Optimize further

hrydgard · Oct 3, 2023 · 4d95250 · 4d95250
1 parent 0260aeb
commit 4d95250
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 12 deletions.
diff --git a/GPU/Common/DrawEngineCommon.cpp b/GPU/Common/DrawEngineCommon.cpp
@@ -683,12 +683,6 @@ bool DrawEngineCommon::ExtendNonIndexedPrim(GEPrimitiveType prim, int vertexCoun
 		return false;
 	}
 
-	bool applySkin = (vertTypeID & GE_VTYPE_WEIGHT_MASK) && decOptions_.applySkinInDecode;
-	if (applySkin) {
-		// TODO: Support this somehow.
-		return false;
-	}
-
 	_dbg_assert_(numDrawInds_ < MAX_DEFERRED_DRAW_INDS);
 	_dbg_assert_(numDrawVerts_ > 0);
 	*bytesRead = vertexCount * dec_->VertexSize();
@@ -707,6 +701,7 @@ bool DrawEngineCommon::ExtendNonIndexedPrim(GEPrimitiveType prim, int vertexCoun
 	dv.vertexCount += vertexCount;
 	dv.indexUpperBound = dv.vertexCount - 1;
 	vertexCountInDrawCalls_ += vertexCount;
+
 	return true;
 }
 
@@ -787,10 +782,6 @@ void DrawEngineCommon::SubmitPrim(const void *verts, const void *inds, GEPrimiti
 		gstate_c.Dirty(DIRTY_TEXTURE_PARAMS);
 		DispatchFlush();
 	}
-
-	if (applySkin) {
-		DecodeVerts(decoded_);
-	}
 }
 
 void DrawEngineCommon::DecodeVerts(u8 *dest) {

diff --git a/GPU/Common/DrawEngineCommon.h b/GPU/Common/DrawEngineCommon.h
@@ -104,6 +104,13 @@ class DrawEngineCommon {
 
 	bool TestBoundingBox(const void *control_points, const void *inds, int vertexCount, u32 vertType);
 
+	void FlushSkin() {
+		bool applySkin = (lastVType_ & GE_VTYPE_WEIGHT_MASK) && decOptions_.applySkinInDecode;
+		if (applySkin) {
+			DecodeVerts(decoded_);
+		}
+	}
+
 	bool ExtendNonIndexedPrim(GEPrimitiveType prim, int vertexCount, u32 vertTypeID, int cullMode, int *bytesRead);
 	void SubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertTypeID, int cullMode, int *bytesRead);
 	template<class Surface>

diff --git a/GPU/GPUCommonHW.cpp b/GPU/GPUCommonHW.cpp
@@ -1038,6 +1038,9 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 			inds = nullptr;
 			if ((vertexType & GE_VTYPE_IDX_MASK) != GE_VTYPE_IDX_NONE) {
 				inds = Memory::GetPointerUnchecked(gstate_c.indexAddr);
+			} else {
+				// We can extend again after submitting a normal draw.
+				canExtend = true;
 			}
 			drawEngineCommon_->SubmitPrim(verts, inds, newPrim, count, vertTypeID, cullMode, &bytesRead);
 			AdvanceVerts(vertexType, count, bytesRead);
@@ -1058,18 +1061,18 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 			break;
 		}
 		case GE_CMD_VADDR:
+			canExtend = false;
 			gstate.cmdmem[GE_CMD_VADDR] = data;
 			gstate_c.vertexAddr = gstate_c.getRelativeAddress(data & 0x00FFFFFF);
-			canExtend = false;
 			break;
 		case GE_CMD_IADDR:
 			gstate.cmdmem[GE_CMD_IADDR] = data;
 			gstate_c.indexAddr = gstate_c.getRelativeAddress(data & 0x00FFFFFF);
 			break;
 		case GE_CMD_OFFSETADDR:
+			canExtend = false;
 			gstate.cmdmem[GE_CMD_OFFSETADDR] = data;
 			gstate_c.offsetAddr = data << 8;
-			canExtend = false;
 			break;
 		case GE_CMD_BASE:
 			gstate.cmdmem[GE_CMD_BASE] = data;
@@ -1124,6 +1127,8 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 				(Memory::ReadUnchecked_U32(target + 12 * 4) >> 24) == GE_CMD_RET &&
 				(target > currentList->stall || target + 12 * 4 < currentList->stall) &&
 				(gstate.boneMatrixNumber & 0x00FFFFFF) <= 96 - 12) {
+				drawEngineCommon_->FlushSkin();
+				canExtend = false;
 				FastLoadBoneMatrix(target);
 			} else {
 				goto bail;
@@ -1145,6 +1150,7 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 	}
 
 bail:
+	drawEngineCommon_->FlushSkin();
 	gstate.cmdmem[GE_CMD_VERTEXTYPE] = vertexType;
 	int cmdCount = src - start;
 	// Skip over the commands we just read out manually.