Merge commit 'b393759a80d17853a26f6cd4eb451709745badd9'

48 files changed, 395 insertions, 138 deletions

diff --git a/luajit/Makefile b/luajit/Makefile
index 900b43b..b23b648 100644
--- a/luajit/Makefile
+++ b/luajit/Makefile
@@ -15,7 +15,7 @@
 
 MAJVER=  2
 MINVER=  0
-RELVER=  2
+RELVER=  3
 VERSION= $(MAJVER).$(MINVER).$(RELVER)
 ABIVER=  5.1
 
diff --git a/luajit/README b/luajit/README
index b9351f6..902a6d2 100644
--- a/luajit/README
+++ b/luajit/README
@@ -1,4 +1,4 @@
-README for LuaJIT 2.0.2
+README for LuaJIT 2.0.3
 -----------------------
 
 LuaJIT is a Just-In-Time (JIT) compiler for the Lua programming language.
diff --git a/luajit/doc/changes.html b/luajit/doc/changes.html
index f58cde8..9841229 100644
--- a/luajit/doc/changes.html
+++ b/luajit/doc/changes.html
@@ -63,7 +63,7 @@ div.major { max-width: 600px; padding: 1em; margin: 1em 0 1em 0; }
 <div id="main">
 <p>
 This is a list of changes between the released versions of LuaJIT.<br>
-The current <span style="color: #0000c0;">stable version</span> is <strong>LuaJIT&nbsp;2.0.2</strong>.<br>
+The current <span style="color: #0000c0;">stable version</span> is <strong>LuaJIT&nbsp;2.0.3</strong>.<br>
 </p>
 <p>
 Please check the
@@ -72,6 +72,42 @@ to see whether newer versions are available.
 </p>
 
 <div class="major" style="background: #d0d0ff;">
+<h2 id="LuaJIT-2.0.3">LuaJIT 2.0.3 &mdash; 2014-03-12</h2>
+<ul>
+<li>Add PS4 port.</li>
+<li>Add support for multilib distro builds.</li>
+<li>Fix OSX build.</li>
+<li>Fix MinGW build.</li>
+<li>Fix Xbox 360 build.</li>
+<li>Improve ULOAD forwarding for open upvalues.</li>
+<li>Fix GC steps threshold handling when called by JIT-compiled code.</li>
+<li>Fix argument checks for <tt>math.deg()</tt> and <tt>math.rad()</tt>.</li>
+<li>Fix <tt>jit.flush(func|true)</tt>.</li>
+<li>Respect <tt>jit.off(func)</tt> when returning to a function, too.</li>
+<li>Fix compilation of <tt>string.byte(s, nil, n)</tt>.</li>
+<li>Fix line number for relocated bytecode after closure fixup</li>
+<li>Fix frame traversal for backtraces.</li>
+<li>Fix ABC elimination.</li>
+<li>Fix handling of redundant PHIs.</li>
+<li>Fix snapshot restore for exit to function header.</li>
+<li>Fix type punning alias analysis for constified pointers</li>
+<li>Fix call unroll checks in the presence of metamethod frames.</li>
+<li>Fix initial maxslot for down-recursive traces.</li>
+<li>Prevent BASE register coalescing if parent uses <tt>IR_RETF</tt>.</li>
+<li>Don't purge modified function from stack slots in <tt>BC_RET</tt>.</li>
+<li>Fix recording of <tt>BC_VARG</tt>.</li>
+<li>Don't access dangling reference to reallocated IR.</li>
+<li>Fix frame depth display for bytecode dump in <tt>-jdump</tt>.</li>
+<li>ARM: Fix register allocation when rematerializing FPRs.</li>
+<li>x64: Fix store to upvalue for lightuserdata values.</li>
+<li>FFI: Add missing GC steps for callback argument conversions.</li>
+<li>FFI: Properly unload loaded DLLs.</li>
+<li>FFI: Fix argument checks for <tt>ffi.string()</tt>.</li>
+<li>FFI/x64: Fix passing of vector arguments to calls.</li>
+<li>FFI: Rehash finalizer table after GC cycle, if needed.</li>
+<li>FFI: Fix <tt>cts-&gt;L</tt> for cdata unsinking in snapshot restore.</li>
+</ul>
+
 <h2 id="LuaJIT-2.0.2">LuaJIT 2.0.2 &mdash; 2013-06-03</h2>
 <ul>
 <li>Fix memory access check for fast string interning.</li>
diff --git a/luajit/doc/ext_ffi_tutorial.html b/luajit/doc/ext_ffi_tutorial.html
index 01fb899..8e4eaf7 100644
--- a/luajit/doc/ext_ffi_tutorial.html
+++ b/luajit/doc/ext_ffi_tutorial.html
@@ -564,7 +564,7 @@ Thus it's not helpful and actually counter-productive to cache
 individual C&nbsp;functions like this:
 </p>
 <pre class="code">
-local <b>funca</b>, <b>funcb</b> = ffi.C.funcb, ffi.C.funcb -- <span style="color:#c00000;">Not helpful!</span>
+local <b>funca</b>, <b>funcb</b> = ffi.C.funca, ffi.C.funcb -- <span style="color:#c00000;">Not helpful!</span>
 local function foo(x, n)
   for i=1,n do <b>funcb</b>(<b>funca</b>(x, i), 1) end
 end
diff --git a/luajit/doc/install.html b/luajit/doc/install.html
index baf1469..f816315 100644
--- a/luajit/doc/install.html
+++ b/luajit/doc/install.html
@@ -127,7 +127,7 @@ operating systems, CPUs and compilers:
 <tr class="odd">
 <td class="compatcpu"><a href="#cross2">ARMv5+<br>ARM9E+</a></td>
 <td class="compatos">GCC 4.2+</td>
-<td class="compatos">GCC 4.2+</td>
+<td class="compatos">GCC 4.2+<br>PSP2 (<a href="#psvita">PS VITA</a>)</td>
 <td class="compatos">GCC 4.2+</td>
 <td class="compatos compatno">&nbsp;</td>
 </tr>
@@ -188,8 +188,8 @@ open a terminal window and change to this directory. Now unpack the archive
 and change to the newly created directory:
 </p>
 <pre class="code">
-tar zxf LuaJIT-2.0.2.tar.gz
-cd LuaJIT-2.0.2</pre>
+tar zxf LuaJIT-2.0.3.tar.gz
+cd LuaJIT-2.0.3</pre>
 <h3>Building LuaJIT</h3>
 <p>
 The supplied Makefiles try to auto-detect the settings needed for your
@@ -458,41 +458,56 @@ ISDKF="-arch armv7 -isysroot $ISDK/SDKs/$ISDKVER"
 make HOST_CC="gcc -m32 -arch i386" CROSS=$ISDKP TARGET_FLAGS="$ISDKF" \
      TARGET_SYS=iOS
 </pre>
+
+<h3 id="consoles">Cross-compiling for consoles</h3>
 <p>
-You can cross-compile for <b id="ps3">PS3</b> using the PS3&nbsp;SDK from
-a Linux host or a Windows host (requires 32&nbsp;bit MinGW (GCC) on the host,
-too). Due to restrictions on consoles, the JIT compiler is disabled and
-only the fast interpreter is built:
+Building LuaJIT for consoles requires both a supported host compiler
+(x86 or x64) and a cross-compiler (to PPC or ARM) from the official
+console SDK.
 </p>
-<pre class="code">
-make HOST_CC="gcc -m32" CROSS=ppu-lv2-
-</pre>
 <p>
-You can cross-compile for <b id="ps4">PS4</b> from a Windows host using
-the PS4&nbsp;SDK (ORBIS) plus 64&nbsp;bit MSVC. Due to restrictions on
-consoles, the JIT compiler is disabled and only the fast interpreter
-is built.
+Due to restrictions on consoles, the JIT compiler is disabled and only
+the fast interpreter is built. This is still faster than plain Lua,
+but much slower than the JIT compiler. The FFI is disabled, too, since
+it's not very useful in such an environment.
 </p>
 <p>
-Open a "Visual Studio .NET Command Prompt" (64&nbsp;bit host compiler),
-<tt>cd</tt> to the directory where you've unpacked the sources and run
-the following commands. This builds a static library <tt>libluajit.a</tt>,
+The following commands build a static library <tt>libluajit.a</tt>,
 which can be linked against your game, just like the Lua library.
 </p>
+<p>
+To cross-compile for <b id="ps3">PS3</b> from a Linux host (requires
+32&nbsp;bit GCC, i.e. multilib Linux/x64) or a Windows host (requires
+32&nbsp;bit MinGW), run this command:
+</p>
+<pre class="code">
+make HOST_CC="gcc -m32" CROSS=ppu-lv2-
+</pre>
+<p>
+To cross-compile for <b id="ps4">PS4</b> from a Windows host,
+open a "Visual Studio .NET Command Prompt" (64&nbsp;bit host compiler),
+<tt>cd</tt> to the directory where you've unpacked the sources and
+run the following commands:
+</p>
 <pre class="code">
 cd src
 ps4build
 </pre>
 <p>
-You can cross-compile for <b id="xbox360">Xbox 360</b> using the
-Xbox&nbsp;360 SDK (MSVC + XEDK). Due to restrictions on consoles, the
-JIT compiler is disabled and only the fast interpreter is built.
+To cross-compile for <b id="psvita">PS Vita</b> from a Windows host,
+open a "Visual Studio .NET Command Prompt" (32&nbsp;bit host compiler),
+<tt>cd</tt> to the directory where you've unpacked the sources and
+run the following commands:
 </p>
+<pre class="code">
+cd src
+psvitabuild
+</pre>
 <p>
-Open a "Visual Studio .NET Command Prompt" (32&nbsp;bit host compiler),
+To cross-compile for <b id="xbox360">Xbox 360</b> from a Windows host,
+open a "Visual Studio .NET Command Prompt" (32&nbsp;bit host compiler),
 <tt>cd</tt> to the directory where you've unpacked the sources and run
-the following commands. This builds a static library <tt>luajit20.lib</tt>,
-which can be linked against your game, just like the Lua library.
+the following commands:
 </p>
 <pre class="code">
 cd src
diff --git a/luajit/doc/luajit.html b/luajit/doc/luajit.html
index 43af69c..ddbc016 100644
--- a/luajit/doc/luajit.html
+++ b/luajit/doc/luajit.html
@@ -38,6 +38,9 @@ table.os1 td {
 table.os2 td {
   color: #ffa040;
 }
+table.os3 td {
+  color: #40ffff;
+}
 table.compiler td {
   color: #2080ff;
   background: #62bf41;
@@ -158,7 +161,10 @@ LuaJIT is Copyright &copy; 2005-2014 Mike Pall, released under the
 <tr><td>Windows</td><td>Linux</td><td>BSD</td><td>OSX</td><td>POSIX</td></tr>
 </table>
 <table class="feature os os2">
-<tr><td><span style="font-size:90%;">Embedded</span></td><td>Android</td><td>iOS</td><td>PS3</td><td>PS4</td><td>Xbox 360</td></tr>
+<tr><td><span style="font-size:90%;">Embedded</span></td><td>Android</td><td>iOS</td></tr>
+</table>
+<table class="feature os os3">
+<tr><td>PS3</td><td>PS4</td><td>PS Vita</td><td>Xbox 360</td></tr>
 </table>
 <table class="feature compiler">
 <tr><td>GCC</td><td>CLANG<br>LLVM</td><td>MSVC</td></tr>
diff --git a/luajit/doc/running.html b/luajit/doc/running.html
index bbbe8be..c01c52d 100644
--- a/luajit/doc/running.html
+++ b/luajit/doc/running.html
@@ -186,7 +186,7 @@ itself. For a description of their options and output format, please
 read the comment block at the start of their source.
 They can be found in the <tt>lib</tt> directory of the source
 distribution or installed under the <tt>jit</tt> directory. By default
-this is <tt>/usr/local/share/luajit-2.0.2/jit</tt> on POSIX
+this is <tt>/usr/local/share/luajit-2.0.3/jit</tt> on POSIX
 systems.
 </p>
 
diff --git a/luajit/dynasm/dasm_x86.lua b/luajit/dynasm/dasm_x86.lua
index f3859e3..5f82f4f 100644
--- a/luajit/dynasm/dasm_x86.lua
+++ b/luajit/dynasm/dasm_x86.lua
@@ -1081,6 +1081,9 @@ local map_op = {
   btr_2 =	"mrqdw:0FB3Rm|miqdw:0FBA6mU",
   bts_2 =	"mrqdw:0FABRm|miqdw:0FBA5mU",
 
+  shld_3 =	"mriqdw:0FA4RmU|mrCqdw:0FA5Rm",
+  shrd_3 =	"mriqdw:0FACRmU|mrCqdw:0FADRm",
+
   rdtsc_0 =	"0F31", -- P1+
   cpuid_0 =	"0FA2", -- P1+
 
@@ -1114,6 +1117,9 @@ local map_op = {
   fucompp_0 =	"DAE9",
   fcompp_0 =	"DED9",
 
+  fldenv_1 =	"x.:D94m",
+  fnstenv_1 =	"x.:D96m",
+  fstenv_1 =	"x.:9BD96m",
   fldcw_1 =	"xw:nD95m",
   fstcw_1 =	"xw:n9BD97m",
   fnstcw_1 =	"xw:nD97m",
@@ -1189,6 +1195,8 @@ local map_op = {
   cvttps2dq_2 =	"rmo:F30F5BrM",
   cvttsd2si_2 =	"rr/do:F20F2CrM|rr/qo:|rx/dq:|rxq:",
   cvttss2si_2 =	"rr/do:F30F2CrM|rr/qo:|rxd:|rx/qd:",
+  fxsave_1 =	"x.:0FAE0m",
+  fxrstor_1 =	"x.:0FAE1m",
   ldmxcsr_1 =	"xd:0FAE2m",
   lfence_0 =	"0FAEE8",
   maskmovdqu_2 = "rro:660FF7rM",
diff --git a/luajit/etc/luajit.pc b/luajit/etc/luajit.pc
index 4823396..d3f0cf3 100644
--- a/luajit/etc/luajit.pc
+++ b/luajit/etc/luajit.pc
@@ -1,7 +1,7 @@
 # Package information for LuaJIT to be used by pkg-config.
 majver=2
 minver=0
-relver=2
+relver=3
 version=${majver}.${minver}.${relver}
 abiver=5.1
 
diff --git a/luajit/src/Makefile b/luajit/src/Makefile
index 72bc912..c558003 100644
--- a/luajit/src/Makefile
+++ b/luajit/src/Makefile
@@ -12,7 +12,7 @@
 
 MAJVER=  2
 MINVER=  0
-RELVER=  2
+RELVER=  3
 ABIVER=  5.1
 NODOTABIVER= 51
 
@@ -122,8 +122,10 @@ XCFLAGS=
 #
 # Use the system provided memory allocator (realloc) instead of the
 # bundled memory allocator. This is slower, but sometimes helpful for
-# debugging. It's helpful for Valgrind's memcheck tool, too. This option
-# cannot be enabled on x64, since the built-in allocator is mandatory.
+# debugging. This option cannot be enabled on x64, since realloc usually
+# doesn't return addresses in the right address range.
+# OTOH this option is mandatory for Valgrind's memcheck tool on x64 and
+# the only way to get useful results from it for all other architectures.
 #XCFLAGS+= -DLUAJIT_USE_SYSMALLOC
 #
 # This define is required to run LuaJIT under Valgrind. The Valgrind
diff --git a/luajit/src/Makefile.dep b/luajit/src/Makefile.dep
index 8ca3315..9e14d61 100644
--- a/luajit/src/Makefile.dep
+++ b/luajit/src/Makefile.dep
@@ -90,7 +90,7 @@ lj_ctype.o: lj_ctype.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_gc.h lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_ctype.h lj_ccallback.h
 lj_debug.o: lj_debug.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_err.h lj_errmsg.h lj_debug.h lj_str.h lj_tab.h lj_state.h lj_frame.h \
- lj_bc.h lj_jit.h lj_ir.h
+ lj_bc.h lj_vm.h lj_jit.h lj_ir.h
 lj_dispatch.o: lj_dispatch.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_err.h lj_errmsg.h lj_func.h lj_str.h lj_tab.h lj_meta.h lj_debug.h \
  lj_state.h lj_frame.h lj_bc.h lj_ff.h lj_ffdef.h lj_jit.h lj_ir.h \
diff --git a/luajit/src/host/buildvm_asm.c b/luajit/src/host/buildvm_asm.c
index 9b28b3b..079e9a8 100644
--- a/luajit/src/host/buildvm_asm.c
+++ b/luajit/src/host/buildvm_asm.c
@@ -286,7 +286,7 @@ void emit_asm(BuildCtx *ctx)
   fprintf(ctx->fp, "\n");
   switch (ctx->mode) {
   case BUILD_elfasm:
-#if !LJ_TARGET_PS3
+#if !(LJ_TARGET_PS3 || LJ_TARGET_PSVITA)
     fprintf(ctx->fp, "\t.section .note.GNU-stack,\"\"," ELFASM_PX "progbits\n");
 #endif
 #if LJ_TARGET_PPCSPE
diff --git a/luajit/src/jit/bc.lua b/luajit/src/jit/bc.lua
index 9192cb9..9df19c7 100644
--- a/luajit/src/jit/bc.lua
+++ b/luajit/src/jit/bc.lua
@@ -41,7 +41,7 @@
 
 -- Cache some library functions and objects.
 local jit = require("jit")
-assert(jit.version_num == 20002, "LuaJIT core/library version mismatch")
+assert(jit.version_num == 20003, "LuaJIT core/library version mismatch")
 local jutil = require("jit.util")
 local vmdef = require("jit.vmdef")
 local bit = require("bit")
diff --git a/luajit/src/jit/bcsave.lua b/luajit/src/jit/bcsave.lua
index 13b79db..122bb24 100644
--- a/luajit/src/jit/bcsave.lua
+++ b/luajit/src/jit/bcsave.lua
@@ -11,7 +11,7 @@
 ------------------------------------------------------------------------------
 
 local jit = require("jit")
-assert(jit.version_num == 20002, "LuaJIT core/library version mismatch")
+assert(jit.version_num == 20003, "LuaJIT core/library version mismatch")
 local bit = require("bit")
 
 -- Symbol name prefix for LuaJIT bytecode.
@@ -69,7 +69,7 @@ local map_arch = {
 
 local map_os = {
   linux = true, windows = true, osx = true, freebsd = true, netbsd = true,
-  openbsd = true, solaris = true,
+  openbsd = true, dragonfly = true, solaris = true,
 }
 
 local function checkarg(str, map, err)
diff --git a/luajit/src/jit/dump.lua b/luajit/src/jit/dump.lua
index 6a77f92..556ce88 100644
--- a/luajit/src/jit/dump.lua
+++ b/luajit/src/jit/dump.lua
@@ -55,7 +55,7 @@
 
 -- Cache some library functions and objects.
 local jit = require("jit")
-assert(jit.version_num == 20002, "LuaJIT core/library version mismatch")
+assert(jit.version_num == 20003, "LuaJIT core/library version mismatch")
 local jutil = require("jit.util")
 local vmdef = require("jit.vmdef")
 local funcinfo, funcbc = jutil.funcinfo, jutil.funcbc
diff --git a/luajit/src/jit/v.lua b/luajit/src/jit/v.lua
index b9e0eb5..197e67c 100644
--- a/luajit/src/jit/v.lua
+++ b/luajit/src/jit/v.lua
@@ -59,7 +59,7 @@
 
 -- Cache some library functions and objects.
 local jit = require("jit")
-assert(jit.version_num == 20002, "LuaJIT core/library version mismatch")
+assert(jit.version_num == 20003, "LuaJIT core/library version mismatch")
 local jutil = require("jit.util")
 local vmdef = require("jit.vmdef")
 local funcinfo, traceinfo = jutil.funcinfo, jutil.traceinfo
diff --git a/luajit/src/lib_aux.c b/luajit/src/lib_aux.c
index 1b01fe0..e88dc7c 100644
--- a/luajit/src/lib_aux.c
+++ b/luajit/src/lib_aux.c
@@ -302,7 +302,7 @@ static int panic(lua_State *L)
 
 #ifdef LUAJIT_USE_SYSMALLOC
 
-#if LJ_64
+#if LJ_64 && !defined(LUAJIT_USE_VALGRIND)
 #error "Must use builtin allocator for 64 bit target"
 #endif
 
diff --git a/luajit/src/lib_ffi.c b/luajit/src/lib_ffi.c
index e789ad6..2549678 100644
--- a/luajit/src/lib_ffi.c
+++ b/luajit/src/lib_ffi.c
@@ -136,6 +136,7 @@ static int ffi_index_meta(lua_State *L, CTState *cts, CType *ct, MMS mm)
 	return 0;
       }
     }
+    copyTV(L, base, L->top);
     tv = L->top-1;
   }
   return lj_meta_tailcall(L, tv);
diff --git a/luajit/src/lib_io.c b/luajit/src/lib_io.c
index 04f0f73..90f310b 100644
--- a/luajit/src/lib_io.c
+++ b/luajit/src/lib_io.c
@@ -426,7 +426,7 @@ LJLIB_CF(io_popen)
 LJLIB_CF(io_tmpfile)
 {
   IOFileUD *iof = io_file_new(L);
-#if LJ_TARGET_PS3 || LJ_TARGET_PS4
+#if LJ_TARGET_PS3 || LJ_TARGET_PS4 || LJ_TARGET_PSVITA
   iof->fp = NULL; errno = ENOSYS;
 #else
   iof->fp = tmpfile();
diff --git a/luajit/src/lib_os.c b/luajit/src/lib_os.c
index f62e8c8..713ec93 100644
--- a/luajit/src/lib_os.c
+++ b/luajit/src/lib_os.c
@@ -7,7 +7,6 @@
 */
 
 #include <errno.h>
-#include <locale.h>
 #include <time.h>
 
 #define lib_os_c
@@ -27,6 +26,10 @@
 #include <stdio.h>
 #endif
 
+#if !LJ_TARGET_PSVITA
+#include <locale.h>
+#endif
+
 /* ------------------------------------------------------------------------ */
 
 #define LJLIB_MODULE_os
@@ -70,7 +73,7 @@ LJLIB_CF(os_rename)
 
 LJLIB_CF(os_tmpname)
 {
-#if LJ_TARGET_PS3 || LJ_TARGET_PS4
+#if LJ_TARGET_PS3 || LJ_TARGET_PS4 || LJ_TARGET_PSVITA
   lj_err_caller(L, LJ_ERR_OSUNIQF);
   return 0;
 #else
@@ -254,6 +257,9 @@ LJLIB_CF(os_difftime)
 
 LJLIB_CF(os_setlocale)
 {
+#if LJ_TARGET_PSVITA
+  lua_pushliteral(L, "C");
+#else
   GCstr *s = lj_lib_optstr(L, 1);
   const char *str = s ? strdata(s) : NULL;
   int opt = lj_lib_checkopt(L, 2, 6,
@@ -265,6 +271,7 @@ LJLIB_CF(os_setlocale)
   else if (opt == 4) opt = LC_MONETARY;
   else if (opt == 6) opt = LC_ALL;
   lua_pushstring(L, setlocale(opt, str));
+#endif
   return 1;
 }
 
diff --git a/luajit/src/lj_alloc.c b/luajit/src/lj_alloc.c
index f856a7a..7c7ec67 100644
--- a/luajit/src/lj_alloc.c
+++ b/luajit/src/lj_alloc.c
@@ -179,23 +179,29 @@ static LJ_AINLINE int CALL_MUNMAP(void *ptr, size_t size)
 
 #if defined(MAP_32BIT)
 
+#if defined(__sun__)
+#define MMAP_REGION_START	((uintptr_t)0x1000)
+#else
 /* Actually this only gives us max. 1GB in current Linux kernels. */
+#define MMAP_REGION_START	((uintptr_t)0)
+#endif
+
 static LJ_AINLINE void *CALL_MMAP(size_t size)
 {
   int olderr = errno;
-  void *ptr = mmap(NULL, size, MMAP_PROT, MAP_32BIT|MMAP_FLAGS, -1, 0);
+  void *ptr = mmap((void *)MMAP_REGION_START, size, MMAP_PROT, MAP_32BIT|MMAP_FLAGS, -1, 0);
   errno = olderr;
   return ptr;
 }
 
-#elif LJ_TARGET_OSX || LJ_TARGET_PS4 || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__sun__)
+#elif LJ_TARGET_OSX || LJ_TARGET_PS4 || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) || defined(__sun__)
 
 /* OSX and FreeBSD mmap() use a naive first-fit linear search.
 ** That's perfect for us. Except that -pagezero_size must be set for OSX,
 ** otherwise the lower 4GB are blocked. And the 32GB RLIMIT_DATA needs
 ** to be reduced to 250MB on FreeBSD.
 */
-#if LJ_TARGET_OSX
+#if LJ_TARGET_OSX || defined(__DragonFly__)
 #define MMAP_REGION_START	((uintptr_t)0x10000)
 #elif LJ_TARGET_PS4
 #define MMAP_REGION_START	((uintptr_t)0x4000)
@@ -232,7 +238,7 @@ static LJ_AINLINE void *CALL_MMAP(size_t size)
       return p;
     }
     if (p != CMFAIL) munmap(p, size);
-#ifdef __sun__
+#if defined(__sun__) || defined(__DragonFly__)
     alloc_hint += 0x1000000;  /* Need near-exhaustive linear scan. */
     if (alloc_hint + size < MMAP_REGION_END) continue;
 #endif
diff --git a/luajit/src/lj_arch.h b/luajit/src/lj_arch.h
index 0196eed..f1c6e5e 100644
--- a/luajit/src/lj_arch.h
+++ b/luajit/src/lj_arch.h
@@ -67,7 +67,8 @@
 #elif defined(__MACH__) && defined(__APPLE__)
 #define LUAJIT_OS	LUAJIT_OS_OSX
 #elif (defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || \
-       defined(__NetBSD__) || defined(__OpenBSD__)) && !defined(__ORBIS__)
+       defined(__NetBSD__) || defined(__OpenBSD__) || \
+       defined(__DragonFly__)) && !defined(__ORBIS__)
 #define LUAJIT_OS	LUAJIT_OS_BSD
 #elif (defined(__sun__) && defined(__svr4__)) || defined(__CYGWIN__)
 #define LUAJIT_OS	LUAJIT_OS_POSIX
@@ -111,6 +112,11 @@
 #define NULL ((void*)0)
 #endif
 
+#ifdef __psp2__
+#define LJ_TARGET_PSVITA	1
+#define LJ_TARGET_CONSOLE	1
+#endif
+
 #if _XBOX_VER >= 200
 #define LJ_TARGET_XBOX360	1
 #define LJ_TARGET_CONSOLE	1
diff --git a/luajit/src/lj_asm.c b/luajit/src/lj_asm.c
index 264649a..2afa92d 100644
--- a/luajit/src/lj_asm.c
+++ b/luajit/src/lj_asm.c
@@ -353,6 +353,7 @@ static Reg ra_rematk(ASMState *as, IRRef ref)
 static int32_t ra_spill(ASMState *as, IRIns *ir)
 {
   int32_t slot = ir->s;
+  lua_assert(ir >= as->ir + REF_TRUE);
   if (!ra_hasspill(slot)) {
     if (irt_is64(ir->t)) {
       slot = as->evenspill;
diff --git a/luajit/src/lj_asm_ppc.h b/luajit/src/lj_asm_ppc.h
index e1a496a..0a3d35e 100644
--- a/luajit/src/lj_asm_ppc.h
+++ b/luajit/src/lj_asm_ppc.h
@@ -49,6 +49,8 @@ static void asm_exitstub_setup(ASMState *as, ExitNo nexits)
 {
   ExitNo i;
   MCode *mxp = as->mctop;
+  if (mxp - (nexits + 3 + MCLIM_REDZONE) < as->mclim)
+    asm_mclimit(as);
   /* 1: mflr r0; bl ->vm_exit_handler; li r0, traceno; bl <1; bl <1; ... */
   for (i = nexits-1; (int32_t)i >= 0; i--)
     *--mxp = PPCI_BL|(((-3-i)&0x00ffffffu)<<2);
diff --git a/luajit/src/lj_asm_x86.h b/luajit/src/lj_asm_x86.h
index fda911e..04b7964 100644
--- a/luajit/src/lj_asm_x86.h
+++ b/luajit/src/lj_asm_x86.h
@@ -325,6 +325,14 @@ static Reg asm_fuseload(ASMState *as, IRRef ref, RegSet allow)
       as->mrm.base = as->mrm.idx = RID_NONE;
       return RID_MRM;
     }
+  } else if (ir->o == IR_KINT64) {
+    RegSet avail = as->freeset & ~as->modset & RSET_GPR;
+    lua_assert(allow != RSET_EMPTY);
+    if (!(avail & (avail-1))) {  /* Fuse if less than two regs available. */
+      as->mrm.ofs = ptr2addr(ir_kint64(ir));
+      as->mrm.base = as->mrm.idx = RID_NONE;
+      return RID_MRM;
+    }
   } else if (mayfuse(as, ref)) {
     RegSet xallow = (allow & RSET_GPR) ? allow : RSET_GPR;
     if (ir->o == IR_SLOAD) {
@@ -361,7 +369,7 @@ static Reg asm_fuseload(ASMState *as, IRRef ref, RegSet allow)
       return RID_MRM;
     }
   }
-  if (!(as->freeset & allow) &&
+  if (!(as->freeset & allow) && !irref_isk(ref) &&
       (allow == RSET_EMPTY || ra_hasspill(ir->s) || iscrossref(as, ref)))
     goto fusespill;
   return ra_allocref(as, ref, allow);
@@ -571,7 +579,7 @@ static void asm_setupresult(ASMState *as, IRIns *ir, const CCallInfo *ci)
       lua_assert(!irt_ispri(ir->t));
       ra_destreg(as, ir, RID_RET);
     }
-  } else if (LJ_32 && irt_isfp(ir->t)) {
+  } else if (LJ_32 && irt_isfp(ir->t) && !(ci->flags & CCI_CASTU64)) {
     emit_x87op(as, XI_FPOP);  /* Pop unused result from x87 st0. */
   }
 }
diff --git a/luajit/src/lj_carith.c b/luajit/src/lj_carith.c
index 9f94091..9a451b2 100644
--- a/luajit/src/lj_carith.c
+++ b/luajit/src/lj_carith.c
@@ -62,7 +62,7 @@ static int carith_checkarg(lua_State *L, CTState *cts, CDArith *ca)
       TValue *o2 = i == 0 ? o+1 : o-1;
       CType *ct = ctype_raw(cts, cdataV(o2)->ctypeid);
       ca->ct[i] = NULL;
-      ca->p[i] = NULL;
+      ca->p[i] = (uint8_t *)strVdata(o);
       ok = 0;
       if (ctype_isenum(ct->info)) {
 	CTSize ofs;
@@ -79,7 +79,7 @@ static int carith_checkarg(lua_State *L, CTState *cts, CDArith *ca)
       }
     } else {
       ca->ct[i] = NULL;
-      ca->p[i] = NULL;
+      ca->p[i] = (void *)(intptr_t)1;  /* To make it unequal. */
       ok = 0;
     }
   }
@@ -234,7 +234,9 @@ static int lj_carith_meta(lua_State *L, CTState *cts, CDArith *ca, MMS mm)
     const char *repr[2];
     int i, isenum = -1, isstr = -1;
     if (mm == MM_eq) {  /* Equality checks never raise an error. */
-      setboolV(L->top-1, 0);
+      int eq = ca->p[0] == ca->p[1];
+      setboolV(L->top-1, eq);
+      setboolV(&G(L)->tmptv2, eq);  /* Remember for trace recorder. */
       return 1;
     }
     for (i = 0; i < 2; i++) {
diff --git a/luajit/src/lj_cconv.c b/luajit/src/lj_cconv.c
index de4938e..90cd36e 100644
--- a/luajit/src/lj_cconv.c
+++ b/luajit/src/lj_cconv.c
@@ -702,6 +702,7 @@ static void cconv_substruct_init(CTState *cts, CType *d, uint8_t *dp,
     } else if (ctype_isxattrib(df->info, CTA_SUBTYPE)) {
       cconv_substruct_init(cts, ctype_rawchild(cts, df),
 			   dp+df->size, o, len, ip);
+      if ((d->info & CTF_UNION)) break;
     }  /* Ignore all other entries in the chain. */
   }
 }
diff --git a/luajit/src/lj_crecord.c b/luajit/src/lj_crecord.c
index da9013f..fe08b93 100644
--- a/luajit/src/lj_crecord.c
+++ b/luajit/src/lj_crecord.c
@@ -794,7 +794,7 @@ again:
     }
   } else if (tref_isstr(idx)) {
     GCstr *name = strV(&rd->argv[1]);
-    if (cd->ctypeid == CTID_CTYPEID)
+    if (cd && cd->ctypeid == CTID_CTYPEID)
       ct = ctype_raw(cts, crec_constructor(J, cd, ptr));
     if (ctype_isstruct(ct->info)) {
       CTSize fofs;
@@ -835,6 +835,7 @@ again:
       CType *cct = ctype_rawchild(cts, ct);
       if (ctype_isstruct(cct->info)) {
 	ct = cct;
+	cd = NULL;
 	if (tref_isstr(idx)) goto again;
       }
     }
@@ -847,8 +848,11 @@ again:
 
   /* Resolve reference for field. */
   ct = ctype_get(cts, sid);
-  if (ctype_isref(ct->info))
+  if (ctype_isref(ct->info)) {
     ptr = emitir(IRT(IR_XLOAD, IRT_PTR), ptr, 0);
+    sid = ctype_cid(ct->info);
+    ct = ctype_get(cts, sid);
+  }
 
   while (ctype_isattrib(ct->info))
     ct = ctype_child(cts, ct);  /* Skip attributes. */
@@ -1317,7 +1321,8 @@ static TRef crec_arith_ptr(jit_State *J, TRef *sp, CType **s, MMS mm)
 }
 
 /* Record ctype arithmetic metamethods. */
-static void crec_arith_meta(jit_State *J, CTState *cts, RecordFFData *rd)
+static TRef crec_arith_meta(jit_State *J, TRef *sp, CType **s, CTState *cts,
+			    RecordFFData *rd)
 {
   cTValue *tv = NULL;
   if (J->base[0]) {
@@ -1338,13 +1343,20 @@ static void crec_arith_meta(jit_State *J, CTState *cts, RecordFFData *rd)
     if (tvisfunc(tv)) {
       J->base[-1] = lj_ir_kfunc(J, funcV(tv)) | TREF_FRAME;
       rd->nres = -1;  /* Pending tailcall. */
-      return;
+      return 0;
     }  /* NYI: non-function metamethods. */
-  } else if ((MMS)rd->data == MM_eq) {
-    J->base[0] = TREF_FALSE;
-    return;
+  } else if ((MMS)rd->data == MM_eq) {  /* Fallback cdata pointer comparison. */
+    if (sp[0] && sp[1] && ctype_isnum(s[0]->info) == ctype_isnum(s[1]->info)) {
+      /* Assume true comparison. Fixup and emit pending guard later. */
+      lj_ir_set(J, IRTG(IR_EQ, IRT_PTR), sp[0], sp[1]);
+      J->postproc = LJ_POST_FIXGUARD;
+      return TREF_TRUE;
+    } else {
+      return TREF_FALSE;
+    }
   }
   lj_trace_err(J, LJ_TRERR_BADTYPE);
+  return 0;
 }
 
 void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
@@ -1357,7 +1369,7 @@ void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
     TRef tr = J->base[i];
     CType *ct = ctype_get(cts, CTID_DOUBLE);
     if (!tr) {
-      goto trymeta;
+      lj_trace_err(J, LJ_TRERR_BADTYPE);
     } else if (tref_iscdata(tr)) {
       CTypeID id = argv2cdata(J, tr, &rd->argv[i])->ctypeid;
       IRType t;
@@ -1387,11 +1399,12 @@ void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
       }
       if (ctype_isenum(ct->info)) ct = ctype_child(cts, ct);
       if (ctype_isnum(ct->info)) {
-	if (t == IRT_CDATA) goto trymeta;
-	if (t == IRT_I64 || t == IRT_U64) lj_needsplit(J);
-	tr = emitir(IRT(IR_XLOAD, t), tr, 0);
-      } else if (!(ctype_isptr(ct->info) || ctype_isrefarray(ct->info))) {
-	goto trymeta;
+	if (t == IRT_CDATA) {
+	  tr = 0;
+	} else {
+	  if (t == IRT_I64 || t == IRT_U64) lj_needsplit(J);
+	  tr = emitir(IRT(IR_XLOAD, t), tr, 0);
+	}
       }
     } else if (tref_isnil(tr)) {
       tr = lj_ir_kptr(J, NULL);
@@ -1411,10 +1424,17 @@ void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
 	  emitir(IRTG(IR_EQ, IRT_STR), tr, lj_ir_kstr(J, str));
 	  ct = ctype_child(cts, cct);
 	  tr = lj_ir_kint(J, (int32_t)ofs);
-	}  /* else: interpreter will throw. */
-      }  /* else: interpreter will throw. */
+	} else {  /* Interpreter will throw or return false. */
+	  ct = ctype_get(cts, CTID_P_VOID);
+	}
+      } else if (ctype_isptr(ct->info)) {
+	tr = emitir(IRT(IR_ADD, IRT_PTR), tr, lj_ir_kintp(J, sizeof(GCstr)));
+      } else {
+	ct = ctype_get(cts, CTID_P_VOID);
+      }
     } else if (!tref_isnum(tr)) {
-      goto trymeta;
+      tr = 0;
+      ct = ctype_get(cts, CTID_P_VOID);
     }
   ok:
     s[i] = ct;
@@ -1422,22 +1442,20 @@ void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
   }
   {
     TRef tr;
-    if ((tr = crec_arith_int64(J, sp, s, (MMS)rd->data)) ||
-	(tr = crec_arith_ptr(J, sp, s, (MMS)rd->data))) {
-      J->base[0] = tr;
-      /* Fixup cdata comparisons, too. Avoids some cdata escapes. */
-      if (J->postproc == LJ_POST_FIXGUARD && frame_iscont(J->L->base-1) &&
-	  !irt_isguard(J->guardemit)) {
-	const BCIns *pc = frame_contpc(J->L->base-1) - 1;
-	if (bc_op(*pc) <= BC_ISNEP) {
-	  setframe_pc(&J2G(J)->tmptv, pc);
-	  J2G(J)->tmptv.u32.lo = ((tref_istrue(tr) ^ bc_op(*pc)) & 1);
-	  J->postproc = LJ_POST_FIXCOMP;
-	}
+    if (!(tr = crec_arith_int64(J, sp, s, (MMS)rd->data)) &&
+	!(tr = crec_arith_ptr(J, sp, s, (MMS)rd->data)) &&
+	!(tr = crec_arith_meta(J, sp, s, cts, rd)))
+      return;
+    J->base[0] = tr;
+    /* Fixup cdata comparisons, too. Avoids some cdata escapes. */
+    if (J->postproc == LJ_POST_FIXGUARD && frame_iscont(J->L->base-1) &&
+	!irt_isguard(J->guardemit)) {
+      const BCIns *pc = frame_contpc(J->L->base-1) - 1;
+      if (bc_op(*pc) <= BC_ISNEP) {
+	setframe_pc(&J2G(J)->tmptv, pc);
+	J2G(J)->tmptv.u32.lo = ((tref_istrue(tr) ^ bc_op(*pc)) & 1);
+	J->postproc = LJ_POST_FIXCOMP;
       }
-    } else {
-    trymeta:
-      crec_arith_meta(J, cts, rd);
     }
   }
 }
diff --git a/luajit/src/lj_debug.c b/luajit/src/lj_debug.c
index 4653a4e..e4ab9c1 100644
--- a/luajit/src/lj_debug.c
+++ b/luajit/src/lj_debug.c
@@ -14,6 +14,7 @@
 #include "lj_state.h"
 #include "lj_frame.h"
 #include "lj_bc.h"
+#include "lj_vm.h"
 #if LJ_HASJIT
 #include "lj_jit.h"
 #endif
@@ -86,7 +87,8 @@ static BCPos debug_framepc(lua_State *L, GCfunc *fn, cTValue *nextframe)
 	if (frame_islua(f)) {
 	  f = frame_prevl(f);
 	} else {
-	  if (frame_isc(f))
+	  if (frame_isc(f) || (LJ_HASFFI && frame_iscont(f) &&
+			       (f-1)->u32.lo == LJ_CONT_FFI_CALLBACK))
 	    cf = cframe_raw(cframe_prev(cf));
 	  f = frame_prevd(f);
 	}
@@ -463,7 +465,7 @@ int lj_debug_getinfo(lua_State *L, const char *what, lj_Debug *ar, int ext)
 	lj_debug_shortname(ar->short_src, name);
 	ar->linedefined = (int)firstline;
 	ar->lastlinedefined = (int)(firstline + pt->numline);
-	ar->what = firstline ? "Lua" : "main";
+	ar->what = (firstline || !pt->numline) ? "Lua" : "main";
       } else {
 	ar->source = "=[C]";
 	ar->short_src[0] = '[';
diff --git a/luajit/src/lj_def.h b/luajit/src/lj_def.h
index 3c43be7..8624aed 100644
--- a/luajit/src/lj_def.h
+++ b/luajit/src/lj_def.h
@@ -111,7 +111,7 @@ typedef uintptr_t BloomFilter;
 #define bloomset(b, x)	((b) |= bloombit((x)))
 #define bloomtest(b, x)	((b) & bloombit((x)))
 
-#if defined(__GNUC__)
+#if defined(__GNUC__) || defined(__psp2__)
 
 #define LJ_NORET	__attribute__((noreturn))
 #define LJ_ALIGN(n)	__attribute__((aligned(n)))
@@ -119,7 +119,7 @@ typedef uintptr_t BloomFilter;
 #define LJ_AINLINE	inline __attribute__((always_inline))
 #define LJ_NOINLINE	__attribute__((noinline))
 
-#if defined(__ELF__) || defined(__MACH__)
+#if defined(__ELF__) || defined(__MACH__) || defined(__psp2__)
 #if !((defined(__sun__) && defined(__svr4__)) || defined(__CELLOS_LV2__))
 #define LJ_NOAPI	extern __attribute__((visibility("hidden")))
 #endif
@@ -150,6 +150,9 @@ static LJ_AINLINE uint32_t lj_fls(uint32_t x)
 #if defined(__arm__)
 static LJ_AINLINE uint32_t lj_bswap(uint32_t x)
 {
+#if defined(__psp2__)
+  return __builtin_rev(x);
+#else
   uint32_t r;
 #if __ARM_ARCH_6__ || __ARM_ARCH_6J__ || __ARM_ARCH_6T2__ || __ARM_ARCH_6Z__ ||\
     __ARM_ARCH_6ZK__ || __ARM_ARCH_7__ || __ARM_ARCH_7A__ || __ARM_ARCH_7R__
@@ -163,6 +166,7 @@ static LJ_AINLINE uint32_t lj_bswap(uint32_t x)
 #endif
   return ((r & 0xff00ffffu) >> 8) ^ lj_ror(x, 8);
 #endif
+#endif
 }
 
 static LJ_AINLINE uint64_t lj_bswap64(uint64_t x)
diff --git a/luajit/src/lj_err.c b/luajit/src/lj_err.c
index db18267..3097795 100644
--- a/luajit/src/lj_err.c
+++ b/luajit/src/lj_err.c
@@ -378,7 +378,7 @@ typedef struct UndocumentedDispatcherContext {
   ULONG64 EstablisherFrame;
   ULONG64 TargetIp;
   PCONTEXT ContextRecord;
-  PEXCEPTION_ROUTINE LanguageHandler;
+  void (*LanguageHandler)(void);
   PVOID HandlerData;
   PUNWIND_HISTORY_TABLE HistoryTable;
   ULONG ScopeIndex;
@@ -518,12 +518,14 @@ static ptrdiff_t finderrfunc(lua_State *L)
     case FRAME_C:
       cf = cframe_prev(cf);
       /* fallthrough */
+    case FRAME_VARG:
+      frame = frame_prevd(frame);
+      break;
     case FRAME_CONT:
 #if LJ_HASFFI
       if ((frame-1)->u32.lo == LJ_CONT_FFI_CALLBACK)
 	cf = cframe_prev(cf);
 #endif
-    case FRAME_VARG:
       frame = frame_prevd(frame);
       break;
     case FRAME_CP:
diff --git a/luajit/src/lj_gc.c b/luajit/src/lj_gc.c
index c2bc397..c856df4 100644
--- a/luajit/src/lj_gc.c
+++ b/luajit/src/lj_gc.c
@@ -631,6 +631,8 @@ static size_t gc_onestep(lua_State *L)
   case GCSsweep: {
     MSize old = g->gc.total;
     setmref(g->gc.sweep, gc_sweep(g, mref(g->gc.sweep, GCRef), GCSWEEPMAX));
+    lua_assert(old >= g->gc.total);
+    g->gc.estimate -= old - g->gc.total;
     if (gcref(*mref(g->gc.sweep, GCRef)) == NULL) {
       gc_shrink(g, L);
       if (gcref(g->gc.mmudata)) {  /* Need any finalizations? */
@@ -643,8 +645,6 @@ static size_t gc_onestep(lua_State *L)
 	g->gc.debt = 0;
       }
     }
-    lua_assert(old >= g->gc.total);
-    g->gc.estimate -= old - g->gc.total;
     return GCSWEEPMAX*GCSWEEPCOST;
     }
   case GCSfinalize:
diff --git a/luajit/src/lj_gdbjit.c b/luajit/src/lj_gdbjit.c
index 5e7fca1..d999986 100644
--- a/luajit/src/lj_gdbjit.c
+++ b/luajit/src/lj_gdbjit.c
@@ -356,6 +356,8 @@ static const ELFheader elfhdr_template = {
   .eosabi = 2,
 #elif defined(__OpenBSD__)
   .eosabi = 12,
+#elif defined(__DragonFly__)
+  .eosabi = 0,
 #elif (defined(__sun__) && defined(__svr4__))
   .eosabi = 6,
 #else
diff --git a/luajit/src/lj_ircall.h b/luajit/src/lj_ircall.h
index af30ba8..92d61df 100644
--- a/luajit/src/lj_ircall.h
+++ b/luajit/src/lj_ircall.h
@@ -86,6 +86,12 @@ typedef struct CCallInfo {
 #define IRCALLCOND_FFI32(x)		NULL
 #endif
 
+#if LJ_TARGET_X86
+#define CCI_RANDFPR	0	/* Clang on OSX/x86 is overzealous. */
+#else
+#define CCI_RANDFPR	CCI_NOFPRCLOBBER
+#endif
+
 #if LJ_SOFTFP
 #define ARG1_FP		2	/* Treat as 2 32 bit arguments. */
 #else
@@ -112,7 +118,7 @@ typedef struct CCallInfo {
   _(ANY,	lj_gc_step_jit,		2,  FS, NIL, CCI_L) \
   _(ANY,	lj_gc_barrieruv,	2,  FS, NIL, 0) \
   _(ANY,	lj_mem_newgco,		2,  FS, P32, CCI_L) \
-  _(ANY,	lj_math_random_step, 1, FS, NUM, CCI_CASTU64|CCI_NOFPRCLOBBER) \
+  _(ANY,	lj_math_random_step, 1, FS, NUM, CCI_CASTU64|CCI_RANDFPR)\
   _(ANY,	lj_vm_modi,		2,  FN, INT, 0) \
   _(ANY,	sinh,			ARG1_FP,  N, NUM, 0) \
   _(ANY,	cosh,			ARG1_FP,  N, NUM, 0) \
diff --git a/luajit/src/lj_jit.h b/luajit/src/lj_jit.h
index b69e4f4..ecd79de 100644
--- a/luajit/src/lj_jit.h
+++ b/luajit/src/lj_jit.h
@@ -279,6 +279,7 @@ typedef struct BPropEntry {
 
 /* Scalar evolution analysis cache. */
 typedef struct ScEvEntry {
+  MRef pc;		/* Bytecode PC of FORI. */
   IRRef1 idx;		/* Index reference. */
   IRRef1 start;		/* Constant start reference. */
   IRRef1 stop;		/* Constant stop reference. */
diff --git a/luajit/src/lj_opt_dce.c b/luajit/src/lj_opt_dce.c
index be80189..fdfe934 100644
--- a/luajit/src/lj_opt_dce.c
+++ b/luajit/src/lj_opt_dce.c
@@ -69,6 +69,7 @@ void lj_opt_dce(jit_State *J)
   if ((J->flags & JIT_F_OPT_DCE)) {
     dce_marksnap(J);
     dce_propagate(J);
+    memset(J->bpropcache, 0, sizeof(J->bpropcache));  /* Invalidate cache. */
   }
 }
 
diff --git a/luajit/src/lj_opt_fold.c b/luajit/src/lj_opt_fold.c
index 8a5b41c..a15f927 100644
--- a/luajit/src/lj_opt_fold.c
+++ b/luajit/src/lj_opt_fold.c
@@ -505,13 +505,14 @@ LJFOLDF(kfold_strref_snew)
   } else {
     /* Reassociate: strref(snew(strref(str, a), len), b) ==> strref(str, a+b) */
     IRIns *ir = IR(fleft->op1);
-    IRRef1 str = ir->op1;  /* IRIns * is not valid across emitir. */
-    lua_assert(ir->o == IR_STRREF);
-    PHIBARRIER(ir);
-    fins->op2 = emitir(IRTI(IR_ADD), ir->op2, fins->op2);  /* Clobbers fins! */
-    fins->op1 = str;
-    fins->ot = IRT(IR_STRREF, IRT_P32);
-    return RETRYFOLD;
+    if (ir->o == IR_STRREF) {
+      IRRef1 str = ir->op1;  /* IRIns * is not valid across emitir. */
+      PHIBARRIER(ir);
+      fins->op2 = emitir(IRTI(IR_ADD), ir->op2, fins->op2); /* Clobbers fins! */
+      fins->op1 = str;
+      fins->ot = IRT(IR_STRREF, IRT_P32);
+      return RETRYFOLD;
+    }
   }
   return NEXTFOLD;
 }
@@ -1825,7 +1826,8 @@ LJFOLDF(merge_eqne_snew_kgc)
   if (len <= FOLD_SNEW_MAX_LEN) {
     IROp op = (IROp)fins->o;
     IRRef strref = fleft->op1;
-    lua_assert(IR(strref)->o == IR_STRREF);
+    if (IR(strref)->o != IR_STRREF)
+      return NEXTFOLD;
     if (op == IR_EQ) {
       emitir(IRTGI(IR_EQ), fleft->op2, lj_ir_kint(J, len));
       /* Caveat: fins/fleft/fright is no longer valid after emitir. */
diff --git a/luajit/src/lj_opt_mem.c b/luajit/src/lj_opt_mem.c
index c8cdc3e..d5f5290 100644
--- a/luajit/src/lj_opt_mem.c
+++ b/luajit/src/lj_opt_mem.c
@@ -740,21 +740,19 @@ retry:
     case ALIAS_MUST:
       /* Emit conversion if the loaded type doesn't match the forwarded type. */
       if (!irt_sametype(fins->t, IR(store->op2)->t)) {
-	IRType st = irt_type(fins->t);
-	if (st == IRT_I8 || st == IRT_I16) {  /* Trunc + sign-extend. */
-	  st |= IRCONV_SEXT;
-	} else if (st == IRT_U8 || st == IRT_U16) {  /* Trunc + zero-extend. */
-	} else if (st == IRT_INT) {
-	  st = irt_type(IR(store->op2)->t);  /* Needs dummy CONV.int.*. */
-	} else {  /* I64/U64 are boxed, U32 is hidden behind a CONV.num.u32. */
-	  goto store_fwd;
+	IRType dt = irt_type(fins->t), st = irt_type(IR(store->op2)->t);
+	if (dt == IRT_I8 || dt == IRT_I16) {  /* Trunc + sign-extend. */
+	  st = dt | IRCONV_SEXT;
+	  dt = IRT_INT;
+	} else if (dt == IRT_U8 || dt == IRT_U16) {  /* Trunc + zero-extend. */
+	  st = dt;
+	  dt = IRT_INT;
 	}
-	fins->ot = IRTI(IR_CONV);
+	fins->ot = IRT(IR_CONV, dt);
 	fins->op1 = store->op2;
-	fins->op2 = (IRT_INT<<5)|st;
+	fins->op2 = (dt<<5)|st;
 	return RETRYFOLD;
       }
-    store_fwd:
       return store->op2;  /* Store forwarding. */
     }
     ref = store->prev;
diff --git a/luajit/src/lj_record.c b/luajit/src/lj_record.c
index c6848fc..12b76bb 100644
--- a/luajit/src/lj_record.c
+++ b/luajit/src/lj_record.c
@@ -421,6 +421,7 @@ static void rec_for_loop(jit_State *J, const BCIns *fori, ScEvEntry *scev,
     J->base[ra+FORL_IDX] = idx = emitir(IRT(IR_ADD, t), idx, step);
   J->base[ra+FORL_EXT] = idx;
   scev->idx = tref_ref(idx);
+  setmref(scev->pc, fori);
   J->maxslot = ra+FORL_EXT+1;
 }
 
@@ -436,7 +437,7 @@ static LoopEvent rec_for(jit_State *J, const BCIns *fori, int isforl)
   IRType t;
   if (isforl) {  /* Handle FORL/JFORL opcodes. */
     TRef idx = tr[FORL_IDX];
-    if (tref_ref(idx) == J->scev.idx) {
+    if (mref(J->scev.pc, const BCIns) == fori && tref_ref(idx) == J->scev.idx) {
       t = J->scev.t.irt;
       stop = J->scev.stop;
       idx = emitir(IRT(IR_ADD, t), idx, J->scev.step);
@@ -1059,7 +1060,7 @@ static void rec_idx_abc(jit_State *J, TRef asizeref, TRef ikey, uint32_t asize)
       lua_assert(irt_isint(J->scev.t) && ir->o == IR_SLOAD);
       stop = numberVint(&(J->L->base - J->baseslot)[ir->op1 + FORL_STOP]);
       /* Runtime value for stop of loop is within bounds? */
-      if ((int64_t)stop + ofs < (int64_t)asize) {
+      if ((uint64_t)stop + ofs < (uint64_t)asize) {
 	/* Emit invariant bounds check for stop. */
 	emitir(IRTG(IR_ABC, IRT_P32), asizeref, ofs == 0 ? J->scev.stop :
 	       emitir(IRTI(IR_ADD), J->scev.stop, ofsref));
@@ -2153,6 +2154,7 @@ void lj_record_setup(jit_State *J)
   memset(J->chain, 0, sizeof(J->chain));
   memset(J->bpropcache, 0, sizeof(J->bpropcache));
   J->scev.idx = REF_NIL;
+  setmref(J->scev.pc, NULL);
 
   J->baseslot = 1;  /* Invoking function is at base[-1]. */
   J->base = J->slot + J->baseslot;
diff --git a/luajit/src/lj_snap.c b/luajit/src/lj_snap.c
index 1c978c2..803533c 100644
--- a/luajit/src/lj_snap.c
+++ b/luajit/src/lj_snap.c
@@ -104,8 +104,6 @@ static BCReg snapshot_framelinks(jit_State *J, SnapEntry *map)
     if (frame_islua(frame)) {
       map[f++] = SNAP_MKPC(frame_pc(frame));
       frame = frame_prevl(frame);
-      if (frame + funcproto(frame_func(frame))->framesize > ftop)
-	ftop = frame + funcproto(frame_func(frame))->framesize;
     } else if (frame_iscont(frame)) {
       map[f++] = SNAP_MKFTSZ(frame_ftsz(frame));
       map[f++] = SNAP_MKPC(frame_contpc(frame));
@@ -114,7 +112,10 @@ static BCReg snapshot_framelinks(jit_State *J, SnapEntry *map)
       lua_assert(!frame_isc(frame));
       map[f++] = SNAP_MKFTSZ(frame_ftsz(frame));
       frame = frame_prevd(frame);
+      continue;
     }
+    if (frame + funcproto(frame_func(frame))->framesize > ftop)
+      ftop = frame + funcproto(frame_func(frame))->framesize;
   }
   lua_assert(f == (MSize)(1 + J->framedepth));
   return (BCReg)(ftop - lim);
diff --git a/luajit/src/lj_state.c b/luajit/src/lj_state.c
index f972fdc..b9eaef4 100644
--- a/luajit/src/lj_state.c
+++ b/luajit/src/lj_state.c
@@ -175,7 +175,7 @@ static void close_state(lua_State *L)
     g->allocf(g->allocd, G2GG(g), sizeof(GG_State), 0);
 }
 
-#if LJ_64
+#if LJ_64 && !(defined(LUAJIT_USE_VALGRIND) && defined(LUAJIT_USE_SYSMALLOC))
 lua_State *lj_state_newstate(lua_Alloc f, void *ud)
 #else
 LUA_API lua_State *lua_newstate(lua_Alloc f, void *ud)
diff --git a/luajit/src/lj_trace.c b/luajit/src/lj_trace.c
index 9e5e400..0f48809 100644
--- a/luajit/src/lj_trace.c
+++ b/luajit/src/lj_trace.c
@@ -607,6 +607,7 @@ static TValue *trace_state(lua_State *L, lua_CFunction dummy, void *ud)
       }
       lj_opt_split(J);
       lj_opt_sink(J);
+      if (!J->loopref) J->cur.snap[J->cur.nsnap-1].count = SNAPCOUNT_DONE;
       J->state = LJ_TRACE_ASM;
       break;
 
diff --git a/luajit/src/luaconf.h b/luajit/src/luaconf.h
index d884a7e..736c0e2 100644
--- a/luajit/src/luaconf.h
+++ b/luajit/src/luaconf.h
@@ -37,7 +37,7 @@
 #endif
 #define LUA_LROOT	"/usr/local"
 #define LUA_LUADIR	"/lua/5.1/"
-#define LUA_LJDIR	"/luajit-2.0.2/"
+#define LUA_LJDIR	"/luajit-2.0.3/"
 
 #ifdef LUA_ROOT
 #define LUA_JROOT	LUA_ROOT
diff --git a/luajit/src/luajit.c b/luajit/src/luajit.c
index 3be170f..4e85908 100644
--- a/luajit/src/luajit.c
+++ b/luajit/src/luajit.c
@@ -592,17 +592,17 @@ static int loadjitmodule(lua_State *L)
   lua_concat(L, 2);
   if (lua_pcall(L, 1, 1, 0)) {
     const char *msg = lua_tostring(L, -1);
-    if (msg && !strncmp(msg, "module ", 7)) {
-    err:
-      l_message(progname,
-		"unknown luaJIT command or jit.* modules not installed");
-      return 1;
-    } else {
-      return report(L, 1);
-    }
+    if (msg && !strncmp(msg, "module ", 7))
+      goto nomodule;
+    return report(L, 1);
   }
   lua_getfield(L, -1, "start");
-  if (lua_isnil(L, -1)) goto err;
+  if (lua_isnil(L, -1)) {
+  nomodule:
+    l_message(progname,
+	      "unknown luaJIT command or jit.* modules not installed");
+    return 1;
+  }
   lua_remove(L, -2);  /* Drop module table. */
   return 0;
 }
diff --git a/luajit/src/luajit.h b/luajit/src/luajit.h
index 7bb9317..be721cf 100644
--- a/luajit/src/luajit.h
+++ b/luajit/src/luajit.h
@@ -30,9 +30,9 @@
 
 #include "lua.h"
 
-#define LUAJIT_VERSION		"LuaJIT 2.0.2"
-#define LUAJIT_VERSION_NUM	20002  /* Version 2.0.2 = 02.00.02. */
-#define LUAJIT_VERSION_SYM	luaJIT_version_2_0_2
+#define LUAJIT_VERSION		"LuaJIT 2.0.3"
+#define LUAJIT_VERSION_NUM	20003  /* Version 2.0.3 = 02.00.03. */
+#define LUAJIT_VERSION_SYM	luaJIT_version_2_0_3
 #define LUAJIT_COPYRIGHT	"Copyright (C) 2005-2014 Mike Pall"
 #define LUAJIT_URL		"http://luajit.org/"
 
diff --git a/luajit/src/psvitabuild.bat b/luajit/src/psvitabuild.bat
new file mode 100644
index 0000000..3991dc6
--- /dev/null
+++ b/luajit/src/psvitabuild.bat
@@ -0,0 +1,93 @@
+@rem Script to build LuaJIT with the PS Vita SDK.

+@rem Donated to the public domain.

+@rem

+@rem Open a "Visual Studio .NET Command Prompt" (32 bit host compiler)

+@rem Then cd to this directory and run this script.

+

+@if not defined INCLUDE goto :FAIL

+@if not defined SCE_PSP2_SDK_DIR goto :FAIL

+

+@setlocal

+@rem ---- Host compiler ----

+@set LJCOMPILE=cl /nologo /c /MD /O2 /W3 /D_CRT_SECURE_NO_DEPRECATE

+@set LJLINK=link /nologo

+@set LJMT=mt /nologo

+@set DASMDIR=..\dynasm

+@set DASM=%DASMDIR%\dynasm.lua

+@set ALL_LIB=lib_base.c lib_math.c lib_bit.c lib_string.c lib_table.c lib_io.c lib_os.c lib_package.c lib_debug.c lib_jit.c lib_ffi.c

+

+%LJCOMPILE% host\minilua.c

+@if errorlevel 1 goto :BAD

+%LJLINK% /out:minilua.exe minilua.obj

+@if errorlevel 1 goto :BAD

+if exist minilua.exe.manifest^

+  %LJMT% -manifest minilua.exe.manifest -outputresource:minilua.exe

+

+@rem Check for 32 bit host compiler.

+@minilua

+@if errorlevel 8 goto :FAIL

+

+@set DASMFLAGS=-D FPU -D HFABI

+minilua %DASM% -LN %DASMFLAGS% -o host\buildvm_arch.h vm_arm.dasc

+@if errorlevel 1 goto :BAD

+

+%LJCOMPILE% /I "." /I %DASMDIR% -DLUAJIT_TARGET=LUAJIT_ARCH_ARM -DLUAJIT_OS=LUAJIT_OS_OTHER -DLUAJIT_DISABLE_JIT -DLUAJIT_DISABLE_FFI -DLJ_TARGET_PSVITA=1 host\buildvm*.c

+@if errorlevel 1 goto :BAD

+%LJLINK% /out:buildvm.exe buildvm*.obj

+@if errorlevel 1 goto :BAD

+if exist buildvm.exe.manifest^

+  %LJMT% -manifest buildvm.exe.manifest -outputresource:buildvm.exe

+

+buildvm -m elfasm -o lj_vm.s

+@if errorlevel 1 goto :BAD

+buildvm -m bcdef -o lj_bcdef.h %ALL_LIB%

+@if errorlevel 1 goto :BAD

+buildvm -m ffdef -o lj_ffdef.h %ALL_LIB%

+@if errorlevel 1 goto :BAD

+buildvm -m libdef -o lj_libdef.h %ALL_LIB%

+@if errorlevel 1 goto :BAD

+buildvm -m recdef -o lj_recdef.h %ALL_LIB%

+@if errorlevel 1 goto :BAD

+buildvm -m vmdef -o jit\vmdef.lua %ALL_LIB%

+@if errorlevel 1 goto :BAD

+buildvm -m folddef -o lj_folddef.h lj_opt_fold.c

+@if errorlevel 1 goto :BAD

+

+@rem ---- Cross compiler ----

+@set LJCOMPILE="%SCE_PSP2_SDK_DIR%\host_tools\build\bin\psp2snc" -c -w -DLUAJIT_DISABLE_FFI -DLUAJIT_USE_SYSMALLOC

+@set LJLIB="%SCE_PSP2_SDK_DIR%\host_tools\build\bin\psp2ld32" -r --output=

+@set INCLUDE=""

+

+"%SCE_PSP2_SDK_DIR%\host_tools\build\bin\psp2as" -o lj_vm.o lj_vm.s

+

+@if "%1" neq "debug" goto :NODEBUG

+@shift

+@set LJCOMPILE=%LJCOMPILE% -g -O0

+@set TARGETLIB=libluajitD.a

+goto :BUILD

+:NODEBUG

+@set LJCOMPILE=%LJCOMPILE% -O2

+@set TARGETLIB=libluajit.a

+:BUILD

+del %TARGETLIB%

+

+%LJCOMPILE% ljamalg.c

+@if errorlevel 1 goto :BAD

+%LJLIB%%TARGETLIB% ljamalg.o lj_vm.o

+@if errorlevel 1 goto :BAD

+

+@del *.o *.obj *.manifest minilua.exe buildvm.exe

+@echo.

+@echo === Successfully built LuaJIT for PS Vita ===

+

+@goto :END

+:BAD

+@echo.

+@echo *******************************************************

+@echo *** Build FAILED -- Please check the error messages ***

+@echo *******************************************************

+@goto :END

+:FAIL

+@echo To run this script you must open a "Visual Studio .NET Command Prompt"

+@echo (32 bit host compiler). The PS Vita SDK must be installed, too.

+:END

diff --git a/luajit/src/vm_ppc.dasc b/luajit/src/vm_ppc.dasc
index da75f55..7ae40ec 100644
--- a/luajit/src/vm_ppc.dasc
+++ b/luajit/src/vm_ppc.dasc
@@ -75,6 +75,29 @@
 |.endif
 |.endmacro
 |
+|.macro clrso, reg
+|.if PPE
+|  li reg, 0
+|  mtxer reg
+|.else
+|  mcrxr cr0
+|.endif
+|.endmacro
+|
+|.macro checkov, reg, noov
+|.if PPE
+|  mfxer reg
+|  add reg, reg, reg
+|  cmpwi reg, 0
+|   li reg, 0
+|   mtxer reg
+|  bgey noov
+|.else
+|  mcrxr cr0
+|  bley noov
+|.endif
+|.endmacro
+|
 |//-----------------------------------------------------------------------
 |
 |// Fixed register assignments for the interpreter.
@@ -2587,7 +2610,7 @@ static void build_subroutines(BuildCtx *ctx)
   |    stw CARG2, DISPATCH_GL(vmstate)(DISPATCH)
   |  savex_ 0,1,2,3
   |   stw CARG1, 0(sp)			// Store extended stack chain.
-  |   mcrxr cr0				// Clear SO flag.
+  |   clrso TMP1
   |  savex_ 4,5,6,7
   |   addi CARG2, sp, 16+32*8+32*4	// Recompute original value of sp.
   |  savex_ 8,9,10,11
@@ -2706,7 +2729,7 @@ static void build_subroutines(BuildCtx *ctx)
   |  cmpwi CARG2, 0
   |   li CARG1, 0
   |  beqlr
-  |  mcrxr cr0			// Clear SO for -2147483648 % -1 and return 0.
+  |  clrso TMP0			// Clear SO for -2147483648 % -1 and return 0.
   |  blr
   |
   |//-----------------------------------------------------------------------
@@ -3309,7 +3332,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
     |4:
     |.if not GPR64
     |  // Potential overflow.
-    |  mcrxr cr0; bley <1		// Ignore unrelated overflow.
+    |  checkov TMP1, <1			// Ignore unrelated overflow.
     |.endif
     |  lus TMP1, 0x41e0			// 2^31.
     |  li TMP0, 0
@@ -3494,7 +3517,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
     |2:
     |  ins_next2
     |4:  // Overflow.
-    |  mcrxr cr0; bley <1		// Ignore unrelated overflow.
+    |  checkov TMP0, <1			// Ignore unrelated overflow.
     |  ins_arithfallback b
     |5:  // FP variant.
     ||if (vk == 1) {
@@ -4692,7 +4715,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
     |  b <1
     if (vk) {
       |6:  // Potential overflow.
-      |  mcrxr cr0; bley <4		// Ignore unrelated overflow.
+      |  checkov TMP0, <4		// Ignore unrelated overflow.
       |  b <2
     }
     |.endif
@@ -4832,7 +4855,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop)
     |  // Traces on PPC don't store the trace number, so use 0.
     |   stw ZERO, DISPATCH_GL(vmstate)(DISPATCH)
     |  lwzx TRACE:TMP2, TMP1, RD
-    |  mcrxr cr0			// Clear SO flag.
+    |  clrso TMP1
     |  lp TMP2, TRACE:TMP2->mcode
     |   stw BASE, DISPATCH_GL(jit_base)(DISPATCH)
     |  mtctr TMP2
diff --git a/luajit/src/vm_x86.dasc b/luajit/src/vm_x86.dasc
index 8a9b536..bd7056f 100644
--- a/luajit/src/vm_x86.dasc
+++ b/luajit/src/vm_x86.dasc
@@ -2198,7 +2198,7 @@ static void build_subroutines(BuildCtx *ctx)
   |.endif
   |
   |.ffunc_nnr math_fmod
-  |1: ; fprem; fnstsw ax; sahf; jp <1
+  |1: ; fprem; fnstsw ax; and ax, 0x400; jnz <1
   |  fpop1
   |  jmp ->fff_resn
   |