Fix stack misalignment fix.

Source/Core/Common/Src/x64Emitter.h

@@ -639,7 +639,7 @@ class XEmitter
 
 	// Pass a register as a parameter.
 	void ABI_CallFunctionR(void *func, Gen::X64Reg reg1);
-	void ABI_CallFunctionRR(void *func, Gen::X64Reg reg1, Gen::X64Reg reg2);
+	void ABI_CallFunctionRR(void *func, Gen::X64Reg reg1, Gen::X64Reg reg2, bool noProlog = false);
 
 	// A function that doesn't have any control over what it will do to regs,
 	// such as the dispatcher, should be surrounded by these.
@@ -652,14 +652,9 @@ class XEmitter
 	void ABI_PushAllCallerSavedRegsAndAdjustStack();
 	void ABI_PopAllCallerSavedRegsAndAdjustStack();
 
-	unsigned int ABI_GetAlignedFrameSize(unsigned int frameSize);
-	void ABI_AlignStack(unsigned int frameSize);
-	void ABI_RestoreStack(unsigned int frameSize);
-
-	// Sets up a __cdecl function.
-	// Only x64 really needs the parameter count.
-	void ABI_EmitPrologue(int maxCallParams);
-	void ABI_EmitEpilogue(int maxCallParams);
+	unsigned int ABI_GetAlignedFrameSize(unsigned int frameSize, bool noProlog = false);
+	void ABI_AlignStack(unsigned int frameSize, bool noProlog = false);
+	void ABI_RestoreStack(unsigned int frameSize, bool noProlog = false);
 
 	#ifdef _M_IX86
 	inline int ABI_GetNumXMMRegs() { return 8; }

Source/Core/Common/Src/x64Thunk.cpp

@@ -91,35 +91,27 @@ void *ThunkManager::ProtectFunction(void *function, int num_params)
 		PanicAlert("Trying to protect functions before the emu is started. Bad bad bad.");
 
 	const u8 *call_point = GetCodePtr();
-	// Make sure to align stack.
 #ifdef _M_X64
-#ifdef _WIN32
-	SUB(64, R(ESP), Imm8(0x28));
-#else
-	SUB(64, R(ESP), Imm8(0x8));
-#endif
+	// Make sure to align stack.
+	ABI_AlignStack(0, true);
 	CALL((void*)save_regs);
 	CALL((void*)function);
 	CALL((void*)load_regs);
-#ifdef _WIN32
-	ADD(64, R(ESP), Imm8(0x28));
-#else
-	ADD(64, R(ESP), Imm8(0x8));
-#endif
+	ABI_RestoreStack(0, true);
 	RET();
 #else
 	CALL((void*)save_regs);
 	// Since parameters are in the previous stack frame, not in registers, this takes some
 	// trickery : we simply re-push the parameters. might not be optimal, but that doesn't really
 	// matter.
-	ABI_AlignStack(num_params * 4);
+	ABI_AlignStack(num_params * 4, true);
 	unsigned int alignedSize = ABI_GetAlignedFrameSize(num_params * 4);
 	for (int i = 0; i < num_params; i++) {
 		// ESP is changing, so we do not need i
-		PUSH(32, MDisp(ESP, alignedSize - 4));
+		PUSH(32, MDisp(ESP, alignedSize));
 	}
 	CALL(function);
-	ABI_RestoreStack(num_params * 4);
+	ABI_RestoreStack(num_params * 4, true);
 	CALL((void*)load_regs);
 	RET();
 #endif

Source/Core/Core/Src/PowerPC/JitCommon/JitAsmCommon.cpp

@@ -167,7 +167,7 @@ void CommonAsmRoutines::GenQuantizedStores() {
 	MOV(64, MComplex(RBX, RCX, SCALE_1, 0), R(RAX));
 	FixupBranch skip_complex = J();
 	SetJumpTarget(too_complex);
-	ABI_CallFunctionRR(thunks.ProtectFunction((void *)&WriteDual32, 2), RAX, RCX); 
+	ABI_CallFunctionRR(thunks.ProtectFunction((void *)&WriteDual32, 2), RAX, RCX, /* noProlog = */ true); 
 	SetJumpTarget(skip_complex);
 	RET();
 #else
@@ -184,10 +184,10 @@ void CommonAsmRoutines::GenQuantizedStores() {
 	FixupBranch arg2 = J();
 	SetJumpTarget(argh);
 	MOV(32, R(EAX), M(((char*)&psTemp)));
-	ABI_CallFunctionRR(thunks.ProtectFunction((void *)&Memory::Write_U32, 2), EAX, ECX); 
+	ABI_CallFunctionRR(thunks.ProtectFunction((void *)&Memory::Write_U32, 2), EAX, ECX, /* noProlog = */ true); 
 	MOV(32, R(EAX), M(((char*)&psTemp)+4));
 	ADD(32, R(ECX), Imm32(4));
-	ABI_CallFunctionRR(thunks.ProtectFunction((void *)&Memory::Write_U32, 2), EAX, ECX); 
+	ABI_CallFunctionRR(thunks.ProtectFunction((void *)&Memory::Write_U32, 2), EAX, ECX, /* noProlog = */ true); 
 	SetJumpTarget(arg2);
 	RET();
 #endif
@@ -206,7 +206,7 @@ void CommonAsmRoutines::GenQuantizedStores() {
 	PACKSSDW(XMM0, R(XMM0));
 	PACKUSWB(XMM0, R(XMM0));
 	MOVD_xmm(R(EAX), XMM0);
-	SafeWriteRegToReg(AX, ECX, 16, 0, false);
+	SafeWriteRegToReg(AX, ECX, 16, 0, false, true);
 
 	RET();
 
@@ -225,7 +225,7 @@ void CommonAsmRoutines::GenQuantizedStores() {
 	PACKSSWB(XMM0, R(XMM0));
 	MOVD_xmm(R(EAX), XMM0);
 
-	SafeWriteRegToReg(AX, ECX, 16, 0, false);
+	SafeWriteRegToReg(AX, ECX, 16, 0, false, true);
 
 	RET();
 
@@ -251,7 +251,7 @@ void CommonAsmRoutines::GenQuantizedStores() {
 	MOV(16, R(AX), M((char*)psTemp + 4));
 
 	BSWAP(32, EAX);
-	SafeWriteRegToReg(EAX, ECX, 32, 0, false);
+	SafeWriteRegToReg(EAX, ECX, 32, 0, false, true);
 
 	RET();
 
@@ -271,7 +271,7 @@ void CommonAsmRoutines::GenQuantizedStores() {
 	MOVD_xmm(R(EAX), XMM0);
 	BSWAP(32, EAX);
 	ROL(32, R(EAX), Imm8(16));
-	SafeWriteRegToReg(EAX, ECX, 32, 0, false);
+	SafeWriteRegToReg(EAX, ECX, 32, 0, false, true);
 
 	RET();
 
@@ -303,11 +303,11 @@ void CommonAsmRoutines::GenQuantizedSingleStores() {
 		// TODO: SafeWriteFloat
 		MOVSS(M(&psTemp[0]), XMM0);
 		MOV(32, R(EAX), M(&psTemp[0]));
-		SafeWriteRegToReg(EAX, ECX, 32, 0, false);
+		SafeWriteRegToReg(EAX, ECX, 32, 0, false, true);
 	} else {
 		MOVSS(M(&psTemp[0]), XMM0);
 		MOV(32, R(EAX), M(&psTemp[0]));
-		SafeWriteRegToReg(EAX, ECX, 32, 0, true);
+		SafeWriteRegToReg(EAX, ECX, 32, 0, true, true);
 	}*/
 
 	const u8* storeSingleU8 = AlignCode4();  // Used by MKWii
@@ -318,7 +318,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores() {
 	MAXSS(XMM0, R(XMM1));
 	MINSS(XMM0, M((void *)&m_255));
 	CVTTSS2SI(EAX, R(XMM0));
-	SafeWriteRegToReg(AL, ECX, 8, 0, true);
+	SafeWriteRegToReg(AL, ECX, 8, 0, true, true);
 	RET();
 
 	const u8* storeSingleS8 = AlignCode4();
@@ -328,7 +328,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores() {
 	MAXSS(XMM0, M((void *)&m_m128));
 	MINSS(XMM0, M((void *)&m_127));
 	CVTTSS2SI(EAX, R(XMM0));
-	SafeWriteRegToReg(AL, ECX, 8, 0, true);
+	SafeWriteRegToReg(AL, ECX, 8, 0, true, true);
 	RET();
 
 	const u8* storeSingleU16 = AlignCode4();  // Used by MKWii
@@ -339,7 +339,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores() {
 	MAXSS(XMM0, R(XMM1));
 	MINSS(XMM0, M((void *)&m_65535));
 	CVTTSS2SI(EAX, R(XMM0));
-	SafeWriteRegToReg(EAX, ECX, 16, 0, true);
+	SafeWriteRegToReg(EAX, ECX, 16, 0, true, true);
 	RET();
 
 	const u8* storeSingleS16 = AlignCode4();
@@ -349,7 +349,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores() {
 	MAXSS(XMM0, M((void *)&m_m32768));
 	MINSS(XMM0, M((void *)&m_32767));
 	CVTTSS2SI(EAX, R(XMM0));
-	SafeWriteRegToReg(EAX, ECX, 16, 0, true);
+	SafeWriteRegToReg(EAX, ECX, 16, 0, true, true);
 	RET();
 
 	singleStoreQuantized = reinterpret_cast<const u8**>(const_cast<u8*>(AlignCode16()));

Source/Core/Core/Src/PowerPC/JitCommon/Jit_Util.cpp

@@ -223,7 +223,7 @@ void EmuCodeBlock::UnsafeWriteRegToReg(X64Reg reg_value, X64Reg reg_addr, int ac
 }
 
 // Destroys both arg registers
-void EmuCodeBlock::SafeWriteRegToReg(X64Reg reg_value, X64Reg reg_addr, int accessSize, s32 offset, bool swap)
+void EmuCodeBlock::SafeWriteRegToReg(X64Reg reg_value, X64Reg reg_addr, int accessSize, s32 offset, bool swap, bool noProlog)
 {
 	if (offset)
 		ADD(32, R(reg_addr), Imm32((u32)offset));
@@ -247,9 +247,9 @@ void EmuCodeBlock::SafeWriteRegToReg(X64Reg reg_value, X64Reg reg_addr, int acce
 	MOV(32, M(&PC), Imm32(jit->js.compilerPC)); // Helps external systems know which instruction triggered the write
 	switch (accessSize)
 	{
-	case 32: ABI_CallFunctionRR(thunks.ProtectFunction(swap ? ((void *)&Memory::Write_U32) : ((void *)&Memory::Write_U32_Swap), 2), reg_value, reg_addr); break;
-	case 16: ABI_CallFunctionRR(thunks.ProtectFunction(swap ? ((void *)&Memory::Write_U16) : ((void *)&Memory::Write_U16_Swap), 2), reg_value, reg_addr); break;
-	case 8:  ABI_CallFunctionRR(thunks.ProtectFunction((void *)&Memory::Write_U8, 2), reg_value, reg_addr);  break;
+	case 32: ABI_CallFunctionRR(thunks.ProtectFunction(swap ? ((void *)&Memory::Write_U32) : ((void *)&Memory::Write_U32_Swap), 2), reg_value, reg_addr, noProlog); break;
+	case 16: ABI_CallFunctionRR(thunks.ProtectFunction(swap ? ((void *)&Memory::Write_U16) : ((void *)&Memory::Write_U16_Swap), 2), reg_value, reg_addr, noProlog); break;
+	case 8:  ABI_CallFunctionRR(thunks.ProtectFunction((void *)&Memory::Write_U8, 2), reg_value, reg_addr, noProlog);  break;
 	}
 	FixupBranch exit = J();
 	SetJumpTarget(fast);

Source/Core/Core/Src/PowerPC/JitCommon/Jit_Util.h

@@ -16,7 +16,7 @@ class EmuCodeBlock : public Gen::XCodeBlock {
 	void UnsafeWriteRegToReg(Gen::X64Reg reg_value, Gen::X64Reg reg_addr, int accessSize, s32 offset = 0, bool swap = true);
 	void UnsafeLoadToEAX(const Gen::OpArg & opAddress, int accessSize, s32 offset, bool signExtend);
 	void SafeLoadToEAX(const Gen::OpArg & opAddress, int accessSize, s32 offset, bool signExtend);
-	void SafeWriteRegToReg(Gen::X64Reg reg_value, Gen::X64Reg reg_addr, int accessSize, s32 offset, bool swap = true);
+	void SafeWriteRegToReg(Gen::X64Reg reg_value, Gen::X64Reg reg_addr, int accessSize, s32 offset, bool swap = true, bool noProlog = false);
 
 	// Trashes both inputs and EAX.
 	void SafeWriteFloatToReg(Gen::X64Reg xmm_value, Gen::X64Reg reg_addr);

Source/Core/VideoCommon/Src/VertexLoader.cpp

@@ -217,7 +217,7 @@ void VertexLoader::CompileVertexTranslator()
 		PanicAlert("Trying to recompile a vertex translator");
 
 	m_compiledCode = GetCodePtr();
-	ABI_EmitPrologue(4);
+	ABI_PushAllCalleeSavedRegsAndAdjustStack();
 
 	// Start loop here
 	const u8 *loop_start = GetCodePtr();
@@ -499,7 +499,8 @@ void VertexLoader::CompileVertexTranslator()
 #endif
 
 	J_CC(CC_NZ, loop_start, true);
-	ABI_EmitEpilogue(4);
+	ABI_PopAllCalleeSavedRegsAndAdjustStack();
+	RET();
 #endif
 	m_NativeFmt->Initialize(vtx_decl);
 }

Source/Core/VideoCommon/Src/x64DLCache.cpp

@@ -409,7 +409,7 @@ void CompileAndRunDisplayList(u32 address, u32 size, CachedDisplayList *dl)
 
 		emitter.AlignCode4();
 		dl->compiled_code = emitter.GetCodePtr();
-		emitter.ABI_EmitPrologue(4);
+		emitter.ABI_PushAllCalleeSavedRegsAndAdjustStack();
 
 		while (g_pVideoData < end)
 		{
@@ -572,7 +572,8 @@ void CompileAndRunDisplayList(u32 address, u32 size, CachedDisplayList *dl)
 				break;
 			}
 		}
-		emitter.ABI_EmitEpilogue(4);
+		emitter.ABI_PopAllCalleeSavedRegsAndAdjustStack();
+		emitter.RET();
 		INCSTAT(stats.numDListsCalled);
 		INCSTAT(stats.thisFrame.numDListsCalled);
 		Statistics::SwapDL();