v4.7.0:

HIGHLIGHTS: - REBLUR: minor performance optimization - SIGMA: numerous improvements and bug fixes - SIGMA: temporal stabilization pass can be disabled, what makes SIGMA more useful for per-light shadow denoising if lights are many DETAILS: - REBLUR: number of steps in "prev-prev" test reduced to 1 (was 2) - SIGMA: taken kernel from REBLUR - SIGMA: per pixel rotations replaced with per frame rotations - SIGMA: tuned random rotators (affects REBLUR too) - SIGMA: improved "umbra in wide penumbra" behavior - SIGMA: clarified usage of "SIGMA_BackEnd_UnpackShadow" - SIGMA: reduced bias - SIGMA: removed "SigmaSettings::blurRadiusScale" - SIGMA: exposed "SigmaSettings::lightDirection" - SIGMA: fixed mismatched behavior between opaque and translucent shadows - SIGMA: anisotropic filtering support (currently only for directional light sources) - SIGMA: reduced flickering in areas with a small blur radius (if TS is on) - SIGMA: exposed "stabilizationStrength" (TS pass is disabled if 0) - SIGMA: reduced umbra blurring when in wide penumbra - NRD INTEGRATION: bug fixes for debug logging - Updated deps - Updated docs
NVIDIAGameWorks · Apr 15, 2024 · f20158d · f20158d
1 parent db4f66f
commit f20158d
Show file tree

Hide file tree

Showing 23 changed files with 239 additions and 189 deletions.
diff --git a/External/MathLib b/External/MathLib
diff --git a/Include/NRD.h b/Include/NRD.h
@@ -28,9 +28,9 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #include <cstddef>
 
 #define NRD_VERSION_MAJOR 4
-#define NRD_VERSION_MINOR 6
-#define NRD_VERSION_BUILD 1
-#define NRD_VERSION_DATE "25 March 2024"
+#define NRD_VERSION_MINOR 7
+#define NRD_VERSION_BUILD 0
+#define NRD_VERSION_DATE "17 April 2024"
 
 #if defined(_MSC_VER)
     #define NRD_CALL __fastcall

diff --git a/Include/NRDDescs.h b/Include/NRDDescs.h
@@ -11,7 +11,7 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #pragma once
 
 #define NRD_DESCS_VERSION_MAJOR 4
-#define NRD_DESCS_VERSION_MINOR 6
+#define NRD_DESCS_VERSION_MINOR 7
 
 static_assert(NRD_VERSION_MAJOR == NRD_DESCS_VERSION_MAJOR && NRD_VERSION_MINOR == NRD_DESCS_VERSION_MINOR, "Please, update all NRD SDK files");
 

diff --git a/Include/NRDSettings.h b/Include/NRDSettings.h
@@ -11,7 +11,7 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #pragma once
 
 #define NRD_SETTINGS_VERSION_MAJOR 4
-#define NRD_SETTINGS_VERSION_MINOR 6
+#define NRD_SETTINGS_VERSION_MINOR 7
 
 static_assert(NRD_VERSION_MAJOR == NRD_SETTINGS_VERSION_MAJOR && NRD_VERSION_MINOR == NRD_SETTINGS_VERSION_MINOR, "Please, update all NRD SDK files");
 
@@ -223,6 +223,7 @@ namespace nrd
         float responsiveAccumulationRoughnessThreshold = 0.0f;
 
         // (normalized %) - stabilizes output, more stabilization improves antilag (clean signals can use lower values)
+        // 0 - disables the stabilization pass
         float stabilizationStrength = 1.0f;
 
         // (normalized %) - represents maximum allowed deviation from local tangent plane
@@ -260,11 +261,16 @@ namespace nrd
 
     struct SigmaSettings
     {
+        // Direction to the light source
+        // IMPORTANT: it is needed only for directional light sources (sun)
+        float lightDirection[3] = {0.0f, 0.0f, 0.0f};
+
         // (normalized %) - represents maximum allowed deviation from local tangent plane
         float planeDistanceSensitivity = 0.005f;
 
-        // [1; 3] - adds bias and stability if > 1
-        float blurRadiusScale = 2.0f;
+        // (normalized %) - stabilizes output, more stabilization improves antilag (clean signals can use lower values)
+        // 0 - disables the stabilization pass and makes denoising spatial only (no history)
+        float stabilizationStrength = 1.0f;
     };
 
     // RELAX

diff --git a/Integration/NRDIntegration.h b/Integration/NRDIntegration.h
@@ -23,8 +23,8 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #include <map>
 
 #define NRD_INTEGRATION_MAJOR 1
-#define NRD_INTEGRATION_MINOR 11
-#define NRD_INTEGRATION_DATE "19 March 2024"
+#define NRD_INTEGRATION_MINOR 12
+#define NRD_INTEGRATION_DATE "17 April 2024"
 #define NRD_INTEGRATION 1
 
 #define NRD_INTEGRATION_DEBUG_LOGGING 0

diff --git a/Integration/NRDIntegration.hpp b/Integration/NRDIntegration.hpp
@@ -312,7 +312,7 @@ void NrdIntegration::CreateResources(uint16_t resourceWidth, uint16_t resourceHe
             m_TransientPoolSize += memoryDesc.size;
 
     #if( NRD_INTEGRATION_DEBUG_LOGGING == 1 )
-        printf("%s %ux%u format=%u mips=%u\n", name, nrdTextureDesc.width, nrdTextureDesc.height, nrdTextureDesc.format, nrdTextureDesc.mipNum);
+        printf("%s format=%u downsampleFactor=%u\n", name, nrdTextureDesc.format, nrdTextureDesc.downsampleFactor);
     #endif
     }
 
@@ -403,7 +403,7 @@ void NrdIntegration::NewFrame()
     NRD_INTEGRATION_ASSERT(m_Instance, "Uninitialized! Did you forget to call 'Initialize'?");
 
 #if( NRD_INTEGRATION_DEBUG_LOGGING == 1 )
-        printf("%s (frame %u) ==============================================================================\n\n", m_Name, frameIndex);
+        printf("%s (frame %u) ==============================================================================\n\n", m_Name, m_FrameIndex);
     #endif
 
     m_DescriptorPoolIndex = m_FrameIndex % m_BufferedFramesNum;
@@ -616,12 +616,7 @@ void NrdIntegration::Dispatch(nri::CommandBuffer& commandBuffer, nri::Descriptor
             if( r.type == nrd::ResourceType::PERMANENT_POOL )
                 printf("P(%u) ", r.indexInPool);
             else if( r.type == nrd::ResourceType::TRANSIENT_POOL )
-            {
-                if (r.mipNum != 1 || r.mipOffset != 0)
-                    printf("T(%u)[%u:%u] ", r.indexInPool, r.mipOffset, r.mipNum);
-                else
-                    printf("T(%u) ", r.indexInPool);
-            }
+                printf("T(%u) ", r.indexInPool);
             else
             {
                 const char* s = nrd::GetResourceTypeString(r.type);

diff --git a/README.md b/README.md
@@ -1,4 +1,4 @@
-# NVIDIA REAL-TIME DENOISERS v4.6.1 (NRD)
+# NVIDIA REAL-TIME DENOISERS v4.7.0 (NRD)
 
 [![Build NRD SDK](https://github.com/NVIDIAGameWorks/RayTracingDenoiser/actions/workflows/build.yml/badge.svg)](https://github.com/NVIDIAGameWorks/RayTracingDenoiser/actions/workflows/build.yml)
 
@@ -18,7 +18,8 @@ For quick starting see *[NRD sample](https://github.com/NVIDIAGameWorks/NRDSampl
 Performance on RTX 4080 @ 1440p (native resolution, default denoiser settings):
 - `REBLUR_DIFFUSE_SPECULAR` - 2.45 ms
 - `RELAX_DIFFUSE_SPECULAR` - 2.90 ms
-- `SIGMA_SHADOW` - 0.30 ms
+- `SIGMA_SHADOW` - 0.30 ms (0.24 mns if temporal stabilization is off)
+- `SIGMA_SHADOW_TRANSLUCENCY` - 0.40 ms (0.30 ms if temporal stabilization is off)
 
 Supported signal types:
 - *RELAX*:
@@ -589,7 +590,7 @@ Denoising is not a panacea or miracle. Denoising works best with ray tracing res
     float3 preintegratedBRDF = PreintegratedBRDF( Rf0, N, V, roughness )
     Denoising( specularRadiance * BRDF ) → NRD( specularRadiance * BRDF / preintegratedBRDF ) * preintegratedBRDF
 
-A good approximation for pre-integrated specular BRDF can be found *[here](https://github.com/NVIDIAGameWorks/Falcor/blob/056f7b7c73b69fa8140d211bbf683ddf297a2ae0/Source/Falcor/Rendering/Materials/Microfacet.slang#L213)*.
+A good approximation for pre-integrated specular BRDF can be found *[here](https://github.com/NVIDIAGameWorks/MathLib/blob/407ecd0d1892d12ee1ec98c3d46cbeed73b79a0d/STL.hlsli#L2147*. Pre-integrated specular BRDF can also be referenced as "specular albedo" or "environment BRDF".
 
 ## COMBINED DENOISING OF DIRECT AND INDIRECT LIGHTING
 
@@ -710,7 +711,7 @@ Hair strands tangent vectors *can't* be used as "normals guide" for *NRD* due to
 
 **[NRD]** Hit distances should come from an importance sampling method. But if denoising of AO/SO is needed, AO/SO can come from cos-weighted (or VNDF) sampling in a tradeoff of IQ.
 
-**[NRD]** Low discrepancy sampling (blue noise) helps to have more stable output in 0.5-1 rpp mode. It's a must for REBLUR-based Ambient and Specular Occlusion denoisers and SIGMA.
+**[NRD]** Low discrepancy sampling (blue noise) helps to get more stable output in 0.5-1 rpp mode. It's a must for REBLUR-based Ambient and Specular Occlusion denoisers and SIGMA.
 
 **[NRD]** It's recommended to set `CommonSettings::accumulationMode` to `RESET` for a single frame, if a history reset is needed. If history buffers are recreated or contain garbage, it's recommended to use `CLEAR_AND_RESET` for a single frame. `CLEAR_AND_RESET` is not free because clearing is done in a compute shader. Render target clears on the application side should be prioritized over this solution.
 
@@ -742,13 +743,11 @@ maxAccumulatedFrameNum > maxFastAccumulatedFrameNum > historyFixFrameNum
 
 **[REBLUR]** *REBLUR* expects hit distances in a normalized form. To avoid mismatching, `REBLUR_FrontEnd_GetNormHitDist` must be used for normalization. Normalization parameters should be passed into *NRD* as `HitDistanceParameters` for internal hit distance denormalization. Some tweaking can be needed here, but in most cases default `HitDistanceParameters` works well. *REBLUR* outputs denoised normalized hit distance, which can be used by the application as ambient or specular occlusion (AO & SO) (see unpacking functions from `NRD.hlsli`).
 
-**[REBLUR]** Intensity antilag parameters need to be carefully tuned. The defaults are good but `AntilagIntensitySettings::sensitivityToDarkness` needs to be tuned for a given HDR range. Initial integration should work with intensity antilag turned off.
-
-**[REBLUR]** Even if antilag is off, it's recommended to tune `AntilagIntensitySettings::sensitivityToDarkness`, because it is used for error estimation.
+**[REBLUR/RELAX]** Antilag parameters need to be carefully tuned. Initial integration should be done with disabled antilag.
 
 **[RELAX]** *RELAX* works well with signals produced by *RTXDI* or very clean high RPP signals. The Sweet Home of *RELAX* is *RTXDI* sample. Please, consider getting familiar with this application.
 
-**[SIGMA]** Using "blue" noise can help to avoid shadow shimmering, it works best if the pattern is static on the screen. Additionally, `blurRadiusScale` can be set to `2-4` to mitigate such problems in complicated cases.
+**[SIGMA]** Using "blue" noise can help to avoid shadow shimmering. It works best if the pattern is static on the screen.
 
 **[SIGMA]** *SIGMA_TRANSLUCENT_SHADOW* can be used for shadow denoising from multiple light sources:
 
@@ -798,3 +797,7 @@ Is this a biased solution? If spatial filtering is off - no, because we just reo
 - if shadows overlap, a separate pass is needed to analyze noisy input and classify pixels as *umbra* - *penumbra* (and optionally *empty space*). Raster shadow maps can be used for this if available
 - it is not recommended to mix 1 cd and 100000 cd lights, since FP32 texture will be needed for a weighted sum.
 In this case, it's better to process the sun and other bright light sources separately.
+
+**[SIGMA]** *SIGMA* can be used for multi-light shadow denoising if applied "per light". `SigmaSettings::stabilizationStrength` can be set to `0` to disable temporal history. It provides the followinmg benefits:
+ - light count independent memory usage
+ - no need to manage history buffers for lights
diff --git a/Resources/Version.h b/Resources/Version.h
@@ -22,7 +22,7 @@ Versioning rules:
 */
 
 #define VERSION_MAJOR                   4
-#define VERSION_MINOR                   6
-#define VERSION_BUILD                   1
+#define VERSION_MINOR                   7
+#define VERSION_BUILD                   0
 
 #define VERSION_STRING STR(VERSION_MAJOR.VERSION_MINOR.VERSION_BUILD encoding=NRD_NORMAL_ENCODING.NRD_ROUGHNESS_ENCODING)
diff --git a/Shaders/Include/Common.hlsli b/Shaders/Include/Common.hlsli
@@ -129,6 +129,34 @@ Usage:
             printf(__VA_ARGS__)
 #endif
 
+//==================================================================================================================
+// KERNELS
+//==================================================================================================================
+
+static const float3 g_Special6[ 6 ] =
+{
+    // https://www.desmos.com/calculator/e5mttzlg6v
+    float3( -0.50 * sqrt( 3.0 ) , -0.50             , 1.0 ),
+    float3(  0.00               ,  1.00             , 1.0 ),
+    float3(  0.50 * sqrt( 3.0 ) , -0.50             , 1.0 ),
+    float3(  0.00               , -0.30             , 0.3 ),
+    float3(  0.15 * sqrt( 3.0 ) ,  0.15             , 0.3 ),
+    float3( -0.15 * sqrt( 3.0 ) ,  0.15             , 0.3 ),
+};
+
+static const float3 g_Special8[ 8 ] =
+{
+    // https://www.desmos.com/calculator/abaqyvswem
+    float3( -1.00               ,  0.00               , 1.0 ),
+    float3(  0.00               ,  1.00               , 1.0 ),
+    float3(  1.00               ,  0.00               , 1.0 ),
+    float3(  0.00               , -1.00               , 1.0 ),
+    float3( -0.25 * sqrt( 2.0 ) ,  0.25 * sqrt( 2.0 ) , 0.5 ),
+    float3(  0.25 * sqrt( 2.0 ) ,  0.25 * sqrt( 2.0 ) , 0.5 ),
+    float3(  0.25 * sqrt( 2.0 ) , -0.25 * sqrt( 2.0 ) , 0.5 ),
+    float3( -0.25 * sqrt( 2.0 ) , -0.25 * sqrt( 2.0 ) , 0.5 )
+};
+
 //==================================================================================================================
 // SHARED FUNCTIONS
 //==================================================================================================================

diff --git a/Shaders/Include/NRD.hlsli b/Shaders/Include/NRD.hlsli
@@ -8,7 +8,7 @@ distribution of this software and related documentation without an express
 license agreement from NVIDIA CORPORATION is strictly prohibited.
 */
 
-// NRD v4.6
+// NRD v4.7
 
 //=================================================================================================================================
 // INPUT PARAMETERS
@@ -56,7 +56,7 @@ float distanceToOccluder:
     - distance to occluder, must follow the rules:
         - NoL <= 0         - 0 ( it's very important )
         - NoL > 0 ( hit )  - hit distance
-        - NoL > 0 ( miss ) - NRD_FP16_MAX
+        - NoL > 0 ( miss ) - >= NRD_FP16_MAX
 
 float tanOfLightAngularRadius:
     - tan( lightAngularSize * 0.5 )
@@ -747,12 +747,12 @@ float2 SIGMA_FrontEnd_PackShadow( float viewZ, float distanceToOccluder, float t
     r.x = 0.0;
     r.y = _NRD_PackViewZ( viewZ );
 
-    if( distanceToOccluder == NRD_FP16_MAX )
+    if( distanceToOccluder >= NRD_FP16_MAX )
         r.x = NRD_FP16_MAX;
     else if( distanceToOccluder != 0.0 )
     {
-        float distanceToOccluderProj = distanceToOccluder * tanOfLightAngularRadius;
-        r.x = min( distanceToOccluderProj, 32768.0 );
+        float penumbraRadius = distanceToOccluder * tanOfLightAngularRadius;
+        r.x = min( penumbraRadius, 32768.0 );
     }
 
     return r;
@@ -762,7 +762,7 @@ float2 SIGMA_FrontEnd_PackShadow( float viewZ, float distanceToOccluder, float t
 float2 SIGMA_FrontEnd_PackShadow( float viewZ, float distanceToOccluder, float tanOfLightAngularRadius, float3 translucency, out float4 out2 )
 {
     // IN_SHADOW_TRANSLUCENCY
-    out2.x = float( distanceToOccluder == NRD_FP16_MAX );
+    out2.x = float( distanceToOccluder >= NRD_FP16_MAX );
     out2.yzw = saturate( translucency );
 
     // IN_SHADOWDATA
@@ -879,16 +879,11 @@ NRD_SG RELAX_BackEnd_UnpackSh( float4 sh0, float4 sh1 )
 //=================================================================================================================================
 
 // OUT_SHADOW_TRANSLUCENCY => X
-//   SIGMA_SHADOW:
-//      float shadowData = SIGMA_BackEnd_UnpackShadow( shadowData );
-//      shadow = shadowData;
+//   SIGMA_SHADOW / SIGMA_SHADOW_TRANSLUCENCY:
+//      float shadow = SIGMA_BackEnd_UnpackShadow( OUT_SHADOW_TRANSLUCENCY );
 //   SIGMA_SHADOW_TRANSLUCENCY:
-//      float4 shadowData = SIGMA_BackEnd_UnpackShadow( shadowData );
-//      float3 finalShadowCommon = lerp( shadowData.yzw, 1.0, shadowData.x ); // or
-//      float3 finalShadowExotic = shadowData.yzw * shadowData.x; // or
-//      float3 finalShadowMoreExotic = shadowData.yzw;
-// IMPORTANT: use "^ 3" to compensate over-blurring ( it really makes the result closer to the reference )
-#define SIGMA_BackEnd_UnpackShadow( color )  ( color * color * color )
+//      float3 translucentShadow = SIGMA_BackEnd_UnpackShadow( OUT_SHADOW_TRANSLUCENCY ).yzw;
+#define SIGMA_BackEnd_UnpackShadow( shadow ) ( shadow * shadow )
 
 //=================================================================================================================================
 // BACK-END - HIGH QUALITY RESOLVE

diff --git a/Shaders/Include/REBLUR_Common.hlsli b/Shaders/Include/REBLUR_Common.hlsli
@@ -14,31 +14,6 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #define REBLUR_BLUR                                     1
 #define REBLUR_POST_BLUR                                2
 
-// Kernels
-static const float3 g_Special6[ 6 ] =
-{
-    // https://www.desmos.com/calculator/e5mttzlg6v
-    float3( -0.50 * sqrt( 3.0 ) , -0.50             , 1.0 ),
-    float3(  0.00               ,  1.00             , 1.0 ),
-    float3(  0.50 * sqrt( 3.0 ) , -0.50             , 1.0 ),
-    float3(  0.00               , -0.30             , 0.3 ),
-    float3(  0.15 * sqrt( 3.0 ) ,  0.15             , 0.3 ),
-    float3( -0.15 * sqrt( 3.0 ) ,  0.15             , 0.3 ),
-};
-
-static const float3 g_Special8[ 8 ] =
-{
-    // https://www.desmos.com/calculator/abaqyvswem
-    float3( -1.00               ,  0.00               , 1.0 ),
-    float3(  0.00               ,  1.00               , 1.0 ),
-    float3(  1.00               ,  0.00               , 1.0 ),
-    float3(  0.00               , -1.00               , 1.0 ),
-    float3( -0.25 * sqrt( 2.0 ) ,  0.25 * sqrt( 2.0 ) , 0.5 ),
-    float3(  0.25 * sqrt( 2.0 ) ,  0.25 * sqrt( 2.0 ) , 0.5 ),
-    float3(  0.25 * sqrt( 2.0 ) , -0.25 * sqrt( 2.0 ) , 0.5 ),
-    float3( -0.25 * sqrt( 2.0 ) , -0.25 * sqrt( 2.0 ) , 0.5 )
-};
-
 // Storage
 
 #define REBLUR_MAX_ACCUM_FRAME_NUM                      63.0
@@ -360,7 +335,9 @@ float2x3 GetKernelBasis( float3 D, float3 N, float NoD, float roughness = 1.0, f
         B = cross( R, T );
 
         float skewFactor = lerp( 0.5 + 0.5 * roughness, 1.0, NoD );
-        T *= lerp( skewFactor, 1.0, anisoFade );
+        skewFactor = lerp( skewFactor, 1.0, anisoFade );
+
+        T *= skewFactor; // TODO: B /= skewFactor?
     }
 
     return float2x3( T, B );

diff --git a/Shaders/Include/REBLUR_Config.hlsli b/Shaders/Include/REBLUR_Config.hlsli
@@ -68,7 +68,7 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #define REBLUR_HIT_DIST_MIN_WEIGHT( smc )                       ( 0.1 * smc ) // was 0.1
 
 #define REBLUR_MAX_PERCENT_OF_LOBE_VOLUME                       0.75
-#define REBLUR_VIRTUAL_MOTION_PREV_PREV_WEIGHT_ITERATION_NUM    2
+#define REBLUR_VIRTUAL_MOTION_PREV_PREV_WEIGHT_ITERATION_NUM    1
 #define REBLUR_COLOR_CLAMPING_SIGMA_SCALE                       2.0 // using smaller values leads to bias if camera rotates slowly due to reprojection instabilities
 #define REBLUR_FIREFLY_SUPPRESSOR_MAX_RELATIVE_INTENSITY        float2( 10.0, 1.1 )
 #define REBLUR_FIREFLY_SUPPRESSOR_RADIUS_SCALE                  0.1