Merge pull request #2186 from Sonicadvance1/aarch64_optimize_paired_slowmem

[AArch64] Optimize slowmem paired stores.

Source/Core/Common/Arm64Emitter.cpp

@@ -2618,6 +2618,7 @@ void ARM64FloatEmitter::SMOV(u8 size, ARM64Reg Rd, ARM64Reg Rn, u8 index)
 	bool b64Bit = Is64Bit(Rd);
 	_assert_msg_(DYNA_REC, Rd < SP, "%s destination must be a GPR!", __FUNCTION__);
 	_assert_msg_(DYNA_REC, size != 64, "%s doesn't support 64bit destination. Use UMOV!", __FUNCTION__);
+	_assert_msg_(DYNA_REC, !b64Bit && size != 32, "%s doesn't support 32bit move to 32bit register. Use UMOV!", __FUNCTION__);
 	u32 imm5 = 0;
 
 	if (size == 8)

Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStorePaired.cpp

@@ -94,9 +94,18 @@ void JitArm64::psq_st(UGeckoInstruction inst)
 	fpr.Lock(Q0, Q1);
 
 	ARM64Reg arm_addr = gpr.R(inst.RA);
+	ARM64Reg VS = fpr.R(inst.RS);
+
 	ARM64Reg scale_reg = W0;
 	ARM64Reg addr_reg = W1;
-	ARM64Reg type_reg = gpr.GetReg();
+	ARM64Reg type_reg = W2;
+
+	BitSet32 gprs_in_use = gpr.GetCallerSavedUsed();
+	BitSet32 fprs_in_use = fpr.GetCallerSavedUsed();
+
+	// Wipe the registers we are using as temporaries
+	gprs_in_use &= BitSet32(~0x40000007);
+	fprs_in_use &= BitSet32(~3);
 
 	LDR(INDEX_UNSIGNED, scale_reg, X29, PPCSTATE_OFF(spr[SPR_GQR0 + inst.I]));
 
@@ -118,13 +127,35 @@ void JitArm64::psq_st(UGeckoInstruction inst)
 	if (update)
 		MOV(arm_addr, addr_reg);
 
-	ARM64Reg VS = fpr.R(inst.RS);
 	m_float_emit.FCVTN(32, D0, VS);
-	MOVI2R(X30, (u64)&asm_routines.pairedStoreQuantized[inst.W * 8]);
-	LDR(X30, X30, ArithOption(EncodeRegTo64(type_reg), true));
-	BLR(X30);
 
-	gpr.Unlock(W0, W1, W2, W30, type_reg);
+	// Inline address check
+	{
+		TST(addr_reg, 6, 1);
+		FixupBranch argh = B(CC_NEQ);
+
+		// Fast
+		MOVI2R(X30, (u64)&asm_routines.pairedStoreQuantized[inst.W * 8]);
+		LDR(EncodeRegTo64(type_reg), X30, ArithOption(EncodeRegTo64(type_reg), true));
+		BLR(EncodeRegTo64(type_reg));
+
+		FixupBranch continue1 = B();
+		SetJumpTarget(argh);
+
+		// Slow
+		MOVI2R(X30, (u64)&asm_routines.pairedStoreQuantized[16 + inst.W * 8]);
+		LDR(EncodeRegTo64(type_reg), X30, ArithOption(EncodeRegTo64(type_reg), true));
+
+		ABI_PushRegisters(gprs_in_use);
+		m_float_emit.ABI_PushRegisters(fprs_in_use, X30);
+		BLR(EncodeRegTo64(type_reg));
+		m_float_emit.ABI_PopRegisters(fprs_in_use, X30);
+		ABI_PushRegisters(gprs_in_use);
+
+		SetJumpTarget(continue1);
+	}
+
+	gpr.Unlock(W0, W1, W2, W30);
 	fpr.Unlock(Q0, Q1);
 }
 

Source/Core/Core/PowerPC/JitArm64/JitAsm.cpp

@@ -107,7 +107,6 @@ void JitArm64AsmRoutineManager::GenerateCommon()
 	ARM64Reg addr_reg = X1;
 	ARM64Reg scale_reg = X0;
 	ARM64FloatEmitter float_emit(this);
-	const u32 GPR_CALLER_SAVE = 0x6007FFFF;
 
 	const u8* loadPairedIllegal = GetCodePtr();
 		BRK(100);
@@ -263,36 +262,27 @@ void JitArm64AsmRoutineManager::GenerateCommon()
 	// Stores
 	const u8* storePairedIllegal = GetCodePtr();
 		BRK(0x101);
-	const u8* storePairedFloat = GetCodePtr();
+	const u8* storePairedFloat;
+	const u8* storePairedFloatSlow;
 	{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
+		storePairedFloat = GetCodePtr();
-		BitSet32 fprs(~3); // All except Q0/Q1
-
-		TST(DecodeReg(addr_reg), 6, 1);
-		FixupBranch argh = B(CC_NEQ);
-
 		float_emit.REV32(8, D0, D0);
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.ST1(64, Q0, 0, addr_reg, SP);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storePairedFloatSlow = GetCodePtr();
-
-		ABI_PushRegisters(gprs);
-		float_emit.ABI_PushRegisters(fprs, X3);
 		float_emit.UMOV(64, X0, Q0, 0);
 		ORR(X0, SP, X0, ArithOption(X0, ST_ROR, 32));
-		MOVI2R(X30, (u64)PowerPC::Write_U64);
+		MOVI2R(X2, (u64)PowerPC::Write_U64);
-		BLR(X30);
+		BR(X2);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
-	const u8* storePairedU8 = GetCodePtr();
-	{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
-		BitSet32 fprs(~3); // All except Q0/Q1
 
+	const u8* storePairedU8;
+	const u8* storePairedU8Slow;
+	{
+		auto emit_quantize = [this, &float_emit, scale_reg]()
+		{
 			MOVI2R(X2, (u64)&m_quantizeTableS);
 			ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
 			float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
@@ -300,30 +290,26 @@ void JitArm64AsmRoutineManager::GenerateCommon()
 			float_emit.FCVTZU(32, D0, D0);
 			float_emit.XTN(16, D0, D0);
 			float_emit.XTN(8, D0, D0);
+		};
 
-		TST(DecodeReg(addr_reg), 6, 1);
+		storePairedU8 = GetCodePtr();
-		FixupBranch argh = B(CC_NEQ);
+		emit_quantize();
-
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.ST1(16, Q0, 0, addr_reg, SP);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storePairedU8Slow = GetCodePtr();
-		ABI_PushRegisters(gprs);
+		emit_quantize();
-		float_emit.ABI_PushRegisters(fprs, X3);
 		float_emit.UMOV(16, W0, Q0, 0);
 		REV16(W0, W0);
-		MOVI2R(X30, (u64)PowerPC::Write_U16);
+		MOVI2R(X2, (u64)PowerPC::Write_U16);
-		BLR(X30);
+		BR(X2);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
-	const u8* storePairedS8 = GetCodePtr();
+	const u8* storePairedS8;
+	const u8* storePairedS8Slow;
+	{
+		auto emit_quantize = [this, &float_emit, scale_reg]()
 		{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
-		BitSet32 fprs(~3); // All except Q0/Q1
-
 			MOVI2R(X2, (u64)&m_quantizeTableS);
 			ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
 			float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
@@ -331,31 +317,27 @@ void JitArm64AsmRoutineManager::GenerateCommon()
 			float_emit.FCVTZS(32, D0, D0);
 			float_emit.XTN(16, D0, D0);
 			float_emit.XTN(8, D0, D0);
+		};
 
-		TST(DecodeReg(addr_reg), 6, 1);
+		storePairedS8 = GetCodePtr();
-		FixupBranch argh = B(CC_NEQ);
+		emit_quantize();
-
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.ST1(16, Q0, 0, addr_reg, SP);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storePairedS8Slow = GetCodePtr();
-		ABI_PushRegisters(gprs);
+		emit_quantize();
-		float_emit.ABI_PushRegisters(fprs, X3);
 		float_emit.UMOV(16, W0, Q0, 0);
 		REV16(W0, W0);
-		MOVI2R(X30, (u64)PowerPC::Write_U16);
+		MOVI2R(X2, (u64)PowerPC::Write_U16);
-		BLR(X30);
+		BR(X2);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
 
-	const u8* storePairedU16 = GetCodePtr();
+	const u8* storePairedU16;
+	const u8* storePairedU16Slow;
+	{
+		auto emit_quantize = [this, &float_emit, scale_reg]()
 		{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
-		BitSet32 fprs(~3); // All except Q0/Q1
-
 			MOVI2R(X2, (u64)&m_quantizeTableS);
 			ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
 			float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
@@ -363,29 +345,26 @@ void JitArm64AsmRoutineManager::GenerateCommon()
 			float_emit.FCVTZU(32, D0, D0);
 			float_emit.XTN(16, D0, D0);
 			float_emit.REV16(8, D0, D0);
+		};
 
-		TST(DecodeReg(addr_reg), 6, 1);
+		storePairedU16 = GetCodePtr();
-		FixupBranch argh = B(CC_NEQ);
+		emit_quantize();
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.ST1(32, Q0, 0, addr_reg, SP);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storePairedU16Slow = GetCodePtr();
-		ABI_PushRegisters(gprs);
+		emit_quantize();
-		float_emit.ABI_PushRegisters(fprs, X3);
 		float_emit.REV32(8, D0, D0);
 		float_emit.UMOV(32, W0, Q0, 0);
-		MOVI2R(X30, (u64)PowerPC::Write_U32);
+		MOVI2R(X2, (u64)PowerPC::Write_U32);
-		BLR(X30);
+		BR(X2);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
-	const u8* storePairedS16 = GetCodePtr(); // Used by Viewtiful Joe's intro movie
+	const u8* storePairedS16; // Used by Viewtiful Joe's intro movie
+	const u8* storePairedS16Slow;
+	{
+		auto emit_quantize = [this, &float_emit, scale_reg]()
 		{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
-		BitSet32 fprs(~3); // All except Q0/Q1
-
 			MOVI2R(X2, (u64)&m_quantizeTableS);
 			ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
 			float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
@@ -393,54 +372,41 @@ void JitArm64AsmRoutineManager::GenerateCommon()
 			float_emit.FCVTZS(32, D0, D0);
 			float_emit.XTN(16, D0, D0);
 			float_emit.REV16(8, D0, D0);
+		};
 
-		TST(DecodeReg(addr_reg), 6, 1);
+		storePairedS16 = GetCodePtr();
-		FixupBranch argh = B(CC_NEQ);
+		emit_quantize();
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.ST1(32, Q0, 0, addr_reg, SP);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storePairedS16Slow = GetCodePtr();
-		ABI_PushRegisters(gprs);
+		emit_quantize();
-		float_emit.ABI_PushRegisters(fprs, X3);
 		float_emit.REV32(8, D0, D0);
 		float_emit.UMOV(32, W0, Q0, 0);
-		MOVI2R(X30, (u64)PowerPC::Write_U32);
+		MOVI2R(X2, (u64)PowerPC::Write_U32);
-		BLR(X30);
+		BR(X2);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
 
-	const u8* storeSingleFloat = GetCodePtr();
+	const u8* storeSingleFloat;
+	const u8* storeSingleFloatSlow;
 	{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
+		storeSingleFloat = GetCodePtr();
-		BitSet32 fprs(~3); // All except Q0/Q1
-
-		TST(DecodeReg(addr_reg), 6, 1);
-		FixupBranch argh = B(CC_NEQ);
-
 		float_emit.REV32(8, D0, D0);
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.STR(32, INDEX_UNSIGNED, D0, addr_reg, 0);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storeSingleFloatSlow = GetCodePtr();
-
-		ABI_PushRegisters(gprs);
-		float_emit.ABI_PushRegisters(fprs, X3);
 		float_emit.UMOV(32, W0, Q0, 0);
-		MOVI2R(X30, (u64)&PowerPC::Write_U32);
+		MOVI2R(X2, (u64)&PowerPC::Write_U32);
-		BLR(X30);
+		BR(X2);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
-	const u8* storeSingleU8 = GetCodePtr();  // Used by MKWii
+	const u8* storeSingleU8;  // Used by MKWii
+	const u8* storeSingleU8Slow;
+	{
+		auto emit_quantize = [this, &float_emit, scale_reg]()
 		{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
-		BitSet32 fprs(~3); // All except Q0/Q1
-
 			MOVI2R(X2, (u64)&m_quantizeTableS);
 			ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
 			float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
@@ -448,28 +414,25 @@ void JitArm64AsmRoutineManager::GenerateCommon()
 			float_emit.FCVTZU(32, D0, D0);
 			float_emit.XTN(16, D0, D0);
 			float_emit.XTN(8, D0, D0);
+		};
 
-		TST(DecodeReg(addr_reg), 6, 1);
+		storeSingleU8 = GetCodePtr();
-		FixupBranch argh = B(CC_NEQ);
+		emit_quantize();
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.ST1(8, Q0, 0, addr_reg);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storeSingleU8Slow = GetCodePtr();
-		ABI_PushRegisters(gprs);
+		emit_quantize();
-		float_emit.ABI_PushRegisters(fprs, X3);
+		float_emit.UMOV(8, W0, Q0, 0);
-		float_emit.UMOV(32, W0, Q0, 0);
+		MOVI2R(X2, (u64)&PowerPC::Write_U8);
-		MOVI2R(X30, (u64)&PowerPC::Write_U8);
+		BR(X2);
-		BLR(X30);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
-	const u8* storeSingleS8 = GetCodePtr();
+	const u8* storeSingleS8;
+	const u8* storeSingleS8Slow;
+	{
+		auto emit_quantize = [this, &float_emit, scale_reg]()
 		{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
-		BitSet32 fprs(~3); // All except Q0/Q1
-
 			MOVI2R(X2, (u64)&m_quantizeTableS);
 			ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
 			float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
@@ -477,85 +440,77 @@ void JitArm64AsmRoutineManager::GenerateCommon()
 			float_emit.FCVTZS(32, D0, D0);
 			float_emit.XTN(16, D0, D0);
 			float_emit.XTN(8, D0, D0);
+		};
 
-		TST(DecodeReg(addr_reg), 6, 1);
+		storeSingleS8 = GetCodePtr();
-		FixupBranch argh = B(CC_NEQ);
+		emit_quantize();
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.ST1(8, Q0, 0, addr_reg);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storeSingleS8Slow = GetCodePtr();
-		ABI_PushRegisters(gprs);
+		emit_quantize();
-		float_emit.ABI_PushRegisters(fprs, X3);
+		float_emit.SMOV(8, W0, Q0, 0);
-		float_emit.SMOV(32, W0, Q0, 0);
+		MOVI2R(X2, (u64)&PowerPC::Write_U8);
-		MOVI2R(X30, (u64)&PowerPC::Write_U8);
+		BR(X2);
-		BLR(X30);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
-	const u8* storeSingleU16 = GetCodePtr();  // Used by MKWii
+	const u8* storeSingleU16;  // Used by MKWii
+	const u8* storeSingleU16Slow;
+	{
+		auto emit_quantize = [this, &float_emit, scale_reg]()
 		{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
-		BitSet32 fprs(~3); // All except Q0/Q1
-
 			MOVI2R(X2, (u64)&m_quantizeTableS);
 			ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
 			float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
 			float_emit.FMUL(32, D0, D0, D1);
 			float_emit.FCVTZU(32, D0, D0);
 			float_emit.XTN(16, D0, D0);
+		};
 
-		TST(DecodeReg(addr_reg), 6, 1);
+		storeSingleU16 = GetCodePtr();
-		FixupBranch argh = B(CC_NEQ);
+		emit_quantize();
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.REV16(8, D0, D0);
 		float_emit.ST1(16, Q0, 0, addr_reg);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storeSingleU16Slow = GetCodePtr();
-		ABI_PushRegisters(gprs);
+		emit_quantize();
-		float_emit.ABI_PushRegisters(fprs, X3);
+		float_emit.UMOV(16, W0, Q0, 0);
-		float_emit.UMOV(32, W0, Q0, 0);
+		MOVI2R(X2, (u64)&PowerPC::Write_U16);
-		MOVI2R(X30, (u64)&PowerPC::Write_U16);
+		BR(X2);
-		BLR(X30);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
-	const u8* storeSingleS16 = GetCodePtr();
+	const u8* storeSingleS16;
+	const u8* storeSingleS16Slow;
+	{
+		auto emit_quantize = [this, &float_emit, scale_reg]()
 		{
-		BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
-		BitSet32 fprs(~3); // All except Q0/Q1
-
 			MOVI2R(X2, (u64)&m_quantizeTableS);
 			ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
 			float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
 			float_emit.FMUL(32, D0, D0, D1);
 			float_emit.FCVTZS(32, D0, D0);
 			float_emit.XTN(16, D0, D0);
+		};
 
-		TST(DecodeReg(addr_reg), 6, 1);
+		storeSingleS16 = GetCodePtr();
-		FixupBranch argh = B(CC_NEQ);
+		emit_quantize();
 		MOVK(addr_reg, ((u64)Memory::logical_base >> 32) & 0xFFFF, SHIFT_32);
 		float_emit.REV16(8, D0, D0);
 		float_emit.ST1(16, Q0, 0, addr_reg);
 		RET(X30);
 
-		SetJumpTarget(argh);
+		storeSingleS16Slow = GetCodePtr();
-		ABI_PushRegisters(gprs);
+		emit_quantize();
-		float_emit.ABI_PushRegisters(fprs, X3);
+		float_emit.SMOV(16, W0, Q0, 0);
-		float_emit.SMOV(32, W0, Q0, 0);
+		MOVI2R(X2, (u64)&PowerPC::Write_U16);
-		MOVI2R(X30, (u64)&PowerPC::Write_U16);
+		BR(X2);
-		BLR(X30);
-		float_emit.ABI_PopRegisters(fprs, X3);
-		ABI_PopRegisters(gprs);
-		RET(X30);
 	}
 
 	pairedStoreQuantized = reinterpret_cast<const u8**>(const_cast<u8*>(AlignCode16()));
-	ReserveCodeSpace(16 * sizeof(u8*));
+	ReserveCodeSpace(32 * sizeof(u8*));
 
+	// Fast
 	pairedStoreQuantized[0] = storePairedFloat;
 	pairedStoreQuantized[1] = storePairedIllegal;
 	pairedStoreQuantized[2] = storePairedIllegal;
@@ -573,4 +528,24 @@ void JitArm64AsmRoutineManager::GenerateCommon()
 	pairedStoreQuantized[13] = storeSingleU16;
 	pairedStoreQuantized[14] = storeSingleS8;
 	pairedStoreQuantized[15] = storeSingleS16;
+
+	// Slow
+	pairedStoreQuantized[16] = storePairedFloatSlow;
+	pairedStoreQuantized[17] = storePairedIllegal;
+	pairedStoreQuantized[18] = storePairedIllegal;
+	pairedStoreQuantized[19] = storePairedIllegal;
+	pairedStoreQuantized[20] = storePairedU8Slow;
+	pairedStoreQuantized[21] = storePairedU16Slow;
+	pairedStoreQuantized[22] = storePairedS8Slow;
+	pairedStoreQuantized[23] = storePairedS16Slow;
+
+	pairedStoreQuantized[24] = storeSingleFloatSlow;
+	pairedStoreQuantized[25] = storePairedIllegal;
+	pairedStoreQuantized[26] = storePairedIllegal;
+	pairedStoreQuantized[27] = storePairedIllegal;
+	pairedStoreQuantized[28] = storeSingleU8Slow;
+	pairedStoreQuantized[29] = storeSingleU16Slow;
+	pairedStoreQuantized[30] = storeSingleS8Slow;
+	pairedStoreQuantized[31] = storeSingleS16Slow;
+
 }