Squashed 'luajit-2.0/' changes from 8715ae6..367cba2

367cba2 DynASM/x86: Add rdpmc instruction.
7991a66 Extend Valgrind suppressions for non-inlined C code.
1393b2f Update build instructions.
f416cf6 OSX: Switch to Clang as the default compiler.
3639ff4 iOS: Disable os.execute() when building for iOS >= 8.0.
7579b16 DynASM/x86: Restrict shld/shrd to operands with same width.
9dc785e DynASM/x86: Fix some SSE instruction templates.
e5f4cb1 FFI: Fix ipairs() recording.
4808e62 Drop marks from replayed instructions when sinking.
2e85af8 Flush symbol tables in jit.dump on trace flush.
ef087aa Fix unsinking check.
076d625 Fix Cygwin support. Note: this is not a supported target.

git-subtree-dir: luajit-2.0
git-subtree-split: 367cba29ea03aa758eb041c3a0c5f1c7616f9a0a

11 files changed, 93 insertions, 63 deletions

diff --git a/doc/install.html b/doc/install.html
index 7a878b1..5dc1289 100644
--- a/doc/install.html
+++ b/doc/install.html
@@ -112,23 +112,23 @@ operating systems, CPUs and compilers:
 </tr>
 <tr class="odd separate">
 <td class="compatcpu">x86 (32 bit)</td>
-<td class="compatos">GCC 4.x<br>GCC 3.4</td>
-<td class="compatos">GCC 4.x<br>GCC 3.4</td>
-<td class="compatos">GCC 4.x<br>GCC 3.4</td>
+<td class="compatos">GCC 4.x+<br>GCC 3.4</td>
+<td class="compatos">GCC 4.x+<br>GCC 3.4</td>
+<td class="compatos">XCode 5.0+<br>Clang</td>
 <td class="compatos">MSVC, MSVC/EE<br>WinSDK<br>MinGW, Cygwin</td>
 </tr>
 <tr class="even">
 <td class="compatcpu">x64 (64 bit)</td>
-<td class="compatos">GCC 4.x</td>
+<td class="compatos">GCC 4.x+</td>
 <td class="compatos">ORBIS (<a href="#ps4">PS4</a>)</td>
-<td class="compatos">GCC 4.x</td>
+<td class="compatos">XCode 5.0+<br>Clang</td>
 <td class="compatos">MSVC + SDK v7.0<br>WinSDK v7.0</td>
 </tr>
 <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+<br>PSP2 (<a href="#psvita">PS VITA</a>)</td>
-<td class="compatos">GCC 4.2+</td>
+<td class="compatos">XCode 5.0+<br>Clang</td>
 <td class="compatos compatno">&nbsp;</td>
 </tr>
 <tr class="even">
@@ -439,8 +439,7 @@ NDKF="--sysroot $NDK/platforms/android-$NDKABI/arch-x86"
 make HOST_CC="gcc -m32" CROSS=$NDKP TARGET_FLAGS="$NDKF"
 </pre>
 <p>
-You can cross-compile for <b id="ios">iOS 3.0+</b> (iPhone/iPad) using the <a href="http://developer.apple.com/devcenter/ios/index.action"><span class="ext">&raquo;</span>&nbsp;iOS SDK</a>.
-The environment variables need to match the iOS SDK version:
+You can cross-compile for <b id="ios">iOS 3.0+</b> (iPhone/iPad) using the <a href="http://developer.apple.com/devcenter/ios/index.action"><span class="ext">&raquo;</span>&nbsp;iOS SDK</a>:
 </p>
 <p style="font-size: 8pt;">
 Note: <b>the JIT compiler is disabled for iOS</b>, because regular iOS Apps
@@ -450,13 +449,11 @@ much slower than the JIT compiler. Please complain to Apple, not me.
 Or use Android. :-p
 </p>
 <pre class="code">
-IXCODE=`xcode-select -print-path`
-ISDK=$IXCODE/Platforms/iPhoneOS.platform/Developer
-ISDKVER=iPhoneOS6.0.sdk
-ISDKP=$ISDK/usr/bin/
-ISDKF="-arch armv7 -isysroot $ISDK/SDKs/$ISDKVER"
-make HOST_CC="gcc -m32 -arch i386" CROSS=$ISDKP TARGET_FLAGS="$ISDKF" \
-     TARGET_SYS=iOS
+ISDKP=$(xcrun --sdk iphoneos --show-sdk-path)
+ICC=$(xcrun --sdk iphoneos --find clang)
+ISDKF="-arch armv7 -isysroot $ISDKP"
+make HOST_CC="clang -m32 -arch i386" CROSS="$(dirname $ICC)/" \
+     TARGET_FLAGS="$ISDKF" TARGET_SYS=iOS
 </pre>
 
 <h3 id="consoles">Cross-compiling for consoles</h3>
diff --git a/dynasm/dasm_x86.lua b/dynasm/dasm_x86.lua
index 7ca061d..3a535f2 100644
--- a/dynasm/dasm_x86.lua
+++ b/dynasm/dasm_x86.lua
@@ -1081,10 +1081,11 @@ 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",
+  shld_3 =	"mriqdw:0FA4RmU|mrC/qq:0FA5Rm|mrC/dd:|mrC/ww:",
+  shrd_3 =	"mriqdw:0FACRmU|mrC/qq:0FADRm|mrC/dd:|mrC/ww:",
 
   rdtsc_0 =	"0F31", -- P1+
+  rdpmc_0 =	"0F33", -- P6+
   cpuid_0 =	"0FA2", -- P1+
 
   -- floating point ops
@@ -1190,7 +1191,7 @@ local map_op = {
   cvtsi2sd_2 =	"rm/od:F20F2ArM|rm/oq:F20F2ArXM",
   cvtsi2ss_2 =	"rm/od:F30F2ArM|rm/oq:F30F2ArXM",
   cvtss2sd_2 =	"rro:F30F5ArM|rx/od:",
-  cvtss2si_2 =	"rr/do:F20F2CrM|rr/qo:|rxd:|rx/qd:",
+  cvtss2si_2 =	"rr/do:F30F2DrM|rr/qo:|rxd:|rx/qd:",
   cvttpd2dq_2 =	"rmo:660FE6rM",
   cvttps2dq_2 =	"rmo:F30F5BrM",
   cvttsd2si_2 =	"rr/do:F20F2CrM|rr/qo:|rx/dq:|rxq:",
@@ -1247,7 +1248,7 @@ local map_op = {
   pcmpgtb_2 =	"rmo:660F64rM",
   pcmpgtd_2 =	"rmo:660F66rM",
   pcmpgtw_2 =	"rmo:660F65rM",
-  pextrw_3 =	"rri/do:660FC5rMU|xri/wo:660F3A15nrMU", -- Mem op: SSE4.1 only.
+  pextrw_3 =	"rri/do:660FC5rMU|xri/wo:660F3A15nRmU", -- Mem op: SSE4.1 only.
   pinsrw_3 =	"rri/od:660FC4rMU|rxi/ow:",
   pmaddwd_2 =	"rmo:660FF5rM",
   pmaxsw_2 =	"rmo:660FEErM",
@@ -1352,7 +1353,7 @@ local map_op = {
   dpps_3 =	"rmio:660F3A40rMU",
   extractps_3 =	"mri/do:660F3A17RmU|rri/qo:660F3A17RXmU",
   insertps_3 =	"rrio:660F3A41rMU|rxi/od:",
-  movntdqa_2 =	"rmo:660F382ArM",
+  movntdqa_2 =	"rxo:660F382ArM",
   mpsadbw_3 =	"rmio:660F3A42rMU",
   packusdw_2 =	"rmo:660F382BrM",
   pblendvb_3 =	"rmRo:660F3810rM",
diff --git a/src/Makefile b/src/Makefile
index 5021e47..9c95c4c 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -24,11 +24,13 @@ NODOTABIVER= 51
 # removing the '#' in front of them. Make sure you force a full recompile
 # with "make clean", followed by "make" if you change any options.
 #
+DEFAULT_CC = gcc
+#
 # LuaJIT builds as a native 32 or 64 bit binary by default.
-CC= gcc
+CC= $(DEFAULT_CC)
 #
 # Use this if you want to force a 32 bit build on a 64 bit multilib OS.
-#CC= gcc -m32
+#CC= $(DEFAULT_CC) -m32
 #
 # Since the assembler part does NOT maintain a frame pointer, it's pointless
 # to slow down the C part by not omitting it. Debugging, tracebacks and
@@ -151,6 +153,29 @@ XCFLAGS=
 ##############################################################################
 
 ##############################################################################
+# Host system detection.
+##############################################################################
+
+ifeq (Windows,$(findstring Windows,$(OS))$(MSYSTEM)$(TERM))
+  HOST_SYS= Windows
+  HOST_RM= del
+else
+  HOST_SYS:= $(shell uname -s)
+  ifneq (,$(findstring MINGW,$(HOST_SYS)))
+    HOST_SYS= Windows
+    HOST_MSYS= mingw
+  endif
+  ifneq (,$(findstring CYGWIN,$(HOST_SYS)))
+    HOST_SYS= Windows
+    HOST_MSYS= cygwin
+  endif
+  # Use Clang for OSX host.
+  ifeq (Darwin,$(HOST_SYS))
+    DEFAULT_CC= clang
+  endif
+endif
+
+##############################################################################
 # Flags and options for host and target.
 ##############################################################################
 
@@ -263,24 +288,9 @@ ifneq (,$(LMULTILIB))
 endif
 
 ##############################################################################
-# System detection.
+# Target system detection.
 ##############################################################################
 
-ifeq (Windows,$(findstring Windows,$(OS))$(MSYSTEM)$(TERM))
-  HOST_SYS= Windows
-  HOST_RM= del
-else
-  HOST_SYS:= $(shell uname -s)
-  ifneq (,$(findstring MINGW,$(HOST_SYS)))
-    HOST_SYS= Windows
-    HOST_MSYS= mingw
-  endif
-  ifneq (,$(findstring CYGWIN,$(HOST_SYS)))
-    HOST_SYS= Windows
-    HOST_MSYS= cygwin
-  endif
-endif
-
 TARGET_SYS?= $(HOST_SYS)
 ifeq (Windows,$(TARGET_SYS))
   TARGET_STRIP+= --strip-unneeded
diff --git a/src/jit/dump.lua b/src/jit/dump.lua
index d15c528..e6ea18a 100644
--- a/src/jit/dump.lua
+++ b/src/jit/dump.lua
@@ -564,6 +564,7 @@ local function dump_trace(what, tr, func, pc, otr, oex)
     end
     if dumpmode.H then out:write("</pre>\n\n") else out:write("\n") end
   else
+    if what == "flush" then symtab, nexitsym = {}, 0 end
     out:write("---- TRACE ", what, "\n\n")
   end
   out:flush()
diff --git a/src/lib_os.c b/src/lib_os.c
index bb5a141..762bb59 100644
--- a/src/lib_os.c
+++ b/src/lib_os.c
@@ -36,7 +36,7 @@
 
 LJLIB_CF(os_execute)
 {
-#if LJ_TARGET_CONSOLE
+#if LJ_NO_SYSTEM
 #if LJ_52
   errno = ENOSYS;
   return luaL_fileresult(L, 0, NULL);
diff --git a/src/lj.supp b/src/lj.supp
index 411f261..acb9e78 100644
--- a/src/lj.supp
+++ b/src/lj.supp
@@ -24,3 +24,18 @@
    Memcheck:Cond
    fun:lj_str_new
 }
+{
+   Optimized string compare
+   Memcheck:Addr4
+   fun:lj_str_fastcmp
+}
+{
+   Optimized string compare
+   Memcheck:Addr1
+   fun:lj_str_fastcmp
+}
+{
+   Optimized string compare
+   Memcheck:Cond
+   fun:lj_str_fastcmp
+}
diff --git a/src/lj_alloc.c b/src/lj_alloc.c
index 7c7ec67..9218c44 100644
--- a/src/lj_alloc.c
+++ b/src/lj_alloc.c
@@ -194,7 +194,7 @@ static LJ_AINLINE void *CALL_MMAP(size_t size)
   return ptr;
 }
 
-#elif LJ_TARGET_OSX || LJ_TARGET_PS4 || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) || defined(__sun__)
+#elif LJ_TARGET_OSX || LJ_TARGET_PS4 || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) || defined(__sun__) || defined(__CYGWIN__)
 
 /* 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,
diff --git a/src/lj_arch.h b/src/lj_arch.h
index d89d116..f1a1160 100644
--- a/src/lj_arch.h
+++ b/src/lj_arch.h
@@ -151,7 +151,11 @@
 #define LJ_ARCH_NAME		"x64"
 #define LJ_ARCH_BITS		64
 #define LJ_ARCH_ENDIAN		LUAJIT_LE
-#define LJ_ABI_WIN		LJ_TARGET_WINDOWS
+#if LJ_TARGET_WINDOWS || __CYGWIN__
+#define LJ_ABI_WIN		1
+#else
+#define LJ_ABI_WIN		0
+#endif
 #define LJ_TARGET_X64		1
 #define LJ_TARGET_X86ORX64	1
 #define LJ_TARGET_EHRETREG	0
@@ -422,6 +426,9 @@
 #if defined(__symbian__)
 #define LUAJIT_NO_EXP2
 #endif
+#if LJ_TARGET_CONSOLE || (LJ_TARGET_IOS && __IPHONE_OS_VERSION_MIN_REQUIRED >= __IPHONE_8_0)
+#define LJ_NO_SYSTEM		1
+#endif
 
 #if defined(LUAJIT_NO_UNWIND) || defined(__symbian__) || LJ_TARGET_IOS || LJ_TARGET_PS3 || LJ_TARGET_PS4
 #define LJ_NO_UNWIND		1
diff --git a/src/lj_err.c b/src/lj_err.c
index 11b07b5..6d8519b 100644
--- a/src/lj_err.c
+++ b/src/lj_err.c
@@ -186,7 +186,7 @@ static void *err_unwind(lua_State *L, void *stopcf, int errcode)
 
 /* -- External frame unwinding -------------------------------------------- */
 
-#if defined(__GNUC__) && !LJ_NO_UNWIND && !LJ_TARGET_WINDOWS
+#if defined(__GNUC__) && !LJ_NO_UNWIND && !LJ_ABI_WIN
 
 /*
 ** We have to use our own definitions instead of the mandatory (!) unwind.h,
@@ -352,7 +352,7 @@ LJ_FUNCA int lj_err_unwind_arm(int state, void *ucb, _Unwind_Context *ctx)
 
 #endif
 
-#elif LJ_TARGET_X64 && LJ_TARGET_WINDOWS
+#elif LJ_TARGET_X64 && LJ_ABI_WIN
 
 /*
 ** Someone in Redmond owes me several days of my life. A lot of this is
@@ -417,7 +417,9 @@ LJ_FUNCA EXCEPTION_DISPOSITION lj_err_unwind_win64(EXCEPTION_RECORD *rec,
     if (cf2) {  /* We catch it, so start unwinding the upper frames. */
       if (rec->ExceptionCode == LJ_MSVC_EXCODE ||
 	  rec->ExceptionCode == LJ_GCC_EXCODE) {
+#if LJ_TARGET_WINDOWS
 	__DestructExceptionObject(rec, 1);
+#endif
 	setstrV(L, L->top++, lj_err_str(L, LJ_ERR_ERRCPP));
       } else if (!LJ_EXCODE_CHECK(rec->ExceptionCode)) {
 	/* Don't catch access violations etc. */
diff --git a/src/lj_ffrecord.c b/src/lj_ffrecord.c
index 69f71ab..3083225 100644
--- a/src/lj_ffrecord.c
+++ b/src/lj_ffrecord.c
@@ -366,11 +366,12 @@ static void LJ_FASTCALL recff_ipairs_aux(jit_State *J, RecordFFData *rd)
 
 static void LJ_FASTCALL recff_ipairs(jit_State *J, RecordFFData *rd)
 {
-  if (!(LJ_52 && recff_metacall(J, rd, MM_ipairs))) {
-    TRef tab = J->base[0];
-    if (tref_istab(tab)) {
+  TRef tr = J->base[0];
+  if (!((LJ_52 || (LJ_HASFFI && tref_iscdata(tr))) &&
+	recff_metacall(J, rd, MM_ipairs))) {
+    if (tref_istab(tr)) {
       J->base[0] = lj_ir_kfunc(J, funcV(&J->fn->c.upvalue[0]));
-      J->base[1] = tab;
+      J->base[1] = tr;
       J->base[2] = lj_ir_kint(J, 0);
       rd->nres = 3;
     }  /* else: Interpreter will throw. */
diff --git a/src/lj_snap.c b/src/lj_snap.c
index 5c870ba..50412bc 100644
--- a/src/lj_snap.c
+++ b/src/lj_snap.c
@@ -26,9 +26,6 @@
 #include "lj_cdata.h"
 #endif
 
-/* Some local macros to save typing. Undef'd at the end. */
-#define IR(ref)		(&J->cur.ir[(ref)])
-
 /* Pass IR on to next optimization in chain (FOLD). */
 #define emitir(ot, a, b)	(lj_ir_set(J, (ot), (a), (b)), lj_opt_fold(J))
 
@@ -73,7 +70,7 @@ static MSize snapshot_slots(jit_State *J, SnapEntry *map, BCReg nslots)
     IRRef ref = tref_ref(tr);
     if (ref) {
       SnapEntry sn = SNAP_TR(s, tr);
-      IRIns *ir = IR(ref);
+      IRIns *ir = &J->cur.ir[ref];
       if (!(sn & (SNAP_CONT|SNAP_FRAME)) &&
 	  ir->o == IR_SLOAD && ir->op1 == s && ref > retf) {
 	/* No need to snapshot unmodified non-inherited slots. */
@@ -404,24 +401,24 @@ static TRef snap_pref(jit_State *J, GCtrace *T, SnapEntry *map, MSize nmax,
 }
 
 /* Check whether a sunk store corresponds to an allocation. Slow path. */
-static int snap_sunk_store2(jit_State *J, IRIns *ira, IRIns *irs)
+static int snap_sunk_store2(GCtrace *T, IRIns *ira, IRIns *irs)
 {
   if (irs->o == IR_ASTORE || irs->o == IR_HSTORE ||
       irs->o == IR_FSTORE || irs->o == IR_XSTORE) {
-    IRIns *irk = IR(irs->op1);
+    IRIns *irk = &T->ir[irs->op1];
     if (irk->o == IR_AREF || irk->o == IR_HREFK)
-      irk = IR(irk->op1);
-    return (IR(irk->op1) == ira);
+      irk = &T->ir[irk->op1];
+    return (&T->ir[irk->op1] == ira);
   }
   return 0;
 }
 
 /* Check whether a sunk store corresponds to an allocation. Fast path. */
-static LJ_AINLINE int snap_sunk_store(jit_State *J, IRIns *ira, IRIns *irs)
+static LJ_AINLINE int snap_sunk_store(GCtrace *T, IRIns *ira, IRIns *irs)
 {
   if (irs->s != 255)
     return (ira + irs->s == irs);  /* Fast check. */
-  return snap_sunk_store2(J, ira, irs);
+  return snap_sunk_store2(T, ira, irs);
 }
 
 /* Replay snapshot state to setup side trace. */
@@ -484,7 +481,7 @@ void lj_snap_replay(jit_State *J, GCtrace *T)
 	} else {
 	  IRIns *irs;
 	  for (irs = ir+1; irs < irlast; irs++)
-	    if (irs->r == RID_SINK && snap_sunk_store(J, ir, irs)) {
+	    if (irs->r == RID_SINK && snap_sunk_store(T, ir, irs)) {
 	      if (snap_pref(J, T, map, nent, seen, irs->op2) == 0)
 		snap_pref(J, T, map, nent, seen, T->ir[irs->op2].op1);
 	      else if ((LJ_SOFTFP || (LJ_32 && LJ_HASFFI)) &&
@@ -518,13 +515,13 @@ void lj_snap_replay(jit_State *J, GCtrace *T)
 	    op2 = emitir_raw(IRT(IR_HIOP, IRT_I64), op2,
 			     snap_pref(J, T, map, nent, seen, (ir+1)->op2));
 	  }
-	  J->slot[snap_slot(sn)] = emitir(ir->ot, op1, op2);
+	  J->slot[snap_slot(sn)] = emitir(ir->ot & ~(IRT_MARK|IRT_ISPHI), op1, op2);
 	} else {
 	  IRIns *irs;
 	  TRef tr = emitir(ir->ot, op1, op2);
 	  J->slot[snap_slot(sn)] = tr;
 	  for (irs = ir+1; irs < irlast; irs++)
-	    if (irs->r == RID_SINK && snap_sunk_store(J, ir, irs)) {
+	    if (irs->r == RID_SINK && snap_sunk_store(T, ir, irs)) {
 	      IRIns *irr = &T->ir[irs->op1];
 	      TRef val, key = irr->op2, tmp = tr;
 	      if (irr->o != IR_FREF) {
@@ -726,7 +723,7 @@ static void snap_unsink(jit_State *J, GCtrace *T, ExitState *ex,
     } else {
       IRIns *irs, *irlast = &T->ir[T->snap[snapno].ref];
       for (irs = ir+1; irs < irlast; irs++)
-	if (irs->r == RID_SINK && snap_sunk_store(J, ir, irs)) {
+	if (irs->r == RID_SINK && snap_sunk_store(T, ir, irs)) {
 	  IRIns *iro = &T->ir[T->ir[irs->op1].op2];
 	  uint8_t *p = (uint8_t *)cd;
 	  CTSize szs;
@@ -759,7 +756,7 @@ static void snap_unsink(jit_State *J, GCtrace *T, ExitState *ex,
     settabV(J->L, o, t);
     irlast = &T->ir[T->snap[snapno].ref];
     for (irs = ir+1; irs < irlast; irs++)
-      if (irs->r == RID_SINK && snap_sunk_store(J, ir, irs)) {
+      if (irs->r == RID_SINK && snap_sunk_store(T, ir, irs)) {
 	IRIns *irk = &T->ir[irs->op1];
 	TValue tmp, *val;
 	lua_assert(irs->o == IR_ASTORE || irs->o == IR_HSTORE ||
@@ -859,7 +856,6 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr)
   return pc;
 }
 
-#undef IR
 #undef emitir_raw
 #undef emitir