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

Jit64: turn 32 bit addresses into offsets

This commit is contained in:
Tillmann Karras 2015-04-20 20:47:51 +02:00
parent 9f15054358
commit a0597f0d62
11 changed files with 82 additions and 74 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
Source/Core/Common/x64Emitter.h
View File

@ -218,17 +218,17 @@ inline OpArg M(const T* ptr) {return OpArg((u64)(const void*)ptr, (int)SCALE_
inline OpArg R(X64Reg value) {return OpArg(0, SCALE_NONE, value);} inline OpArg R(X64Reg value) {return OpArg(0, SCALE_NONE, value);}
inline OpArg MatR(X64Reg value) {return OpArg(0, SCALE_ATREG, value);} inline OpArg MatR(X64Reg value) {return OpArg(0, SCALE_ATREG, value);}
inline OpArg MDisp(X64Reg value, int offset) inline OpArg MDisp(X64Reg value, ptrdiff_t offset)
{ {
return OpArg((u32)offset, SCALE_ATREG, value); return OpArg(offset, SCALE_ATREG, value);
} }
inline OpArg MComplex(X64Reg base, X64Reg scaled, int scale, int offset) inline OpArg MComplex(X64Reg base, X64Reg scaled, int scale, ptrdiff_t offset)
{ {
return OpArg(offset, scale, base, scaled); return OpArg(offset, scale, base, scaled);
} }
inline OpArg MScaled(X64Reg scaled, int scale, int offset) inline OpArg MScaled(X64Reg scaled, int scale, ptrdiff_t offset)
{ {
if (scale == SCALE_1) if (scale == SCALE_1)
return OpArg(offset, SCALE_ATREG, scaled); return OpArg(offset, SCALE_ATREG, scaled);
@ -247,17 +247,10 @@ inline OpArg Imm32(u32 imm) {return OpArg(imm, SCALE_IMM32);}
inline OpArg Imm64(u64 imm) {return OpArg(imm, SCALE_IMM64);} inline OpArg Imm64(u64 imm) {return OpArg(imm, SCALE_IMM64);}
inline OpArg ImmPtr(const void* imm) {return Imm64((u64)imm);} inline OpArg ImmPtr(const void* imm) {return Imm64((u64)imm);}
inline u32 PtrOffset(const void* ptr, const void* base) inline bool FitsInS32(const ptrdiff_t distance)
{ {
s64 distance = (s64)ptr-(s64)base; return distance < 0x80000000LL &&
if (distance >= 0x80000000LL || distance >= -0x80000000LL;
distance < -0x80000000LL)
{
_assert_msg_(DYNA_REC, 0, "pointer offset out of range");
return 0;
}
return (u32)distance;
} }
//usage: int a[]; ARRAY_OFFSET(a,10) //usage: int a[]; ARRAY_OFFSET(a,10)

16
Source/Core/Core/DSP/Jit/DSPJitUtil.cpp
View File

@ -20,14 +20,16 @@ void DSPEmitter::dsp_reg_stack_push(int stack_reg)
AND(8, R(AL), Imm8(DSP_STACK_MASK)); AND(8, R(AL), Imm8(DSP_STACK_MASK));
MOV(8, M(&g_dsp.reg_stack_ptr[stack_reg]), R(AL)); MOV(8, M(&g_dsp.reg_stack_ptr[stack_reg]), R(AL));
X64Reg tmp1; X64Reg tmp1, tmp2;
gpr.getFreeXReg(tmp1); gpr.getFreeXReg(tmp1);
gpr.getFreeXReg(tmp2);
//g_dsp.reg_stack[stack_reg][g_dsp.reg_stack_ptr[stack_reg]] = g_dsp.r[DSP_REG_ST0 + stack_reg]; //g_dsp.reg_stack[stack_reg][g_dsp.reg_stack_ptr[stack_reg]] = g_dsp.r[DSP_REG_ST0 + stack_reg];
MOV(16, R(tmp1), M(&g_dsp.r.st[stack_reg])); MOV(16, R(tmp1), M(&g_dsp.r.st[stack_reg]));
MOVZX(64, 8, RAX, R(AL)); MOVZX(64, 8, RAX, R(AL));
MOV(16, MComplex(EAX, EAX, 1, MOV(64, R(tmp2), ImmPtr(g_dsp.reg_stack[stack_reg]));
PtrOffset(&g_dsp.reg_stack[stack_reg][0],nullptr)), R(tmp1)); MOV(16, MComplex(tmp2, EAX, SCALE_2, 0), R(tmp1));
gpr.putXReg(tmp1); gpr.putXReg(tmp1);
gpr.putXReg(tmp2);
} }
//clobbers: //clobbers:
@ -37,13 +39,15 @@ void DSPEmitter::dsp_reg_stack_pop(int stack_reg)
{ {
//g_dsp.r[DSP_REG_ST0 + stack_reg] = g_dsp.reg_stack[stack_reg][g_dsp.reg_stack_ptr[stack_reg]]; //g_dsp.r[DSP_REG_ST0 + stack_reg] = g_dsp.reg_stack[stack_reg][g_dsp.reg_stack_ptr[stack_reg]];
MOV(8, R(AL), M(&g_dsp.reg_stack_ptr[stack_reg])); MOV(8, R(AL), M(&g_dsp.reg_stack_ptr[stack_reg]));
X64Reg tmp1; X64Reg tmp1, tmp2;
gpr.getFreeXReg(tmp1); gpr.getFreeXReg(tmp1);
gpr.getFreeXReg(tmp2);
MOVZX(64, 8, RAX, R(AL)); MOVZX(64, 8, RAX, R(AL));
MOV(16, R(tmp1), MComplex(EAX, EAX, 1, MOV(64, R(tmp2), ImmPtr(g_dsp.reg_stack[stack_reg]));
PtrOffset(&g_dsp.reg_stack[stack_reg][0],nullptr))); MOV(16, R(tmp1), MComplex(tmp2, EAX, SCALE_2, 0));
MOV(16, M(&g_dsp.r.st[stack_reg]), R(tmp1)); MOV(16, M(&g_dsp.r.st[stack_reg]), R(tmp1));
gpr.putXReg(tmp1); gpr.putXReg(tmp1);
gpr.putXReg(tmp2);
//g_dsp.reg_stack_ptr[stack_reg]--; //g_dsp.reg_stack_ptr[stack_reg]--;
//g_dsp.reg_stack_ptr[stack_reg] &= DSP_STACK_MASK; //g_dsp.reg_stack_ptr[stack_reg] &= DSP_STACK_MASK;

37
Source/Core/Core/PowerPC/Jit64/JitAsm.cpp
View File

@ -103,9 +103,9 @@ void Jit64AsmRoutineManager::Generate()
// optimizations safe, because IR and DR are usually set/cleared together. // optimizations safe, because IR and DR are usually set/cleared together.
// TODO: Branching based on the 20 most significant bits of instruction // TODO: Branching based on the 20 most significant bits of instruction
// addresses without translating them is wrong. // addresses without translating them is wrong.
u64 icache = (u64)jit->GetBlockCache()->iCache.data(); u8* icache = jit->GetBlockCache()->iCache.data();
u64 icacheVmem = (u64)jit->GetBlockCache()->iCacheVMEM.data(); u8* icacheVmem = jit->GetBlockCache()->iCacheVMEM.data();
u64 icacheEx = (u64)jit->GetBlockCache()->iCacheEx.data(); u8* icacheEx = jit->GetBlockCache()->iCacheEx.data();
u32 mask = 0; u32 mask = 0;
FixupBranch no_mem; FixupBranch no_mem;
FixupBranch exit_mem; FixupBranch exit_mem;
@ -117,13 +117,13 @@ void Jit64AsmRoutineManager::Generate()
no_mem = J_CC(CC_NZ); no_mem = J_CC(CC_NZ);
AND(32, R(RSCRATCH), Imm32(JIT_ICACHE_MASK)); AND(32, R(RSCRATCH), Imm32(JIT_ICACHE_MASK));
if (icache <= INT_MAX) if (FitsInS32(PPCSTATE_OFS(icache)))
{ {
MOV(32, R(RSCRATCH), MDisp(RSCRATCH, (s32)icache)); MOV(32, R(RSCRATCH), MPIC(icache, RSCRATCH));
} }
else else
{ {
MOV(64, R(RSCRATCH2), Imm64(icache)); MOV(64, R(RSCRATCH2), ImmPtr(icache));
MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH)); MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH));
} }
@ -132,13 +132,14 @@ void Jit64AsmRoutineManager::Generate()
TEST(32, R(RSCRATCH), Imm32(JIT_ICACHE_VMEM_BIT)); TEST(32, R(RSCRATCH), Imm32(JIT_ICACHE_VMEM_BIT));
FixupBranch no_vmem = J_CC(CC_Z); FixupBranch no_vmem = J_CC(CC_Z);
AND(32, R(RSCRATCH), Imm32(JIT_ICACHE_MASK)); AND(32, R(RSCRATCH), Imm32(JIT_ICACHE_MASK));
if (icacheVmem <= INT_MAX)
if (FitsInS32(PPCSTATE_OFS(icacheVmem)))
{ {
MOV(32, R(RSCRATCH), MDisp(RSCRATCH, (s32)icacheVmem)); MOV(32, R(RSCRATCH), MPIC(icacheVmem, RSCRATCH));
} }
else else
{ {
MOV(64, R(RSCRATCH2), Imm64(icacheVmem)); MOV(64, R(RSCRATCH2), ImmPtr(icacheVmem));
MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH)); MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH));
} }
@ -149,14 +150,13 @@ void Jit64AsmRoutineManager::Generate()
TEST(32, R(RSCRATCH), Imm32(JIT_ICACHE_EXRAM_BIT)); TEST(32, R(RSCRATCH), Imm32(JIT_ICACHE_EXRAM_BIT));
FixupBranch no_exram = J_CC(CC_Z); FixupBranch no_exram = J_CC(CC_Z);
AND(32, R(RSCRATCH), Imm32(JIT_ICACHEEX_MASK)); AND(32, R(RSCRATCH), Imm32(JIT_ICACHEEX_MASK));
if (FitsInS32(PPCSTATE_OFS(icacheEx)))
if (icacheEx <= INT_MAX)
{ {
MOV(32, R(RSCRATCH), MDisp(RSCRATCH, (s32)icacheEx)); MOV(32, R(RSCRATCH), MPIC(icacheEx, RSCRATCH));
} }
else else
{ {
MOV(64, R(RSCRATCH2), Imm64(icacheEx)); MOV(64, R(RSCRATCH2), ImmPtr(icacheEx));
MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH)); MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH));
} }
@ -169,16 +169,17 @@ void Jit64AsmRoutineManager::Generate()
TEST(32, R(RSCRATCH), R(RSCRATCH)); TEST(32, R(RSCRATCH), R(RSCRATCH));
FixupBranch notfound = J_CC(CC_L); FixupBranch notfound = J_CC(CC_L);
//grab from list and jump to it //grab from list and jump to it
u64 codePointers = (u64)jit->GetBlockCache()->GetCodePointers(); const u8** codePointers = jit->GetBlockCache()->GetCodePointers();
if (codePointers <= INT_MAX) if (FitsInS32(PPCSTATE_OFS(codePointers)))
{ {
JMPptr(MScaled(RSCRATCH, SCALE_8, (s32)codePointers)); JMPptr(MPIC(codePointers, RSCRATCH, SCALE_8));
} }
else else
{ {
MOV(64, R(RSCRATCH2), Imm64(codePointers)); MOV(64, R(RSCRATCH2), ImmPtr(codePointers));
JMPptr(MComplex(RSCRATCH2, RSCRATCH, SCALE_8, 0)); JMPptr(MComplex(RSCRATCH2, RSCRATCH, SCALE_8, 0));
} }
SetJumpTarget(notfound); SetJumpTarget(notfound);
//Ok, no block, let's jit //Ok, no block, let's jit
@ -271,7 +272,7 @@ void Jit64AsmRoutineManager::GenerateCommon()
CMP(32, R(ABI_PARAM2), Imm32(0xCC008000)); CMP(32, R(ABI_PARAM2), Imm32(0xCC008000));
FixupBranch skip_fast_write = J_CC(CC_NE, false); FixupBranch skip_fast_write = J_CC(CC_NE, false);
MOV(32, RSCRATCH, M(&m_gatherPipeCount)); MOV(32, RSCRATCH, M(&m_gatherPipeCount));
MOV(8, MDisp(RSCRATCH, (u32)&m_gatherPipe), ABI_PARAM1); MOV(8, MPIC(&m_gatherPipe, RSCRATCH), ABI_PARAM1);
ADD(32, 1, M(&m_gatherPipeCount)); ADD(32, 1, M(&m_gatherPipeCount));
RET(); RET();
SetJumpTarget(skip_fast_write); SetJumpTarget(skip_fast_write);

2
Source/Core/Core/PowerPC/Jit64/Jit_Integer.cpp
View File

@ -45,7 +45,7 @@ void Jit64::GenerateOverflow()
//rare). //rare).
static const u8 ovtable[4] = {0, 0, XER_SO_MASK, XER_SO_MASK}; static const u8 ovtable[4] = {0, 0, XER_SO_MASK, XER_SO_MASK};
MOVZX(32, 8, RSCRATCH, PPCSTATE(xer_so_ov)); MOVZX(32, 8, RSCRATCH, PPCSTATE(xer_so_ov));
MOV(8, R(RSCRATCH), MDisp(RSCRATCH, (u32)(u64)ovtable)); MOV(8, R(RSCRATCH), MPIC(ovtable, RSCRATCH));
MOV(8, PPCSTATE(xer_so_ov), R(RSCRATCH)); MOV(8, PPCSTATE(xer_so_ov), R(RSCRATCH));
SetJumpTarget(exit); SetJumpTarget(exit);
} }

6
Source/Core/Core/PowerPC/Jit64/Jit_LoadStorePaired.cpp
View File

@ -132,13 +132,13 @@ void Jit64::psq_stXX(UGeckoInstruction inst)
{ {
// One value // One value
CVTSD2SS(XMM0, fpr.R(s)); CVTSD2SS(XMM0, fpr.R(s));
CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)asm_routines.singleStoreQuantized)); CALLptr(MPIC(asm_routines.singleStoreQuantized, RSCRATCH, SCALE_8));
} }
else else
{ {
// Pair of values // Pair of values
CVTPD2PS(XMM0, fpr.R(s)); CVTPD2PS(XMM0, fpr.R(s));
CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)asm_routines.pairedStoreQuantized)); CALLptr(MPIC(asm_routines.pairedStoreQuantized, RSCRATCH, SCALE_8));
} }
if (update && jo.memcheck) if (update && jo.memcheck)
@ -306,7 +306,7 @@ void Jit64::psq_lXX(UGeckoInstruction inst)
AND(32, R(RSCRATCH2), gqr); AND(32, R(RSCRATCH2), gqr);
MOVZX(32, 8, RSCRATCH, R(RSCRATCH2)); MOVZX(32, 8, RSCRATCH, R(RSCRATCH2));
CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)(&asm_routines.pairedLoadQuantized[w * 8]))); CALLptr(MPIC(&asm_routines.pairedLoadQuantized[w * 8], RSCRATCH, SCALE_8));
MemoryExceptionCheck(); MemoryExceptionCheck();
CVTPS2PD(fpr.RX(s), R(XMM0)); CVTPS2PD(fpr.RX(s), R(XMM0));

4
Source/Core/Core/PowerPC/Jit64/Jit_SystemRegisters.cpp
View File

@ -458,7 +458,7 @@ void Jit64::mtcrf(UGeckoInstruction inst)
SHR(32, R(RSCRATCH), Imm8(28 - (i * 4))); SHR(32, R(RSCRATCH), Imm8(28 - (i * 4)));
if (i != 0) if (i != 0)
AND(32, R(RSCRATCH), Imm8(0xF)); AND(32, R(RSCRATCH), Imm8(0xF));
MOV(64, R(RSCRATCH), MScaled(RSCRATCH, SCALE_8, (u32)(u64)m_crTable)); MOV(64, R(RSCRATCH), MPIC(m_crTable, RSCRATCH, SCALE_8));
MOV(64, PPCSTATE(cr_val[i]), R(RSCRATCH)); MOV(64, PPCSTATE(cr_val[i]), R(RSCRATCH));
} }
} }
@ -493,7 +493,7 @@ void Jit64::mcrxr(UGeckoInstruction inst)
// [SO OV CA 0] << 3 // [SO OV CA 0] << 3
SHL(32, R(RSCRATCH), Imm8(4)); SHL(32, R(RSCRATCH), Imm8(4));
MOV(64, R(RSCRATCH), MDisp(RSCRATCH, (u32)(u64)m_crTable)); MOV(64, R(RSCRATCH), MPIC(m_crTable, RSCRATCH));
MOV(64, PPCSTATE(cr_val[inst.CRFD]), R(RSCRATCH)); MOV(64, PPCSTATE(cr_val[inst.CRFD]), R(RSCRATCH));
// Clear XER[0-3] // Clear XER[0-3]

46
Source/Core/Core/PowerPC/Jit64Common/Jit64AsmCommon.cpp
View File

@ -24,10 +24,8 @@ void CommonAsmRoutines::GenFifoWrite(int size)
const void* start = GetCodePtr(); const void* start = GetCodePtr();
// Assume value in RSCRATCH // Assume value in RSCRATCH
u32 gather_pipe = (u32)(u64)GPFifo::m_gatherPipe;
_assert_msg_(DYNA_REC, gather_pipe <= 0x7FFFFFFF, "Gather pipe not in low 2GB of memory!");
MOV(32, R(RSCRATCH2), M(&GPFifo::m_gatherPipeCount)); MOV(32, R(RSCRATCH2), M(&GPFifo::m_gatherPipeCount));
SwapAndStore(size, MDisp(RSCRATCH2, gather_pipe), RSCRATCH); SwapAndStore(size, MPIC(GPFifo::m_gatherPipe, RSCRATCH2), RSCRATCH);
ADD(32, R(RSCRATCH2), Imm8(size >> 3)); ADD(32, R(RSCRATCH2), Imm8(size >> 3));
MOV(32, M(&GPFifo::m_gatherPipeCount), R(RSCRATCH2)); MOV(32, M(&GPFifo::m_gatherPipeCount), R(RSCRATCH2));
RET(); RET();
@ -68,8 +66,8 @@ void CommonAsmRoutines::GenFrsqrte()
SHR(64, R(RSCRATCH), Imm8(37)); SHR(64, R(RSCRATCH), Imm8(37));
AND(32, R(RSCRATCH), Imm32(0x7FF)); AND(32, R(RSCRATCH), Imm32(0x7FF));
IMUL(32, RSCRATCH, MScaled(RSCRATCH_EXTRA, SCALE_4, (u32)(u64)MathUtil::frsqrte_expected_dec)); IMUL(32, RSCRATCH, MPIC(MathUtil::frsqrte_expected_dec, RSCRATCH_EXTRA, SCALE_4));
MOV(32, R(RSCRATCH_EXTRA), MScaled(RSCRATCH_EXTRA, SCALE_4, (u32)(u64)MathUtil::frsqrte_expected_base)); MOV(32, R(RSCRATCH_EXTRA), MPIC(MathUtil::frsqrte_expected_base, RSCRATCH_EXTRA, SCALE_4));
SUB(32, R(RSCRATCH_EXTRA), R(RSCRATCH)); SUB(32, R(RSCRATCH_EXTRA), R(RSCRATCH));
SHL(64, R(RSCRATCH_EXTRA), Imm8(26)); SHL(64, R(RSCRATCH_EXTRA), Imm8(26));
OR(64, R(RSCRATCH2), R(RSCRATCH_EXTRA)); // vali |= (s64)(frsqrte_expected_base[index] - frsqrte_expected_dec[index] * (i % 2048)) << 26; OR(64, R(RSCRATCH2), R(RSCRATCH_EXTRA)); // vali |= (s64)(frsqrte_expected_base[index] - frsqrte_expected_dec[index] * (i % 2048)) << 26;
@ -136,11 +134,11 @@ void CommonAsmRoutines::GenFres()
AND(32, R(RSCRATCH), Imm32(0x3FF)); // i % 1024 AND(32, R(RSCRATCH), Imm32(0x3FF)); // i % 1024
AND(32, R(RSCRATCH2), Imm8(0x1F)); // i / 1024 AND(32, R(RSCRATCH2), Imm8(0x1F)); // i / 1024
IMUL(32, RSCRATCH, MScaled(RSCRATCH2, SCALE_4, (u32)(u64)MathUtil::fres_expected_dec)); IMUL(32, RSCRATCH, MPIC(MathUtil::fres_expected_dec, RSCRATCH2, SCALE_4));
ADD(32, R(RSCRATCH), Imm8(1)); ADD(32, R(RSCRATCH), Imm8(1));
SHR(32, R(RSCRATCH), Imm8(1)); SHR(32, R(RSCRATCH), Imm8(1));
MOV(32, R(RSCRATCH2), MScaled(RSCRATCH2, SCALE_4, (u32)(u64)MathUtil::fres_expected_base)); MOV(32, R(RSCRATCH2), MPIC(MathUtil::fres_expected_base, RSCRATCH2, SCALE_4));
SUB(32, R(RSCRATCH2), R(RSCRATCH)); SUB(32, R(RSCRATCH2), R(RSCRATCH));
SHL(64, R(RSCRATCH2), Imm8(29)); SHL(64, R(RSCRATCH2), Imm8(29));
OR(64, R(RSCRATCH2), R(RSCRATCH_EXTRA)); // vali |= (s64)(fres_expected_base[i / 1024] - (fres_expected_dec[i / 1024] * (i % 1024) + 1) / 2) << 29 OR(64, R(RSCRATCH2), R(RSCRATCH_EXTRA)); // vali |= (s64)(fres_expected_base[i / 1024] - (fres_expected_dec[i / 1024] * (i % 1024) + 1) / 2) << 29
@ -199,7 +197,7 @@ void CommonAsmRoutines::GenMfcr()
// SO: Bit 61 set; set flag bit 0 // SO: Bit 61 set; set flag bit 0
// LT: Bit 62 set; set flag bit 3 // LT: Bit 62 set; set flag bit 3
SHR(64, R(cr_val), Imm8(61)); SHR(64, R(cr_val), Imm8(61));
OR(32, R(dst), MScaled(cr_val, SCALE_4, (u32)(u64)m_flagTable)); OR(32, R(dst), MPIC(m_flagTable, cr_val, SCALE_4));
} }
RET(); RET();
@ -247,7 +245,7 @@ void CommonAsmRoutines::GenQuantizedStores()
const u8* storePairedU8 = AlignCode4(); const u8* storePairedU8 = AlignCode4();
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS)); MOVQ_xmm(XMM1, MPIC(m_quantizeTableS, RSCRATCH2));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
#ifdef QUANTIZE_OVERFLOW_SAFE #ifdef QUANTIZE_OVERFLOW_SAFE
MINPS(XMM0, M(m_65535)); MINPS(XMM0, M(m_65535));
@ -262,7 +260,7 @@ void CommonAsmRoutines::GenQuantizedStores()
const u8* storePairedS8 = AlignCode4(); const u8* storePairedS8 = AlignCode4();
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS)); MOVQ_xmm(XMM1, MPIC(m_quantizeTableS, RSCRATCH2));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
#ifdef QUANTIZE_OVERFLOW_SAFE #ifdef QUANTIZE_OVERFLOW_SAFE
MINPS(XMM0, M(m_65535)); MINPS(XMM0, M(m_65535));
@ -278,7 +276,7 @@ void CommonAsmRoutines::GenQuantizedStores()
const u8* storePairedU16 = AlignCode4(); const u8* storePairedU16 = AlignCode4();
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS)); MOVQ_xmm(XMM1, MPIC(m_quantizeTableS, RSCRATCH2));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
if (cpu_info.bSSE4_1) if (cpu_info.bSSE4_1)
@ -310,7 +308,7 @@ void CommonAsmRoutines::GenQuantizedStores()
const u8* storePairedS16 = AlignCode4(); const u8* storePairedS16 = AlignCode4();
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS)); MOVQ_xmm(XMM1, MPIC(m_quantizeTableS, RSCRATCH2));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
#ifdef QUANTIZE_OVERFLOW_SAFE #ifdef QUANTIZE_OVERFLOW_SAFE
MINPS(XMM0, M(m_65535)); MINPS(XMM0, M(m_65535));
@ -355,7 +353,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores()
const u8* storeSingleU8 = AlignCode4(); // Used by MKWii const u8* storeSingleU8 = AlignCode4(); // Used by MKWii
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS)); MULSS(XMM0, MPIC(m_quantizeTableS, RSCRATCH2));
XORPS(XMM1, R(XMM1)); XORPS(XMM1, R(XMM1));
MAXSS(XMM0, R(XMM1)); MAXSS(XMM0, R(XMM1));
MINSS(XMM0, M(&m_255)); MINSS(XMM0, M(&m_255));
@ -365,7 +363,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores()
const u8* storeSingleS8 = AlignCode4(); const u8* storeSingleS8 = AlignCode4();
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS)); MULSS(XMM0, MPIC(m_quantizeTableS, RSCRATCH2));
MAXSS(XMM0, M(&m_m128)); MAXSS(XMM0, M(&m_m128));
MINSS(XMM0, M(&m_127)); MINSS(XMM0, M(&m_127));
CVTTSS2SI(RSCRATCH, R(XMM0)); CVTTSS2SI(RSCRATCH, R(XMM0));
@ -374,7 +372,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores()
const u8* storeSingleU16 = AlignCode4(); // Used by MKWii const u8* storeSingleU16 = AlignCode4(); // Used by MKWii
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS)); MULSS(XMM0, MPIC(m_quantizeTableS, RSCRATCH2));
XORPS(XMM1, R(XMM1)); XORPS(XMM1, R(XMM1));
MAXSS(XMM0, R(XMM1)); MAXSS(XMM0, R(XMM1));
MINSS(XMM0, M(m_65535)); MINSS(XMM0, M(m_65535));
@ -384,7 +382,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores()
const u8* storeSingleS16 = AlignCode4(); const u8* storeSingleS16 = AlignCode4();
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS)); MULSS(XMM0, MPIC(m_quantizeTableS, RSCRATCH2));
MAXSS(XMM0, M(&m_m32768)); MAXSS(XMM0, M(&m_m32768));
MINSS(XMM0, M(&m_32767)); MINSS(XMM0, M(&m_32767));
CVTTSS2SI(RSCRATCH, R(XMM0)); CVTTSS2SI(RSCRATCH, R(XMM0));
@ -484,7 +482,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
} }
CVTDQ2PS(XMM0, R(XMM0)); CVTDQ2PS(XMM0, R(XMM0));
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS)); MOVQ_xmm(XMM1, MPIC(m_dequantizeTableS, RSCRATCH2));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
RET(); RET();
@ -495,7 +493,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
UnsafeLoadRegToRegNoSwap(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 8, 0); // RSCRATCH_EXTRA = 0x000000xx UnsafeLoadRegToRegNoSwap(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 8, 0); // RSCRATCH_EXTRA = 0x000000xx
CVTSI2SS(XMM0, R(RSCRATCH_EXTRA)); CVTSI2SS(XMM0, R(RSCRATCH_EXTRA));
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS)); MULSS(XMM0, MPIC(m_dequantizeTableS, RSCRATCH2));
UNPCKLPS(XMM0, M(m_one)); UNPCKLPS(XMM0, M(m_one));
RET(); RET();
@ -523,7 +521,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
} }
CVTDQ2PS(XMM0, R(XMM0)); CVTDQ2PS(XMM0, R(XMM0));
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS)); MOVQ_xmm(XMM1, MPIC(m_dequantizeTableS, RSCRATCH2));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
RET(); RET();
@ -534,7 +532,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
UnsafeLoadRegToRegNoSwap(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 8, 0, true); UnsafeLoadRegToRegNoSwap(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 8, 0, true);
CVTSI2SS(XMM0, R(RSCRATCH_EXTRA)); CVTSI2SS(XMM0, R(RSCRATCH_EXTRA));
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS)); MULSS(XMM0, MPIC(m_dequantizeTableS, RSCRATCH2));
UNPCKLPS(XMM0, M(m_one)); UNPCKLPS(XMM0, M(m_one));
RET(); RET();
@ -557,7 +555,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
} }
CVTDQ2PS(XMM0, R(XMM0)); CVTDQ2PS(XMM0, R(XMM0));
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS)); MOVQ_xmm(XMM1, MPIC(m_dequantizeTableS, RSCRATCH2));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
RET(); RET();
@ -568,7 +566,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
UnsafeLoadRegToReg(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 16, 0, false); UnsafeLoadRegToReg(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 16, 0, false);
CVTSI2SS(XMM0, R(RSCRATCH_EXTRA)); CVTSI2SS(XMM0, R(RSCRATCH_EXTRA));
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS)); MULSS(XMM0, MPIC(m_dequantizeTableS, RSCRATCH2));
UNPCKLPS(XMM0, M(m_one)); UNPCKLPS(XMM0, M(m_one));
RET(); RET();
@ -590,7 +588,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
} }
CVTDQ2PS(XMM0, R(XMM0)); CVTDQ2PS(XMM0, R(XMM0));
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS)); MOVQ_xmm(XMM1, MPIC(m_dequantizeTableS, RSCRATCH2));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
RET(); RET();
@ -601,7 +599,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
UnsafeLoadRegToReg(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 16, 0, true); UnsafeLoadRegToReg(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 16, 0, true);
CVTSI2SS(XMM0, R(RSCRATCH_EXTRA)); CVTSI2SS(XMM0, R(RSCRATCH_EXTRA));
SHR(32, R(RSCRATCH2), Imm8(5)); SHR(32, R(RSCRATCH2), Imm8(5));
MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS)); MULSS(XMM0, MPIC(m_dequantizeTableS, RSCRATCH2));
UNPCKLPS(XMM0, M(m_one)); UNPCKLPS(XMM0, M(m_one));
RET(); RET();

4
Source/Core/Core/PowerPC/Jit64IL/IR_X86.cpp
View File

@ -1611,7 +1611,7 @@ static void DoWriteCode(IRBuilder* ibuild, JitIL* Jit, u32 exitAddress)
Jit->OR(32, R(RSCRATCH), Imm8(w << 3)); Jit->OR(32, R(RSCRATCH), Imm8(w << 3));
Jit->MOV(32, R(RSCRATCH_EXTRA), regLocForInst(RI, getOp1(I))); Jit->MOV(32, R(RSCRATCH_EXTRA), regLocForInst(RI, getOp1(I)));
Jit->CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)(Jit->asm_routines.pairedLoadQuantized))); Jit->CALLptr(MPIC(Jit->asm_routines.pairedLoadQuantized, RSCRATCH, SCALE_8));
Jit->MOVAPD(reg, R(XMM0)); Jit->MOVAPD(reg, R(XMM0));
RI.fregs[reg] = I; RI.fregs[reg] = I;
regNormalRegClear(RI, I); regNormalRegClear(RI, I);
@ -1669,7 +1669,7 @@ static void DoWriteCode(IRBuilder* ibuild, JitIL* Jit, u32 exitAddress)
Jit->MOV(32, R(RSCRATCH_EXTRA), regLocForInst(RI, getOp2(I))); Jit->MOV(32, R(RSCRATCH_EXTRA), regLocForInst(RI, getOp2(I)));
Jit->MOVAPD(XMM0, fregLocForInst(RI, getOp1(I))); Jit->MOVAPD(XMM0, fregLocForInst(RI, getOp1(I)));
Jit->CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)(Jit->asm_routines.pairedStoreQuantized))); Jit->CALLptr(MPIC(Jit->asm_routines.pairedStoreQuantized, RSCRATCH, SCALE_8));
if (RI.IInfo[I - RI.FirstI] & 4) if (RI.IInfo[I - RI.FirstI] & 4)
fregClearInst(RI, getOp1(I)); fregClearInst(RI, getOp1(I));
if (RI.IInfo[I - RI.FirstI] & 8) if (RI.IInfo[I - RI.FirstI] & 8)

12
Source/Core/Core/PowerPC/JitCommon/JitBase.h
View File

@ -43,6 +43,18 @@
// to address as much as possible in a one-byte offset form. // to address as much as possible in a one-byte offset form.
#define RPPCSTATE RBP #define RPPCSTATE RBP
namespace Gen
{
inline OpArg MPIC(const void* address, X64Reg scale_reg, int scale = SCALE_1)
{
ptrdiff_t offset = PPCSTATE_OFS(address);
_dbg_assert_(DYNA_REC, FitsInS32(offset));
return MComplex(RPPCSTATE, scale_reg, scale, offset);
}
}
// Use these to control the instruction selection // Use these to control the instruction selection
// #define INSTRUCTION_START FallBackToInterpreter(inst); return; // #define INSTRUCTION_START FallBackToInterpreter(inst); return;
// #define INSTRUCTION_START PPCTables::CountInstruction(inst); // #define INSTRUCTION_START PPCTables::CountInstruction(inst);

4
Source/Core/Core/PowerPC/JitCommon/Jit_Util.cpp
View File

@ -173,11 +173,11 @@ private:
u32 all_ones = (1ULL << sbits) - 1; u32 all_ones = (1ULL << sbits) - 1;
if ((all_ones & mask) == all_ones) if ((all_ones & mask) == all_ones)
{ {
MoveOpArgToReg(sbits, MDisp(RSCRATCH, 0)); MoveOpArgToReg(sbits, MatR(RSCRATCH));
} }
else else
{ {
m_code->MOVZX(32, sbits, m_dst_reg, MDisp(RSCRATCH, 0)); m_code->MOVZX(32, sbits, m_dst_reg, MatR(RSCRATCH));
m_code->AND(32, R(m_dst_reg), Imm32(mask)); m_code->AND(32, R(m_dst_reg), Imm32(mask));
if (m_sign_extend) if (m_sign_extend)
m_code->MOVSX(32, sbits, m_dst_reg, R(m_dst_reg)); m_code->MOVSX(32, sbits, m_dst_reg, R(m_dst_reg));

4
Source/Core/Core/PowerPC/JitCommon/Jit_Util.h
View File

@ -14,8 +14,8 @@ namespace MMIO { class Mapping; }
// We offset by 0x80 because the range of one byte memory offsets is // We offset by 0x80 because the range of one byte memory offsets is
// -0x80..0x7f. // -0x80..0x7f.
#define PPCSTATE(x) MDisp(RPPCSTATE, \ #define PPCSTATE_OFS(x) ((u8*)(x) - (u8*)&PowerPC::ppcState - 0x80)
(int) ((char *) &PowerPC::ppcState.x - (char *) &PowerPC::ppcState) - 0x80) #define PPCSTATE(x) MDisp(RPPCSTATE, PPCSTATE_OFS(&PowerPC::ppcState.x))
// In case you want to disable the ppcstate register: // In case you want to disable the ppcstate register:
// #define PPCSTATE(x) M(&PowerPC::ppcState.x) // #define PPCSTATE(x) M(&PowerPC::ppcState.x)
#define PPCSTATE_LR PPCSTATE(spr[SPR_LR]) #define PPCSTATE_LR PPCSTATE(spr[SPR_LR])