x86: Remove x87 support from interpreter.

SSE2 required from now on.

src/Makefile

@@ -42,13 +42,10 @@ CCOPT= -O2 -fomit-frame-pointer
 #
 # Target-specific compiler options:
 #
-# x86 only: it's recommended to compile at least for i686. Better yet,
-# compile for an architecture that has SSE2, too (-msse -msse2).
-#
 # x86/x64 only: For GCC 4.2 or higher and if you don't intend to distribute
 # the binaries to a different machine you could also use: -march=native
 #
-CCOPT_x86= -march=i686
+CCOPT_x86= -march=i686 -msse -msse2 -mfpmath=sse
 CCOPT_x64=
 CCOPT_arm=
 CCOPT_ppc=
@@ -394,11 +391,6 @@ DASM_AFLAGS+= -D VER=$(subst LJ_ARCH_VERSION_,,$(filter LJ_ARCH_VERSION_%,$(subs
 ifeq (Windows,$(TARGET_SYS))
   DASM_AFLAGS+= -D WIN
 endif
-ifeq (x86,$(TARGET_LJARCH))
-  ifneq (,$(findstring __SSE2__ 1,$(TARGET_TESTARCH)))
-    DASM_AFLAGS+= -D SSE
-  endif
-else
 ifeq (x64,$(TARGET_LJARCH))
   DASM_ARCH= x86
 else
@@ -423,7 +415,6 @@ ifeq (ppc,$(TARGET_LJARCH))
 endif
 endif
 endif
-endif
 
 DASM_FLAGS= $(DASM_XFLAGS) $(DASM_AFLAGS)
 DASM_DASC= vm_$(DASM_ARCH).dasc

src/lib_jit.c

@@ -538,18 +538,14 @@ static uint32_t jit_cpudetect(lua_State *L)
   uint32_t features[4];
   if (lj_vm_cpuid(0, vendor) && lj_vm_cpuid(1, features)) {
 #if !LJ_HASJIT
-#define JIT_F_CMOV	1
 #define JIT_F_SSE2	2
 #endif
-    flags |= ((features[3] >> 15)&1) * JIT_F_CMOV;
     flags |= ((features[3] >> 26)&1) * JIT_F_SSE2;
 #if LJ_HASJIT
     flags |= ((features[2] >> 0)&1) * JIT_F_SSE3;
     flags |= ((features[2] >> 19)&1) * JIT_F_SSE4_1;
     if (vendor[2] == 0x6c65746e) {  /* Intel. */
-      if ((features[0] & 0x0ff00f00) == 0x00000f00)  /* P4. */
+      if ((features[0] & 0x0fff0ff0) == 0x000106c0)  /* Atom. */
-	flags |= JIT_F_P4;  /* Currently unused. */
-      else if ((features[0] & 0x0fff0ff0) == 0x000106c0)  /* Atom. */
 	flags |= JIT_F_LEA_AGU;
     } else if (vendor[2] == 0x444d4163) {  /* AMD. */
       uint32_t fam = (features[0] & 0x0ff00f00);
@@ -562,14 +558,8 @@ static uint32_t jit_cpudetect(lua_State *L)
   }
   /* Check for required instruction set support on x86 (unnecessary on x64). */
 #if LJ_TARGET_X86
-#if !defined(LUAJIT_CPU_NOCMOV)
-  if (!(flags & JIT_F_CMOV))
-    luaL_error(L, "CPU not supported");
-#endif
-#if defined(LUAJIT_CPU_SSE2)
   if (!(flags & JIT_F_SSE2))
-    luaL_error(L, "CPU does not support SSE2 (recompile without -DLUAJIT_CPU_SSE2)");
+    luaL_error(L, "CPU with SSE2 required");
-#endif
 #endif
 #elif LJ_TARGET_ARM
 #if LJ_HASJIT
@@ -631,11 +621,7 @@ static void jit_init(lua_State *L)
   uint32_t flags = jit_cpudetect(L);
 #if LJ_HASJIT
   jit_State *J = L2J(L);
-#if LJ_TARGET_X86
+  J->flags = flags | JIT_F_ON | JIT_F_OPT_DEFAULT;
-  /* Silently turn off the JIT compiler on CPUs without SSE2. */
-  if ((flags & JIT_F_SSE2))
-#endif
-    J->flags = flags | JIT_F_ON | JIT_F_OPT_DEFAULT;
   memcpy(J->param, jit_param_default, sizeof(J->param));
   lj_dispatch_update(G(L));
 #else
@@ -645,6 +631,7 @@ static void jit_init(lua_State *L)
 
 LUALIB_API int luaopen_jit(lua_State *L)
 {
+  jit_init(L);
   lua_pushliteral(L, LJ_OS_NAME);
   lua_pushliteral(L, LJ_ARCH_NAME);
   lua_pushinteger(L, LUAJIT_VERSION_NUM);
@@ -657,7 +644,6 @@ LUALIB_API int luaopen_jit(lua_State *L)
   LJ_LIB_REG(L, "jit.opt", jit_opt);
 #endif
   L->top -= 2;
-  jit_init(L);
   return 1;
 }
 

src/lj_asm.c

@@ -1730,7 +1730,7 @@ static void asm_setup_regsp(ASMState *as)
       break;
     case IR_FPMATH:
 #if LJ_TARGET_X86ORX64
-      if (ir->op2 == IRFPM_EXP2) {  /* May be joined to lj_vm_pow_sse. */
+      if (ir->op2 == IRFPM_EXP2) {  /* May be joined to pow. */
 	ir->prev = REGSP_HINT(RID_XMM0);
 #if !LJ_64
 	if (as->evenspill < 4)  /* Leave room for 16 byte scratch area. */

src/lj_jit.h

@@ -14,18 +14,16 @@
 
 /* CPU-specific JIT engine flags. */
 #if LJ_TARGET_X86ORX64
-#define JIT_F_CMOV		0x00000010
+#define JIT_F_SSE2		0x00000010
-#define JIT_F_SSE2		0x00000020
+#define JIT_F_SSE3		0x00000020
-#define JIT_F_SSE3		0x00000040
+#define JIT_F_SSE4_1		0x00000040
-#define JIT_F_SSE4_1		0x00000080
+#define JIT_F_PREFER_IMUL	0x00000080
-#define JIT_F_P4		0x00000100
+#define JIT_F_SPLIT_XMM		0x00000100
-#define JIT_F_PREFER_IMUL	0x00000200
+#define JIT_F_LEA_AGU		0x00000200
-#define JIT_F_SPLIT_XMM		0x00000400
-#define JIT_F_LEA_AGU		0x00000800
 
 /* Names for the CPU-specific flags. Must match the order above. */
-#define JIT_F_CPU_FIRST		JIT_F_CMOV
+#define JIT_F_CPU_FIRST		JIT_F_SSE2
-#define JIT_F_CPUSTRING		"\4CMOV\4SSE2\4SSE3\6SSE4.1\2P4\3AMD\2K8\4ATOM"
+#define JIT_F_CPUSTRING		"\4SSE2\4SSE3\6SSE4.1\3AMD\2K8\4ATOM"
 #elif LJ_TARGET_ARM
 #define JIT_F_ARMV6_		0x00000010
 #define JIT_F_ARMV6T2_		0x00000020

src/lj_vm.h

@@ -49,12 +49,14 @@ LJ_ASMF void lj_vm_exit_handler(void);
 LJ_ASMF void lj_vm_exit_interp(void);
 
 /* Internal math helper functions. */
-#if LJ_TARGET_X86ORX64 || LJ_TARGET_PPC
+#if LJ_TARGET_PPC
 #define lj_vm_floor	floor
 #define lj_vm_ceil	ceil
 #else
 LJ_ASMF double lj_vm_floor(double);
+#if !LJ_TARGET_X86ORX64
 LJ_ASMF double lj_vm_ceil(double);
+#endif
 #if LJ_TARGET_ARM
 LJ_ASMF double lj_vm_floor_sf(double);
 LJ_ASMF double lj_vm_ceil_sf(double);

src/msvcbuild.bat

@@ -35,6 +35,7 @@ if exist minilua.exe.manifest^
 @if errorlevel 8 goto :X64
 @set DASMFLAGS=-D WIN -D JIT -D FFI
 @set LJARCH=x86
+@set LJCOMPILE=%LJCOMPILE% /arch:SSE2
 :X64
 minilua %DASM% -LN %DASMFLAGS% -o host\buildvm_arch.h vm_x86.dasc
 @if errorlevel 1 goto :BAD