Render 3DGS debug splats as quads

MVS/3DGS-D3D12/shaders/DebugPointsPS.hlsl

@@ -1,4 +1,19 @@
-float4 MainPS(float4 position : SV_Position, float4 color : COLOR0) : SV_Target0
+struct PixelInput
 {
-    return color;
+    float4 position : SV_Position;
+    float4 color : COLOR0;
+    float2 localPosition : TEXCOORD0;
+};
+
+float4 MainPS(PixelInput input) : SV_Target0
+{
+    float alpha = exp(-dot(input.localPosition, input.localPosition));
+    alpha = saturate(alpha * input.color.a);
+
+    if (alpha < (1.0 / 255.0))
+    {
+        discard;
+    }
+
+    return float4(input.color.rgb * alpha, alpha);
 }

MVS/3DGS-D3D12/shaders/DebugPointsVS.hlsl

@@ -17,23 +17,34 @@ struct VertexOutput
 {
     float4 position : SV_Position;
     float4 color : COLOR0;
+    float2 localPosition : TEXCOORD0;
 };
 
 VertexOutput MainVS(uint vertexId : SV_VertexID, uint instanceId : SV_InstanceID)
 {
-    VertexOutput output;
+    VertexOutput output = (VertexOutput)0;
     uint splatIndex = gOrderBuffer[instanceId];
     PreparedSplatView view = gPreparedViews[splatIndex];
     float4 color = UnpackPreparedColor(view);
 
     if (view.clipPosition.w <= 0.0)
     {
-        output.position = float4(2.0, 2.0, 2.0, 1.0);
+        const float nanValue = asfloat(0x7fc00000);
-        output.color = 0.0;
+        output.position = float4(nanValue, nanValue, nanValue, nanValue);
         return output;
     }
 
+    float2 quadPosition = float2(vertexId & 1, (vertexId >> 1) & 1) * 2.0 - 1.0;
+    quadPosition *= 2.0;
+
+    float2 scaledAxis1 = view.axis1 * gSettings.w;
+    float2 scaledAxis2 = view.axis2 * gSettings.w;
+    float2 deltaScreenPosition =
+        (quadPosition.x * scaledAxis1 + quadPosition.y * scaledAxis2) * 2.0 / gScreenParams.xy;
+
     output.position = view.clipPosition;
-    output.color = float4(color.rgb, 1.0);
+    output.position.xy += deltaScreenPosition * view.clipPosition.w;
+    output.color = color;
+    output.localPosition = quadPosition;
     return output;
 }

MVS/3DGS-D3D12/src/App.cpp

@@ -209,7 +209,7 @@ FrameConstants BuildFrameConstants(uint32_t width, uint32_t height, uint32_t spl
     constants.settings[0] = static_cast<float>(splatCount);
     constants.settings[1] = 1.0f; // opacity scale
     constants.settings[2] = 3.0f; // SH order
-    constants.settings[3] = 1.0f;
+    constants.settings[3] = 0.9f;
     return constants;
 }
 
@@ -1108,7 +1108,7 @@ bool App::InitializeDebugDrawResources() {
 
     GraphicsPipelineDesc pipelineDesc = {};
     pipelineDesc.pipelineLayout = m_debugPipelineLayout;
-    pipelineDesc.topologyType = static_cast<uint32_t>(PrimitiveTopologyType::Point);
+    pipelineDesc.topologyType = static_cast<uint32_t>(PrimitiveTopologyType::Triangle);
     pipelineDesc.renderTargetCount = 1;
     pipelineDesc.renderTargetFormats[0] = static_cast<uint32_t>(Format::R8G8B8A8_UNorm);
     pipelineDesc.depthStencilFormat = static_cast<uint32_t>(Format::D24_UNorm_S8_UInt);
@@ -1117,6 +1117,11 @@ bool App::InitializeDebugDrawResources() {
     pipelineDesc.rasterizerState.cullMode = static_cast<uint32_t>(CullMode::None);
     pipelineDesc.depthStencilState.depthTestEnable = false;
     pipelineDesc.depthStencilState.depthWriteEnable = false;
+    pipelineDesc.blendState.blendEnable = true;
+    pipelineDesc.blendState.srcBlend = static_cast<uint32_t>(BlendFactor::One);
+    pipelineDesc.blendState.dstBlend = static_cast<uint32_t>(BlendFactor::InvSrcAlpha);
+    pipelineDesc.blendState.srcBlendAlpha = static_cast<uint32_t>(BlendFactor::One);
+    pipelineDesc.blendState.dstBlendAlpha = static_cast<uint32_t>(BlendFactor::InvSrcAlpha);
 
     pipelineDesc.vertexShader.fileName = ResolveShaderPath(L"DebugPointsVS.hlsl").wstring();
     pipelineDesc.vertexShader.entryPoint = L"MainVS";
@@ -1908,8 +1913,8 @@ void App::RenderFrame(bool captureScreenshot) {
     m_commandList.SetPipelineState(m_debugPipelineState);
     RHIDescriptorSet* debugSets[] = { m_debugDescriptorSet };
     m_commandList.SetGraphicsDescriptorSets(0, 1, debugSets, m_debugPipelineLayout);
-    m_commandList.SetPrimitiveTopology(PrimitiveTopology::PointList);
+    m_commandList.SetPrimitiveTopology(PrimitiveTopology::TriangleStrip);
-    m_commandList.Draw(1u, m_gaussianSceneData.splatCount, 0u, 0u);
+    m_commandList.Draw(4u, m_gaussianSceneData.splatCount, 0u, 0u);
 
     if (captureScreenshot) {
         AppendTrace("RenderFrame: close+execute capture pre-screenshot");