ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Linux: Added more correct __asm__ qualifiers and conditions; including __volatile__ on a lot of asm code (it should really be the default behavior and non-vlatile the specifier, but whatever >_<), and added register clobber specifiers. Might help unbreak some of GCC 4.4's optimization problems, although VIFdma's uber-hack SSE optimization looks like a real problem. 

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@1964 96395faa-99c1-11dd-bbfe-3dabce05a288
This commit is contained in:
Jake.Stine 2009-10-05 11:05:11 +00:00
parent a8b55baa38
commit b71b837efa
13 changed files with 150 additions and 138 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

86
common/include/Utilities/lnx_memzero.h
View File

@ -102,64 +102,64 @@ static __forceinline void memset_8( void *dest )
return; return;
case 3: case 3:
__asm__ __asm__ volatile
( (
".intel_syntax noprefix\n" ".intel_syntax noprefix\n"
"cld\n" "cld\n"
// "mov edi, %[dest]\n" // "mov edi, %[dest]\n"
// "mov eax, %[data32]\n" // "mov eax, %[data32]\n"
"stosd\n"
"stosd\n"
"stosd\n"
".att_syntax\n"
:
// Input specifiers: D - edi, a -- eax, c ecx
: [dest]"D"(dest), [data32]"a"(data32)
: "memory"
);
return;
case 4:
__asm__ volatile
(
".intel_syntax noprefix\n"
"cld\n"
// "mov edi, %[dest]\n"
// "mov eax, %[data32]\n"
"stosd\n"
"stosd\n"
"stosd\n"
"stosd\n"
".att_syntax\n"
:
: [dest]"D"(dest), [data32]"a"(data32)
: "memory"
);
return;
case 5:
__asm__ volatile
(
".intel_syntax noprefix\n"
"cld\n"
// "mov edi, %[dest]\n"
// "mov eax, %[data32]\n"
"stosd\n"
"stosd\n"
"stosd\n" "stosd\n"
"stosd\n" "stosd\n"
"stosd\n" "stosd\n"
".att_syntax\n" ".att_syntax\n"
: :
: [dest]"D"(dest), [data32]"a"(data32) : [dest]"D"(dest), [data32]"a"(data32)
// D - edi, a -- eax, c ecx : "memory"
:
);
return;
case 4:
__asm__
(
".intel_syntax noprefix\n"
"cld\n"
// "mov edi, %[dest]\n"
// "mov eax, %[data32]\n"
"stosd\n"
"stosd\n"
"stosd\n"
"stosd\n"
".att_syntax\n"
:
: [dest]"D"(dest), [data32]"a"(data32)
:
);
return;
case 5:
__asm__
(
".intel_syntax noprefix\n"
"cld\n"
// "mov edi, %[dest]\n"
// "mov eax, %[data32]\n"
"stosd\n"
"stosd\n"
"stosd\n"
"stosd\n"
"stosd\n"
".att_syntax\n"
:
: [dest]"D"(dest), [data32]"a"(data32)
:
); );
return; return;
default: default:
__asm__ __asm__ volatile
( (
".intel_syntax noprefix\n" ".intel_syntax noprefix\n"
"cld\n" "cld\n"
@ -170,7 +170,7 @@ static __forceinline void memset_8( void *dest )
".att_syntax\n" ".att_syntax\n"
: :
: [remdat]"c"(remdat), [dest]"D"(dest), [data32]"a"(data32) : [remdat]"c"(remdat), [dest]"D"(dest), [data32]"a"(data32)
: : "memory"
); );
return; return;
} }

7
common/src/x86emitter/cpudetect.cpp
View File

@ -27,12 +27,9 @@ __aligned16 x86CPU_INFO x86caps;
static s32 iCpuId( u32 cmd, u32 *regs ) static s32 iCpuId( u32 cmd, u32 *regs )
{ {
#ifdef _MSC_VER #ifdef _MSC_VER
__asm __asm xor ecx, ecx; // ecx should be zero for CPUID(4)
{
xor ecx, ecx; // ecx should be zero for CPUID(4)
}
#else #else
__asm__ ( "xor %ecx, %ecx" ); __asm__ __volatile__ ( "xor %ecx, %ecx" );
#endif #endif
__cpuid( (int*)regs, cmd ); __cpuid( (int*)regs, cmd );

16
common/src/x86emitter/tools.cpp
View File

@ -65,7 +65,7 @@ __forceinline void FreezeMMXRegs(int save)
emms emms
} }
#else #else
__asm__( __asm__ volatile(
".intel_syntax noprefix\n" ".intel_syntax noprefix\n"
"movq [%[g_globalMMXData]+0x00], mm0\n" "movq [%[g_globalMMXData]+0x00], mm0\n"
"movq [%[g_globalMMXData]+0x08], mm1\n" "movq [%[g_globalMMXData]+0x08], mm1\n"
@ -76,7 +76,7 @@ __forceinline void FreezeMMXRegs(int save)
"movq [%[g_globalMMXData]+0x30], mm6\n" "movq [%[g_globalMMXData]+0x30], mm6\n"
"movq [%[g_globalMMXData]+0x38], mm7\n" "movq [%[g_globalMMXData]+0x38], mm7\n"
"emms\n" "emms\n"
".att_syntax\n" : : [g_globalMMXData]"r"(g_globalMMXData) ".att_syntax\n" : : [g_globalMMXData]"r"(g_globalMMXData) : "memory"
); );
#endif #endif
@ -105,7 +105,7 @@ __forceinline void FreezeMMXRegs(int save)
emms emms
} }
#else #else
__asm__( __asm__ volatile(
".intel_syntax noprefix\n" ".intel_syntax noprefix\n"
"movq mm0, [%[g_globalMMXData]+0x00]\n" "movq mm0, [%[g_globalMMXData]+0x00]\n"
"movq mm1, [%[g_globalMMXData]+0x08]\n" "movq mm1, [%[g_globalMMXData]+0x08]\n"
@ -116,7 +116,7 @@ __forceinline void FreezeMMXRegs(int save)
"movq mm6, [%[g_globalMMXData]+0x30]\n" "movq mm6, [%[g_globalMMXData]+0x30]\n"
"movq mm7, [%[g_globalMMXData]+0x38]\n" "movq mm7, [%[g_globalMMXData]+0x38]\n"
"emms\n" "emms\n"
".att_syntax\n" : : [g_globalMMXData]"r"(g_globalMMXData) ".att_syntax\n" : : [g_globalMMXData]"r"(g_globalMMXData) : "memory"
); );
#endif #endif
} }
@ -154,7 +154,7 @@ __forceinline void FreezeXMMRegs(int save)
} }
#else #else
__asm__( __asm__ volatile(
".intel_syntax noprefix\n" ".intel_syntax noprefix\n"
"movaps [%[g_globalXMMData]+0x00], xmm0\n" "movaps [%[g_globalXMMData]+0x00], xmm0\n"
"movaps [%[g_globalXMMData]+0x10], xmm1\n" "movaps [%[g_globalXMMData]+0x10], xmm1\n"
@ -164,7 +164,7 @@ __forceinline void FreezeXMMRegs(int save)
"movaps [%[g_globalXMMData]+0x50], xmm5\n" "movaps [%[g_globalXMMData]+0x50], xmm5\n"
"movaps [%[g_globalXMMData]+0x60], xmm6\n" "movaps [%[g_globalXMMData]+0x60], xmm6\n"
"movaps [%[g_globalXMMData]+0x70], xmm7\n" "movaps [%[g_globalXMMData]+0x70], xmm7\n"
".att_syntax\n" : : [g_globalXMMData]"r"(g_globalXMMData) ".att_syntax\n" : : [g_globalXMMData]"r"(g_globalXMMData) : "memory"
); );
#endif // _MSC_VER #endif // _MSC_VER
@ -196,7 +196,7 @@ __forceinline void FreezeXMMRegs(int save)
} }
#else #else
__asm__( __asm__ volatile(
".intel_syntax noprefix\n" ".intel_syntax noprefix\n"
"movaps xmm0, [%[g_globalXMMData]+0x00]\n" "movaps xmm0, [%[g_globalXMMData]+0x00]\n"
"movaps xmm1, [%[g_globalXMMData]+0x10]\n" "movaps xmm1, [%[g_globalXMMData]+0x10]\n"
@ -206,7 +206,7 @@ __forceinline void FreezeXMMRegs(int save)
"movaps xmm5, [%[g_globalXMMData]+0x50]\n" "movaps xmm5, [%[g_globalXMMData]+0x50]\n"
"movaps xmm6, [%[g_globalXMMData]+0x60]\n" "movaps xmm6, [%[g_globalXMMData]+0x60]\n"
"movaps xmm7, [%[g_globalXMMData]+0x70]\n" "movaps xmm7, [%[g_globalXMMData]+0x70]\n"
".att_syntax\n" : : [g_globalXMMData]"r"(g_globalXMMData) ".att_syntax\n" : : [g_globalXMMData]"r"(g_globalXMMData) : "memory"
); );
#endif // _MSC_VER #endif // _MSC_VER

50
pcsx2/IPU/yuv2rgb.cpp
View File

@ -55,7 +55,7 @@ enum
BCb_COEFF = 0x40 BCb_COEFF = 0x40
}; };
static volatile const __aligned16 SSE2_Tables sse2_tables = static const __aligned16 SSE2_Tables sse2_tables =
{ {
{0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000}, // c_bias {0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000}, // c_bias
{16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16}, // y_bias {16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16}, // y_bias
@ -211,8 +211,10 @@ ihatemsvc:
cmp esi, 64 cmp esi, 64
jne tworows jne tworows
} }
#elif defined(__GNUC__) #elif defined(__GNUC__)
__asm__(
__asm__ __volatile__ (
".intel_syntax noprefix\n" ".intel_syntax noprefix\n"
"mov eax, 1\n" "mov eax, 1\n"
"xor esi, esi\n" "xor esi, esi\n"
@ -220,8 +222,8 @@ ihatemsvc:
// Use ecx and edx as base pointers, to allow for Mod/RM form on memOps. // Use ecx and edx as base pointers, to allow for Mod/RM form on memOps.
// This saves 2-3 bytes per instruction where these are used. :) // This saves 2-3 bytes per instruction where these are used. :)
"mov ecx, offset %c[yuv2rgb_temp]\n" //"mov ecx, offset %c[yuv2rgb_temp]\n"
"mov edx, offset %c[sse2_tables]+64\n" //"mov edx, offset %c[sse2_tables]+64\n"
".align 16\n" ".align 16\n"
"tworows:\n" "tworows:\n"
@ -237,29 +239,29 @@ ihatemsvc:
// unfortunately I don't think this will matter despite being // unfortunately I don't think this will matter despite being
// technically potentially a little faster, but this is // technically potentially a little faster, but this is
// equivalent to an add or sub // equivalent to an add or sub
"pxor xmm2, xmmword ptr [edx+%c[C_BIAS]]\n" // xmm2 <-- 8 x (Cb - 128) << 8 "pxor xmm2, xmmword ptr [%[sse2_tables]+%c[C_BIAS]]\n" // xmm2 <-- 8 x (Cb - 128) << 8
"pxor xmm0, xmmword ptr [edx+%c[C_BIAS]]\n" // xmm0 <-- 8 x (Cr - 128) << 8 "pxor xmm0, xmmword ptr [%[sse2_tables]+%c[C_BIAS]]\n" // xmm0 <-- 8 x (Cr - 128) << 8
"movaps xmm1, xmm0\n" "movaps xmm1, xmm0\n"
"movaps xmm3, xmm2\n" "movaps xmm3, xmm2\n"
"pmulhw xmm1, xmmword ptr [edx+%c[GCr_COEFF]]\n" "pmulhw xmm1, xmmword ptr [%[sse2_tables]+%c[GCr_COEFF]]\n"
"pmulhw xmm3, xmmword ptr [edx+%c[GCb_COEFF]]\n" "pmulhw xmm3, xmmword ptr [%[sse2_tables]+%c[GCb_COEFF]]\n"
"pmulhw xmm0, xmmword ptr [edx+%c[RCr_COEFF]]\n" "pmulhw xmm0, xmmword ptr [%[sse2_tables]+%c[RCr_COEFF]]\n"
"pmulhw xmm2, xmmword ptr [edx+%c[BCb_COEFF]]\n" "pmulhw xmm2, xmmword ptr [%[sse2_tables]+%c[BCb_COEFF]]\n"
"paddsw xmm1, xmm3\n" "paddsw xmm1, xmm3\n"
// store for the next line; looking at the code above // store for the next line; looking at the code above
// compared to the code below, I have to wonder whether // compared to the code below, I have to wonder whether
// this was worth the hassle // this was worth the hassle
"movaps xmmword ptr [ecx], xmm0\n" "movaps xmmword ptr [%[yuv2rgb_temp]], xmm0\n"
"movaps xmmword ptr [ecx+16], xmm1\n" "movaps xmmword ptr [%[yuv2rgb_temp]+16], xmm1\n"
"movaps xmmword ptr [ecx+32], xmm2\n" "movaps xmmword ptr [%[yuv2rgb_temp]+32], xmm2\n"
"jmp ihategcctoo\n" "jmp ihategcctoo\n"
".align 16\n" ".align 16\n"
"onerow:\n" "onerow:\n"
"movaps xmm0, xmmword ptr [ecx]\n" "movaps xmm0, xmmword ptr [%[yuv2rgb_temp]]\n"
"movaps xmm1, xmmword ptr [ecx+16]\n" "movaps xmm1, xmmword ptr [%[yuv2rgb_temp]+16]\n"
"movaps xmm2, xmmword ptr [ecx+32]\n" "movaps xmm2, xmmword ptr [%[yuv2rgb_temp]+32]\n"
"ihategcctoo:\n" "ihategcctoo:\n"
"movaps xmm3, xmm0\n" "movaps xmm3, xmm0\n"
@ -267,13 +269,13 @@ ihatemsvc:
"movaps xmm5, xmm2\n" "movaps xmm5, xmm2\n"
"movaps xmm6, xmmword ptr [mb8+edi]\n" "movaps xmm6, xmmword ptr [mb8+edi]\n"
"psubusb xmm6, xmmword ptr [edx+%c[Y_BIAS]]\n" "psubusb xmm6, xmmword ptr [%[sse2_tables]+%c[Y_BIAS]]\n"
"movaps xmm7, xmm6\n" "movaps xmm7, xmm6\n"
"psllw xmm6, 8\n" // xmm6 <- Y << 8 for pixels 0,2,4,6,8,10,12,14 "psllw xmm6, 8\n" // xmm6 <- Y << 8 for pixels 0,2,4,6,8,10,12,14
"pand xmm7, xmmword ptr [edx+%c[Y_MASK]]\n" // xmm7 <- Y << 8 for pixels 1,3,5,7,9,11,13,15 "pand xmm7, xmmword ptr [%[sse2_tables]+%c[Y_MASK]]\n" // xmm7 <- Y << 8 for pixels 1,3,5,7,9,11,13,15
"pmulhuw xmm6, xmmword ptr [edx+%c[Y_COEFF]]\n" "pmulhuw xmm6, xmmword ptr [%[sse2_tables]+%c[Y_COEFF]]\n"
"pmulhuw xmm7, xmmword ptr [edx+%c[Y_COEFF]]\n" "pmulhuw xmm7, xmmword ptr [%[sse2_tables]+%c[Y_COEFF]]\n"
"paddsw xmm0, xmm6\n" "paddsw xmm0, xmm6\n"
"paddsw xmm3, xmm7\n" "paddsw xmm3, xmm7\n"
@ -283,7 +285,7 @@ ihatemsvc:
"paddsw xmm5, xmm7\n" "paddsw xmm5, xmm7\n"
// round // round
"movaps xmm6, xmmword ptr [edx+%c[ROUND_1BIT]]\n" "movaps xmm6, xmmword ptr [%[sse2_tables]+%c[ROUND_1BIT]]\n"
"paddw xmm0, xmm6\n" "paddw xmm0, xmm6\n"
"paddw xmm1, xmm6\n" "paddw xmm1, xmm6\n"
"paddw xmm2, xmm6\n" "paddw xmm2, xmm6\n"
@ -343,11 +345,11 @@ ihatemsvc:
:[C_BIAS]"i"(C_BIAS), [Y_BIAS]"i"(Y_BIAS), [Y_MASK]"i"(Y_MASK), :[C_BIAS]"i"(C_BIAS), [Y_BIAS]"i"(Y_BIAS), [Y_MASK]"i"(Y_MASK),
[ROUND_1BIT]"i"(ROUND_1BIT), [Y_COEFF]"i"(Y_COEFF), [GCr_COEFF]"i"(GCr_COEFF), [ROUND_1BIT]"i"(ROUND_1BIT), [Y_COEFF]"i"(Y_COEFF), [GCr_COEFF]"i"(GCr_COEFF),
[GCb_COEFF]"i"(GCb_COEFF), [RCr_COEFF]"i"(RCr_COEFF), [BCb_COEFF]"i"(BCb_COEFF), [GCb_COEFF]"i"(GCb_COEFF), [RCr_COEFF]"i"(RCr_COEFF), [BCb_COEFF]"i"(BCb_COEFF),
[yuv2rgb_temp]"i"(yuv2rgb_temp), [sse2_tables]"i"(&sse2_tables) [yuv2rgb_temp]"r"(yuv2rgb_temp), [sse2_tables]"r"(&sse2_tables)
: : "eax", "ebx", "esi", "edi", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7", "memory"
); );
#else #else
#error Unsupported compiler # error Unsupported compiler
#endif #endif
FreezeXMMRegs(0); FreezeXMMRegs(0);

9
pcsx2/VifDma.cpp
View File

@ -646,6 +646,11 @@ static void VIFunpack(u32 *data, vifCode *v, unsigned int size, const unsigned i
int writemask; int writemask;
u32 oldcycle = -1; u32 oldcycle = -1;
// yay evil .. let's just set some XMM registers in the middle of C code
// and "hope" they get preserved, in spite of the fact that x86-32 ABI specifies
// these as "clobberable" registers (so any printf or something could decide to
// clobber them, and has every right to... >_<) --air
#ifdef _MSC_VER #ifdef _MSC_VER
if (VIFdmanum) if (VIFdmanum)
{ {
@ -658,6 +663,10 @@ static void VIFunpack(u32 *data, vifCode *v, unsigned int size, const unsigned i
__asm movaps XMM_COL, xmmword ptr [g_vifmask.Col0] __asm movaps XMM_COL, xmmword ptr [g_vifmask.Col0]
} }
#else #else
// I'd add volatile to these, but what's the point? This code already breaks
// like 5000 coveted rules of binary interfacing regardless, and is only working by
// the miracles and graces of a profound deity (or maybe it doesn't -- linux port
// *does* have stability issues, especially in GCC 4.4). --air
if (VIFdmanum) if (VIFdmanum)
{ {
__asm__(".intel_syntax noprefix\n" __asm__(".intel_syntax noprefix\n"

2
pcsx2/x86/iMisc.cpp
View File

@ -46,7 +46,7 @@ void SetCPUState(u32 sseMXCSR, u32 sseVUMXCSR)
#ifdef _MSC_VER #ifdef _MSC_VER
__asm ldmxcsr g_sseMXCSR; // set the new sse control __asm ldmxcsr g_sseMXCSR; // set the new sse control
#else #else
__asm__("ldmxcsr %[g_sseMXCSR]" : : [g_sseMXCSR]"m"(g_sseMXCSR) ); __asm__ __volatile__("ldmxcsr %[g_sseMXCSR]" : : [g_sseMXCSR]"m"(g_sseMXCSR) );
#endif #endif
//g_sseVUMXCSR = g_sseMXCSR|0x6000; //g_sseVUMXCSR = g_sseMXCSR|0x6000;
} }

25
pcsx2/x86/iR3000A.cpp
View File

@ -657,22 +657,29 @@ static __forceinline s32 recExecuteBlock( s32 eeCycles )
pop ebx pop ebx
} }
#else #else
__asm__ __asm__ __volatile__
( (
// We should be able to rely on GAS syntax (the register clobber list) as a
// replacement for manual push/pop of unpreserved registers.
//
// EBP note: As I feared, EBP is "required" for C++ excepion handling in Linux, and trying
// to issue a clobber specifier for it causes an error. We really need to find a way to
// disable EBP regalloc in iCore. --air
".intel_syntax noprefix\n" ".intel_syntax noprefix\n"
"push ebx\n" //"push ebx\n"
"push esi\n" //"push esi\n"
"push edi\n" //"push edi\n"
"push ebp\n" "push ebp\n"
"call iopDispatcherReg\n" "call iopDispatcherReg\n"
"pop ebp\n" "pop ebp\n"
"pop edi\n" //"pop edi\n"
"pop esi\n" //"pop esi\n"
"pop ebx\n" //"pop ebx\n"
".att_syntax\n" ".att_syntax\n"
); : : : "eax", "ebx", "ecx", "edx", "esi", "edi", "memory" );
#endif #endif
return psxBreak + psxCycleEE; return psxBreak + psxCycleEE;
@ -836,7 +843,7 @@ static void checkcodefn()
#ifdef _MSC_VER #ifdef _MSC_VER
__asm mov pctemp, eax; __asm mov pctemp, eax;
#else #else
__asm__("movl %%eax, %[pctemp]" : : [pctemp]"m"(pctemp) ); __asm__ __volatile__("movl %%eax, %[pctemp]" : [pctemp]"m="(pctemp) );
#endif #endif
Console.WriteLn("iop code changed! %x", pctemp); Console.WriteLn("iop code changed! %x", pctemp);
} }

4
pcsx2/x86/iVif.cpp
View File

@ -104,7 +104,7 @@ static void __forceinline UseOldMaskCode(u32* &vif1masks, u32 &mask)
u8* p0 = (u8*)&s_maskarr[mask&15][0]; u8* p0 = (u8*)&s_maskarr[mask&15][0];
u8* p1 = (u8*)&s_maskarr[(mask>>4)&15][0]; u8* p1 = (u8*)&s_maskarr[(mask>>4)&15][0];
__asm__(".intel_syntax noprefix\n" __asm__ __volatile__(".intel_syntax noprefix\n"
"movaps xmm0, [%0]\n" "movaps xmm0, [%0]\n"
"movaps xmm1, [%1]\n" "movaps xmm1, [%1]\n"
"movaps xmm2, xmm0\n" "movaps xmm2, xmm0\n"
@ -121,6 +121,6 @@ static void __forceinline UseOldMaskCode(u32* &vif1masks, u32 &mask)
"movq [%2+40], xmm3\n" "movq [%2+40], xmm3\n"
"movhps [%2+48], xmm2\n" "movhps [%2+48], xmm2\n"
"movhps [%2+56], xmm3\n" "movhps [%2+56], xmm3\n"
".att_syntax\n" : : "r"(p0), "r"(p1), "r"(vif1masks) ); ".att_syntax\n" : : "r"(p0), "r"(p1), "r"(vif1masks) : "memory" );
} }
#endif #endif

43
pcsx2/x86/ix86-32/iR5900-32.cpp
View File

@ -601,22 +601,29 @@ static void recExecute()
#else // _MSC_VER #else // _MSC_VER
__asm__ __asm__ __volatile__
( (
// We should be able to rely on GAS syntax (the register clobber list) as a
// replacement for manual push/pop of unpreserved registers.
// EBP note: As I feared, EBP is "required" for C++ excepion handling in Linux, and trying
// to issue a clobber specifier for it causes an error. We really need to find a way to
// disable EBP regalloc in iCore. --air
".intel_syntax noprefix\n" ".intel_syntax noprefix\n"
"push ebx\n" //"push ebx\n"
"push esi\n" //"push esi\n"
"push edi\n" //"push edi\n"
"push ebp\n" //"push ebp\n"
"call DispatcherReg\n" "call DispatcherReg\n"
"pop ebp\n" //"pop ebp\n"
"pop edi\n" //"pop edi\n"
"pop esi\n" //"pop esi\n"
"pop ebx\n" //"pop ebx\n"
".att_syntax\n" ".att_syntax\n"
); : : : "eax", "ebx", "ecx", "edx", "esi", "edi", "memory" );
#endif #endif
} }
catch( Exception::ForceDispatcherReg& ) catch( Exception::ForceDispatcherReg& )
@ -679,21 +686,11 @@ void recClear(u32 addr, u32 size)
BASEBLOCKEX* pexblock; BASEBLOCKEX* pexblock;
BASEBLOCK* pblock; BASEBLOCK* pblock;
//why the hell?
#if 1
// necessary since recompiler doesn't call femms/emms // necessary since recompiler doesn't call femms/emms
#ifdef __INTEL_COMPILER #ifdef _MSC_VER
__asm__("emms"); asm emms;
#else #else
#ifdef _MSC_VER __asm__ __volatile__("emms");
if (x86caps.has3DNOWInstructionExtensions) __asm femms;
else __asm emms;
#else
if( x86caps.has3DNOWInstructionExtensions )__asm__("femms");
else
__asm__("emms");
#endif
#endif
#endif #endif
if ((addr) >= maxrecmem || !(recLUT[(addr) >> 16] + (addr & ~0xFFFFUL))) if ((addr) >= maxrecmem || !(recLUT[(addr) >> 16] + (addr & ~0xFFFFUL)))

4
plugins/spu2-x/src/Mixer.cpp
View File

@ -299,7 +299,7 @@ static s32 __forceinline GetNoiseValues()
#else #else
__asm__ ( __asm__ (
".intel_syntax\n" ".intel_syntax\n"
"MOV %%eax,%0\n" "MOV %%eax,%1\n"
"ROR %%eax,5\n" "ROR %%eax,5\n"
"XOR %%eax,0x9a\n" "XOR %%eax,0x9a\n"
"MOV %%ebx,%%eax\n" "MOV %%ebx,%%eax\n"
@ -308,7 +308,7 @@ static s32 __forceinline GetNoiseValues()
"XOR %%eax,%%ebx\n" "XOR %%eax,%%ebx\n"
"ROR %%eax,3\n" "ROR %%eax,3\n"
"MOV %0,%%eax\n" "MOV %0,%%eax\n"
".att_syntax\n" : :"r"(Seed)); ".att_syntax\n" : "m="(Seed) : "m"(Seed));
#endif #endif
return retval; return retval;
} }

4
plugins/zerospu2/voices.cpp
View File

@ -207,7 +207,7 @@ static void __forceinline GetNoiseValues(s32& VD)
#else #else
__asm__ ( __asm__ (
".intel_syntax\n" ".intel_syntax\n"
"MOV %%eax,%0\n" "MOV %%eax,%1\n"
"ROR %%eax,5\n" "ROR %%eax,5\n"
"XOR %%eax,0x9a\n" "XOR %%eax,0x9a\n"
"MOV %%ebx,%%eax\n" "MOV %%ebx,%%eax\n"
@ -216,7 +216,7 @@ static void __forceinline GetNoiseValues(s32& VD)
"XOR %%eax,%%ebx\n" "XOR %%eax,%%ebx\n"
"ROR %%eax,3\n" "ROR %%eax,3\n"
"MOV %0,%%eax\n" "MOV %0,%%eax\n"
".att_syntax\n" : :"r"(Seed)); ".att_syntax\n" : "r="(Seed) :"r"(Seed));
#endif #endif
} }