diff --git a/GPU/Software/DrawPixel.cpp b/GPU/Software/DrawPixel.cpp
index f28e908e45..3caa911067 100644
--- a/GPU/Software/DrawPixel.cpp
+++ b/GPU/Software/DrawPixel.cpp
@@ -155,7 +155,7 @@ static inline void SetPixelColor(GEBufferFormat fmt, int fbStride, int x, int y,
 static inline bool AlphaTestPassed(const PixelFuncID &pixelID, int alpha) {
 	const u8 ref = pixelID.alphaTestRef;
 	if (pixelID.hasAlphaTestMask)
-		alpha &= gstate.getAlphaTestMask();
+		alpha &= pixelID.cached.alphaTestMask;
 
 	switch (pixelID.AlphaTestFunc()) {
 	case GE_COMP_NEVER:
@@ -209,7 +209,7 @@ static inline bool ColorTestPassed(const Vec3<int> &color) {
 
 static inline bool StencilTestPassed(const PixelFuncID &pixelID, u8 stencil) {
 	if (pixelID.hasStencilTestMask)
-		stencil &= gstate.getStencilTestMask();
+		stencil &= pixelID.cached.stencilTestMask;
 	u8 ref = pixelID.stencilTestRef;
 	switch (pixelID.StencilTestFunc()) {
 	case GE_COMP_NEVER:
@@ -239,7 +239,7 @@ static inline bool StencilTestPassed(const PixelFuncID &pixelID, u8 stencil) {
 	return true;
 }
 
-static inline u8 ApplyStencilOp(GEBufferFormat fmt, GEStencilOp op, u8 old_stencil) {
+static inline u8 ApplyStencilOp(GEBufferFormat fmt, uint8_t stencilReplace, GEStencilOp op, u8 old_stencil) {
 	switch (op) {
 	case GE_STENCILOP_KEEP:
 		return old_stencil;
@@ -248,7 +248,7 @@ static inline u8 ApplyStencilOp(GEBufferFormat fmt, GEStencilOp op, u8 old_stenc
 		return 0;
 
 	case GE_STENCILOP_REPLACE:
-		return gstate.getStencilTestRef();
+		return stencilReplace;
 
 	case GE_STENCILOP_INVERT:
 		return ~old_stencil;
@@ -427,20 +427,21 @@ void SOFTRAST_CALL DrawSinglePixel(int x, int y, int z, int fog, Vec4IntArg colo
 		if (pixelID.DepthClear())
 			SetPixelDepth(x, y, pixelID.cached.depthbufStride, z);
 	} else if (pixelID.stencilTest) {
+		const uint8_t stencilReplace = pixelID.hasStencilTestMask ? pixelID.cached.stencilRef : pixelID.stencilTestRef;
 		if (!StencilTestPassed(pixelID, stencil)) {
-			stencil = ApplyStencilOp(fbFormat, pixelID.SFail(), stencil);
+			stencil = ApplyStencilOp(fbFormat, stencilReplace, pixelID.SFail(), stencil);
 			SetPixelStencil(fbFormat, pixelID.cached.framebufStride, targetWriteMask, x, y, stencil);
 			return;
 		}
 
 		// Also apply depth at the same time.  If disabled, same as passing.
 		if (pixelID.DepthTestFunc() != GE_COMP_ALWAYS && !DepthTestPassed(pixelID.DepthTestFunc(), x, y, pixelID.cached.depthbufStride, z)) {
-			stencil = ApplyStencilOp(fbFormat, pixelID.ZFail(), stencil);
+			stencil = ApplyStencilOp(fbFormat, stencilReplace, pixelID.ZFail(), stencil);
 			SetPixelStencil(fbFormat, pixelID.cached.framebufStride, targetWriteMask, x, y, stencil);
 			return;
 		}
 
-		stencil = ApplyStencilOp(fbFormat, pixelID.ZPass(), stencil);
+		stencil = ApplyStencilOp(fbFormat, stencilReplace, pixelID.ZPass(), stencil);
 	} else {
 		if (pixelID.DepthTestFunc() != GE_COMP_ALWAYS && !DepthTestPassed(pixelID.DepthTestFunc(), x, y, pixelID.cached.depthbufStride, z)) {
 			return;
diff --git a/GPU/Software/DrawPixelX86.cpp b/GPU/Software/DrawPixelX86.cpp
index fd313cb1ee..7a0d4fbc7c 100644
--- a/GPU/Software/DrawPixelX86.cpp
+++ b/GPU/Software/DrawPixelX86.cpp
@@ -425,14 +425,13 @@ bool PixelJitCache::Jit_AlphaTest(const PixelFuncID &id) {
 	}
 
 	if (id.hasAlphaTestMask) {
-		// Unfortunate, we'll need gstate to load the mask.
+		// Unfortunate, we'll need pixelID to load the mask.
 		// Note: we leave the ALPHA purpose untouched and free it, because later code may reuse.
-		X64Reg gstateReg = GetGState();
+		X64Reg idReg = GetPixelID();
 		X64Reg maskedReg = regCache_.Alloc(RegCache::GEN_TEMP0);
 
-		// The mask is >> 16, so we load + 2.
-		MOVZX(32, 8, maskedReg, MDisp(gstateReg, offsetof(GPUgstate, alphatest) + 2));
-		regCache_.Unlock(gstateReg, RegCache::GEN_GSTATE);
+		MOVZX(32, 8, maskedReg, MDisp(idReg, offsetof(PixelFuncID, cached.alphaTestMask)));
+		UnlockPixelID(idReg);
 		AND(32, R(maskedReg), R(alphaReg));
 		regCache_.Unlock(alphaReg, RegCache::GEN_SRC_ALPHA);
 
@@ -642,11 +641,11 @@ bool PixelJitCache::Jit_StencilAndDepthTest(const PixelFuncID &id) {
 	Describe("StencilAndDepth");
 	X64Reg maskedReg = stencilReg;
 	if (id.hasStencilTestMask) {
-		X64Reg gstateReg = GetGState();
+		X64Reg idReg = GetPixelID();
 		maskedReg = regCache_.Alloc(RegCache::GEN_TEMP0);
 		MOV(32, R(maskedReg), R(stencilReg));
-		AND(8, R(maskedReg), MDisp(gstateReg, offsetof(GPUgstate, stenciltest) + 2));
-		regCache_.Unlock(gstateReg, RegCache::GEN_GSTATE);
+		AND(8, R(maskedReg), MDisp(idReg, offsetof(PixelFuncID, cached.stencilTestMask)));
+		UnlockPixelID(idReg);
 	}
 
 	bool success = true;
@@ -738,7 +737,6 @@ bool PixelJitCache::Jit_StencilTest(const PixelFuncID &id, RegCache::Reg stencil
 		return true;
 	}
 
-	bool hadGStateReg = regCache_.Has(RegCache::GEN_GSTATE);
 	bool hadColorOffReg = regCache_.Has(RegCache::GEN_COLOR_OFF);
 	bool hadIdReg = regCache_.Has(RegCache::GEN_ID);
 
@@ -751,9 +749,7 @@ bool PixelJitCache::Jit_StencilTest(const PixelFuncID &id, RegCache::Reg stencil
 		Discard();
 	}
 
-	// If we allocated either gstate or colorOff in the conditional, forget.
-	if (!hadGStateReg && regCache_.Has(RegCache::GEN_GSTATE))
-		regCache_.Change(RegCache::GEN_GSTATE, RegCache::GEN_INVALID);
+	// If we allocated either id or colorOff in the conditional, forget.
 	if (!hadColorOffReg && regCache_.Has(RegCache::GEN_COLOR_OFF))
 		regCache_.Change(RegCache::GEN_COLOR_OFF, RegCache::GEN_INVALID);
 	if (!hadIdReg && regCache_.Has(RegCache::GEN_ID))
@@ -816,7 +812,6 @@ bool PixelJitCache::Jit_DepthTestForStencil(const PixelFuncID &id, RegCache::Reg
 		break;
 	}
 
-	bool hadGStateReg = regCache_.Has(RegCache::GEN_GSTATE);
 	bool hadColorOffReg = regCache_.Has(RegCache::GEN_COLOR_OFF);
 	bool hadIdReg = regCache_.Has(RegCache::GEN_ID);
 
@@ -825,9 +820,7 @@ bool PixelJitCache::Jit_DepthTestForStencil(const PixelFuncID &id, RegCache::Reg
 	success = success && Jit_WriteStencilOnly(id, stencilReg);
 	Discard();
 
-	// If we allocated either gstate or colorOff in the conditional, forget.
-	if (!hadGStateReg && regCache_.Has(RegCache::GEN_GSTATE))
-		regCache_.Change(RegCache::GEN_GSTATE, RegCache::GEN_INVALID);
+	// If we allocated either id or colorOff in the conditional, forget.
 	if (!hadColorOffReg && regCache_.Has(RegCache::GEN_COLOR_OFF))
 		regCache_.Change(RegCache::GEN_COLOR_OFF, RegCache::GEN_INVALID);
 	if (!hadIdReg && regCache_.Has(RegCache::GEN_ID))
@@ -859,9 +852,9 @@ bool PixelJitCache::Jit_ApplyStencilOp(const PixelFuncID &id, GEStencilOp op, Re
 	case GE_STENCILOP_REPLACE:
 		if (id.hasStencilTestMask) {
 			// Load the unmasked value.
-			X64Reg gstateReg = GetGState();
-			MOVZX(32, 8, stencilReg, MDisp(gstateReg, offsetof(GPUgstate, stenciltest) + 1));
-			regCache_.Unlock(gstateReg, RegCache::GEN_GSTATE);
+			X64Reg idReg = GetPixelID();
+			MOVZX(32, 8, stencilReg, MDisp(idReg, offsetof(PixelFuncID, cached.stencilRef)));
+			UnlockPixelID(idReg);
 		} else {
 			MOV(8, R(stencilReg), Imm8(id.stencilTestRef));
 		}
diff --git a/GPU/Software/FuncId.cpp b/GPU/Software/FuncId.cpp
index 19ab3fa436..db73995a4d 100644
--- a/GPU/Software/FuncId.cpp
+++ b/GPU/Software/FuncId.cpp
@@ -203,6 +203,14 @@ void ComputePixelFuncID(PixelFuncID *id) {
 	id->cached.maxz = gstate.getDepthRangeMax();
 	id->cached.framebufStride = gstate.FrameBufStride();
 	id->cached.depthbufStride = gstate.DepthBufStride();
+
+	if (id->hasStencilTestMask) {
+		// Without the mask applied, unlike the one in the key.
+		id->cached.stencilRef = gstate.getStencilTestRef();
+		id->cached.stencilTestMask = gstate.getStencilTestMask();
+	}
+	if (id->hasAlphaTestMask)
+		id->cached.alphaTestMask = gstate.getAlphaTestMask();
 }
 
 std::string DescribePixelFuncID(const PixelFuncID &id) {
diff --git a/GPU/Software/FuncId.h b/GPU/Software/FuncId.h
index 598f62cd78..382a48c2d6 100644
--- a/GPU/Software/FuncId.h
+++ b/GPU/Software/FuncId.h
@@ -57,6 +57,9 @@ struct PixelFuncID {
 		uint16_t framebufStride;
 		uint16_t depthbufStride;
 		GELogicOp logicOp;
+		uint8_t stencilRef;
+		uint8_t stencilTestMask;
+		uint8_t alphaTestMask;
 	} cached;
 
 	union {