v4.8.1:

HIGHLIGHTS: - bug fixes and improvements DETAILS: - REBLUR_SH: fixed mis-normalized weighted sum for "diffSh" and "specSh" in "HistoryFix" pass (the regression appeared after finding a better initial weight for the center pixel) - REBLUR_SH: minor performance optimization in "Temporal Stabilization" pass - REBLUR_SH: fixed "touching" of "specSh.w" in "HistoryFix" pass, which is "modified roughness" and must not be processed in any way
NVIDIAGameWorks · May 16, 2024 · 4398316 · 4398316
1 parent 8d3c245
commit 4398316
Show file tree

Hide file tree

Showing 7 changed files with 22 additions and 14 deletions.
diff --git a/Include/NRD.h b/Include/NRD.h
@@ -11,7 +11,7 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 /*
 CREDITS:
     Developed by:
-        Dmitry Zhdan (dzhdan@nvidia.com)
+        Dmitrii Zhdan (dzhdan@nvidia.com)
         Tim Cheblokov (ttcheblokov@nvidia.com)
 
     Special thanks:
@@ -29,8 +29,8 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 
 #define NRD_VERSION_MAJOR 4
 #define NRD_VERSION_MINOR 8
-#define NRD_VERSION_BUILD 0
-#define NRD_VERSION_DATE "9 May 2024"
+#define NRD_VERSION_BUILD 1
+#define NRD_VERSION_DATE "16 May 2024"
 
 #if defined(_MSC_VER)
     #define NRD_CALL __fastcall

diff --git a/README.md b/README.md
@@ -1,4 +1,4 @@
-# NVIDIA REAL-TIME DENOISERS v4.8.0 (NRD)
+# NVIDIA REAL-TIME DENOISERS v4.8.1 (NRD)
 
 [![Build NRD SDK](https://github.com/NVIDIAGameWorks/RayTracingDenoiser/actions/workflows/build.yml/badge.svg)](https://github.com/NVIDIAGameWorks/RayTracingDenoiser/actions/workflows/build.yml)
 

diff --git a/Resources/Version.h b/Resources/Version.h
@@ -23,6 +23,6 @@ Versioning rules:
 
 #define VERSION_MAJOR                   4
 #define VERSION_MINOR                   8
-#define VERSION_BUILD                   0
+#define VERSION_BUILD                   1
 
 #define VERSION_STRING STR(VERSION_MAJOR.VERSION_MINOR.VERSION_BUILD encoding=NRD_NORMAL_ENCODING.NRD_ROUGHNESS_ENCODING)
diff --git a/Shaders/Include/REBLUR_Common_SpecularSpatialFilter.hlsli b/Shaders/Include/REBLUR_Common_SpecularSpatialFilter.hlsli
@@ -196,7 +196,7 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
             #ifdef REBLUR_SH
                 float4 sh = gIn_SpecSh.SampleLevel( gNearestClamp, checkerboardUvScaled, 0 );
                 sh = Denanify( w, sh );
-                specSh.xyz += sh.xyz * w; // see TA
+                specSh.xyz += sh.xyz * w;
             #endif
         }
 

diff --git a/Shaders/Include/REBLUR_HistoryFix.hlsli b/Shaders/Include/REBLUR_HistoryFix.hlsli
@@ -125,7 +125,11 @@ NRD_EXPORT void NRD_CS_MAIN( int2 threadPos : SV_GroupThreadId, int2 pixelPos :
             #ifdef REBLUR_PERFORMANCE_MODE
                 sumd = 1.0 + 1.0 / ( 1.0 + gMaxAccumulatedFrameNum ) - diffNonLinearAccumSpeed;
             #endif
+
             diff *= sumd;
+            #ifdef REBLUR_SH
+                diffSh *= sumd;
+            #endif
 
             [unroll]
             for( j = -2; j <= 2; j++ )
@@ -314,7 +318,11 @@ NRD_EXPORT void NRD_CS_MAIN( int2 threadPos : SV_GroupThreadId, int2 pixelPos :
             #ifdef REBLUR_PERFORMANCE_MODE
                 sums = 1.0 + 1.0 / ( 1.0 + gMaxAccumulatedFrameNum ) - specNonLinearAccumSpeed;
             #endif
+
             spec *= sums;
+            #ifdef REBLUR_SH
+                specSh.xyz *= sums;
+            #endif
 
             [unroll]
             for( j = -2; j <= 2; j++ )
@@ -372,15 +380,15 @@ NRD_EXPORT void NRD_CS_MAIN( int2 threadPos : SV_GroupThreadId, int2 pixelPos :
                     #ifdef REBLUR_SH
                         float4 sh = gIn_SpecSh[ pos ];
                         sh = Denanify( w, sh );
-                        specSh.xyz += sh.xyz * w; // see TA
+                        specSh.xyz += sh.xyz * w;
                     #endif
                 }
             }
 
             sums = STL::Math::PositiveRcp( sums );
             spec *= sums;
             #ifdef REBLUR_SH
-                specSh *= sums;
+                specSh.xyz *= sums;
             #endif
         }
 

diff --git a/Shaders/Include/REBLUR_TemporalAccumulation.hlsli b/Shaders/Include/REBLUR_TemporalAccumulation.hlsli
@@ -672,7 +672,7 @@ NRD_EXPORT void NRD_CS_MAIN( int2 threadPos : SV_GroupThreadId, int2 pixelPos :
             float4 specShResult = lerp( smbShSpec, vmbShSpec, virtualHistoryAmount );
 
             // ( Optional ) Output modified roughness to assist AA during SG resolve
-            specShResult.w = roughnessModified; // IMPORTANT: should not be blurred
+            specShResult.w = roughnessModified; // IMPORTANT: must not be blurred! this is why "specSh.xyz" is used ~everywhere
         #endif
 
         float specAccumSpeed = lerp( smbSpecAccumSpeed, vmbSpecAccumSpeed, virtualHistoryAmount );

diff --git a/Shaders/Include/REBLUR_TemporalStabilization.hlsli b/Shaders/Include/REBLUR_TemporalStabilization.hlsli
@@ -89,8 +89,8 @@ NRD_EXPORT void NRD_CS_MAIN( int2 threadPos : SV_GroupThreadId, int2 pixelPos :
 
         #ifdef REBLUR_SH
             float4 specSh = s_SpecSh[ smemPos.y ][ smemPos.x ];
-            float4 specShM1 = specSh;
-            float4 specShM2 = specSh * specSh;
+            float3 specShM1 = specSh.xyz;
+            float3 specShM2 = specSh.xyz * specSh.xyz;
         #endif
 
         float2 specMin = NRD_INF;
@@ -133,7 +133,7 @@ NRD_EXPORT void NRD_CS_MAIN( int2 threadPos : SV_GroupThreadId, int2 pixelPos :
                 specM2 += s * s;
 
                 #ifdef REBLUR_SH
-                    float4 sh = s_SpecSh[ pos.y ][ pos.x ];
+                    float3 sh = s_SpecSh[ pos.y ][ pos.x ].xyz;
                     specShM1 += sh;
                     specShM2 += sh * sh;
                 #endif
@@ -181,7 +181,7 @@ NRD_EXPORT void NRD_CS_MAIN( int2 threadPos : SV_GroupThreadId, int2 pixelPos :
         #ifdef REBLUR_SH
             specShM1 *= invSum;
             specShM2 *= invSum;
-            float4 specShSigma = GetStdDev( specShM1, specShM2 );
+            float3 specShSigma = GetStdDev( specShM1, specShM2 );
         #endif
 
         // RCRS
@@ -407,7 +407,7 @@ NRD_EXPORT void NRD_CS_MAIN( int2 threadPos : SV_GroupThreadId, int2 pixelPos :
         specHistory = STL::Color::Clamp( specM1, specSigma * specTemporalAccumulationParams.y, specHistory );
         float4 specResult = lerp( spec, specHistory, specHistoryWeight );
         #ifdef REBLUR_SH
-            specShHistory = STL::Color::Clamp( specShM1, specShSigma * specTemporalAccumulationParams.y, specShHistory );
+            specShHistory.xyz = STL::Color::Clamp( specShM1, specShSigma * specTemporalAccumulationParams.y, specShHistory.xyz );
             float4 specShResult = lerp( specSh, specShHistory, specHistoryWeight );
 
             // ( Optional ) Output modified roughness to assist AA during SG resolve