Cycles: update build configurations to include CUDA sm_35 architecture. When using

a compiler older than CUDA 5.0 it will give a warning and skip this architecture.

4 files changed, 22 insertions, 14 deletions

diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 1f96ed0ae83..27c54af5153 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -304,7 +304,7 @@ public:
 			}
 		}
 		else {
-			/* CUDA 4.x */
+			/* CUDA 5.x */
 			if(major == 1) {
 				/* sm_1x */
 				arch_flags = "--maxrregcount=24 --opencc-options -OPT:Olimit=0 --use_fast_math";
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index 5e9dd15b812..2fa1393e935 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -130,6 +130,12 @@ if(WITH_CYCLES_CUDA_BINARIES)
 	string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\2" CUDA_VERSION_MINOR ${NVCC_OUT})
 	set(CUDA_VERSION "${CUDA_VERSION_MAJOR}${CUDA_VERSION_MINOR}")
 
+	# warn for other versions
+	if(CUDA_VERSION MATCHES "50")
+	else()
+		message(WARNING "CUDA version ${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR} detected, build may succeed but only CUDA 5.0 is officially supported")
+	endif()
+
 	# build for each arch
 	set(cuda_sources kernel.cu ${SRC_HEADERS} ${SRC_SVM_HEADERS} ${SRC_CLOSURE_HEADERS} ${SRC_UTIL_HEADERS})
 	set(cuda_cubins)
@@ -139,12 +145,6 @@ if(WITH_CYCLES_CUDA_BINARIES)
 
 		set(cuda_version_flags "-D__KERNEL_CUDA_VERSION__=${CUDA_VERSION}")
 
-		# warn for other versions
-		if(CUDA_VERSION MATCHES "50")
-		else()
-			message(STATUS "CUDA version ${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR} detected, build may succeed but only CUDA 5.0 is officially supported")
-		endif()
-
 		# build flags depending on CUDA version and arch
 		if(CUDA_VERSION LESS 50)
 			# CUDA 4.x
@@ -176,13 +176,17 @@ if(WITH_CYCLES_CUDA_BINARIES)
 			set(cuda_math_flags "--use_fast_math")
 		endif()
 		
-		add_custom_command(
-			OUTPUT ${cuda_cubin}
-			COMMAND ${CUDA_NVCC_EXECUTABLE} -arch=${arch} -m${CUDA_BITS} --cubin ${CMAKE_CURRENT_SOURCE_DIR}/kernel.cu -o ${CMAKE_CURRENT_BINARY_DIR}/${cuda_cubin} --ptxas-options="-v" ${cuda_arch_flags} ${cuda_version_flags} ${cuda_math_flags} -I${CMAKE_CURRENT_SOURCE_DIR}/../util -I${CMAKE_CURRENT_SOURCE_DIR}/svm -DCCL_NAMESPACE_BEGIN= -DCCL_NAMESPACE_END= -DNVCC
-			DEPENDS ${cuda_sources})
+		if(CUDA_VERSION LESS 50 AND ${arch} MATCHES "sm_35")
+			message(WARNING "Can't build kernel for CUDA sm_35 architecture, skipping")
+		else()
+			add_custom_command(
+				OUTPUT ${cuda_cubin}
+				COMMAND ${CUDA_NVCC_EXECUTABLE} -arch=${arch} -m${CUDA_BITS} --cubin ${CMAKE_CURRENT_SOURCE_DIR}/kernel.cu -o ${CMAKE_CURRENT_BINARY_DIR}/${cuda_cubin} --ptxas-options="-v" ${cuda_arch_flags} ${cuda_version_flags} ${cuda_math_flags} -I${CMAKE_CURRENT_SOURCE_DIR}/../util -I${CMAKE_CURRENT_SOURCE_DIR}/svm -DCCL_NAMESPACE_BEGIN= -DCCL_NAMESPACE_END= -DNVCC
+				DEPENDS ${cuda_sources})
 
-		delayed_install("${CMAKE_CURRENT_BINARY_DIR}" "${cuda_cubin}" ${CYCLES_INSTALL_PATH}/lib)
-		list(APPEND cuda_cubins ${cuda_cubin})
+			delayed_install("${CMAKE_CURRENT_BINARY_DIR}" "${cuda_cubin}" ${CYCLES_INSTALL_PATH}/lib)
+			list(APPEND cuda_cubins ${cuda_cubin})
+		endif()
 	endforeach()
 
 	add_custom_target(cycles_kernel_cuda ALL DEPENDS ${cuda_cubins})
diff --git a/intern/cycles/kernel/SConscript b/intern/cycles/kernel/SConscript
index 353ec1ce9d8..6459c3ed183 100644
--- a/intern/cycles/kernel/SConscript
+++ b/intern/cycles/kernel/SConscript
@@ -88,6 +88,10 @@ if env['WITH_BF_CYCLES_CUDA_BINARIES']:
 
 		# build flags depending on CUDA version and arch
         if cuda_version < 50:
+            if arch == "sm_35":
+                print("Can't build kernel for CUDA sm_35 architecture, skipping")
+                continue
+
             # CUDA 4.x
             if arch.startswith("sm_1"):
                 # sm_1x
diff --git a/intern/cycles/util/util_types.h b/intern/cycles/util/util_types.h
index 4fd1e9d8807..a67c55acf3d 100644
--- a/intern/cycles/util/util_types.h
+++ b/intern/cycles/util/util_types.h
@@ -517,7 +517,7 @@ __device_inline const __m128 shuffle_swap(const __m128& a, const shuffle_swap_t&
 
 #else
 
-/* somewhat slower version for SSE3 */
+/* somewhat slower version for SSE2 */
 typedef int shuffle_swap_t;
 
 __device_inline const shuffle_swap_t shuffle_swap_identity(void)