Started work on luminance pyramid and acculumate pipelines.

Added keywords for customization options to the compute shaders using multi_compile_local.

Assets/Resources/FSR2/ffx_fsr2_accumulate_pass.compute

@@ -1,5 +1,13 @@
 #pragma kernel CS
 
+#pragma multi_compile_local __ FFX_HALF
+#pragma multi_compile_local __ FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
+#pragma multi_compile_local __ FFX_FSR2_OPTION_HDR_COLOR_INPUT
+#pragma multi_compile_local __ FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_INVERTED_DEPTH
+#pragma multi_compile_local __ FFX_FSR2_OPTION_APPLY_SHARPENING
+
 #define FFX_GPU         // Compiling for GPU
 #define FFX_HLSL        // Compile for plain HLSL
 

Assets/Resources/FSR2/ffx_fsr2_autogen_reactive_pass.compute

@@ -1,5 +1,13 @@
 #pragma kernel CS
 
+#pragma multi_compile_local __ FFX_HALF
+#pragma multi_compile_local __ FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
+#pragma multi_compile_local __ FFX_FSR2_OPTION_HDR_COLOR_INPUT
+#pragma multi_compile_local __ FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_INVERTED_DEPTH
+#pragma multi_compile_local __ FFX_FSR2_OPTION_APPLY_SHARPENING
+
 #define FFX_GPU         // Compiling for GPU
 #define FFX_HLSL        // Compile for plain HLSL
 

Assets/Resources/FSR2/ffx_fsr2_compute_luminance_pyramid_pass.compute

@@ -1,5 +1,13 @@
 #pragma kernel CS
 
+//#pragma multi_compile_local __ FFX_HALF  // causes a hard-coded error message from the shader include ¯\_(ツ)_/¯
+#pragma multi_compile_local __ FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
+#pragma multi_compile_local __ FFX_FSR2_OPTION_HDR_COLOR_INPUT
+#pragma multi_compile_local __ FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_INVERTED_DEPTH
+#pragma multi_compile_local __ FFX_FSR2_OPTION_APPLY_SHARPENING
+
 #define FFX_GPU         // Compiling for GPU
 #define FFX_HLSL        // Compile for plain HLSL
 #define SPD_NO_WAVE_OPERATIONS  // Wave operations require shader model 6.0; this works with #pragma use_dxc but only on D3D12

Assets/Resources/FSR2/ffx_fsr2_depth_clip_pass.compute

@@ -1,5 +1,13 @@
 #pragma kernel CS
 
+#pragma multi_compile_local __ FFX_HALF
+#pragma multi_compile_local __ FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
+#pragma multi_compile_local __ FFX_FSR2_OPTION_HDR_COLOR_INPUT
+#pragma multi_compile_local __ FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_INVERTED_DEPTH
+#pragma multi_compile_local __ FFX_FSR2_OPTION_APPLY_SHARPENING
+
 #define FFX_GPU         // Compiling for GPU
 #define FFX_HLSL        // Compile for plain HLSL
 

Assets/Resources/FSR2/ffx_fsr2_lock_pass.compute

@@ -1,5 +1,13 @@
 #pragma kernel CS
 
+#pragma multi_compile_local __ FFX_HALF
+#pragma multi_compile_local __ FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
+#pragma multi_compile_local __ FFX_FSR2_OPTION_HDR_COLOR_INPUT
+#pragma multi_compile_local __ FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_INVERTED_DEPTH
+#pragma multi_compile_local __ FFX_FSR2_OPTION_APPLY_SHARPENING
+
 #define FFX_GPU         // Compiling for GPU
 #define FFX_HLSL        // Compile for plain HLSL
 

Assets/Resources/FSR2/ffx_fsr2_prepare_input_color_pass.compute

@@ -1,5 +1,13 @@
 #pragma kernel CS
 
+#pragma multi_compile_local __ FFX_HALF
+#pragma multi_compile_local __ FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
+#pragma multi_compile_local __ FFX_FSR2_OPTION_HDR_COLOR_INPUT
+#pragma multi_compile_local __ FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_INVERTED_DEPTH
+#pragma multi_compile_local __ FFX_FSR2_OPTION_APPLY_SHARPENING
+
 #define FFX_GPU         // Compiling for GPU
 #define FFX_HLSL        // Compile for plain HLSL
 

Assets/Resources/FSR2/ffx_fsr2_rcas_pass.compute

@@ -1,7 +1,12 @@
 // Each #kernel tells which function to compile; you can have many kernels
 #pragma kernel CS
-//#pragma kernel CS_HALF CS=CS_HALF FFX_HALF
-//#pragma kernel CS_HDR CS=CS_HDR FFX_FSR2_OPTION_HDR_COLOR_INPUT
+
+#pragma multi_compile_local __ FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
+#pragma multi_compile_local __ FFX_FSR2_OPTION_HDR_COLOR_INPUT
+#pragma multi_compile_local __ FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_INVERTED_DEPTH
+#pragma multi_compile_local __ FFX_FSR2_OPTION_APPLY_SHARPENING
 
 // Some global defines are needed
 #define FFX_GPU         // Compiling for GPU

Assets/Resources/FSR2/ffx_fsr2_reconstruct_previous_depth_pass.compute

@@ -1,5 +1,13 @@
 #pragma kernel CS
 
+#pragma multi_compile_local __ FFX_HALF
+#pragma multi_compile_local __ FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
+#pragma multi_compile_local __ FFX_FSR2_OPTION_HDR_COLOR_INPUT
+#pragma multi_compile_local __ FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_INVERTED_DEPTH
+#pragma multi_compile_local __ FFX_FSR2_OPTION_APPLY_SHARPENING
+
 #define FFX_GPU         // Compiling for GPU
 #define FFX_HLSL        // Compile for plain HLSL
 

Assets/Resources/FSR2/ffx_fsr2_tcr_autogen_pass.compute

@@ -1,5 +1,13 @@
 #pragma kernel CS
 
+#pragma multi_compile_local __ FFX_HALF
+#pragma multi_compile_local __ FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
+#pragma multi_compile_local __ FFX_FSR2_OPTION_HDR_COLOR_INPUT
+#pragma multi_compile_local __ FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
+#pragma multi_compile_local __ FFX_FSR2_OPTION_INVERTED_DEPTH
+#pragma multi_compile_local __ FFX_FSR2_OPTION_APPLY_SHARPENING
+
 #define FFX_GPU         // Compiling for GPU
 #define FFX_HLSL        // Compile for plain HLSL
 

Assets/Scripts/Fsr2Context.cs

@@ -55,6 +55,7 @@ namespace FidelityFX
             // Set defaults
             _firstExecution = true;
             _resourceFrameIndex = 0;
+            
             Constants.displaySize = _contextDescription.DisplaySize;
             
             // Generate the data for the LUT
@@ -91,11 +92,36 @@ namespace FidelityFX
 
         private void InitPipelines()
         {
-            _rcasPipeline = new Fsr2RcasPipeline(_contextDescription.Callbacks, _fsr2ConstantsBuffer, _rcasConstantsBuffer);
+            _computeLuminancePyramidPipeline = new Fsr2ComputeLuminancePyramidPipeline(_contextDescription.Callbacks, _contextDescription.Flags, _fsr2ConstantsBuffer);
+            _accumulatePipeline = new Fsr2AccumulatePipeline(_contextDescription.Callbacks, _contextDescription.Flags, _fsr2ConstantsBuffer);
+            _accumulateSharpenPipeline = new Fsr2AccumulateSharpenPipeline(_contextDescription.Callbacks, _contextDescription.Flags, _fsr2ConstantsBuffer);
+            _rcasPipeline = new Fsr2RcasPipeline(_contextDescription.Callbacks, _contextDescription.Flags, _fsr2ConstantsBuffer, _rcasConstantsBuffer);
+        }
+        
+        public void Destroy()
+        {
+            DisposePipeline(ref _tcrAutogeneratePipeline);
+            DisposePipeline(ref _generateReactivePipeline);
+            DisposePipeline(ref _computeLuminancePyramidPipeline);
+            DisposePipeline(ref _rcasPipeline);
+            DisposePipeline(ref _accumulateSharpenPipeline);
+            DisposePipeline(ref _accumulatePipeline);
+            DisposePipeline(ref _lockPipeline);
+            DisposePipeline(ref _reconstructPreviousDepthPipeline);
+            DisposePipeline(ref _depthClipPipeline);
+            
+            DestroyConstantBuffer(ref _rcasConstantsBuffer);
+            DestroyConstantBuffer(ref _spdConstantsBuffer);
+            DestroyConstantBuffer(ref _fsr2ConstantsBuffer);
+
+            _commandBuffer.Dispose();
+            _commandBuffer = null;
         }
 
         public void Dispatch(Fsr2.DispatchDescription dispatchParams)
         {
+            // TODO: validation & debug checking
+            
             _commandBuffer.Clear();
             
             // TODO: Should probably use a CommandBuffer here, to queue up all of the commands and dispatch them in one go
@@ -136,9 +162,9 @@ namespace FidelityFX
             _fsr2ConstantsBuffer.SetData(_fsr2ConstantsArray);
             _spdConstantsBuffer.SetData(_spdConstantsArray);
             
-            // // Compute luminance pyramid
-            // _commandBuffer.DispatchCompute(_computeLuminancePyramidShader, _computeLuminancePyramidKernel, dispatchThreadGroupCount.x, dispatchThreadGroupCount.y, 1);
-            //
+            // Compute luminance pyramid
+            _computeLuminancePyramidPipeline.ScheduleDispatch(_commandBuffer, dispatchParams, dispatchThreadGroupCount.x, dispatchThreadGroupCount.y);
+
             // // Reconstruct previous depth
             // _commandBuffer.DispatchCompute(_reconstructPreviousDepthShader, _reconstructPreviousDepthKernel, dispatchSrcX, dispatchSrcY, 1);
             //
@@ -148,10 +174,14 @@ namespace FidelityFX
             // // Lock
             // _commandBuffer.DispatchCompute(_lockShader, _lockKernel, dispatchSrcX, dispatchSrcY, 1);
             //
-            // // Accumulate
-            // _commandBuffer.DispatchCompute(_accumulateShader, _accumulateKernel, dispatchDstX, dispatchDstY, 1);    // TODO: accumulate + sharpen
 
-            if (dispatchParams.EnableSharpening)
+            bool sharpenEnabled = dispatchParams.EnableSharpening;
+            
+            // Accumulate
+            var accumulatePipeline = sharpenEnabled ? _accumulateSharpenPipeline : _accumulatePipeline;
+            accumulatePipeline.ScheduleDispatch(_commandBuffer, dispatchParams, dispatchDstX, dispatchDstY);
+
+            if (sharpenEnabled)
             {
                 // Compute the constants
                 SetupRcasConstants(dispatchParams);
@@ -322,26 +352,6 @@ namespace FidelityFX
             }
         }
 
-        public void Destroy()
-        {
-            DisposePipeline(ref _tcrAutogeneratePipeline);
-            DisposePipeline(ref _generateReactivePipeline);
-            DisposePipeline(ref _computeLuminancePyramidPipeline);
-            DisposePipeline(ref _rcasPipeline);
-            DisposePipeline(ref _accumulateSharpenPipeline);
-            DisposePipeline(ref _accumulatePipeline);
-            DisposePipeline(ref _lockPipeline);
-            DisposePipeline(ref _reconstructPreviousDepthPipeline);
-            DisposePipeline(ref _depthClipPipeline);
-            
-            DestroyConstantBuffer(ref _rcasConstantsBuffer);
-            DestroyConstantBuffer(ref _spdConstantsBuffer);
-            DestroyConstantBuffer(ref _fsr2ConstantsBuffer);
-
-            _commandBuffer.Dispose();
-            _commandBuffer = null;
-        }
-
         /// <summary>
         /// The FSR2 C++ codebase uses floats bitwise converted to ints to pass sharpness parameters to the RCAS shader.
         /// This is not possible in C# without enabling unsafe code compilation, so to avoid that we instead use a table of precomputed values.

Assets/Scripts/Fsr2Pipeline.cs

@@ -7,14 +7,16 @@ namespace FidelityFX
 {
     internal abstract class Fsr2Pipeline: IDisposable
     {
+        private readonly Fsr2Callbacks _callbacks;
+        protected readonly ComputeBuffer Constants;
+        
         protected ComputeShader ComputeShader;
         protected int KernelIndex;
 
-        private readonly Fsr2Callbacks _callbacks;
-
-        protected Fsr2Pipeline(Fsr2Callbacks callbacks)
+        protected Fsr2Pipeline(Fsr2Callbacks callbacks, ComputeBuffer constants)
         {
             _callbacks = callbacks;
+            Constants = constants;
         }
 
         public virtual void Dispose()
@@ -24,25 +26,32 @@ namespace FidelityFX
         
         public abstract void ScheduleDispatch(CommandBuffer commandBuffer, Fsr2.DispatchDescription dispatchParams, int dispatchX, int dispatchY);
         
-        protected void LoadComputeShader(string name)
+        protected void LoadComputeShader(string name, Fsr2.InitializationFlags flags)
         {
-            LoadComputeShader(name, ref ComputeShader, out KernelIndex);
+            LoadComputeShader(name, flags, ref ComputeShader, out KernelIndex);
         }
         
-        protected void LoadComputeShader(string name, ref ComputeShader shaderRef, out int kernelIndex)
+        private void LoadComputeShader(string name, Fsr2.InitializationFlags flags, ref ComputeShader shaderRef, out int kernelIndex)
         {
             if (shaderRef == null)
                 shaderRef = _callbacks.LoadComputeShader(name);
 
             kernelIndex = shaderRef.FindKernel("CS");
+            
+            if ((flags & Fsr2.InitializationFlags.EnableHighDynamicRange) != 0) shaderRef.EnableKeyword("FFX_FSR2_OPTION_HDR_COLOR_INPUT");
+            if ((flags & Fsr2.InitializationFlags.EnableDisplayResolutionMotionVectors) == 0) shaderRef.EnableKeyword("FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS");
+            if ((flags & Fsr2.InitializationFlags.EnableMotionVectorsJitterCancellation) != 0) shaderRef.EnableKeyword("FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS");
+            if ((flags & Fsr2.InitializationFlags.EnableDepthInverted) != 0) shaderRef.EnableKeyword("FFX_FSR2_OPTION_INVERTED_DEPTH");
+            // TODO: enable FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE if the device capabilities allow (default subgroup size == 32 or 64)
+            // TODO: enable FFX_HALF if FP16 is supported (except RCAS)
         }
 
-        protected void UnloadComputeShader()
+        private void UnloadComputeShader()
         {
             UnloadComputeShader(ref ComputeShader);
         }
         
-        protected void UnloadComputeShader(ref ComputeShader shaderRef)
+        private void UnloadComputeShader(ref ComputeShader shaderRef)
         {
             if (shaderRef == null)
                 return;
@@ -52,18 +61,78 @@ namespace FidelityFX
         }
     }
 
+    internal class Fsr2ComputeLuminancePyramidPipeline : Fsr2Pipeline
+    {
+        public Fsr2ComputeLuminancePyramidPipeline(Fsr2Callbacks callbacks, Fsr2.InitializationFlags flags, ComputeBuffer constants)
+            : base(callbacks, constants)
+        {
+            LoadComputeShader("FSR2/ffx_fsr2_compute_luminance_pyramid_pass", flags);
+        }
+
+        public override void ScheduleDispatch(CommandBuffer commandBuffer, Fsr2.DispatchDescription dispatchParams, int dispatchX, int dispatchY)
+        {
+            //throw new NotImplementedException();
+        }
+    }
+
+    internal class Fsr2AccumulatePipeline : Fsr2Pipeline
+    {
+        public Fsr2AccumulatePipeline(Fsr2Callbacks callbacks, Fsr2.InitializationFlags flags, ComputeBuffer constants)
+            : base(callbacks, constants)
+        {
+            LoadComputeShader("FSR2/ffx_fsr2_accumulate_pass", flags);
+        }
+
+        public override void ScheduleDispatch(CommandBuffer commandBuffer, Fsr2.DispatchDescription dispatchParams, int dispatchX, int dispatchY)
+        {
+            //throw new NotImplementedException();
+        }
+    }
+
+    internal class Fsr2AccumulateSharpenPipeline : Fsr2AccumulatePipeline
+    {
+        private readonly ComputeShader _shaderCopy;
+        
+        public Fsr2AccumulateSharpenPipeline(Fsr2Callbacks callbacks, Fsr2.InitializationFlags flags, ComputeBuffer constants)
+            : base(callbacks, flags, constants)
+        {
+            // Simply loading the accumulate_pass compute shader will give us the same instance as the non-sharpen pipeline
+            // So we have to clone the shader instance and set the extra keyword on the new copy
+            _shaderCopy = UnityEngine.Object.Instantiate(ComputeShader);
+            foreach (var keyword in ComputeShader.enabledKeywords)
+            {
+                _shaderCopy.EnableKeyword(keyword.name);
+            }
+            _shaderCopy.EnableKeyword("FFX_FSR2_OPTION_APPLY_SHARPENING");
+        }
+
+        public override void ScheduleDispatch(CommandBuffer commandBuffer, Fsr2.DispatchDescription dispatchParams, int dispatchX, int dispatchY)
+        {
+            // Temporarily swap around the shaders so that the dispatch will bind and execute the correct one
+            ComputeShader tmp = ComputeShader;
+            ComputeShader = _shaderCopy;
+            base.ScheduleDispatch(commandBuffer, dispatchParams, dispatchX, dispatchY);
+            ComputeShader = tmp;
+        }
+
+        public override void Dispose()
+        {
+            // Since we instantiated this copy, we have to destroy it instead of unloading the shader resource
+            UnityEngine.Object.Destroy(_shaderCopy);
+            base.Dispose();
+        }
+    }
+    
     internal class Fsr2RcasPipeline : Fsr2Pipeline
     {
-        private readonly ComputeBuffer _fsr2Constants;
         private readonly ComputeBuffer _rcasConstants;
         
-        public Fsr2RcasPipeline(Fsr2Callbacks callbacks, ComputeBuffer fsr2Constants, ComputeBuffer rcasConstants)
-            : base(callbacks)
+        public Fsr2RcasPipeline(Fsr2Callbacks callbacks, Fsr2.InitializationFlags flags, ComputeBuffer constants, ComputeBuffer rcasConstants)
+            : base(callbacks, constants)
         {
-            _fsr2Constants = fsr2Constants;
             _rcasConstants = rcasConstants;
             
-            LoadComputeShader("FSR2/ffx_fsr2_rcas_pass");
+            LoadComputeShader("FSR2/ffx_fsr2_rcas_pass", flags);
         }
 
         public override void ScheduleDispatch(CommandBuffer commandBuffer, Fsr2.DispatchDescription dispatchParams, int dispatchX, int dispatchY)
@@ -72,7 +141,7 @@ namespace FidelityFX
             ComputeShader.SetTexture(KernelIndex, "r_input_exposure", dispatchParams.Exposure);
             ComputeShader.SetTexture(KernelIndex, "r_rcas_input", dispatchParams.Input);
             ComputeShader.SetTexture(KernelIndex, "rw_upscaled_output", dispatchParams.Output);
-            ComputeShader.SetConstantBuffer("cbFSR2", _fsr2Constants, 0, Marshal.SizeOf<Fsr2.Fsr2Constants>());
+            ComputeShader.SetConstantBuffer("cbFSR2", Constants, 0, Marshal.SizeOf<Fsr2.Fsr2Constants>());
             ComputeShader.SetConstantBuffer("cbRCAS", _rcasConstants, 0, Marshal.SizeOf<Fsr2.RcasConstants>());
 
             commandBuffer.DispatchCompute(ComputeShader, KernelIndex, dispatchX, dispatchY, 1);