Yaya48/dolphin
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Improve code and clarify parameters to ABI_Push/PopRegistersAndAdjustStack.

Browse Source 
- Factor common work into a helper function.
- Replace confusingly named "noProlog" with "rsp_alignment".  Now that
x86 is not supported, we can just specify it explicitly as 8 for
clarity.
- Add the option to include more frame size, which I'll need later.
- Revert a change by magumagu in March which replaced MOVAPD with MOVUPD
on account of 32-bit Windows, since it's no longer supported.  True,
apparently recent processors don't execute the former any faster if the
pointer is, in fact, aligned, but there's no point using MOVUPD for
something that's guaranteed to be aligned...

(I discovered that GenFrsqrte and GenFres were incorrectly passing false
to noProlog - they were, in fact, functions without prologs, the
original meaning of the parameter - which caused the previous change to
break.  This is now fixed.)
This commit is contained in:
comex committed 2014-09-08 00:58:56 -04:00
1 parent 6ea82790ba
commit 2dafbfb3ef
9 files changed
+96 -72
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

Source/Core/Core/PowerPC/Jit64/Jit.cpp
+2 -2
View File
@@ -495,9 +495,9 @@ const u8* Jit64::DoJit(u32 em_address, PPCAnalyst::CodeBuffer *code_buf, JitBloc
js.fifoBytesThisBlock -= 32;
MOV(32, PPCSTATE(pc), Imm32(jit->js.compilerPC)); // Helps external systems know which instruction triggered the write
u32 registersInUse = CallerSavedRegistersInUse();
ABI_PushRegistersAndAdjustStack(registersInUse, false);
ABI_PushRegistersAndAdjustStack(registersInUse, 0);
ABI_CallFunction((void *)&GPFifo::CheckGatherPipe);
ABI_PopRegistersAndAdjustStack(registersInUse, false);
ABI_PopRegistersAndAdjustStack(registersInUse, 0);
}
u32 function = HLE::GetFunctionIndex(ops[i].address);
Source/Core/Core/PowerPC/Jit64/Jit_LoadStore.cpp
+6 -6
View File
@@ -116,11 +116,11 @@ void Jit64::lXXx(UGeckoInstruction inst)
FixupBranch noIdle = J_CC(CC_NZ);
u32 registersInUse = CallerSavedRegistersInUse();
ABI_PushRegistersAndAdjustStack(registersInUse, false);
ABI_PushRegistersAndAdjustStack(registersInUse, 0);
ABI_CallFunctionC((void *)&PowerPC::OnIdle, PowerPC::ppcState.gpr[a] + (s32)(s16)inst.SIMM_16);
ABI_PopRegistersAndAdjustStack(registersInUse, false);
ABI_PopRegistersAndAdjustStack(registersInUse, 0);
// ! we must continue executing of the loop after exception handling, maybe there is still 0 in r0
//MOV(32, PPCSTATE(pc), Imm32(js.compilerPC));
@@ -285,9 +285,9 @@ void Jit64::dcbz(UGeckoInstruction inst)
// supposedly there are, at least for some MMU titles. Let's be careful and support it to be sure.
MOV(32, M(&PC), Imm32(jit->js.compilerPC));
u32 registersInUse = CallerSavedRegistersInUse();
ABI_PushRegistersAndAdjustStack(registersInUse, false);
ABI_PushRegistersAndAdjustStack(registersInUse, 0);
ABI_CallFunctionR((void *)&Memory::ClearCacheLine, RSCRATCH);
ABI_PopRegistersAndAdjustStack(registersInUse, false);
ABI_PopRegistersAndAdjustStack(registersInUse, 0);
FixupBranch exit = J();
SetJumpTarget(fast);
@@ -374,7 +374,7 @@ void Jit64::stX(UGeckoInstruction inst)
MOV(32, PPCSTATE(pc), Imm32(jit->js.compilerPC));
u32 registersInUse = CallerSavedRegistersInUse();
ABI_PushRegistersAndAdjustStack(registersInUse, false);
ABI_PushRegistersAndAdjustStack(registersInUse, 0);
switch (accessSize)
{
case 32:
@@ -387,7 +387,7 @@ void Jit64::stX(UGeckoInstruction inst)
ABI_CallFunctionAC((void *)&Memory::Write_U8, gpr.R(s), addr);
break;
}
ABI_PopRegistersAndAdjustStack(registersInUse, false);
ABI_PopRegistersAndAdjustStack(registersInUse, 0);
if (update)
gpr.SetImmediate32(a, addr);
return;
Source/Core/Core/PowerPC/JitCommon/JitAsmCommon.cpp
+7 -6
View File
@@ -110,9 +110,9 @@ void CommonAsmRoutines::GenFrsqrte()
SetJumpTarget(complex1);
SetJumpTarget(complex2);
SetJumpTarget(complex3);
ABI_PushRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, false);
ABI_PushRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, 8);
ABI_CallFunction((void *)&MathUtil::ApproximateReciprocalSquareRoot);
ABI_PopRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, false);
ABI_PopRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, 8);
RET();
}
@@ -169,9 +169,9 @@ void CommonAsmRoutines::GenFres()
SetJumpTarget(complex1);
SetJumpTarget(complex2);
ABI_PushRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, false);
ABI_PushRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, 8);
ABI_CallFunction((void *)&MathUtil::ApproximateReciprocal);
ABI_PopRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, false);
ABI_PopRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, 8);
RET();
}
@@ -258,9 +258,10 @@ void CommonAsmRoutines::GenQuantizedStores()
SwapAndStore(64, MComplex(RMEM, RSCRATCH_EXTRA, SCALE_1, 0), RSCRATCH);
FixupBranch skip_complex = J(true);
SetJumpTarget(too_complex);
ABI_PushRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, true);
// RSP alignment here is 8 due to the call.
ABI_PushRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, 8);
ABI_CallFunctionR((void *)&WriteDual32, RSCRATCH_EXTRA);
ABI_PopRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, true);
ABI_PopRegistersAndAdjustStack(QUANTIZED_REGS_TO_SAVE, 8);
SetJumpTarget(skip_complex);
RET();
Source/Core/Core/PowerPC/JitCommon/JitBackpatch.cpp
+5 -7
View File
@@ -56,10 +56,8 @@ const u8 *TrampolineCache::GetReadTrampoline(const InstructionInfo &info, u32 re
X64Reg dataReg = (X64Reg)info.regOperandReg;
// It's a read. Easy.
// It ought to be necessary to align the stack here. Since it seems to not
// affect anybody, I'm not going to add it just to be completely safe about
// performance.
ABI_PushRegistersAndAdjustStack(registersInUse, true);
// RSP alignment here is 8 due to the call.
ABI_PushRegistersAndAdjustStack(registersInUse, 8);
if (addrReg != ABI_PARAM1)
MOV(32, R(ABI_PARAM1), R((X64Reg)addrReg));
@@ -91,7 +89,7 @@ const u8 *TrampolineCache::GetReadTrampoline(const InstructionInfo &info, u32 re
MOV(32, R(dataReg), R(ABI_RETURN));
}
ABI_PopRegistersAndAdjustStack(registersInUse, true);
ABI_PopRegistersAndAdjustStack(registersInUse, 8);
RET();
return trampoline;
}
@@ -115,7 +113,7 @@ const u8 *TrampolineCache::GetWriteTrampoline(const InstructionInfo &info, u32 r
// PC is used by memory watchpoints (if enabled) or to print accurate PC locations in debug logs
MOV(32, PPCSTATE(pc), Imm32(pc));
ABI_PushRegistersAndAdjustStack(registersInUse, true);
ABI_PushRegistersAndAdjustStack(registersInUse, 8);
MOVTwo(64, ABI_PARAM1, dataReg, ABI_PARAM2, addrReg, ABI_PARAM3);
@@ -140,7 +138,7 @@ const u8 *TrampolineCache::GetWriteTrampoline(const InstructionInfo &info, u32 r
break;
}
ABI_PopRegistersAndAdjustStack(registersInUse, true);
ABI_PopRegistersAndAdjustStack(registersInUse, 8);
RET();
return trampoline;
Source/Core/Core/PowerPC/JitCommon/Jit_Util.cpp
+9 -9
View File
@@ -204,9 +204,9 @@ private:
void CallLambda(int sbits, const std::function<T(u32)>* lambda)
{
m_code->ABI_PushRegistersAndAdjustStack(m_registers_in_use, false);
m_code->ABI_PushRegistersAndAdjustStack(m_registers_in_use, 0);
m_code->ABI_CallLambdaC(lambda, m_address);
m_code->ABI_PopRegistersAndAdjustStack(m_registers_in_use, false);
m_code->ABI_PopRegistersAndAdjustStack(m_registers_in_use, 0);
MoveOpArgToReg(sbits, R(ABI_RETURN));
}
@@ -305,7 +305,7 @@ void EmuCodeBlock::SafeLoadToReg(X64Reg reg_value, const Gen::OpArg & opAddress,
}
else
{
ABI_PushRegistersAndAdjustStack(registersInUse, false);
ABI_PushRegistersAndAdjustStack(registersInUse, 0);
switch (accessSize)
{
case 64: ABI_CallFunctionC((void *)&Memory::Read_U64, address); break;
@@ -313,7 +313,7 @@ void EmuCodeBlock::SafeLoadToReg(X64Reg reg_value, const Gen::OpArg & opAddress,
case 16: ABI_CallFunctionC((void *)&Memory::Read_U16_ZX, address); break;
case 8: ABI_CallFunctionC((void *)&Memory::Read_U8_ZX, address); break;
}
ABI_PopRegistersAndAdjustStack(registersInUse, false);
ABI_PopRegistersAndAdjustStack(registersInUse, 0);
MEMCHECK_START
@@ -350,7 +350,7 @@ void EmuCodeBlock::SafeLoadToReg(X64Reg reg_value, const Gen::OpArg & opAddress,
FixupBranch fast = J_CC(CC_Z, true);
ABI_PushRegistersAndAdjustStack(registersInUse, false);
ABI_PushRegistersAndAdjustStack(registersInUse, 0);
switch (accessSize)
{
case 64:
@@ -366,7 +366,7 @@ void EmuCodeBlock::SafeLoadToReg(X64Reg reg_value, const Gen::OpArg & opAddress,
ABI_CallFunctionA((void *)&Memory::Read_U8_ZX, addr_loc);
break;
}
ABI_PopRegistersAndAdjustStack(registersInUse, false);
ABI_PopRegistersAndAdjustStack(registersInUse, 0);
MEMCHECK_START
@@ -470,9 +470,9 @@ void EmuCodeBlock::SafeWriteRegToReg(X64Reg reg_value, X64Reg reg_addr, int acce
FixupBranch fast = J_CC(CC_Z, true);
// PC is used by memory watchpoints (if enabled) or to print accurate PC locations in debug logs
MOV(32, PPCSTATE(pc), Imm32(jit->js.compilerPC));
bool noProlog = (0 != (flags & SAFE_LOADSTORE_NO_PROLOG));
size_t rsp_alignment = (flags & SAFE_LOADSTORE_NO_PROLOG) ? 8 : 0;
bool swap = !(flags & SAFE_LOADSTORE_NO_SWAP);
ABI_PushRegistersAndAdjustStack(registersInUse, noProlog);
ABI_PushRegistersAndAdjustStack(registersInUse, rsp_alignment);
switch (accessSize)
{
case 64:
@@ -488,7 +488,7 @@ void EmuCodeBlock::SafeWriteRegToReg(X64Reg reg_value, X64Reg reg_addr, int acce
ABI_CallFunctionRR((void *)&Memory::Write_U8, reg_value, reg_addr, false);
break;
}
ABI_PopRegistersAndAdjustStack(registersInUse, noProlog);
ABI_PopRegistersAndAdjustStack(registersInUse, rsp_alignment);
FixupBranch exit = J();
SetJumpTarget(fast);
UnsafeWriteRegToReg(reg_value, reg_addr, accessSize, 0, swap);