTANCE,
+                                 NODE_VECTOR_MATH_DOT_PRODUCT));
+}
+
+void register_node_type_sh_vect_math(void)
+{
+  static bNodeType ntype;
+
+  sh_node_type_base(&ntype, SH_NODE_VECTOR_MATH, "Vector Math", NODE_CLASS_CONVERTOR, 0);
+  node_type_socket_templates(&ntype, sh_node_vector_math_in, sh_node_vector_math_out);
+  node_type_label(&ntype, node_vector_math_label);
+  node_type_gpu(&ntype, gpu_shader_vector_math);
+  node_type_update(&ntype, node_shader_update_vector_math);
+
+  nodeRegisterType(&ntype);
+}
diff --git a/source/blender/nodes/shader/nodes/node_shader_volume_info.c b/source/blender/nodes/shader/nodes/node_shader_volume_info.c
new file mode 100644
index 00000000000..98c44dd590b
--- /dev/null
+++ b/source/blender/nodes/shader/nodes/node_shader_volume_info.c
@@ -0,0 +1,56 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * The Original Code is Copyright (C) 2005 Blender Foundation.
+ * All rights reserved.
+ */
+
+#include "../node_shader_util.h"
+
+static bNodeSocketTemplate sh_node_volume_info_out[] = {
+    {SOCK_RGBA, 0, N_("Color")},
+    {SOCK_FLOAT, 0, N_("Density")},
+    {SOCK_FLOAT, 0, N_("Flame")},
+    {SOCK_FLOAT, 0, N_("Temperature")},
+    {-1, 0, ""},
+};
+
+static int node_shader_gpu_volume_info(GPUMaterial *mat,
+                                       bNode *node,
+                                       bNodeExecData *UNUSED(execdata),
+                                       GPUNodeStack *in,
+                                       GPUNodeStack *out)
+{
+
+  return GPU_stack_link(mat,
+                        node,
+                        "node_volume_info",
+                        in,
+                        out,
+                        GPU_builtin(GPU_VOLUME_DENSITY),
+                        GPU_builtin(GPU_VOLUME_FLAME),
+                        GPU_builtin(GPU_VOLUME_TEMPERATURE));
+}
+
+void register_node_type_sh_volume_info(void)
+{
+  static bNodeType ntype;
+
+  sh_node_type_base(&ntype, SH_NODE_VOLUME_INFO, "Volume Info", NODE_CLASS_INPUT, 0);
+  node_type_socket_templates(&ntype, NULL, sh_node_volume_info_out);
+  node_type_gpu(&ntype, node_shader_gpu_volume_info);
+
+  nodeRegisterType(&ntype);
+}
diff --git a/source/blender/nodes/texture/nodes/node_texture_math.c b/source/blender/nodes/texture/nodes/node_texture_math.c
index 2eb32e0addc..b1d67a5a953 100644
--- a/source/blender/nodes/texture/nodes/node_texture_math.c
+++ b/source/blender/nodes/texture/nodes/node_texture_math.c
@@ -46,10 +46,10 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
     case NODE_MATH_ADD:
       *out = in0 + in1;
       break;
-    case NODE_MATH_SUB:
+    case NODE_MATH_SUBTRACT:
       *out = in0 - in1;
       break;
-    case NODE_MATH_MUL:
+    case NODE_MATH_MULTIPLY:
       *out = in0 * in1;
       break;
     case NODE_MATH_DIVIDE: {
@@ -62,19 +62,19 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       }
       break;
     }
-    case NODE_MATH_SIN: {
+    case NODE_MATH_SINE: {
       *out = sinf(in0);
       break;
     }
-    case NODE_MATH_COS: {
+    case NODE_MATH_COSINE: {
       *out = cosf(in0);
       break;
     }
-    case NODE_MATH_TAN: {
+    case NODE_MATH_TANGENT: {
       *out = tanf(in0);
       break;
     }
-    case NODE_MATH_ASIN: {
+    case NODE_MATH_ARCSINE: {
       /* Can't do the impossible... */
       if (in0 <= 1 && in0 >= -1) {
         *out = asinf(in0);
@@ -84,7 +84,7 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       }
       break;
     }
-    case NODE_MATH_ACOS: {
+    case NODE_MATH_ARCCOSINE: {
       /* Can't do the impossible... */
       if (in0 <= 1 && in0 >= -1) {
         *out = acosf(in0);
@@ -94,11 +94,11 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       }
       break;
     }
-    case NODE_MATH_ATAN: {
+    case NODE_MATH_ARCTANGENT: {
       *out = atan(in0);
       break;
     }
-    case NODE_MATH_POW: {
+    case NODE_MATH_POWER: {
       /* Only raise negative numbers by full integers */
       if (in0 >= 0) {
         out[0] = pow(in0, in1);
@@ -114,7 +114,7 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       }
       break;
     }
-    case NODE_MATH_LOG: {
+    case NODE_MATH_LOGARITHM: {
       /* Don't want any imaginary numbers... */
       if (in0 > 0 && in1 > 0) {
         *out = log(in0) / log(in1);
@@ -124,7 +124,7 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       }
       break;
     }
-    case NODE_MATH_MIN: {
+    case NODE_MATH_MINIMUM: {
       if (in0 < in1) {
         *out = in0;
       }
@@ -133,7 +133,7 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       }
       break;
     }
-    case NODE_MATH_MAX: {
+    case NODE_MATH_MAXIMUM: {
       if (in0 > in1) {
         *out = in0;
       }
@@ -147,7 +147,7 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       break;
     }
 
-    case NODE_MATH_LESS: {
+    case NODE_MATH_LESS_THAN: {
       if (in0 < in1) {
         *out = 1.0f;
       }
@@ -157,7 +157,7 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       break;
     }
 
-    case NODE_MATH_GREATER: {
+    case NODE_MATH_GREATER_THAN: {
       if (in0 > in1) {
         *out = 1.0f;
       }
@@ -167,7 +167,7 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       break;
     }
 
-    case NODE_MATH_MOD: {
+    case NODE_MATH_MODULO: {
       if (in1 == 0.0f) {
         *out = 0.0f;
       }
@@ -177,12 +177,12 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       break;
     }
 
-    case NODE_MATH_ABS: {
+    case NODE_MATH_ABSOLUTE: {
       *out = fabsf(in0);
       break;
     }
 
-    case NODE_MATH_ATAN2: {
+    case NODE_MATH_ARCTAN2: {
       *out = atan2(in0, in1);
       break;
     }
@@ -197,7 +197,7 @@ static void valuefn(float *out, TexParams *p, bNode *node, bNodeStack **in, shor
       break;
     }
 
-    case NODE_MATH_FRACT: {
+    case NODE_MATH_FRACTION: {
       *out = in0 - floorf(in0);
       break;
     }
diff --git a/source/blender/physics/BPH_mass_spring.h b/source/blender/physics/BPH_mass_spring.h
index 9666f9ba9ed..c0ceff4d8cf 100644
--- a/source/blender/physics/BPH_mass_spring.h
+++ b/source/blender/physics/BPH_mass_spring.h
@@ -33,7 +33,6 @@ struct Depsgraph;
 struct Implicit_Data;
 struct ListBase;
 struct Object;
-struct VoxelData;
 
 typedef enum eMassSpringSolverStatus {
   BPH_SOLVER_SUCCESS = (1 << 0),
diff --git a/source/blender/physics/intern/BPH_mass_spring.cpp b/source/blender/physics/intern/BPH_mass_spring.cpp
index 95c16c2f033..07cb3370eec 100644
--- a/source/blender/physics/intern/BPH_mass_spring.cpp
+++ b/source/blender/physics/intern/BPH_mass_spring.cpp
@@ -393,7 +393,7 @@ BLI_INLINE void cloth_calc_spring_force(ClothModifierData *clmd, ClothSpring *s)
     if (s->type & CLOTH_SPRING_TYPE_SEWING) {
       /* TODO: verify, half verified (couldn't see error)
        * sewing springs usually have a large distance at first so clamp the force so we don't get
-       * tunnelling through collision objects. */
+       * tunneling through collision objects. */
       BPH_mass_spring_force_spring_linear(data,
                                           s->ij,
                                           s->kl,
diff --git a/source/blender/physics/intern/implicit.h b/source/blender/physics/intern/implicit.h
index a88a5785fee..32416fa01ab 100644
--- a/source/blender/physics/intern/implicit.h
+++ b/source/blender/physics/intern/implicit.h
@@ -186,8 +186,6 @@ bool BPH_mass_spring_force_spring_goal(struct Implicit_Data *data,
 
 struct HairGrid;
 
-struct VoxelData;
-
 #define MAX_HAIR_GRID_RES 256
 
 struct HairGrid *BPH_hair_volume_create_vertex_grid(float cellsize,
diff --git a/source/blender/python/bmesh/bmesh_py_types_meshdata.c b/source/blender/python/bmesh/bmesh_py_types_meshdata.c
index 0aa01ddb594..2e15c1d9ce0 100644
--- a/source/blender/python/bmesh/bmesh_py_types_meshdata.c
+++ b/source/blender/python/bmesh/bmesh_py_types_meshdata.c
@@ -20,8 +20,8 @@
 /** \file
  * \ingroup pybmesh
  *
- * This file defines customdata types which can't be accessed as primitive
- * python types such as MDeformVert, MLoopUV, MTexPoly
+ * This file defines custom-data types which can't be accessed as primitive
+ * python types such as #MDeformVert, #MLoopUV.
  */
 
 #include <Python.h>
@@ -515,8 +515,8 @@ static PySequenceMethods bpy_bmdeformvert_as_sequence = {
     NULL,                          /* sq_concat */
     NULL,                          /* sq_repeat */
 
-    /* note: if this is set PySequence_Check() returns True,
-     * but in this case we dont want to be treated as a seq */
+    /* Note: if this is set #PySequence_Check() returns True,
+     * but in this case we don't want to be treated as a seq. */
     NULL, /* sq_item */
 
     NULL,                                  /* sq_slice */
diff --git a/source/blender/python/intern/CMakeLists.txt b/source/blender/python/intern/CMakeLists.txt
index fc945562c98..a5f71e92438 100644
--- a/source/blender/python/intern/CMakeLists.txt
+++ b/source/blender/python/intern/CMakeLists.txt
@@ -231,6 +231,10 @@ if(WITH_OPENAL)
   add_definitions(-DWITH_OPENAL)
 endif()
 
+if(WITH_OPENSUBDIV)
+  add_definitions(-DWITH_OPENSUBDIV)
+endif()
+
 if(WITH_SDL)
   list(APPEND INC_SYS
     ${SDL_INCLUDE_DIR}
diff --git a/source/blender/python/intern/bpy_app_build_options.c b/source/blender/python/intern/bpy_app_build_options.c
index a841e974e85..afb2f6b3636 100644
--- a/source/blender/python/intern/bpy_app_build_options.c
+++ b/source/blender/python/intern/bpy_app_build_options.c
@@ -46,6 +46,7 @@ static PyStructSequence_Field app_builtopts_info_fields[] = {
     {(char *)"audaspace", NULL},
     {(char *)"international", NULL},
     {(char *)"openal", NULL},
+    {(char *)"opensubdiv", NULL},
     {(char *)"sdl", NULL},
     {(char *)"sdl_dynload", NULL},
     {(char *)"jack", NULL},
@@ -190,6 +191,12 @@ static PyObject *make_builtopts_info(void)
   SetObjIncref(Py_False);
 #endif
 
+#ifdef WITH_OPENSUBDIV
+  SetObjIncref(Py_True);
+#else
+  SetObjIncref(Py_False);
+#endif
+
 #ifdef WITH_SDL
   SetObjIncref(Py_True);
 #else
diff --git a/source/blender/python/intern/bpy_app_translations.c b/source/blender/python/intern/bpy_app_translations.c
index a48cd742448..3c1dbfba72e 100644
--- a/source/blender/python/intern/bpy_app_translations.c
+++ b/source/blender/python/intern/bpy_app_translations.c
@@ -890,7 +890,7 @@ PyObject *BPY_app_translations_struct(void)
 
 void BPY_app_translations_end(void)
 {
-  /* Incase the object remains in a module's namespace, see T44127. */
+  /* In case the object remains in a module's name-space, see T44127. */
 #ifdef WITH_INTERNATIONAL
   _clear_translations_cache();
 #endif
diff --git a/source/blender/python/intern/bpy_interface.c b/source/blender/python/intern/bpy_interface.c
index 71bc01d6b98..a73e4034ca6 100644
--- a/source/blender/python/intern/bpy_interface.c
+++ b/source/blender/python/intern/bpy_interface.c
@@ -808,7 +808,7 @@ int BPY_context_member_get(bContext *C, const char *member, bContextDataResult *
     ptr = &(((BPy_StructRNA *)item)->ptr);
 
     // result->ptr = ((BPy_StructRNA *)item)->ptr;
-    CTX_data_pointer_set(result, ptr->id.data, ptr->type, ptr->data);
+    CTX_data_pointer_set(result, ptr->owner_id, ptr->type, ptr->data);
     CTX_data_type_set(result, CTX_DATA_TYPE_POINTER);
     done = true;
   }
@@ -834,7 +834,7 @@ int BPY_context_member_get(bContext *C, const char *member, bContextDataResult *
           BLI_addtail(&result->list, link);
 #endif
           ptr = &(((BPy_StructRNA *)list_item)->ptr);
-          CTX_data_list_add(result, ptr->id.data, ptr->type, ptr->data);
+          CTX_data_list_add(result, ptr->owner_id, ptr->type, ptr->data);
         }
         else {
           CLOG_INFO(BPY_LOG_CONTEXT,
diff --git a/source/blender/python/intern/bpy_msgbus.c b/source/blender/python/intern/bpy_msgbus.c
index 9db4201c252..630a0392dbd 100644
--- a/source/blender/python/intern/bpy_msgbus.c
+++ b/source/blender/python/intern/bpy_msgbus.c
@@ -269,7 +269,7 @@ static PyObject *bpy_msgbus_subscribe_rna(PyObject *UNUSED(self), PyObject *args
   /* Note: we may want to have a way to pass this in. */
   bContext *C = (bContext *)BPy_GetContext();
   struct wmMsgBus *mbus = CTX_wm_message_bus(C);
-  wmMsgParams_RNA msg_key_params = {{{0}}};
+  wmMsgParams_RNA msg_key_params = {{0}};
 
   wmMsgSubscribeValue msg_val_params = {0};
 
@@ -342,7 +342,7 @@ static PyObject *bpy_msgbus_publish_rna(PyObject *UNUSED(self), PyObject *args,
   /* Note: we may want to have a way to pass this in. */
   bContext *C = (bContext *)BPy_GetContext();
   struct wmMsgBus *mbus = CTX_wm_message_bus(C);
-  wmMsgParams_RNA msg_key_params = {{{0}}};
+  wmMsgParams_RNA msg_key_params = {{0}};
 
   if (py_msgbus_rna_key_from_py(py_sub, &msg_key_params, error_prefix) == -1) {
     return NULL;
diff --git a/source/blender/python/intern/bpy_operator_wrap.c b/source/blender/python/intern/bpy_operator_wrap.c
index 8f2f08c7c37..2e88a2a5b06 100644
--- a/source/blender/python/intern/bpy_operator_wrap.c
+++ b/source/blender/python/intern/bpy_operator_wrap.c
@@ -58,7 +58,7 @@ static void operator_properties_init(wmOperatorType *ot)
   /* set the default property: ot->prop */
   {
     /* Picky developers will notice that 'bl_property' won't work with inheritance
-     * get direct from the dict to avoid raising a load of attribute errors (yes this isnt ideal)
+     * get direct from the dict to avoid raising a load of attribute errors (yes this isn't ideal)
      * - campbell. */
     PyObject *py_class_dict = py_class->tp_dict;
     PyObject *bl_property = PyDict_GetItem(py_class_dict, bpy_intern_str_bl_property);
diff --git a/source/blender/python/intern/bpy_rna.c b/source/blender/python/intern/bpy_rna.c
index 9a8d8c5ec21..12005b92388 100644
--- a/source/blender/python/intern/bpy_rna.c
+++ b/source/blender/python/intern/bpy_rna.c
@@ -139,7 +139,7 @@ static void id_release_gc(struct ID *id)
       if (PyType_IsSubtype(Py_TYPE(ob), &pyrna_struct_Type) ||
           PyType_IsSubtype(Py_TYPE(ob), &pyrna_prop_Type)) {
         BPy_DummyPointerRNA *ob_ptr = (BPy_DummyPointerRNA *)ob;
-        if (ob_ptr->ptr.id.data == id) {
+        if (ob_ptr->ptr.owner_id == id) {
           pyrna_invalidate(ob_ptr);
           // printf("freeing: %p %s, %.200s\n", (void *)ob, id->name, Py_TYPE(ob)->tp_name);
           // i++;
@@ -308,7 +308,7 @@ static bool rna_disallow_writes = false;
 
 static bool rna_id_write_error(PointerRNA *ptr, PyObject *key)
 {
-  ID *id = ptr->id.data;
+  ID *id = ptr->owner_id;
   if (id) {
     const short idcode = GS(id->name);
     /* May need more ID types added here. */
@@ -907,7 +907,7 @@ static PyObject *pyrna_struct_str(BPy_StructRNA *self)
 
 static PyObject *pyrna_struct_repr(BPy_StructRNA *self)
 {
-  ID *id = self->ptr.id.data;
+  ID *id = self->ptr.owner_id;
   PyObject *tmp_str;
   PyObject *ret;
 
@@ -918,21 +918,21 @@ static PyObject *pyrna_struct_repr(BPy_StructRNA *self)
 
   tmp_str = PyUnicode_FromString(id->name + 2);
 
-  if (RNA_struct_is_ID(self->ptr.type)) {
+  if (RNA_struct_is_ID(self->ptr.type) && (id->flag & LIB_PRIVATE_DATA) == 0) {
     ret = PyUnicode_FromFormat(
         "bpy.data.%s[%R]", BKE_idcode_to_name_plural(GS(id->name)), tmp_str);
   }
   else {
     const char *path;
-    path = RNA_path_from_ID_to_struct(&self->ptr);
+    ID *real_id = NULL;
+    path = RNA_path_from_real_ID_to_struct(G_MAIN, &self->ptr, &real_id);
     if (path) {
-      if (GS(id->name) == ID_NT) { /* Nodetree paths are not accurate. */
-        ret = PyUnicode_FromFormat("bpy.data...%s", path);
-      }
-      else {
-        ret = PyUnicode_FromFormat(
-            "bpy.data.%s[%R].%s", BKE_idcode_to_name_plural(GS(id->name)), tmp_str, path);
+      if (real_id != id) {
+        Py_DECREF(tmp_str);
+        tmp_str = PyUnicode_FromString(real_id->name + 2);
       }
+      ret = PyUnicode_FromFormat(
+          "bpy.data.%s[%R].%s", BKE_idcode_to_name_plural(GS(real_id->name)), tmp_str, path);
 
       MEM_freeN((void *)path);
     }
@@ -1019,7 +1019,7 @@ static PyObject *pyrna_prop_str(BPy_PropertyRNA *self)
 
 static PyObject *pyrna_prop_repr_ex(BPy_PropertyRNA *self, const int index_dim, const int index)
 {
-  ID *id = self->ptr.id.data;
+  ID *id = self->ptr.owner_id;
   PyObject *tmp_str;
   PyObject *ret;
   const char *path;
@@ -1033,20 +1033,23 @@ static PyObject *pyrna_prop_repr_ex(BPy_PropertyRNA *self, const int index_dim,
 
   tmp_str = PyUnicode_FromString(id->name + 2);
 
-  path = RNA_path_from_ID_to_property_index(&self->ptr, self->prop, index_dim, index);
+  /* Note that using G_MAIN is absolutely not ideal, but we have no access to actual Main DB from
+   * here. */
+  ID *real_id = NULL;
+  path = RNA_path_from_real_ID_to_property_index(
+      G_MAIN, &self->ptr, self->prop, index_dim, index, &real_id);
 
   if (path) {
-    const char *data_delim = (path[0] == '[') ? "" : ".";
-    if (GS(id->name) == ID_NT) { /* Nodetree paths are not accurate. */
-      ret = PyUnicode_FromFormat("bpy.data...%s", path);
-    }
-    else {
-      ret = PyUnicode_FromFormat("bpy.data.%s[%R]%s%s",
-                                 BKE_idcode_to_name_plural(GS(id->name)),
-                                 tmp_str,
-                                 data_delim,
-                                 path);
+    if (real_id != id) {
+      Py_DECREF(tmp_str);
+      tmp_str = PyUnicode_FromString(real_id->name + 2);
     }
+    const char *data_delim = (path[0] == '[') ? "" : ".";
+    ret = PyUnicode_FromFormat("bpy.data.%s[%R]%s%s",
+                               BKE_idcode_to_name_plural(GS(real_id->name)),
+                               tmp_str,
+                               data_delim,
+                               path);
 
     MEM_freeN((void *)path);
   }
@@ -1982,7 +1985,7 @@ static int pyrna_py_to_prop(
           return -1;
         }
         else if ((value != Py_None) && ((flag & PROP_ID_SELF_CHECK) &&
-                                        ptr->id.data == ((BPy_StructRNA *)value)->ptr.id.data)) {
+                                        ptr->owner_id == ((BPy_StructRNA *)value)->ptr.owner_id)) {
           PyErr_Format(PyExc_TypeError,
                        "%.200s %.200s.%.200s ID type does not support assignment to itself",
                        error_prefix,
@@ -3447,7 +3450,7 @@ static PyObject *pyrna_struct_subscript(BPy_StructRNA *self, PyObject *key)
     return NULL;
   }
 
-  return BPy_IDGroup_WrapData(self->ptr.id.data, idprop, group);
+  return BPy_IDGroup_WrapData(self->ptr.owner_id, idprop, group);
 }
 
 static int pyrna_struct_ass_subscript(BPy_StructRNA *self, PyObject *key, PyObject *value)
@@ -3543,7 +3546,7 @@ static PyObject *pyrna_struct_items(BPy_PropertyRNA *self)
     return PyList_New(0);
   }
 
-  return BPy_Wrap_GetItems(self->ptr.id.data, group);
+  return BPy_Wrap_GetItems(self->ptr.owner_id, group);
 }
 
 PyDoc_STRVAR(pyrna_struct_values_doc,
@@ -3571,7 +3574,7 @@ static PyObject *pyrna_struct_values(BPy_PropertyRNA *self)
     return PyList_New(0);
   }
 
-  return BPy_Wrap_GetValues(self->ptr.id.data, group);
+  return BPy_Wrap_GetValues(self->ptr.owner_id, group);
 }
 
 PyDoc_STRVAR(pyrna_struct_is_property_set_doc,
@@ -4708,9 +4711,9 @@ PyDoc_STRVAR(pyrna_struct_get_id_data_doc,
 static PyObject *pyrna_struct_get_id_data(BPy_DummyPointerRNA *self)
 {
   /* Used for struct and pointer since both have a ptr. */
-  if (self->ptr.id.data) {
+  if (self->ptr.owner_id) {
     PointerRNA id_ptr;
-    RNA_id_pointer_create((ID *)self->ptr.id.data, &id_ptr);
+    RNA_id_pointer_create((ID *)self->ptr.owner_id, &id_ptr);
     return pyrna_struct_CreatePyObject(&id_ptr);
   }
 
@@ -4894,7 +4897,7 @@ static PyObject *pyrna_struct_get(BPy_StructRNA *self, PyObject *args)
     idprop = IDP_GetPropertyFromGroup(group, key);
 
     if (idprop) {
-      return BPy_IDGroup_WrapData(self->ptr.id.data, idprop, group);
+      return BPy_IDGroup_WrapData(self->ptr.owner_id, idprop, group);
     }
   }
 
@@ -4937,7 +4940,7 @@ static PyObject *pyrna_struct_pop(BPy_StructRNA *self, PyObject *args)
     idprop = IDP_GetPropertyFromGroup(group, key);
 
     if (idprop) {
-      PyObject *ret = BPy_IDGroup_WrapData(self->ptr.id.data, idprop, group);
+      PyObject *ret = BPy_IDGroup_WrapData(self->ptr.owner_id, idprop, group);
       IDP_RemoveFromGroup(group, idprop);
       return ret;
     }
@@ -5787,7 +5790,7 @@ static PyObject *pyrna_param_to_py(PointerRNA *ptr, PropertyRNA *prop, void *dat
              * and will break if a function returns a pointer from
              * another ID block, watch this! - it should at least be
              * easy to debug since they are all ID's */
-            RNA_pointer_create(ptr->id.data, ptype, *(void **)data, &newptr);
+            RNA_pointer_create(ptr->owner_id, ptype, *(void **)data, &newptr);
           }
         }
 
@@ -5903,7 +5906,7 @@ static PyObject *pyrna_func_call(BPy_FunctionRNA *self, PyObject *args, PyObject
   /* include the ID pointer for pyrna_param_to_py() so we can include the
    * ID pointer on return values, this only works when returned values have
    * the same ID as the functions. */
-  RNA_pointer_create(self_ptr->id.data, &RNA_Function, self_func, &funcptr);
+  RNA_pointer_create(self_ptr->owner_id, &RNA_Function, self_func, &funcptr);
 
   pyargs_len = PyTuple_GET_SIZE(args);
   pykw_len = kw ? PyDict_Size(kw) : 0;
@@ -6987,7 +6990,7 @@ static void pyrna_subtype_set_rna(PyObject *newclass, StructRNA *srna)
 
   /* Add staticmethods and classmethods. */
   {
-    const PointerRNA func_ptr = {{NULL}, srna, NULL};
+    const PointerRNA func_ptr = {NULL, srna, NULL};
     const ListBase *lb;
     Link *link;
 
@@ -7285,8 +7288,8 @@ PyObject *pyrna_struct_CreatePyObject(PointerRNA *ptr)
   // PyC_ObSpit("NewStructRNA: ", (PyObject *)pyrna);
 
 #ifdef USE_PYRNA_INVALIDATE_WEAKREF
-  if (ptr->id.data) {
-    id_weakref_pool_add(ptr->id.data, (BPy_DummyPointerRNA *)pyrna);
+  if (ptr->owner_id) {
+    id_weakref_pool_add(ptr->owner_id, (BPy_DummyPointerRNA *)pyrna);
   }
 #endif
   return (PyObject *)pyrna;
@@ -7334,8 +7337,8 @@ PyObject *pyrna_prop_CreatePyObject(PointerRNA *ptr, PropertyRNA *prop)
   pyrna->prop = prop;
 
 #ifdef USE_PYRNA_INVALIDATE_WEAKREF
-  if (ptr->id.data) {
-    id_weakref_pool_add(ptr->id.data, (BPy_DummyPointerRNA *)pyrna);
+  if (ptr->owner_id) {
+    id_weakref_pool_add(ptr->owner_id, (BPy_DummyPointerRNA *)pyrna);
   }
 #endif
 
@@ -7358,7 +7361,7 @@ PyObject *pyrna_id_CreatePyObject(ID *id)
 bool pyrna_id_FromPyObject(PyObject *obj, ID **id)
 {
   if (pyrna_id_CheckPyObject(obj)) {
-    *id = ((BPy_StructRNA *)obj)->ptr.id.data;
+    *id = ((BPy_StructRNA *)obj)->ptr.owner_id;
     return true;
   }
   else {
@@ -8557,9 +8560,10 @@ static PyObject *pyrna_register_class(PyObject *UNUSED(self), PyObject *py_class
   }
 
   if (PyDict_GetItem(((PyTypeObject *)py_class)->tp_dict, bpy_intern_str_bl_rna)) {
-    PyErr_SetString(PyExc_ValueError,
-                    "register_class(...): "
-                    "already registered as a subclass");
+    PyErr_Format(PyExc_ValueError,
+                 "register_class(...): "
+                 "already registered as a subclass '%.200s'",
+                 ((PyTypeObject *)py_class)->tp_name);
     return NULL;
   }
 
diff --git a/source/blender/python/intern/bpy_rna_anim.c b/source/blender/python/intern/bpy_rna_anim.c
index 7a3499d0295..5e535d0e3ce 100644
--- a/source/blender/python/intern/bpy_rna_anim.c
+++ b/source/blender/python/intern/bpy_rna_anim.c
@@ -88,7 +88,7 @@ static int pyrna_struct_anim_args_parse_ex(PointerRNA *ptr,
                    path);
       return -1;
     }
-    else if (ptr->id.data != r_ptr.id.data) {
+    else if (ptr->owner_id != r_ptr.owner_id) {
       PyErr_Format(PyExc_ValueError, "%.200s path spans ID blocks", error_prefix, path);
       return -1;
     }
@@ -370,7 +370,7 @@ PyObject *pyrna_struct_keyframe_insert(BPy_StructRNA *self, PyObject *args, PyOb
     return PyBool_FromLong(result);
   }
   else {
-    ID *id = self->ptr.id.data;
+    ID *id = self->ptr.owner_id;
     ReportList reports;
     short result;
 
@@ -451,7 +451,7 @@ PyObject *pyrna_struct_keyframe_delete(BPy_StructRNA *self, PyObject *args, PyOb
     }
 
     if (prop) {
-      ID *id = ptr.id.data;
+      ID *id = ptr.owner_id;
       NlaStrip *strip = (NlaStrip *)ptr.data;
       FCurve *fcu = list_find_fcurve(&strip->fcurves, RNA_property_identifier(prop), index);
 
@@ -502,7 +502,7 @@ PyObject *pyrna_struct_keyframe_delete(BPy_StructRNA *self, PyObject *args, PyOb
     BKE_reports_init(&reports, RPT_STORE);
 
     result = delete_keyframe(
-        G.main, &reports, (ID *)self->ptr.id.data, NULL, group_name, path_full, index, cfra, 0);
+        G.main, &reports, (ID *)self->ptr.owner_id, NULL, group_name, path_full, index, cfra, 0);
     MEM_freeN((void *)path_full);
 
     if (BPy_reports_to_error(&reports, PyExc_RuntimeError, true) == -1) {
@@ -548,7 +548,7 @@ PyObject *pyrna_struct_driver_add(BPy_StructRNA *self, PyObject *args)
     BKE_reports_init(&reports, RPT_STORE);
 
     result = ANIM_add_driver(&reports,
-                             (ID *)self->ptr.id.data,
+                             (ID *)self->ptr.owner_id,
                              path_full,
                              index,
                              CREATEDRIVER_WITH_FMODIFIER,
@@ -559,7 +559,7 @@ PyObject *pyrna_struct_driver_add(BPy_StructRNA *self, PyObject *args)
     }
 
     if (result) {
-      ID *id = self->ptr.id.data;
+      ID *id = self->ptr.owner_id;
       AnimData *adt = BKE_animdata_from_id(id);
       FCurve *fcu;
 
@@ -629,7 +629,7 @@ PyObject *pyrna_struct_driver_remove(BPy_StructRNA *self, PyObject *args)
 
     BKE_reports_init(&reports, RPT_STORE);
 
-    result = ANIM_remove_driver(&reports, (ID *)self->ptr.id.data, path_full, index, 0);
+    result = ANIM_remove_driver(&reports, (ID *)self->ptr.owner_id, path_full, index, 0);
 
     if (path != path_full) {
       MEM_freeN((void *)path_full);
diff --git a/source/blender/python/intern/bpy_rna_driver.c b/source/blender/python/intern/bpy_rna_driver.c
index e9e8d05aa74..a8d8252b231 100644
--- a/source/blender/python/intern/bpy_rna_driver.c
+++ b/source/blender/python/intern/bpy_rna_driver.c
@@ -91,7 +91,7 @@ bool pyrna_driver_is_equal_anim_rna(const PathResolvedRNA *anim_rna, const PyObj
     const PointerRNA *ptr_a = &anim_rna->ptr;
     const PointerRNA *ptr_b = &(((const BPy_StructRNA *)py_anim_rna)->ptr);
 
-    if ((ptr_a->id.data == ptr_b->id.data) && (ptr_a->type == ptr_b->type) &&
+    if ((ptr_a->owner_id == ptr_b->owner_id) && (ptr_a->type == ptr_b->type) &&
         (ptr_a->data == ptr_b->data)) {
       return true;
     }
diff --git a/source/blender/python/mathutils/mathutils_Quaternion.c b/source/blender/python/mathutils/mathutils_Quaternion.c
index 23fd65319a6..267971408bf 100644
--- a/source/blender/python/mathutils/mathutils_Quaternion.c
+++ b/source/blender/python/mathutils/mathutils_Quaternion.c
@@ -1015,7 +1015,7 @@ static PyObject *Quaternion_matmul(PyObject *q1, PyObject *q2)
   return NULL;
 }
 /*------------------------obj @= obj------------------------------
- * inplace quaternion multiplication */
+ * in-place quaternion multiplication */
 static PyObject *Quaternion_imatmul(PyObject *q1, PyObject *q2)
 {
   float quat[QUAT_SIZE];
diff --git a/source/blender/python/mathutils/mathutils_Vector.c b/source/blender/python/mathutils/mathutils_Vector.c
index 490a1d9dd76..aa7cbadde14 100644
--- a/source/blender/python/mathutils/mathutils_Vector.c
+++ b/source/blender/python/mathutils/mathutils_Vector.c
@@ -786,10 +786,8 @@ static PyObject *Vector_to_track_quat(VectorObject *self, PyObject *args)
     return NULL;
   }
 
-  /*
-   * flip vector around, since vectoquat expect a vector from target to tracking object
-   * and the python function expects the inverse (a vector to the target).
-   */
+  /* Flip vector around, since #vec_to_quat expect a vector from target to tracking object
+   * and the python function expects the inverse (a vector to the target). */
   negate_v3_v3(vec, self->vec);
 
   vec_to_quat(quat, vec, track, up);
diff --git a/source/blender/python/mathutils/mathutils_geometry.c b/source/blender/python/mathutils/mathutils_geometry.c
index a6dded4ee8b..13d36e5af91 100644
--- a/source/blender/python/mathutils/mathutils_geometry.c
+++ b/source/blender/python/mathutils/mathutils_geometry.c
@@ -1213,7 +1213,7 @@ static PyObject *M_Geometry_tessellate_polygon(PyObject *UNUSED(self), PyObject
   PyObject *polyLine, *polyVec;
   int i, len_polylines, len_polypoints, ls_error = 0;
 
-  /* display listbase */
+  /* Display #ListBase. */
   ListBase dispbase = {NULL, NULL};
   DispList *dl;
   float *fp; /*pointer to the array of malloced dl->verts to set the points from the vectors */
@@ -1299,7 +1299,7 @@ static PyObject *M_Geometry_tessellate_polygon(PyObject *UNUSED(self), PyObject
     BKE_displist_fill(&dispbase, &dispbase, NULL, false);
 
     /* The faces are stored in a new DisplayList
-     * that's added to the head of the listbase */
+     * that's added to the head of the #ListBase. */
     dl = dispbase.first;
 
     tri_list = PyList_New(dl->parts);
@@ -1510,7 +1510,7 @@ static PyObject *M_Geometry_convex_hull_2d(PyObject *UNUSED(self), PyObject *poi
 
 /* Return a PyObject that is a list of lists, using the flattened list array
  * to fill values, with start_table and len_table giving the start index
- * and length of the toplevel_len sublists
+ * and length of the toplevel_len sub-lists.
  */
 static PyObject *list_of_lists_from_arrays(int *array,
                                            int *start_table,
diff --git a/source/blender/render/extern/include/RE_engine.h b/source/blender/render/extern/include/RE_engine.h
index dee8dbfaa24..12e9123b5cb 100644
--- a/source/blender/render/extern/include/RE_engine.h
+++ b/source/blender/render/extern/include/RE_engine.h
@@ -33,7 +33,6 @@
 
 struct BakePixel;
 struct Depsgraph;
-struct IDProperty;
 struct Main;
 struct Object;
 struct Render;
diff --git a/source/blender/render/extern/include/RE_pipeline.h b/source/blender/render/extern/include/RE_pipeline.h
index 93b85b6b96a..dc18bac541f 100644
--- a/source/blender/render/extern/include/RE_pipeline.h
+++ b/source/blender/render/extern/include/RE_pipeline.h
@@ -28,11 +28,9 @@
 #include "DNA_vec_types.h"
 #include "DEG_depsgraph.h"
 
-struct Depsgraph;
 struct Image;
 struct ImageFormatData;
 struct Main;
-struct NodeBlurData;
 struct Object;
 struct RenderData;
 struct RenderResult;
@@ -41,7 +39,6 @@ struct Scene;
 struct StampData;
 struct ViewLayer;
 struct bMovieHandle;
-struct bNodeTree;
 
 /* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ */
 /* this include is what is exposed of render to outside world */
@@ -256,7 +253,7 @@ void RE_SetView(struct Render *re, float mat[4][4]);
 /* get current view and window transform */
 void RE_GetViewPlane(struct Render *re, rctf *r_viewplane, rcti *r_disprect);
 
-/* set the render threads based on the commandline and autothreads setting */
+/* set the render threads based on the command-line and autothreads setting */
 void RE_init_threadcount(Render *re);
 
 bool RE_WriteRenderViewsImage(struct ReportList *reports,
diff --git a/source/blender/render/extern/include/RE_render_ext.h b/source/blender/render/extern/include/RE_render_ext.h
index 6771eb53abd..bca3b749192 100644
--- a/source/blender/render/extern/include/RE_render_ext.h
+++ b/source/blender/render/extern/include/RE_render_ext.h
@@ -31,9 +31,6 @@
 struct Depsgraph;
 struct ImagePool;
 struct MTex;
-struct Render;
-struct Scene;
-struct ViewLayer;
 
 /* render_texture.c */
 /* used by particle.c, effect.c, editmesh_modes.c and brush.c, returns 1 if rgb, 0 otherwise */
diff --git a/source/blender/render/extern/include/RE_shader_ext.h b/source/blender/render/extern/include/RE_shader_ext.h
index c5e3575d1f9..792b2b7e071 100644
--- a/source/blender/render/extern/include/RE_shader_ext.h
+++ b/source/blender/render/extern/include/RE_shader_ext.h
@@ -41,10 +41,8 @@ typedef struct BakeImBufuserData {
 } BakeImBufuserData;
 
 /* node shaders... */
-struct ImBuf;
 struct ImagePool;
 struct MTex;
-struct Object;
 struct Tex;
 
 /* this one uses nodes */
diff --git a/source/blender/render/intern/source/bake_api.c b/source/blender/render/intern/source/bake_api.c
index c61c9b4dd7a..4e5a83555eb 100644
--- a/source/blender/render/intern/source/bake_api.c
+++ b/source/blender/render/intern/source/bake_api.c
@@ -700,7 +700,7 @@ void RE_bake_pixels_populate(Mesh *me,
   }
   else {
     int uv_id = CustomData_get_named_layer(&me->ldata, CD_MLOOPUV, uv_layer);
-    mloopuv = CustomData_get_layer_n(&me->ldata, CD_MTFACE, uv_id);
+    mloopuv = CustomData_get_layer_n(&me->ldata, CD_MLOOPUV, uv_id);
   }
 
   if (mloopuv == NULL) {
diff --git a/source/blender/render/intern/source/imagetexture.c b/source/blender/render/intern/source/imagetexture.c
index e9175d8d024..9672184cec8 100644
--- a/source/blender/render/intern/source/imagetexture.c
+++ b/source/blender/render/intern/source/imagetexture.c
@@ -267,7 +267,7 @@ int imagewrap(Tex *tex,
 
   if (texres->nor) {
     if (tex->imaflag & TEX_NORMALMAP) {
-      /* qdn: normal from color
+      /* Normal from color:
        * The invert of the red channel is to make
        * the normal map compliant with the outside world.
        * It needs to be done because in Blender
diff --git a/source/blender/render/intern/source/pipeline.c b/source/blender/render/intern/source/pipeline.c
index 6284fa2d119..b4d0c2147f2 100644
--- a/source/blender/render/intern/source/pipeline.c
+++ b/source/blender/render/intern/source/pipeline.c
@@ -1992,8 +1992,8 @@ static int render_initialize_from_main(Render *re,
   winx = (rd->size * rd->xsch) / 100;
   winy = (rd->size * rd->ysch) / 100;
 
-  /* We always render smaller part, inserting it in larger image is compositor bizz,
-   * it uses disprect for it. */
+  /* We always render smaller part, inserting it in larger image is compositor business,
+   * it uses 'disprect' for it. */
   if (scene->r.mode & R_BORDER) {
     disprect.xmin = rd->border.xmin * winx;
     disprect.xmax = rd->border.xmax * winx;
diff --git a/source/blender/windowmanager/WM_api.h b/source/blender/windowmanager/WM_api.h
index cf87afa2a34..c7b18adf9b1 100644
--- a/source/blender/windowmanager/WM_api.h
+++ b/source/blender/windowmanager/WM_api.h
@@ -53,19 +53,15 @@ struct PropertyRNA;
 struct ScrArea;
 struct ViewLayer;
 struct bContext;
-struct bToolRef_Runtime;
 struct rcti;
 struct wmDrag;
 struct wmDropBox;
 struct wmEvent;
-struct wmEventHandler;
 struct wmEventHandler_Keymap;
 struct wmEventHandler_UI;
 struct wmGenericUserData;
 struct wmGesture;
 struct wmJob;
-struct wmMsgSubscribeKey;
-struct wmMsgSubscribeValue;
 struct wmOperator;
 struct wmOperatorType;
 struct wmPaintCursor;
@@ -162,7 +158,8 @@ enum {
   WM_WINDOW_RENDER = 1,
   WM_WINDOW_USERPREFS,
   WM_WINDOW_DRIVERS,
-  // WM_WINDOW_FILESEL // UNUSED
+  WM_WINDOW_INFO,
+  WM_WINDOW_FILESEL,
 };
 
 struct wmWindow *WM_window_open(struct bContext *C, const struct rcti *rect);
@@ -496,6 +493,8 @@ bool WM_operator_properties_checker_interval_test(const struct CheckerIntervalPa
 #define WM_FILESEL_FILENAME (1 << 2)
 #define WM_FILESEL_FILEPATH (1 << 3)
 #define WM_FILESEL_FILES (1 << 4)
+/* Show the properties sidebar by default. */
+#define WM_FILESEL_SHOW_PROPS (1 << 5)
 
 /* operator as a python command (resultuing string must be freed) */
 char *WM_operator_pystring_ex(struct bContext *C,
diff --git a/source/blender/windowmanager/WM_types.h b/source/blender/windowmanager/WM_types.h
index 195822efd7e..7fdbf79248b 100644
--- a/source/blender/windowmanager/WM_types.h
+++ b/source/blender/windowmanager/WM_types.h
@@ -117,7 +117,6 @@ struct ID;
 struct ImBuf;
 struct bContext;
 struct wmEvent;
-struct wmMsgBus;
 struct wmOperator;
 struct wmWindowManager;
 
diff --git a/source/blender/windowmanager/gizmo/intern/wm_gizmo_intern.h b/source/blender/windowmanager/gizmo/intern/wm_gizmo_intern.h
index f0771437518..00df6edef22 100644
--- a/source/blender/windowmanager/gizmo/intern/wm_gizmo_intern.h
+++ b/source/blender/windowmanager/gizmo/intern/wm_gizmo_intern.h
@@ -22,8 +22,6 @@
 #define __WM_GIZMO_INTERN_H__
 
 struct BLI_Buffer;
-struct GHashIterator;
-struct GizmoGeomInfo;
 struct wmGizmoMap;
 struct wmKeyConfig;
 
diff --git a/source/blender/windowmanager/gizmo/intern/wm_gizmo_target_props.c b/source/blender/windowmanager/gizmo/intern/wm_gizmo_target_props.c
index f5ae5427ac7..094fdf3f514 100644
--- a/source/blender/windowmanager/gizmo/intern/wm_gizmo_target_props.c
+++ b/source/blender/windowmanager/gizmo/intern/wm_gizmo_target_props.c
@@ -84,6 +84,7 @@ void WM_gizmo_target_property_def_rna_ptr(wmGizmo *gz,
 
   /* if gizmo evokes an operator we cannot use it for property manipulation */
   BLI_assert(gz->op_data == NULL);
+  BLI_assert(prop != NULL);
 
   gz_prop->type = gz_prop_type;
 
@@ -101,6 +102,9 @@ void WM_gizmo_target_property_def_rna(
 {
   const wmGizmoPropertyType *gz_prop_type = WM_gizmotype_target_property_find(gz->type, idname);
   PropertyRNA *prop = RNA_struct_find_property(ptr, propname);
+  if (prop == NULL) {
+    RNA_warning("%s: %s.%s not found", __func__, RNA_struct_identifier(ptr->type), propname);
+  }
   WM_gizmo_target_property_def_rna_ptr(gz, gz_prop_type, ptr, prop, index);
 }
 
diff --git a/source/blender/windowmanager/gizmo/wm_gizmo_fn.h b/source/blender/windowmanager/gizmo/wm_gizmo_fn.h
index 242b0f6de80..58b58fa01d0 100644
--- a/source/blender/windowmanager/gizmo/wm_gizmo_fn.h
+++ b/source/blender/windowmanager/gizmo/wm_gizmo_fn.h
@@ -25,6 +25,8 @@
 
 #include "BLI_compiler_attrs.h"
 
+struct wmMsgBus;
+
 /* wmGizmoGroup */
 typedef bool (*wmGizmoGroupFnPoll)(const struct bContext *,
                                    struct wmGizmoGroupType *) ATTR_WARN_UNUSED_RESULT;
diff --git a/source/blender/windowmanager/gizmo/wm_gizmo_wmapi.h b/source/blender/windowmanager/gizmo/wm_gizmo_wmapi.h
index 33c0d305e59..cc9ccc5f4bb 100644
--- a/source/blender/windowmanager/gizmo/wm_gizmo_wmapi.h
+++ b/source/blender/windowmanager/gizmo/wm_gizmo_wmapi.h
@@ -31,11 +31,9 @@
 #ifndef __WM_GIZMO_WMAPI_H__
 #define __WM_GIZMO_WMAPI_H__
 
-struct wmEventHandler;
 struct wmEventHandler_Gizmo;
 struct wmEventHandler_Op;
 struct wmGizmoMap;
-struct wmOperator;
 struct wmOperatorType;
 
 /* -------------------------------------------------------------------- */
diff --git a/source/blender/windowmanager/intern/wm_event_system.c b/source/blender/windowmanager/intern/wm_event_system.c
index 832bf1e9b6d..7192b25a7d5 100644
--- a/source/blender/windowmanager/intern/wm_event_system.c
+++ b/source/blender/windowmanager/intern/wm_event_system.c
@@ -2259,7 +2259,7 @@ static int wm_handler_operator_call(bContext *C,
 
     if (ot && wm_operator_check_locked_interface(C, ot)) {
       bool use_last_properties = true;
-      PointerRNA tool_properties = {{0}};
+      PointerRNA tool_properties = {0};
 
       bToolRef *keymap_tool = ((handler_base->type == WM_HANDLER_TYPE_KEYMAP) ?
                                    ((wmEventHandler_Keymap *)handler_base)->keymap_tool :
@@ -2339,46 +2339,36 @@ static int wm_handler_fileselect_do(bContext *C,
 
   switch (val) {
     case EVT_FILESELECT_FULL_OPEN: {
-      ScrArea *sa;
+      wmWindow *win = CTX_wm_window(C);
+      const int sizex = 1020 * UI_DPI_FAC;
+      const int sizey = 600 * UI_DPI_FAC;
+
+      if (WM_window_open_temp(C,
+                              WM_window_pixels_x(win) / 2,
+                              WM_window_pixels_y(win) / 2,
+                              sizex,
+                              sizey,
+                              WM_WINDOW_FILESEL) != NULL) {
+        ScrArea *area = CTX_wm_area(C);
+        ARegion *region_header = BKE_area_find_region_type(area, RGN_TYPE_HEADER);
 
-      /* sa can be null when window A is active, but mouse is over window B
-       * in this case, open file select in original window A. Also don't
-       * use global areas. */
-      if (handler->context.area == NULL || ED_area_is_global(handler->context.area)) {
-        bScreen *screen = CTX_wm_screen(C);
-        sa = (ScrArea *)screen->areabase.first;
-      }
-      else {
-        sa = handler->context.area;
-      }
+        BLI_assert(area->spacetype == SPACE_FILE);
 
-      if (sa->full) {
-        /* ensure the first area becomes the file browser, because the second one is the small
-         * top (info-)area which might be too small (in fullscreens we have max two areas) */
-        if (sa->prev) {
-          sa = sa->prev;
-        }
-        ED_area_newspace(C, sa, SPACE_FILE, true); /* 'sa' is modified in-place */
-        /* we already had a fullscreen here -> mark new space as a stacked fullscreen */
-        sa->flag |= (AREA_FLAG_STACKED_FULLSCREEN | AREA_FLAG_TEMP_TYPE);
-      }
-      else if (sa->spacetype == SPACE_FILE) {
-        sa = ED_screen_state_toggle(C, CTX_wm_window(C), sa, SCREENMAXIMIZED);
+        region_header->flag |= RGN_FLAG_HIDDEN;
+        /* Header on bottom, AZone triangle to toggle header looks misplaced at the top */
+        region_header->alignment = RGN_ALIGN_BOTTOM;
+
+        /* settings for filebrowser, sfile is not operator owner but sends events */
+        sfile = (SpaceFile *)area->spacedata.first;
+        sfile->op = handler->op;
+
+        ED_fileselect_set_params(sfile);
       }
       else {
-        sa = ED_screen_full_newspace(C, sa, SPACE_FILE); /* sets context */
+        BKE_report(&wm->reports, RPT_ERROR, "Failed to open window!");
+        return OPERATOR_CANCELLED;
       }
 
-      /* note, getting the 'sa' back from the context causes a nasty bug where the newly created
-       * 'sa' != CTX_wm_area(C). removed the line below and set 'sa' in the 'if' above */
-      /* sa = CTX_wm_area(C); */
-
-      /* settings for filebrowser, sfile is not operator owner but sends events */
-      sfile = (SpaceFile *)sa->spacedata.first;
-      sfile->op = handler->op;
-
-      ED_fileselect_set_params(sfile);
-
       action = WM_HANDLER_BREAK;
       break;
     }
@@ -2390,14 +2380,27 @@ static int wm_handler_fileselect_do(bContext *C,
       BLI_remlink(handlers, handler);
 
       if (val != EVT_FILESELECT_EXTERNAL_CANCEL) {
-        ScrArea *sa = CTX_wm_area(C);
+        for (wmWindow *win = wm->windows.first; win; win = win->next) {
+          if (WM_window_is_temp_screen(win)) {
+            bScreen *screen = WM_window_get_active_screen(win);
+            ScrArea *file_sa = screen->areabase.first;
 
-        if (sa->full) {
-          ED_screen_full_prevspace(C, sa);
-        }
-        /* user may have left fullscreen */
-        else {
-          ED_area_prevspace(C, sa);
+            BLI_assert(file_sa->spacetype == SPACE_FILE);
+
+            if (BLI_listbase_is_single(&file_sa->spacedata)) {
+              wmWindow *ctx_win = CTX_wm_window(C);
+              wm_window_close(C, wm, win);
+              CTX_wm_window_set(C, ctx_win);  // wm_window_close() NULLs.
+            }
+            else if (file_sa->full) {
+              ED_screen_full_prevspace(C, file_sa);
+            }
+            else {
+              ED_area_prevspace(C, file_sa);
+            }
+
+            break;
+          }
         }
       }
 
diff --git a/source/blender/windowmanager/intern/wm_files.c b/source/blender/windowmanager/intern/wm_files.c
index 189c4f18613..af841f20620 100644
--- a/source/blender/windowmanager/intern/wm_files.c
+++ b/source/blender/windowmanager/intern/wm_files.c
@@ -77,6 +77,7 @@
 #include "BKE_context.h"
 #include "BKE_global.h"
 #include "BKE_idprop.h"
+#include "BKE_library_override.h"
 #include "BKE_main.h"
 #include "BKE_packedFile.h"
 #include "BKE_report.h"
@@ -100,6 +101,7 @@
 #include "ED_datafiles.h"
 #include "ED_fileselect.h"
 #include "ED_image.h"
+#include "ED_outliner.h"
 #include "ED_screen.h"
 #include "ED_view3d.h"
 #include "ED_util.h"
@@ -1203,6 +1205,8 @@ static ImBuf *blend_file_thumb(const bContext *C,
   /* will be scaled down, but gives some nice oversampling */
   ImBuf *ibuf;
   BlendThumbnail *thumb;
+  wmWindowManager *wm = CTX_wm_manager(C);
+  wmWindow *windrawable_old = wm->windrawable;
   char err_out[256] = "unknown";
 
   /* screen if no camera found */
@@ -1236,6 +1240,9 @@ static ImBuf *blend_file_thumb(const bContext *C,
   /* gets scaled to BLEN_THUMB_SIZE */
   Depsgraph *depsgraph = CTX_data_ensure_evaluated_depsgraph(C);
 
+  /* Offscreen drawing requires a drawable window context. */
+  wm_window_make_drawable(wm, CTX_wm_window(C));
+
   if (scene->camera) {
     ibuf = ED_view3d_draw_offscreen_imbuf_simple(depsgraph,
                                                  scene,
@@ -1268,6 +1275,14 @@ static ImBuf *blend_file_thumb(const bContext *C,
                                           err_out);
   }
 
+  /* Reset to old drawable. */
+  if (windrawable_old) {
+    wm_window_make_drawable(wm, windrawable_old);
+  }
+  else {
+    wm_window_clear_drawable(wm);
+  }
+
   if (ibuf) {
     float aspect = (scene->r.xsch * scene->r.xasp) / (scene->r.ysch * scene->r.yasp);
 
@@ -1351,10 +1366,13 @@ static bool wm_file_write(bContext *C, const char *filepath, int fileflags, Repo
     }
   }
 
-  /* Call pre-save callbacks befores writing preview,
+  /* Call pre-save callbacks before writing preview,
    * that way you can generate custom file thumbnail. */
   BLI_callback_exec(bmain, NULL, BLI_CB_EVT_SAVE_PRE);
 
+  /* Enforce full override check/generation on file save. */
+  BKE_main_override_library_operations_create(bmain, true);
+
   /* blend file thumbnail */
   /* Save before exit_editmode, otherwise derivedmeshes for shared data corrupt T27765. */
   /* Main now can store a '.blend' thumbnail, useful for background mode
@@ -1967,6 +1985,10 @@ static int wm_homefile_read_exec(bContext *C, wmOperator *op)
     }
   }
 
+  if (G.fileflags & G_FILE_NO_UI) {
+    ED_outliner_select_sync_from_all_tag(C);
+  }
+
   return OPERATOR_FINISHED;
 }
 
@@ -2222,6 +2244,9 @@ static int wm_open_mainfile__open(bContext *C, wmOperator *op)
   BKE_report_print_level_set(op->reports, RPT_WARNING);
 
   if (success) {
+    if (G.fileflags & G_FILE_NO_UI) {
+      ED_outliner_select_sync_from_all_tag(C);
+    }
     return OPERATOR_FINISHED;
   }
   else {
@@ -2310,7 +2335,7 @@ static void wm_open_mainfile_ui(bContext *UNUSED(C), wmOperator *op)
 
 void WM_OT_open_mainfile(wmOperatorType *ot)
 {
-  ot->name = "Open Blender File";
+  ot->name = "Open";
   ot->idname = "WM_OT_open_mainfile";
   ot->description = "Open a Blender file";
 
@@ -2492,7 +2517,7 @@ void WM_OT_recover_auto_save(wmOperatorType *ot)
                                  FILE_BLENDER,
                                  FILE_OPENFILE,
                                  WM_FILESEL_FILEPATH,
-                                 FILE_LONGDISPLAY,
+                                 FILE_HORIZONTALDISPLAY,
                                  FILE_SORT_TIME);
 }
 
@@ -2626,7 +2651,7 @@ void WM_OT_save_as_mainfile(wmOperatorType *ot)
 {
   PropertyRNA *prop;
 
-  ot->name = "Save As Blender File";
+  ot->name = "Save As";
   ot->idname = "WM_OT_save_as_mainfile";
   ot->description = "Save the current file in the desired location";
 
@@ -2832,6 +2857,12 @@ static uiBlock *block_create_autorun_warning(struct bContext *C,
   /* Buttons */
   uiBut *but;
   uiLayout *split = uiLayoutSplit(layout, 0.0f, true);
+  uiLayoutSetScaleY(split, 1.2f);
+
+  /* empty space */
+  col = uiLayoutColumn(split, false);
+  uiItemS(col);
+
   col = uiLayoutColumn(split, false);
 
   /* Allow reload if we have a saved file.
@@ -2872,12 +2903,9 @@ static uiBlock *block_create_autorun_warning(struct bContext *C,
                            TIP_("Enable scripts"));
     UI_but_func_set(but, wm_block_autorun_warning_enable_scripts, block, NULL);
   }
+  UI_but_drawflag_disable(but, UI_BUT_TEXT_LEFT);
 
-  /* empty space between buttons */
   col = uiLayoutColumn(split, false);
-  uiItemS(col);
-
-  col = uiLayoutColumn(split, 1);
   but = uiDefIconTextBut(block,
                          UI_BTYPE_BUT,
                          0,
@@ -2894,8 +2922,10 @@ static uiBlock *block_create_autorun_warning(struct bContext *C,
                          0,
                          TIP_("Continue using file without Python scripts"));
   UI_but_func_set(but, wm_block_autorun_warning_ignore, block, NULL);
+  UI_but_drawflag_disable(but, UI_BUT_TEXT_LEFT);
+  UI_but_flag_enable(but, UI_BUT_ACTIVE_DEFAULT);
 
-  UI_block_bounds_set_centered(block, 10);
+  UI_block_bounds_set_centered(block, 14 * U.dpi_fac);
 
   return block;
 }
@@ -2998,7 +3028,7 @@ static void wm_block_file_close_cancel_button(uiBlock *block, wmGenericCallback
 static void wm_block_file_close_discard_button(uiBlock *block, wmGenericCallback *post_action)
 {
   uiBut *but = uiDefIconTextBut(
-      block, UI_BTYPE_BUT, 0, 0, IFACE_("Discard Changes"), 0, 0, 0, UI_UNIT_Y, 0, 0, 0, 0, 0, "");
+      block, UI_BTYPE_BUT, 0, 0, IFACE_("Don't Save"), 0, 0, 0, UI_UNIT_Y, 0, 0, 0, 0, 0, "");
   UI_but_func_set(but, wm_block_file_close_discard, block, post_action);
   UI_but_drawflag_disable(but, UI_BUT_TEXT_LEFT);
 }
@@ -3018,8 +3048,28 @@ static uiBlock *block_create__close_file_dialog(struct bContext *C, struct ARegi
 {
   wmGenericCallback *post_action = (wmGenericCallback *)arg1;
   Main *bmain = CTX_data_main(C);
-
   uiStyle *style = UI_style_get();
+  uiFontStyle *fs = &style->widgetlabel;
+
+  /* Filename */
+  const char *blendfile_pathpath = BKE_main_blendfile_path(bmain);
+  char filename[FILE_MAX];
+  if (blendfile_pathpath[0] != '\0') {
+    BLI_split_file_part(blendfile_pathpath, filename, sizeof(filename));
+    BLI_path_extension_replace(filename, sizeof(filename), "");
+  }
+  else {
+    STRNCPY(filename, IFACE_("Untitled"));
+  }
+
+  /* Title */
+  char title[FILE_MAX + 100];
+  UI_text_clip_middle_ex(
+      fs, filename, U.widget_unit * 9, U.widget_unit * 2, sizeof(filename), '\0');
+  BLI_snprintf(title, sizeof(title), TIP_("Save changes to \"%s\" before closing?"), filename);
+  int title_width = MAX2(UI_fontstyle_string_width(fs, title), U.widget_unit * 22);
+
+  /* Create dialog */
   uiBlock *block = UI_block_begin(C, ar, close_file_dialog_name, UI_EMBOSS);
 
   UI_block_flag_enable(
@@ -3032,25 +3082,25 @@ static uiBlock *block_create__close_file_dialog(struct bContext *C, struct ARegi
                                      UI_LAYOUT_PANEL,
                                      10,
                                      2,
-                                     U.widget_unit * 24,
+                                     U.widget_unit * 2 + title_width,
                                      U.widget_unit * 6,
                                      0,
                                      style);
 
   /* Title */
-  bool blend_file_is_saved = BKE_main_blendfile_path(bmain)[0] != '\0';
-  if (blend_file_is_saved) {
-    uiItemL(layout, "This file has unsaved changes.", ICON_INFO);
-  }
-  else {
-    uiItemL(layout, "This file has not been saved yet.", ICON_INFO);
-  }
+  uiItemL(layout, title, ICON_ERROR);
 
   /* Image Saving */
   ReportList reports;
   BKE_reports_init(&reports, RPT_STORE);
   uint modified_images_count = ED_image_save_all_modified_info(C, &reports);
 
+  LISTBASE_FOREACH (Report *, report, &reports.list) {
+    uiLayout *row = uiLayoutRow(layout, false);
+    uiLayoutSetRedAlert(row, true);
+    uiItemL(row, report->message, ICON_CANCEL);
+  }
+
   if (modified_images_count > 0) {
     char message[64];
     BLI_snprintf(message,
@@ -3058,6 +3108,7 @@ static uiBlock *block_create__close_file_dialog(struct bContext *C, struct ARegi
                  (modified_images_count == 1) ? "Save %u modified image" :
                                                 "Save %u modified images",
                  modified_images_count);
+    uiItemS(layout);
     uiDefButBitC(block,
                  UI_BTYPE_CHECKBOX,
                  1,
@@ -3075,13 +3126,9 @@ static uiBlock *block_create__close_file_dialog(struct bContext *C, struct ARegi
                  "");
   }
 
-  LISTBASE_FOREACH (Report *, report, &reports.list) {
-    uiItemL(layout, report->message, ICON_ERROR);
-  }
-
   BKE_reports_clear(&reports);
 
-  uiItemL(layout, "", ICON_NONE);
+  uiItemS_ex(layout, 3.0f);
 
   /* Buttons */
 #ifdef _WIN32
@@ -3090,10 +3137,12 @@ static uiBlock *block_create__close_file_dialog(struct bContext *C, struct ARegi
   const bool windows_layout = false;
 #endif
 
-  uiLayout *split = uiLayoutSplit(layout, 0.0f, true);
-
   if (windows_layout) {
     /* Windows standard layout. */
+
+    uiLayout *split = uiLayoutSplit(layout, 0.18f, true);
+    uiLayoutSetScaleY(split, 1.2f);
+
     uiLayout *col = uiLayoutColumn(split, false);
     uiItemS(col);
 
@@ -3108,20 +3157,24 @@ static uiBlock *block_create__close_file_dialog(struct bContext *C, struct ARegi
   }
   else {
     /* macOS and Linux standard layout. */
-    uiLayout *col = uiLayoutColumn(split, false);
+
+    uiLayout *split = uiLayoutSplit(layout, 0.0f, true);
+    uiLayoutSetScaleY(split, 1.2f);
+
+    uiLayout *col = uiLayoutColumn(split, true);
     wm_block_file_close_discard_button(block, post_action);
 
-    col = uiLayoutColumn(split, false);
+    col = uiLayoutColumn(split, true);
     uiItemS(col);
 
-    col = uiLayoutColumn(split, false);
+    col = uiLayoutColumn(split, true);
     wm_block_file_close_cancel_button(block, post_action);
 
     col = uiLayoutColumn(split, false);
     wm_block_file_close_save_button(block, post_action);
   }
 
-  UI_block_bounds_set_centered(block, 10);
+  UI_block_bounds_set_centered(block, 14 * U.dpi_fac);
   return block;
 }
 
diff --git a/source/blender/windowmanager/intern/wm_files_link.c b/source/blender/windowmanager/intern/wm_files_link.c
index 5a6606984ba..2cd2cf0636a 100644
--- a/source/blender/windowmanager/intern/wm_files_link.c
+++ b/source/blender/windowmanager/intern/wm_files_link.c
@@ -561,7 +561,7 @@ static void wm_link_append_properties_common(wmOperatorType *ot, bool is_link)
 
 void WM_OT_link(wmOperatorType *ot)
 {
-  ot->name = "Link from Library";
+  ot->name = "Link";
   ot->idname = "WM_OT_link";
   ot->description = "Link from a Library .blend file";
 
@@ -585,7 +585,7 @@ void WM_OT_link(wmOperatorType *ot)
 
 void WM_OT_append(wmOperatorType *ot)
 {
-  ot->name = "Append from Library";
+  ot->name = "Append";
   ot->idname = "WM_OT_append";
   ot->description = "Append from a Library .blend file";
 
diff --git a/source/blender/windowmanager/intern/wm_init_exit.c b/source/blender/windowmanager/intern/wm_init_exit.c
index 29cb02888ac..d08f4e236ee 100644
--- a/source/blender/windowmanager/intern/wm_init_exit.c
+++ b/source/blender/windowmanager/intern/wm_init_exit.c
@@ -524,6 +524,7 @@ void WM_exit_ex(bContext *C, const bool do_python)
 
   BKE_addon_pref_type_free();
   BKE_keyconfig_pref_type_free();
+  BKE_material_gpencil_default_free();
 
   wm_operatortype_free();
   wm_dropbox_free();
diff --git a/source/blender/windowmanager/intern/wm_operator_props.c b/source/blender/windowmanager/intern/wm_operator_props.c
index fcb55d3f801..663e4adf06b 100644
--- a/source/blender/windowmanager/intern/wm_operator_props.c
+++ b/source/blender/windowmanager/intern/wm_operator_props.c
@@ -65,19 +65,27 @@ void WM_operator_properties_filesel(wmOperatorType *ot,
        0,
        "Default",
        "Automatically determine display type for files"},
-      {FILE_SHORTDISPLAY,
-       "LIST_SHORT",
-       ICON_SHORTDISPLAY,
+      {FILE_VERTICALDISPLAY,
+       "LIST_VERTICAL",
+       ICON_SHORTDISPLAY, /* Name of deprecated short list */
        "Short List",
        "Display files as short list"},
-      {FILE_LONGDISPLAY,
-       "LIST_LONG",
-       ICON_LONGDISPLAY,
+      {FILE_HORIZONTALDISPLAY,
+       "LIST_HORIZONTAL",
+       ICON_LONGDISPLAY, /* Name of deprecated long list */
        "Long List",
        "Display files as a detailed list"},
       {FILE_IMGDISPLAY, "THUMBNAIL", ICON_IMGDISPLAY, "Thumbnails", "Display files as thumbnails"},
       {0, NULL, 0, NULL, NULL},
   };
+  static const EnumPropertyItem file_action_types[] = {
+      {FILE_OPENFILE,
+       "OPENFILE",
+       0,
+       "Open",
+       "Use the file browser for opening files or a directory"},
+      {FILE_SAVE, "SAVE", 0, "Save", "Use the file browser for saving a file"},
+  };
 
   if (flag & WM_FILESEL_FILEPATH) {
     RNA_def_string_file_path(ot->srna, "filepath", NULL, FILE_MAX, "File Path", "Path to file");
@@ -99,6 +107,15 @@ void WM_operator_properties_filesel(wmOperatorType *ot,
     RNA_def_property_flag(prop, PROP_HIDDEN | PROP_SKIP_SAVE);
   }
 
+  if ((flag & WM_FILESEL_SHOW_PROPS) == 0) {
+    prop = RNA_def_boolean(ot->srna,
+                           "hide_props_region",
+                           true,
+                           "Hide Operator Properties",
+                           "Collapse the region displaying the operator settings");
+    RNA_def_property_flag(prop, PROP_HIDDEN | PROP_SKIP_SAVE);
+  }
+
   if (action == FILE_SAVE) {
     /* note, this is only used to check if we should highlight the filename area red when the
      * filepath is an existing file. */
@@ -186,6 +203,9 @@ void WM_operator_properties_filesel(wmOperatorType *ot,
   prop = RNA_def_enum(
       ot->srna, "sort_method", rna_enum_file_sort_items, sort, "File sorting mode", "");
   RNA_def_property_flag(prop, PROP_HIDDEN | PROP_SKIP_SAVE);
+
+  prop = RNA_def_enum(ot->srna, "action_type", file_action_types, action, "Action Type", "");
+  RNA_def_property_flag(prop, PROP_HIDDEN | PROP_SKIP_SAVE);
 }
 
 static void wm_operator_properties_select_action_ex(wmOperatorType *ot,
diff --git a/source/blender/windowmanager/intern/wm_operators.c b/source/blender/windowmanager/intern/wm_operators.c
index 29ecf798b3c..9657347a1c4 100644
--- a/source/blender/windowmanager/intern/wm_operators.c
+++ b/source/blender/windowmanager/intern/wm_operators.c
@@ -368,7 +368,7 @@ static const char *wm_context_member_from_ptr(bContext *C, const PointerRNA *ptr
       continue;
     }
 
-    if (ptr->id.data == ctx_item_ptr.id.data) {
+    if (ptr->owner_id == ctx_item_ptr.owner_id) {
       if ((ptr->data == ctx_item_ptr.data) && (ptr->type == ctx_item_ptr.type)) {
         /* found! */
         member_found = identifier;
@@ -402,13 +402,13 @@ static const char *wm_context_member_from_ptr(bContext *C, const PointerRNA *ptr
 {
   const char *member_id = NULL;
 
-  if (ptr->id.data) {
+  if (ptr->owner_id) {
 
 #  define CTX_TEST_PTR_ID(C, member, idptr) \
     { \
       const char *ctx_member = member; \
       PointerRNA ctx_item_ptr = CTX_data_pointer_get(C, ctx_member); \
-      if (ctx_item_ptr.id.data == idptr) { \
+      if (ctx_item_ptr.owner_id == idptr) { \
         member_id = ctx_member; \
         break; \
       } \
@@ -420,7 +420,7 @@ static const char *wm_context_member_from_ptr(bContext *C, const PointerRNA *ptr
       const char *ctx_member = member; \
       const char *ctx_member_full = member_full; \
       PointerRNA ctx_item_ptr = CTX_data_pointer_get(C, ctx_member); \
-      if (ctx_item_ptr.id.data && cast(ctx_item_ptr.id.data) == idptr) { \
+      if (ctx_item_ptr.owner_id && (ID *)cast(ctx_item_ptr.owner_id) == idptr) { \
         member_id = ctx_member_full; \
         break; \
       } \
@@ -447,19 +447,19 @@ static const char *wm_context_member_from_ptr(bContext *C, const PointerRNA *ptr
     } \
     (void)0
 
-    switch (GS(((ID *)ptr->id.data)->name)) {
+    switch (GS(ptr->owner_id->name)) {
       case ID_SCE: {
-        CTX_TEST_PTR_ID(C, "scene", ptr->id.data);
+        CTX_TEST_PTR_ID(C, "scene", ptr->owner_id);
         break;
       }
       case ID_OB: {
-        CTX_TEST_PTR_ID(C, "object", ptr->id.data);
+        CTX_TEST_PTR_ID(C, "object", ptr->owner_id);
         break;
       }
       /* from rna_Main_objects_new */
       case OB_DATA_SUPPORT_ID_CASE: {
 #  define ID_CAST_OBDATA(id_pt) (((Object *)(id_pt))->data)
-        CTX_TEST_PTR_ID_CAST(C, "object", "object.data", ID_CAST_OBDATA, ptr->id.data);
+        CTX_TEST_PTR_ID_CAST(C, "object", "object.data", ID_CAST_OBDATA, ptr->owner_id);
         break;
 #  undef ID_CAST_OBDATA
       }
@@ -467,18 +467,18 @@ static const char *wm_context_member_from_ptr(bContext *C, const PointerRNA *ptr
 #  define ID_CAST_OBMATACT(id_pt) \
     (give_current_material(((Object *)id_pt), ((Object *)id_pt)->actcol))
         CTX_TEST_PTR_ID_CAST(
-            C, "object", "object.active_material", ID_CAST_OBMATACT, ptr->id.data);
+            C, "object", "object.active_material", ID_CAST_OBMATACT, ptr->owner_id);
         break;
 #  undef ID_CAST_OBMATACT
       }
       case ID_WO: {
 #  define ID_CAST_SCENEWORLD(id_pt) (((Scene *)(id_pt))->world)
-        CTX_TEST_PTR_ID_CAST(C, "scene", "scene.world", ID_CAST_SCENEWORLD, ptr->id.data);
+        CTX_TEST_PTR_ID_CAST(C, "scene", "scene.world", ID_CAST_SCENEWORLD, ptr->owner_id);
         break;
 #  undef ID_CAST_SCENEWORLD
       }
       case ID_SCR: {
-        CTX_TEST_PTR_ID(C, "screen", ptr->id.data);
+        CTX_TEST_PTR_ID(C, "screen", ptr->owner_id);
 
         SpaceLink *space_data = CTX_wm_space_data(C);
 
@@ -539,7 +539,7 @@ char *WM_prop_pystring_assign(bContext *C, PointerRNA *ptr, PropertyRNA *prop, i
 
   if (lhs == NULL) {
     /* fallback to bpy.data.foo[id] if we dont find in the context */
-    lhs = RNA_path_full_property_py(ptr, prop, index);
+    lhs = RNA_path_full_property_py(CTX_data_main(C), ptr, prop, index);
   }
 
   if (!lhs) {
@@ -850,7 +850,7 @@ static uiBlock *wm_enum_search_menu(bContext *C, ARegion *ar, void *arg)
            NULL);
 
   /* Move it downwards, mouse over button. */
-  UI_block_bounds_set_popup(block, 6, (const int[2]){0, -UI_UNIT_Y});
+  UI_block_bounds_set_popup(block, 0.3f * U.widget_unit, (const int[2]){0, -UI_UNIT_Y});
 
   UI_but_focus_on_enter_event(win, but);
 
@@ -1111,7 +1111,7 @@ static uiBlock *wm_block_create_redo(bContext *C, ARegion *ar, void *arg_op)
 
   /* if register is not enabled, the operator gets freed on OPERATOR_FINISHED
    * ui_apply_but_funcs_after calls ED_undo_operator_repeate_cb and crashes */
-  assert(op->type->flag & OPTYPE_REGISTER);
+  BLI_assert(op->type->flag & OPTYPE_REGISTER);
 
   UI_block_func_handle_set(block, wm_block_redo_cb, arg_op);
   layout = UI_block_layout(
@@ -1126,18 +1126,15 @@ static uiBlock *wm_block_create_redo(bContext *C, ARegion *ar, void *arg_op)
   if (op->type->flag & OPTYPE_MACRO) {
     for (op = op->macro.first; op; op = op->next) {
       uiTemplateOperatorPropertyButs(
-          C, layout, op, UI_BUT_LABEL_ALIGN_SPLIT_COLUMN, UI_TEMPLATE_OP_PROPS_SHOW_TITLE);
-      if (op->next) {
-        uiItemS(layout);
-      }
+          C, layout, op, UI_BUT_LABEL_ALIGN_NONE, UI_TEMPLATE_OP_PROPS_SHOW_TITLE);
     }
   }
   else {
     uiTemplateOperatorPropertyButs(
-        C, layout, op, UI_BUT_LABEL_ALIGN_SPLIT_COLUMN, UI_TEMPLATE_OP_PROPS_SHOW_TITLE);
+        C, layout, op, UI_BUT_LABEL_ALIGN_NONE, UI_TEMPLATE_OP_PROPS_SHOW_TITLE);
   }
 
-  UI_block_bounds_set_popup(block, 4, NULL);
+  UI_block_bounds_set_popup(block, 6 * U.dpi_fac, NULL);
 
   return block;
 }
@@ -1217,7 +1214,8 @@ static uiBlock *wm_block_dialog_create(bContext *C, ARegion *ar, void *userData)
   }
 
   /* center around the mouse */
-  UI_block_bounds_set_popup(block, 4, (const int[2]){data->width / -2, data->height / 2});
+  UI_block_bounds_set_popup(
+      block, 6 * U.dpi_fac, (const int[2]){data->width / -2, data->height / 2});
 
   UI_block_active_only_flagged_buttons(C, ar, block);
 
@@ -1245,7 +1243,7 @@ static uiBlock *wm_operator_ui_create(bContext *C, ARegion *ar, void *userData)
 
   UI_block_func_set(block, NULL, NULL, NULL);
 
-  UI_block_bounds_set_popup(block, 4, NULL);
+  UI_block_bounds_set_popup(block, 6 * U.dpi_fac, NULL);
 
   UI_block_active_only_flagged_buttons(C, ar, block);
 
@@ -1516,7 +1514,7 @@ static uiBlock *wm_block_search_menu(bContext *C, ARegion *ar, void *userdata)
            NULL);
 
   /* Move it downwards, mouse over button. */
-  UI_block_bounds_set_popup(block, 6, (const int[2]){0, -UI_UNIT_Y});
+  UI_block_bounds_set_popup(block, 0.3f * U.widget_unit, (const int[2]){0, -UI_UNIT_Y});
 
   return block;
 }
@@ -3011,7 +3009,7 @@ static void WM_OT_redraw_timer(wmOperatorType *ot)
 /** \} */
 
 /* -------------------------------------------------------------------- */
-/** \name Reporet Memory Statistics
+/** \name Report Memory Statistics
  *
  * Use for testing/debugging.
  * \{ */
diff --git a/source/blender/windowmanager/intern/wm_splash_screen.c b/source/blender/windowmanager/intern/wm_splash_screen.c
index 8629997030f..d3f7661a008 100644
--- a/source/blender/windowmanager/intern/wm_splash_screen.c
+++ b/source/blender/windowmanager/intern/wm_splash_screen.c
@@ -113,31 +113,37 @@ static void wm_block_splash_add_label(uiBlock *block, const char *label, int x,
 static void wm_block_splash_add_labels(uiBlock *block, int x, int y)
 {
   /* Version number. */
-  const char *version_suffix = NULL;
+  const char *version_cycle = NULL;
   bool show_build_info = true;
 
   if (STREQ(STRINGIFY(BLENDER_VERSION_CYCLE), "alpha")) {
-    version_suffix = " Alpha";
+    version_cycle = " Alpha";
   }
   else if (STREQ(STRINGIFY(BLENDER_VERSION_CYCLE), "beta")) {
-    version_suffix = " Beta";
+    version_cycle = " Beta";
   }
   else if (STREQ(STRINGIFY(BLENDER_VERSION_CYCLE), "rc")) {
-    version_suffix = " Release Candidate";
+    version_cycle = " Release Candidate";
     show_build_info = false;
   }
   else if (STREQ(STRINGIFY(BLENDER_VERSION_CYCLE), "release")) {
-    version_suffix = STRINGIFY(BLENDER_VERSION_CHAR);
+    version_cycle = STRINGIFY(BLENDER_VERSION_CHAR);
     show_build_info = false;
   }
 
+  const char *version_cycle_number = "";
+  if (strlen(STRINGIFY(BLENDER_VERSION_CYCLE_NUMBER))) {
+    version_cycle_number = " " STRINGIFY(BLENDER_VERSION_CYCLE_NUMBER);
+  }
+
   char version_buf[256] = "\0";
   BLI_snprintf(version_buf,
                sizeof(version_buf),
-               "v %d.%d%s",
+               "v %d.%d%s%s",
                BLENDER_VERSION / 100,
                BLENDER_VERSION % 100,
-               version_suffix);
+               version_cycle,
+               version_cycle_number);
 
   wm_block_splash_add_label(block, version_buf, x, &y);
 
diff --git a/source/blender/windowmanager/intern/wm_window.c b/source/blender/windowmanager/intern/wm_window.c
index d17b8817691..91173804b18 100644
--- a/source/blender/windowmanager/intern/wm_window.c
+++ b/source/blender/windowmanager/intern/wm_window.c
@@ -482,8 +482,8 @@ void wm_window_title(wmWindowManager *wm, wmWindow *win)
     }
 
     /* Informs GHOST of unsaved changes, to set window modified visual indicator (macOS)
-     * and to give hint of unsaved changes for a user warning mechanism in case of OS
-     * application terminate request (e.g. OS Shortcut Alt+F4, Cmd+Q, (...), or session end). */
+     * and to give hint of unsaved changes for a user warning mechanism in case of OS application
+     * terminate request (e.g. OS Shortcut Alt+F4, Command+Q, (...), or session end). */
     GHOST_SetWindowModifiedState(win->ghostwin, (GHOST_TUns8)!wm->file_saved);
   }
 }
@@ -658,7 +658,7 @@ void wm_window_ghostwindows_ensure(wmWindowManager *wm)
 
   BLI_assert(G.background == false);
 
-  /* no commandline prefsize? then we set this.
+  /* No command-line prefsize? then we set this.
    * Note that these values will be used only
    * when there is no startup.blend yet.
    */
@@ -807,6 +807,7 @@ wmWindow *WM_window_open_temp(bContext *C, int x, int y, int sizex, int sizey, i
   ScrArea *sa;
   Scene *scene = CTX_data_scene(C);
   ViewLayer *view_layer = CTX_data_view_layer(C);
+  eSpace_Type space_type = SPACE_EMPTY;
   const char *title;
 
   /* convert to native OS window coordinates */
@@ -888,14 +889,24 @@ wmWindow *WM_window_open_temp(bContext *C, int x, int y, int sizex, int sizey, i
   CTX_wm_area_set(C, sa);
 
   if (type == WM_WINDOW_RENDER) {
-    ED_area_newspace(C, sa, SPACE_IMAGE, false);
+    space_type = SPACE_IMAGE;
   }
   else if (type == WM_WINDOW_DRIVERS) {
-    ED_area_newspace(C, sa, SPACE_GRAPH, false);
+    space_type = SPACE_GRAPH;
+  }
+  else if (type == WM_WINDOW_USERPREFS) {
+    space_type = SPACE_USERPREF;
+  }
+  else if (type == WM_WINDOW_FILESEL) {
+    space_type = SPACE_FILE;
+  }
+  else if (type == WM_WINDOW_INFO) {
+    space_type = SPACE_INFO;
   }
   else {
-    ED_area_newspace(C, sa, SPACE_USERPREF, false);
+    BLI_assert(false);
   }
+  ED_area_newspace(C, sa, space_type, false);
 
   ED_screen_change(C, screen);
   ED_screen_refresh(CTX_wm_manager(C), win); /* test scale */
@@ -917,6 +928,9 @@ wmWindow *WM_window_open_temp(bContext *C, int x, int y, int sizex, int sizey, i
   else if (sa->spacetype == SPACE_GRAPH) {
     title = IFACE_("Blender Drivers Editor");
   }
+  else if (sa->spacetype == SPACE_INFO) {
+    title = IFACE_("Blender Info Log");
+  }
   else {
     title = "Blender";
   }
diff --git a/source/blender/windowmanager/message_bus/intern/wm_message_bus_rna.c b/source/blender/windowmanager/message_bus/intern/wm_message_bus_rna.c
index 73758e2d98f..0c2ce9783ec 100644
--- a/source/blender/windowmanager/message_bus/intern/wm_message_bus_rna.c
+++ b/source/blender/windowmanager/message_bus/intern/wm_message_bus_rna.c
@@ -50,7 +50,7 @@ static uint wm_msg_rna_gset_hash(const void *key_p)
   //  printf("%s\n", RNA_struct_identifier(params->ptr.type));
   uint k = void_hash_uint(params->ptr.type);
   k ^= void_hash_uint(params->ptr.data);
-  k ^= void_hash_uint(params->ptr.id.data);
+  k ^= void_hash_uint(params->ptr.owner_id);
   k ^= void_hash_uint(params->prop);
   return k;
 }
@@ -59,7 +59,7 @@ static bool wm_msg_rna_gset_cmp(const void *key_a_p, const void *key_b_p)
   const wmMsgParams_RNA *params_a = &((const wmMsgSubscribeKey_RNA *)key_a_p)->msg.params;
   const wmMsgParams_RNA *params_b = &((const wmMsgSubscribeKey_RNA *)key_b_p)->msg.params;
   return !((params_a->ptr.type == params_b->ptr.type) &&
-           (params_a->ptr.id.data == params_b->ptr.id.data) &&
+           (params_a->ptr.owner_id == params_b->ptr.owner_id) &&
            (params_a->ptr.data == params_b->ptr.data) && (params_a->prop == params_b->prop));
 }
 static void wm_msg_rna_gset_key_free(void *key_p)
@@ -100,7 +100,7 @@ static void wm_msg_rna_update_by_id(struct wmMsgBus *mbus, ID *id_src, ID *id_ds
   while (BLI_gsetIterator_done(&gs_iter) == false) {
     wmMsgSubscribeKey_RNA *key = BLI_gsetIterator_getKey(&gs_iter);
     BLI_gsetIterator_step(&gs_iter);
-    if (key->msg.params.ptr.id.data == id_src) {
+    if (key->msg.params.ptr.owner_id == id_src) {
 
       /* GSet always needs updating since the key changes. */
       BLI_gset_remove(gs, key, NULL);
@@ -123,10 +123,10 @@ static void wm_msg_rna_update_by_id(struct wmMsgBus *mbus, ID *id_src, ID *id_ds
       if (BLI_listbase_is_empty(&key->head.values)) {
         /* Remove, no reason to keep. */
       }
-      else if (key->msg.params.ptr.data == key->msg.params.ptr.id.data) {
+      else if (key->msg.params.ptr.data == key->msg.params.ptr.owner_id) {
         /* Simple, just update the ID. */
         key->msg.params.ptr.data = id_dst;
-        key->msg.params.ptr.id.data = id_dst;
+        key->msg.params.ptr.owner_id = id_dst;
         remove = false;
       }
       else {
@@ -177,7 +177,7 @@ static void wm_msg_rna_remove_by_id(struct wmMsgBus *mbus, const ID *id)
   while (BLI_gsetIterator_done(&gs_iter) == false) {
     wmMsgSubscribeKey_RNA *key = BLI_gsetIterator_getKey(&gs_iter);
     BLI_gsetIterator_step(&gs_iter);
-    if (key->msg.params.ptr.id.data == id) {
+    if (key->msg.params.ptr.owner_id == id) {
       /* Clear here so we can decrement 'messages_tag_count'. */
       for (wmMsgSubscribeValueLink *msg_lnk = key->head.values.first, *msg_lnk_next; msg_lnk;
            msg_lnk = msg_lnk_next) {
@@ -226,7 +226,7 @@ void WM_msg_publish_rna_params(struct wmMsgBus *mbus, const wmMsgParams_RNA *msg
   CLOG_INFO(WM_LOG_MSGBUS_PUB,
             2,
             "rna(id='%s', %s.%s)",
-            msg_key_params->ptr.id.data ? ((ID *)msg_key_params->ptr.id.data)->name : none,
+            msg_key_params->ptr.owner_id ? ((ID *)msg_key_params->ptr.owner_id)->name : none,
             msg_key_params->ptr.type ? RNA_struct_identifier(msg_key_params->ptr.type) : none,
             msg_key_params->prop ? RNA_property_identifier((PropertyRNA *)msg_key_params->prop) :
                                    none);
@@ -237,7 +237,7 @@ void WM_msg_publish_rna_params(struct wmMsgBus *mbus, const wmMsgParams_RNA *msg
 
   /* Support anonymous subscribers, this may be some extra overhead
    * but we want to be able to be more ambiguous. */
-  if (msg_key_params->ptr.id.data || msg_key_params->ptr.data) {
+  if (msg_key_params->ptr.owner_id || msg_key_params->ptr.data) {
     wmMsgParams_RNA msg_key_params_anon = *msg_key_params;
 
     /* We might want to enable this later? */
@@ -250,7 +250,7 @@ void WM_msg_publish_rna_params(struct wmMsgBus *mbus, const wmMsgParams_RNA *msg
       msg_key_params_anon.prop = msg_key_params->prop;
     }
 
-    msg_key_params_anon.ptr.id.data = NULL;
+    msg_key_params_anon.ptr.owner_id = NULL;
     msg_key_params_anon.ptr.data = NULL;
     if ((key = WM_msg_lookup_rna(mbus, &msg_key_params_anon))) {
       WM_msg_publish_with_key(mbus, &key->head);
@@ -292,7 +292,7 @@ void WM_msg_subscribe_rna_params(struct wmMsgBus *mbus,
   CLOG_INFO(WM_LOG_MSGBUS_SUB,
             3,
             "rna(id='%s', %s.%s, info='%s')",
-            msg_key_params->ptr.id.data ? ((ID *)msg_key_params->ptr.id.data)->name : none,
+            msg_key_params->ptr.owner_id ? ((ID *)msg_key_params->ptr.owner_id)->name : none,
             msg_key_params->ptr.type ? RNA_struct_identifier(msg_key_params->ptr.type) : none,
             msg_key_params->prop ? RNA_property_identifier((PropertyRNA *)msg_key_params->prop) :
                                    none,
@@ -303,7 +303,7 @@ void WM_msg_subscribe_rna_params(struct wmMsgBus *mbus,
 
   if (msg_val_params->is_persistent) {
     if (msg_key->msg.params.data_path == NULL) {
-      if (msg_key->msg.params.ptr.data != msg_key->msg.params.ptr.id.data) {
+      if (msg_key->msg.params.ptr.data != msg_key->msg.params.ptr.owner_id) {
         /* We assume prop type can't change. */
         msg_key->msg.params.data_path = RNA_path_from_ID_to_struct(&msg_key->msg.params.ptr);
       }
@@ -339,14 +339,14 @@ void WM_msg_subscribe_ID(struct wmMsgBus *mbus,
                          const wmMsgSubscribeValue *msg_val_params,
                          const char *id_repr)
 {
-  wmMsgParams_RNA msg_key_params = {{{NULL}}};
+  wmMsgParams_RNA msg_key_params = {NULL};
   RNA_id_pointer_create(id, &msg_key_params.ptr);
   WM_msg_subscribe_rna_params(mbus, &msg_key_params, msg_val_params, id_repr);
 }
 
 void WM_msg_publish_ID(struct wmMsgBus *mbus, ID *id)
 {
-  wmMsgParams_RNA msg_key_params = {{{NULL}}};
+  wmMsgParams_RNA msg_key_params = {NULL};
   RNA_id_pointer_create(id, &msg_key_params.ptr);
   WM_msg_publish_rna_params(mbus, &msg_key_params);
 }
diff --git a/source/blender/windowmanager/message_bus/wm_message_bus.h b/source/blender/windowmanager/message_bus/wm_message_bus.h
index 4a8bce52992..4ee087593ac 100644
--- a/source/blender/windowmanager/message_bus/wm_message_bus.h
+++ b/source/blender/windowmanager/message_bus/wm_message_bus.h
@@ -23,7 +23,6 @@
 
 #include <stdio.h>
 
-struct GSet;
 struct ID;
 struct bContext;
 struct wmMsg;
@@ -173,7 +172,7 @@ void WM_msg_subscribe_static(struct wmMsgBus *mbus,
 /* wm_message_bus_rna.c */
 
 typedef struct wmMsgParams_RNA {
-  /** when #PointerRNA.data & id.data are NULL. match against all. */
+  /** when #PointerRNA.data & owner_id are NULL. match against all. */
   PointerRNA ptr;
   /** when NULL, match against any property. */
   const PropertyRNA *prop;
@@ -233,7 +232,7 @@ void WM_msg_publish_ID(struct wmMsgBus *mbus, struct ID *id);
 
 #define WM_msg_publish_rna_prop(mbus, id_, data_, type_, prop_) \
   { \
-    wmMsgParams_RNA msg_key_params_ = {{{0}}}; \
+    wmMsgParams_RNA msg_key_params_ = {{0}}; \
     _WM_MESSAGE_EXTERN_BEGIN; \
     extern PropertyRNA rna_##type_##_##prop_; \
     _WM_MESSAGE_EXTERN_END; \
@@ -244,7 +243,7 @@ void WM_msg_publish_ID(struct wmMsgBus *mbus, struct ID *id);
   ((void)0)
 #define WM_msg_subscribe_rna_prop(mbus, id_, data_, type_, prop_, value) \
   { \
-    wmMsgParams_RNA msg_key_params_ = {{{0}}}; \
+    wmMsgParams_RNA msg_key_params_ = {{0}}; \
     _WM_MESSAGE_EXTERN_BEGIN; \
     extern PropertyRNA rna_##type_##_##prop_; \
     _WM_MESSAGE_EXTERN_END; \
diff --git a/source/blender/windowmanager/wm_window.h b/source/blender/windowmanager/wm_window.h
index 90c580818c9..b0329f7415e 100644
--- a/source/blender/windowmanager/wm_window.h
+++ b/source/blender/windowmanager/wm_window.h
@@ -24,10 +24,6 @@
 #ifndef __WM_WINDOW_H__
 #define __WM_WINDOW_H__
 
-struct EnumPropertyItem;
-struct PointerRNA;
-struct PropertyRNA;
-struct wmEvent;
 struct wmOperator;
 
 /* *************** internal api ************** */
diff --git a/source/creator/CMakeLists.txt b/source/creator/CMakeLists.txt
index f74fd57252d..1cd0a6661f8 100644
--- a/source/creator/CMakeLists.txt
+++ b/source/creator/CMakeLists.txt
@@ -820,18 +820,6 @@ elseif(WIN32)
     )
   endif()
 
-  if(NOT CMAKE_CL_64)
-    install(
-      FILES ${LIBDIR}/thumbhandler/lib/BlendThumb.dll
-      DESTINATION "."
-    )
-  endif()
-
-  install( # x86 builds can run on x64 Windows, so this is required at all times
-    FILES ${LIBDIR}/thumbhandler/lib/BlendThumb64.dll
-    DESTINATION "."
-  )
-
   install(
     FILES
       ${CMAKE_SOURCE_DIR}/release/windows/batch/blender_debug_gpu.cmd
diff --git a/source/creator/creator.c b/source/creator/creator.c
index 9eb1578001c..6fd472d24c5 100644
--- a/source/creator/creator.c
+++ b/source/creator/creator.c
@@ -400,8 +400,8 @@ int main(int argc,
   /* background render uses this font too */
   BKE_vfont_builtin_register(datatoc_bfont_pfb, datatoc_bfont_pfb_size);
 
-  /* Initialize ffmpeg if built in, also needed for bg mode if videos are
-   * rendered via ffmpeg */
+  /* Initialize ffmpeg if built in, also needed for background-mode if videos are
+   * rendered via ffmpeg. */
   BKE_sound_init_once();
 
   init_def_material();
diff --git a/tests/python/CMakeLists.txt b/tests/python/CMakeLists.txt
index 4793010885e..2125694701c 100644
--- a/tests/python/CMakeLists.txt
+++ b/tests/python/CMakeLists.txt
@@ -497,61 +497,68 @@ if(WITH_CYCLES OR WITH_OPENGL_RENDER_TESTS)
   elseif(NOT EXISTS "${TEST_SRC_DIR}/render/shader")
     MESSAGE(STATUS "Disabling render tests because tests folder does not exist at ${TEST_SRC_DIR}")
   else()
-    macro(add_cycles_render_test subject)
-      if(WITH_CYCLES)
+    set(render_tests
+      bsdf
+      denoise
+      denoise_animation
+      displacement
+      hair
+      image_colorspace
+      image_data_types
+      image_mapping
+      image_texture_limit
+      integrator
+      light
+      mesh
+      motion_blur
+      render_layer
+      reports
+      shader
+      shadow_catcher
+      sss
+      volume
+    )
+
+    if(WITH_OPENGL_RENDER_TESTS)
+      list(APPEND render_tests grease_pencil)
+    endif()
+
+    if(WITH_CYCLES)
+      foreach(render_test bake;${render_tests})
         add_python_test(
-          cycles_${subject}
+          cycles_${render_test}
           ${CMAKE_CURRENT_LIST_DIR}/cycles_render_tests.py
           -blender "$<TARGET_FILE:blender>"
-          -testdir "${TEST_SRC_DIR}/render/${subject}"
+          -testdir "${TEST_SRC_DIR}/render/${render_test}"
           -idiff "${OPENIMAGEIO_IDIFF}"
           -outdir "${TEST_OUT_DIR}/cycles"
         )
-      endif()
+      endforeach()
+    endif()
 
-      if(WITH_OPENGL_RENDER_TESTS AND (NOT ${subject} MATCHES "bake"))
+    if(WITH_OPENGL_RENDER_TESTS)
+      foreach(render_test ${render_tests})
         add_python_test(
-          eevee_${subject}_test
+          eevee_${render_test}_test
           ${CMAKE_CURRENT_LIST_DIR}/eevee_render_tests.py
           -blender "$<TARGET_FILE:blender>"
-          -testdir "${TEST_SRC_DIR}/render/${subject}"
+          -testdir "${TEST_SRC_DIR}/render/${render_test}"
           -idiff "${OPENIMAGEIO_IDIFF}"
           -outdir "${TEST_OUT_DIR}/eevee"
         )
+      endforeach()
 
+      foreach(render_test ${render_tests})
         add_python_test(
-          workbench_${subject}_test
+          workbench_${render_test}_test
           ${CMAKE_CURRENT_LIST_DIR}/workbench_render_tests.py
           -blender "$<TARGET_FILE:blender>"
-          -testdir "${TEST_SRC_DIR}/render/${subject}"
+          -testdir "${TEST_SRC_DIR}/render/${render_test}"
           -idiff "${OPENIMAGEIO_IDIFF}"
           -outdir "${TEST_OUT_DIR}/workbench"
         )
-      endif()
-    endmacro()
-    add_cycles_render_test(bake)
-    add_cycles_render_test(bsdf)
-    add_cycles_render_test(denoise)
-    add_cycles_render_test(denoise_animation)
-    add_cycles_render_test(displacement)
-    if(WITH_OPENGL_RENDER_TESTS)
-      add_cycles_render_test(grease_pencil)
+      endforeach()
     endif()
-    add_cycles_render_test(hair)
-    add_cycles_render_test(image_colorspace)
-    add_cycles_render_test(image_data_types)
-    add_cycles_render_test(image_mapping)
-    add_cycles_render_test(image_texture_limit)
-    add_cycles_render_test(integrator)
-    add_cycles_render_test(light)
-    add_cycles_render_test(mesh)
-    add_cycles_render_test(motion_blur)
-    add_cycles_render_test(render_layer)
-    add_cycles_render_test(reports)
-    add_cycles_render_test(shader)
-    add_cycles_render_test(shadow_catcher)
-    add_cycles_render_test(sss)
-    add_cycles_render_test(volume)
   endif()
 endif()
 
diff --git a/tests/python/modules/render_report.py b/tests/python/modules/render_report.py
index 6e3223140ce..58eae834879 100755
--- a/tests/python/modules/render_report.py
+++ b/tests/python/modules/render_report.py
@@ -431,6 +431,7 @@ class Report:
 
             # Run process
             crash = False
+            output = None
             try:
                 output = subprocess.check_output(command)
             except subprocess.CalledProcessError as e:
@@ -440,7 +441,8 @@ class Report:
 
             if verbose:
                 print(" ".join(command))
-                print(output.decode("utf-8"))
+                if output:
+                    print(output.decode("utf-8"))
 
             # Detect missing filepaths and consider those errors
             for filepath, output_filepath in zip(remaining_filepaths[:], output_filepaths):