A bunch more WIP JIT work.

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@1779 8ced0084-cf51-0410-be5f-012b33b47a6e

5 files changed, 203 insertions, 222 deletions

diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/IR.cpp b/Source/Core/Core/Src/PowerPC/Jit64IL/IR.cpp
index 7db7c6e954..a06431f8d1 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/IR.cpp
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/IR.cpp
@@ -245,6 +245,13 @@ InstLoc IRBuilder::FoldZeroOp(unsigned Opcode, unsigned extra) {
 			GRegCache[extra] = EmitZeroOp(LoadGReg, extra);
 		return GRegCache[extra];
 	}
+	if (Opcode == LoadFReg) {
+		// Reg load folding: if we already loaded the value,
+		// load it again
+		if (!FRegCache[extra])
+			FRegCache[extra] = EmitZeroOp(LoadFReg, extra);
+		return FRegCache[extra];
+	}
 	if (Opcode == LoadCarry) {
 		if (!CarryCache)
 			CarryCache = EmitZeroOp(LoadGReg, extra);
@@ -270,6 +277,14 @@ InstLoc IRBuilder::FoldUOp(unsigned Opcode, InstLoc Op1, unsigned extra) {
 		GRegCacheStore[extra] = EmitUOp(StoreGReg, Op1, extra);
 		return GRegCacheStore[extra];
 	}
+	if (Opcode == StoreFReg) {
+		FRegCache[extra] = Op1;
+		if (FRegCacheStore[extra]) {
+			*FRegCacheStore[extra] = 0;
+		}
+		FRegCacheStore[extra] = EmitUOp(StoreFReg, Op1, extra);
+		return FRegCacheStore[extra];
+	}
 	if (Opcode == StoreCarry) {
 		CarryCache = Op1;
 		if (CarryCacheStore) {
@@ -286,6 +301,10 @@ InstLoc IRBuilder::FoldUOp(unsigned Opcode, InstLoc Op1, unsigned extra) {
 		CRCacheStore[extra] = EmitUOp(StoreCR, Op1, extra);
 		return CRCacheStore[extra];
 	}
+	if (Opcode == CompactMRegToPacked) {
+		if (getOpcode(*Op1) == ExpandPackedToMReg)
+			return getOp1(Op1);
+	}
 
 	return EmitUOp(Opcode, Op1, extra);
 }
@@ -441,6 +460,8 @@ InstLoc IRBuilder::FoldInterpreterFallback(InstLoc Op1, InstLoc Op2) {
 	for (unsigned i = 0; i < 32; i++) {
 		GRegCache[i] = 0;
 		GRegCacheStore[i] = 0;
+		FRegCache[i] = 0;
+		FRegCacheStore[i] = 0;
 	}
 	CarryCache = 0;
 	CarryCacheStore = 0;
@@ -610,9 +631,10 @@ static X64Reg fregFindFreeReg(RegInfo& RI) {
 	for (unsigned i = 0; i < FRegAllocSize; i++)
 		if (RI.fregs[FRegAllocOrder[i]] == 0)
 			return FRegAllocOrder[i];
-	// XMM0/1 are scratch, so we don't allocate it
-	fregSpill(RI, XMM7);
-	return XMM7;
+	static unsigned nextReg = 0;
+	X64Reg reg = FRegAllocOrder[nextReg++ % FRegAllocSize];
+	fregSpill(RI, reg);
+	return reg;
 }
 
 static OpArg regLocForInst(RegInfo& RI, InstLoc I) {
@@ -1016,10 +1038,12 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 		case SExt16:
 		case BSwap32:
 		case BSwap16:
+		case Cntlzw:
 		case DupSingleToMReg:
 		case DoubleToSingle:
 		case ExpandPackedToMReg:
 		case CompactMRegToPacked:
+		case FPNeg:
 			if (thisUsed)
 				regMarkUse(RI, I, getOp1(I), 1);
 			break;
@@ -1062,6 +1086,9 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 		case ICmpSgt:
 		case FSMul:
 		case FSAdd:
+		case FPAdd:
+		case FPMul:
+		case FPSub:
 		case InsertDoubleInMReg:
 			if (thisUsed) {
 				regMarkUse(RI, I, getOp1(I), 1);
@@ -1237,6 +1264,17 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 			regNormalRegClear(RI, I);
 			break;
 		}
+		case Cntlzw: {
+			if (!thisUsed) break;
+			X64Reg reg = regUReg(RI, I);
+			Jit->MOV(32, R(ECX), Imm32(63));
+			Jit->BSR(32, reg, regLocForInst(RI, getOp1(I)));
+			Jit->CMOVcc(32, reg, R(ECX), CC_Z);
+			Jit->XOR(32, R(reg), Imm8(31));
+			RI.regs[reg] = I;
+			regNormalRegClear(RI, I);
+			break;
+		}
 		case And: {
 			if (!thisUsed) break;
 			regEmitBinInst(RI, I, &Jit64::AND);
@@ -1447,6 +1485,17 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 			fregNormalRegClear(RI, I);
 			break;
 		}
+		case FPNeg: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			static const u32 GC_ALIGNED16(psSignBits[4]) = 
+				{0x80000000, 0x80000000};
+			Jit->PXOR(reg, M((void*)&psSignBits));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
 		case LoadFReg: {
 			if (!thisUsed) break;
 			X64Reg reg = fregFindFreeReg(RI);
@@ -1488,6 +1537,33 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 			fregNormalRegClear(RI, I);
 			break;
 		}
+		case FPAdd: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->ADDPS(reg, fregLocForInst(RI, getOp2(I)));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
+		case FPMul: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->MULPS(reg, fregLocForInst(RI, getOp2(I)));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
+		case FPSub: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->SUBPS(reg, fregLocForInst(RI, getOp2(I)));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
 		case CInt32:
 		case CInt16: {
 			if (!thisUsed) break;
@@ -1579,11 +1655,19 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 			if (!thisUsed) break;
 			// FIXME: Optimize!
 			InstLoc Op = I - 1 - (*I >> 8);
-			X64Reg reg = regFindFreeReg(RI);
-			Jit->MOV(32, R(reg), regLocForInst(RI, Op));
-			RI.regs[reg] = I;
-			if (RI.IInfo[I - RI.FirstI] & 4)
-				regClearInst(RI, Op);
+			if (isFResult(*Op)) {
+				X64Reg reg = fregFindFreeReg(RI);
+				Jit->MOVAPD(reg, fregLocForInst(RI, Op));
+				RI.fregs[reg] = I;
+				if (RI.IInfo[I - RI.FirstI] & 4)
+					fregClearInst(RI, Op);
+			} else {
+				X64Reg reg = regFindFreeReg(RI);
+				Jit->MOV(32, R(reg), regLocForInst(RI, Op));
+				RI.regs[reg] = I;
+				if (RI.IInfo[I - RI.FirstI] & 4)
+					regClearInst(RI, Op);
+			}
 			break;
 		}
 		case Nop: break;
diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/IR.h b/Source/Core/Core/Src/PowerPC/Jit64IL/IR.h
index 4fafd65b81..092831ed03 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/IR.h
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/IR.h
@@ -41,6 +41,7 @@ namespace IREmitter {
 		SExt16,
 		BSwap32,
 		BSwap16,
+		Cntlzw, // Count leading zeros
 		Load8,  // These loads zext
 		Load16,
 		Load32,
@@ -143,19 +144,26 @@ namespace IREmitter {
 		ForceToDouble,
 		ForceToMReg,
 #endif
+		FResult_Start,
 		LoadSingle,
 		LoadDouble,
 		LoadPaired, // This handles quantizers itself
-		StorePaired,
 		DoubleToSingle,
 		DupSingleToMReg,
+		DupSingleToPacked,
 		InsertDoubleInMReg,
 		ExpandPackedToMReg,
 		CompactMRegToPacked,
 		LoadFReg,
-		StoreFReg,
 		FSMul,
 		FSAdd,
+		FPAdd,
+		FPMul,
+		FPSub,
+		FPNeg,
+		FResult_End,
+		StorePaired,
+		StoreFReg,
 
 		// "Trinary" operators
 		// FIXME: Need to change representation!
@@ -189,37 +197,9 @@ namespace IREmitter {
 		return getOpcode(i) >= CInt16 && getOpcode(i) <= CInt32;
 	}
 
-	unsigned inline isUnary(Inst i) {
-		return getOpcode(i) >= SExt8 && getOpcode(i) <= BSwap16;
-	}
-
-	unsigned inline isBinary(Inst i) {
-		return getOpcode(i) >= Add && getOpcode(i) <= ICmpCRUnsigned;
-	}
-
-	unsigned inline isMemLoad(Inst i) {
-		return getOpcode(i) >= Load8 && getOpcode(i) <= Load32;
-	}
-
-	unsigned inline isMemStore(Inst i) {
-		return getOpcode(i) >= Store8 && getOpcode(i) <= Store32;
-	}
-
-	unsigned inline isRegLoad(Inst i) {
-		return getOpcode(i) >= LoadGReg && getOpcode(i) <= LoadCR;
-	}
-
-	unsigned inline isRegStore(Inst i) {
-		return getOpcode(i) >= LoadGReg && getOpcode(i) <= LoadCR;
-	}
-
-	unsigned inline isBranch(Inst i) {
-		return getOpcode(i) >= BranchUncond &&
-		       getOpcode(i) <= BranchCond;
-	}
-
-	unsigned inline isInterpreterFallback(Inst i) {
-		return getOpcode(i) == InterpreterFallback;
+	unsigned inline isFResult(Inst i) {
+		return getOpcode(i) > FResult_Start && 
+		       getOpcode(i) < FResult_End;
 	}
 
 	InstLoc inline getOp1(InstLoc i) {
@@ -360,6 +340,9 @@ namespace IREmitter {
 		InstLoc EmitSExt8(InstLoc op1) {
 			return FoldUOp(SExt8, op1);
 		}
+		InstLoc EmitCntlzw(InstLoc op1) {
+			return FoldUOp(Cntlzw, op1);
+		}
 		InstLoc EmitICmpCRSigned(InstLoc op1, InstLoc op2) {
 			return FoldBiOp(ICmpCRSigned, op1, op2);
 		}
@@ -405,6 +388,9 @@ namespace IREmitter {
 		InstLoc EmitDupSingleToMReg(InstLoc val) {
 			return FoldUOp(DupSingleToMReg, val);
 		}
+		InstLoc EmitDupSingleToPacked(InstLoc val) {
+			return FoldUOp(DupSingleToPacked, val);
+		}
 		InstLoc EmitInsertDoubleInMReg(InstLoc val, InstLoc reg) {
 			return FoldBiOp(InsertDoubleInMReg, val, reg);
 		}
@@ -420,6 +406,18 @@ namespace IREmitter {
 		InstLoc EmitFSAdd(InstLoc op1, InstLoc op2) {
 			return FoldBiOp(FSAdd, op1, op2);
 		}
+		InstLoc EmitFPAdd(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FPAdd, op1, op2);
+		}
+		InstLoc EmitFPMul(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FPMul, op1, op2);
+		}
+		InstLoc EmitFPSub(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FPSub, op1, op2);
+		}
+		InstLoc EmitFPNeg(InstLoc op1) {
+			return FoldUOp(FPNeg, op1);
+		}
 		InstLoc EmitDoubleToSingle(InstLoc op1) {
 			return FoldUOp(DoubleToSingle, op1);
 		}
@@ -439,6 +437,8 @@ namespace IREmitter {
 			for (unsigned i = 0; i < 32; i++) {
 				GRegCache[i] = 0;
 				GRegCacheStore[i] = 0;
+				FRegCache[i] = 0;
+				FRegCacheStore[i] = 0;
 			}
 			CarryCache = 0;
 			CarryCacheStore = 0;
@@ -458,6 +458,8 @@ namespace IREmitter {
 		InstLoc curReadPtr;
 		InstLoc GRegCache[32];
 		InstLoc GRegCacheStore[32];
+		InstLoc FRegCache[32];
+		InstLoc FRegCacheStore[32];
 		InstLoc CarryCache;
 		InstLoc CarryCacheStore;
 		InstLoc CRCache[8];
diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Integer.cpp b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Integer.cpp
index 9be273564c..7a741528b2 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Integer.cpp
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Integer.cpp
@@ -179,39 +179,32 @@
 
 	void Jit64::subfic(UGeckoInstruction inst)
 	{
-		if(Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITIntegerOff)
-			{Default(inst); return;} // turn off from debugger
-
-		INSTRUCTION_START;
-		int a = inst.RA, d = inst.RD;
-		gpr.FlushLockX(ECX);
-		gpr.Lock(a, d);
-		gpr.LoadToX64(d, a == d, true);
-		int imm = inst.SIMM_16;
-		MOV(32, R(EAX), gpr.R(a));
-		NOT(32, R(EAX));
-		ADD(32, R(EAX), Imm32(imm + 1));
-		MOV(32, gpr.R(d), R(EAX));
-		//GenerateCarry(ECX);
-		gpr.UnlockAll();
-		gpr.UnlockAllX();
-		// This instruction has no RC flag
+		Default(inst);
+		return;
+		// FIXME: Disabling until I figure out subfcx
+		IREmitter::InstLoc val, test, c;
+		c = ibuild.EmitIntConst(inst.SIMM_16);
+		val = ibuild.EmitSub(c, ibuild.EmitLoadGReg(inst.RA));
+		ibuild.EmitStoreGReg(val, inst.RD);
+		test = ibuild.EmitICmpUgt(val, c);
+		ibuild.EmitStoreCarry(test);
 	}
 
 	void Jit64::subfcx(UGeckoInstruction inst) 
 	{
-		INSTRUCTION_START;
 		Default(inst);
 		return;
-		/*
-		u32 a = m_GPR[_inst.RA];
-		u32 b = m_GPR[_inst.RB];
-		m_GPR[_inst.RD] = b - a;
-		SetCarry(a == 0 || Helper_Carry(b, 0-a));
-
-		if (_inst.OE) PanicAlert("OE: subfcx");
-		if (_inst.Rc) Helper_UpdateCR0(m_GPR[_inst.RD]);
-		*/
+		// FIXME: Figure out what the heck is going wrong here...
+		if (inst.OE) PanicAlert("OE: subfcx");
+		IREmitter::InstLoc val, test, lhs, rhs;
+		lhs = ibuild.EmitLoadGReg(inst.RB);
+		rhs = ibuild.EmitLoadGReg(inst.RA);
+		val = ibuild.EmitSub(lhs, rhs);
+		ibuild.EmitStoreGReg(val, inst.RD);
+		test = ibuild.EmitICmpUgt(rhs, lhs);
+		ibuild.EmitStoreCarry(test);
+		if (inst.Rc)
+			ComputeRC(ibuild, val);
 	}
 
 	void Jit64::subfex(UGeckoInstruction inst) 
@@ -394,33 +387,14 @@
 
 	void Jit64::rlwnmx(UGeckoInstruction inst)
 	{
-		if(Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITIntegerOff)
-			{Default(inst); return;} // turn off from debugger
-
-		INSTRUCTION_START;
-		int a = inst.RA, b = inst.RB, s = inst.RS;
-		if (gpr.R(a).IsImm())
-		{
-			Default(inst);
-			return;
-		}
-
-		u32 mask = Helper_Mask(inst.MB, inst.ME);
-		gpr.FlushLockX(ECX);
-		gpr.Lock(a, b, s);
-		MOV(32, R(EAX), gpr.R(s));
-		MOV(32, R(ECX), gpr.R(b));
-		AND(32, R(ECX), Imm32(0x1f));
-		ROL(32, R(EAX), R(ECX));
-		AND(32, R(EAX), Imm32(mask));
-		MOV(32, gpr.R(a), R(EAX));
-		gpr.UnlockAll();
-		gpr.UnlockAllX();
+		INSTRUCTION_START
+		unsigned mask = Helper_Mask(inst.MB, inst.ME);
+		IREmitter::InstLoc val = ibuild.EmitLoadGReg(inst.RS);
+		val = ibuild.EmitRol(val, ibuild.EmitLoadGReg(inst.RB));
+		val = ibuild.EmitAnd(val, ibuild.EmitIntConst(mask));
+		ibuild.EmitStoreGReg(val, inst.RA);
 		if (inst.Rc)
-		{
-			MOV(32, R(EAX), gpr.R(a));
-			CALL((u8*)asm_routines.computeRc);
-		}
+			ComputeRC(ibuild, val);
 	}
 
 	void Jit64::negx(UGeckoInstruction inst)
@@ -509,30 +483,9 @@
 	// count leading zeroes
 	void Jit64::cntlzwx(UGeckoInstruction inst)
 	{
-		if(Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITIntegerOff)
-			{Default(inst); return;} // turn off from debugger
-
-		INSTRUCTION_START;
-		int a = inst.RA;
-		int s = inst.RS;
-		if (gpr.R(a).IsImm() || gpr.R(s).IsImm() || s == a)
-		{
-			Default(inst);
-			return;
-		}
-		gpr.Lock(a,s);
-		gpr.LoadToX64(a,false);
-		BSR(32, gpr.R(a).GetSimpleReg(), gpr.R(s));
-		FixupBranch gotone = J_CC(CC_NZ);
-		MOV(32, gpr.R(a), Imm32(63));
-		SetJumpTarget(gotone);
-		XOR(32, gpr.R(a), Imm8(0x1f));  // flip order
-		gpr.UnlockAll();
-
+		IREmitter::InstLoc val = ibuild.EmitLoadGReg(inst.RS);
+		val = ibuild.EmitCntlzw(val);
+		ibuild.EmitStoreGReg(val, inst.RA);
 		if (inst.Rc)
-		{
-			MOV(32, R(EAX), gpr.R(a));
-			CALL((u8*)asm_routines.computeRc);
-			// TODO: Check PPC manual too
-		}
+			ComputeRC(ibuild, val);
 	}
diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_LoadStore.cpp b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_LoadStore.cpp
index 06f21b354d..fcd512f232 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_LoadStore.cpp
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_LoadStore.cpp
@@ -88,8 +88,10 @@ void Jit64::lXz(UGeckoInstruction inst)
 	IREmitter::InstLoc addr = ibuild.EmitIntConst(inst.SIMM_16);
 	if (inst.RA)
 		addr = ibuild.EmitAdd(addr, ibuild.EmitLoadGReg(inst.RA));
+	if (inst.OPCD & 1)
+		ibuild.EmitStoreGReg(addr, inst.RA);
 	IREmitter::InstLoc val;
-	switch (inst.OPCD)
+	switch (inst.OPCD & -2)
 	{
 	case 32: val = ibuild.EmitLoad32(addr); break; //lwz	
 	case 40: val = ibuild.EmitLoad16(addr); break; //lhz
diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Paired.cpp b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Paired.cpp
index dd2f95ba25..b0b2a3cf58 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Paired.cpp
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Paired.cpp
@@ -159,70 +159,31 @@
 	add a,b,a
 	*/
 
-	//There's still a little bit more optimization that can be squeezed out of this
-	void Jit64::tri_op(int d, int a, int b, bool reversible, void (XEmitter::*op)(X64Reg, OpArg))
-	{
-		fpr.Lock(d, a, b);
-		
-		if (d == a)
-		{
-			fpr.LoadToX64(d, true);
-			(this->*op)(fpr.RX(d), fpr.R(b));
-		}
-		else if (d == b && reversible)
-		{
-			fpr.LoadToX64(d, true);
-			(this->*op)(fpr.RX(d), fpr.R(a));
-		}
-		else if (a != d && b != d) 
-		{
-			//sources different from d, can use rather quick solution
-			fpr.LoadToX64(d, false);
-			MOVAPD(fpr.RX(d), fpr.R(a));
-			(this->*op)(fpr.RX(d), fpr.R(b));
-		}
-		else if (b != d)
-		{
-			fpr.LoadToX64(d, false);
-			MOVAPD(XMM0, fpr.R(b));
-			MOVAPD(fpr.RX(d), fpr.R(a));
-			(this->*op)(fpr.RX(d), Gen::R(XMM0));
-		}
-		else //Other combo, must use two temps :(
-		{
-			MOVAPD(XMM0, fpr.R(a));
-			MOVAPD(XMM1, fpr.R(b));
-			fpr.LoadToX64(d, false);
-			(this->*op)(XMM0, Gen::R(XMM1));
-			MOVAPD(fpr.RX(d), Gen::R(XMM0));
-		}
-		ForceSinglePrecisionP(fpr.RX(d));
-		fpr.UnlockAll();
-	}
-
 	void Jit64::ps_arith(UGeckoInstruction inst)
-	{	
-		if(Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITPairedOff)
-			{Default(inst); return;} // turn off from debugger
-		INSTRUCTION_START;
-		if (inst.Rc) {
+	{
+		if (inst.Rc || (inst.SUBOP5 != 21 && inst.SUBOP5 != 20 && inst.SUBOP5 != 25)) {
 			Default(inst); return;
 		}
+		IREmitter::InstLoc val = ibuild.EmitLoadFReg(inst.FA), rhs;
+		if (inst.SUBOP5 == 25)
+			rhs = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FC));
+		else
+			rhs = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FB));
+		val = ibuild.EmitCompactMRegToPacked(val);
+		
 		switch (inst.SUBOP5)
 		{
-		case 18: tri_op(inst.FD, inst.FA, inst.FB, false, &XEmitter::DIVPD); break; //div
-		case 20: tri_op(inst.FD, inst.FA, inst.FB, false, &XEmitter::SUBPD); break; //sub 
-		case 21: tri_op(inst.FD, inst.FA, inst.FB, true,  &XEmitter::ADDPD); break; //add
-		case 23://sel
-			Default(inst);
+		case 20:
+			val = ibuild.EmitFPSub(val, rhs);
 			break;
-		case 24://res
-			Default(inst);
+		case 21:
+			val = ibuild.EmitFPAdd(val, rhs);
 			break;
-		case 25: tri_op(inst.FD, inst.FA, inst.FC, true, &XEmitter::MULPD); break; //mul
-		default:
-			_assert_msg_(DYNA_REC, 0, "ps_arith WTF!!!");
+		case 25:
+			val = ibuild.EmitFPMul(val, rhs);
 		}
+		val = ibuild.EmitExpandPackedToMReg(val);
+		ibuild.EmitStoreFReg(val, inst.FD);
 	}
 
 	void Jit64::ps_sum(UGeckoInstruction inst)
@@ -347,58 +308,37 @@
 	//TODO: add optimized cases
 	void Jit64::ps_maddXX(UGeckoInstruction inst)
 	{
-		if(Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITPairedOff)
-			{Default(inst); return;} // turn off from debugger
-		INSTRUCTION_START;
-		if (inst.Rc) {
+		if (inst.Rc || (inst.SUBOP5 != 28 && inst.SUBOP5 != 29 && inst.SUBOP5 != 30)) {
 			Default(inst); return;
 		}
-		int a = inst.FA;
-		int b = inst.FB;
-		int c = inst.FC;
-		int d = inst.FD;
-		fpr.Lock(a,b,c,d);
-
-		MOVAPD(XMM0, fpr.R(a));
+		
+		IREmitter::InstLoc val = ibuild.EmitLoadFReg(inst.FA), op2, op3;
+		val = ibuild.EmitCompactMRegToPacked(val);
 		switch (inst.SUBOP5)
 		{
-		case 14: //madds0
-			MOVDDUP(XMM1, fpr.R(c));
-			MULPD(XMM0, R(XMM1));
-			ADDPD(XMM0, fpr.R(b));
+		case 28: {//msub
+			op2 = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FC));
+			val = ibuild.EmitFPMul(val, op2);
+			op3 = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FB));
+			val = ibuild.EmitFPSub(val, op3);
 			break;
-		case 15: //madds1
-			MOVAPD(XMM1, fpr.R(c));
-			SHUFPD(XMM1, R(XMM1), 3); // copy higher to lower
-			MULPD(XMM0, R(XMM1));
-			ADDPD(XMM0, fpr.R(b));
-			break;
-		case 28: //msub
-			MULPD(XMM0, fpr.R(c));
-			SUBPD(XMM0, fpr.R(b));
-			break;
-		case 29: //madd
-			MULPD(XMM0, fpr.R(c));
-			ADDPD(XMM0, fpr.R(b));
-			break;
-		case 30: //nmsub
-			MULPD(XMM0, fpr.R(c));
-			SUBPD(XMM0, fpr.R(b));
-			XORPD(XMM0, M((void*)&psSignBits));
+		}
+		case 29: {//madd
+			op2 = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FC));
+			val = ibuild.EmitFPMul(val, op2);
+			op3 = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FB));
+			val = ibuild.EmitFPAdd(val, op3);
 			break;
-		case 31: //nmadd
-			MULPD(XMM0, fpr.R(c));
-			ADDPD(XMM0, fpr.R(b));
-			XORPD(XMM0, M((void*)&psSignBits));
+		}
+		case 30: {//nmsub
+			op2 = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FC));
+			val = ibuild.EmitFPMul(val, op2);
+			op3 = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FB));
+			val = ibuild.EmitFPSub(val, op3);
+			val = ibuild.EmitFPNeg(val);
 			break;
-		default:
-			_assert_msg_(DYNA_REC, 0, "ps_maddXX WTF!!!");
-			//Default(inst);
-			//fpr.UnlockAll();
-			return;
 		}
-		fpr.LoadToX64(d, false);
-		MOVAPD(fpr.RX(d), Gen::R(XMM0));
-		ForceSinglePrecisionP(fpr.RX(d));
-		fpr.UnlockAll();
+		}
+		val = ibuild.EmitExpandPackedToMReg(val);
+		ibuild.EmitStoreFReg(val, inst.FD);
 	}