Centralize EstimatePerVertexCost and Execute_BlockTransferStart

GPU/Common/DrawEngineCommon.h

@@ -22,6 +22,7 @@
 
 #include "Common/CommonTypes.h"
 
+#include "GPU/GPUState.h"
 #include "GPU/Common/GPUDebugInterface.h"
 #include "GPU/Common/VertexDecoderCommon.h"
 
@@ -56,6 +57,36 @@ public:
 
 	std::vector<std::string> DebugGetVertexLoaderIDs();
 	std::string DebugGetVertexLoaderString(std::string id, DebugShaderStringType stringType);
+
+	int EstimatePerVertexCost() {
+		// TODO: This is transform cost, also account for rasterization cost somehow... although it probably
+		// runs in parallel with transform.
+
+		// Also, this is all pure guesswork. If we can find a way to do measurements, that would be great.
+
+		// GTA wants a low value to run smooth, GoW wants a high value (otherwise it thinks things
+		// went too fast and starts doing all the work over again).
+
+		int cost = 20;
+		if (gstate.isLightingEnabled()) {
+			cost += 10;
+
+			for (int i = 0; i < 4; i++) {
+				if (gstate.isLightChanEnabled(i))
+					cost += 10;
+			}
+		}
+
+		if (gstate.getUVGenMode() != GE_TEXMAP_TEXTURE_COORDS) {
+			cost += 20;
+		}
+		int morphCount = gstate.getNumMorphWeights();
+		if (morphCount > 1) {
+			cost += 5 * morphCount;
+		}
+		return cost;
+	}
+
 protected:
 	// Preprocessing for spline/bezier
 	u32 NormalizeVertices(u8 *outPtr, u8 *bufPtr, const u8 *inPtr, int lowerBound, int upperBound, u32 vertType);

GPU/Directx9/DrawEngineDX9.h

@@ -135,36 +135,6 @@ public:
 	void SetupVertexDecoder(u32 vertType);
 	void SetupVertexDecoderInternal(u32 vertType);
 
-	// This requires a SetupVertexDecoder call first.
-	int EstimatePerVertexCost() {
-		// TODO: This is transform cost, also account for rasterization cost somehow... although it probably
-		// runs in parallel with transform.
-
-		// Also, this is all pure guesswork. If we can find a way to do measurements, that would be great.
-
-		// GTA wants a low value to run smooth, GoW wants a high value (otherwise it thinks things
-		// went too fast and starts doing all the work over again).
-
-		int cost = 20;
-		if (gstate.isLightingEnabled()) {
-			cost += 10;
-
-			for (int i = 0; i < 4; i++) {
-				if (gstate.isLightChanEnabled(i))
-					cost += 10;
-			}
-		}
-
-		if (gstate.getUVGenMode() != GE_TEXMAP_TEXTURE_COORDS) {
-			cost += 20;
-		}
-		if (dec_ && dec_->morphcount > 1) {
-			cost += 5 * dec_->morphcount;
-		}
-
-		return cost;
-	}
-
 	// So that this can be inlined
 	void Flush() {
 		if (!numDrawCalls)

GPU/Directx9/GPU_DX9.cpp

@@ -332,7 +332,7 @@ static const CommandTableEntry commandTable[] = {
 
 	// Changes that trigger data copies. Only flushing on change for LOADCLUT must be a bit of a hack...
 	{GE_CMD_LOADCLUT, FLAG_FLUSHBEFOREONCHANGE | FLAG_EXECUTE, &GPU_DX9::Execute_LoadClut},
-	{GE_CMD_TRANSFERSTART, FLAG_FLUSHBEFORE | FLAG_EXECUTE | FLAG_READS_PC},
+	{GE_CMD_TRANSFERSTART, FLAG_FLUSHBEFORE | FLAG_EXECUTE | FLAG_READS_PC, &GPU_DX9::Execute_BlockTransferStart},
 
 	// We don't use the dither table.
 	{GE_CMD_DITH0},
@@ -1311,7 +1311,6 @@ void GPU_DX9::Execute_Generic(u32 op, u32 diff) {
 		Execute_Prim(op, diff);
 		break;
 
-		// The arrow and other rotary items in Puzbob are bezier patches, strangely enough.
 	case GE_CMD_BEZIER:
 		Execute_Bezier(op, diff);
 		break;
@@ -1452,17 +1451,9 @@ void GPU_DX9::Execute_Generic(u32 op, u32 diff) {
 	case GE_CMD_TRANSFERSIZE:
 		break;
 
-	case GE_CMD_TRANSFERSTART:  // Orphis calls this TRXKICK
-		{
-			// TODO: Here we should check if the transfer overlaps a framebuffer or any textures,
-			// and take appropriate action. This is a block transfer between RAM and VRAM, or vice versa.
-			// Can we skip this entirely on SkipDraw? It skips some things internally.
-			DoBlockTransfer(gstate_c.skipDrawReason);
-
-			// Fixes Gran Turismo's funky text issue, since it overwrites the current texture.
-			gstate_c.textureChanged = TEXCHANGE_UPDATED;
-			break;
-		}
+	case GE_CMD_TRANSFERSTART:
+		Execute_BlockTransferStart(op, diff);
+		break;
 
 	case GE_CMD_TEXSIZE0:
 		Execute_TexSize0(op, diff);

GPU/Directx9/ShaderManagerDX9.cpp

@@ -272,7 +272,7 @@ static void ConvertProjMatrixToD3DThrough(Matrix4x4 &in) {
 	in.translateAndScale(Vec3(xoff, yoff, 0.5f), Vec3(1.0f, 1.0f, 0.5f));
 }
 
-void ShaderManagerDX9::PSUpdateUniforms(int dirtyUniforms) {
+void ShaderManagerDX9::PSUpdateUniforms(u32 dirtyUniforms) {
 	if (dirtyUniforms & DIRTY_TEXENV) {
 		PSSetColorUniform3(CONST_PS_TEXENV, gstate.texenvcolor);
 	}
@@ -324,7 +324,7 @@ void ShaderManagerDX9::PSUpdateUniforms(int dirtyUniforms) {
 	}
 }
 
-void ShaderManagerDX9::VSUpdateUniforms(int dirtyUniforms) {
+void ShaderManagerDX9::VSUpdateUniforms(u32 dirtyUniforms) {
 	// Update any dirty uniforms before we draw
 	if (dirtyUniforms & DIRTY_PROJMATRIX) {
 		Matrix4x4 flippedMatrix;

GPU/Directx9/ShaderManagerDX9.h

@@ -137,8 +137,8 @@ public:
 	std::string DebugGetShaderString(std::string id, DebugShaderType type, DebugShaderStringType stringType);
 
 private:
-	void PSUpdateUniforms(int dirtyUniforms);
-	void VSUpdateUniforms(int dirtyUniforms);
+	void PSUpdateUniforms(u32 dirtyUniforms);
+	void VSUpdateUniforms(u32 dirtyUniforms);
 	void PSSetColorUniform3Alpha255(int creg, u32 color, u8 alpha);
 	void PSSetColorUniform3(int creg, u32 color);
 	void PSSetFloat(int creg, float value);

GPU/GLES/DrawEngineGLES.h

@@ -139,36 +139,6 @@ public:
 	void SetupVertexDecoder(u32 vertType);
 	inline void SetupVertexDecoderInternal(u32 vertType);
 
-	// This requires a SetupVertexDecoder call first.
-	int EstimatePerVertexCost() {
-		// TODO: This is transform cost, also account for rasterization cost somehow... although it probably
-		// runs in parallel with transform.
-
-		// Also, this is all pure guesswork. If we can find a way to do measurements, that would be great.
-
-		// GTA wants a low value to run smooth, GoW wants a high value (otherwise it thinks things
-		// went too fast and starts doing all the work over again).
-
-		int cost = 20;
-		if (gstate.isLightingEnabled()) {
-			cost += 10;
-
-			for (int i = 0; i < 4; i++) {
-				if (gstate.isLightChanEnabled(i))
-					cost += 10;
-			}
-		}
-
-		if (gstate.getUVGenMode() != GE_TEXMAP_TEXTURE_COORDS) {
-			cost += 20;
-		}
-		if (dec_ && dec_->morphcount > 1) {
-			cost += 5 * dec_->morphcount;
-		}
-
-		return cost;
-	}
-
 	// So that this can be inlined
 	void Flush() {
 		if (!numDrawCalls)

GPU/GLES/GPU_GLES.cpp

@@ -335,7 +335,7 @@ static const CommandTableEntry commandTable[] = {
 
 	// Changes that trigger data copies. Only flushing on change for LOADCLUT must be a bit of a hack...
 	{GE_CMD_LOADCLUT, FLAG_FLUSHBEFOREONCHANGE | FLAG_EXECUTE, 0, &GPU_GLES::Execute_LoadClut},
-	{GE_CMD_TRANSFERSTART, FLAG_FLUSHBEFORE | FLAG_EXECUTE | FLAG_READS_PC, 0, &GPU_GLES::Execute_BlockTransferStart},
+	{GE_CMD_TRANSFERSTART, FLAG_FLUSHBEFORE | FLAG_EXECUTE | FLAG_READS_PC, 0, &GPUCommon::Execute_BlockTransferStart},
 
 	// We don't use the dither table.
 	{GE_CMD_DITH0},
@@ -1512,16 +1512,6 @@ void GPU_GLES::Execute_BoneMtxData(u32 op, u32 diff) {
 	gstate.boneMatrixNumber = (GE_CMD_BONEMATRIXNUMBER << 24) | (num & 0x7F);
 }
 
-void GPU_GLES::Execute_BlockTransferStart(u32 op, u32 diff) {
-	// TODO: Here we should check if the transfer overlaps a framebuffer or any textures,
-	// and take appropriate action. This is a block transfer between RAM and VRAM, or vice versa.
-	// Can we skip this on SkipDraw?
-	DoBlockTransfer(gstate_c.skipDrawReason);
-
-	// Fixes Gran Turismo's funky text issue, since it overwrites the current texture.
-	gstate_c.textureChanged = TEXCHANGE_UPDATED;
-}
-
 void GPU_GLES::Execute_Generic(u32 op, u32 diff) {
 	u32 cmd = op >> 24;
 	u32 data = op & 0xFFFFFF;
@@ -1543,7 +1533,6 @@ void GPU_GLES::Execute_Generic(u32 op, u32 diff) {
 		Execute_Prim(op, diff);
 		break;
 
-	// The arrow and other rotary items in Puzbob are bezier patches, strangely enough.
 	case GE_CMD_BEZIER:
 		Execute_Bezier(op, diff);
 		break;
@@ -1692,7 +1681,7 @@ void GPU_GLES::Execute_Generic(u32 op, u32 diff) {
 	case GE_CMD_TRANSFERSIZE:
 		break;
 
-	case GE_CMD_TRANSFERSTART:  // Orphis calls this TRXKICK
+	case GE_CMD_TRANSFERSTART:
 		Execute_BlockTransferStart(op, diff);
 		break;
 

GPU/GLES/GPU_GLES.h

@@ -140,7 +140,6 @@ public:
 	void Execute_TgenMtxData(u32 op, u32 diff);
 	void Execute_BoneMtxNum(u32 op, u32 diff);
 	void Execute_BoneMtxData(u32 op, u32 diff);
-	void Execute_BlockTransferStart(u32 op, u32 diff);
 
 	// Using string because it's generic - makes no assumptions on the size of the shader IDs of this backend.
 	std::vector<std::string> DebugGetShaderIDs(DebugShaderType shader) override;

GPU/GLES/ShaderManagerGLES.cpp

@@ -547,12 +547,6 @@ void LinkedShader::UpdateUniforms(u32 vertType, const ShaderID &vsid) {
 		glUniform2fv(u_fogcoef, 1, fogcoef);
 	}
 
-	// Texturing
-
-	// If this dirty check is changed to true, Frontier Gate Boost works in texcoord speedhack mode.
-	// This means that it's not a flushing issue.
-	// It uses GE_TEXMAP_TEXTURE_MATRIX with GE_PROJMAP_UV a lot.
-	// Can't figure out why it doesn't dirty at the right points though...
 	if (dirty & DIRTY_UVSCALEOFFSET) {
 		const float invW = 1.0f / (float)gstate_c.curTextureWidth;
 		const float invH = 1.0f / (float)gstate_c.curTextureHeight;

GPU/GPUCommon.cpp

@@ -1153,6 +1153,16 @@ void GPUCommon::Execute_BoundingBox(u32 op, u32 diff) {
 	}
 }
 
+void GPUCommon::Execute_BlockTransferStart(u32 op, u32 diff) {
+	// TODO: Here we should check if the transfer overlaps a framebuffer or any textures,
+	// and take appropriate action. This is a block transfer between RAM and VRAM, or vice versa.
+	// Can we skip this on SkipDraw?
+	DoBlockTransfer(gstate_c.skipDrawReason);
+
+	// Fixes Gran Turismo's funky text issue, since it overwrites the current texture.
+	gstate_c.textureChanged = TEXCHANGE_UPDATED;
+}
+
 void GPUCommon::ExecuteOp(u32 op, u32 diff) {
 	const u32 cmd = op >> 24;
 

GPU/GPUCommon.h

@@ -83,6 +83,7 @@ public:
 	void Execute_Bezier(u32 op, u32 diff);
 	void Execute_Spline(u32 op, u32 diff);
 	void Execute_BoundingBox(u32 op, u32 diff);
+	void Execute_BlockTransferStart(u32 op, u32 diff);
 
 	u64 GetTickEstimate() override {
 #if defined(_M_X64) || defined(__ANDROID__)

GPU/GPUState.h

@@ -393,6 +393,7 @@ struct GPUgstate {
 	bool isModeThrough() const { return (vertType & GE_VTYPE_THROUGH) != 0; }
 	bool areNormalsReversed() const { return reversenormals & 1; }
 	bool isSkinningEnabled() const { return ((vertType & GE_VTYPE_WEIGHT_MASK) != GE_VTYPE_WEIGHT_NONE); }
+	int getNumMorphWeights() const { return ((vertType & GE_VTYPE_MORPHCOUNT_MASK) >> GE_VTYPE_MORPHCOUNT_SHIFT) + 1; }
 
 	GEPatchPrimType getPatchPrimitiveType() const { return static_cast<GEPatchPrimType>(patchprimitive & 3); }
 	bool isPatchNormalsReversed() const { return patchfacing & 1; }

GPU/GeDisasm.cpp

@@ -135,7 +135,6 @@ void GeDisassembleOp(u32 pc, u32 op, u32 prev, char *buffer, int bufsize) {
 		}
 		break;
 
-	// The arrow and other rotary items in Puzbob are bezier patches, strangely enough.
 	case GE_CMD_BEZIER:
 		{
 			int bz_ucount = data & 0xFF;
@@ -521,7 +520,7 @@ void GeDisassembleOp(u32 pc, u32 op, u32 prev, char *buffer, int bufsize) {
 			break;
 		}
 
-	case GE_CMD_TRANSFERSTART:  // Orphis calls this TRXKICK
+	case GE_CMD_TRANSFERSTART:
 		if (data & ~1)
 			snprintf(buffer, bufsize, "Block transfer start: %d (extra %x)", data & 1, data & ~1);
 		else

GPU/Null/NullGpu.cpp

@@ -77,7 +77,6 @@ void NullGPU::ExecuteOp(u32 op, u32 diff) {
 		}
 		break;
 
-	// The arrow and other rotary items in Puzbob are bezier patches, strangely enough.
 	case GE_CMD_BEZIER:
 		{
 			int bz_ucount = data & 0xFF;
@@ -92,7 +91,6 @@ void NullGPU::ExecuteOp(u32 op, u32 diff) {
 			int sp_vcount = (data >> 8) & 0xFF;
 			int sp_utype = (data >> 16) & 0x3;
 			int sp_vtype = (data >> 18) & 0x3;
-			//drawSpline(sp_ucount, sp_vcount, sp_utype, sp_vtype);
 			DEBUG_LOG(G3D,"DL DRAW SPLINE: %i x %i, %i x %i", sp_ucount, sp_vcount, sp_utype, sp_vtype);
 		}
 		break;

GPU/Software/SoftGpu.h

@@ -21,7 +21,7 @@
 #include "GPU/Common/GPUDebugInterface.h"
 #include "thin3d/thin3d.h"
 
-typedef struct {
+struct FormatBuffer {
 	union {
 		u8 *data;
 		u16 *as16;
@@ -43,7 +43,7 @@ typedef struct {
 	inline u32 Get32(int x, int y, int stride) {
 		return as32[x + y * stride];
 	}
-} FormatBuffer;
+};
 
 class ShaderManagerGLES;
 

GPU/Vulkan/DrawEngineVulkan.h

@@ -94,36 +94,6 @@ public:
 	void SetupVertexDecoder(u32 vertType);
 	void SetupVertexDecoderInternal(u32 vertType);
 
-	// This requires a SetupVertexDecoder call first.
-	int EstimatePerVertexCost() {
-		// TODO: This is transform cost, also account for rasterization cost somehow... although it probably
-		// runs in parallel with transform.
-
-		// Also, this is all pure guesswork. If we can find a way to do measurements, that would be great.
-
-		// GTA wants a low value to run smooth, GoW wants a high value (otherwise it thinks things
-		// went too fast and starts doing all the work over again).
-
-		int cost = 20;
-		if (gstate.isLightingEnabled()) {
-			cost += 10;
-
-			for (int i = 0; i < 4; i++) {
-				if (gstate.isLightChanEnabled(i))
-					cost += 10;
-			}
-		}
-
-		if (gstate.getUVGenMode() != GE_TEXMAP_TEXTURE_COORDS) {
-			cost += 20;
-		}
-		if (dec_ && dec_->morphcount > 1) {
-			cost += 5 * dec_->morphcount;
-		}
-
-		return cost;
-	}
-
 	// So that this can be inlined
 	void Flush(VkCommandBuffer cmd) {
 		if (!numDrawCalls)

GPU/Vulkan/GPU_Vulkan.cpp

@@ -1301,16 +1301,6 @@ void GPU_Vulkan::Execute_BoneMtxData(u32 op, u32 diff) {
 	gstate.boneMatrixNumber = (GE_CMD_BONEMATRIXNUMBER << 24) | (num & 0x7F);
 }
 
-void GPU_Vulkan::Execute_BlockTransferStart(u32 op, u32 diff) {
-	// TODO: Here we should check if the transfer overlaps a framebuffer or any textures,
-	// and take appropriate action. This is a block transfer between RAM and VRAM, or vice versa.
-	// Can we skip this on SkipDraw?
-	DoBlockTransfer(gstate_c.skipDrawReason);
-
-	// Fixes Gran Turismo's funky text issue, since it overwrites the current texture.
-	gstate_c.textureChanged = TEXCHANGE_UPDATED;
-}
-
 void GPU_Vulkan::Execute_Generic(u32 op, u32 diff) {
 	u32 cmd = op >> 24;
 	u32 data = op & 0xFFFFFF;
@@ -1332,7 +1322,6 @@ void GPU_Vulkan::Execute_Generic(u32 op, u32 diff) {
 		Execute_Prim(op, diff);
 		break;
 
-		// The arrow and other rotary items in Puzbob are bezier patches, strangely enough.
 	case GE_CMD_BEZIER:
 		Execute_Bezier(op, diff);
 		break;
@@ -1481,7 +1470,7 @@ void GPU_Vulkan::Execute_Generic(u32 op, u32 diff) {
 	case GE_CMD_TRANSFERSIZE:
 		break;
 
-	case GE_CMD_TRANSFERSTART:  // Orphis calls this TRXKICK
+	case GE_CMD_TRANSFERSTART:
 		Execute_BlockTransferStart(op, diff);
 		break;
 

GPU/Vulkan/GPU_Vulkan.h

@@ -127,7 +127,6 @@ public:
 	void Execute_TgenMtxData(u32 op, u32 diff);
 	void Execute_BoneMtxNum(u32 op, u32 diff);
 	void Execute_BoneMtxData(u32 op, u32 diff);
-	void Execute_BlockTransferStart(u32 op, u32 diff);
 
 	// Using string because it's generic - makes no assumptions on the size of the shader IDs of this backend.
 	std::vector<std::string> DebugGetShaderIDs(DebugShaderType shader) override;

Qt/Debugger/debugger_displaylist.cpp

@@ -674,7 +674,7 @@ QString Debugger_DisplayList::DisassembleOp(u32 pc, u32 op, u32 prev, const GPUg
 			return QString("Block Transfer Rect Size: %1 x %2").arg(w).arg(h);
 		}
 
-	case GE_CMD_TRANSFERSTART:  // Orphis calls this TRXKICK
+	case GE_CMD_TRANSFERSTART:
 		{
 		return QString("Block Transfer Start : %1").arg(data ? "32-bit texel size" : "16-bit texel size");
 		}