v4.9.1:

HIGHLIGHTS:
- improvements and bug fixes

DETAILS:
- NRD: improved per-frame random rotations offer a better distribution of samples across frames (barely visible in OCCLUSION and SIGMA)
- REFERENCE: split-screen and resolution scaling related bug fixes
- NRD INTEGRATION: updated NRI
- updated MathLib

Include/NRD.h

@@ -29,8 +29,8 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 
 #define NRD_VERSION_MAJOR 4
 #define NRD_VERSION_MINOR 9
-#define NRD_VERSION_BUILD 0
-#define NRD_VERSION_DATE "1 August 2024"
+#define NRD_VERSION_BUILD 1
+#define NRD_VERSION_DATE "14 August 2024"
 
 #if defined(_MSC_VER)
     #define NRD_CALL __fastcall

Integration/NRDIntegration.hpp

@@ -11,7 +11,7 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #include "NRDIntegration.h"
 
 static_assert(NRD_VERSION_MAJOR >= 4 && NRD_VERSION_MINOR >= 5, "Unsupported NRD version!");
-static_assert(NRI_VERSION_MAJOR >= 1 && NRI_VERSION_MINOR >= 125, "Unsupported NRI version!");
+static_assert(NRI_VERSION_MAJOR >= 1 && NRI_VERSION_MINOR >= 140, "Unsupported NRI version!");
 
 #ifdef _WIN32
     #define alloca _alloca
@@ -133,7 +133,7 @@ void NrdIntegration::CreatePipelines()
     uint32_t samplerOffset = 0;
     uint32_t textureOffset = 0;
     uint32_t storageTextureAndBufferOffset = 0;
-    if (m_NRI->GetDeviceDesc(*m_Device).graphicsAPI == nri::GraphicsAPI::VULKAN)
+    if (m_NRI->GetDeviceDesc(*m_Device).graphicsAPI == nri::GraphicsAPI::VK)
     {
         const nrd::LibraryDesc& nrdLibraryDesc = nrd::GetLibraryDesc();
         constantBufferOffset = nrdLibraryDesc.spirvBindingOffsets.constantBufferOffset;

README.md

@@ -1,4 +1,4 @@
-# NVIDIA REAL-TIME DENOISERS v4.9.0 (NRD)
+# NVIDIA REAL-TIME DENOISERS v4.9.1 (NRD)
 
 [![Build NRD SDK](https://github.com/NVIDIAGameWorks/RayTracingDenoiser/actions/workflows/build.yml/badge.svg)](https://github.com/NVIDIAGameWorks/RayTracingDenoiser/actions/workflows/build.yml)
 

Resources/Version.h

@@ -23,6 +23,6 @@ Versioning rules:
 
 #define VERSION_MAJOR                   4
 #define VERSION_MINOR                   9
-#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)

Shaders.cfg

@@ -166,7 +166,7 @@ REBLUR_Specular_SplitScreen.cs.hlsl -T cs
 REBLUR_Specular_TemporalAccumulation.cs.hlsl -T cs
 REBLUR_Specular_TemporalStabilization.cs.hlsl -T cs
 REBLUR_Validation.cs.hlsl -T cs
-REFERENCE_SplitScreen.cs.hlsl -T cs
+REFERENCE_Copy.cs.hlsl -T cs
 REFERENCE_TemporalAccumulation.cs.hlsl -T cs
 RELAX_ClassifyTiles.cs.hlsl -T cs
 RELAX_Diffuse_AntiFirefly.cs.hlsl -T cs

Shaders/Resources/REFERENCE_SplitScreen.resources.hlsli → Shaders/Resources/REFERENCE_Copy.resources.hlsli

@@ -8,7 +8,9 @@ distribution of this software and related documentation without an express
 license agreement from NVIDIA CORPORATION is strictly prohibited.
 */
 
-NRD_CONSTANTS_START( REFERENCE_SplitScreenConstants )
+NRD_CONSTANTS_START( REFERENCE_CopyConstants )
+    NRD_CONSTANT( float2, gRectSizeInv )
+    NRD_CONSTANT( float, gSplitScreen )
     NRD_CONSTANT( float, gDebug ) // only for availability in Common.hlsl
     NRD_CONSTANT( float, gViewZScale ) // only for availability in Common.hlsl
 NRD_CONSTANTS_END
@@ -22,11 +24,11 @@ NRD_OUTPUTS_START
 NRD_OUTPUTS_END
 
 // Macro magic
-#define REFERENCE_SplitScreenGroupX 16
-#define REFERENCE_SplitScreenGroupY 16
+#define REFERENCE_CopyGroupX 16
+#define REFERENCE_CopyGroupY 16
 
 // Redirection
 #undef GROUP_X
 #undef GROUP_Y
-#define GROUP_X REFERENCE_SplitScreenGroupX
-#define GROUP_Y REFERENCE_SplitScreenGroupY
+#define GROUP_X REFERENCE_CopyGroupX
+#define GROUP_Y REFERENCE_CopyGroupY

Shaders/Resources/REFERENCE_TemporalAccumulation.resources.hlsli

@@ -10,8 +10,6 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 
 NRD_CONSTANTS_START( REFERENCE_TemporalAccumulationConstants )
     NRD_CONSTANT( uint2, gRectOrigin )
-    NRD_CONSTANT( float2, gRectSizeInv )
-    NRD_CONSTANT( float, gSplitScreen )
     NRD_CONSTANT( float, gAccumSpeed )
     NRD_CONSTANT( float, gDebug ) // only for availability in Common.hlsl
     NRD_CONSTANT( float, gViewZScale ) // only for availability in Common.hlsl

Shaders/Source/REFERENCE_SplitScreen.cs.hlsl → Shaders/Source/REFERENCE_Copy.cs.hlsl

@@ -11,12 +11,15 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #include "NRD.hlsli"
 #include "ml.hlsli"
 
-#include "REFERENCE_SplitScreen.resources.hlsli"
+#include "REFERENCE_Copy.resources.hlsli"
 
 #include "Common.hlsli"
 
 [numthreads( GROUP_X, GROUP_Y, 1 )]
 NRD_EXPORT void NRD_CS_MAIN( uint2 pixelPos : SV_DispatchThreadId )
 {
-    gOut_Output[ pixelPos ] = gIn_Input[ pixelPos ];
+    float2 pixelUv = float2( pixelPos + 0.5 ) * gRectSizeInv;
+
+    if( pixelUv.x > gSplitScreen )
+        gOut_Output[ pixelPos ] = gIn_Input[ pixelPos ];
 }

Shaders/Source/REFERENCE_TemporalAccumulation.cs.hlsl

@@ -18,12 +18,9 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 [numthreads( GROUP_X, GROUP_Y, 1 )]
 NRD_EXPORT void NRD_CS_MAIN( uint2 pixelPos : SV_DispatchThreadId )
 {
-    float2 pixelUv = float2( pixelPos + 0.5 ) * gRectSizeInv;
-
     float4 input = gIn_Input[ pixelPos ];
     float4 history = gInOut_History[ pixelPos ];
     float4 result = lerp( history, input, gAccumSpeed );
 
-    gInOut_History[ pixelPos ] = pixelUv.x > gSplitScreen ? result : input;
+    gInOut_History[ pixelPos ] = result;
 }
-

Source/Denoisers/Reference.hpp

@@ -9,7 +9,7 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 */
 
 #include "../Shaders/Resources/REFERENCE_TemporalAccumulation.resources.hlsli"
-#include "../Shaders/Resources/REFERENCE_SplitScreen.resources.hlsli"
+#include "../Shaders/Resources/REFERENCE_Copy.resources.hlsli"
 
 void nrd::InstanceImpl::Add_Reference(DenoiserData& denoiserData)
 {
@@ -40,7 +40,7 @@ void nrd::InstanceImpl::Add_Reference(DenoiserData& denoiserData)
 
         PushOutput( AsUint(ResourceType::OUT_SIGNAL) );
 
-        AddDispatch( REFERENCE_SplitScreen, REFERENCE_SplitScreen, 1 ); // TODO: rename to Copy
+        AddDispatch( REFERENCE_Copy, REFERENCE_Copy, 1 );
     }
 
     #undef DENOISER_NAME
@@ -56,7 +56,9 @@ void nrd::InstanceImpl::Update_Reference(const DenoiserData& denoiserData)
 
     const ReferenceSettings& settings = denoiserData.settings.reference;
 
-    if (m_WorldToClip != m_WorldToClipPrev || m_CommonSettings.accumulationMode != AccumulationMode::CONTINUE)
+    if (m_WorldToClip != m_WorldToClipPrev || m_CommonSettings.accumulationMode != AccumulationMode::CONTINUE ||
+        m_CommonSettings.rectSize[0] != m_CommonSettings.rectSizePrev[0] ||
+        m_CommonSettings.rectSize[1] != m_CommonSettings.rectSizePrev[1])
         m_AccumulatedFrameNum = 0;
     else
         m_AccumulatedFrameNum = min(m_AccumulatedFrameNum + 1, settings.maxAccumulatedFrameNum);
@@ -66,13 +68,13 @@ void nrd::InstanceImpl::Update_Reference(const DenoiserData& denoiserData)
     { // ACCUMULATE
         REFERENCE_TemporalAccumulationConstants* consts = (REFERENCE_TemporalAccumulationConstants*)PushDispatch(denoiserData, AsUint(Dispatch::ACCUMULATE));
         consts->gRectOrigin     = uint2(m_CommonSettings.rectOrigin[0], m_CommonSettings.rectOrigin[1]);
-        consts->gRectSizeInv    = float2(1.0f / float(rectW), 1.0f / float(rectH));
-        consts->gSplitScreen    = m_CommonSettings.splitScreen;
         consts->gAccumSpeed     = 1.0f / (1.0f + float(m_AccumulatedFrameNum));
         consts->gDebug          = m_CommonSettings.debug;
     }
 
     { // COPY
-        PushDispatch(denoiserData, AsUint(Dispatch::COPY));
+        REFERENCE_CopyConstants* consts = (REFERENCE_CopyConstants*)PushDispatch(denoiserData, AsUint(Dispatch::COPY));
+        consts->gRectSizeInv    = float2(1.0f / float(rectW), 1.0f / float(rectH));
+        consts->gSplitScreen    = m_CommonSettings.splitScreen;
     }
 }

Source/InstanceImpl.cpp

@@ -283,20 +283,13 @@ nrd::Result nrd::InstanceImpl::SetCommonSettings(const CommonSettings& commonSet
 
     memcpy(&m_CommonSettings, &commonSettings, sizeof(commonSettings));
 
-    // Rotators
-    float4 rndAngle = Rng::Hash::GetFloat4(m_RngState) * radians(360.0f);
+    // Rotators (respecting sample patterns symmetry)
+    float angle1 = Sequence::Weyl1D(0.5f, commonSettings.frameIndex) * radians(90.0f);
+    m_Rotator_PrePass = Geometry::GetRotator(angle1);
 
-    float ca = cos( rndAngle.x );
-    float sa = sin( rndAngle.x );
-    m_Rotator_PrePass = float4( ca, sa, -sa, ca );
-
-    ca = cos( rndAngle.y );
-    sa = sin( rndAngle.y );
-    m_Rotator_Blur = float4( ca, sa, -sa, ca );
-
-    ca = cos( rndAngle.z );
-    sa = sin( rndAngle.z );
-    m_Rotator_PostBlur = float4( ca, sa, -sa, ca );
+    float angle2 = Sequence::Weyl1D(0.0f, commonSettings.frameIndex) * radians(90.0f);
+    m_Rotator_Blur = Geometry::GetRotator(angle2);
+    m_Rotator_PostBlur = Geometry::GetRotator(angle2 + radians(45.0f));
 
     // Main matrices
     m_ViewToClip = float4x4

Source/InstanceImpl.h

@@ -208,12 +208,6 @@ namespace nrd
         void Add_Reference(DenoiserData& denoiserData);
         void Update_Reference(const DenoiserData& denoiserData);
 
-        void Add_SpecularReflectionMv(DenoiserData& denoiserData);
-        void Update_SpecularReflectionMv(const DenoiserData& denoiserData);
-
-        void Add_SpecularDeltaMv(DenoiserData& denoiserData);
-        void Update_SpecularDeltaMv(const DenoiserData& denoiserData);
-
     // Internal
     public:
         inline InstanceImpl(const StdAllocator<uint8_t>& stdAllocator) :
@@ -230,8 +224,6 @@ namespace nrd
             , m_ActiveDispatches(GetStdAllocator())
             , m_IndexRemap(GetStdAllocator())
         {
-            Rng::Hash::Initialize(m_RngState, 106937, 69);
-
             m_ConstantDataUnaligned = m_StdAllocator.allocate(CONSTANT_DATA_SIZE + sizeof(float4));
 
             // IMPORTANT: underlying memory for constants must be aligned, as well as any individual SSE-type containing member,
@@ -316,7 +308,6 @@ namespace nrd
         Vector<DispatchDesc> m_ActiveDispatches;
         Vector<uint16_t> m_IndexRemap;
         Timer m_Timer;
-        uint32_t m_RngState = {};
         InstanceDesc m_Desc = {};
         CommonSettings m_CommonSettings = {};
         float4x4 m_ViewToClip = float4x4::Identity();

Source/Reference.cpp

@@ -13,17 +13,17 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 // REFERENCE
 #ifdef NRD_EMBEDS_DXBC_SHADERS
     #include "REFERENCE_TemporalAccumulation.cs.dxbc.h"
-    #include "REFERENCE_SplitScreen.cs.dxbc.h"
+    #include "REFERENCE_Copy.cs.dxbc.h"
 #endif
 
 #ifdef NRD_EMBEDS_DXIL_SHADERS
     #include "REFERENCE_TemporalAccumulation.cs.dxil.h"
-    #include "REFERENCE_SplitScreen.cs.dxil.h"
+    #include "REFERENCE_Copy.cs.dxil.h"
 #endif
 
 #ifdef NRD_EMBEDS_SPIRV_SHADERS
     #include "REFERENCE_TemporalAccumulation.cs.spirv.h"
-    #include "REFERENCE_SplitScreen.cs.spirv.h"
+    #include "REFERENCE_Copy.cs.spirv.h"
 #endif
 
 #include "Denoisers/Reference.hpp"