Merge pull request #2048 from CookiePLMonster/vertex-stream-cache-optimization

Optimize vertex stream cache for partial buffer access

src/core/hle/D3D8/Direct3D9/Direct3D9.cpp

@@ -6864,7 +6864,7 @@ void CxbxDrawIndexed(CxbxDrawContext &DrawContext)
 	// for an explanation on the function of the BaseVertexIndex, MinVertexIndex, NumVertices and StartIndex arguments.
 	HRESULT hRet = g_pD3DDevice->DrawIndexedPrimitive(
 		/* PrimitiveType = */EmuXB2PC_D3DPrimitiveType(DrawContext.XboxPrimitiveType),
-		BaseVertexIndex,
+		/* BaseVertexIndex, = */-CacheEntry.LowIndex, // Base vertex index has been accounted for in the stream conversion, now we need to "un-offset" the index buffer
 		/* MinVertexIndex = */CacheEntry.LowIndex,
 		/* NumVertices = */(CacheEntry.HighIndex - CacheEntry.LowIndex) + 1,
 		/* startIndex = DrawContext.dwStartVertex = */0,
@@ -7516,9 +7516,6 @@ xbox::void_xt WINAPI xbox::EMUPATCH(D3DDevice_DrawVertices)
 			// Draw quadlists using a single 'quad-to-triangle mapping' index buffer :
 			// Assure & activate that special index buffer :
 			CxbxAssureQuadListD3DIndexBuffer(/*NrOfQuadIndices=*/DrawContext.dwVertexCount);
-			// This API's StartVertex argument is multiplied by vertex stride and added to the start of the vertex buffer;
-			// BaseVertexIndex offers the same functionality on host :
-			UINT BaseVertexIndex = DrawContext.dwStartVertex;
 			// Convert quad vertex count to triangle vertex count :
 			UINT NumVertices = QuadToTriangleVertexCount(DrawContext.dwVertexCount);
 			// Convert quad primitive count to triangle primitive count :
@@ -7528,7 +7525,7 @@ xbox::void_xt WINAPI xbox::EMUPATCH(D3DDevice_DrawVertices)
 			// Emulate drawing quads by drawing each quad with two indexed triangles :
 			HRESULT hRet = g_pD3DDevice->DrawIndexedPrimitive(
 				/*PrimitiveType=*/D3DPT_TRIANGLELIST,
-				BaseVertexIndex,
+				/*BaseVertexIndex=*/0, // Base vertex index has been accounted for in the stream conversion
 				/*MinVertexIndex=*/0,
 				NumVertices,
 				/*startIndex=*/0,
@@ -7542,7 +7539,7 @@ xbox::void_xt WINAPI xbox::EMUPATCH(D3DDevice_DrawVertices)
 			// if (StartVertex > 0) LOG_TEST_CASE("StartVertex > 0 (non-quad)"); // Verified test case : XDK Sample (PlayField)
 			HRESULT hRet = g_pD3DDevice->DrawPrimitive(
 				EmuXB2PC_D3DPrimitiveType(DrawContext.XboxPrimitiveType),
-				DrawContext.dwStartVertex,
+				/*StartVertex=*/0, // Start vertex has been accounted for in the stream conversion
 				DrawContext.dwHostPrimitiveCount
 			);
 			DEBUG_D3DRESULT(hRet, "g_pD3DDevice->DrawPrimitive");
@@ -7553,8 +7550,8 @@ xbox::void_xt WINAPI xbox::EMUPATCH(D3DDevice_DrawVertices)
 				LOG_TEST_CASE("X_D3DPT_LINELOOP"); // TODO : Text-cases needed
 
 				assert(DrawContext.dwBaseVertexIndex == 0); // if this fails, it needs to be added to LowIndex and HighIndex :
-				INDEX16 LowIndex = (INDEX16)DrawContext.dwStartVertex;
+				INDEX16 LowIndex = 0;
-				INDEX16 HighIndex = (INDEX16)(DrawContext.dwStartVertex + DrawContext.dwHostPrimitiveCount);
+				INDEX16 HighIndex = (INDEX16)(DrawContext.dwHostPrimitiveCount);
 				// Draw the closing line using a helper function (which will SetIndices)
 				CxbxDrawIndexedClosingLine(LowIndex, HighIndex);
 				// NOTE : We don't restore the previously active index buffer

src/core/hle/D3D8/XbVertexBuffer.cpp

@@ -261,9 +261,7 @@ void CxbxVertexBufferConverter::ConvertStream
 	UINT HostStreamNumber = XboxStreamNumber; // Use Xbox stream index on host
 	uint8_t *pXboxVertexData = xbox::zeroptr;
 	UINT uiXboxVertexStride = 0;
-	UINT uiVertexCount = 0;
 	UINT uiHostVertexStride = 0;
-	DWORD dwHostVertexDataSize = 0;
 	uint8_t *pHostVertexData = nullptr;
 	IDirect3DVertexBuffer *pNewHostVertexBuffer = nullptr;
 
@@ -275,9 +273,7 @@ void CxbxVertexBufferConverter::ConvertStream
 
 		pXboxVertexData = (uint8_t *)pDrawContext->pXboxVertexStreamZeroData;
 		uiXboxVertexStride = pDrawContext->uiXboxVertexStreamZeroStride;
-		uiVertexCount = pDrawContext->VerticesInBuffer; 
 		uiHostVertexStride = (bNeedVertexPatching) ? pVertexShaderStreamInfo->HostVertexStride : uiXboxVertexStride;
-		dwHostVertexDataSize = uiVertexCount * uiHostVertexStride;
 	} else {
 		xbox::X_STREAMINPUT& XboxStreamInput = GetXboxVertexStreamInput(XboxStreamNumber);
 		xbox::X_D3DVertexBuffer *pXboxVertexBuffer = XboxStreamInput.VertexBuffer;
@@ -298,14 +294,11 @@ void CxbxVertexBufferConverter::ConvertStream
 
 		pXboxVertexData += XboxStreamInput.Offset;
 		uiXboxVertexStride = XboxStreamInput.Stride;
-        // Set a new (exact) vertex count
-		uiVertexCount = pDrawContext->VerticesInBuffer;
 		// Dxbx note : Don't overwrite pDrawContext.dwVertexCount with uiVertexCount, because an indexed draw
 		// can (and will) use less vertices than the supplied nr of indexes. Thix fixes
 		// the missing parts in the CompressedVertices sample (in Vertex shader mode).
 
 		uiHostVertexStride = (bNeedVertexPatching) ? pVertexShaderStreamInfo->HostVertexStride : uiXboxVertexStride;
-		dwHostVertexDataSize = uiVertexCount * uiHostVertexStride;
 
 		// Copy stream for patching and caching.
 		bNeedStreamCopy = true;
@@ -327,7 +320,17 @@ void CxbxVertexBufferConverter::ConvertStream
     }
 
     // Now we have enough information to hash the existing resource and find it in our cache!
-    DWORD xboxVertexDataSize = uiVertexCount * uiXboxVertexStride;
+    // To avoid hashing and converting unused vertices, identify the "interesting" region
+    // basing on the index/starting vertex data 
+    if (pDrawContext->pXboxIndexData != nullptr) {
+        pXboxVertexData += (pDrawContext->dwBaseVertexIndex + pDrawContext->LowIndex) * uiXboxVertexStride;
+    } else {
+        pXboxVertexData += pDrawContext->dwStartVertex * uiXboxVertexStride;
+    }
+
+    const UINT uiVertexCount = pDrawContext->NumVerticesToUse;
+    const DWORD dwHostVertexDataSize = uiVertexCount * uiHostVertexStride;
+    const DWORD xboxVertexDataSize = uiVertexCount * uiXboxVertexStride;
     uint64_t vertexDataHash = ComputeHash(pXboxVertexData, xboxVertexDataSize);
     uint64_t pVertexShaderSteamInfoHash = 0;
 
@@ -634,9 +637,6 @@ void CxbxVertexBufferConverter::Apply(CxbxDrawContext *pDrawContext)
 
 	m_pCxbxVertexDeclaration = CxbxGetVertexDeclaration();
 
-	// If we are drawing from an offset, we know that the vertex count must have
-	// 'offset' vertices before the first drawn vertices
-	pDrawContext->VerticesInBuffer = pDrawContext->dwStartVertex + pDrawContext->dwVertexCount;
 	// When this is an indexed draw, take the index buffer into account
 	if (pDrawContext->pXboxIndexData) {
 		// Is the highest index in this buffer not set yet?
@@ -646,11 +646,13 @@ void CxbxVertexBufferConverter::Apply(CxbxDrawContext *pDrawContext)
 			LOG_TEST_CASE("HighIndex == 0"); // TODO : If this is never hit, replace entire block by assert(pDrawContext->HighIndex > 0);
 			WalkIndexBuffer(pDrawContext->LowIndex, pDrawContext->HighIndex, pDrawContext->pXboxIndexData, pDrawContext->dwVertexCount);
 		}
-		// Convert highest index (including the base offset) into a count
+		// Convert the range of indices into a count
-		DWORD dwHighestVertexCount = pDrawContext->dwBaseVertexIndex + pDrawContext->HighIndex + 1;
+		pDrawContext->NumVerticesToUse = pDrawContext->HighIndex - pDrawContext->LowIndex + 1;
-		// Use the biggest vertex count that can be reached
+	}
-		if (pDrawContext->VerticesInBuffer < dwHighestVertexCount)
+	else {
-			pDrawContext->VerticesInBuffer = dwHighestVertexCount;
+		// If we are drawing from an offset, we know that the vertex count must have
+		// 'offset' vertices before the first drawn vertices
+		pDrawContext->NumVerticesToUse = pDrawContext->dwVertexCount;
 	}
 
     // Get the number of streams

src/core/hle/D3D8/XbVertexBuffer.h

@@ -40,7 +40,7 @@ typedef struct _CxbxDrawContext
 	IN	   PWORD				 pXboxIndexData; // Set by D3DDevice_DrawIndexedVertices, D3DDevice_DrawIndexedVerticesUP and HLE_draw_inline_elements
 	IN	   DWORD				 dwBaseVertexIndex; // Set to g_Xbox_BaseVertexIndex in D3DDevice_DrawIndexedVertices
 	IN	   INDEX16				 LowIndex, HighIndex; // Set when pXboxIndexData is set
-	IN	   size_t				 VerticesInBuffer; // Set by CxbxVertexBufferConverter::Apply
+	IN	   UINT 				 NumVerticesToUse; // Set by CxbxVertexBufferConverter::Apply
     // Data if Draw...UP call
     IN PVOID                     pXboxVertexStreamZeroData;
     IN UINT                      uiXboxVertexStreamZeroStride;