The-Homebrew-Cloud/dolphin
2
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-01-10 08:09:26 +01:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #4533 from lioncash/jits

Browse Source 
Jit64/JitIL: Get rid of trivial global jit variable usages
This commit is contained in:
Mat M 2016-12-20 21:09:50 -05:00 committed by GitHub
commit 8192af9b3d
6 changed files with 24 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
Source/Core/Core/PowerPC/Jit64/Jit_FloatingPoint.cpp
View File

@ -189,15 +189,15 @@ void Jit64::fp_arith(UGeckoInstruction inst)
// If both the inputs are known to have identical top and bottom halves, we can skip the MOVDDUP // If both the inputs are known to have identical top and bottom halves, we can skip the MOVDDUP
// at the end by // at the end by
// using packed arithmetic instead. // using packed arithmetic instead.
bool packed = inst.OPCD == 4 || (inst.OPCD == 59 && jit->js.op->fprIsDuplicated[a] && bool packed = inst.OPCD == 4 ||
jit->js.op->fprIsDuplicated[arg2]); (inst.OPCD == 59 && js.op->fprIsDuplicated[a] && js.op->fprIsDuplicated[arg2]);
// Packed divides are slower than scalar divides on basically all x86, so this optimization isn't // Packed divides are slower than scalar divides on basically all x86, so this optimization isn't
// worth it in that case. // worth it in that case.
// Atoms (and a few really old CPUs) are also slower on packed operations than scalar ones. // Atoms (and a few really old CPUs) are also slower on packed operations than scalar ones.
if (inst.OPCD == 59 && (inst.SUBOP5 == 18 || cpu_info.bAtom)) if (inst.OPCD == 59 && (inst.SUBOP5 == 18 || cpu_info.bAtom))
packed = false; packed = false;
bool round_input = single && !jit->js.op->fprIsSingle[inst.FC]; bool round_input = single && !js.op->fprIsSingle[inst.FC];
bool preserve_inputs = SConfig::GetInstance().bAccurateNaNs; bool preserve_inputs = SConfig::GetInstance().bAccurateNaNs;
X64Reg dest = INVALID_REG; X64Reg dest = INVALID_REG;
@ -241,10 +241,9 @@ void Jit64::fmaddXX(UGeckoInstruction inst)
int c = inst.FC; int c = inst.FC;
int d = inst.FD; int d = inst.FD;
bool single = inst.OPCD == 4 || inst.OPCD == 59; bool single = inst.OPCD == 4 || inst.OPCD == 59;
bool round_input = single && !jit->js.op->fprIsSingle[c]; bool round_input = single && !js.op->fprIsSingle[c];
bool packed = bool packed = inst.OPCD == 4 || (!cpu_info.bAtom && single && js.op->fprIsDuplicated[a] &&
inst.OPCD == 4 || (!cpu_info.bAtom && single && jit->js.op->fprIsDuplicated[a] && js.op->fprIsDuplicated[b] && js.op->fprIsDuplicated[c]);
jit->js.op->fprIsDuplicated[b] && jit->js.op->fprIsDuplicated[c]);
fpr.Lock(a, b, c, d); fpr.Lock(a, b, c, d);
@ -635,7 +634,7 @@ void Jit64::frspx(UGeckoInstruction inst)
FALLBACK_IF(inst.Rc); FALLBACK_IF(inst.Rc);
int b = inst.FB; int b = inst.FB;
int d = inst.FD; int d = inst.FD;
bool packed = jit->js.op->fprIsDuplicated[b] && !cpu_info.bAtom; bool packed = js.op->fprIsDuplicated[b] && !cpu_info.bAtom;
fpr.Lock(b, d); fpr.Lock(b, d);
OpArg src = fpr.R(b); OpArg src = fpr.R(b);

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

@ -279,7 +279,7 @@ void Jit64::dcbx(UGeckoInstruction inst)
// Check whether a JIT cache line needs to be invalidated. // Check whether a JIT cache line needs to be invalidated.
LEA(32, value, MScaled(addr, SCALE_8, 0)); // addr << 3 (masks the first 3 bits) LEA(32, value, MScaled(addr, SCALE_8, 0)); // addr << 3 (masks the first 3 bits)
SHR(32, R(value), Imm8(3 + 5 + 5)); // >> 5 for cache line size, >> 5 for width of bitset SHR(32, R(value), Imm8(3 + 5 + 5)); // >> 5 for cache line size, >> 5 for width of bitset
MOV(64, R(tmp), ImmPtr(jit->GetBlockCache()->GetBlockBitSet())); MOV(64, R(tmp), ImmPtr(GetBlockCache()->GetBlockBitSet()));
MOV(32, R(value), MComplex(tmp, value, SCALE_4, 0)); MOV(32, R(value), MComplex(tmp, value, SCALE_4, 0));
SHR(32, R(addr), Imm8(5)); SHR(32, R(addr), Imm8(5));
BT(32, R(value), R(addr)); BT(32, R(value), R(addr));
@ -355,7 +355,7 @@ void Jit64::dcbz(UGeckoInstruction inst)
SwitchToFarCode(); SwitchToFarCode();
SetJumpTarget(slow); SetJumpTarget(slow);
} }
MOV(32, M(&PC), Imm32(jit->js.compilerPC)); MOV(32, M(&PC), Imm32(js.compilerPC));
BitSet32 registersInUse = CallerSavedRegistersInUse(); BitSet32 registersInUse = CallerSavedRegistersInUse();
ABI_PushRegistersAndAdjustStack(registersInUse, 0); ABI_PushRegistersAndAdjustStack(registersInUse, 0);
ABI_CallFunctionR(PowerPC::ClearCacheLine, RSCRATCH); ABI_CallFunctionR(PowerPC::ClearCacheLine, RSCRATCH);

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

@ -105,7 +105,7 @@ void Jit64::stfXXX(UGeckoInstruction inst)
if (single) if (single)
{ {
if (jit->js.op->fprIsStoreSafe[s]) if (js.op->fprIsStoreSafe[s])
{ {
CVTSD2SS(XMM0, fpr.R(s)); CVTSD2SS(XMM0, fpr.R(s));
} }

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

@ -84,7 +84,7 @@ void Jit64::ps_muls(UGeckoInstruction inst)
int d = inst.FD; int d = inst.FD;
int a = inst.FA; int a = inst.FA;
int c = inst.FC; int c = inst.FC;
bool round_input = !jit->js.op->fprIsSingle[c]; bool round_input = !js.op->fprIsSingle[c];
fpr.Lock(a, c, d); fpr.Lock(a, c, d);
switch (inst.SUBOP5) switch (inst.SUBOP5)
{ {

10
Source/Core/Core/PowerPC/Jit64Common/Jit64Base.cpp
View File

@ -54,7 +54,7 @@ bool Jitx86Base::BackPatch(u32 emAddress, SContext* ctx)
TrampolineInfo& info = it->second; TrampolineInfo& info = it->second;
u8* exceptionHandler = nullptr; u8* exceptionHandler = nullptr;
if (jit->jo.memcheck) if (jo.memcheck)
{ {
auto it2 = m_exception_handler_at_loc.find(codePtr); auto it2 = m_exception_handler_at_loc.find(codePtr);
if (it2 != m_exception_handler_at_loc.end()) if (it2 != m_exception_handler_at_loc.end())
@ -68,13 +68,13 @@ bool Jitx86Base::BackPatch(u32 emAddress, SContext* ctx)
// into the original code if necessary to ensure there is enough space // into the original code if necessary to ensure there is enough space
// to insert the backpatch jump.) // to insert the backpatch jump.)
jit->js.generatingTrampoline = true; js.generatingTrampoline = true;
jit->js.trampolineExceptionHandler = exceptionHandler; js.trampolineExceptionHandler = exceptionHandler;
// Generate the trampoline. // Generate the trampoline.
const u8* trampoline = trampolines.GenerateTrampoline(info); const u8* trampoline = trampolines.GenerateTrampoline(info);
jit->js.generatingTrampoline = false; js.generatingTrampoline = false;
jit->js.trampolineExceptionHandler = nullptr; js.trampolineExceptionHandler = nullptr;
u8* start = info.start; u8* start = info.start;

16
Source/Core/Core/PowerPC/Jit64IL/JitIL.cpp
View File

@ -354,8 +354,8 @@ void JitIL::Cleanup()
{ {
// SPEED HACK: MMCR0/MMCR1 should be checked at run-time, not at compile time. // SPEED HACK: MMCR0/MMCR1 should be checked at run-time, not at compile time.
if (MMCR0.Hex || MMCR1.Hex) if (MMCR0.Hex || MMCR1.Hex)
ABI_CallFunctionCCC(PowerPC::UpdatePerformanceMonitor, js.downcountAmount, ABI_CallFunctionCCC(PowerPC::UpdatePerformanceMonitor, js.downcountAmount, js.numLoadStoreInst,
jit->js.numLoadStoreInst, jit->js.numFloatingPointInst); js.numFloatingPointInst);
} }
void JitIL::WriteExit(u32 destination) void JitIL::WriteExit(u32 destination)
@ -518,8 +518,8 @@ const u8* JitIL::DoJit(u32 em_address, PPCAnalyst::CodeBuffer* code_buf, JitBloc
js.blockStart = em_address; js.blockStart = em_address;
js.fifoBytesSinceCheck = 0; js.fifoBytesSinceCheck = 0;
js.curBlock = b; js.curBlock = b;
jit->js.numLoadStoreInst = 0; js.numLoadStoreInst = 0;
jit->js.numFloatingPointInst = 0; js.numFloatingPointInst = 0;
PPCAnalyst::CodeOp* ops = code_buf->codebuffer; PPCAnalyst::CodeOp* ops = code_buf->codebuffer;
@ -610,7 +610,7 @@ const u8* JitIL::DoJit(u32 em_address, PPCAnalyst::CodeBuffer* code_buf, JitBloc
if (type == HLE::HLE_HOOK_REPLACE) if (type == HLE::HLE_HOOK_REPLACE)
{ {
MOV(32, R(EAX), PPCSTATE(npc)); MOV(32, R(EAX), PPCSTATE(npc));
jit->js.downcountAmount += jit->js.st.numCycles; js.downcountAmount += js.st.numCycles;
WriteExitDestInOpArg(R(EAX)); WriteExitDestInOpArg(R(EAX));
break; break;
} }
@ -625,7 +625,7 @@ const u8* JitIL::DoJit(u32 em_address, PPCAnalyst::CodeBuffer* code_buf, JitBloc
ibuild.EmitFPExceptionCheck(ibuild.EmitIntConst(ops[i].address)); ibuild.EmitFPExceptionCheck(ibuild.EmitIntConst(ops[i].address));
} }
if (jit->js.fifoWriteAddresses.find(js.compilerPC) != jit->js.fifoWriteAddresses.end()) if (js.fifoWriteAddresses.find(js.compilerPC) != js.fifoWriteAddresses.end())
{ {
ibuild.EmitExtExceptionCheck(ibuild.EmitIntConst(ops[i].address)); ibuild.EmitExtExceptionCheck(ibuild.EmitIntConst(ops[i].address));
} }
@ -648,10 +648,10 @@ const u8* JitIL::DoJit(u32 em_address, PPCAnalyst::CodeBuffer* code_buf, JitBloc
} }
if (opinfo->flags & FL_LOADSTORE) if (opinfo->flags & FL_LOADSTORE)
++jit->js.numLoadStoreInst; ++js.numLoadStoreInst;
if (opinfo->flags & FL_USE_FPU) if (opinfo->flags & FL_USE_FPU)
++jit->js.numFloatingPointInst; ++js.numFloatingPointInst;
} }
} }