Merge pull request #333 from EvgeniiG/master

Add a stencil copy, fix the deferred CS and enable it by default

Assets/ScriptableRenderPipeline/HDRenderPipeline/HDRenderPipeline.cs

 
         readonly GBufferManager m_gbufferManager = new GBufferManager();
 
+        Material m_CopyStencilBuffer;
+
         // Various set of material use in render loop
         Material m_FilterAndCombineSubsurfaceScattering;
         // Old SSS Model >>>
         readonly RenderTargetIdentifier m_DistortionBufferRT;
 
         private RenderTexture m_CameraDepthStencilBuffer = null;
-        private RenderTexture m_CameraDepthStencilBufferCopy = null;
+        private RenderTexture m_CameraDepthBufferCopy = null;
+        private RenderTexture m_CameraStencilBufferCopy = null; // Currently, it's manually copied using a pixel shader, and optimized to only contain the SSS bit
-        private RenderTargetIdentifier m_CameraDepthStencilBufferCopyRT;
+        private RenderTargetIdentifier m_CameraDepthBufferCopyRT;
+        private RenderTargetIdentifier m_CameraStencilBufferCopyRT;
 
         // Post-processing context and screen-space effects (recycled on every frame to avoid GC alloc)
         readonly PostProcessRenderContext m_PostProcessContext;
         // Currently we use only 2 bits to identify the kind of lighting that is expected from the render pipeline
         // Usage is define in LightDefinitions.cs
         [Flags]
-        public enum StencilBits
+        public enum StencilBitMask
-            Lighting = 3,                    // 0
-            All = 255                        // 0xFF
+            Clear    = 0,                    // 0x0
+            Lighting = 3,                    // 0x3  - 2 bit
+            All      = 255                   // 0xFF - 8 bit
         }
 
         // Detect when windows size is changing
             CreateSssMaterials(sssSettings.useDisneySSS);
             // <<< Old SSS Model
 
+            m_CopyStencilBuffer           = Utilities.CreateEngineMaterial("Hidden/HDRenderPipeline/CopyStencilBuffer");
             m_CameraMotionVectorsMaterial = Utilities.CreateEngineMaterial("Hidden/HDRenderPipeline/CameraMotionVectors");
 
             InitializeDebugMaterials();
         };
 #endif
 
-        void CreateDepthBuffer(Camera camera)
+        void CreateDepthStencilBuffer(Camera camera)
         {
             if (m_CameraDepthStencilBuffer != null)
             {
 
             if (NeedDepthBufferCopy())
             {
-                if (m_CameraDepthStencilBufferCopy != null)
+                if (m_CameraDepthBufferCopy != null)
-                    m_CameraDepthStencilBufferCopy.Release();
+                    m_CameraDepthBufferCopy.Release();
-                m_CameraDepthStencilBufferCopy = new RenderTexture(camera.pixelWidth, camera.pixelHeight, 24, RenderTextureFormat.Depth);
-                m_CameraDepthStencilBufferCopy.filterMode = FilterMode.Point;
-                m_CameraDepthStencilBufferCopy.Create();
-                m_CameraDepthStencilBufferCopyRT = new RenderTargetIdentifier(m_CameraDepthStencilBufferCopy);
+                m_CameraDepthBufferCopy = new RenderTexture(camera.pixelWidth, camera.pixelHeight, 24, RenderTextureFormat.Depth);
+                m_CameraDepthBufferCopy.filterMode = FilterMode.Point;
+                m_CameraDepthBufferCopy.Create();
+                m_CameraDepthBufferCopyRT = new RenderTargetIdentifier(m_CameraDepthBufferCopy);
+            }
+
+            if (NeedStencilBufferCopy())
+            {
+                if (m_CameraStencilBufferCopy != null)
+                {
+                    m_CameraStencilBufferCopy.Release();
+                }
+                m_CameraStencilBufferCopy = new RenderTexture(camera.pixelWidth, camera.pixelHeight, 0, RenderTextureFormat.R8);
+                m_CameraStencilBufferCopy.filterMode = FilterMode.Point;
+                m_CameraStencilBufferCopy.Create();
+                m_CameraStencilBufferCopyRT = new RenderTargetIdentifier(m_CameraStencilBufferCopy);
             }
         }
 
 
             if (resolutionChanged || m_CameraDepthStencilBuffer == null)
             {
-                CreateDepthBuffer(camera);
+                CreateDepthStencilBuffer(camera);
             }
 
             if (resolutionChanged || m_LightLoop.NeedResize())
             return SystemInfo.graphicsDeviceType != GraphicsDeviceType.PlayStation4;
         }
 
-        Texture GetDepthTexture()
+        bool NeedStencilBufferCopy()
+        {
+            // Currently, Unity does not offer a way to bind the stencil buffer as a texture in a compute shader.
+            // Therefore, it's manually copied using a pixel shader, and optimized to only contain the SSS bit.
+            return m_DebugDisplaySettings.renderingDebugSettings.enableSSSAndTransmission;
+        }
+
+        RenderTargetIdentifier GetDepthTexture()
-            if (NeedDepthBufferCopy())
-                return m_CameraDepthStencilBufferCopy;
-            else
-                return m_CameraDepthStencilBuffer;
+            return NeedDepthBufferCopy() ? m_CameraDepthBufferCopy : m_CameraDepthStencilBuffer;
+        }
+
+        RenderTargetIdentifier GetStencilTexture()
+        {
+            return NeedStencilBufferCopy() ? m_CameraStencilBufferCopyRT : m_CameraDepthStencilBufferRT;
         }
 
         private void CopyDepthBufferIfNeeded(CommandBuffer cmd)
                 {
                     using (new Utilities.ProfilingSample("Copy depth-stencil buffer", cmd))
                     {
-                        cmd.CopyTexture(m_CameraDepthStencilBufferRT, m_CameraDepthStencilBufferCopyRT);
+                        cmd.CopyTexture(m_CameraDepthStencilBufferRT, m_CameraDepthBufferCopyRT);
+        }
+
+        private void PrepareAndBindStencilTexture(CommandBuffer cmd)
+        {
+            if (NeedStencilBufferCopy())
+            {
+                using (new Utilities.ProfilingSample("Copy StencilBuffer", cmd))
+                {
+                    Utilities.DrawFullScreen(cmd, m_CopyStencilBuffer, m_CameraStencilBufferCopyRT, m_CameraDepthStencilBufferRT);
+                }
+            }
+
+            cmd.SetGlobalTexture("_StencilTexture", GetStencilTexture());
         }
 
         public void UpdateCommonSettings()
             {
                 CopyDepthBufferIfNeeded(cmd);
             }
+
+            // Required for the SSS pass.
+            PrepareAndBindStencilTexture(cmd);
 
             if (m_DebugDisplaySettings.IsDebugMaterialDisplayEnabled())
             {
                 return;
             }
 
-            RenderTargetIdentifier[] colorRTs = { m_CameraColorBufferRT, m_CameraSubsurfaceBufferRT };
+            RenderTargetIdentifier[] colorRTs     = { m_CameraColorBufferRT, m_CameraSubsurfaceBufferRT };
+            RenderTargetIdentifier   depthTexture = GetDepthTexture();
-                m_LightLoop.RenderDeferredLighting(hdCamera, cmd, m_DebugDisplaySettings, colorRTs, m_CameraDepthStencilBufferRT, new RenderTargetIdentifier(GetDepthTexture()), true);
+                m_LightLoop.RenderDeferredLighting(hdCamera, cmd, m_DebugDisplaySettings, colorRTs, m_CameraDepthStencilBufferRT, depthTexture, true);
-            m_LightLoop.RenderDeferredLighting(hdCamera, cmd, m_DebugDisplaySettings, colorRTs, m_CameraDepthStencilBufferRT, new RenderTargetIdentifier(GetDepthTexture()), false);
+            m_LightLoop.RenderDeferredLighting(hdCamera, cmd, m_DebugDisplaySettings, colorRTs, m_CameraDepthStencilBufferRT, depthTexture, false);
         }
 
         // Combines specular lighting and diffuse lighting with subsurface scattering.

Assets/ScriptableRenderPipeline/HDRenderPipeline/HDRenderPipelineAsset.asset

   tileSettings:
     enableTileAndCluster: 1
     enableSplitLightEvaluation: 1
-    enableComputeLightEvaluation: 0
+    enableComputeLightEvaluation: 1
     enableComputeLightVariants: 0
     enableComputeMaterialVariants: 0
     enableClustered: 1

Assets/ScriptableRenderPipeline/HDRenderPipeline/Lighting/TilePass/TilePass.cs

 
             public void RenderDeferredLighting( HDCamera hdCamera, CommandBuffer cmd,
                                                 DebugDisplaySettings debugDisplaySettings,
-                                                RenderTargetIdentifier[] colorBuffers, RenderTargetIdentifier depthStencilBuffer, RenderTargetIdentifier depthStencilTexture,
+                                                RenderTargetIdentifier[] colorBuffers, RenderTargetIdentifier depthStencilBuffer, RenderTargetIdentifier depthTexture,
+
+                if (m_TileSettings.enableComputeLightEvaluation && outputSplitLighting)
+                {
+                    // The CS is always in the MRT mode. Do not execute the same shader twice.
+                    return;
+                }
 
                 // TODO: To reduce GC pressure don't do concat string here
                 using (new Utilities.ProfilingSample((m_TileSettings.enableTileAndCluster ? "TilePass - Deferred Lighting Pass" : "SinglePass - Deferred Lighting Pass") + (outputSplitLighting ? " MRT" : ""), cmd))
                                 // Pass global parameters to compute shader
                                 // TODO: get rid of this by making global parameters visible to compute shaders
                                 PushGlobalParams(camera, cmd, shadeOpaqueShader, kernel);
+                                hdCamera.SetupComputeShader(shadeOpaqueShader, cmd);
 
                                 // TODO: Update value like in ApplyDebugDisplaySettings() call. Sadly it is high likely that this will not be keep in sync. we really need to get rid of this by making global parameters visible to compute shaders
                                 cmd.SetComputeIntParam(shadeOpaqueShader, "_DebugViewMaterial", Shader.GetGlobalInt("_DebugViewMaterial"));
                                 cmd.SetComputeBufferParam(shadeOpaqueShader, kernel, "g_vLightListGlobal", bUseClusteredForDeferred ? s_PerVoxelLightLists : s_LightList);
 
-                                cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "_MainDepthTexture", depthStencilTexture);
+                                cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "_MainDepthTexture", depthTexture);
                                 cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "_GBufferTexture0", Shader.PropertyToID("_GBufferTexture0"));
                                 cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "_GBufferTexture1", Shader.PropertyToID("_GBufferTexture1"));
                                 cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "_GBufferTexture2", Shader.PropertyToID("_GBufferTexture2"));
                                 cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "_LtcDisneyDiffuseMatrix", Shader.GetGlobalTexture("_LtcDisneyDiffuseMatrix"));
                                 cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "_LtcMultiGGXFresnelDisneyDiffuse", Shader.GetGlobalTexture("_LtcMultiGGXFresnelDisneyDiffuse"));
 
-                                hdCamera.SetupComputeShader(shadeOpaqueShader, cmd);
-
                                 cmd.SetComputeMatrixParam(shadeOpaqueShader, "g_mInvScrProjection", Shader.GetGlobalMatrix("g_mInvScrProjection"));
                                 cmd.SetComputeIntParam(shadeOpaqueShader, "_UseTileLightList", Shader.GetGlobalInt("_UseTileLightList"));
 
                                 cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "_IESArray", IESArrayTexture ? IESArrayTexture : m_DefaultTexture2DArray);
                                 cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "_SkyTexture", skyTexture ? skyTexture : m_DefaultTexture2DArray);
 
-                                // Since we need the stencil test, the compute path does not currently support SSS.
-                                cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "combinedLightingUAV", colorBuffers[0]);
+                                // Set SSS parameters.
+                                cmd.SetComputeIntParam(   shadeOpaqueShader, "_EnableSSSAndTransmission", Shader.GetGlobalInt(       "_EnableSSSAndTransmission"));
+                                cmd.SetComputeIntParam(   shadeOpaqueShader, "_TexturingModeFlags",       Shader.GetGlobalInt(       "_TexturingModeFlags"));
+                                cmd.SetComputeIntParam(   shadeOpaqueShader, "_TransmissionFlags",        Shader.GetGlobalInt(       "_TransmissionFlags"));
+                                cmd.SetComputeFloatParams(shadeOpaqueShader, "_ThicknessRemaps",          Shader.GetGlobalFloatArray("_ThicknessRemaps"));
+                                // We are currently supporting two different SSS mode: Jimenez (with 2-Gaussian profile) and Disney
+                                // We have added the ability to switch between each other for subsurface scattering, but for transmittance this is more tricky as we need to add
+                                // shader variant for forward, gbuffer and deferred shader. We want to avoid this.
+                                // So for transmittance we use Disney profile formulation (that we know is more correct) in both case, and in the case of Jimenez we hack the parameters with 2-Gaussian parameters (Ideally we should fit but haven't find good fit) so it approximately match.
+                                // Note: Jimenez SSS is in cm unit whereas Disney is in mm unit making an inconsistency here to compare model side by side
+                                cmd.SetComputeVectorArrayParam(shadeOpaqueShader, "_ShapeParams",       Shader.GetGlobalVectorArray("_ShapeParams"));
+                                cmd.SetComputeVectorArrayParam(shadeOpaqueShader, "_TransmissionTints", Shader.GetGlobalVectorArray("_TransmissionTints"));
+
+                                cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "specularLightingUAV", colorBuffers[0]);
+                                cmd.SetComputeTextureParam(shadeOpaqueShader, kernel, "diffuseLightingUAV",  colorBuffers[1]);
 
                                 // always do deferred lighting in blocks of 16x16 (not same as tiled light size)
 

Assets/ScriptableRenderPipeline/HDRenderPipeline/Lighting/TilePass/shadeopaque.compute

 #pragma kernel ShadeOpaque_Indirect_Clustered_Variant14         SHADE_OPAQUE_ENTRY=ShadeOpaque_Indirect_Clustered_Variant14             USE_CLUSTERED_LIGHTLIST     USE_INDIRECT    VARIANT=14
 #pragma kernel ShadeOpaque_Indirect_Clustered_Variant15         SHADE_OPAQUE_ENTRY=ShadeOpaque_Indirect_Clustered_Variant15             USE_CLUSTERED_LIGHTLIST     USE_INDIRECT    VARIANT=15
 
-//#pragma #pragma enable_d3d11_debug_symbols
-
-// Split lighting is required for the SSS pass.
-// Not currently possible since we need to access the stencil buffer from the compute shader.
-// #pragma multi_compile _ OUTPUT_SPLIT_LIGHTING
 
 #define LIGHTLOOP_TILE_PASS 1
 #define LIGHTLOOP_TILE_DIRECT 1
 
 DECLARE_GBUFFER_TEXTURE(_GBufferTexture);
 
-#ifdef OUTPUT_SPLIT_LIGHTING
-    RWTexture2D<float4> specularLightingUAV;
-    RWTexture2D<float3> diffuseLightingUAV;
-#else
-    RWTexture2D<float4> combinedLightingUAV;
-#endif
+RWTexture2D<float3> diffuseLightingUAV;
+RWTexture2D<float4> specularLightingUAV;
 
 CBUFFER_START(UnityShadeOpaque)
 	uint g_TileListOffset;
 #endif
 
     float depth = LOAD_TEXTURE2D(_MainDepthTexture, posInput.unPositionSS).x;
+
+    // TODO: add an early-out using HiZ/HiS or our light tile data.
+    if (depth == 0) { return; }
+
     UpdatePositionInput(depth, _InvViewProjMatrix, _ViewProjMatrix, posInput);
     float3 V = GetWorldSpaceNormalizeViewDir(posInput.positionWS);
 
     float3 specularLighting;
     LightLoop(V, posInput, preLightData, bsdfData, bakeDiffuseLighting, featureFlags, diffuseLighting, specularLighting);
 
-#ifdef OUTPUT_SPLIT_LIGHTING
-    specularLightingUAV[pixelCoord] = float4(specularLighting, 1.0);
-    diffuseLightingUAV[pixelCoord]  = diffuseLighting;
-#else
-    combinedLightingUAV[pixelCoord] = float4(diffuseLighting + specularLighting, 1.0);
-#endif
+    if (_EnableSSSAndTransmission != 0 && bsdfData.materialId == MATERIALID_LIT_SSS)
+    {
+        // We SSSSS is enabled with use split lighting.
+        // SSSSS algorithm need to know which pixels contribute to SSS and which doesn't. We could use the stencil for that but it mean that it will increase the cost of SSSSS
+        // A simpler solution is to add a slight contribution here that isn't visible (here we chose fp16 min (which is also fp11 and fp10 min).
+        // The SSSSS algorithm will check if diffuse lighting is black and discard the pixel if it is the case
+        diffuseLighting.r = max(diffuseLighting.r, HFLT_MIN);
+
+        specularLightingUAV[pixelCoord] = float4(specularLighting, 1.0);
+        diffuseLightingUAV[pixelCoord]  = diffuseLighting;
+    }
+    else
+    {
+        specularLightingUAV[pixelCoord] = float4(diffuseLighting + specularLighting, 1.0);
+    }
 }
 

Assets/ScriptableRenderPipeline/HDRenderPipeline/Material/Lit/Lit.hlsl

 #if defined(SHADERPASS) && (SHADERPASS == SHADERPASS_LIGHT_TRANSPORT) // In case of GI pass don't modify the diffuseColor
     if (0)
 #else
-    if (_EnableSSSAndTransmission > 0) // If we globally disable SSS effect, don't modify diffuseColor
+    if (_EnableSSSAndTransmission != 0) // If we globally disable SSS effect, don't modify diffuseColor
 #endif
     {
         // We modify the albedo here as this code is used by all lighting (including light maps and GI).

Assets/ScriptableRenderPipeline/HDRenderPipeline/Material/Lit/Resources/CombineSubsurfaceScattering.shader

         {
             Stencil
             {
-                Ref  1 // StencilBits.SSS
+                Ref  1 // StencilLightingUsage.SplitLighting
                 Comp Equal
                 Pass Keep
             }

Assets/ScriptableRenderPipeline/ShaderLibrary/AreaLighting.hlsl

             float a = cosOmega * acos(x) - z;                       // y*ArcCos[-y*Sqrt[(1/x-1)/(1-y^2)]]-Sqrt[(1-y^2)*(x/(1-x))-y^2]*(1/x-1)
             float b = atan(y);                                      // ArcTan[Sqrt[(1-y^2)*(x/(1-x))-y^2]]
 
-            // Replacing max() with saturate() results in a 12 cycle SGPR forwarding stall on PS4.
-            return max(INV_PI * (a * sinSqSigma + b), 0);           // (a/Pi)*x+(b/Pi)
+            return saturate(INV_PI * (a * sinSqSigma + b));
         }
     #endif
 #endif

Assets/ScriptableRenderPipeline/HDRenderPipeline/Material/Lit/Resources/CopyStencilBuffer.shader

+Shader "Hidden/HDRenderPipeline/CopyStencilBuffer"
+{
+    SubShader
+    {
+        Pass
+        {
+            Stencil
+            {
+                Ref  1 // StencilLightingUsage.SplitLighting
+                Comp Equal
+                Pass Keep
+            }
+
+            Cull   Off
+            ZTest  Always
+            ZWrite Off
+            Blend  Off
+
+            HLSLPROGRAM
+            #pragma target 4.5
+            #pragma only_renderers d3d11 ps4 metal // TEMP: until we go further in dev
+            // #pragma enable_d3d11_debug_symbols
+
+            #pragma vertex Vert
+            #pragma fragment Frag
+
+            #include "../../../../ShaderLibrary/Common.hlsl"
+            #include "../../../ShaderConfig.cs.hlsl"
+            #include "../../../ShaderVariables.hlsl"
+            #include "../../../Lighting/LightDefinition.cs.hlsl"
+
+            struct Attributes
+            {
+                uint vertexID : SV_VertexID;
+            };
+
+            struct Varyings
+            {
+                float4 positionCS : SV_Position;
+            };
+
+            Varyings Vert(Attributes input)
+            {
+                Varyings output;
+                output.positionCS = GetFullScreenTriangleVertexPosition(input.vertexID);
+                return output;
+            }
+
+            // Should use HiS and therefore be faster than a GPU memcpy().
+            float4 Frag(Varyings input) : SV_Target // use SV_StencilRef in D3D 11.3+
+            {
+                return float4(STENCILLIGHTINGUSAGE_SPLIT_LIGHTING, 0, 0, 0);
+            }
+            ENDHLSL
+        }
+    }
+    Fallback Off
+}

Assets/ScriptableRenderPipeline/HDRenderPipeline/Material/Lit/Resources/CopyStencilBuffer.shader.meta

+fileFormatVersion: 2
+guid: 7fd941b2d9d2a39429de64bde023932c
+timeCreated: 1499946987
+licenseType: Pro
+ShaderImporter:
+  externalObjects: {}
+  defaultTextures: []
+  userData: 
+  assetBundleName: 
+  assetBundleVariant: