Initial port of SGSR2 2-pass compute shaders to HLSL (untested)

Packages/com.unity.postprocessing@3.2.2/PostProcessing/PostProcessResources.asset

@@ -150,6 +150,10 @@ MonoBehaviour:
       tcrAutoGenPass: {fileID: 7200000, guid: 75cdc6ef23f08ed498d4da511923fcea, type: 3}
       debugViewPass: {fileID: 7200000, guid: cb24a71d54164c54eb5e86839acd48c5, type: 3}
     sgsr2Upscaler:
-      convert: {fileID: 7200000, guid: a41757aacd8b70e42a4001d514bfbe53, type: 3}
-      activate: {fileID: 7200000, guid: d7de362950af6fe4e90da7d6e32f9826, type: 3}
-      upscale: {fileID: 7200000, guid: 5d28d29787492b74aa736a21f70572c7, type: 3}
+      twoPassCompute:
+        convert: {fileID: 7200000, guid: 073ee927fbee25841a31cf364834071c, type: 3}
+        upscale: {fileID: 7200000, guid: d7bacd7d04c6521499bef936d93921cc, type: 3}
+      threePassCompute:
+        convert: {fileID: 7200000, guid: a41757aacd8b70e42a4001d514bfbe53, type: 3}
+        activate: {fileID: 7200000, guid: d7de362950af6fe4e90da7d6e32f9826, type: 3}
+        upscale: {fileID: 7200000, guid: 5d28d29787492b74aa736a21f70572c7, type: 3}

Packages/com.unity.postprocessing@3.2.2/PostProcessing/Runtime/Effects/Upscaling/SGSR2/SGSR2.cs

@@ -3,7 +3,7 @@ using System.Collections;
 using System.Collections.Generic;
 using UnityEngine;
 
-public class SGSR2
+public static class SGSR2
 {
     [Serializable]
     public struct Params
@@ -26,10 +26,22 @@ public class SGSR2
     [Serializable]
     public class Shaders
     {
-        public ComputeShader convert;
+        public TwoPassCompute twoPassCompute;
+        public ThreePassCompute threePassCompute;
         
-        public ComputeShader activate;
+        [Serializable]
+        public class TwoPassCompute
+        {
+            public ComputeShader convert;
+            public ComputeShader upscale;
+        }
         
-        public ComputeShader upscale;
+        [Serializable]
+        public class ThreePassCompute
+        {
+            public ComputeShader convert;
+            public ComputeShader activate;
+            public ComputeShader upscale;
+        }
     }
 }

Packages/com.unity.postprocessing@3.2.2/PostProcessing/Runtime/Effects/Upscaling/SGSR2/Shaders/2_pass_cs.meta

@@ -0,0 +1,8 @@
+fileFormatVersion: 2
+guid: def18d58a2ff64f44a2d9f73e487a689
+folderAsset: yes
+DefaultImporter:
+  externalObjects: {}
+  userData: 
+  assetBundleName: 
+  assetBundleVariant: 

Packages/com.unity.postprocessing@3.2.2/PostProcessing/Runtime/Effects/Upscaling/SGSR2/Shaders/2_pass_cs/sgsr2_convert.compute

@@ -0,0 +1,136 @@
+#pragma kernel CS
+
+//============================================================================================================
+//
+//
+//                  Copyright (c) 2024, Qualcomm Innovation Center, Inc. All rights reserved.
+//                              SPDX-License-Identifier: BSD-3-Clause
+//
+//============================================================================================================
+
+float2 decodeVelocityFromTexture(float2 ev) {
+    const float inv_div = 1.0f / (0.499f * 0.5f);
+    float2 dv;
+    dv.xy = ev.xy * inv_div - 32767.0f / 65535.0f * inv_div;
+    //dv.z = uintBitsToFloat((uint(round(ev.z * 65535.0f)) << 16) | uint(round(ev.w * 65535.0f)));
+    return dv;
+}
+
+Texture2D           InputColor                  : register(t0);
+Texture2D<float>    InputDepth                  : register(t1);
+Texture2D<float2>   InputVelocity               : register(t2);
+RWTexture2D<float4> MotionDepthClipAlphaBuffer  : register(u0);
+RWTexture2D<uint>   YCoCgColor                  : register(u1);
+
+cbuffer Params : register(b0)
+{
+    uint2                renderSize;
+    uint2                displaySize;
+    float2               renderSizeRcp;
+    float2               displaySizeRcp;
+    float2               jitterOffset;
+    float2               padding1;
+    float4               clipToPrevClip[4];
+    float                preExposure;
+    float                cameraFovAngleHor;
+    float                cameraNear;
+    float                MinLerpContribution;
+    uint                 bSameCamera;
+    uint                 reset;
+};
+
+SamplerState s_PointClamp   : register(s0);
+SamplerState s_LinearClamp  : register(s1);
+
+[numthreads(8, 8, 1)]
+void CS(uint3 globalInvocationID : SV_DispatchThreadID)
+{
+    float Exposure_co_rcp = preExposure;
+    float2 ViewportSizeInverse = displaySizeRcp.xy;
+    uint2 InputPos = globalInvocationID.xy;
+
+    float2 gatherCoord = float2(globalInvocationID.xy) * ViewportSizeInverse;
+    float2 ViewportUV = gatherCoord + 0.5f * ViewportSizeInverse;
+
+    //derived from ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h
+    //FindNearestDepth
+
+    float4 topleft = InputDepth.GatherRed(s_PointClamp, gatherCoord);
+    float2 v10 = float2(ViewportSizeInverse.x*2.0, 0.0);
+    float4 topRight = InputDepth.GatherRed(s_PointClamp, (gatherCoord+v10));
+    float2 v12 = float2(0.0, ViewportSizeInverse.y*2.0);
+    float4 bottomLeft = InputDepth.GatherRed(s_PointClamp, (gatherCoord+v12));
+    float2 v14 = float2(ViewportSizeInverse.x*2.0, ViewportSizeInverse.y*2.0);
+    float4 bottomRight = InputDepth.GatherRed(s_PointClamp, (gatherCoord+v14));
+    float maxC = max(max(max(topleft.y,topRight.x),bottomLeft.z),bottomRight.w);
+    float topleft4 = max(max(max(topleft.y,topleft.x),topleft.z),topleft.w);
+    float topLeftMax9 = max(bottomLeft.w,max(max(maxC,topleft4),topRight.w));
+
+    float depthclip = 0.0;
+    if (maxC > 1.0e-05f)
+    {
+        float topRight4 = max(max(max(topRight.y,topRight.x),topRight.z),topRight.w);
+        float bottomLeft4 = max(max(max(bottomLeft.y,bottomLeft.x),bottomLeft.z),bottomLeft.w);
+        float bottomRight4 = max(max(max(bottomRight.y,bottomRight.x),bottomRight.z),bottomRight.w);
+
+        float Wdepth = 0.0;
+        float Ksep = 1.37e-05f;
+        float Kfov = cameraFovAngleHor;
+        float diagonal_length = length(float2(renderSize));
+        float Ksep_Kfov_diagonal = Ksep * Kfov * diagonal_length;
+
+        float Depthsep = Ksep_Kfov_diagonal * (1.0 - maxC);
+        float EPSILON = 1.19e-07f;
+        Wdepth += clamp((Depthsep / (abs(maxC - topleft4) + EPSILON)), 0.0, 1.0);
+        Wdepth += clamp((Depthsep / (abs(maxC - topRight4) + EPSILON)), 0.0, 1.0);
+        Wdepth += clamp((Depthsep / (abs(maxC - bottomLeft4) + EPSILON)), 0.0, 1.0);
+        Wdepth += clamp((Depthsep / (abs(maxC - bottomRight4) + EPSILON)), 0.0, 1.0);
+        depthclip = clamp(1.0f - Wdepth*0.25, 0.0, 1.0);
+    }
+
+    //refer to ue/fsr2 PostProcessFFX_FSR2ConvertVelocity.usf, and using nearest depth for dilated motion
+
+    float2 EncodedVelocity = InputVelocity[InputPos];
+
+    float2 motion;
+    if (EncodedVelocity.x > 0.0)
+    {
+        motion = decodeVelocityFromTexture(EncodedVelocity.xy);
+    }
+    else
+    {
+#ifdef REQUEST_NDC_Y_UP
+        float2 ScreenPos = float2(2.0f * ViewportUV.x - 1.0f, 1.0f - 2.0f * ViewportUV.y);
+#else
+         float2 ScreenPos = float2(2.0f * ViewportUV - 1.0f);
+#endif
+        float3 Position = float3(ScreenPos, topLeftMax9);    //this_clip
+        float4 PreClip = clipToPrevClip[3] + ((clipToPrevClip[2] * Position.z) + ((clipToPrevClip[1] * ScreenPos.y) + (clipToPrevClip[0] * ScreenPos.x)));
+        float2 PreScreen = PreClip.xy / PreClip.w;
+        motion = Position.xy - PreScreen;
+    }
+
+    motion = EncodedVelocity;
+
+    ////////////compute luma
+    float3 Colorrgb = InputColor[InputPos].xyz;
+
+    ///simple tonemap
+    float ColorMax = max(max(Colorrgb.x, Colorrgb.y), Colorrgb.z) + Exposure_co_rcp;
+    Colorrgb /= ColorMax;
+
+    float3 Colorycocg;
+    Colorycocg.x = 0.25 * (Colorrgb.x + 2.0 * Colorrgb.y + Colorrgb.z);
+    Colorycocg.y = clamp(0.5 * Colorrgb.x + 0.5 - 0.5 * Colorrgb.z, 0.0, 1.0);
+    Colorycocg.z = clamp(Colorycocg.x + Colorycocg.y - Colorrgb.x, 0.0, 1.0);
+
+    //now color YCoCG all in the range of [0,1]
+    uint x11 = uint(Colorycocg.x * 2047.5);
+    uint y11 = uint(Colorycocg.y * 2047.5);
+    uint z10 = uint(Colorycocg.z * 1023.5);
+
+    YCoCgColor[InputPos] = ((x11 << 21u) | (y11 << 10u)) | z10;
+
+    float4 v29 = float4(motion, depthclip, ColorMax);
+    MotionDepthClipAlphaBuffer[InputPos] = v29;
+}

Packages/com.unity.postprocessing@3.2.2/PostProcessing/Runtime/Effects/Upscaling/SGSR2/Shaders/2_pass_cs/sgsr2_convert.compute.meta

@@ -0,0 +1,7 @@
+fileFormatVersion: 2
+guid: 073ee927fbee25841a31cf364834071c
+ComputeShaderImporter:
+  externalObjects: {}
+  userData: 
+  assetBundleName: 
+  assetBundleVariant: 

Packages/com.unity.postprocessing@3.2.2/PostProcessing/Runtime/Effects/Upscaling/SGSR2/Shaders/2_pass_cs/sgsr2_upscale.compute

@@ -0,0 +1,332 @@
+#pragma kernel CS
+
+//============================================================================================================
+//
+//
+//                  Copyright (c) 2024, Qualcomm Innovation Center, Inc. All rights reserved.
+//                              SPDX-License-Identifier: BSD-3-Clause
+//
+//============================================================================================================
+
+float FastLanczos(float base)
+{
+    float y = base - 1.0f;
+    float y2 = y * y;
+    float y_temp = 0.75f * y + y2;
+    return y_temp * y2;
+}
+
+float3 DecodeColor(uint sample32)
+{
+    uint x11 = sample32 >> 21u;
+    uint y11 = sample32 & (2047u << 10u);
+    uint z10 = sample32 & 1023u;
+    float3 samplecolor;
+    samplecolor.x = (float(x11) * (1.0 / 2047.5));
+    samplecolor.y = (float(y11) * (4.76953602e-7)) - 0.5;
+    samplecolor.z = (float(z10) * (1.0 / 1023.5)) - 0.5;
+
+    return samplecolor;
+}
+
+Texture2D           PrevHistoryOutput           : register(t0);
+Texture2D           MotionDepthClipAlphaBuffer  : register(t1);
+Texture2D<uint>     YCoCgColor                  : register(t2);
+RWTexture2D<float4> SceneColorOutput            : register(u0);
+RWTexture2D<float4> HistoryOutput               : register(u1);
+
+cbuffer Params : register(b0)
+{
+    uint2                renderSize;
+    uint2                displaySize;
+    float2               renderSizeRcp;
+    float2               displaySizeRcp;
+    float2               jitterOffset;
+    float2               padding1;
+    float4               clipToPrevClip[4];
+    float                preExposure;
+    float                cameraFovAngleHor;
+    float                cameraNear;
+    float                MinLerpContribution;
+    uint                 bSameCamera;
+    uint                 reset;
+};
+
+SamplerState s_PointClamp   : register(s0);
+SamplerState s_LinearClamp  : register(s1);
+
+[numthreads(8, 8, 1)]
+void CS(uint3 globalInvocationID : SV_DispatchThreadID)
+{
+    float Biasmax_viewportXScale = min(float(displaySize.x) / float(renderSize.x), 1.99);  //Biasmax_viewportXScale
+    float scalefactor = min(20.0, pow((float(displaySize.x) / float(renderSize.x)) * (float(displaySize.y) / float(renderSize.y)), 3.0));
+    float f2 = preExposure;            //1.0;   //preExposure
+    float2 HistoryInfoViewportSizeInverse = displaySizeRcp;
+    float2 HistoryInfoViewportSize = float2(displaySize);
+    float2 InputJitter = jitterOffset;
+    float2 InputInfoViewportSize = float2(renderSize);
+    float2 Hruv = (float2(globalInvocationID.xy) + 0.5f) * HistoryInfoViewportSizeInverse;
+    float2 Jitteruv;
+    Jitteruv.x = clamp(Hruv.x + (InputJitter.x * HistoryInfoViewportSizeInverse.x), 0.0, 1.0);
+    Jitteruv.y = clamp(Hruv.y + (InputJitter.y * HistoryInfoViewportSizeInverse.y), 0.0, 1.0);
+
+    int2 InputPos = int2(Jitteruv * InputInfoViewportSize);
+    float4 mda = MotionDepthClipAlphaBuffer.SampleLevel(s_LinearClamp, Jitteruv, 0).xyzw;
+    float2 Motion = mda.xy;
+
+    ///ScreenPosToViewportScale&Bias
+    float2 PrevUV;
+    PrevUV.x = clamp(-0.5 * Motion.x + Hruv.x, 0.0, 1.0);
+#ifdef REQUEST_NDC_Y_UP
+    PrevUV.y = clamp(0.5 * Motion.y + Hruv.y, 0.0, 1.0);
+#else
+    PrevUV.y = clamp(-0.5 * Motion.y + Hruv.y, 0.0, 1.0);
+#endif
+
+    float depthfactor = mda.z;
+    float ColorMax = mda.w;
+
+    float4 History = PrevHistoryOutput.SampleLevel(s_LinearClamp, PrevUV, 0);
+    float3 HistoryColor = History.xyz;
+    float Historyw = History.w;
+    float Wfactor = clamp(abs(Historyw), 0.0, 1.0);
+
+    /////upsample and compute box
+    float4 Upsampledcw = 0.0f;
+    float kernelfactor = clamp(Wfactor + float(reset), 0.0, 1.0);
+    float biasmax = Biasmax_viewportXScale - Biasmax_viewportXScale * kernelfactor;
+    float biasmin = max(1.0f, 0.3 + 0.3 * biasmax);
+    float biasfactor = max(0.25f * depthfactor, kernelfactor);
+    float kernelbias = lerp(biasmax, biasmin, biasfactor);
+    float motion_viewport_len = length(Motion * HistoryInfoViewportSize);
+    float curvebias = lerp(-2.0, -3.0, clamp(motion_viewport_len * 0.02, 0.0, 1.0));
+
+    float3 rectboxcenter = 0.0f;
+    float3 rectboxvar = 0.0f;
+    float rectboxweight = 0.0f;
+    float2 srcpos = float2(InputPos) + 0.5f - InputJitter;
+    float2 srcOutputPos = Hruv * InputInfoViewportSize;
+
+    kernelbias *= 0.5f;
+    float kernelbias2 = kernelbias * kernelbias;
+    float2 srcpos_srcOutputPos = srcpos - srcOutputPos;
+
+    int2 InputPosBtmRight = 1 + InputPos;
+    float2 gatherCoord = float2(InputPos) * renderSizeRcp;
+    uint btmRight = YCoCgColor[InputPosBtmRight].x;
+    uint4 topleft = YCoCgColor.GatherRed(s_PointClamp, gatherCoord);
+    uint2 topRight = 0;
+    uint2 bottomLeft = 0;
+
+    uint sameCameraFrmNum = bSameCamera;
+
+    if (sameCameraFrmNum!=0u)
+    {
+        topRight = YCoCgColor.GatherRed(s_PointClamp, gatherCoord + float2(renderSizeRcp.x, 0.0)).yz;
+        bottomLeft = YCoCgColor.GatherRed(s_PointClamp, gatherCoord + float2(0.0, renderSizeRcp.y)).xy;
+    }
+    else
+    {
+        uint2 btmRight = YCoCgColor.GatherRed(s_PointClamp, gatherCoord + float2(renderSizeRcp.x, renderSizeRcp.y)).xz;
+        bottomLeft.y = btmRight.x;
+        topRight.x = btmRight.y;
+    }
+
+    float3 rectboxmin;
+    float3 rectboxmax;
+    {
+        float3 samplecolor = DecodeColor(bottomLeft.y);
+        float2 baseoffset = srcpos_srcOutputPos + float2(0.0, 1.0);
+        float baseoffset_dot = dot(baseoffset, baseoffset);
+        float base = clamp(baseoffset_dot * kernelbias2, 0.0f, 1.0f);
+        float weight = FastLanczos(base);
+        Upsampledcw += float4(samplecolor * weight, weight);
+        float boxweight = exp(baseoffset_dot * curvebias);
+        rectboxmin = samplecolor;
+        rectboxmax = samplecolor;
+        float3 wsample = samplecolor * boxweight;
+        rectboxcenter += wsample;
+        rectboxvar += (samplecolor * wsample);
+        rectboxweight += boxweight;
+    }
+    {
+        float3 samplecolor = DecodeColor(topRight.x);
+        float2 baseoffset = srcpos_srcOutputPos + float2(1.0, 0.0);
+        float baseoffset_dot = dot(baseoffset, baseoffset);
+        float base = clamp(baseoffset_dot * kernelbias2, 0.0f, 1.0f);
+        float weight = FastLanczos(base);
+        Upsampledcw += float4(samplecolor * weight, weight);
+        float boxweight = exp(baseoffset_dot * curvebias);
+        rectboxmin = min(rectboxmin, samplecolor);
+        rectboxmax = max(rectboxmax, samplecolor);
+        float3 wsample = samplecolor * boxweight;
+        rectboxcenter += wsample;
+        rectboxvar += (samplecolor * wsample);
+        rectboxweight += boxweight;
+    }
+    {
+        float3 samplecolor = DecodeColor(topleft.x);
+        float2 baseoffset = srcpos_srcOutputPos + float2(-1.0, 0.0);
+        float baseoffset_dot = dot(baseoffset, baseoffset);
+        float base = clamp(baseoffset_dot * kernelbias2, 0.0f, 1.0f);
+        float weight = FastLanczos(base);
+        Upsampledcw += float4(samplecolor * weight, weight);
+        float boxweight = exp(baseoffset_dot * curvebias);
+        rectboxmin = min(rectboxmin, samplecolor);
+        rectboxmax = max(rectboxmax, samplecolor);
+        float3 wsample = samplecolor * boxweight;
+        rectboxcenter += wsample;
+        rectboxvar += (samplecolor * wsample);
+        rectboxweight += boxweight;
+    }
+    {
+        float3 samplecolor = DecodeColor(topleft.y);
+        float2 baseoffset = srcpos_srcOutputPos;
+        float baseoffset_dot = dot(baseoffset, baseoffset);
+        float base = clamp(baseoffset_dot * kernelbias2, 0.0f, 1.0f);
+        float weight = FastLanczos(base);
+        Upsampledcw += float4(samplecolor * weight, weight);
+        float boxweight = exp(baseoffset_dot * curvebias);
+        rectboxmin = min(rectboxmin, samplecolor);
+        rectboxmax = max(rectboxmax, samplecolor);
+        float3 wsample = samplecolor * boxweight;
+        rectboxcenter += wsample;
+        rectboxvar += (samplecolor * wsample);
+        rectboxweight += boxweight;
+    }
+    {
+        float3 samplecolor = DecodeColor(topleft.z);
+        float2 baseoffset = srcpos_srcOutputPos + float2(0.0, -1.0);
+        float baseoffset_dot = dot(baseoffset, baseoffset);
+        float base = clamp(baseoffset_dot * kernelbias2, 0.0f, 1.0f);
+        float weight = FastLanczos(base);
+        Upsampledcw += float4(samplecolor * weight, weight);
+        float boxweight = exp(baseoffset_dot * curvebias);
+        rectboxmin = min(rectboxmin, samplecolor);
+        rectboxmax = max(rectboxmax, samplecolor);
+        float3 wsample = samplecolor * boxweight;
+        rectboxcenter += wsample;
+        rectboxvar += (samplecolor * wsample);
+        rectboxweight += boxweight;
+    }
+
+    if (sameCameraFrmNum!=0u)
+    {
+        {
+            float3 samplecolor = DecodeColor(btmRight);
+            float2 baseoffset = srcpos_srcOutputPos + float2(1.0, 1.0);
+            float baseoffset_dot = dot(baseoffset, baseoffset);
+            float base = clamp(baseoffset_dot * kernelbias2, 0.0, 1.0);
+            float weight = FastLanczos(base);
+            Upsampledcw += float4(samplecolor * weight, weight);
+            float boxweight = exp(baseoffset_dot * curvebias);
+            rectboxmin = min(rectboxmin, samplecolor);
+            rectboxmax = max(rectboxmax, samplecolor);
+            float3 wsample = samplecolor * boxweight;
+            rectboxcenter += wsample;
+            rectboxvar += (samplecolor * wsample);
+            rectboxweight += boxweight;
+        }
+        {
+            float3 samplecolor = DecodeColor(bottomLeft.x);
+            float2 baseoffset = srcpos_srcOutputPos + float2(-1.0, 1.0);
+            float baseoffset_dot = dot(baseoffset, baseoffset);
+            float base = clamp(baseoffset_dot * kernelbias2, 0.0f, 1.0f);
+            float weight = FastLanczos(base);
+            Upsampledcw += float4(samplecolor * weight, weight);
+            float boxweight = exp(baseoffset_dot * curvebias);
+            rectboxmin = min(rectboxmin, samplecolor);
+            rectboxmax = max(rectboxmax, samplecolor);
+            float3 wsample = samplecolor * boxweight;
+            rectboxcenter += wsample;
+            rectboxvar += (samplecolor * wsample);
+            rectboxweight += boxweight;
+        }
+        {
+            float3 samplecolor = DecodeColor(topRight.y);
+            float2 baseoffset = srcpos_srcOutputPos + float2(1.0, -1.0);
+            float baseoffset_dot = dot(baseoffset, baseoffset);
+            float base = clamp(baseoffset_dot * kernelbias2, 0.0f, 1.0f);
+            float weight = FastLanczos(base);
+            Upsampledcw += float4(samplecolor * weight, weight);
+            float boxweight = exp(baseoffset_dot * curvebias);
+            rectboxmin = min(rectboxmin, samplecolor);
+            rectboxmax = max(rectboxmax, samplecolor);
+            float3 wsample = samplecolor * boxweight;
+            rectboxcenter += wsample;
+            rectboxvar += (samplecolor * wsample);
+            rectboxweight += boxweight;
+        }
+
+        {
+            float3 samplecolor = DecodeColor(topleft.w);
+            float2 baseoffset = srcpos_srcOutputPos + float2(-1.0, -1.0);
+            float baseoffset_dot = dot(baseoffset, baseoffset);
+            float base = clamp(baseoffset_dot * kernelbias2, 0.0f, 1.0f);
+            float weight = FastLanczos(base);
+            Upsampledcw += float4(samplecolor * weight, weight);
+            float boxweight = exp(baseoffset_dot * curvebias);
+            rectboxmin = min(rectboxmin, samplecolor);
+            rectboxmax = max(rectboxmax, samplecolor);
+            float3 wsample = samplecolor * boxweight;
+            rectboxcenter += wsample;
+            rectboxvar += (samplecolor * wsample);
+            rectboxweight += boxweight;
+        }
+    }
+
+    rectboxweight = 1.0 / rectboxweight;
+    rectboxcenter *= rectboxweight;
+    rectboxvar *= rectboxweight;
+    rectboxvar = sqrt(abs(rectboxvar - rectboxcenter * rectboxcenter));
+
+    Upsampledcw.xyz =  clamp(Upsampledcw.xyz / Upsampledcw.w, rectboxmin-0.05f, rectboxmax+0.05f);
+    Upsampledcw.w = Upsampledcw.w * (1.0f / 3.0f) ;
+
+    float OneMinusWfactor = 1.0f - Wfactor;
+
+    float baseupdate = OneMinusWfactor - OneMinusWfactor * depthfactor;
+    baseupdate = min(baseupdate, lerp(baseupdate, Upsampledcw.w *10.0f, clamp(10.0f* motion_viewport_len, 0.0, 1.0)));
+    baseupdate = min(baseupdate, lerp(baseupdate, Upsampledcw.w, clamp(motion_viewport_len *0.05f, 0.0, 1.0)));
+    float basealpha = baseupdate;
+
+    const float EPSILON = 1.192e-07f;
+    float boxscale = max(depthfactor, clamp(motion_viewport_len * 0.05f, 0.0, 1.0));
+    float boxsize = lerp(scalefactor, 1.0f, boxscale);
+    float3 sboxvar = rectboxvar * boxsize;
+    float3 boxmin = rectboxcenter - sboxvar;
+    float3 boxmax = rectboxcenter + sboxvar;
+    rectboxmax = min(rectboxmax, boxmax);
+    rectboxmin = max(rectboxmin, boxmin);
+
+    float3 clampedcolor = clamp(HistoryColor, rectboxmin, rectboxmax);
+    float startLerpValue = MinLerpContribution; //MinLerpContribution; //MinLerpContribution;
+    if ((abs(mda.x) + abs(mda.y)) > 0.000001) startLerpValue = 0.0;
+    float lerpcontribution = (any(rectboxmin > HistoryColor) || any(HistoryColor > rectboxmax)) ? startLerpValue : 1.0f;
+
+    HistoryColor = lerp(clampedcolor, HistoryColor, clamp(lerpcontribution, 0.0, 1.0));
+    float basemin = min(basealpha, 0.1f);
+    basealpha = lerp(basemin, basealpha, clamp(lerpcontribution, 0.0, 1.0));
+
+    ////blend color
+    float alphasum = max(EPSILON, basealpha + Upsampledcw.w);
+    float alpha = clamp(Upsampledcw.w / alphasum + float(reset), 0.0, 1.0);
+    Upsampledcw.xyz = lerp(HistoryColor, Upsampledcw.xyz, alpha);
+    
+    HistoryOutput[globalInvocationID.xy] = float4(Upsampledcw.xyz, Wfactor);
+
+    ////ycocg to rgb
+    float x_z = Upsampledcw.x - Upsampledcw.z;
+    Upsampledcw.xyz = float3(
+        clamp(x_z + Upsampledcw.y, 0.0, 1.0),
+        clamp(Upsampledcw.x + Upsampledcw.z, 0.0, 1.0),
+        clamp(x_z - Upsampledcw.y, 0.0, 1.0));
+
+    float compMax = max(Upsampledcw.x, Upsampledcw.y);
+    compMax = clamp(max(compMax, Upsampledcw.z), 0.0f, 1.0f);
+    float scale = preExposure /  ((1.0f + 600.0f / 65504.0f) - compMax);
+
+    if (ColorMax > 4000.0f) scale = ColorMax;
+    Upsampledcw.xyz = Upsampledcw.xyz * scale;
+    SceneColorOutput[globalInvocationID.xy] = Upsampledcw;
+}

Packages/com.unity.postprocessing@3.2.2/PostProcessing/Runtime/Effects/Upscaling/SGSR2/Shaders/2_pass_cs/sgsr2_upscale.compute.meta

@@ -0,0 +1,7 @@
+fileFormatVersion: 2
+guid: d7bacd7d04c6521499bef936d93921cc
+ComputeShaderImporter:
+  externalObjects: {}
+  userData: 
+  assetBundleName: 
+  assetBundleVariant: 

Packages/com.unity.postprocessing@3.2.2/PostProcessing/Runtime/Effects/Upscaling/SGSR2Upscaler.cs

@@ -53,7 +53,7 @@ namespace UnityEngine.Rendering.PostProcessing
             parms.renderSizeRcp = new Vector2(1.0f / parms.renderSize.x, 1.0f / parms.renderSize.y);
             parms.displaySizeRcp = new Vector2(1.0f / parms.displaySize.x, 1.0f / parms.displaySize.y);
             parms.jitterOffset = config.JitterOffset;
-            parms.clipToPrevClip = Matrix4x4.identity; // TODO: clipToPrevClip
+            parms.clipToPrevClip = Matrix4x4.identity; // TODO: clipToPrevClip => (previous_view_proj * inv_vp)
             parms.preExposure = config.preExposure;
             parms.cameraFovAngleHor = Mathf.Tan(context.camera.fieldOfView * Mathf.Deg2Rad * 0.5f) * (float)parms.renderSize.x / parms.renderSize.y;
             parms.cameraNear = context.camera.nearClipPlane;
@@ -84,7 +84,7 @@ namespace UnityEngine.Rendering.PostProcessing
 
         private void Convert(CommandBuffer cmd, PostProcessRenderContext context, Upscaling config)
         {
-            var shader = context.resources.computeShaders.sgsr2Upscaler.convert;
+            var shader = context.resources.computeShaders.sgsr2Upscaler.threePassCompute.convert;
             int kernelIndex = shader.FindKernel("CS");
             
             cmd.SetComputeConstantBufferParam(shader, "Params", _paramsBuffer, 0, Marshal.SizeOf<SGSR2.Params>());
@@ -103,7 +103,7 @@ namespace UnityEngine.Rendering.PostProcessing
 
         private void Activate(CommandBuffer cmd, PostProcessRenderContext context)
         {
-            var shader = context.resources.computeShaders.sgsr2Upscaler.activate;
+            var shader = context.resources.computeShaders.sgsr2Upscaler.threePassCompute.activate;
             int kernelIndex = shader.FindKernel("CS");
             uint frameIndex = _frameCount % 2;
             
@@ -122,7 +122,7 @@ namespace UnityEngine.Rendering.PostProcessing
 
         private void Upscale(CommandBuffer cmd, PostProcessRenderContext context)
         {
-            var shader = context.resources.computeShaders.sgsr2Upscaler.upscale;
+            var shader = context.resources.computeShaders.sgsr2Upscaler.threePassCompute.upscale;
             int kernelIndex = shader.FindKernel("CS");
             uint frameIndex = _frameCount % 2;