- Fixed convert matrix bug happening on other vendors GPU Devices.

- Fixed wrong render target format for visiblity buffer pass

Author: AnisB

sdk/src/dx12/dx12_backend_command_buffer.cpp

@@ -1894,9 +1894,9 @@ namespace d3d12
 			{
 				// DestInfo
 				{
-					0,																												// DestSize to be populated by driver implementation
+					optimal ? 0 : width * height * sizeof(float16_t),																												// DestSize to be populated by driver implementation
 					optimal ? D3D12_LINEAR_ALGEBRA_MATRIX_LAYOUT_MUL_OPTIMAL : D3D12_LINEAR_ALGEBRA_MATRIX_LAYOUT_COLUMN_MAJOR,     // convert to mul optimal layout
-					0,																												// stride is ignored since optimal layout is implementation dependent
+					optimal ? 0 : width * sizeof(float16_t),																												// stride is ignored since optimal layout is implementation dependent
 					width,																											// number of rows in weight matrix to be converted
 					height,																											// number of columns in weight matrix to converted
 					D3D12_LINEAR_ALGEBRA_DATATYPE_FLOAT16																			// convert to float16_t datatype

sdk/src/network/tsnc.cpp

@@ -171,8 +171,8 @@ void TSNC::upload_network(CommandQueue cmdQ, CommandBuffer cmdB)
             if (m_CVS)
             {
                 mlp::upload_and_convert_matrices(m_Device, cmdQ, cmdB, (char*)cpuMLP.mlp0Buffer.data(), cpuMLP.mlp0Width, cpuMLP.mlp0Height, m_Nwk.mlp.weight0Buffer, m_Nwk.mlp.weight0OptimalBuffer, setIdx * cpuMLP.mlp0Width * cpuMLP.mlp0Height * sizeof(float16_t));
-                mlp::upload_and_convert_matrices(m_Device, cmdQ, cmdB, (char*)cpuMLP.mlp1Buffer.data(), cpuMLP.mlp1Width, cpuMLP.mlp1Height, m_Nwk.mlp.weight1Buffer, m_Nwk.mlp.weight1OptimalBuffer, setIdx * cpuMLP.mlp0Width * cpuMLP.mlp0Height * sizeof(float16_t));
-                mlp::upload_and_convert_matrices(m_Device, cmdQ, cmdB, (char*)cpuMLP.mlp2Buffer.data(), cpuMLP.mlp2Width, cpuMLP.mlp2Height, m_Nwk.mlp.weight2Buffer, m_Nwk.mlp.weight2OptimalBuffer, setIdx * cpuMLP.mlp0Width * cpuMLP.mlp0Height * sizeof(float16_t));
+                mlp::upload_and_convert_matrices(m_Device, cmdQ, cmdB, (char*)cpuMLP.mlp1Buffer.data(), cpuMLP.mlp1Width, cpuMLP.mlp1Height, m_Nwk.mlp.weight1Buffer, m_Nwk.mlp.weight1OptimalBuffer, setIdx * cpuMLP.mlp1Width * cpuMLP.mlp1Height * sizeof(float16_t));
+                mlp::upload_and_convert_matrices(m_Device, cmdQ, cmdB, (char*)cpuMLP.mlp2Buffer.data(), cpuMLP.mlp2Width, cpuMLP.mlp2Height, m_Nwk.mlp.weight2Buffer, m_Nwk.mlp.weight2OptimalBuffer, setIdx * cpuMLP.mlp2Width * cpuMLP.mlp2Height * sizeof(float16_t));
             }
             else
             {

sdk/src/render_pipeline/dino_renderer.cpp

@@ -54,7 +54,7 @@ void DinoRenderer::initialize(uint64_t hInstance, const CommandLineOptions& opti
 
     // Create the graphics components
     graphics::setup_graphics_api(GraphicsAPI::DX12);
-    //d3d12::device::enable_debug_layer();
+    // graphics::device::enable_debug_layer();
     graphics::device::enable_experimental_features();
     m_Device = graphics::device::create_graphics_device();
     m_Window = graphics::window::create_window(m_Device, (uint64_t)hInstance, 1920, 1080, "Intel TSNC (DX12)");

sdk/src/render_pipeline/skinned_mesh_renderer.cpp

@@ -94,7 +94,7 @@ void SkinnedMeshRenderer::reload_shaders(const std::string& shaderLibrary)
         GraphicsPipelineDescriptor gpd;
         gpd.includeDirectories.push_back(shaderLibrary);
         gpd.filename = shaderLibrary + "\\Mesh\\VisibilityPass.graphics";
-        gpd.rtFormat[0] = TextureFormat::R16_UInt;
+        gpd.rtFormat[0] = TextureFormat::R32_UInt;
         gpd.depthStencilState.enableDepth = true;
         gpd.depthStencilState.depthtest = DepthTest::LEqual;
         gpd.depthStencilState.depthWrite = true;