Merge pull request #16132 from unknownbrackets/softgpu-blend

softjit: Fix dst blend shift

GPU/GPUCommon.cpp

@@ -780,6 +780,9 @@ void GPUCommon::ResetMatrices() {
 		matrixVisible.proj[i] = toFloat24(gstate.projMatrix[i]);
 	for (size_t i = 0; i < ARRAY_SIZE(gstate.tgenMatrix); ++i)
 		matrixVisible.tgen[i] = toFloat24(gstate.tgenMatrix[i]);
+
+	// Assume all the matrices changed, so dirty things related to them.
+	gstate_c.Dirty(DIRTY_WORLDMATRIX | DIRTY_VIEWMATRIX | DIRTY_PROJMATRIX | DIRTY_TEXMATRIX | DIRTY_FRAGMENTSHADER_STATE | DIRTY_BONE_UNIFORMS);
 }
 
 u32 GPUCommon::EnqueueList(u32 listpc, u32 stall, int subIntrBase, PSPPointer<PspGeListArgs> args, bool head) {
@@ -1341,7 +1344,8 @@ void GPUCommon::Execute_Iaddr(u32 op, u32 diff) {
 }
 
 void GPUCommon::Execute_Origin(u32 op, u32 diff) {
-	gstate_c.offsetAddr = currentList->pc;
+	if (currentList)
+		gstate_c.offsetAddr = currentList->pc;
 }
 
 void GPUCommon::Execute_Jump(u32 op, u32 diff) {
@@ -2175,6 +2179,11 @@ void GPUCommon::Execute_BlockTransferStart(u32 op, u32 diff) {
 }
 
 void GPUCommon::Execute_WorldMtxNum(u32 op, u32 diff) {
+	if (!currentList) {
+		gstate.worldmtxnum = (GE_CMD_WORLDMATRIXNUMBER << 24) | (op & 0xF);
+		return;
+	}
+
 	// This is almost always followed by GE_CMD_WORLDMATRIXDATA.
 	const u32_le *src = (const u32_le *)Memory::GetPointerUnchecked(currentList->pc + 4);
 	u32 *dst = (u32 *)(gstate.worldMatrix + (op & 0xF));
@@ -2225,6 +2234,11 @@ void GPUCommon::Execute_WorldMtxData(u32 op, u32 diff) {
 }
 
 void GPUCommon::Execute_ViewMtxNum(u32 op, u32 diff) {
+	if (!currentList) {
+		gstate.viewmtxnum = (GE_CMD_VIEWMATRIXNUMBER << 24) | (op & 0xF);
+		return;
+	}
+
 	// This is almost always followed by GE_CMD_VIEWMATRIXDATA.
 	const u32_le *src = (const u32_le *)Memory::GetPointerUnchecked(currentList->pc + 4);
 	u32 *dst = (u32 *)(gstate.viewMatrix + (op & 0xF));
@@ -2273,6 +2287,11 @@ void GPUCommon::Execute_ViewMtxData(u32 op, u32 diff) {
 }
 
 void GPUCommon::Execute_ProjMtxNum(u32 op, u32 diff) {
+	if (!currentList) {
+		gstate.projmtxnum = (GE_CMD_PROJMATRIXNUMBER << 24) | (op & 0xF);
+		return;
+	}
+
 	// This is almost always followed by GE_CMD_PROJMATRIXDATA.
 	const u32_le *src = (const u32_le *)Memory::GetPointerUnchecked(currentList->pc + 4);
 	u32 *dst = (u32 *)(gstate.projMatrix + (op & 0xF));
@@ -2322,6 +2341,11 @@ void GPUCommon::Execute_ProjMtxData(u32 op, u32 diff) {
 }
 
 void GPUCommon::Execute_TgenMtxNum(u32 op, u32 diff) {
+	if (!currentList) {
+		gstate.texmtxnum = (GE_CMD_TGENMATRIXNUMBER << 24) | (op & 0xF);
+		return;
+	}
+
 	// This is almost always followed by GE_CMD_TGENMATRIXDATA.
 	const u32_le *src = (const u32_le *)Memory::GetPointerUnchecked(currentList->pc + 4);
 	u32 *dst = (u32 *)(gstate.tgenMatrix + (op & 0xF));
@@ -2339,7 +2363,8 @@ void GPUCommon::Execute_TgenMtxNum(u32 op, u32 diff) {
 			if (dst[i] != newVal) {
 				Flush();
 				dst[i] = newVal;
-				gstate_c.Dirty(DIRTY_TEXMATRIX);
+				// We check the matrix to see if we need projection.
+				gstate_c.Dirty(DIRTY_TEXMATRIX | DIRTY_FRAGMENTSHADER_STATE);
 			}
 			if (++i >= end) {
 				break;
@@ -2370,6 +2395,11 @@ void GPUCommon::Execute_TgenMtxData(u32 op, u32 diff) {
 }
 
 void GPUCommon::Execute_BoneMtxNum(u32 op, u32 diff) {
+	if (!currentList) {
+		gstate.boneMatrixNumber = (GE_CMD_BONEMATRIXNUMBER << 24) | (op & 0x7F);
+		return;
+	}
+
 	// This is almost always followed by GE_CMD_BONEMATRIXDATA.
 	const u32_le *src = (const u32_le *)Memory::GetPointerUnchecked(currentList->pc + 4);
 	u32 *dst = (u32 *)(gstate.boneMatrix + (op & 0x7F));

GPU/GPUState.cpp

@@ -127,9 +127,6 @@ void GPUgstate::Reset() {
 	memset(gstate.tgenMatrix, 0, sizeof(gstate.tgenMatrix));
 	memset(gstate.boneMatrix, 0, sizeof(gstate.boneMatrix));
 
-	if (gpu)
-		gpu->ResetMatrices();
-
 	savedContextVersion = 1;
 }
 
@@ -143,6 +140,7 @@ void GPUgstate::Save(u32_le *ptr) {
 	u32_le *cmds = ptr + 17;
 	for (size_t i = 0; i < ARRAY_SIZE(contextCmdRanges); ++i) {
 		for (int n = contextCmdRanges[i].start; n <= contextCmdRanges[i].end; ++n) {
+			// We'll run ReapplyGfxState after this to process dirtying.
 			*cmds++ = cmdmem[n];
 		}
 	}
@@ -171,11 +169,11 @@ void GPUgstate::Save(u32_le *ptr) {
 		cmds = SaveMatrix(cmds, GE_MTX_PROJECTION, ARRAY_SIZE(projMatrix), GE_CMD_PROJMATRIXNUMBER, GE_CMD_PROJMATRIXDATA);
 		cmds = SaveMatrix(cmds, GE_MTX_TEXGEN, ARRAY_SIZE(tgenMatrix), GE_CMD_TGENMATRIXNUMBER, GE_CMD_TGENMATRIXDATA);
 
-		*cmds++ = boneMatrixNumber;
-		*cmds++ = worldmtxnum;
-		*cmds++ = viewmtxnum;
-		*cmds++ = projmtxnum;
-		*cmds++ = texmtxnum;
+		*cmds++ = boneMatrixNumber & 0xFF00007F;
+		*cmds++ = worldmtxnum & 0xFF00000F;
+		*cmds++ = viewmtxnum & 0xFF00000F;
+		*cmds++ = projmtxnum & 0xFF00000F;
+		*cmds++ = texmtxnum & 0xFF00000F;
 		*cmds++ = GE_CMD_END << 24;
 	}
 }
@@ -251,11 +249,11 @@ void GPUgstate::Restore(u32_le *ptr) {
 		cmds = LoadMatrix(cmds, projMatrix, ARRAY_SIZE(projMatrix));
 		cmds = LoadMatrix(cmds, tgenMatrix, ARRAY_SIZE(tgenMatrix));
 
-		boneMatrixNumber = *cmds++;
-		worldmtxnum = *cmds++;
-		viewmtxnum = *cmds++;
-		projmtxnum = *cmds++;
-		texmtxnum = *cmds++;
+		boneMatrixNumber = (*cmds++) & 0xFF00007F;
+		worldmtxnum = (*cmds++) & 0xFF00000F;
+		viewmtxnum = (*cmds++) & 0xFF00000F;
+		projmtxnum = (*cmds++) & 0xFF00000F;
+		texmtxnum = (*cmds++) & 0xFF00000F;
 	}
 
 	if (gpu)

GPU/Software/DrawPixelX86.cpp

@@ -1159,7 +1159,7 @@ bool PixelJitCache::Jit_AlphaBlend(const PixelFuncID &id) {
 			if (id.AlphaBlendEq() == GE_BLENDMODE_MUL_AND_SUBTRACT_REVERSE)
 				PXOR(dstReg, R(dstReg));
 		} else if (id.AlphaBlendDst() == PixelBlendFactor::ONE) {
-			if (blendState.dstColorAsFactor)
+			if (blendState.dstColorAsFactor || blendState.usesDstAlpha)
 				PSRLW(dstReg, 4);
 		}
 

GPU/Software/Rasterizer.cpp

@@ -130,6 +130,12 @@ void ComputeRasterizerState(RasterizerState *state) {
 		state->minFilt = gstate.isMinifyFilteringEnabled();
 		state->magFilt = gstate.isMagnifyFilteringEnabled();
 		state->textureProj = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX;
+		if (state->textureProj) {
+			// We may be able to optimize this off.  This is actually kinda common.
+			if (gstate.tgenMatrix[2] == 0.0f && gstate.tgenMatrix[5] == 0.0f && gstate.tgenMatrix[8] == 0.0f && gstate.tgenMatrix[11] == 1.0f) {
+				state->textureProj = false;
+			}
+		}
 	}
 
 	state->shadeGouraud = gstate.getShadeMode() == GE_SHADE_GOURAUD;

GPU/Software/SoftGpu.cpp

@@ -1127,8 +1127,11 @@ void SoftGPU::Execute_TgenMtxData(u32 op, u32 diff) {
 	if (num < 12) {
 		u32 *target = (u32 *)&gstate.tgenMatrix[num];
 		u32 newVal = op << 8;
-		// No dirtying, read during vertex read.
-		*target = newVal;
+		if (newVal != *target) {
+			*target = newVal;
+			// This is mainly used in vertex read, but also affects if we enable texture projection.
+			dirtyFlags_ |= SoftDirty::RAST_TEX;
+		}
 	}
 
 	// Doesn't wrap to any other matrix.
@@ -1196,6 +1199,11 @@ bool SoftGPU::GetMatrix24(GEMatrixType type, u32_le *result, u32 cmdbits) {
 	return true;
 }
 
+void SoftGPU::ResetMatrices() {
+	GPUCommon::ResetMatrices();
+	dirtyFlags_ |= SoftDirty::TRANSFORM_MATRIX | SoftDirty::RAST_TEX;
+}
+
 void SoftGPU::Execute_ImmVertexAlphaPrim(u32 op, u32 diff) {
 	GPUCommon::Execute_ImmVertexAlphaPrim(op, diff);
 	// We won't flush as often as hardware renderers, so we want to flush right away.

GPU/Software/SoftGpu.h

@@ -194,6 +194,7 @@ class SoftGPU : public GPUCommon {
 	void Execute_BoneMtxData(u32 op, u32 diff);
 
 	bool GetMatrix24(GEMatrixType type, u32_le *result, u32 cmdbits) override;
+	void ResetMatrices() override;
 
 	void Execute_ImmVertexAlphaPrim(u32 op, u32 diff);