Merge pull request #19743 from hrydgard/vertex-decoder-minor-cleanup

Vertex decoder code cleanup

GPU/Common/DrawEngineCommon.cpp

@@ -133,11 +133,10 @@ void DrawEngineCommon::NotifyConfigChanged() {
 
 	useHWTransform_ = g_Config.bHardwareTransform;
 	useHWTessellation_ = UpdateUseHWTessellation(g_Config.bHardwareTessellation);
-	decOptions_.applySkinInDecode = g_Config.bSoftwareSkinning;
 }
 
 u32 DrawEngineCommon::NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, u32 vertType, int *vertexSize) {
-	const u32 vertTypeID = GetVertTypeID(vertType, gstate.getUVGenMode(), decOptions_.applySkinInDecode);
+	const u32 vertTypeID = GetVertTypeID(vertType, gstate.getUVGenMode(), applySkinInDecode_);
 	VertexDecoder *dec = GetVertexDecoder(vertTypeID);
 	if (vertexSize)
 		*vertexSize = dec->VertexSize();
@@ -185,7 +184,7 @@ void DrawEngineCommon::DispatchSubmitImm(GEPrimitiveType prim, TransformedVertex
 	}
 
 	int bytesRead;
-	uint32_t vertTypeID = GetVertTypeID(vtype, 0, decOptions_.applySkinInDecode);
+	uint32_t vertTypeID = GetVertTypeID(vtype, 0, applySkinInDecode_);
 
 	bool clockwise = !gstate.isCullEnabled() || gstate.getCullMode() == cullMode;
 	SubmitPrim(&temp[0], nullptr, prim, vertexCount, vertTypeID, clockwise, &bytesRead);
@@ -303,10 +302,10 @@ bool DrawEngineCommon::TestBoundingBox(const void *vdata, const void *inds, int
 			}
 			// TODO: Avoid normalization if just plain skinning.
 			// Force software skinning.
-			bool wasApplyingSkinInDecode = decOptions_.applySkinInDecode;
-			decOptions_.applySkinInDecode = true;
+			bool wasApplyingSkinInDecode = applySkinInDecode_;
+			applySkinInDecode_ = true;
 			NormalizeVertices((u8 *)corners, temp_buffer, (const u8 *)vdata, indexLowerBound, indexUpperBound, vertType);
-			decOptions_.applySkinInDecode = wasApplyingSkinInDecode;
+			applySkinInDecode_ = wasApplyingSkinInDecode;
 
 			IndexConverter conv(vertType, inds);
 			for (int i = 0; i < vertexCount; i++) {
@@ -985,7 +984,7 @@ bool DrawEngineCommon::SubmitPrim(const void *verts, const void *inds, GEPrimiti
 		}
 	}
 
-	bool applySkin = (vertTypeID & GE_VTYPE_WEIGHT_MASK) && decOptions_.applySkinInDecode;
+	bool applySkin = dec_->skinInDecode;
 
 	DeferredInds &di = drawInds_[numDrawInds_++];
 	di.inds = inds;
@@ -1038,10 +1037,17 @@ bool DrawEngineCommon::SubmitPrim(const void *verts, const void *inds, GEPrimiti
 	return true;
 }
 
-void DrawEngineCommon::DecodeVerts(u8 *dest) {
+void DrawEngineCommon::BeginFrame() {
+	applySkinInDecode_ = g_Config.bSoftwareSkinning;
+}
+
+void DrawEngineCommon::DecodeVerts(VertexDecoder *dec, u8 *dest) {
+	if (!numDrawVerts_) {
+		return;
+	}
 	// Note that this should be able to continue a partial decode - we don't necessarily start from zero here (although we do most of the time).
 	int i = decodeVertsCounter_;
-	int stride = (int)dec_->GetDecVtxFmt().stride;
+	int stride = (int)dec->GetDecVtxFmt().stride;
 	for (; i < numDrawVerts_; i++) {
 		DeferredVerts &dv = drawVerts_[i];
 
@@ -1056,7 +1062,7 @@ void DrawEngineCommon::DecodeVerts(u8 *dest) {
 		}
 
 		// Decode the verts (and at the same time apply morphing/skinning). Simple.
-		dec_->DecodeVerts(dest + numDecodedVerts_ * stride, dv.verts, &dv.uvScale, indexLowerBound, indexUpperBound);
+		dec->DecodeVerts(dest + numDecodedVerts_ * stride, dv.verts, &dv.uvScale, indexLowerBound, indexUpperBound);
 		numDecodedVerts_ += indexUpperBound - indexLowerBound + 1;
 	}
 	decodeVertsCounter_ = i;

GPU/Common/DrawEngineCommon.h

@@ -53,12 +53,6 @@ enum FBOTexState {
 	FBO_TEX_READ_FRAMEBUFFER,
 };
 
-inline uint32_t GetVertTypeID(uint32_t vertType, int uvGenMode, bool skinInDecode) {
-	// As the decoder depends on the UVGenMode when we use UV prescale, we simply mash it
-	// into the top of the verttype where there are unused bits.
-	return (vertType & 0xFFFFFF) | (uvGenMode << 24) | (skinInDecode << 26);
-}
-
 struct SimpleVertex;
 namespace Spline { struct Weight2D; }
 
@@ -83,6 +77,8 @@ public:
 
 	void Init();
 
+	virtual void BeginFrame();
+
 	void SetGPUCommon(GPUCommon *gpuCommon) {
 		gpuCommon_ = gpuCommon;
 	}
@@ -114,9 +110,8 @@ public:
 	bool TestBoundingBoxThrough(const void *vdata, int vertexCount, u32 vertType);
 
 	void FlushSkin() {
-		bool applySkin = (lastVType_ & GE_VTYPE_WEIGHT_MASK) && decOptions_.applySkinInDecode;
-		if (applySkin) {
-			DecodeVerts(decoded_);
+		if (dec_->skinInDecode) {
+			DecodeVerts(dec_, decoded_);
 		}
 	}
 
@@ -160,7 +155,7 @@ protected:
 	virtual bool UpdateUseHWTessellation(bool enabled) const { return enabled; }
 	void UpdatePlanes();
 
-	void DecodeVerts(u8 *dest);
+	void DecodeVerts(VertexDecoder *dec, u8 *dest);
 	int DecodeInds();
 
 	// Preprocessing for spline/bezier
@@ -314,6 +309,8 @@ protected:
 	bool anyCCWOrIndexed_ = 0;
 	bool anyIndexed_ = 0;
 
+	bool applySkinInDecode_ = false;
+
 	// Vertex collector state
 	IndexGenerator indexGen;
 	int numDecodedVerts_ = 0;

GPU/Common/SplineCommon.cpp

@@ -507,7 +507,7 @@ void DrawEngineCommon::SubmitCurve(const void *control_points, const void *indic
 	if (indices)
 		GetIndexBounds(indices, num_points, vertType, &index_lower_bound, &index_upper_bound);
 
-	VertexDecoder *origVDecoder = GetVertexDecoder(GetVertTypeID(vertType, gstate.getUVGenMode(), decOptions_.applySkinInDecode));
+	VertexDecoder *origVDecoder = GetVertexDecoder(GetVertTypeID(vertType, gstate.getUVGenMode(), applySkinInDecode_));
 	*bytesRead = num_points * origVDecoder->VertexSize();
 
 	// Simplify away bones and morph before proceeding
@@ -574,7 +574,7 @@ void DrawEngineCommon::SubmitCurve(const void *control_points, const void *indic
 		gstate_c.uv.vOff = 0;
 	}
 
-	uint32_t vertTypeID = GetVertTypeID(vertTypeWithIndex16, gstate.getUVGenMode(), decOptions_.applySkinInDecode);
+	uint32_t vertTypeID = GetVertTypeID(vertTypeWithIndex16, gstate.getUVGenMode(), applySkinInDecode_);
 	int generatedBytesRead;
 	if (output.count)
 		DispatchSubmitPrim(output.vertices, output.indices, PatchPrimToPrim(surface.primType), output.count, vertTypeID, true, &generatedBytesRead);

GPU/Common/VertexDecoderCommon.cpp

@@ -1125,7 +1125,7 @@ void VertexDecoder::SetVertexType(u32 fmt, const VertexDecoderOptions &options,
 		DEBUG_LOG(Log::G3D, "VTYPE: THRU=%i TC=%i COL=%i POS=%i NRM=%i WT=%i NW=%i IDX=%i MC=%i", (int)throughmode, tc, col, pos, nrm, weighttype, nweights, idx, morphcount);
 	}
 
-	skinInDecode = weighttype != 0 && options.applySkinInDecode;
+	skinInDecode = weighttype != 0 && VertTypeIDSkinInDecode(fmt);
 
 	if (weighttype) { // && nweights?
 		weightoff = size;
@@ -1182,7 +1182,8 @@ void VertexDecoder::SetVertexType(u32 fmt, const VertexDecoderOptions &options,
 
 		// NOTE: That we check getUVGenMode here means that we must include it in the decoder ID!
 		// throughmode is automatically included though, because it's part of the vertType.
-		if (!throughmode && (gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_COORDS || gstate.getUVGenMode() == GE_TEXMAP_UNKNOWN)) {
+		GETexMapMode mode = VertTypeIDUVGenMode(fmt);
+		if (!throughmode && (mode == GE_TEXMAP_TEXTURE_COORDS || mode == GE_TEXMAP_UNKNOWN)) {
 			if (g_DoubleTextureCoordinates)
 				steps_[numSteps_++] = morphcount == 1 ? tcstep_prescale_remaster[tc] : tcstep_prescale_morph_remaster[tc];
 			else

GPU/Common/VertexDecoderCommon.h

@@ -325,9 +325,22 @@ typedef void (*JittedVertexDecoder)(const u8 *src, u8 *dst, int count, const UVS
 struct VertexDecoderOptions {
 	bool expandAllWeightsToFloat;
 	bool expand8BitNormalsToFloat;
-	bool applySkinInDecode;
 };
 
+inline uint32_t GetVertTypeID(uint32_t vertType, int uvGenMode, bool skinInDecode) {
+	// As the decoder depends on the UVGenMode when we use UV prescale, we simply mash it
+	// into the top of the verttype where there are unused bits.
+	return (vertType & 0xFFFFFF) | (uvGenMode << 24) | (skinInDecode << 26);
+}
+
+inline bool VertTypeIDSkinInDecode(uint32_t vertType) {
+	return ((vertType >> 26) & 1) != 0;
+}
+
+inline GETexMapMode VertTypeIDUVGenMode(uint32_t vertType) {
+	return (GETexMapMode)((vertType >> 24) & 3);
+}
+
 class VertexDecoder {
 public:
 	// A jit cache is not mandatory.
@@ -512,7 +525,6 @@ public:
 
 	// Returns a pointer to the code to run.
 	JittedVertexDecoder Compile(const VertexDecoder &dec, int32_t *jittedSize);
-	bool DescribeCodePtr(const u8 *ptr, std::string &name) const;
 	void Clear();
 
 	void Jit_WeightsU8();

GPU/D3D11/DrawEngineD3D11.cpp

@@ -239,6 +239,8 @@ ID3D11InputLayout *DrawEngineD3D11::SetupDecFmtForDraw(D3D11VertexShader *vshade
 }
 
 void DrawEngineD3D11::BeginFrame() {
+	DrawEngineCommon::BeginFrame();
+
 	pushVerts_->Reset();
 	pushInds_->Reset();
 
@@ -282,7 +284,7 @@ void DrawEngineD3D11::Flush() {
 		int vertexCount;
 		int maxIndex;
 		bool useElements;
-		DecodeVerts(decoded_);
+		DecodeVerts(dec_, decoded_);
 		DecodeIndsAndGetData(&prim, &vertexCount, &maxIndex, &useElements, false);
 		gpuStats.numUncachedVertsDrawn += vertexCount;
 
@@ -303,7 +305,7 @@ void DrawEngineD3D11::Flush() {
 
 		D3D11VertexShader *vshader;
 		D3D11FragmentShader *fshader;
-		shaderManager_->GetShaders(prim, dec_, &vshader, &fshader, pipelineState_, useHWTransform, useHWTessellation_, decOptions_.expandAllWeightsToFloat, decOptions_.applySkinInDecode);
+		shaderManager_->GetShaders(prim, dec_, &vshader, &fshader, pipelineState_, useHWTransform, useHWTessellation_, decOptions_.expandAllWeightsToFloat, applySkinInDecode_);
 		ID3D11InputLayout *inputLayout = SetupDecFmtForDraw(vshader, dec_->GetDecVtxFmt(), dec_->VertexType());
 		context_->PSSetShader(fshader->GetShader(), nullptr, 0);
 		context_->VSSetShader(vshader->GetShader(), nullptr, 0);
@@ -346,12 +348,15 @@ void DrawEngineD3D11::Flush() {
 		}
 	} else {
 		PROFILE_THIS_SCOPE("soft");
-		if (!decOptions_.applySkinInDecode) {
-			decOptions_.applySkinInDecode = true;
-			lastVType_ |= (1 << 26);
-			dec_ = GetVertexDecoder(lastVType_);
+		VertexDecoder *swDec = dec_;
+		if (swDec->nweights != 0) {
+			u32 withSkinning = lastVType_ | (1 << 26);
+			if (withSkinning != lastVType_) {
+				swDec = GetVertexDecoder(withSkinning);
+			}
 		}
-		DecodeVerts(decoded_);
+
+		DecodeVerts(swDec, decoded_);
 		int vertexCount = DecodeInds();
 
 		bool hasColor = (lastVType_ & GE_VTYPE_COL_MASK) != GE_VTYPE_COL_NONE;
@@ -400,7 +405,7 @@ void DrawEngineD3D11::Flush() {
 		const Lin::Vec3 scale(gstate_c.vpWidthScale, -gstate_c.vpHeightScale, gstate_c.vpDepthScale * 0.5f);
 		swTransform.SetProjMatrix(gstate.projMatrix, gstate_c.vpWidth < 0, gstate_c.vpHeight < 0, trans, scale);
 
-		swTransform.Transform(prim, dec_->VertexType(), dec_->GetDecVtxFmt(), numDecodedVerts_, &result);
+		swTransform.Transform(prim, swDec->VertexType(), swDec->GetDecVtxFmt(), numDecodedVerts_, &result);
 		// Non-zero depth clears are unusual, but some drivers don't match drawn depth values to cleared values.
 		// Games sometimes expect exact matches (see #12626, for example) for equal comparisons.
 		if (result.action == SW_CLEAR && everUsedEqualDepth_ && gstate.isClearModeDepthMask() && result.depth > 0.0f && result.depth < 1.0f)
@@ -416,7 +421,7 @@ void DrawEngineD3D11::Flush() {
 		ApplyDrawState(prim);
 
 		if (result.action == SW_NOT_READY)
-			swTransform.BuildDrawingParams(prim, vertexCount, dec_->VertexType(), inds, RemainingIndices(inds), numDecodedVerts_, VERTEX_BUFFER_MAX, &result);
+			swTransform.BuildDrawingParams(prim, vertexCount, swDec->VertexType(), inds, RemainingIndices(inds), numDecodedVerts_, VERTEX_BUFFER_MAX, &result);
 		if (result.setSafeSize)
 			framebufferManager_->SetSafeSize(result.safeWidth, result.safeHeight);
 
@@ -425,7 +430,7 @@ void DrawEngineD3D11::Flush() {
 		if (result.action == SW_DRAW_INDEXED) {
 			D3D11VertexShader *vshader;
 			D3D11FragmentShader *fshader;
-			shaderManager_->GetShaders(prim, dec_, &vshader, &fshader, pipelineState_, false, false, decOptions_.expandAllWeightsToFloat, true);
+			shaderManager_->GetShaders(prim, swDec, &vshader, &fshader, pipelineState_, false, false, decOptions_.expandAllWeightsToFloat, true);
 			context_->PSSetShader(fshader->GetShader(), nullptr, 0);
 			context_->VSSetShader(vshader->GetShader(), nullptr, 0);
 			shaderManager_->UpdateUniforms(framebufferManager_->UseBufferedRendering());
@@ -478,7 +483,6 @@ void DrawEngineD3D11::Flush() {
 				framebufferManager_->ApplyClearToMemory(scissorX1, scissorY1, scissorX2, scissorY2, clearColor);
 			}
 		}
-		decOptions_.applySkinInDecode = g_Config.bSoftwareSkinning;
 	}
 
 	ResetAfterDrawInline();

GPU/D3D11/DrawEngineD3D11.h

@@ -75,12 +75,12 @@ public:
 	void InitDeviceObjects();
 	void DestroyDeviceObjects();
 
-	void BeginFrame();
+	void BeginFrame() override;
 
 	void Flush() override;
 
 	void FinishDeferred() {
-		DecodeVerts(decoded_);
+		DecodeVerts(dec_, decoded_);
 	}
 
 	void NotifyConfigChanged() override;

GPU/Directx9/DrawEngineDX9.cpp

@@ -219,6 +219,7 @@ static uint32_t SwapRB(uint32_t c) {
 }
 
 void DrawEngineDX9::BeginFrame() {
+	DrawEngineCommon::BeginFrame();
 	lastRenderStepId_ = -1;
 }
 
@@ -256,7 +257,7 @@ void DrawEngineDX9::Flush() {
 		int vertexCount;
 		int maxIndex;
 		bool useElements;
-		DecodeVerts(decoded_);
+		DecodeVerts(dec_, decoded_);
 		DecodeIndsAndGetData(&prim, &vertexCount, &maxIndex, &useElements, false);
 		gpuStats.numUncachedVertsDrawn += vertexCount;
 
@@ -276,7 +277,7 @@ void DrawEngineDX9::Flush() {
 		ApplyDrawState(prim);
 		ApplyDrawStateLate();
 
-		VSShader *vshader = shaderManager_->ApplyShader(true, useHWTessellation_, dec_, decOptions_.expandAllWeightsToFloat, decOptions_.applySkinInDecode, pipelineState_);
+		VSShader *vshader = shaderManager_->ApplyShader(true, useHWTessellation_, dec_, decOptions_.expandAllWeightsToFloat, applySkinInDecode_, pipelineState_);
 		ComPtr<IDirect3DVertexDeclaration9> pHardwareVertexDecl;
 		SetupDecFmtForDraw(dec_->GetDecVtxFmt(), dec_->VertexType(), &pHardwareVertexDecl);
 
@@ -301,12 +302,14 @@ void DrawEngineDX9::Flush() {
 			}
 		}
 	} else {
-		if (!decOptions_.applySkinInDecode) {
-			decOptions_.applySkinInDecode = true;
-			lastVType_ |= (1 << 26);
-			dec_ = GetVertexDecoder(lastVType_);
+		VertexDecoder *swDec = dec_;
+		if (swDec->nweights != 0) {
+			u32 withSkinning = lastVType_ | (1 << 26);
+			if (withSkinning != lastVType_) {
+				swDec = GetVertexDecoder(withSkinning);
+			}
 		}
-		DecodeVerts(decoded_);
+		DecodeVerts(swDec, decoded_);
 		int vertexCount = DecodeInds();
 
 		bool hasColor = (lastVType_ & GE_VTYPE_COL_MASK) != GE_VTYPE_COL_NONE;
@@ -361,7 +364,7 @@ void DrawEngineDX9::Flush() {
 		const Lin::Vec3 scale(gstate_c.vpWidthScale, gstate_c.vpHeightScale, gstate_c.vpDepthScale * 0.5f);
 		swTransform.SetProjMatrix(gstate.projMatrix, gstate_c.vpWidth < 0, gstate_c.vpHeight > 0, trans, scale);
 
-		swTransform.Transform(prim, dec_->VertexType(), dec_->GetDecVtxFmt(), numDecodedVerts_, &result);
+		swTransform.Transform(prim, swDec->VertexType(), swDec->GetDecVtxFmt(), numDecodedVerts_, &result);
 		// Non-zero depth clears are unusual, but some drivers don't match drawn depth values to cleared values.
 		// Games sometimes expect exact matches (see #12626, for example) for equal comparisons.
 		if (result.action == SW_CLEAR && everUsedEqualDepth_ && gstate.isClearModeDepthMask() && result.depth > 0.0f && result.depth < 1.0f)
@@ -376,13 +379,13 @@ void DrawEngineDX9::Flush() {
 		ApplyDrawState(prim);
 
 		if (result.action == SW_NOT_READY)
-			swTransform.BuildDrawingParams(prim, vertexCount, dec_->VertexType(), inds, RemainingIndices(inds), numDecodedVerts_, VERTEX_BUFFER_MAX, &result);
+			swTransform.BuildDrawingParams(prim, vertexCount, swDec->VertexType(), inds, RemainingIndices(inds), numDecodedVerts_, VERTEX_BUFFER_MAX, &result);
 		if (result.setSafeSize)
 			framebufferManager_->SetSafeSize(result.safeWidth, result.safeHeight);
 
 		ApplyDrawStateLate();
 
-		VSShader *vshader = shaderManager_->ApplyShader(false, false, dec_, decOptions_.expandAllWeightsToFloat, true, pipelineState_);
+		VSShader *vshader = shaderManager_->ApplyShader(false, false, swDec, decOptions_.expandAllWeightsToFloat, true, pipelineState_);
 
 		if (result.action == SW_DRAW_INDEXED) {
 			if (result.setStencil) {
@@ -414,7 +417,6 @@ void DrawEngineDX9::Flush() {
 				framebufferManager_->ApplyClearToMemory(scissorX1, scissorY1, scissorX2, scissorY2, clearColor);
 			}
 		}
-		decOptions_.applySkinInDecode = g_Config.bSoftwareSkinning;
 	}
 
 	ResetAfterDrawInline();

GPU/Directx9/DrawEngineDX9.h

@@ -65,13 +65,13 @@ public:
 	void InitDeviceObjects();
 	void DestroyDeviceObjects();
 
-	void BeginFrame();
+	void BeginFrame() override;
 
 	// So that this can be inlined
 	void Flush() override;
 
 	void FinishDeferred() {
-		DecodeVerts(decoded_);
+		DecodeVerts(dec_, decoded_);
 	}
 
 protected:

GPU/GLES/DrawEngineGLES.cpp

@@ -147,6 +147,8 @@ void DrawEngineGLES::ClearInputLayoutMap() {
 }
 
 void DrawEngineGLES::BeginFrame() {
+	DrawEngineCommon::BeginFrame();
+
 	FrameData &frameData = frameData_[render_->GetCurFrame()];
 	frameData.pushIndex->Begin();
 	frameData.pushVertex->Begin();
@@ -264,7 +266,7 @@ void DrawEngineGLES::Flush() {
 
 	GEPrimitiveType prim = prevPrim_;
 
-	Shader *vshader = shaderManager_->ApplyVertexShader(CanUseHardwareTransform(prim), useHWTessellation_, dec_, decOptions_.expandAllWeightsToFloat, decOptions_.applySkinInDecode || !CanUseHardwareTransform(prim), &vsid);
+	Shader *vshader = shaderManager_->ApplyVertexShader(CanUseHardwareTransform(prim), useHWTessellation_, dec_, decOptions_.expandAllWeightsToFloat, applySkinInDecode_ || !CanUseHardwareTransform(prim), &vsid);
 
 	GLRBuffer *vertexBuffer = nullptr;
 	GLRBuffer *indexBuffer = nullptr;
@@ -272,9 +274,9 @@ void DrawEngineGLES::Flush() {
 	uint32_t indexBufferOffset = 0;
 
 	if (vshader->UseHWTransform()) {
-		if (decOptions_.applySkinInDecode && (lastVType_ & GE_VTYPE_WEIGHT_MASK)) {
+		if (applySkinInDecode_ && (lastVType_ & GE_VTYPE_WEIGHT_MASK)) {
 			// If software skinning, we're predecoding into "decoded". So make sure we're done, then push that content.
-			DecodeVerts(decoded_);
+			DecodeVerts(dec_, decoded_);
 			uint32_t size = numDecodedVerts_ * dec_->GetDecVtxFmt().stride;
 			u8 *dest = (u8 *)frameData.pushVertex->Allocate(size, 4, &vertexBuffer, &vertexBufferOffset);
 			memcpy(dest, decoded_, size);
@@ -282,13 +284,13 @@ void DrawEngineGLES::Flush() {
 			// Figure out how much pushbuffer space we need to allocate.
 			int vertsToDecode = ComputeNumVertsToDecode();
 			u8 *dest = (u8 *)frameData.pushVertex->Allocate(vertsToDecode * dec_->GetDecVtxFmt().stride, 4, &vertexBuffer, &vertexBufferOffset);
-			DecodeVerts(dest);
+			DecodeVerts(dec_, dest);
 		}
 
 		int vertexCount;
 		int maxIndex;
 		bool useElements;
-		DecodeVerts(decoded_);
+		DecodeVerts(dec_, decoded_);
 		DecodeIndsAndGetData(&prim, &vertexCount, &maxIndex, &useElements, false);
 
 		if (useElements) {
@@ -328,12 +330,14 @@ void DrawEngineGLES::Flush() {
 		}
 	} else {
 		PROFILE_THIS_SCOPE("soft");
-		if (!decOptions_.applySkinInDecode) {
-			decOptions_.applySkinInDecode = true;
-			lastVType_ |= (1 << 26);
-			dec_ = GetVertexDecoder(lastVType_);
+		VertexDecoder *swDec = dec_;
+		if (swDec->nweights != 0) {
+			u32 withSkinning = lastVType_ | (1 << 26);
+			if (withSkinning != lastVType_) {
+				swDec = GetVertexDecoder(withSkinning);
+			}
 		}
-		DecodeVerts(decoded_);
+		DecodeVerts(swDec, decoded_);
 		int vertexCount = DecodeInds();
 
 		bool hasColor = (lastVType_ & GE_VTYPE_COL_MASK) != GE_VTYPE_COL_NONE;
@@ -387,7 +391,7 @@ void DrawEngineGLES::Flush() {
 		const bool invertedY = gstate_c.vpHeight * (params.flippedY ? 1.0 : -1.0f) < 0;
 		swTransform.SetProjMatrix(gstate.projMatrix, gstate_c.vpWidth < 0, invertedY, trans, scale);
 
-		swTransform.Transform(prim, dec_->VertexType(), dec_->GetDecVtxFmt(), numDecodedVerts_, &result);
+		swTransform.Transform(prim, swDec->VertexType(), swDec->GetDecVtxFmt(), numDecodedVerts_, &result);
 		// Non-zero depth clears are unusual, but some drivers don't match drawn depth values to cleared values.
 		// Games sometimes expect exact matches (see #12626, for example) for equal comparisons.
 		if (result.action == SW_CLEAR && everUsedEqualDepth_ && gstate.isClearModeDepthMask() && result.depth > 0.0f && result.depth < 1.0f)
@@ -403,7 +407,7 @@ void DrawEngineGLES::Flush() {
 		ApplyDrawState(prim);
 
 		if (result.action == SW_NOT_READY)
-			swTransform.BuildDrawingParams(prim, vertexCount, dec_->VertexType(), inds, RemainingIndices(inds), numDecodedVerts_, VERTEX_BUFFER_MAX, &result);
+			swTransform.BuildDrawingParams(prim, vertexCount, swDec->VertexType(), inds, RemainingIndices(inds), numDecodedVerts_, VERTEX_BUFFER_MAX, &result);
 		if (result.setSafeSize)
 			framebufferManager_->SetSafeSize(result.safeWidth, result.safeHeight);
 
@@ -447,7 +451,6 @@ void DrawEngineGLES::Flush() {
 			}
 			gstate_c.Dirty(DIRTY_BLEND_STATE);  // Make sure the color mask gets re-applied.
 		}
-		decOptions_.applySkinInDecode = g_Config.bSoftwareSkinning;
 	}
 
 bail:

GPU/GLES/DrawEngineGLES.h

@@ -81,7 +81,7 @@ public:
 
 	void ClearTrackedVertexArrays() override {}
 
-	void BeginFrame();
+	void BeginFrame() override;
 	void EndFrame();
 
 	// So that this can be inlined

GPU/Software/TransformUnit.cpp

@@ -63,7 +63,7 @@ SoftwareDrawEngine::~SoftwareDrawEngine() {}
 
 void SoftwareDrawEngine::NotifyConfigChanged() {
 	DrawEngineCommon::NotifyConfigChanged();
-	decOptions_.applySkinInDecode = true;
+	applySkinInDecode_ = true;
 }
 
 void SoftwareDrawEngine::Flush() {
@@ -973,8 +973,8 @@ bool TransformUnit::GetCurrentSimpleVertices(int count, std::vector<GPUDebugVert
 
 	VertexDecoder vdecoder;
 	VertexDecoderOptions options{};
-	options.applySkinInDecode = true;
-	vdecoder.SetVertexType(gstate.vertType, options);
+	u32 vertTypeID = GetVertTypeID(gstate.vertType, gstate.getUVGenMode(), true);
+	vdecoder.SetVertexType(vertTypeID, options);
 
 	if (!Memory::IsValidRange(gstate_c.vertexAddr, (indexUpperBound + 1) * vdecoder.VertexSize()))
 		return false;

GPU/Vulkan/DrawEngineVulkan.cpp

@@ -164,6 +164,8 @@ void DrawEngineVulkan::DeviceRestore(Draw::DrawContext *draw) {
 }
 
 void DrawEngineVulkan::BeginFrame() {
+	DrawEngineCommon::BeginFrame();
+
 	lastPipeline_ = nullptr;
 
 	// These will be re-bound if needed, let's not let old bindings linger around too long.
@@ -258,9 +260,9 @@ void DrawEngineVulkan::Flush() {
 		uint32_t vbOffset;
 
 		VkBuffer vbuf = VK_NULL_HANDLE;
-		if (decOptions_.applySkinInDecode && (lastVType_ & GE_VTYPE_WEIGHT_MASK)) {
+		if (applySkinInDecode_ && (lastVType_ & GE_VTYPE_WEIGHT_MASK)) {
 			// If software skinning, we're predecoding into "decoded". So make sure we're done, then push that content.
-			DecodeVerts(decoded_);
+			DecodeVerts(dec_, decoded_);
 			VkDeviceSize size = numDecodedVerts_ * dec_->GetDecVtxFmt().stride;
 			u8 *dest = (u8 *)pushVertex_->Allocate(size, 4, &vbuf, &vbOffset);
 			memcpy(dest, decoded_, size);
@@ -269,7 +271,7 @@ void DrawEngineVulkan::Flush() {
 			int vertsToDecode = ComputeNumVertsToDecode();
 			// Decode directly into the pushbuffer
 			u8 *dest = pushVertex_->Allocate(vertsToDecode * dec_->GetDecVtxFmt().stride, 4, &vbuf, &vbOffset);
-			DecodeVerts(dest);
+			DecodeVerts(dec_, dest);
 		}
 
 		int vertexCount;
@@ -302,7 +304,7 @@ void DrawEngineVulkan::Flush() {
 			VulkanFragmentShader *fshader = nullptr;
 			VulkanGeometryShader *gshader = nullptr;
 
-			shaderManager_->GetShaders(prim, dec_, &vshader, &fshader, &gshader, pipelineState_, true, useHWTessellation_, decOptions_.expandAllWeightsToFloat, decOptions_.applySkinInDecode);
+			shaderManager_->GetShaders(prim, dec_, &vshader, &fshader, &gshader, pipelineState_, true, useHWTessellation_, decOptions_.expandAllWeightsToFloat, applySkinInDecode_);
 			_dbg_assert_msg_(vshader->UseHWTransform(), "Bad vshader");
 			VulkanPipeline *pipeline = pipelineManager_->GetOrCreatePipeline(renderManager, pipelineLayout_, pipelineKey_, &dec_->decFmt, vshader, fshader, gshader, true, 0, framebufferManager_->GetMSAALevel(), false);
 			if (!pipeline || !pipeline->pipeline) {
@@ -381,14 +383,16 @@ void DrawEngineVulkan::Flush() {
 		}
 	} else {
 		PROFILE_THIS_SCOPE("soft");
-		if (!decOptions_.applySkinInDecode) {
-			decOptions_.applySkinInDecode = true;
-			lastVType_ |= (1 << 26);
-			dec_ = GetVertexDecoder(lastVType_);
+		VertexDecoder *swDec = dec_;
+		if (swDec->nweights != 0) {
+			u32 withSkinning = lastVType_ | (1 << 26);
+			if (withSkinning != lastVType_) {
+				swDec = GetVertexDecoder(withSkinning);
+			}
 		}
 		int prevDecodedVerts = numDecodedVerts_;
 
-		DecodeVerts(decoded_);
+		DecodeVerts(swDec, decoded_);
 		int vertexCount = DecodeInds();
 
 		bool hasColor = (lastVType_ & GE_VTYPE_COL_MASK) != GE_VTYPE_COL_NONE;
@@ -444,7 +448,7 @@ void DrawEngineVulkan::Flush() {
 		const Lin::Vec3 scale(gstate_c.vpWidthScale, gstate_c.vpHeightScale, gstate_c.vpDepthScale * 0.5f);
 		swTransform.SetProjMatrix(gstate.projMatrix, gstate_c.vpWidth < 0, gstate_c.vpHeight < 0, trans, scale);
 
-		swTransform.Transform(prim, dec_->VertexType(), dec_->GetDecVtxFmt(), numDecodedVerts_, &result);
+		swTransform.Transform(prim, swDec->VertexType(), swDec->GetDecVtxFmt(), numDecodedVerts_, &result);
 		// Non-zero depth clears are unusual, but some drivers don't match drawn depth values to cleared values.
 		// Games sometimes expect exact matches (see #12626, for example) for equal comparisons.
 		if (result.action == SW_CLEAR && everUsedEqualDepth_ && gstate.isClearModeDepthMask() && result.depth > 0.0f && result.depth < 1.0f)
@@ -452,7 +456,7 @@ void DrawEngineVulkan::Flush() {
 
 		if (result.action == SW_NOT_READY) {
 			// decIndex_ here is always equal to inds currently, but it may not be in the future.
-			swTransform.BuildDrawingParams(prim, vertexCount, dec_->VertexType(), inds, RemainingIndices(inds), numDecodedVerts_, VERTEX_BUFFER_MAX, &result);
+			swTransform.BuildDrawingParams(prim, vertexCount, swDec->VertexType(), inds, RemainingIndices(inds), numDecodedVerts_, VERTEX_BUFFER_MAX, &result);
 		}
 
 		if (result.setSafeSize)
@@ -480,9 +484,9 @@ void DrawEngineVulkan::Flush() {
 				VulkanFragmentShader *fshader = nullptr;
 				VulkanGeometryShader *gshader = nullptr;
 
-				shaderManager_->GetShaders(prim, dec_, &vshader, &fshader, &gshader, pipelineState_, false, false, decOptions_.expandAllWeightsToFloat, true);
+				shaderManager_->GetShaders(prim, swDec, &vshader, &fshader, &gshader, pipelineState_, false, false, decOptions_.expandAllWeightsToFloat, true);
 				_dbg_assert_msg_(!vshader->UseHWTransform(), "Bad vshader");
-				VulkanPipeline *pipeline = pipelineManager_->GetOrCreatePipeline(renderManager, pipelineLayout_, pipelineKey_, &dec_->decFmt, vshader, fshader, gshader, false, 0, framebufferManager_->GetMSAALevel(), false);
+				VulkanPipeline *pipeline = pipelineManager_->GetOrCreatePipeline(renderManager, pipelineLayout_, pipelineKey_, &swDec->decFmt, vshader, fshader, gshader, false, 0, framebufferManager_->GetMSAALevel(), false);
 				if (!pipeline || !pipeline->pipeline) {
 					// Already logged, let's bail out.
 					ResetAfterDraw();
@@ -563,7 +567,6 @@ void DrawEngineVulkan::Flush() {
 				framebufferManager_->ApplyClearToMemory(scissorX1, scissorY1, scissorX2, scissorY2, result.color);
 			}
 		}
-		decOptions_.applySkinInDecode = g_Config.bSoftwareSkinning;
 	}
 
 	ResetAfterDrawInline();
@@ -581,7 +584,6 @@ void DrawEngineVulkan::ResetAfterDraw() {
 	vertexCountInDrawCalls_ = 0;
 	decodeIndsCounter_ = 0;
 	decodeVertsCounter_ = 0;
-	decOptions_.applySkinInDecode = g_Config.bSoftwareSkinning;
 	gstate_c.vertexFullAlpha = true;
 }
 

GPU/Vulkan/DrawEngineVulkan.h

@@ -136,7 +136,7 @@ public:
 		return pipelineLayout_;
 	}
 
-	void BeginFrame();
+	void BeginFrame() override;
 	void EndFrame();
 
 	void DirtyAllUBOs();

UI/ImDebugger/ImGe.cpp

@@ -1360,15 +1360,13 @@ void ImGeStateWindow::Draw(ImConfig &cfg, ImControl &control, GPUDebugInterface
 				if (!gpuDebug->GetCurrentSimpleVertices(rowCount_, vertices, indices)) {
 					rowCount_ = 0;
 				}
-				VertexDecoderOptions options{};
-				// TODO: Maybe an option?
-				options.applySkinInDecode = true;
-
 				auto buildVertexTable = [&](bool raw) {
 					// Ignore indices for now.
 					if (ImGui::BeginTable("rawverts", VERTEXLIST_COL_COUNT + 1, tableFlags)) {
 						static VertexDecoder decoder;
-						decoder.SetVertexType(state.vertType, options);
+						u32 vertTypeID = GetVertTypeID(state.vertType, state.getUVGenMode(), true);
+						VertexDecoderOptions options{};
+						decoder.SetVertexType(vertTypeID, options);
 
 						static const char * const colNames[] = {
 							"Index",

Windows/GEDebugger/TabVertices.cpp

@@ -175,8 +175,8 @@ int CtrlVertexList::GetRowCount() {
 	}
 	VertexDecoderOptions options{};
 	// TODO: Maybe an option?
-	options.applySkinInDecode = true;
-	decoder->SetVertexType(state.vertType, options);
+	u32 vertTypeID = GetVertTypeID(state.vertType, state.getUVGenMode(), true);
+	decoder->SetVertexType(vertTypeID, options);
 	return rowCount_;
 }
 

unittest/TestVertexJit.cpp

@@ -51,7 +51,7 @@ public:
 	void Reset() {
 		memset(src_, 0, BUFFER_SIZE);
 		memset(dst_, 0, BUFFER_SIZE);
-		memset(&options_, 0, sizeof(options_));
+		options_ = {};
 		delete dec_;
 		dec_ = nullptr;
 		indexLowerBound_ = 0;
@@ -65,7 +65,7 @@ public:
 		if (needsReset_) {
 			Reset();
 		}
-		memcpy(&options_, &opts, sizeof(options_));
+		options_ = opts;
 	}
 
 	void SetIndexLowerBound(const int lower) {
@@ -543,7 +543,6 @@ static bool TestVertexColor565() {
 static bool TestVertex8Skin() {
 	VertexDecoderTestHarness dec;
 	VertexDecoderOptions opts{};
-	opts.applySkinInDecode = true;
 	dec.SetOptions(opts);
 
 	for (int i = 0; i < 8 * 12; ++i) {
@@ -558,13 +557,14 @@ static bool TestVertex8Skin() {
 	gstate.boneMatrix[20] = 5.0f;
 
 	int vtype = GE_VTYPE_POS_8BIT | GE_VTYPE_NRM_8BIT | GE_VTYPE_WEIGHT_8BIT | (1 << GE_VTYPE_WEIGHTCOUNT_SHIFT);
+	u32 vertTypeID = GetVertTypeID(vtype, 0, true);  // enable skinning in decode
 
 	dec.Add8(128 + 64, 128 - 64);
 	dec.Add8(127, 0, 128);
 	dec.Add8(127, 0, 128);
 
 	for (int jit = 0; jit <= 1; ++jit) {
-		dec.Execute(vtype, 0, jit == 1);
+		dec.Execute(vertTypeID, 0, jit == 1);
 		dec.AssertFloat("TestVertex8Skin-Nrm", (2.0f * 1.5f + 1.0f * 0.5f) * 127.0f / 128.0f, 0.0f, 2.0f * 5.0f * -1.0f);
 		dec.AssertFloat("TestVertex8Skin-Pos", (2.0f * 1.5f + 1.0f * 0.5f) * 127.0f / 128.0f, 0.0f, 2.0f * 5.0f * -1.0f);
 	}
@@ -575,7 +575,6 @@ static bool TestVertex8Skin() {
 static bool TestVertex16Skin() {
 	VertexDecoderTestHarness dec;
 	VertexDecoderOptions opts{};
-	opts.applySkinInDecode = true;
 	dec.SetOptions(opts);
 
 	for (int i = 0; i < 8 * 12; ++i) {
@@ -590,13 +589,14 @@ static bool TestVertex16Skin() {
 	gstate.boneMatrix[20] = 5.0f;
 
 	int vtype = GE_VTYPE_POS_16BIT | GE_VTYPE_NRM_16BIT | GE_VTYPE_WEIGHT_16BIT | (1 << GE_VTYPE_WEIGHTCOUNT_SHIFT);
+	u32 vertTypeID = GetVertTypeID(vtype, 0, true);  // enable skinning in decode
 
 	dec.Add16(32768 + 16384, 32768 - 16384);
 	dec.Add16(32767, 0, 32768);
 	dec.Add16(32767, 0, 32768);
 
 	for (int jit = 0; jit <= 1; ++jit) {
-		dec.Execute(vtype, 0, jit == 1);
+		dec.Execute(vertTypeID, 0, jit == 1);
 		dec.AssertFloat("TestVertex16Skin-Nrm", (2.0f * 1.5f + 1.0f * 0.5f) * 32767.0f / 32768.0f, 0.0f, 2.0f * 5.0f * -1.0f);
 		dec.AssertFloat("TestVertex16Skin-Pos", (2.0f * 1.5f + 1.0f * 0.5f) * 32767.0f / 32768.0f, 0.0f, 2.0f * 5.0f * -1.0f);
 	}
@@ -607,7 +607,6 @@ static bool TestVertex16Skin() {
 static bool TestVertexFloatSkin() {
 	VertexDecoderTestHarness dec;
 	VertexDecoderOptions opts{};
-	opts.applySkinInDecode = true;
 	dec.SetOptions(opts);
 
 	for (int i = 0; i < 8 * 12; ++i) {
@@ -622,13 +621,14 @@ static bool TestVertexFloatSkin() {
 	gstate.boneMatrix[20] = 5.0f;
 
 	int vtype = GE_VTYPE_POS_FLOAT | GE_VTYPE_NRM_FLOAT | GE_VTYPE_WEIGHT_FLOAT | (1 << GE_VTYPE_WEIGHTCOUNT_SHIFT);
+	u32 vertTypeID = GetVertTypeID(vtype, 0, true);  // enable skinning in decode
 
 	dec.AddFloat(1.5f, 0.5f);
 	dec.AddFloat(1.0f, 0, -1.0f);
 	dec.AddFloat(1.0f, 0, -1.0f);
 
 	for (int jit = 0; jit <= 1; ++jit) {
-		dec.Execute(vtype, 0, jit == 1);
+		dec.Execute(vertTypeID, 0, jit == 1);
 		dec.AssertFloat("TestVertexFloatSkin-Nrm", (2.0f * 1.5f + 1.0f * 0.5f) * 1.0f, 0.0f, 2.0f * 5.0f * -1.0f);
 		dec.AssertFloat("TestVertexFloatSkin-Pos", (2.0f * 1.5f + 1.0f * 0.5f) * 1.0f, 0.0f, 2.0f * 5.0f * -1.0f);
 	}