Merge pull request #21881 from hrydgard/render-code-cleanup

Misc render code cleanup - remove cached UVScale, etc

GPU/Common/DrawEngineCommon.cpp

@@ -232,7 +232,8 @@ bool DrawEngineCommon::TestBoundingBox(const void *vdata, const void *inds, int
 			// TODO: Avoid normalization if just plain skinning.
 			// Force software skinning.
 			const u32 vertTypeID = GetVertTypeID(vertType, gstate.getUVGenMode(), true);
-			::NormalizeVertices(corners, temp_buffer, (const u8 *)vdata, indexLowerBound, indexUpperBound, dec, vertType);
+			UVScale uvScale{};  // We don't care about UV.
+			::NormalizeVertices(corners, temp_buffer, (const u8 *)vdata, indexLowerBound, indexUpperBound, uvScale, dec, vertType);
 			IndexConverter conv(vertType, inds);
 			for (int i = 0; i < vertexCount; i++) {
 				verts[i * 3] = corners[conv(i)].pos.x;
@@ -891,7 +892,7 @@ bool DrawEngineCommon::SubmitPrim(const void *verts, const void *inds, GEPrimiti
 		numDrawVerts_ = numDrawVerts + 1;  // Increment the uncached variable
 		dv.verts = verts;
 		dv.vertexCount = vertexCount;
-		dv.uvScale = gstate_c.uv;
+		dv.uvScale = LoadUVScaleOffset(gstate);
 		// Does handle the unindexed case.
 		GetIndexBounds(inds, vertexCount, vertTypeID, &dv.indexLowerBound, &dv.indexUpperBound);
 	}
@@ -1064,6 +1065,7 @@ Mat4F32 ComputeFinalProjMatrix() {
 		0.0f,
 	};
 
+	// TODO: Simply use Mat4F32 m(gstate_c.worldviewproj);
 	Mat4F32 wv = Mul4x3By4x4(Mat4x3F32(gstate.worldMatrix), Mat4F32::Load4x3(gstate.viewMatrix));
 	Mat4F32 m = Mul4x4By4x4(wv, Mat4F32(gstate.projMatrix));
 	// NOTE: Applying the translation actually works pre-divide, since W is also affected.

GPU/Common/FragmentShaderGenerator.cpp

@@ -154,10 +154,7 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
 
 	ReplaceBlendType replaceBlend = static_cast<ReplaceBlendType>(id.Bits(FS_BIT_REPLACE_BLEND, 3));
 
-	bool blueToAlpha = false;
-	if (replaceBlend == ReplaceBlendType::REPLACE_BLEND_BLUE_TO_ALPHA) {
-		blueToAlpha = true;
-	}
+	const bool blueToAlpha = replaceBlend == ReplaceBlendType::REPLACE_BLEND_BLUE_TO_ALPHA;
 
 	bool isModeClear = id.Bit(FS_BIT_CLEARMODE);
 
@@ -854,8 +851,8 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
 			// This happens before fog is applied.
 			*uniformMask |= DIRTY_TEX_ALPHA_MUL;
 
-			// We only need a clamp if the color will be further processed. Otherwise the hardware color conversion will clamp for us.
-			// TODO: Clamp is so cheap that this probably is pretty meaningless.
+			// We only need a clamp if the color will be further processed directly in the shader.
+			// Otherwise the hardware color conversion will clamp for us (unless we start using FP formats for fake HDR but then we probably don't want the clamp...)
 			if (enableFog || enableColorTest || replaceBlend != REPLACE_BLEND_NO || simulateLogicOpType != LOGICOPTYPE_NORMAL || colorWriteMask || blueToAlpha) {
 				WRITE(p, "  v.rgb = clamp(v.rgb * u_texNoAlphaMul.y, 0.0, 1.0);\n");
 			} else {

GPU/Common/FramebufferManagerCommon.cpp

@@ -313,7 +313,7 @@ void FramebufferManagerCommon::EstimateDrawingSize(u32 fb_address, int fb_stride
 	DEBUG_LOG(Log::G3D, "Est: %08x V: %ix%i, R: %ix%i, S: %ix%i, STR: %i, THR:%i, Z:%08x = %ix%i %s", fb_address, viewport_width,viewport_height, region_width, region_height, scissor_width, scissor_height, fb_stride, gstate.isModeThrough(), gstate.isDepthWriteEnabled() ? gstate.getDepthBufAddress() : 0, drawing_width, drawing_height, margin ? " (margin!)" : "");
 }
 
-void GetFramebufferHeuristicInputs(FramebufferHeuristicParams *params, const GPUgstate &gstate) {
+void GetFramebufferHeuristicInputs(FramebufferHeuristicParams *params, const GEState &gstate) {
 	// GetFramebufferHeuristicInputs is only called from rendering, and thus, it's VRAM.
 	params->fb_address = gstate.getFrameBufRawAddress() | 0x04000000;
 	params->fb_stride = gstate.FrameBufStride();

GPU/Common/FramebufferManagerCommon.h

@@ -195,10 +195,10 @@ struct FramebufferHeuristicParams {
 	int16_t scissorBottom;
 };
 
-struct GPUgstate;
-extern GPUgstate gstate;
+struct GEState;
+extern GEState gstate;
 
-void GetFramebufferHeuristicInputs(FramebufferHeuristicParams *params, const GPUgstate &gstate);
+void GetFramebufferHeuristicInputs(FramebufferHeuristicParams *params, const GEState &gstate);
 
 enum BindFramebufferColorFlags {
 	BINDFBCOLOR_SKIP_COPY = 0,

GPU/Common/GPUDebugInterface.cpp

@@ -621,7 +621,7 @@ bool GEExpressionFunctions::parseSymbol(char *str, uint32_t &symbolValue) {
 }
 
 uint32_t GEExpressionFunctions::getReferenceValue(uint32_t referenceIndex) {
-	GPUgstate state = gpu_->GetGState();
+	GEState state = gpu_->GetGState();
 	if (referenceIndex < 0x100) {
 		GECmdFormat fmt = GECmdInfoByCmd(GECommand(referenceIndex)).cmdFmt;
 		uint32_t value = state.cmdmem[referenceIndex];

GPU/Common/GPUStateUtils.h

@@ -59,7 +59,7 @@ bool NeedsTestDiscard();
 bool IsDepthTestEffectivelyDisabled();
 bool IsStencilTestOutputDisabled();
 
-inline bool CanForceBilinear(const GPUgstate &gstate) {
+inline bool CanForceBilinear(const GEState &gstate) {
 	return ((!gstate.isColorTestEnabled() || IsColorTestTriviallyTrue()) && (!gstate.isAlphaTestEnabled() || IsAlphaTestTriviallyTrue()));
 }
 

GPU/Common/ShaderUniforms.cpp

@@ -254,7 +254,7 @@ void BoneUpdateUniforms(UB_VS_Bones *ub, uint64_t dirtyUniforms) {
 	}
 }
 
-void UpdateFogCoef(const GPUgstate &state, float fogCoef[2]) {
+void UpdateFogCoef(const GEState &state, float fogCoef[2]) {
 	fogCoef[0] = getFloat24(gstate.fog1);
 	fogCoef[1] = getFloat24(gstate.fog2);
 	// The PSP just ignores infnan here (ignoring IEEE), so take it down to a valid float.

GPU/Common/ShaderUniforms.h

@@ -123,10 +123,10 @@ void BoneUpdateUniforms(UB_VS_Bones *ub, uint64_t dirtyUniforms);
 uint32_t PackLightControlBits();
 uint32_t PackDepalBits();
 
-void UpdateFogCoef(const GPUgstate &state, float fogCoef[2]);
+void UpdateFogCoef(const GEState &state, float fogCoef[2]);
 
 // This happens so much that I want it inline.
-inline void UpdateUVScaleOff(const GPUgstate &state, float uvScaleOff[4]) {
+inline void UpdateUVScaleOff(const GEState &state, float uvScaleOff[4]) {
 	float widthFactor;
 	float heightFactor;
 	if (gstate_c.textureIsFramebuffer) {

GPU/Common/SoftwareTransformCommon.cpp

@@ -80,7 +80,7 @@ static void RotateUV(TransformedVertex v[4]) {
 	}
 }
 
-static bool ShouldApplySpriteBorderFix(const GPUgstate &gstate) {
+static bool ShouldApplySpriteBorderFix(const GEState &gstate) {
 	return gstate.isMagnifyFilteringEnabled() && gstate.isAlphaBlendEnabled() && gstate.getBlendFuncA() != GE_SRCBLEND_FIXA && gstate.isTextureAlphaUsed();
 }
 
@@ -407,7 +407,7 @@ SoftwareTransformAction RunSoftwareTransform(SoftwareTransformParams &params, in
 
 // Modifies the vertices in-place. Applies viewport and projection.
 // TODO: SIMD.
-static void ProjectVertices(const GPUgstate &gstate, TransformedVertex *transformed, int vertexCount) {
+static void ProjectVertices(const GEState &gstate, TransformedVertex *transformed, int vertexCount) {
 #if 0
 	Lin::Vec3 vpOffset(gstate.getViewportXCenter(), gstate.getViewportYCenter(), gstate.getViewportZCenter());
 	Lin::Vec3 vpScale(gstate.getViewportXScale(), gstate.getViewportYScale(), gstate.getViewportZScale());
@@ -1366,13 +1366,13 @@ static bool ExpandPoints(int vertexCount, int &maxIndex, int vertsSize, u16 *&in
 // The rest of the transform pipeline like lighting will go as normal, either hardware or software.
 // The implementation is initially a bit inefficient but shouldn't be a big deal.
 // An intermediate buffer of not-easy-to-predict size is stored at bufPtr.
-u32 NormalizeVertices(SimpleVertex *sverts, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, const VertexDecoder *dec, u32 vertType) {
+u32 NormalizeVertices(SimpleVertex *sverts, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, const UVScale &uvScale, const VertexDecoder *dec, u32 vertType) {
 	// First, decode the vertices into a GPU compatible format. This step can be eliminated but will need a separate
 	// implementation of the vertex decoder.
 	// Actually if software transform is off, we could enforce it in the vertex decoder lookup before calling this,
 	// avoiding having to implement it again below.
 	const int count = upperBound + 1 - lowerBound;
-	dec->DecodeVerts(bufPtr, inPtr + lowerBound * dec->VertexSize(), &gstate_c.uv, count);
+	dec->DecodeVerts(bufPtr, inPtr + lowerBound * dec->VertexSize(), &uvScale, count);
 
 	// OK, morphing eliminated but bones still remain to be taken care of.
 	// Let's do a partial software transform where we only do skinning.
@@ -1543,12 +1543,10 @@ bool GetCurrentDrawAsDebugVertices(DrawEngineCommon *drawEngine, GECommand cmd,
 	const int stride = (int)dec->GetDecVtxFmt().stride;
 	vertsTemp.resize(stride * verticesToDecode + 32);  // Add some padding bytes for "over-writes".
 
-	UVScale uvScale{};
-	LoadUVScaleOffsetVec(gstate).Store(&uvScale.uScale);
-
 	const u8 *startPos = verts + indexLowerBound * dec->VertexSize();
 
 	bool savedVertexFullAlpha = gstate_c.vertexFullAlpha;
+	const UVScale uvScale = LoadUVScaleOffset(gstate);
 	dec->DecodeVerts(vertsTemp.data(), startPos, &uvScale, verticesToDecode);
 	gstate_c.vertexFullAlpha = savedVertexFullAlpha;
 

GPU/Common/SoftwareTransformCommon.h

@@ -83,7 +83,7 @@ SoftwareTransformAction RunSoftwareTransform(SoftwareTransformParams &params, in
 class DrawEngineCommon;
 
 // Slow. See description in the cpp file.
-u32 NormalizeVertices(SimpleVertex *sverts, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, const VertexDecoder *dec, u32 vertType);
+u32 NormalizeVertices(SimpleVertex *sverts, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, const UVScale &uvScale, const VertexDecoder *dec, u32 vertType);
 
 // In the returned data, you should subtract the value of lowerIndexBound from the indices to get the actual vertex index in the vertices array.
 // This is because some draws in some games use very large indices, but they only use a small range of them in each PRIM submission.

GPU/Common/SplineCommon.cpp

@@ -159,6 +159,8 @@ private:
 	}
 
 	static void CalcWeights(float t, const float *knots, const KnotDiv &div, Weight &w) {
+		// TODO: This SSE code doesn't look like it's worth it. We need to parallelize across another
+		// dimension.
 #ifdef _M_SSE
 		const __m128 knot012 = _mm_loadu_ps(knots);
 		const __m128 t012 = _mm_sub_ps(_mm_set_ps1(t), knot012);
@@ -463,8 +465,9 @@ void DrawEngineCommon::SubmitCurve(const void *control_points, const void *indic
 	u16 index_lower_bound = 0;
 	u16 index_upper_bound = num_points - 1;
 	IndexConverter ConvertIndex(vertType, indices);
-	if (indices)
+	if (indices) {
 		GetIndexBounds(indices, num_points, vertType, &index_lower_bound, &index_upper_bound);
+	}
 
 	u32 vertTypeID = GetVertTypeID(vertType, gstate.getUVGenMode(), applySkinInDecode_);
 	VertexDecoder *origVDecoder = GetVertexDecoder(vertTypeID);
@@ -485,7 +488,8 @@ void DrawEngineCommon::SubmitCurve(const void *control_points, const void *indic
 	}
 
 	const u32 origVertType = vertType;
-	vertType = ::NormalizeVertices(simplified_control_points, temp_buffer, (u8 *)control_points, index_lower_bound, index_upper_bound, origVDecoder, vertType);
+	UVScale neutralUVScale{1.0f, 1.0f, 0.0f, 0.0f};  // Avoid rescaling UV during normalization, it will happen anyway later (in DispatchSubmitPrim). Although ideally we should avoid running Decode at all there.
+	vertType = ::NormalizeVertices(simplified_control_points, temp_buffer, (u8 *)control_points, index_lower_bound, index_upper_bound, neutralUVScale, origVDecoder, vertType);
 
 	VertexDecoder *vdecoder = GetVertexDecoder(vertType);
 
@@ -500,8 +504,9 @@ void DrawEngineCommon::SubmitCurve(const void *control_points, const void *indic
 		ERROR_LOG(Log::G3D, "Failed to allocate space for control point pointers, skipping curve draw");
 		return;
 	}
-	for (int idx = 0; idx < num_points; idx++)
+	for (int idx = 0; idx < num_points; idx++) {
 		points[idx] = simplified_control_points + (indices ? ConvertIndex(idx) : idx);
+	}
 
 	OutputBuffers output;
 	output.vertices = (SimpleVertex *)(decoded_ + DECODED_VERTEX_BUFFER_SIZE / 2);
@@ -513,23 +518,15 @@ void DrawEngineCommon::SubmitCurve(const void *control_points, const void *indic
 	surface.Init(maxVerts);
 
 	ControlPoints cpoints(points, num_points, managedBuf);
-	if (cpoints.IsValid())
+	if (cpoints.IsValid()) {
+		// Run the tessellation!
 		SoftwareTessellation(output, surface, origVertType, cpoints);
-	else
+	} else {
 		ERROR_LOG(Log::G3D, "Failed to allocate space for control point values, skipping curve draw");
+	}
 
 	u32 vertTypeWithIndex16 = (vertType & ~GE_VTYPE_IDX_MASK) | GE_VTYPE_IDX_16BIT;
 
-	UVScale prevUVScale;
-	if (origVertType & GE_VTYPE_TC_MASK) {
-		// We scaled during Normalize already so let's turn it off when drawing.
-		prevUVScale = gstate_c.uv;
-		gstate_c.uv.uScale = 1.0f;
-		gstate_c.uv.vScale = 1.0f;
-		gstate_c.uv.uOff = 0;
-		gstate_c.uv.vOff = 0;
-	}
-
 	vertTypeID = GetVertTypeID(vertTypeWithIndex16, gstate.getUVGenMode(), applySkinInDecode_);
 	int generatedBytesRead;
 	if (output.count) {
@@ -537,11 +534,7 @@ void DrawEngineCommon::SubmitCurve(const void *control_points, const void *indic
 		DispatchSubmitPrim(output.vertices, output.indices, PatchPrimToPrim(surface.primType), output.count, vertTypeID, true, &generatedBytesRead, flags);
 	}
 
-	if (flushOnParams_)
+	if (flushOnParams_) {
 		Flush();
-
-	if (origVertType & GE_VTYPE_TC_MASK) {
-		// If analysis says this is uninitialized, it's wrong.
-		gstate_c.uv = prevUVScale;
 	}
 }

GPU/Common/SplineCommon.h

@@ -27,7 +27,7 @@
 
 #include "Core/Config.h"
 
-#define HALF_CEIL(x) (x + 1) / 2 // Integer ceil = (int)ceil((float)x / 2.0f)
+inline int half_ceil(int x) { return (x + 1) / 2; } // Integer ceil = (int)ceil((float)x / 2.0f)
 
 class SimpleBufferManager;
 
@@ -61,8 +61,8 @@ struct SurfaceInfo {
 			break;
 		case SplineQuality::MEDIUM_QUALITY:
 			// Don't cut below 2, though.
-			if (tess_u > 2) tess_u = HALF_CEIL(tess_u);
-			if (tess_v > 2) tess_v = HALF_CEIL(tess_v);
+			if (tess_u > 2) tess_u = half_ceil(tess_u);
+			if (tess_v > 2) tess_v = half_ceil(tess_v);
 			break;
 		default:
 			break;
@@ -140,7 +140,8 @@ struct SplineSurface : public SurfaceInfo {
 };
 
 struct Weight {
-	float basis[4], deriv[4];
+	float basis[4];
+	float deriv[4];
 };
 
 template<class T>
@@ -207,10 +208,10 @@ struct NAME { \
 	} \
 };
 
-template<typename Func, int NumParams, class Dispatcher> 
+template<typename Func, int NumParams, class Dispatcher>
 class TemplateParameterDispatcher {
 
-	/* Store all combinations of template functions into an array */
+	// Store all combinations of template functions into an array
 	template<int LoopCount, int Index = 0, bool ...Params>
 	struct Initializer {
 		static void Init(Func funcs[]) {
@@ -218,7 +219,7 @@ class TemplateParameterDispatcher {
 			Initializer<LoopCount - 1, (Index << 1) + 0, false, Params...>::Init(funcs); // false
 		}
 	};
- 	/* Specialized for terminates the recursive loop */
+	// Specialized for terminates the recursive loop
 	template<int Index, bool ...Params>
 	struct Initializer<0, Index, Params...> {
 		static void Init(Func funcs[]) {
@@ -226,19 +227,19 @@ class TemplateParameterDispatcher {
 		}
 	};
 
-private: 
-	Func funcs[1 << NumParams]; /* Function pointers array */ 
-public: 
-	TemplateParameterDispatcher() { 
-		Initializer<NumParams>::Init(funcs); 
-	} 
- 
-	Func GetFunc(const bool params[]) const { 
- 		/* Convert bool parameters to index of the array */ 
-		int index = 0; 
-		for (int i = 0; i < NumParams; ++i) 
-			index |= params[i] << i; 
- 
-		return funcs[index]; 
-	} 
+private:
+	Func funcs[1 << NumParams]; /* Function pointers array */
+public:
+	TemplateParameterDispatcher() {
+		Initializer<NumParams>::Init(funcs);
+	}
+
+	Func GetFunc(const bool params[]) const {
+		// Convert bool parameters to index of the array
+		int index = 0;
+		for (int i = 0; i < NumParams; ++i) {
+			index |= params[i] << i;
+		}
+		return funcs[index];
+	}
 };

GPU/Common/TextureCacheCommon.cpp

@@ -2301,7 +2301,7 @@ static bool CanDepalettizeBufferAs(GETextureFormat texFormat, GEBufferFormat buf
 // But we only do it if the mask/shift exactly matches a color channel, else something different might be going
 // on and we definitely don't want to interpolate.
 // Great enhancement for Test Drive and Manhunt 2.
-static bool CanUseSmoothDepal(const GPUgstate &gstate, GEBufferFormat framebufferFormat, const ClutTexture &clutTexture) {
+static bool CanUseSmoothDepal(const GEState &gstate, GEBufferFormat framebufferFormat, const ClutTexture &clutTexture) {
 	for (int i = 0; i < ClutTexture::MAX_RAMPS; i++) {
 		if (gstate.getClutIndexStartPos() == clutTexture.rampStarts[i] &&
 			gstate.getClutIndexMask() < clutTexture.rampLengths[i]) {

GPU/Debugger/Stepping.cpp

@@ -68,7 +68,7 @@ static bool lastWasFramebuffer;
 static u32 pauseSetCmdValue;
 
 // This is used only to highlight differences. Should really be owned by the debugger.
-static GPUgstate lastGState;
+static GEState lastGState;
 
 const char *PauseActionToString(PauseAction action) {
 	switch (action) {
@@ -291,7 +291,7 @@ bool GPU_FlushDrawing() {
 	return true;
 }
 
-const GPUgstate &LastState() {
+const GEState &LastState() {
 	return lastGState;
 }
 

GPU/Debugger/Stepping.h

@@ -47,5 +47,5 @@ namespace GPUStepping {
 	bool GPU_FlushDrawing();
 
 	// Can be used to highlight differences in a debugger.
-	const GPUgstate &LastState();
+	const GEState &LastState();
 };

GPU/GPUCommon.cpp

@@ -1325,8 +1325,6 @@ void GPUCommon::FlushImm() {
 
 	SetDrawType(DRAW_PRIM, immPrim_);
 
-	gstate_c.UpdateUVScaleOffset();
-
 	VirtualFramebuffer *vfb = nullptr;
 	if (framebufferManager_) {
 		bool changed;
@@ -1707,7 +1705,7 @@ u32 GPUCommon::GetIndexAddress() {
 	return gstate_c.indexAddr;
 }
 
-const GPUgstate &GPUCommon::GetGState() {
+const GEState &GPUCommon::GetGState() {
 	return gstate;
 }
 

GPU/GPUCommon.h

@@ -30,7 +30,7 @@ class TextureCacheCommon;
 class DrawEngineCommon;
 class GraphicsContext;
 struct PspGeListArgs;
-struct GPUgstate;
+struct GEState;
 class PointerWrap;
 struct VirtualFramebuffer;
 
@@ -221,7 +221,7 @@ public:
 	u32 GetRelativeAddress(u32 data);
 	u32 GetVertexAddress();
 	u32 GetIndexAddress();
-	const GPUgstate &GetGState();
+	const GEState &GetGState();
 	void SetCmdValue(u32 op);
 
 	DisplayList* getList(int listid) {

GPU/GPUCommonHW.cpp

@@ -967,8 +967,6 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 		canExtend = false;
 	}
 
-	gstate_c.UpdateUVScaleOffset();
-
 	// cull mode
 	int cullMode = gstate.getCullMode();
 
@@ -1190,22 +1188,18 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 			// when texscale commands are in line with the prims like this, they actually have an effect
 			// and requires us to stop extending strips anyway.
 			gstate.cmdmem[GE_CMD_TEXSCALEU] = data;
-			gstate_c.uv.uScale = getFloat24(data);
 			canExtend = false;
 			break;
 		case GE_CMD_TEXSCALEV:
 			gstate.cmdmem[GE_CMD_TEXSCALEV] = data;
-			gstate_c.uv.vScale = getFloat24(data);
 			canExtend = false;
 			break;
 		case GE_CMD_TEXOFFSETU:
 			gstate.cmdmem[GE_CMD_TEXOFFSETU] = data;
-			gstate_c.uv.uOff = getFloat24(data);
 			canExtend = false;
 			break;
 		case GE_CMD_TEXOFFSETV:
 			gstate.cmdmem[GE_CMD_TEXOFFSETV] = data;
-			gstate_c.uv.vOff = getFloat24(data);
 			canExtend = false;
 			break;
 		case GE_CMD_TEXLEVEL:
@@ -1333,7 +1327,6 @@ void GPUCommonHW::Execute_Bezier(u32 op, u32 diff) {
 	gstate_c.submitType = SubmitType::BEZIER;
 
 	int bytesRead = 0;
-	gstate_c.UpdateUVScaleOffset();
 	drawEngineCommon_->SubmitCurve(control_points, indices, surface, gstate.vertType, &bytesRead, "bezier");
 
 	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_UVSCALEOFFSET);
@@ -1405,7 +1398,6 @@ void GPUCommonHW::Execute_Spline(u32 op, u32 diff) {
 	gstate_c.submitType = SubmitType::SPLINE;
 
 	int bytesRead = 0;
-	gstate_c.UpdateUVScaleOffset();
 	drawEngineCommon_->SubmitCurve(control_points, indices, surface, gstate.vertType, &bytesRead, "spline");
 
 	gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VERTEXSHADER_STATE | DIRTY_UVSCALEOFFSET);

GPU/GPUState.cpp

@@ -27,7 +27,7 @@
 #include "GPU/GPUState.h"
 
 // This must be aligned so that the matrices within are aligned.
-alignas(16) GPUgstate gstate;
+alignas(16) GEState gstate;
 // Let's align this one too for good measure.
 alignas(16) GPUStateCache gstate_c;
 
@@ -102,7 +102,7 @@ static const u32_le *LoadMatrix(const u32_le *cmds, float *mtx, int sz) {
 	return cmds;
 }
 
-void GPUgstate::Reset() {
+void GEState::Reset() {
 	memset(gstate.cmdmem, 0, sizeof(gstate.cmdmem));
 	for (int i = 0; i < 256; i++) {
 		gstate.cmdmem[i] = i << 24;
@@ -120,7 +120,7 @@ void GPUgstate::Reset() {
 	gstate_c.Dirty(DIRTY_WORLD_VIEW_PROJ_MATRIX | DIRTY_VIEW_PROJ_MATRIX | DIRTY_CULL_MATRIX);
 }
 
-void GPUgstate::Save(u32_le *ptr) {
+void GEState::Save(u32_le *ptr) {
 	// Not sure what the first 10 values are, exactly, but these seem right.
 	ptr[5] = gstate_c.vertexAddr;
 	ptr[6] = gstate_c.indexAddr;
@@ -168,7 +168,7 @@ void GPUgstate::Save(u32_le *ptr) {
 	}
 }
 
-void GPUgstate::FastLoadBoneMatrix(u32 addr) {
+void GEState::FastLoadBoneMatrix(u32 addr) {
 	const u32 *src = (const u32 *)Memory::GetPointerUnchecked(addr);
 	u32 num = boneMatrixNumber;
 	u32 *dst = (u32 *)(boneMatrix + (num & 0x7F));
@@ -204,7 +204,7 @@ void GPUgstate::FastLoadBoneMatrix(u32 addr) {
 	gstate.boneMatrixNumber = (GE_CMD_BONEMATRIXNUMBER << 24) | (num & 0x00FFFFFF);
 }
 
-void GPUgstate::Restore(const u32_le *ptr) {
+void GEState::Restore(const u32_le *ptr) {
 	// Not sure what the first 10 values are, exactly, but these seem right.
 	gstate_c.vertexAddr = ptr[5];
 	gstate_c.indexAddr = ptr[6];
@@ -292,7 +292,6 @@ void GPUStateCache::DoState(PointerWrap &p) {
 		textureSolidAlpha = old.textureSolidAlpha;
 		vertexFullAlpha = old.vertexFullAlpha;
 		skipDrawReason = old.skipDrawReason;
-		uv = old.uv;
 
 		savedContextVersion = 0;
 	} else {
@@ -310,6 +309,8 @@ void GPUStateCache::DoState(PointerWrap &p) {
 
 		Do(p, skipDrawReason);
 
+		// Legacy, remove in the next bump.
+		UVScale uv{};
 		Do(p, uv);
 
 		bool oldFlipTexture = false;

GPU/GPUState.h

@@ -26,12 +26,11 @@
 #include "GPU/GPU.h"
 #include "GPU/ge_constants.h"
 #include "GPU/Common/ShaderCommon.h"
-#include "Common/Math/SIMDHeaders.h"
 #include "Common/Math/lin/vec3.h"
 
 class PointerWrap;
 
-struct GPUgstate {
+struct GEState {
 	// Getting rid of this ugly union in favor of the accessor functions
 	// might be a good idea....
 	union {
@@ -514,7 +513,7 @@ enum class SubmitType {
 	SPLINE,
 };
 
-extern GPUgstate gstate;
+extern GEState gstate;
 
 struct GPUStateCache {
 	bool Use(u32 flags) const { return (useFlags_ & flags) != 0; } // Return true if ANY of flags are true.
@@ -588,21 +587,6 @@ struct GPUStateCache {
 		return useFlags_;
 	}
 
-	void UpdateUVScaleOffset() {
-#if defined(_M_SSE)
-		__m128i values = _mm_slli_epi32(_mm_load_si128((const __m128i *)&gstate.texscaleu), 8);
-		_mm_storeu_si128((__m128i *)&uv, values);
-#elif PPSSPP_ARCH(ARM_NEON)
-		const uint32x4_t values = vshlq_n_u32(vld1q_u32((const u32 *)&gstate.texscaleu), 8);
-		vst1q_u32((u32 *)&uv, values);
-#else
-		uv.uScale = getFloat24(gstate.texscaleu);
-		uv.vScale = getFloat24(gstate.texscalev);
-		uv.uOff = getFloat24(gstate.texoffsetu);
-		uv.vOff = getFloat24(gstate.texoffsetv);
-#endif
-	}
-
 private:
 	u32 useFlags_;
 public:
@@ -625,8 +609,6 @@ public:
 
 	int skipDrawReason;
 
-	UVScale uv;
-
 	bool bgraTexture;
 	bool needShaderTexClamp;
 	bool textureIsArray;

GPU/GPUStateSIMDUtil.h

@@ -3,14 +3,16 @@
 #include "Common/Math/CrossSIMD.h"
 #include "GPU/GPUState.h"
 
-inline Vec4F32 LoadViewportOffsetVec(const GPUgstate &gstate) {
+inline Vec4F32 LoadViewportOffsetVec(const GEState &gstate) {
 	// We ignore the last member.
 	return Vec4F32::LoadF24x4(&gstate.viewportxcenter);
 }
-inline Vec4F32 LoadViewportScaleVec(const GPUgstate &gstate) {
+inline Vec4F32 LoadViewportScaleVec(const GEState &gstate) {
 	// We ignore the last member.
 	return Vec4F32::LoadF24x4(&gstate.viewportxscale);
 }
-inline Vec4F32 LoadUVScaleOffsetVec(const GPUgstate &gstate) {
-	return Vec4F32::LoadF24x4(&gstate.texscaleu);
+inline UVScale LoadUVScaleOffset(const GEState &gstate) {
+	UVScale uvScale;
+	Vec4F32::LoadF24x4(&gstate.texscaleu).Store(&uvScale.uScale);
+	return uvScale;
 }

GPU/Software/SoftGpu.cpp

@@ -863,7 +863,6 @@ void SoftGPU::Execute_Prim(u32 op, u32 diff) {
 
 	cyclesExecuted += EstimatePerVertexCost() * count;
 	int bytesRead;
-	gstate_c.UpdateUVScaleOffset();
 	drawEngine_->transformUnit.SetDirty(dirtyFlags_);
 	drawEngine_->transformUnit.SubmitPrimitive(verts, indices, prim, count, gstate.vertType, &bytesRead, drawEngine_);
 	dirtyFlags_ = drawEngine_->transformUnit.GetDirty();
@@ -916,7 +915,6 @@ void SoftGPU::Execute_Bezier(u32 op, u32 diff) {
 	SetDrawType(DRAW_BEZIER, PatchPrimToPrim(surface.primType));
 
 	int bytesRead = 0;
-	gstate_c.UpdateUVScaleOffset();
 	drawEngine_->transformUnit.SetDirty(dirtyFlags_);
 	drawEngineCommon_->SubmitCurve(control_points, indices, surface, gstate.vertType, &bytesRead, "bezier");
 	dirtyFlags_ = drawEngine_->transformUnit.GetDirty();
@@ -970,7 +968,6 @@ void SoftGPU::Execute_Spline(u32 op, u32 diff) {
 	SetDrawType(DRAW_SPLINE, PatchPrimToPrim(surface.primType));
 
 	int bytesRead = 0;
-	gstate_c.UpdateUVScaleOffset();
 	drawEngine_->transformUnit.SetDirty(dirtyFlags_);
 	drawEngineCommon_->SubmitCurve(control_points, indices, surface, gstate.vertType, &bytesRead, "spline");
 	dirtyFlags_ = drawEngine_->transformUnit.GetDirty();

GPU/Software/TransformUnit.cpp

@@ -29,6 +29,7 @@
 #include "GPU/Common/VertexDecoderCommon.h"
 #include "GPU/Common/SoftwareTransformCommon.h"
 #include "GPU/Common/VertexReader.h"
+#include "GPU/GPUStateSIMDUtil.h"
 #include "Common/Math/SIMDHeaders.h"
 #include "GPU/Software/BinManager.h"
 #include "GPU/Software/Clipper.h"
@@ -482,7 +483,8 @@ public:
 			GetIndexBounds(indices, vertex_count, vertex_type, &lowerBound_, &upperBound_);
 		if (vertex_count != 0) {
 			const int count = upperBound_ - lowerBound_ + 1;
-			vdecoder.DecodeVerts(base, (const u8 *)vertices + vdecoder.VertexSize() * lowerBound_, &gstate_c.uv, count);
+			const UVScale uvScale = LoadUVScaleOffset(gstate);
+			vdecoder.DecodeVerts(base, (const u8 *)vertices + vdecoder.VertexSize() * lowerBound_, &uvScale, count);
 		}
 
 		// If we're only using a subset of verts, it's better to decode with random access (usually.)

UI/ImDebugger/ImGe.cpp

@@ -1643,7 +1643,7 @@ void ImGeStateWindow::Draw(ImConfig &cfg, ImControl &control, GPUCommon *gpuDebu
 					bool sectionOpen = false;
 					for (size_t i = 0; i < numRows; i++) {
 						const GECmdInfo &info = GECmdInfoByCmd(rows[i].cmd);
-						const GPUgstate &lastState = GPUStepping::LastState();
+						const GEState &lastState = GPUStepping::LastState();
 						bool diff = lastState.cmdmem[rows[i].cmd] != gstate.cmdmem[rows[i].cmd];
 
 						if (rows[i].header) {

Windows/GEDebugger/GEDebugger.cpp

@@ -377,7 +377,7 @@ void CGEDebugger::SetupPreviews() {
 	}
 }
 
-void CGEDebugger::DescribePrimaryPreview(const GPUgstate &state, char desc[256]) {
+void CGEDebugger::DescribePrimaryPreview(const GEState &state, char desc[256]) {
 	if (primaryTrackX_ < primaryBuffer_->GetStride() && primaryTrackY_ < primaryBuffer_->GetHeight()) {
 		u32 pix = primaryBuffer_->GetRawPixel(primaryTrackX_, primaryTrackY_);
 		DescribePixel(pix, primaryBuffer_->GetFormat(), primaryTrackX_, primaryTrackY_, desc);
@@ -410,7 +410,7 @@ void CGEDebugger::DescribePrimaryPreview(const GPUgstate &state, char desc[256])
 	}
 }
 
-void CGEDebugger::DescribeSecondPreview(const GPUgstate &state, char desc[256]) {
+void CGEDebugger::DescribeSecondPreview(const GEState &state, char desc[256]) {
 	if (secondTrackX_ != 0xFFFFFFFF) {
 		uint32_t x = secondTrackX_;
 		uint32_t y = secondTrackY_;
@@ -557,7 +557,7 @@ void CGEDebugger::UpdatePreviews() {
 		return;
 	}
 
-	GPUgstate state{};
+	GEState state{};
 
 	if (gpu != nullptr) {
 		state = gpu->GetGState();
@@ -663,7 +663,7 @@ int CGEDebugger::HasTabIndex(GEDebuggerTab *tab, GETabPosition pos) {
 	return -1;
 }
 
-u32 CGEDebugger::TexturePreviewFlags(const GPUgstate &state) {
+u32 CGEDebugger::TexturePreviewFlags(const GEState &state) {
 	if (state.isTextureAlphaUsed() && !forceOpaque_) {
 		return SimpleGLWindow::ALPHA_BLEND | SimpleGLWindow::RESIZE_BEST_CENTER;
 	} else {
@@ -671,7 +671,7 @@ u32 CGEDebugger::TexturePreviewFlags(const GPUgstate &state) {
 	}
 }
 
-void CGEDebugger::UpdatePrimaryPreview(const GPUgstate &state) {
+void CGEDebugger::UpdatePrimaryPreview(const GEState &state) {
 	bool bufferResult = false;
 	u32 flags = SimpleGLWindow::ALPHA_IGNORE | SimpleGLWindow::RESIZE_SHRINK_CENTER;
 
@@ -724,7 +724,7 @@ void CGEDebugger::UpdatePrimaryPreview(const GPUgstate &state) {
 	}
 }
 
-void CGEDebugger::UpdateSecondPreview(const GPUgstate &state) {
+void CGEDebugger::UpdateSecondPreview(const GEState &state) {
 	bool bufferResult = false;
 
 	SetupPreviews();
@@ -784,7 +784,7 @@ void CGEDebugger::PrimaryPreviewHover(int x, int y) {
 		desc[0] = 0;
 	} else if (x < 0 || y < 0) {
 		// This means they left the area.
-		GPUgstate state{};
+		GEState state{};
 		if (gpu != nullptr) {
 			state = gpu->GetGState();
 		}
@@ -812,7 +812,7 @@ void CGEDebugger::SecondPreviewHover(int x, int y) {
 		desc[0] = 0;
 	} else if (x < 0 || y < 0) {
 		// This means they left the area.
-		GPUgstate state{};
+		GEState state{};
 		if (gpu != nullptr) {
 			state = gpu->GetGState();
 		}
@@ -834,7 +834,7 @@ void CGEDebugger::SecondPreviewHover(int x, int y) {
 }
 
 void CGEDebugger::UpdateTextureLevel(int level) {
-	GPUgstate state{};
+	GEState state{};
 	if (gpu != nullptr) {
 		state = gpu->GetGState();
 	}

Windows/GEDebugger/GEDebugger.h

@@ -43,7 +43,7 @@ class TabStateSettings;
 class TabVertices;
 class TabMatrices;
 class TabStateWatch;
-struct GPUgstate;
+struct GEState;
 
 enum class GETabPosition {
 	LEFT = 1,
@@ -100,8 +100,8 @@ protected:
 private:
 	void SetupPreviews();
 	void UpdatePreviews();
-	void UpdatePrimaryPreview(const GPUgstate &state);
-	void UpdateSecondPreview(const GPUgstate &state);
+	void UpdatePrimaryPreview(const GEState &state);
+	void UpdateSecondPreview(const GEState &state);
 	static u32 PrimPreviewOp();
 	void UpdatePrimPreview(u32 op, int which);
 	void CleanupPrimPreview();
@@ -109,8 +109,8 @@ private:
 	void UpdateSize(WORD width, WORD height);
 	void SavePosition();
 	void UpdateTextureLevel(int level);
-	void DescribePrimaryPreview(const GPUgstate &state, char desc[256]);
-	void DescribeSecondPreview(const GPUgstate &state, char desc[256]);
+	void DescribePrimaryPreview(const GEState &state, char desc[256]);
+	void DescribeSecondPreview(const GEState &state, char desc[256]);
 	void PrimaryPreviewHover(int x, int y);
 	void SecondPreviewHover(int x, int y);
 	void PreviewExport(const GPUDebugBuffer *buffer);
@@ -122,7 +122,7 @@ private:
 	int HasTabIndex(GEDebuggerTab *tab, GETabPosition pos);
 	void CheckTabMessage(TabControl *t, GETabPosition pos, LPARAM lParam);
 
-	u32 TexturePreviewFlags(const GPUgstate &state);
+	u32 TexturePreviewFlags(const GEState &state);
 
 	SimpleGLWindow *primaryWindow = nullptr;
 	SimpleGLWindow *secondWindow = nullptr;

Windows/GEDebugger/TabVertices.cpp

@@ -328,7 +328,7 @@ bool CtrlMatrixList::OnColPrePaint(int row, int col, LPNMLVCUSTOMDRAW msg) {
 	return false;
 }
 
-bool CtrlMatrixList::ColChanged(const GPUgstate &lastState, const GPUgstate &state, int row, int col) {
+bool CtrlMatrixList::ColChanged(const GEState &lastState, const GEState &state, int row, int col) {
 	union {
 		float f;
 		uint32_t u;
@@ -340,7 +340,7 @@ bool CtrlMatrixList::ColChanged(const GPUgstate &lastState, const GPUgstate &sta
 	return newVal.u != oldVal.u;
 }
 
-bool CtrlMatrixList::GetValue(const GPUgstate &state, int row, int col, float &val) {
+bool CtrlMatrixList::GetValue(const GEState &state, int row, int col, float &val) {
 	if (!gpu || row < 0 || row >= MATRIXLIST_ROW_COUNT || col < 0 || col >= MATRIXLIST_COL_COUNT)
 		return false;
 

Windows/GEDebugger/TabVertices.h

@@ -83,8 +83,8 @@ protected:
 	bool OnColPrePaint(int row, int col, LPNMLVCUSTOMDRAW msg) override;
 
 private:
-	bool GetValue(const GPUgstate &state, int row, int col, float &val);
-	bool ColChanged(const GPUgstate &lastState, const GPUgstate &state, int row, int col);
+	bool GetValue(const GEState &state, int row, int col, float &val);
+	bool ColChanged(const GEState &lastState, const GEState &state, int row, int col);
 	void ToggleBreakpoint(int row);
 	void PromptBreakpointCond(int row);
 };

unittest/TestVertexJit.cpp

@@ -27,6 +27,8 @@
 #include "unittest/TestVertexJit.h"
 #include "unittest/UnitTest.h"
 
+const UVScale g_uvScale{1.0f, 1.0f, 0.0f, 0.0f};
+
 class VertexDecoderTestHarness {
 	static const int BUFFER_SIZE = 64 * 65536;
 	static const int ROUNDS = 200;
@@ -37,9 +39,6 @@ public:
 		src_ = new u8[BUFFER_SIZE];
 		dst_ = new u8[BUFFER_SIZE];
 		cache_ = new VertexDecoderJitCache();
-
-		gstate_c.uv.uScale = 1.0f;
-		gstate_c.uv.vScale = 1.0f;
 	}
 	~VertexDecoderTestHarness() {
 		delete [] src_;
@@ -69,7 +68,7 @@ public:
 	void Execute(int vtype, int count, bool useJit) {
 		SetupExecute(vtype, useJit);
 
-		dec_->DecodeVerts(dst_, src_, &gstate_c.uv, count);
+		dec_->DecodeVerts(dst_, src_, &g_uvScale, count);
 	}
 
 	double ExecuteTimed(int vtype, int count, bool useJit) {
@@ -79,7 +78,7 @@ public:
 		double st = time_now_d();
 		do {
 			for (int j = 0; j < ROUNDS; ++j) {
-				dec_->DecodeVerts(dst_, src_, &gstate_c.uv, count);
+				dec_->DecodeVerts(dst_, src_, &g_uvScale, count);
 				++total;
 			}
 		} while (time_now_d() - st < 0.5);