And a bit more WIP JIT work, mostly floating-point improvements. It

should be getting to the point where the performance is reasonably
comparable to the current JIT on most workloads. I'm not sure of the 
exact comparisons because I haven't done much in terms of checking the 
performance versus the current JIT on floating-point heavy stuff. The 
floating point stuff is still relatively unoptimized, and there are still a
few relatively common functions that aren't being compiled yet, but the most 
critical work is done.



git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@1807 8ced0084-cf51-0410-be5f-012b33b47a6e

5 files changed, 262 insertions, 188 deletions

diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/IR.cpp b/Source/Core/Core/Src/PowerPC/Jit64IL/IR.cpp
index 791e7f377c..356718cc73 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/IR.cpp
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/IR.cpp
@@ -305,6 +305,10 @@ InstLoc IRBuilder::FoldUOp(unsigned Opcode, InstLoc Op1, unsigned extra) {
 		if (getOpcode(*Op1) == ExpandPackedToMReg)
 			return getOp1(Op1);
 	}
+	if (Opcode == DoubleToSingle) {
+		if (getOpcode(*Op1) == DupSingleToMReg)
+			return getOp1(Op1);
+	}
 
 	return EmitUOp(Opcode, Op1, extra);
 }
@@ -590,6 +594,11 @@ static OpArg fregLocForSlot(RegInfo& RI, unsigned slot) {
 	return M(&FSlotSet[slot*16]);
 }
 
+// Used for accessing the top half of a spilled double
+static OpArg fregLocForSlotPlusFour(RegInfo& RI, unsigned slot) {
+	return M(&FSlotSet[slot*16+4]);
+}
+
 static unsigned fregCreateSpill(RegInfo& RI, InstLoc I) {
 	unsigned newSpill = ++RI.numFSpills;
 	RI.IInfo[I - RI.FirstI] |= newSpill << 16;
@@ -679,6 +688,16 @@ static X64Reg regEnsureInReg(RegInfo& RI, InstLoc I) {
 	return loc.GetSimpleReg();
 }
 
+static X64Reg fregEnsureInReg(RegInfo& RI, InstLoc I) {
+	OpArg loc = fregLocForInst(RI, I);
+	if (!loc.IsSimpleReg()) {
+		X64Reg newReg = fregFindFreeReg(RI);
+		RI.Jit->MOVAPD(newReg, loc);
+		loc = R(newReg);
+	}
+	return loc.GetSimpleReg();
+}
+
 static void regSpillCallerSaved(RegInfo& RI) {
 	regSpill(RI, EDX);
 	regSpill(RI, ECX);
@@ -1044,6 +1063,8 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 		case ExpandPackedToMReg:
 		case CompactMRegToPacked:
 		case FPNeg:
+		case FSNeg:
+		case FDNeg:
 			if (thisUsed)
 				regMarkUse(RI, I, getOp1(I), 1);
 			break;
@@ -1052,6 +1073,7 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 		case Load32:
 			regMarkMemAddress(RI, I, getOp1(I), 1);
 			break;
+		case LoadDouble:
 		case LoadSingle:
 		case LoadPaired:
 			regMarkUse(RI, I, getOp1(I), 1);
@@ -1086,9 +1108,17 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 		case ICmpSgt:
 		case FSMul:
 		case FSAdd:
+		case FSSub:
+		case FDMul:
+		case FDAdd:
+		case FDSub:
 		case FPAdd:
 		case FPMul:
 		case FPSub:
+		case FPMerge00:
+		case FPMerge01:
+		case FPMerge10:
+		case FPMerge11:
 		case InsertDoubleInMReg:
 			if (thisUsed) {
 				regMarkUse(RI, I, getOp1(I), 1);
@@ -1104,6 +1134,7 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 			regMarkMemAddress(RI, I, getOp2(I), 2);
 			break;
 		case StoreSingle:
+		case StoreDouble:
 		case StorePaired:
 			regMarkUse(RI, I, getOp1(I), 1);
 			regMarkUse(RI, I, getOp2(I), 2);
@@ -1417,6 +1448,21 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 			regNormalRegClear(RI, I);
 			break;
 		}
+		case LoadDouble: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOV(32, R(ECX), regLocForInst(RI, getOp1(I)));
+			Jit->ADD(32, R(ECX), Imm8(4));
+			RI.Jit->UnsafeLoadRegToReg(ECX, ECX, 32, 0, false);
+			Jit->MOVD_xmm(reg, R(ECX));
+			Jit->MOV(32, R(ECX), regLocForInst(RI, getOp1(I)));
+			RI.Jit->UnsafeLoadRegToReg(ECX, ECX, 32, 0, false);
+			Jit->MOVD_xmm(XMM0, R(ECX));
+			Jit->PUNPCKLDQ(reg, R(XMM0));
+			RI.fregs[reg] = I;
+			regNormalRegClear(RI, I);
+			break;
+		}
 		case LoadPaired: {
 			if (!thisUsed) break;
 			regSpill(RI, EAX);
@@ -1449,6 +1495,30 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 				regClearInst(RI, getOp2(I));
 			break;
 		}
+		case StoreDouble: {
+			regSpill(RI, EAX);
+			// FIXME: Use 64-bit where possible
+			// FIXME: Use unsafe write with pshufb where possible
+			unsigned fspill = fregGetSpill(RI, getOp1(I));
+			if (!fspill) {
+				// Force the value to spill, so we can use
+				// memory operations to load it
+				fspill = fregCreateSpill(RI, getOp1(I));
+				X64Reg reg = fregLocForInst(RI, getOp1(I)).GetSimpleReg();
+				RI.Jit->MOVAPD(fregLocForSlot(RI, fspill), reg);
+			}
+			Jit->MOV(32, R(EAX), fregLocForSlotPlusFour(RI, fspill));
+			Jit->MOV(32, R(ECX), regLocForInst(RI, getOp2(I)));
+			RI.Jit->SafeWriteRegToReg(EAX, ECX, 32, 0);
+			Jit->MOV(32, R(EAX), fregLocForSlot(RI, fspill));
+			Jit->MOV(32, R(ECX), regLocForInst(RI, getOp2(I)));
+			RI.Jit->SafeWriteRegToReg(EAX, ECX, 32, 4);
+			if (RI.IInfo[I - RI.FirstI] & 4)
+				fregClearInst(RI, getOp1(I));
+			if (RI.IInfo[I - RI.FirstI] & 8)
+				regClearInst(RI, getOp2(I));
+			break;
+		}
 		case StorePaired: {
 			regSpill(RI, EAX);
 			regSpill(RI, EDX);
@@ -1501,6 +1571,28 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 			fregNormalRegClear(RI, I);
 			break;
 		}
+		case FSNeg: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			static const u32 GC_ALIGNED16(ssSignBits[4]) = 
+				{0x80000000};
+			Jit->PXOR(reg, M((void*)&ssSignBits));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
+		case FDNeg: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			static const u64 GC_ALIGNED16(ssSignBits[2]) = 
+				{0x8000000000000000ULL};
+			Jit->PXOR(reg, M((void*)&ssSignBits));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
 		case FPNeg: {
 			if (!thisUsed) break;
 			X64Reg reg = fregFindFreeReg(RI);
@@ -1522,8 +1614,8 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 		}
 		case StoreFReg: {
 			unsigned ppcreg = *I >> 16;
-			Jit->MOVAPD(XMM0, fregLocForInst(RI, getOp1(I)));
-			Jit->MOVAPD(M(&PowerPC::ppcState.ps[ppcreg]), XMM0);
+			Jit->MOVAPD(M(&PowerPC::ppcState.ps[ppcreg]),
+				      fregEnsureInReg(RI, getOp1(I)));
 			fregNormalRegClear(RI, I);
 			break;
 		}
@@ -1553,6 +1645,42 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 			fregNormalRegClear(RI, I);
 			break;
 		}
+		case FSSub: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->SUBSS(reg, fregLocForInst(RI, getOp2(I)));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
+		case FDMul: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->MULSD(reg, fregLocForInst(RI, getOp2(I)));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
+		case FDAdd: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->ADDSD(reg, fregLocForInst(RI, getOp2(I)));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
+		case FDSub: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->SUBSD(reg, fregLocForInst(RI, getOp2(I)));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
 		case FPAdd: {
 			if (!thisUsed) break;
 			X64Reg reg = fregFindFreeReg(RI);
@@ -1580,6 +1708,47 @@ static void DoWriteCode(IRBuilder* ibuild, Jit64* Jit, bool UseProfile) {
 			fregNormalRegClear(RI, I);
 			break;
 		}
+		case FPMerge00: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->PUNPCKLDQ(reg, fregLocForInst(RI, getOp2(I)));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
+		case FPMerge01: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			// Note reversed operands!
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp2(I)));
+			Jit->MOVAPD(XMM0, fregLocForInst(RI, getOp1(I)));
+			Jit->MOVSS(reg, R(XMM0));
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
+		case FPMerge10: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->MOVAPD(XMM0, fregLocForInst(RI, getOp2(I)));
+			Jit->MOVSS(reg, R(XMM0));
+			Jit->SHUFPS(reg, R(reg), 0xF1);
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
+		case FPMerge11: {
+			if (!thisUsed) break;
+			X64Reg reg = fregFindFreeReg(RI);
+			Jit->MOVAPD(reg, fregLocForInst(RI, getOp1(I)));
+			Jit->PUNPCKLDQ(reg, fregLocForInst(RI, getOp2(I)));
+			Jit->SHUFPD(reg, R(reg), 0x1);
+			RI.fregs[reg] = I;
+			fregNormalRegClear(RI, I);
+			break;
+		}
 		case CInt32:
 		case CInt16: {
 			if (!thisUsed) break;
diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/IR.h b/Source/Core/Core/Src/PowerPC/Jit64IL/IR.h
index 4db1100791..92f30d8081 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/IR.h
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/IR.h
@@ -157,13 +157,24 @@ namespace IREmitter {
 		LoadFReg,
 		FSMul,
 		FSAdd,
+		FSSub,
+		FSNeg,
 		FPAdd,
 		FPMul,
 		FPSub,
 		FPNeg,
+		FDMul,
+		FDAdd,
+		FDSub,
+		FDNeg,
+		FPMerge00,
+		FPMerge01,
+		FPMerge10,
+		FPMerge11,
 		FResult_End,
 		StorePaired,
 		StoreSingle,
+		StoreDouble,
 		StoreFReg,
 
 		// "Trinary" operators
@@ -380,6 +391,9 @@ namespace IREmitter {
 		InstLoc EmitStoreSingle(InstLoc value, InstLoc addr) {
 			return FoldBiOp(StoreSingle, value, addr);
 		}
+		InstLoc EmitStoreDouble(InstLoc value, InstLoc addr) {
+			return FoldBiOp(StoreDouble, value, addr);
+		}
 		InstLoc EmitStorePaired(InstLoc value, InstLoc addr, unsigned quantReg) {
 			return FoldBiOp(StorePaired, value, addr, quantReg);
 		}
@@ -410,6 +424,24 @@ namespace IREmitter {
 		InstLoc EmitFSAdd(InstLoc op1, InstLoc op2) {
 			return FoldBiOp(FSAdd, op1, op2);
 		}
+		InstLoc EmitFSSub(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FSSub, op1, op2);
+		}
+		InstLoc EmitFSNeg(InstLoc op1) {
+			return FoldUOp(FSNeg, op1);
+		}
+		InstLoc EmitFDMul(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FDMul, op1, op2);
+		}
+		InstLoc EmitFDAdd(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FDAdd, op1, op2);
+		}
+		InstLoc EmitFDSub(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FDSub, op1, op2);
+		}
+		InstLoc EmitFDNeg(InstLoc op1) {
+			return FoldUOp(FDNeg, op1);
+		}
 		InstLoc EmitFPAdd(InstLoc op1, InstLoc op2) {
 			return FoldBiOp(FPAdd, op1, op2);
 		}
@@ -419,6 +451,18 @@ namespace IREmitter {
 		InstLoc EmitFPSub(InstLoc op1, InstLoc op2) {
 			return FoldBiOp(FPSub, op1, op2);
 		}
+		InstLoc EmitFPMerge00(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FPMerge00, op1, op2);
+		}
+		InstLoc EmitFPMerge01(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FPMerge01, op1, op2);
+		}
+		InstLoc EmitFPMerge10(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FPMerge10, op1, op2);
+		}
+		InstLoc EmitFPMerge11(InstLoc op1, InstLoc op2) {
+			return FoldBiOp(FPMerge11, op1, op2);
+		}
 		InstLoc EmitFPNeg(InstLoc op1) {
 			return FoldUOp(FPNeg, op1);
 		}
diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_FloatingPoint.cpp b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_FloatingPoint.cpp
index f64a5e1556..8837cc86ff 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_FloatingPoint.cpp
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_FloatingPoint.cpp
@@ -28,31 +28,40 @@
 
 	void Jit64::fp_arith_s(UGeckoInstruction inst)
 	{
-		if (inst.Rc || inst.OPCD != 59 || inst.SUBOP5 != 25) {
+		if (inst.Rc || inst.OPCD != 59 || (inst.SUBOP5 != 25 && inst.SUBOP5 != 20 && inst.SUBOP5 != 21)) {
 			Default(inst); return;
 		}
 		IREmitter::InstLoc val = ibuild.EmitLoadFReg(inst.FA);
-		val = ibuild.EmitDoubleToSingle(val);
-		bool dupe = inst.OPCD == 59;
 		switch (inst.SUBOP5)
 		{
 		case 25: //mul
-			val = ibuild.EmitFSMul(val, ibuild.EmitDoubleToSingle(ibuild.EmitLoadFReg(inst.FC)));
+			val = ibuild.EmitFDMul(val, ibuild.EmitLoadFReg(inst.FC));
+			break;
 		case 18: //div
 		case 20: //sub
+			val = ibuild.EmitFDSub(val, ibuild.EmitLoadFReg(inst.FB));
+			break;
 		case 21: //add
+			val = ibuild.EmitFDAdd(val, ibuild.EmitLoadFReg(inst.FB));
+			break;
 		case 23: //sel
 		case 24: //res
 		default:
 			_assert_msg_(DYNA_REC, 0, "fp_arith_s WTF!!!");
 		}
-		val = ibuild.EmitDupSingleToMReg(val);
+
+		if (inst.OPCD == 59) {
+			val = ibuild.EmitDoubleToSingle(val);
+			val = ibuild.EmitDupSingleToMReg(val);
+		} else {
+			val = ibuild.EmitInsertDoubleInMReg(val, ibuild.EmitLoadFReg(inst.FD));
+		}
 		ibuild.EmitStoreFReg(val, inst.FD);
 	}
 
 	void Jit64::fmaddXX(UGeckoInstruction inst)
 	{
-		if (inst.Rc || inst.OPCD != 59 || inst.SUBOP5 != 29) {
+		if (inst.Rc || inst.OPCD != 59) {
 			Default(inst); return;
 		}
 
@@ -61,7 +70,12 @@
 		IREmitter::InstLoc val = ibuild.EmitLoadFReg(inst.FA);
 		val = ibuild.EmitDoubleToSingle(val);
 		val = ibuild.EmitFSMul(val, ibuild.EmitDoubleToSingle(ibuild.EmitLoadFReg(inst.FC)));
-		val = ibuild.EmitFSAdd(val, ibuild.EmitDoubleToSingle(ibuild.EmitLoadFReg(inst.FB)));
+		if (inst.SUBOP5 & 1)
+			val = ibuild.EmitFSAdd(val, ibuild.EmitDoubleToSingle(ibuild.EmitLoadFReg(inst.FB)));
+		else
+			val = ibuild.EmitFSSub(val, ibuild.EmitDoubleToSingle(ibuild.EmitLoadFReg(inst.FB)));
+		if (inst.SUBOP5 & 2)
+			val = ibuild.EmitFSNeg(val);
 		val = ibuild.EmitDupSingleToMReg(val);
 		ibuild.EmitStoreFReg(val, inst.FD);
 	}
@@ -78,57 +92,6 @@
 
 	void Jit64::fcmpx(UGeckoInstruction inst)
 	{
-		if(Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITFloatingPointOff)
-			{Default(inst); return;} // turn off from debugger
-		INSTRUCTION_START;
-		if (jo.fpAccurateFlags)
-		{
-			Default(inst);
-			return;
-		}
-		bool ordered = inst.SUBOP10 == 32;
-	/*
-	double fa =	rPS0(_inst.FA);
-	double fb =	rPS0(_inst.FB);
-	u32 compareResult;
-
-	if(IsNAN(fa) ||	IsNAN(fb))	compareResult = 1; 
-	else if(fa < fb)			compareResult = 8;	
-	else if(fa > fb)			compareResult = 4;
-	else						compareResult = 2;
-
-	FPSCR.FPRF = compareResult;
-	CR = (CR & (~(0xf0000000 >> (_inst.CRFD * 4)))) | (compareResult <<	((7	- _inst.CRFD) *	4));
-*/
-		int a = inst.FA;
-		int b = inst.FB;
-		int crf = inst.CRFD;
-		int shift = crf * 4;
-		//FPSCR
-		//XOR(32,R(EAX),R(EAX));
-
-		fpr.Lock(a,b);
-		if (a != b)
-			fpr.LoadToX64(a, true);
-
-		// USES_CR
-		if (ordered)
-			COMISD(fpr.R(a).GetSimpleReg(), fpr.R(b));
-		else
-			UCOMISD(fpr.R(a).GetSimpleReg(), fpr.R(b));
-		FixupBranch pLesser  = J_CC(CC_B);
-		FixupBranch pGreater = J_CC(CC_A);
-		// _x86Reg == 0
-		MOV(8, M(&PowerPC::ppcState.cr_fast[crf]), Imm8(0x2));
-		FixupBranch continue1 = J();
-		// _x86Reg > 0
-		SetJumpTarget(pGreater);
-		MOV(8, M(&PowerPC::ppcState.cr_fast[crf]), Imm8(0x4));
-		FixupBranch continue2 = J();
-		// _x86Reg < 0
-		SetJumpTarget(pLesser);
-		MOV(8, M(&PowerPC::ppcState.cr_fast[crf]), Imm8(0x8));
-		SetJumpTarget(continue1);
-		SetJumpTarget(continue2);
-		fpr.UnlockAll();
+		Default(inst);
+		return;
 	}
diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_LoadStoreFloating.cpp b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_LoadStoreFloating.cpp
index fca175b33c..0c13c4e05d 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_LoadStoreFloating.cpp
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_LoadStoreFloating.cpp
@@ -65,121 +65,26 @@ void Jit64::lfs(UGeckoInstruction inst)
 
 void Jit64::lfd(UGeckoInstruction inst)
 {
-	if (Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITLoadStoreFloatingOff)
-		{Default(inst); return;} // turn off from debugger	
-	INSTRUCTION_START;
-
-	int d = inst.RD;
-	int a = inst.RA;
-	if (!a) 
-	{
-		Default(inst);
-		return;
-	}
-	s32 offset = (s32)(s16)inst.SIMM_16;
-	gpr.FlushLockX(ABI_PARAM1);
-	gpr.Lock(a);
-	MOV(32, R(ABI_PARAM1), gpr.R(a));
-	// TODO - optimize. This has to load the previous value - upper double should stay unmodified.
-	fpr.LoadToX64(d, true);
-	fpr.Lock(d);
-	X64Reg xd = fpr.RX(d);
-	if (cpu_info.bSSSE3) {
-#ifdef _M_X64
-		MOVQ_xmm(XMM0, MComplex(RBX, ABI_PARAM1, SCALE_1, offset));
-#else
-		AND(32, R(ABI_PARAM1), Imm32(Memory::MEMVIEW32_MASK));
-		MOVQ_xmm(XMM0, MDisp(ABI_PARAM1, (u32)Memory::base + offset));
-#endif
-		PSHUFB(XMM0, M((void *)bswapShuffle1x8Dupe));
-		MOVSD(xd, R(XMM0));
-	} else {
-#ifdef _M_X64
-		MOV(64, R(EAX), MComplex(RBX, ABI_PARAM1, SCALE_1, offset));
-		BSWAP(64, EAX);
-		MOV(64, M(&temp64), R(EAX));
-		MOVSD(XMM0, M(&temp64));
-		MOVSD(xd, R(XMM0));
-#else
-		AND(32, R(ABI_PARAM1), Imm32(Memory::MEMVIEW32_MASK));
-		MOV(32, R(EAX), MDisp(ABI_PARAM1, (u32)Memory::base + offset));
-		BSWAP(32, EAX);
-		MOV(32, M((void*)((u32)&temp64+4)), R(EAX));
-		MOV(32, R(EAX), MDisp(ABI_PARAM1, (u32)Memory::base + offset + 4));
-		BSWAP(32, EAX);
-		MOV(32, M(&temp64), R(EAX));
-		MOVSD(XMM0, M(&temp64));
-		MOVSD(xd, R(XMM0));
-#if 0
-		// Alternate implementation; possibly faster
-		AND(32, R(ABI_PARAM1), Imm32(Memory::MEMVIEW32_MASK));
-		MOVQ_xmm(XMM0, MDisp(ABI_PARAM1, (u32)Memory::base + offset));
-		PSHUFLW(XMM0, R(XMM0), 0x1B);
-		PSRLW(XMM0, 8);
-		MOVSD(xd, R(XMM0));
-		MOVQ_xmm(XMM0, MDisp(ABI_PARAM1, (u32)Memory::base + offset));
-		PSHUFLW(XMM0, R(XMM0), 0x1B);
-		PSLLW(XMM0, 8);
-		POR(xd, R(XMM0));
-#endif
-#endif
-	}
-	gpr.UnlockAll();
-	gpr.UnlockAllX();
-	fpr.UnlockAll();
+	IREmitter::InstLoc addr = ibuild.EmitIntConst(inst.SIMM_16), val;
+	if (inst.RA)
+		addr = ibuild.EmitAdd(addr, ibuild.EmitLoadGReg(inst.RA));
+	val = ibuild.EmitLoadFReg(inst.RD);
+	val = ibuild.EmitInsertDoubleInMReg(ibuild.EmitLoadDouble(addr), val);
+	ibuild.EmitStoreFReg(val, inst.RD);
+	return;
 }
 
 
 void Jit64::stfd(UGeckoInstruction inst)
 {
-	if (Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITLoadStoreFloatingOff)
-		{Default(inst); return;} // turn off from debugger	
-	INSTRUCTION_START;
-
-	int s = inst.RS;
-	int a = inst.RA;
-	if (!a)
-	{
-		Default(inst);
-		return;
-	}
-	s32 offset = (s32)(s16)inst.SIMM_16;
-	gpr.FlushLockX(ABI_PARAM1);
-	gpr.Lock(a);
-	fpr.Lock(s);
-	MOV(32, R(ABI_PARAM1), gpr.R(a));
-#ifdef _M_IX86
-	AND(32, R(ABI_PARAM1), Imm32(Memory::MEMVIEW32_MASK));
-#endif
-	if (cpu_info.bSSSE3) {
-		MOVAPD(XMM0, fpr.R(s));
-		PSHUFB(XMM0, M((void *)bswapShuffle1x8));
-#ifdef _M_X64
-		MOVQ_xmm(MComplex(RBX, ABI_PARAM1, SCALE_1, offset), XMM0);
-#else
-		MOVQ_xmm(MDisp(ABI_PARAM1, (u32)Memory::base + offset), XMM0);
-#endif
-	} else {
-#ifdef _M_X64
-		fpr.LoadToX64(s, true, false);
-		MOVSD(M(&temp64), fpr.RX(s));
-		MOV(64, R(EAX), M(&temp64));
-		BSWAP(64, EAX);
-		MOV(64, MComplex(RBX, ABI_PARAM1, SCALE_1, offset), R(EAX));
-#else
-		fpr.LoadToX64(s, true, false);
-		MOVSD(M(&temp64), fpr.RX(s));
-		MOV(32, R(EAX), M(&temp64));
-		BSWAP(32, EAX);
-		MOV(32, MDisp(ABI_PARAM1, (u32)Memory::base + offset + 4), R(EAX));
-		MOV(32, R(EAX), M((void*)((u32)&temp64 + 4)));
-		BSWAP(32, EAX);
-		MOV(32, MDisp(ABI_PARAM1, (u32)Memory::base + offset), R(EAX));
-#endif
-	}
-	gpr.UnlockAll();
-	gpr.UnlockAllX();
-	fpr.UnlockAll();
+	IREmitter::InstLoc addr = ibuild.EmitIntConst(inst.SIMM_16),
+			   val  = ibuild.EmitLoadFReg(inst.RS);
+	if (inst.RA)
+		addr = ibuild.EmitAdd(addr, ibuild.EmitLoadGReg(inst.RA));
+	if (inst.OPCD & 1)
+		ibuild.EmitStoreGReg(addr, inst.RA);
+	ibuild.EmitStoreDouble(val, addr);
+	return;
 }
 
 
diff --git a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Paired.cpp b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Paired.cpp
index b0b2a3cf58..b34a14c2d2 100644
--- a/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Paired.cpp
+++ b/Source/Core/Core/Src/PowerPC/Jit64IL/Jit_Paired.cpp
@@ -270,42 +270,35 @@
 	//TODO: find easy cases and optimize them, do a breakout like ps_arith
 	void Jit64::ps_mergeXX(UGeckoInstruction inst)
 	{
-		if(Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITPairedOff)
-			{Default(inst); return;} // turn off from debugger
-		INSTRUCTION_START;
 		if (inst.Rc) {
 			Default(inst); return;
 		}
-		int d = inst.FD;
-		int a = inst.FA;
-		int b = inst.FB;
-		fpr.Lock(a,b,d);
 
-		MOVAPD(XMM0, fpr.R(a));
+		IREmitter::InstLoc val = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FA)),
+				   rhs = ibuild.EmitCompactMRegToPacked(ibuild.EmitLoadFReg(inst.FB));
+
 		switch (inst.SUBOP10)
 		{
 		case 528: 
-			UNPCKLPD(XMM0, fpr.R(b)); //unpck is faster than shuf
+			val = ibuild.EmitFPMerge00(val, rhs);
 			break; //00
 		case 560:
-			SHUFPD(XMM0, fpr.R(b), 2); //must use shuf here
+			val = ibuild.EmitFPMerge01(val, rhs);
 			break; //01
 		case 592:
-			SHUFPD(XMM0, fpr.R(b), 1);
+			val = ibuild.EmitFPMerge10(val, rhs);
 			break; //10
 		case 624:
-			UNPCKHPD(XMM0, fpr.R(b));
+			val = ibuild.EmitFPMerge11(val, rhs);
 			break; //11
 		default:
 			_assert_msg_(DYNA_REC, 0, "ps_merge - invalid op");
 		}
-		fpr.LoadToX64(d, false);
-		MOVAPD(fpr.RX(d), Gen::R(XMM0));
-		fpr.UnlockAll();
+		val = ibuild.EmitExpandPackedToMReg(val);
+		ibuild.EmitStoreFReg(val, inst.FD);
 	}
 
 
-	//TODO: add optimized cases
 	void Jit64::ps_maddXX(UGeckoInstruction inst)
 	{
 		if (inst.Rc || (inst.SUBOP5 != 28 && inst.SUBOP5 != 29 && inst.SUBOP5 != 30)) {